The Role of Primary Cancer Diagnosis in Determining Costs

The Role of Primary Cancer Diagnosis in Determining Costs and Caregiver Burden Associated with

Palliative Home Care

by

Ruby Redmond-Misner

A thesis submitted in conformity with the requirements for the degree of Master of Science

Institute of Health Policy, Management and Evaluation University of Toronto

© Copyright by Ruby Redmond-Misner 2014

ii

The Role of Primary Cancer Diagnosis in Determining Costs and Caregiver Burden Associated with Palliative Home Care

Ruby Redmond-Misner

Master of Science

Institute of Health Policy, Management and Evaluation University of Toronto

2014

Abstract Community-based palliative home care (CBPHC) has been repeatedly endorsed for expansion

in Ontario. The objective of this thesis is to assess the role of primary cancer diagnosis in the

costs and caregiver burden associated with CBPHC.

Patients from two Ontario CBPHC programs were categorized using the World Health

Organization’s International Classification of Diseases (10th Ed.), differentiating tumour sites (i.e.

breast) rather than tumour type (i.e. carcinoma). Diagnosis was integrated into econometric

models alongside other important covariates, identified by conceptual frameworks of cost and

burden, in order to ascertain its role in both.

Findings suggest that there are differential care demands associated with recipients of CBPHC

that are related to primary cancer diagnosis. Several tumour sites emerged as significant drivers

of costs and caregiver burden. This is useful for the level of care and associated strain that can

be expected upon admission into CBPHC, the maintenance CBPHC and its unpaid caregivers,

and the personalization of care.

iii

Acknowledgements

In culminating my Master’s degree, I have so many people to thank. Thank you to anybody that has encouraged me and supported me along the way.

I gratefully acknowledge my thesis supervisor, Dr. Peter C. Coyte, for your guidance and for always reviewing my work promptly and thoroughly. I could not have completed my degree so quickly without your dedication. I also thank my committee member, Dr. Audrey Laporte, for econometric counsel, encouragement and for involving me with the Canadian Centre for Health Economics.

My sincerest appreciation goes to the members of my examination committee: Drs. Jan Barnsley (University of Toronto) and Daryl Bainbridge (McMaster). Thank you for your time and for providing valuable insight and feedback.

I would like to thank those at the Canadian Centre for Applied Research in Cancer Control (ARCC) who facilitated my incredibly helpful mock defense. Thank you Drs. Jeffrey Hoch, Wanrudee Isaranuwatchai and Rebecca Mercer and Jaclyn Beca for asking challenging questions and giving useful advice.

Thank you to my fellow students for being so encouraging and such great role models. I felt very welcome as soon as I started at IHPME and I wish you all the best. We will be in touch.

I gratefully acknowledge Adrian Rohit Dass for his econometric counsel, recommendations and revisions of my use of Allison’s hybrid model.

Last but not least, I thank my mother Kelly, aunt Karen, and sister Rachel for providing support and comic relief throughout and prior to my pursuit of higher education. Mom and Karen, thank you for the sacrifices you have made for Rachel and I to have privileges that were not afforded to you. We love and appreciate you very much.

iv

Table of Contents Acknowledgements ....................................................................................................................... iii

List of Tables ................................................................................................................................. vi

List of Figures ............................................................................................................................... vii

List of Appendices ....................................................................................................................... viii

Chapter 1 ....................................................................................................................................... 1 1.1 What is palliative care? ........................................................................................................ 1 1.2 Background: Palliative home care in Canada ...................................................................... 2 1.3 Research purpose and rationale ......................................................................................... 5 1.4 Research questions and hypotheses .................................................................................. 6

Chapter 2 ....................................................................................................................................... 9 2.1 Search strategy .................................................................................................................... 9 2.2 Palliative home care cost ................................................................................................... 12 2.3 Caregiver burden in palliative care .................................................................................... 19 2.4 Neoplasm specific palliative care research ....................................................................... 27 2.5 Summary ........................................................................................................................... 32

Chapter 3 ..................................................................................................................................... 33 3.1 Overview of the larger study .............................................................................................. 33 3.2 Data ................................................................................................................................... 34

3.2.1 Data origin ................................................................................................................... 34 3.2.2 Data description .......................................................................................................... 34 3.2.3 Variable description ..................................................................................................... 36 3.2.4 Variable assessment ................................................................................................... 42 3.2.5 Variable inclusion ........................................................................................................ 42 3.2.6 Categorical variable interpretation and baseline ......................................................... 43

3.3 Descriptive statistics .......................................................................................................... 44 3.4 Cost analyses .................................................................................................................... 44

3.4.1 Logarithmic transformation and quadratic forms ......................................................... 44 3.4.2 Aggregated costs ........................................................................................................ 45 3.4.3 Disaggregated cost ..................................................................................................... 48

3.5 Caregiver burden analysis ................................................................................................. 51 3.6 Diagnostics ........................................................................................................................ 52

3.6.1 Hausman test .............................................................................................................. 52 3.6.2 Allison’s hybrid method ............................................................................................... 52 3.6.3 Standard diagnostics .................................................................................................. 54

3.7 Summary ........................................................................................................................... 55

Chapter 4 ..................................................................................................................................... 56 4.1 Descriptive statistics .......................................................................................................... 56 4.2 Cost analyses .................................................................................................................... 58

4.2.1 Aggregated costs ........................................................................................................ 59

v

4.2.2 Service-specific cost analysis results .......................................................................... 66 4.3 Caregiver burden ............................................................................................................... 73 4.4 Diagnostics/Additional information .................................................................................... 74 4.5 Summary ........................................................................................................................... 78

Chapter 5 ..................................................................................................................................... 81 5.1 Findings vis-à-vis the hypotheses ..................................................................................... 82 5.2 Comparisons and inferences from the literature ................................................................ 85 5.3 Policy implications ............................................................................................................. 90 5.4 Limitations .......................................................................................................................... 95 5.5 Summary ........................................................................................................................... 96

Chapter 6 ..................................................................................................................................... 97 6.1 Thesis summary ................................................................................................................ 97 6.2 Future work ...................................................................................................................... 100

References ................................................................................................................................ 104

vi

List of Tables

Table 1: ICD-10 categories ......................................................................................................... 37

Table 2: AHCR framework for the identification of palliative care costs ...................................... 38

Table 3: Caregiver burden scale in end-of-life care .................................................................... 40

Table 4: Categorical variable baselines ...................................................................................... 44

Table 5: Summary of patient and caregiver demographics ......................................................... 57

Table 6: Hausman test results, final estimator(s) ........................................................................ 59

Table 7: Determinants of societal costs of CBPHC ..................................................................... 60

Table 8: Determinants of CBPHC Ministry of Health costs ......................................................... 62

Table 9: Determinants of CBPHC unpaid caregiver costs .......................................................... 65

Table 10: Determinants of receiving any medications and their cost .......................................... 67

Table 11: Probability of hospitalization among CBPHC recipients .............................................. 69

Table 12: Determinants of having a home appointment and its associated cost ........................ 71

Table 13: Determinants of caregiver burden in CBPHC ............................................................. 73

Table 14: Significant findings across all analyses ....................................................................... 78

vii

List of Figures

Figure 1: PRISMA flow chart for home care cost literature search ............................................. 13

Figure 2: PRISMA flow chart for caregiver burden literature search ........................................... 20

Figure 3: PRISMA flow chart for palliative oncology literature search ........................................ 28

Figure 4: Observations per time period ....................................................................................... 35

Figure 5: Charlson comorbidity index .......................................................................................... 41

Figure 6: Algorithm for deciding which estimator(s) to use ......................................................... 54

Figure 7: Raw and log-positive distribution of public medication cost ......................................... 76

Figure 8: Raw and log-positive distribution of public home appointment cost ............................ 77

Figure 9: Raw and log scale burden scores ................................................................................ 78

viii

List of Appendices Appendix A ................................................................................................................................ 117

Appendix B ................................................................................................................................ 123

Appendix C ................................................................................................................................ 129

Appendix D ................................................................................................................................ 134

Appendix E ................................................................................................................................ 137

Appendix F ................................................................................................................................ 138

Appendix G ................................................................................................................................ 139

Appendix H ................................................................................................................................ 141

Appendix I ................................................................................................................................. 143

1

Chapter 1 Introduction

The objective of this thesis is to comprehensively analyze the financial and caregiver

burdens associated with community-based palliative home care (CBPHC) for people

with cancer so as to identify significant determinants. This will be done with special

attention paid to the primary cancer diagnosis of the patients, which is hypothesized to

have a role in palliative care needs and outcomes. These are defined by solid tumour

site (i.e. breast or brain) as opposed to type (i.e. carcinoma or sarcoma) in congruence

with the World Health Organization’s (WHO) International Classification of Diseases (10th

Ed.) (ICD-10). There is little in the existing literature that assesses multiple neoplastic

primary diagnoses simultaneously to allow comparative inferences to be made. In fact,

treating all patients with cancer as a homogenous group is the norm.

1.1 What is palliative care?

A note on terminology

The term palliative care is often used interchangeably with end-of-life (EoL) care (Health

Canada, 2009). It is important to define precisely due to the ambiguity and obscurity

introduced by this latter term. The broad goal of palliative care is to improve quality of life

(QoL) for people facing terminal illnesses through pain and symptom management,

normalizing death and supporting those affected by the impending death into

bereavement (WHO, 2014; Health Canada, 2009). Referral into palliative care may

therefore commence at the time that curative treatment has stopped, shifting to symptom

management while the patient has some time left to live. For palliative patients with

cancer, this is often when neoplastic malignancies have metastasized, relapsed or are

inoperable (Gaertner et al, 2011). Conversely, counseling and other supports may be

offered to family members and unpaid caregivers beyond the death of the patient (WHO,

2014; Health Canada, 2009). These are points in the palliative trajectory that would not

constitute the EoL, which is why this term can obscure palliative care into something that

is provided only when death is imminent.

Chapter 1: Introduction R. Redmond-Misner

2

The Temmy Latner Centre for Palliative Care (TLCPC), from which data for this thesis

were derived, provides an idea of the services that go into providing palliative care at

home (2013). An interdisciplinary team of doctors, nurses, personal support workers

(PSWs), grief counselors, spiritual community members and volunteers may visit the

patient in their home to assist with activities of daily living (ADLs) or provide medical

care. The final, fundamental members of CBPHC teams (CBPHCTs) are unpaid

caregivers: family and friends who care for the patients in lieu of the formal CBPHCT.

Unpaid caregivers are becoming increasingly central to facilitating health care at home

(Guerriere, 2012), which is an important piece of information to precede the following

overview of palliative home care in Canada.

1.2 Background: Palliative home care in Canada

Current context

Restructuring of health care in Canada has been partially characterized by shifting

services into the community (Guerriere, 2012; Laville et al, 2007; Spalding, 2005;

Skinner, 2005). This is true of palliative care, which is commonly used by cancer patients

with terminal prognoses (Alesi et al, 2011; Balducci, 2003; Brink et al, 2008). While

concerns about the practicality of shifting a greater volume of services into the

community are emerging, it remains a popular option among influential health and public

policy research institutions. Recent endorsements for the further development of CBPHC

include the Health Council of Canada’s (2013) Progress Report, Health Quality Ontario’s

(HQO) (2012) Report on Ontario’s Health System, the Ontario Seniors’ Secretariat’s

(2013) Action Plan for Seniors, and Drummond et al’s (2012) Commission on the Reform

of Ontario’s Public Services.

Improving palliative care is a global public health priority, on the agendas of the WHO

and United Nations (UN) (Broad et al, 2013). Home- and community-based service

shifting has been proposed and implemented as a means of improving health care

sustainability internationally (Docherty et al, 2008; Drummond et al, 2012), which will

become apparent in the literature review. The Organization for Economic Cooperation

and Development (OECD) has written extensively on this as part of a broader shifting of


3

responsibility for health and social services onto community members, referring to

unpaid service providers as the “social economy” (Greffe, 2007).

Broad issues being raised

Ontario’s health critics, across political parties, express the idea that a demographic

crisis with serious implications for palliative care is pending. “Unless something is done

in the next few years, we’re going to be in a crisis situation on a number of fronts

including palliative care” said Christine Elliott, Ontario Progressive Conservative health

critic (Nash et al, 2013). France Gelinas, the Ontario New Democratic Party (NDP)

health critic (Nash et al, 2013), made a similar statement. Apprehension about an

impending demographic crisis is driven by unprecedented aging, cancer incidence and

life expectancies allowing for more chronic illness, but also by labour force changes

wherein most people work and people have fewer children (Jiwani, 2003; Kirkey, 2010;

Nash et al, 2013; Dennis et al, 2011). These trends jeopardize primary sources of

unpaid care: spouses and children (Seow, 2009). Yet we are still moving in a direction

that increases patient dependence on unpaid care.

Variability in the accessibility of public home care services that do exist is also a

concern. While some Local Health Integration Networks (LHINs) have fully staffed

CBPHCTs, others have very little in the way of home care (Yu, 2011). Residing outside

the protection of the Canada Health Act (Yu, 2011), the precarious position of CBPHC

prompts ongoing inquiry about the role of the state, public-private financing, unpaid

family caregivers, and models of delivery spanning the past two decades (Baranek,

2000). The Canadian Cancer Society (CCS) has attributed regional inconsistencies in

institutional and home death on the national level to the inconsistent structure of CBPHC

(Kirkey, 2010). With no comprehensive, province-wide plan for CBPHC, regional

disparities exist within Ontario as well (Yu, 2011; Kirkey, 2010). Home care receipt in

Ontario is sensitive to a number of factors including socio-economic status (SES)

(Motiwala et al, 2006), the availability of unpaid caregivers (Aoun et al, 2013), and home

care availability related to rurality and centrality (Skinner, 2005; Funk et al, 2010;

Kuluski, 2010; Bainbridge et al, 2011).

Families that do access home care reportedly experience high financial and

psychosocial burden. The CCS estimates costs of care for families facilitating home care


4

to be higher than $1,000 a month premised solely on the items that inpatient care would

provide: medications, nutritional supplements, special meals, vitamins, diapers,

dressings, aids for bathing and so on (Kirkey, 2010). This estimate does not include lost

income from time off work, use of vacation time or, in some cases, complete job loss

(Haley, 2003). Research has found that outcomes and quality of palliative care are

largely dependent on the stability and wellness of the caregiver. The catch twenty-two is

that it is compromised by their participation in caregiving (Brink, 2008; Cain et al, 2004).

The unpaid caregivers of palliative patients are more vulnerable to stress and

depression than their non-caregiver (Haley, 2003; Funk et al, 2010) and non-palliative

caregiver counterparts (Williams et al, 2014). The CCS advocates for extending

caregiver tax benefits from six to twenty-six weeks (Kirkey, 2010). In the June 2014

provincial election, a caregiver tax credit was on the platform of the NDP (Campion-

Smith, 2014), demonstrating the political momentum and attention being drawn by

community-based services and unpaid caregivers.

Despite growing interest in the burden placed on unpaid caregivers (Docherty et al,

2008; Bachner, 2007; Ryn et al, 2011; Parker Oliver, 2013), many cost studies in

CBPHC adopt a Ministry of Health (MoH) perspective if for no other reason than a data

deficit (Klinger et al, 2010; Cartoni et al, 2007). This excludes a considerable proportion

of the spending and labour time that is actually required to facilitate CBPHC. It has been

estimated that 80-90% of in-home care is provided by unpaid caregivers in Canada

(Haley et al, 2003). The palliative care sector in Canada generally is highly reliant on

volunteer labour. The Hospice Association of Ontario (HAO) reports that volunteers

provide 600,000 hours of work annually in over 400 communities (Shephard, 2004). The

value of unpaid care provided at home, if it were provided by paid workers, has been

estimated from $1 billion for cancer patients over seventy in the United States (Hayman

et al, 2001) to $35 billion for all unpaid care in Australia (Stajduhar et al, 2007). Other

developed nations using this type of home care model have given similar estimates

(Aoun et al, 2005). Projects assessing determinants of cost that adopt a societal

perspective – accounting for all costs irrespective of payer (Neumann, 2009) – are

therefore in demand.

Nature of existing findings

The determinants of cost and caregiver burden often overlap given that these are not


5

mutually exclusive dimensions of the palliative trajectory. Many are socio-demographic

and related to the living situation of the patient (Guerriere et al, 2010; Chai et al, 2013;

Gardiner et al, 2014), the comorbidity of the patient (Glajchen, 2012) and its associated

demands on their unpaid caregiver (Hirdes et al, 2012). Differences in cost have also

been noted across disease groups (Enguidanos et al, 2005), and this brings us to the

research purpose and rationale. “Cancer” is often grouped together and rarely treated as

distinct diseases requiring variable attention beyond curative treatment, but this may be

a significant variable in predicting the magnitude of cost and caregiver burden

associated with palliative cancer patients.

1.3 Research purpose and rationale

A key to stabilizing and improving CBPHC is identifying and mediating factors

associated with service use and caregiver burden, which may impinge on one another.

The objective of this thesis is to comprehensively analyze the determinants of the

financial and caregiver burdens associated with CBPHC for people with cancer. This will

be done with consideration for different stakeholders in this service and their cost

perspectives. Several gaps in the literature will be touched upon in this thesis, including

1) analysis of cost from multiple payer perspectives and at the micro-level, 2)

longitudinal analysis of caregiver burden, and finally, 3) to analyze these outcomes

taking into consideration specific primary cancer diagnoses (categorized by tumour

site vis-à-vis the ICD-10). In previous analyses of palliative care outcomes for this

population, the presence of any cancer and cancer stage have been used as clinical

indicators (Motiwala et al, 2006; Fairfield et al, 2012; Sussman et al, 2011; Sims et al,

1997).

The College of Physicians and Surgeons of Ontario’s (CPSO) (2002) “Decision-making

for the End of Life” policy statement asserts that emergency service use is often related

to the caregivers’ reservations about care tasks. Caregivers report clinical uncertainty

regarding medications, disease progression, side effects, symptoms, or whether to call

for assistance, as one of the most stressful dimensions of unpaid caregiving (Ryn et al,

2011). Palliative oncologic literature suggests that the aforementioned points of obscurity

may differ from malignancy to malignancy (Alesi et al, 2011; Dennis et al, 2011; Von


6

Roenn et al, 2011; Janjan, 2011). This research can speak to whether differential care

needs related to primary diagnosis persist beyond curative stages of treatment.

Palliative and oncologic care have previously been conceptualized as mutually exclusive

disciplines (Ramchandran et al, 2013), but literature embracing their interaction has

followed Temel et al’s (2010) seminal paper regarding mixed palliative-oncologic

treatment for patients with metastatic non-small cell lung cancer (NSCLC). This

illuminated the heterogeneity of cancer patient needs in the palliative trajectory. In 2011,

the journal Oncology dedicated an entire volume to the reconciliation of these disciplines

to improve QoL and individual-specific palliative care (Dennis et al, 2011; Von Roenn et

al, 2011; Janjan, 2011). Metastasized NSCLC has been the focus of the succeeding

literature (Ryn et al, 2011; Irwin et al, 2012) in addition to haematologic malignancies

that are recognized as having uniquely unpredictable and intensive care needs

(Simoens et al, 2010; Cartoni et al, 2007). While some studies have assessed palliative

outcomes for one neoplasm subgroup (Temel et al, 2010; Bergman et al, 2009;

Ceilleachair et al, 2011), there is little that assesses multiple neoplastic primary

diagnoses simultaneously, allowing comparative inferences to be made.

There is motive and logic to investigate the role of primary cancer diagnosis in the

magnitude of palliative home care burden, both financial and caregiver-related. In

addition to the unique needs of NSCLC and haematologic patients that has been

investigated, the Canadian Institute of Health Information (CIHI) recently reported that

lung and colorectal cancer patients are highly represented among deaths in acute

settings (CIHI, 2013) which are also associated with high cost (Simoens et al, 2010).

Given the CPSO’s (2002) attribution of EoL hospital admissions to caregiver burden, this

also suggests that the magnitude of burden and service use differs depending on

disease site. Disease specific care requirements may demand different levels of

commitment or complexity to be handled by unpaid caregivers in an informal setting.

Neoplasm specific Standard Operating Procedures (SOPs) for palliation have been

developed by Gaertner et al (2011) for nineteen malignancies.

1.4 Research questions and hypotheses

This thesis is focused on the following research questions:

Do primary cancer diagnoses, as defined by the ICD-10, have a determinant role in


7

palliative home care cost controlling for other pertinent predictors (i.e. SES, home care

agency, etc.)? and

Do primary cancer diagnoses, as defined by the ICD-10, play a determinant role in

caregiver burden controlling for other pertinent predictors (i.e. SES, home care agency,

etc.)?

Hypothesizing about CBPHC cost

Based on both existing studies and the Andersen and Newman model of health services

use (Section 3.1.3), it is expected that cost will be driven by both clinical and socio-

demographic variables. Comorbidity scores, marital status and living arrangement have

been found to be significant in societal perspective studies (Guerriere et al, 2010; Chai

et al, 2013). Differences across different diseases (cancer and other) have been found in

Enguidanos et al’s (2005) MoH perspective study. It was also found that SES, which is

indicated by education and employment status in these data, drove cost from an unpaid

caregiver perspective (Gardiner et al, 2014). Therefore, these are the variables that are

expected to show significance in the societal perspective model that integrates all of

these payers. However, when isolating the payer perspectives, there may be a

differential effect attributable to them. These covariates correspond most strongly with

Andersen and Newman’s ‘predisposing’ (i.e. demographics) and ‘need’ (i.e. comorbidity)

factors (1973).

With respect to statistical analysis of aggregated and service-specific (disaggregated)

costs of home care, it is hypothesized that:

H1: The influence of primary cancer diagnosis (need factor) on costs will be statistically

significant. (Primary)

H2: Costs will be driven by caregiver burden, which in turn will be driven by primary

cancer diagnosis. (Primary)

H3: Socio-demographic variables, or predisposing factors, will be statistically

significant, particularly SES indicators, marital status and patient living

arrangement. (Secondary)


8

Hypothesizing about caregiver burden

Much of the burden articulated by caregivers in the literature review is reportedly driven

by uncertainty surrounding patient-specific care tasks (Docherty et al, 2008; Parker

Oliver et al, 2014). Palliative oncologic literature suggests that these care tasks differ

from patient to patient based on primary cancer diagnosis, and therefore this variable is

expected to have a significant predictive role in caregiver burden. Time spent providing

care has been positively associated with caregiver burden (Hirdes et al, 2012) and it is

expected that the analysis will reproduce this finding. Living arrangement and marital

status have also been found to be influential; being married to or living with the patient

can especially limit opportunities for respite (Glajchen, 2012).

With respect to statistical analysis of caregiver burden scores, it is hypothesized that:

H4: The influence of primary diagnosis (need factor) in caregiver burden will be

statistically significant. (Primary)

H5: Caregiver burden will be driven by time spent providing care, which will be

determined by primary diagnosis in the analysis of unpaid caregiver cost. (Primary)

In the following chapter, a more in-depth description of the literature that informed these

hypotheses is given. This is followed by a description of the methodologies used to test

the hypotheses and the results that were derived. The results, and whether or not the

hypotheses proved to be true, will be discussed in the final chapter of the thesis. These

results will inform future research by indicating whether the inclusion of primary cancer

diagnosis adds anything to the analysis. The analyses use data collected through a

much larger project; a detailed description is contained in Section 3.1.

9

Chapter 2 Literature Review

This literature review takes a systematic approach that is thematically stratified to

capture the multi-faceted, multi-stakeholder landscape that is faced by decision-makers

with respect to the future of CBPHC. The methodology used is conducive to research

using various designs (Section 2.1). The three upcoming sections focus on 1.) the cost

of palliative home care (Section 2.2), 2.) caregiver burden (Section 2.3), and 3.)

indications from palliative oncologic research that primary diagnosis could play a role in

both cost and caregiver burden (Section 2.4).

In addition to the Andersen and Newman model that is explained in the Methodology

(Section 3.1.3), this review of the literature illuminates independent variables other than

primary diagnosis that have been found significant for dependent variables of cost and

caregiver burden. It therefore informs what to include in analytic models, as well as what

could be expected with regard to the hypotheses. These are searched separately

because they are measuring different dependent variables. The third section discusses

an independent variable: primary cancer diagnosis, and describes the existing research

base that lead to hypothesizing that it will have significant influence over these

dependent variables. Whether or not this is the case could be informative and useful to

those working in palliative care, which is very often utilized by people with cancer.

2.1 Search strategy

Overview of each section of the review

1.) Palliative home care cost (Section 2.2)

Given political and demographic contexts that favour the further development of CBPHC,

it is important to understand as many aspects of its relative cost from as many

stakeholder perspectives as possible. It can be expected that existing studies use

different methodologies and assess different aspects of cost, be it determinants of cost,

comparative cost, or standalone approximations of cost. Studies considering these

Chapter 2: Literature Review R. Redmond-Misner

10

dimensions of cost for palliative home care specifically were included. Search terms

included “palliative,” “home care,” “cost” and “community-based.” Further diversification

of terminology did not produce different or new results. Hand-searched journals included

Health Economics, Palliative Medicine and Journal of Palliative Care. Hand-searching

journals is a snowball technique used to acquire articles that were not identified through

the search by looking through journals that published other suitable articles.

2.) Caregiver burden in palliative care (Section 2.3)

Community-based health care is characterized by reliance on unpaid sources of labour

(Greffe, 2007). This is often in the form of family members with little to no training in

providing health care (Docherty et al, 2008; Parker Oliver et al, 2014). Caregiver burden

is an important consideration for CBPHC, especially if it is expanded in coming years. It

is important to understand the needs of unpaid caregivers and predictors of having a

psychologically, physically or economically detrimental caregiving experience. It can be

expected that existing studies use different methodologies and assess different aspects

and determinants of caregiver burden. Studies addressing the determinants of

psychological, physical or economic caregiver burden were included. Search terms used

included “end-of-life,” “home care,” “palliative” and “caregiver burden.” Diversifying

terminology further did not produce new findings and all articles may be found with

combinations of these search terms. Hand-searched journals included Palliative

Medicine, Journal of Palliative Care and Social Science & Medicine.

3.) Insight from palliative oncologic literature

People with cancer are the most intensive consumers of palliative home care (Alesi et al,

2011; Brink et al, 2008). In the light of recent findings that specific malignancies require

disease-specific treatment and respond differently than other cancers, particularly

NSCLC and haematologic tumours (Temel et al, 2010; Simoens et al, 2010; Tzala,

2005), improved cooperation between palliative and oncologic disciplines is occurring.

Palliative care is becoming increasingly specific and tailored to the individual patient, and

this is true for their specific primary diagnosis as well (Gaertner et al, 2011). Studies that

discussed palliative care processes specific to a disease site were included. Search

terms used included palliative, “supportive care,” “palliative care,” “cancer,” “oncology,”


11

“integration” and “end-of-life.” Further diversification of terminology did not produce new

articles. Hand-searched journals included Oncology and Supportive Oncology.

Review strategy

Research on home care cost, caregiver burden and palliative oncology use variable

study designs and outcome measures, often not in the form of a randomized controlled

trial (RCT). The review reconciles the approaches taken by Docherty et al (2008),

Gardiner et al (2013) and Rodby et al (2014) to accommodate non-RCT studies.

Reviewing the literature this way involves appraising journal articles and charting the

design, sample, setting, focus, methods, outcomes, results and weaknesses of selected

studies. This allows for the inclusion of cohort, case control and qualitative studies that

may represent reality more accurately than the controlled environment created by

inclusion criteria for RCTs (external validity) (Coates, 2012) and be nuanced by

qualitative factors such as caregiver perspective. Considered articles were critically

appraised using tools created by Oxford University’s Critical Appraisal Skills Programme

(CASP) (Appendix A). This is the method by which abstracts were included or excluded.

This section discusses parts of the review process that are unanimous across subtopics,

whereas topic-specific details are held in its designated section. Databases searched

include Cochrane Library, MEDLINE, PubMed, Embase, EconLit and the Cumulative

Index to Nursing and Allied Health Literature (CINAHL) up to June 2014. Articles moved

through stages of Identification, Screening, Eligibility and Inclusion in accordance with

the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA,

2014). For each section, an initial search for systematic reviews in the Cochrane Library

took place to inform the time period component of the inclusion criteria. Various Boolean

searches were done until results were saturated with duplicates. Nothing prior to the

year 2000 is included, as prior to this was thoroughly searched and/or deemed irrelevant

(Simoens et al, 2010). All are English language articles limited to palliative home care

for adult patients, unless another setting had relevance to home care.

This review is vulnerable to several types of bias. Positive and significant findings are

more likely to be (a) published (publication bias), (b) published quickly (time lag bias), (c)

published more than once (multiple publication bias), and (d) cited (citation bias)

(Hopewell et al, 2009; Cochrane Collaboration, 2011). The articles within the review, too,


12

have associated and documented biases. Non-English language publications,

commentaries, editorials, letters and special articles were not excluded from the search,

but very rarely made it to the final review.

2.2 Palliative home care cost

Two systematic reviews on this topic were found through this search. Simoens et al

(2010) searched articles from 2000-2009 focusing on the cost of palliative care in any

setting. They report many of their included studies to be relatively old, taking a hospital

perspective, and internationally variable (whether this is a limitation is up for debate).

The 2000 cut off is due to the questionable relevance and usability of older research,

and this review adopts the same cut off for this reason. Gardiner et al (2014) review the

value of home-based unpaid caregiving, searching articles from database inception to

2012. This review will focus on home-based palliative care, consider all cost

perspectives, and search from 2000-2014. Therefore, while there is a possibility for

some overlap with these reviews, there are also unsearched years and cost

perspectives that were not reached by them.

Figure 1 depicts the process by which articles were selected, culminating in eighteen

articles from thirteen journals being charted and synthesized. These include two

systematic reviews (11%), a qualitative/narrative review (5.5%) and fifteen original

studies (83%), representing 4,879 subjects (patients, caregivers, patient-caregiver dyads

or other providers) controlling for double publication. This was allowed in two instances

because the analyses were different and relevant. All cohorts are palliative but have

variable diseases. Eight countries are represented by the original studies. Thematically,

the articles produce comparative cost, specific estimates of cost or payer shares of the

cost, determinants of cost, and Lavergne et al (2011) consider the generalizability of

such results.

Discussion of whether home care was more or less costly than conventional treatment

was a common topic in the existing literature. Many of the findings are contradictory and

the methodologies are informative; it becomes apparent that the payer perspective

adopted can drastically change the results. This discrepancy is what leads to research


13

findings that make this mode of palliative care delivery attractive to decision-makers and,

conversely, research findings that generate concern around sustainability and pressure

on unpaid caregivers.

Wong et al’s (2013) prospective cohort study included forty-four advanced heart failure

(HF) patients in Singapore with prognoses of less than one year between 2008-2010.

Figure 1: PRISMA flow chart for home care cost literature search

Cost reduction was defined as a decline in hospitalizations, and 71% of the patients

experienced reduced hospitalization after admission to an advanced care program

(ACP) delivered at home. This is a hospital or Ministry of Health cost perspective,

however, the additional cost of home visits and how this may counteract reductions in

hospitalization, even within this perspective, is not described.

Iden

tific

atio

n S

cree

ning

E

ligib

ility

In

clud

ed

Records identified through database searching (N=675)

Records identified through other sources (N=10)

Records after duplicates removed (N=300)

Records screened (N=300)

(N=600)

Records excluded (n=264)

(N=564)

Full-text articles assessed for eligibility (N=36)

Articles excluded (N=18)

Double publication (N=1) No applicable design (N=6) Minimal relevance (N=11)

Studies included (N=18)


14

Simoens et al (2010) conducted a systematic review of the costs associated with

different palliative care settings. Findings that lead them to suggest that home care was

less costly were related to reduced hospitalizations, similarly to Wong et al (2014). They

caution that this may not be true for patients with haematological malignancies who

require frequent transfusions. While this conclusion was not the sole purpose of the

review, the data for which this particular conclusion was drawn was among the most

outdated.

Shnoor et al’s (2007) case control cohort study followed 146 patients with metastatic

cancer who lived with their families and received either conventional (N=73) or home

care (N=72) in 2003 in Israel. They were not randomized into these groups but receiving

their chosen type of care. The average cost to the health system for the last two months

of life for home and conventional care was found to be $3,467 and $12,434 ($US 2003,

average exchange to $CDN = 1.3), respectively. While the patients are matched by

disease, there is a selection bias in that patients selecting home care are opting for

palliation whilst conventional care was more focused on life-extension. This

compromises comparability and implies more intensive treatment for what is defined as

the conventional care group. The disparity in cost was tied to a difference in the number

of treatments received.

Enguidanos et al’s (2005) case control study included 159 home care recipients and 139

controls diagnosed with cancer, congestive heart failure (CHF) or chronic obstructive

pulmonary disease (COPD). Patients in the intervention group were found more likely to

die at home across all primary diagnoses and to reduce costs by approximately $5936

for people with cancer, $11,325 for people with COPD and $8,445 ($US 2001, average

exchange to $CDN = 1.5) for people with CHF. The study took place in the United

States, meaning that these costs correspond with what patients might have been billed.

This is a unique instance where the health care savings are attributable to the patient

and caregiver, however, the compensatory unpaid caregiving or out-of-pocket costs that

replaced formal services are not considered.

Cartoni et al’s (2007) cohort study included 144 patients enrolled in a specialized home

care program for haematologic malignancies between 2004-2006 in Italy. Their cost

analysis included health care providers, materials and medicines, transfusions and

laboratory (blood chemistry and microbiology). The home care costs associated with


15

various stages of haematologic cancer were found to be lower than the corresponding

hospital charges.

Brumley et al’s (2007) RCT compared usual care (N=152) to usual care plus in-home

palliative care (N=145) in Colorado and Hawaii. They found the intervention group to be

more satisfied, more likely to die at home, and less likely to visit the emergency

department or be admitted to the hospital. Admissions to the emergency department or

hospital were associated with higher costs. However, new costs for the additional

services and unpaid care are not considered. There are also ethical issues surrounding

offering additional services only to some patients; many studies in this field are restricted

to case controls where the patients select their preferred treatment option for this reason

(Shnoor et al, 2007).

At this point in the review, it is notable that all studies assessing comparative cost that

fall into the “less costly” list used Ministry of Health and hospital perspectives. There is a

lack of consensus in the literature, with several studies arguing it to be more costly. The

first of which to be included here is Tzala et al’s (2005) case control study and cost-

minimization analysis that included 27 home care recipients and 25 controls with

haematologic cancers in 2002 in Greece. Interestingly, this study also adopts the

perspective of the hospital. However, home care recipients had to be more frequently

monitored for full blood count and blood cross-tests in order to proceed at home while

the hospital cohort was only tested during admission. Transfusion requirements were the

same in both groups. This does not consider the additional cost of unpaid caregiving and

is not generalizable to other palliative patients, but illuminates the potential for costly

increased surveillance in CBPHC. It is also notable for the broader thesis that the

magnitude of cost is attributed to this particular cancer diagnosis.

Ostgathe et al’s (2008) qualitative study derives cost estimates for the palliative care of

patients with NSCLC by attributing costs to the home care, hospital care and day care

projections of a team of experts. Consensus was formed among an oncologist, surgeon,

palliative specialist and radiotherapist participating on the negotiation committee of a

hospital finance department in Germany. The cost of the first and follow-up visits was the

same for all scenarios and home care was believed to reduce hospitalization. However,

home care had a higher overall estimate due to projections of longer, more frequent

visits and travel costs, even if the patients lived close by. These estimates are


16

hypothetical rather than premised on empirical data, and therefore cannot actually verify

that higher appointment costs counteract reduced hospitalizations and vice versa. This is

again a hospital perspective on cost.

Jacobs et al’s (2011) prospective cohort study included 192 patient-caregiver dyads

receiving home care between 2005-2006 in five Canadian cities: Halifax, Montreal,

Winnipeg, Edmonton and Victoria. The focus of the study is the economic loss of unpaid

caregivers. They found that 9% of families incurred economic losses greater than 10% of

their pre-study income, and low-income status increased from 27-40%. It is argued that

this is a greater cost to caregivers than would be alternative palliative care settings,

however there is no control group. Some patients in the cohort had multiple unpaid

caregivers that might minimize one another’s economic burden, therefore these

estimates may be conservative. At this point in the review, it is notable that studies

presenting home care as more costly, although not always, tend to factor in unpaid

labour. This illuminates the importance of cost perspective and the significant difference

that consideration for unpaid caregivers makes in study conclusions.

Specific estimates and payer shares of cost

Specific monetary values were estimated by six studies. In addition to the estimates of

Shnoor et al (2007) ($1,733.50/month to the Israeli health system), estimates of cost or

payer shares are given by Klinger et al (2011), Guerriere et al (2010), Dumont et al

(2009), and Chai et al (2014). Guerriere et al (2010) estimate the average monthly cost

for palliative home care patients with cancer in Ontario, Canada, to be approximately

$24,549. Family caregivers account for 70% of this total. Their costing method used a

societal perspective guided by the Ambulatory Home Care Record (AHCR). Dumont et

al (2009) found the share of the cost of resource utilization by palliative home care

recipients to be 71.3%, 26.6% and 1.6% among the public health care system, families,

and not-for-profit organizations respectively. Costing was guided by the Canadian

Coordinating Office for Health Technology Assessment (HTA). This considers multiple

payers, but only goods and services that are explicitly paid for whereas the former study

incorporated the economic cost of caregiver time. Chai et al (2014), using members of

the same cohort and the same costing method as Guerriere et al (2010), estimate the

cost share among unpaid caregivers time, the Ministry of Health and out-of-pocket

expenses at proportions of 77%, 21% and 2%, respectively.


17

The societal estimates are considerably higher than that produced by Klinger et al’s

(2010) hospital perspective. Klinger et al’s (2011) cohort study included Ministry of

Health cost data associated with ninety-five cancer patients over a fifteen-month period

in 2005 and 2006. Data was gathered from the Community Care Access Centre (CCAC),

Enhanced Palliative Care Team (EPCT) fee schedule and Ontario Health Insurance Plan

(OHIP) fee schedule. They found the overall cost to be $1.626 million; $17,112 per

patient and $117 per patient day. Estimates are based on a retrospective assessment of

administrative data. The Ontario Drug Benefit (ODB), out-of-pocket spending, unpaid

caregiving and emergency services are not included. This is also unable to address

variability in cost across patients.

Determinants of home care expenditures

Clinical determinants will be discussed first. Simoens et al (2010) and Cartoni et al

(2007) both identified the transfusion requirements of people living with haematologic

malignancies as a significant driver of costs. The closer monitoring of blood count and

blood tests for this population was what caused Tzala et al (2005) to deem home care

more costly than inpatient care, where patients can be less frequently tested. This is not

generalizable to all palliative care recipients or non-haematologic cancer patients, but

that affirms the research question and hypotheses. Multiple studies found expenses to

rise as the patient came closer to death (Dumont et al, 2010; Guerriere et al, 2010;

Gardiner et al, 2014). Guerriere et al (2010) found higher costs among patients with

lower physical functioning. These studies have already been synopsized here.

In terms of socio-demographic determinants, Chai et al (2013) found that the public

share of costs was significantly lower for older and married home care recipients. These

people may have more unpaid care available to them through their spouse and their

spouse may be more likely to be retired. Conversely, it was found by Guerriere et al

(2010) that, from a societal perspective, costs were higher among patients who were

living with someone. This may, too, be related to higher availability of unpaid care,

translating into higher unpaid costs.

Finally, in considering the cumulative implications of these studies, Lavergne et al’s

(2011) Canadian case control study suggests potential for selection bias in study

recruitment. This has already been alluded to in the Introduction (Section 1.2) with

respect to determinants of receiving home care at all, and is certainly relevant to this


18

thesis. With the intention of assessing the generalizability of community-based program

cost estimates, Lavergne et al (2011) compared eligible participants who agreed to the

study, eligible participants who refused, the entirety of program enrollees and all cancer

decedents in the provincial cancer registry. They found that eligible and receptive

patients were generally younger, enrolled for longer, and lived closer to the palliative

care program than those who declined or were not eligible. Therefore many studies will

be vulnerable to selection bias, and many that rely on informal reporting may have recall

or social desirability biases (Guerriere et al, 2010).

Summary

Thematically, the literature available through the searched databases discussed the

comparative and actual cost of home care, payer shares of this cost and determinants of

cost. Whether home care was deemed more or less expensive than other settings was

largely dependent on the payer perspective used, with hospital and Ministry of Health

perspectives tending to find that these programs are cost saving and societal and

caregiver perspectives finding it to be more expensive. Similarly, standalone estimates

of the cost of home care were drastically different depending on which of these

perspectives was taken, with the societal estimates being considerably higher than the

ministerial ones.

Determinants of cost were found to be both clinical and socio-demographic in nature,

with certain diagnoses and patient-caregiver characteristics driving higher or lower

service utilization and cost. Enguidanos et al (2005) found cancer patients to have lower

savings than patients with heart conditions, but did not distinguish among cancer

patients. This will be controlled for in the analyses of this thesis. Lower physical function,

found to exacerbate cost by Guerriere et al (2010), will be controlled for with comorbidity

scores generated using a validated comorbidity measure (see Section 3.1.3). Marital

status and living arrangements, found to be significant by Guerriere et al (2010) and

Chai et al (2013), will be included as well. SES, illuminated by Gardiner et al’s (2014)

systematic review, will be controlled for by caregiver employment status and caregiver

and patient education levels. Finally, ethnicity was found to be significant (Gardiner et al,

2014); ethnicity, race and migration status are not in these data, indicating a weakness

of the analyses in later chapters.


19

In addition to the technical weaknesses of this review outlined in the Search Strategy

(Section 2.1), there are gaps in the literature that the broader thesis will attempt to

address. This includes a dearth of literature looking at disaggregated (service-specific)

costs as opposed to aggregated public or unpaid expenses. Most studies adopt the

former Ministry of Health perspective which, too, is a weakness. The societal perspective

that is commonly endorsed for health economics research is lacking (Nuemann et al,

2009; Weinstein et al, 1996). Finally, while there is a proportionately high amount of

work focusing on cancer patients – potentially the most intensive users of palliative

home care (Alesi et al, 2011; Brink et al, 2008) – there is little that analyzes their clinical

characteristics and tumour sites comparatively or at all. The two diagnostic groups

currently represented are haematologic malignancies (Tzala et al, 2005) and NSCLC

(Temel et al, 2010). The charts for this section of the review are found in Appendix B.

2.3 Caregiver burden in palliative care

Community-based health care is characterized by reliance on unpaid sources of labour

(Greffe, 2007). This is often in the form of family members with little to no training in

providing health care (Docherty et al, 2008; Parker Oliver et al, 2014). Caregiver burden

is a weakness of CBPHC should it be expanded in coming years, thus it is important to

understand the needs of unpaid caregivers and predictors of having a psychologically,

physically or economically detrimental caregiving experience.

One systematic review on this topic was found through this search, done by Glajchen

(2012). This review drew on articles published between 1963-2011 from PubMed and

CINAHL only, focusing on the physical aspect of caregiver burden. The cut off for this

review will be the year 2000, again due to questionable relevance, and the overlap with

Glajchen’s (2012) review is justified by their narrower focus on physical outcomes only.

This review contributes consideration for psychological and economic impacts, and the

unsearched years from 2011 to 2014.

Figure 2 depicts the process by which articles were selected, culminating in eighteen

articles in fourteen journals being charted and synthesized. These include one

systematic review (5.5%) and seventeen original studies (94.5%), representing 6973


20

units of patient-caregiver dyads, bereaved caregivers or other service providers

participating in caregiver burden related studies. All care was palliative, with most

patients having cancer. Eleven countries are represented by the original studies.

Thematically, the articles discuss health problems associated with caregiver burden,

unmet needs reported by caregivers, and determinants of satisfaction and burden.

Figure 2: PRISMA flow chart for caregiver burden literature search

Health problems associated with caregiver burden

Health problems associated with unpaid caregiving are first identified with Glajchen’s

(2012) systematic review of international studies considering physical health impacts of

unpaid caregiving. It was found that more than half of caregivers in these studies

reported health problems including heart problems, hypertension and arthritis. These

Iden

tific

atio

n S

cree

ning

E

ligib

ility

In

clud

ed



Records after duplicates removed (N=~1000)

Records screened (N=~1000)

(N=600)

Records excluded (n=~971)

(N=564)



Minimal relevance/unoriginal (commentary) (N=9)



21

issues were linked to assisting cancer patients with ADLs (eating, dressing, bathing,

etc.), instrumental ADLs (IADLs) (cooking, shopping, providing transportation, etc.), and

performing tasks formerly done by home care nurses (dispensing medication, changing

patches, monitoring symptoms, etc.). The physical health problems were driven by the

patient’s cancer stage, disability, mobility, dependency and pain. Living with the patient

was found to diminish opportunities for respite and enhance the magnitude of physical

impairment experienced by oncology caregivers.

Götze et al’s (2014) cohort study of 106 patient-caregiver dyads in Germany measured

psychological distress among caregivers using the Hospital Anxiety and Depression

Scale (HADS) and European Organization for Research and Treatment of Cancer’s

(EORTC) Quality of Life Questionnaire (QLQ C-15-PAL). They found 33% of caregivers

to have high anxiety and 28% to have depression, which was highly correlated with

patient anxiety and depression, financial burden and low social support. Spousal

caregivers tended to have the highest psychological distress.The authors caution, with

regard to the relationship with patient depression, that the HADS instrument uses

anhedonia as an indicator for depression that may not be suitable for the EoL, when this

is a very common symptom.

Grov et al’s (2006) cross-sectional analysis of ninety-six Norwegian caregivers in

palliative home care measured burden using dimensions of the Caregiver Reaction

Assessment (CRA). They assessed dimensions of family support, self-esteem, finances

and impact on daily schedule as dependent variables with a series of health measures

as independent variables. These included physical QoL, anxiety, depression and social

support. It was found that depression had the strongest relationship with these aspects

of caregiver burden, if not constituting a dimension of burden in itself. This study is

vulnerable to bias via self-selection into the study. Many declined because they were too

tired, making these findings potentially conservative.

Kenny et al’s (2010) observational study assessed the health-related QoL (HRQoL) of

178 home-based caregivers during 2005-2006 in Australia. Using a cross-sectional

survey administered to the general population, it was found that unpaid caregivers had

comparatively better physical health and worse mental health. While the mental health

finding is unsurprising and congruent with other research, the finding of better physical

health contrasts the studies that have found physical detriments to caregivers (Glajchen,


22

2012). Weaknesses of this study include that that the survey is inherently retrospective

and therefore vulnerable to recall bias. Also, the cross-sectional nature of the data

inhibits them from seeing changes in health over time. While they have better health in

that static instance, longitudinal observation might illuminate more rapid deterioration

among unpaid caregivers.

Unmet needs of unpaid caregivers

The unmet needs reported by informal caregivers are first identified here with Aoun et

al’s (2005) observational study that administered surveys to twenty caregivers, six

volunteers and twenty-three service providers in Australia. Lack of information,

communication, service provision and support from community services were among the

prominent unmet needs. Caregivers wished they had learned how to comfort patients,

what to expect, how to deal with symptoms and how to acquire aids (walking frames,

wheelchairs, etc.). There is potential for selection bias that caregivers who were

particularly devoid of information were more likely to participate in this study. At this point

in the review, it is helpful for cross-chapter coherence to recall the broader research

rationale which touched on informational deficits. The care tasks and symptoms to which

caregivers are responding may differ across diagnoses and thus manifest in a different

caregiver burden.

Docherty et al’s (2008) systematic review focused on the unmet informational needs

reported by unpaid caregivers. They critically appraised the included studies in terms of

the strength of the evidence, and found the strongest evidence for pain and symptom

management informational needs. As a result, they emphasize the importance of

effective communication between patient, caregiver and service provider, and a

secondary need for welfare and social support information. The included studies were

ethnically homogenous, small-scale and largely focused on cancer patients.

Sheehy-Skeffington et al’s (2013) qualitative observational study held caregiver focus

groups in Ireland to discuss their experiences managing medications. Polypharmacy

emerged as a significant burden and the importance of vivid instructions was

emphasized. Caregivers coveted the ability to give medications as needed for symptom

control, and a lack of clinical insight was a barrier to doing this. While this is congruent

with the findings of Docherty et al (2008), it is likely that there is more to learn in terms of


23

polypharmacy information deficits. With only three focus groups, this study may not have

achieved saturation. These participants also hail from a region with no specialist

inpatient palliative unit, meaning caregivers in other regions or countries may have better

access to instruction from health service providers.

Parker Oliver et al’s (2014) observational study of 146 bereaved caregivers in the United

States found similar themes during their interview process. Only thirty-eight (26%)

interviewees discussed pain and symptom management-related concerns, and their

responses were thematically analyzed. Difficulty with the administration of medications,

uncertainty about side effects and insecurity with pain assessment culminated with

frustration toward the health care team. This was related to feelings that important

aspects of caregiving, namely the responsibility of actually treating the patient, were

never explicitly communicated. Because this study was retrospective, there is potential

for recall bias. The low proportion of caregivers discussing this need does not

convincingly show it to be a prominent concern without the validation of other studies.

Ryn et al’s (2011) observational study included 677 home-based caregivers of colorectal

and lung cancer patients in the United States. The cohort responded to a self-

administered survey and those who reported providing more than 50% of informal care

to the patient were included. In addition to ADLs and IADLs, caregivers provided cancer-

specific care such as watching for treatment side effects (68%), helping manage pain,

nausea or fatigue (47%), administering medicine (34%), deciding whether to call a

doctor (30%), deciding whether medicine was needed (29%), and changing bandages

(19%). Half of caregivers reported that they did not not get the training they perceived as

necessary. Questionnaires were done approximately four months into bereavement and

responses are vulnerable to recall bias.

Determinants of caregiver burden

Service related determinants of caregiver burden or satisfaction are first identified with

Diwan et al’s (2004) observational study of 150 caregivers of palliative patients with

dementia in the United States, aimed at identifying challenges associated specifically

with dementia. Data was collected through interviews using the Caregiver Strain Index

(CSI), Revised Memory and Behaviour Problems Checklist and the Katz Index of ADLs.

Included dimensions of the CSI were role, personal and emotional. Instead of dementia


24

specific struggles, it was found that perceived lack of support from the health care team

enhanced personal and emotional strain. The authors note that this was an ethnically

homogenous sample.

Guerriere et al’s (2013) prospective cohort study included 104 caregivers of CBPHC

patients in Ontario and assessed caregiver satisfaction with home-based nursing and

physician care. Interviews were conducted bi-weekly from the patient’s admission into

the program until their death, and satisfaction was assessed using the Quality of End-of-

Life Care and Satisfaction with Treatment (QUEST) questionnaire. They found that

characteristics of the service providers were significant predictors of overall satisfaction,

including “always spent enough time,” “never arrived late,” “never been hard to reach,”

“always responded quickly,” and several others. Due to small sample size, there were

limitations around including too many variables in the statistical analysis, which may

have led to omitted variable bias.

Hirdes et al’s (2012) cross-sectional analysis included 3,929 patients assessed using the

interRAI pilot from 2007-2009 in Ontario. The service related determinants of caregiver

burden and satisfaction identified through this study included the specific home care

agency, hospitalizations and nursing visits. They also identified patient and caregiver

related predictors. Patient clinical instability, depressive symptoms, cognitive impairment

and positive outlook were significantly associated with caregiver distress levels. Hours of

unpaid care also determined caregiver burden. Unfortunately, the interRAI does not

include caregiver demographics such as sex and age and the data used for this analysis

was cross-sectional rather than longitudinal. In Kenny et al’s (2010) previously described

study, they found patient symptom severity, but not caregiver time input, to be predictive

of burden.

Carlsson et al’s (2003) Swedish observational study featured 183 caregivers of patients

in home care. Caregiver questionnaires and patient medical records were analyzed.

They found that caregivers of patients who died at home, while they experienced more

sleep deprivation, felt there was a more positive impact on patient QoL than did

caregivers of patients who died in other settings. They were more satisfied with their own

achievement. Caregivers of patients dying elsewhere and with lower satisfaction

reported that patients had to leave home due to acutely developing symptoms that they


25

were unable to manage. This is a retrospective study, making the reports of caregivers

vulnerable to recall bias.

Hudson et al’s (2006) prospective cohort study in Australia included 35 primary

caregivers of people with cancer receiving home care. Data were obtained at entry and a

five-week follow-up using the HADS measure of preparedness, competence, social

support, anxiety and self-efficacy. Self-reported anxiety at admission was associated

with caregivers having lower levels of psychosocial functioning five weeks later.

Therefore the baseline mental health status of informal caregivers can predict

susceptibility to caregiver burden and potentially identify vulnerable caregivers in

advance. This was a secondary analysis of existing data, thus some independent

variables of interest were not included, and this is also a particularly small sample size.

Singer et al’s (2005) case control study in Israel found that facilitating care at home

came with considerable financial and emotional burden. However, through 159

interviews with caregivers of patients with and without access to home care, it was found

that home-based caregivers were still more satisfied with their caregiving experience for

having facilitated the preference of the patient. This is congruent with the findings of

Carlsson et al (2003). There was a relatively low response rate for this study leading to

potential selection and convenience biases, and most caregivers were not spoken to

until 6 months into bereavement, leading to potential recall bias or social desirability

bias.

Wasner et al’s (2013) qualitative study with twenty-seven caregivers of patients with

malignant brain tumours in Germany assessed QoL vis-à-vis burden of care. Only eight

patient-caregiver dyads used home care, however, this diagnostic group is a novel

amidst the literature. Caregiver QoL was most determined by the burden of care, the

patient’s mental state and cognitive impairment. The sample, particularly the home care

sample (N=8), is obviously very small. Due to critical conditions, many patients were

unable to complete the HADS assessment of their mental health. This may indicate brain

cancer as a vulnerable population with high ADL and IADL needs.

Summary

In conclusion, the literature currently available through the searched databases

discusses health problems associated with caregiver burden, unmet needs reported by


26

caregivers, and determinants of satisfaction and burden. Unpaid caregivers are

susceptible to worsening of their physical and mental wellbeing, though there was less of

a consensus around physical health with Kenny et al (2010) finding it better than the

general population in their cross-sectional analysis. The health of the caregiver tends to

deteriorate with that of the patient, which was found using diverse validated burden

measurement instruments (HADS, EORTC, CRA, etc.). Many of the unmet needs of

caregivers are related to information and guidance around providing care. The

heightened involvement of health professionals was among determinants of satisfaction

for caregivers. Determinants of burden were related to service utilization, patient

condition and the amount of time they spent providing care.

The findings of this review will inform variable inclusion for the analyses and illuminate

missing variables. Glajchen (2012) found disability and mobility to drive caregiver

burden, which is adequately captured by the comorbidity measure used in this thesis.

They also found that the living arrangements of patients were significant determinants of

caregiver burden, which, too, is in these data. Götze et al (2014) found marital status

influential, while Hirdes et al (2012) identified the home care agency, hours of unpaid

care and hospitalizations to be important, which are also in these data. However, patient

depression was also found to be important (Götze et al, 2014) and is not adequately

captured by the comorbidity measure. Guerriere et al (2013) found characteristics of

service providers to drive burden and satisfaction, which these data also do not have.

Gardiner et al’s (2014) paper, included in the previous section, found variability in

caregiver strain to be tied to SES and ethnicity. Carlsson et al (2003) found place of

death significant in their retrospective study of caregiver satisfaction. While these data

contain places of death, it will not be included as predictive of burden due to temporal

order; this may impact satisfaction in bereavement, as was observed by Carlsson et al

(2003), but cannot logically affect burden before death has happened as is the case with

prospectively gathered data.

Finally, in addition to the limitations outlined in the Search Strategy (Section 2.1), there

are gaps in the current literature that weaken this review. The broader thesis will attempt

to address some of these. Common weaknesses across the existing studies include

small sample size, homogenous cohorts, and a lack of longitudinal data tracking the size

and determinants of caregiver burden over the palliative trajectory (Wasner et al, 2013;

Kenny et al, 2010; Docherty et al, 2008). Three longitudinal studies were included, one


27

of which only had one follow-up. While caregivers are interested in disease specific care

information, most studies do not control for primary diagnoses or thoroughly describe

their sample clinically. Rather comorbidity-type measures are used, which are inherently

secondary to the primary diagnosis. This review also brings to light the physical health

problems associated with caregiving, which the analyses found here will not be able to

speak to. The charts for this section of the review are found in Appendix C.

2.4 Neoplasm specific palliative care research

Palliative care is becoming increasingly specific and tailored to the individual patient, and

this is true for their specific primary diagnosis as well (Gaertner et al, 2011). One

systematic review on this topic was found through this search, Lester et al (2012), which

focuses on specific palliation regimens for NSCLC. A modest increase in survival is

associated with this disease specific approach based on 14 RCTs. The contribution of

these findings to this thesis is the fundamental and broad suggestion that symptoms and

palliation needs vary from malignancy to malignancy, and may therefore be influential in

predicting cost and caregiver burden. This review will consider articles discussing

malignancy-specific symptom control so as to find any traces of this ostensible gap in

the home care cost (Section 2.2) and caregiver burden (Section 2.3) literature.

Figure 3 depicts the process by which articles were selected, culminating in seven

articles in seven journals being charted and synthesized. These include one systematic

review (14%), one narrative review (14%) and five original studies (72%), representing

538 study subjects and three countries. These proportions speak to the topics’ relative

scarcity compared to the former two topics of discussion. Much of the palliative

oncologic literature has followed Temel et al’s seminal article published in 2010, thus this

research base is relatively small and currently overwhelmed by commentary and review

articles, generally written by physicians. Given that this is the case, many articles were

excluded for “No applicable design,” but some of these will be discussed. First, charted

articles will be described, all of which discuss palliative care for specific neoplasms.

Bakitas et al (2013) interviewed thirty-five oncology clinicians in the United States in their

qualitative study to gage their perspectives on caring for advanced cancer patients. They


28

generally agreed with the early integration of palliative care, and identified haematologic

patients as “different” than other cancer patients due to the unpredictability and rapidity

with which symptoms can develop. It is consequently more difficult to live at home or

receive informal care from unpaid caregivers without complications and rehospitalization.

This is consistent with previous articles in this chapter and fortifies the existing literature.

Weaknesses include that the oncologists in this study work in a palliative oncologic team

setting that is very uncommon and may not representative of American oncologists

generally. They are also based in an ethnically and racially homogenous region.

Figure 3: PRISMA flow chart for palliative oncology literature search

Bukki et al’s (2013) observational study assessed symptoms and treatment intensity

over the palliative trajectory among ninety-six decedents in Germany between 2009-

2011. Forty-three had haematologic cancers and forty-nine had solid tumours while four

Iden

tific

atio

n S

cree

ning

E

ligib

ility

In

clud

ed



Records after duplicates removed (N=2307)

Records screened (N=2307)

(N=600)

Records excluded (n=2267)

(N=564)



No applicable design (N=33)



29

had unknown primaries. Data on the last two weeks of life were available for sixty-two

patients, of which forty-seven received aggressive EoL care and fifteen used palliative

care services. While palliative radiotherapy and antineoplastic treatments at the EoL

have been shown as helpful among the NSCLC population, they were associated with

hospital death and greater symptoms in this cohort. As a retrospective chart review, the

study may be vulnerable to assessor bias. It is also a cross-sectional analysis and small

sample. This study groups solid tumours together as is seemingly customary in existing

research.

Fauci et al’s (2012) chart review studied care utilization and reason for hospitalization

during the last six months of life among 268 American decedents with gynecologic

cancers between 2007-2010. About 58% of them received antineoplastic treatments

during the last six months of life and 81% had at least one admission to the hospital. The

most common complaint for this particular neoplastic category was gastrointestinal

complications. They were generally referred to palliative care relatively late, dying thirty

days after referral on average. Earlier referral for this population was promoted to reduce

hospitalizations. This, too, is an ethnically homogenous cohort that was analyzed via

retrospective chart review.

Gaertner et al (2011) worked with an interdisciplinary group (palliative care, oncology,

radiotherapy) to specify timing and early integration approaches to palliative care for

nineteen malignancies. The objective was to outline disease specific approaches to

palliative cancer care. SOPs were generated that outlined a disease-specific point in

each disease trajectory to initiate early integration of palliative care with disease-specific

green and red flags. Many malignancies shared metastasis as a green flag for palliation

referral, however, others do not and secondary green flags are entirely variable. While

the merits of early integration are widely accepted, these SOPs are merely suggestions

made by the authors and are not similarly widely accepted or standard procedures. They

have not been implemented, tested or validated. Notably, they do discriminate between

solid tumours rather than homogenizing them.

Lester et al’s (2012) systematic review of NSCLC-specific palliative care needs

synthesized results from 14 RCTs administering radiotherapy to the intervention group.

They did not find strong evidence supporting symptom control; conversely, they found

high dose regimens to worsen toxicity. However, they did find evidence that it extends


30

survival for the patients with these higher doses. Consequently, for this particular cancer

subpopulation, it is a trade-off to consider. Focus on this treatment has followed Temel

et al’s (2010) research and contributes to that existing literature. It is included here

because it contributes to the evidencing of variability in symptom control across

malignancies, therefore justifying the inclusion of a diagnostic predictor variable.

Manitta et al’s (2010) narrative review focused on haematological cancer patients in

palliative care. Evidence suggests that these patients access palliative and home care

much less frequently. They are also more likely to die in hospital as they require many

more treatments, namely regular blood transfusions. These patients present a well-

known challenge to prognostication due to the speed at which an event may become

fatal. This literature review is non-systematic and may have assessor bias, publication

bias or English language bias.

McGrath et al’s (2007) study, too, focuses on haematologic patients in palliative care

and entailed interviews with 25 palliative care nurses. They reported need for the

integration of disease-specific palliative care in haematology. This further established the

distinction of haematologic cancers from other primary cancer diagnoses, but also

perpetuates the lack of diagnostic diversity.

Temel et al’s (2010) RCT included 151 patients with NSCLC in the United States with

palliative radiotherapy extended to randomly selected participants. The outcome was

QoL rather than life extension, as it has been in similar RCTs. QoL was measured at

baseline and week twelve using the Functional Assessment of Cancer Therapy-Lung

(FACT-L) and HADS. It was found that patients referred into early palliation had a better

QoL and were less likely to experience depression than those receiving standard care.

The study took place in a specialized tertiary care site with a group of thoracic oncologist

care providers and the patient group is ethnically and racially homogenous.

Non-original review articles

There is a plethora of review and special articles that emerged with this search that were

not conducive to the typical review process, but were considered due to lack of literature.

These articles do not present hard evidence, but the input of people working in the field

vis-à-vis the evidence that does exist.


31

There is a general consensus across these articles, often written by oncologists and

other doctors, that palliative cancer patients will follow unique trajectories determined in

part by their oncologic characteristics (Von Roenn et al, 2011; Rocque et al, 2013;

Rangachari et al, 2013; Ramchandran et al, 2013; Dennis et al, 2011; Janjan, 2011;

Flechl et al, 2013). Janjan et al (2011) provides an example of pulmonary emboli (PE)

being much more common in lung cancer patients, implicating an alternative appropriate

caregiver response to the symptom of mild chest pain than could generally be advised.

Flechl et al (2013) discuss the unique symptoms associated with glioblastoma

multiforme (brain cancer), including cognitive deficits that can compromise patient ability

to communicate symptoms. One might also note that they are all relatively recent, with

the earliest belonging to the previously mentioned Oncology volume dedicated to this

topic in 2011.

Studies presenting diagnosis-specific considerations for palliative care delivery cannot

explicitly speak to cost and caregiver burden. Because it has not been explicitly studied

in the known literature, potentiality for determinacy of primary cancer diagnosis can only

be inferred by reconciling the fragmented bodies of literature presented in the three

sections. In lieu of diagnostic diversity in existing literature, the analyses in this thesis

cannot only address this gap but illuminate whether such distinctions are necessary and

informative, or if proceeding with the convention of homogenizing solid tumours is

sufficient and inconsequential. This gap in the research literature has only been

thoroughly looked into with respect to NSCLC and haematologic tumours at this point.

Finally, this analysis will not be able to address the ethnic homogeneity that is found yet

again in this review (Bakitas et al, 2013; Fauci et al, 2012; Temel et al, 2010).

The broader thesis will attempt to address some of these. Common weaknesses across

the existing studies include small sample size and samples that are both

demographically and clinically homogenous. There is disproportionate focus on

haematologic malignancies and NSCLC, though Gaertner et al (2011) illuminate need to

identify other vulnerable patient groups within cancer. Single-neoplasm studies are

useful in illuminating differences between primary diagnoses, but statistical comparative

inferences cannot be made. This study assembles a clinically diverse population in the

same program and dataset. The charts for this section of the review are found in

Appendix D.


32

2.5 Summary

This literature review looked at palliative home care cost research, caregiver burden

research, and palliative oncologic research. This was done so that inferences could be

made in order to justify the analysis of primary cancer diagnosis as it impacts on cost

and caregiver burden. Also, so that it could inform the inclusion of independent variables

alongside primary cancer diagnosis in the analytic models. This will help to form the

most comprehensive possible analyses to draw inference about the role of diagnosis

from.

Variable inclusion is also guided by the Andersen and Newman model (Section 3.1.3),

which the cumulative findings of this review generally corroborate. This model

conceptualizes determinants of service use, not caregiver burden. Findings in the first

section identified ‘need factors,’ including having cancer (no distinction) (Enguidanos et

al, 2005) and lower physical function (Guerriere et al, 2010), as determinant of home

care cost. While these analyses did not include primary cancer diagnoses

(conceptualized here as the ICD-10 categories), the third section expressed differential

treatments and expectations depending on this patient characteristic (Lester et al, 2012;

Gaertner et al, 2011; Manitta et al, 2010). The analyses can speak to whether or not

such differential treatment processes result in significantly different service use and

thereby cost. ‘Predisposing factors’ were also identified as significant drivers of cost,

including marital status, living arrangement (Guerriere et al, 2010; Chai, 2013), and SES

(Gardiner et al, 2014).

The caregiver burden search also corresponded with the conceptual framework. An

‘enabling factor,’ home care agency, was found to be significant by Hirdes et al (2012).

‘Need factor’ of comorbidity and ‘predisposing factors’ of marital status and living

arrangement (Glajchen, 2012) were also found to be significant and are able to be

implemented given the data. Caregivers also reported difficulty with care tasks (Parker

Oliver et al, 2014; Docherty et al, 2008) which is where it coincides with the third section,

where research suggested that these tasks are variable depending on diagnosis (Janjan

et al, 2011; Flechl et al, 2013). One predisposing factor that is underrepresented,

potentially significant, and cannot be addressed by this study is race and ethnicity

(Temel et al, 2010; Fauci et al, 2012; Diwan et al, 2004; Docherty et al, 2008).

33

Chapter 3 Methodology

3.1 Overview of the larger study

The data used for this thesis are part of a larger study of caregiver burden over the

palliative trajectory and place of death among CBPHC patients (Guerriere, 2012). The

primary investigators are Drs. Denise Guerriere and Peter C. Coyte . Co-investigators

are Drs. Amna Husain (MD), Denise Marshal (MD), Hsien Seow (PhD), Kevin Brazil

(PhD), Eric Nauenberg (PhD) and Julie Darnay.

The study is focused largely on place of death, particularly home death (Guerriere,

2012). Shifting end-of life careand the site of death away from institutional settings to the

family home is a major part of the incentive to establish CBPHC. It is also a common

preference among patients. The study sought to identify modifiable factors that

influenced place of death among cancer patients in home care. In addition, it sought to

identify differences in resource utilization between patients dying in institutional settings

versus dying at home. Similarly to what was seen with health care cost estimates in the

literature review, the proposal evokes the importance of perspective; home death is

often associated with inherently lower cost, and this is perhaps true to the Ministry of

Health, but it may be more complex when factoring in unpaid caregiving.

The Andersen and Newman model was used to inform predictors of place of death and

therefore the data that was collected. The use of the framework by those who collected

the data is what enabled this study to have the majority of the desirable covariates be in

the data. The patients in this sample died in long-term care facilities (LTC), hospices,

hospitals or at home.

This is a prospective cohort study with data collected bi-weekly over the course of the

palliative trajectory. Patients and caregivers were recruited from Toronto’s Temmy

Latner Centre for Palliative Care at Mount Sinai Hospital and Hamilton Niagara

Haldimand Brant LHIN Community Palliative Care Teams. Collecting longitudinal data is

intended to enable assessment of cost of an entire episode of palliative care and also to

factor in the differential length of treatment, which was exhibited in the data description

(Figure 4). Caregivers recruited into the study had to be a) the primary caregiver, b)

Chapter 3: Methodology R. Redmond-Misner

34

caregiver to a patient with cancer, c) fluent in English and d) > 18 years of age. The

criteria for cancer was imposed because 85% of admissions into the program live with

cancer, and also to ensure diagnostic homogeneity (Guerriere, 2012).

Ethics approval has been gained for these data through the University of Toronto.

Standard procedures were followed to keep patients and caregivers anonymous and

encrypted in all analytic datasets.

3.2 Data

3.2.1 Data origin

Data were collected during 2011 and 2012 from caregivers of patients admitted into two

CBPHC programs in Ontario. They include the TLCPC at Mount Sinai Hospital in

Toronto and the Hamilton Niagara Haldimand Brant (HNHB) LHIN’s CBPHCT in

Hamilton.

3.2.2 Data description

Researchers collected the TLCPC and HNHB LHIN data via successive interviews

(N=1,940) with caregivers of patients (N=327) on a bi-weekly basis from admission into

the program until death. Prospective cohort studies, whilst vulnerable to social

desirability bias in caregiver reporting (Guerriere et al, 2010), are resilient against the

recall bias associated with retrospective studies (Kenny et al, 2010). Selection biases

may exist given that the data collection is contingent on willingness to partake. In Jacob

et al’s (2011) study, multiple caregivers per patient threatened to obscure and bias their

results, but this is not common in this sample.

There are both challenges and advantages with these data. The collection process is

subject to the unpredictability of the palliative trajectory; some patients exceed their

prognosis while others are in the program for a very short time. Patients enter and exit

the data at different times. Using the months or years to define time, in this context,

would be a meaningless measure. It would not tell us where any of the patients are in

their palliative trajectory. This has been addressed by using a time variable that


35

measures every two weeks prior to death, resulting in an unbalanced panel. This is to

say that, of forty-three time periods spanning 602 days before death, not every patient

appears in each one. Recall the distinction between EoL and palliative care that was

made in Section 1.1; some patients are in palliation long before the EoL. Figure 4 shows

the number of interviews conducted for each time period. The quantity dwindles off as

the distance from death becomes greater, which is not surprising. Ultimately, it makes

much more sense to assess costs across patients defining time this way than to

compare the costs incurred in an arbitrary month. If patients did not die or dropped out

during the study, resulting in no recorded date of death, their observations cannot be

used because the time period is not calculable.

Figure 4: Observations per time period

In light of the additional complexity associated with panel data, it also has considerable

analytic advantages in terms of robustness. One of the most common weaknesses in the

literature review were the limitations to accuracy in cross-sectional data; namely, that

they represent a static health status or other outcome (Aoun et al, 2005). It is unknown

whether associations that appear in cross-sectional analyses hold true over time, while

relationships that hold true over the palliative trajectory are more convincing. Panel

estimators and their error terms control for unobserved heterogeneity and systematic

drop out (Hill et al, 2011). The trouble with utilizing the unbalanced panel for the sake of

the time series element is that the majority of the explanatory variables are time-

invariant, categorical variables that cannot be incorporated into some conventional

050

100

150

200

250

Careg

ivers

Interv

iewed

0-14

15-28

29-42

43-56

57-70

71-84

85-98

99-11

211

3-126

127-1

4014

1-154

155-1

6816

9-182

183-1

9619

7-210

211-2

2422

5-238

239-2

5225

3-266

267-2

8028

1-294

295-3

0830

9-322

323-3

3633

7-350

351-3

6436

5-378

379-3

9239

3-406

407-4

2042

1-434

435-4

4844

9-462

462-4

7647

7-490

491-5

0450

5-518

519-5

3253

3-546

547-5

6056

1-574

575-5

8858

9-602

Time Period (Days Before Death)

Observations per Time Period


36

econometric models for panel data such as fixed effects (FE) regressions (Hill et al,

2011). Predictors of interest here are often socio-demographic and clinical in nature and

therefore do not change over time like the dependent variables do.

3.2.3 Variable description

Andersen and Newman model of health services utilization

The Andersen and Newman framework for the use of health services informed data and

variable collection (Andersen et al, 1973). This framework articulates three broad

determinants of health service use including a.) predisposing factors (i.e. demographics),

b.) enabling factors (i.e. community characteristics, employment) and c.) need factors

(i.e. comorbidity, illness severity) (Guerriere, 2012). As will be shown in the discussion of

the Ambulatory Home Care Record (AHCR), these costs are generated based on the

services that the patient has used in a given time period. Therefore the extrapolation of

this model onto cost is not a big jump. The Andersen and Newman model has since

been updated (Andersen, 1995), however the three main factors given here are still the

primary drivers of utilization.

a.) Predisposing factors

Predisposing factors are generally demographic and speak to the socio-cultural

characteristics that might inform service use. Data were collected for the socio-

demographic predictors established in previous research (Skinner, 2005; Motiwala,

2006; Aoun et al, 2013; Kuluski, 2010; Bainbridge et al. 2011). Sex, education, age, and

marital status were collected for all patients and caregivers. The living arrangement of

the patient, relationship of the caregiver to the patient, and caregiver burden scores were

also collected. Caregiver burden scores were generated using the Canadian Caregiver

Burden Scale in End-of-Life Care (CBS-EOLC) (Dumont et al, 2008).

b.) Enabling factors

Enabling factors are those that might enable/disable one from obtaining care, including

community characteristics, availability of services and accessibility/affordability of

services. The city of residence, which corresponds with which program the patient was

enrolled in, is a potentially explanatory community characteristic. This has been found to


37

be significant by Hirdes et al (2012). Covariates that may effect caregivers’ ability to

provide care include caregiver employment status, days spent caregiving overnight, and

caregiver burden scores.

c.) Need factors

Patient comorbidity and their primary cancer diagnosis were both collected. Comorbidity

scores were generated using the Charlson comorbidity index. Primary diagnoses were

coded using the WHO’s ICD-10, which provides categories that are premised on the site

of the tumour. Haematologic malignancies are not represented here, as this

categorization reflects the prominent cancer types in the sample. Table 1 gives their

composition.

Table 1: ICD-10 categories

ICD-10 Directory Patients (Obs.) Reference “Malignant neoplasms of respiratory and intrathoracic organs” (C30-39)

75 (398) Lung

“Malignant neoplasms of urinary tract” (C64-68) 17 (95) Urinary “Malignant neoplasms of digestive organs” (C15-26) 94 (432) Digestive “Malignant neoplasms of eye, brain and other parts of central nervous system” (C69-72)

14 (68) Brain

“Malignant neoplasm of breast” (C50) 33 (159) Breast “Malignant neoplasms of female genital organs” (C51-58)

22 (150) Gynecologic

“Malignant neoplasms of male genital organs” (C60-63)

18 (119) Male organs

Other 37 (241) Other

The presence of two cancer sites or an unknown primary diagnosis could complicate this

approach, but such problems were very limited. Patients with unknown primaries were

categorized as ‘other.’ Patients with multiple disease sites were rare and had multiple

disease sites in the same category, thus it did not complicate their categorization. For

example, one patient had malignancies of the pancreas and duodenum (digestive); one

had malignancies of the bladder and kidney (urinary); another had malignancies of the

bronchus and lung.


38

Conceptual framework for caregiver burden

Many existing conceptual frameworks for caregiver burden focus on dementia, namely

Alzheimer’s disease (Pallett, 1990; Conde-Sala, 2010). Though the conceptual

framework used by Giverns et al (2004) is largely in agreement with Alzheimer’s oriented

models such as the stress process model (Conde-Sala, 2010), it is intended for patients

with cancer. In fact, it is similar to the Andersen and Newman framework in terms of the

independent variables it prioritizes.

Given et al’s (2004) approach considers patient and caregiver characteristics as well as

patient symptom experience. It emphasizes a.) patient and family characteristics (i.e.

demographics, relationship to patient), which is also prioritized by Pallett (1990), Conde-

Sala et al (2010) and Andersen and Newman (1973); b.) ‘care situation’ or sources of

demand on the caregiver (i.e. comorbidity); and c.) ‘care process,’ which does not

correspond with the independent variables but the prospective cohort study design that

captures the longitudinal care process. Their study collected many of the same variables

assembled by the researchers who collected these data. Both a.) and b.) correspond

well with Andersen and Newman’s ‘predisposing’ and ‘need’ factors, indicating that many

of the independent variables are going to be the same with this dependent variable as

with costs. What is missing is the ‘enabling’ factors that implicate the home care agency,

however, due to the broader conceptualization of these themes on behalf of Given et al

(2004), this may be considered part of the care situation.

Ambulatory home care record

The cost of care was estimated using the AHCR (©Coyte & Guerriere 1998; Guerriere et

al, 2011), a prospective questionnaire that measures the cost of ambulatory and home-

based care from a societal perspective (Chai et al, 2013). Costs fall into categories of

public, private and unpaid care costs, shown in Table 2.

Table 2: AHCR framework for the identification of palliative care costs

Expenditure category Resource Public costs Ambulatory

All services financed by the government Healthcare professional appointments Clinic visits


39

Inpatient Home

Laboratory and diagnostic tests Treatment (chemotherapy and radiation) Medications Supplies and equipment Emergency room visits Hospitalizations Nursing home Hospice care Home care: nursing, personal

support/homemaking, occupational therapy, physiotherapy, oxygen therapy, diagnostic tests

Private costs Out-of-pocket Third-party insurance

All healthcare costs not publicly insured and/or financed

Healthcare professional appointments Home caregivers Travel expenses Medications Supplies and equipment Insurance payments Healthcare appointments Medications Hospitalizations Supplies and equipment

Unpaid care costs Time devoted by family, friends/neighbours to caregiving

Time lost from paid market labour Time lost from leisure/household work

(Chai et al, 2013)

Psychometric properties refer the ability of the tool to actually measure the construct of

interest consistently. The psychometric properties of the AHCR have been validated in a

CIHR-funded study focusing on adults with Cystic Fibrosis (Guerriere et al, 2006). It has

since been used in more than ten studies including respondents of diverse age, care

setting and clinical condition as well as in multiple countries (Guerriere, 2012).

Caregiver burden scale in end-of-life care

Caregiver burden was assessed using the CBS-EOLC (Dumont et al, 2008) that was

developed among Canadian caregivers of family members or friends with cancer in

palliative care (Dumont et al, 2008; Guerriere, 2012). Items in the questionnaire relate to

emotional, social and physical burden but not financial burden. Higher scores reflect

stronger caregiver agreement with a series of statements shown in Table 3.


40

Dumont et al (2008) have validated the psychometric properties of the CBS-EOLC and

its reliability with fatigue and depression. It is a sixteen item questionnaire measuring

caregivers’ “cognitive appraisal of the gap between potential assistance and support

perceived to be accessible for dealing with the demands of providing care” (Guerriere,

2012, p. 7).

Table 3: Caregiver burden scale in end-of-life care

How often do you experience this feeling in your role as caregiver?

Currently…

Never From time to time

Fairly often

Very often

1. Do you ever find that the tasks required in caring for the sick person are too demanding?

1 2 3 4

2. Do you ever feel emotionally exhausted? 1 2 3 4 3. Do you ever feel that you no longer have the strength to care for the ill person?

1 2 3 4

4. Do you ever feel unable to go on? 1 2 3 4 5. Do you feel overwhelmed by everything that has happened to you?

1 2 3 4

6. Do you have the impression that your role as caregiver is making you physically ill?

1 2 3 4

7. Do you ever feel emotionally drained? 1 2 3 4 8. Do you ever feel that you are no longer capable of caring for the ill person?

1 2 3 4

9. Do you ever feel physically exhausted? 1 2 3 4 10. Are you ever afraid that you won’t be able to hold out much longer?

1 2 3 4

11. Do you feel like you are at the end of your rope?

1 2 3 4

12. Are you comfortable with the type of care you must provide the ill person with?

1 2 3 4

13. Do you ever feel discouraged by all the tasks you have to accomplish?

1 2 3 4

14. Do you ever think that caregiving is too demanding an experience for you?

1 2 3 4

15. Do you ever have the impression that you have lost control over your life?

1 2 3 4

16. Do you ever have the impression that you carry to heavy a burden?

1 2 3 4

(Dumont et al, 2008)

It is common to use an adaptation of the Zaritt burden interview (Zaritt et al, 1980), which

is a widely used tool and the preferred measure of the American Psychological

Association (Hébert et al, 2000; Lai, 2007). However that is because it is intended for

dementias (Guerriere, 2012). The CBS-EOLC was preferable because it was directly


41

applicable to this population and caregiver environment as opposed to this widely used

measure that is technically intended for dementia caregivers.

Charlson comorbidity index

The Charlson comorbidity index is a prospectively applicable tool for measuring

comorbid conditions in longitudinal studies (Charlson et al, 1987). This, too, is a

validated and widely used measure that is delineated and endorsed by the WHO

(Sundararajan et al, 2004). The index, shown in Figure 6, factors neoplastic

malignancies into the comorbidity score but attributes the same number of points to any

solid tumour, leukemia or lymphoma and any metastasized disease. Thus it does not

distinguish among primary diagnoses or measure the same thing as the categorical

diagnostic variable created using the ICD-10.

Figure 5: Charlson comorbidity index

Scoring: Comorbidity (Apply 1 point to each unless otherwise noted) 1. Myocardial Infarction 2. Congestive Heart Failure 3. Peripheral Vascular Disease 4. Cerebrovascular Disease 5. Dementia 6. COPD 7. Connective Tissue Disease 8. Peptic Ulcer Disease 9. Diabetes Mellitus (1 point uncomplicated, 2 points if end‐organ

damage) 10. Moderate to Severe Chronic Kidney Disease (2 points) 11. Hemiplegia (2 points) 12. Leukemia (2 points) 13. Malignant Lymphoma (2 points) 14. Solid Tumor (2 points, 6 points if metastatic) 15. Liver Disease (1 point mild, 3 points if moderate to severe) 16. AIDS (6 points)

Scoring: Age

1. Age <40 years: 0 points 2. Age 41‐50 years: 1 points 3. Age 51‐60 years: 2 points 4. Age 61‐70 years: 3 points 5. Age 71‐80 years: 4 points

Source: http://www.uroweb.org/fileadmin/livesurgery/Charlson_Comorbidity_Index.pdf……..


42

The points attributed to each condition in Figure 6 can be used to calculate comorbidity

scores that reflect the extent to which the patient has comorbid conditions, but also to

calculate life expectancy (Sundararajan et al, 2004). That is not how it is used here. It is

used as the formerly described representation of comorbidity severity and predictor of

costs and caregiver burden vis-à-vis the conceptual frameworks that posit its potential

role in both. It has been used this way in similar studies (Guerriere et al, 2010).

3.2.4 Variable assessment

Regression techniques function under the assumption that independent variables are not

collinear (Hill et al, 2011). In addition to the literature review and conceptual frameworks,

independent variable inclusion will be guided by a series of tests to derive correlation

coefficients. Correlation matrices were generated using Pearson’s correlation coefficient

for continuous variables (detailed description: Appendix E), phi coefficients for binaries

(detailed description: Appendix F), Cramér’s V test for association between nominal

variables (Appendix F), and polychoric coefficients for correlation between continuous,

ordered and binary variables (see Ekström, 2008). Results were used to inform

inclusion/exclusion of independent variables in the econometric models.

These correlation coefficients have all originated with the work of Pearson (1900; 1907;

1913; Ekström, 2008). While they are calculated differently, they are interpreted the

same way (Sanyal et al, 2009). For all of these correlation coefficients, -0.3 to 0.3 is

considered indicative of little or no association, -/+0.3 to -/+0.7 is considered weak

association, and -/+0.8 to -/+1.0 is considered a strong association (Sanyal et al, 2009).

3.2.5 Variable inclusion

Collinearity testing informed which variables were included in the final models. All

correlation coefficients outlined in Section 3.1.4 (phi, Cramér’s V, Pearson, polychoric)

are reported in Appendix G. While some authors have expressed acceptability for

correlations as strong as 0.7 (Sanyal et al, 2009), coefficients greater than 0.35 were

generally excluded. This was the case for patient and caregiver sex (phi=-0.5412);

patient living arrangement and marital status (Cramér’s V=0.5757); patient marital status

and caregiver relationship to the patient (Cramér’s V=0.5691); caregiver relationship to

the patient and patient living arrangement (Cramér’s V=0.5209); ministerial cost and


43

emergency room visits (polychoric=0.4802); societal cost and emergency room visits

(polychoric=0.4017); hospitalization and emergency room visits (polychoric=0.6509); and

caregiver days overnight and hospitalization (polychoric=0.7773). All aggregated costs

were highly correlated, as can be expected (polychoric & Pearson > 0.7). The

independent variables in the final models do not exceed correlation coefficients of 0.35.

Several data generating processes with different combinations of correlatively safe

independent variables were assessed for higher R2 values. These values indicate the

proportion of variability in the data that the model can explain (Hill et al, 2011). The

combination that consistently produced higher R2 values included site (Toronto or

Hamilton), patient age (and a quadratic form, patient age squared), comorbidity score,

caregiver education, patient education, caregiver employment, ICD-10 category, patient

living arrangement and caregiver sex. Most of these had correlation coefficients < 0.3,

but patient living arrangement was chosen at the expense of relationship to the patient

and caregiver sex is included at the expense of patient sex, with which they were highly

collinear and only one could be included at a time. Caregiver burden scores will be

included as a predictor in the cost regressions and time-spent caregiving will be included

as a predictor of caregiver burden scores.

3.2.6 Categorical variable interpretation and baseline

Many of the independent variables are categorical. These covariates are incorporated

into linear regressions using dummy coding, 1 if the condition is true and 0 otherwise.

The interpretation of the results, specifically beta coefficients (βx), is different from

continuous variables wherein the direction the relationship with the dependent variable is

expressed (Hill et al, 2011). For categorical variables, coefficients express the difference

in means for the indicated subgroup, and ‘otherwise’. If the coefficient for caregiver sex

were negative for male caregivers in a regression on out-of-pocket cost, it would suggest

that their out-of-pocket expenses were lower than their female caregiver counterparts.

For non-binary categorical variables, a dummy variable is created for each condition and

one category is excluded to avoid perfect multicollinearity. The excluded category

essentially serves as a baseline to compare other categories to the way that male was

just interpreted vis-à-vis female. The baseline categories for the categorical independent

variables are shown in Table 4.


44

Table 4: Categorical variable baselines

Variable Baseline Site Toronto Caregiver education High school Caregiver employment Retired Primary diagnosis Lung Patient education High school Caregiver sex Female Patient living arrangement With spouse

3.3 Descriptive statistics

Descriptive statistics (frequencies) were generated in Stata 13 (StataCorp, 2013) for

demographic characteristics of both patients and caregivers.

3.4 Cost analyses

3.4.1 Logarithmic transformation and quadratic forms

Financial variables are often positively skewed or heteroskedastic due to a lack of non-

negative values (Hill et al, 2011; Frees 2010; Manning et al, 2001). Logarithmic

transformation is used to normalize the distribution of the dependent variable, thus

satisfying the distributional assumption and ensuring the best linear unbiased estimators

(BLUE) (Hill et al, 20111; Manning et al, 2001). This method is easily applicable to the

aggregated costs, which do not contain zeros.

Frequently, cost variables will contain zeros that are problematic for the truncation of the

variable via logarithmic transformation. Given how the logarithm is calculated, it cannot

be practically used for zero or negative values. In national-level databases, for example

the Medicare reimbursement data used by Beeuwkes Buntin et al (2003), many people

will not have received any reimbursement ($0.00). This skews the data but these

observations are also dropped by logarithmic transformation. This will become a


45

prominent consideration in the modeling of disaggregated costs that do contain zeros in

Section 3.3.3.

Although lines of fit are straight, forms with log, quadratic or cubic transformed variables

have a different slope at every point (Hill et al, 2011). This changes how the output is

interpreted. In log-log regression output, the βx parameter reflects the elasticity of y to x

rather than slope (Hill et al, 2011). In a linear-log model, the slope is βx/y, meaning that

the slope declines or becomes less steep as y increases. Log-linear models, like the

ones in this thesis, exhibit the inverse. If βx > 0, the slope becomes steeper as y

increases (Hill et al, 2011).

In discussing the linearizing of potentially nonlinear components of the data generating

process, it is also notable that a quadratic form will be included for any age used as an

independent variable. Quadratic forms represent curvilinear relationships. The direction

that age drives many variables, such as income or health status, can peak and decline

as age grows rather than following a consistent +/- causal trajectory (Hill et al, 2011).

This nonlinearity can be addressed by using a polynomial model with a quadratic form,

age squared, in addition to age. This is conventional practice, though it may be a more

contestable assumption in the context of palliative or EoL care given that it has been

shown that costs rise as death approaches (i.e. as age rises) (Guerriere et al, 2010) and

patients are often not in the sample long enough for their age to change. However, such

a peak and decline may exist across patients.

3.4.2 Aggregated costs

Payer perspectives

The literature review revealed considerable disagreement surrounding the cost of home

care that was seemingly rooted in the payer perspective adopted by the author. The

AHCR is conducive to observing these mutually exclusive perspectives independently as

well as accumulatively, representing a societal perspective that acknowledges cost

irrespective of payer. The direction and magnitude of determinacy attributable to the

independent variables – particularly the variable of interest, primary diagnosis – may

differ depending on payer. Aggregated costs will be analyzed from the perspective of the

a.) Ministry of Health, b.) unpaid caregivers, and c.) society. This can also illuminate the

branch of societal costs driving the coefficients from that data generating process.


46

a.) Ministry of Health perspective

This dependent variable constitutes aggregated components of the public cost of home

care vis-à-vis the AHCR (Table 4), including healthcare professional appointments, clinic

visits, laboratory and diagnostic tests, treatment (chemotherapy and radiation),

medications, supplies and equipment, emergency room visits, hospitalizations, nursing

home and hospice care, home care nursing, personal support and homemaking,

occupational therapy, physiotherapy and oxygen therapy (Chai et al, 2013). These costs

were generated using fee-for-service rates for physician and laboratory costs; Toronto

CCAC rates for home-based provider services; the Ontario Drug Benefit (ODB) for

medication costs.

b.) Unpaid caregiver perspective

This dependent variable constitutes aggregated components of the unpaid cost of home

care vis-à-vis the AHCR (Table 4). This includes time devoted by the primary caregiver

to caregiving, time lost from paid market labour and time lost from leisure and household

work. It will also include the out-of-pocket spending on behalf of unpaid caregivers

(travel expenses, medications, supplies etc.). Time losses were valued using the human

capital approach, which takes the average earning by age and gender to value the time

withdrawn from leisure and work (Guerriere, 2012; Rice, 1989).

c.) Societal perspective

This dependent variable aggregates all components of the AHCR (public, private [i.e.

third-party insurance], and unpaid care) (Table 4). These were collected bi-weekly.

Estimator

The advantage of panel over cross-sectional data is the potential to observe the

dependent and independent variables over time, within- and between-individuals (Hill et

al, 2011). Cross-sectional multiple regression analysis summarizes the data generating

process as:

(ln)yi = β0 + β1x1 + β2x2 + … βnxn + 𝜀i. (1)


47

The dependent variable, (ln)yi, is estimated as a function of the y-intercept or constant

term, β0, the beta coefficients (βn) and values for each independent variable (xn), and an

error term accounting for unobservable noise, 𝜀i.

In the context of this study, we are interested in how x (particularly cancer diagnosis)

effects y across individuals in each group. We want to know how the patients in each

subgroup differ from patients in other subgroups as opposed to how each patient differs

from each other patient. This is permitted by the random effects (RE) model, which

accounts for unit variance, 𝜇i, outside of and in addition to the error term, 𝜀it. RE may be

written as:

(ln)yit = β̂0 + β1x1it + β2x2it + … βnxnit + 𝜀it + 𝜇i, (2)

stratifying the observations for (ln)y and xn by time series, t, in addition to cross-

sectional, i, elements (Hill et al, 2011). When estimating the effect of x on y in the fixed

effects (FE) model, variance over time is examined within individual units rather than

among or between them (Hill et al, 2011). The coefficient estimates are based on

changes in y and x over time within each person, thus the FE estimator cannot generate

estimates for variables that are time-invariant within individuals (Hill et al, 2011). This

can be considered an advantage in some contexts, but would be detrimental in the

context of wanting to observe the determinacy of variables that are heterogeneous

across the sample, but invariant on the unit level (i.e. most demographic variables). This

is the case for these analyses, where the RE estimator is strongly preferred.

However, one does not simply use the RE estimator due to one of its fundamental

assumptions: that the individual random effect and explanatory variables are

independent. Unobserved individual characteristics are captured by the error

component, 𝜇i, if not explicitly accounted for in the regression. If these uncontrolled

characteristics are correlated with the variables in the regression, it violates the

assumption that cov(𝜇i, xnit) = 0 (Hill et al, 2011). FE does not assume this, meaning that

that estimator is consistent whether or not a correlation exists. If the correlation does not

exist, however, the RE estimator is consistent and more efficient (Hill et al, 2011). This

is evaluated using the Hausman test (Section 3.5.1). This test is done by storing and

comparing the estimates from RE and FE models, so both must be run to evaluate the

robustness of RE. In the event that RE rejects the null hypothesis of the Hausman test


48

(that RE and FE estimates do not differ), both results will be presented alongside a

modification outlined by Allison (2009) (Section 3.5.2).

3.4.3 Disaggregated cost

Estimator

This section looks at selected service-specific costs for the same logic of dividing cost

perspectives: to assess use of more specific services and better illuminate what might

drive the coefficients in aggregated cost analysis, namely the Ministry of Health

aggregate. This demonstrates a more vivid relationship between independent variables

and more tangible sources of spending. The unpaid arm is clearly associated with

unpaid caregiving time, while the Ministry of Health expenses are spread across a

plethora of services. Hospitalization, publicly financed home appointments and publicly

provided medications will be dependent variables. The former two, in the literature

review, were regarded as costly enough to independently determine the cost-

effectiveness of home care (Ostgathe et al, 2008; Wong et al, 2013).

Cost variables with heavy right-tails and a spike of zero values were alluded to in

Section 3.1.1 (re: logarithms). This is very common for dependent variables in health

economics and will be the case with these disaggregated costs (Beeukes Buntin et al,

2004; Neelon et al, N/A; Manning et al, 2001). This is commonly dealt with using two-

part or frequency-severity models in both health and actuarial sciences (Frees, 2010).

This approach analyzes any usage,

𝑦!" =1 𝑖𝑓 𝑦!" > 00 𝑖𝑓 𝑦!" = 0 , (3)

and its probability using probit or logit links (Part 1) separately from variability in the

costs greater than zero (Part 2):

𝑦!" = 𝑚!" 𝑖𝑓 𝑦!" > 0 . 𝑖𝑓 𝑦!" = 0 . (4)

Part 2 uses generalized linear models (GLM) on raw-scale cost or OLS (in this case,

generalized least squares [GLS] for panel data) on log-scale cost (Beeukes Buntin et al,


49

2004). The probability of any cost, 𝑃𝑟𝑜𝑏(𝑦 > 0), and what the patient is expected to

incur, 𝐸(𝑦|𝑦 > 0), can theoretically give the overall expectation of y accommodating the

zeros:

𝐸 𝑦 𝑥 = 𝑃𝑟𝑜𝑏 𝑦 > 0 ∗ 𝐸 𝑦 𝑦 > 0 . (5)

There are compelling reasons for selecting this method, which gives each part

independent and useful meaning. Cragg (1971) first used a two-part model with differing

independent variables in each part, and the usefulness of this aspect was extrapolated

onto health care data by Mullahy (1998). Using tobit models that have a single latent

variable and the same predictors has been problematized because the decision to

consume a service may be generated differently than the magnitude of consumption.

This argument is based on the principal-agent model, which conceptualizes Part 1 as

representative of patient (principal) choice to seek health services, and Part 2 as

representative of the frequency of visits that is determined by the health professional

(agent) (Mora et al, 2013).

Longitudinal two-part models appear infrequently in the literature but simply augment

traditional two-part form with corresponding panel models. Olsen & Schafer (2001) and

Liu et al (2011) employed RE logit links in the first part and GLMs in the second part.

Mora et al (2013) use a RE probit for the first part followed by a GLM panel regression

on raw scale cost > 0, and determine the appropriate link and distribution functions (i.e.

gamma, Poisson or Guassian) using the Park (1966) test (2013). The use of GLMs is

common because they analyze the raw-scale cost, and there are notable concerns

about retransforming to original scale in log-linear GLS models in the presence of

heteroskedasticity. Special retransformation techniques have been offered, namely

Duan’s (1983) smearing estimator (Manning et al, 1998; Beeukes Buntin et al, 2004). In

this particular instance, obtaining the raw-scale expected value of y or coefficients is not

of interest so much as the most precise coefficients to draw conclusions about the

predictor variables and comparative determinacy of ICD-10 categories. Given this

objective, RE regression is a highly efficient and desirable estimator (Hill et al, 2011).

Available panel probit options for the first part include RE and population-averaged (PA).

These predict different population parameters for which RE is preferable (Hu et al,

1998). For example, the PA probit estimates the probability of an average married


50

person experiencing the outcome compared to the average unmarried person, whereas

RE estimates give the probability of the outcome for a married person compared to the

same person experiencing the outcome if not married, taking into account their other

characteristics (Neuhaus, 1992; Neuhaus et al, 1991). The RE probit model, first

specified by Heckman and Willis (1976), can be written as

𝑦!"∗ = 𝑥!"𝛽 + 𝜇! + 𝜀!" (Guilkey et al, 1993). (6)

The observed dependent variable in (3), yit, takes the value of one if the service has

been used and zero if not. 𝑦!"∗ in equation (6) denotes an unobserved latent variable

premised on yit that is linearly related to k explanatory regressors, xit. β represents k

deterministic coefficients in a k x 1 vector (Bertschek et al, 1998). xit constitutes a 1 x k

vector of explanatory variables. Similar to the RE regression explained in Section 3.3.1,

𝜇i and 𝜀!" are independent error terms controlling for effects not accounted for by

regressors; 𝜇i within individuals and 𝜀!" across individuals (Guilkey et al, 1993).

FE and RE panel regressions will be run on log-transformed costs in the second part

and a Hausman test will be conducted to test whether FE or RE is the preferred

specification. In the presence of heteroskedastic residuals or rejection of the null

hypothesis of the Hausman test, the modification proposed by Allison (2009) will be

implemented (Section 3.5.2). Cost components for analyses include those associated

specifically with hospitalization (public), medication (public), and home appointments

(public). These particular cost components were chosen due to relatively high costs

amidst the zeros (i.e. one hospitalization = $5,558), therefore representing a potentially

more strenuous expense when use does occur. Hospitalizations and public home

appointments were of particular interest in the literature review. These also demonstrate

more specific sources of cost in the Ministry of Health arm of the aggregated cost of

home care, which is dispersed over many services. There were other costs

characterized by attribute of high cost among the zeros (out-of-pocket travel, out-of-

pocket home appointments, public supplies), but they were unable to yield statistically

significant results.


51

3.5 Caregiver burden analysis

Estimator

Caregiver burden score is a count variable on a scale from 1-64 representing how many

aspects of caregiver burden the caregiver has reported for a given time period and how

intensely they rank them on the Likert scale in Table 3. Hausman, Hall and Griliches

established FE and RE Poisson models in 1984, extending control for unobserved

heterogeneity in panel data to count variables (Wooldridge, 2012). Linear models, when

the dependent variable only takes nonnegative integer values, is not ideal because they

tend to follow a Poisson distribution and cannot be normalized by logarithmic

transformation (Wooldridge, 2012). Therefore, a RE Poisson model will be applied to this

dependent variable. This model can handle heteroskedasticity, discrete outcomes, the

panel aspect of the data, and generates appropriate non-negative predictions

(Wooldridge, 2012). The Hausman test can be similarly applied to FE and RE estimates

to ascertain the reliability of RE, which is preferable for the inclusion of time-invariant

observations within individuals.

The model is structured as

𝑦!"|𝒙! , 𝑐!~𝑃𝑜𝑖𝑠𝑠𝑜𝑛[𝑐!𝑚 𝒙!" ,𝜷! ] (Wooldridge, 2012), (7)

where yit is the dependent variable, 𝒙! is the observed explanatory variables (1 x k

vector), 𝑐! is the unobserved and time constant effect and 𝑚 𝒙!" ,𝜷! is the model for the

conditional mean (𝜷! is a k x 1 vector of parameters). An alternative but similar model

would be the RE negative binomial model. The negative binomial model can relax the

distributional assumption of Poisson (mean = variance) in the case that the dependent

variable is over-dispersed (Wooldridge, 2012). Both models were explored but results

did not differ.


52

3.6 Diagnostics

3.6.1 Hausman test

To ensure the RE model is consistent, the Hausman test will be applied to the estimates

from FE and RE versions of each regression. This tests the null hypothesis, H0, that 𝜇i

are independent from the independent variables, with rejection of H0 indicating the

alternative hypothesis, Ha, that 𝜇i are correlated with independent variables. It does this

by comparing the parameters from the FE and RE models.

The catch is that, irrespective of the Hausman test results, the FE estimator is

completely undesirable for the purpose of this thesis. Hausman is a notoriously sensitive

test that will reject the null hypothesis for any reason that causes the coefficients from

RE and FE models to differ (Wooldridge, 2012). Considering that the preliminary FE

estimatation will drop many of the independent variables, it is expected that the

Hausman test will reject in some instances, even if it is attributable to an artificial

exaggeration of coefficient differences induced by different groups of independent

variables (and consequently, degrees of freedom) in each model. This dilemma has

been acknowledged and analyzed extensively by Allison (2009), who has proposed

redeeming modifications to RE model specification.

3.6.2 Allison’s hybrid method

Allison’s hybrid method is a slight modification to RE specification. It is simply the

implementation of RE with the mean and deviation from the mean of time-varying

covariates entered as separate independent variables.

The RE and FE models differ with respect to their use of between- and within-person

variability. Premising coefficients on within-individual variability is what prevents the FE

model from generating estimates for time-invariant covariates with no within-individual

variability; their deviations from the mean will be zero. RE estimates on time-varying

variables are a weighted average of “within” and “between” coefficients. Recall equation

(2) for RE, and observe the structure of FE:

(ln)yit = β0i + β1x1it + β2x2it + … βnxnit + 𝜀it. (8)


53

all intercepts, β0i, are different for individuals and the coefficients, βn, are constant (Hill et

al, 2011). Individual intercepts are included to “control” for individual-specific covariates

that do not change over time, thus there is no term for unit variance, negating the

assumption that cov(𝜇i, xnit) = 0. This makes FE more consistent than RE, with RE’s

consistency being compromised by violation of that assumption. If the correlation does

not exist, however, the RE estimator is consistent and more efficient (Hill et al, 2011).

Failure of the Hausman test can be an instance when FE would be preferable to RE, or

alternatively, if the researcher wants to drop all time-invariant covariates (both observed

and unobserved). However, in the case that time-invariant covariates are important, the

Hausman test rejected, and pooled cross-sections are highly undesirable, Allison’s

hybrid method can be used to generate FE-type estimates (that are robust to cov(𝜇i, xnit)

≠ 0). It also provides ways to infer whether or not the estimates are being really affected

or altered by presenting the results alongside RE and FE.

Allison’s (2009) hybrid method involves separating within- from between-person

variation in the RE model by subtracting individual-specific means from the time-varying

covariates to obtain their deviations from the mean, as well as including the means as

explanatory variables. Because putting the data in deviation form is part of FE

estimation, this gives the best of both worlds: obtaining FE-type estimates for time-

varying covariates, which are cleansed of all time-invariant variables (observed and

unobserved), while still obtaining estimates on the time-invariant ones (Allison, 2009).

The variation across individuals is being controlled for by the individual-specific means

(Allison, 2009). The deviation coefficients can be interpreted as if they were FE

coefficients.

By including the time-invariant, individual-specific means and time-varying deviations

from the mean in the RE specification, you can make inferences from the coefficients.

The coefficients on the FE-type RE estimates in the hybrid model and the actual FE

estimates should be identical in balanced panels. In unbalanced panels, they won’t be

identical, but they should be similar (Allison, 2009). Similarity between these FE-type RE

estimates and actual FE estimates indicates that RE is behaving despite having failed

the Hausman test.

Whether or not RE is appropriate can also be evaluated by looking at the congruency

between the deviation and mean variables in the hybrid regression. The coefficients on


54

the mean and deviation should be similar if the assumptions of the RE model are

correct, namely that 𝜇i, is uncorrelated with xnit (Allison, 2009). When the coefficients on

the mean and deviation of the time-varying covariates are substantially different, it offers

insight into the appropriateness of a standard RE model.

In conclusion, Allison’s hybrid model offers reliable estimates on all of the independent

variables that allows one to circumvent some of the challenges associated with the RE

and FE models. The challenge with the RE model is meeting its assumption that 𝜇i is

uncorrelated with xnit, the violation of which compromises the consistently of the results.

When this assumption is violated, one would typically resort to FE. The challenge with

FE is that it only considers within-individual variability, and consequently drops all time-

invariant variables. It does not consider variance across-individuals which is undesirable

as well. Therefore, the Allison hybrid approach seems to represent an appropriate

alternative approach to estimation. The algorithm that decides whether or not Allison’s

hybrid method is required is shown in Figure 6.

Figure 6: Algorithm for deciding which estimator(s) to use

If the RE model passes the Hausman test thus satisfying important assumptions and

being the most efficient estimator, no modifications will be made to model specification

and only RE results will be reported. If the Hausman test rejects the null hypothesis, RE

estimates will be presented alongside Allison’s hybrid and FE estimates.

3.6.3 Standard diagnostics

In addition to logarithmic transformations intended to correct heteroskedasticity, the

panel/cluster-robust standard error estimator will be applied to the final models.

Run Hausman test

Accept Reject

Report robust RE regression estimates

Report robust RE, FE and hybrid estimates

Run RE and FE regressions


55

The robust models will undergo the Breusch and Pagan Lagrangian multiplier test for

random effects. While the Hausman test assesses the reliability of RE, this tests the

appropriateness of RE. The H0 of this test is that there is no difference across units

(patients) or no “panel effect” (Torres-Reyna, 2007). Rejecting this hypothesis validates

the appropriateness of a panel regression for these data, whereas accepting the null

suggests that a cross-sectional ordinary least squares (OLS) or pooled regression would

suffice and yield similar results (Torres-Reyna, 2007).

3.7 Summary

This chapter has discussed the independent and dependent variables, conceptual

frameworks, and analytic estimators that will be applied to the data. The independent

variables will generally adhere to socio-demographic, clinical, and service/community-

related predictors. This has been informed by the literature, the Andersen and Newman

framework, and the caregiver burden framework used by Given et al (2004).

The following chapter will reveal how these predictors determine cost from three payer

perspectives: societal, Ministry of Health, and informal caregiver. In addition, service

specific costs will be analyzed using two-part models to see if further variation emerges

and potentially illuminate relationships between different independent variables and

different components of the costs associated with CBPHC. Hospitalization, public

medication, and public home appointments will be looked at. Finally, their influence over

caregiver burden will be analyzed.

56

Chapter 4 Results

The results section will discuss the socio-demographic characteristics of the sample

followed by the results of the analyses and diagnostics described in Chapter 3. Chapter

3 also explained the conceptual frameworks used to identify important independent

variables, with one of the most consistent determinants in the generation of costs and

burden being ‘need’ factors related to symptom severity and clinical characteristics

(Andersen & Newman, 1973; Givens, 2004). This brings us back to the focus of the

thesis which concerns the role of the primary cancer diagnoses as outlined by the ICD-

10 diagnostic codes, which distinguishes among tumour sites (i.e. breast, brain) as

opposed to tumour types (i.e. carcinoma, sarcoma), in determining cost and burden.

This chapter presents findings from analyses of a) aggregate costs of CBPHC from three

stakeholder perspectives (society, the Ministry of Health, and unpaid caregivers)

(Section 4.2.1), b) service-specific public costs associated with CBPHC (medication,

hospitalization, home appointments) (Section 4.2.2), and c) caregiver burden scores

generated using the CBS-EOLC (Dumont et al, 2008) (Section 4.2). Intrathoracic

(primarily lung) cancers served as the baseline category for the ICD-10 variable.

4.1 Descriptive statistics

As mentioned in the data description section (Section 3.1.2), there was an attrition rate

of 4.6% (N=15) related to dropout or outliving the study period. The time periods are

based on the date of death, which prohibited the calculation of time periods and

integration into the analysis for those who dropped out or outlived the study. Table 4

reports descriptive data on 312 of 327 patients and their primary caregivers who

remained in the study through their palliative trajectory.

The socio-demographic composition of the cohort is consistent with and evocative of

some of the trends outlined by Seow (2009) and Jiwani (2003). Spouses (46.8%) and

children (40.6%) of the patient constituted an overwhelming proportion of the unpaid

caregivers. These sources of unpaid labour may decline in coming years due to an

abating propensity to rear children and fewer opportunities for spouses to retire.

Chapter 4: Results R. Redmond-Misner

57

Prospects for retirement among Ontario’s population are stagnating as incomes have not

grown with the cost of living (Yalnizyan, 2007). With that said, being retired was the most

common employment status among caregivers (35.8%), followed by full-time work

(30.2%). Most dyads were from Toronto (66%). The majority (52%) of patients had an

educational level of high school or less, while most caregivers had a post-secondary

education. The sex distribution of the patients was virtually equal, but the majority of

caregivers were women (64.5%). Caregivers were generally younger than the patients;

the average age was 58 for caregivers (range: 20-94) and 71 for patients (range: 27-96).

Diverse living arrangements are represented in these data. Living with a spouse was by

far the most common (59%), and the proportion living with children (16.3%) indicates

that most patients who received unpaid care from their children did not actually live with

them. Only 16.7% of patients lived alone, which is an underrepresented group in home

care and therefore in existing research (Aoun et al, 2013).

Table 5: Summary of patient and caregiver demographics

Demographic Characteristics

Patients (N=312) Caregivers (N=318) N n (%) N n (%)

Sex Male Female

312 148 (47.4%) 164 (52.6%)

312 104 (33.3%) 208 (66.7%)

Education Any college High school or less Any university Post grad Missing

312 54 (17.3%)

164 (52.6%) 63 (20.2%) 31 (9.9%)

-

312 47 (15.1%) 91 (29.2%) 90 (28.8%) 52 (16.7%) 32 (10.3%)

Site Toronto Hamilton

312 206 (66%) 106 (34%)

312 206 (66%) 106 (34%)

Marital status Divorced/separated Never married Widow Married Missing

312 33 (10.6%) 10 (3.2%)

82 (26.3%) 187 (59.9%)

-

312 28 (9%)

26 (8.3%) 6 (1.9%)

243 (77.9%) 9 (2.9%)

Relation to patient Child Spouse Friend Sibling Other family Other

–

– – – – – –

312 131 (42%)

142 (45.5%) 8 (2.6%)

11 (3.5%) 2 (0.6%)

18 (5.8%)


58

Patient living arrangement Lives w spouse Lives w children Lives w other Lives alone

312 184 (59%) 51 (16.3%)

25 (8%) 52 (16.7%)

– – – – –

Caregiver employment Disability Full-time Missing Not employed On leave Part-time Retired

– – – – – – – –

312 8 (2.5%)

95 (30.4%) 21 (6.7%)

37 (11.6%) 19 (6.1%) 21 (6.7%)

111 (35.6%)

While only one caregiver per patient was included in the analysis, some patients did

have more than one caregiver. The few secondary caregivers were removed for several

reasons, namely a) for consistency in reporting and b) to ensure the caregiver burden

reports associated with each patient come from the same caregiver. In panel analysis,

the dataset needs to be stratified by units (here, patient-caregiver dyads). If the panel

was set using patient IDs, keeping all the caregivers in the data, there would be no

control for caregivers (i.e. two caregivers for the same patient would be counted as if

they were one person). By treating them as dyads with the same ID, the panel controls

for both patients and caregivers. In only considering the reports of the primary caregiver,

it is notable that Jacobs et al’s (2011) found that multiple caregivers to the same patient

can offset one another’s costs, thereby lowering the report of the primary caregivers.

The costs reported may therefore be conservative estimates.

4.2 Cost analyses

Overall, it is evident that primary cancer diagnosis had an influential role in the costs

associated with CBPHC. This section will first go through the determinants of aggregate

costs, adopting the perspectives of society, the Ministry of Health and informal

caregivers. It will then discuss service-specific costs including medication costs (public),

hospitalizations, and home appointments (public). Each dependent variable has its own

subsection, within which results from the regressions will be reported. Diagnostics and

additional information pertaining to each of the set of estimates are at the end of the

chapter. While the analyses produce a plethora of statistics, only some of them will be


59

thoroughly discussed as significant or relevant to the research question and purpose.

Appendix H contains an explanation of the lesser discussed statistics reported.

A note on Hausman test results

The algorithm for selecting estimator(s) in the Methodology chapter showed that the

adopted approach is contingent on the result of the Hausman test. The Hausman test

determines whether Random Effects (RE) specification alone is sufficient, therefore

additional estimates derived from alternative specifications (Allison’s hybrid and fixed

effects(FE)) are provided when the Hausman test indicates that RE may be biased

(‘Reject’). While this is a diagnostic test, most of which will be presented at the end of

the chapter, Hausman test results precede the finalized estimations presented here to

illustrate why each approach was taken. By dependent variable, Table 6 gives the model

specifications that can be expected for each dependent variable. This is to show the

general direction for the chapter, but a fuller discussion is provided in the Diagnostics

section.

Table 6: Hausman test results, final estimator(s)

Dependent variable Accept/reject Hausman Estimation Societal perspective cost Reject (Prob > Chi2 = .0004) RE, FE, hybrid Ministy perspective cost Reject (Prob > Chi2 = .0011) RE, FE, hybrid Unpaid caregiver costs Reject (Prob > Chi2 = .0154) RE, FE, hybrid Public medication costs Accept (Prob > Chi2 = .8848) RE (two-part) Hospitalizations N/A (No FE probit) RE Probit only Public home appointment costs Accept (Prob > Chi2 = .0053) RE (two-part) Caregiver burden score Accept (Prob > Chi2 = .5030) RE Poisson

4.2.1 Aggregated costs

Societal perspective

It is apparent from the regression models reported in Table 7 (RE, FE, Allison’s hybrid)

that the brain cancer group was associated with significantly higher societal costs.

Statistically significant findings also emerged from the other covariates that were

indicated by the Andersen and Newman framework. ‘Enabling’ factors of city of

residence and days spent over night were found to be significant; lower societal costs

were found for residents of Hamilton and costs were higher when caregivers spent more


60

days overnight. Living with children was associated with higher costs, as opposed to

living with spouses (baseline), with others or alone. Caregiver burden was also strongly

significant, driving costs higher as caregiver burden increased. These results were

invariant to model specification and are therefore quite robust. The percentage of

variability accounted for by the model is given by R2 values, which are as high as 20% in

the hybrid specification.

Table 7: Determinants of societal costs of CBPHC (RE, FE & Allison’s hybrid)

Dependent Variable: (ln)Societal Cost Independent Variable RE Coef. Hybrid Coef. FE Coef. Site

Hamilton

-.1130987*

-.1816811*

Patient age Patient age2

-.0309784 .0002126

-.0329119 .0002262

Comorbidity score .0209617 .0315624 Caregiver burden

Mean Deviation

.0307182*** .0152143** .0368713***

.0368737***

Caregiver education (Baseline: < High school)

College Any university Post grad Missing

-.0650912 -.1423874 -.0727493 -.0556227

-.0787003 -.1244516 -.0474793 .0251493

Caregiver employment (Baseline: Retired)

Disability Full time Missing Not employed On leave Part time

.1227241 -.1367954 -.1463374 .0818035 .1760111 -.0689561

.0818017 -.1375489 -.0165935 .1201743 .1985637 -.0685794

Cancer type (Baseline: Intrathoracic)

Digestive organs Female organs Brain Breast Male organs Urinary tract Other

.1038947

.1606436 .4971476***

.081204 -.0930862 -.0655134 .1213257

.157179 .2164067

.5292978*** .0679875 -.0725099 -.0020805 .1528889

Patient education (Baseline: < High school)

College University Post grad

.1257612 .109098

.1064619

.1294936

.0636265

.0480916


61

Table 7 exhibits that there are few time-varying covariates in the data generation

process. The implication of few time-varying covariates is simply that, in instances where

Allison’s hybrid and FE coefficients are presented, there is little to compare between

them. The two time-varying covariates, days spent overnight and caregiver burden

scores, are where FE and Allison’s FE-type parameters (deviation) can be compared.

These results exhibit strong similarity between FE and Allison’s FE-type parameters,

which is what is being looked for in augmenting the RE estimator. Similarity

demonstrates that RE estimation is not producing drastically different results than the FE

estimation. Granted similarity between Allison’s hybrid RE and FE estimates, both the

RE and hybrid results are interpretable. Discrepancies between them can speak to the

sensitivity of the results (i.e. if a variable is significant in one specification and not in the

other).

Ministry of Health perspective

In Table 8, brain cancer emerges as a predictive characteristic for Ministry of Health

costs, as do female cancers (primarily ovarian in this sample). This suggests that brain

1 The meaning of asterisks beside coefficients will remain consistent throughout.

Caregiver sex Male

-.1136043

-.0652709

Patient living (Baseline: With spouse)

Alone With children With others

.1462281 .2662788* -.2818415

.1238164 .2649985* -.3058154

Days overnight Mean Deviation

.0776996*** .0458273** .0798558***

.0800069***

Intercept 8.794154 9.162876 7.669735 Overall R2 Between R2 Within R2 Wald Chi2 Prob > Chi2 Σe Σu Rho

.1854 (18.5%)

.1945 (19.5%)

.1258 (12.6%) 295.23

.000 .47238208 .47546074 .50324804

.1963 (19.6%)

.2067 (20.7%)

.1282 (12.8%) 264.2 .000

.47191569

.47895282

.50740033

.0807 (8%)

.0521 (5%) .1285 (12.8%)

.47238208

.62232207

.63444625 *p=<.05 **p=<.005 ***p=<.0001


62

cancer’s significance in total societal costs (β=0.497, p=<.000) may have been related to

Ministry of Health costs. The finding around female cancers is also interesting because it

demonstrates that the break-down of costs can illuminate relationships that become

diluted when the costs are aggregated. While female cancers do not seem to have

significantly higher total societal cost, they apparently incurred higher public cost.

As a component of societal costs, Ministry of Health costs derive some similar

associations with the rest of the independent variables. The results show caregiver

burden, days overnight and living with children are also positively associated with

Ministry of Health costs. While the findings in the previous model held irrespective of the

estimation approach used, one finding in Table 6 is sensitive to model specification.

Specifically, female cancers are positively associated with Ministry of Health costs, but it

is only significant in the hybrid model. This relationship is not as robust.

Table 8: Determinants of CBPHC Ministry of Health costs (RE, FE & Allison’s hybrid)

Dependent Variable: (ln)Ministry of Health Cost Independent variable RE Coef. Hybrid Coef. FE Coef. Site

Hamilton

-.0196268

-.0573724


-.0154474 .0001132

-.0219836 .0001607

Comorbidity score .0606988 .0712779 Caregiver burden

Mean Deviation

.0349555*** .0045776

.0363205***

.0364323***



-.0638455 -.0290597 .1080564 .0092523

.0654584

.0355507

.1875039 -.0720691


Disability Full time Missing Not employed

.0707613 -.0312437 -.1778579 .1898731

.0509877 -.0205308 -.1606884 .2265385


63

On leave Part time

.0166669 -.0059202

.0461801

.0378089 Cancer type (Baseline: Intrathoracic)


.1543349

.2978083 .5552936*

.12887 -.0135634 .115685

.2179945

.226099 .3903798* .6255724** .1235337 .0004234 .184768

.2503838



.2121052

.0717804 .143288

.2279943

.0486106 .122122

Caregiver sex Male

.0017382

.0139336



.4286443* .3111334** -.0460779

.4146304* .3155617** -.0799712


.1341513*** .1309083*** .1338707***

.1338186***


.1031 (10%) .147 (14.7%) .0909 (9.1%)

336.47 .000

.69313764

.59739958

.42622137

.1311 (13%)

.1709 (17%) .0914 (9%)

185.51 .000

.69250399

.59272823 .4228326

.0402 (4%)

.0402 (4%) .0918 (9.2%)

.69313764 .7939662 .5674917

There are several differences in the determinants of Ministry of Health costs compared

to those that drove societal costs. Living alone emerged as a determinant of increased

Ministry of Health costs despite not having produced any significant determinacy in total

societal costs. While the Hamilton site was associated with lower societal costs, it was

not associated with lower Ministry of Health costs compared to the Toronto site. This

suggests that a.) while patients living alone did not have higher costs overall, a lack of


64

unpaid caregiving might mask the fact that they actually use public services more than

patients living with someone else, b.) city of residence/home care agency may be more

closely linked with unpaid caregiving costs than the costs borne by the Ministry of

Health, and c.) that different variables do affect different stakeholders differently.

Unpaid caregiver cost perspective

The results of the unpaid caregiver cost analyses are presented in Table 9. They fit

logically with the determinants of aggregated societal (Table 7) and Ministry of Health

(Table 8) costs and further illustrate the rationale for an examination of these differential

cost perspectives. The Hamilton site, which was associated with lower societal cost but

had no effect on Ministry of Health costs, was associated with lower caregiver costs than

their Torontonian counterparts. Living alone, which drove Ministry of Health costs

upwards, was negatively, but non-significantly associated with unpaid caregiver costs.

Patients living with others, who were associated with lower societal costs but had no

effect on Ministry of Health costs, were significantly associated with lower unpaid

caregiver costs. Significantly lower unpaid cost among patients living with others

suggests that less unpaid care could explain their markedly lower societal costs.

Caregivers on leave from work were associated with higher unpaid caregiver costs.

There are also consistencies across societal, Ministry of Health and unpaid caregiver

costs. In contrast to the findings discussed in the previous paragraph where Ministry of

Health and unpaid caregiver care costs were oppositely driven by certain characteristics

(i.e. living alone has a higher Ministry of Health and lower unpaid caregiver cost), some

characteristics drove expenses upward irrespective of payer. These include higher

caregiver burden, more days overnight, living with children as opposed to spouses,

others or alone, and having brain cancer as opposed to the other ICD-10 denominations

found in the sample. Driving cost unanimously upward suggests that despite being more

intensive users of public services, patients with brain cancer (primarily glioblastoma

multiforme in this sample) were also receiving more unpaid care than most people in the

program. The same is true for patients whose caregiver was experiencing high burden,

patients whose caregiver spent many days overnight, or those who lived with children.


65

Table 9: Determinants of CBPHC unpaid caregiver costs (RE, FE & Allison’s hybrid)

Dependent Variable: (ln)Unpaid Caregiver Cost Independent variable RE Coef. Hybrid Coef. FE Coef. Site

Hamilton

-.2318017*

-.2794294*


-.0457998 .0003089

-.0438953* .0002923*

Comorbidity score -.0033099 .0084026 Caregiver burden

Mean Deviation

.0316818*** .0158273** .0374144***

.0375295***



-.1705292 -.2368761 -.2312876 .0007709

-.2089305 -.2503827 -2148301 .139797



.2261023 -.1784909 -.1874397 .0021671 .3016255* -.122424

.2074949 -.185674

-.0737118 .0445025 .3284185* -.1491987



.0790303 .004515

.5322952** .0793724 -.1582612 -.2442645 .1138028

.1323618

.0503773 .545301** .0644224 -.1278603 -.1653247 .1638022



.0226484

.1078051

.0359023

.0169526

.0324579 -.0786854

Caregiver sex Male

-.1471362

-.1651238

Patient living


66

(Baseline: With spouse) Alone With children With others

-.1797427 .2795824* -.5030078*

-.2130987 .2687588

-.5258394**


.0313074*** -.0536556* .0379052***

.0358283***


.1967 (19.7%)

.1938 (19.4%) .0857 (8.6%)

1714.75 .000

.60266598

.73980838 .5298672

.1889 (18.9%) .1916 (19%) .0868 (8.7%)

193.32 .000

.60247802

.64579173

.53465727

.0688 (6.9%)

.0278 (2.8%)

.0884 (8.8%)

.60266598

.82545998

.65229791

4.2.2 Service-specific cost analysis results

Public medication costs

In order to accommodate the distribution of costs where many patients did not incur any

costs, public medication costs were analyzed using a RE probit model for the probability

of any cost (0 if no costs were incurred, 1 otherwise) and RE generalized least squares

(GLS) regression to assess costs greater than zero. Publicly financed medications were

received by 242 (77.5%) patients at some point in their palliative trajectory. The bi-

weekly cost ranged from $0 to $5,200 and had an average of $119. The results are

sparse in terms of statistically significant findings, and are given in Table 10.

Cancers of the digestive and gynecologic organs reduced the probability of receiving any

medication, while patients with the highest level of education were associated with

significantly higher medication costs when costs were incurred. Cancer diagnosis has

been a consistent determinant in the analyses, though there is now variability in which

category is significant and in what direction. The most consistent diagnostic driver so far,

brain cancer, is not significant here. This lack of effect suggests that medications are not

behind brain cancer patient’s elevated public cost.


67

Table 10: Determinants of receiving any medications and their cost (RE probit and GLS)

Dependent Variable

Any cost incurred (binary)

(ln)Public medication cost > 0

Independent variable RE Probit RE GLS Site

Hamilton

-.1470933

-.0689471 Patient age Patient age2

.0764394 -.0004008

.0126442 -.0001443

Comorbidity score -.0434624 .0146193 Caregiver burden .0045518 .0105105 Caregiver education (Baseline: < High school)


-.0119788 -.1857371 .2453136 .4181955

.2915087

.0689206

.2084064

.2224894 Caregiver employment (Baseline: Retired)


.8117905

.2977489 -.3998645 .3976619 -.1481287 .5322104

.2716082

.2127579 -.3591463 .3759479 .2036561 .0445465



-.433949* -.7921458* .5306406 -.4280103 -.0943561 -.4676685 -.272698

.1942796 -.038057 .2672141 -.2487115 -.0427441 -.0694487 .0079689



.147035 -.1476198 -.4347936

.2825997

.1088122 .0942958*

Caregiver sex Male

.0177658

-.0234747



68

Hospitalization cost

Hospitalization costs took on one of two values: zero and the cost of a single

hospitalization ($5,558). This is because no patients experienced more than one

hospitalization in a given two-week time period. As a result, only a RE probit is required

given that there is no variability in the cost that would be analyzed using a RE GLS

model. Over the palliative care trajectory, 109 (34%) patients were hospitalized, making

this a less common but very costly aspect of CBPHC.

The RE probit results for the probability of hospitalization are given in Table 11. RE

probit is not subject to the Hausman test because a FE version of probit is not available

(StataCorp, 2013). Non-linear FE estimation is associated with practical and

methodological shortcomings, namely the “incidental parameters problem” that is

believed to irredeemably bias non-linear FE estimates (Greene, 2002). Hospitalizations

were significantly higher for patients with cancers of the digestive organs, caregivers

experiencing higher burden, and those who lived alone. In the aggregated costs, living

alone was only a significant predictor of Ministry of Health spending. The association

between living alone and Ministry of Health cost may be attributable in part to

heightened use of hospitals.


-.1261921 .1390012 .1292038

.0810581

.0339595 -.1645129

Intercept -2.95923 Overall R2 Between R2 Within R2 Wald Chi2 Prob > Chi2 Σe Σu Rho (ln)Σu2

46.91 .0254

.8745786 .4333917 -.2680262

.0497 (5%) .1062 (10.6%)

.0013 (1%) 40.46 .0767

1.0729366 .74413165 .32478376


69

Table 11: Probability of hospitalization among CBPHC recipients

Dependent variable: Hospitalization (binary) Independent variable Coef. Site

Hamilton

-.339328 .0905324

Caregiver burden .045269*** Caregiver education (Baseline: < High school)


-.0916529 -.1254921 -.1897629 .1348325


Disability Full time Missing Not employed On leave Part time Student

.034076 -.3741487 -.3188431 .0390696 -.4577477 .0210105 .1900453



.4441604* -.0052286 -.0976603 -.4734185 -.0269657 .323222 .437184



.25128 .1078051 .0359023

Caregiver sex Male

-.0076969


Alone With children

1.172126*** .4668778


70

With others .0013056 Intercept -.5239812 Wald Chi2 Prob > Chi2 Σu Rho (ln)Σu2

77.42 .000

.7413276

.3546583

.7413276

The rationale behind the service-specific cost analyses was to illustrate more specific

drivers of the aggregate Ministry of Health costs. The aggregated cost to the Ministry of

Health is more ambiguous because it is spread over many services. While variability in

hospitalization costs could not be analyzed (due to lack of variability), the probability of

incurring such costs was assessed. Several covariates that drove Ministry of Health

costs were associated with higher probability of hospitalization in this analysis. However,

it is noticeable that brain cancer is not significant for this service. Similar to medication

costs, this suggests that hospitalizations do not drive the positive and significant

association brain cancer has with Ministry of Health costs.

Public home appointment cost

Home appointments were used by 310 (99%) of the patients at some point in their

palliative trajectory. Bi-weekly costs ranged from $0 to $17,000, with an average of

$955. One of the benefits of two-part analysis was the use of different independent

variables in the data generation process of the two dependent variables (Mullahy, 1998;

Cragg, 1971; Mora et al, 2013). This was performed by using a different diagnosis

variable in the first and second part. Separate variables were used because all urinary

tract cancer patients had home appointments in all time periods, skewing their probit

coefficients drastically. A seven-category ICD-10 variable was used in the probit

specification where urinary tract cancer patients were categorized as “other” to

circumvent this issue. Lung cancer was still used as the baseline comparator.

The results of the RE probit on the incidence of home appointments and on RE

generalized least squares (GLS) regression on costs are presented in Table 12, showing

several significant findings. Both caregiver burden and female cancers negatively

influenced the probability of a home appointment, yet they positively determined costs.

Paradoxical findings such as this are the justification of separate analyses. Cragg (1971)


71

and Mullahy (1998) first used two-part analysis to use differing independent variables in

each part premised on the suspicion that these aspects of cost (any usage vs. extent of

usage) would be generated differently. Mora et al (2013) used the principal-agent model

to explain that each component may be determined by different people (i.e. any usage

by the patient/caregiver and the extent of service use by doctors) whose decisions are

influenced differently by the independent variables. The findings pertaining to caregiver

burden and gynecologic cancers suggest that whether or not a public home appointment

occurs, and the cost of the care provided at the home appointment, are determined

differently. While these patients may not have had home appointments as consistently

over time, they required extensive care or multiple appointments in the time periods

when they did. Patients who had post-graduate educations, lived in Hamilton and lived

alone were also associated with fewer visits.

The cost of services provided at home appointments are positively associated with both

female and brain cancers. Brain cancer has been a consistently positive driver of

aggregate costs, but was not significant when it came to medication or hospitalization

costs. Public home appointments however, contribute at least partially to their higher

costs. While gynecologic malignancies did not drive societal or unpaid costs, they were a

significant predictor of Ministry of Health costs. This result suggests that the positive

impact of female malignancies on Ministry of Health expenses could also be related to

public home appointments.

Table 12: Determinants of having a home appointment and its associated cost (RE probit and GLS)

Dependent variable

Any cost incurred (binary)

(ln)Home appointment cost > 0

Independent variable Probit GLS Site

Hamilton

-.838226*


.0154411

.0001287 .041956

-.0002556 Comorbidity score -.1885314 .0527401 Caregiver burden -.0283992* .0268329*** Caregiver education (Baseline: < High school)

College Any university

.5572194 -.0273343

-.0635957 -.1174621


72

Post grad Missing

.8871087 -.5419388

.0346318

.0193709 Caregiver employment (Baseline: Retired)


-.5469074 .1663302 .3165967 .7937325 .7487401 .4491194

.287002 .0720758 -.0857068 .0904596 .1862101 -.1497837



-.5728867 -1.09672* -.3543498 .094809

.7776071

1.071584

.254821 .5518785* .6220284** .3191997 -.110943 .526638

.1035797 Patient education (Baseline: < High school)


-.1404788 -.8395315 -1.314628*

-.0598004 -.0000761 .0000666

Caregiver sex Male

.9573216*

-.0520232



-1.672275** -.9444785 -.9393589

.171116 .1492152 .2009693

Intercept 3.830204 Overall R2 Between R2 Within R2 Wald Chi2 Prob > Chi2 Σe Σu Rho (ln)Σu2

25.59 .5695

1.084338 .5403967 .1619397

.0733 (7%) .0789 (7.9%) .0382 (3.8%)

68.51 .000

.73604573

.64634202

.43538168


73

4.3 Caregiver burden

The results of the caregiver burden score analysis are shown in Table 13. The highest

observation on this 64-point scale was 63, with a mean of 26. Three primary diagnoses

were significant and positive drivers of the caregiver burden score: digestive,

gynecologic and urinary tract cancers. There was consistency between the diagnostic

categories that drove caregiver burden and those that drive hospitalizations (digestive),

and public home appointments (gynecologic). It is notable that, while it did not

significantly drive home appointment costs, urinary tract cancers were combined with

‘other’ for the probit regression on home appointment incidence because the frequency

with which they used that service was 100%. Burden is also strongly positively related to

the time loss of the caregiver and higher among caregivers with post-graduate

educations. Interestingly, patient living arrangements were not significant despite

literature citing living with the patient as having negative implications for respite, and

therefore, caregiver burden (Glajchen, 2012). The most consistently explanatory

diagnostic category – brain cancer – is also insignificant.

Table 13: Determinants of caregiver burden in CBPHC (RE Poisson)

Dependent variable: Caregiver burden score Independent variable Coef. Site

Hamilton


-.0012702 .0000111

Comorbidity score .0087657 Time loss .00004*** Caregiver education (Baseline: < High school)


.0133564

.0809493 .1130036* .0780259


Disability Full time Missing Not employed

-.0412655 .0231466 .0598248 .0625991


74

On leave Part time Student

-.0085026 .0291817 -.1638782



.10859* .1407698* .0349128 .0274689 .0558959 .1654493* .064661



.0249797 -.0306019 -.0644647

Caregiver sex Male

.0108964



.0059566 -.0081153 -.0070816

Intercept 3.003243***

Wald Chi2 Prob > Chi2 (ln)α Α

126.17 .000

-2.867784 .0568247

4.4 Diagnostics/Additional information

This section will discuss the diagnostics outlined in Section 3.5 and give additional

information about the analyses where appropriate.

Societal cost: Additional analysis information

The Hausman test applied to RE and FE analyses of societal costs rejected the null


75

hypothesis (H0) that the difference in coefficients between estimators was not systematic

(Prob > Chi2 = 0.0004). The alternative hypothesis indicates potential for predictions to

be inconsistent and biased, however the Hausman test will reject the null for any reason

that causes difference in RE and FE estimates, including the different degrees of

freedom (DF) after FE drops most covariates (RE=29 DF, FE=6 DF). What this test picks

up is the difference in RE and FE estimates; if the assumptions underlying the use of RE

hold, the estimates should be similar (Hill et al, 2011). In the succeeding subsections,

the interpretation of the Hausman test will remain the same.

In order to assess the accuracy of the RE results, which can still be valid despite the

result of the Hausman test, the RE results are presented alongside Allison’s hybrid

model and a FE model in Table 7. The results indicate that RE is behaving consistently

with FE. Coefficients on days overnight and caregiver burden, the only time-varying

covariates in this regression, agree across the standard RE, FE, and Allison’s hybrid FE-

type parameters on deviations from the mean in an RE specification.

Standard diagnostics were performed including graphing residuals and fitted values

(Appendix I), robust standard errors and the Breusch & Pagan Lagrange multiplier. All

GLS coefficients in this chapter are derived from models with heteroskedasticity-robust

standard errors. The Breusch & Pagan Lagrange multiplier test for the RE model

rejected the null hypothesis that there is no random effect (Prob > chibar2 = 0.000) in the

generation of CBPHC’s societal cost. The presence of random effects means that the

RE model is preferable to and more precise than a pooled or OLS model, and

consequently, the optimal model for this particular analysis. In the succeeding

subsections, the diagnostic interpretation of these statistics will be the same.

Ministry of Health cost: Additional analysis information

The RE model has rejected the null hypothesis of the Hausman test (Prob > Chi2 =

0.0011) despite otherwise affirming diagnostics. The residuals of the RE model are

homoskedastic and the fitted values slightly over predict (Appendix I); the variability

explained by the data generation process is much less for this aggregated cost (R2 =

10%). The coefficients in Table 8 show that the hybrid’s deviation variables’ and FE

parameters agree, which suggests that the RE estimator is behaving. The Breusch and

Pagan test detected random effect (Prob < chibar2 = 0.000), validating the

appropriateness of the estimator.


76

Unpaid caregiver cost: Additional analysis information

The unpaid care analysis rejected the Hausman test (Prob > Chi2 = 0.0154) and as well

as the Breusch and Pagan test (Prob < chibar2 = 0.000). The Hausman test result

means that the RE model should be augmented with Allison’s hybrid approach and

presented alongside FE estimates. The Breusch and Pagan test results indicate that

there are random effects in the panel and that RE is an appropriate estimator. The

residuals produce slightly more heteroskedasticity than the past two models (Appendix

I), which the use of Huber-White standard errors control for. Consistency across the RE,

FE and hybrid estimates are shown in Table 9. These models had a considerably higher

R2 than that reported in the Ministry of Health analysis (~19%).

Publicly funded medication: Additional analysis information

For the second part (GLS on costs > 0), the RE model accepted the null hypothesis of

the Hausman test (0.8848). Therefore, only the robust RE model is reported in Table 10.

The Breusch and Pagan Lagrange multiplier confirmed RE as the appropriate estimator

(Prob < chibar2 = 0.000). The graphs in Figure 10 show the distribution of the dependent

variables in the RE probit and GLS models. The left-hand-side (LHS) graph

demonstrates the high proportion of zeros in the data that needed to be modeled

separately. The right-hand-side (RHS) graph shows that the logarithms of the positive

values form an appropriate distribution for regression analysis. The overall R2 statistic for

the RE GLS is low (5%), but it is higher for the variability between individuals (10%).

Figure 7: Raw and log-positive distribution of public medication cost

0.1

.2.3

.4De

nsity

0 2 4 6 8Log-Transformed Medication Cost

Log Positive Public Medication Cost

0.0

01.0

02.0

03.0

04.0

05De

nsity

0 1000 2000 3000 4000 5000Medication Cost

Public Cost of Medications


77

Public home appointment: Additional analysis information

The second part of the model did not reject the Hausman test (Prob > Chi2 = 0.0053)

meaning that only the RE estimator is used on the log positive observations. The

Breusch and Pagan test confirmed that it is again the most appropriate model. Figure 9

shows the distributions of the dependent variables, demonstrating that there were zeros

requiring separate modeling (LHS) and that the logarithm of the positive observations

formed an appropriate distribution for the estimator (RHS). The overall and between-

individual R2 statistics on the RE GLS are both ~7%.

Figure 8: Raw and log-positive distribution of public home appointment cost

Caregiver burden scores: Additional analysis information

RE Poisson is subject to the Hausman test, and the null hypothesis was

accepted (Prob > Chi2 = 0.5030). As shown in Figure 10 (LHS), the distribution of the

variable is skewed, but because it is not genuinely continuous, cannot be addressed

through logarithmic transformation as shown by the graph on the RHS. The data is count

in nature and conforms to a Poisson distribution, although they are slightly over-

dispersed (variance < mean). Running a negative binomial that relaxes distributional

assumptions made no difference in results.

0.1

.2.3

.4.5

Dens

ity

2 4 6 8 10Log-Transformed Home Appointment Cost

Log Home Appointment Cost >0

02.

0e-0

44.

0e-0

46.

0e-0

48.

0e-0

4De

nsity

0 5000 10000 15000 20000Public Home Appointment Cost

Raw Scale Home Appointment Cost


78

Figure 9: Raw and log scale burden scores

4.5 Summary

This chapter discussed the results of econometric analyses of aggregate costs, service-

specific (disaggregated) costs, and caregiver burden associated with CBPHC. A detailed

summary of significant findings is given by Table 14, including dependent variables,

estimators, significant independent variables, their level of significance and their

direction of association (+/–). It was found that, amidst other influential predictors,

primary cancer diagnosis was a significant driver of both cost and burden.

Table 14: Significant findings across all analyses

Dependent Variable

Estimator Significant covariates (+/–)

Significance level

Societal cost RE/Hybrid Hamilton (–)2 Burden (+)3 Brain (+) Living w/ children (+) Days overnight (+)3

p < .05 p < .000 p < .000 p < .05

p < .000 Ministry of Health cost

RE/Hybrid Burden (+)3 Brain (+) Living alone (+) Living w/ children (+) Days overnight (+)3 Female cancers (+)2

p < .000 p < .005 p < .05

p < .005 p < .000 p < .05

2 Sensitive to model specification. 3 Also significant in FE estimation.

0.0

5.1

Dens

ity

20 30 40 50 60Caregiver Burden Score

Caregiver Burden Score Distribution

0.5

11.

52

2.5

Dens

ity

2.5 3 3.5 4Log-Transformed Burden Score

Log Caregiver Burden Distribution


79

Unpaid caregiver cost

RE/Hybrid Hamilton (–) On leave (+) Brain (+) Living w/ others (–) Living w/ children (+)2 Days overnight (+)3 Burden (+)3

p < .05 p < .05

p < .005 p < .005 p < .05

p < .000 p < .000

Public medications

RE probit Digestive cancers (–) Female cancers (–)

p < .05 p < .05 p < .05 RE Post-grad education

(caregiver) (+) Hospitalization RE probit Burden (+)

Digestive (+) Living alone (+)

p < .000 p < .05

p < .000 Public home appointment cost

RE probit Hamilton (–) Burden (–) Female cances (–) Post-grad education (caregiver) (–) Male (caregiver) (–) Living alone (–)

p < .05 p < .05 p < .05 p < .05

p < .05

p < .005 p < .000 p < .05

p < .005

RE Burden (+) Female cancers (+) Brain (+)

Caregiver burden RE Poisson Time loss (+) Post-grad education (caregiver) (+) Digestive cancers (+) Urinary cancers (+) Female cancers (+)

p < .000 p < .05 p < .05 p < .05 p < .05 p < .05

Given the context and previous literature described earlier in the thesis, these findings

are of interest for several reasons. Differences in how costs are generated across

payers speaks to the trend in the literature reporting opposing conclusions that adopt

different payer perspectives. Conversely, some drivers of cost are consistent across

payer perspectives; this is true for caregiver burden and brain cancer.

The literature review found a strong focus on haematologic malignancies and NSCLC.

These results are able to speak to alternative denominations of cancer that have not

been as thoroughly reviewed in the palliative care literature. Specialized programming

and treatment have resulted from the neoplasm-specific palliative care research that has

taken place around haematologic and NSCLC (Temel et al, 2010; Cartoni, 2007).


80

Therefore, these findings may be of interest to providers, decision-makers looking to

foster a more individual-specific palliative care experience, or those interested in the

establishment of palliative care as a specialized field (Shadd, 2008).

The following chapters will discuss how these results correspond to previous research,

the hypotheses, generalizability and future work, It is evident that primary cancer

diagnosis is an important driver of both economic costs (irrespective of perspective) and

caregiver burden associated with CBPHC. This suggests that cancer site has a role in

shaping the experience of CBPHC.

81

Chapter 5 Discussion

The objective of this thesis was to comprehensively analyze the financial and caregiver

burdens associated with CBPHC for people with cancer so as to identify key

determinants. Specifically, whether or not primary cancer diagnosis (ICD-10 cancer

categories) influenced the magnitude of costs and/or burden was of interest given the

paucity of research in those areas. The results showed that, for this Ontarian cohort,

some cancer sites played a role in the costs and caregiver burden associated with

CBPHC, from the perspectives of all considered stakeholders (society, the Ministry of

Health and unpaid caregivers). This chapter offers an overview of the findings vis-à-vis

the hypotheses stated in Chapter 1, followed by a discussion of the findings and the

literature. This leads into policy implications, and the chapter is concluded with

disclosure of the study limitations.

In Chapter 4, the results showed that the costs and caregiver burden associated with

CBPHC were significantly associated with four of the ICD-10 categories: brain, digestive,

gynecologic and urinary tract malignancies. Brain cancer was positively associated with

the aggregated cost of CBPHC from all payer perspectives and the service-specific cost

of home appointments (publicly funded); digestive cancers were associated with

increased probability of hospitalization and higher caregiver burden; gynecologic

cancers were associated with higher home appointment cost, caregiver burden and

Ministry of Health cost (albeit sensitive to model specification); urinary tract cancer was

associated with higher caregiver burden. The socio-demographic and care-related (i.e.

home care agency, days overnight, etc.) covariates also played significant roles that

were generally consistent with the conceptual frameworks and literature that influenced

their inclusion. Some of the hypotheses outlined in Chapter 1 pertained to these socio-

demographic covariates and they will be discussed with regard to the applicable

hypothesis in Section 5.1, whereas the ICD-10 variable of interest will be discussed

more in-depth.

Chapter 6: Conclusion R. Redmond-Misner

82

5.1 Findings vis-à-vis the hypotheses

While the hypotheses were informed by previous research (Section 1.4), the Andersen

and Newman framework (Section 3.1.3) informed data collection and the type of

variables conceptualized as generating health service use. In this study, service use is

represented by its monetary cost. The framework posits that determinants of health

service use include predisposing, enabling and needs-based factors. Variables from all

of these subsets were significant. Predisposing factors of caregiver employment status,

education and patient living arrangement were predictive of all aggregated costs and the

probability of hospitalization. The one enabling factor that was included in the statistical

models, which was a dummy for which home care agency the patient was enrolled in,

was significant for societal and unpaid costs. Need factors included the diagnostic

variable and comorbidity scores, of which the diagnostic categories were consistently

significant. This latter type of predictor has been described as “illness severity”

(Guerriere, 2012), thus it is notable that the diagnostic variable does not actually

attribute differential severities to the categories and treats them as nominal. This section

will describe the results pertaining to specific independent variables in relation to the

specific hypotheses.

H1: The influence of primary diagnosis on cost will be statistically significant.

The diagnostic variable was statistically significant for all cost regressions. However, the

magnitude, direction and ICD-10 category of significance varied. Malignancies of the

brain and gynecologic organs were both important drivers of each cost component, while

malignancies of the digestive organs significantly drove the probability of hospitalization.

These three diagnostic categories were the only significant diagnostic variables in the

cost or service usage equations. Brain cancer was consistently positive across all payer

perspectives and for public home appointment costs. Gynecologic cancers were

significant determinants of aggregated Ministry of Health costs (albeit sensitive to model

specification and therefore not as robust), and for the cost of public home appointments.

Most of the patients in the female organs cancer category had ovarian cancer.

H2: Costs will be driven by caregiver burden, which will be driven by primary diagnosis.

Evidence supported this two-part hypothesis. Aggregate costs were positively driven by


83

caregiver burden scores irrespective of payer, as were the disaggregated costs. The

probability of a hospitalization was also higher as burden scores rose. However, while

costs were most consistently driven by brain cancer, caregiver burden was not. This is

further discussed under H4. Caregiver burden may conceptualized as an ‘enabling’ factor

in the Andersen and Newman framework, because the extent to which the caregiver

feels burdened may impact on their ability to provide care.

H3: Socio-demographic variables will be statistically significant predictors of cost,

particularly SES indicators, marital status and patient living arrangement.

The former two hypotheses were more central to the research question and have fewer

sources of verification and justification in previous literature. The suspicion that socio-

demographic variables will be significant, particularly SES and living arrangement, is

based much more on the results of previous research (Gardiner et al, 2014; Guerriere et

al, 2010; Chai et al, 2013). However the SES indicators – education and employment

status – were not overly predictive in the analysis. One category from each, post-

graduate education and ‘on leave’ employment status, produced significant results. Post-

graduate educated caregivers were associated with higher public medication cost and

higher caregiver burden; caregivers ‘on leave’ were associated with greater unpaid

caregiver cost. The post-graduate caregiver education level was also associated with

significantly fewer home appointments, which suggests that burden may be greater

among dyads receiving fewer visits from the CBPHCT.

Patient living arrangements were also consistent drivers of cost. Patients living alone,

who are not frequently observed in existing research, had a significant and positive

effect on Ministry of Health costs. Living alone was also associated with a higher

probability of hospitalization, which helps to explain their elevated cost from the Ministry

of Health’s perspective. Patients living with children were associated with higher costs

from all perspectives, though the source of those positive coefficients was not further

illuminated by the analysis of service-specific costs. Patients living with others

experienced smaller unpaid costs. Because the possibility of living with multiple people

is implied by the term “others,” it is perhaps possible that the obscured/lowered cost

reporting that was attributed to multiple caregivers by Jacobs et al (2011) is happening

here.


84

In addition to these variables, city of residence was a significant determinant of unpaid

caregiver time. Hamilton was associated with less societal and less unpaid caregiver

costs than Toronto. Because all caregivers were valued the same way (human capital

approach, described in Section 3.4.2), this difference must be attributable to less time

spent caregiving. This suggests that Toronto and Hamilton residents may differ in

unobserved ways. The experience of home care is expected to differ depending on

location due to the level of services available (Kuluski, 2010). Other explanations may be

that Toronto residents had higher wages that allowed them to work less overall; perhaps

the Toronto program was busier resulting in more necessity for unpaid caregivers to

supplement care; or perhaps suburban Hamilton residents were more likely to know their

neighbours and have help from more people which offset the demand on the primary

caregiver.

H4: The influence of primary diagnosis in caregiver burden will be of statistically

significant.

Caregiver burden scores were associated with malignancies of digestive organs, female

organs and the urinary tract. Interestingly, urinary tract cancers did not determine

anything else. This is part of a trend of disassociation between the cancer-related

determinants of cost and burden that was alluded to under H2; brain cancer, which drove

all costs, did not drive caregiver burden at all. Digestive and female organs were also

associated with use of public services including hospitalization and home appointments.

Many of the patients in the ‘digestive’ category lived with colorectal cancer, which CIHI

(2013) found to be highly represented among hospital deaths. CPSO (2002) associated

hospital admission with caregiver burden. In Given et al’s (2004) conceptual framework

for assessing caregiver burden of people caring for cancer patients, this would be

considered a ‘care situation’ factor that determines the demands on the caregiver.

With regard to urinary tract cancers seemingly only driving caregiver burden, recall that

the urinary tract cancer category had to be moved into ‘other’ in the home appointment

probit regression because the frequency with which they used that service was 100%

which was skewing the estimates. Despite high frequency, their actual costs related to

home appointments were not found to be significant.

H5: Caregiver burden will be driven by time spent providing care, which will be


85

determined by primary diagnosis in the analysis of unpaid caregiver cost.

The caregiver burden score was driven significantly by time spent providing care, which

was positively associated with brain cancer. However, as outlined in H4, caregiver

burden was also driven by the primary diagnosis of the patient. Unpaid caregiver cost

(which is primarily time cost) and caregiver burden were driven by slightly different

diagnoses in the regressions where they were the dependent variables. Burden was

associated with malignancies of the digestive and female organs, which is most

congruent with the determinants of public cost. It was not associated with brain cancer,

which was the only diagnostic category to predict unpaid caregiver costs to a significant

degree. This suggests that, in addition to having a causal relationship with the amount of

unpaid time devoted to the patient, burden also has a close relationship with public

costs. This is further congruent with statements and the CPSO (2002) and findings

under H4 that indicate that worsening/intensification of caregiver burden instigates the

use of public services or hospitals. This usage trajectory may therefore be associated

more strongly with some diagnoses than others.

5.2 Comparisons and inferences from the literature

The primary research question of this thesis pertained to whether or not primary

diagnosis (defined by tumour site as per the ICD-10) played a role in costs and caregiver

burden associated with CBPHC. CBPHC is a growing service and evolving policy area

that is not yet reinforced by stable funding. In fact, it is increasingly reliant on unpaid

care because funding has not grown proportionately with demand following health care

restructuring (TLCPC. 2014). Therefore, establishing points of vulnerability can inform

admission and referral into these programs by understanding predictors of excessive

caregiver burden, hospitalization, or demand for public home appointments in advance.

In Ostgathe et al’s (2008) study, a panel of experts used prior knowledge of NSCLC

treatment to project what palliative home care would cost. Neoplasm-specific services

have been been developed and implemented in the case of haematologic malignancies

in response to their higher propensity for service use and neoplasm-specific needs (i.e.

transfusions, blood laboratory, etc.) (Cartoni et al, 2007).


86

Brain cancer is the most glaring subgroup in terms of elevated cost across all considered

cost dimensions. The majority of these patients lived with glioblastoma multiforme.

Gliomas are the most common subgroup of primary brain tumours, glioblastoma being

stage IV (Davis et al, 2011). There are unique and significant symptoms associated with

brain cancer from the perspective of both health professionals and unpaid caregivers.

Focal neurologic deficit, cognitive deficit and seizures are often already present at the

time of diagnosis (Flechl et al, 2013). The rapidity of symptom development is

unpredictable, echoing the concerns that underlie special focus on haematologic

malignancies (Bakitas et al, 2013). Depending on the location of the tumour, the

memory, personality or physical function of the patient can be suddenly compromised

(Cahill et al, 2011). People living with glioblastoma multiforme are usually bedbound at

the EoL, for as short as a week or as long as three months (Flechl et al, 2013). This is

consistent and fits logically with Wasner et al’s (2013) study in which the brain cancer

patients were unable to complete mental health assessments due to critical conditions

and had high I/ADL needs. Additional services sought by this particular group include

cognitive rehabilitation services (Davis et al, 2011). Thus there are several

characteristics of this tumour site that might explain why it differs from others in terms of

cost. Declining cognition impacts on caregivers emotionally and can have implications

for communicating symptoms according to Flechl et al (2013), although this was not

corroborated by the caregiver burden analysis.

Gynecologic malignancies did emerge as predictive of caregiver burden. This may be

because while brain cancer patients are potentially bedridden, many gynecologic

patients are actively pursuing various medical treatments during the palliative trajectory

according to Fauci et al (2012). The sample in their study was also largely comprised of

ovarian cancer sufferers. More than half of patients were undergoing radiation and

chemotherapy in the last six months of life, and more than 80% were hospitalized at

least once (maximum 14 times). The patients commonly suffered gastrointestinal

complications and required surgical procedures related to that, namely paracentesis

(removal of excess peritoneal fluid) (Fauci et al, 2012). Therefore being associated only

with Ministry of Health costs makes sense, and higher caregiver burden may be related

to the travel, time and stress of the patient undergoing more procedures during their

palliative trajectory even if they do not translate into exponentially more time.

Gynecologic cancers were also associated with higher publicly financed home


87

appointments; the doctors in Ostgathe et al’s (2008) study regarded home appointments

as costly enough in their own rite to make home care more costly than conventional

treatment.

Malignancies of digestive organs were shown to be important determinants of both

hospitalization and caregiver burden. Many of the patients in this category had colorectal

cancer. Hospitalization generally was a common focus when assessing the relative

success or cost of home care (Wong et al, 2013; Simoens et al, 2010). The finding that

colorectal cancer patients had a higher probability of being hospitalized is congruent with

the paucity of literature concerning this group. CIHI (2013) recently reported that

colorectal cancers are highly represented in hospital deaths. CPSO (2002) attributed

EoL hospital admissions to caregiver burden, it is interesting that this diagnostic

category drives both. The CIHI (2013) report also cited lung cancer amidst high EoL

hospitalization; in this analysis, lung cancer was the baseline category that digestive

cancer patients were significantly more probable than, so that part of the report is not

corroborated. While the CPSO’s (2002) attribution of EoL hospital admission to caregiver

burden and trouble with interpreting symptoms (also reported by Docherty et al, 2008

and Parker Oliver et al, 2014) is being drawn on to explain this trend, there is little in the

literature that explicitly tries to answer why this might be the case.

The final ICD-10 category that played a significant role was that of urinary tract cancers

on caregiver burden. The literature is especially sparse for this population, although

Ershler (2003) points out that it is particularly prevalent in older patients (80+). These

latter categories indicate areas for future research. The remaining groups that did not

diverge significantly from the expenditure or burden level associated with lung cancer

include male genital cancers, breast cancer, and ‘other.’ The results relating to the

variable of interest, ICD-10 category, imply that what enrollees can expect from palliative

home care will be dependent on the primary diagnosis of the patient. This information

could be helpful in terms of referral and predicting vulnerable caregivers (i.e. colorectal

cancer) and foreseeing higher demand on the CBPHCT (i.e. brain cancer).

The variability in cancer types that drive cost and burden seemingly validate the primary

hypotheses and answer the research question by showing that significant variability

does exist between patients with different types of cancer. The other clinical indicator,

comorbidity scores, had no significant influence. This suggests that in addition to being


88

inherently secondary to the patient’s primary affliction, comorbidities are secondary in

their capacity to determine cost and caregiver burden. Yet this has been a more

common measure in the existing literature. It has also been found to be significant in

previous palliative care research (Masucci, 2011; Guerriere, 2012).

Cancer diagnoses constitute a ‘need factor’ in the broader conceptual framework that

also includes predisposing and enabling factors (Andersen & Newman, 1973). While the

Andersen and Newman framework applies to health services use, Given et al (2004)

outlined similar predictors of caregiver burden. There are additional notable associations

with cost and burden that were derived from the analysis. Living alone enhanced the

probability of hospitalization among CBPHC patients. This is a ‘predisposing’ factor that

can also be found in the literature as a source of potential vulnerability in home care

(Aoun et al, 2013). In the aggregated costs, living alone was only a significant predictor

of Ministry of Health spending. That may be attributable in part to heightened use of

hospitals, which may be related to unavailability of unpaid caregivers. The unavailability

of unpaid caregivers has been cited as a barrier to accessing home care at all in past

literature (Aoun et al, 2013). The Hamilton program enrollees produced negative

coefficients across the board. This is an enabling factor and may be attributable to

unobserved program characteristics that were found to impact on caregiver burden by

Guerriere et al (2013). Differences across home care agencies were also found by

Hirdes et al (2012).

Another aspect of the literature that the results resonate with is that estimations of cost

are highly sensitive to payer perspective. This has been particularly true in the case of

conventional and home care cost estimations that do or do not consider the cost of

unpaid care. Most Ministry of Health studies, including Wong et al (2013) and Klinger et

al (2010) find home care less expensive, whereas societal and caregiver perspective

studies find it more costly (Jacobs et al, 2011). Not only are estimates going to be

different, but these results indicate that the data generation process will be too. This

means that, while it seems that primary diagnosis plays a role, different predictors may

have different implications for different payers. For example, the aggregated cost results

presented here indicate that Toronto residents receive much more unpaid care while the

difference in cost is insignificant to the Ministry of Health or home care program.

Caregiver burden on the other hand inflates costs across the board and is in everyone’s

interest to curb. So does living with children of the patient. Conversely, while living alone


89

drives public cost, it does not place strain on an unpaid caregiver.

Despite some disjointedness in the significant predictors of cost and burden, caregiver

burden is a strong predictor of the costs associated with CBPHC. Caregivers are

invaluable to the sustainability of home care. Thus determinants of their burden are

useful for stabilizing this mode of delivery for palliative care. Caregiver burden was

strongly positively related to the time loss of the caregiver, although findings to the

contrary have also been made (Kenny et al, 2010). The rise in informal caregiving is not

an extraneous effect of the shift to community-based services inherent in the

restructuring of health case, but an intentional part of the plan (Church et al, 2002; Yu,

2011). Their preservation is of interest for all stakeholders. In Ward-Griffin et al’s (2012)

study, caregivers reported acting as coworkers with the palliative care team and felt that

attention on their needs was primarily for the purpose of ensuring that they continued

working as caregivers. Interestingly, patient living arrangement was not significant

despite literature citing living with the patient as having negative implications for respite

and therefore caregiver burden (Glajchen, 2012).

Many predictors that were found to be significant in the literature were able to be

included as covariates in the analyses, but some variables of determinacy were unable

to be included. This is true for the role of provider characteristics (Guerriere et al, 2013)

and ethnicity (Gardiner et al, 2014). The latter was reported as a weakness of several

studies (Temel et al, 2010; Docherty et al, 2008). While the non-diagnostic variables

were not of paramount interest, they were included in the statistical models in order to

form the most comprehensive and informed data generation process to assess primary

diagnosis amidst. The role of primary diagnosis emerged among the majority of

significant determinants established in previous literature. This fortifies the finding and

congruency with past research also suggests that this is not a particularly unique or

abnormal home care sample. That said, some potentially important covariates from the

literature could not be assessed.

Some of the trends in the data also speak to the policy concerns that were discussed at

the outset of the thesis. A large proportion of the caregivers were children and spouses;

sources of unpaid care that are expected to gradually lessen due to retirement

postponement and people having fewer children (Jiwani, 2003). As CBPHC shifts toward

a source of labour that is projected to be sparse, identifying predictors of strain on them


90

can be helpful with planning, even at the agency level. Conversely, due to health care

restructuring, agencies themselves struggle to spread their resources across all the

demand (TLCPC, 2014), therefore it is also important to identify characteristics

associated with their strain. The following section will elaborate on the policy situation

that CBPHC is currently in and how this pertains to the research findings.

5.3 Policy implications

Home care is currently in a position where funding is not growing proportionately with

demand (TLCPC, 2014). It is therefore of interest to avoid high costs and high burden so

as to maintain unpaid caregivers (DeMiglio et al, 2012). Profiling patients to circumvent

financial strain and navigate provision, for all stakeholders and providers, has

manifested in personalized medicine (Kamal et al, 2013; Unroe et al, 2013) and

advocacy for the specialization of palliative care (Shadd, 2008). Personalized medicine

posits that the management of patients should be nuanced by specific demographic

characteristics and risk factors in order to predict burdens in advance. This study

revealed three subgroups that are associated with greater burden for service providers.

Brain cancer was associated with higher service use and unpaid caregiver time;

digestive cancers (primarily colorectal in this cohort) had a higher probability of

hospitalization; brain and gynecologic cancers (primarily ovarian in this sample)

predicted home appointments; gynecologic and digestive cancers predicted caregiver

burden felt by unpaid caregivers.

Home care’s lack of recognition in Canada’s national insurance plan leaves it to the

discretion of the province to organize (DiMiglio et al, 2012). Prominent representatives

such as Dr. Larry Librach of the Temmy Latner Centre for Palliative Care and Palliative

Care Council in Toronto have argued that palliative care needed distinct policy

representation (Ogilvie, 1998). The Romanow and Kirby reports both called for home-

based palliative care to become an insured service under the Canada Health Act

(Romanow, 2002; Kirby, 2002). In coming years, palliative home care specifically can be

expected to rise due to the demographic and epidemiologic composition of Canada.


91

Prior acknowledgement of risk factors for future diseases or complications, as opposed

to treating existing problems as they come, can be an avenue for prevention as well as

cost saving through the prevention of things like unforeseen hospitalization.

Understanding the service usages associated with different ICD-10 denominations of

neoplastic malignancy could enhance the personalization and specificity of CBPHC.

Gaertner et al (2011) propose a disease-specific approach that entails standard

operating procedures (SOPs) for nineteen disease sites (categorized similarly to those

here). To date, internationally, specialized programming has been created for patients

with haematologic cancers (Cartoni et al, 2007), and specialized treatment regimens

have been developed for patients with NSCLC (Temel et al, 2010). Conversely, if some

cancers are associated with particularly high cost and/or caregiver burden, perhaps

there are better settings for them to receive care.

The senior’s population will double by 2036 and cancer is the leading cause of death in

Canada (Kirkey, 2010). Improving palliative and EoL care and setting up sustainable

infrastructure for it is a global public health priority on the agendas of both the WHO and

UN (Broad et al, 2013). The current funding model for CBPHC and the institutions that

people might use to in lieu of home care, such as hospices, relies on charitable

donations and unpaid labour (Nash et al, 2013). The goal of shifting care to unpaid

community members (Spalding, 2005; Yu, 2011) combined with the aforementioned

demographic prognoses means that the importance of CBPCTs (DeMiglio et al, 2012)

and unpaid caregivers (Haley, 2003) are going to rise even more substantially (Seow,

2009). Currently, many unpaid caregivers of patients with cancer lose considerable

savings and require health services themselves after bereavement (Bachner, 2007).

In the Ontario context, the province is divided into fourteen LHINs within which are

CCACs that facilitate community-based health care services on an even more

compartmentalized level (Klinger et al, 2011). CCAC’s were created in Ontario as the

single access point for home care services, palliative or not (Ogilvie, 1998; Seow, 2009).

Not-for-profit and for-profit agencies compete for CCAC home care service contracts by

which the CCAC acts as their case manager, point of access and coordinator (DiMiglio

et al, 2012). This was preceded by Pain and Symptom Management Coordinators

responsible for forming palliative care resource teams, but they too experienced financial

constraints and were implemented by the province under the assumption that teams

could be formed without using additional resources (Howell, 2003). Thus to achieve the


92

sustainability evoked by the WHO and UN at present, this service is contingent on

austere use of their resources. This can be guided and aided through profiling recipients

and their associated strain on providers, including those that are unpaid.

The intention of capping future costs fuels policy interest in gradually transforming

volunteers, previously conceptualized as complementary to public services, into

legitimate service providers and self-sustaining public service substitutes (Eakin et al,

2009; Stajduhar et al, 2010). The results found here suggest that this is happening. For

the government to reimburse informal caregivers in the ways that other interest groups

have proposed, such as the Canadian Cancer Society or the NDP, would defeat the

purpose of the shift to community-based care. While community-based care might drive

up invisible unpaid costs, there are potential savings to be derived from these

alternatives from a ministerial perspective. Baranek (2000) argues that because

representation for societal spending on community-based care is more weakly

organized, government interests have been dominant in policy outcomes which favour

the public portion of the funding mechanics. In a survey conducted among Ontario’s

fourteen CCACs, the goal of “optimal health, independence and/or quality of life” was

second only to “cost-effective resource use,” bespeaking the thin spread of resources in

palliative care (King, 2002, p. 65). Thus the need to distribute resources optimally is

certainly felt by and of interest to service providers.

In moving palliative care in a more complex, specialized and encompassing direction,

recognizing it as a specialization has been advocated (Shadd, 2008;). This could

reformulate palliative care to further accommodate personalized care, give it more

legitimacy to compete for funding, and perhaps lead to the national insurance once

advocated by Romanow (2002) and Kirby (2002). “Common assessment tools,

standards of practice, central inventories of resources, shared information systems,

acute PC designed beds, and common access points to care” are necessary not only for

effective provision of palliative care, but capacity to prove effectiveness and better

qualify for funding consideration (Bainbridge et al, 2011, p. 275). Howell’s (2003)

research concluded that “Home as the place of care at the end-of-life and place of death

can be supported if there is a willingness to allocate the maximum level of services

allowable through … [the Ministry of Health]” (p. xiv). The current logic of finite and

intermittent block funding could be related to state hesitance to legally commit to the

unknown future costs of palliative care. In the mean time, professionals at the program


93

level can only assess the suitability of the patient and presumably the intensity of their

needs in the palliative trajectory based on socio-demographic and clinical aspects of

their profile.

Some of the solutions, such as neoplasm-specific standard procedures for palliative care

(Gaertner et al, 2011), do imply relatively specialized and sophisticated services.

CBPCTs are intended to have physicians, nurses, psycho-spiritual counseling,

bereavement counseling, administrative staff and case managers working through

CCACs, but this is not always the case. The scope of a respective LHIN’s CBPCT is

dependent on their capacity to finance this (Bainbridge et al, 2011; Seow, 2009). If some

CBPCTs struggle to finance full teams, it is likely that they will face barriers in moving

toward or implementing the type of neoplasm-specific programming seen with

haematologic malignancies and NSCLC. Some regions with less resources may have

partial teams limited to one nurse (Nash et a, 2013) or restrict the times services are

available to a day time model as it is (with the ideal availability being 24/7) (Bainbridge,

2011; Sussman et al, 2011). It is also possible that with enough cost-driving barriers

experienced simultaneously, certain regions cannot feasibly or cost-effectively be

provided with a CBPCT at all (Kuluski, 2010). Recognizing palliative care as a

specialization may attract more funding and personnel to enable this (Shadd, 2008).

In these highly sensitive areas, it can be expected that the burden on unpaid caregivers

will be greater. So determining characteristics associated with high burden or service

use can illuminate the practicality of referring someone to home care. Some of the

predictors found in previous literature included time spent providing care (Hides et al,

2012), living arrangement and marital status (Glajchen et al, 2012). Urban and rural

capacity for development of community-based services is one of the most prominent

disparities among Ontario LHINs (Schuklenk, 2011). This study identified gynecologic

and digestive malignancies as predictive of caregiver burden, presumably due to

associated symptoms and demands of care. The often foreign clinical demands of home

care have been reported as a stressor by unpaid caregivers (Docherty et al, 2008); the

CPSO (2002) attributed EoL hospitalization to the panic that can be induced by rapidly

developing symptoms. These types of cancer were also associated with elevated

hospitalization and home appointment cost. Therefore the profiling of patients including

the risks associated with their malignancy may reduce caregiver burden. Going forward,

the primary diagnosis of the patient could be indicative at the onset of palliative care


94

whether or not home care would be exceedingly strenuous on any providers, especially

unpaid caregivers. This neoplasm-specific SOPs outlined by Gaerner et al (2011) that

attempt to customize the palliative trajectory according to primary diagnosis.

The potential for the results of this study to inform policy and practice is relatively broad.

These data pertain to two Ontario CBPHC programs, but the care demands associated

with a given cancer site can speak to palliative home care for people with cancer

anywhere. The WHO’s World Cancer Report (Stewart & Wild, 2014) recently stated that

cancer is expected to surge globally over the next twenty years should no major

preventative measures be taken (including behavioural changes of the patients, i.e.

smoking). Cancer rates also rise with an aging population, and many developed

countries are currently experiencing this demographic trend (Hume & Christensen,

2014). The literature review included studies from several countries (Canada, US,

Singapore, Israel, Italy, Germany and Greece). This type of care is emerging in

response to the growing demand in order to circumvent some of the cost by sharing the

responsibility for care with community members. The ICD-10 defines cancer by tumour

site (i.e. breast, brain) rather than tumour type (i.e. sarcoma, carcinoma), which is also

beneficial and preferable for many reasons. This is how both patients with cancer and

specialized cancer organizations typically identify themselves (i.e. Canadian Breast

Cancer Foundation, Prostate Cancer Canada, Brain Tumour Foundation of Canada,

etc.). It is also how previous neoplasm-specific research has defined the type of

neoplasms they are focusing on (Temel et al, 2010; Gaertner et al, 2011).

In conclusion, the extent of the usefulness of these findings is contingent on policy

decisions. Currently, this knowledge can be utilized for personalized medicine; the use of

risk factors and patient characteristics to assess suitability of the patient for the service

and to foresee their potential complications. To implement more specialized

programming, as has been done with haematologic malignancies or NSCLC, may

require more infrastructural support for the palliative care field.


95

5.4 Limitations

There are several limitations to this study. Firstly, generalizability may be limited by

sample size, regional differences in the level of CBPHC available, and unknown

differences between recipients and non-recipients. The sample size and number of

observations in these data are generous meaning that the results are generally

dependable, especially with the validation of previous literature. It is simply with regard

to the diagnostic variable of interest, for which there is little research (hence this thesis),

that further research and validation is needed. There is no control group to show how

this population compares to non-recipients or how representative it is of the general

population. It has been shown in previous research that receiving home care at all is

sensitive to a number of factors including SES (Motiwala et al, 2006), the availability of

unpaid caregivers (Aoun et al, 2013), and home care availability related to rurality and

centrality (Skinner, 2005; Funk et al, 2010; Kuluski, 2010; Bainbridge et al, 2011).

However, it presumably reflects the recipients of home care, for whom the main

inferences from the findings are intended (i.e. caregivers of patients with digestive organ

cancers in home care are vulnerable to burden).

Second, these data and analyses are missing potentially important variables. Ethnicity

was found to be an important predictor by Gardiner et al (2014). Several studies in the

literature review reported ethnically homogenous samples as a limitation to

generalizability, and this study cannot contribute to filling this gap either. However, it is

notable that lack of data does not necessarily equate to a racially/ethnically homogenous

sample, with Toronto being a very diverse city (Guerriere, 2012). It is simply

unrepresented in the data and can therefore not be spoken to, even if ethnic diversity did

exist.

Third, the costs in this study are reported by unpaid caregivers which opens the data up

to several types of bias including selection bias, social desirability bias and recall bias

(Guerriere et al, 2010). In Wasner et al (2013) and Grov et al’s (2006) studies, patients

and caregivers experiencing worse symptoms were unable to participate. The people

who are receptive to bi-weekly interviews may therefore represent dyads with lesser

burden or lesser costs. Comorbidity scores were not found to be significant and this

could always be because patients with the worst comorbidities did not partake. The

AHCR has been validated, with caregiver reports closely matching administrative data


96

and demonstrating minimal impact for this potential bias (Guerriere et al, 2006).

Moreover, because the data were gathered prospectively, they are much more resilient

against recall biases found in retrospective studies. The interviews could have been

conducted even closer together to further defend against this type of bias, but two week

intervals were chosen by the research team who collected this data in order to mediate

between minimizing recall bias and minimizing interview burden for the participating

caregivers.

5.5 Summary

This chapter discussed the findings as they pertained to the hypotheses, the literature,

and policy implications. The research question was whether or not primary cancer

diagnosis would impact costs and caregiver burden, thus the broad hypothesis (broken

down into 5 hypotheses in Sections 1.4 and 5.1) was that primary diagnosis would play a

role in both. This broad hypothesis was supported by the evidence. The literature was

used to interpret the findings surrounding brain, digestive and gynecologic cancers and

their relationships with cost, hospitalization, public home appointments and caregiver

burden. The findings had implications for policy and personalized medicine, wherein they

could inform the planning of palliative trajectories taking into account these risk factors

so as to curb the strain on providers. This is important to the maintenance and

sustainability of this service given the current policy context that is characterized by

minimal funding and reliance on unpaid care.

97

Chapter 6 Conclusion

The objective of this thesis was to comprehensively analyze the financial and caregiver

burdens associated with CBPHC for people with cancer and evaluate the role of their

primary diagnosis. Primary cancer diagnosis (as defined by the ICD-10) was found to be

a consistently significant determinant of both CBPHC costs and caregiver burden. In the

concluding chapter, a summary of the thesis will be given, followed by recommendations

for future work.

6.1 Thesis summary

The central hypothesis of this thesis was that primary cancer diagnosis would play a

significant role in the generation of CBPHC cost and caregiver burden. The conceptual

framework articulated by Andersen and Newman posits that health services use

(represented by cost in this study) is determined through socio-demographic, clinical and

community-related factors. Many conceptual frameworks for caregiver burden include

similar predictors (Givens et al, 2004). Clinical factors in previous analytic models have

used comorbidity scores, the presence of any cancer, and cancer stage (Motiwala et al,

2006; Fairfield et al, 2012; Sussman et al, 2011; Sims et al, 1997). Many studies also

focused on one disease site, with much of the disease-specific studies focusing on

haematologic malignancies and NSCLC (Temel et al, 2010; Cartoni et al, 2007). While

still controlling for comorbidity, this study integrated a nominal cancer variable defined by

the ICD-10 categories that the patients fell into. The ICD-10 distinguishes among tumour

(i.e. breast, brain) sites as opposed to tumour types (i.e. carcinoma, sarcoma). This does

not only represent previously underrepresented neoplasm subgroups, but allows nine

disease sites to be observed side-by-side for comparative inference among brain,

breast, male, female, digestive, urinary, lung and other cancers.

This research was informed by the literature on home care costs, caregiver burden, and

palliative oncologic care. The literature on the cost of home care included research

discussing the comparative and actual cost of home care, payer shares of this cost, and

determinants of cost. Whether home care was deemed more or less expensive, and


98

standalone estimates of its cost, were drastically different depending on the payer

perspective that is used. It was therefore decided to analyze cost from different payer

perspectives, which may be generated differently when it comes to determinants of cost.

This is allowable through the data which uses the comprehensive Ambulatory Home

Care Record for costing, which adopts a societal perspective but categorizes costs

according to payer (considered here: societal cost, Ministry of Health cost, and unpaid

caregiver cost). Determinants of cost in the existing literature included physical function

(represented here by comorbidity scores) (Guerriere et al, 2010), SES (represented here

by education and caregiver employment status) (Chai et al, 2013), marital status and

living arrangements (Guerriere et al, 2010). The analytic models presented in this thesis

were able to control for these important predictors alongside the key variable of interest,

namely, primary cancer diagnosis. One variable that was missing was ethnicity.

The literature on caregiver burden focuses on problems associated with caregiver

burden, unmet needs reported by caregivers, and determinants of satisfaction and

burden. Determinants of burden were related to service utilization, patient condition and

the amount of time they spent providing care. Disability, living arrangements, marital

status, SES, hours of unpaid care, and home care agency were the variables found to

be significant in the previous literature that were controlled for alongside primary cancer

diagnosis (Glajchen, 2012; Götze et al, 2014; Hirdes et al, 2012). It was found that in

addition to the psychosocial burden that is documented through the caregiver burden

scale, caregivers also experience physical burden and health problems (Glajchen, 2012)

that are not represented by the caregiver burden scale. Gardiner et al (2014) also tied

ethnicity to caregiver burden, which was once again not able to be included in the

analyses. Carlsson et al (2003) found place of death to predict burden among bereaved

caregivers in their retrospective study; while this data was collected, the prospective

data collection means that place of death was not known when reporting burden and

therefore was not included as a covariate.

The literature discussing palliative care for oncologic patients was sought for justification

of the inclusion of the diagnostic variables. While neoplasm-specific palliative care is a

relatively new and sparse research area, some interesting findings were derived.

Malignancy-specific symptom control was interpreted as a trace of this ostensible gap in

the literature. The contribution of these findings to this thesis is the fundamental and

broad suggestion that symptoms and palliation needs vary from malignancy to


99

malignancy, and may therefore be influential in predicting cost and caregiver burden. It

seemed that the NSCLC and haematologic cancers were the most thoroughly

researched diagnoses to date, however Gaertner et al (2011) presented customized

SOPs for nineteen malignancies which reinforced the idea of analyzing them separately.

The analyses in this thesis notonly address this gap but illuminate whether such

distinctions are necessary and informative, or if proceeding with the convention of

homogenizing solid tumours is sufficient and inconsequential. Finally, ethnic

homogeneity was cited several times in the literature (Bakitas et al, 2013; Fauci et al,

2012; Temel et al, 2010), which cannot be addressed by this thesis.

Costs were analyzed taking advantage of the detail of the Ambulatory Home Care

Record. Payer perspectives, which were suspected to produce differential results based

on the literature, were divided into Ministry of Health, unpaid caregiver, and societal.

This helped to illuminate the branch of societal costs driving the coefficients in the

societal cost regression, and demonstrated how public and unpaid dimensions of home

care might stand in for one another (i.e. living alone is associated with higher Ministry of

Health cost, but lower unpaid care cost). Being that Ministry of Health costs are spread

across a variety of services, some smaller components of that cost were looked at

separately to derive what specific service usage might drive Ministry of Health cost

(particularly, publicly provided medication, hospitalization, and home appointments).

Caregiver burden scores were calculated using the CBS-EOLC, which measures

psychosocial burden (Dumont et al, 2008). Items in the questionnaire relate to

emotional, social and financial burden but not physical symptoms or changes.

Conceptual frameworks of caregiver burden, directed at both dementias and cancers,

prioritize patient and caregiver characteristics as well as the demands of care related to

patient symptoms (Givens et al, 2004). Thus this analysis, as well as the cost analyses,

were controlling for many of the most important covariates when assessing the role of

primary cancer diagnosis.

It was found that, amidst other influential predictors, primary cancer diagnosis was

frequently a driver of both costs and burden. Particularly, brain cancer drove aggregated

cost irrespective of payer, while female and digestive cancers drove both hospitalization

and burden. These are the only ICD-10 categories that were significant predictors in the

analyses. The literature review found a strong focus on haematologic and NSCLC.


100

These results are able to speak to alternative denominations of cancer that are

underrepresented. Specialized programming and treatment have resulted from the

research that has taken place around palliative care for haematologic and NSCLC.

Therefore, these findings may be of interest to providers, decision-makers looking to

foster a more individual-specific palliative care experience or further specialize the

palliative care field (Shadd, 2008). Currently, it is particularly useful for profiling the

recipients of this care in advance so that complications leading to hospitalizations or

caregiver burden can be mediated and anticipated. Restructuring of health care in

Canada has been partially characterized by shifting services to the community

(Guerriere, 2012; Laville et al, 2007; Spalding, 2005; Skinner, 2005). While there are

demographic signs that the patient base is going to grow as the caregiver population

shrinks, further development of palliative home care has been endorsed in several

influential policy reports and is likely the direction that Ontario is going in. Recent

endorsements for the further development of community-based palliative home care

(CBPHC) include the Health Council of Canada’s (2013) Progress Report, Health Quality

Ontario’s (HQO) (2012) Report on Ontario’s Health System, the Ontario Seniors’

Secretariat’s (2013) Action Plan for Seniors, and Drummond et al’s (2012) Commission

on the Reform of Ontario’s Public Services.

6.2 Future work

To build on this work, there are several avenues that future research could take:

verification of the results, clinical interpretation of the results, caregiver interpretation of

the results, comparative analysis and places of death analysis.

Verification of the findings

Future work should endeavor to verify these results by analyzing the ICD-10 diagnoses

among other samples to enhance generalizability. It would be useful to diversify the

home care context within Canada, as the programs studied here are in metropolitan

areas with more resources than their rural counterparts. The studies that emerged in the

literature review were internationally variable, but had many consistencies across the

caregivers of home care recipients. Thus it would also be useful and complementary to


101

yield international results. The further development and preservation of community-

based services is of international interest as trends of an aging population and growing

cancer incidence are not limited to Canada (Stewart & Wild, 2014). Future analytic

models could surpass the comprehensiveness of these ones should they include

additional variables of interest that were missing here. The inclusion of ethnicity could be

important and explanatory for both cost and caregiver burden analyses (Gardiner et al,

2014). For caregiver burden analyses, provider characteristics such as those in the

Quality of End-of-Life Care and Satisfaction with Treatment (QUEST) questionnaire

would be ideal to include as per the work done by Guerriere et al (2013).

Clinical interpretation of the findings

This study constitutes an analysis and attainment of patterns in cost and caregiver

burden across ICD-10 diagnostic categories. The purpose of this is so that they can be

useful and interpretable to people shaping policies, working and providing care in

CBPHC. There would be definite value in the clinician interpretation of the results; what

is it about brain cancer that makes it more costly in the home care context? What is it

that makes home appointments more frequent among brain and gynecologic cancers?

What is it that makes hospitalization more frequent among gynecologic and digestive

cancer patients? In-depth clinical insight into the differences among these neoplasms

could enhance personalized palliative care and inform more specialized care, as has

been implemented for NSCLC and haematologic malignancies. Previous studies have

premised their recommendations for (Gaertner et al, 2011) and anticipated demands

associated with home care (Ostgathe et al, 2008) entirely on professional/clinician

opinions and projections.

Caregiver interpretation of the findings

Caregiver experiences with symptom management tasks emerged as a source of

uncertainty and stress in the literature (Docherty et al, 2008; Parker Oliver, 2014). This is

part of why it was expected that clinical characteristics that determine those symptoms

would drive caregiver burden differentially. Care tasks are used as a broad, generalizing

term, but specific burdensome dimensions of caregiving have been identified in the case

of polypharmacy (Sheehy-Skeffington et al, 2013). Future work could pursue a more in-

depth explanation of the results by eliciting the interpretation and specific experiences of


102

unpaid caregivers of patients with different malignancies. The two subgroups that were

associated with higher burden in this study were gynecologic and digestive cancers.

Unpaid caregivers could help to interpret why these malignancies may be more

burdensome. In the case of digestive cancers, which were significantly associated with

hospitalization, they may be able to illuminate specific explanations for that elevated

probability. This, combined with the interpretation of clinicians, could shape an idea of

the palliative trajectories that characterize these differential costs and further validate or

otherwise nuance the results.

Comparative analysis/extension to other settings

It is important to keep in mind that the analysis and findings of this thesis can only speak

to variability among a specific group of palliative home care patients. While some

characteristics have been identified as “more costly” or associated with “more burden,”

this is only within recipients of CBPHC. Whether or not these more expensive and

intensive patients exceed the cost or demand on caregivers experienced in other

settings is unknown. It was found by Singer et al (2005) that unpaid caregivers in the

home setting, despite being more sleep deprived and possibly contributing more time,

were still happier due to the facilitation of the patient’s preference. Therefore it would

also be interesting to look at the role of primary diagnosis across settings.

Places of death analysis

While costs and caregiver burden are of prominent interest in CBPHC research, so is

place of death (Brink et al, 2008). From a Ministry of Health perspective, CBPHC is

appealing for curbing institutional death, but it is also appealing to patients for facilitating

often-preferred home death. This is a commonly studied outcome in the field that also

has not been assessed vis-à-vis primary diagnosis. It would be complementary to this

topic and to personalized palliative care to further assess the role of primary diagnosis

through its impact on place of death.

This research found primary cancer diagnosis to be a consistently significant

determinant of CBPHC cost and associated unpaid caregiver burden. In a fragile sector

that must prioritize the maintenance of unpaid caregivers and optimize the distribution of

resources, it is incredibly useful to be aware of characteristics that are predictive of these

aspects of care. As Ontario shifts palliative care into the community, patients are being


103

discharged faster at higher levels of complexity (Laporte & Rudoler, 2013). By profiling

vulnerable candidates for home care, patients who are at much greater risk of

hospitalization, high caregiver burden or excessive service use may be foreseen. This is

not only of interest to precariously funded home care programs to be able to form a

preliminary expectation, but also for caregivers to understand the magnitude of burden

they may be more likely to experience caring for someone with gynecologic, digestive or

urinary tract cancers.

104

References Alesi, E., Fletcher, D., Muir, C., Beveridge, R. & Smith, T.J. (2011). Palliative Care and

Oncology Partnerships in Real Practice. Oncology, vol. 25(13), 1287-1293.

Allison, PD. (2009). Fixed Effects Regression Models. SAGE Publications, University of Pennsylvania: Pennsylvania, USA.

Andersen, R. & Newman, JF. (1973). Societal and Individual Determinants of Medical Care Utilization in the United States. The Milbank Memorial Fund Quarterly: Health and Society, vol. 51(1), 95-124.

Aoun, S. & Skett, K. (2013). A longitudinal study of end-of-life preferences of terminally-ill people who live alone. Health and Social Care in the Community, vol. 21(5), 530-535.

Aoun, S., Kristjanson, L., Currow D. & Hudson P. (2005). Caregiving for the terminally ill: at what cost? Palliative Medicine, vol. 19, 551-555.

Bachner, Y. (2007). Psychometric Properties of a Modified Version of the Caregiver Reaction Assessment Scale Measuring Caregiving and Post-Caregiving Reactions of Caregivers of Cancer Patients. Journal of Palliative Care, vol. 23(2), 80-86.

Bainbridge, D., Brazil, K., Krueger, P., Ploeg, J., Taniguchi, A. & Darnay, J. (2011). Evaluating Program Integration and the Rise in Collaboration: Case study of a palliative care network. Journal of Palliative Care, vol. 27(4), 270-278.

Bakitas, M., Doyle Lyons, K., Hegel, M.T., & Ahles, T. (2013). Oncologists’ perspectives on concurrent palliative care in a National Cancer Institute-designated comprehensive cancer center. Palliative and Supportive Care, vol. 11, 415-423.

Balducci, L. (2003). New Paradigms for Treating Elderly Patients With Cancer: The Comprehensive Geriatric Assessment and Guidelines for Supportive Care. The Journal of Supportive Oncology, vol. 1(2), 30-7.

Balducci, L. (2003). The Elderly Patient With Cancer: New Approaches for Improved Outcomes. The Journal of Supportive Oncology, vol. 1(2), 3-4.

Baranek, P. (2000). Long Term Care Reform in Ontario: The Influence of Ideas, Institutions, and Interests on the Public/Private Mix. Dissertation, University of Toronto: Toronto, ON.

Beeuwkes Buntin, M. & Zaslavsky, AM. (2003). Too much ado about two-part models and transformation? Comparing methods of modeling Medicare expenditures. Journal of Health Economics, vol. 23, 525–542.

Bergman, J., Kwan, L., Fink, A., Connor, SE., & Litwin MS. (2009). Hospice and Emergency Room Use by Disadvantaged Men Dying of Prostate Cancer. The Journal of Urology, vol. 181, 2084-2089,

Bertschek, I. & Lechner, M. (1998). Convenient estimators for the panel probit model.

References R. Redmond-Misner

105

Journal of Econometrics, vol. 87, 329-371.

Brink, P. & Frise Smith, T. (2008). Determinants of Home Death in Palliative Home Care: Using the interRAI Palliative Care to Assess End-of-Life Care. American Journal of Hospice & Palliative Medicine, vol. 25(4), 263-70.

Broad, J., Gott, M., Kim, H., Boyd, M., Chen, H. & Connolly, M. (2013). Where do people die? An international comparison of the percentage of deaths occurring in hospital and residential aged care settings in 45 populations, using published and available statistics. International Journal of Public Health, vol. 58, 257-267.

Brumley, R., Enguidanos, S., Jamison, P., Seitz, R., Morgenstern, N., Saito, S., McIlwane, J., Hillary, K. & Gonzalez, J. (2007). Increased Satisfaction with Care and Lower Costs: Results of a Randomized Trial of In-Home Palliative Care. The American Geriatrics Society, vol. 55(7), 995-1000.

Bukki, J., Scherbel, J., Stiel, S., Klein, C., Meidenbauer, N. & Ostgathe, C. (2013). Palliative care needs, symptoms, and treatment intensity along the disease trajectory in medical oncology outpatients: a retrospective chart review. Support Care Cancer, vol. 21, 1743-1730.

Cain, R., Maclean, M. & Sellick, S. (2004). Giving support and getting help: Informal caregivers’ experiences with palliative care services. Palliative and Supportive Care, vol. 2, 265-272.

Campion-Smith, B. (May 22, 2014). Ontario NDP pledge tax credit for family caregivers. The Toronto Star: Toronto, ON. Retrieved from http://www.thestar.com/news/ontario_election/2014/05/22/ontario_ndp_pledge_tax_credit_for_family_caregivers.html.

Canadian Institute for Health Information (CIHI). (2013). End-of-Life Hospital Care for Cancer Patients. Retrieved from https://secure.cihi.ca/free_products/Cancer_Report_EN_web_April2013.pdf. Accessed on December 6, 2013.

Carlsson, ME. & Rollison, B. (2003). A comparison of patients dying at home and patients dying at a hospice: Sociodemographic factors and caregivers’ experiences. Palliative and Supportive Care, vol. 1, 33–39.

Cartoni, C., Brunetti, GA., D’Elia, GM., Breccia, M., Niscola, P., Marini, MG., Nastri, A., Alimena, G., Mandelli, F. & Foà, R. (2007). Cost analysis of a domiciliary program of supportive and palliative care for patients with hematologic malignancies. The Hematology Journal, vol. 92(5), 666-673.

Ceilleachair, O., Finn, C., Deady, S., Carsin, E. & Sharp, L. (2011). Have developments in palliative care services impacted on place of death of colorectal cancer patients in Ireland? A population-based study. Irish Journal of Medical Science, vol. 180, 91-96.

Chai, H., Guerriere, DN., Zagorski, B. and Coyte, PC. (2014). The magnitude, share and determinants of unpaid care costs for home-based palliative care service provision


106

in Toronto, Canada. Health and Social Care in the Community, vol. 22(1), 30-39.

Chai, H., Guerriere, D., Zagorski, B., Kennedy, J. & Coyte, P. (2013). The Size, Share, and Predictors of Publicly Financed Healthcare Costs in the Home Setting over the Palliative Care Trajectory. Journal of Palliative Care, vol. 29(3), 154-163.

Charlson, ME., Pompei, P., Ales, KL. & MacKenzie, CR. (1987). A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. Journal of Chronic Diseases, vol. 40(5), 373-383.

Church, J., Saunders, D., Wanke, M., Pong, R., Spooner, C. & Dorgan, M. (2002). Citizen Participation in Health Decision-Making: Past Experience and Future Prospects. Journal of Public Health Policy, vol. 23(1), 12-33.

Coates, V. (2012). Randomised controlled trials: still going for gold in 2012? Royal College of Nursing: UK. Retrieved from https://www.rcn.org.uk/__data/assets/pdf_file/0008/449135/Coates_keynote_research_2012.pdf.

Cochrane Collaboration. (2011). Cochrane Handbook for Systematic Reviews of Interventions (Version 5.1.0). Retrieved from http://handbook.cochrane.org.

College of Physicians and Surgeons of Ontario (CPSO). (2002). Decision-making for the End of Life (Policy Statement #1-06). Retrieved from http://www.cpso.on.ca/uploadedFiles/policies/policies/policyitems/End%20of%20Life.pdf.

Conde-Sala, JL., Garre-Olmo, J., Turro-Garriga, O., Vilalta-Franch, J. & Lopez-Pousa, S. (2010). Differential features of burden between spouse and adult-child caregivers of patients with Alzheimer’s disease: an exploratory comparative design. International Journal of Nursing Studies, 47(10), 1262-1273.

Cragg, John G. (1971). Some statistical models for limited dependent variables with application to the demand for durable goods. Econometrica, vol. 39(5), 829-844.

Cramér, H. (1946). Mathematical Methods of Statistics. Princeton University Press: United States.

DeMiglio, L. & Williams, A. (2012). Shared care: the barriers encountered by community-based palliative care teams in Ontario, Canada. Health and Social Care in the Community, vol. 20(4), 420-429.

DeMiglio, L. & Williams, A. (2012). Factors Enabling Shared Care with Primary Healthcare Providers in Community Settings: The experiences of interdisciplinary palliative care teams. Journal of Palliative Care, vol. 28(4), 282-290.

Dennis, K., Lawrence Librach, S. & Chow, E. (2011). Palliative Care Oncology Leads to Better Care. Oncology, vol. 25(13), 1271-1275.

Diwan, S., Hougham GW. & Sachs, GA. (2004). Strain Experienced by Caregivers of Dementia Patients Receiving Palliative Care: Findings from the Palliative


107

Excellence in Alzheimer Care Efforts (PEACE) Program. Journal of Palliative Medicine, vol. 7(6), 797-807.

Docherty, A., Owens, A., Asadi-Lari, M., Petchey, R., Williams, J. & Carter, Y. (2008). Knowledge and information needs of informal caregivers in palliative care: a qualitative systematic review. Palliative Medicine, vol. 22, 153-71.

Drummond, D., Giroux, D., Pigott, S. & Stephenson, C. (2012). Commission on the Reform of Ontario’s Public Services. Retrieved from http://www.fin.gov.on.ca/en/reformcommission/.

Duan, N. (1983). Smearing estimate: a nonparametric retransformation method. Journal of the American Statistical Association, vol. 78, 605-610.

Dumont, S., Jacobs, P., Turcotte, V., Anderson, D. & Harel, F. (2010). The trajectory of palliative care costs over the last 5 months of life: a Canadian longitudinal study. Palliative Medicine, vol. 24(6), 630-640.

Dumont, S., Jacobs, P., Fassbender, K., Anderson, D. Turcotte, V., & Harel F. (2009). Costs associated with resource utilization during the palliative phase of care: a Canadian perspective. Palliative Medicine, 23(8) 708–717, 10.1177/0269216309346546.

Ekström, J. (2008). An empirical polychoric correlation coefficient. Jan Wallander and Tom Hedelius Research Foundation, project P2008-0102:1.

Enguidanos, SM., Cherin, D. & Brumley, R. (2005). Home-Based Palliative Care Study: Site of Death, and Costs of Medical Care for Patients with Congestive Heart Failure, Chronic Obstructive Pulmonary Disease, and Cancer. Journal of Social Work in End-of-Life & Palliative Care, vol. 1(3), 37-55.

Ershler, W. (2003). Cancer: A Disease of the Elderly. Supportive Oncology, vol. 1(2), 5-10.

Everitt, BS. (2002). The Cambridge Dictionary of Statistics (2nd Ed.). Cambridge University Press: Cambridge, UK.

Fairfield, K., Murray, K., Wierman, H., Han,P., Hallen, S., Miesfeldt, S., Trimble, E., Warren, J., & Earle, C. (2012). Disparities in hospice care among older women dying with ovarian cancer. Gynecologic Oncology, vol. 125, 14-18.

Fauci, J., Schneider, K., Walters, C., Boone, J., Whitworth, J., Killian, E. & Straughn, JM. (2012). The utilization of palliative care in gynecologic oncology patients near the end of life. Gynecologic Oncology, vol. 127, 175-179.

Flechl, B., Ackerl, M., Sax, C., Oberndorfer, S., Calabek, B., Sizoo, E., Reijneveld, J., Crevenna, R., Keilani, M., Gaiger, A., Dieckman, K., Preusser, M., Taphoorn, M. & Marosi, C. (2013). The caregivers’ perspective on the end-of-life phase of glioblastoma patients. Journal of Neurooncology, vol. 112, 403-411.

Frees, E.W. (2010). Regression Modeling with Actuarial and Financial Applications.


108

Cambridge University Press: Cambridge, UK.

Fukahori, H., Yamamoto-Mitani, N., Sugiyama, T., Sugai, T. & Kai, I. (2010). Psychometric properties of the Caregiver Burden Scale for Family Caregivers with Relatives in Nursing Homes: Scale development. Japan Journal of Nursing Science, vol. 7, 136-147.

Funk, L., Stajduhar, K., Toye, C., Aoun, S., Grande, G. & Todd, C. (2010). Part 2: Home-based family caregiving at the end of life: a comprehensive review of published qualitative research (1998-2008). Palliative Medicine, vol. 24(6), 594-607.

Gaertner, J., Wolf, J., Hallek, M., Glossmann, J. & Voltz R. (2011). Standardizing integration of palliative care into comprehensive cancer therapy—a disease specific approach. Support Care Cancer, vol. 19, 1037-1043.

Wang, G., Cheng, Q., Wang, Y., Deng, Y., Ren, R., Xu, W., Zeng, J. & Chen, S. (2008). The Metric Properties of Zarit Caregiver Burden Scale: Validation Study of a Chinese Version. Alzheimer Disease & Associated Disorders, vol. 22(4), 321-326.

Gardiner, C., Brereton, L., Frey, R., Wilkinson-Meyers, L. & Gott, M. (2014). Exploring the financial impact of caring for family members receiving palliative and end-of-life care: A systematic review of the literature. Palliative Medicine, vol. 28(5), 375-390.

Gibbons, R. & Hedeker, D. (1994). Application of Random-Effects Probit Regression Models. Journal of Consulting and Clinical Psychology, vol. 62(6), 285-296.

Given, B. & Given, C. (1991). Family Caregivers of Cancer Patients. In Hubbard, SM., Greene, P. & Knobs, MT (Eds.), Current Issues in Cancer Nursing Practice (pp. 1-9). Philadelphia, USA: J.B. Lippincott Co.

Given, B., Wyatt, G. & Rahbar, M. (2004). Burden and Depression Among Caregivers of Patients with Cancer at the End-of-Life. Oncology Nursing Forum, vol. 31(6), 1105-1117.

Glajchen, M. (2012). Physical Well-Being of Oncology Caregivers: An Important Quality-of-Life Domain. Seminars in Oncology Nursing, vol 28(4), 226-235.

Götze, H., Brähler, E., Gansera, L., Polze, N. & Köhler, N. (2014). Psychological distress and quality of life of palliative cancer patients and their caring relatives during home care. Support Care Cancer, DOI: 10.1007/s00520-014-2257-5.

Greene, W. (2002). The Bias of the Fixed Effects Estimator in Nonlinear Models. Department of Economics, Stern School of Business, New York University: New York, USA.

Greffe, X. (2007). The Role of the Social Economy in Local Development. In OECD. (2007). Social Economy: Building Inclusive Economies (pp. 91-117).

Grov, EK., Fossa, SD., Sorebo, O. & Dahl, AA. (2006). Primary caregivers of cancer patients in the palliative phase: A path analysis of variables influencing their burden. Social Science & Medicine, vol. 63, 2429-2439.


109

Guerriere, DN. (2012). Predictors of Place of Death and Family Caregiver Burden Across the Home-Based Palliative Care Trajectory. Proposal.

Guerriere, DN. & Coyte, PC. (2011). The Ambulatory Home Care Record: A Methodological Framework for Economic Analyses in End-of-Life Care. Journal of Aging Research, vol. 2011, http://dx.doi.org/10.4061/2011/374237.

Guerriere, DN., Zagorski, B. & Coyte, PC. (2013). Family caregiver satisfaction with home-based nursing and physician care over the palliative care trajectory: Results from a longitudinal survey questionnaire. Palliative Medicine, vol. 27(7) 632–638.

Guerriere, DN., Ungar, WJ., Corey, M., Croxford, R., Tranmer, JE., Tullis, E. & Coyte, PC. (2006). Evaluation of the ambulatory home care record: agreement between self-reports and administrative data. International Journal of Technology Assessment in Health Care, vol. 22(2), 203-210.

Guerriere, DN., Zagorski, B., Fassbender, K., Masucci, L., Librach, L. & Coyte, PC. (2010). Cost variations in ambulatory and home-based palliative care. Palliative Medicine, vol. 24(5), 523-532.

Haley, W. (2003). Family Caregivers of Elderly Patients With Cancer: Understanding and Minimizing the Burden of Care. Supportive Oncology, vol. 1(2), 25-29.

Hayman, J., Langa, K., Kabeto, M., Katz, S., DeManner, S., Chemew, M., Slavin, M. & Fendrick, M. (2001). Estimating the Cost of Informal Caregiving for Elderly Patients with Cancer. Journal of Clinical Oncology, vol. 19(13), 3219-3225.

Health Canada. (2009). Palliative and End-of-Life Care. Retrieved from http://www.hc-sc.gc.ca/hcs-sss/palliat/index-eng.php.

Health Council of Canada. (2013). Progress Report 2013: Health care renewal in Canada. Retrieved from http://www.healthcouncilcanada.ca/rpt_det.php?id=481.

Health Quality Ontario (HQO). (2012). Report on Ontario’s Health System. Retrieved from http://www.hqontario.ca/portals/0/Documents/pr/qmonitor-full-report-2012-en.pdf.

Hérbert, R., Bravo, G., & Préville, M. (2000). Reliability, validity, and reference values of the Zarit Burden Interview for assessing informal caregivers of community-dwelling older persons with dementia. Canadian Journal on Aging, vol. 19, 494-507.

Hill, R., Griffiths, W. & Lim, G. (2011). Principles of Econometrics (4th Ed.). John Wiley & Sons, Inc: Hoboken, New Jersey, USA.

Hirdes, J.P., Freeman, S., Frise Smith, T., & Stolee, P. (2012). Predictors of caregiver distress among palliative home care clients in Ontario: Evidence based on the interRAI Palliative Care. Palliative and Supportive Care, vol. 10, 155-163.

Hopewell, S., Loudon, K., Clarke, MJ., Oxman, AD. & Dickersin, K. (2009). Publication bias in clinical trials due to statistical significance or direction of trial results.


110

Cochrane Database of Systematic Reviews, vol. 21(1), 10.1002/14651858.MR000006.pub3.

Hu, F., Goldberg, J., Hedeker, D., Flay, B. & Pentz, M. (1998). Comparison of Population-Averaged and Subject-Specific Approaches for Analyzing Repeated Binary Outcomes. American Journal of Epidemiology, vol. 147(7), 694-703.

Hudson, PL., Hayman-White, K., Aranda, S. & Kristjanson, LJ. (2006). Predicting Family Caregiver Psychosocial Functioning in Palliative Care. Journal of Palliative Care, vol. 22(3), 133-140.

Hume, T. & Christensen, J. (2014). WHO: Imminent global cancer ‘disaster’ reflects aging, lifestyle factors. Retrieved from http://www.cnn.com/2014/02/04/health/who-world-cancer-report/.

Iecovich, E. (2012). Psychometric properties of the Hebrew version of the Zarit Caregiver Burden Scale short version. Age & Mental Health, vol. 16(2), 254-263.

Irwin, K., Greer, J., Khatib, J., Temel, J. & Pirl, W. (2012). Early palliative care and metastatic non-small cell lung cancer: Potential mechanisms of prolonged survival. Chronic Respiratory Disease, vol. 10(1), 35-47.

Jacobs, P., Dumont, S., Turcotte, V. & Anderson, D. (2011). Evaluating the Economic Loss of Caregiving for Palliative Care Patients. Journal of Palliative Care, vol. 27(3), 210-215.

Jiwani, I. (2003). Transformation of Health Organizations and Rhetoric of Accountability in Community-Based Health Care Practices in Ontario. Dissertation, York University: Toronto, ON.

Kamal, A., Nipp, R., Bull, J. & Abernethy, A. (2013). Towards Achieving Personalized Palliative Care: Predicting Symptom Burden in Community-Dwelling Patients With Advanced Ilnness (SA527-B). Journal of Pain and Symptom Management, vol. 45(2), 411-412.

Kao, H. & Acton, G. (2006). Conceptualization and psychometric properties of the caregiver burden scale in Taiwan. Issues in Mental Health Nursing, vol. 27, 853-866.

Kenny, PM., Hall, JP., Zapart, S. & Davis, PR. (2010). Informal Care and Home-Based Palliative Care: The Health-Related Quality of Life of Carers. Journal of Pain and Symptom Management, vol. 40(1), 35-48.

King, M. (2002). Analysis of Case Management as A Service Delivery Model in Ontario’s Home Care Sector. Dissertation, University of Toronto: Toronto, ON.

Kirkey, S. (May 20, 2010). End-of-life care an uncertain ‘patchwork’; Canadian cancer society report; Agency calls for a national caregiver strategy and access to high-quality palliative programs. The Gazette; Montreal, QUE.

Klinger, C., Howell, D., Marshall, D., Zakus, D., Brazil, K. & Deber, R. (2011). Resource


111

utilization and cost analyses of home-based palliative care service provision: The Niagara West End-of-Life Shared-Care Project. Palliative Medicine, vol. 27(2) 115-122.

Kuluski, K. (2010). Homecare or Long-term Care? The Balance of Care in Urban and Rural Northwestern Ontario. Dissertation, University of Toronto: Toronto, ON.

Janjan, N. (2011). Palliative Care: Meaningful Benefit in Oncology Care. Oncology, vol. 25(13), 1244-1256.

Lai, DWL. (2007). Validation of the Zarit Burden Interview for Chinese Canadian caregivers. Social Work Research, vol. 31, 45-53.

Laporte, A. & Rudoler, D. (2013). Assessing Ontario’s Personal Support Worker Registry. Health Reform Observer – Observatoire des Réformes de Santé, vol. 1(1), Article 2.

Lavergne, MR., Johnston, GM., Gao, J., Dumont, S. & Burge, FI. (2011). Exploring Generalizability in a Study of Costs for Community-Based Palliative Care. Journal of Pain and Symptom Management, vol. 41(4), 779-787.

Laville, J., Levesque, B. & Mendell, M. (2007). Diverse Approaches and Practices in Europe and Canada. In OECD. (2007). Social Economy: Building Inclusive Economies (pp. 155-81).

Lester, JF., Macbeth, F., Toy, E. & Coles, B. (2012). Palliative radiotherapy regimens for non-small cell lung cancer. Cochrane Database of Systematic Reviews, Issue 4. Art. No.: CD002143. DOI: 10.1002/14651858.CD002143.pub2.

Liu, L., Cowen, ME., Strawderman, RL. & Shih, YT. (2011). A Flexible Two-Part Random Effects Model for Correlated Medical Costs. Journal of Health Economics, vol. 29(1), 110–123.

Manitta, VJ., Philip, JAM. & Cole-Sinclair, MF. (2010). Palliative Care and the Hemato-Oncological Patient: Can We Live Together? A Review of the Literature. Journal of Palliative Medicine, vol. 13(8), 1021-1025.

Manning, WG. (1998). The logged dependent variable, heteroscedasticity, and the retransformation problem. Journal of Health Economics, vol. 17, 283-295.

Manning, WG., Basu, A. & Mullahy, J. (2005).Generalized modeling approaches to risk adjustment of skewed outcomes data. Journal of Health Economics, vol. 24(3), 465-488.

Manning, WG. & Mullahy, J. (2001). Estimating log models: to transform or not to transform? Journal of Health Economics, vol. 20(4), 461-494.

Masucci, L. (2011). Health Care Service Provision Over the Palliative Care Trajectory. Thesis, University of Toronto: Toronto, ON.

McGrath, PD., & Holewa, HA. (2007). Description of an Australian Model for End-of-Life


112

Care in Patients With Hematologic Malignancies. Oncology Nursing Forum, vol. 34(1), 79-85.

Mora, T., Gil, J. & Sicras-Mainar, A. (2013). The influence of BMI, obesity and overweight on medical costs: a panel data perspective. Health, Econometrics and Data Group: University of York, UK. Retrieved from york.ac.uk/res/herc/hedgwp.

Motiwala, S., Croxford, R., Guerriere, D. & Coyte, P. (2006). Predictors of Place of Death for Seniors in Ontario: A Population-Based Cohort Analysis. Canadian Journal on Aging, vol. 25(4), 363-371.

Mullahy, J. (1998). Much ado about two: reconsidering retransformation and the two-part model in health econometrics. Journal of Health Economics, vol. 17:247–281.

Nash, M., Cunha, J., Mueller, L., Edwards, B. & Jackson, E. (October 10, 2013). The way forward: Ontario is on the cusp of a revolution in the way the province offers palliative care, but the plan hinges on government support. Kanata EMC: Kanata, ON.

Neelon, B. & O’Malley, AJ. (N/A). Two-Part Models for Zero-Modified Count and Semicontinuous Data. Department of Biostatistics and Bioinformatics, Duke University School of Medicine: Durhan, North Carolina, USA.

Neuhaus, JM. (1992). Statistical methods for longitudinal and clustered designs with binary responses. Statistical Methods in Medical Research, vol. 1, 249-273.

Neuhaus, J. M., J. D. Kalbfleisch, and W. W. Hauck. 1991. A comparison of cluster-specific and population-averaged approaches for analyzing correlated binary data. International Statistical Review 59: 25–35.

Neumann, P. (2009). Costing and perspective in published cost-effectiveness analysis. Medical Care, vol. 47(7), S28-32.

Nymphas, L. & Wu, S. (2014). Measurement structure of the caregiver burden scale: Findings from a national community survey in Taiwan. Geriatrics Gerontoloty, vol. 14, 176-184.

Ogilvie, D. (1998). Palliative care in Canada facing a real crisis: only 10% of patients who require home care are getting it. Medical Post, vol. 34(18), 5.

Olsen, MK. & Schafer, JL. (2001) A two-part random effects model for semicontinuous longitudinal data. Journal of the American Statistical Association 2001;96:730–745.

Ontario Senior’s Secretariat, The. (2013). Ontario’s Action Plan for Seniors. Retrieved from https://dr6j45jk9xcmk.cloudfront.net/documents/215/ontarioseniorsactionplan-en-20130204.pdf.

Ostgathe, C., Walshe, R., Wolf, J., Hallek, M. & Voltz, R. (2008). A cost calculation model for specialist palliative care for patients with non-small cell lung cancer in a tertiary centre. Support Care Cancer, vol.16, 501–506.


113

Pallett, P. (1990). Conceptual Framework for Studying Family Caregiver Burden in Alzheimer’s-Type Dementia. Journal of Nursing Scholarship, vol. 22(1), 52-58.

Park, RE. (1966). Estimation with Heteroscedastic Error Terms. Econometrica, vol. 34(4), 888.

Parker Oliver, D., Wittenberg-Lyles, E., Washington, K., Kruse, R., Albright, D., Baldwin, P., Boxer, A. & Demiris, G. (2013). Hospice Caregivers’ Experiences With Pain Management: “I’m Not a Doctor, and I Don’t Know if I Helped Her Go Faster or Slower”. Journal of Pain and Symptom Management, vol. 46(6), 846-859.

Pearson, K. (1900). Mathematical contributions to the theory of evolution: On the correlation of characters not quantitatively measurable. Philosophical Transactions of the Royal Society of Mathematical, Physical & Engineering Sciences, vol. 195, 1-47.

Pearson, K. (1907). Mathematical contributions to the theory of evolution: On further methods of determining correlation. Drapers’ Company Research Memoirs, Biometric Series (Volume 4). London: Cambridge University Press.

Pearson, K. & Heron, D. (1913). On theories of association. Biometrika, vol. 9, 159-315.

Preferred Reporting Items for Systematic Reviews (PRISMA). (2014). Transparent Reporting of Systematic Reviews and Meta-Analysis. Retrieved from http://www.prisma-statement.org/statement.htm.

Ramchandran, K. & Von Roenn, J. (2013). Palliative Care Always. Oncology, Review Article. Retrieved from http://www.cancernetwork.com/palliative-and-supportive-care/palliative-care-always.

Rangachari, D. & Smith, TJ. (2013). Integrating Palliative Care in Oncology The Oncologist as a Primary Palliative Care Provider. The Cancer Journal, vol. 19(5), 373-378.

Rice, DM. (1989). Cost of injury in the United States. University of California, Institute for Health and Aging: San Francisco, CA.

Rocque, GB. & Cleary, JF. (2013). Palliative care reduces morbidity and mortality in cancer. Nature Reviews Clinical Oncology, vol. 10, 80-89.

Rodby, KA., Turin, S., Jacobs, RJ., Cruz, JF., Hassid, VJ., Kolokythas, A. & Antony, AK. (2014). Advances in oncologic head and neck reconstruction: Systematic review and future considerations of virtual surgical planning and computer aided design/ computer aided modeling. Journal of Plastic, Reconstructive & Aesthetic Surgery, http://dx.doi.org/10.1016/j.bjps.2014.04.038.

Romanow, R. (2002). Building on values: the future of health care in Canada. Final report of the Commission on the Future of Health Care in Canada. Saskatoon: Commission on the Future of Health Care in Canada.


114

Ryn, M., Sanders, S., Kahn, K., van Houtven, C., Griffin, J., Martin, M., Atienza, A., Phelan, S., Finstad, D. & Rowland, J. (2011). Objective burden, resources, and other stressors among informal cancer caregivers: a hidden quality issue? Pyscho-Oncology, vol. 20, 44-52.

Sanyal, P. & Babu, S. (2009). Food Security, Poverty and Nutrition Policy Analysis: Statistical Methods and Applications (1st Ed.). Academic Press, Elsevier: London, UK & Amsterdam, NL.

Schuklenk, U., van Delden, J., Downie, J., McLean, S., Upshur, R. & Weinstock, D. (2011). The Royal Society of Canada Expert Panel: End-of-Life Decision Making. The Academies of Arts, Humanities and Sciences of Canada. Accessed November 2013. Retrievable from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3265521/pdf/bioe0025-0001.pdf.

Seow, H. (2009). The Use of End-of-Life Homecare Services in Ontario, Canada: Is it associated with using less acute care services in late life? Dissertation, John Hopkins University: Baltimore, USA.

Shadd, J. (2008). Should palliative care be a specialty?: yes. Canadian Family Physician, vol. 54(6), 840-846.

Sheehy-Skeffington, B., McLean, S., Bramwell, M., O’Leary, N. & O’Gorman, A. (2013). Caregivers Experiences of Managing Medications for Palliative Care Patients at the End of Life: A Qualitative Study. American Journal of Hospice & Palliative Medicine, vol. 31(2), 148-154.

Shephard, T. (May 15, 2004). Etobicoke MPP calls for palliative care standards, policies. Etobicoke Guardian: Etobicoke, ON.

Shnoor, Y., Szlaifer, M., Aoberman, AS. & Bentur, N. (2007). The Cost of Home Hospice Care for Terminal Patients in Israel. American Journal of Hospice & Palliative Medicine, vol. 24(4), 284-290.

Simoens, S., Kutten, B., Keirse, E., Vanden Berghe, P., Beguin, C., Desmedt, M., Deveugele, M., Leonard, C., Paulus, D. & Menten, J. (2010). The Costs of Treating Terminal Patients. Journal of Pain and Symptom Management, vol. 40(3), 436-448.

Sims, A., Radford, J., Doran, K. & Page, H. (1997). Social class variation in place of cancer death. Palliative Medicine, vol. 11, 369-373.

Singer, Y., Bachner, YG., Shvartzman, P. & Carmel, S. (2005). Home Death—The Caregivers’ Experiences. Journal of Pain and Symptom Management, vol. 30(1), 70-74.

Skinner, M. (2005). Voluntarism and Long-Term Care in the Countryside: Exploring the Implications of Health Care Restructuring for Voluntary Sector Providers in Rural Ontario. Dissertation, Queen’s University: Kingston, ON.

Spalding, K. (2005). Policy by Default: How changes in Ontario’s home care sector have


115

impacted providers, children and families. Dissertation, University of Toronto: Toronto, ON.

Stajduhar, K., Funk, L., Toye, C., Grande, G., Aoun, S. & Todd, C. (2010). Part 1: Home-based family caregiving at the end of life: a comprehensive review of published quantitative research (1998-2008). Palliative Medicine, vol. 24(6), 573-593.

StataCorp. 2013. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP.

Sundararajan, V., Henderson, R., Perry, C., Muggivan, A., Quan, H. & Ghali, W. (2004). New ICD-10 version o the Charlson Comorbidity Index predicted in-hospital mortality. Journal of Clinical Oncology, vol. 57, 1288-1294.

Sussman, J., Barbera, L., Howell, D., Yang, J., Husain, A., Librach, S., Viola, R. & Walker, H. (2011). Health system characteristics of quality care delivery: A comparative case study examination of palliative care for cancer patients in four regions in Ontario, Canada. Palliative Medicine, vol. 26(4), 322-335.

Temel, J., Greer, J., Muzikansky, A., Gallagher, E., Admane, D., Jackson, V., Dahlin, C., Blinderman, C., Jacobsen, J., Pirl, W., Billings, J. & Lynch, T. (2010). Early Palliative Care for Patients with Metastatic Non-Small-Cell Lung Cancer. The New England Journal of Medicine, vol. 363(8), 733-744

Temmy Latner Centre for Palliative Care. (2013). What is palliative care? Mount Sinai Hospital: Toronto, ON. Retrieved from http://www.tlcpc.org/palliative-care.

Torres-Reyna, O. (2007). Panel Data Analysis: Fixed & Random Effects. Data & Statistical Services, Princeton University: New Jersey, USA. Retrieved from http://www.princeton.edu/~otorres/Panel101.pdf.

Tzala, S., Lord, J., Ziras, N., Repousis, P., Potamianou, A. & Tzala, E. (2005). Cost of home palliative care compared with conventional hospital care for patients with haematological cancers in Greece. European Journal of Health Economics, vol. 6, 102-106.

Unroe, K. & Meier, D. (2013). Research Priorities in Geriatric Palliative Care: Policy Initiatives. Journal of Palliative Medicine, vol. 16(12),

Von Roenn, J. & Temel, J. (2011). The Integration of Palliative Care and Oncology: the Evidence. Oncology, vol. 25(13), 1258-1266.

Ward-Griffin, C., McWilliam, C. & Oudshoorn, A. (2012). Relational Experiences of Family Caregivers Providing Home-Based End-of-Life Care. Journal of Family Nursing, vol. 18(4), 491-516.

Wasner, M., Paal, P. & Borasio, G.D. (2013). Psychosocial Care for the Caregivers of Primary Malignant Brain Tumor Patients. Journal of Social Work in End-of-Life & Palliative Care, vol. 9(1), 74-95.


116

Weinstein, MC., Siegel, JE., Gold, MG., Kamlet, MS. & Russel, LB. (1996). Recommendations of the panel of cost-effectiveness in health and medicine. Journal of the American Medical Association, vol. 276(15), 1253-1258.

Williams, AM., Wang, L. & Kitchen, P. (2014). Differential impacts of caregiving across three caregiver groups in Canada: end-of-life care, long-term care and short-term care. Health and Social Care in the Community, vol. 22(2), 187-196.

Wong, RCC., Tan, PT., Seow, YH., Aziz, S., Oo, N., Seow, SC., Seah, A., & Chai, P. (2013). Home-Based Advance Care Programme is Effective in Reducing Hospitalizations of Advanced Heart Failure Patients: A Clinical and Healthcare Cost Study. Annals Academy of Medicine Singapore, vol. 42, 466-471.

Wooldridge, J. (2012). Introductory Econometrics: A Modern Approach (5th Ed.). South-Western College Publishing: Boston, Massachusetts, USA.

World Health Organization. (2014). WHO Definition of Palliative Care. Retrieved from http://www.who.int/cancer/palliative/definition/en/.

Yalnizyan, A. (2007). Ontario’s Growing Gap. Canadian Centre for Policy Alternatives. Toronto, ON.

Yu, J. (2011). Home Care Utilization Patterns Among the Elderly Population: a Case Study of Ontario. Thesis, Queen’s University: Kingston, ON.

Zarit, SH., Reever, KE. & Bach-Peterson, J. (1980). Relatives of the Impaired Elderly: Correlates of Feelings of Burden. The Gerontologist, vol. 20(6), 649-655.

117

Appendix A Critical Appraisal Skills Program (CASP) tools4

TOOL 1: 12 questions to help you make sense of cohort study5 Three broad issues need to be considered when appraising a cohort study: • Are the results of the study valid? (A) • What are the results? (B) • Will the results help locally? (C)

(A) Are the results of the study valid? Screening questions 1. Did the study address a clearly focused issue? ☐Yes ☐Can’t tell ☐No HINT: A question can be ‘focused’ In terms of • The population studied • The risk factors studied • The outcomes considered • Is it clear whether the study tried to detect a beneficial or harmful effect? 2. Was the cohort recruited in an acceptable way? ☐Yes ☐Can’t tell ☐No HINT: Look for selection bias which might compromise the generalizability of the findings: • Was the cohort representative of a defined population? • Was there something special about the cohort? • Was everybody included who should have been included? Is it worth continuing? Detailed questions 3. Was the exposure accurately measured to minimize bias? ☐Yes ☐Can’t tell ☐No HINT: Look for measurement or classification bias: • Did they use subjective or objective measurements? • Do the measurements truly reflect what you want them to (have they been validated)? • Were all the subjects classified into exposure groups using the same procedure? 4. Was the outcome accurately measured to minimize bias? ☐Yes ☐Can’t tell ☐No HINT: Look for measurement or classification bias: • Did they use subjective or objective measurements? • Do the measures truly reflect what you want them to (have they been validated)? • Has a reliable system been established for detecting all the cases (for measuring disease occurrence)? • Were the measurement methods similar in the different groups? • Were the subjects and/or the outcome assessor blinded to exposure (does this matter)? 4 This follows the questionnaire verbatim (including punctuation), but spelling has been changed to Canadian where applicable. 5 The succeeding tools will only include questions distinct from Tool 1a.

Appendices R. Redmond-Misner

118

5. (a) Have the authors identified all important confounding factors? ☐Yes ☐Can’t tell ☐No List the ones you think might be important, that the author missed. 5. (b) Have they taken account of the confounding factors in the design and/or analysis? ☐Yes ☐Can’t tell ☐No List: HINT: Look for restriction in design, and techniques e.g. modeling, stratified-, regression-, or sensitivity analysis to correct, control or adjust for confounding factors 6. (a) Was the follow up of subjects complete enough? ☐Yes ☐Can’t tell ☐No 6. (b) Was the follow up of subjects long enough ☐Yes ☐Can’t tell ☐No HINT: Consider • The good or bad effects should have had long enough to reveal themselves • The persons that are lost to follow-up may have different outcomes than those available for assessment • In an open or dynamic cohort, was there anything special about the outcome of the people leaving, or the exposure of the people entering the cohort?

(B) What are the results? 7. What are the results of this study? HINT: Consider • What are the bottom line results? • Have they reported the rate or the proportion between the exposed/unexposed, the ratio/the rate difference? • How strong is the association between exposure and outcome? • What is the absolute risk reduction (ARR)? 8. How precise are the results? HINT: Look for the range of the confidence intervals, if given. 9. Do you believe the results? ☐Yes ☐Can’t tell ☐No HINT: Consider • Big effect is hard to ignore! • Can it be due to bias, chance or confounding? • Are the design and methods of this study sufficiently flawed to make the results unreliable? • Bradford Hills criteria (e.g. time sequence, dose-response gradient, biological plausibility, consistency)

(C) Will the results help locally? 10. Can the results be applied to the local population? ☐Yes ☐Can’t tell ☐No HINT: Consider whether • A cohort study was the appropriate method to answer this question • The subjects covered in this study could be sufficiently different from your population to cause concern • Your local setting is likely to differ much from that of the study • You can quantify the local benefits and harms


119

11. Do the results of this study fit with other available evidence? ☐Yes ☐Can’t tell ☐No 12. What are the implications of this study for practice? HINT: Consider • One observational study rarely provides sufficiently robust evidence to recommend changes to clinical practice or within health policy decision making • For certain questions observational studies provide the only evidence • Recommendations from observational studies are always stronger when supported by other evidence

TOOL 2: 11 questions to help you make sense of case control study

(A) Are the results of the study valid? Screening questions 1. Same as Tool 1. 2. Did the authors use an appropriate method to answer their question? ☐Yes ☐Can’t tell ☐No HINT: Consider • Is a case control study an appropriate way of answering the question under the circumstances? (Is the outcome rare or harmful) • Did it address the study question? Is it worth continuing? Detailed questions 3. Were the cases recruited in an acceptable way? ☐Yes ☐Can’t tell ☐No HINT: We are looking for selection bias which might compromise validity of the findings • Are the cases defined precisely? • Were the cases representative of a defined population? (geographically and/or temporally?) • Was there an established reliable system for selecting all the cases • Are they incident or prevalent? • Is there something special about the cases? • Is the time frame of the study relevant to disease/exposure? • Was there a sufficient number of cases selected? • Was there a power calculation? 4. Were the controls selected in an acceptable way? ☐Yes ☐Can’t tell ☐No HINT: We are looking for selection bias which might compromise the generalizability of the findings • Were the controls representative of defined population (geographically and/or temporally) • Was there something special about the controls? • Was the non-response high? Could non-respondents be different in any way? • Are they matched, population based or randomly selected? • Was there a sufficient number of controls selected?


120

5. Was the exposure accurately measured to minimize bias? ☐Yes ☐Can’t tell ☐No HINT: We are looking for measurement, recall or classification bias • Was the exposure clearly defined and accurately measured? • Did the authors use subjective or objective measurements? • Do the measures truly reflect what they are supposed to measure? (Have they been validated?) • Were the measurement methods similar in the cases and controls? • Did the study incorporate blinding where feasible? • Is the temporal relation correct? (Does the exposure of interest precede the outcome?) 6. (a) What confounding factors have the authors accounted for? List: HINT: List the ones you think might be important, that the author missed. • Genetic • Environmental • Socio-economic 6. (b) Same as question 5. (b) in Tool 1.

(B) What are the results? 7-9. Same as Tool 1.

(C) Will the results help locally? 10-11. Same as Tool 1.

TOOL 3: 11 questions to help you make sense of qualitative studies

Screening questions 1. Was there a clear statement of the aims of this research? ☐Yes☐Can’t tell ☐No HINT: Consider • What was the goal of the research? • Why it was thought important? • Its relevance 2. Is a qualitative methodology appropriate? ☐Yes ☐Can’t tell ☐No HINT: Consider • If the research seeks to interpret or illuminate the actions and/or subjective experiences of research participants • Is qualitative research the right methodology for addressing the research goal? Is it worth continuing? Detailed questions 3. Was the research design appropriate to address the aims of the research? ☐Yes ☐Can’t tell ☐No HINT: Consider • If the researcher has justified the research design (e.g. have they discussed how they decided which method to use)?


121

4. Was the recruitment strategy appropriate to the aims of the research? ☐Yes ☐Can’t tell ☐No HINT: Consider • If the researcher has explained how the participants were selected • If they explained why the participants they selected were the most appropriate to provide access to the type of knowledge sought by the study • If there are any discussions around recruitment (e.g. why some people chose not to take part) 5. Was the data collected in a way that addressed the research issue? ☐Yes ☐Can’t tell ☐No HINT: Consider • If the setting for data collection was justified • If it is clear how data were collected (e.g. focus group, semi-structured interview etc.) • If the researcher has justified the methods chosen • If the researcher has made the methods explicit (e.g. for interview method, is there an indication of how interviews were conducted, or did they use a topic guide)? • If methods were modified during the study. If so, has the researcher explained how and why? • If the form of data is clear (e.g. tape recordings, video material, notes etc) • If the researcher has discussed saturation of data 6. Has the relationship between researcher and participants been adequately considered? ☐Yes ☐Can’t tell ☐No HINT: Consider • If the researcher critically examined their own role, potential bias and influence during (a) Formulation of the research questions (b) Data collection, including sample recruitment and choice of location • How the researcher responded to events during the study and whether they considered the implications of any changes in the research design 7. Have ethical issues been taken into consideration? ☐Yes ☐Can’t tell ☐No HINT: Consider • If there are sufficient details of how the research was explained to participants for the reader to assess whether ethical standards were maintained • If the researcher has discussed issues raised by the study (e.g. issues around informed consent or confidentiality or how they have handled the effects of the study on the participants during and after the study) • If approval has been sought from the ethics committee 8. Was the data analysis sufficiently rigorous? ☐Yes ☐Can’t tell ☐No HINT: Consider • If there is an in-depth description of the analysis process • If thematic analysis is used. If so, is it clear how the categories/themes were derived from the data? • Whether the researcher explains how the data presented were selected from the original sample to demonstrate the analysis process


122

• If sufficient data are presented to support the findings • To what extent contradictory data are taken into account • Whether the researcher critically examined their own role, potential bias and influence during analysis and selection of data for presentation 9. Is there a clear statement of findings? ☐Yes ☐Can’t tell ☐No HINT: Consider • If the findings are explicit • If there is adequate discussion of the evidence both for and against the researchers arguments • If the researcher has discussed the credibility of their findings (e.g. triangulation, respondent validation, more than one analyst) • If the findings are discussed in relation to the original research question 10. How valuable is the research? HINT: Consider • If the researcher discusses the contribution the study makes to existing knowledge or understanding e.g. do they consider the findings in relation to current practice or policy?, or relevant research-based literature? • If they identify new areas where research is necessary • If the researchers have discussed whether or how the findings can be transferred to other populations or considered other ways the research may be used Additional tools and full questionnaires are available at http://www.casp-uk.net/#!casp-tools-checklists/c18f8.


123

Appendix B Detailed characteristics of home care cost papers

Wea

knes

ses

Sm

all s

ampl

e.

Gen

eral

izab

ility

. D

oes

not c

alcu

late

ac

tual

eco

nom

ic

impa

ct.

Doe

s no

t acc

ount

fo

r pot

entia

l ext

ra

cost

of

hom

e

visi

ts.

1 di

seas

e gr

oup.

No

cont

rol g

roup

. H

ospi

tal

pers

pect

-iv

e on

ly.

Sm

all s

ampl

e.

Des

pite

resu

lt, h

ome

care

stil

l hig

hly

pr

efer

red

by

mos

t pa

tient

s.

Onl

y un

til 2

009.

Stu

dies

per

tain

-

ing

to h

ome

care

ar

e ou

tdat

ed.

Stu

dies

per

tain

ing

to

hom

e ca

re a

re

inte

rnat

iona

l.

Res

ults

71%

exp

erie

nced

redu

ced

hosp

italiz

atio

n.

5% h

ad m

ore

frequ

ent

adm

issi

ons.

95

% h

ad n

o w

orse

ning

.

Adj

uste

d fo

r ear

ly d

eath

(e

arly

dea

th =

no/

few

ho

spita

lizat

ions

).

Tran

sdus

ions

con

sist

ent i

n

bo

th g

roup

s, b

ut h

ome

gr

oup

had

to b

e re

gula

rly

mon

itore

d

for f

ull b

lood

co

unt a

nd b

lood

cro

ss-te

sts

in

orde

r to

proc

eed

at h

ome,

w

hich

mad

e th

em

mor

e ex

pens

ive

(hos

pita

l gro

up

only

test

ed a

t adm

issi

on

an

d w

hen

nece

ssar

y

ther

eafte

r).

1998

Spa

in d

ata

sugg

este

d

hom

e ca

re c

heap

er (v

ia

hosp

italiz

atio

n re

duct

ion)

.

Isra

eli s

tudy

sho

wed

that

sp

ecia

lized

hom

e ca

re w

as

30%

che

aper

than

non

-sp

ecia

lized

.

Foun

d th

at h

ome

care

cos

t de

pend

ed o

n th

e di

seas

e

of

th

e pa

tient

(i.e

. ha

emat

olog

ic p

atie

nts

and

trans

fusi

ons)

.

Out

com

e

Hos

pita

lizat

ions

Cos

t (H

ospi

tal

pers

pect

ive)

Con

clus

ions

abo

ut th

e co

st o

f pal

liativ

e ca

re in

va

rious

set

tings

(N/A

)

Met

hods

Dat

a co

llect

ion

from

regi

stry

of

end-

stag

e H

F pa

tient

s re

crui

ted

into

pal

liativ

e ca

re

prog

ram

from

200

8-10

. C

olle

cted

d

emog

raph

ics,

m

edic

atio

ns, l

ab re

sults

from

H

F da

taba

se e

lect

roni

c ho

spita

l rec

ords

.

Sta

tistic

al a

naly

sis

of h

ospi

tal

adm

issi

ons

befo

re a

nd a

fter

enro

lmen

t in

palli

ativ

e ca

re.

Pat

ient

recr

uit J

an-J

une

2002

. E

stim

ated

cos

t per

pat

ient

ex

clud

ing

trans

fusi

ons

and

drug

s th

at w

ere

com

mon

to

both

. Dat

a co

llect

ed fr

om

hosp

ital r

ecor

ds in

fina

nce

depa

rtmen

t. H

ospi

tal s

chem

e in

clud

ed h

otel

cos

t, ut

ilitie

s,

laun

dry,

food

, med

ical

sta

ff,

lab

test

s et

c. b

ased

on

bed

days

. Hom

e sc

hem

e in

clud

es

1-da

y cl

inic

, nur

se v

isits

, lab

te

sts.

Sys

tem

atic

revi

ew o

f

Pub

Med

, Coc

hran

e, E

conL

it an

d ot

hers

. App

rais

e m

etho

ds.

Syn

thes

ize

resu

lts. A

rticl

es

publ

ishe

d 20

00-0

9.

Exc

lude

d st

udie

s th

at

quan

tifie

d ut

iliza

tion

but d

id n

ot

conv

ert i

t to

cost

s.

Focu

s

Adv

ance

d he

art

failu

re p

atie

nts

with

1-

year

exp

ecte

d

surv

ival

; doe

s ho

me-

base

d pa

lliat

ive

care

re

duce

ho

spita

lizat

ion

and

thus

cos

ts?

EoL

hae

mat

olog

ic

canc

er p

opul

atio

n; is

it

chea

per t

o ca

re

for

them

at h

ome

or in

ho

spita

l? H

ome

patie

nts

atte

nd 1

-

day

clin

ic fo

r tra

nsfu

sion

, whi

le

inpa

tient

s oc

cupy

ho

spita

l bed

.

Cos

t of p

allia

tive

ca

re.

Setti

ng

Sin

gapo

re

2008

-10

Mul

tidis

cipl

i-nar

y

palli

ativ

e ho

me

care

pro

gram

; “a

dvan

ced

car

e

prog

ram

” (A

CP

)

Gre

ece

Inte

rnat

iona

l

Sam

ple

39%

mal

e.

Adv

ance

d he

art

failu

re (H

F) in

New

Yor

k H

eart

Ass

ocia

tion’

s (N

YH

A) c

lass

III o

r IV

. With

or w

ithou

t im

plan

tatio

n de

vice

s. C

ompl

iant

w

ith d

iet r

estri

ctio

ns

52 p

atie

nts

with

ha

emat

olog

ic

canc

ers.

25

in

hosp

ital

(con

vent

iona

l ca

re),

27 in

hom

e ca

re

Inte

rnat

iona

l

Des

ign

Pro

spec

tive

co

hort

stud

y

Cas

e co

ntro

l/

cost

m

inim

izat

ion

Sys

tem

atic

re

view

Art

icle

Won

g et

al

(201

3)

AN

NA

LS

Aca

dem

y of

M

edic

ine

Sin

gapo

re

Tzal

a et

al

(200

5)

Eur

opea

n Jo

urna

l of

Hea

lth

Eco

nom

ics

Sim

oens

et a

l (2

010)

Jour

nal o

f Pai

n an

d S

ympt

om

Man

agem

ent


124

Wea

knes

ses

MLR

: Did

they

log

y or

no

t? S

tepw

ise.

Sel

ectio

n bi

as b

etw

een

pat

ient

s w

ho c

hoos

e to

go

int

o H

C a

nd th

ose

who

go

in

to li

fe-e

xten

sion

.

Not

long

itudi

nal d

ata,

ju

st 1

sum

. Pat

ient

s m

atch

ed o

nly

by d

isea

se

and

resi

dent

ial s

ituat

ion.

Pro

spec

tive

– th

is is

ex

pect

ed u

tiliz

atio

n an

d

cost

for t

hese

sp

ecia

lized

ser

vice

s. It

is n

ot b

ased

on

em

piric

al d

ata.

Not

from

a s

ocie

tal

pers

pect

ive.

Doe

s no

t ad

dres

s w

heth

er

redu

ctio

n in

ho

spita

lizat

ion

wou

ld

coun

tera

ct th

e hi

gher

estim

ates

for h

ome

care

.

Cos

ting

met

hod

vuln

erab

le to

con

test

; re

lies

on p

oten

tial

earn

ings

rath

er th

an

actu

al e

arni

ngs.

Gen

eral

izab

ility

.

Rec

all a

nd s

ocia

l de

sira

bilit

y bi

ases

.

Res

ults

Sig

nific

ant d

iffer

ence

in

num

ber o

f tre

atm

ents

. A

vera

ge c

ost o

f hom

e ca

re $

3467

, and

$12,

434

for

co

nven

tiona

l (ov

er 2

mon

ths)

.

Old

er p

atie

nts

chea

per.

The

cost

s of

the

first

visi

t

and

follo

w-u

p

visi

t wer

e es

timat

ed fo

r

all s

cena

rios/

mod

ules

(h

ospi

tal,

hom

e, d

ay-

care

, inp

atie

nt).

Hom

e ca

re th

ough

t to

prev

ent h

ospi

taliz

atio

n

but

had

an

over

all

hi

gher

es

timat

e du

e to

ex

pect

ed lo

nger

and

m

ore

frequ

ent

visi

ts

an

d tra

vel c

ost (

even

with

ass

umpt

ion

that

they

all

live

clos

e).

Pub

licly

fina

nced

cos

ts

acco

unt f

or ~

20%

of t

otal

co

sts

and

incr

ease

d to

war

d de

ath.

Pub

lic c

osts

ar

e dr

iven

by

soci

o-de

mog

raph

ic (o

lder

or

mar

ried

= lo

wer

sha

re o

f co

st fo

r pub

lic) a

nd c

linic

al

char

acte

ristic

s.

Out

com

e

Cos

t

Cos

t es

timat

es

Pub

licly

fina

nced

co

st

Met

hods

Pat

ient

s re

crui

ted

in 2

003.

Com

pare

d co

st o

f car

e fo

r all

pa

tient

s du

ring

the

final

2

m

onth

s of

life

. All

form

al

treat

men

ts a

nd s

ervi

ces

from

the

heal

th s

yste

m w

ere

incl

uded

in th

e co

stin

g,

ob

tain

ed fr

om th

e qu

ality

cont

rol u

nit o

f Cla

lit H

ealth

S

ervi

ces

afte

r the

ir de

ath.

Con

sens

us p

roce

ss a

mon

g on

colo

gist

, sur

geon

, pal

liativ

e sp

ecia

list a

nd ra

dio-

ther

apis

t pa

rtici

patin

g on

neg

otia

tion

com

mitt

ee o

f hos

pita

l fin

ance

de

partm

ent t

o es

timat

e th

e ut

iliza

tion

and

cost

for h

ospi

tal-

base

d, d

ay c

are

and

hom

e

care

.

Cos

ts w

ere

seem

ingl

y

estim

ated

from

a m

inis

teria

l pe

rspe

ctiv

e. U

sed

real

pric

ing

data

for h

ealth

pro

fess

iona

ls.

Bi-w

eekl

y in

terv

iew

s (to

tal

667)

Mul

tiple

regr

essi

on

an

alys

is (l

og -

linea

r).

Focu

s

Hom

e ca

re v

s.

conv

entio

nal c

are

cost

for p

atie

nts

dy

ing

with

m

etas

tatic

can

cer.

Est

imat

ion

of c

ost

give

n th

e ut

iliza

tion

expe

cted

by

expe

rt

parti

cipa

nts.

Pub

licly

fina

nced

co

sts

of p

allia

tive

care

.

Setti

ng

Isra

el

Ger

man

y

Toro

nto,

Ont

ario

, C

anad

a

Sam

ple

146

patie

nts

with

m

etas

tatic

can

cer

73 c

onve

ntio

nal 7

2 ho

me

care

Pat

ient

s in

bot

h gr

oups

live

d w

ith

thei

r fam

ilies

55%

men

NS

CLC

pat

ient

s

(N u

nkno

wn)

129

care

give

rs o

f pa

lliat

ive

patie

nts

from

200

5-20

07

Des

ign

Cas

e co

ntro

l/ co

hort

stud

y

Qua

litat

ive

Pro

spec

tive

coho

rt st

udy

Art

icle

Shn

oor e

t al (

2007

)

Am

eric

an J

ourn

al o

f H

ospi

ce &

Pal

liativ

e M

edic

ine

Ost

gath

e et

al

(200

8)

Sup

port

Car

e C

ance

r

Cha

i et a

l (20

13)

Jour

nal o

f Pal

liativ

e C

are


125

Wea

knes

ses

Lite

ratu

re re

view

non

-sy

stem

atic

; pot

entia

l

sele

ctio

n bi

as.

Arg

ues

that

pub

lic c

over

age

is

mos

t cos

t-effi

cien

t but

do

es n

ot e

xact

ly p

rove

it.

Doe

s no

t spe

ak e

qual

ly to

bo

th a

rgum

ents

or a

t all

to

the

argu

men

t for

priv

ate.

Doe

s no

t rep

ort o

n ge

nera

lizab

ility

of t

he H

alifa

x fin

ding

s to

oth

er g

eogr

aphi

c ar

eas

or ti

me

perio

ds.

Doe

s no

t dis

cuss

thei

r ow

n lim

itatio

ns.

Elig

ibili

ty c

riter

ia m

ay li

mit

gene

raliz

abili

ty.

Util

izat

ion

base

d on

re

trosp

ectiv

e re

view

on

ad

min

dat

a. O

ntar

io D

rug

Ben

efit

(OD

B),

out-o

f-

po

cket

spe

ndin

g or

em

erge

ncy

serv

ices

are

not

re

cord

ed.

No

care

give

r cos

t.

Res

ults

Priv

ate

finan

cing

of c

omm

unity

-bas

ed

serv

ices

incr

ease

s he

alth

car

e co

st in

lo

ng-te

rm th

roug

h in

crea

sed

acut

e

care

ut

iliza

tion.

Incr

ease

d in

dire

ct

cost

s in

the

form

of c

areg

iver

bu

rden

/redu

ced

labo

r mar

ket

parti

cipa

tion

for

c

areg

iver

s.

Stu

dy p

artic

ipan

ts w

ere:

you

nger

, en

rolle

d lo

nger

, liv

ed c

lose

r to

the

P

CP

, and

wer

e m

ore

likel

y to

hav

e ca

ncer

th

an th

e ot

her c

lient

s;

You

nger

, liv

ed c

lose

r to

the

PC

P, a

nd

wer

e m

ore

likel

y to

hav

e ha

d ra

diot

hera

py/ m

edic

al o

ncol

ogy

co

nsul

ts t

han

othe

r can

cer d

eced

ents

; Li

ved

~42

days

long

er a

fter d

iagn

osis

an

d liv

ed c

lose

r to

the

PC

P th

an

elig

ible

clie

nts

who

de

clin

ed. M

ay

mis

estim

ate

cost

of c

arin

g

for a

ll pe

ople

nee

ding

pal

liativ

e ca

re

be

caus

e st

udy

subj

ects

m te

nd to

be

youn

ger a

nd c

lose

r to

the

PC

P.

Tota

l cos

ts fo

r all

patie

nt-r

elat

ed

serv

ices

foun

d to

be

$1,6

25,6

58.0

7

(CA

N 2

007)

; $17

,112

.19/

patie

nt a

nd

$117

.95/

patie

nt d

ay. O

n pa

r with

per

di

em fu

ndin

g fo

r LTC

and

low

er th

an

hosp

itals

as

assi

gned

in O

ntar

io.

Out

com

e

Con

clus

ions

ab

out t

he

focu

s

Diff

eren

ces

amon

g gr

oups

Res

ourc

e ut

iliza

tion/

co

st

Met

hods

Lite

ratu

re re

view

(non

-sy

stem

atic

).

Com

pare

s el

igib

le

parti

cipa

nts

who

agr

ee to

th

e st

udy

to th

ose

who

w

ere

elig

ible

but

refu

sed,

th

e en

tiret

y of

pro

gram

en

rolle

es, a

nd a

ll ca

ncer

de

cede

nts

in th

e

prov

inci

al c

ance

r reg

istry

.

Com

pare

d by

sex

, age

, di

stan

ce to

PC

P,

ho

useh

old

inco

me,

and

su

rviv

al ti

me

afte

r can

cer

diag

nosi

s if

appl

icab

le.

Cos

ts w

ere

track

ed o

ver

15 m

onth

s be

twee

n Ja

nuar

y 20

05-M

arch

200

6 us

ing

CC

AC

, E

PC

T an

d O

HIP

fee

sche

dule

s.

Cos

t-ana

lysi

s fo

rm o

f ec

onom

ic e

valu

atio

n.

Focu

s

The

in/d

irect

co

sts

asso

ciat

ed

with

com

mun

ity-

base

d se

rvic

es in

Can

ada.

Gen

eral

izab

ility

of

com

mun

ity-

base

d pr

ogra

ms

cost

est

imat

es.

Rol

e of

sel

ectio

n bi

as, e

ligib

ility

cr

iteria

, ref

usal

bi

as.

The

cost

of

com

mun

ity-b

ased

pa

lliat

ive

care

in

Ont

ario

from

MoH

pe

rspe

ctiv

e.

Setti

ng

Can

ada

Hal

ifax,

Nov

a

Sco

tia,

Can

ada

Ont

ario

, C

anad

a

Sam

ple

Pap

ers

stud

ying

co

mm

unity

-bas

ed c

are

in C

anad

a

50 p

atie

nts

refe

rred

to

the

palli

ativ

e ca

re

prog

ram

in H

alifa

x be

twee

n Fe

b 7,

200

5 an

d N

ov 2

9, 2

005;

co

mpa

red

to10

10

clie

nts

in th

e P

CP

who

w

ere

not s

ubje

cts

of

this

stu

dy.

45/5

0 ha

d ca

ncer

95 p

atie

nts

in s

hare

d-ca

re/ h

ome-

base

d pa

lliat

ive

care

in ru

ral

Ont

ario

.

87%

had

can

cer.

Ave

rage

age

71

Des

ign

Nar

rativ

e re

view

Cas

e co

ntro

l

Coh

ort s

tudy

Art

icle

Lyzw

insk

i (20

12)

Pal

liativ

e an

d S

uppo

rtive

Car

e

Lave

rgne

et a

l (2

011)

Jour

nal o

f Pai

n an

d S

ympt

om

Man

agem

ent

Klin

ger e

t al

(201

1)

Pal

liativ

e M

edic

ine


126

Wea

knes

ses

Upp

er li

mit

on in

com

e ca

tego

ries

may

hav

e be

en lo

w ($

61,0

00+)

or

had

a lo

t of v

aria

bilit

y.

Did

not

mea

sure

tim

e sp

an o

f car

e be

fore

the

prog

ram

; con

serv

ativ

e es

timat

es. S

ome

patie

nts

had

mul

tiple

ca

regi

vers

(no

t co

ntro

lled

for)

.

Rel

ianc

e on

sel

f-rep

orts

of

car

egiv

er (s

ocia

l de

sira

bilit

y bi

as).

Not

gen

eral

izab

le to

ot

her t

ypes

of p

allia

tive

care

. Met

hods

of c

ostin

g ca

regi

ver t

ime

vuln

erab

le to

con

test

.

Eng

lish

only

.

Mos

t gre

y lit

erat

ure.

UK

ba

sed.

Met

hodo

logi

cally

qu

estio

nabl

e ar

ticle

s in

clud

ed.

Gen

eral

izab

ility

.

Res

ults

9% o

f fam

ilies

incu

rred

ec

onom

ic lo

sses

in

exce

ss o

f 10%

of t

heir

pre-

stud

y gr

oss

annu

al

inco

me.

Low

-inco

me

stat

us in

crea

sed

from

27

% to

40%

.

Mea

n m

onth

ly c

ost p

er

patie

nt $

24,5

49 (2

008

CD

N$)

. Fam

ily

care

give

rs c

onst

itute

70

% o

f tha

t. C

osts

are

gr

eate

r for

pat

ient

s cl

oser

to d

eath

, with

lo

wer

phy

sica

l fu

nctio

ning

, and

who

liv

ed w

ith s

omeo

ne.

Fina

ncia

l cos

t of c

arin

g

for s

omeo

ne a

t the

EoL

is

subs

tant

ial a

nd o

f 3

sour

ces:

wor

k-re

late

d,

out-o

f-poc

k-et

, and

tim

e.

Thes

e re

sulte

d in

maj

or

life

chan

ges

for

care

give

rs. S

train

was

as

soci

ated

with

dis

ease

st

age,

SE

S a

nd

et

hnic

ity.

Out

com

e

Eco

nom

ic

loss

Tota

l cos

t

Con

clus

ions

ab

out

info

rmal

cost

of c

are

Met

hods

Sam

ples

of p

allia

tive

ca

re p

atie

nts

livin

g at

ho

me

and

thei

r mai

n in

form

al c

areg

iver

s re

crui

ted

from

5 c

ities

an

d in

terv

iew

ed b

i- w

eekl

y. P

artic

ipan

ts

aske

d to

pro

vide

det

ails

ab

out t

heir

expe

nses

and

abse

nces

from

wor

k th

at re

late

d to

the

patie

nt’s

con

ditio

n.

Dat

a co

llect

ed fr

om

enro

llmen

t unt

il de

ath

via

inte

rvie

ws

with

ca

regi

vers

. Mul

tivar

iabl

e re

gres

sion

use

d to

as

sess

det

erm

inan

ts o

f va

riabi

lity

in th

e to

tal

cost

s.

Sys

tem

atic

revi

ew o

f M

ED

LIN

E, C

ochr

ane,

E

conL

it, C

INA

HL,

E

mba

se. A

ppra

ise

met

hods

. Syn

thes

ize.

Arti

cles

from

ince

ptio

n

of

data

base

to

2012

.

Focu

s

Impa

ct o

f eco

nom

ic

loss

on

info

rmal

ca

regi

vers

vi

s- à

-vis

Sta

tistic

s C

anad

a’s

low

-I nc

ome

cut o

ff.

Est

imat

ing

the

soci

etal

cos

t of

hom

e-ba

sed

pa

lliat

ive

care

and

de

term

inan

ts

of

varia

bilit

y.

Cos

t of u

npai

d ca

re in

palli

ativ

e ho

me

ca

re.

Setti

ng

Can

ada

Ont

ario

,

Can

ada

Inte

rnat

iona

l

Sam

ple

192

fam

ily u

nits

pr

ovid

ing

care

at

hom

e fo

r a te

rmin

al

patie

nt fr

om

Janu

ary

2005

-O

ctob

er 2

006

in

Hal

ifax,

Mon

treal

, W

inni

peg,

E

dmon

ton

and

Vic

toria

136

patie

nt-

care

give

r dy

ads

in

Ont

ario

hom

e ca

re

prog

ram

s6

Inte

rnat

iona

l

Des

ign

Pro

spec

tive

co

hort

stud

y

Pro

spec

tive

coho

rt st

udy

Sys

tem

atic

revi

ew

Art

icle

Jaco

bs e

t al

(201

1)

Pal

liativ

e C

are

Gue

rrie

re e

t al

(201

0)

Pal

liativ

e M

edic

ine

Gar

dine

r et a

l (2

014)

Pal

liativ

e M

edic

ine

6 Dotted borders indicate instances of double-publication.


127

Wea

knes

ses

The

time

spen

t at h

ome

inst

ead

of in

patie

nt c

are

is n

ot

com

pens

ated

for u

sing

a

soci

etal

per

spec

tive/

acco

untin

g fo

r the

cos

t of c

areg

ivin

g. C

ost

is le

ss b

ecau

se th

ey p

ay fo

r all

the

inpa

tient

ser

vice

redu

cing

ap

plic

abili

ty.

Non

-ran

dom

(sel

ectio

n bi

as).

Sm

all s

ampl

e.

Diff

eren

ces

in s

ocia

l pr

ogra

ms/

polic

ies

betw

een

the

5 pr

ovin

ces

are

not t

aken

into

ac

coun

t.

Mai

nly

canc

er p

atie

nts;

trea

ted

as h

omog

enou

s gr

oup.

Lim

ited

to p

atie

nts

in P

CP

s,

mos

tly c

ance

r pat

ient

s.

Gen

eral

izab

ility

que

stio

nabl

e.

Soc

ial d

esira

bilit

y bi

as- s

ome

cost

s co

llect

ed fr

om p

atie

nts

and

care

give

rs.

Res

ults

Pat

ient

s in

trea

tmen

t gro

up

mor

e lik

ely

to d

ie a

t hom

e an

d re

duce

d c

osts

acr

oss

all

prim

ary

diag

nose

s, w

ith s

ome

diffe

renc

e in

sav

ing

betw

een

them

. Can

cer m

ost c

omm

on.

40%

non

-whi

te. C

ance

r sav

es

$593

6. C

OP

D s

aves

$11

325.

C

HF

save

s $8

445

(US

- pat

ient

pa

ys).

The

publ

ic h

ealth

car

e sy

stem

, th

e fa

mily

, and

not

-for-

prof

it

orga

niza

tions

sus

tain

ed

resp

ectiv

ely

71.3

%,

26.

6%,

an

d 1.

6% o

f the

mea

n to

tal

co

st p

er

patie

nt.

Tota

l cos

t of c

are

inc

reas

ed

from

the

fifth

to

the

last

mon

th o

f life

. A la

rge

part

of

th

is w

as a

ttrib

utab

le to

inpa

tient

car

e. T

he la

rges

t in

crea

se in

out

patie

nt c

are

co

sts

was

ob

serv

ed fo

r hom

e ca

re. I

nfor

mal

car

e co

sts

are

high

est f

or th

e la

st 3

mon

ths.

Out

com

e

Pla

ce o

f dea

th a

nd

cost

of

car

e

Tota

l cos

t/cos

t-

shar

e am

ong

publ

ic

heal

th c

are

syst

em,

th

e fa

mily

and

non

-pr

ofits

whe

re

appl

icab

le

Mon

thly

soc

ieta

l

cost

Met

hods

Trea

tmen

t gro

up re

ceiv

ed

inte

rdis

cipl

inar

y ho

me-

base

d P

C a

nd c

ontro

l rec

eive

d

usua

l ser

vice

s bt

w 1

999

and

2001

. Com

pare

d co

ntro

lling

fo

r pat

ient

dem

o-gr

aphi

cs

and

sym

ptom

sev

erity

(MLR

). C

osts

from

per

spec

tive

of th

e pr

ovid

er K

aise

r Per

man

ente

.

Pro

spec

tive

rese

arch

des

ign

w

ith re

peat

ed m

easu

res.

In

terv

iew

s at

2 w

eek

in

terv

als.

Sta

tistic

al a

naly

ses.

C

ostin

g w

ith re

com

men

datio

- n

of th

e C

anad

ian

Coo

rdin

atin

g O

ffice

for H

TA &

Dru

mm

ond;

el

emen

ts o

f a) q

uant

ities

used

and

b) u

nit c

osts

. S

ocie

tal p

ersp

ectiv

e.

Pro

spec

tive

rese

arch

des

ign

w/ r

epea

ted

mea

sure

s.

Inte

rvie

ws

at 2

wee

k in

terv

als.

Sta

tistic

al a

naly

ses

on m

onth

-by-

mon

th b

asis

for

5 m

onth

s pr

ior t

o de

ath.

Soc

ieta

l per

spec

tive.

Focu

s

Pla

ce o

f dea

th a

nd

cost

of s

ervi

ces

by

prim

ary

diag

nosi

s fo

r pa

tient

s re

ceiv

ing

hom

e-ba

sed

palli

ativ

e ca

re c

ompa

red

to

usua

l car

e.

Cos

t of

reso

urce

ut

iliza

tion

by p

allia

tive

care

reci

pien

ts li

ving

at

hom

e.

Cos

t of

res

ourc

e ut

iliza

tion

over

the

palli

ativ

e tra

ject

ory

(com

mun

ity-b

ased

pr

ogra

ms)

.

Setti

ng

Cal

iforn

ia,

Uni

ted

Sta

tes

Can

ada

Can

ada

Sam

ple

Term

inal

ly il

l pa

tient

s; 1

59 in

ho

me

care

and

139

co

ntro

ls. D

iagn

osed

w

ith c

ance

r, co

nges

tive

hear

t fa

ilure

(CH

F), o

r ch

roni

c ob

stru

ctiv

e pu

lmon

ary

dise

ase

(CO

PD

)

248

patie

nt-

care

give

r dya

ds in

P

CP

s fro

m J

anua

ry

2005

-Dec

embe

r 20

06 in

in H

alifa

x,

Mon

treal

, Win

nipe

g,

Edm

onto

n an

d V

icto

ria

160

patie

nt-

care

give

r dya

ds i

n P

CP

s fro

m J

anua

ry

2005

-Oct

ober

200

6 in

H

alifa

x, M

ontre

al,

Win

nipe

g,

Edm

onto

n an

d V

icto

ria

Des

ign

Cas

e co

ntro

l

Coh

ort s

tudy

Coh

ort s

tudy

Art

icle

Eng

uida

nos

et a

l (2

005)

Jour

nal o

f Soc

ial

Wor

k in

EoL

&

Pal

liativ

e C

are

Dum

ont e

t al

(200

9)

Pal

liativ

e M

edic

ine

Dum

ont e

t al

(201

0)

Pal

liativ

e M

edic

ine


128

Wea

knes

ses

Cos

ting

met

hod

vuln

erab

le to

con

test

; re

lies

on p

oten

tial

earn

ings

rath

er th

an

actu

al e

arni

ngs.

Gen

eral

izab

ility

.

Rec

all a

nd s

ocia

l de

sira

bilit

y bi

as.

Sm

all s

ampl

e.

No

cont

rol g

roup

.

Min

iste

rial c

ost o

nly.

Nov

el m

odel

for

haem

atol

ogic

pat

ient

s;

non-

gene

raliz

able

, ha

rd to

ver

ify o

r co

nfid

ently

end

orse

th

is ty

pe o

f mod

el fo

r ad

optio

n. N

eeds

mor

e re

sear

ch in

this

are

a.

Low

repr

esen

tatio

n

for m

inor

ity g

roup

s.

Gen

eral

izab

ility

. R

elyi

ng o

n ho

me

deat

h, n

ot P

OD

pr

efer

ence

, as

mea

sure

men

t of

patie

nt p

refe

renc

e.

Res

ults

Ave

rage

mon

thly

cos

t was

$14

924

(201

1 C

DN

$) p

er p

atie

nt in

la

st y

ear o

f life

. Unp

aid

care

givi

ng

cost

s– $

11 3

34 –

acc

ount

for 7

7%

follo

wed

by

publ

ic c

osts

($32

11;

21%

) and

out-o

f-poc

ket (

$379

; 2%

). M

onth

ly

cost

s in

crea

sed

with

pro

xim

ity to

de

ath.

Dis

char

ged

early

and

term

inal

phas

e re

quire

d th

e m

ost h

ome

visi

ts (a

vg 2

7.2

and

24.1

), tra

nsfu

sion

s (6

.1 a

nd 6

.8) a

nd d

ays

of

car

e (2

2.8

and

19.7

). M

ean

mon

thly

cos

ts d

eter

min

ed b

y

dise

ase

stat

us a

nd tr

ansf

usio

n re

quire

men

ts. M

MC

for t

erm

inal

pa

tient

s (4

,232

.50€

) and

thos

e di

scha

rged

ear

ly (3

,986

.40€

) wer

e hi

gher

than

thos

e fo

r adv

ance

d (2

,303

.80€

) and

chr

onic

pat

ient

s (1

,488

,30€

). H

C c

ost w

as lo

wer

than

the

corr

espo

ndin

g ho

spita

l ch

arge

s, b

ut e

xcee

ded

the

dist

rict

fare

s fo

r the

HC

of c

ance

r pat

ient

s.

Inte

rven

tion

grou

p re

porte

d gr

eate

r im

prov

emen

t in

satis

fact

ion

at 3

0 an

d 90

day

s, w

ere

mor

e lik

ely

to d

ie

at h

ome,

less

like

ly to

vis

it th

e em

erge

ncy

depa

rtmen

t or b

e ad

mitt

ed to

the

hosp

ital r

esul

ting

in

low

er c

osts

.

Out

com

e

Unp

aid

ca

regi

ving

cos

t, pu

blic

cos

t,

out-o

f-poc

ket

co

st

Rec

ours

e

utili

zatio

n co

st

Sat

isfa

ctio

n,

utili

zatio

n, P

OD

, co

st

Met

hods

Pro

spec

tive

coho

rt st

udy

recr

uite

d pa

rtici

pant

s fro

m P

CP

an

d in

terv

iew

ed b

y-w

eekl

y fro

m a

dmis

sion

unt

il de

ath.

C

ostin

g do

ne u

sing

Am

bula

tory

H

ome

Car

e R

ecor

d. S

ocie

tal

pers

pect

ive.

Sta

tistic

al a

naly

ses.

Dat

a de

scrib

ed fr

om c

linic

al

reco

rds

and

data

base

whe

re

al

l h

ome

activ

ities

wer

e re

cord

ed. C

osts

per

tain

ed to

he

alth

car

e pr

ovid

ers,

mat

eria

ls

and

med

icin

es, t

rans

fusi

ons,

an

d la

bora

tory

(blo

od c

hem

istry

, m

icro

biol

ogy)

. Pat

ient

s di

vide

d in

to te

rmin

al p

hase

, adv

ance

d ph

ase,

chr

onic

pha

se a

nd

“dis

char

ged

early

with

cur

able

di

seas

e.”

152

rece

ive

usua

l car

e an

d 14

5 re

ceiv

e th

e in

terv

entio

n pl

us

usua

l car

e. T

est h

ypot

hesi

s

that

the

new

trea

tmen

t wou

ld

impr

ove

satis

fact

ion

and

cost

.

Focu

s

The

mag

nitu

de,

sh

are

and

dete

rmin

ants

of

palli

ativ

e ho

me

care

cost

s th

at a

re

un

paid

car

egiv

ing.

Cos

t ana

lysi

s o

f ho

me

care

fo

r PC

pa

tient

s w

ith

haem

atol

ogic

m

alig

nanc

ies

spec

ifica

lly.

Usu

al c

are

vs.

usua

l car

e pl

us in

-ho

me

palli

ativ

e ca

re.

Setti

ng

Toro

nto,

C

anad

a

Italy

Col

orad

o

(N

=14)

and

H

awai

i (N

=150

)

Sam

ple

169

patie

nt-c

areg

iver

dy

ads

with

prim

ary

canc

er d

iagn

oses

fro

m T

oron

to (M

ount

S

inai

) btw

Jul

y 20

05

and

Sep

tem

ber 2

007

144

patie

nts

with

ha

emat

olog

ic

mal

igna

ncie

s of

va

riabl

e di

seas

e

stat

us b

tw 2

004-

2006

298

hom

e-bo

und,

term

inal

ly il

l pat

ient

s w

ith C

OP

D a

nd C

HF

Des

ign

Pro

spec

tive

coho

rt st

udy

Coh

ort s

tudy

RC

T

Art

icle

Cha

i et a

l (20

14)

Hea

lth a

nd

S

ocia

l C

are

in

the

Com

mun

ity

Car

toni

et a

l (2

007)

Hae

mat

olog

ica/

the

hem

atol

ogy

jour

nal

Bru

mle

y et

al

(200

7)

Jour

nal o

f the

A

mer

icn

Ger

iatri

cs S

ocie

ty


129

Appendix C Detailed characteristics of caregiver burden papers

Wea

knes

ses

Syn

thes

izes

exi

stin

g

evid

ence

but

do

es

not p

rovi

de m

uch

new

.

Non

-sys

tem

atic

revi

ew,

incl

usio

n cr

iteria

un

clea

r, in

clus

ion/

se

lect

ion

bias

.

Pat

ient

s w

ith n

o

care

give

rs e

xclu

de

(not

the

poin

t of t

his

st

udy,

but

onc

e ag

ain

thei

r out

com

es a

re

mis

sing

).

Com

para

bilit

y; m

ost

stud

ies

are

prim

arily

w

hite

sub

ject

s (th

ough

st

udyi

ng a

noth

er

popu

latio

n is

a

poss

ible

stre

ngth

). Fi

ndin

gs a

re

cons

iste

nt w

ith o

ther

st

udie

s.

Res

ults

Car

egiv

ers

repo

rt un

met

nee

ds fo

r in

form

atio

n, c

omm

unic

atio

n, s

ervi

ce p

rovi

sion

an

d su

ppor

t fro

m h

ealth

and

com

mun

ity

serv

ices

. Wan

t to

know

how

to p

rovi

de

prac

tical

car

e, h

ow to

com

fort

the

patie

nt,

wha

t to

expe

ct, h

ow to

dea

l with

sym

ptom

s,

how

to a

cqui

re a

ids

like

wal

king

fram

es a

nd

whe

elch

airs

etc

. Stu

dy re

veal

ed (i

) bar

riers

to

seek

ing

help

; (ii)

lack

of r

esea

rch-

base

d in

terv

entio

ns fo

cuse

d on

redu

cing

the

nega

tive

aspe

cts

of c

areg

ivin

g; a

nd (i

ii)

impe

dim

ents

to e

ffect

ive

polic

y an

d se

rvic

e de

velo

pmen

t for

fam

ily c

areg

iver

s. L

ack

out

outc

ome

evid

ence

am

ong

care

give

rs- e

thic

s pr

oble

ms

w/ R

CTs

. C

areg

iver

s of

pat

ient

s w

ho d

ied

at h

ome

thou

ght t

hat i

t had

a m

ore

posi

tive

influ

ence

on

the

patie

nts’

QoL

than

was

the

case

in th

e ot

her g

roup

s, w

ere

mor

e sa

tisfie

d w

ith th

eir

own

achi

evem

ent,

but

als

o ex

perie

nced

the

mos

t sle

ep d

epriv

atio

n. c

areg

iver

opi

nion

of

why

pat

ient

s ne

eded

hos

pice

car

e w

as

near

ly th

e sa

me

in g

roup

s 1

and

2:

acut

ely

deve

lopi

ng s

ympt

oms.

Fact

or a

naly

sis

of th

e C

SI

(rol

e, p

erso

nal,

and

emot

iona

l stra

in) f

ound

pat

ient

pro

blem

be

havi

ors

pred

ict a

ll ty

pes

of c

areg

iver

stra

in.

Per

ceiv

ed la

ck o

f sup

port

from

the

heal

th

care

team

pre

dict

ed p

erso

nal a

nd e

mot

iona

l st

rain

. Hig

her i

ncom

e pr

edic

ted

role

stra

in.

Pat

ient

func

tiona

l lim

itatio

ns p

redi

cted

pe

rson

al a

nd ro

le s

train

.

Out

com

e

Com

mon

ly

repo

rted

expe

rienc

es a

nd

unm

et

need

s

Car

egiv

er

expe

rienc

es

Car

egiv

er S

train

In

dex

(CS

I)

scor

es

Met

hods

Lite

ratu

re re

view

(non

-sy

stem

atic

) plu

s

su

rvey

of h

ome

care

pr

ovid

ers

to

valid

ate/

cont

est t

he

findi

ngs

of th

e

lit

erat

ure

revi

ew.

Pat

ient

s re

crui

ted

from

ad

vanc

ed p

allia

tive

hom

e ca

re te

ams

and

a

hosp

ice

in U

ppsa

la

durin

g 1

year

(vag

ue)

stud

y pe

riod.

A

ques

tionn

aire

was

m

aile

d to

car

egiv

ers

an

d m

edic

al re

cord

s w

ere

anal

yzed

.

Dat

a co

llect

ed th

roug

h st

ruct

ured

inte

rvie

ws

w

ith a

ll ca

regi

vers

usin

g C

areg

iver

Stra

in

Inde

x, R

evis

ed

M

emor

y an

d

B

ehav

iour

Pro

blem

s C

heck

list a

nd th

e K

atz

Inde

x of

Act

iviti

es o

f D

aily

Liv

ing

to

unde

rsta

nd s

ituat

iona

l ch

arac

teris

tics.

Focu

s

Bar

riers

to

addr

essi

ng th

e un

met

nee

ds o

f ca

regi

vers

in

palli

ativ

e ho

me

care

.

Doe

s ca

regi

ver

perc

eptio

n of

pa

lliat

ive

care

di

ffer d

epen

ding

on

pla

ce o

f car

e

and

deat

h?

Iden

tifyi

ng th

e is

sues

/ ch

arac

teris

tics

of

pro

vidi

ng

palli

ativ

e ca

re

to

peo

ple

with

de

men

tia

spec

ifica

lly.

Setti

ng

Aus

tralia

Upp

sala

, S

wed

en

Uni

ted

S

tate

s

Sam

ple

20 c

areg

iver

s, 6

vo

lunt

eers

and

23

serv

ice

prov

ider

s su

bmitt

ed

desc

ribin

g th

eir

expe

rienc

e an

d un

met

nee

ds.

Maj

ority

fem

ale

wiv

es/d

augh

ters

ca

ring

for s

omeo

ne

with

can

cer o

r de

men

tia

Car

egiv

ers

to

patie

nts

in h

ome

care

who

die

d a

t ho

me

(N =

63) (

1),

at

hos

pice

(N=

51)

(2),

and

care

in

hosp

ice

dyin

g at

ho

spic

e (N

=69)

(3)

150

prim

arily

bla

ck

care

give

rs o

f pa

tient

s w

ith

dem

entia

in

palli

ativ

e ca

re

Des

ign

Lite

ratu

re

revi

ew/

C

ohor

t stu

dy

Cas

e co

ntro

l

Coh

ort s

tudy

Art

icle

Aou

n et

al

(200

5)

Pal

liativ

e M

edic

ine

Car

lsso

n et

al

(200

3)

Pal

liativ

e an

d S

uppo

rtive

C

are

Diw

an e

t al

(200

4)

Jour

nal o

f P

allia

tive

Med

icin

e


130

Wea

knes

ses

Lack

s si

gns

of th

orou

gh

re

view

i.e.

sta

ges

of t

he

sear

ch, p

roce

ss o

f arti

cle

sele

ctio

n, h

ow m

any

artic

les

re

view

ed.

Lack

of l

ongi

tudi

nal d

ata

on

phys

ical

bur

den.

Sam

ple

size

.

Non

-long

itudi

nal.

HA

DS

inst

rum

ent n

ot

appr

opria

te fo

r eve

ryon

e be

caus

e an

hedo

nia

is a

co

mm

on E

oL s

ympt

om,

Sat

urat

ion

may

not

hav

e be

en

achi

eved

Pos

sibi

lity

that

car

egiv

ers

who

de

clin

ed h

ad d

iffer

ent

expe

rienc

es.

Res

ults

Ove

r ½ o

f hom

e ca

regi

vers

repo

rt he

alth

pro

blem

s,

mai

nly

hear

t pr

oble

ms,

hyp

erte

nsio

n, a

nd a

rthrit

is.

Sta

ge o

f can

cer,

exte

nt o

f p

atie

nt

disa

bilit

y, h

ighe

r per

sona

l car

e ne

eds

of

the

patie

nt, l

ack

of p

atie

nt m

obili

ty,

patie

nt d

epen

denc

y

in in

stru

men

tal

activ

ities

ass

ocia

ted

with

adv

erse

ph

ysic

al c

areg

iver

out

-com

es.

Pat

ient

s’ a

nd c

areg

iver

s’ a

nxie

ty a

nd

depr

essi

on s

core

s w

ere

sign

ifica

ntly

co

rrel

ated

. 33%

of c

areg

iver

s ha

d hi

gh a

nxie

ty a

nd 2

8 %

dep

ress

ion.

¼

patie

nts

had

clin

ical

ly re

leva

nt a

nxie

ty

and ½

had

clin

ic-a

lly s

ympt

omat

ic

depr

essi

on s

core

s.

Them

es: t

he s

igni

fican

t bur

den

of

poly

phar

mac

y; th

e po

sitiv

e im

pact

of

subc

utan

eous

infu

sion

s; th

e va

lue

of

bein

g ab

le to

giv

e m

edic

atio

ns a

s ne

eded

for s

ympt

om c

ontro

l

Out

com

e

Phy

sica

l

heal

th

need

s

of

care

give

rs

Psy

chol

ogic

al

dist

ress

Exp

erie

nces

re

porte

d by

ca

regi

vers

Met

hods

Rev

iew

of P

ubM

ed a

nd

CIN

AH

L us

ing

key

te

rms

“Car

egiv

er b

urde

n,” “

phys

ical

w

ell-b

eing

” and

“adv

erse

ph

ysic

al o

utco

mes

.”

QoL

was

col

lect

ed w

ith th

e E

OR

TC Q

LQ -C

15-P

AL

(pat

ient

s) a

nd th

e S

hort

Form

-8 H

ealth

Sur

vey

(car

egiv

ers)

; Psy

chol

ogic

al

dist

ress

with

Hos

pita

l Anx

iety

an

d D

epre

ssio

n S

cale

(H

AD

S)

Them

atic

con

tent

ana

lysi

s of

fo

cus

grou

ps. D

iscu

ss

poly

phar

mac

y th

e us

e of

sy

ringe

s an

d as

-nee

ded

med

s by

info

rmal

car

egiv

ers.

Focu

s

Iden

tifyi

ng p

hysi

cal

he

alth

nee

ds

and

vuln

erab

ilitie

s o

f pa

lliat

ive

onco

logy

ca

regi

vers

.

Psy

chol

ogic

al d

istre

ss

and

QoL

for

pa

tient

s an

d ca

regi

vers

in

palli

ativ

e c

ance

r ho

me

care

.

Car

egiv

er e

xper

ienc

e m

anag

ing

med

icat

ions

for

palli

ativ

e ca

re

patie

nts.

Setti

ng

Inte

rnat

iona

l

Leip

zig,

Ger

man

y

Nor

th E

ast R

egio

n,

Irela

nd

Sam

ple

Lite

ratu

re

disc

ussi

ng h

ealth

im

plic

atio

ns o

f ca

regi

ving

for

care

give

rs

106

patie

nt

care

give

r dy

ads

in p

allia

tive

onco

logi

c h

ome

care

68%

fem

ale

3 fo

cus

grou

ps

(N=1

8) o

f be

reav

ed

care

give

rs w

ho

care

d fo

r so

meo

ne a

t ho

me

Des

ign

Sys

tem

atic

re

view

Coh

ort s

tudy

Qua

litat

ive

Art

icle

Gla

jche

n (2

012)

Sem

inar

s in

O

ncol

ogy

Nur

sing

Göt

ze (2

014)

Sem

inar

s in

O

ncol

ogy

Nur

sing

She

ehy-

Ske

ffing

ton

et a

l (2

013)

Am

eric

an J

ourn

al

of H

ospi

ce &

P

allia

tive

Med

icin

e


131

Wea

knes

ses

Sel

f-sel

ectio

n in

favo

ur o

f tho

se

with

ene

rgy/

inte

rest

in c

ance

r and

ca

regi

ving

. Sel

ectio

n bi

as. M

ay

be m

ore

dyad

s w

ithou

t con

flict

th

an w

ith. M

any

decl

ined

beca

use

they

wer

e to

o tir

ed.

Res

ults

may

not

refle

ct e

very

one.

S

mal

l sam

ple.

Car

egiv

er s

atis

fact

ion

rath

er th

an

patie

nt- t

his

is p

artia

lly b

ecau

se

patie

nts

coul

dn’t

resp

ond

by th

e

EoL

. Res

ults

may

not

be

gene

raliz

able

to o

ther

pop

ulat

ions

.

Due

to s

ize

of s

ampl

e, li

mite

d in

nu

mbe

r of v

aria

bles

that

cou

ld b

e co

ncur

rent

ly a

sses

sed.

Pot

entia

l om

itted

var

iabl

e bi

as (u

sed

back

war

d st

epw

ise

to c

ount

er

this

). Th

e in

terR

AI P

C d

oes

not i

nclu

de

care

give

r sex

or a

ge. H

as b

een

show

n in

oth

er re

sear

ch th

at

olde

r car

egiv

ers

have

wor

se

perc

eive

d he

alth

and

incr

ease

d pr

escr

iptio

n dr

ug u

se.

Cro

ss-s

ectio

nal d

ata

limits

abi

lity

to s

ee te

mpo

ral o

rder

of

asso

ciat

ions

foun

d.

Res

ults

Sup

ports

the

findi

ng th

at o

nly

care

give

rs’ d

epre

ssio

n ha

s a

dire

ct

sign

ifica

nt a

ssoc

iatio

n w

ith c

areg

iver

bu

rden

, and

sho

ws

tha

t the

effe

cts

of

the

othe

r ind

epen

dent

var

iabl

es o

n bu

rden

are

med

iate

d th

roug

h de

pres

sion

. In

Mod

el 2

, anx

iety

and

de

pres

sion

are

med

iatin

g fa

ctor

s be

twee

n 3

IVs

(soc

ial s

uppo

rt-

depr

essi

on, p

hysi

cal h

ealth

- anx

iety

) an

d ca

regi

ver b

urde

n, a

nd 1

2% o

f the

varia

nce

is e

xpla

ined

.

Eac

h of

the

nine

qua

lity

of c

are

para

met

ers

wer

e co

nsis

tent

ly fo

und

to

be s

igni

fican

t pre

dict

ors

of o

vera

ll

satis

fact

ion

with

pal

liativ

e ca

re. T

hey

w

ere

“alw

ays

spen

t eno

ugh

time,

” “n

ever

arr

ived

late

,” “n

ever

bee

n ha

rd

to re

ach,

” “ne

ver s

eem

ed d

istra

cted

,” “a

lway

s w

illin

g to

list

en,”

“nev

er

treat

ed m

ore

as a

dis

ease

than

a

pers

on,”

“alw

ays

show

ed p

erso

nal

conc

ern,

” “al

way

s re

spon

ded

quic

kly,

” “n

ever

igno

red.

” C

areg

iver

dis

tress

was

evi

dent

for

22

% o

f pa

lliat

ive

hom

e ca

re c

lient

s.

Clin

ical

inst

abili

ty d

epre

ssiv

e

sym

ptom

s, c

ogni

tive

impa

irmen

t, an

d po

sitiv

e ou

tlook

iden

tifie

d as

as

clie

nt-

leve

l pre

dict

ors.

Ser

vice

use

/ pro

vide

r va

riabl

es p

redi

ctin

g ca

regi

ver d

istre

ss

incl

ude

the

spec

ific

hom

e ca

re a

genc

y,

hosp

italiz

atio

ns in

the

last

90

day

s,

and

nurs

ing

visi

ts.

Out

com

e

Car

egiv

er b

urde

n

Sat

isfa

ctio

n w

ith

hom

e c

are

expe

rienc

e

Indi

cato

rs o

f ca

regi

ver d

istre

ss

Met

hods

Wea

knes

ses

Sec

onda

ry a

naly

sis

of a

n ex

istin

g da

tase

t- no

t all

varia

bles

of

pred

icte

d va

riabl

es w

ere

incl

uded

a

prio

ri. C

lust

er m

etho

ds

not

supp

orte

d by

ext

ensi

ve s

tatis

tical

re

ason

ing

(no

defin

itive

test

exi

sts

to d

eter

min

e if

true

clus

terin

g is

pr

esen

t in

the

data

; ge

nera

lizab

ility

.

Faci

litie

s lim

ited

to 1

are

a

in

Japa

n.

Vas

t var

iatio

n in

pos

t-be

reav

emen

t (6

mon

ths

– 2½

ye

ars)

, rec

all b

ias.

Cro

ss-s

ectio

nal,

cann

ot s

ee h

ow it

ch

ange

s ov

er ti

me.

Can

not s

ee

caus

al d

irect

ion

as w

ell.

Car

er’s

men

tal s

tate

may

influ

ence

as

sess

men

t of

self-

repo

rted

mea

sure

Rec

all b

ias

Res

ults

Sel

f-rep

orte

d an

xiet

y an

d

com

pete

nce

subs

cale

tota

l sco

res

at

the

time

of

com

men

cem

ent

wer

e as

soci

ated

with

ca

regi

vers

at r

isk

of lo

wer

leve

ls o

f ps

ycho

soci

al fu

nctio

ning

5 w

eeks

late

r. P

ossi

ble

to id

entif

y vu

lner

able

car

egiv

ers

early

.

Fact

or a

naly

sis

resu

lted

in 2

9 it

ems

and

8 fa

ctor

s: B

urde

n of

Car

e, C

once

rns

abou

t H

ome

Car

e D

octo

r, B

alan

ce o

f Wor

k an

d C

are,

Pat

ient

’s P

ain

and

Con

ditio

n, C

once

rns

abou

t Vis

iting

Nur

se, C

once

rns

abou

t Hom

e C

are

Ser

vice

, Rel

atio

nshi

p be

twee

n Fa

mily

C

areg

iver

s an

d th

eir F

amili

es, a

nd F

uner

al

Pre

para

tions

.

Car

ers

foun

d to

hav

e be

tter p

hysi

c-al

hea

lth

and

wor

se m

enta

l hea

lth th

an th

e ge

nera

l po

pula

tion.

35%

repo

rted

thei

r hea

lth to

be

wor

se th

an it

was

on

e ye

ar a

go. H

RQ

OL

of

care

rs w

hose

he

alth

had

det

erio

r-at

ed in

the

pr

evio

us y

ear w

as a

ssoc

iate

d w

ith th

e pa

tient

’s

care

nee

ds b

ut n

ot ti

me

in

put,

unlik

e th

e ca

rers

repo

rting

st

able

hea

lth’s

HR

QO

L.

Car

er’c

cla

ssifi

ed a

s st

able

if th

ey re

porte

d

thei

r hea

lth th

e “s

ame”

as

the

prio

r y

ear.

Out

com

e

Pre

dict

ing

th

e vu

lner

abili

ty o

f in

form

al c

areg

iver

s

Com

plet

ed F

DS

The

HR

QO

L su

rvey

resu

lts o

f hom

e-ba

sed

palli

ativ

e ca

regi

vers

Met

hods

Dat

a ob

tain

ed a

t the

sta

rt of

ho

me-

base

d pa

lliat

ive

care

an

d 5-

wee

k fo

llow

-ups

in

clud

ing

inst

rum

ents

m

easu

ring

prep

ared

ness

, co

mpe

tenc

e, s

ocia

l sup

port,

an

xiet

y an

d se

lf-ef

ficac

y of

ca

regi

vers

(HA

DS

).

Dis

tribu

ted

Fam

ily’s

Diff

icul

ty

Sca

le (F

DS

) sur

vey,

whi

ch

was

der

ived

from

p

ilot

inte

rvie

ws

and

lit

revi

ews,

to

395

bere

aved

car

egiv

ers.

S

tatis

tical

ana

lyse

s.

The

cros

s-se

ctio

nal s

tudy

us

ed th

e S

hort

Form

-36

Hea

lth S

urve

y to

mea

sure

H

RQ

OL.

Thi

s su

rvey

is

adm

inis

tere

d to

the

gene

ral

popu

latio

n an

d m

easu

res

ph

ysic

al fu

nctio

n, b

odily

pai

n,

gene

ral h

ealth

, vita

lity,

soc

ial

func

tion,

em

otio

nal f

unct

ion,

m

enta

l hea

lth.

Focu

s

Pre

dict

ors

of

care

give

r ps

ycho

soci

al fu

nctio

n.

Fam

ily d

iffic

ulty

with

E

oL h

ome

c

are.

Hea

lth-r

elat

ed Q

oL o

f ca

regi

vers

.

Dim

ensi

ons

of th

e C

areg

iver

R

eact

ion

Ass

essm

ent:

self-

es

teem

, lac

k of

fam

ily s

uppo

rt,

finan

ces,

impa

ct o

n da

ily

sche

dule

; wer

e us

ed a

s th

e de

pend

ent v

aria

ble.

Inde

pend

ent

varia

bles

wer

e te

sted

in th

e m

odel

s: p

atie

nt p

ain,

fatig

ue,

an

d na

usea

; car

egiv

ers’

phy

sica

l Q

oL, a

nxie

ty a

nd d

epre

ssio

n,

an

d so

cial

sup

port.

Par

tial

le

ast s

quar

es p

ath

anal

ysis

.

Car

egiv

ers

inte

rvie

wed

biw

eekl

y fro

m h

ome

care

ad

mis

sion

unt

il de

ath.

S

atis

fact

ion

asse

ssed

usi

ng

th

e Q

ualit

y of

End

-of-L

ife c

are

and

Sat

isfa

ctio

n w

ith T

reat

men

t (Q

UE

ST)

que

stio

nnai

re. M

LR

used

to d

eter

min

e th

e ex

tent

to

whi

ch d

emog

raph

ic, q

ualit

y of

ca

re, a

nd s

ervi

ce re

late

d

varia

bles

pre

dict

ed s

atis

fact

ion.

Ass

essm

ents

per

form

ed b

y

case

man

ager

s du

ring

the

inte

rRA

I p

ilot i

mpl

emen

tatio

n

in 2

007-

09. M

ultiv

aria

te

anal

ysis

.

Focu

s

How

var

iabl

es

cont

ribut

ing

to

care

give

r bu

rden

ar

e co

rrel

ated

with

eac

h ot

her.

Car

egiv

er

satis

fact

ion

with

ho

me-

base

d

palli

ativ

e nu

rsin

g/

phys

icia

n ca

re.

Det

erm

inan

ts o

f ca

regi

ver b

urde

n am

ong

hom

e

ca

re c

lient

s.

Setti

ng

Nor

way

Ont

ario

, Can

ada

Ont

ario

, Can

ada

Sam

ple

96 c

areg

iver

s of

ca

ncer

pat

ient

s

in p

allia

tive

hom

e ca

re

104

fam

ily

care

give

rs o

f pa

lliat

ive

hom

e-ba

sed

patie

nts

All

reci

pien

ts o

f co

mm

unity

-

base

d pa

lliat

ive

care

who

wer

e as

sess

ed w

ith

th

e in

terR

AI P

C

Des

ign

Coh

ort s

tudy

Coh

ort s

tudy

Coh

ort s

tudy

Art

icle

Gro

v et

al (

2006

)

Soc

ial S

cien

ce &

M

edic

ine

Gue

rrie

re e

t al

(201

3)

Pal

liativ

e M

edic

ine

Hird

es e

t al

(201

2)

Pal

liativ

e an

d S

uppo

rtive

Car

e


132

Wea

knes

ses

Low

resp

onse

rate

.

No

rand

omiz

atio

n (s

elec

tion/

conv

enie

nce

bias

).

Mos

t par

ticip

ants

wer

e no

t sp

oken

to u

ntil

6 m

onth

s in

to

bere

avem

ent (

reca

ll bi

as).

Gen

eral

izab

ility

.

care

give

rs n

ot li

ving

with

the

patie

nt le

ss a

cces

sibl

e to

the

trial

. Ext

erna

l val

idity

.

Mis

sing

dat

a du

e to

dea

th o

f th

e pa

tient

. Exc

lude

d da

ta

whe

re th

e pe

rson

die

d be

fore

th

e 4-

9- 1

2-w

eek

follo

w-u

ps

Attr

ition

hig

her t

han

expe

cted

, red

uced

pow

er.

Sm

all s

ampl

e si

ze

Ver

y sm

all h

ome

care

sa

mpl

e G

erm

an la

ngua

ge

only

. The

SE

IQoL

-DW

too

dem

andi

ng fo

r som

e pa

rtici

pant

s w

ho re

fuse

d (N

=4).

Dep

ress

ion

and

anxi

ety

of

the

patie

nt c

ould

not

be

asse

ssed

due

to to

o cr

itica

l

of c

ondi

tion

(8/2

7 fil

led

out

HA

DS

).

Res

ults

Dea

th a

t hom

e oc

curr

ed fo

r 80.

3% o

f pa

tient

s w

ith a

cces

s to

hom

ecar

e an

d

20.5

% o

f tho

se w

ithou

t acc

ess.

Des

pite

ca

ring

for a

love

d on

e at

hom

e be

ing

a gr

eate

r fin

anci

al a

nd e

mot

iona

l bur

den,

th

ere

was

mor

e sa

tisfa

ctio

n w

ith th

e ca

ring

expe

rienc

e of

thos

e w

hose

love

d on

es

di

ed a

t hom

e an

d ha

d ac

cess

to th

e ho

mec

are

prog

ram

. HC

pre

fera

ble

to m

ost

care

give

rs.

Sco

res

on th

e G

HQ

-28

fell

belo

w th

e th

resh

old

of 5

/6 in

a th

ird o

f par

ticip

ants

in

each

tria

l arm

at a

ny fo

llow

-up

poin

t. M

ean

scor

es in

the

inte

rven

tion

grou

p w

ere

lo

wer

at a

ll tim

e po

ints

but

diff

eren

ces

wer

e no

t sig

nific

ant.

No

diffe

renc

e w

as

obse

rved

in s

econ

dary

out

com

es. C

arer

s re

ceiv

ing

the

inte

rven

tion

repo

rted

qual

itativ

e be

nefit

.

Fifty

-nin

e pe

rcen

t did

not

rece

ive

any

finan

cial

aid

for h

ome

care

, 33%

had

in

crea

sed

risk

for p

sych

osom

atic

pro

blem

s,

45%

had

anx

iety

, and

33%

incr

ease

d de

pres

sion

leve

ls. T

he c

areg

iver

’s Q

oL

w

as m

ost s

trong

ly a

ffect

ed b

y th

e bu

rden

of c

are

(p <

.001

) and

the

patie

nt’s

men

tal s

tate

(p <

.03)

. The

mos

t

chal

leng

ing

aspe

ct fo

r th

e ca

regi

vers

was

th

e co

gniti

ve im

pairm

ent a

mon

g th

e P

MB

T pa

tient

s. P

artic

ular

ly e

xhau

stin

g an

d tro

ublin

g w

ere

chan

ges

in th

e pa

tient

s’

pers

onal

ity a

nd a

ggre

ssiv

e (u

npre

dict

able

) be

havi

or.

Out

com

e

Sat

isfa

ctio

n an

d ex

perie

nce

with

hom

e

deat

h

Car

egiv

er d

istre

ss

at

4-w

eek,

9-w

eek

and

12-w

eek

follo

w-

up

Car

egiv

er d

istre

ss

Met

hods

A to

tal o

f 159

ca

regi

vers

wer

e in

terv

iew

ed, 7

6

from

the

hom

e pa

lliat

ive

prog

ram

an

d 83

who

had

no

acce

ss to

a

palli

ativ

e ca

re

prog

ram

. Dat

a co

llect

ed a

nd

anal

yzed

.

Inte

rven

tion

(s

peci

alis

t pal

liativ

e ca

re s

ervi

ces)

of 6

w

eekl

y vi

sits

by

a tra

ined

adv

isor

pr

ovid

ed to

the

inte

rven

tion

grou

p.

2001

-200

3.

Inte

rvie

ws

co

nduc

ted

with

ca

regi

vers

and

th

emat

ic c

onte

nt

anal

ysis

with

tra

nscr

ipt d

ata.

Min

i-Men

tal S

tate

E

valu

atio

n (M

MS

E)

for p

atie

nts.

Focu

s

Car

egiv

er e

xper

ienc

e

of h

ome

deat

h w

ith

an

d w

ithou

t ac

cess

to

hom

e ca

re

Car

egiv

er d

istre

ss in

H

C w

ith a

nd w

ithou

t sp

ecia

list

ser

vice

s

Dis

tress

and

bur

den

of c

are

asso

ciat

ed

w

ith b

rain

tum

our

pa

tient

spe

cific

ally

Setti

ng

Neg

ev, I

srae

l

Lond

on, U

K

Ger

man

y

Sam

ple

240

care

give

rs o

f pa

tient

s w

ho d

ied

in a

ho

mec

are

prog

ram

an

d 40

4 ca

regi

vers

of

patie

nts

who

die

d

with

no

acce

ss to

HC

co

ntac

ted

1999

-200

1

1271

info

rmal

ca

regi

vers

sco

ring

over

5 o

n th

e

Gen

eral

Hea

lth

Que

stio

nnai

re (G

HQ

-28

)

27 c

areg

iver

s of

pe

ople

with

prim

ary

m

alig

nant

bra

in

tum

ours

(PM

BT)

(onl

y 8

of th

ese

used

hom

e ca

re) f

rom

200

3-20

09

Des

ign

Cas

e co

ntro

l

RC

T

Coh

ort s

tudy

Art

icle

Sin

ger e

t al

(200

5)

Jour

nal o

f Pai

n an

d S

ympt

om

Man

agem

ent

Wal

sh e

t al

(200

7)

Brit

ish

Jour

nal o

f P

sych

iatry

Was

ner e

t al

(201

3)

Jour

nal o

f Soc

ial

Wor

k in

EoL

&

Pal

liativ

e C

are


133

Wea

knes

ses

Ther

e w

as li

mite

d lit

erat

ure

on n

on-c

ance

r co

nditi

ons

and

the

care

givi

ng in

form

atio

n ne

eds

of b

lack

and

m

inor

ity e

thni

c po

pula

tions

. Ove

rall,

the

evid

ence

bas

e w

as

pred

omin

antly

de

scrip

tive

and

dom

inat

ed b

y sm

all-

scal

e st

udie

s, li

miti

n ge

nera

lizab

ility

.

May

not

be

gene

raliz

able

to

hos

pice

car

e- th

ese

peop

le re

ceiv

ed c

are

in

thei

r h

omes

. R

etro

spec

tive;

re

call

bias

. Hom

ogen

ous

popu

latio

n;

gene

raliz

abili

ty.

Can

cer n

ot n

eces

saril

y te

rmin

al.

Res

ults

The

evi

denc

e w

as s

trong

est i

n re

latio

n to

pai

n m

anag

emen

t, w

here

in

adeq

uaci

es in

car

egiv

er k

now

ledg

e an

d th

e im

porta

nce

of e

duca

tion

wer

e em

phas

ized

. The

sig

nific

ance

of

effe

ctiv

e co

mm

unic

atio

n an

d in

form

atio

n sh

arin

g be

twee

n pa

tient

, ca

regi

ver a

nd s

ervi

ce p

rovi

der w

as

also

em

phas

ized

. The

evi

denc

e fo

r ot

her c

areg

iver

kno

wle

dge

and

info

rmat

ion

need

s, fo

r exa

mpl

e in

re

latio

n to

wel

fare

and

soc

ial s

uppo

rt,

was

wea

ker.

Five

them

es w

ere

iden

tifie

d in

the

da

ta in

clud

ing

diffi

culty

with

ad

min

istra

tion

of p

ain

med

icin

es,

conc

erns

abo

ut s

ide

effe

cts

of

med

icat

ions

, ins

ecur

ity w

ith p

ain

asse

ssm

ent,

frust

ratio

ns w

ith

com

mun

icat

ion

amon

g he

alth

car

e te

am m

embe

rs, a

nd m

emor

ies

of

unre

lieve

d pa

in. G

uilt

if fe

lt th

at d

eath

w

as re

late

d to

ove

r med

icat

ion.

Ove

r hal

f of c

areg

iver

s (5

5%) c

ared

fo

r a p

atie

nt w

ith m

etas

tatic

dis

ease

, se

vere

com

orbi

dity

, or u

nder

goin

g cu

rren

t tre

atm

ent.

Bes

ides

ass

istin

g w

ith a

ctiv

ities

of d

aily

livi

ng, c

areg

iver

s pr

ovid

ed c

ance

r-sp

ecifi

c ca

re s

uch

as

wat

chin

g fo

r tre

atm

ent s

ide

effe

cts

(68%

), he

lpin

g m

anag

e pa

in, n

ause

a or

fat

igue

(47%

), ad

min

iste

ring

med

icin

e (3

4%),

deci

ding

whe

ther

to

call

a do

ctor

(30%

), de

cidi

ng w

heth

er

med

icin

e w

as n

eede

d (2

9%),

and

chan

ging

ban

dage

s (1

9%).

How

ever

, ha

lf of

car

egiv

ers

repo

rted

not g

ettin

g tra

inin

g pe

rcei

ved

as n

eces

sary

.

Out

com

e

Stu

dy

co

nclu

sion

s re

gard

ing

the

focu

s

Bar

riers

to p

ain

man

agem

ent

Bur

den

and

reso

urce

s of

ca

regi

vers

Met

hods

Incl

uded

pee

r-re

view

ed jo

urna

l ar

ticle

s pr

esen

ting

orig

inal

re

sear

ch. A

val

idat

ed s

yste

mat

ic

revi

ew m

etho

dolo

gy fo

r ass

essi

ng

disp

arat

e ev

iden

ce w

as u

sed

in

orde

r to

assi

gn s

core

s to

diff

eren

t as

pect

s of

eac

h st

udy

(intro

duct

ion

and

aim

s, m

etho

d an

d da

ta, s

ampl

ing,

dat

a an

alys

is,

ethi

cs a

nd b

ias,

find

ings

/resu

lts,

trans

fera

bilit

y/ge

nera

lizab

ility

, im

plic

atio

ns a

nd u

sefu

lnes

s).

Sem

i-stru

ctur

ed in

terv

iew

s w

ith

146

care

give

rs p

rovi

ded

data

for

the

stud

y. R

espo

nses

to s

even

qu

estio

ns a

skin

g fo

r a ra

nkin

g of

en

d-of

-life

pai

n m

anag

emen

t in

dica

ted

a le

ss th

an id

eal

expe

rienc

e. A

vaila

ble

narr

ativ

es

from

38

care

give

rs w

ere

anal

yzed

fo

r the

mes

rela

ted

to fu

rther

un

ders

tand

ing

of th

e co

ncer

ns.

Car

egiv

ers

of s

even

ge

ogra

phic

ally

and

inst

itutio

nally

de

fined

coh

orts

of n

ewly

di

agno

sed

colo

rect

al a

nd lu

ng

canc

er p

atie

nts

com

plet

ed s

elf-

adm

inis

tere

d qu

estio

nnai

res

(n56

77).

We

com

bine

d th

is

info

rmat

ion

with

pat

ient

sur

vey

and

char

t abs

tract

ion

data

and

fo

cuse

d on

car

egiv

ers

who

re

porte

d pr

ovid

ing,

unp

aid,

at

leas

t 50%

of t

he p

atie

nt’s

in

form

al c

ance

r car

e.

Focu

s

Kno

wle

dge

and

info

rmat

ion

need

s

of i

nfor

mal

ca

regi

vers

in

palli

ativ

e ca

re.

Car

egiv

er

expe

rienc

e w

ith

pa

in m

anag

emen

t.

Bur

den,

reso

urce

s an

d st

ress

ors

am

ong

info

rmal

ca

ncer

car

egiv

ers.

Setti

ng

Inte

rnat

iona

l

Mid

wes

tern

US

7 U

S re

gion

s

Sam

ple

34 s

tudi

es

146

care

give

rs o

f pe

ople

in h

ospi

ce

hom

e ca

re

677

prim

ary

care

give

rs o

f pe

ople

with

te

rmin

al c

ance

r

Des

ign

Sys

tem

atic

re

view

Coh

ort s

tudy

Coh

ort

stud

y

Art

icle

Doc

herty

et

al

(200

8)

Pal

liativ

e M

edic

ine

Par

ker O

liver

et

al (2

014)

Jour

nal o

f Pai

n an

d S

ympt

om

Man

agem

ent

Ryn

et a

l (20

11)

Psy

cho-

Onc

olog

y


134

Appendix D Detailed characteristics of palliative oncology papers

Wea

knes

ses

The

dura

tion

and

dept

h of

pal

liativ

e ca

re in

this

set

ting

is

atyp

ical

of o

ncol

ogy

prac

tices

in th

e U

S

(gen

eral

izab

ility

). R

ural

but

aca

dem

ic

canc

er c

ente

rs

Clin

icia

ns a

nd

patie

nts

are

et

hnic

ally

an

raci

ally

ho

mog

enou

s

Mis

sing

dat

a m

ay

lead

to b

ias.

Var

iabl

es n

ot

com

plet

ely

inde

pend

ent

(end

ogen

eity

).

Ret

rosp

ectiv

e st

udy-

as

sess

or b

ias.

Non

rand

om.

Eth

nica

lly

hom

ogen

ous.

Onl

y th

e in

fo fo

r th

at h

ospi

tal.

Res

ults

Sel

f-ass

essm

ent c

ompr

ised

4 th

emes

:

(1) t

reat

ing

the

who

le p

atie

nt, (

2)

focu

sing

on

qual

ity v

ersu

s qu

antit

y of

life

, (3

) “so

me

patie

nts

just

wan

t to

fight

,” an

d (4

) hel

ping

with

tran

sitio

ns; t

imin

g is

ev

eryt

hing

. 5 th

emes

com

pris

ed v

iew

s

on th

e ro

le o

f pal

liativ

e ca

re: (

1) “r

efer

ea

rly a

nd o

ften,

” (2)

refe

rral

cha

lleng

es:

“Pal

liativ

e” e

qual

s “h

ospi

ce”;

“Hem

e pa

tient

s ar

e di

ffere

nt,”

(3) p

allia

tive

care

as

con

sult-

ants

or c

o-m

anag

ers,

(4)

palli

ativ

e ca

re “s

hare

s th

e lo

ad,”

and

(5)

EN

AB

LE II

faci

litat

ed in

tegr

atio

n.

Mea

n ag

e at

dea

th w

as 6

2.4

year

s (r

ange

24–

83).

Mea

n du

ratio

n of

tre

atm

ent w

as 1

7.9

mon

ths

(ran

ge 1

–12

9). 4

7 pa

tient

s re

ceiv

ed a

ggre

ssiv

e E

oL c

are

whi

ch w

as s

trong

ly a

ssoc

iate

d w

ith h

ospi

tal d

eath

. 15

patie

nts

havi

ng

used

pal

liativ

e ca

re s

ervi

ces

or d

ying

in

a pa

lliat

ive

care

uni

t (P

CU

) had

few

er

sym

ptom

s an

d in

terv

entio

ns a

t the

EoL

. H

avin

g ad

dres

sed

EoL

issu

es w

as

corr

elat

ed w

ith fe

wer

pro

cedu

res

durin

g th

e la

st 3

day

s.

Mos

t pat

ient

s w

ere

whi

te (7

6.9%

) and

had

ov

aria

n ca

ncer

(56.

7%).

155

(57.

8%)

unde

rwen

t ant

i-can

cer t

hera

py w

ith

chem

othe

rapy

, 19

(7.1

%) w

ere

treat

ed

with

radi

atio

n, a

nd 1

7 pa

tient

s (6

.3%

) un

derw

ent b

oth.

218

pat

ient

s (8

1.3%

)

had

at le

ast o

ne a

dmis

sion

(ran

ge 0

–14)

. Th

e m

ost c

omm

on re

ason

for a

dmis

sion

w

as g

astro

inte

stin

al c

ompl

aint

s (3

7.1%

) an

d pr

oced

ures

(18.

3%).

157

(58.

6%)

unde

r-w

ent a

t lea

st o

ne p

roce

dure

dur

ing

the

last

6

mon

ths

of li

fe (r

ange

0–1

1).

Out

com

e

Per

spec

tives

on c

oncu

rren

t on

colo

gy/

palli

ativ

e ca

re

Trea

tmen

t in

tens

ity,

pl

ace

of d

eath

, ut

iliza

tion

Car

e

ut

iliza

tion/

re

ason

for

hosp

ital

adm

issi

ons

Met

hods

Qua

litat

ive

inte

rvie

ws

with

35

onco

logy

cl

inic

ians

abo

ut th

eir

appr

oach

to p

atie

nts

w

ith a

dvan

ced

canc

er

and

the

effe

ct o

f the

E

NA

BLE

(Edu

catio

n,

Nur

ture

, Adv

ise,

Bef

ore

Life

End

s) II

RC

T (b

roug

ht h

ospi

ce

conc

epts

to c

ance

r pa

tient

s ea

rly in

thei

r di

seas

e).

Ret

rosp

ectiv

e ch

art

revi

ew w

as c

ondu

cted

in

univ

ersi

ty c

ance

r clin

ic

dece

dent

s. A

naly

sis

of

plac

e of

dea

th, p

allia

tive

care

util

izat

ion,

prio

r EoL

di

scus

sion

, and

soc

-ial

back

grou

nd w

ith

sym

ptom

bur

den

and

treat

men

t int

ensi

ty

(out

com

e va

riabl

es).

Ret

rosp

ectiv

e ch

art r

evie

w

of p

atie

nts

with

a

diag

nosi

s of

a

gyne

colo

gic

mal

igna

ncy.

A

bstra

cted

dat

a

in

clud

ed d

emog

raph

ics,

ad

mis

sion

and

pr

oced

ural

his

tory

, use

of a

nti-c

ance

r the

rapy

, an

d pa

lliat

ive

care

ut

iliza

tion

durin

g th

e la

st

6

mon

ths

of li

fe.

Focu

s

Onc

olog

ist

pers

pect

ive

on

carin

g fo

r ad

vanc

ed,

palli

ativ

e

canc

er p

atie

nts.

Pat

ient

nee

ds,

sym

ptom

s an

d tre

atm

ent

inte

nsity

alo

ng

dise

ase

traje

ctor

y in

m

edic

al

onco

logy

ou

tpat

ient

s.

Pal

liativ

e ca

re

for g

ynec

olog

ic

mal

igna

ncie

s.

Setti

ng

Leba

non,

N

ew

Ham

pshi

re

(NC

-CC

) &

Whi

te R

iver

Ju

nctio

n,

Ver

mon

t, U

S

(bot

h ru

ral)

Erla

ngen

-N

ümbe

rg,

Ger

man

y

Ala

bam

a, U

S

Sam

ple

35 o

ncol

ogy

clin

icia

ns

from

NC

CC

an

d V

AM

C

wor

king

in

an in

tegr

ated

pa

lliat

ive

onco

logi

c,

team

set

ting

96

dece

dent

s dy

ing

btw

20

09 a

nd

2011

from

ca

ncer

268

de

cede

nts

w

ith

gyne

colo

gic

canc

ers

btw

20

07-2

010

Des

ign

Qua

litat

ive

Obs

erva

tion

al, c

ross

-se

ctio

nal

Cha

rt re

view

, cr

oss-

se

ctio

nal

Art

icle

Bak

itas

et a

l (2

013)

Pal

liativ

e

and

Sup

porti

ve

Car

e

Buk

ki e

t al

(201

3)

Sup

port

C

are

Can

cer

Fauc

i et a

l (2

012)

Gyn

ecol

ogic

O

ncol

ogy


135

Wea

knes

ses

Mer

its o

f EI g

ener

ally

w

idel

y re

cogn

ized

,

but t

hese

spe

cific

S

OP

s no

t nec

essa

rily

(they

are

new

)

Eng

lish

lang

uage

bi

as?

Trie

d to

avo

id

the

othe

r sys

tem

atic

re

view

bia

ses

Res

ults

SO

Ps

wer

e de

velo

ped

for

19 m

alig

nanc

ies

(a) t

o id

entif

y a

dise

ase-

spec

ific

poin

t in

each

dis

ease

tra

ject

ory

to in

itiat

e E

I (“

gree

n fla

gs”)

and

to

prov

ide

(b)

a c

lear

de

linea

tion

and

sem

antic

di

ffere

ntia

tion

of P

C

assi

gnm

ents

[“pa

lliat

ive

care

” v

s. “s

uppo

rtive

” or

“pal

liativ

e th

erap

ies”

(“gr

een”

vs

. “re

d fla

gs”)

]. P

. 103

9

for S

OP

gre

en fl

ags.

Ther

e is

no

stro

ng

evid

ence

that

any

regi

men

gi

ves

grea

ter p

allia

tion.

H

ighe

r dos

e re

gim

ens

give

m

ore

acut

e to

xici

ty,

espe

cial

ly e

soph

agiti

s.

Ther

e is

evi

denc

e fo

r a

mod

est i

ncre

ase

in

surv

ival

(5%

at 1

yea

r and

3%

at 2

yea

rs) i

n pa

tient

s w

ith b

ette

r per

form

ance

st

atus

(PS

) giv

en h

ighe

r do

se ra

diot

hera

py.

Som

e re

gim

ens

are

asso

ciat

ed

with

an

incr

ease

d ris

k of

ra

diat

ion

mye

litis

.

Out

com

e

Sta

ndar

d op

erat

ing

proc

edur

e (S

OP

)

for s

peci

fic

mal

igna

ncie

s

RC

T fin

ding

s

Met

hods

A w

orki

ng g

roup

(a) s

peci

fy th

e tim

ing

of e

arly

inte

grat

ion

and

(b) s

peci

fy P

C a

ssig

nmen

ts b

y (c

) pro

vidi

ng m

ore

clea

r-cu

t se

man

tic a

nd c

linic

al

defin

ition

s.

Ann

ual u

pdat

e of

trea

tmen

t gu

idel

ines

(SO

P) f

or e

ach

mal

igna

ncy,

the

need

for

dise

ase-

spec

ific

EI S

OP

s w

as

iden

tifie

d.

The

elec

troni

c da

ta- b

ases

M

ED

LIN

E, E

MB

AS

E, C

ance

rlit

and

the

Coc

hran

e C

entra

l R

egis

ter o

f Con

trolle

d Tr

ials

, re

fere

nce

lists

, han

d-se

arch

ing

of jo

urna

ls a

nd

conf

eren

ce

proc

eedi

ngs,

and

dis

cuss

ion

w

ith e

xper

ts w

e-re

use

d to

id

entif

y po

tent

ially

elig

ible

tri

als,

pub

lishe

d an

d un

publ

ishe

d.

Focu

s

Dis

ease

spe

cific

ap

proa

ch to

pa

lliat

ive

canc

er

care

.

NS

CLC

pal

liativ

e ra

diot

hera

py.

Setti

ng

Col

ogne

, G

erm

any

Inte

rnat

iona

l

Sam

ple

Inte

rdis

cipl

inar

y w

orki

ng g

roup

(PC

, on

colo

gy,

radi

othe

rapy

, etc

.)

14 R

CTs

re: p

allia

tive

radi

othe

rapy

in

NS

CLC

Des

ign

Qua

litat

ive

Sys

tem

atic

re

view

Art

icle

Gae

rtner

et a

l (2

011)

Sup

port

Car

e C

ance

r

Lest

er e

t al

(201

2)

Coc

hran

e Li

brar

y


136

Wea

knes

ses

Non

-sys

tem

atic

(Sel

ectio

n bi

as?

Pub

licat

ion

bias

? E

nglis

h la

ngua

ge b

ias?

)

Not

clin

ical

ly s

peci

fic a

s th

e tit

le

wou

ld im

ply.

Non

rand

om.

Rec

all/s

ocia

l des

irabi

lity

bias

es?

Per

form

ed a

t a s

ingl

e, te

rtiar

y ca

re s

ite w

ith a

spe

cial

ized

gr

oup

of th

orac

ic o

ncol

ogy

prov

ider

s an

d pa

lliat

ive

care

cl

inic

ians

, lim

iting

ge

nera

lizat

ion.

Sam

ple

lack

ed

dive

rsity

with

resp

ect t

o ra

ce

and

ethn

ic g

roup

. Did

not

den

y pa

lliat

ive

care

con

sulta

tions

to

parti

cipa

nts

rece

ivin

g st

anda

rd

care

, and

a s

mal

l min

ority

of

patie

nts

in th

e st

anda

rd c

are

grou

p w

ere

seen

by

the

palli

ativ

e ca

re te

am.

Res

ults

Cur

rent

evi

denc

e su

gges

ts th

at p

atie

nts

with

ha

emat

olog

ical

mal

igna

ncie

s ac

cess

palli

ativ

e ca

re s

ervi

ces

less

freq

uent

ly. F

or

thos

e w

ho d

o, it

tend

s to

occ

ur la

ter i

n th

eir

illne

ss th

an th

eir s

olid

tum

our c

ount

erpa

rt-s.

Th

ese

patie

nts

are

mor

e lik

ely

to d

ie in

ho

spita

l fo

llow

ing

esca

latin

g in

terv

entio

ns.

An

epis

odic

app

roac

h ac

cord

ing

to n

eeds

ra

ther

than

pro

gnos

is m

ay b

e m

ore

valu

able

, as

hae

mat

olog

ic p

atie

nts

intro

duce

diff

icul

ty

in p

rogn

ostic

atio

n; e

chni

cal n

atur

e an

d co

mpl

icat

ions

of t

reat

men

t; sp

eed

of c

hang

e

to a

term

inal

eve

nt; n

eed

for p

atho

logy

test

ing

and

trans

fusi

on o

f blo

od p

rodu

cts

as

deat

h ap

proa

ches

;.

Nur

sing

insi

ghts

indi

cate

that

an

unde

rsta

ndin

g of

end

-of-l

ife c

are

in

haem

atol

ogy

need

s to

be

set i

n a

trilo

gy o

f ov

erla

ppin

g m

odel

s (la

bele

d fu

nctio

nal,

evol

ving

, and

refra

ctor

y) th

at a

ddre

ss th

e co

mpl

exity

of i

ssue

s as

soci

ated

with

pr

ofes

sion

al a

nd h

ospi

tal c

ultu

re.

Of t

he 1

51 p

atie

nts

who

und

erw

ent

rand

omiz

atio

n, 2

7 di

ed b

y 12

wee

ks a

nd 1

07

(86%

of t

he re

mai

ning

pat

ient

s) c

ompl

eted

as

sess

men

ts. P

atie

nts

assi

gned

to e

arly

pa

lliat

ive

care

had

a b

ette

r qua

lity

of li

fe th

an

did

patie

nts

assi

gned

to s

tand

ard

care

. In

addi

tion,

few

er p

atie

nts

in th

e pa

lliat

ive

care

gr

oup

than

in th

e st

anda

rd c

are

grou

p ha

d de

pres

sive

sym

ptom

s (1

6% v

s. 3

8%, P

=

0.01

). D

espi

te th

e fa

ct th

at fe

wer

pat

ient

s in

th

e ea

rly p

allia

tive

care

gro

up th

an in

the

stan

dard

car

e gr

oup

rece

ived

agg

ress

ive

end-

of-li

fe c

are

(33%

vs.

54%

, P =

0.0

5),

med

ian

surv

ival

was

long

er.

Out

com

e

Pal

liativ

e ca

re

co

ncer

ns s

peci

fic to

ha

emat

olog

ic

patie

nts

Nur

se

pe

rspe

ctiv

e

QoL

Met

hods

Lite

ratu

re re

view

Rec

ent h

aem

atol

ogy

clin

ical

gu

idel

ines

reco

mm

end

that

pa

lliat

ive

care

spe

cial

ists

sh

ould

hav

e ce

ntra

l rol

es in

ha

emat

o-on

colo

gy te

ams.

25

nurs

ing

inte

rvie

ws

durin

g tw

o-ye

ar q

ualit

ativ

e st

udy.

Ran

dom

ly a

ssig

ned

patie

nts

to

rece

ive

early

pal

liativ

e

care

inte

grat

ed w

ith s

tand

ard

onco

logi

c c

are

or s

tand

ard

onco

logi

c ca

re a

lone

. QoL

as

sess

ed a

t bas

elin

e an

d 12

w

eeks

late

r usi

ng F

unct

iona

l A

sses

smen

t of C

ance

r Th

erap

y-Lu

ng (F

AC

T-L)

and

H

AD

S.

Focu

s

Hae

mat

olog

ical

pa

tient

s in

pa

lliat

ive

care

.

Pal

liativ

e ca

re fo

r ha

emat

olog

ic

mal

igna

ncie

s.

Con

curr

ent

palli

ativ

e an

d on

colo

gic

treat

men

t fo

r pat

ient

s w

ith

NS

CLC

Setti

ng

Inte

rnat

iona

l

Aus

tralia

Uni

ted

S

tate

s

Sam

ple

N/A

25 n

urse

s

151

patie

nts

with

m

etas

tatic

N

SC

LC

Des

ign

Nar

rativ

e

revi

ew

Qua

litat

ive

RC

T

Art

icle

Man

itta

et a

l (2

010)

Jour

nal o

f P

allia

tive

Med

icin

e

McG

rath

et a

l (2

007)

Onc

olog

y

Nur

sing

For

um

Tem

el e

t al

(201

0)

New

Eng

land

Jo

urna

l of

Med

icin

e


137

Appendix E Pearson Correlation

Pearson’s product-moment correlation coefficient (r) uses the variance and

covariance of two continuous variables to attempt to draw a line of fit between them and

derive their correlation. The sample variance of x is calculated as the average squared

difference from the mean:

𝑣𝑎𝑟 𝑥 = ! !!!! !

!!!, (1)

by subtracting the mean from each observation, squaring the differences (so that

negative and positive deviations do not cancel each other out), summing them and

dividing by N-1. The square root of the variance gives the standard deviation. Sample

covariance of x and y is given by

𝑐𝑜𝑣 𝑥, 𝑦 = !( !!!! !!!! )!!!

. (2)

Pearson’s sample correlation is given by

𝑟!" =!"# !,!

!"# ! !"#(!), (3)

dividing the covariance by the standard deviation of each variable:

𝑟!" =! !!!! !!!!

! !!!! ! ! !!!! !. (4)

This measures the linear relationship between the variables which is why it is only

suitable for continuous variables. Pearson’s r is always between -1 and 1 and can speak

to the direction of the relationship as well. To derive a population correlation, all terms

divided by N-1 are instead divided by 1.


138

Appendix F Phi coefficient and Cramér’s V test

Phi coefficients (mean square contingency coefficients), ∅, measure association

between binary variables and have the same interpretation as the Pearson correlation

(Sanyal et al, 2009). Ranging from -1 to 1, it is the ratio of the Pearson chi-square (x2)

statistic to the total number of observations:

∅ = 𝑥!/𝑁 (or ∅! = !!

!), (1)

where x2 is a measure of association between the two variables. So if we have a 2x2

contingency table for random binary variables x and y,

y = 1 y = 0 Total

x = 1 n1 n4 n7

x = 0 n2 n5 n8

Total n3 n6 n

∅ = !!!!!!!!!!!!!!!!!

(Everitt, 2002). (2)

Cramér’s V test is a “rescaling of phi” to handle tables larger than 2x2 (Cramér,

1946):

𝑉 = 𝑥!/𝑁(𝑘 − 1) , (3)

where N is the number of observations and k is the smaller number of rows and columns

(Sanyal et al, 2009, p. 71). For 2x2 tables, Cramér’s V returns the phi coefficient

because k = 2.


139

Appendix G Correlation coefficients

Phi Caregiver sex Patient sex Site Caregiver sex 1 Patient sex -.5412 1 Site .1748 -.14551 1

Cramér’s V D

iagn

osis

Car

egiv

er

educ

atio

n

Patie

nt

educ

atio

n

Car

egiv

er

mar

ital

Patie

nt

mar

ital

Patie

nt

livin

g

Car

egiv

er

empl

oym

ent

Car

egiv

er

rela

tion

Diagnosis 1 Caregiver education

.224

1

Patient education

.2221 .3568 1

Caregiver marital

.1964 .2748 .1179 1

Patient marital

.2222 .1376 .1724 .3834 1

Patient living

.2201 .1783 .1596 .3744 .5757 1

Caregiver employment

.2287 .2969 .2302 .2783 .2826 .2758 1

Caregiver relation

.2011 .1782 .1999 .333 .5691 .5209 .2406 1

Polychoric (1)

Com

orbi

dity

Car

egiv

er

burd

en

Car

egiv

er

sex

Patie

nt se

x

Site

Em

erge

ncy

visi

ts

Day

s ov

erni

ght

Car

egiv

er

age

Patie

nt a

ge

Comorbidity 1 Caregiver burden

-.0441 1

Caregiver sex .1285 -.1186 1 Patient sex .1701 .2311 -.4263 1 Site -.0075 .2513 .158 -.2763 1 Emergency visits

-.0139 .1274 -.1046 -.0083 -.0195 1

Days overnight

.0359 .168 .0572 .0088 .0834 .3759 1

Caregiver age -.0231 -.019 .0689 .1101 .0116 -.2486 -.0657 1 Patient age .0714 -.018 -.0602 -.0546 .0911 -.0216 -.0278 .1856 1 (ln)Time cost .0769 .2503 -.1276 .0349 .1685 .1002 .0903 .0079 -.1224 (ln)Ministerial cost

.1214 .1924 .0757 -.108 .2251 .4802 .3979 -.1655 .0025

(ln)Societal .0764 .279 -.0499 -.0714 .3111 .4017 .315 -.0769 -.0438


140

cost (ln)Meds -.0399 .1164 .0288 .0406 .0779 .0569 .1121 -.0721 -.0919 Hospitalization .0617 .2409 -.0286 .057 .1445 .6509 .7773 -.1464 -.0841

Polychoric (2) (ln

)Tim

e co

st

(ln) M

inis

teria

l co

st

(ln)S

ocie

tal c

ost

(ln)M

eds

Hos

pita

lizat

ion

(ln)Time cost 1 (ln)Ministerial cost .2932 1 (ln)Societal cost .7358 .7962 1 (ln) meds .0148 .3164 .1851 1 Hospitalization .2204 .953 .888 .1728 1

Pearson O

ut-o

f-po

cket

m

eds

Out

-of-

pock

et

trave

l

Out

-of-

pock

et

supp

lies

Publ

ic m

eds

Publ

ic te

sts

Publ

ic

appo

intm

ents

Publ

ic

hosp

italiz

atio

n

Publ

ic

emer

genc

y ro

om

(ln)S

ocie

tal

(ln)M

inis

teria

l

Out-of-pocket meds

1

Out-of-pocket travel

.0382 1

Out-of-pocket supplies

.0448 .0159 1

Public meds -.0095 .0089 1 Public tests .0519 .2061 .0125 .016 1 Public appointments

.0275 .1884 .0602 .0571 .1401 1

Public hospitalization

.0616 .3737 -.0165 .0189 -.011 .1766 1

Public emerg-ency room

.0211 .17 -.0139 .0582 .0226 .3176 .3378 1

(ln)Societal .0383 .2331 .0893 .1398 .0261 .0845 .3825 .192 1 (ln)Ministerial .025 .2336 .0091 .2229 .0637 .1221 .5032 .2375 .7795 1 (ln)Time cost .1364 .1528 .1084 .0325 -.0038 .007 .1017 .0815 .7966 .2477


141

Appendix H Explanation of lesser-discussed statistics

Statistic Estimator Interpretation

Coefficients GLS7 Log-linear: for a change in x by 1, expect y to change by 100 x β (%). For categorical variables, it is the difference in impact of the x condition on y from the baseline category.

Probit The predicted probability can be calculated using these coefficients. Positive coefficients indicate increasing the probability of y=1 and vice versa.

Poisson For a 1 unit change in x, the difference in the expected count (y) is expected to change by the coefficient on x.

Wald Chi2 All with the same interpretation

This tests the H0 that at least one coefficient is not equal to 0.

Prob > Chi2 All with the same interpretation

This gives the probability of obtaining the Wald Chi2 statistic if the predictor variables had no effect.

P-value All with the same interpretation

The probability of the coefficient under the null hypothesis that the independent variable has no effect. Significant p-values that reject the null hypothesis are generally associated with standard errors > 2.

Σ𝜇 GLS and probit (same interpretation)

The standard deviation of the random effect (panel-level standard deviation); taking the log and squaring it gives you (ln)Σ𝜇2. When Σ𝜇 = 0, the panel component is unimportant and the results are no

7 RE, FE and hybrid RE.


142

different than a pooled model.

Σ𝜀 GLS Standard deviation of 𝜀it.

Rho GLS Rho is the fraction of variance attributable to the 𝜇i term in RE models.

R2 GLS Overall, between and within

Breusch & Pagan GLS Panel GLS considers unobservable heterogeneity across and within individuals over time that cross-sectional analysis cannot capture. Breusch & Pagan tests the H0 that there are no RE, in which case a pooled OLS could do just as adequate a job of analyzing the panel. Rejection of H0 means that GLS is the optimal estimator.

(ln)Σu2 Probit Logarithm of the random effect standard deviation (panel variance).

𝛼 Poisson This is an under/over-dispersion coefficient. Poisson distribution assumes that mean=variance of the dependent variable. Variance greater than the mean is over-dispersion, and the inverse is under-dispersion.

(ln)𝛼 Poisson Logarithm of 𝛼.


143

Appendix I Residuals and fitted values (graphs)

RE Regression on societal cost of CBPHC

RE regression on Ministry of Health cost of CBPHC

RE regression on unpaid caregiver cost of CBPHC

56

78

910

Pred

icted

Val

ues

5 6 7 8 9 10Log-Transformed Societal Cost

Linear prediction Fitted values

Random Effects Regression on Societal Cost: Fitted Values

0.1

.2.3

.4.5

Den

sity

-6 -4 -2 0 2Residuals

Random Effects Regression on Ministerial Cost: Residuals

02

46

810

Pred

icte

d Va

lues

0 2 4 6 8 10Log-Transformed Informal Cost


Random Effects Regression on Informal Cost: Fitted Values

0.2

.4.6

.8De

nsity

-3 -2 -1 0 1 2Residuals

Random Effects Regression on Societal Cost: Residuals

02

46

810

Pre

dict

ed V

alue

s

0 2 4 6 8 10Log-Transformed Ministerial Cost


Random Effects Regression on Ministerial Cost: Fitted Values0

.2.4

.6D

ensi

ty

-6 -4 -2 0 2Residuals

Random Effects Regression on Informa Cost: Residuals