Embed Size (px)
Citation preview
A population based study of lung cancer in Norway –the importance of resection rate and factors associated
with treatment and survival
Yngvar Nilssen, MSc
PhD thesis
Faculty of Medicine
University of Oslo
© Yngvar Nilssen, 2016 Series of dissertations submitted to the Faculty of Medicine, University of Oslo ISBN 978-82-8333-258-2 All rights reserved. No part of this publication may be reproduced or transmitted, in any form or by any means, without permission. Cover: Hanne Baadsgaard Utigard Printed in Norway: 07 Media AS – www.07.no
3
Contents
Acknowledgements ............................................................................................................................... .. 4
List of papers ............................................................................................................................... ............ 6
Abbreviations ............................................................................................................................... ........... 7
Introduction............................................................................................................................... .............. 9
Background............................................................................................................................... ........... 9
Incidence, prevalence and mortality................................................................................................... 9
Aetiology............................................................................................................................... ............. 11
Anatomy and histology...................................................................................................................... 13
Treatment............................................................................................................................... ........... 14
Prognostic factors.............................................................................................................................. 16
Survival ............................................................................................................................... ............... 23
Aims of the study............................................................................................................................... .... 25
Material and methods........................................................................................................................... 26
Data sources ............................................................................................................................... ....... 26
Data linkage............................................................................................................................... ........ 28
Classification of variables .................................................................................................................. 29
Statistical methods ............................................................................................................................ 33
Main results............................................................................................................................... ............ 40
Lung cancer survival in Norway (Paper I) .......................................................................................... 40
Lung cancer treatment (Paper II) ...................................................................................................... 41
Resection in relation to survival (Paper III) ....................................................................................... 41
Discussion............................................................................................................................... ............... 42
Overview of results............................................................................................................................ 42
Methodological considerations......................................................................................................... 42
Discussion of the results.................................................................................................................... 50
Conclusion ............................................................................................................................... .............. 58
Future perspectives............................................................................................................................... 59
Errors in published papers .................................................................................................................... 61
References............................................................................................................................... .............. 62
Papers I III............................................................................................................................... ............... 76
4
Acknowledgements
This project was conducted at the Cancer Registry of Norway, Institute of Population based Cancer
Research, in Oslo from 2013 to 2016. I am grateful to everyone who has contributed in any way to
my PhD thesis, but some deserve extra acknowledgement.
First, I would like to thank the Cancer Registry of Norway, led by Director Giske Ursin, for giving me
the opportunity to write this thesis. The working facilities have been excellent. I would also like to
thank the South Eastern Norway Regional Health Authority for understanding the importance of this
work through their generous financial grant.
I would like to thank my main supervisor and initiator of this project, Bjørn Møller, for his
outstanding supervision over the last three years. Through his understanding, patience and ability to
challenge me, as well as, his willingness to share his superb expertise in the field of statistics and
cancer epidemiology, Bjørn has been able to guide me through this period in a way I have found to
be nothing less than perfect. Further, my questions and suggestions have always been received with
a positive attitude, something I have appreciated. I could not have asked for a better supervisor.
Thank you to my project group that consisted of Bjørn Møller, Trond Eirik Strand, Lars Fjellbirkeland,
Kristian Bartnes, Odd Terje Brustugun (co author on Paper II), Xue Qin Yu and Dianne L. O’Connell.
This well organised, multi disciplinary group consisted of senior epidemiologists, clinicians and
statisticians who all contributed to the content of this thesis by providing constructive feedback, all
within reasonable timeframes.
5
I would also like to thank Cancer Council New South Wales, Australia, for hosting me from October
2014 to September 2015, and particularly Professor Dianne L. O’Connell for granting my wish to work
abroad as a part of my project. I am grateful that I could discuss project challenges and learn from
the expertise and knowledge she and Qin possess. An extra thanks to everyone in the Health Services
Research group. They all welcomed and included me in a way that was overwhelming, warm, and
beyond every expectation I had for my stay.
I want to thank all my fellow PhD students at the CRN for all their contributions throughout the
different phases of my project, for providing a good working environment in our team office, and for
the relaxing chats and breaks.
Thanks to the house statistician Tor Åge Myklebust and the Stata expert Bjarte Aagnes for
outstanding assistance to any and all statistical and Stata related challenges. Their help has been
invaluable.
Finally, I would like to thank to my dearest family and friends for always believing, motivating and
supporting me, even at times when I struggled to see the light at the end of the tunnel.
6
List of papers
Paper I:
Y. Nilssen, T.E. Strand, L. Fjellbirkeland, K. Bartnes, B. Møller. Lung cancer survival in Norway, 1997–
2011: from nihilism to optimism. European Respiratory Journal 2016 Jan;47(1):275 87.
Paper II:
Yngvar Nilssen, Trond Eirik Strand, Lars Fjellbirkeland, Kristian Bartnes, Odd Terje Brustugun, Dianne
L O’Connell, Xue Qin Yu, Bjørn Møller. Lung cancer treatment is influenced by income, education, age
and place of residence in a country with universal health coverage. International Journal of Cancer
2016 Mar 15;138(6):1350 60.
Paper III:
Yngvar Nilssen, Trond Eirik Strand, Lars Fjellbirkeland, Kristian Bartnes, Dianne L O’Connell, Xue Qin
Yu, Bjørn Møller. Resection rates and regional differences in survival: a national population based
study of non small cell lung cancer in Norway. [Submitted]
7
Abbreviations
ALK – Anaplastic lymphoma kinase
CCI – Charlson comorbidity index
CI – Confidence interval
CRN – Cancer Registry of Norway
cTNM – Clinical tumour, node, metastasis
EGFR – Epidermal growth factor receptor
EOD – Extent of disease
Gy – Gray
HR – Hazard ratio
ICD O 3 – International Classification of Disease for Oncology, 3rd Edition
ICD 10 – International Classification of Disease, 10th Edition
MAR – Missing at random
MCAR – Missing completely at random
MNAR – Missing not at random
NPR – Norwegian Patient Register
NSCLC – Non small cell lung cancer
PET – Positron emission tomography
pTNM – Pathological tumour, node, metastasis
SBRT – Stereotactic body radiation therapy
8
SCC – Squamous cell carcinoma
SCLC – Small cell lung cancer
SES – Socioeconomic status
SSB – Statistics Norway
TNM – Tumour, node, metastasis
9
Introduction
Background
Lung cancer is the most common cancer in the world with 1.8 million new cases in 2013, which
accounts for 13% of all new cancer diagnoses (1). There has been a lot of stigmatisation related to
lung cancer, and a nihilistic attitude characterised the field for years (2 5). However, an indication of
an upward trend in survival has been observed in Norway (6). The relationships between prognostic
factors and survival, as well as, predictors and treatment, have previously been reported by studies
that have used selected groups of lung cancer patients, or hospital materials (7 10). Therefore, the
present project was initiated by the Cancer Registry of Norway (CRN), with funding from the South
Eastern Norway Regional Health Authority, to use a national population based material to (i) study
prognostic factors and the improvement in lung cancer survival in Norway, (ii) describe predictors for
treatment for lung cancer and (iii) explore the relationship between resection rates and survival,
while studying geographical variation.
Incidence, prevalence and mortality
In Norway, 3 019 new lung cancer cases (9.5% of all new cancers) were diagnosed in 2014, making
lung cancer the third most common cancer type, after prostate and breast (6). Lung cancer is the
most common cause of death from cancer worldwide, responsible for approximately 1.6 million
deaths in 2013 (1). In Norway, lung cancer was responsible for 2 158 deaths in 2014, which is more
than prostate cancer and breast cancer deaths combined, and it was also responsible for almost as
many years of life lost as colon, breast and prostate cancers combined (6, 11).
10
Figure 1: Age standardised lung cancer incidence (C33 34) in the Nordic countries (excl. Iceland)
among men and women.
Footnote: Swedish rates are not directly comparable to those from the other Nordic countries, since
the Cancer Registry of Sweden does not include information about cancer patients diagnosed based
on death certificate only. Source: NordCan (12, 13).
Historically, the age standardised incidence rates among men with lung cancer in Norway reached a
plateau in the 1990s, which was 10 years and 20 years after the observed peaks in Denmark and
Finland, respectively (Figure 1). For women, the overall incidence is still increasing and the trend in
Norway is similar to that in Denmark, but steeper than those observed in Sweden and Finland. When
comparing Nordic lung cancer incidence trends stratified by age groups, a decreasing trend was
observed in women under 65, and men under 80, while the incidence continued to increase among
older female patients (12, 13). A recent study also showed that large differences in age specific lung
0
10
20
30
40
50
60
70
80
90
rate
per
100
000
1943
1953
1963
1973
1983
1993
2003
2013
Year of diagnosis
Men
0
10
20
30
40
50
60
70
80
90
rate
per
100
000
1943
1953
1963
1973
1983
1993
2003
2013
Year of diagnosis
Women
Sweden
Norway
Finland
Denmark
11
cancer incidence does exist between counties in Norway (14). Since lung cancer is a lethal disease,
the mortality rate follows the incidence rate closely, with an overall estimated mortality to incidence
ratio of 0.91 and 0.85 among Norwegian men and women, respectively (15). At the end of 2014,
there were 6 619 people alive after a lung cancer diagnosis in Norway, and out of these, less than
20% (1 197) were diagnosed more than five years ago (6).
Aetiology
Lung cancer is one of the few cancer types where the aetiology is known for the majority of the
cases. The study by Doll and Hill in 1950 established the association between lung cancer and
smoking (16). It is estimated that approximately 90% of all lung cancer cases are related to smoking
(17). Other well known causes for lung cancer are radon, asbestos and occupational exposures, as
well as, both indoor (e.g. solid fuel combustion, environmental tobacco smoking) and outdoor air
pollution (18).
The risk for lung cancer increases with the number of cigarettes smoked, number of years smoking
and if a person started smoking at an early age (19). The close relation between smoking and lung
cancer can be observed in how the historical incidence of lung cancer follows a similar shaped curve
as the smoking prevalence, with a time lag. While the prevalence of daily smokers among men aged
16–74 in Norway has steadily decreased from 42% in 1973 to 16% in 2012, the decrease among
women did not start until the end of the 1990s (20). The same report showed that the decrease in
smoking prevalence has been largest among people aged between 16–24, regardless of gender (20).
Smoking habits have been shown to vary by region in Norway. The national average of daily smokers
from 2008 to 2012 in Norway was 19%, with the prevalence varying from 14% in the region of Oslo to
12
28% in Finnmark (21). It is also well known that smoking habits are strongly related to socioeconomic
status (SES) in the population. Smoking prevalence among people with an elementary school
education (34%) is four times higher than those with a university degree or similar (8%) in Norway
(Figure 2) (20).
Figure 2: The proportion of Norwegian population aged 16–74 that are current smokers, stratified by
level of education, 1976 2015.
0
10
20
30
40
50
Prop
ortio
n cu
rren
t sm
okin
g (%
)
1975 1980 1985 1990 1995 2000 2005 2010 2015
Year
Low
Intermediate
High
Level of education
13
Figure 3: Anatomy of the respiratory system, showing the trachea and both lungs with their lobes
and airways. Lymph nodes are also illustrated.
Footnote: For the National Cancer Institute © 2006 Terese Winslow, U.S. Govt. has certain rights
Anatomy and histology
The lung is anatomically divided into lobes, three on the right and two on the left side (Figure 3).
From the trachea, there is a main bronchus going into each of the lungs. Lung cancer is characterised
by uncontrolled growth of abnormal cells, which do not develop new healthy lung tissue. According
to histological type, lung cancer is broadly divided into two main groups: non small cell lung cancer
(NSCLC) and small cell lung cancer (SCLC). Both international and Norwegian data report that NSCLC,
with the most common types being adenocarcinomas, squamous cell carcinomas (SCC) and large cell
14
carcinomas, represents 80–85% of all lung cancer diagnoses (22, 23). Large differences in histological
type between genders have also been observed, as 28% and 42% of all lung cancers are
adenocarcinomas in men and women, respectively and 44% of men and 25% of women with lung
cancer have SCC (24, 25). These differences are likely to be caused by an earlier historical
introduction to smoking among men compared with women (26). SCLC is named after the size of the
tumour cells. These tumours are often located at the centre of the lung, and tend to grow and spread
quickly.
Treatment
Possible treatment modalities for lung cancer patients are surgical resection, radiotherapy, and
chemotherapy. Chemotherapy has become more personalised in recent years (27). Combinations of
these modalities are also possible. The treatment decision is based on the histopathological
diagnosis, which may be supplemented by an immunohistochemical and cancer genome (mutational)
examination to obtain a more specific subgroup of histology. In addition, the localisation and spread
of the tumour (using the International Classification of Disease for Oncology [ICD O 3] and clinical
stage I–IVa), the patient’s performance status, presence of comorbidities, as well as, the patient’s
own preferences are considered important when deciding the treatment (27).
Among patients diagnosed between 2010 and 2014 in Norway, 44% were diagnosed with distant
spread (6). An advanced stage of disease reduces the possibility for curative treatment and the
probability of achieving cure. Patients diagnosed with NSCLC in stage I or II are candidates for surgery
with curative intent. It is known that approximately 20% of lung cancer patients in Norway are
resected every year (23, 29). Adjuvant platinum based chemotherapy is offered for patients in stage
a Classified according to Staging Manual in Thoracic Oncology into stage I: (T1, N0, M0), stage II: (T2, N0, M0),stage III: (T1, T2, N1, M0) or (T1, T2, N2, M0) or (T3, N0, N1, N2, M0), and stage IV: (T4, any N, M0) or (any T,N3, M0) or (any T, any N, M1) (28).
15
II. The majority of resections performed are (bi ) lobectomies, where one or two of the lobes are
surgically removed. Other resection alternatives include pneumonectomy and sub lobar resection.
Neoadjuvant or adjuvant radiotherapy, i.e. radiotherapy given before or after surgery, respectively,
can also be used, if required. If the patient is not considered a surgical candidate due to technical or
medical inoperability, either traditional radiotherapy or stereotactic body radiation therapy (SBRT),
can be offered. During SBRT, high dose radiation is directly aimed at the tumour from multiple
angles. Compared to traditional radiotherapy, the benefits of SBRT include better preservation of the
normal tissue surrounding the tumour and improved survival. However, this technique is only
possible in N0 situations, i.e. no lymph node metastases. Due to the heterogeneity of patients
diagnosed in stage III, the differentiation between treatment decisions is based on T and N stage.
While the debate continues as to whether or not patients with stage III disease and limited N2
metastases should undergo surgical resection, Norwegian national guidelines recommend that this
group of patients receive radiotherapy in concomitant combination with chemotherapy.
For stage IV patients, treatment should be given with life prolonging or palliative intent, i.e.
chemotherapy, palliative radiotherapy, or a combination of the two. Cytostatic treatment with
cisplatin or carboplatin, in combination with vinorelbine, gemcitabine, docetaxel, paclitaxel or
pemetrexed, is considered standard first line therapy in patients with no genetic aberrations
amenable for targeted therapy. Approximately 10% of patients harbour mutations in the epidermal
growth factor receptor (EGFR) gene and should be treated with first line erlotinib, gefitinib or
afatanib. Crizotinib is the treatment of choice in the approximately 5% of patients with changes in
the anaplastic lymphoma kinase (ALK) gene. These targeted therapies are all oral drugs, containing a
kinase inhibitor, and are taken daily as long as they provide benefit to the patient. All patients will
inevitably relapse, and approved second line therapies include docetaxel or pemetrexed for non
mutated patients. Novel kinase inhibitors are preferable for patients with known mutations.
16
Mainly SCLC patients with stage I (<10%) are considered candidates for surgery with adjuvant
chemotherapy, given as with four courses of cisplatin in combination with etoposide. All resected
SCLC patients should also undergo prophylactic brain irradiation. For all other SCLC patients (>90%),
surgery is considered futile. For non resectable patients with limited spread of disease, four courses
of cisplatin and etoposide should be given, with 3 weeks of radiotherapy intercalated between
courses two and three. For SCLC patients with extended disease and good general health, four
courses of platinum and etoposide are recommended, while for patients with reduced general
health, four courses of doxorubicin, cyclophosphamide and vincristine are recommended.
Prognostic factors
Prognostic factors (or predictors) are defined as variables that can account for some of the
heterogeneity in the course of the disease and the ultimate outcome for the patients (30, 31).
Understanding these factors may help answer a wide range of important questions, such as
predicting the outcome or prognosis for individual patients, and providing information about possible
differences in the quality and health care services provided between sub groups of patients (32). For
the purpose of this thesis, prognostic factors are divided into three different groups: tumour ,
patient and treatment related factors. Tumour related factors are those directly related to the
biological aspects of the tumour, such as extent of disease (EOD) and histology. Patient related
factors are not directly related to the cancer, but more specific to the individual, such as gender, SES,
comorbidity, smoking status and area of residence. Treatment related factors include the treatment
modality, procedure and the expertise of the clinician (33). While it is possible to quantify tumour
and patient related factors and whether or not patients receive surgery, other factors that are
associated with quality of treatment and personal experiences, cannot as easily be measured.
17
Tumour related
Extent of disease
Tumour stage is an important prognostic factor, as the 5 year survival rates range from 50% (stage
IA) to 2% (stage IV), and from 73% (stage IA) to 13% (stage IV), according to clinical tumour, node,
metastasis (cTNM) and pathological tumour, node, metastasis (pTNM), respectively (34). A study
from 2012 including lung cancer patients from Australia, Canada, Denmark, Norway, Sweden and the
UK, showed significant differences in stage distribution between the countries, as well as, significant
survival differences between the different stages (35).
Histology
A number of studies have examined the association between histological group and survival. A study
from 2012 showed that histology may be an independent prognostic factor (36). While a Norwegian
study considering all lung cancer patients found no difference between SCC and adenocarcinoma,
they found a 12% increased relative risk of death when comparing large cell carcinoma with
adenocarcinoma (37). Another Norwegian study, only including resected patients, found a 44% and
33% increased mortality when comparing adenocarcinoma and large cell carcinoma, respectively,
with SCC (38). Results from the International Association for the Study of Lung Cancer (IASLC)
identified histology to be an independent prognostic factor among resected patients (39). While
large cell carcinoma was associated with a 19% increased mortality compared to SCC, a difference
only observed among men, the results slightly favoured SCC compared to adenocarcinoma (hazard
ratio [HR] = 0.86, p<0.0001). A study using population based data from seven regions in Spain,
reported varying 5 year relative survival estimates with histological type, favouring SCC and
adenocarcinoma, while SCLC patients had the worst prognosis (40). Therefore, histological subgroup
should be considered a potential prognostic factor when studying survival.
18
Patient related
Socioeconomic status
One of the patient related prognostic factors, SES, is intended to measure a person’s social position
(41). SES is difficult to measure and therefore, income, education, marital status and/or area of
residence are commonly used as proxies. In Norway, individual level information regarding both
income and education is available from Statistics Norway, however, there may be situations where
agreement between these measures and a person’s SES is poor. For example, patients who are not
working would be registered with a low income and hence categorised with a low SES. However,
these patients may have partners who are working and who can financially support both. Hence, the
patient’s personal income as a proxy for SES would cause a misclassification, while including
information about the total household income would be more appropriate.
A recent systematic review and meta analysis showed that lung cancer patients with high SES are
more likely to receive both surgery and chemotherapy, while the influence of SES on whether the
patient receives radiotherapy remains inconclusive (42). These results were found both in SCLC and
NSCLC patients (42 52). When reviewing literature on the relationship between SES and survival, it
was reported that these survival estimates were ambiguous (53). However, studies from Sweden,
Denmark and England have reported that the survival of lung cancer patients is affected by the
patients’ SES (43 45, 49, 54, 55). Differences in lifestyle, culture and behaviour (e.g. smoking habits),
may also be related to SES, and these may further influence the patients’ health. Similar to countries
like Sweden, Denmark and England, Norway has a universal healthcare system where healthcare is
equally available to everyone independent of social factors and area of residence. This contrasts with
the insurance based system in the US, and therefore, care should be taken when comparing the SES
estimates of Norway, with those from non universal healthcare systems, where the differences
between SES groups are expected to be larger.
19
Smoking status
In addition to being the dominating aetiological factor, smoking has been studied in relation to
cancer recurrence and survival. A review article from 2013 showed that patients who continue to
smoke after a lung cancer diagnosis are 1.9 times more likely to get a recurrent tumour, 2.3 times
more likely to get a second tumour, and have 2.9 times higher overall mortality than patients who
quit smoking at the date of diagnosis (56). This review also found that for patients receiving palliative
treatment, smoking cessation at time of diagnosis was associated with improved pulmonary function,
weight gain and better overall quality of life. While there are some inconsistencies in reported
results, a systematic review and meta analysis showed that among early stage NSCLC patients,
continuing smoking compared with smoking cessation was associated with almost a 3 fold increase in
all cause mortality, while the comparable increase for limited SCLC was almost 2 fold (57 60). Other
studies have shown that there is a significant positive effect on survival of being a never smoker
compared to an ever smoker, with estimates varying from a 5–50% reduction in risk of death (61,
62). However, among patients with stage II and III, the results are inconclusive (63, 64).
Comorbidity
As a large proportion of the lung cancer patients are current or former smokers, they are more likely
to have reduced lung function, in addition to a number of other conditions (65). These other
conditions are called comorbidities, and are often summarised into a score or index used in
epidemiological studies. The most commonly used comorbidity index is called the Charlson
comorbidity index (CCI), which gives a score to 17 different chronic diseases based on their severity
(66). This score includes chronic obstructive pulmonary disease, the most common comorbidity for
lung cancer, as well as, myocardial infarction, congestive heart failure, peripheral vascular disease,
cerebrovascular disease, dementia, rheumatic disease, peptic ulcer disease, mild liver disease,
20
diabetes with and without chronic complications, hemiplegia or paraplegia, renal disease, cancer,
moderate or severe liver disease, metastatic cancer and AIDS/HIV (66, 67). From national and
international guidelines, it is known that information about the patient’s general health, lung
function and comorbidities are important when making a treatment decision. While it would be
optimal to have information on all three factors, comorbidity information which often serves as a
proxy of general health, is most accessible in large national population based materials. One study
showed that compared with patients who have a low level of comorbidity, patients with a high level
had a 35% increased 1 year mortality and a 26% increased 5 year mortality (68). A literature review
on the association between comorbidity and lung cancer survival found that having more
comorbidities was associated with a 10 50% increased mortality (69). However, as the prognosis for
lung cancer is considered poor, it has been shown that having comorbidities is of relatively low
prognostic importance. (70, 71).
Symptoms
The most common symptoms that lung cancer patients display are progressive shortness of breath,
coughing (blood), chest pain/oppression, hoarseness or loss of voice and pneumonia (72).
Unfortunately, these symptoms are most likely not to be present until the tumour has metastasised
beyond the primary site. Vague symptoms can lead to later patient contact with a doctor, and hence
later diagnosis. There have been studies examining the presence and duration of symptoms in
relation to survival. A Korean study showed that among NSCLC patients there was a significant
reduction in the risk of death (odds ratio [OR]: 0.23 95% confidence interval [CI] 0.22–0.52) for
asymptomatic compared to symptomatic patients at the time of diagnosis, while no effect was
observed among SCLC patients (73). In a local centre study from England where they considered 5
year survival among all resected NSCLC patients between 2000 and 2009, there was no significant
difference between asymptomatic and symptomatic patients (74). In India they found that patients
who had symptoms for less than one month before diagnosis had a greater than 50% reduction in
21
the risk of death (HR = 0.44 95% CI 0.26–0.74) during the next 30 months (75). A systematic review
from 2009 and a review article from 2014 identified the presence of symptoms to be a significant
negative prognostic factor for lung cancer outcome (76, 77). However, others have shown that
persons living with symptoms for a short period of time had worse prognosis compared to those
living longer with symptoms. Hence, it is not only the presence of symptoms that is important to
consider, but also its duration.
Gender
Gender differences in regard to survival have been studied extensively, and the results consistently
show that women have a better prognosis than men (78). Previous studies from Norway showed that
women had a 14% and 41% improved survival compared to men, when analysing all lung cancer
patients and resected patients, respectively (38, 79). The latter result is comparable to a study from
the United States (US) that reported a 50% lower 30 day post operative mortality among women
compared with men (80). Other studies examined the gender differences among NSCLC and SCLC
patients, separately, and found a 15–20% higher survival among women in both groups (81 83). A
Polish study found that men had a 15% increased risk of death compared to women (84). Another
study of NSCLC patients found better overall survival as well as stage specific survival among women
(85). Unless these results are (residually) confounded by factors that could not be adjusted
(sufficiently) for, it is important to adjust for gender as a prognostic factor. An example of a possible
confounder is smoking, as it is well known that smoking habits differ between men and women.
Therefore, it is important to interpret the previous results with care, and keep in mind that there
may be other associations disguised as a gender difference.
22
Area of residence
Another patient related prognostic factor is area of residence. The results of a number of European
studies have shown that the probability of getting treatment differs within countries (29, 43, 86 88).
The proportion of patients being resected varied between 15–31% in Norway, 3–18% in England, 13–
24% in Denmark and 8–16% in Ireland. Marked survival differences between regions in Norway,
Sweden and England have been reported (43, 88 90). From England, it is known that the regional
differences in survival decreased between the periods 1991 to 1995 and 2001 to 2006, both among
men and women (91). Even though the geographical differences in overall survival got smaller, two
studies showed that among NSCLC patients there is still significant variation in survival (43, 92).
Hence, area of residence seems to be an important prognostic factor for survival.
Treatment related
The 5 year overall survival among all lung cancer patients is approximately 15%, however, resected
patients experience a 5 year survival ranging from 20% to 80% depending on their stage (93). A
Norwegian population based study, examining prognostic factors among resected patients, showed
that having a more extensive procedure than lobectomy was associated with a worse prognosis (38).
Even though studies showed significant association between resection rates and survival, the optimal
proportion of patients who should be resected has not been found. In addition, the varying
definitions of a resection rate between countries make it difficult to compare results (94 96). For
patients receiving radical radiotherapy in stages I and II, the 5 year survival was 17% according to a
published Cochrane report (97). If left untreated, barely any of these lung cancer patients would be
alive after three years (98).
Further, a strong association between overall survival and regional variation in resection rates has
been reported (43, 92). Results from the Lung Cancer Register of Central Sweden, observed that the
23
risk of death was 20–40% higher in other county centres than the reference centre (88). However,
after adjusting for treatment (surgery, radiotherapy, chemotherapy), county of residence was no
longer considered a prognostic factor.
Survival
Worldwide, while survival after a lung cancer diagnosis has been considered poor for decades, there
have been recent indications of a promising positive trend in survival (99 101). In addition, variation
in survival estimates between countries has been observed (15, 99, 100, 102 107). In 2013 in
Norway, the median survival time was 8.2 and 12.3 months for men and women, respectively (27).
Excluding Iceland due to its small population which would affect the relative survival estimates by
random variation from year to year, there have historically been differences between the Nordic
countries in survival estimates among men and women. The NordCan database shows that for the
period from 2009 to 2013, the 5 year relative survival estimates varied from 10 to 15% and from 16
to 19% among men and women, respectively (Figure 4) (12, 13). The survival estimates were 15% for
men and 19% for women in Norway during this period. Norway and Sweden had the highest survival
estimates, while Denmark and Finland had the lowest, according to the NordCan database (12, 13).
Varying 5 year relative survival has been observed in regions with similar universal healthcare
systems, ranging from 8.7% in England to 20.1% in Manitoba, Canada, for the period from 1995 to
2007 (103). It was argued that the low survival in England can be attributed to the fact that cancer
patients seek contact with their doctors at a later stage of disease (104). The EUROCARE 5 study
estimated the 5 year European mean relative survival of lung cancer to be 13.0%, which was the
poorest of the ten index cancer sites. The CONCORD 2 study, which incorporates worldwide
information from 279 population based registries in 67 different countries including over 25 million
cancer patients, reported a 5 year relative lung cancer survival of <20% all over Europe, 15–19% in
North America and 7–9% in some parts of Asia, in the period from 1995 to 2009 (100).
24
Figure 4: Showing the improvement in 5 year relative survival from 1999–2003 to 2004–2008, as well
as, from 2004–2008 to 2009–2013 in the Nordic countries (excl. Iceland) among men and women.
Source: NordCan (12, 13).
Sweden
Norway
Finland
Denmark
0 5 10 15 20
5−year relative survival (%)
Men
Sweden
Norway
Finland
Denmark
0 5 10 15 20
5−year relative survival (%)
1999−2003 to 2004−2008 2004−2008 to 2009−2013
Women
25
Aims of the study
The aims of the present study were to:
examine changes in survival, and patient , tumour and treatment related factors affecting
survival, among resected and non resected lung cancer patients (Paper I).
identify subgroups of age, gender, SES, histology and treatment in relation to improvement
in survival over the last decade (Paper I).
examine and quantify the association between possible predictors and surgical treatment,
radical radiotherapy or palliative radiotherapy for lung cancer patients (Paper II).
examine if regional variation in survival exists among lung cancer patients in Norway and, if it
does, can the variation be explained by varying resection rates (Paper III).
explore the relationship between survival and resection rate, and investigate whether an
optimal resection rate can be identified (Paper III).
26
Material and methods
Data sources
This national, population based study involves three sources of information: Cancer Registry of
Norway (CRN), Statistics Norway (SSB), and Norwegian Patient Register (NPR). A unique personal
identification number has been assigned to every Norwegian citizen since 1964. This personal
identification number allows linkage of information on all Norwegian citizens across institutions and
national health registries.
Cancer Registry of Norway
It is mandatory for all hospitals, pathology laboratories and general practitioners in Norway to report
all newly diagnosed malignant neoplasms to the CRN. The CRN has data on cancers in Norway dating
back to 1953. The CRN also receives death certificates for all patients with a cancer diagnosis from
the Cause of Death Registry, which is operated by the Norwegian Institute of Public Health. Using the
personal identification number, the CRN is linked monthly to the National Population Register to
update vital status (death or emigration), and three times per year with the NPR to ensure
completeness of cancer cases (6). The quality (i.e. comparability, completeness, validity and
timeliness) of the data in the CRN has been evaluated to be high (108).
Statistics Norway
Statistics Norway was established in 1876 and is a governmental entity that falls under the Ministry
of Finance (109). It is considered to be an independent scientific institution as it decides when and
what to publish. Its main objective is to publish statistics about Norwegian society regarding many
different areas, such as population, health, finance and education. Statistics Norway does not directly
collect data from the population regarding education and income, but receives that information from
other relevant administrative registers. The errors in these data are considered to be negligible (110).
27
Every Norwegian citizen has to declare his income and wealth annually to the tax authorities, who
also collect these data from employers, banks etc. These data are then transferred to Statistics
Norway. In addition to tax files, Statistics Norway also collect various tax free transfers from other
administrative registers, primarily from The Labour and Welfare Administration (110, 111). When
Statistics Norway publishes information about education, it uses information from “Nasjonal
utdanningsdatabase”, “Nasjonal vitnemålsdatabase”, “Helsepersonellregisteret” and
“Utlendingsdatabasen”. For immigrants with unknown education, a small proportion of persons are
directly contacted to collect this information (112). Then SSB classifies the education level based on
the Norwegian Standard Classification of Education (113).
Norwegian Patient Register
The Norwegian Patient Register (NPR) falls under the Norwegian Directorate of Health and is a
national health registry covering all sectors of specialised health care services. Reporting to NPR is
mandatory, and the register includes data on all patients treated in Norwegian government funded
institutions. Personal identification numbers have only been reported to the NPR from 2008
onwards. This enables researchers and health planners to follow the disease trajectory of patients
between sectors and hospitals. In addition, alignment of data and validation with other national
health registries are made feasible. The NPR data consist of three main sources for statistics: visits for
medical treatment for in and outpatients at publicly financed hospitals, private hospitals and private
specialist practices. It is important to have data from all these sources, as the government purchases
medical treatment from private hospitals and private specialist practices as a supplement to services
at the public hospitals. The NPR does not include data on privately financed hospital treatments,
however, in 2008 only around 0.5% of all health care services were provided by these hospitals (114).
The basic data unit in the NPR is hospital visits. However, when a patient is transferred between
wards at the same hospital, the individual data records are aggregated. Each episode of national
28
hospital data contains one or more diagnoses, coded according to the International Classification of
Diseases, 10th Edition (ICD 10) classification.
Data linkage
In this study, the national, population based data registered in the CRN were used to identify all
patients diagnosed with malignant neoplasm of bronchus and lung (ICD 10 code C34) between 1
January 1997 and 31 December 2011 in Norway (n = 34 157). In order to have a homogeneous group
of lung cancer patients, those with tracheal cancer (C33) were excluded (n = 49). If a patient was
registered with multiple tumours, only the first case within ICD 10 group C34 was included. For
example, if a patient was diagnosed with a tumour in one lobe and later diagnosed with another
independent tumour in a different lobe, only the first diagnosis would be included in this study.
Patients registered as dead before diagnosis, and patients whose diagnosis was solely based on
death certificate and autopsy, were excluded from the study (n = 886). Information from Statistics
Norway was linked with indicators for SES, i.e. education, personal and household incomes.
Education was measured by the highest achieved education at the year of diagnosis, while personal
and household incomes were measured the year prior to lung cancer diagnosis. Education and
personal income were available for the whole study period, while household income was only
available after 2003. Finally, information about co existing diseases (i.e. comorbidities) during one
year prior to diagnosis was obtained from the NPR, but this was only available for patients diagnosed
after 1 January 2009.
In Paper I, data on all lung cancer patients registered in the CRN from 1997 to 2011 were used. The
rationale behind starting the study period in 1997 was that this was the first year radiotherapy data
on a national level were available. In Paper II, all lung cancer patients identified in the CRN from 2002
to 2011 were included, as health trusts were defined and came into effect in Norway from January 1
29
2002, and since regional variation was an aim of the study (115). For Paper III, only NSCLC patients
who were diagnosed from 2002 to 2011 were included. Patients with SCLC rarely undergo surgery, so
they were excluded in this study (27). Figure 5 shows the linkage of the different data sources and
shows which patients were included in the different papers.
Figure 5: Flowchart showing the data linkage and study populations for papers I–III.
Classification of variables
The following information was available for the lung cancer patients in the study: date of diagnosis,
date of death or last observation, age, gender, EOD, education, personal income, household income
(after 2004), histology, topography, date and type of treatment (surgery and radical or palliative
radiotherapy), laterality, comorbidity (after 2008), symptom duration and smoking status (2004–
2010). In addition, due to the extensive quality control work done by H. Rostad and T.E. Strand
internally at the CRN, surgical procedure, pTNM, tumour size and resection margin information were
also available for resected patients.
30
Radiotherapy
The CRN receives electronic records from all radiotherapy centres annually, with information about
ICD 10 group, date of treatment, treatment intention, total radiation dose and the number of
fractions. The possible treatment intentions are curative, local control, prophylactic or palliative, and
for the purpose of this thesis the first three intentions were grouped together as radical (116). For
95% of the lung cancer cohort, the intention was known, while for the last 5% of patients, total
radiation dose given was used to categorise them as either radical or palliative. For NSCLC patients
who did not undergo surgery, a total radiation dose over 60 Gray (Gy) was considered as radical, and
the comparable limit for resected patients was 50 Gy. For SCLC patients, radiation doses of 42 Gy or
higher were classified as radical. Any radiation dose lower than those described above, was classified
as palliative.
Histology
For Papers I and III, information about histology was obtained using all information registered in the
CRN and by choosing the most informative and specific subgroup. For Paper II, information about the
most specific histological subgroup registered in the CRN before the time of resection was used for
the resected patients. However, if the patients were not resected, the approach for classifying
histology was the same in Paper II as it was in Papers I and III.
Histology was classified based on the 2004 version of the World Health Organization classification as
SCC, adenocarcinoma, small cell carcinoma, large cell carcinoma, other specified carcinomas,
carcinoma not specified and unknown histology (72). Table 1 shows which ICD O 3 codes that were
included in the different histology groups. Since sarcomas are biologically different from the other
tumours, and because the number of sarcoma patients is very low (0.2%), these tumours were not
included in this study.
31
Table 1: Shows which morphology codes (ICD O 3) are included in the different histological groups.
Histological group: ICD O 3 Codes:Squamous cell carcinoma 8050–8076Adenocarcinoma 8140, 8211, 8230–8231, 8250–8260, 8333, 8341, 8480–8490,
8550, 8560, 8570, 8574
Small cell carcinoma 8040–8045Large cell carcinoma 8012–8031, 8310Other specified carcinoma 8046, 8082, 8123, 8200, 8240, 8244, 8246, 8249, 8430Carcinoma, not specified 8010, 8032, 8033Unknown 6900, 6999, 8000, 8001
Extent of disease
It is important to include information about EOD both for analysing prognosis and likelihood of
treatment for lung cancer patients. In the CRN, the variable describing EOD at the time of diagnosis is
grouped into localised, regional, metastatic or unknown according to the condensed tumour, node,
metastasis (TNM) status (117). Localised tumours are defined as tumours with no direct growth into
neighbouring tissue, lymph nodes or organs, however, there can be micro invasive growth or
carcinoma with the beginning of a microscopically infiltrating tumour. Regional tumours are defined
as tumours with metastasis to regional lymph nodes or microscopic/macroscopic growth into
neighbouring tissue. Metastatic disease involves metastasis to distant lymph nodes or metastasis to
organs in the same/different part of the body as the primary tumour. The EOD is coded as unknown
if a metastasis is found but the location of the primary tumour is uncertain.
Before 2008, EOD was coded as unknown if the coding was solely based on a pathology report, i.e. no
valid clinical notification, and there was no information about metastasis at the time of diagnosis.
After 2008, these cases were coded as localised if they received curative surgery. To obtain
consistency in the data and to avoid bias in the analyses, all stage information post 2008 were
considered unknown. This approach led to some methodological challenges when it came to
32
analysing changes in either survival (Papers I, III) and treatment over time (Paper II), which will be
addressed later.
Socioeconomic status
All the analyses included education and income, both serving as proxies for SES. In Papers I and III,
personal income was used as a proxy for SES, while in Paper II, household income was used. While
both personal and household incomes were explored as predictors in Paper II, it was concluded that
household income was a better predictor of treatment. This was especially noted for women; while
for men using personal or household income, did not change the results markedly. Education was
categorised based on the number of years of education: low (1–9 years, lower secondary school),
intermediate (10–12 years, upper secondary school) and high (12+ years, university or similar). Data
on both personal and household incomes were defined based on the percentiles. The cut points
were set at the 33rd (low) and 66th (high) percentiles, and were redefined every year to adjust for
the increase in income over time. In addition, when redefining these cut points for personal income,
gender was taken into account.
Health trust
In 2011, Norway consisted of 21 health trusts, which are responsible for general healthcare and
management of all patients residing in its geographical catchment area. If a health trust does not
provide certain services (e.g. lung cancer resection), these patients are referred to another health
trust that offer the appropriate treatment. The study variable denoting health service region (health
trust) was chosen instead of county, because it represents the actual catchment area of the different
treatment institutions. Health trust is an explanatory variable that is based on the patient’s place of
residence at the time of diagnosis, independent of where the patient was treated.
33
Comorbidity
Comorbidity information was measured using a modified version of the CCI, which was constructed
by using diagnostic codes (ICD 10) from hospitalisations within one year prior to, and including, the
date of diagnosis for a list of 17 chronic diseases. A score was determined for each of a patient’s
recorded comorbid disease based on its severity, and the combination of these scores resulted in a
modified CCI. The index was categorised into: “no hospital admissions before lung cancer diagnosis"
(CCI = 1), low (CCI = 0), intermediate (CCI = 1, 2) and high (CCI 3) (66, 118).
Statistical methods
In addition to standard descriptive statistics, a number of different statistical methods were applied
in this study.
Cox proportional hazard regression
In Paper I, Cox proportional hazard regression analyses were performed to identify and examine the
effects that different prognostic factors have on 1 year survival among the following three groups of
lung cancer patients: all patients diagnosed, non resected patients, and resected patients. Cox
regression was also used in Paper I to identify which sub groups of patients had the largest
improvement in survival over time. The underlying assumption (i.e. the proportional hazard
assumption) of this model is that the explanatory covariates are multiplicatively related to the
baseline hazard, and that the ratio of the hazards comparing groups with different values of the
explanatory variables, remains constant over time. The Cox regression model can generally be
expressed as 0 1 1( ) ( )exp( ... )n nh t h t x x , where ( )h t is the hazard function, 0 ( )h t is the
baseline hazard function and 1,..., nx x are the covariates with their corresponding parameters
1,..., n (119).
34
Logistic regression
In Paper II, three separate multivariable logistic regression models were estimated to investigate how
different covariates influence the odds of receiving surgical treatment, as well as, curative and
palliative radiotherapy, as a patient’s first treatment within one year of a lung cancer diagnosis. In
each of the three models, the dichotomous outcome was defined as either receiving treatment or
not. The logistic regression model can be expressed as 0 1 1ln( ) ...1 n np x xp
, where p is
the success probability of the event of interest, in this case, the patient receiving treatment, 0 is the
intercept which is the estimate as all covariates are zero, 1,..., nx x are the covariates and 1,..., n are
the corresponding parameters (120).
Net survival and excess mortality
Net cancer survival is the probability of surviving in a hypothetical world where cancer is the only
possible cause of death (121). It provides a measure of the excess mortality associated with being
diagnosed with cancer. Cause specific survival and relative survival are two ways of estimating net
survival. The major advantage of using a relative survival framework, is that it does not require any
cause of death information, as this information cannot always be trusted, especially for older
patients ( 85 years) (122). Net survival can be described as1
( )1( )* ( )
ni
i i
S tNS tn S t
, where ( )iS t is the
all cause survival for patient i weighted using the inverse of the cumulative expected survival for a
comparable, lung cancer free individual ( *( )iS t ). This comparable individual is found by matching
the lung cancer patient cohort with an age, gender and calendar year stratified Norwegian lifetable
obtained from Statistics Norway. The lifetable used in this thesis was not adjusted for smoking status
35
in the general population, however, it has been shown that additional adjustments may have little
effect on the net survival estimates (123). Five year net survival was estimated using the method
proposed by Pohar Perme in 2011, implemented in the Stata command strs, for both resected
and non resected lung cancer patients separately in Paper I (124, 125)b. In Paper III, the relative
excess risk of death among patients with localised, regional and metastatic disease by health trust
was modelled. The excess risk of death, ( )t at time t can be expressed as ( ) ( ) ( *)t h t h t , where
( )h t is the all cause mortality rate experienced by the patients and ( *)h t is the corresponding
expected mortality rate. Comparing the excess risk of death for the different health trusts to that of
the entire country resulted in estimates of the relative excess risk of death. In order to account for
case mix differences between health trusts that might affect survival, a Poisson regression model
was used to adjust for available explanatory variables. The Poisson regression model estimated the
expected number of deaths due to causes other than lung cancer. These were estimated using the
general population mortality rates (125, 126). The general form of a Poisson regression with a
logarithmic link function can be written as 1 1log(E( | )) ... n nY x x x , where Y represents
the number of counts of an event, is a scaling variable called the offset term which is used to make
the different groups comparable and 1,..., nx x are covariates with their corresponding parameters
1,..., n . In Paper III, the link function proposed by Dickman et al. in 2004 was used when modelling
relative excess risk of death (125).
Competing risk
While net survival aims to estimate the hypothetical survival probability in a world where cancer is
the only possible cause of death, competing risk can be used to estimate real world survival
probabilities, i.e. survival probabilities in a situation where other causes of death also exists (127). In
b In Papers I and III the term relative survival was used to denote net survival.
36
Paper II, the cumulative incidence using the Aalen Johansen estimator, i.e. the probability of patients
experiencing surgery and radical or palliative radiotherapy as their first treatment within one year of
lung cancer diagnosis, was estimated using Stata’s stcompet command, under the assumption that
any of the other treatment modalities (competing risks) could happen (128, 129).
Joinpoint regression
A joinpoint is a point where two linear lines with different slopes meet. By using the Joinpoint
Regression Program available from Surveillance, Epidemiology, and End Results Program (SEER), it is
possible to analyse linear trends, identify changes in linear trends and determine the number of
significant joinpoints (130). The program fits the simplest model as a straight line, i.e. zero joinpoints,
and then determines if additional joinpoints should be added to the model. In Paper III where the
relationship between surgery and survival was studied, the relative excess risks of death with
standard error, as well as, the resection rates were estimated for each health trust per year. Using
these estimates in the joinpoint program to plot the relative excess risk of death against the
resection rates, an optimal rate of resected patients, both among patients with localised and regional
diseases, was sought. The program calculated the annual percentage change, which in this case was
interpreted as the average change in relative excess risk of death as the resection rate category
increased by one unit. Hence, an inflection point in the relative excess risk of death where the
survival stabilises or declines, for increasing levels of resection, would identify an optimal resection
rate.
Multiple imputation
In medical and epidemiological research there will be missing information in the data for a number of
reasons. The degree of missing information varies from dataset to dataset and from study to study.
37
There may be many possible reasons for data to be missing, but three kinds of missing mechanisms
are identified in the literature: missing completely at random (MCAR), missing at random (MAR) and
missing not at random (MNAR). For data to be MCAR, the information missing should be completely
independent from both observed and unobserved data. Data are considered MAR if the probability
of missing data does not depend on unobserved data. And finally, data are MNAR if the probability of
missing data does depend on unobserved data.
Historically, there have been different ways to deal with missing information, e.g. complete case,
mean imputation, last observation carried forward and treating missing data as a new category (131).
If missing data are handled inadequately, the statistical analyses will lead to biased and/or inefficient
estimates. Treating the unknown data as a separate category has been shown to be a poor option,
even when the data are MCAR, since severe bias can arise in parameter estimations (132, 133). Using
the complete case approach, which is deleting all observations with unknown information, will lead
to correct and unbiased estimates, but only if the missing data are MCAR. However, as part of the
given data will be excluded, the statistical power of the analysis will decrease.
Imputation as an alternative approach to handle missing data needs to be carefully considered. Using
a single imputation approach, i.e. replacing the missing observation with a single value, will result in
standard error estimates that are too small. In contrast, multiple imputation procedures have
become widely accepted as a standard approach to handling missing data. In order to use this
methodology and obtain unbiased and efficient estimates, the nature of the missing data is
important. To achieve asymptotically unbiased estimates, it is important that the data meet at least
the MAR assumption. Multiple imputation can be performed even on MNAR data, however, then the
mechanism of missing data needs to be modelled as well.
38
Multiple imputation is generally performed in three steps: (i) generating multiple (m ) imputed data
sets, (ii) analysing each of the imputed data sets and (iii) pooling the estimates from the different
analyses together. Multiple imputation by chained equations (MICE) is an approach to constructm
imputed datasets. These are based on a set of imputation models, i.e. one model for each variable
with missing values. The initial step involves filling the missing values in each variable with a random
replacement from the observed data. Then the missing values of a variable, say 1y , will be regressed
on the other variables 2 n,...,y y , using the individuals where 1y is observed. Initial missing values of
1y is replaced with simulated draws from the obtained posterior predictive distribution of 1y . The
same procedure follows for 2y , which is regressed on 1 3, ,..., ny y y restricted to the individuals with
observed 2y , using the imputed values on 1y .Values for missing 2y is replaced by draws from the
posterior predictive distribution of 2y .This is repeated for all other variables and the process is often
called a cycle. In order to stabilise the results, several cycles are performed resulting in one imputed
dataset. This procedure is then repeated m times to give m imputed data sets. The second step is
analysing the m datasets separately and is usually easy as standard analysis tools can be used.
Finally, using Rubin’s rule, the estimates and variance covariance matrix from the m different
imputed datasets can be pooled together. The combined estimate of the individual obtained
estimates ( j ) can be found by taking the average over all imputations, i.e.1
1ˆm
jjm
. The
combined variance covariance matrix includes the components describing both within imputation
variability (i.e. variation between the different imputed dataset) and between imputation (i.e.
reflection of the uncertainty due to missing information). The total variance obtained can be
expressed as1ˆvar( ) (1 )W Bm
, where the within imputation variance j1
1 var( )m
jW
mand
the between imputation variance 2
1
1 ˆ( )1
m
jj
Bm
. To find a sufficient number of imputations,
m should be chosen to fulfil 100*m the percentage of incomplete cases in the data (134, 135).
39
It is possible to use multiple imputation with different kinds of variables, i.e. continuous, binary and
categorical variables. Categorical variables can be modelled using a multinomial logistic regression.
When choosing which variables to use in the imputation model, it is important that all variables,
explanatory and outcome, are included in the same form as they appear in the final analysis model. If
the analysis is based on a survival model, the outcome variables to include are a censoring indicator
and some function of follow up time. An alternative to using the follow up time is to use the Nelson
Aalen cumulate hazard estimate (136, 137).
In all papers, multiple imputation was used to impute missing values on the variables education,
income, EOD, histology and smoking status. In addition, laterality, tumour size and resection
procedure were imputed in Paper I, while symptom status was imputed in Paper II. All of these
variables were categorical variables, and therefore, multinomial logistic regressions were performed
using the actual variables from the analyses model as explanatory variables. Unfortunately, there is
no formal way to test the underlying assumption of data being MAR, and the best approach is to
condition the variables with missing values on the available variables with a known or plausibly
important association (138). In the data, there was no reason to suspect that there was any
association between the missing structure of the variables and their true values or with any other
variables that was not adjusted for in the imputation model.
40
Main results
Lung cancer survival in Norway (Paper I)
From 1997 to 2011, the 1 and 5 year relative survival for all lung cancer patients increased from
35.4% to 47.7% and 11.6% to 17.5%, respectively. The 1 and 5 year relative survival among resected
patients increased from 77.2% to 93.3% and 47.0% to 62.2%, respectively. The corresponding figures
for non resected patients were 28.4% to 37.0% and 3.6% to 6.3%, respectively. The HRs for death
among resected and non resected patients up to two years after resection/diagnosis showed that
these were consistently lower in 2004–2011 compared to 1997–2003, varying from 0.50 (95% CI
0.37–0.68) to 0.71 (95% CI 0.62–0.82) for resected patients and 0.75 (95% CI 0.70–0.81) to 0.88 (95%
CI 0.83–0.94) for non resected patients. The largest improvements in survival occurred among
resected, as well as, adenocarcinoma patients, while patients aged 80 years or older experienced the
smallest increase.
There were several factors affecting the improvement in survival during the study period from 1997
to 2011. The waiting time (median number of days) from diagnosis to resection increased from 26 to
34 days. Diagnostic improvements were also made during this period, which can be seen by the fact
that more patients received their diagnosis based not only on a histological examination but also
with additional examinations. Hence, the proportion of patients receiving diagnosis based only on
histological examination decreased from 68% in 1997 to 30% in 2011. The proportion of molecular
genetic examinations (including EGFR testing) increased from being less than 1% up until 2009 to
26% in 2011.
41
Lung cancer treatment (Paper II)
The resection rate among lung cancer patients in Norway remained fairly constant around 18% while
the proportion of radical radiotherapy administered increased from 8.6% to 14.1% in 2002–2011. The
proportion of patients not receiving either surgery or radiotherapy decreased from 50.0% to 38.7% in
the same period.
Older patients (i.e. patients aged 80+), patients with low household income, and patients from
certain health trusts, were less likely to receive any treatment. Compared to patients with a high
level of education, patients with a low level were found to have a lower odds of resection. Having a
low level of education was identified as a negative predictor for receiving surgery. A smoking history
was positively associated with both radical and palliative radiotherapy, while comorbidity and
symptoms prior to diagnosis were independently associated with receiving palliative radiotherapy.
Resection in relation to survival (Paper III)
The existence of regional variation in survival among NSCLC patients in Norway was established in
the study period (2002–2011) both among patients with localised and regional spread of disease.
These differences between regions persisted after adjusting for case mix which included information
about whether or not patients underwent resection.
For patients with localised disease an increasing resection rate was associated with a monotone
decrease in the relative excess risk of death, while among patients with regional disease a point was
identified where a further increase in resection would not contribute to any survival benefit.
42
Discussion
Overview of results
The results confirmed that survival after a lung cancer diagnosis for all patients has improved over
the last 15 years in Norway, with a larger improvement seen among resected than non resected
patients (Paper I). When investigating the predictors for receiving different lung cancer treatments,
SES variables and place of residence were identified as independent predictive factors (Paper II).
Differences in survival between geographical regions could not be explained by regional differences
in resection rate (Paper III).
Methodological considerations
All studies have analysed risk by either analysing survival in relation to surgery and other prognostic
factors, or the chance of receiving treatment.
Study design
Within epidemiological studies, the two main branches of studies are observational and
experimental. Observational studies examine the individuals of interest without any intervention,
only based on recording, classifying, counting and performing statistical analyses (18). These studies
can further be divided into descriptive and analytical, where analytical studies usually measure the
effect of different risk factors in relation to a specified outcome. Further, there are two types of
analytical studies, i.e. case control and cohort studies. A cohort is defined as a “group of individuals
who are followed or traced over a period of time” (139). Starting from a given time, the relationship
between different factors and the outcome of interest are explored for the cohort. There are two
types of cohort studies, which differ in when the data are collected, i.e. prospectively and historically
(140). All papers contributing to this thesis are historical cohort studies. This kind of study is
43
important in order to reveal possible differences in health services or in care given, and the results
can be used to help decision makers change health policies. For example, the results of Paper II,
identified groups of patients who were less likely to receive treatment, and therefore, efforts should
be made to address the inequity. And in Paper III, an optimal resection rate was found where
performing additional surgery has no further survival benefit, and this result may affect the
recommended patterns of surgical care.
Validity
The validity of an epidemiological study is usually divided into internal and external validity (141).
Internal validity means that the inference performed in the study is valid for the study subjects
themselves. However, (mainly) three different types of biases can impair a study’s internal validity,
namely, selection bias, information bias and confounding (discussed later). Since an unselected
population based lung cancer cohort was used, the results from the papers are considered valid and
representative for the Norwegian population. External validity refers to the applicability of the results
found within the study population to other populations, i.e. the generalisability of the results.
External validity assumes internal validity, but also relies on comparability of characteristics between
the study and target groups. The results that were obtained from these data regarding lung cancer
patients in Norway can be considered externally valid and comparable to other populations that have
similar demographics and a universal health care system, where treatment and care are equally
available for everyone, independent of social factors.
Selection bias is a systematic error that may be a result of the procedure of selecting participants for
a study. This bias will occur if the association between the exposure and the disease differs among
patients that are and are not included. The mandatory reporting about all new cancer cases to the
CRN prevents the data from suffering from selection bias. As all lung cancer patients registered in the
44
CRN were included in Papers I and II, these studies are very unlikely to be affected by selection bias.
The results from the third paper are only valid for NSCLC patients, however, within this group there
have not been any further selections. Another type of bias called information bias is related to
measurement errors registered for the patients. That is, the covariates or outcome variables may be
of different quality, and thus misclassification or measurement error will vary between the
comparison groups. As the quality of the data in the CRN is considered to be high, the risk of
information bias is considerably reduced. However, random errors can still occur and are related to
typing errors and data processing. For example, if a coder at the CRN is in doubt of what is written on
the clinical report regarding stage or histology, he will make a decision regarding the classification of
this patient. However, it is unlikely that there are systematic errors. In all the papers, stage
information has been adjusted for as a possible confounder. Due to the changes in the coding
practice (described earlier) after 2008, our results for the period 2002–2011 would be influenced by
information bias. However, all information regarding stage was considered as unknown after 2008 in
order to minimise this kind of bias.
The definition of a confounder is a variable that is correlated to the dependent variable and causally
linked to the outcome (18, 142). Confounding variables either falsely create an association that do
not really exist, or hide an already existing relation between the groups being compared. In the data
analysis process there are two ways of dealing with confounders: by stratification or by adjustment in
a statistical regression model (139). In the papers both techniques have been used, e.g. in Paper I,
the analyses were stratified by resection status and then adjusted for possible confounders, such as
age and stage.
In Paper I, the effect of education level on survival for a cohort of lung cancer patients was examined.
Survival depends on the stage of the disease, and therefore, performing a multivariable regression
45
without stage, could be confounded if those with lower education level have more advanced stage at
time of diagnosis. Therefore, in all the papers, the internal coding practice at the CRN which divides
stage into localised, regional, and metastatic, was used. A stratified analysis may still be affected by
confounding within the three strata, as it will only be able to control for confounding between, and
not within the different strata. This within strata confounding is also referred to as residual
confounding. Residual confounding may appear in any epidemiological study when a certain
confounding variable is not sufficiently adjusted for. A way of testing if an observed effect is affected
by residual confounding is to first perform a multivariable regression excluding the confounding
variable of interest, and then to examine the change in the estimates when the confounding variable
is included. If the estimate for the variable of interest changes towards a smaller effect, it indicates
that confounding still exists within the variable that was adjusted for. For example, stage information
(condensed TNM) was used in the regression model in Paper I. If instead information about the
patients’ TNM, which is a more detailed grouping of stage, was available, the estimates for education
could move closer to 1. This would indicate that the observed effect of education, when adjusted for
stage categorised in broad groups, could be a result of residual confounding. However, when stage
was included in the multivariable model, the estimates for intermediate and high education went
from 0.92 (95% CI 0.89–0.95) and 0.87 (95% CI 0.83–0.92) to 0.91 (95% CI 0.88–0.94) and 0.85 (95%
CI 0.81–0.89), respectively, indicating that the estimates were not likely to be a result of residual
confounding.
Statistical methods
Multiple imputation was used in all three papers to handle the difficulties related to changes in
coding practice on tumour stage in the CRN. Tumour stage was considered as missing for 100% of the
patients from 2009 to 2011, as there was no way to re code these patients similar to those coded
before 2009. Can one really use multiple imputations when missing a whole year of information? To
46
validate the method used in the papers, sensitivity analyses based on observed, historical data from
2002 to 2008 were performed comparing tumour stage distribution in two different scenarios: 1)
imputation on the patients that were actually missing tumour stage, and 2) imputation on data that
came from sequentially deleting all information about tumour stage for entire yearly cohorts. The
first scenario took the data as it is registered in the CRN and imputed the unknown stage information
based on the other known variables of the patients. The second scenario was performed by re coding
all stage information for patients diagnosed in 2002 as missing. To obtain the imputed stage
distribution for this cohort, stage information regarding patients diagnosed from 2003 to 2008 and
other available variables for the 2002 cohort were used. The same procedure was then applied
sequentially for all cohorts in years 2003 to 2008.
47
Figure 6: Comparing 1 year survival when dealing with missing stage information in two different
ways. First, multiple imputation is used to impute stage data the way it has been registered at the
Cancer Registry of Norway (method 1). Second, multiple imputation was used to impute stage when
assuming that all information about stage was missing for entire years (method 2).
Since the outcome in Papers I and III were related to survival and prognosis, the survival estimates in
these two scenarios were compared. The distribution and survival for the different tumour stages at
different years are shown in Figures 6 and 7.
0
10
20
30
40
50
60
70
80
90
100
1-ye
ar s
urvi
val (
%)
2002
2003
2004
2005
2006
2007
2008
Year of diagnosis
Localised - method 1
Localised - method 2
Regional - method 1
Regional - method 2
Metastasis - method 1
Metastasis - method 2
48
Figure 7: Comparing the stage distribution when dealing with missing stage information in two
different ways. First, multiple imputation is used to impute stage data the way it has been registered
at the Cancer Registry of Norway (method 1). Second, multiple imputation was used to impute stage
when assuming that all information about stage was missing for entire years (method 2).
Figures 6 and 7 show that survival and the distribution for stage for the two scenarios were similar,
except for the marginal difference in the stage distributions in 2002 and 2008. This indicates that the
imputed stage information is reliable, and that using multiple imputation techniques to deal with
whole years of missing data seems reasonable.
0
5
10
15
20
25
30
35
40
45
50
55
60
65
Prop
ortio
n (%
)
2002
2003
2004
2005
2006
2007
2008
Year of diagnosis
Localised - method 1
Localised - method 2
Regional - method 1
Regional - method 2
Metastasis - method 1
Metastasis - method 2
49
Strengths and limitations of register based studies
In this thesis the results of three Norwegian population based studies, of which two analysed all lung
cancer patients, and one was restricted to NSCLC patients, are presented. Since it is mandatory for
Norwegian hospitals to report all new cancer cases to the CRN, and since the CRN receives
information about emigration from the National Population Register, a negligible number of patients
were lost to follow up, and the CRN data are considered close to complete. Hence, it was possible to
analyse different aspects of an unselected cohort of lung cancer patients registered in a high quality
register. Further, due to the unique personal identification number, linkage of high quality individual
level SES and comorbidity data was possible. The ability to link individual level data for income and
education is possibly the greatest strength of this study when compared to other studies that have
use area specific measurements as a proxy for a patient’s SES.
On the other hand, there are still limitations and areas where the registry can improve. First, the
registry has limited cancer treatment information available. Additional clinical information would be
useful in studies examining prognostic factors and survival. Two examples of useful information are
surgery procedure and resection margin. Treatment procedure information would be important to
know as it would be possible for researchers to evaluate the differences in survival by procedure in a
population based material. Information about the resection margin would enable researchers to not
only examine if certain subgroups of patients have better resection outcomes than others, but also
to evaluate the quality of resections performed in different regions. Second, stage information for a
patient is important when determining what kind of treatment the patient should receive, as well as,
his prognosis. At the CRN the tumour stage information is coded and classified in house according to
the condensed TNM system, i.e. localised, regional and metastatic spread of disease. However, the
TNM system is the internationally agreed upon standard in describing and categorising cancer stages
and it is being maintained by the Union for International Cancer Control (143). Both national and
50
international guidelines use the TNM system to guide their recommendations. Further, it was
recently reported that comparing stage at diagnosis in six high income nations using national
population based cancer registry data was challenging due to the use of different measures of stage
(144). Having information about cTNM and pTNM would make the epidemiological studies more
comparable internationally and more interpretable clinically. Hence, it is a clear limitation of the CRN
data that only condensed TNM is used instead of TNM.
Discussion of the results
An improvement in lung cancer survival was found in Norway, which was larger than what was
observed in most other countries and registries worldwide (100). Also, the results showed that
survival improved among both non resected and resected patients, indicating that the explanation is
multifactorial and that the improvement possibly started as a consequence of increased attention
and focus on this group of patients. This included the establishment of the Norwegian Lung Cancer
Group in 1987 who prepared, revised and later updated national guidelines, which are published by
the Norwegian Directorate of Health (29, 79, 145 148). The historical changes that occurred can be
divided into two groups; diagnosis and treatment related changes. Diagnosis related changes
include increased ad hoc screening (i.e. more widespread use of image diagnostics when lung cancer
is suspected), improved immunohistochemistry, introduction of positron emission tomography (PET)
scan and better methods for analysing molecular genetic factors. The start of multidisciplinary team
meetings, the introduction and updated national guidelines, and the centralisation of examination
and surgical treatment contributed to both a more precise diagnosis and improved treatment of
patients. The introductions of SBRT and adjuvant chemotherapy are examples of treatment related
changes (149 151).
51
The results of Paper I showed that significantly fewer diagnoses were based on a plain histological
examination at the end of the study period compared to earlier. This was offset by an increase in the
use of more advanced methods like immunohistochemistry and molecular genetics as these became
available and improved. With the advances in immunohistochemistry, the pathologists have been
able to more accurately classify tumour cells, which is of paramount importance for choosing the
correct treatment. In addition, an increased use of computed tomography (CT) scanning – leading to
more ad hoc screening – and a more widespread use of PET scanning – revealing possible distant
metastases – are both contributing to an earlier and more precise diagnosis (152). Molecular
genetics is used for personalised medicine, which had its entry in lung cancer care in Norway around
2009 (153). Lung cancer patients are tested for an EGFR gene mutation, and research has shown that
advanced and recurrent NSCLC EGFR positive patients that were treated with gefitinib instead of
traditional chemotherapy experienced an extended progression free survival, i.e. the median time
from date of administration of treatment until the date of disease progression or death (151, 154,
155). The median progression free survival times were reported to be 9.2 and 6.3 months among
EGFR positive patients receiving gefitinib and those receiving a combination of cisplatin and
docetaxel, respectively (156). In addition, testing for the ALK gene mutation can be used as another
example of how molecular genetics testing has affected survival. A randomised study from 2013
showed that there was a significant reduced risk of progression or death among ALK positive
advanced lung cancer patients receiving crizotinib compared to chemotherapy as a second line
treatment (HR = 0.49 95% CI 0.37–0.64) (157). In addition, they showed that the progression free
survival time was 7.7 months in the crizotinib group and 3.0 months in the chemotherapy group.
It was also observed in Paper I that stage distribution was approximately constant over time. An
increase in ad hoc screening was observed during this period in Norway, and this may have led to an
earlier diagnosis and hence a shift towards a distribution with more patients with localised stage
52
disease (158). However, this has probably been offset by an increase in, and better examination of
possible metastases.
There was gradual centralisation of lung cancer treatment in Norway during the 2000s. While, less
than 20% of the resected patients received surgery at a high volume hospital in 2003, more than 80%
did in 2011 (23). After the study period, this trend continued and 87% were resected at a high
volume hospital in 2013. This centralisation, in addition to the introduction of multidisciplinary team
meetings and more diagnostic work up, may have contributed to the increase in waiting time for
surgery (149). The date of diagnosis registered in the CRN is the first of the following: date on the
clinical report, date of first histological verification, or date of death. It can be argued that the
increased time to surgery is affected by the extended diagnostics examination period. This is likely as
more diagnostic work was done in the last years of the study period. Hence, one cannot conclude
that the true waiting time for surgery has increased. A study from 2006 reported that the increased
waiting time for resection was a consequence of better diagnostics (159). Despite the increased time
to surgery, survival has still improved significantly. This can be seen as an indication that treatment
and/or the selection of patients for surgery have become better. It was indicated in a Norwegian lung
cancer study from 2007 that patients resected at a high volume hospital (>19 resections/year) may
have a lower 30 day post operative mortality (OR: 0.76, p value = 0.076), even if the result did not
achieve statistical significance (9).
It can by hypothesised that centralisation of lung cancer care may lead to an increased case load for
the clinicians and surgeons, which would provide addition experience and further improve the
resection quality, as well as, the anaesthetic and pre , and post operative care. The introductions of
cisplatin based adjuvant chemotherapy in 2005, and SBRT in 2008 were the two major treatment
specific changes that happened in the 2000s (151, 160 164). Both are likely to have contributed to
53
the observed improved survival as other studies have shown significant effects of these interventions
for selected groups (165).
The improvement in survival has been explained by advances in diagnostics and treatment, however,
some caution is required when interpreting these results. Due to the development of new diagnostic
methods, patients at the end of the study period may have been diagnosed earlier compared to
patients diagnosed at the beginning of the period. This difference between the times of diagnoses is
called a lead time, and the artificial improvement one may observe in survival, if the patients die at
the same point in time, is called lead time bias. In Paper I, it is possible that lead time bias was
introduced and that it may have influenced the improvement in survival. While it cannot be ruled out
that some of the improvement observed was a result of lead time bias, it is unlikely that it can
explain all improvement as there have been several treatment related changes that have positively
affected survival. One could question whether the observed improvement in survival could be the
result of an earlier diagnosis, causing previously undetected cancers to be categorised as localised,
previously defined localised patients with the most advanced spread of disease to be categorised as
regional, and the previously defined regional patients with the most advanced spread to be
categorised as metastatic. This would lead to an improved stage specific survival, as patients with the
worst prognosis shift to a more advanced group, where they would be the patients with the best
prognosis. However, there would not be an overall observed improvement for the whole group. This
is known as the Will Rogers phenomenon (166). In Paper I, since an improvement of survival was
observed for the whole group of lung cancer patients, the results seem valid (166). The increasing
difference between incidence and mortality among both men and women further supports that
there has been a true improvement in survival (6).
54
Paper II found that patients classified with high education and high income were around 30% and
60% more likely to receive surgery, respectively, compared to patients in the corresponding low
categories. Patients with high income were around 50% more likely to receive radiotherapy
compared to low income patients. A possible reason for this difference is that patients with high
education and income may be more health literate and able to seek information individually
compared to those in the low education and income groups. This may make them more critical to the
information they are given and they may be better able to discuss their options with their doctor. It
can be further hypothesised that this group of patients may have higher self confidence, stronger
assertiveness, and interest in their own healthcare. These can all contribute to the way they interact
with the health system, so that in situations where the doctor is ambivalent, the patient can push for
a resection. Consequently, unnecessary medical care may be used on patients where the benefit
might be minimal, i.e. there is no survival benefit from the surgery. However, the results from Paper I
showed that survival was significantly better among patients with high compared to low SES in the
same period. This superior prognosis was observed among resected and non resected patients, and
was thus an indication that the suspicion about a high rate of futile resections among high SES groups
was less likely.
Another important consideration is the lack of information on which patients are active workers and
which are pensioners at the time of diagnosis. This information could be important as there may be
patients with high income that were grouped as low income because they were registered with their
pension income the year before diagnosis. This can cause the observed effect for income in relation
to receiving surgical treatment to be biased, i.e. one might observe a weakened effect (closer to 1). A
way to adjust for this possible issue would be to include the patient’s annual income in a time period
prior to diagnosis, or to have precise information about which patients are pensioners and which are
workers. In Norway, the earliest age when a person can retire and receive his pension is 62 years, but
55
there are many people who choose to keep working and do not retire until they reach the age of 70,
even if the normal pension age in Norway is 67 years. To examine the potential problem of having a
combination of workers and pensioners in the same income group, a stratified analysis was
performed on the following three age bands: patients under 63 years (the majority of patients
assumed to be workers), 63–69 years (likely to be a mixture of pensioners and workers) and 70+ (the
majority of patients assumed to be pensioners). For patients under 63 years, odds ratios for receiving
surgery were 1.27 (95% CI 1.03–1.55) and 1.41 (95% CI 1.13–1.78), comparing intermediate and high
income against low income, respectively. The comparable numbers for 63–69 years were 1.56 (95%
CI 1.26–1.94) and 1.82 (95 CI 1.42–2.33) and for patients aged 70+ years the comparable numbers
were 1.20 (95% CI 1.07–1.35) and 1.25 (95% CI 1.02–1.52). These results show a clear effect of having
high income, however, it seems that the effect is smaller in the 70+ years group compared to those
under 63 years. This indicates that as patients get older, the positive effect of having a high income
fades. Since the effect of having high income is larger in the mixed group compared to the other two
groups, the heterogeneity of patients aged 63–69 does not seem to be problematic. Hence the
reported results appeared to be valid and not significantly biased by the misclassification of income
among pensioners.
Paper II showed that there were significant differences in the likelihood of receiving surgery and
radiotherapy based on where the patient lives. It was observed that patients living in a health trust
that had a reduced chance of receiving surgery had an increased chance of getting radical
radiotherapy. Hence, it seems like doctors and institutions in different health trusts have different
views on what they consider to be the best pattern of care for the patient. International literature
has also found that large within country variations in the likelihood of getting treatment exist (29, 43,
46, 87, 167). In Paper III the relationship between survival and resection was explored further, and
the results found significant geographical variation in survival. This is similar to what was found in a
56
Swedish and English study, but they also found that survival differences disappeared after adjusting
for whether the patient received treatment or not (43, 88). In this study, the variation in survival
could not be explained by the differences in resection rates between health trusts. The English study
reported that the proportion of patients undergoing surgery and their survival depended on patient
(e.g. patient’s decision making preferences and processes), physician (e.g. propensity to operate on
a patient where the appropriateness of resection is uncertain), and institutional factors (e.g.
availability of specialist thoracic surgical expertise). Since information about the physician and
institutional factors is not available for this study, one cannot rule out the possibility that the
observed trends reflect differences in pattern of care between the regions. Hence, some hospitals
and regions may have better quality of care and better surgical practice for all lung cancer patients.
The results of this study were adjusted for case mix to rule out the possibility that differences in the
cohort of patients could explain the results.
In the third paper where the existence of an optimal resection rate was analysed, it was established
that among patients with localised disease, a higher resection rate were associated with better
survival. However, among patients with regional spread, there was an upper limit for when
performing more resections seemed futile. Using this information it appears that some hospitals may
be more aggressive in performing resections on patients for whom the benefit is questionable. In
practice, stage III patients are divided into two groups where those diagnosed with clinical stage IIIA
are rarely considered candidates for surgery and those diagnosed with stage IIIB are not considered
at all. Paper III reported that 51.9% and 27.2% of resected patients in regions with high (>52%) and
low ( 52%) resection rates, respectively, were diagnosed with pathological stage IIIA. This shows that
an aggressive selection of patients for surgery occurred in regions with high resection rates.
57
Other possibilities that may explain regional variation in survival are the changes in lung cancer care
that have occurred at different times in the different regions of Norway. One study that examined
the utilisation and effectiveness of third generation chemotherapy among advanced NSCLC patients
in Norway observed significant survival improvement comparing the period before and after
introduction of vinorelbine, and the study noted that there were substantial geographical variation in
the uptake of chemotherapy use (168). While this study was conducted using data from 1994 to
2005, one can suspect that this time lag between regions can also be applied to more recent medical
advances.
58
Conclusion
Paper I:
Between 1997 and 2011, the overall survival for both resected and non resected lung cancer patients
in Norway has improved. The reason for this is multifactorial and contributing factors include an
increased attention on the group of lung cancer patients in general, more accurate diagnosis tools
and improvements in treatment.
Paper II:
The overall resection rate for all lung cancer patients remained approximately constant at around
18%, while radical radiotherapy treatment increased from 8.6% in 2002 to 14.1% in 2011. Although
Norway is an egalitarian country with a GINI index (i.e. a measure of the extent to which the
distribution of income among individuals or households within an economy deviates from a perfectly
equal distribution) in the upper quartile and has a free, universal healthcare system, lung cancer
patients with low SES and increasing age are less likely to receive any treatment (169). There were
also regional differences in the likelihood of receiving treatment. These results pose the question
whether equal health care is provided, independent of place of residence, age and socioeconomic
status.
Paper III:
The survival for non small cell lung cancer patients differs significantly between the different regions
in Norway during the period from 2002 to 2011, both among patients with localised and regional
spread of disease. Neither case mix nor differences in resection rates could explain these survival
differences. For patients with localised disease, an optimal resection rate was not apparent, but for
patients with regional spread of disease, an inflection point in the resection rate was found, beyond
which there was no improvement in survival.
59
Future perspectives
Even though its survival has improved in recent years, lung cancer is still one of the most lethal
cancer types. Around half of the diagnosed patients have metastatic spread at diagnosis and only
20% of all patients are resected. The focus in the following years should mainly (continue to) be on
primary prevention. This can be done through more and/or better restrictions on tobacco smoking
and stronger encouragement with better accompanying help for current smokers to quit smoking.
Screening is a secondary prevention for lung cancer and it has been analysed in a number of
international studies. A systematic review showed that there was no reduction in the number of
deaths from lung cancer after using chest radiography for screening (170). A randomised controlled
trial from the US showed a significant relative reduction of 20.0% and 6.7% in lung cancer mortality
and all cause mortality, respectively, however, the rate of false positives was above 96% (171). It will
be interesting to follow the discussion and the progress in the field of lung cancer screening.
The results of this study identified that there are differences in how likely patients are to receive
treatment depending on their SES. It should be an aim to equate these groups by adjusting the
treatment aggressiveness among patients in the low SES group. The greater likelihood of treatment
in high SES patients was observed when focusing on surgery and radiotherapy, but the results of a
study examining predictors for receiving chemotherapy among advanced stage lung cancer patients
would be valuable to gain a complete picture of the treatment pattern in Norway.
Due to its gradual introduction in Norway after 2008, the proportion of patients in this study who
received SBRT was low. Since the clinical register for lung cancer collects more specific stage
information (cTNM), it will be interesting to analyse and compare lung cancer survival after the
60
introduction of SBRT, and to be able to examine surgery in more detail at different tumour stages
using a national population based material.
Through the work of this thesis it is apparent that while the data quality at the CRN is considered to
be high, there is room for significant improvement, especially when it comes to clinical variables such
as stage. The continuity of how this is registered and keeping the level of missing data as low as
possible should be a clear focus. With the establishment of the National Clinical Register for Lung
Cancer in 2013, additional and relevant clinical information will become available and this data will
be important in future epidemiological studies.
61
Errors in published papers
Paper I:
Table 2: The multivariable confidence intervals regarding the variable smoking for all patients and the
resected patients should have been the following: Current: 1.18 (95% CI 1.09–1.28) instead of 1.18
(95% CI 0.94–1.03) and Former: 1.06 (95% CI 0.98–1.16) instead of 1.06 (95% CI 0.97–1.13) for all
diagnosed patients. For resected patients the multivariable estimates should have been Current: 1.25
(95% CI 0.87–1.80) instead of 1.25 (95% CI 0.87–1.01) and Former: 1.07 (95% CI 0.73–1.57) instead of
1.07 (95% CI 0.73–1.26).
Paper II:
Table 2: The last category of the variable Comorbidity was supposed to be CCI and not PRI as
published.
62
References
1. Ferlay J, Soerjomataram I, Ervik M, Dikshit R, Eser S, Mathers C, Rebelo M, Parkin DM,Forman D, Bray F. GLOBOCAN 2012 v1.0, Cancer Incidence and Mortality Worldwide: IARCCancerBase No. 11 [Internet]. Lyon, France: International Agency for Research on Cancer; 2013.Available from: http://globocan.iarc.fr, accessed 16/02/2016.
2. Aasebo U, Bremnes R. Er lungekreftbehandlingen preget av nihilisme? [Is treatment of lungcancer characterized by nihilism?]. Tidsskr Nor Laegeforen. 1998 May 20;118(13):2056.
3. Chambers SK, Dunn J, Occhipinti S, Hughes S, Baade P, Sinclair S, Aitken J, Youl P, O'ConnellDL. A systematic review of the impact of stigma and nihilism on lung cancer outcomes. BMC Cancer.2012 May 20;12:184.
4. Bremnes RM, Aasebo U. Lungekreft en lavstatussykdom? [Lung cancer a low statusdisease?]. Tidsskr Nor Laegeforen. 2002 Sep 30;122(23):2257.
5. Leira HO. Mye lungekreft – lite forskning [A lot of lung cancer little research]. Tidsskr NorLaegeforen. 2014 Dec 9;134(23 24):2287 8.
6. Cancer Registry of Norway. Cancer in Norway 2014 Cancer incidence, mortality, survival andprevalence in Norway. Oslo: Norway; CRo, 2014.
7. Kawaguchi T, Takada M, Kubo A, Matsumura A, Fukai S, Tamura A, Saito R, Maruyama Y,Kawahara M, Ignatius Ou SH. Performance status and smoking status are independent favorableprognostic factors for survival in non small cell lung cancer: a comprehensive analysis of 26,957patients with NSCLC. J Thorac Oncol. 2010 May;5(5):620 30.
8. Ou SH, Ziogas A, Zell JA. Prognostic factors for survival in extensive stage small cell lungcancer (ED SCLC): the importance of smoking history, socioeconomic and marital statuses, andethnicity. J Thorac Oncol. 2009 Jan;4(1):37 43.
9. Strand TE, Rostad H, Damhuis RA, Norstein J. Risk factors for 30 day mortality after resectionof lung cancer and prediction of their magnitude. Thorax. 2007 Nov;62(11):991 7.
10. Piccirillo JF, Tierney RM, Costas I, Grove L, Spitznagel EL J. Prognostic Importance ofComorbidity in a Hospital Based Cancer Registry. JAMA. 2004 May 26;291(20):2441 7.
11. Brustugun OT, Møller B, Helland A. Years of life lost as a measure of cancer burden on anational level. Br J Cancer. 2014 Aug 26;111(5):1014 20.
12. Engholm G, Ferlay J, Christensen N, Kejs AMT, Johannesen TB, Khan S, Leinonen MK, MilterMC, Ólafsdóttir E, Petersen T, Trykker H, HH S. NORDCAN: Cancer Incidence, Mortality, Prevalenceand Survival in the Nordic Countries, Version 7.2 (16.12.2015) [Internet]. Association of the NordicCancer Registries. Danish Cancer Society. Available from: http://www.ancr.nu, accessed on16/02/2016.
63
13. Engholm G, Ferlay J, Christensen N, Bray F, Gjerstorff ML, Klint A, Kotlum JE, Olafsdottir E,Pukkala E, Storm HH. NORDCAN a Nordic tool for cancer information, planning, quality control andresearch. Acta Oncol. 2010 Jun;49(5):725 36.
14. Grimsrud TK, Skaug HK, Larsen IK. Lung cancer changes in incidence by gender, age andcounty of residence 1984 2013. Tidsskr Nor Laegeforen. 2015 Nov 3;135(20):1844 9.
15. Janssen Heijnen ML, Coebergh JW. The changing epidemiology of lung cancer in Europe. LungCancer. 2003 Sep;41(3):245 58.
16. Doll R, Hill AB. Smoking and carcinoma of the lung; preliminary report. Br Med J. 1950 Sep30;2(4682):739 48.
17. Tyczynski JE, Bray F, Parkin DM. Lung cancer in Europe in 2000: epidemiology, prevention,and early detection. Lancet Oncol. 2003 Jan;4(1):45 55.
18. Adami HO, Hunter D, Trichopoulos D, editors. Textbook of Cancer Epidemiology. Second ed.New York: Oxford University Press; 2008.
19. IARC Working Group on the Evaluation of Carcinogenic Risks to Humans. Personal habits andindoor combustions. A reveiw of human carcinogens. IARC Monogr Eval Carcinog Risks Hum.2012;100E(Pt E):1 538.
20. The Norwegian Directorate of Health. Tal om tobakk 1973 2012. [Tobacco numbers 19732012]. Oslo: The Norwegian Directorate of Health; 2013.
21. Statistics Norway. Røykevaner, 2015 [Internet]. Oslo: Statistics Norway; 2016 [cited 201622.02]. Available from: https://www.ssb.no/helse/statistikker/royk/aar/2016 01 14.
22. National Cancer Institute. SEER Stat Fact Sheets: Lung and Bronchus Cancer [Internet].Bethesda (MD): National Cancer Institute; [cited 2016 15.02]. Available from:http://seer.cancer.gov/statfacts/html/lungb.html.
23. Norwegian Lung Cancer Registry. Årsrapport 2013 2014 [Yearly report 2013 2014]. Oslo:Norwegian Lung Cancer Registry; 2014.
24. Syrigos K, Nutting C, Roussos C, editors. Tumors of the Chest Biology, Diagnosis andManagement. New York: Springer; 2006.
25. Parkin D, Whelan S, Ferlay J, Teppo L, Thomas D, editors. Cancer Incidence in Five Continents,vol VIII (IARC Scientific Publications No. 155). Lyon: International Agency for Research on Cancer;2002.
26. Meza R, Meernik C, Jeon J, Cote ML. Lung cancer incidence trends by gender, race andhistology in the United States, 1973 2010. PLoS One. 2015 Mar 30;10(3):e0121323.
64
27. The Norwegian Directorate of Health. Nasjonalt handlingsprogram med retningslinjer fordiagnostikk, behandling or oppfølgning av lungekreft, mesoteliom og thymom [National guidelinesfor diagnosis, treatment and monitoring of lung cancer, mesothelioma and thymoma]. Oslo: TheNorwegian Directorate of Health; 2015.
28. Goldshaw P, editor. Staging Manual in Thoracic Oncology. Florida: International Associationfor the Study of Lung Cancer; 2009.
29. Strand TE, Bartnes K, Rostad H. National trends in lung cancer surgery. Eur J CardiothoracSurg. 2012 Aug;42(2):355 8.
30. Bonita R, Beaglehole R, Kjellström T. Basic Epidemiology. Second ed. Geneva: WHO Press;2006.
31. Gospodarowicz MK, O'Sullivan B, Sobin LH, editors. Prognostic Factors in Cancer. Third ed.New Jersey: Wiley Liss; 2006.
32. Birim O, Kappetein AP, van Klaveren RJ, Bogers AJ. Prognostic factors in non small cell lungcancer surgery. Eur J Surg Oncol. 2006 Feb;32(1):12 23.
33. Jazieh AR, Hussain M, Howington JA, Spencer HJ, Husain M, Grismer JT, Read RC. Prognosticfactors in patients with surgically resected stages I and II non small cell lung cancer. Ann Thorac Surg.2000 Oct;70(4):1168 71.
34. Goldstraw P, Crowley J, Chansky K, Giroux DJ, Groome PA, Rami Porta R, Postmus PE, RuschV, Sobin L. The IASLC Lung Cancer Staging Project: proposals for the revision of the TNM stagegroupings in the forthcoming (seventh) edition of the TNM Classification of malignant tumours. JThorac Oncol. 2007 Aug;2(8):706 14.
35. Walters S, Maringe C, Coleman MP, Peake MD, Butler J, Young N, Bergstrom S, Hanna L,Jakobsen E, Kolbeck K, Sundstrom S, Engholm G, Gavin A, Gjerstorff ML, Hatcher J, Johannesen TB,Linklater KM, McGahan CE, Steward J, Tracey E, Turner D, Richards MA, Rachet B. Lung cancersurvival and stage at diagnosis in Australia, Canada, Denmark, Norway, Sweden and the UK: apopulation based study, 2004 2007. Thorax. 2013 Jun;68(6):551 64.
36. Paesmans M. Prognostic and predictive factors for lung cancer. Breathe. 2012 Dec 1;7(2):11322.
37. Sagerup CM, Smastuen M, Johannesen TB, Helland A, Brustugun OT. Increasing age andcarcinoma not otherwise specified: a 20 year population study of 40,118 lung cancer patients. JThorac Oncol. 2012 Jan;7(1):57 63.
38. Strand TE, Rostad H, Møller B, Norstein J. Survival after resection for primary lung cancer: apopulation based study of 3211 resected patients. Thorax. 2006 Aug;61(8):710 5.
39. Chansky K, Sculier JP, Crowley JJ, Giroux D, Van Meerbeeck J, Goldstraw P. The InternationalAssociation for the Study of Lung Cancer Staging Project: prognostic factors and pathologic TNMstage in surgically managed non small cell lung cancer. J Thorac Oncol. 2009 Jul;4(7):792 801.
65
40. Salmeron D, Chirlaque MD, Isabel Izarzugaza M, Sanchez MJ, Marcos Gragera R, Ardanaz E,Galceran J, Mateos A, Navarro C. Lung cancer prognosis in Spain: the role of histology, age and sex.Respir Med. 2012 Sep;106(9):1301 8.
41. Galobardes B, Shaw M, Lawlor DA, Lynch JW, Davey Smith G. Indicators of socioeconomicposition (part 1). J Epidemiol Community Health. 2006 Jan;60(1):7 12.
42. Forrest LF, Adams J, Wareham H, Rubin G, White M. Socioeconomic inequalities in lungcancer treatment: systematic review and meta analysis. PLoS Med. 2013 Feb 5;10(2):e1001376.
43. Riaz SP, Luchtenborg M, Jack RH, Coupland VH, Linklater KM, Peake MD, Møller H. Variationin surgical resection for lung cancer in relation to survival: population based study in England 20042006. Eur J Cancer. 2012 Jan;48(1):54 60.
44. Berglund A, Holmberg L, Tishelman C, Wagenius G, Eaker S, Lambe M. Social inequalities innon small cell lung cancer management and survival: a population based study in central Sweden.Thorax. 2010 Apr;65(4):327 33.
45. Berglund A, Lambe M, Luchtenborg M, Linklater K, Peake MD, Holmberg L, Møller H. Socialdifferences in lung cancer management and survival in South East England: a cohort study. BMJOpen. 2012;2(3).
46. Crawford SM, Sauerzapf V, Haynes R, Zhao H, Forman D, Jones AP. Social and geographicalfactors affecting access to treatment of lung cancer. Br J Cancer. 2009 Sep 15;101(6):897 901.
47. Jack RH, Gulliford MC, Ferguson J, Møller H. Explaining inequalities in access to treatment inlung cancer. J Eval Clin Pract. 2006 Oct;12(5):573 82.
48. Johnson AM, Hines RB, Johnson JA 3rd, Bayakly AR. Treatment and survival disparities in lungcancer: the effect of social environment and place of residence. Lung Cancer. 2014 Mar;83(3):401 7.
49. Powell HA, Tata LJ, Baldwin DR, Potter VA, Stanley RA, Khakwani A, Hubbard RB. Treatmentdecisions and survival for people with small cell lung cancer. Br J Cancer. 2014 Feb 18;110(4):908 15.
50. Forrest LF, White M, Rubin G, Adams J. The role of patient, tumour and system factors insocioeconomic inequalities in lung cancer treatment: population based study. Br J Cancer. 2014 Jul29;111(3):608 18.
51. Currow DC, You H, Aranda S, McCaughan BC, Morrell S, Baker DF, Walton R, Roder DM. Whatfactors are predictive of surgical resection and survival from localised non small cell lung cancer?Med J Aust. 2014 Oct 20;201(8):475 80.
52. Kaergaard Starr L, Osler M, Steding Jessen M, Lidegaard Frederiksen B, Jakobsen E, OsterlindK, Schuz J, Johansen C, Oksbjerg Dalton S. Socioeconomic position and surgery for early stage nonsmall cell lung cancer: A population based study in Denmark. Lung Cancer. 2013 Mar;79(3):262 9.
66
53. Herndon JE 2nd, Kornblith AB, Holland JC, Paskett ED. Patient education level as a predictorof survival in lung cancer clinical trials. J Clin Oncol. 2008 Sep 1;26(25):4116 23.
54. Forrest LF, Adams J, Rubin G, White M. The role of receipt and timeliness of treatment insocioeconomic inequalities in lung cancer survival: population based, data linkage study. Thorax.2015 Feb;70(2):138 45.
55. Dalton SO, Steding Jessen M, Jakobsen E, Mellemgaard A, Osterlind K, Schuz J, Johansen C.Socioeconomic position and survival after lung cancer: Influence of stage, treatment and comorbidityamong Danish patients with lung cancer diagnosed in 2004 2010. Acta Oncol. 2015 May;54(5):797804.
56. Andreas S, Rittmeyer A, Hinterthaner M, Huber RM. Smoking cessation in lung cancerachievable and effective. Dtsch Arztebl Int. 2013 Oct;110(43):719 24.
57. Baser S, Shannon VR, Eapen GA, Jimenez CA, Onn A, Lin E, Morice RC. Smoking cessation afterdiagnosis of lung cancer is associated with a beneficial effect on performance status. Chest. 2006Dec;130(6):1784 90.
58. Dobson Amato KA, Hyland A, Reed R, Mahoney MC, Marshall J, Giovino G, Bansal Travers M,Ochs Balcom HM, Zevon MA, Cummings KM, Nwogu C, Singh AK, Chen H, Warren GW, Reid M.Tobacco Cessation May Improve Lung Cancer Patient Survival. J Thorac Oncol. 2015 Jul;10(7):1014 9.
59. Parsons A, Daley A, Begh R, Aveyard P. Influence of smoking cessation after diagnosis of earlystage lung cancer on prognosis: systematic review of observational studies with meta analysis. BMJ.2010 Jan 21;340:b5569.
60. Warren GW, Kasza KA, Reid ME, Cummings KM, Marshall JR. Smoking at diagnosis andsurvival in cancer patients. International journal of cancer Journal international du cancer. 2013 Jan15;132(2):401 10.
61. Kawaguchi T, Matsumura A, Fukai S, Tamura A, Saito R, Zell JA, Maruyama Y, Ziogas A,Kawahara M, Ignatius Ou SH. Japanese ethnicity compared with Caucasian ethnicity and neversmoking status are independent favorable prognostic factors for overall survival in non small cell lungcancer: a collaborative epidemiologic study of the National Hospital Organization Study Group forLung Cancer (NHSGLC) in Japan and a Southern California Regional Cancer Registry databases. JThorac Oncol. 2010 Jul;5(7):1001 10.
62. Sardari Nia P, Weyler J, Colpaert C, Vermeulen P, Van Marck E, Van Schil P. Prognostic valueof smoking status in operated non small cell lung cancer. Lung Cancer. 2005 Mar;47(3):351 9.
63. Kogure Y, Ando M, Saka H, Chiba Y, Yamamoto N, Asami K, Hirashima T, Seto T, Nagase S,Otsuka K, Yanagihara K, Takeda K, Okamoto I, Aoki T, Takayama K, Yamasaki M, Kudoh S, Katakami N,Miyazaki M, Nakagawa K. Histology and smoking status predict survival of patients with advancednon small cell lung cancer. Results of West Japan Oncology Group (WJOG) Study 3906L. J ThoracOncol. 2013 Jun;8(6):753 8.
67
64. Maeda R, Yoshida J, Ishii G, Hishida T, Nishimura M, Nagai K. The prognostic impact ofcigarette smoking on patients with non small cell lung cancer. J Thorac Oncol. 2011 Apr;6(4):735 42.
65. Gridilli C, Audisio R, editors. Management of Lung Cancer in Older People. London: Springer;2013.
66. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognosticcomorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373 83.
67. Iachina M, Jakobsen E, Møller H, Luchtenborg M, Mellemgaard A, Krasnik M, Green A. Theeffect of different comorbidities on survival of non small cells lung cancer patients. Lung. 2015Apr;193(2):291 7.
68. Deleuran T, Thomsen RW, Nørgaard M, Jacobsen JB, Rasmussen TR, Søgaard M. Comorbidityand survival of Danish lung cancer patients from 2000 2011: a population based cohort study. ClinEpidemiol. 2013 Nov 1;5(Suppl 1):31 8.
69. Søgaard M, Thomsen RW, Bossen KS, Sørensen HT, Nørgaard M. The impact of comorbidityon cancer survival: a review. Clin Epidemiol. 2013 Nov 1;5(Suppl 1):3 29.
70. Read WL, Tierney RM, Page NC, Costas I, Govindan R, Spitznagel EL, Piccirillo JF. Differentialprognostic impact of comorbidity. J Clin Oncol. 2004 Aug 1;22(15):3099 103.
71. Bauml J, Mick R, Zhang Y, Watt CD, Vachani A, Aggarwal C, Evans T, Langer C. Determinants ofsurvival in advanced non small cell lung cancer in the era of targeted therapies. Clin Lung Cancer.2013 Sep;14(5):581 91.
72. Travis WD, Brambilla E, Muller Hermelink HK, Harris CC, editors. World Health OrganizationClassification of Tumours. Pathology and Genetics of Tumours of the Lung, Pleura, Thymus and Heart.Lyon: IARC Press; 2004.
73. In KH, Kwon YS, Oh IJ, Kim KS, Jung MH, Lee KH, Kim SY, Ryu JS, Lee SY, Jeong ET, Lee SY, YumHK, Lee CG, Kim WS, Zo JI, Kim H, Kim YW, Kim SK, Lee JC, Kim YC. Lung cancer patients who areasymptomatic at diagnosis show favorable prognosis: a korean Lung Cancer Registry Study. LungCancer. 2009 May;64(2):232 7.
74. Sheel AR, McShane J, Poullis MP. Survival of patients with or without symptoms undergoingpotentially curative resections for primary lung cancer. Ann Thorac Surg. 2013 Jan;95(1):276 84.
75. Mahesh PA, Archana S, Jayaraj BS, Patil S, Chaya SK, Shashidhar HP, Sunitha BS, PrabhakarAK. Factors affecting 30 month survival in lung cancer patients. Indian J Med Res. 2012Oct;136(4):614 21.
76. Cuyun Carter G, Barrett AM, Kaye JA, Liepa AM, Winfree KB, John WJ. A comprehensivereview of nongenetic prognostic and predictive factors influencing the heterogeneity of outcomes inadvanced non small cell lung cancer. Cancer Manag Res. 2014 Oct 23;6:437 49.
68
77. Montazeri A. Quality of life data as prognostic indicators of survival in cancer patients: anoverview of the literature from 1982 to 2008. Health Qual Life Outcomes. 2009 Dec 23;7:102.
78. Harichand Herdt S, Ramalingam SS. Gender associated differences in lung cancer: clinicalcharacteristics and treatment outcomes in women. Semin Oncol. 2009 Dec;36(6):572 80.
79. Sagerup CM, Smastuen M, Johannesen TB, Helland A, Brustugun OT. Sex specific trends inlung cancer incidence and survival: a population study of 40,118 cases. Thorax. 2011 Apr;66(4):301 7.
80. Tong BC, Kosinski AS, Burfeind WR Jr, Onaitis MW, Berry MF, Harpole DH, Jr., D'Amico TA. Sexdifferences in early outcomes after lung cancer resection: analysis of the Society of ThoracicSurgeons General Thoracic Database. J Thorac Cardiovasc Surg. 2014 Jul;148(1):13 8.
81. Visbal AL, Williams BA, Nichols FC, 3rd, Marks RS, Jett JR, Aubry MC, Edell ES, Wampfler JA,Molina JR, Yang P. Gender differences in non small cell lung cancer survival: an analysis of 4,618patients diagnosed between 1997 and 2002. Ann Thorac Surg. 2004 Jul;78(1):209 15; discussion 15.
82. Osterlind K, Andersen PK. Prognostic factors in small cell lung cancer: multivariate modelbased on 778 patients treated with chemotherapy with or without irradiation. Cancer Res. 1986Aug;46(8):4189 94.
83. Govindan R, Page N, Morgensztern D, Read W, Tierney R, Vlahiotis A, Spitznagel EL, PiccirilloJ. Changing epidemiology of small cell lung cancer in the United States over the last 30 years: analysisof the surveillance, epidemiologic, and end results database. J Clin Oncol. 2006 Oct 1;24(28):4539 44.
84. Radzikowska E, Glaz P, Roszkowski K. Lung cancer in women: age, smoking, histology,performance status, stage, initial treatment and survival. Population based study of 20 561 cases.Ann Oncol. 2002 Jul;13(7):1087 93.
85. Cerfolio RJ, Bryant AS, Scott E, Sharma M, Robert F, Spencer SA, Garver RI. Women withpathologic stage I, II, and III non small cell lung cancer have better survival than men. Chest. 2006Dec;130(6):1796 802.
86. Walsh PM, Comber H. Patterns of care and survival of cancer patients in Ireland 1994 to2001: time trends and regional variation for breast, colorectal, lung and prostate cancer. SummaryReport. Ireland: National Cancer Registry, 2006.
87. Sant M, Minicozzi P, Allemani C, Cirilli C, Federico M, Capocaccia R, Budroni M, Candela P,Crocetti E, Falcini F, Ferretti S, Fusco M, Giacomin A, La Rosa F, Mangone L, Natali M, Leon MP, TrainaA, Tumino R, Zambon P. Regional inequalities in cancer care persist in Italy and can influence survival.Cancer Epidemiol. 2012 Dec;36(6):541 7.
88. Myrdal G, Lamberg K, Lambe M, Stahle E, Wagenius G, Holmberg L. Regional differences intreatment and outcome in non small cell lung cancer: a population based study (Sweden). LungCancer. 2009 Jan;63(1):16 22.
89. Rahimi A, Borgan E, Aagnes B, Møller B. Cancer survival by county and health region inNorway, 2000 2009. Oslo: Cancer Registry of Norway; 2012.
69
90. Skyrud KD, Bray F, Eriksen MT, Nilssen Y, Møller B. Regional variations in cancer survival:Impact of tumour stage, socioeconomic status, comorbidity and type of treatment in Norway.International journal of cancer Journal international du cancer. 2016 May 1;138(9):2190 200.
91. Walters S, Quaresma M, Coleman MP, Gordon E, Forman D, Rachet B. Geographical variationin cancer survival in England, 1991 2006: an analysis by Cancer Network. J Epidemiol CommunityHealth. 2011 Nov;65(11):1044 52.
92. Nur U, Quaresma M, De Stavola B, Peake M, Rachet B. Inequalities in non small cell lungcancer treatment and mortality. J Epidemiol Community Health. 2015 Oct;69(10):985 92.
93. Edge SB, Byrd DR, Compton CC, Fritz AG, Greene FL, Trotti A, editors. AJCC Cancer StagingManual. Seventh ed. New York: Springer; 2010.
94. Strand TE. More emphasis on resection rates! J Thorac Oncol. 2012 Jul;7(7):1067 8.
95. Thorsteinsson H, Alexandersson A, Oskarsdottir GN, Skuladottir R, Isaksson HJ, Jonsson S,Gudbjartsson T. Resection rate and outcome of pulmonary resections for non small cell lung cancer:a nationwide study from Iceland. J Thorac Oncol. 2012 Jul;7(7):1164 9.
96. Dansk Lunge Cancer Gruppe & Dansk Lunge Cancer Register. Årsrapport 2008 [Annual report2008]. Århus: Dansk Lunge Cancer Gruppe & Dansk Lunge Cancer Register; 2008.
97. Rowell NP, Williams CJ. Radical radiotherapy for stage I/II non small cell lung cancer inpatients not sufficiently fit for or declining surgery (medically inoperable): a systematic review.Thorax. 2001 Aug;56(8):628 38.
98. Vrdoljak E, Mise K, Sapunar D, Rozga A, Marusic M. Survival analysis of untreated patientswith non small cell lung cancer. Chest. 1994 Dec;106(6):1797 800.
99. De Angelis R, Sant M, Coleman MP, Francisci S, Baili P, Pierannunzio D, Trama A, Visser O,Brenner H, Ardanaz E, Bielska Lasota M, Engholm G, Nennecke A, Siesling S, Berrino F, Capocaccia R.Cancer survival in Europe 1999 2007 by country and age: results of EUROCARE 5 a population basedstudy. Lancet Oncol. 2014 Jan;15(1):23 34.
100. Allemani C, Weir HK, Carreira H, Harewood R, Spika D, Wang XS, Bannon F, Ahn JV, JohnsonCJ, Bonaventure A, Marcos Gragera R, Stiller C, Azevedo e Silva G, Chen WQ, Ogunbiyi OJ, Rachet B,Soeberg MJ, You H, Matsuda T, Bielska Lasota M, Storm H, Tucker TC, Coleman MP. Globalsurveillance of cancer survival 1995 2009: analysis of individual data for 25,676,887 patients from279 population based registries in 67 countries (CONCORD 2). Lancet. 2015 Mar 14;385(9972):9771010.
101. van der Drift MA, Karim Kos HE, Siesling S, Groen HJ, Wouters MW, Coebergh JW, de Vries E,Janssen Heijnen ML. Progress in standard of care therapy and modest survival benefits in thetreatment of non small cell lung cancer patients in the Netherlands in the last 20 years. J ThoracOncol. 2012 Feb;7(2):291 8.
70
102. Hakulinen T, Engholm G, Gislum M, Storm HH, Klint A, Tryggvadottir L, Bray F. Trends in thesurvival of patients diagnosed with cancers in the respiratory system in the Nordic countries 19642003 followed up to the end of 2006. Acta Oncol. 2010 Jun;49(5):608 23.
103. Coleman MP, Forman D, Bryant H, Butler J, Rachet B, Maringe C, Nur U, Tracey E, Coory M,Hatcher J, McGahan CE, Turner D, Marrett L, Gjerstorff ML, Johannesen TB, Adolfsson J, Lambe M,Lawrence G, Meechan D, Morris EJ, Middleton R, Steward J, Richards MA. Cancer survival inAustralia, Canada, Denmark, Norway, Sweden, and the UK, 1995 2007 (the International CancerBenchmarking Partnership): an analysis of population based cancer registry data. Lancet. 2011 Jan8;377(9760):127 38.
104. Holmberg L, Sandin F, Bray F, Richards M, Spicer J, Lambe M, Klint A, Peake M, Strand TE,Linklater K, Robinson D, Møller H. National comparisons of lung cancer survival in England, Norwayand Sweden 2001 2004: differences occur early in follow up. Thorax. 2010 May;65(5):436 41.
105. Francisci S, Minicozzi P, Pierannunzio D, Ardanaz E, Eberle A, Grimsrud TK, Knijn A, PastorinoU, Salmeron D, Trama A, Sant M. Survival patterns in lung and pleural cancer in Europe 1999 2007:Results from the EUROCARE 5 study. Eur J Cancer. 2015 Oct;51(15):2242 53.
106. Eberle A, Jansen L, Castro F, Krilaviciute A, Luttmann S, Emrich K, Holleczek B, Nennecke A,Katalinic A, Brenner H. Lung cancer survival in Germany: A population based analysis of 132,612 lungcancer patients. Lung Cancer. 2015 Dec;90(3):528 33.
107. Akhtar Danesh N, Finley C. Temporal trends in the incidence and relative survival of nonsmall cell lung cancer in Canada: A population based study. Lung Cancer. 2015 Oct;90(1):8 14.
108. Larsen IK, Smastuen M, Johannesen TB, Langmark F, Parkin DM, Bray F, Møller B. Data qualityat the Cancer Registry of Norway: an overview of comparability, completeness, validity andtimeliness. Eur J Cancer. 2009 May;45(7):1218 31.
109. Statistics Norway. Statistics Norway in brief [Internet]. Oslo: Statistics Norway; [cited 201625.02]. Available from: http://www.ssb.no/en/omssb/om oss/kort om ssb.
110. Statistics Norway. Income and wealth statistics for households, 2014 [Internet]. Oslo:Statistics Norway; 2015 [cited 2016 25.02]. Available from: http://www.ssb.no/en/inntekt ogforbruk/statistikker/ifhus/aar/2015 12 16?fane=om#content.
111. Statistics Norway. Dette er Statistisk sentralbyrå statistikk og analyser til nytte forsamfunnet [This is Statistics Norway statistics and analysis for the benefit of the society]. Oslo:Statistics Norway; 2014.
112. Statistics Norway. Befolkningens utdannigsnivå, manglende opplysninger om innvandrere2013 [The populations level of education, lack of information about immigrants in 2013] [Internet].Statistics Norway; 2015 [cited 2016 25.02]. Available from: http://www.ssb.no/utdanning/artikler ogpublikasjoner/befolkningens utdanningsniva manglende opplysninger om innvandrere 2013.
71
113. Statistics Norway. Population's level of education, 1 October 2014 [Internet]. Oslo: StatisticsNorway; [cited 2016 25.02]. Available from:http://www.ssb.no/en/utdanning/statistikker/utniv/aar/2015 06 18?fane=om.
114. Kastpersen S, Kalseth B. Omfang og Utvikling av det Selv betalende Markedet for PrivateSpesialisthelsetjenester i Norge [Scope and Development of the Private Spending on Healthcare forPrivate Healthcare in Norway]. Trondheim: SINTEF; 2010.
115. The Norwegian Directorate of Health. Definisjonsvedlegg SAMDATA Spesialisthelsetjenesten2013 [Definition Appendix SAMDATA Specialist Healthcare 2013]. Norway: The NorwegianDirectorate of Health; 2013.
116. Asli LM, Kvaloy SO, Jetne V, Myklebust TA, Levernes SG, Tveit KM, Green TO, Johannesen TB.Utilization of radiation therapy in Norway after the implementation of the national cancer plan anational, population based study. Int J Radiat Oncol Biol Phys. 2014 Nov 1;90(3):707 14.
117. Tyczynski J, Démaret E, Parkin D, editors. Standards and Guidelines for Cancer Registration inEurope, vol I (IARC Technical Publication No. 40). Lyon: IARC; 2003.
118. Nilssen Y, Strand TE, Wiik R, Bakken IJ, Yu XQ, O'Connell DL, Møller B. Utilizing nationalpatient register data to control for comorbidity in prognostic studies. Clin Epidemiol. 2014 Oct24;6:395 404.
119. Cox DR. Regression Models and Life Tables. Royal Statistical Society. 1972;34(2):187 220.
120. Hosmer DW, Lemeshow S. Applied Logistic Regression. Second ed. New York: WileyInterscience Publication; 2005.
121. Lambert PC, Dickman PW, Rutherford MJ. Comparison of different approaches to estimatingage standardized net survival. BMC Med Res Methodol. 2015 Aug 15;15:64.
122. Skyrud KD, Bray F, Møller B. A comparison of relative and cause specific survival by cancersite, age and time since diagnosis. International journal of cancer Journal international du cancer.2014 Jul 1;135(1):196 203.
123. Hinchliffe SR, Rutherford MJ, Crowther MJ, Nelson CP, Lambert PC. Should relative survivalbe used with lung cancer data? Br J Cancer. 2012 May 22;106(11):1854 9.
124. Perme MP, Stare J, Esteve J. On estimation in relative survival. Biometrics. 2012Mar;68(1):113 20.
125. Dickman PW, Sloggett A, Hills M, Hakulinen T. Regression models for relative survival. StatMed. 2004 Jan 15;23(1):51 64.
126. Yu XQ, O'Connell DL, Gibberd RW, Smith DP, Dickman PW, Armstrong BK. Estimating regionalvariation in cancer survival: a tool for improving cancer care. Cancer Causes Control. 2004Aug;15(6):611 8.
72
127. Fine JP, Gray RJ. A Proportional Hazards Model for the Subdistribution of a Competing Risk. JAm Stat Assoc. 1999 Jun;94:496 509.
128. Aalen O, Johansen S. An empirical transition matrix for non homogeneous Markov chainsbased on censored observations. Scand J Stat. 1978;5:141 50.
129. Coviello V, Boggess M. Cumulative incidence estimation in the presence of competing risks.The Stata Journal. 2004;4(2):103 12.
130. Kim HJ, Fay MP, Feuer EJ, Midthune DN. Permutation tests for joinpoint regression withapplications to cancer rates. Stat Med. 2000 Feb 15;19(3):335 51.
131. van Buuren S. Flexible Imputation of Missing Data. Florida: Taylor & Francis Group; 2012.
132. Moons KG, Donders RA, Stijnen T, Harrell FE, Jr. Using the outcome for imputation of missingpredictor values was preferred. J Clin Epidemiol. 2006 Oct;59(10):1092 101.
133. Vach W, Blettner M. Biased estimation of the odds ratio in case control studies due to theuse of ad hoc methods of correcting for missing values for confounding variables. Am J Epidemiol.1991 Oct 15;134(8):895 907.
134. Bodner TE. What improves with increasing missing data imputations? Struct Equ Modeling.2008 Oct;15(4):651 75.
135. White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues andguidance for practice. Stat Med. 2011 Feb 20;30(4):377 99.
136. Falcaro M, Nur U, Rachet B, Carpenter JR. Estimating excess hazard ratios and net survivalwhen covariate data are missing: strategies for multiple imputation. Epidemiology. 2015May;26(3):421 8.
137. White IR, Royston P. Imputing missing covariate values for the Cox model. Stat Med. 2009 Jul10;28(15):1982 98.
138. Woodward M. Epidemiology Study Design and Data Analysis. Third ed. Florida: Taylor &Francis Group; 2014.
139. Rothman KJ. Epidemiology An Introduction. Second ed. New York: Oxford University Press;2012.
140. Breslow NE, Day NE. Statistical Methods in Cancer Research The Design and Analysis ofCohort Studies. Lyon: International Agenct for Research on Cancer; 1987.
141. Rothman KJ, Greenland S. Modern Epidemiology. Third ed. Philadelphia: Lippincott Williamsand Wilkins; 2008.
73
142. Last JM, editor. A Dictionary of Epidemiology. Fourth ed. New York: Oxford University Press;2001.
143. Sobin L, Gospodarowicz M, Wittekind C, editors. TNM Classification of Malignant Tumours.Seventh ed. New Jersey: Wiley Blackwell; 2009.
144. Walters S, Maringe C, Butler J, Brierley JD, Rachet B, Coleman MP. Comparability of stagedata in cancer registries in six countries: lessons from the International Cancer BenchmarkingPartnership. International journal of cancer Journal international du cancer. 2013 Feb 1;132(3):67685.
145. Bremnes RM, von Plessen C, Sundstrom S. Lungekreft mer aktuelt enn noen gang [Lungcancer of more current interest than before]. Tidsskr Nor Laegeforen. 2006 Aug 10;126(15):1940.
146. Rostad H, Naalsund A, Norstein J, Jacobsen R, Aalokken TM. Er behandlingen av lungekreft iNorge god nok? [Is the treatment of lung cancer in Norway adequate?]. Tidsskr Nor Laegeforen. 2002Sep 30;122(23):2258 62.
147. Strom HH, Bremnes RM, Sundstrom SH, Helbekkmo N, Flotten O, Aasebo U. Concurrentpalliative chemoradiation leads to survival and quality of life benefits in poor prognosis stage III nonsmall cell lung cancer: a randomised trial by the Norwegian Lung Cancer Study Group. Br J Cancer.2013 Sep 17;109(6):1467 75.
148. Norsk Lunge Cancer Gruppe [Internet]. Oslo: Norsk Lunge Cancer Gruppe; 2010 [cited 201618.02]. Available from: http://www.nlcg.no/.
149. Nilssen Y, Strand TE, Fjellbirkeland L, Bartnes K, Møller B. Lung cancer survival in Norway,1997 2011: from nihilism to optimism. Eur Respir J. 2016 Jan;47(1):275 87.
150. Matsuo Y, Chen F, Hamaji M, Kawaguchi A, Ueki N, Nagata Y, Sonobe M, Morita S, Date H,Hiraoka M. Comparison of long term survival outcomes between stereotactic body radiotherapy andsublobar resection for stage I non small cell lung cancer in patients at high risk for lobectomy: Apropensity score matching analysis. Eur J Cancer. 2014 Nov;50(17):2932 8.
151. Mok TS, Wu YL, Thongprasert S, Yang CH, Chu DT, Saijo N, Sunpaweravong P, Han B, MargonoB, Ichinose Y, Nishiwaki Y, Ohe Y, Yang JJ, Chewaskulyong B, Jiang H, Duffield EL, Watkins CL, ArmourAA, Fukuoka M. Gefitinib or carboplatin paclitaxel in pulmonary adenocarcinoma. N Engl J Med. 2009Sep 3;361(10):947 57.
152. Hicks RJ, Kalff V, MacManus MP, Ware RE, Hogg A, McKenzie AF, Matthews JP, Ball DL. (18)FFDG PET provides high impact and powerful prognostic stratification in staging newly diagnosed nonsmall cell lung cancer. J Nucl Med. 2001 Nov;42(11):1596 604.
153. Brustugun OT, Helland A, Fjellbirkeland L, Kleinberg L, Ariansen S, Jebsen P, Scott H, DonnemT, Bremnes R, Berg T, Gronberg BH, Dai HY, Wahl SG, Mangseth K, Helgeland L. Mutasjonstesting vedikke småcellet lungekreft [Mutation testing for non small cell lung cancer]. Tidsskr Nor Laegeforen.2012 Apr 30;132(8):952 5.
74
154. Sandelin M, Berglund A, Sundstrom M, Micke P, Ekman S, Bergqvist M, Bergstrom S, Koyi H,Branden E, Janson C, Botling J. Patients with Non small Cell Lung Cancer Analyzed for EGFR:Adherence to Guidelines, Prevalence and Outcome. Anticancer Res. 2015 Jul;35(7):3979 85.
155. Lee CK, Wu YL, Ding PN, Lord SJ, Inoue A, Zhou C, Mitsudomi T, Rosell R, Pavlakis N, Links M,Gebski V, Gralla RJ, Yang JC. Impact of Specific Epidermal Growth Factor Receptor (EGFR) Mutationsand Clinical Characteristics on Outcomes After Treatment With EGFR Tyrosine Kinase InhibitorsVersus Chemotherapy in EGFR Mutant Lung Cancer: A Meta Analysis. J Clin Oncol. 2015 Jun10;33(17):1958 65.
156. Mitsudomi T, Morita S, Yatabe Y, Negoro S, Okamoto I, Tsurutani J, Seto T, Satouchi M, TadaH, Hirashima T, Asami K, Katakami N, Takada M, Yoshioka H, Shibata K, Kudoh S, Shimizu E, Saito H,Toyooka S, Nakagawa K, Fukuoka M. Gefitinib versus cisplatin plus docetaxel in patients with nonsmall cell lung cancer harbouring mutations of the epidermal growth factor receptor (WJTOG3405):an open label, randomised phase 3 trial. Lancet Oncol. 2010 Feb;11(2):121 8.
157. Shaw AT, Kim DW, Nakagawa K, Seto T, Crino L, Ahn MJ, De Pas T, Besse B, Solomon BJ,Blackhall F, Wu YL, Thomas M, O'Byrne KJ, Moro Sibilot D, Camidge DR, Mok T, Hirsh V, Riely GJ, IyerS, Tassell V, Polli A, Wilner KD, Janne PA. Crizotinib versus chemotherapy in advanced ALK positivelung cancer. N Engl J Med. 2013 Jun 20;368(25):2385 94.
158. Almèn A, Friberg E, Wodmark A, Olerud H. Radiology in Norway anno 2008. Trends inexamination frequency and collective effective dose to the population. StrålevernRapport 2010:12.Østerås: Authority; NRP, 2010.
159. Strand TE, Rostad H, Sorum R, Solberg S, Norstein J. Ventetid på operasjon for lungekreft[Waiting time for surgery of lung cancer]. Tidsskr Nor Laegeforen. 2006 Aug 10;126(15):1894 7.
160. Aarts MJ, van den Borne BE, Biesma B, Kloover JS, Aerts JG, Lemmens VE. Improvement inpopulation based survival of stage IV NSCLC due to increased use of chemotherapy. Internationaljournal of cancer Journal international du cancer. 2015 Mar 1;136(5):E387 95.
161. Solda F, Lodge M, Ashley S, Whitington A, Goldstraw P, Brada M. Stereotactic radiotherapy(SABR) for the treatment of primary non small cell lung cancer; systematic review and comparisonwith a surgical cohort. Radiother Oncol. 2013 Oct;109(1):1 7.
162. Zhang B, Zhu F, Ma X, Tian Y, Cao D, Luo S, Xuan Y, Liu L, Wei Y. Matched pair comparisons ofstereotactic body radiotherapy (SBRT) versus surgery for the treatment of early stage non small celllung cancer: a systematic review and meta analysis. Radiother Oncol. 2014 Aug;112(2):250 5.
163. Ricardi U, Badellino S, Filippi AR. Stereotactic body radiotherapy for early stage lung cancer:History and updated role. Lung Cancer. 2015 Dec;90(3):388 96.
164. Haque W, Szeja S, Tann A, Kalra S, Teh BS. Changes in Treatment Patterns and OverallSurvival in Patients With Early Stage Non Small Cell Lung Cancer in the United States After theIncorporation of Stereotactic Ablative Radiation Therapy: A Population based Analysis. Am J ClinOncol. 2016 Jan 14.
75
165. Arriagada R, Bergman B, Dunant A, Le Chevalier T, Pignon JP, Vansteenkiste J. Cisplatin basedadjuvant chemotherapy in patients with completely resected non small cell lung cancer. N Engl JMed. 2004 Jan 22;350(4):351 60.
166. Feinstein AR, Sosin DM, Wells CK. The Will Rogers phenomenon. Stage migration and newdiagnostic techniques as a source of misleading statistics for survival in cancer. N Engl J Med. 1985Jun 20;312(25):1604 8.
167. Wouters MW, Siesling S, Jansen Landheer ML, Elferink MA, Belderbos J, Coebergh JW,Schramel FM. Variation in treatment and outcome in patients with non small cell lung cancer byregion, hospital type and volume in the Netherlands. Eur J Surg Oncol. 2010 Sep;36 Suppl 1:S83 92.
168. von Plessen C, Strand TE, Wentzel Larsen T, Omenaas E, Wilking N, Sundstrom S, Sorenson S.Effectiveness of third generation chemotherapy on the survival of patients with advanced non smallcell lung cancer in Norway: a national study. Thorax. 2008 Oct;63(10):866 71.
169. The World Bank. GINI index (World Bank estimate) [Internet]. Washington (DC): The WorldBank; 2015 [cited 2015 07.10]. Available from: http://data.worldbank.org/indicator/SI.POV.GINI.
170. Detterbeck FC, Mazzone PJ, Naidich DP, Bach PB. Screening for lung cancer: Diagnosis andmanagement of lung cancer, 3rd ed: American College of Chest Physicians evidence based clinicalpractice guidelines. Chest. 2013 May;143(5 Suppl):e78S 92S.
171. Aberle DR, Adams AM, Berg CD, Black WC, Clapp JD, Fagerstrom RM, Gareen IF, Gatsonis C,Marcus PM, Sicks JD. Reduced lung cancer mortality with low dose computed tomographicscreening. N Engl J Med. 2011 Aug 4;365(5):395 409.
