Journal of Critical Care (2007) 22, 290–295

Interhospital transfer of critically ill patients:Demographic and outcomes comparison withnontransferred intensive care unit patientsB ,BB

Andrea D. Hill MSca,⁎, Evelyn Vingilis PhDb,c, Claudio M. Martin MD, MSc, FRCPCd,e,Kathleen Hartford PhDc,f, Kathy N. Speechley PhDc,g

aDepartment of Medicine, London Health Sciences Centre, Ontario, CanadabDepartment of Family Medicine, University of Western Ontario, Ontario, CanadacDepartment of Epidemiology and Biostatistics, University of Western Ontario, Ontario, CanadadCentre for Critical Illness Research, Lawson Health Research Institute, London Health Sciences Centre, Ontario, CanadaeDepartment of Medicine, University of Western Ontario, Ontario, CanadafSchool of Nursing, University of Western Ontario, Ontario, CanadagDepartment of Paediatrics, University of Western Ontario, Ontario, Canada

0d

Keywords:Intensive care;Health services;Accessibility;Critically ill

AbstractPurpose: We examined the association between access to intensive care services and mortality in acohort of critically ill patients.Materials and Methods: We conducted an observational study involving 6298 consecutive admissionsto the intensive care units (ICUs) of a tertiary care hospital. Data including demographics, admissionsource, and outcomes were collected on all patients. Admission source was classified as “transfer” forpatients admitted to the ICU from other hospitals, “ER” for patients admitted from the emergency room,and “ward” for patients admitted from non-ICU inpatient wards.Results: Transfer patients had higher crude ICUand hospital mortality rates comparedwith emergency roomadmissions (crude odds ratio [OR], 1.51; 95% confidence interval [CI], 1.32-1.75). After adjusting for age,sex, diagnosis, comorbidities, and acute physiology scores, the difference in ICU mortality remainedsignificant (OR, 1.30; 95% CI, 1.09-1.56); however, hospital mortality did not (OR, 1.19; 95% CI, 1.00-1.41). Compared with ward patients, transfer from other hospitals was associated with lower hospitalmortality after adjusting for severity of illness and other case-mix variables (OR, 0.81; 95% CI, 0.68-0.95).Conclusions:We found some evidence to suggest that differential access to intensive care services impactsmortality within this case mix of patients. These findings may have implications for current efforts tocentralize and regionalize critical care services.© 2007 Elsevier Inc. All rights reserved.

☆ Work for this study was performed at the London Health Sciences Centre, Ontario, Canada, and the University of Western Ontario.☆☆ Partial support was received from Physicians Services Incorporated Foundation.⁎ Corresponding author. Tel.: +1 416 340 4800x2765.E-mail address: andrea.hill@uhn.on.ca (A.D. Hill).

883-9441/$ – see front matter © 2007 Elsevier Inc. All rights reserved.oi:10.1016/j.jcrc.2007.06.002

291Interhospital transfer of critically ill patients

1. Introduction excluded patients admitted from all other sources (including

Critically ill patients presenting at smaller community orrural hospitals, where intensive care services are either notavailable or limited, are appropriately transferred to theintensive care units (ICUs) of tertiary care centers. Interest inthe association between access to intensive care services andoutcomes for critically ill patients has fuelled investigationsexamining the impact of transfer status (whether a patientwas transferred from another hospital or not) on patient andhospital outcomes [1-3]. In addition to being sparse,available evidence examining outcomes for ICU transferand nontransfer patients is also conflicting [1-5]. Investiga-tions to date lack methodological rigor. In particular, smallsample sizes [2,5] and inadequate control for potentialconfounders [3,6] limit inferences. To date, only 2 studieshave used a validated severity-of-illness index for ICUpatients to compare risk-adjusted outcomes for transferredand nontransferred patients [4,7]. Evidence from severalother patient populations, including surgical and internalmedicine, suggests that compared with nontransferredpatients, transferred patients have poorer outcomes, includ-ing higher mortality, morbidity, and costs [8-10].

Given current emphasis on hospital mergers and con-solidation of specialized services within tertiary centers, thetransportation of patients between hospitals will continue tobe an established aspect of health care delivery systems, andit will be increasingly important to establish whethervariations in access to intensive care services impact patientoutcomes. The aim of the present article was to examinewhether differences in hospital mortality exist betweentransfer and nontransfer patients admitted to a tertiary carecenter after adjusting for case mix and severity of illness.

2. Methods

2.1. Patient population

A retrospective cohort study was conducted in the2 adult ICUs at the London Health Sciences Centre, a787-bed tertiary care teaching hospital in Ontario, Canada.These units accounted for all ICU beds in the hospital, with26 and 30 beds, respectively. As the only tertiary careteaching hospital in Southwestern Ontario, it is the primaryICU referral center for this region, with most referralscoming from community and rural hospitals within a 200-km radius. Although no standard retrieval service existsduring the study period, most transfers to the study hospitalsused land ambulance staffed with paramedics and criticalcare nurses.

Consecutive patients admitted to the ICUs from theemergency room (ER), general wards, or other hospitalsbetween January 1, 1998, and December 31, 2002, and whowere at least 16 years old, were included in the study. We

the operating room and transfers between the ICUs) andpatients with missing outcome and severity-of-illness data.For patients readmitted to the ICUs during the samehospitalization, only the first admission was considered foranalyses. We classified patients admitted to ICU from otherhospitals as “transfer” admissions, whereas those admittedfrom the emergency ERs and general inpatient wards wereclassified as “ER” and “ward” admissions, respectively.

2.2. Sample size

Assuming a 35% risk of hospital mortality in transferpatients [11], a sample size of 585 patients was needed ineach group to detect a 10% difference in crude hospitalmortality, with a 2-sided α of 0.017 and power of 90%. Theα was conservatively set at 0.017 (0.05/3) to account forcomparisons between 3 groups.

2.3. Data collection

Since 1995, a minimal data set has been collectedprospectively on all admissions to the 2 ICUs as part of thetertiary center's involvement in a registry-based network[12]. Data for the 5-year study period were extracted fromthis database and included the following variables: sex; age;ICU admission diagnosis classified according to previouslyreported nomenclature [13]; admission source; comorbidconditions (ie, end-stage heart failure, end-stage respiratoryfailure, leukemia, cirrhosis, AIDS, cancer, hepatic failure,immunosuppresion and lymphoma), as previously defined[13]; hospital and ICU length of stay (LOS); admissionseverity of illness as measured by the Acute Physiology andChronic Health Evaluation (APACHE) II score [14]; andvital status at ICU and hospital discharge (dead/alive). Forpatients who were transferred to the study ICUs from othernetwork hospitals, we also extracted data from the sendinghospital, including APACHE II score, admission diagnosis,date of admission, and LOS in the sending hospital. Thereliability of the database has previously been establishedusing independent reabstraction [15].

2.4. Statistical analyses

Data are presented as means ± SD, median, andinterquartile range (IQR) and percentages as appropriate.For categorical variables, comparisons were made using χ2

tests for comparing proportions. Continuous variables werecompared using analysis of variance with Tukey honestlysignificantly difference used to make pair-wise compar-isons. The Kruskal-Wallis test was used to compare ICUand hospital LOS among the 3 groups. For patientstransferred from other network hospitals, the paired t testwas used to compare APACHE II scores from the sendingand study ICUs.

292 A.D. Hill et al.

Independent associations between admission source andICU and hospital mortality were examined using logisticregression analyses. The acute physiology score (APS)component of the admission APACHE II score [14], age, sex,admission diagnoses, and comorbidities were used to adjustfor severity-of-illness and case-mix differences in theseanalyses. Given an inadequate sample size to compare andadjust for all ICU diagnostic categories, ICU diagnosis wascollapsed into 8 mutually exclusive categories according topublished reports [13]. Separate analyses were conducted forthe 3 diagnoses that accounted for the greatest number ofhospital deaths. The odds ratios (ORs) and respective 95%confidence intervals (CIs) are reported.

Logistic regression analyses that included the study ICUto which patients were admitted confirmed that there was nodifference in mortality between the 2 ICUs; therefore, allresults are presented with the 2 ICUs combined. Statisticalanalyses were performed using SAS system, version 8 (SASInstitute, Cary, NC).

3. Results

The database contained 7067 patients whose admissionsource to the ICUs was the ER, general wards, or otherhospitals. After exclusion of patients with missingAPACHE II scores (109 [1.5%]) and readmitted patients

Table 1 Characteristics of study patients by admission source

Variable Transfer (n = 1452)

Male, n (%) 816 (56.2)Mean age ± SD ⁎ (y) 57.9 ±16.9Mean admission APACHE IIscore ± SD ⁎

23.4 ± 9.2

Mean admission APS ± SD † 19.4 ± 8.7ICU admission diagnosis, n (%)Respiratory 249 (17.2)Gastrointestinal 143 (9.9)Cardiovascular 344 (23.7)Neurologic 287 (19.8)Metabolic 81 (5.5)Renal 82 (5.7)Sepsis 111 (7.6)Trauma 128 (8.8)Miscellaneous 27 (1.8)One or more comorbidconditions, n (%)

224 (15.4)

ICU mortality, n (%) 448 (30.9)Hospital mortality, n (%) 509 (35.1)ICU LOS, median (IQR) 3.2 (6.7)Hospital LOS, median (IQR) 10.0 (16.5)

a P values refer to tests of overall equality of means/medians between the groor χ2 analysis (categorical variables).

⁎ P b .05, significantly different among the 3 groups (Tukey test).† P b .05, significantly different for ER patients compared with transfer and

(660 [9.3%]), 6298 (89.1%) patients were available for theanalyses. Table 1 details the demographic and clinicalcharacteristics of these patients. All patients had a medicaladmission diagnosis. However, the distribution of admis-sion diagnoses varied, with the most common diagnosisfor transfer patients being cardiovascular and neurologicdiseases. Admission APACHE II scores differed signifi-cantly as a function of admission source (P b 0.0001).The Tukey post hoc test identified that ER patients hadsignificantly lower admission APACHE II scores relativeto ward and transfer patients (mean, 20.8 ± 9.7, 25.0 ±9.4, and 23.4 ± 9.2, respectively; P b 0.01). There weresmall but significant differences in age between ward andER patients (Table 1).

The median LOS in the ICU for this cohort of patients was2.6 days and differed significantly across the 3 groups(Table 1). As expected given their earlier stay in the ward,patients admitted from the ward had the longest hospitalLOS (Table 1).

The overall crude ICU and hospital mortality rates were29.1% and 34.3%, respectively. Table 2 details theunadjusted and adjusted ORs for ICU and hospitalmortality. In the unadjusted analyses, relative to ERpatients, transfer patients were more likely to die in thehospital (unadjusted OR, 1.47; 95% CI, 1.28-1.69).Logistic regression controlling for age, sex, comorbidities,admission diagnosis, and APS demonstrated that admissionto the ICU from other hospitals was associated with an

ER (n = 2775) Ward (n = 2071) P a

1640 (59.1) 1231 (58.6) 0.1856.2 ±20.2 62.7 ±15.9 b0.000120.8 ± 9.7 25.0 ± 9.4 b0.0001

17.2 ± 8.9 19.9 ± 8.7 b0.0001

514 (18.5) 770 (37.2) b0.0001109 (3.9) 170 (8.2)627 (22.6) 656 (31.7)423 (15.2) 168 (8.1)283 (10.2) 26 (1.3)27 (0.97) 38 (1.8)141 (5.1) 153 (7.4)604 (21.8) 32 (1.6)47 (1.7) 58 (2.8)331 (11.9) 529 (25.5) b0.0001

630 (22.7) 756 (36.5) b0.0001743 (26.8) 909 (43.9) b0.00011.9 (3.7) 3.3 (6.5) b0.00017.0 (12.0) 15.0 (21.0) b0.0001

ups using analysis of variance or Kruskal-Wallis test (continuous variables)

ward patients (Tukey test).

Table 3 Adjusted OR for ICU and hospital mortality bydiagnoses (top 3 diagnoses accounting for hospital death)

Transfer relative Transfer relative

Table 2 Unadjusted and adjusted ORs for ICU and hospitalmortality

Unadjusted OR(95% CI)

Adjusted OR a

(95% CI)P b

Overall ICU mortalityTransfer vsER

1.52 (1.32-1.75) 1.30 (1.09-1.56) 0.004

Transfer vsward

0.78 (0.67-0. 90) 0.91 (0.77-1.09) 0.31

Overall hospital mortalityTransfer vsER

1.47 (1.28-1.69) 1.19 (1.00-1.41) 0.05

Transfer vsward

0.69 (0.60-0.79) 0.81 (0.68-0.95) 0.01

a Adjusted for sex, age, comorbid conditions, ICU admissiondiagnoses, and APS.

b For adjusted OR.

293Interhospital transfer of critically ill patients

increased odds of ICU death relative to patients admitteddirectly to the ICU from the ER (adjusted OR, 1.30; 95%CI, 1.09-1.56). However, the adjusted odds of hospitalmortality for transfer patients relative to ER patients wasnot significant (adjusted OR, 1.19; 95% CI, 1.00-1.41).Relative to ward patients, transfer patients had lower oddsof dying in the hospital (adjusted OR, 0.81; 95% CI, 0.68-0.95). Separate analyses conducted for the 3 diagnoses thataccounted for the greatest number of hospital deaths foundthat compared with patients admitted to the ICU from theER, transfer patients with respiratory diagnoses had 68%higher odds of dying in the hospital (adjusted OR, 1.68;95% CI, 1.15-2.47) (Table 3).

Of the 1452 transfer patients, 121 (8.3%) were transferredfrom hospitals participating in the registry-based network(network transfers). Network and nonnetwork transferpatients were comparable with respect to demographicsand severity-of-illness score (data not shown). Median timeto transfer patients from network member hospitals to thestudy ICUs was 2 days (IQR, 4 days), with 75% of transfersoccurring within 5 days. Acute Physiology and ChronicHealth Evaluation II scores calculated for the first day ofadmission at the sending hospital was significantly lowerthan that calculated on the first day of admission to thereceiving hospital (mean, 21.6 ± 8.0 and 24.0 ± 9.3,respectively; P = .002).

to ER admissions to ward admissionsaAdjusted OR (95% CI)

ICU mortalityNeurologic 1.12 (0.76-1.64) 1.32 (0.84-2.06)Respiratory 2.16 (1.42-3.31) 0.98 (0.71-1.41)Cardiovascular 0.99 (0.70-1.42) 0.78 (0.55-1.1)Hospital mortalityNeurologic 1.05 (0.72-1.53) 1.05 (0.68-1.61)Respiratory 1.68 (1.15-2.47) 0.94 (0.66-1.27)Cardiovascular 1.08 (0.78-1.50) 0.78 (0.57-1.07)

a Adjusted for age, sex, comorbidity, and APS.

4. Discussion

Our findings demonstrated that there were importantdifferences in mortality between ICU patients admitted fromthe hospital's general wards or ER and those transferred fromother hospitals. We found that crude ICU and hospitalmortality rates were significantly higher in transfer patientscompared with patients admitted to the ICU from the ER.The difference in ICU mortality remained significant after

adjusting for age, sex, APS, comorbidities, and admissiondiagnosis; however, hospital mortality differences betweenthe ER and transfer patients were not statistically significantafter adjusting for these covariates.

Despite recent recommendations to regionalize criticalcare services [16-18], little empirical data that compare theoutcomes for patients transferred from other hospitals withpatients admitted from within the referral hospital exist.Similar to other reports [1,2,11,19], we did not findsignificantly higher crude hospital mortality rates fortransfer patients relative to patients admitted from thewards. In examining the issue of accessibility to tertiaryintensive care services, Surgenor and colleagues [2] foundthat hospital mortality rates were not significantly differentbetween transfer patients and patients admitted directly fromthe general ward to the ICU. Goh and colleagues [1] alsofound no significant differences for intensive care pediatricpatients admitted from the ward compared with patientsadmitted from other hospitals. In contrast to the presentstudy, these earlier reports were limited to comparisonsbetween transfer patients and patients admitted to the ICUfrom the general wards.

Similar to the present findings, Rosenberg and colleagues[4] found that transfer patients had a higher ICU mortalitythan ER admissions after adjusting for case mix and severityof illness. However, their observed 2-fold increase in odds ofICU death is higher than seen in the present study. Thedifferent findings across these 2 studies may relate todifferences in patient population. As noted by those authors,a high proportion of transfer patients had more complexdisorders, such as hepatic failure, compared with the patientsin the present study.

Alternative explanations for the increased ICU mortalityobserved for transfer patients vs ER patients in the presentstudy may include selection bias or unmeasured severity ofillness. In the former case, given their failure to respond toprior therapy, transfer patients may represent patients whoare less likely to respond to further treatment and maytherefore be at increased risk of mortality, independent of

294 A.D. Hill et al.

severity of illness or other factors [20]. In the latter case,treatment before and during the transfer may improvepreexisting physiologic abnormalities without concomitantimprovement in the disease process that led to the ICUadmission and may result in underestimation of severity-of-illness measures that rely on physiologic data for theircalculation (such as the APACHE II score). This is referredto as lead-time bias and has been proposed as anexplanation for outcome differences between transferpatients and patients admitted from inpatient sources[21,22]. Although we did not have illness severity scoresjust before transfer, we did have sending hospital admissionAPACHE II scores for a subset of these transfer patients.Based on the higher APACHE II scores found onadmission to the study ICU, our data suggest that lead-time bias may not be an adequate explanation for thedifferences in mortality.

Although risk-adjusted hospital mortality for transferpatients was not significantly different from that of patientsadmitted from the ER, the current study results do notpreclude a significant influence of transfer status on hospitalmortality for individual diagnoses. After adjusting forcovariates, transfer patients with respiratory diagnoses had66% greater odds of dying in-hospital relative to patientsadmitted from the ER. It is plausible that disparities in theproportion of admissions for particular subtypes of respira-tory diagnoses may account for the observed differences.That is, referring health care providers may be more likely totransfer sicker patients, which may preferentially select forsubtypes of respiratory diagnoses that are associated with anincreased risk of death. Given small samples, the data do notallow for examination of individual subtypes within eachmajor diagnostic category.

It is interesting to note that compared with transfer andER patients, the ward patients had higher adjusted hospitalmortality. Although admissions to the ICU from the generalwards are not the focus of this article, these findings mayhave implications for efforts to improve the outcomes forpatients at the study hospital. Although case-mix and lead-time bias may also contribute to this finding, it may alsoindicate the need for strategies for earlier identification ofward patients who are at risk for deterioration andsubsequent transfer to the ICU through such efforts asoutreach teams [23,24].

Several limitations to this study should be noted. First,the data used in the study are from a single tertiary carecenter; therefore, these results may not be generalizable toother hospitals with different patient populations and carepractices. Second, we did not have information about “donot resuscitate” orders or efforts to limit care, such aswithdrawal of life support. These are important factors tomeasure because they may influence mortality outcomes[25,26]. Third, we did not have information from thesending hospital for all transfer patients. Data such asreasons for transfer, location of the patient in the originalhospital before transfer, LOS in original hospital, and the

timing of referral, would have allowed for improvedcharacterization of transfer patients and determination ofthe appropriate comparison group (ER or ward) forassessment of mortality differences. Finally, comparisonswere restricted to ICU and hospital mortality. Informationabout mortality occurring post–hospital discharge was notavailable. In particular, it is plausible that differentialdischarges to other acute and chronic care centers mayhave led to attenuation of the mortality rates observed.Reports examining the impact of discharges to other healthcare facilities on estimates of in-hospital mortality have beeninconsistent [27,28].

Limitations not withstanding, this study improves onprevious investigations by using validated methods to adjustfor case mix and severity of illness. To date, only 2 otherstudies have used similar risk-adjustment measures tocompare outcomes for transfer and nontransfer patients[4,7]. In addition, an important finding of our study was thatoutcomes for ICU transfer patients were not worse thanoutcomes of patients admitted from the general wards, and,in fact, transfer patients were less likely to die in thehospital. Furthermore, this is the first study to describe risk-adjusted mortality for individual diagnostic categories in thecomparison of ICU patients admitted from the ER, generalwards, or other hospitals. Although inferences are limited bysample size, the observation that transfer patients withrespiratory diagnoses were more likely to die in the ICUthan ER patients with respiratory etiologies is important andwarrants further investigation. Ideally, we would comparepatients transferred directly from other hospital ERs with theER comparison group. This level of detail was not available,but inferences are still possible. Because the transferpopulation is an aggregate of patients from the ER andwards of the original hospital, the risk-adjusted analysissuggests that outcomes might be improved by either havingmore resources and services available locally or improvedrecognition and transfer of appropriate patients. The fact thatpatients were potentially admitted to the sending hospitalward may be a consequence of system-related factors anddoes not invalidate our analysis. In addition, our data on thesubset of transfer patients for whom sending hospitalseverity-of-illness scores was available demonstrated higheracuity on admission to the receiving ICU. Although thismay indicate appropriate transfer of a patient with aworsening condition, it highlights the need for researchthat prospectively monitors the transfer of care for the entirepatient encounter.

5. Conclusion

On the basis of these results, it would appear that highermortality may be limited to comparisons between transferpatients and patients admitted directly from the ER to theICU. The implication is that differences in access to

295Interhospital transfer of critically ill patients

intensive care services may impact outcomes within thiscase mix of patients. Although patient transfers areinevitable in current health care systems, this practiceshould be prospectively monitored using data on the entirepatient encounter. In particular, attention should be givento specific diagnostic groups of patients who may havepoorer outcomes.

References

[1] Goh AY, Abdel-Latif M, Lum LC, Abu-Bakar MN. Outcome ofchildren with different accessibility to tertiary pediatric intensive carein a developing country—a prospective cohort study. Intensive CareMed 2003;29:97-102.

[2] Surgenor SD, Corwin HL, Clerico T. Survival of patients transferred totertiary intensive care from rural community hospitals. Crit Care 2001;5:100-4.

[3] Flabouris A. Patient referral and transportation to a regional tertiaryICU: patient demographics, severity of illness and outcome compar-ison with non-transported patients. Anaesth Intensive Care 1999;27:385-90.

[4] Rosenberg AL, Hofer TP, Strachan C, Watts CM, Hayward RA.Accepting critically ill transfer patients: adverse effect on a referralcenter's outcome and benchmark measures. Ann Intern Med 2003;138:882-90.

[5] Duke GJ, Green JV. Outcome of critically ill patients undergoinginterhospital transfer. Med J Aust 2001;174:122-5.

[6] Ridley S, Carter R. The effects of secondary transport on critically illpatients. Anaesthesia 1989;44:822-7.

[7] Combes A, Luyt CE, Trouillet JL, Chastre J, Gibert C. Adverse effecton a referral intensive care unit's performance of accepting patientstransferred from another intensive care unit. Crit Care Med 2005;33:705-10.

[8] Clough JD, Kay R, Gombeski Jr WR, Nickelson DE, Loop FD.Mortality of patients transferred to a tertiary care hospital. Cleve Clin JMed 1993;60:449-54.

[9] Bernard AM, Hayward RA, Rosevear J, Chun H, McMahon LF.Comparing the hospitalizations of transfer and non-transfer patients inan academic medical center. Acad Med 1996;71:262-6.

[10] Riggs JE, Libell DP, Hobbs GR. In-hospital stroke mortality, hospitaltransfers, and referral bias at a rural academic medical center. J RuralHealth 2002;18:294-7.

[11] Escarce JJ, Kelley MA. Admission source to the medical intensive careunit predicts hospital death independent of APACHE II score. JAMA1990;264:2389-94.

[12] Keenan SP, Martin CM, Kossuth JD, Eberhard J, Sibbald WJ.The Critical Care Research Network: a partnership in community-

based research and research transfer. J Eval Clin Pract 2000;6:15-22.

[13] Knaus WA, Wagner DP, Draper EA, Zimmerman JE, Bergner M,Bastos PG, et al. The APACHE III prognostic system. Risk predictionof hospital mortality for critically ill hospitalized adults. Chest 1991;100:1619-36.

[14] Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: aseverity of disease classification system. Crit Care Med 1985;13:818-29.

[15] Chen LM, Martin CM, Morrison TL, Sibbald WJ. Interobservervariability in data collection of the APACHE II score in teaching andcommunity hospitals. Crit Care Med 1999;27:1999-2004.

[16] Hughes CM. Regionalization of critical care medicine. Crit Care Med1995;23:1153-5.

[17] Clark K. Regionalization: economies of scale for acute critical care.J Nurs Adm 2002;32:173-4.

[18] Anonymous. Consensus report for regionalization of services forcritically ill or injured children. Council of the Society of Critical CareMedicine. Crit Care Med 2000;28:236-9.

[19] Durairaj L, Will JG, Torner JC, Doebbeling BN. Prognostic factorsfor mortality following interhospital transfers to the medicalintensive care unit of a tertiary referral center. Crit Care Med 2003;31:1981-6.

[20] Gordon HS, Rosenthal GE. Impact of interhospital transfers onoutcomes in an academic medical center. Implications for profilinghospital quality.

[21] Dragsted L, Jorgensen J, Jensen NH, Bonsing E, Jacobsen E, KnausWA, et al. Interhospital comparisons of patient outcome fromintensive care: importance of lead-time bias. Crit Care Med 1989;17:418-22.

[22] Rapoport J, Teres D, Lemeshow S, Harris D. Timing of intensive careunit admission in relation to ICU outcome. Crit Care Med 1990;18:1231-5.

[23] Hillman K. Critical care without walls. Curr Opin Crit Care 2002;8:594-9.

[24] Ball C, Kirkby M, Williams S. Effect of the critical care outreach teamon patient survival to discharge from hospital and readmission to criti-cal care: non-randomised population based study. BMJ 2001;327:1014.

[25] Kollef MH, Ward S. The influence of access to a private attendingphysician on the withdrawal of life-sustaining therapies in the intensivecare unit. Crit Care Med 1999;27:2125-32.

[26] Jayes RL, Zimmerman JE, Wagner DP, Knaus WA. Variations in theuse of do-not-resuscitate orders in ICUs. Findings from a nationalstudy. Chest 1996;110:1332-9.

[27] Carey JS, Parker JP, Robertson JM, Misbach GA, Fisher AL. Hospitaldischarge to other healthcare facilities: impact on in-hospital mortality.J Am Coll Surg 2003;197:806-12.

[28] Rosenthal GE, Baker DW, Norris DG, Way LE, Harper DL, Snow RJ.Relationships between in-hospital and 30-day standardized hospitalmortality: implications for profiling hospitals. Health Serv Res 2000;34:1449-68.