a comparison of patient characteristics and survival in two trauma centres located in different...

A comparison of patient characteristics and survival in twotrauma centres located in di�erent countries

John Templetona,*, Peter A. Oakleya, Gilbert MacKenzieb, Alexandra L. Cooka,Dawn Brandc, Richard J. Mullinsc, Donald D. Trunkeyc

aDepartment of Trauma Research, North Sta�ordshire Hospital NHS Trust, Princes Road, Hartshill, Stoke-on-Trent ST4 7LN, UKbCentre for Medical Statistics, Keele University, Sta�ordshire, UK

cOregon Health Sciences University, Department of Surgery, L223A, 3181 S.W. Sam Jackson Park Rd, Portland, OR 97201-3098, USA

Accepted 13 March 2000

Abstract

Introduction: The aim of the study was to compare patient characteristics and mortality in severely injured patients in two

trauma centres located in di�erent countries, allowing for di�erences in case-mix. It represents a direct bench-marking exercisebetween the trauma centres at the North Sta�ordshire Hospital (NSH), Stoke-on-Trent, UK and the Oregon Health SciencesUniversity (OHSU) Hospital, Portland, Oregon, USA.Methods: Patients of all ages admitted to the two hospitals during 1995 and 1996 with an Injury Severity Score >15 were

included, except for those who died in the emergency departments. Twenty-three factors were studied, including the InjurySeverity Score, Glasgow Coma Score, mechanism of injury and anatomical site of injury. Outcome analysis was based onmortality at discharge.

Results: The pattern of trauma di�ered signi®cantly between Stoke and Portland. Patients from Stoke tended to be older,presented with a lower conscious level and a lower systolic blood pressure and were intubated less frequently before arriving athospital. Mortality depended on similar factors in both centres, especially age, highest AIS score, systolic blood pressure and

Glasgow Coma Score.The crude analysis of mortality showed a highly signi®cant odds-ratio of 1.64 in Stoke compared with Portland. Single-factor

adjustments were made for the above four factors, which had a similar in¯uence on mortality in both centres. Adjusting for the

®rst three factors individually did not alter the odds-ratio, which stayed in the range 1.53±1.59 and remained highly signi®cant.Adjusting for the Glasgow Coma Score reduced the odds-ratio to 0.82 and rendered it non-signi®cant. In a multi-factor logisticregression model incorporating all of the factors shown to in¯uence mortality in either centre, the odds-ratio was 1.7 but wasnot signi®cant.

Conclusion: The analysis illustrates the limitations and pitfalls of making crude outcome comparisons between centres. Highlysigni®cant di�erences in crude mortality were rendered non-signi®cant by case-mix adjustments, supporting the null hypothesisthat the two centres were equally e�ective in terms of this short-term indicator of outcome. To achieve a meaningful comparison

between centres, adjustments must be made for the factors which a�ect mortality. 7 2000 Elsevier Science Ltd. All rightsreserved.

1. Introduction

The designation of selected hospitals as trauma

centres is an important part of trauma system plan-ning. Expertise is maintained more easily by providinga high volume of experience and concentratingresources allows full tertiary trauma care to be rapidlyavailable at all times. Evidence indicates that urbantrauma systems have been e�ective in the UnitedStates [1±4] but corresponding evidence is lacking in

Injury, Int. J. Care Injured 31 (2000) 493±501

0020-1383/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved.

PII: S0020-1383(00 )00029 -2

www.elsevier.com/locate/injury

* Corresponding author. Tel.: +44-1782-554379; fax: +44-1782-

554627.

the United Kingdom [5]. In a comment on regionaltrauma system development in the UK, Yates pointedout that the cause, frequency and demographics oftrauma vary signi®cantly between the two countries[6].

Previous comparisons between trauma centres havetended to rely on methods such as TRISS, usingnorms derived from pooled data [7]. Such studies havebeen limited by the di�culty of validating data from alarge number of hospitals as well as by the averagingprocess itself. An alternative approach is a direct com-parison with an equivalent centre. Although few suchstudies have been reported [8], this benchmarkingapproach is becoming popular as a quality improve-ment strategy.

This study compares two urban trauma centres, onein the United States and one in the UK. Both centreshave made substantial investments in trauma registries.Their databases contain details abstracted from themedical records by trained personnel. This informationallowed a direct comparison to be made between thetwo centres, using methods which adjust for theobserved di�erences in case-mix.

1.1. The North Sta�ordshire Hospital

The North Sta�ordshire Hospital (NSH) served as apilot trauma centre in the United Kingdom between1990 and 1994 [9,10]. It is a large acute hospital (1300beds) in an area with a local population of 500,000people. It is surrounded by ®ve district general hospi-tals, resulting in a total catchment population of about1.7 million. Since 1990, the NSH has served as a ter-tiary referral centre for severely injured patients in theNorth West Midlands. Although some patients withsevere trauma were taken directly to the centre ratherthan to the nearest hospital, there was no legal require-ment for such triage. Most patients were taken to thenearest hospital and, if necessary, transferred later tothe centre.

The North Sta�ordshire Hospital Trauma ResearchDepartment collects data on all injured patients whostay longer than 72 h in hospital, are admitted to theICU or are transferred from other hospitals, togetherwith all deaths. These criteria with minor quali®cationsare required by the national Trauma Audit andResearch Network (TARN), previously known as theMajor Trauma Outcome Study (MTOS-UK) Group[11,12], to which the NSH submits data.

1.2. Oregon Health Sciences University

The state of Oregon passed legislation in 1985,which assigned authority over designation of traumacentres to the State Health Division. In 1988, twoLevel I trauma centres were designated in Portland,

one of which was Oregon Health Sciences University(OHSU). The Portland area is the state's largestmetropolitan region with a population of approxi-mately 1 million. All seriously injured patients in thisregion who met triage criteria were transported fromthe scene directly to one of the two Level I traumacentres. These centres also received patients transferredfrom other hospitals throughout the state of Oregon,which has a population of approximately 3 million[13].

The OHSU Trauma Registry collects data on allpatients entered into the trauma system by pre-hospitalpersonnel and brought directly to the OHSU emer-gency department, as well as other patients transferredto the OHSU trauma service from other hospitals. TheOHSU trauma registry also includes injured patientswho present to and are admitted from the OHSUemergency department following injury without beingentered into the trauma system by pre-hospital person-nel.

2. Patients and methods

2.1. Factors and outcomes studied

The cohort selected for study were patients admittedto the two trauma centres who had an Injury SeverityScore (ISS) [14] over 15. Included in the analysis werepatients who presented to the trauma centres from thescene of injury and those who were transferred thereafter ®rst being treated at another hospital. Thepatients were all treated in the years 1995 and 1996.Those patients who were dead on arrival at the emer-gency department or died in the emergency departmentwere excluded from the analysis.

In both institutions, the trauma registry was main-tained by trained abstractors of the medical records.They recorded the patients' diagnoses using the evi-dence of direct observation, imaging, surgical pro-cedures and autopsy. Injuries in the six ISS bodyregions (head and neck, face, chest, abdomen, limbsand external) were scored as 1 Ð minor, 2 Ð moder-ate, 3 Ð serious, 4 Ð severe, 5 Ð critical or 6 Ðunsurvivable, according to the Abbreviated InjuryScale (AIS) coding system [15]. Each patient's highestthree AIS scores were squared and summed to calcu-late the ISS.

The information on 23 items recorded in the traumaregistries of the two institutions were selected foranalysis (Table 1). Trauma registry personnel from thetwo centres met to agree that equivalent de®nitionswere being used. The two datasets were then mergedfor analysis. Twenty-three factors were studied singlyand in combination to compare the patient character-istics, to identify the determinants of mortality in each

J. Templeton et al. / Injury, Int. J. Care Injured 31 (2000) 493±501494

centre and ®nally to compare the mortality betweencentres.

2.2. Statistical methods

A variety of conventional univariate, parametric andnon-parametric statistical techniques were used tocompare outcomes in the two centres. The multiple lin-ear logistic regression model [16], which allows thee�ects of continuous and categorical factors to be eval-uated simultaneously, was used for multi-factor ana-lyses. This model was used repeatedly to compare thepattern of trauma in the two centres and later to com-pare mortality in the two centres. With more than 20factors studied, automatic variable selection procedures(typically the stepwise forward and backward algor-ithms) were used to suggest best ®tting logistic models.These procedures were augmented by ®tting additionalmodels suggested by clinical insight. The 5% level ofstatistical signi®cance was used in tests of statisticalhypotheses and in the construction of con®dence inter-vals.

3. Results

3.1. Patient characteristics: Stoke vs Portland

Tables 2 and 3 and Fig. 1 show single factor com-parisons between the two centres. Table 2 containssimple demographic information, while Table 3 focuseson information relating to the processes of care, such

as length of stay in the ICU or in hospital. Fig. 1 illus-trates the di�erences in the body regions injured.

Of the 23 factors studied, only seven showed no sig-ni®cant di�erence between Stoke and Portland. Thesewere year, quarter, day of admission, sex, medical his-tory, ISS and the proportion of patients admitted withfacial injuries.

In all other respects patients in Stoke di�ered signi®-cantly, often highly so, from patients in Portland.Stoke patients were older on average (40.2 years com-pared with 35.6 years), were less likely to have beeninvolved in a road tra�c accident (51.5% comparedwith 64.6%) and had more falls (35.4% comparedwith 15.2%). They experienced a lower rate of pene-trating injury (1.5% compared with 12.3%), were intu-bated before arrival in the emergency departmentmuch less frequently than patients in Portland (4.1%compared with 22.9%) and were more likely to havebeen transferred from another hospital (39.2% com-pared with 32.9%). In the emergency department,Stoke patients were found to have lower systolic bloodpressures (134.2 vs 137.8 mm Hg) and lower GlasgowComa Scores (10.5 vs 12.9 for the mean; 12.5 vs 14 forthe median). The overall pattern of the site of theinjury was also di�erent. Stoke patients had a higherrate of head and neck injuries (89.5% compared with74.2%) and lower rates at all other sites, except facialinjury where there was no di�erence.

With regard to treatment, Stoke patients spent onaverage 3.6 days in the intensive care unit comparedwith 5.0 days for Portland patients. When consideringonly those patients who were admitted to ICU, the

Table 1

Case-mix factors extracted from the trauma registries

1. Year of admission

2. Quarter

3. Day

4. Age

5. Sex

6. Whether or not there was a signi®cant past medical history

7. Mechanism of injury

8. Whether intubated or not before arrival in the emergency department

9. Systolic blood pressure (SBP) in the emergency department

10. Type of trauma (blunt or penetrating)

11. The presence or absence of injury to the head and neck

12. The presence or absence of injury to the face

13. The presence or absence of injury to the chest

14. The presence or absence of injury to the abdomen and pelvic contents

15. The presence or absence of injury to the extremities

16. The presence or absence of injury to the external regions

17. Whether or not multiply-injured (i.e. involving more than one body region)

18. Whether transferred from another hospital

19. Injury Severity Score (ISS)

20. Glasgow Coma Score (GCS) on arrival in the emergency department

21. Highest Abbreviated Injury Scale (AIS) code (in any body region)

22. Number of days spent in the intensive care unit (ICU)

23. Length of stay in hospital

J. Templeton et al. / Injury, Int. J. Care Injured 31 (2000) 493±501 495

average lengths of stay were similar (6.15 days inStoke compared with 6.56 in Portland). The di�erenceis explained by a higher proportion of patients beingadmitted to the ICU in Portland (76.0% comparedwith 58.2% in Stoke). In terms of the total stay in thetrauma centre, Stoke patients were hospitalised for sig-ni®cantly longer than their Portland counterparts (18.2days compared with 11.2 days, on average; medianstay was 10 and 8 days, respectively).

3.2. Determinants of mortality

As mortality may depend on di�erent factors ineach centre, the determinants of mortality were investi-gated in the two centres separately, before making acomparison.

3.2.1. StokeOn single factor testing, those who died were signi®-

Table 2

Single factor comparisons between Stoke and Portland

Continuous factors

Factor Statistics Stoke Portland Probabilitya

n 390 796

Age (yr) Mean (2SE) 40.2 (1.19) 35.6 (0.73) p=0.008

Median 35.0 31.5

Systolic BP (mm/Hg) Mean (2SE) 134.2 (1.52) 137.8 (1.13) p=0.004

Median 132.0 139.0

ISS Mean (2SE) 25.0 (0.44) 26.3 (0.39) p=0.122

Median 25.0 25.0

GCS Mean (2SE) 10.5 (0.23) 12.9 (0.12) p<0.001

Median 12.5 14.0

ICU stay (days) Mean (2SE) 3.6 (0.32) 5.0 (0.25) p<0.001

Median 1.0 2.0

Length of stay (days) Mean (2SE) 18.2 (1.20) 11.2 (0.36) p<0.001

Median 10.0 8.0

a Based on non-parametric test of ranks (Kruskal±Wallis).

Fig. 1. Comparison of body regions injured.


cantly older (47.5 years compared with 38.6, p =0.006), and had higher Injury Severity Scores (30.5compared with 23.8, p < 0.001), and lower GlasgowComa Scores (6.6 compared with 11.3, p < 0.001).Patients who died spent signi®cantly longer in intensivecare and had a shorter hospital stay (Table 5). Mor-tality was signi®cantly related to four other factors.Stoke patients who were intubated (16 in total) had ahigher death rate (75% compared with 15%, p <0.001), as did patients with co-existing medical history(30.1% compared with 13.5%, p< 0.001) and patients

with a highest AIS score of 5 or more (32.5% com-pared with 7%, p<0.001).

A multi-factor analysis based on the logistic re-gression model was performed (Table 4), excludingtime spent in intensive care and length of stay in hos-pital, as they may be considered to be consequences ofcare processes rather than causes of mortality. Four ofthe factors identi®ed in the single factor analysis Ðage, highest AIS code, medical history and GCS Ðwere found to be independently related to mortality.The e�ects of three additional factors were unmaskedin this analysis Ð systolic blood pressure, mechanismof injury and the presence of head & neck injury. Theanalysis suggests that Stoke patients who fall have asigni®cantly lower mortality (than those involved inroad tra�c accidents) when the e�ects of other factorsin the multiple logistic equation are taken intoaccount. Similarly patients with a head & neck injurywere more likely to die. The magnitude of both ofthese adjusted e�ects is far from marginal.

3.2.2. PortlandThe analysis was repeated in Portland. Those who

died were marginally older on average (39.9 years com-pared with 35.0, p = 0.07: t-test, unequal variances).This result borders on, but does not reach, statisticalsigni®cance. Patients who died also had a higher ISS(36.0 compared with 25.0, p < 0.001), a lower GCS(9.5 compared with 13.3, p < 0.001) and a shorterlength of hospital stay (5.1 days compared with 11.1, p< 0.001). However, unlike the pattern in Stoke, therewas no di�erence in the time spent in the ICU (medianstay 2 days for each group).

Mortality was also signi®cantly related to the mech-anism of injury but the pattern di�ered from Stoke Ðrelative to road tra�c accidents, assaults were morelikely to result in death ( p = 0.003). In Portland, thedeath rate was signi®cantly higher among patients inwhom the trauma was penetrating (23.5% comparedwith 9.7%, p < 0.001), who had been intubated (33%compared with 5.0%, p< 0.001) and who had a high-est AIS score of 5 or more (27.8% compared with3.0%, p < 0.001). As in Stoke, patients who weretransferred had a signi®cantly lower mortality rate(8.0% compared with 13.1%). Mortality was higher inpatients with no history of co-existing medical con-ditions (12.7% compared with 7.7%, p=0.055).

The results of the multi-factor analysis, based on thelogistic regression model, are shown in Table 5. Fiveof the eight risk factors identi®ed in the single factoranalysis Ð GCS, highest AIS code, intubation status,whether transferred and mechanism Ð were found tobe independently related to mortality: the e�ects oftrauma type and ISS were completely explained by theother factors in the model. The multi-factor analysisunmasked the e�ects of age and systolic blood press-

Table 3

Single factor comparisons between Stoke and Portland

Selected categorical factors

Factor Stoke Portland Probabilitya

n=390

(%)

n=796

(%)

Year

1995 186 (47.7) 423 (53.1) p=0.078

1996 204 (52.3) 373 (46.9)

Quarter

January±March 88 (22.6) 159 (20.0) p=0.251

April±June 92 (23.6) 181 (22.7)

July±September 105 (26.9) 259 (32.5)

October±December 105 (26.9) 197 (24.7)

Day

Sunday 61 (15.6) 121 (15.2) p=0.073

Monday 60 (15.4) 104 (13.1)

Tuesday 45 (11.5) 118 (14.8)

Wednesday 63 (16.2) 107 (13.4)

Thursday 52 (13.3) 83 (10.4)

Friday 60 (15.4) 118 (14.8)

Saturday 49 (12.6) 145 (18.2)

Sex

Male 280 (71.8) 562 (70.6) p=0.671

Female 110 (28.2) 234 (29.4)

Mechanism

Road accident 201 (51.5) 514 (64.6) p<0.001

Falls 138 (35.4) 121 (15.2)

Assault 31 (7.9) 110 (13.8)

Other 20 (5.1) 51 (6.4)

Trauma type

Blunt 383 (98.5) 698 (87.7) p<0.001

Penetrating 6 (1.5) 98 (12.3)

Intubation

No 374 (95.9) 614 (77.1) p<0.001

Yes 16 (4.1) 182 (22.9)

Signi®cant past medical history

No 297 (76.2) 600 (75.4) p=0.770

Yes 93 (23.8) 196 (24.6)

Transferred

No 237 (60.8) 534 (67.1) p=0.032

Yes 153 (39.2) 262 (32.9)

Highest AIS score

3 78 (20.0) 204 (25.6) p=0.008

4 152 (39.0) 322 (40.5)

5 159 (40.8) 259 (32.5)

6 1 (0.3) 11 (1.4)

a Based on the null hypothesis of no di�erence.


ure Ð both previously judged non-signi®cant. In Port-land, as in Stoke, mortality increased with age andwith lower systolic blood pressure. The relatively puz-zling ®nding of a positive medical history being associ-ated with a reduced mortality was also con®rmed inthe multi-factor analysis.

3.3. Mortality comparison: Stoke vs Portland

In Portland, the crude mortality rate among studypatients was 11.4% (91/796), whilst in Stoke it was17.4% (68/390), w 2=8.1, df=1, p = 0.004. Theobserved di�erence in mortality, without making anyadjustments for case-mix, yields an odds-ratio of 1.64in Stoke compared with Portland, which was highlysigni®cant ( p=0.005).

The determinants of mortality in the two centres dif-

fer both qualitatively and quantitatively in terms ofthe factors studied. Moreover, not only is mortality inPortland related to a slightly di�erent set of factors,but even when some of the factors are shared, thenature of the relationship may di�er substantiallybetween the centres (e.g. mechanism of injury andmedical history). On the other hand, the relationshipbetween mortality and age, GCS, systolic blood press-ure and the highest AIS code in Stoke is quantitativelysimilar to that observed in Portland (Tables 4 and 5).

As comparisons based on crude mortality rates arepotentially misleading, the strategy adopted was toadjust for the e�ects of the factors identi®ed in theprevious logistic regression analyses. The rates wereadjusted in two stages: ®rstly, for those factors whichhave quantitatively similar e�ects in Stoke and Port-land and secondly for factors which in¯uence mortality

Table 4

Mortality in Stoke

Logistic regression modela: multi-factor analysis

Factor Coe�cient Standard error Probability

Age +0.0469 0.0098 p<0.0001

Mechanismb

Falls ÿ1.7539 0.5028 p=0.0005

Assault ÿ0.0182 0.6672 p=0.9782

Other ÿ0.1252 0.8347 p=0.8808

Systolic BP ÿ0.0120 0.0055 p=0.0297

Presence of head & neck injuryc +1.8237 0.6762 p=0.0070

Highest AIS score +1.0754 0.2947 p=0.0003

Signi®cant past medical historyc +1.1934 0.3891 p=0.0022

GCS ÿ0.2864 0.0480 p<0.0001

a Dead=1, Alive=0.b Road accident used as reference category.c Present=1, Absent=0.

Table 5

Mortality in Portland

Logistic regression modela: multi-factor analysis

Factor Coe�cient Standard error Probability

Age +0.0377 0.0084 p<0.0001

Mechanismb

Falls +0.8753 0.4688 p=0.0619

Assault +0.9771 0.4079 p=0.0166

Other +0.3048 0.6163 p=0.6209

Systolic BP ÿ0.0163 0.0047 p=0.0005

Presence of abdominal injuryc ÿ0.7053 0.3274 p=0.0312

Highest AIS score +1.8150 0.2847 p<0.0001

Signi®cant past medical historyc ÿ3.8393 0.7867 p<0.0001

GCS ÿ0.1494 0.0348 p<0.0001

Intubated before arrivalc +1.7230 0.3320 p<0.0001

Transferredc ÿ0.7017 0.3494 p=0.0046

a Dead=1, Alive=0.b Road accident used as reference category.c Present=1, Absent=0.


in only one centre or which di�er in their relationshipwith mortality between centres. In the latter case, in-teraction terms were introduced to account for thedi�erential mortality between the centres.

Table 6 shows the results of the adjustment process.The ®rst row of the table reproduces the crude odds-ratio for mortality of 1.64 showing the signi®cantlyhigher death rate in Stoke. Adjusting for age, systolicblood pressure and highest AIS code individually didnot alter the odds-ratio, which stayed in the range of1.53±1.59 and remained highly signi®cant. Whenadjusted for the GCS, the odds-ratio is halved to 0.82and rendered non-signi®cant ( p = 0.31). Thus, the in-¯uence of this single factor provides one possible ex-planation for the observed di�erence in mortalitybetween Stoke and Portland. Although the in¯uence ofthe ®rst three factors (age, highest AIS code and systo-lic blood pressure) considered singly did not explainthe observed di�erences in crude mortality, ®ttingthem together reduced the odds-ratio to 1.34 with lossof signi®cance ( p = 0.13). Fitting GCS, highest AIScode, age and systolic blood pressure together reducedthe odds-ratio further (odds-ratio=0.75, p=0.18), butnot signi®cantly so.

In the full model, all of the factors which in¯uencedmortality were ®tted (GCS, age, SBP, highest AIScode, intubation status on arrival at hospital, transfer,head and neck injury, abdominal and pelvic contentsinjury, mechanism of injury and medical history). Twointeraction terms were included in the full logistic re-gression analysis to take into account the opposinge�ects of two factors (medical history and mechanismof injury) in the two centres. The likelihood of the fullmodel matched the sum of the likelihoods obtainedfrom the separate mortality models ®tted in Stoke andPortland. The ®nal adjusted odds-ratio was 1.7, whichwas not signi®cant ( p=0.11).

4. Discussion

This is the ®rst trans-Atlantic comparative study ofits type, comparing patient characteristic and mortalityin two trauma centres. The results should be inter-preted in the context of the current development of thetwo trauma systems [17].

Trauma systems are intended to bene®t severelyinjured patients by implementing comprehensive healthcare policies throughout a de®ned catchment region.They aim to improve care at all stages from pre-hospi-tal management through to long-term rehabilitation[18]. Trauma systems may take many years to mature.Moreover, measures of the e�ectiveness of care in thecentre are insu�cient to judge the e�ectiveness of thesystem as a whole. In a region served by a maturetrauma system, the majority of seriously injuredpatients in a region are treated at trauma centres.

What is not available in this study is the proportionof patients with serious injuries who were treated inhospitals other than trauma centres. Measures of theextent of implementation of regional trauma systemsin the two regions have been previously published.Mullins reported that 80% of seriously injured patientsin Portland were hospitalised in a Level I traumacentre and concluded the system in that city has beene�ectively implemented. In contrast, trauma systemdevelopment was hindered in the North West Mid-lands region of England by the lack of legislative auth-ority. Consequently, the proportion of seriouslyinjured patients triaged directly to the trauma centrewas reported by Nicholl as only 39%. Thus, while thepresent study compared the mortality of severelyinjured patients treated in the trauma centres, it didnot compare the risk of death for all cases of severetrauma hospitalised throughout the two regions.

Hospital mortality as a robust indicator of out-

Table 6

Adjusted mortality Ð Stoke compared with Portland

Adjustment factor(s) Adjusted between centre coe�cient Standard error Adjusted odds-ratio Probability

No adjustment for case mix

None +0.4923 0.1738 1.64 p=0.0046

Single factors

Age +0.4305 0.1758 1.53 p=0.0143

Highest AIS +0.4340 0.1872 1.54 p=0.0204

Systolic BP +0.4641 0.1774 1.59 p=0.0089

GCS ÿ0.2042 0.2015 0.82 p=0.3109

Combinations of main factors

Age+Highest AIS+Systolic BP +0.2957 0.1961 1.34 p=0.1315

Age+Highest AIS+Systolic BP+GCS ÿ0.2903 0.2206 0.75 p=0.1882

Full model

Adjusting for all factors which a�ect mortality (see text) +0.5455 0.3452 1.70 p=0.1141


come has been challenged [19]. Some patients diesoon after discharge. Di�erent health care systemsare under di�erent pressures to discharge patientsand have di�erent options for continuing care. Mor-tality after hospital discharge may vary as a result.Despite these limitations, there is much to learnfrom the direct comparison. The patient character-istics di�ered more than was expected. With a fewexceptions, the determinants of mortality were simi-lar, with the GCS exercising the largest e�ect inboth centres. By adjusting for the factors whichwere shown to in¯uence mortality, the odds-ratiofor mortality in Stoke compared with Portlandunderwent large variations. Apart from adjustmentswhich included the GCS, the propensity was forPortland outcomes to be better than those in Stoke,often signi®cantly. On the other hand, adjusting forthe GCS tended to nullify or reverse the discre-pancy in the outcomes, though never allowing Stokeoutcomes to be signi®cantly better. In the ®nallogistic regression model, which represents a fairoverall comparison, the odds-ratio rose to its high-est value but was not signi®cant. The lack of sig-ni®cance supports the null hypothesis that the twocentres were equally e�ective in terms of mortalityas a short-term indicator of outcome.

These analyses demonstrate that comparisonsbased on crude mortality are potentially misleading.Considerable care should be exercised when compar-ing mortality rates between centres. Adjustmentsshould be made for case-mix factors which are, evi-dentially, related to mortality. However, it is gener-ally impracticable to adjust for all possible subsetsof case-mix factors. Accordingly, the particular fac-tors identi®ed may not be unique. In this study, the®nal logistic model matched in likelihood the separ-ate mortality models for Stoke and Portland, indi-cating that no information had been lost byanalysing the combined model. Various alternativecase-mix models were ®tted and the results obtainedwere similar to those in Table 6.

As in any observational study, the choice of fac-tors for investigation varies. When the number offactors is large, the di�culties of interpreting multi-factor adjustments should not be under-estimated.Nevertheless, it is only by taking into account thein¯uence of the important case-mix factors thatmeaningful comparisons can be made. This studyidenti®es the GCS as a critical factor.

At a time when league tables are becoming fash-ionable as indicators of performance, we must beaware of the limitations of crude comparisons. Thisapplies no less to purchasers of health care than toclinicians. On the other hand, careful comparisonsbetween centres, using validated data and a rigorousanalysis of confounding factors, help identify de-

®ciencies and provide an invaluable insight intohow outcome is in¯uenced.

As trauma systems mature and preventable deathsbecome infrequent, mortality, even when adjusted forcase-mix, is likely to become a poor discriminator. Tobe more discerning in future comparisons betweencentres, analyses of morbidity and comparisons of theprocesses of care will be required.

Acknowledgements

Sponsorship: Work for this study by Dr. Mullinswas supported by grant R49/CCR-006283 from theUS Public Health Service, Centers for Disease Controland Prevention, National Center for Injury Preventionand Control, Atlanta, GA. The authors are solely re-sponsible for the contents of the article, and theopinions do not necessarily represent the views of theCenters for Disease Control and Prevention.

References

[1] West JG, Cales RH, Gazzaniga AB. Impact of regionalization:

the Orange County experience. Archives of Surgery

1983;118:740±4.

[2] Shackford SR, Hollingworth-Fridlund P, Cooper GR, Eastman

AB. The e�ect of regionalization upon quality of trauma care

as assessed by concurrent audit before and after institution of a

trauma system: a preliminary report. Journal of Trauma

1986;24:812±20.

[3] Mullins RJ, Veum-Stone J, Helfand M, et al. Outcome of hospi-

talized injured patients after institution of a trauma system in

an urban area. Journal of the American Medical Association

1994;271:1919±24.

[4] Guss DA, Meyer FT, Baxt WG, et al. The impact of a regiona-

lized trauma system on trauma care in San Diego County.

Annals of Emergency Medicine 1989;18:1141±5.

[5] Nicholl J, Turner J. E�ectiveness of a regional trauma system in

reducing mortality from major trauma: before and after study.

British Medical Journal 1997;315:1349±54.

[6] Yates DW. Regional trauma systems. British Medical Journal

1997;315:1321±2 (Editorial).

[7] Boyd CR, Tolson MA, Copes WS. Evaluating trauma care: the

TRISS method. Journal of Trauma 1987;27:370±8.

[8] Boulanger BR, McLellan BA, Sharkey PW, et al. A Canadian

regional trauma unit versus an American Level I trauma center.

Journal of Trauma 1993;35:261±6.

[9] Redmond A, Johnstone S, Maryosh J, Templeton J. A trauma

centre in the UK. Annals of the Royal College of Surgeons of

England 1993;75:317±20.

[10] Redmond AD. A trauma center for the United Kingdom.

Annals of Emergency Medicine 1993;22:1584±8.

[11] Yates DW, Woodford M, Hollis S. Preliminary analysis of the

care of injured patients in 33 British hospitals: ®rst report of the

United Kingdom major trauma outcome study. British Medical

Journal 1992;305:737±40.

[12] Roberts I, Campbell F, Hollis S, Yates DW. Reducing death

rates in children and young adults: the contribution of hospital

care. British Medical Journal 1996;313:1239±41.

[13] Mullins RJ, Veum-Stone J, Hedges JR, et al. Journal of


Trauma 1996;40:In¯uence of a statewide trauma system on lo-

cation of hospitalization and outcome of injured patients, 536±

546.

[14] Baker SP, O'Neill B, Haddon Jr W, Long WB. The injury

severity score: a method for describing patients with multiple

injuries and evaluating emergency care. Journal of Trauma

1974;14:187±96.

[15] Baker SP, O'Neill B, Haddon Jr W, et al. The Abbreviated

Injury Scale Ð 1990 revision. Des Pleines, IL: Association for

the Advancement of Automotive Medicine, 1990.

[16] Cox DR. Analysis of binary data. London: Methuen, 1970.

[17] Bazzoli GJ, Madura KJ, Cooper GF, MacKenzie EJ,

Maier RV. Progress in the development of trauma systems

in the United States: results of a national survey. Journal

of the American Medical Association 1995;273:395±

401.

[18] Trunkey DD. Towards optimal trauma care. American College

of Surgeons Bulletin 1984;69:1±6.

[19] Mullins RJ, Mann NC, Hedges JR, et al. Adequacy of hospital

discharge status as a measure of outcome among injured

patients. Journal of the American Medical Association

1998;279:1727±31.


a comparison of patient characteristics and survival in two trauma centres located in different...

Documents