- Email: [email protected]
Controlling for the severity of injuries in emergency medicine research
Controlling for the Severity of Injuries in Emergency Medicine Research JANE G. MURPHY, PHD,* C. GENE CAYTEN, MD, MPH,*t WILLIAM M. STAHL, MD*+ The injury severity score (ISS) and age have been used retrospectively lo control for trauma severity. Other control variables such as the revised trauma score (RTS) and the TRISS method (which estimates the pmbablllty of survival for each patient) addltlonally require that values of blood pressure, Glasgow coma scale, and respiratory rate, be recorded in the emergency department. The authors question when the RTS, IS& the ISS and age, or the probablllty of survival calculated using the TRISS method should be used to control for sevarlty of injuries in trauma research. Relations between predictor variables and (1) survival to hospital dlscharge, (2) hospital length of stay for survivors, and (3) length of ICU stay were compared by cause of Injury: penetrating, motor vehicle accident, low fall, or other blunt. Data were collected over 12 months for 2,914 consecutive adult patlents who died or stayed In five nontrauma and three trauma centers for 48 hours or more. For survival, the false-negative rates of probability of survival calculated using the TRISS method were appmxlmately half that of the ISS and age; no variable adequately explained survlval among those with low falls. Combinations of ISS, RTS, and age explained the most varlatlon in lengths of hospital stay among survivors, while IS8 explained the most variation In lengths of intensive care unit (ICU) stay. Researchers should consider the ISS with RTS and age to control for severity when lengths of hospltal or ICU stay are studied. The TRISS method should be used in studies of survival. In both cases, the RTS which requires data collection in the emergency department must be calculated. (Am J Emerg Med lggO;8:484-491. 0 1990 by W.8. Saunders Company.)
Because those with injuries make up a large proportion of patients treated in emergency departments (EDs), injuries should be a major focus of research in emergency medicine. Much of the research done to date has tested hypotheses regarding relationships between injured patients’ outcomes and variations in aspects of patient care or trauma systems. As these studies have attempted to define relationships among variables more precisely, the importance of statistically controlling for the severity of patients’ injuries has become more evident. A number of indexes are available for use as severity controls. Three have been widely used. The injury severity score (ISS) was designed by Baker et al’ to express the overall effect of injuries to several body systems. The ISS
can be used in retrospective analyses and can be combined with age to control for severity. The revised trauma score (RTS) developed by Champion et al’” requires data regarding the patient’s physiologic status soon after the injuryeither in the field or in the ED. The TRISS methodology, also developed by Champion et a1,2-6 estimates each patients’ probability of survival. This method combines the RTS, ISS, and age using coefficients developed from the Major Trauma Outcome Study (MTOS). In general, researchers decide which index to use as a control based largely on reports in the literature regarding the relative abilities of those available to explain variation in the dependent variable of interest-the larger the proportion of variation explained, the more useful the index as a control. The literature is inconclusive in this regard in the case of controhing for injury severity. Previous studies of the relationships between the ISS and survival, length of hospital stay, and level of disability at discharge have yielded variable results.’ For example, while Baker et al’ reported a correlation of 0.49 between ISS and mortality among motor vehicle accident victims, Oreskovich et al* found no significant relationship between ISS and mortality among older patients suffering falls. Neither the RTS nor the TRISS method have been extensively tested with outcome variables other than survival. No study has used a single data set to prospectively compare the abilities of these indexes to explain variation in a variety of patient outcome variables. In this study the following question was addressed: When should the RTS, ISS, the ISS and age, or the probability of survival calculated using the TRISS method be used to control for severity of injuries in trauma research? To answer this question, we compared the predictive abilities of each index with respect to three dependent variables: patients’ survival to hospital discharge, hospital length of stay among survivors, and intensive care length of stay among survivors spending at least 1 day in an intensive care unit (ICU).
METHODS Settings andSelection ofCases
From the ‘Institute for Trauma and Emergency Care, New York Medical College Valhalla, NY, TDepartment of Surgery, Our Lady of Mercy Medical Center, Bronx, NY, *Department of Surgery, Lincoln Hospital, Bronx, NY. Supported by the Centers for Disease Control, Public Health Services, Department of Health and Human Services, Atlanta, Georgia, grant number R49CCR202487 (“EMS System Factors: Motor Vehicle versus Other Injuries”). Presented at the UAEM Meetings, May 1989, San Diego, CA. Address reprint requests to Dr Murphy: Institute for Trauma and Emergency Care, New York Medical College, Valhalla, NY. Key Words: Injury Severity Score, Revised Trauma Score, Severity Index, Trauma Research, TRISS method. 0 1990 by W.B. Saunders Company. 07358757/90/0806-0003$5.00/O 484
The study was conducted by the Institute for Trauma and Emergency Care (ITEC) at the New York Medical College (NYMC). Eight hospitals, all affiliates of NYMC, participated in the research. A summary of characteristics of the hospitals is in Table 1. Hospitals ranged in size from 200 to 813 beds. Three of the eight hospitals are offtcially designated Level I trauma centers. Average numbers of trauma admissions per month ranged from 32 to 248, while average numbers of adult major trauma admissions eligible for the study ranged from 12 to 75. Logs of all ED admissions in each hospital were reviewed by one of three trained nurse-abstractor between July 1, 1987
MURPHY, CAYTEN, STAHL l EMERGENCY MEDICINE RESEARCH
TABLE1. Summary of Characteristics of Participating Hospitals
Hospital 1 2 3 4 5 6
No. of Beds
Official Trauma Center
Average No. Trauma AdmlMo
Average No. Adult Major Trauma/MO.
630 613 636 305 200 458 336 222
Yes Yes Yes No No No No No
248 127 ::
75 57 28 14 13 27 34 12
32 75 64 36
and June 30, 1988. Patients 13 years of age or older with injuries were included if they either died or stayed in the hospital for 48 hours or more; patients with major burns or only distal fractures were excluded.
Data Collected A study record was opened for each patient upon his/her identification as eligible. Information from prehospital and ED records was abstracted at that time and was entered on standardized forms. The complete hospital record was reviewed following the patient’s discharge, and final information regarding the patient’s stay and condition at discharge were recorded. A wide range of data elements were collected regarding the care of each patient. For the purpose of this article, important elements included demographic data, information regarding the cause of the injury, and data required to calculate the RTS, ISS and the probability of survival using the TRISS method.
ISS, RTS, and TRISS The Abbreviated Injury Scale (AIS) assesses injury severity on a scale of 1 (minor) to 6 (fatal) in each of several body regions. (AISwas used in this study.) Scores are subjective assessments of severity made by expert raters. Ratings are completed on the basis of reviewing emergency department and hospital records. Baker et al’ designed the ISS to be an anatomic indicator of the overall effect of injuries to several areas of the body. Each of six regions is scored with the highest AIS value given to any injury in that area. AIS values for the three highest scoring body regions then are squared and summed to form the ISS. Baker et al’ found that the ISS correlated more closely with mortality than did the maximum AIS value. The RTS (a refinement of the Trauma Score) was developed by Champion, Sacco, et al.‘& The Score includes three physiological parameters (Glasgow Coma Scale, systolic blood pressure, and respiratory rate) recorded soon after the patient’s injury, generally in a prehospital or ED setting. The coefficients used in the formula were developed using regression analyses based on 2,000 Washington Hospital Center patients and tested using the more than 25,000 patients in the MTOS. (The American College of Surgeons Committee on Trauma initiated the MTOS in 1982. The study includes data submitted voluntarily by trauma centers across the United States and Canada; criteria used by hospitals to enter patients vary.)
485
The TRISS methodology also was developed by Champion, Sacco, et a1.2-6 It allows the probability of survival of any one patient to be estimated given a formula that includes the patient’s ISS, age, and RTS. Coefficients for each component were developed using regression analyses of the MTOS data. Taken across a set of patients, the method allows differences in actual numbers of survivors (or nonsurvivors) to be compared with the numbers predicted based on the MTOS data set.
Statistical Analyses The statistical significance of differences between characteristics of those for whom complete data were and were not available were evaluated using x2 and f-tests. AIS codes were assigned on the basis of the subjective assessments of trained nurse raters. American Association of Automotive Medicine (AAAM) (1985) coding guidelines were followed. These generally parallel the MTOS guidelines for 1988.” However, three differences between MTOS coding and ours deserve description. Under MTOS, initial loss of consciousness is not scored, named vessel lacerations (not further specified) are coded as “major,” and hemomediastinum is not coded but is used to upgrade. Using AAAM guidelines, we scored initial loss of consciousness, gave a lower code to named vessel lacerations (not further specified), and coded hemomediastinum. To our knowledge, there have been no studies of the effect of such differences in guidelines on AIS scores; however, it seems unlikely that the effect is great. The reliability of AIS coding among raters is critical to the generalizability of results. Following the method developed by MacKenzie et ak9 K and weighted K were used to evaluate levels of agreement with regard to AIS coding, and the intraclass correlation coefficient was used to assess the reliability of ISS coding. It was not feasible to test the reliability of RTS coding since those data were entered by ED personnel rather than by our nurse raters. Previous studies of the reliability of Trauma Score coding by Morris et al” and by Moreau et al’* have demonstrated high levels of interrater reliability in field situations. Champion et al4 believe that the simplification represented by the RTS (particularly by including respiration rate rather than “respiratory expansion”) should improve reliability even further. Demographic data were stratified by cause of injury: penetrating, including gunshot and stab wounds; motor vehicle accidents (MVAs), including occupants and pedestrians; low falls (on level surface or from low steps); and, other blunt, including falls from heights and assaults with blunt objects. Groups were compared using x2 analyses or analyses of variance. Logistic regression was used to evaluate relationships between the dichotomous variable “survived to discharge” and the predictor variables (probability of survival, RTS, ISS, ISS and age). False-positive and false-negative rates were calculated based on resubstitution using the results of the logistic regressions. Patients with survival functions greater than or equal to 0.5000 were predicted to live; patients with survival functions less than 0.5000 were predicted to die. False positives were defined as survivors predicted to die; false negatives were deaths predicted to survive.6
AMERICAN
486
JOURNAL
OF EMERGENCY
Pearson product moment correlation coefficients were developed to evaluate bivariate relationships among TRISSderived probabilities of survival, RTS, and ISS and hospital outcomes-lengths of hospital and ICU stays. Squared values of correlation coefficients are provided in tables to estimate the strengths of the linear relationships between variables. Ninety-five percent confidence intervals also are given. Stepwise regression was used to develop multivariate models predictive of lengths of hospital and ICU stays. Potential predictor variables were RTS, ISS, and age. Partial R’s (showing contributions to the overall model by individual variables) and model R’s are given. F-tests for the significance of individual variables and for the overall significance of the model were calculated. Throughout the analyses, age was used both as a continuous variable and as a dichotomous variable (~55 years v ~55 years of age as in the TRISS). Analyses were stratified by cause of injury and by whether or not care was rendered in a Level I trauma center hospital. Length of hospital stay analyses included only patients who stayed in a hospital for 190 days or less and who were discharged alive; length of ICU stay analyses included those in the preceeding group who spent at least one day in an ICU. Even with these restrictions, length of hospital stay and days in an ICU were skewed toward higher values; log transformations of both were used in correlation and regression analyses. RESULTS During the 12-month study period, 8224 patients with injuries were admitted to the eight hospitals. Of these, 2986 (36.3%) were eligible for inclusion in the study. Some element or elements of data required to calculate the probability of survival using the TRISS method were missing for 72 (2.4%) of the eligible cases. Those for whom probabilities of survival were versus were not calculable were statistically
TABLE 2.
Summary of Patients’ Characteristics
MEDICINE
n Volume 8, Number 6 n November 1990
similar with regard to sex, cause of injury, survival rate, Glasgow Coma Scale score in the ED, complications, and number of ICU days. However, those for whom probabilities could not be calculated were significantly younger, and had significantly higher ISS scores (17.6 v 11.4) and longer hospital stays than did those for whom probabilities of survival could be calculated. Interrater agreement between pairs of nurse raters with regard to AIS and ISS coding was tested on 63 medical records. For AIS coding, K ranged from 0.61 to 0.70; weighted K ranged from 0.68 to 0.79. The intraclass correlation coefficient, used to evaluate the reliability of ISS coding, was 0.77. These results are comparable to those reported by MacKenzie et al9 and suggest very good agreement among nurse raters. Table 2 summarizes characteristics of the 2914 patients included in the analyses by cause of injury. Twenty percent (595) of patients suffered penetrating injuries. The mean age among these patients was lower than were average ages in other groups. The proportions of males, and black and hispanic patients were greater. Overall, those with penetrating injuries had the shortest average hospital and ICU stays. Ninety percent of these patients survived. Twenty-seven percent (785 patients) were involved in MVAs. On average, these patients were older than those with penetrating injuries but younger than those with low falls. Over half the patients were white and two-thirds were men. Nearly 93% of patients in MVAs survived. Low falls were the most frequent causes of injuries, accounting for 994 patients or 34% of the total. These patients had the highest average age (74.8 years). Over 80% were white and two-thirds were women. On average, these patients had low ISS scores and high Glasgow Coma Scale scores; however, their average hospital and ICU stays were protracted. Over 93% of patients with low falls survived. Nineteen percent of patients (540) suffered other blunt injuries. Seventy-five percent were men and nearly 40% were
by Injury Cause
Variable*
Penetrating
MVA
Low Falls
Other Blunt
No. of patients Sex (No. (%))t Male Female Race (No. (%))t White Black Hispanic Other Age’tS RTStS -t* Glasgow Coma Scoret* Hospital Days (n = 2,704)t$ ICU Days (n = 613)$§ Survival: No. and rate (%)t
595
785
994
540
530 (89.1) 65 (10.9)
532 (67.8) 253 (32.2)
326 (32.8) 668 (67.2)
427 (79.1) 113 (20.9)
421 153 182 29 39.1 7.3 13.8 13.6 15.1 6.8 729
834 78 77 5 74.8 7.8 8.8 14.6 19.7 7.2 933
213 163 153 11 39.6 7.5 11.1 14.0 13.7 7.2 509
59 264 258 14 29.9 7.0 14.0 13.6 10.2 4.5 534
(9.9) (44.4) (43.4) (2.4) + 10.7 2 2.1 t 10.8 + 3.6 2 17.1 2 6.3 (89.7)
(53.6) (19.5) (23.2) (3.7) + 18.9 ” 1.5 5 11.6 r 3.3 2 20.4 f 10.3 (92.6)
(83.9) (7.8) (7.7) (0.5) ? 16.8 * 0.5 2 4.2 + 1.6 2 16.4 f 8.7 (93.9)
(39.4) (30.2) (28.3) (2.0) 2 18.6 * 1.1 2 9.6 ? 2.8 t 22.1 2 8.6 (94.3)
* Hospital days and complications based on patients who were in the hospital for 190 days or less; ICU days are based on the number of patients spending time in an ICU. X2 or analysis of variance tests were used to evaluate the significance of differences, t P < 0.01 $ Mean + SD. 8 P < .05
MURPHY,
CAYTEN,
STAHL
W EMERGENCY
MEDICINE
RESEARCH
white. These patients were similar to those patients in MVAs with regard to average age, anatomic and physiological indicators, and lengths of hospital stays. Nearly 95% survived. Table 3 summarizes distributions of cases by levels of age, ISS, RTS, and probability of survival. Over 85% of patients with penetrating injuries were under age 40 while nearly 75% of patients suffering low falls were 70 years or older. Those in MVAs and having other blunt injuries were more evenly distributed among the age categories. Nearly 40% of those with penetrating injuries had ISS values of 16 or above while over 90% of patients suffering from low falls had ISS values of less than 13. Again, those in MVAs and having other blunt injuries were more evenly distributed among categories. Over 80% of those in each category had RTS values of greater than 7 and probabilities of survival greater than 0.95. Nearly 8% of those with penetrating injuries had probabilities of survival of less than 20%. The relations between predictors and survival to discharge are shown in Tables 4 and 5. The false positive rates of all predictors were very low-0.03 or below. The TRISSgenerated probabilities of survival had the lowest falsenegative rates among those with penetrating injuries, in MVAs, or having other blunt injuries: 0.23, 0.45, and 0.48, respectively. These rates were reasonably stable when trauma center status was considered (Table 5). The falsenegative rates of the RTS taken alone approached those achieved by probabilities of survival, while those for the ISS with or without age/age score were high. All of the predictor variables had false-negative rates of 0.95 or above for patients having low falls. Bivariate relations between predictors and hospital outTABLE 3.
Distributions
Variable
of Cases
by Age,
ISS Score, Penetrating
No. of patients
and TRISS
487
TABLE 4. False-Positive/False-Negative Survival Predictors
of
Variable
Penetrating
MVA
Falls
Other Blunt
No. of patients Prob. Survival (TRISS) RTS ISS ISS and Age ISS and Age Score
595
705
994
540
0.02/0.23 0.02/0.28 0.0310.75 0.02!0.77
0.01/0.45 0.01/0.40 0.0110.66 0.02/0.68
0.01/0.90 0.00/l .oo 0.00/l .oa 0.00/l .oo
0.0110.46 0.01/0.55 0.0210.71 0.01/0.61
0.03/0.75
O.OLTYO.64
0.00/l .oo
0.0110.64
Low
NOTES. False-positive/false-negative rates calculated based on resubstitution using logistic regression functions. Patients with survival functions greater than or equal to 0.5000 were predicted to live: patients with survival functions less than 0.5000 were predicted to die. Age Score was dichotomized: age greater than or equal to 55 years vs less than 55 years. False-positive rate = % of survivors predicted to die. False-negative rate = % of deaths predicted to survive.
comes are shown in Tables 6 and 7. The ISS showed the strongest linear correlation with length of hospital stay both overall and when trauma center status was taken into account; the RTS generally showed the weakest correlation with hospital lengths of stay. There was little overlap between the confidence intervals for ISS v RTS or probability of survival when the full data set was considered. More overlap occurred when sample sizes were reduced by considering trauma center status. It is important to note that no predictor was strongly associated with length of hospital stay among those having low falls. Relationships with lengths of ICU stay were less clear. The ISS had the strongest linear correlation with ICU stay among those in MVAs, and those having low falls or blunt injuries. However, there was some overlap in the confidence (Probability
of Survival)
MVA
Low Falls
785
994
595
Age 13-29 30-39 40-49 50-59 60-69 70+ ISS Score
Rates
Other
Blunt 540
349 164 54 18 6 4
(58.7) (27.6) (9.1) (3.0) (1 .O) (0.7)
331 144 102 73 60 75
(41.7) (17.8) (13.3) (9.3) (7.5) (10.3)
24 41 37 60 104 728
(2.4) (4.1) (3.7) (6.0) (10.5) (73.2)
201 150 58 50 30 51
(37.2) (27.8) (10.7) (9.3) (5.6) (9.4)
1-8 9-12 13-15 16-19 20+ RTS
162 155 42 88 148
(27.2) (26.0) (7.1) (14.8) (24.9)
257 201 87 76 164
(34.0) (26.5) (11.3) (9.5) (18.7)
240 655 34 36 29
(24.1) (65.9) (3.4) (3.6) (2.9)
225 164 29 50 72
(41.7) (30.4) (5.4) (9.3) (13.3)
O-l 2-3 4-5 6-7 27 Prob. Survival
40 17 14 28 496
(6.7) (2.9) (2.4) (4.7) (63.4)
21 18 43 42 661
(2.7) (2.3) (5.5) (5.4) (84.2)
2 2 15 19 956
(0.2) (0.2) (1.5) (1.9) (96.2)
7 6 26 27 474
(1.3) (1.1) (4.8) (5.0) (87.8)
45 15 4 24 507
(7.6) (2.5) (0.7) (4.0) (85.2)
29 18 21 48 669
(2.5) (3.0) (2.0) (7.5) (85.0)
3 2 7 35 947
(0.3) (0.2) (0.7) (3.5) (95.3)
9 7 9 21 494
(I .7) (1.3) (1.7) (3.9) (91.5)
co.20 0.20-0.49 0.50-0.79 0.80-0.95 >0.95
(TRISS)
AMERICAN
488
TABLE 5. False-Positive/False-Negative Predictors by Trauma Center Status
JOURNAL
Rates of Survival
Penetrating
MVA
Falls
Other Blunt
513
508
360
300
0.02/0.24 0.02/0.30 0.03/0.76 0.03/0.78 0.0410.76
0.02/0.47 0.01/0.47 0.02/0.63 0.02IO.65 0.03/0.61
0.0111 .oo 0.0010.96 0.01/l .oo 0.01/0.96 0.00/1 .oo
0.0110.46 0.01/0.58 0.01/0.54 0.0110.62 0.00/0.54
82
277
634
152
0.0110.23 0.0110.43 0.01/0.75 0.0110.77 +
0.01/0.43 0.0110.43 0.01/0.71 0.01/0.86 0.0110.71
0.00/0.95 0.00/0.95 0.00/l .oo 0.00/l .oo 0.00/l .oo
0.00/0.57 0.00/0.57 0.02/i .oo 0.01/1.00 0.0111 .oo
Low
Variable Trauma Centers No. of patients Prob. Survival (TRW) RTS ISS ISS & Age ES 8. Age Score Nontrauma Centers No. of patients Prob. Survival (TRISS) RTS ISS ISS 8. Age ISS & Age Score
OF EMERGENCY
NOTES. False-positive/false-negative rates calculated based on resubstitution using logistic regression functions. Patients with survival functions greater than or equal to 0.5000 were predicted to live: patients with survival functions less than 0.5000 were predicted to die. Age Score was dichotomized: age greater than or equal to 55 years vs less than 55 years. False positive rate = % of survivors predicted to die. False negative rate = % of deaths predicted to survive. + Parameters for Age Score regarded to be infinite.
intervals for probabilities of survival and IS.5 scores, while squared correlation coeffkients for these predictors were nearly identical among those having penetrating injuries. Small numbers of cases precluded an analysis of ICU stay stratified by trauma center status. Multivariate relationships between RTS, ISS, age/age score, and hospital outcomes are shown in Tables 8 and 9. For the total sample, proportions of variation in lengths of hospital stays explained by combinations of variables ranged from 0.32 for those with other blunt trauma to 0.15 for those having low falls. Explanatory power was reasonably stable across trauma center status. TABLE 6.
Bivariate
Relations
Variable Length of Hospital No. Prob. Survival (TRISS) RTS ISS Length of KU Stay No. Prob. Survival (TRISS) RTS ISS
Between
Predictors
and Hospital
MEDICINE
n Volume 8, Number 6 n November
1990
The ISS was the first variable to enter all models with length of hospital stay as the dependent variable. Sixty percent to 90% of the total variation explained by each model was attributable to the ISS alone. The RTS entered all models with the exception of those for low falls patients and for patients with other blunt trauma cared for in nontrauma centers. Age or age score entered most models as the third variable. Comparing models for the same data set using age v age score showed small differences in explanatory power (usually less than one percent) when one or the other agerelated variable was used. With respect to lengths of ICU stays, ISS again was the first variable to enter all models-and was the only variable that significantly explained variation in stays among patients having low falls. RTS added to the explanatory power of models for penetrating injury patients and for those in MVAs. DISCUSSION This research began with the question: When should the RTS, ISS, the ISS and age, or the probability of survival calculated using the TRISS method be used to control for severity of injuries in trauma research? The results indicate that the answer depends on what outcome variable is being addressed. Probability of survival calculated using the TRISS method was the best discriminator between survival/nonsurvival to hospital discharge. The false-negative rates for probability of survival were somewhat lower than those for RTS and approximately half those of the ISS, and ISS and age/age score. The range of false-negative rates (23% to 48%) for three of the four injury cause categories parallel the 30% to 40% rates reported as “typical” by Sacco et aL6 The rates were generally steady across injury types and trauma center designation categories-with one major exception. Neither probaOutcomes
Penetrating
WA
Low Falls
Other Blunt
534 0.08 (0.04/0.12) 0.09 (0.05/0.14) 0.22 (0.17/0.28)
728 0.15 (0.1 l/0.20) 0.13 (0.09/0.18) 0.24 (0.19/0.30)
933 0.03 (0.0110.05) co.01 (<0.01/0.01) 0.11 (0.0710.14)
509 0.10 (0.05/0.16) 0.04 (0.01/0.07) 0.28 (0.21/0.34)
132 0.17 (0.07/0.29) 0.13 (0.05/0.26) 0.16 (0.08/0.28)
275 0.22 (0.13/0.30) 0.21 (0.13/0.30) 0.31 (0.21/0.40)
Stay
99 0.06 (0.01/0.17) 0.02 (0.01/0.10) 0.20 (0.08/0.36)
107 0.16 (0.05/0.29) 0.06 (0.01/0.18 0.36 (0.20/0.49)
NOTES. Squared correlations (P) and (lower/upper 95% confidence intervals) between Prob. (probability of) Survival, RTS and ISS, and hospital outcomes. Length of hospital stay analyses included only patients discharged alive; ICU analyses included only patients spending time in an ICU and discharged alive. Log transformations of both outcome variables were used. In all cases, relationships between lengths of hospital and ICU stay, and RTS and Probability of Survival were negative; relationships between lengths of hospital and ICU stay, and ISS were positive. All coefficients were significant at the 0.01 level or beyond, except where ? is cO.01.
MURPHY, CAYTEN, STAHL I EMERGENCY MEDICINE RESEARCH
TABLE7.
Bivariate Relations Between Predictors and Hospital Outcomes by Trauma Center Status
Variable Trauma Centers Lengrh of Hospital Stay No. Prob. Survival (TRISS) RTS ISS Nontrauma Centers Length of Hospital Stay No. Prob. Survival (TRISS) RTS ISS
489
Penetrating
MVA
459 0.08 (0.03/0.12) 0.08 (0.04/0.13) 0.21 (0.13/0.30)
459 0.13 (0.08/0.19) 0.12 (0.07/0.18) 0.26 (0.19/0.32)
336 0.04 (O.Ol/O.OS) co.01 (<0.01/0.03) 0.11 (0.05/0.18)
75 0.16 (0.04/0.32) 0.20 (0.06/0.38) 0.32 (0.14/0.48)
269 0.20 (0.1 l/0.29) 0.13 (0.06/0.21) 0.20 (0.1 l/0.29)
597 0.04 (0.02/0.07) co.01 (~0.0110.01) 0.10 (0.06/0.15)
Low Falls
Other Blunt
364 0.09 (0.04/0.15) 0.05 (0.01/0.08) 0.27 (0.1910.35)
145 0.21 (0.08/0.30) 0.01 (<0.01/0.05) 0.28 (0.16/0.40)
NOTES. Squared correlations (P) and (lower/upper 95% confidence intervals) between Prob. (probability of) Survival, RTS and ISS, and hospital outcomes. Length of hospital stay analyses included only patients discharged alive; ICU analyses included only patients spending time in an ICU and discharged alive. Log transformations of both outcome variables were used. In all cases, relationships between lengths of hospital and ICU stay, and RTS and Probability of Survival were negative; relationships between lengths of hospital and ICU stay, and ISS were positive. All coefficients were significant at the 0.01 level or beyond, except where P is ~0.01.
bility of survival, RTS, ISS, nor ISS and age/age score adequately predicted survival among those having low falls. Variability in length of hospital stay among survivors was best explained by combinations of ISS, RTS, and either age or age score. Between 15% and 39% of variation in length of stay was explained by combinations of these variables. Explanatory power was weakest among those injured in low falls. The ISS was the first variable to enter all models; partial R*s for RTS and age/age score were less than half-and usually less than one fourth-those for ISS. ISS was most closely associated with lengths of intensive care stays among survivors spending at least 1 day in an ICU. RTS also entered the regression models for penetrating injuries and MVAs. Again, explanatory power was weakest among those having low falls. The fact that the probability of survival calculated with the TRISS method was the best predictor of actual survival with RTS a close second in predictive abilities is not surprising. Using the MTOS data set, Champion et a12-6developed the coefficients underlying the probability of survival and RTS calculations by conducting regression analyses with survival as the dependent variable. However, it is significant that the strong relations between probability of survival and RTS with actual survival remained stable across trauma center status and three of the four categories of injury cause. A variety of statistical techniques have been used to evaluate the relation between ISS and survival, making comparisons among results difficult. For example, Semmlow and ConeI studied victims of both vehicular and nonvehicular trauma in Illinois. Probit analyses of relationships between ISS values and percent mortality showed a strong monotonic relation between variables. Copes et alI4 grouped ISS scores into categories or “bins”; they also reported a strong positive relation between increasing ISS scores and mortality in all injury groups. Studying only those injured in MVAs,
Baker et al’ reported a correlation coefficient of 0.49 between ISS and survival. Using logistic regression to analyze survival, we found the ISS (with or without age/age score) to have low false-positive rates but unacceptably high falsenegative rates. This relationship may have been masked by statistical tests used in previous work. None of the potential predictor variables discriminated well between survivahnonsurvival among those suffering low falls. It is unclear what proportion of patients in the MTOS data set had low falls. However, some hospitals participating in the MTOS may not consider these patients to be trauma patients; differences in definitions of trauma in the MTOS study may explain the lack of association between probability of survival and the RTS and actual survival among patients with low falls in the present study. The relation between ISS and survival among older patients, and particularly those suffering falls, has been under debate. Oreskovich et al* studied patients over 70 years of age with multiple injuries. Patients with falls constituted the largest proportion of patients and had the highest death rate. No strong relation between ISS values and survival was detected. Long et al” studied over 2,000 trauma victims; all who died with ISS scores of 15 or less were 69 years of age or older and were victims of falls. Baker et al’ also noted that increases in mortality among older patients was most notable when injuries were least severe, ie, when ISS values were lowest. However, Finelli et all6 found the average ISS of elderly nonsurvivors of traumatic injuries to be two-to-three times that of survivors. Within our sample, the mean ISS for nonsurviving low falls patients (average age, 74.8 years), was approximately 40% higher than that of survivors (P < 0.01); but, we also found a 100% false-negative rate for ISS as a predictor of survival. To our knowledge, the relations between probability of survival and RTS, and hospital or ICU days have not been
AMERICAN JOURNAL OF EMERGENCY MEDICINE W Volume 8. Number 6 n November 1990
490
TABLE8. Relations Between Predictors and Hospital Outcomes
Injury Cause Length of Hospital Stay Penetrating (534) MVA (728) Low Falls (933) Other Blunt (509) Length
Variables Included in Most Predictive Model/Partial r* ISS/O.219 RTS10.034 IWO.244 RTS10.027 Age/O.016 Iss10.107 Age/O.047 ISS10.278 Age Score/O.030 RTS/O.OOS
Model r2 0.253 0.207
0.154 0.317
of ICU Stay
Penetrating (132) MVA (275) Low Falls (99) Other Blunt (107)
0.244
IWO.160 RTSl0.084 IWO.308 FITS/O.058 Iss10.200
0.200
IWO.359
0.359
0.366
NOTES.Variables are listed in the order in which they entered stepwise regression analyses. Length of hospital stay analyses included only patients discharged alive: ICU analyses included only patients spending time in an ICU and discharged alive. Log transformations of both outcome variables were used. Age Score was dichotomized as <55 years vs. 255 years of age. Age was a continuous variable. Two sets of regression equations were developed for each injury cause; one used age with RTS and ES, while the other used Age Score with RTS and ISS. In all cases, relationships between lengths of hospital and ICU stay, and RTS and Probability of Survival were negative; relationships between lengths of hospital and ICU stay, and ISS and age/age score were positive. All variables entered were significant at 0.01 or beyond: all models were significant at Pr < 0.0001 (F-tests).
No other research has attempted to stratify analyses of indexes by trauma center designation. Our results indicate that the indexes tested perform equally well across types of hospitals. The research has limits. Data required to calculate probabilities of survival were unavailable for less than 3% of patients. It seems unlikely that the results have been biased by the omission of this small number of patients, but we cannot estimate the effect precisely. In addition, although the sample was large (nearly 3,000 cases), the subsamples were relatively small. For that reason, the relations between lengths of ICU stays and the various predictor variables could not be analyzed with respect to trauma center status. Con& dence intervals of correlation coefficients would likely have been more distinct if a larger sample had been available. The results of the present study suggest two important topics for further investigation. First, a severity index that better explains variation in survival among those with low falls is needed. In our sample, the group having low falls was relatively homogeneous with regard to age, ISS. and RTS values. It seems likely that an index that can discriminate between survivors and nonsurvivors will include comorbidity indicators such as numbers of complications, and indicators of the existence and severity of preexisting conditions, which might show more variation.‘6,‘8 Relations between survival rates of patients having low falls and predictor variables will be difficult to explain until such an index is developed. Second, the results of this study are suggestive with reTABLE9. Hospital
evaluated. Given that the TRISS method and the RTS were developed as predictors of survival, it is not surprising that they did not perform well as independent predictors of variables related to hospitalization. In addition, our analysis of hospital and ICU days included only survivors while probability of survival and the RTS were developed using information concerning both survivors and nonsurvivors. Other researchers have reported significant relations between the ISS and hospital length of stay. Bull” found that ISS values separated survivors into groups with statistically different mean lengths of hospitalizations. Semmlow and ConeI reported an approximately linear increase in length of stay with increasing ISS values. Neither tested the explanatory abilities of the ISS in combination with RTS, age or age score; however, we found that these variables significantly increased the proportion of variation explained in length of hospital stays. The relation between numbers of ICU days among survivors spending at least 1 day in an ICU and the several independent variables had not been tested previously. ISS proved to be most closely related to ICU days with RTS adding to the explanatory power among those in MVAs and with penetrating injuries. Both age (a continuous variable) and age score (a dichotomous variable) were tested in the various models developed. No clear pattern emerged. We found little difference in explanatory power between models using age v age score suggesting that researchers should test both and use whichever provides an optimum model.
Relationships Between Predictors Stay: by Trauma Center Status’
Injury Cause
and Length
Variables Included in Most Predictive Model/Partial r2
of
Model r*
Trauma Centers
Penetrating (459) MVA (459) Low Falls (336) Other Blunt (364) Penetrating (75) MVA (269) Low Falls (597) Other Blunt (145)
ISSiO.207 RTW0.032 ISS10.263 RTWO.081 Age/O.013 Iss/o.115 Age/O.066 lSWO.274 Age Score/O.024 RTSI0.014 ISS10.316 RTSI0.062 ISWO.196 RTW0.048 Age/O.037 Iss10.100 Age/O.036 ISS10.282 Age/O.086
0.239 0.294
0.180 0.305
0.378 0.281
0.136 0.368
NOTES.Variables are listed in the order in which they entered stepwise regression analyses. Length of hospital stay analyses included only patients discharged alive; ICU analyses included only patients spending time in an ICU and discharged alive. Log transformations of both outcome variables were used. Age Score was dichotomized as ~55 years vs. 255 years of age. Age was a continuous variable. Two sets of regression equations were developed for each injury cause; one used age with RTS and ISS, while the other used Age Score with RTS and ISS. In all cases, relationships between lengths of hospital and ICU stay, and RTS and Probability of Survival were negative; relationships between lengths of hospital and ICU stay, and ISS and age/age score were positive. All variables entered were significant at 0.01 or beyond: all models were significant at Pr < 0.0001 (F-tests).
MURPHY, CAYTEN, STAHL n EMERGENCY MEDICINE RESEARCH
spect to the possible development of a TRISS-type predictor of lengths of hospital and ICU stays. With the increasing concerns in our society concerning hospital resource allocation it is important to understand how well existing measures predict hospital outcomes. We found that the variables included in the TRISS method (ISS, RTS, and age score) explained reasonably large proportions of variation in these hospital outcome variables. However, our data set was not sufficiently large to develop and to test a predictive model. A larger study would contribute significantly by providing a model that predicted either categories of lengths of stay or diagnostic related groups (DRG) outliers. Such research also might test whether different coefficients for components of the RTS would enhance the measure’s prediction of hospital outcomes. CONCLUSION Emergency medicine researchers concerned with injuries must control for the severity of those injuries. Relations among patient outcome variables, and variations in patient care or trauma system factors will be obscured or distorted unless severity is statistically controlled. No research existed to guide researchers in the selection of a control measure. We compared the amounts of variation explained by the TRISS-derived probability of survival, the RTS, the ISS and the ISS with age in regard to three dependent variables: survival, length of hospitalization among survivors, and length of ICU stay among those spending at least 1 day in an ICU. The results indicate that the selection of a control for injury severity depends on the outcome variable of interest. The ISS (which can be calculated retrospectively from hospital records) and age can be used to control for severity if hospital length of stay among survivors or days spent in an ICU are of concern. However, the explanatory power of these variables will be enhanced in most cases by the addition of the RTS to the model. Moreover, if survival is to be studied among patients with injuries other than low falls, probabilities of survival-which include the RTS-should be developed. Calculation of the RTS requires the careful collection of data regarding patients’ blood pressures, Glasgow Coma Scale scores, and respiratory rates in the ED. If these physiologic parameters are not recorded in the ED, the opportunity to control for severity in investigations into factors affecting patient survival will be lost.
491
The authors wish to thank Bea Clanton, RN, Karen Ende, RN, Gail Murray, RN, Sara Nealon, RN, and Deborah Vierra, RN, for collecting the data reported. We also thank Marvin Glasser, PhD, for statistical advice, Dan Byrne, MA, for establishing and maintaining the data processing system, and Nancy Corbett for technical support.
REFERENCES 1. Baker SP, O’Neill B, Haddon W, et al: The injury severity score: A method for describing patients with multiple injuries and evaluating emergency care. J Trauma 1974;14:187-198 2. Boyd CR, Tolson MA, Copes WS: Evaluating Trauma Care: The TRISS method. J Trauma 1987;27:370-378 3. Champion HR, Sacco WG: Measurement of injury severity and its practical application. Trauma Q 1984;1:25-38 4. Champion HR, Sacco WJ, Copes WS, et al. A revision of the trauma score. J Trauma 1989;29:623-629 5. Champion HR, Sacco WJ, Hunt TK: Trauma severity scoring to predict mortality. World J Surg 1983;7:4-11 6. Sacco WJ, Jameson JW, Copes WS, et al: Progress toward a new injury severity characterization: Severity profiles. Comput Biol Med 1986;18:419-429 7. MacKenzie EJ: Injury severity scales: Overview and directions for future research. Am J Emerg Med 1984;2:537-549 8. Oreskovich MR, Howard JD, Copass MK, et al: Geriatric Trauma: Injury patterns and outcome. J Trauma 1984;24:565572 9. MacKenzie EJ, Shapiro S, Eastham JN: The abbreviated injury scale and injury severity score: Levels of inter- and intrarater reliabilitv. Med Care 1985:23:823-836 10. Lawnick iM: Personal communication, October, 1989 re: MTOS coding guidelines for 1988. Washington Hospital Center, Washington, DC 11. Morris JA, Auerbach PS, Marshall GA, et al: The trauma score as a triage tool in the prehospital setting. JAMA 1986;256: 1319-1325 12. Moreau M, Gainer PS, Champion H, et al: Application of the trauma score in the prehospital setting. Ann Emerg Med 1985;14:1049-1054 13. Semmlow JG, Cone R: Utility of the injury severity scale: A confirmation. Health Serv Res 1976;11:45-52 14. Copes WS, Champion HR, Sacco WJ, et al: The injury severity score revisited. J Trauma 1988;28:69-77 15. Long WB, Bachulis BL, Hynes GD: Accuracy and relationships of mechanisms of injury, trauma score, and injury severity score in identifying major trauma. Am J Surg 1987;15:481-484 16. Finelli FC, Jonsson J, Champion HR, et al: A case control study for major trauma in geriatric patients. J Trauma 1989;29:541-548 17. Bull JP: The injury severity score of road traffic casualties in relation to mortality, time of death, hospital treatment time and disability. Accid Anal Prev 1975;7:249-255 18. MacKenzie EJ, Morris JA, Edelstein SL: Effect of preexisting disease on length of hospital stay in trauma patients. J Trauma 1989;29:757-765
Because those with injuries make up a large proportion of patients treated in emergency departments (EDs), injuries should be a major focus of research in emergency medicine. Much of the research done to date has tested hypotheses regarding relationships between injured patients’ outcomes and variations in aspects of patient care or trauma systems. As these studies have attempted to define relationships among variables more precisely, the importance of statistically controlling for the severity of patients’ injuries has become more evident. A number of indexes are available for use as severity controls. Three have been widely used. The injury severity score (ISS) was designed by Baker et al’ to express the overall effect of injuries to several body systems. The ISS
can be used in retrospective analyses and can be combined with age to control for severity. The revised trauma score (RTS) developed by Champion et al’” requires data regarding the patient’s physiologic status soon after the injuryeither in the field or in the ED. The TRISS methodology, also developed by Champion et a1,2-6 estimates each patients’ probability of survival. This method combines the RTS, ISS, and age using coefficients developed from the Major Trauma Outcome Study (MTOS). In general, researchers decide which index to use as a control based largely on reports in the literature regarding the relative abilities of those available to explain variation in the dependent variable of interest-the larger the proportion of variation explained, the more useful the index as a control. The literature is inconclusive in this regard in the case of controhing for injury severity. Previous studies of the relationships between the ISS and survival, length of hospital stay, and level of disability at discharge have yielded variable results.’ For example, while Baker et al’ reported a correlation of 0.49 between ISS and mortality among motor vehicle accident victims, Oreskovich et al* found no significant relationship between ISS and mortality among older patients suffering falls. Neither the RTS nor the TRISS method have been extensively tested with outcome variables other than survival. No study has used a single data set to prospectively compare the abilities of these indexes to explain variation in a variety of patient outcome variables. In this study the following question was addressed: When should the RTS, ISS, the ISS and age, or the probability of survival calculated using the TRISS method be used to control for severity of injuries in trauma research? To answer this question, we compared the predictive abilities of each index with respect to three dependent variables: patients’ survival to hospital discharge, hospital length of stay among survivors, and intensive care length of stay among survivors spending at least 1 day in an intensive care unit (ICU).
METHODS Settings andSelection ofCases
From the ‘Institute for Trauma and Emergency Care, New York Medical College Valhalla, NY, TDepartment of Surgery, Our Lady of Mercy Medical Center, Bronx, NY, *Department of Surgery, Lincoln Hospital, Bronx, NY. Supported by the Centers for Disease Control, Public Health Services, Department of Health and Human Services, Atlanta, Georgia, grant number R49CCR202487 (“EMS System Factors: Motor Vehicle versus Other Injuries”). Presented at the UAEM Meetings, May 1989, San Diego, CA. Address reprint requests to Dr Murphy: Institute for Trauma and Emergency Care, New York Medical College, Valhalla, NY. Key Words: Injury Severity Score, Revised Trauma Score, Severity Index, Trauma Research, TRISS method. 0 1990 by W.B. Saunders Company. 07358757/90/0806-0003$5.00/O 484
The study was conducted by the Institute for Trauma and Emergency Care (ITEC) at the New York Medical College (NYMC). Eight hospitals, all affiliates of NYMC, participated in the research. A summary of characteristics of the hospitals is in Table 1. Hospitals ranged in size from 200 to 813 beds. Three of the eight hospitals are offtcially designated Level I trauma centers. Average numbers of trauma admissions per month ranged from 32 to 248, while average numbers of adult major trauma admissions eligible for the study ranged from 12 to 75. Logs of all ED admissions in each hospital were reviewed by one of three trained nurse-abstractor between July 1, 1987
MURPHY, CAYTEN, STAHL l EMERGENCY MEDICINE RESEARCH
TABLE1. Summary of Characteristics of Participating Hospitals
Hospital 1 2 3 4 5 6
No. of Beds
Official Trauma Center
Average No. Trauma AdmlMo
Average No. Adult Major Trauma/MO.
630 613 636 305 200 458 336 222
Yes Yes Yes No No No No No
248 127 ::
75 57 28 14 13 27 34 12
32 75 64 36
and June 30, 1988. Patients 13 years of age or older with injuries were included if they either died or stayed in the hospital for 48 hours or more; patients with major burns or only distal fractures were excluded.
Data Collected A study record was opened for each patient upon his/her identification as eligible. Information from prehospital and ED records was abstracted at that time and was entered on standardized forms. The complete hospital record was reviewed following the patient’s discharge, and final information regarding the patient’s stay and condition at discharge were recorded. A wide range of data elements were collected regarding the care of each patient. For the purpose of this article, important elements included demographic data, information regarding the cause of the injury, and data required to calculate the RTS, ISS and the probability of survival using the TRISS method.
ISS, RTS, and TRISS The Abbreviated Injury Scale (AIS) assesses injury severity on a scale of 1 (minor) to 6 (fatal) in each of several body regions. (AISwas used in this study.) Scores are subjective assessments of severity made by expert raters. Ratings are completed on the basis of reviewing emergency department and hospital records. Baker et al’ designed the ISS to be an anatomic indicator of the overall effect of injuries to several areas of the body. Each of six regions is scored with the highest AIS value given to any injury in that area. AIS values for the three highest scoring body regions then are squared and summed to form the ISS. Baker et al’ found that the ISS correlated more closely with mortality than did the maximum AIS value. The RTS (a refinement of the Trauma Score) was developed by Champion, Sacco, et al.‘& The Score includes three physiological parameters (Glasgow Coma Scale, systolic blood pressure, and respiratory rate) recorded soon after the patient’s injury, generally in a prehospital or ED setting. The coefficients used in the formula were developed using regression analyses based on 2,000 Washington Hospital Center patients and tested using the more than 25,000 patients in the MTOS. (The American College of Surgeons Committee on Trauma initiated the MTOS in 1982. The study includes data submitted voluntarily by trauma centers across the United States and Canada; criteria used by hospitals to enter patients vary.)
485
The TRISS methodology also was developed by Champion, Sacco, et a1.2-6 It allows the probability of survival of any one patient to be estimated given a formula that includes the patient’s ISS, age, and RTS. Coefficients for each component were developed using regression analyses of the MTOS data. Taken across a set of patients, the method allows differences in actual numbers of survivors (or nonsurvivors) to be compared with the numbers predicted based on the MTOS data set.
Statistical Analyses The statistical significance of differences between characteristics of those for whom complete data were and were not available were evaluated using x2 and f-tests. AIS codes were assigned on the basis of the subjective assessments of trained nurse raters. American Association of Automotive Medicine (AAAM) (1985) coding guidelines were followed. These generally parallel the MTOS guidelines for 1988.” However, three differences between MTOS coding and ours deserve description. Under MTOS, initial loss of consciousness is not scored, named vessel lacerations (not further specified) are coded as “major,” and hemomediastinum is not coded but is used to upgrade. Using AAAM guidelines, we scored initial loss of consciousness, gave a lower code to named vessel lacerations (not further specified), and coded hemomediastinum. To our knowledge, there have been no studies of the effect of such differences in guidelines on AIS scores; however, it seems unlikely that the effect is great. The reliability of AIS coding among raters is critical to the generalizability of results. Following the method developed by MacKenzie et ak9 K and weighted K were used to evaluate levels of agreement with regard to AIS coding, and the intraclass correlation coefficient was used to assess the reliability of ISS coding. It was not feasible to test the reliability of RTS coding since those data were entered by ED personnel rather than by our nurse raters. Previous studies of the reliability of Trauma Score coding by Morris et al” and by Moreau et al’* have demonstrated high levels of interrater reliability in field situations. Champion et al4 believe that the simplification represented by the RTS (particularly by including respiration rate rather than “respiratory expansion”) should improve reliability even further. Demographic data were stratified by cause of injury: penetrating, including gunshot and stab wounds; motor vehicle accidents (MVAs), including occupants and pedestrians; low falls (on level surface or from low steps); and, other blunt, including falls from heights and assaults with blunt objects. Groups were compared using x2 analyses or analyses of variance. Logistic regression was used to evaluate relationships between the dichotomous variable “survived to discharge” and the predictor variables (probability of survival, RTS, ISS, ISS and age). False-positive and false-negative rates were calculated based on resubstitution using the results of the logistic regressions. Patients with survival functions greater than or equal to 0.5000 were predicted to live; patients with survival functions less than 0.5000 were predicted to die. False positives were defined as survivors predicted to die; false negatives were deaths predicted to survive.6
AMERICAN
486
JOURNAL
OF EMERGENCY
Pearson product moment correlation coefficients were developed to evaluate bivariate relationships among TRISSderived probabilities of survival, RTS, and ISS and hospital outcomes-lengths of hospital and ICU stays. Squared values of correlation coefficients are provided in tables to estimate the strengths of the linear relationships between variables. Ninety-five percent confidence intervals also are given. Stepwise regression was used to develop multivariate models predictive of lengths of hospital and ICU stays. Potential predictor variables were RTS, ISS, and age. Partial R’s (showing contributions to the overall model by individual variables) and model R’s are given. F-tests for the significance of individual variables and for the overall significance of the model were calculated. Throughout the analyses, age was used both as a continuous variable and as a dichotomous variable (~55 years v ~55 years of age as in the TRISS). Analyses were stratified by cause of injury and by whether or not care was rendered in a Level I trauma center hospital. Length of hospital stay analyses included only patients who stayed in a hospital for 190 days or less and who were discharged alive; length of ICU stay analyses included those in the preceeding group who spent at least one day in an ICU. Even with these restrictions, length of hospital stay and days in an ICU were skewed toward higher values; log transformations of both were used in correlation and regression analyses. RESULTS During the 12-month study period, 8224 patients with injuries were admitted to the eight hospitals. Of these, 2986 (36.3%) were eligible for inclusion in the study. Some element or elements of data required to calculate the probability of survival using the TRISS method were missing for 72 (2.4%) of the eligible cases. Those for whom probabilities of survival were versus were not calculable were statistically
TABLE 2.
Summary of Patients’ Characteristics
MEDICINE
n Volume 8, Number 6 n November 1990
similar with regard to sex, cause of injury, survival rate, Glasgow Coma Scale score in the ED, complications, and number of ICU days. However, those for whom probabilities could not be calculated were significantly younger, and had significantly higher ISS scores (17.6 v 11.4) and longer hospital stays than did those for whom probabilities of survival could be calculated. Interrater agreement between pairs of nurse raters with regard to AIS and ISS coding was tested on 63 medical records. For AIS coding, K ranged from 0.61 to 0.70; weighted K ranged from 0.68 to 0.79. The intraclass correlation coefficient, used to evaluate the reliability of ISS coding, was 0.77. These results are comparable to those reported by MacKenzie et al9 and suggest very good agreement among nurse raters. Table 2 summarizes characteristics of the 2914 patients included in the analyses by cause of injury. Twenty percent (595) of patients suffered penetrating injuries. The mean age among these patients was lower than were average ages in other groups. The proportions of males, and black and hispanic patients were greater. Overall, those with penetrating injuries had the shortest average hospital and ICU stays. Ninety percent of these patients survived. Twenty-seven percent (785 patients) were involved in MVAs. On average, these patients were older than those with penetrating injuries but younger than those with low falls. Over half the patients were white and two-thirds were men. Nearly 93% of patients in MVAs survived. Low falls were the most frequent causes of injuries, accounting for 994 patients or 34% of the total. These patients had the highest average age (74.8 years). Over 80% were white and two-thirds were women. On average, these patients had low ISS scores and high Glasgow Coma Scale scores; however, their average hospital and ICU stays were protracted. Over 93% of patients with low falls survived. Nineteen percent of patients (540) suffered other blunt injuries. Seventy-five percent were men and nearly 40% were
by Injury Cause
Variable*
Penetrating
MVA
Low Falls
Other Blunt
No. of patients Sex (No. (%))t Male Female Race (No. (%))t White Black Hispanic Other Age’tS RTStS -t* Glasgow Coma Scoret* Hospital Days (n = 2,704)t$ ICU Days (n = 613)$§ Survival: No. and rate (%)t
595
785
994
540
530 (89.1) 65 (10.9)
532 (67.8) 253 (32.2)
326 (32.8) 668 (67.2)
427 (79.1) 113 (20.9)
421 153 182 29 39.1 7.3 13.8 13.6 15.1 6.8 729
834 78 77 5 74.8 7.8 8.8 14.6 19.7 7.2 933
213 163 153 11 39.6 7.5 11.1 14.0 13.7 7.2 509
59 264 258 14 29.9 7.0 14.0 13.6 10.2 4.5 534
(9.9) (44.4) (43.4) (2.4) + 10.7 2 2.1 t 10.8 + 3.6 2 17.1 2 6.3 (89.7)
(53.6) (19.5) (23.2) (3.7) + 18.9 ” 1.5 5 11.6 r 3.3 2 20.4 f 10.3 (92.6)
(83.9) (7.8) (7.7) (0.5) ? 16.8 * 0.5 2 4.2 + 1.6 2 16.4 f 8.7 (93.9)
(39.4) (30.2) (28.3) (2.0) 2 18.6 * 1.1 2 9.6 ? 2.8 t 22.1 2 8.6 (94.3)
* Hospital days and complications based on patients who were in the hospital for 190 days or less; ICU days are based on the number of patients spending time in an ICU. X2 or analysis of variance tests were used to evaluate the significance of differences, t P < 0.01 $ Mean + SD. 8 P < .05
MURPHY,
CAYTEN,
STAHL
W EMERGENCY
MEDICINE
RESEARCH
white. These patients were similar to those patients in MVAs with regard to average age, anatomic and physiological indicators, and lengths of hospital stays. Nearly 95% survived. Table 3 summarizes distributions of cases by levels of age, ISS, RTS, and probability of survival. Over 85% of patients with penetrating injuries were under age 40 while nearly 75% of patients suffering low falls were 70 years or older. Those in MVAs and having other blunt injuries were more evenly distributed among the age categories. Nearly 40% of those with penetrating injuries had ISS values of 16 or above while over 90% of patients suffering from low falls had ISS values of less than 13. Again, those in MVAs and having other blunt injuries were more evenly distributed among categories. Over 80% of those in each category had RTS values of greater than 7 and probabilities of survival greater than 0.95. Nearly 8% of those with penetrating injuries had probabilities of survival of less than 20%. The relations between predictors and survival to discharge are shown in Tables 4 and 5. The false positive rates of all predictors were very low-0.03 or below. The TRISSgenerated probabilities of survival had the lowest falsenegative rates among those with penetrating injuries, in MVAs, or having other blunt injuries: 0.23, 0.45, and 0.48, respectively. These rates were reasonably stable when trauma center status was considered (Table 5). The falsenegative rates of the RTS taken alone approached those achieved by probabilities of survival, while those for the ISS with or without age/age score were high. All of the predictor variables had false-negative rates of 0.95 or above for patients having low falls. Bivariate relations between predictors and hospital outTABLE 3.
Distributions
Variable
of Cases
by Age,
ISS Score, Penetrating
No. of patients
and TRISS
487
TABLE 4. False-Positive/False-Negative Survival Predictors
of
Variable
Penetrating
MVA
Falls
Other Blunt
No. of patients Prob. Survival (TRISS) RTS ISS ISS and Age ISS and Age Score
595
705
994
540
0.02/0.23 0.02/0.28 0.0310.75 0.02!0.77
0.01/0.45 0.01/0.40 0.0110.66 0.02/0.68
0.01/0.90 0.00/l .oo 0.00/l .oa 0.00/l .oo
0.0110.46 0.01/0.55 0.0210.71 0.01/0.61
0.03/0.75
O.OLTYO.64
0.00/l .oo
0.0110.64
Low
NOTES. False-positive/false-negative rates calculated based on resubstitution using logistic regression functions. Patients with survival functions greater than or equal to 0.5000 were predicted to live: patients with survival functions less than 0.5000 were predicted to die. Age Score was dichotomized: age greater than or equal to 55 years vs less than 55 years. False-positive rate = % of survivors predicted to die. False-negative rate = % of deaths predicted to survive.
comes are shown in Tables 6 and 7. The ISS showed the strongest linear correlation with length of hospital stay both overall and when trauma center status was taken into account; the RTS generally showed the weakest correlation with hospital lengths of stay. There was little overlap between the confidence intervals for ISS v RTS or probability of survival when the full data set was considered. More overlap occurred when sample sizes were reduced by considering trauma center status. It is important to note that no predictor was strongly associated with length of hospital stay among those having low falls. Relationships with lengths of ICU stay were less clear. The ISS had the strongest linear correlation with ICU stay among those in MVAs, and those having low falls or blunt injuries. However, there was some overlap in the confidence (Probability
of Survival)
MVA
Low Falls
785
994
595
Age 13-29 30-39 40-49 50-59 60-69 70+ ISS Score
Rates
Other
Blunt 540
349 164 54 18 6 4
(58.7) (27.6) (9.1) (3.0) (1 .O) (0.7)
331 144 102 73 60 75
(41.7) (17.8) (13.3) (9.3) (7.5) (10.3)
24 41 37 60 104 728
(2.4) (4.1) (3.7) (6.0) (10.5) (73.2)
201 150 58 50 30 51
(37.2) (27.8) (10.7) (9.3) (5.6) (9.4)
1-8 9-12 13-15 16-19 20+ RTS
162 155 42 88 148
(27.2) (26.0) (7.1) (14.8) (24.9)
257 201 87 76 164
(34.0) (26.5) (11.3) (9.5) (18.7)
240 655 34 36 29
(24.1) (65.9) (3.4) (3.6) (2.9)
225 164 29 50 72
(41.7) (30.4) (5.4) (9.3) (13.3)
O-l 2-3 4-5 6-7 27 Prob. Survival
40 17 14 28 496
(6.7) (2.9) (2.4) (4.7) (63.4)
21 18 43 42 661
(2.7) (2.3) (5.5) (5.4) (84.2)
2 2 15 19 956
(0.2) (0.2) (1.5) (1.9) (96.2)
7 6 26 27 474
(1.3) (1.1) (4.8) (5.0) (87.8)
45 15 4 24 507
(7.6) (2.5) (0.7) (4.0) (85.2)
29 18 21 48 669
(2.5) (3.0) (2.0) (7.5) (85.0)
3 2 7 35 947
(0.3) (0.2) (0.7) (3.5) (95.3)
9 7 9 21 494
(I .7) (1.3) (1.7) (3.9) (91.5)
co.20 0.20-0.49 0.50-0.79 0.80-0.95 >0.95
(TRISS)
AMERICAN
488
TABLE 5. False-Positive/False-Negative Predictors by Trauma Center Status
JOURNAL
Rates of Survival
Penetrating
MVA
Falls
Other Blunt
513
508
360
300
0.02/0.24 0.02/0.30 0.03/0.76 0.03/0.78 0.0410.76
0.02/0.47 0.01/0.47 0.02/0.63 0.02IO.65 0.03/0.61
0.0111 .oo 0.0010.96 0.01/l .oo 0.01/0.96 0.00/1 .oo
0.0110.46 0.01/0.58 0.01/0.54 0.0110.62 0.00/0.54
82
277
634
152
0.0110.23 0.0110.43 0.01/0.75 0.0110.77 +
0.01/0.43 0.0110.43 0.01/0.71 0.01/0.86 0.0110.71
0.00/0.95 0.00/0.95 0.00/l .oo 0.00/l .oo 0.00/l .oo
0.00/0.57 0.00/0.57 0.02/i .oo 0.01/1.00 0.0111 .oo
Low
Variable Trauma Centers No. of patients Prob. Survival (TRW) RTS ISS ISS & Age ES 8. Age Score Nontrauma Centers No. of patients Prob. Survival (TRISS) RTS ISS ISS 8. Age ISS & Age Score
OF EMERGENCY
NOTES. False-positive/false-negative rates calculated based on resubstitution using logistic regression functions. Patients with survival functions greater than or equal to 0.5000 were predicted to live: patients with survival functions less than 0.5000 were predicted to die. Age Score was dichotomized: age greater than or equal to 55 years vs less than 55 years. False positive rate = % of survivors predicted to die. False negative rate = % of deaths predicted to survive. + Parameters for Age Score regarded to be infinite.
intervals for probabilities of survival and IS.5 scores, while squared correlation coeffkients for these predictors were nearly identical among those having penetrating injuries. Small numbers of cases precluded an analysis of ICU stay stratified by trauma center status. Multivariate relationships between RTS, ISS, age/age score, and hospital outcomes are shown in Tables 8 and 9. For the total sample, proportions of variation in lengths of hospital stays explained by combinations of variables ranged from 0.32 for those with other blunt trauma to 0.15 for those having low falls. Explanatory power was reasonably stable across trauma center status. TABLE 6.
Bivariate
Relations
Variable Length of Hospital No. Prob. Survival (TRISS) RTS ISS Length of KU Stay No. Prob. Survival (TRISS) RTS ISS
Between
Predictors
and Hospital
MEDICINE
n Volume 8, Number 6 n November
1990
The ISS was the first variable to enter all models with length of hospital stay as the dependent variable. Sixty percent to 90% of the total variation explained by each model was attributable to the ISS alone. The RTS entered all models with the exception of those for low falls patients and for patients with other blunt trauma cared for in nontrauma centers. Age or age score entered most models as the third variable. Comparing models for the same data set using age v age score showed small differences in explanatory power (usually less than one percent) when one or the other agerelated variable was used. With respect to lengths of ICU stays, ISS again was the first variable to enter all models-and was the only variable that significantly explained variation in stays among patients having low falls. RTS added to the explanatory power of models for penetrating injury patients and for those in MVAs. DISCUSSION This research began with the question: When should the RTS, ISS, the ISS and age, or the probability of survival calculated using the TRISS method be used to control for severity of injuries in trauma research? The results indicate that the answer depends on what outcome variable is being addressed. Probability of survival calculated using the TRISS method was the best discriminator between survival/nonsurvival to hospital discharge. The false-negative rates for probability of survival were somewhat lower than those for RTS and approximately half those of the ISS, and ISS and age/age score. The range of false-negative rates (23% to 48%) for three of the four injury cause categories parallel the 30% to 40% rates reported as “typical” by Sacco et aL6 The rates were generally steady across injury types and trauma center designation categories-with one major exception. Neither probaOutcomes
Penetrating
WA
Low Falls
Other Blunt
534 0.08 (0.04/0.12) 0.09 (0.05/0.14) 0.22 (0.17/0.28)
728 0.15 (0.1 l/0.20) 0.13 (0.09/0.18) 0.24 (0.19/0.30)
933 0.03 (0.0110.05) co.01 (<0.01/0.01) 0.11 (0.0710.14)
509 0.10 (0.05/0.16) 0.04 (0.01/0.07) 0.28 (0.21/0.34)
132 0.17 (0.07/0.29) 0.13 (0.05/0.26) 0.16 (0.08/0.28)
275 0.22 (0.13/0.30) 0.21 (0.13/0.30) 0.31 (0.21/0.40)
Stay
99 0.06 (0.01/0.17) 0.02 (0.01/0.10) 0.20 (0.08/0.36)
107 0.16 (0.05/0.29) 0.06 (0.01/0.18 0.36 (0.20/0.49)
NOTES. Squared correlations (P) and (lower/upper 95% confidence intervals) between Prob. (probability of) Survival, RTS and ISS, and hospital outcomes. Length of hospital stay analyses included only patients discharged alive; ICU analyses included only patients spending time in an ICU and discharged alive. Log transformations of both outcome variables were used. In all cases, relationships between lengths of hospital and ICU stay, and RTS and Probability of Survival were negative; relationships between lengths of hospital and ICU stay, and ISS were positive. All coefficients were significant at the 0.01 level or beyond, except where ? is cO.01.
MURPHY, CAYTEN, STAHL I EMERGENCY MEDICINE RESEARCH
TABLE7.
Bivariate Relations Between Predictors and Hospital Outcomes by Trauma Center Status
Variable Trauma Centers Lengrh of Hospital Stay No. Prob. Survival (TRISS) RTS ISS Nontrauma Centers Length of Hospital Stay No. Prob. Survival (TRISS) RTS ISS
489
Penetrating
MVA
459 0.08 (0.03/0.12) 0.08 (0.04/0.13) 0.21 (0.13/0.30)
459 0.13 (0.08/0.19) 0.12 (0.07/0.18) 0.26 (0.19/0.32)
336 0.04 (O.Ol/O.OS) co.01 (<0.01/0.03) 0.11 (0.05/0.18)
75 0.16 (0.04/0.32) 0.20 (0.06/0.38) 0.32 (0.14/0.48)
269 0.20 (0.1 l/0.29) 0.13 (0.06/0.21) 0.20 (0.1 l/0.29)
597 0.04 (0.02/0.07) co.01 (~0.0110.01) 0.10 (0.06/0.15)
Low Falls
Other Blunt
364 0.09 (0.04/0.15) 0.05 (0.01/0.08) 0.27 (0.1910.35)
145 0.21 (0.08/0.30) 0.01 (<0.01/0.05) 0.28 (0.16/0.40)
NOTES. Squared correlations (P) and (lower/upper 95% confidence intervals) between Prob. (probability of) Survival, RTS and ISS, and hospital outcomes. Length of hospital stay analyses included only patients discharged alive; ICU analyses included only patients spending time in an ICU and discharged alive. Log transformations of both outcome variables were used. In all cases, relationships between lengths of hospital and ICU stay, and RTS and Probability of Survival were negative; relationships between lengths of hospital and ICU stay, and ISS were positive. All coefficients were significant at the 0.01 level or beyond, except where P is ~0.01.
bility of survival, RTS, ISS, nor ISS and age/age score adequately predicted survival among those having low falls. Variability in length of hospital stay among survivors was best explained by combinations of ISS, RTS, and either age or age score. Between 15% and 39% of variation in length of stay was explained by combinations of these variables. Explanatory power was weakest among those injured in low falls. The ISS was the first variable to enter all models; partial R*s for RTS and age/age score were less than half-and usually less than one fourth-those for ISS. ISS was most closely associated with lengths of intensive care stays among survivors spending at least 1 day in an ICU. RTS also entered the regression models for penetrating injuries and MVAs. Again, explanatory power was weakest among those having low falls. The fact that the probability of survival calculated with the TRISS method was the best predictor of actual survival with RTS a close second in predictive abilities is not surprising. Using the MTOS data set, Champion et a12-6developed the coefficients underlying the probability of survival and RTS calculations by conducting regression analyses with survival as the dependent variable. However, it is significant that the strong relations between probability of survival and RTS with actual survival remained stable across trauma center status and three of the four categories of injury cause. A variety of statistical techniques have been used to evaluate the relation between ISS and survival, making comparisons among results difficult. For example, Semmlow and ConeI studied victims of both vehicular and nonvehicular trauma in Illinois. Probit analyses of relationships between ISS values and percent mortality showed a strong monotonic relation between variables. Copes et alI4 grouped ISS scores into categories or “bins”; they also reported a strong positive relation between increasing ISS scores and mortality in all injury groups. Studying only those injured in MVAs,
Baker et al’ reported a correlation coefficient of 0.49 between ISS and survival. Using logistic regression to analyze survival, we found the ISS (with or without age/age score) to have low false-positive rates but unacceptably high falsenegative rates. This relationship may have been masked by statistical tests used in previous work. None of the potential predictor variables discriminated well between survivahnonsurvival among those suffering low falls. It is unclear what proportion of patients in the MTOS data set had low falls. However, some hospitals participating in the MTOS may not consider these patients to be trauma patients; differences in definitions of trauma in the MTOS study may explain the lack of association between probability of survival and the RTS and actual survival among patients with low falls in the present study. The relation between ISS and survival among older patients, and particularly those suffering falls, has been under debate. Oreskovich et al* studied patients over 70 years of age with multiple injuries. Patients with falls constituted the largest proportion of patients and had the highest death rate. No strong relation between ISS values and survival was detected. Long et al” studied over 2,000 trauma victims; all who died with ISS scores of 15 or less were 69 years of age or older and were victims of falls. Baker et al’ also noted that increases in mortality among older patients was most notable when injuries were least severe, ie, when ISS values were lowest. However, Finelli et all6 found the average ISS of elderly nonsurvivors of traumatic injuries to be two-to-three times that of survivors. Within our sample, the mean ISS for nonsurviving low falls patients (average age, 74.8 years), was approximately 40% higher than that of survivors (P < 0.01); but, we also found a 100% false-negative rate for ISS as a predictor of survival. To our knowledge, the relations between probability of survival and RTS, and hospital or ICU days have not been
AMERICAN JOURNAL OF EMERGENCY MEDICINE W Volume 8. Number 6 n November 1990
490
TABLE8. Relations Between Predictors and Hospital Outcomes
Injury Cause Length of Hospital Stay Penetrating (534) MVA (728) Low Falls (933) Other Blunt (509) Length
Variables Included in Most Predictive Model/Partial r* ISS/O.219 RTS10.034 IWO.244 RTS10.027 Age/O.016 Iss10.107 Age/O.047 ISS10.278 Age Score/O.030 RTS/O.OOS
Model r2 0.253 0.207
0.154 0.317
of ICU Stay
Penetrating (132) MVA (275) Low Falls (99) Other Blunt (107)
0.244
IWO.160 RTSl0.084 IWO.308 FITS/O.058 Iss10.200
0.200
IWO.359
0.359
0.366
NOTES.Variables are listed in the order in which they entered stepwise regression analyses. Length of hospital stay analyses included only patients discharged alive: ICU analyses included only patients spending time in an ICU and discharged alive. Log transformations of both outcome variables were used. Age Score was dichotomized as <55 years vs. 255 years of age. Age was a continuous variable. Two sets of regression equations were developed for each injury cause; one used age with RTS and ES, while the other used Age Score with RTS and ISS. In all cases, relationships between lengths of hospital and ICU stay, and RTS and Probability of Survival were negative; relationships between lengths of hospital and ICU stay, and ISS and age/age score were positive. All variables entered were significant at 0.01 or beyond: all models were significant at Pr < 0.0001 (F-tests).
No other research has attempted to stratify analyses of indexes by trauma center designation. Our results indicate that the indexes tested perform equally well across types of hospitals. The research has limits. Data required to calculate probabilities of survival were unavailable for less than 3% of patients. It seems unlikely that the results have been biased by the omission of this small number of patients, but we cannot estimate the effect precisely. In addition, although the sample was large (nearly 3,000 cases), the subsamples were relatively small. For that reason, the relations between lengths of ICU stays and the various predictor variables could not be analyzed with respect to trauma center status. Con& dence intervals of correlation coefficients would likely have been more distinct if a larger sample had been available. The results of the present study suggest two important topics for further investigation. First, a severity index that better explains variation in survival among those with low falls is needed. In our sample, the group having low falls was relatively homogeneous with regard to age, ISS. and RTS values. It seems likely that an index that can discriminate between survivors and nonsurvivors will include comorbidity indicators such as numbers of complications, and indicators of the existence and severity of preexisting conditions, which might show more variation.‘6,‘8 Relations between survival rates of patients having low falls and predictor variables will be difficult to explain until such an index is developed. Second, the results of this study are suggestive with reTABLE9. Hospital
evaluated. Given that the TRISS method and the RTS were developed as predictors of survival, it is not surprising that they did not perform well as independent predictors of variables related to hospitalization. In addition, our analysis of hospital and ICU days included only survivors while probability of survival and the RTS were developed using information concerning both survivors and nonsurvivors. Other researchers have reported significant relations between the ISS and hospital length of stay. Bull” found that ISS values separated survivors into groups with statistically different mean lengths of hospitalizations. Semmlow and ConeI reported an approximately linear increase in length of stay with increasing ISS values. Neither tested the explanatory abilities of the ISS in combination with RTS, age or age score; however, we found that these variables significantly increased the proportion of variation explained in length of hospital stays. The relation between numbers of ICU days among survivors spending at least 1 day in an ICU and the several independent variables had not been tested previously. ISS proved to be most closely related to ICU days with RTS adding to the explanatory power among those in MVAs and with penetrating injuries. Both age (a continuous variable) and age score (a dichotomous variable) were tested in the various models developed. No clear pattern emerged. We found little difference in explanatory power between models using age v age score suggesting that researchers should test both and use whichever provides an optimum model.
Relationships Between Predictors Stay: by Trauma Center Status’
Injury Cause
and Length
Variables Included in Most Predictive Model/Partial r2
of
Model r*
Trauma Centers
Penetrating (459) MVA (459) Low Falls (336) Other Blunt (364) Penetrating (75) MVA (269) Low Falls (597) Other Blunt (145)
ISSiO.207 RTW0.032 ISS10.263 RTWO.081 Age/O.013 Iss/o.115 Age/O.066 lSWO.274 Age Score/O.024 RTSI0.014 ISS10.316 RTSI0.062 ISWO.196 RTW0.048 Age/O.037 Iss10.100 Age/O.036 ISS10.282 Age/O.086
0.239 0.294
0.180 0.305
0.378 0.281
0.136 0.368
NOTES.Variables are listed in the order in which they entered stepwise regression analyses. Length of hospital stay analyses included only patients discharged alive; ICU analyses included only patients spending time in an ICU and discharged alive. Log transformations of both outcome variables were used. Age Score was dichotomized as ~55 years vs. 255 years of age. Age was a continuous variable. Two sets of regression equations were developed for each injury cause; one used age with RTS and ISS, while the other used Age Score with RTS and ISS. In all cases, relationships between lengths of hospital and ICU stay, and RTS and Probability of Survival were negative; relationships between lengths of hospital and ICU stay, and ISS and age/age score were positive. All variables entered were significant at 0.01 or beyond: all models were significant at Pr < 0.0001 (F-tests).
MURPHY, CAYTEN, STAHL n EMERGENCY MEDICINE RESEARCH
spect to the possible development of a TRISS-type predictor of lengths of hospital and ICU stays. With the increasing concerns in our society concerning hospital resource allocation it is important to understand how well existing measures predict hospital outcomes. We found that the variables included in the TRISS method (ISS, RTS, and age score) explained reasonably large proportions of variation in these hospital outcome variables. However, our data set was not sufficiently large to develop and to test a predictive model. A larger study would contribute significantly by providing a model that predicted either categories of lengths of stay or diagnostic related groups (DRG) outliers. Such research also might test whether different coefficients for components of the RTS would enhance the measure’s prediction of hospital outcomes. CONCLUSION Emergency medicine researchers concerned with injuries must control for the severity of those injuries. Relations among patient outcome variables, and variations in patient care or trauma system factors will be obscured or distorted unless severity is statistically controlled. No research existed to guide researchers in the selection of a control measure. We compared the amounts of variation explained by the TRISS-derived probability of survival, the RTS, the ISS and the ISS with age in regard to three dependent variables: survival, length of hospitalization among survivors, and length of ICU stay among those spending at least 1 day in an ICU. The results indicate that the selection of a control for injury severity depends on the outcome variable of interest. The ISS (which can be calculated retrospectively from hospital records) and age can be used to control for severity if hospital length of stay among survivors or days spent in an ICU are of concern. However, the explanatory power of these variables will be enhanced in most cases by the addition of the RTS to the model. Moreover, if survival is to be studied among patients with injuries other than low falls, probabilities of survival-which include the RTS-should be developed. Calculation of the RTS requires the careful collection of data regarding patients’ blood pressures, Glasgow Coma Scale scores, and respiratory rates in the ED. If these physiologic parameters are not recorded in the ED, the opportunity to control for severity in investigations into factors affecting patient survival will be lost.
491
The authors wish to thank Bea Clanton, RN, Karen Ende, RN, Gail Murray, RN, Sara Nealon, RN, and Deborah Vierra, RN, for collecting the data reported. We also thank Marvin Glasser, PhD, for statistical advice, Dan Byrne, MA, for establishing and maintaining the data processing system, and Nancy Corbett for technical support.
REFERENCES 1. Baker SP, O’Neill B, Haddon W, et al: The injury severity score: A method for describing patients with multiple injuries and evaluating emergency care. J Trauma 1974;14:187-198 2. Boyd CR, Tolson MA, Copes WS: Evaluating Trauma Care: The TRISS method. J Trauma 1987;27:370-378 3. Champion HR, Sacco WG: Measurement of injury severity and its practical application. Trauma Q 1984;1:25-38 4. Champion HR, Sacco WJ, Copes WS, et al. A revision of the trauma score. J Trauma 1989;29:623-629 5. Champion HR, Sacco WJ, Hunt TK: Trauma severity scoring to predict mortality. World J Surg 1983;7:4-11 6. Sacco WJ, Jameson JW, Copes WS, et al: Progress toward a new injury severity characterization: Severity profiles. Comput Biol Med 1986;18:419-429 7. MacKenzie EJ: Injury severity scales: Overview and directions for future research. Am J Emerg Med 1984;2:537-549 8. Oreskovich MR, Howard JD, Copass MK, et al: Geriatric Trauma: Injury patterns and outcome. J Trauma 1984;24:565572 9. MacKenzie EJ, Shapiro S, Eastham JN: The abbreviated injury scale and injury severity score: Levels of inter- and intrarater reliabilitv. Med Care 1985:23:823-836 10. Lawnick iM: Personal communication, October, 1989 re: MTOS coding guidelines for 1988. Washington Hospital Center, Washington, DC 11. Morris JA, Auerbach PS, Marshall GA, et al: The trauma score as a triage tool in the prehospital setting. JAMA 1986;256: 1319-1325 12. Moreau M, Gainer PS, Champion H, et al: Application of the trauma score in the prehospital setting. Ann Emerg Med 1985;14:1049-1054 13. Semmlow JG, Cone R: Utility of the injury severity scale: A confirmation. Health Serv Res 1976;11:45-52 14. Copes WS, Champion HR, Sacco WJ, et al: The injury severity score revisited. J Trauma 1988;28:69-77 15. Long WB, Bachulis BL, Hynes GD: Accuracy and relationships of mechanisms of injury, trauma score, and injury severity score in identifying major trauma. Am J Surg 1987;15:481-484 16. Finelli FC, Jonsson J, Champion HR, et al: A case control study for major trauma in geriatric patients. J Trauma 1989;29:541-548 17. Bull JP: The injury severity score of road traffic casualties in relation to mortality, time of death, hospital treatment time and disability. Accid Anal Prev 1975;7:249-255 18. MacKenzie EJ, Morris JA, Edelstein SL: Effect of preexisting disease on length of hospital stay in trauma patients. J Trauma 1989;29:757-765