12 Scandinavian Journal of Surgery 91: 1222, 2002

H. R. Champion

TRAUMA SCORING
H. R. Champion
Surgery and Military & Emergency Medicine, Uniformed Services University of the Health Sciences, Bethesda, Maryland, U.S.A.

ABSTRACT

Severity scales to characterize the nature and extent of injury are important adjuncts to trauma care systems, trauma research and many of the elements of a complete public health approach to injury. This article provides a brief overview of severity scale development over the past 30 years during which the science to support such initiatives has matured substantially. Anatomical, physiological, intensive care, composite and complex scales and models now abound and are being applied to a variety of tasks with increasing precision. Trauma registries enable the meaningful aggregation of data for the development and testing of models. Future challenges are identified as are potentially fruitful avenues of research.
Key words: Trauma; trauma scores; injury scoring; Injury Severity Score; TRISS

INTRODUCTION Measurement and tabulation of the severity of injury is an essential prerequisite both to effective trauma care and clinical research. Furthermore, programmatic and public policy issues related to injury are not easily addressed without understandable renditions of the epidemiology and the nature and severity of the problem. Also, both civilian mass casualty contingency planning and military expeditionary medical resourcing benefit immensely from knowledge of the number, nature and severity of anticipated casualties. Several trauma severity scales are used to quantify injury severity and describe patient severity casemix. Once injury is thus characterized, the results can be applied to triage, resourcing, research, and quality of care assessment. In many cases, indices facilitate precise communication about often complex problems. Severity classifications can be nominal, ordinal or interval in nature. Many injury severity characteriCorrespondence: Howard. R. Champion, M.D. 954 Melvin Road Annapolis, Maryland 21403, U.S.A. Email: HRCHAMPION@aol.com

zations are nominal scales where verbal definitions are used to place injuries in various dissimilar bins of severity or complexity. They are valuable in facilitating communication between interested parties and allowing for consistent casemix definitions to be adopted by a variety of investigators. Ordinal scales (Ordinal: any positive whole number defining a things place in a series) assign a number to states of severity. Most cancer classifications are ordinal scales. In injury, many classifications of fractures (1) are examples, as is the Glasgow Coma Scale (2). Triage classifications and many orthopaedic, neurosurgical and general injury classifications fall into this group. Interval scales also assign numbers, but there is an implicit expectation of some consistency in the intervals between the numbers. The Abbreviated Injury Scale (AIS) describes anatomical injury, but the 5 in head has a different outcome from a 5 in the abdomen and the interval between a 2 and a 3 or a 3 and a 4 also varies from body area to body area. Few if any injury scoring systems are truly consistent interval scales. A score is the number, grade or rating. A scale is the system of measurement. Thus, both the Trauma Score and the Pediatric Trauma Score should have been named as scales as the Glasgow Coma Scale was so correctly identified.

Trauma scoring

13 TABLE 1 Some measures of discrimination.

A model is created when a scale or score is used to predict or correlate with an outcome measure, e.g., survival. Models refine the ability of scores or scales to be used for comparison of groups of patients or in determining thresholds which allow patients to be placed in various bins for the purpose of treatment, triage or comparative analysis. Although each scale in and as of itself may also be used as a predictive model, injury, in its complex involvement of many organs and anatomical areas with time dependent physiological sequelae after anatomical rearrangement requires more than a single, simple descriptor for its characterization. Models may be logistic regression based, Bayesian or Hybrid. Many neural nets have complex bases for models, e.g., back propagation, radial, etc. Composite models such as AIS derivatives (see below) plus the Trauma Score (3), e.g., (TRISS) (4) are often used to fashion a predictive model and are composed of the determinants deemed important in predicting the outcome measure being used. It is important to understand that such predictive models are not deterministic for individual outcome. They give the probability of an outcome for a given patient. Thus, a patient may have a probability of death Pd of .6 or 60 % risk of dying. For every 10 identical patients, 6 will die and 4 will live if it is a perfect model. But which group does this patient fall into: the 60 % dead or the 40 % alive? Only chance determines the answer. Probabilistic models require different statistical approaches to deterministic models. Probabilistic models should always be evaluated for their discrimination and their calibration. Discrimination refers to the ability of the model to distinguish between outcomes e.g. survivors from deaths. Measures of discrimination are shown in Table 1. Perfect discrimination would assign a 1 to all those that died and 0 to those that did not. Calibration measures how accurately the model predicts the probability of death of given interval sets. A perfectly calibrated model would give the same probability of death for any given group of cases as the observed proportion of deaths. Concordance is often used as a measure of discrimination. Each survivor/non-survivor pair is assigned a score of 1 where a survivor has probability of survival strictly greater than that of the non-survivor, 0,5 where the survival probabilities are equal and 0 where the non-survivor has a greater probability of survival than the survivor. The concordance is the sum of these values divided by the total number of survivor/non-survivor pairs. A perfectly discriminating model has a concordance of 1. Data sets can be compared using concordance by using test/design set methodologies or by bootstrapping techniques. Data sets can be compared for calibration using calibration curves and the Hosmer-Lemshow statistic (5). The H-L statistic is a single summary measure of the calibration of a model. It is based on comparing the observed and estimated probability outcome of patients grouped by estimated mortality or decile. The resulting statistic has a chi-square distri-

Sensitivity Specificity False Positive Rate False Negative Rate Positive Predictive Power Misclassification Rate Area under receiver operating characteristic curve Concordance

bution with degrees of freedom equal to 2 less than the number of groups. The smaller the H-L statistic the better the fit. Another thing to consider when evaluating scores and models is their validity: face, concurrent, construct, content and predictive. Since there are a wide variety of applications for injury severity scoring systems, it is important that the choice of the severity score scale, index, or model accurately match the application. Thus, for triage, which places a patient in one or three or four bins of severity, a relatively simple tool may be all that is required. For controlling for casemix and comparing outcomes over time or in different centers or in therapeutic arms of a study, a more scientifically robust tool might be in order. Misapplication of a simple tool to a complex task creates an opportunity for unwarranted extrapolations and poor science. ANATOMICAL INDICES
AIS, ISS, AP, NISS, MAP AND OTHERS

Indices that score anatomic injury severity have been widely applied in epidemiologic studies. When used alone, their correlation with outcome is lower than when combined with measures of physiologic states. The Abbreviated Injury Scale (AIS) (6) was developed by engineers to rate and compare blunt injuries from road vehicle accidents and has undergone several modifications since its introduction in 1971. We are currently updating AIS -95 (1998 version). It is called AIS 2000, but may not be out until 2003. The AIS scores individual injuries and classifies them into one of six categories, each with an associated severity score: minor [1], moderate [2], serious [3] severe [4] critical [5], and fatal [6]. The AIS has been shown to relate to risk of death in each body region. The severity scores were subjectively assigned by experts thus, there is some internal inconsistency, e.g., a 3 for one body region does not have the same risk of death as a 3 in another region. The Injury Severity Score (ISS) (7) was derived to summarize the severity of the condition of multiply injured patients. The ISS is the sum of squares of the highest AIS grades in each of the three most severely injured body regions. ISS = a2 + b2 + c2

14 TABLE 2 Measures for each index (Ref. 19). Discrimination Measures Concordance (Area Under ROC Curve) 86 86 87 87 87 88 82 84 85 85 Calibration H-L Statistica

H. R. Champion

ISS NISS mAP APS ICSS (NCTR) ICISS (PTOS) ICD/ISS ICD/NISS ICD/mAP ICD/APS
a

384 040 029 070 244 370 283 133 097 094

H-L statistic computed by comparing predicted versus observed number of survivors across strata defined by equal range fixed Ps cut-off values (e.g., 0.00.09; 0100.19, etc.) Injury Severity Score New Injury Severity Score Maximum Anatomical Profile Score Anatomical Profile Score

ISS NISS mAP APS

xx When prefixed by ICD means ICDMAP and ICD-10CM code were used to derive ISS ICD derived ISS NCTR. North Carolina Dataset ICISS & ICD based ISS PTOS dataset (Table 2 is published with permission from Lippincott Williams & Wilkins, Baltimore, MD, USA)

Limitations of the ISS have been described (8). These limitations result from the use of a one-dimensional score to represent the spectrum of injury locations and severities and from the ISS definition that excludes all but the most serious injury in any body region. The ISS also considers equal serious injuries of the same AIS severity to any body region. Thus, the ISS does not provide a reliable basis for characterizing injury severity, inflates the variability of outcome estimates, and obscures comparisons of certain patient populations. It is particularly frail for ballistic penetrating injury where there are many injuries in a single body area, e.g., gunshot to abdomen, and in populations with high proportions of head injuries. Nevertheless, the ISS was the first widely used scale to capture multiple injuries. It is still widely used. Because of ISS limitations, a multidimensional characterization was sought that considers the number, location, and severity of anatomic injuries and their influence on outcome. The Anatomic Profile characterization uses four variables to describe a patients injuries: A, B, and C summarizes serious (AIS = 3) injuries to the head and neck, thorax, and other body regions respectively, and D describes all nonserious injuries to any body region. The scores are combined using an Euclidean Distance Model viz. the square root of the sum of the squares

number of injuries increases. Using the A, B, C, D methodology, survival probabilities can be estimated by multiple logistic regression or from the percentage of survivors among clusters of patients with similar injury constellations. This anatomic injury score is called the Anatomic Profile (AP). The maximum AP(mAP) (9) double scores the maximum AIS injury score allowing for its dominant effect in multiple injuries. Another improvement has been the development of the new Injury Severity Score (NISS) which overcomes some of the shortcomings of the Injury Severity Score (ISS) allowing severe injuries in multiple body areas to be counted (10). Anatomic injury scales have also been based on versions of the International Classification of Diseases (ICD) used throughout the world as the prime nomenclature system for disease and injury. ICD 9CM (CM stands for Clinical Modification) (11) has many discrete definitions of, for instance fractures, and relatively poor definition for vascular and solid organ injuries, e.g., liver injuries to which are a major cause of early death after injury thus it lacks face validity. There have been several attempts to circumnavigate the lack of discrimination in the ICD nominal system and the labor intensive duplicative coding that developing AIS requires. Codes based on the ICD 9-CM have included the Estimated Survival Probability Index (12), the Anatomic Index and ICISS (13, 14). The PEBL Code provides greater specificity for penetrating injuries. It was derived from data on 3,600 Vietnam combat casualties and remains the most detailed anatomic code (15), but is not used. A computerized conversion (ICDMAP) (16) has been developed that assigns ICD 9-CM diagnoses to AIS-90 and then AIS-95 severity scores, although the conversion has some limitations primarily because of the range of severities associated with some ICD 9CM codes. It is widely used to convert administrative data sets coded in ICD-9CM into AIS and ISS. ICD 10 is now in use. Its clinical modification will be available soon. Severity models based on this will follow. ICD-10CM is much more detailed than ICD9CM and may improve the ability of ICD to be used as an element in models predicting outcome, and in a fashion that does not require the labor intensive AIS coding system. Recent years have seen competitive claims from the various authors of various anatomical injury severity scoring methodologies (17, 18). We, therefore, compared a number of different offerings ICD-based and non-ICD-based, AIS generated scores (19) (Table 2). These data show a general difference in the discrimination measure between ICD-based scores and AIS coded scores. The former also showed poor calibration. The new ISS (NISS) and Anatomic Profile (AP and MAP) scales are the best. Researchers continue to identify limitations with existing formulae, but replacements have yet to get traction (20, 21, 22).

A2 + B2 + C2 + D2
allowing for decreasing influence of injuries as the

Trauma scoring TABLE 3 Revised Trauma Score variables (Ref. 28). Glasgow Coma Score 1315 0912 068 045 03 Systolic BP (mmHg) 8> 89 7689 5075 0149 0 Resp. rate (min) 1029 8> 29 69 15 0 Coded Value 4 3 2 1 0

15

(Published with permission from Lippincott Williams & Wilkins, Baltimore, MD, USA)

PHYSIOLOGICAL SCALES
TRAUMA SCORE, REVISED TRAUMA SCORE, CRAMS SCALE

If anatomic scores characterize the static component of injury, then physiologic scores measure the acute dynamic component. This is important since most deaths from injury occur within hours (23), and are due to exsanguinations, brain or spinal cord injury or respiratory failure. In 1980, consensus physician peer review of severity indices resulted in the Trauma Score. The Trauma Score modified the Triage Index (24) to include systolic blood pressure and respiratory rate and used the Glasgow Coma Scale (GCS) to assess the degree of coma. The Trauma Score ranged from 1 (worst prognosis) to 16 (best prognosis) and is computed by summing points assigned to component variables. The Trauma Score was a useful predictor of outcome for patients with blunt or penetrating injuries (25). Field assessments determined that the Trauma Score user had high inter-rater reliability (26). It was tested in the military (27). The Revised Trauma Score (RTS) (28) (Table 3) was developed to mitigate limitations of the Trauma Score. Field use of the Trauma Score revealed that capillary refill and respiratory expansion were difficult to assess, especially at night, and that retractive respiratory expansion was always difficult to observe. Furthermore, dichotomous variables have limited utility. When retrospectively correlated with patient outcome, the Trauma Score was also found to underestimate the severity of the condition of some head-injured patients (i.e., the GCS was underweighted). In developing the RTS, measurement scales for systolic blood pressure and respiratory rate were divided into five intervals, chosen so that their associated survival probabilities approximated those of the widely used Glasgow Coma Scale intervals. A value from 0 to 4 was assigned each interval. For triage, coded values of Glasgow Coma Scale, systolic blood pressure, and respiratory rate are added to yield the triage version (labeled T-RTS) of the RTS. T-RTS takes on integer values from 0 to 12. The American College of Surgeons triage guidelines, which identify patients to be triaged to trauma centers whose sum of RTS coded variable values is less than or equal to 11, demonstrate a substantial gain in sensitivity with only a modest loss in specificity when compared with the previously recommended

Trauma-Score-based guidelines. The T-RTS also triages nearly all deaths (97.2 percent) to trauma centers. Another simplification of the Trauma Score was developed for field triage and added information on torsal injury. The CRAMS Scale (29), measures five components: circulation, respiration, abdominal injury, and motor and speech responses. In CRAMS, a value is assigned to each variable, depending on whether it is normal (2), mildly abnormal (1), or severely abnormal (0). Scores of 9 or 10 are defined as minor trauma, e.g., patient discharged home, and scores of 8 or less are defined as major trauma, e.g., patient died in emergency department or went to the operating room for general surgery or neurosurgery. In field tests, CRAMS accurately discriminated between major and minor trauma, with 92 percent sensitivity and 98 percent specificity (i.e., two bin discrimination). Comparisons of physiologic scores for triage have had varying results and have been strongly influenced by differing case severity mixes, inconsistent triage cut-off points, and varying definitions of patients needing trauma center treatment. An analysis of triage scales has shown that the Trauma Score and CRAMS have similar predictive values. Both, however, were found to have high false-negative triage rates. As a tool for prehospital triage, the effectiveness of the Trauma Score is enhanced when it is used with additional information, e.g., on mechanism of injury, and extent and location of identifiable anatomic injury. American College of Surgeons Committee on Trauma guidelines (Fig. 1) emphasize the importance of evaluating anatomic damage, even in the presence of normal physiology, and factors such as age (< 5 and > 55) and pre-existing disease that might lower the threshold at which patients should be triaged to a trauma center. A Pediatric Trauma Score (PTS) (30) has been developed that grades six components commonly seen in pediatric trauma. A 0 percent mortality has been found in patients with a PTS higher than 8, whereas mortality increases to 30 percent for patients with a PTS lower than or equal to 8. PTS triage guidelines recommend that a child with a PTS lower than or equal to 8 should be triaged to a level I trauma center, as defined by the American College of Surgeons. An examination of the utility of the PTS found significant correlations with survival, Injury Severity Score and other measures, although it demonstrated

16

H. R. Champion

Fig. 1. Field triage decision scheme (from Resources for optimal care of the injured 1999, with permission from American College of Surgeons).

Trauma scoring

17 TABLE 4 TRISS Calculation.

no advantages over the Revised Trauma Score in pediatric patients (31). As with anatomical scales, greater precision continues to be sought with physiologic scales. The reaction level scale (RLS 85) was developed by Starmark et al. (32) RLS 85 is an 8 point scale which removes the redundancy in the Glasgow Coma Scale between best verbal and best motor response. It compares favorably when tested against the Glasgow Coma Scale (33, 34). The PolyTrauma Score (PS) was developed in 1983 in Hanover. It was based on an analysis of 696 trauma patients treated between 1975 and 1986. The revised version used data from 389 patients treated between 1984 and 1990. The revised (PS) adds up to 90 points. It includes physiological data, the Glasgow Coma Scale, PaO2, FiO2, base excess and anatomical information on injuries to abdomen, extremities, thorax and pelvis. Each injury receives a specific score. Age is also factored into the PolyTrauma Score. Patients with a score of less than 20 are considered to have a mortality risk of less than 10 %. Patients with a score between 21 and 34 have a mortality risk of 20 %. Patients with a score of 35 to 48.7, 38 % and injuries beyond 48 % have a mortality risk of 65 %. The PS represented a system of practical reliable scoring for estimating mortality and morbidity of blunt trauma patients in the emergency room (35). COMBINATION INDICES/MODELS
TRAUMA SCORE-INJURY SEVERITY SCORE METHODOLOGY

A 25 year old male with a GSW to the chest received the following injuries: (coded in AIS-98). GSW to the chest Perforation RLL, LLL lung Perforation diaphragm Bilateral hemopneumothorax Through and through laceration of liver Lacerated duodenum GSW R thigh ISS = 16 + 9 + 1 = 26 AIS-98 ISS Body region

441450.4 4 440604.3 3 included in lung injury code 541824.3 541020.2 816002.1 3 2 1

Physiologic variables on admission were: SBP 80 mmHg RR 28/min GCS 15 RTS = 6.8174 Probability of Survival = 92.7 %

TRISS Development of the Trauma Score-Injury Severity Score (TRISS) in 1981 gave clinicians a way to identify major trauma patients with unexpected outcomes and to compare patient outcomes among institutions while controlling for differences in injury severity. TRISS, a severity index based on the patients Revised Trauma Score at admission to the emergency department, ISS, patient age and mechanism of injury is the basis of the Major Trauma Outcome Study methodology. TRISS has been adapted for assessing the impact of aeromedical emergency and prehospital care, for comparing outcomes in both adult and pediatric populations, and for documenting improvement of trauma patient outcomes over time or between countries and states. The key mathematical element of TRISS is the logistic function [1] Ps = 1/(1 + eb) where Ps is an estimate of a patients survival probability and [2] b = bo + b1(RTS) + b2(ISS) + b3(AGE) and RTS = the admission Revised Trauma Score, ISS = the Injury Severity Score and AGE = 0 for age less than 55 years, AGE = 1 for age greater than or equal to 55.

The bs are regression coefficients. For a patient group, e.g., adults over 15 years old with blunt injuries, equation [1] is used to assign a survival probability to each combination of RTS, ISS, and AGE. For outcome evaluation, the RTS is computed from coded values (04) of the Glasgow Coma Scale (GCS), systolic blood pressure (SBP), and respiratory rate (RR) (see Table 3). Regression weights were derived for the coded RTS variables using data from the Major Trauma Outcome Study patients. Those weights are: GCS: 0.9368 / SBP: 0.7326 / RR: 0.2908. The RTS is the sum of the product of the weights above and the coded values of the associated variables. The RTS takes on values between 0 and (approximately) 8. Evaluations have shown that the RTS is a more reliable predictor of survival and death than the Trauma Score. The weight given to the Glasgow Coma Scale attests to the importance of coma to outcome. The TRISS coefficients for equation [2] are (again based on MTOS data)
Blunt Penetrating Constant (b0) RTS (b1) 0.4499 0.8085 2.5355 0.9934 ISS (b2) 0.0835 0.0651 Age (b3) 1.7430 1.1360

Examples of TRISS calculations are shown in Tables 4 and 5. The Major Trauma Outcome study utilized equation [1] in two ways to evaluate patient outcomes. The first method, PRE, supports quality assurance activities. Fig. 2 shows a PRE chart, a scattergram for a set of patients (all in the same age group) characterized by RTSs (ordinate) and ISSs (abscissa). Survivors and nonsurvivors are indicated by Ls and Ds, respectively. Patients whose ISS-RTS coordinates are

18

H. R. Champion

Fig. 2. Sample PRE chart.

TABLE 5 TRISS Calculation. 65 year old female in MVA received the following injuries: AIS-98 ISS Multiple cerebral contusions LOC 1 hour Maxilla fracture Fractured R ribs #4, 5, 6 R hemothorax Bilateral femur fractures ISS = 9 + 9 + 9 = 27 140611.3 3 included in cerebral contusion code 250800.2 2 450222.3 3 included in rib fracture code 851800.3 3

Physiologic variables on admission were: SBP 60 mmHg RR 20 min GCS 13 RTS = 6.376 Probability of Survival = 67.2 %

on the diagonal line have an estimated survival probability of 0.50. This P50 line is determined by setting b = 0 in equation [2]. Ls (survivors) above the line and Ds (no survivors) below it represent patients with unexpected outcomes worthy of review. Not all such patients are therapeutic successes or failures. Because of limitations in its component indices (particularly ISS), such estimates of survival probability must be reviewed by clinicians. A systematic review of the unexpected outcomes leads not only to suggestions for improvement in trauma care but also to potential improvements in severity scores.

The second method, DEF, is a statistical assessment of patient outcome. Imagine lines parallel to the PRE P50 line covering every decile with L&D patients lying between all the lines. DEF compares a study population to a baseline population by counting the L/D ratio between the lines. Unlike PRE, DEF accounts explicitly for differences in patient injury severity mix and for the statistical significance of the comparison. In DEF, the statistic z compares the number of survivors in a sample (hospital) to that expected from a baseline population or norm. The statistic z is the ratio (A E)/S, where A and E are the actual and expected numbers of survivors for the study sample, and S is a scale factor that accounts for statistical variation. E is computed as Pi, where Pi is the survival probability for in the ith patient in the study population. Pi is computed from the logistic TRISS model (equation [1]) based on the data from the baseline population. The scale factor S is given by Pi (I Pj). Depending on whether the survival rate in the study sample is greater or less than that predicted by TRISS from the baseline data, z may be positive or negative. Absolute values of z greater than 1.96 are statistically significant (p < 0.05). Nonsignificant z scores can result from either small sample sizes for which the z statistic has limited statistical power or from outcomes that are indistinguishable from the norm. For large samples, high (in absolute value) and statistically significant z values can result from small and clinically insignificant departures from expectations. Thus, institutional performance cannot be inferred from the magnitude of statistically significant z scores. W quantifies the clin-

Trauma scoring

19

ical significance of statistically meaningful differences between the actual (A) and expected (E) number of survivors: W = 100 (A E)/N where A and E are defined as in z, and N is the number of patients in the institutions sample. For institutions with significant positive (negative) z scores, W is the average increase (decrease) in the number of survivors per 100 patients treated compared with norm expectations. Statistics z and W provided a more balanced description of patient outcomes in the context of an institutions contemporaries. TRISS has been the pre-eminent trauma outcome prediction model for the past 20 years. It is used throughout the world to compare patient outcomes in a consistent fashion. Its greatest frailty is related to the Injury Severity Score (ISS). For that reason, ISS was replaced in the TRISS formulation by AP to create ASCOT. ASCOT not only replaces the ISS with Anatomical Profile (AP), but changes age from being a dichotomous to a continuous variable (36). When comparing ASCOT and TRISS, the ASCOT performs much better on outcome prediction than TRISS. However its complexity has deterred many from implementing it and TRISS still remains the mainstay of comparative analysis of trauma patients. A study reporting the replacement of ISS with NISS in TRISS would be a worthwhile contribution. OUTCOMES
DISABILITY OUTCOMES

well as for its measurement properties, such as standardization, discriminatory power, reliability, and validity. The FIM was found to have face validity and to be reliable. Three of its assessments (feeding, locomotion, and expression) are included in the Major Trauma Outcome Study database. Other non-mortality scales include the widely used SF 36 scale. This has been modified by adding measures of cognitive function for its application to injury patient populations. For the past decade a number of attempts to develop a disability scale have been lead by Mackenzie et al. Efforts to develop the Functional Capacity Index (FCI) are ongoing. In this case, severity of a limitation of function is defined in terms of its impact and overall functioning in everyday living. Thus, the FCI looks at eating, excretory and sexual function, walking, hand and arm movement, bending, lifting, vision, hearing, speech and cognitive function. Such a validated scale would be of great value in crafting comprehensive health care approaches to injury.

TRIAGE AND TRAUMA SCORING The triage of trauma patients is necessary when a small proportion of hospitals receive victims of major injury in a regional trauma system. The triage decision involves determination of the threshold of injury severity that merits prompt access to the special resources of a trauma center care. The other application of triage is the priority determination of mass casualties and effective matching of their needs with the limited available resources. A significant proportion of patients with severe injuries and at risk of dying can be identified at the scene of the accident by their abnormal physiology. There have been many attempts to aggregate the various physiologic parameters into a single entity. The Trauma Score has been widely applied and field-tested for this purpose. Simpler versions, such as the CRAMS score, have also been advocated. The Revised Trauma Score physiologic threshold variables identifies 10 percent mortality on arrival at trauma centers. A far more difficult problem in triage is the identification of patients with significant anatomic injury whose physiologic status is close to normal at the scene shortly after an accident. There have been many attempts to refine tools that would provide accurate triage. The tool developed by the authors for the American College of Surgeons Guidelines has stood the test of time for nearly 20 years (Fig. 1).

Only a small proportion of injured patients are at risk of dying certainly less than 10 % of those hospitalized. Depending on the nature and type of injuries, however, disabilities and impairments can persist for many years. Thus, scales that relate to outcomes other than mortality are important in evaluating the public health impact of injury. For obvious reasons, most such scales focus on central nervous system and musculoskeletal injuries. The Glasgow Outcome Score (GOS) (37) has been widely used since its development in 1975, despite its recognized inadequacy to quantitatively reflect subtle changes during recovery from severe head injury. The Disability Rating Scale has been advocated as a single instrument to chart the progress of severely head injured patients from coma to community. The Admission Disability Rating Scale has been related to clinical outcome 1 year after injury and to electrophysiologic measures of brain dysfunction. Non-trauma related factors (e.g., educational level, type of employment, income, and family support) have also been shown to affect the functional recovery of general trauma patients. In 1983, the Functional Independent Measure (FIM) was developed to assess the assistance required for a disabled person to effectively perform self-care, sphincter management, mobility, locomotion, communication, and social and cognitive skills. The Functional Independent Measure was pilot-tested by clinicians in 50 United States facilities in 1985. The scale was assessed for ease of use and utility as

SEVERITY SCALES FOR SEVERELY ILL PATIENTS About the same time that injury severity indices were being developed for hospitalized injured patients similar efforts were underway to provide outcome based characterization of seriously ill patients in the intensive care setting. The challenges in this patient

20

H. R. Champion

population are somewhat different to but in no way less than those of providing adequate models to characterize hospitalized injured patients in general. The intensive care environment provides care for a variety of medical and surgical patients. Some of the surgical patients are trauma patients. The dynamics of care of surgical and medical patients are somewhat different in that surgical intervention often provides a watershed event for improving the patients trajectory towards wellness (or death). In 1981, the Acute, Physiological And Chronic Health Evaluation (APACHE) scoring system was introduced by Knaus et al (38). It consisted of 34 physiological variables each awarded a value between 0 and 4 depending on the severity of the disturbance. The sum of the worse value at each variable during the initial 24 hours after ICU admission yielded the acute physiologic score (APS). The system was complex to use and was not found to track surgical or trauma patients very well. APACHE II was derived by summing a 12-variable acute physiologic score (APS) and points awarded for age and chronic health problems. APACHE II has been widely used to control for casemix in intensive care units. Its value in surgical and trauma patients is still questioned. APACHE III is a proprietary prognostic system developed from APACHE II. The number of physiological variables was extended from 12 to 17. The variable ranges were adjusted and weighted. The APACHE III was proposed as a system to serve as a frame of reference to assess quality of ICU care. Some institutions use it to track outcome in this fashion. Other scoring systems in use for ICU patients include the Sepsis Score described in 1983 by Elebute and Stoner (39),the Sepsis Severity Score described by Stevens (40), the Multiple Organ Failure score introduced by Goris in 1985 (41) and the Multiple Organ Dysfunction score developed by Marshall et al in 1995 (42). Most of these systems score different organ system functions and aggregate them on some mathematical basis to prognosticate for individual patients or predict outcome for overall sets of patients. In addition to these multiple organ scales, there are specific scales used to score pulmonary function. Few have been validated and have limited value in predicting outcome. USE OF SEVERITY SCORING IN TRAUMA REGISTRIES

Numerous institutional, local, and statewide registries have been developed to monitor hospital trauma care, to help make decisions about allocation of hospital resources, and to provide information to county and local accrediting agencies. Existing registries vary greatly in the data they contain, the training of the personnel who record and enter data, the sophistication and versatility of the system, and the reports and information they can generate. Often the quality and completeness of the data raise serious questions regarding their value for anything, let alone as a research tool. The Major Trauma Outcome Study (MTOS) opened the way to interinstitutional collaboration in trauma indices. Since that time, many states or counties have obligatory or local registries from their trauma centers. In addition, similar registries have developed in the U.K., Germany, Austria, Norway, Australia and a number of other countries. The American College of Surgeons has assumed the mantel of pooling data from trauma centers in the United States and has the TRACS and the National Trauma Databases. Trauma indices and their applications are an essential element of a public health approach to injury and will continue to evolve in their accuracy and usefulness. Indices of injury severity are central to triage, to outcome evaluation and quality assurance. With the growing importance of their applications, severity scales are being more closely scrutinized for their accuracy and validity. Limitations discovered through analyses of registry data have motivated the development of improved measures. However, further improvements are needed. For example, more streamlined physiologic scores for triage, which maintain or improve the predictive power of current indices perhaps by trending data would be of great value. TRISS requires values for all component variables to derive survival probability estimates. However, it is frequently not possible to obtain best verbal response or respiratory rate from intubated patients, who must then be excluded from an institutions PRE and DEF analyses. Those exclusions can seriously bias outcome evaluations and reduce the benefit derived for quality assurance activities. Alternative means for quantitatively evaluating such patients are needed. TRISS coefficients based on NISS should be developed. Continued injury research can lead to the needed refinements in indices and methods and, thus, in their applications to triage, outcome evaluation, and quality assurance activities. USE OF SEVERITY SCORING IN RESEARCH

Although organized systems of medical reporting of injury have been in use since the early l980s, no comprehensive data source has evolved for the epidemiologic study of injury and resultant impairment. Such information has been found to be severely lacking and must be acquired from many sources such as hospital records, police reports, and governmental and research organizations. Inherent inconsistencies and inadequacies of these data sources limit their use for injury research.

The heterogeneity of trauma patient populations severely hampers both retrospective and prospective studies. Injury severity scores together with tight use of operational definitions for entry and exclusion criteria to studies can be of significant value in study design and thus enhance the ability of such studies to draw valid and generalizable conclusions. The TRISS methodology in particular has been used in numerous studies as a means of controlling

Trauma scoring

21

for casemix differences in different populations or over time and thus to normalize for injury severity and control for its effect on outcome. It is not perfect at this task, but it is more widely used than any tool other than tight case definition criteria which severely limit the populations. TRISS has been used throughout the world in this fashion and allows for a reasonably effective mechanism by which anatomy, physiology, age and mechanism of injury can be taken into account as to their influence on outcome when some other independent variable is being studied. The demands of acute care research and the difficulties in implementing research protocols in nocturnal weekend hours are primary factors in why acute care resuscitation research has attracted so little effort. This is so even though substantial research questions require to be answered not the least of which include the optimum resuscitation fluids, appropriate early resuscitation therapeutic endpoints and the value of blood substitute hemoglobin solutions in early acute resuscitation. Studies on patients in shock following injury require definition of a homogenous target population within this highly heterogenous subset of trauma patients. Without such attempts research in this domain will often be an exercise in futility. CHALLENGES The major challenges of trauma registries and injury severity scoring are as follows: 1. Unavailable data. It is particularly difficult to measure the GCS in patients who are intubated or paralyzed. This situation has led to approximations, few of which are valid. Thus, a number of researchers are exploring ways of measuring coma on intubated patients. 2. Missing data. It is notoriously difficult to get accurate and complete data in the early phases of care of the acutely injured individuals both in the prehospital and inhospital phases of care. Conventions for handling missing data and approximating their effect need to be developed. It is unfortunate that the largest sources of missing data are always the most critical patients. 3. Development and implementation of outcome measures other than death. It is hoped that Functional Capacity Index under development will address this problem. 4. Complex and military injuries. The current conventions for scoring anatomic injury do not relate well to the multiple etiologies and multiple injuries encountered in combat. War is estimated to be the 8th leading cause of death and disability of all diseases and injuries by the year 2020. Efforts are underway at the present time to develop an International Early Conflict Care database. If lessons learned from its civilian counterparts hold true, this latter effort will provide tremendous insight into the nature and severity of injuries, outcome and time dependency of theses injuries. However, be-

cause of the epidemiological and etiological differences of injury and the challenges of providing care in a tactical setting, substantial differences are likely to be encountered. This international collaborative effort should provide information to many countries. THE FUTURE Trauma severity scales albeit imperfect have established themselves in the patina of health care delivery as an essential tool for many research and public health initiatives. Severity scoring systems are a natural partner, with trauma registries and other databases designed to aggregate information on injured patients for a variety of applications, and provide the large numbers from which to develop valid models and statistically valid analyses. As the accuracy of injury severity scoring systems improves, they will have increased applications in risk assessment and decision support. It has already been determined that the risk of multiple organ failure sequelae of major injury increases with increasing injury severity. Combinations of anatomical and physiological scales, age and preexisting conditions are important in general classification of injury severity against survival probability. In the future, an identified threshold of injury severity will be used to trigger early assessment of indicators of high risk of serious sequelae of injury. These indicators may well include a risk model based on cytokine response and/or shock dose. The latter would be a measure which takes into account not only the severity of shock, but the length of time of hypoperfusion and the residual status after resuscitation as measured by lactate or base deficit. Such a model would isolate high risk patients allowing for institution of early prophylaxis against multiple organ failure and other highly lethal consequences of severe injury. Such models would also enable consistent case definition for studies of subsets of severe patients and decrease the statistical noise that bedevils much of the research into severely injured patients at this time. In conclusion, injury severity scores are an established part of trauma care as are trauma registries. The two work in conjunction as a basis for probabilistic models for a variety of scientific applications. Severity scoring systems are a work in process. Future developments will refine them and add scientific precision to their application. REFERENCES
01. Muller ME, Nazarian S, Koch P, Schataker J: The comprehensive classification of fractures of long bones. Berlin Heidelberg: Springer-Verlag, 1990 02. Teasdale G, Jennett B: Assessment of coma and impaired consciousness. A practical scale. Lancet 1974;2:81 03. Champion HR, Sacco WJ, Carnazzo AJ, et al.: Trauma score. Crit Care Med 1981;9:672 04. Champion HR, Sacco WJ, Hunt TK: Trauma severity scoring to predictor mortality. Wld J Surg 1983;7:411 05. Hosmer DW, Lemeshow S: Applied logistic regression. New

22

H. R. Champion 23. Sausia A, Moore FA, Moore EE, et al.: Epidemiology of trauma deaths: A reassessment. J Trauma 1995;38:18,593 24. Champion HR, Sacco WJ, Hannan DS, et al.: Assessment of injury severity: the triage index. Crit Care Med 1980;8:201 25. Champion HR, Sacco WJ: The trauma score as applied to penetrating injury. Ann Emerg Med 1984;13:6 26. Moreau M, Gainer P, Champion HR, Sacco WJ: Application of the trauma score in the prehospital setting. Ann Emer Med 1985;14:1049 27. Sacco WJ, Champion HR, Henderson JV: Implementation of severity scores in Navy casualty care. 17th Hawaii International Conference on System Sciences. Honolulu, Hawaii, 1984 28. Champion HR, Sacco WJ, Copes WS, Gann DS, Gennarelli TA, Flanagan ME: A revision of the trauma score. J Trauma 1989;29:623629 29. Gormican SP: CRAMS scale: Field triage of trauma victims. Ann Emerg Med 1982;11:132 30. Tepas JJ, Mollitt DL, Gans BM, DiScala C: The pediatric trauma score as a predictor of injury severity in the injured child. J Pediatr Surg 1987;22:1418 31. Eichelberger M, Gotschall C, Sacco W, et al.: A comparison of Trauma Score, the Revised Trauma Score, and the Pediatric Trauma Score. Ann Emerg Med 1989;18:1053 32. Lindsay KW, Teasdale GM, Knill-jones RP, Murray L: Observer variability in assessing the clinical features of subarachnoid haemorrhage. J Neurosurg 1983;58:5762 33. Lipowski ZJ: Delirium. Acute confusional states. Oxford University Press, 1990, Oxford. 34. Malmgren K, Bilting M, et al.: A compound score for estimating the influence of inattention and somnolence during the intracarotid amobarbital test. Epilepsy Res 1992;12:253259 35. Pilz G, Kaab S, Kreuzer E, Werdan K: Evaluation of definitions and parameters for sepsis assessment in patients after cardiac surgery. Infection 1994;22:817 36. Champion HR, Copes WS, Sacco WJ, Frey CF, Holcroft JW, Hoyt DB, Weigelt JA: Improved predictions from a severity characterization of trauma (ASCOT) over trauma and injury severity score (TRISS): results of an independent evaluation. J Trauma 1996;40:4249 37. Jennett B, Teasdale G, et al.: Predicting outcome in individual patients after severe head injury. Lancet 1976;1031 38. Knaus WA, Zimmerman JE, Wagner DP, et al.: APACHE acute physiology and health evaluation: A physiologically based classification system. Crit Care Med 1981;9:591597 39. Elebute EA, Stoner HB: The grading of sepsis; Br J Surg 1983; 983;70:293 40. Stevens LE: Gauging the severity of surgical sepsis. Arch Surg 1983;118:11901192 41. Goris RJA, te Boekhorst TPA, et al.: Multiple organ failure: generalized autodestructive inflammation? Arch Surg 1985;120: 11091115 42. Marshall JC, Cook DJ, et al.: Multiple organ dysfunction score: A reliable descriptor of a complex clinical outcome. Crit Care Med 1995;23:16381652

York, New York. John Wiley & Sons, Inc. 1989 06. American Medical Association Committee on the Medical Aspects of Automotive Safety: Rating the severity of tissue damage: The abbreviated scale. JAMA 1971;215:277 07. Baker SP, ONeill B, Haddon W, et al.: The injury severity score: An update. J Trauma 1974;14:187 08. Copes WS, Champion HR, Sacco WJ, Lawnick MM, Keast SL, Bain LW: The injury severity score revisited. J Trauma 1988;28: 6977 09. Copes WS, Champion HR, Sacco WJ, et al.: Progress in characterizing anatomic injury. J Trauma 1977;30:12001207 10. Osler T, Baker SP, Long W: A modification the Injury Severity Score that both improves accuracy and simplifies scoring. J Trauma. 1997;43:922926 11. US Department of Health and Human Services, Public Health Service, Health Care Financing Administration. International classification of diseases, 9th revision, clinical modification, sixth revision. October 1, 1997 12. Levy PS, Mullner R, Goldberg J, Gelfand H: The estimated survival probability index of trauma severity. Health Serv Res 1978;13:2835 13. Champion HR, Sacco WJ, Lepper RL, et al.: An anatomic index of injury severity. J Trauma 1980;20:197 14. Osler T, Rutledge R, Deis J, Bedrick E. ICISS: an international classification to disease-based injury severity score. J Trauma 1996;41:380388 15. Merkler J, Mickiewicz A, Milholland A: PEBL: A code for penetrating and blunt trauma based on the HICDA Index. Technical Report ARCSC-TR78054 CSL, Aberdeen Proving Ground, Md. October 1978 16. Mackenzie EJ, Sacco WJ, and Colleagues: ICDMAP-90: a users guide. Baltimore, Md: The Johns Hopkins University School of Public Health and Tri-Analytics, Inc, 1997 17. Rutledge R, Hoyt DB, Eastman AB, et al. Comparison of the injury severity score and ICD-9 diagnosis codes as predictors of outcome in injury: analysis of 44,032 patients. J Trauma 1997;42:477489 18. Rutledge R, Osler T, Emery S, Kromhout-Schiro S: The end of ISS and the TRISS: ICISS outperforms both ISS and TRISS as predictors of trauma patient survival, hospital charges and hospital length of stay. J Trauma 1998;44:4149 19. Sacco WJ, MacKenzie EJ, Champion HR, Davis EG, Buckman RF: Comparison of alternative methods for assessing injury severity based on anatomic descriptors. J Trauma 1999; 47:441447 20. Cayten CG, Stahl WM, Murphy JG, Agarwal N, Bryne DW: Limitations of the TRISS method for interhospital comparisons: A multihospital study. J Trauma 1991;1:31,471482 21. Markle J, Cayten CG, Byrne DW, Moy F, Murphy G: Comparison between TRISS and ASCOT methods in controlling for injury severity. J Trauma 1992;33:326332 22. Osler TM, Rogers FB, Badger GJ, Healey M, Vane DW, Shackford SR: A simple mathematical modification of TRISS markedly improves calibration. Abstract: American Association for the Surgery of Trauma Annual Meeting, Seattle, Wa., September 2001 (Meeting cancelled)

Received: September 11, 2001