Turner M. Osler

TRISS has reigned as the preeminent trauma outcome prediction model for 20 years. Despite this endorsement, the calibration of TRISS has been poor in most data sets where it has been examined. We hypothesized that the lack of calibration of TRISS was because of the inappropriate mathematical specification of the model that TRISS is based on, rather than the predictors in the model. In particular, we hypothesized that the nonlinearity of the Injury Severity Score (ISS) in the log odds of death was responsible for the poor calibration of TRISS, and further, that this nonlinearity could be corrected by the simple addition of an ISS squared term to the TRISS model. We examined ISS in the log odds of mortality for linearity in one large trauma data set, the National Pediatric Trauma Registry (NPTR) (n = 53,113 from 1985-1996; mortality, 1.3%); and two small data sets, the University of New Mexico (UNM) (n = 3,142 from 1991-1995; mortality, 8.6%) and Portland, Oregon (PORT) (n = 2,916 from 1990-1994; mortality, 1.75%). In addition, in the NPTR we compared the calibration of TRISS models with and without linearity in the log odds of death. In the NPTR, ISS was profoundly nonlinear in the log odds of death for both blunt and penetrating trauma (p < 0.001). Moreover, the overall calibration of the TRISS model for the NPTR data was significantly improved when the nonlinearity of ISS was corrected by the addition of a quadratic ISS term as demonstrated by a 70% reduction (improvement) in the Hosmer-Lemeshow statistic. Interestingly, the addition of the ISS squared term did not affect the discrimination of the model. The log odds of survival in the UNM and PORT data sets were also better modeled when an ISS squared term was added (UNM, p = 0 0.052; PORT, p = 0.014), but improvements in the Hosmer-Lemeshow statistic were smaller, possibly because of the small size of these data sets. The TRISS model for outcome prediction currently uses ISS in a mathematically inappropriate way that impairs the calibration, but not the discrimination, of its predictions. If TRISS is to continue as the prediction standard for trauma, a quadratic ISS term must be added to the model. In the future, outcome prediction models should undergo thorough statistical modeling and evaluation before being released. Injury severity descriptors other than ISS (such as ASCOT, ICISS, or NISS) may require other modeling techniques to optimize the calibration of survival models that use these injury scores.

Pediatric trauma centers (PTCs) were developed to improve the survival of injured children, but it is currently unknown if children admitted to PTCs are more likely to survive than those admitted to adult trauma centers (ATCs). Fifty-three thousand one hundred thirteen pediatric trauma cases from 22 PTCs and 31 ATCs included in the National Pediatric Trauma Registry were reviewed to evaluate survival rates at PTCs and ATCs. Overall, 1,259 children died. The raw mortality rate was lower at PTCs (1.81% of 32,554 children) than at ATCs (3.88% of 18,368 children). However, patients admitted to ATCs were more severely injured. When Injury Severity Score, Pediatric Trauma Score, mechanism (blunt or penetrating), gender, age, clustering, and American College of Surgeons (ACS) verification status were controlled for using a single logistic regression model, there was no statistically significant difference in survival between PTCs and ATCs (odds ratio, 1.02; 95% confidence interval, 0.83–1.26;p = 0.587). A similar comparison of the 12 ACS-verified trauma centers with the 41 nonverified centers showed verification to be associated with improved survival (odds ratio, 0.75; 95% confidence interval, 0.58–0.97;p = 0.013). Although PTCs have higher overall survival rates than ATCs, this difference disappears when the analysis controls for Injury Severity Score, Pediatric Trauma Score, age, mechanism, and ACS verification status. ACS-verified centers have significantly higher survival rates than do unverified centers.

Previous work has demonstrated that the International Classification of Diseases 9th Revision (ICD-9) Based Illness Severity Score (ICISS) methodology developed by Rutledge and Osler can perform well in this role as a severity adjustment tool in trauma patients. The purpose of the present study was to extend this previous work to determine the ability of ICISS to predict outcomes in all types of hospitalized patients. The ICISS methodology was used to derive predictions of survival, length of hospital stay, and hospital charges in the entire study population. A total of 821,455 hospitalized patients in North Carolina in 1996 had complete data available for analysis. The overall hospital mortality rate was 2.9%. ICISS was an accurate predictor of hospital survival in all hospitalized patients (accuracy 95.9%, sensitivity 97.2%, and specificity 52.7%.) The area of the receiver operator characteristic curve was 0.93. By adding to the model, the area under the receiver operator characteristic curve increased to 0.95. ICISS also explained a large amount of the variance in hospital stay and charges (R2 = 0.38 and 0.56, respectively, P < .0001). This study extends previous work suggesting that ICISS may be an important improvement over other presently available severity adjustment models. If these findings are confirmed in comparison with other predictive tools, ICISS may find an important place in assessing illness severity.

Risk stratification of patients in the intensive care unit (ICU) is an important tool because it permits comparison of patient populations for research and quality control. Unfortunately, currently available scoring systems were developed primarily in medical ICUs and have only mediocre performance in surgical ICUs. Moreover, they are very expensive to purchase and use. We conceived a simple risk-stratification tool for the surgical ICU that uses readily available International Classification of Diseases, Ninth Revision, codes to predict outcome. Called ICISS (International Classification of Disease Illness Severity Score), our score is the product of the survival risk ratios (obtained from an independent data set) for all International Classification of Diseases, Ninth Revision, diagnosis codes. A total of 5,322 noncardiac patients admitted to a surgical ICU during an 8-year period had their Acute Physiology and Chronic Health Evaluation (APACHE) II scores compared with their ICISS as predictors of outcome (survival/nonsurvival, length of stay, and charges). ICISS proved to be a much better predictor of survival than APACHE (receiver operating characteristic (ROC) APACHE = 0.806; Hosmer-Lemeshow (HL) APACHE = 22.56; ROC ICISS = 0.892; HL ICISS = 12.06) or the APACHE survival probability (ROC = 0.836; HL = 34.47). These differences were highly statistically significant (p < 0.001). ICISS was also better correlated with ICU length of stay (APACHE R2 = 0.06; ICISS R2 = 0.32) and ICU charges (APACHE R2 = 0.07; ICISS R2 = 0.39). When combined in a logistic model with ICISS, APACHE II added slightly to the predictive power of ICISS alone (combined ROC = 0.903) but degraded the calibration of the model (combined HL = 16.29; p = 0.038). Because ICISS is both more accurate and much less expensive to calculate than APACHE II score, ICISS should replace APACHE II score as the standard risk stratification tool in surgical ICUs.

The Injury Severity Score (ISS) has served as the standard summary measure of anatomic injury for more than 20 years. Nevertheless, the ISS has an idiosyncrasy that both impairs its predictive power and complicates its calculation. We present here a simple modification of the ISS called the New Injury Severity Score (NISS), which significantly outperforms the venerable but dated ISS as a predictor of mortality. Retrospective calculation of NISS and comparison of NISS with prospectively calculated ISS. The NISS is defined as the sum of the squares of the Abbreviated Injury Scale scores of each of a patient's three most severe Abbreviated Injury Scale injuries regardless of the body region in which they occur. NISS values were calculated for every patient in two large independent data sets: 3,136 patients treated during a 4-year period at the American College of Surgeons' Level I trauma center in Albuquerque, New Mexico, and 3,449 patients treated during a 4-year period at the American College of Surgeons' Level I trauma center at the Emanuel Hospital in Portland, Oregon. The power of NISS to predict mortality was then compared with previously calculated ISS values for the same patients in each of the two data sets. We find that NISS is not only simple to calculate but more predictive of survival as well (Albuquerque: receiver operating characteristic (ROC) ISS = 0.869, ROC NISS = 0.896, p < 0.001; Portland: ROC ISS = 0.896, ROC NISS = 0.907, p < 0.004). Moreover, NISS provides a better fit throughout its entire range of prediction (Hosmer Lemeshow statistic for Albuquerque ISS = 29.12, NISS = 8.88; Hosmer Lemeshow statistic for Portland ISS = 83.48, NISS = 19.86). NISS should replace ISS as the standard summary measure of human trauma.

We view with concern the claim of Geerts et al. (Sept. 5 issue)¹ that the potent anticoagulant low-molecular-weight heparin can safely be used as routine prophylaxis against deep venous thrombosis in trauma victims. The study compared low-dose heparin with low-molecular-weight heparin in a cohort of 265 patients. The authors found that deep venous thrombosis occurred at a higher rate in the low-dose–heparin group (60 of 136 patients, or 44 percent) than in the low-molecular-weight–heparin group (40 of 129, or 31 percent; P = 0.014). Thus, 18 (60 minus 31 percent of 136) cases of deep venous thrombosis may have been prevented in the group that received low-molecular-weight heparin. This reduction in the rate of deep venous thrombosis was purchased, however, at the price of five episodes of severe hemorrhage in the low-molecular-weight–heparin group, including one subdural hematoma requiring operative drainage. Geerts et al. note that only one episode of bleeding (a nosebleed) occurred in the low-dose–heparin “control” group but claim that there was no difference between the two groups on the basis of a Fisher's exact test P value of 0.12.