Mathieu Gagné1, Lynne Moore, Claudia Beaudoin, Brice Lionel Batomen Kuimi, Marie-Josée Sirois. 1. From the Bureau d'information et d'études en santé des populations (M.G., C.B.), Institut national de santé publique du Québec; Département de médecine sociale et préventive (M.G., L.M., C.B., B.L.B.K.), Faculté de médecine, Université Laval; Axe Santé des Populations et pratiques Optimales en Santé (Population Health and Optimal Health Practices Research Unit) (L.M., B.L.B.K., M-J.S.), Traumatologie - Urgence - Soins intensifs (Trauma - Emergency - Critical Care Medicine), Centre de Recherche du Centre Hospitalier Universitaire (CHU) de Québec (Hôpital de l'Enfant-Jésus), Université Laval, Québec (Qc), Canada; and Département de réadaptation (M-J.S.), Faculté de médecine, Université Laval.
Abstract
BACKGROUND: The International Classification of Diseases (ICD) is the main classification system used for population-based injury surveillance activities but does not contain information on injury severity. ICD-based injury severity measures can be empirically derived or mapped, but no single approach has been formally recommended. This study aimed to compare the performance of ICD-based injury severity measures to predict in-hospital mortality among injury-related admissions. METHODS: A systematic review and a meta-analysis were conducted. MEDLINE, EMBASE, and Global Health databases were searched from their inception through September 2014. Observational studies that assessed the performance of ICD-based injury severity measures to predict in-hospital mortality and reported discriminative ability using the area under a receiver operating characteristic curve (AUC) were included. Metrics of model performance were extracted. Pooled AUC were estimated under random-effects models. RESULTS: Twenty-two eligible studies reported 72 assessments of discrimination on ICD-based injury severity measures. Reported AUC ranged from 0.681 to 0.958. Of the 72 assessments, 46 showed excellent (0.80 ≤ AUC < 0.90) and 6 outstanding (AUC ≥ 0.90) discriminative ability. Pooled AUC for ICD-based Injury Severity Score (ICISS) based on the product of traditional survival proportions was significantly higher than measures based on ICD mapped to Abbreviated Injury Scale (AIS) scores (0.863 vs. 0.825 for ICDMAP-ISS [p = 0.005] and ICDMAP-NISS [p = 0.016]). Similar results were observed when studies were stratified by the type of data used (trauma registry or hospital discharge) or the provenance of survival proportions (internally or externally derived). However, among studies published after 2003 the Trauma Mortality Prediction Model based on ICD-9 codes (TMPM-9) demonstrated superior discriminative ability than ICISS using the product of traditional survival proportions (0.850 vs. 0.802, p = 0.002). Models generally showed poor calibration. CONCLUSION: ICISS using the product of traditional survival proportions and TMPM-9 predict mortality more accurately than those mapped to AIS codes and should be preferred for describing injury severity when ICD is used to record injury diagnoses. LEVEL OF EVIDENCE: Systematic review and meta-analysis, level III.
BACKGROUND: The International Classification of Diseases (ICD) is the main classification system used for population-based injury surveillance activities but does not contain information on injury severity. ICD-based injury severity measures can be empirically derived or mapped, but no single approach has been formally recommended. This study aimed to compare the performance of ICD-based injury severity measures to predict in-hospital mortality among injury-related admissions. METHODS: A systematic review and a meta-analysis were conducted. MEDLINE, EMBASE, and Global Health databases were searched from their inception through September 2014. Observational studies that assessed the performance of ICD-based injury severity measures to predict in-hospital mortality and reported discriminative ability using the area under a receiver operating characteristic curve (AUC) were included. Metrics of model performance were extracted. Pooled AUC were estimated under random-effects models. RESULTS: Twenty-two eligible studies reported 72 assessments of discrimination on ICD-based injury severity measures. Reported AUC ranged from 0.681 to 0.958. Of the 72 assessments, 46 showed excellent (0.80 ≤ AUC < 0.90) and 6 outstanding (AUC ≥ 0.90) discriminative ability. Pooled AUC for ICD-based Injury Severity Score (ICISS) based on the product of traditional survival proportions was significantly higher than measures based on ICD mapped to Abbreviated Injury Scale (AIS) scores (0.863 vs. 0.825 for ICDMAP-ISS [p = 0.005] and ICDMAP-NISS [p = 0.016]). Similar results were observed when studies were stratified by the type of data used (trauma registry or hospital discharge) or the provenance of survival proportions (internally or externally derived). However, among studies published after 2003 the Trauma Mortality Prediction Model based on ICD-9 codes (TMPM-9) demonstrated superior discriminative ability than ICISS using the product of traditional survival proportions (0.850 vs. 0.802, p = 0.002). Models generally showed poor calibration. CONCLUSION: ICISS using the product of traditional survival proportions and TMPM-9 predict mortality more accurately than those mapped to AIS codes and should be preferred for describing injury severity when ICD is used to record injury diagnoses. LEVEL OF EVIDENCE: Systematic review and meta-analysis, level III.