International classification of disease based injury severity score (ICISS): A data linkage study of hospital and death data in Victoria, Australia

Abstract

BackgroundSurveillance of severe injury incidence and prevalence using ICD-based injury severity scores (ICISS) requires valid, locally applicable diagnosis-specific survival probabilities (DSPs). This study aims to derive and validate ICISS in Victoria, Australia, and compare various ICISS methodologies in terms of accuracy and calculated severe injury prevalence. MethodsThis study used injury admissions (ICD-10-AM coded) from the Victorian Admitted Episodes Database (VAED) linked with death data (Cause of Death – Unit Record Files: CODURF). Using design data (July 2008 – June 2014; n = 720,759), various ICISS scales were derived, based on (i) in-hospital and (ii) three-month mortality. These scales were applied to testing data (July 2014 – December 2016; n = 334,363). Logistic regression modelling was used to determine model discrimination and calibration. ResultsIn the design data, there were 6,337(0.9%) hospital deaths and 17,514(2.4%) three-months deaths; in the testing data, there were 2,700(0.8%) hospital deaths and 8,425(2.5%) three-month deaths. Newly developed ICISS scales had acceptable to outstanding discrimination, with Area Under the Curve ranging from 0.758 to 0.910. Age-specific ICISS scales were superior to general ICISS scales in model discrimination but inferior in model calibration. Calculated severe injury (ICISS ≤0.941) prevalence in the testing data ranged from 2% to 24%, depending on which mortality outcomes were used to calculate DRGs. ConclusionsThis study provides local, validated ICISS scores that can be used in Victoria. It is recommended that age group stratified ICISS based on the worst-injury method is used. From the comparison of various ICISS scores, reflecting the range of ICISS permutations that are currently in use, care should be taken to compare ICISS methodology before comparing severe injury prevalence per population, injury cause, and time trends.