Validation of a Rule-Based ICD-10-CM Algorithm to Detect Fall Injuries in Medicare Data

Abstract

Abstract Background Diagnosis-code-based algorithms to identify fall injuries in Medicare data are useful for ascertaining outcomes in interventional and observational studies. However, these algorithms have not been validated against a fully external reference standard, in ICD-10-CM, or in Medicare Advantage (MA) data. Methods We linked self-reported fall injuries leading to medical attention (FIMA) from the Strategies to Reduce Injuries and Develop Confidence in Elders (STRIDE) trial (reference standard) to Medicare fee-for-service (FFS) and MA data from 2015–19. We measured the area under the receiver operating characteristic curve (AUC) based on sensitivity and specificity of a diagnosis-code-based algorithm against the reference standard for presence or absence of ≥1 FIMA within a specified window of dates, varying the window size to obtain points on the curve. We stratified results by source (FFS vs MA), trial arm (intervention vs control), and STRIDE’s 10 participating health care systems. Results Both reference standard data and Medicare data were available for 4 941 (of 5 451) participants. The reference standard and algorithm identified 2 054 and 2 067 FIMA, respectively. The algorithm had 45% sensitivity (95% confidence interval [CI]: 43%–47%) and 99% specificity (95% CI: 99%–99%) to identify reference standard FIMA within the same calendar month. The AUC was 0.79 (95% CI: 0.78–0.81) and was similar by FFS or MA data source and by trial arm but showed variation among STRIDE health care systems (AUC range by health care system, 0.71 to 0.84). Conclusions An ICD-10-CM algorithm to identify fall injuries demonstrated acceptable performance against an external reference standard, in both MA and FFS data.