Predictive validity of the All Patients Refined Diagnosis Related Group modifiers for costs and outcomes from intracranial hemorrhage

OBJECTIVE The All Patients Refined Diagnosis Related Group (APR-DRG) modifiers—severity of illness (SOI) and risk of mortality (ROM)—inform hospital reimbursement nationally. The ubiquitous APR-DRG data bear the potential to inform public health research; however, the algorithms that generate these modifiers are proprietary and therefore should be independently verified. This study evaluated the predictive value of APR-DRG modifiers for the outcomes and costs of intracranial hemorrhage. METHODS The New York Statewide Planning and Research Cooperative System databases were accessed and searched for the intracranial hemorrhage Diagnosis Related Group in records from 2012 to 2020. Receiver operating characteristic and multiple logistic regressions characterized the predictive validity of the APR-DRG modifiers for patient outcomes. One-way ANOVA compared costs and charges between SOI and ROM designations. RESULTS Among 46,019 patients, 12,627 (27.4%) died. The mean ± SEM costs per patient were $21,342 ± $145 and the mean ± SEM charges per patient were $68,117 ± $408. For prediction of mortality, the area under the curve (AUC) was 0.74 for SOI and 0.83 for ROM. For prediction of discharge to a facility, AUC was 0.62 for SOI and 0.64 for ROM. Regression analysis showed that ROM was a strong predictor of mortality, while SOI was a weak predictor; both were modest predictors of discharge to a facility. SOI and ROM were significant predictors of costs and charges. CONCLUSIONS Compared with the prior studies, the authors identified several limitations of APR-DRG modifiers, including low specificity, modest AUC, and limited outcomes prediction. This report supports the limited use of APR-DRG modifiers in independent research on intracranial hemorrhage epidemiology and reimbursement and advocates for general caution in their use for evaluation of neurosurgical disease.