Development and validation of a prognostic index for mental health and substance use disorder burden

Abstract

ObjectiveTo develop an accessible index which quantifies MHSUD burden among patients of Veterans Affairs hospitals. MethodWe used 21 disorder categories provided by the diagnostic and statistical manual (DSM) to characterize diagnoses among primary care (PC) patients. For each patient, we generated counts of unique disorder categories present during the PC encounter or in the year prior. We used these counts to generate multiple indexes, which we compared in a 60% training sample of our population. Using model fit statistics generated from ordered multinomial logistic regressions, we identified the subset of DSM categories which, structured as index, were most predictive of MHSUD hospitalization and death. We validated and fine-tuned the form of the selected index in the full population using measures of calibration and discrimination. ResultsIn model development, the index (I-6) which best fit the data (R2 = 0.191) included the following six disorder categories: substance use, depressive, psychotic, bipolar, trauma, and personality. When applied in the full population and weighted by disorder severity, this index demonstrated good predictive discrimination for MHSUD death (C = 0.66) and hospitalization (C = 0.88) and was well calibrated in comparisons of observed versus predicted outcomes. ConclusionsWe recommend the I-6 as a parsimonious and effective tool for MHSUD burden risk adjustment.