Profiling Substance Abuse Provider Trends in Health Care Delivery Systems

In profiling of health care providers, distinguishing extreme behavior from random variation with incomplete risk adjustment requires multiple observations on each profiling unit. Unlike typical health care delivery system studies, the US Department of Veterans Affairs (VA) information system does provide longer time series of data at individual, institutional and system levels. This information resource allows us to develop incisive profiling analyses that isolate and highlight system wide improvements and institution-specific profiles in the context of risk adjustments using several covariates. This is illustrated here in the context of substance abuse care. One common process monitor for systems delivering substance abuse care is follow-up outpatient care within a certain number of days after inpatient substance abuse discharges. The VA system provides ten years of such data, at the individual level, and we employ this to build hierarchical models that profile providers within the system. Our models use logistic regression, longitudinal random effects models at the individual patient level, combined with simple time series models of institutional effects across years. This structure effectively captures variability across hospitals within each year as well as systematic dependencies within hospitals from year to year. Analysis depends on Markov chain Monte Carlo methods to derive posterior inferences for all parameters. Results indicate significant system wide improvement in the monitor in addition to large amounts of variation in this improvement across medical centers. Covariates such as age of patient, VA treatment priority, and diagnoses (where psychotic patients have lower return rates) help to illustrate important potential new health policy interventions and the outcomes of previous interventions.