Keywords: Bayesian; multi-outcome evaluation; hierarchical model
Health care evaluations are multi-faceted; we wish to gauge a program’s impact on not only health care costs but also utilization and quality of care. In a typical approach, we evaluate the program’s impact on each outcome independently, a fragmented strategy that increases the risk of both false positive and false negative findings. Estimating impacts simultaneously in a hierarchical Bayesian model can strengthen our conclusions by capitalizing on relationships among outcomes that typical approaches assume away. These relationships reinforce trends and negate chance findings by placing a single outcome in the context of other, related outcomes, much as more standard Bayesian approaches place a single subgroup in the context of the overall sample. Through a case study from the evaluation of the Comprehensive Primary Care Initiative (CPC), a Center for Medicare and Medicaid Innovation (CMMI) demonstration, we illustrate how borrowing strength across outcomes produces more precise estimates and a more comprehensive understanding of the intervention’s overall impact by integrating the contextual information we use to assess traditional inference into the statistical model itself.