Using instrumental variables regression to evaluate Medicaid disease management program effectiveness: an exploratory analysis

In disease management (DM) evaluations, causal inferences about the effects of participation may be subject to omitted variables bias (OVB), which could compromise policy makers' efforts to accurately determine how effective these programs are in treating chronically ill patients. Instrumental variables (IV) regression using patient three-digit ZIP codes as instruments has been offered as a solution to OVB in DM evaluations. Our main objective was to investigate the utility of this method by using it to evaluate the effects of a Medicaid diabetes DM program on annual diabetes-related costs, emergency room visits, and hospital days. The IV regression estimates were compared with those obtained from similar ordinary least squares (OLS) regressions. Our analysis indicated that the IV estimates may be unreliable due to the failure of some of the fundamental assumptions on which they are based. The appropriateness of this particular IV method may depend on several factors including the study sample, the results of statistical validity tests, and the specific policy questions that the analyses address. Our study concerns a method that could be used to inform health policy debate over whether DM programs provide services that result in improved patient outcomes for subpopulations of individuals whose participation decision was influenced by geographic region. In light of the current economic crisis that the federal government and states are facing, it is imperative that policy makers and those who advise them fully understand these issues when using this procedure to determine if DM programs represent an appropriate investment of public money. Key words: Medicaid, disease management evaluation, instrumental variables regression, ZIP code instruments, omitted variables bias.
Chronic diseases, such as diabetes, hypertension, asthma, and coronary heart disease are serious conditions that affect approximately 25 percent of the US population and account for more than 75 percent [...]