Interpreting IV Estimates with Treatment Effect Heterogeneity: ACE/ARBs & Race

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Treatment rates vary dramatically across patients grouped by race, age, and gender. If treatment effects are heterogeneous across patients, though, it is not clear from treatment rate comparisons alone whether treatment rates should be increased or decreased for specific patient groups. Theorists have suggested that estimates of the "local average treatment effect" LATE produced by instrumental variable (IV) estimators provide evidence to assess whether treatment rates should be modified. However, alternative IV estimators exist that have distinct approaches to estimate LATEs across patient groups. No methodological research has contrasted the ability of alternative IV estimators to estimate LATEs across patient groups. Since 2005 over 375 articles have been published in healthcare using IV estimators. Methodological research is needed to ensure proper inferences are made from IV estimates when evaluating the comparative effectiveness of treatments for distinct patient groups. This research will first perform simulation modeling to assess the properties of alternative IV estimators to estimate treatment effectiveness within patient groups when treatment effects are heterogeneous both across and within these groups. Next, we will estimate treatment effectiveness using alternative IV estimators for a treatment thought to have heterogeneous treatment effects both within and across patient groups and interpret these estimates through the prism of our simulation findings. Guidelines suggest that renin- angiotensin system antagonists (ACE/ARBs) be used for secondary prevention post-acute myocardial infarction (AMI), but controlled trial data suggests that ACE/ARB effectiveness is lower for African Americans than whites. In practice, African American ACE/ARB utilization rates are significantly lower post AMI than whites. Are these rate differences by race justified by the clinical evidence or are clinicians underutilizing ACE/ARBs in African Americans? To investigate this question we will use Medicare claims from the Centers for Medicare & Medicaid Services (CMS) Chronic Condition Data Warehouse (CCW) for patients with primary diagnosis of AMI that have Medicare "Part D" prescription drug coverage. In addition, we will perform chart abstractions for a portion of our AMI sample to assess the extent that factors unmeasured in Medicare claims confound our IV estimates and whether our estimates can be interpreted as bounds of true treatment effects with each race group. PUBLIC HEALTH RELEVANCE: Assessing the comparative effectiveness of treatments in practice often requires analysis of observational healthcare databases. The properties of methods available to estimate comparative effectiveness from observational data are often unclear. The goal of this research is to provide clarity to healthcare policy-makers as to the treatment effect inferences that can be made from instrumental variable (IV) estimators.