Causal Inference in Health Services Research

Abstract

In Chapter 1, we introduce a common subject of health services research, the "volume-outcomes relationship," or the study of the relationship between patient outcomes and hospital and/or physician case volume. We describe and emulate four hypothetical randomized trials ("target trials") in order to demonstrate methods that can be used to determine the effect on mortality of individuals undergoing pancreatic resection for malignancy. In doing so, we highlight the need to carefully consider well-defined causal contrasts, positivity violations, and multiple treatment versions. We elucidate where prior analyses fit into and deviate from this framework. In Chapter 2, we note that prior analyses of the volume-outcomes relationship often fail to draw a distinction between interventions on patients selecting physicians with certain case-volumes and interventions on the case-volume of physicians. We specify four target trials that could be used to estimate the effect on post-operative patient mortality of intervening on the operative volume of surgeons performing coronary artery bypass grafting operations. We demonstrate how to implement the analysis with an approach that contrasts sharply with that of the patient intervention from Chapter 1. In Chapter 3, we address another common substantive topic in health services research, interventions that affect hospital readmission after discharge from an inpatient service. Using recently developed methods of separable effects, we focus on emulating target trials by incorporating competing events into the analyses in an interpretable way.