Health Disparities
Combining Target Trial Emulation with Causal Decomposition to Estimate Intervention Effects on Health Disparities Chloe R. Bennett* Chloe Bennett Bennett Uniformed Services University of the Health Sciences
Researchers often ask how much health disparities could be reduced under more equitable conditions, but “equity” can be operationalized in many different ways, each representing distinct interventions. Causal decomposition analysis provides principled guidance for distinguishing which covariates represent inequitable barriers (constituting the disparity) versus legitimate sources of variation (appropriate for standardization) and allows researchers to estimate disparity reduction under hypothetical equitable scenarios. Target trial emulation provides a framework for defining such an intervention by requiring explicit definition of a protocol—including eligibility criteria, treatment assignment mechanisms, time zero, and follow-up periods—thereby transforming abstract equity goals into concrete, evaluable interventions. Using continuous glucose monitor (CGM) access and racial disparities in diabetes outcomes as an example, different equity framings lead to fundamentally different target trial designs. Equity as universal access would result in examining outcomes if all patients across racial groups received CGMs. Equity as equalized access rates would examine outcomes if the disadvantaged group’s CGM access matched the advantaged group’s existing rate. Equity as need-based allocation would provide CGMs to anyone exceeding clinical thresholds (e.g., A1C > 8%), eliminating socioeconomic barriers to clinically-indicated care. Each scenario requires different target trial specifications and yields different disparity reduction estimates. By combining causal decomposition with target trial emulation, researchers can systematically evaluate multiple operationalizations of equity, identifying which specific, implementable interventions would most effectively reduce disparities while maintaining rigorous causal identification.