Primary Submission Category: Mediation Analysis, Mechanisms
Efficient estimation of pathway effects mediated by intermediate events in multi-state models
Authors: Yuhao Deng, Haoyu Wei, Donglin Zeng, Rui Song, Xiao-Hua Zhou,
Presenting Author: Yuhao Deng*
Cardiovascular and microvascular events are leading causes of death in patients with type 2 diabetes. While a randomized controlled trial indicated that X reduces cardiovascular and microvascular risks as well as mortality, the mechanism by which X prevents vascular events and death is not fully understood. In this work, we consider hypothetical interventions in each transition of disease progression. Through such interventions, we distinguish the effects along specific pathways from the total effect on each event. Our proposed framework enables three key applications: estimating path-specific treatment effects, identifying which events are influenced by treatment, and inferring dynamic treatment strategies. Based on multi-state models, we derive multiply robust, nonparametrically efficient estimators for the counterfactual cumulative incidences and treatment effects, accompanied by inference procedures. By analyzing data from a randomized controlled trial, we find that X significantly reduces the risk of non-fatal expanded major adverse cardiovascular events as well as microvascular events. Beyond existing results, our new methods recover the following potentially useful clinical findings. The reduction in all-cause mortality associated with X is primarily mediated by its effects on expanded major adverse cardiovascular events, and importantly, sustained adherence to X is crucial for achieving an effective reduction in cardiovascular risk