A1342
Title: Dynamic causal mediation analysis for micro-randomized trials
Authors: Tianchen Qian - University of California Irvine (United States) [presenting]
Abstract: Mobile health (mHealth) interventions aim to deliver real-time, personalized behavioral support, premised on the idea that prompts drive short-term behavior changes that produce long-term health benefits. Testing this requires causal mediation analysis, but applying it to micro-randomized trials (MRTs), where individuals are randomized hundreds of times, poses two challenges: (1) the large number of treatment occasions (e.g., 210 decision points in HeartSteps with only 37 participants) makes it difficult to define and model causal effects on a distal outcome; and (2) the many potential mediators, since a treatment may influence the outcome through all future mediators, complicate decomposition into interpretable pathways. Natural direct and indirect excursion effects address these challenges by using marginal contrasts to define each treatment occasion's effect on a distal outcome and focusing mediation on the most immediate mediator. This captures the most meaningful pathways in mHealth (prompt -> immediate behavioral target -> long-term health goal) while improving the plausibility of identification assumptions. Efficient influence functions are derived and a multiply robust estimator is proposed that accommodates flexible modeling and is doubly robust in the MRT context. Applied to HeartSteps, early direct effects of activity suggestions are large but decline rapidly, while indirect effects, mediated through increased immediate walking, remain stable throughout.
