A1309
Title: Estimation of average treatment effect for survival outcomes with continuous treatment in observational studies
Authors: Qi Zheng - University of Louisville (United States) [presenting]
Triparna Poddar - University of Louisville (United States)
Maiying Kong - University of Louisville (United States)
Abstract: In healthcare research, where extensive observational datasets such as claims data and electronic health records are abundant, researchers often aim to explore both the effects of treatments and the mechanisms by which these effects occur. While recent literature on causal effects in survival analyses typically concentrates on binary or multiple treatment scenarios, studies involving continuous treatment settings remain comparatively underexplored. Prompted by the need to assess the impact of blood lead levels on mortality among older adults in the United States, this project investigates the estimation of the average treatment effect (ATE) of continuous treatment on time-to-event outcomes. Estimating the ATE directly is proposed using an accelerated failure time-based marginal structural model (AFT-MSM). To tackle multiple confounding factors and censoring issues, the inverse probability of treatment weighting (IPTW) method is utilized, complemented by censoring weights. This approach has been rigorously validated through theoretical examinations and comprehensive simulation studies, affirming its validity and effectiveness. Additionally, the analysis suggests that the current regulatory level for blood lead is safe regarding mortality risk.
