Time-Varying IV Marginal Structural Quantile Models for Censored Survival Data

This project develops statistical methods for estimating population-level causal effects of time-varying treatments on censored survival outcomes when treatment decisions are affected by unmeasured confounding. The proposed approach combines time-varying instrumental-variable weighting with a marginal structural quantile model, allowing treatment effects to vary across the survival-time distribution rather than being summarized by a single mean or hazard ratio. Estimation is based on a recursively defined censored quantile estimating equation, with treatment, instrumental-variable, and compliance-related nuisance functions estimated from observed data. Statistical uncertainty is assessed using a perturbation bootstrap that re-estimates the nuisance functions in each replicate. The method will be evaluated through extensive Monte Carlo simulations under different sample sizes, treatment-effect patterns, and censoring levels, and may subsequently be illustrated using a suitable de-identified secondary dataset. Project outputs will include simulation datasets, summary tables, figures, analysis code, computational logs, and manuscripts. No directly identifiable personal data will be generated or collected during the simulation phase.