Instrumental variable estimation in semi-parametric additive hazards models.

Area of Science:

• Biostatistics
• Survival Analysis
• Econometrics

Background:

• Instrumental variable (IV) methods are crucial for unbiased estimation when unmeasured confounders are present.
• Additive hazard models are suitable for censored survival data, accommodating time-dependent covariates.
• Existing IV adaptations to survival models have limitations, particularly with complex covariate effects.

Purpose of the Study:

• To prove the asymptotic normality of two-stage residual inclusion (2SRI) estimators in semi-parametric additive hazard models.
• To extend 2SRI methods for both time-independent and time-dependent covariate effects.
• To provide methods for estimating conditional survival functions and simultaneous confidence bands.

Main Methods:

• Developed and proved asymptotic normality for 2SRI estimators within a semi-parametric additive hazard framework.
• Considered both continuous and binary exposure variables.
• Proposed a resampling scheme for simultaneous confidence bands of the conditional survival function.

Main Results:

• The 2SRI method allows direct estimation of time-independent covariate effects without restricting residual influence.
• The proposed methods demonstrate favorable performance compared to existing techniques in simulations.
• The methods are particularly effective in scenarios with exposure-dependent censoring.

Conclusions:

• The 2SRI method is a robust approach for causal inference in semi-parametric additive hazard models.
• The developed techniques enhance the analysis of censored survival data with potential unmeasured confounding.
• The study provides valuable tools for biostatistical and econometric research involving survival data.