Causal Proportional Hazards Estimation with a Binary Instrumental Variable.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Unmeasured confounding complicates causal effect estimation in survival analysis.
• Existing instrumental variable (IV) methods are limited in nonlinear survival models with censored data.

Purpose of the Study:

• To develop a simple instrumental variable (IV) method for estimating causal effects in proportional hazards models with right-censored survival data.
• To address challenges in applying IV methods to complex survival settings, including informative censoring.

Main Methods:

• Developed a novel causal hazard ratio estimator using an intuitive inverse weighting scheme based on a special IV characterization.
• Established asymptotic properties and provided plug-in variance estimators for the proposed IV method.
• Validated the method through extensive simulation studies and applied it to the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer screening trial data.

Main Results:

• The proposed IV method effectively estimates causal effects in the presence of unmeasured confounding and informative noncompliance.
• The method demonstrated robustness in simulation studies across various survival data complexities.
• Application to the PLCO trial provided insights into the causal effect of sigmoidoscopy screening on colorectal cancer survival.

Conclusions:

• The developed instrumental variable (IV) method offers a broadly applicable and statistically rigorous approach for causal inference in complex survival data.
• This method enhances the ability to delineate true treatment effects, even with unmeasured confounding and censoring.
• The approach is implementable in standard statistical software, facilitating its use in epidemiological and clinical research.