Regression-based Proximal Causal Inference for Right-censored Time-to-event Data.

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Unmeasured confounding poses a significant challenge to drawing valid causal conclusions from observational data.
• Proximal causal inference offers a framework to address confounding bias using negative control variables.
• Existing regression methods for proximal causal inference do not adequately cover right-censored time-to-event outcomes.

Purpose of the Study:

• To develop and validate a novel proximal causal inference regression method for right-censored survival data.
• To extend the application of proximal causal inference to time-to-event outcomes, a previously unaddressed area.
• To provide a robust statistical framework for analyzing observational data with potential unmeasured confounding.

Main Methods:

• A two-stage regression approach is proposed for right-censored survival data.
• The method is based on an additive hazard structural model.
• Theoretical justifications are provided for various types of negative control outcomes (continuous, count, time-to-event).

Main Results:

• The proposed method effectively addresses unmeasured confounding in time-to-event data.
• The approach is demonstrated using real-world data on the effectiveness of right heart catheterization.
• The methodology is implemented in the open-access R package "pci2s".

Conclusions:

• The novel two-stage regression proximal causal inference method provides a valuable tool for analyzing survival data.
• This approach enhances the credibility of causal inferences from observational studies with time-to-event outcomes.
• The availability of the "pci2s" R package facilitates the application of this advanced methodology.