Multiple comparisons for survival data with propensity score adjustment.

Area of Science:

• Biostatistics
• Survival Analysis
• Causal Inference

Background:

• Clinical and observational studies often involve comparing multiple treatment or prognostic groups with right-censored survival data.
• Existing methods may not adequately address multiple comparisons or causal inference in such complex survival data scenarios.

Purpose of the Study:

• To develop and extend statistical methods for multiple comparisons with a control (MCC) and multiple comparisons with the best (MCB) for survival outcomes.
• To incorporate propensity score adjustments for causal inference, reducing confounding bias in survival data analysis.
• To provide robust testing procedures and simultaneous confidence intervals for causal statements in survival studies.

Main Methods:

• Extension of MCC and MCB approaches from general linear models to survival outcomes using a propensity-score-stratified Cox proportional hazards model.
• Specification of assumptions for causal inference within a potential outcome framework for survival data.
• Development of testing procedures and simultaneous confidence intervals for comparing multiple groups with censored survival data.

Main Results:

• The proposed methods effectively extend MCC and MCB for survival data with right censoring and confounding.
• The study provides a framework for causal inference on survival outcomes, including testing and confidence intervals.
• Application to real cancer study data and a simulation study demonstrate the utility and validity of the developed methods.

Conclusions:

• The developed statistical framework enhances the ability to make causal statements when comparing multiple groups with censored survival data.
• Propensity score adjustment within a Cox model is effective for addressing confounding in these comparisons.
• The methods offer valuable tools for researchers in clinical and observational studies analyzing survival outcomes.