Covariate-adjusted survival analyses in propensity-score matched samples: Imputing potential time-to-event outcomes

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Propensity score matching is a common method for estimating treatment effects in observational studies.
• Combining matching with regression adjustment has shown superior statistical properties for continuous outcomes.
• Survival outcomes are prevalent in biomedical and epidemiological research, necessitating specialized methods.

Purpose of the Study:

• To propose a novel method integrating regression adjustment and propensity score matching for survival data analysis.
• To estimate survival curves and hazard ratios accurately using observational data.
• To address the challenge of analyzing time-to-event outcomes in the presence of confounding.

Main Methods:

• Developed a method to impute potential outcomes under control for matched treated subjects.
• Utilized accelerated failure time parametric survival models or Cox proportional hazard models fitted to control subjects.
• Applied fitted models to treated subjects to simulate missing control outcomes for comprehensive survival analysis.

Main Results:

• Simulations demonstrated the repeated-sampling bias of the proposed methods.
• Nearest neighbor matching combined with the proposed method showed decreased bias compared to crude analyses.
• The method was illustrated using an example of beta-blocker prescription in heart failure patients.

Conclusions:

• The proposed method effectively combines propensity score matching and regression adjustment for survival data.
• This integrated approach offers improved accuracy in estimating survival curves and hazard ratios from observational studies.
• The method provides a valuable tool for epidemiological and clinical research involving time-to-event outcomes.