Estimating propensity scores and causal survival functions using prevalent survival data

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Prevalent survival data collection can lead to incomplete covariate observation.
• This missingness is associated with failure times, biasing causal inference.
• Existing methods do not fully account for biases from prevalent sampling and potential outcomes.

Purpose of the Study:

• To develop semiparametric methods for estimating propensity scores and causal survival functions.
• To address bias arising from incomplete covariate data in prevalent survival studies.
• To provide accurate causal estimates in observational health data.

Main Methods:

• Semiparametric estimation of propensity scores using adjusted logistic regression.
• Development of causal survival function estimation accounting for two sources of missingness.
• Application to Surveillance, Epidemiology, and End Results (SEER)-Medicare data.

Main Results:

• Corrected propensity scores can be obtained with specific intercept adjustments in logistic regression.
• The proposed methods adjust for bias from both potential outcomes and prevalent sampling.
• Accurate causal survival estimation is achievable with the developed techniques.

Conclusions:

• The developed semiparametric approaches effectively handle missing covariate data in prevalent survival analysis.
• Adjusting for biases is crucial for valid causal inference in such settings.
• These methods enhance the reliability of causal estimates derived from observational health datasets.