Causal estimation using semiparametric transformation models under prevalent sampling.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Prevalent survival data presents unique challenges, including prevalent sampling bias and systematic imbalances in observational data.
• Standard causal estimation methods may yield biased results when applied to such complex datasets.

Purpose of the Study:

• To develop and present methods for causal estimation in semiparametric transformation models tailored for prevalent survival data.
• To propose analytical procedures for estimating the causal survival function while addressing prevalent sampling bias and observational data imbalances.

Main Methods:

• Estimation of semiparametric transformation models and covariate distributions.
• Development of a unified approach to simultaneously correct for prevalent sampling bias and balance systematic differences in observational data.
• Utilizing empirical process techniques to establish large sample properties of the proposed estimation procedures.

Main Results:

• Simulation studies demonstrate that standard analyses without proper adjustment lead to biased causal inference.
• The proposed unified approach effectively corrects for prevalent sampling bias and balances systematic differences.
• The methods are validated through simulation studies and applied to real-world cancer registry data.

Conclusions:

• The proposed methods provide a robust framework for accurate causal survival function estimation in prevalent data.
• Addressing both prevalent sampling bias and observational imbalances is critical for reliable causal inference in epidemiological studies.
• The application to SEER and Medicare data highlights the practical utility of these methods for breast cancer research.