Adjusting for time-varying confounders in survival analysis using structural nested cumulative survival time models

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Assessing causal effects of time-varying exposures on survival is complex due to time-varying confounding.
• Standard survival methods and existing models like structural nested accelerated failure time models (AFTMs) have limitations, including biased estimates, instability, and estimation difficulties.

Purpose of the Study:

• To introduce the structural nested cumulative survival time model (SNCSTM) for improved causal effect estimation.
• To provide efficient and robust estimators for the SNCSTM that overcome limitations of existing methods.

Main Methods:

• Introduction of the structural nested cumulative survival time model (SNCSTM).
• Development of three estimators for the SNCSTM, including two efficient, double robust, closed-form estimators.
• Fitting the SNCSTM using standard generalized linear model software.

Main Results:

• The proposed SNCSTM estimators avoid the artificial recensoring issues of AFTMs.
• The estimators are more stable than those relying on inverse probability of exposure weighting.
• A simulation study and real-world data from the UK Cystic Fibrosis Registry demonstrate the model's performance.

Conclusions:

• The SNCSTM provides a robust and efficient approach to analyzing time-varying exposures and confounders in survival analysis.
• The SNCSTM offers advantages over existing structural nested cumulative failure time models.
• The developed estimators are practical for use with standard statistical software.