Incorporating time-dependent covariates in survival analysis using the LVAR method.

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• The Cox proportional hazards model is a cornerstone of survival analysis.
• Accurate covariate data is essential for reliable model performance.
• Observed data often requires inference for time-dependent covariates.

Purpose of the Study:

• To introduce and evaluate the Last Value Auto-Regressed (LVAR) estimation method.
• To compare LVAR against existing methods for inferring time-dependent covariates.
• To assess the impact of different time-dependent covariate processes on estimation accuracy.

Main Methods:

• Developed the Last Value Auto-Regressed (LVAR) estimation method.
• Conducted a simulation study to compare LVAR with established estimation techniques.
• Analyzed performance based on mean square error for time-dependent covariate effects.

Main Results:

• The LVAR method demonstrated a smaller mean square error compared to other approaches.
• This improvement was observed across various simulated time-dependent covariate processes.
• LVAR provides a more precise estimation of covariate effects over time.

Conclusions:

• The LVAR method is a valuable advancement for handling time-dependent covariates in survival analysis.
• It offers improved accuracy in estimating covariate effects, particularly within the Cox proportional hazards framework.
• This method enhances the reliability of survival models when dealing with dynamic covariate data.