Survival analysis using auxiliary variables via multiple imputation, with application to AIDS clinical trial data

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Censored observations in survival analysis limit data utility.
• Auxiliary variables can potentially recover information from censored data.
• Accurate statistical methods are crucial for analyzing clinical trial outcomes.

Purpose of the Study:

• To develop and evaluate a multiple imputation approach for utilizing auxiliary variables in survival analysis.
• To apply this method to AIDS clinical trial data using CD4 counts as a time-dependent auxiliary variable.
• To assess the impact of this imputation technique on treatment effect estimation and statistical efficiency.

Main Methods:

• Developed a joint model incorporating a hierarchical change-point model for CD4 counts and a time-dependent proportional hazards model for AIDS events.
• Employed Markov chain Monte Carlo (MCMC) methods for multiple imputation of event times for censored observations.
• Combined results from imputed datasets using standard multiple imputation techniques.

Main Results:

• Multiple imputation resulted in minor changes to the estimated effect of Zidovudine (ZDV) but yielded smaller standard errors compared to analyzing observed data alone.
• Sensitivity analyses confirmed the robustness of qualitative findings across various imputation models.
• Simulation studies indicated improved efficiency and partial correction for dependent censoring.

Conclusions:

• The proposed multiple imputation method effectively utilizes auxiliary variables to enhance survival analysis with censored data.
• The approach offers improved statistical efficiency and robustness in clinical trial settings.
• Further research is needed to address the compatibility between the primary analysis and the imputation model.