Joint modeling of outcome, observation time, and missingness

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Modeling

Background:

• Clinical trials often feature single post-randomization measurements that can be missing or taken before trial completion.
• Existing methods like last observation carried forward (LOCF) may not adequately account for the timing of measurements and missing data processes.

Purpose of the Study:

• To develop and evaluate a novel statistical modeling framework for clinical trials with single, time-to-event, and missing response data.
• To compare the performance of the proposed model against the conventional LOCF approach.

Main Methods:

• A latent variable structure within a multivariate Gaussian distribution is employed to simultaneously model the outcome, time to measurement, and missingness.
• A simulation study assesses the behavior of maximum likelihood estimators.
• The model is illustrated using data from a congestive heart failure treatment trial.

Main Results:

• The proposed model provides a framework for analyzing complex missing data scenarios in clinical trials.
• Simulation results indicate the behavior of the maximum likelihood estimators.
• Comparison with LOCF highlights potential differences in parameter estimation and inference.

Conclusions:

• The developed modeling framework offers a statistically rigorous approach for handling single, potentially missing or early, response measurements in clinical trials.
• This method provides a more nuanced analysis compared to simpler imputation techniques like LOCF.
• The approach is applicable to real-world clinical trial data, as demonstrated in a heart failure study.