Standard and reference-based conditional mean imputation.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Longitudinal Data Analysis

Background:

• Clinical trials with longitudinal outcomes frequently encounter missing data due to missed assessments or intercurrent events.
• Current methods often use Bayesian random multiple imputation and Rubin's rules to address missing data and align analyses with targeted estimands.

Purpose of the Study:

• To propose and justify deterministic conditional mean imputation combined with the jackknife as a novel alternative for handling missing data in clinical trials.
• To demonstrate the method's applicability to both missing-at-random and reference-based imputation scenarios.

Main Methods:

• Deterministic conditional mean imputation coupled with jackknife for statistical inference.
• Application in a real-world clinical trial and a simulation study to evaluate performance.

Main Results:

• The proposed method yields consistent treatment effect estimates comparable to Bayesian approaches.
• It provides reliable frequentist inference, including accurate standard error estimation and controlled Type I error rates.
• The method is replicable and free from Monte Carlo error, unlike random sampling-based techniques.

Conclusions:

• Deterministic conditional mean imputation with the jackknife offers a robust and reproducible alternative for analyzing longitudinal clinical trial data with missing outcomes.
• This approach ensures valid statistical inference and consistent treatment effect estimation.