Methods for Analysis of Pre-Post Data in Clinical Research: A Comparison of Five Common Methods

Area of Science:

• Biostatistics
• Statistical Modeling
• Clinical Trial Analysis

Background:

• Repeated measures data are frequently summarized into pre-post-treatment measurements for analysis.
• Existing statistical methods for estimating treatment effects include ANOVA, ANCOVA, and LMM.
• Different modeling approaches exist, such as using post-treatment scores or change scores as outcomes.

Purpose of the Study:

• To compare the effectiveness of five common statistical methods for analyzing pre-post-treatment data.
• To evaluate methods based on simulation studies and theoretical variance derivations.
• To identify the most precise and powerful method for estimating treatment effects in repeated measures designs.

Main Methods:

• Analysis of Variance (ANOVA) using post-treatment scores or change scores.
• Analysis of Covariance (ANCOVA) using post-treatment scores or change scores.
• Linear Mixed Modeling (LMM), with and without Kenward-Rogers adjustment.

Main Results:

• All five methods produced unbiased treatment effect estimates.
• ANCOVA modeling, whether using change scores or post-treatment scores, demonstrated superior precision, 95% coverage probability, and statistical power.
• Simulations and theoretical derivations supported the effectiveness of ANCOVA methods.

Conclusions:

• ANCOVA modeling of either change scores or post-treatment scores is the most effective approach for analyzing pre-post-treatment data.
• These ANCOVA methods provide the best balance of precision, coverage, and power for treatment effect estimation.
• The study illustrates these comparisons using a real-world clinical data example.