Missing data in clinical trials: control-based mean imputation and sensitivity analysis.

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Pharmaceutical Research

Background:

• Randomized trials often assess treatment effects on endpoints unaffected by rescue medication.
• Missing data due to treatment discontinuation or rescue medication use complicates analysis.
• Standard methods like mixed-effects models may yield biased results by overestimating treatment benefits.

Purpose of the Study:

• To propose and evaluate alternative methods for handling missing data in clinical trials.
• To provide a more accurate estimation of the between-treatment difference in population means.
• To address the limitations of standard missing data assumptions in drug versus placebo studies.

Main Methods:

• Proposed an alternative approach replacing missing means with placebo means or conservative estimates.
• Introduced a "dropout equals failure" analysis using quantile regression.
• Utilized simulation results and three real-world examples for validation.
• All methods adjust for baseline covariates and target the same estimand.

Main Results:

• The commonly used mixed-effects model repeated measures analysis can overestimate treatment effects.
• The proposed methods offer more conservative and potentially less biased estimates of the true treatment effect.
• Sensitivity analyses using tipping point frameworks provide robust results.

Conclusions:

• Standard missing data handling can lead to exaggerated treatment effect estimates.
• Alternative imputation strategies, including placebo means or "dropout equals failure" approaches, are recommended.
• These methods improve the reliability of treatment effect estimation in clinical trials with missing data.