Handling non-ignorable missing intimal hyperplasia data - Lessons from the VEST trial.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Cardiovascular Surgery Research

Background:

• Non-ignorable missing data is a common challenge in clinical trials, particularly in studies like the VEST trial assessing saphenous vein graft outcomes.
• Graft occlusion, leading to missing intimal hyperplasia (IH) data, is a negative outcome, making the missing data non-ignorable (missing not at random - MNAR).

Purpose of the Study:

• To develop and evaluate a novel statistical method to address MNAR data in the VEST trial.
• To assess the performance of this new method against existing approaches through simulation studies.

Main Methods:

• A two-part method combining penalized multiple imputation with a modified Wilcoxon signed-rank test was developed for MNAR scenarios.
• Simulation studies were conducted to evaluate the method's power and Type I error rate compared to alternatives.

Main Results:

• The novel method demonstrated no Type I error inflation and adequate power under trial assumptions.
• Power decreased with missing data exceeding 20% using double penalization due to treatment effect underestimation.
• Penalized multiple imputation alone was more powerful for balanced missing data, while the combined method was superior for unbalanced MNAR data.

Conclusions:

• The developed method is suitable for handling non-ignorable missing data in clinical trials, particularly in cardiovascular studies.
• The approach is robust even when missing data is balanced across treatment arms, as observed in the VEST trial.
• This methodology offers a valuable tool for other research settings facing similar missing data challenges.