A cautionary tale: dealing with missing data in clinical trials for rheumatic diseases

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Data Science

Background:

• Missing data are prevalent in clinical trials, necessitating robust statistical methods for accurate analysis.
• Inappropriate handling of missing data can lead to biased results and flawed conclusions.

Purpose of the Study:

• To evaluate various statistical techniques for addressing missing data in clinical trials.
• To provide guidance on interpreting conflicting results from different missing data analysis methods.

Main Methods:

• Application of eleven distinct statistical imputation methods to a placebo-controlled randomized trial in diffuse cutaneous systemic sclerosis.
• Assessment of methods including complete case analysis, available case analysis, last observation carried forward (LOCF), multiple imputation, and mixed-effects models.
• Exploration of a joint likelihood-based model for handling data not missing at random.

Main Results:

• Complete case analysis and LOCF methods rely on assumptions often unmet in practice.
• Multiple imputation and mixed-effects models attempt to account for patient variability but may not fully address missingness mechanisms.
• The joint likelihood model offers a more realistic approach by explicitly handling data not missing at random, potentially reducing bias.

Conclusions:

• Various statistical approaches exist for handling missing data in clinical trials, each with strengths and limitations.
• When analyses yield conflicting results, careful consideration of the underlying assumptions and model appropriateness is crucial.
• Recommendations are provided for navigating discrepancies and reaching reliable conclusions from clinical trial data with missing values.