Statistical recommendations for count, binary, and ordinal data in rare disease cross-over trials.

Area of Science:

• Biostatistics
• Clinical Trials
• Rare Diseases

Background:

• Statistical method recommendations for rare disease trials, particularly crossover designs, are limited.
• An international consortium of statisticians compared state-of-the-art methodologies using an illustrative dataset from epidermolysis bullosa research.
• The comparison focused on count, binary, and ordinal outcome variables to inform recommendations.

Purpose of the Study:

• To compare various statistical methodologies for rare disease crossover trials.
• To provide evidence-based recommendations for selecting appropriate statistical methods based on outcome variable type and study goals.
• To identify methods that optimize statistical power and account for specific design features like period and carry-over effects.

Main Methods:

• Parametric methods (model averaging).
• Semiparametric methods (generalized estimating equations type [GEE-like]).
• Nonparametric methods (generalized pairwise comparisons [GPC] and a marginal model via nparLD).

Main Results:

• No single method uniformly outperformed others across all outcome types.
• The prioritized unmatched GPC method demonstrated high power, particularly for prioritizing clinically relevant time points.
• Model averaging and GEE-like methods showed favorable results for binary outcomes and properly accounted for period/carry-over effects.
• The nonparametric marginal model achieved high power for ordinal outcomes, even with small sample sizes, and accommodated longitudinal/interaction effects.

Conclusions:

• The optimal statistical method depends on balancing statistical power, accounting for crossover/period/carry-over effects, and prioritizing clinically relevant time points.
• Specific methods like GPC, model averaging, GEE-like, and nonparametric marginal models offer distinct advantages in different rare disease trial scenarios.
• Further research and clear guidelines are needed to support robust statistical analysis in rare disease crossover trials.