Impacts of Random and Fixed Effect Models on Type I and Type II Errors in Bioequivalence Hypotheses in Crossover Trial Designs: A Comprehensive Simulation Study

Bioequivalence (BE) is the fundamental regulatory requirement for the approval of generic drug products. Despite global harmonization efforts, such as the ICH M13A guideline, subtle differences in statistical modeling—specifically the treatment of “subject” as a Fixed vs. Random effect—persist between the EMA and US-FDA frameworks.

This study provides a comprehensive evaluation of these two statistical paradigms on Type I Error (TIE) and Statistical Power (Type II Error) across varying sample sizes, intra-subject variability (CV), and inter-period correlations.

1.89 million Monte Carlo simulations were conducted for a standard 2-sequence, 2-period (2 × 2) crossover design. Parameters included: sample size (n = 12, 24, 32), intra-subject CV (15%, 25%, 30%), inter-period correlation ($r$: 0.30, 0.60, 0.90), and geometric mean ratios (GMR: 1.00–1.10). Data were generated using MNORMAL software assuming log-normal distribution. Analysis followed EMA (Fixed) and FDA (Random) ANOVA specifications.

Both models effectively controlled Type I error within the nominal 5% level; however, statistical power was significantly higher under the FDA Random Effect model than under the EMA Fixed Effect model in borderline scenarios (GMR 1.03–1.05). Notably, increasing inter-period correlation substantially amplified statistical power, with effects comparable to doubling the sample size, underscoring its critical role in crossover bioequivalence study design.

Understanding these model discrepancies is vital for global drug development in the ICH M13A (2025) era. The study highlights that the FDA’s model is slightly more permissive, reducing manufacturer risk without compromising patient safety. Accounting for inter-period correlation in power calculations is highly recommended for efficient trial design.