Novel Scaled Bioequivalence Limits with Leveling-off Properties

Abstract

(1) To develop novel scaled bioequivalence (BE) limits with levelling-off properties based solely on variability considerations and (2) to evaluate their performance in comparison to the classic unscaled BE limits 0.80-1.25, the expanded BE limits 0.75-1.33 and the recently proposed Geometric Mean Ratio (GMR)-dependent scaled BE limits BELscW (Karalis et al., Eur. J. Pharm. Sci., 26:54-61, 2005). Two model functions were used to ensure the gradual change of the BE limits from a starting value towards a predefined plateau value. Plots of the new BE limits and extreme GMR values ensuring BE as a function of the coefficient of variation (CV) were constructed. Two-period crossover BE studies with 12, 24, or 36 subjects were simulated assuming CV values from 10 to 60%. Power curves were constructed by recording the percentage of accepted BE studies as the true GMR was raised from 1.00 to 1.50. The percentage of the true GMR within the simulated BE limits vs. true GMR was used to evaluate the estimation accuracy of the scaled methods. Depending on the parameters' values of the model functions, the scaled BE limits exhibit different performance. Four new scaled BE limits, showing favourable performance for the evaluation of average BE are presented. At low variability levels two of the novel BE limits show similar performance to the 0.80-1.25 criterion, while the other two (as expected from their design) appear to be less permissive. At high CV values (30, 40%) all new BE limits exhibit much higher statistical power than the 0.80-1.25 criterion. They show almost identical behavior with the expanded 0.75-1.33 limits and appear to be less permissive than BELscW. Finally, the percentage of the true GMR within the simulated BE limits vs. true GMR shows a sharp decline. Due to the absence of the GMR factor in the model functions a more accurate estimation of the new scaled BE limits, compared to BELscW, is observed. The new scaled BE limits appear to be highly effective at all levels of variation investigated and present satisfactory estimation accuracy.