Inflation of Type I Error in the Evaluation of Scaled Average Bioequivalence, and a Method for its Control.

Abstract

To verify previously reported findings for the European Medicines Agency's method for Average Bioequivalence with Expanding Limits (ABEL) for assessing highly variable drugs and to extend the assessment for other replicate designs in a wide range of sample sizes and CVs. To explore the properties of a new modified method which maintains the consumer risk ≤0.05 in all cases. Monte-Carlo simulations of three different replicate designs covering a wide range of sample sizes and intra-subject variabilities were performed. At the switching variability of CV wR 30% the consumer risk is substantially inflated to up to 9.2%, which translates into a relative increase of up to 84%. The critical region of inflated type I errors ranges approximately from CV wR 25 up to 45%. The proposed method of iteratively adjusting α maintains the consumer risk at the desired level of ≤5% independent from design, variability, and sample size. Applying the European Medicines Agency's ABEL method at the nominal level of 0.05 inflates the type I error to an unacceptable degree, especially close to a CV wR of 30%. To control the type I error nominal levels ≤0.05 should be employed. Iteratively adjusting α is suggested to find optimal levels of the test.