Comparison of dissolution profiles: 90 % confidence intervals of different f₂ estimators using bootstrap methodology...

The most widely used method to compare dissolution profiles is the similarity factor f₂ method. When the regulatory requirements to apply the f₂ method are not fulfilled, alternative methods should be used. In the current study two commonly used methods, 90% confidence intervals (CI) of different f₂ estimators using bootstrap methodology and the Euclidean Distance of the Non-standardized Expected (EDNE) values, are compared using two different simulation approaches. For the first approach, the reference and test population profiles were simulated based on the multivariate normal distribution with different target population f₂ values, variability, and sample sizes. For each pair of randomly simulated profiles, 90% CI of various f₂ estimators and the EDNE values were calculated. For the second approach, the first-order release model-based simulation, one million individual dissolution profiles were simulated for the reference and test populations with different variability and predefined target population f₂ values, random samples of different sizes were taken from those populations to obtain 90% CI of the same f₂ estimators and the EDNE values. The whole process was repeated 10000 times for both approaches to evaluate the type I error and statistical power of the methods by calculating the percentages of replicates where the dissolution profiles are similar. When the true populations of test and reference profiles are not similar, this percentage of similarity represents the type I error; when the true populations of test and reference profiles are similar, this percentage represents the statistical power. The results shows that the EDNE method has much higher statistical power than the bootstrap f₂ methods, but the associated type I errors are also unacceptably higher, making it unsuitable for regulatory adoption. The best method is the 90% CI of the expected f₂, therefore, this method is recommended. In addition, sample sizes should be increased to account for the low statistical power when using bootstrap methods.