Comparison of simulated cumulative drug versus time data sets with indices

The objectives of this study were twofold. First, to clarify the applicability of the similarity factor, f(2), the difference factor, f(1), and the Rescigno index, xi(i), in the comparison of cumulative drug vs. time data sets. Second, to assess the possibility for these indices to be used on a confidence interval basis. Theoretical profiles as well as errorless and errant cumulative % data sets were simulated using the Weibull function. At various variability levels, 12-fold and 3-fold replicated reference and test data sets were generated from the errant data sets. The 90% confidence intervals were constructed for the median of the index (non-parametric confidence intervals, NPCIs) and for the evaluated index based on the 5th and 95th percentiles of 1000 index values estimated from bootstrapped samples (bootstrapped confidence intervals, BSCIs). It was observed that at low variability levels, i.e. when mean data sets can be used, all indices could be used for the comparison of cumulative data sets. At high variability levels, only the BSCIs of f(2) included the actual index value for the 12-fold replicated data sets. However, deviations of the low limits of NPCIs of f(2) from the actual index values were similar to corresponding deviations of BSCIs. When 3-fold replicated data sets were used, both NPCIs and BSCIs of all indices were generally reliable but much larger than that of 12 replications. In conclusion, the time period for the evaluation of the indices cannot be theoretically justified because indices change continuously with time. Cutoff points for their evaluation must be decided on a case-by-case basis. If theoretically possible, evaluation of the indices should be done by using areas. With highly variable data, BSCIs are preferred because compared to NPCIs they are less dependent on the number of replications. When n=12, BSCIs of f(2) are comparatively more reliable. When n=3, BSCIs of the indices tested in this study have similar performances.