Experimental and statistical approaches in method cross-validation to support pharmacokinetic decisions

A case study of experimental and statistical approaches for cross-validating and examining the equivalence of two ligand binding assay (LBA) methods that were employed in pharmacokinetic (PK) studies is presented. The impact of changes in methodology based on the intended use of the methods was assessed. The cross-validation processes included an experimental plan, sample size selection, and statistical analysis with a predefined criterion of method equivalence. The two methods were deemed equivalent if the ratio of mean concentration fell within the 90% confidence interval (0.80-1.25). Statistical consideration of method imprecision was used to choose the number of incurred samples (collected from study animals) and conformance samples (spiked controls) for equivalence tests. The difference of log-transformed mean concentration and the 90% confidence interval for two methods were computed using analysis of variance. The mean concentration ratios of the two methods for the incurred and spiked conformance samples were 1.63 and 1.57, respectively. The 90% confidence limit was 1.55-1.72 for the incurred samples and 1.54-1.60 for the spiked conformance samples; therefore, the 90% confidence interval was not contained within the (0.80-1.25) equivalence interval. When the PK parameters of two studies using each of these two methods were compared, we determined that the therapeutic exposure, AUC(0-168) and Cmax, from Study A/Method 1 was approximately twice that of Study B/Method 2. We concluded that the two methods were not statistically equivalent and that the magnitude of the difference was reflected in the PK parameters in the studies using each method. This paper demonstrates the need for method cross-validation whenever there is a switch in bioanalytical methods, statistical approaches in designing the cross-validation experiments and assessing results, or interpretation of the impact of PK data.