(T1530-02-08) Cross Validation of Pharmacokinetic Assays: Experimental and Statistical Design

Purpose: The purpose is to present the strategy to cross validate two different pharmacokinetic (PK) assays that measure the same drug.

Methods: Two case studies are presented. In the first case study, the equivalency of results between 2 labs was assessed. A bioanalytical method was initially validated at a laboratory in the U.S. (BioA Lab 1) and, subsequently, transferred and validated at an ex-US laboratory (BioA Lab 2). Both laboratories were utilized to measure clinical PK serum samples from the same global trial. In the second study, the equivalency of results from a multiplexing immunoaffinity (IA) liquid chromatography tandem mass spectrometry (IA LC-MS/MS) assay and a previous/historical enzyme-linked immunosorbent assay (ELISA) method were assessed. One hundred incurred samples (within informed consent) should be selected and used for the PK assay cross validation. The incurred study samples over the applicable range of concentrations (≥ 3x LLOQ – Cmax) are selected using stratified sampling with a disproportionate allocation scheme based on the four quartiles (Q) of in-study concentration levels. Thirty samples with concentrations in Q1 (lowest concentrations), 20 samples with concentrations in Q2, 20 samples with concentrations in Q3, and 30 samples with concentrations in Q4 (highest concentrations) are sampled. The samples are assayed once in the two assay methods. The two methods are considered equivalent if the percent differences in the lower and upper bound limits of the 90% confidence interval (CI) are both within ±30%. If the criterion is not met, the two methods are determined to not be equivalent. A Bland-Altman plot of the percent difference of the sample concentrations versus the mean concentration of each sample is created to help investigate the existence of any concentration dependent biases between the measurements. Furthermore, the 90% CI of the slope of the concentration percent difference–mean concentration curve is evaluated to further identify a concentration dependent bias. If “0” is contained within the 90% CI of the slope, there is no bias. If “0” is not contained within the 90% CI of the slope, re-evaluate each quartile (Q1, Q2 and Q3 [combined], and Q4) according to the first criterion to determine if a concentration dependent bias exists. If the 90% CI for the lower and upper bound limits of the percent differences for each quartile are within ±30%, there is no concentration dependent bias.

Results: In the first case study, the mean percent difference of the sample set (n = 100) was -3.82%, with 90% CI lower and upper bound limits of -5.89% and -1.76%, respectively. The percent differences of the lower and upper bound limits of the 90% CI fell within the pre-specified criteria of ±30%. The slope of the percent difference mean concentration curve was determined to be -0.00001 (90% CI lower and upper bound limits of -0.00005 and 0.00003, respectively). Zero was contained within the 90% CI lower and upper bound limits of the slope; therefore, no concentration dependent bias was observed. In the second case study, the mean percent difference of the sample set (n = 100) was -10.9%, with 90% CI lower and upper bound limits of -14.0% and -7.70%, respectively. The percent differences of the lower and upper bound limits of the 90% CI were contained within the criteria of ±30%. The slope of the percent difference mean concentration curve was determined to be -0.00012, however, the 90% CI lower and upper bound limits were -0.00017 and -0.00007, respectively. Zero was not contained within the 90% CI lower and upper bound limits of the slope; therefore, the samples were again split into quartiles by mean concentration and the percent differences by quartile were evaluated for potential concentration dependent bias. The mean percent difference by quartile was 3.82% (90% CI lower and upper bound -3.05% and 10.7%, respectively) for Quartile 1, -17.9% (90% CI lower and upper bound -21.2% and -14.6%, respectively) for Quartile 2 and 3, and -16.1% (90% CI lower and upper bound -20.6% and -11.6%, respectively) for Quartile 4. The 90% CI for the lower and upper bound of the % differences for all quartiles fell within the criteria of ±30%. Although, based on the quartile analysis, the mid and high concentrations still quantified lower in the IA LC-MS/MS; there is no concentration dependent bias at a level that would impact PK.

Conclusion: The cross validation experiments comparing the two different labs, the U.S. Lab and ex-US lab, and the two different assay formats, an ELISA and an IA LC-MS/MS, using incurred samples, demonstrated that the two labs or two methods are equivalent. The cross validation strategy allows for a robust assessment of assay comparability. The strategy also allows for subgroup analyses of concentration levels and potentially other variables that may not be otherwise considered.