Background: The Clinical Pharmacology Quality Assurance (CPQA) program provides semiannual proficiency testing (PT) of antiretroviral analytes for 11 US and international clinical pharmacology laboratories (CPLs) to ensure interlaboratory comparability. In this article, we provide estimates of the main sources of variability and assess the accuracy of the algorithm for the assessment of performance.
Methods: Descriptive statistics are reported from 13 PT rounds from 2010 to 2016. Eight of the most common antiretroviral analytes were examined. Variance components analysis was used to rank the relative contributions of CPLs, antiretroviral analyte, and concentration category (low, medium, and high) to bias and variability using mixed models. Binary classification metrics of the PT assessment algorithm are calculated in comparison with a model using 95% prediction limits around estimated regression equations.
Results: CPLs provided 4109 reported concentrations of 65 unique samples for each of the 8 antiretroviral analytes across 13 PT rounds. Individual CPL accounted for the greatest amount of total variability (4.4%). Individual CPL and analyte combination (interaction) accounted for the greatest amount of bias (8.1%). Analyte alone accounted for 0.5% or less for total variability and bias. Overall, using a ±20% acceptance window around the final target, 97% of individual reported concentrations were scored acceptable, and 96% of antiretroviral/round scores were deemed satisfactory. Comparison with the regression model gave 100% sensitivity but only 34.47% specificity. Narrowing the acceptance window to ±15% improved specificity to 84.47% while maintaining a 99.17% sensitivity.
Conclusions: The current CPQA PT scoring algorithm that use a ±20% acceptance window seems to suffer from a low specificity and may be too lenient. A stricter ±15% acceptance window would increase specificity and overall accuracy while lowering the overall pass rate by only 3%.