B-095 Determination of Error Rates with Duplicate Analysis in Anti-Drug Antibody Testing

Abstract Background Duplicate measurements can be used to characterize assay imprecision, identify outliers, and estimate error rates. The benefits of duplicate analysis, however, need to be balanced against the increased financial cost. The objective of this study was to evaluate the need for continued replicate measurement in a laboratory developed assay measuring anti-drug antibodies to infliximab (ATI). To this end, duplicate patient sample measurements from the ATI assay over a two-year period were analyzed to calculate outlier and error frequencies. Methods ATI were measured by electrochemiluminescence (ECL) on the MesoScale Discovery (MSD) platform, with a standard curve from 20–1250 U/mL using commercially available monoclonal ATIs. All patient samples tested for ATI over a 27-month period of time (n = 9582) were subject to inclusion in this study. Only patient samples with ATI levels within the analytical measuring range (AMR) of the assay, however, were selected for further analysis. Samples were divided equally into 7 bins across the measuring range based on mean replicate concentration. Within-run imprecision was determined for each bin using the standard deviation (combined from all replicates using sum of squares) and mean concentration derived from all replicates. Outliers were defined as replicate measurements with a percent difference greater than ±3 SD of the calculated assay imprecision. Errors were defined as samples which had replicate values on opposite sides of the clinical cutoff of positive ATI (≥50 U/mL) and a percent difference greater than 3SD of the assay imprecision. Results Of the 9582 total patient samples analyzed, 2467 had ATI values within the AMR and 1655 had levels over the clinical cut-off of 50 U/mL (17.3% positivity rate). Within-run imprecision determined from the mean and combined SD of the duplicate measurements ranged from 15.7% at the lowest concentration bin (20–27.5 U/mL) to 10.0% at the highest concentration bin (358–1243 U/mL). The mean percentage of outliers in each bin ranged from 0.8%–3.4% resulting in an overall outlier frequency of 2.1% when considering all samples within the AMR. There were 66 samples within the AMR which demonstrated qualitative discrepancies between the replicates. However, only 21 (31.8%) of these samples, had a percent agreement greater than 3SD of the assay imprecision resulting in an overall error rate of 0.2% when considering all 9582 samples. Conclusion In this study, the analy