Statistical Approaches for Establishing Appropriate Immunogenicity Assay Cut Points: Impact of Sample Distribution, Sample Size, and Outlier Removal

• Published in: The AAPS journal 2023-04, Vol.25 (3), p.37-37, Article 37
• Main Authors: Garlits, John, McAfee, Sean, Taylor, Jessica-Ann, Shum, Enoch, Yang, Qi, Nunez, Emily, Kameron, Kristina, Fenech, Keilah, Rodriguez, Jacqueline, Torri, Albert, Chen, Jihua, Sumner, Giane, Partridge, Michael A.
• Format: Article
• Language: English
• Subjects: Antibodies; Biochemistry; Biological products; Biomedical and Life Sciences; Biomedicine; Biotechnology; Comparative analysis; Humans; Pharmacology/Toxicology; Pharmacy; Research Article; Sample Size; Viral antibodies
• ISSN: 1550-7416; 1550-7416
• DOI: 10.1208/s12248-023-00806-5

The statistical assessments needed to establish anti-drug antibody (ADA) assay cut points (CPs) can be challenging for bioanalytical scientists. Poorly established CPs that are too high could potentially miss treatment emergent ADA or, when set too low, result in detection of responses that may have no clinical relevance. We evaluated 16 validation CP datasets generated with ADA assays at Regeneron’s bioanalytical laboratory and compared results obtained from different CP calculation tools. We systematically evaluated the impact of various factors on CP determination including biological and analytical variability, number of samples for capturing biological variability, outlier removal methods, and the use of parametric vs. non-parametric CP determination. In every study, biological factors were the major component of assay response variability, far outweighing the contribution from analytical variability. Non-parametric CP estimations resulted in screening positivity in drug-naïve samples closer to the targeted rate (5%) and were less impacted by skewness. Outlier removal using the boxplot method with an interquartile range (IQR) factor of 3.0 resulted in screening positivity close to the 5% targeted rate when applied to entire drug-naïve dataset. In silico analysis of CPs calculated using different sample sizes showed that using larger numbers of individuals resulted in CP estimates closer to the CP of the entire population, indicating a larger sample size (~ 150) for CP determination better represents the diversity of the study population. Finally, simpler CP calculations, such as the boxplot method performed in Excel, resulted in CPs similar to those determined using complex methods, such as random-effects ANOVA. Graphical Abstract