A Non-Normal Bayesian Model for the Estimation and Comparison of Immunogenicity Screening Assay Cut-Points

Anti-drug antibodies (ADAs) elicited by potentially immunogenic therapeutic agents represent a serious safety concern. Monitoring of subject ADA levels has become an integral part of clinical programs. Monitoring typically follows a tiered approach that begins with the initial screening of subject serum samples. Screening assays employ a cut-point (CP) to classify samples as ADA positive or negative. Regulators expect screening assay CPs to be set to assure a 5% false positive rate. Traditional approaches to CP determination use common statistical tools that assume normality. Because normality is often rejected, even after log transformation and deletion of extreme values, less powerful non-parametric methods are often substituted. This chapter motivates and describes the use of a hierarchical exponential Gaussian distribution model, which accommodates non-normality and thus, in principle, is more powerful and less dependent on the removal of extreme values. Bayesian implementation of this model is straightforward and provides an informative probability-based approach for CP estimation and CP comparison across studies.