Online Program

Return to main conference page
Wednesday, September 25
Wed, Sep 25, 1:15 PM - 2:30 PM
Thurgood Marshall South
Immunogenicity Assay Cut Point Determination

Statistical concerns for cut point determination (301048)

*Meiyu Shen, Office of Biostatistics, Center for Drug Evaluation and Research, FDA 

Anti-drug antibodies (ADA) may impact the safety and effectiveness of therapeutic proteins, peptides, and oligonucleotides. Therefore, FDA requests that drug developers test for the presence of ADA during clinical trials. ADA assays are usually run as threshold assays, with the threshold, called the cut point, statistically calculated using results of biological samples obtained from treatment naïve subjects. The first step in assessing for ADA is to screen subjects before, during, and after treatment in the clinical trial using a high throughput screening assay. Samples that screen positive then undergo confirmatory testing, and if confirmed positive may undergo additional characterization testing. The cut point of the screening assay is the level of response at or above which a sample is defined to be positive. The Food and Drug Administration Guidance for Industry on Immunogenicity Testing of Therapeutic Protein Products —Developing and Validating Assays for Anti-Drug Antibody Detection recommends the cut point to be an upper 95 percentile of the results from 50 individual treatment-naïve subjects’ samples. These samples along with negative and positive controls are analyzed by 2-3 analysts performing 2-3 assay runs each, generally resulting in approximately 300 observations. Various experimental designs are extensively discussed in the literature. Methods using analysis of variance and a mixed effect model are widely used to determine the effects of analyst, run, and plate on the response. A commonly used approach to calculating the cut point treats all observations as independent, but this approach is flawed because there are replicate observations for subjects and analysts. Another way to calculate the cut point is to average the results from each donor across runs, but this approach loses some information on inter-assay variability. Furthermore, the cut point determined by a sample quantile does not consider the variability of the sample quantile. Therefore, the sample percentile does not assure at least 5% false-positive rate (FPR) with a high confidence level (e.g., 90%) when the sample size is not sufficiently enough. With these concerns, we discuss the use a lower confidence limit for a percentile of the repeated measurement data as the cut point instead. We propose to calculate the cut point from repeated measurement data using a random effect model which captures variability of inter-assay and inter-subject, and measurement errors. In this talk, we will discuss appropriate use of negative control data, pre-processing of in-study samples data (normalized by negative control or not), and a statistical method for calculating the cut point from the repeated measurement data. We will compare these methods for the immunogenicity screening assay cut-point determination.