Non-Normal Random Effects Models for Immunogenicity Assay Cut Point Determination

Modeling immunogenecity data to establish screening bioassays cut point

The response of immunogenecity anti-drug antibody (ADA) generally includes biological and analytical variability. The nature of biological and analytical variations may lead to a variety of symmetric and asymmetric ADA data. As a result, current statistical methods may yield unreliable results because these methods assume special types of symmetric or asymmetric ADA data. In this paper, we survey and compare parametric models that are useful for analyzing a variety of asymmetric data that have rarely been used to calculate assay cut points. These models include symmetric distributions as limiting case; therefore, they are useful in the analysis of a variety of symmetric data. We also investigate two nonparametric approaches that have received little attention in screening cut point calculations. A simulation study was conducted to compare the performance of the methods. We evaluate the methods using four published different types of data, and make recommendations concerning the use of the methods.

Method validation of a bridging immunoassay in combination with acid-dissociation and bead treatment for detection of anti-drug antibody

Anti-drug antibody (ADA) positivity is correlated with disease relapse risk when treated with monoclonal antibody (mAb) therapeutics. ADA evaluation can assist with interpreting pharmacokinetic, pharmacological, and toxicology results. Here, we established an ADA assay based on two steps of acid dissociation combined with a bridging immunoassay to provide a comprehensive validation strategy. The three-tiered sample analysis process included screening, confirmation, and titration assays using therapeutic HLX26 (targeting lymphocyte activation gene-3 [LAG-3]) as an example. The cut points were determined by testing 50 individual normal human serum samples, including screening cut point (SCP) (SNR: 1.08), confirmatory cut point (CCP) (% inhibition: 12.65), and titration cut point (TCP) (sample-to-noise ratio [SNR]: 1.17). The assay sensitivity, low positive control (LPC), and high positive control (HPC) titer acceptable range were also set up as 33.0 ng/mL, 41.0 ng/mL, and 320-1280, respectively. After full validation, both the intra-assay and inter-assay precision testing passed with coefficient of variations (CVs) < 20%. The assay enabled excellent drug tolerance up to 768.0 μg/mL at the HPC level and 291.0 μg/mL at the LPC level, while the tolerance of target interference was up to 74.0 ng/mL of soluble LAG3. Moreover, no false-positive results were observed in the presence of 5% hemolyzed serum samples and 150 mg/dL of triglyceride in the serum samples, no hook effect was observed, and the stability performed normally under room temperature for 24 h, 2-8 °C for 7 d, and six freeze/thaw cycles. In summary, this ADA assay is feasible and could be used for evaluating the immunogenicity of HLX26 in clinical trials.

Methods and applications of percentile estimation

Percentile is ubiquitous in statistics and plays a significant role in the day-to-day statistical application. FDA Guidance for Industry: Assay Development for Immunogenicity Testing of Therapeutic Protein Products (2016) recommends the use of a lower confidence limit of the percentile of the negative subject population as the cut point to guarantee a pre-specified false-positive rate with high confidence. Shen proposed and compared an exact t approach with some approximated approaches. However, the exact t approach might be compromised by computational time and complexity. In this article, we proposed to use a UMOVER method as a potential alternative for percentile estimation for one application to screening and confirmatory cut point determination due to its easy implementation and similar performance to the exact t approach. The applications and performance comparison with different approaches are investigated and discussed. Furthermore, we extended the proposed method for the comparison of the percentile of the test product and percentile of the reference product followed by numerical studies.

Elucidation of the statistical factors that influence anti-drug antibody cut point setting through a multi-laboratory study

Aim: Appropriateness of anti-drug antibody (ADA) assay is critical for immunogenicity assessment of biopharmaceuticals. Although cut point setting in ADA assay has a large impact on the results, a standard statistical approach for its setting has not been well established. Methodology: In this multi-laboratory study, to elucidate factors influencing the cut point setting, we compared the statistical approaches and calculated cut points for multiple datasets of ADA assays using the individual procedure employed at each laboratory. Conclusion: We showed that outlier exclusion, false-positive rate and investigating data distribution have the greatest impact on both screening and confirmatory cut points. Our results would be useful for industry researchers and regulators engaged in immunogenicity assessment of biopharmaceuticals.

Immunogenicity assay cut point determination using nonparametric tolerance limit

The newly released FDA guidance on immunogenicity assay development and validation recommends use of a lower confidence limit of the percentile of the negative subject population as the cut point in order to guarantee a pre-specified false positive rate with high confidence. The limit is, in essence, a lower tolerance limit. Although in literature several methods are available for determining the tolerance limit, they either fail to take into account the repeated measurement of the data from a typical immunogenicity assay quantification/validation experiment or rely heavily on normality assumption of the data, which is rarely correct. As a result, the methods may result in biased estimates of the cut point, causing the false positive rate to be either lower or higher than expected. To overcome this drawback, we propose two non-parametric methods under repeated measure data structure and without normal distribution assumption. Simulation studies were carried to compare the performance of the two non-parametric approaches with the current methods. The results of the simulation studies show that one of the two nonparametric methods outperforms all the other methods and provides a satisfactory coverage probability even with moderate sample sizes. In addition, it is simple and straightforward to implement. Therefore, it is a preferred method for immunogenicity assay cut point determination.