Evaluating the Practice of Assuming Parallelism in Relative Potency Determination with Four-Parameter Logistic Regression

• Activity Number: 129
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Quality and Productivity Section
• Author(s): Liping Song* and Robert Capen
• Companies: Merck and Merck Research Laboratories
• Keywords: four parameter logistic regression ; parallelism ; relative potency ; similarity ; assay variability ; weighting function

Abstract

Four-parameter logistic regression (4-PL) is a common statistical model used in bioassay to determine the potency of a sample relative to the reference standard. Due to both the natural characteristics of biological substances and the complexity of biological reactions, it is normally required to confirm similarity or parallelism of the dilution response curves before calculating the relative potency. However, sometimes the laboratory may omit the confirmation step, since confirming parallelism is not straightforward for 4-PL models. In this presentation, we evaluate the impact of parallelism evaluation on the bias and variability of relative potency estimates with different assay variability and weighting using simulated data and real data. The results show that at the early stage of assay development it is strongly recommended to evaluate parallelism with a proper weighting function. But once the assay system is well-controlled, there is little impact to the bias or variability in the relative potency estimates if parallelism is simply assumed even for approximately parallel curves, partial curves, or even curves with hooked effects.