Statistical Quality Control of Radioimmunoassays¹

TLDR

RIA and CPB assays generate dose‑response curves with severe nonlinearity and non‑uniform, non‑normal residual variance, rendering classic least‑squares methods unsuitable and making logit linearization increase variance heterogeneity. The authors employed an empirical quality‑control system to estimate the stability, precision, and reproducibility of these assays. These methods enable sequential monitoring of assay performance and yield valid confidence limits for potency estimates.

Abstract

Radioimmunoassay (RIA) and competitive protein binding (CPB) assays provide dose response curves which show severe nonlinearity and the residual (error) variance is non-uniform and not normally distributed. Accordingly, several classic "least squares" statistical procedures commonly used for bioassay data are not directly applicable. Linearization may be obtained by the use of the logit transformation, but this increases non-uniformity of variance. We have utilized an empirical quality control system in order to obtain estimates of the stability, precision and reproducibility of the radioimmunoassays and CPB assays. These methods provide sequential monitoring of the performance of the assays, and permit the determination of valid estimates of confidence limits on potency estimates.