Observations on the automated calculation of radioimmunoassay results.

Radioimmunoassay standard curves derived by four different automated calculation methods are compared with those derived manually for five assays of clinical importance (thyroxine, triiodothyronine, thyrotrophin, alphafetoprotein, and human growth hormone). The three curve fitting methods (linear or cubic regression on log x versus logit y, cubic regression on log x versus y) produced considerable distortion of the manually derived curves, which in some cases could have impaired the accuracy of an estimate in a clinical sample and altered the clinical interpretation. Distortion patterns varied with the particular antigen assayed and with experimental conditions. In contrast, a simple linear interpolation method produced little distortion and gave results which were close to those derived manually in each assay examined.