The Error of Replicated Potency Estimates in a Biological Assay Method of the Parallel Line Type

Abstract

The agreement observed among repeated estimates of log potency (M) derived from two modifications of a chick comb androgen assay method was compared with the error predicted by the usual formulas. In the first modification, the variance observed among 169 replicate M''s for 75 urine extracts was three times as large as predicted, providing highly significant evidence of large interassay variation. In the second modification of the method, in which the vehicle for the androgenic substances was changed from ether to ethanol, there was no evidence of interassay error among duplicate M''s for 28 urine extracts; the observed error of 54 M''s for 23 synthetic steroids was 55% greater than predicted. This improvement, which was probably due to the change in the experimental procedure, could not have been detected without an examination of replicate estimates. The observed error of replicate M values in a parathyroid hormone assay with relatively imprecise slopes was in good agreement with prediction. It was concluded that comparison of replicate estimates is essential for the full evaluation of an assay method and for assessing the effect of changes in assay procedure. A method of including the observed interassay error in computing confidence limits was described.