The determination of detection limits for insulin antibody assays.

In order to define the detection limit of a radioimmunoassay for insulin-antibody a correction was made for binding in the presence of an excess of unlabelled insulin and assay precision was calculated. One hundred forty control sera were assessed; all were islet cell antibody negative. For each sample, binding of 125-I human insulin was determined both with and without excess unlabelled insulin, subtraction of the latter acting as a correction. The distribution of uncorrected binding was skewed while corrected binding was normally distributed, (mean (SD) = 0.149 (0.298%)) Precision, defined as the mean of the standard deviations of replicates, was 0.263%. Detection limits calculated from the estimate of precision (0.263%) or from the standard deviation of the corrected binding (0.298%) were similar. Two hundred thirty sera from insulin-treated patients were studied. Precision was plotted as a 'precision profile' and the detection limit calculated from the precision for binding of less than 1% [0.261%]; 88% of the sera were positive [cut-off 1.3%, p less than 0.01]. We conclude that corrected binding is normally distributed in antibody-negative sera and that an estimate of assay precision can be used to define the detection limit of the assay.