Comparison of four different algorithms for the calculation of radioimmunoassay standard curves.

Standard curves from 150 determinations of 10 different hormone radioimmunoassays have been evaluated using the following four procedures: logit transformation as performed by Rodbard, Healy's 4-component-logit, spline approximation and a linear interpolation program. The results, calculated by computer, have been compared with manual analysis as the refernce method. The influece of various correction factors has been studied by artificial insertion of outliers. Approximately two thirds of all investigated curves could be calculated with almost equal success using all 4 methods. In calculating the remaining third of the standard curves, each method showed characteristic errors which depended upon the reliability of the assay data and the kind of outliers. The results suggest that the smoothing by spine function is the most flexible method in approximating a radioimmunoassay standard curve. In comparison with the other methods it is less influenced by random errors and fits the characteristic and symmetry of the ideal curve most exactly. The 4-component logit which has been extended by data screening is an improvement of the original logit transformation. Certain reservations exist, however, in elimination of outliers because of the dispersion of the data points, and also because of the arbitrarily set thresholds.