Statistical Aspects of Radioimmunoassay

Abstract

The logit-log method (Rodbard and Lewald 1970) remains the most popular method for RIA data reduction in use today (Fig. 1). It has the virtue of simplicity (Rodbard 1979). The method can be implemented graphically, using special logit-log graph paper. It can be implemented using a small programmable handheld calculator or computer (Davis et al. 1980). Even the smallest microcomputers can easily perform this method with proper weighting and detailed statistical analysis. A weighted linear regression can even be performed by special “macro” programs for popular spreadsheet programs. The logit-log method has been incorporated into numerous commercial systems (β- and γ-counters). When the logit-log method “works” (which is probably about 90%–95% of the time) everything is fine. Unfortunately, in about 5%–10% of assays, the logit-log method fails to provide an adequate description of the RIA dose-response curve. What should the assayist do when the logit-log method fails? Since numerous alternatives are available, each with its own advantages and limitations, the assayist is often faced with a bewildering situation (Rodbard 1979). Are we to try all possible methods, using a trial-and-error approach? When will we be able to say that we have a “good” method, an “adequate” or “optimal” method? Is there a systematic approach? Are we to be restricted to the curve-fitting methods implemented in commercial, turnkey, “black box” approaches to data analysis? This chapter will attempt to provide a simple, rational approach to the question, “what should I do when the logit-log method fails?”