A procedure for weight and model selection in assay development
The problem of weight and model selection for standard curve in the calibration of an instrument for assay development is considered. An instrument is usually calibrated by studying the relationship between a number of known standards and their corresponding responses. The relationship between known standards and their responses is usually described by a standard curve. Based on the established standard curve, an unknown sample can be determined. In practice, since the responses may have different variabilities at different known standards and the relatioship between the standards and their corresponding responses may not be linear, it is important to select an appropriate model with an appropriate weight in establishing a standard curve for obtaining an accurate and reliable assay result. In this paper, criteria for weight and model selection are proposed to choose an appropriate statistical model among five commonly used models with three different possible weights. An example concerning the validation of a plasma assay of a pharmaceutical compound is presented to illustrate the use of the proposed criteria.