A revised approach to data modelling for patient classification to determine analytical performance specifications...

A widely accepted approach to setting Analytical Performance Specifications (APS) in laboratory medicine is the Milan models. The Milan Model 1b is an approach in which APS are developed based on patient classification or clinical decision-making, noting that these factors are linked to the probability of patient outcomes. A common approach is to model the effect of assay performance on the clinical classification of patients. The goal is that the effect of assay imprecision and bias can be assessed against the alternative patient classification rates seen with the changes. Modelling in this area has frequently used values from an original data set as the initial values for clinical classification and then assessed changes in classification against this background. In this paper an alternative approach is proposed. The classification of subjects in a study can be seen as one of many possible outcomes, noting that if the study was repeated with the same individuals and the same analytical method, a percentage of the population would be differently classified due to within-subject biological variation and random analytical variation. The approach presented here is to first quantify the baseline expected alternate classification rate in a study if it were repeated, and then consider changes in assay performance relative to that used in the original study on the expected alternate classification rate.