A calibration problem in statistical diagnosis: The clinical amalgamation problem

Abstract

To obtain a sufficiently large diagnostic training set for differential diagnosis it is often necessary to use cases from two or more clinics. Important questions that then arise are whether there are any differences between clinics in their methods of measuring the diagnostic features of cases, to what extent ignoring such differences invalidates any diagnostic system devised for these clinics and in what manner interclinic calibration information may be employed to make efficient use of the complete set of calibrative and diagnostic data. The modelling of such complex diagnostic situations is discussed, problems of statistical methodology arising are resolved, and the methods devised are illustrated by a particular problem of amalgamation for diagnostic purposes of data from different clinics.