The selection of laboratory tests for organ dysfunctions using multivariate statistics

Abstract

There is a need for selecting good combinations of laboratory tests for various medical decision situations. Multivariate statistical methods are better tools for this type of selection than multiple univariate comparisons of candidate tests. Linear discriminant analysis is a suitable method. The best selection strategies are stepwise selection and evaluation of all possible subsets. Guiding criteria in the selection process may be measures of statistical separation between the clinical groups or empirical probabilities of correct reallocation. The latter criterion has the advantage that an optimum subset size may be identified.