A multivariate approach to laboratory practice

Abstract

The application of multivariate statistical technics to laboratory data analysis is a recent development in clinical pathology. The proliferation in number and type of diagnostic tests requires a simple method to extract predictors of clinical states from data generated by existing laboratory procedures, as well as a method to assess the usefulness of new tests as they are proposed. A series of algorithms can select a "best predictor" subset of laboratory tests and verify this reduced group's predictive diagnostic worth. The established predictive ability of existing laboratory tests can serve as a reference scale when a newly available laboratory test is evaluated. The proposed new test is included with the best predictors already identified and confirmed. In multivariate analysis the usefulness of the proposed new procedure in the predictive diagnosis scheme is determined.