Clinical Chemistry, Vol 13, 931-940, Copyright © 1967 by the American Association for Clinical Chemistry

Applications of Computer Produced Frequency Distribution Curves

II. Evaluation of the Diagnostic Significance of Test Results by Multidimensional Analysis

¹ Departments of Pathology and of Biological Chemistry, The University of Michigan Medical School, and from the Department of Biostatistics, The University of Michigan School of Public Health, Ann Arbor, Mich.

A method is described for determining the diagnostic significance of test results. It is based upon the relation of a test value to a frequency distribution curve of patients with a particular disease, and to a curve of others without that disease. When the two curves overlap at the test value found, the results are evaluated in terms of the probability of the patient corresponding to either of the two curves. The evaluation of results in terms of probabilities is more descriptive than the classification of all answers as either "normal" or "abnormal" and permits the combination of probabilities obtained from a number of tests into a single, resultant probability. The most probable diagnosis of a given patient, based upon all the tests done, or the most discriminating combination of tests for the diagnosis of any disease can be determined. The method permits compensation for such factors as a patient's age and sex, or the stage of a disease. It is nonparametric, and can be performed automatically with a data-processing system.