The Use of Clustering Techniques for the Classification of Psychiatric Patients
JOHN S. STRAUSS M.D.1,
JOHN J. BARTKO Ph.D.2, and
WILLIAM T. CARPENTER JR. M.D.3
1 Director, Clinical Psychiatry Research Programs, Rochester University School of Medicine, 260 Crittenden Boulevard, Rochester, New York 14642.
2 Mathematical Statistician, Biometry Branch, National Institute of Mental Health, Bethesda, Maryland
3 Acting Chief, Psychiatric Assessment Section, National Institute of Mental Health, Bethesda, Maryland
The sensitivity of the classifications produced by cluster analysis to type of data preparation and choice of program has been demonstrated. Because of this sensitivity, it is helpful to determine at least that a given cluster program can reproduce known groups in a stable way utilizing data of the type to be analysed in the real sample. Once this has been accomplished, analysis of the real data can be performed. In the application of clustering techniques to problems of psychiatric classification, the results of the analysis of real data can be related to diagnostic categories and characteristic profiles. If these appear promising, the next step is the difficult task of validation of the groups using criteria different from the kind of data from which the patients were originally classified.
Submitted on March 27, 1972
