Data Mining for Hidden Groups in Hospital Populations

A data mining case study is presented illustrating exploration for hidden inpatient subpopulations. Some of the major characteristics of data mining and their relation to business goals are reviewed. Six subpopulations are tentatively identified. Two subpopulations correspond to mental illness and alcohol/drug abuse. Four other subpopulations reflect medical and surgical diseases, segregated along lines of major diagnostic category, age, numbers of other diagnoses, and income. Some suggestions are presented for using cluster analysis to find potential hidden groups, and for the interpretation and validation of these groups using visualization and discriminant analysis techniques.