Discriminating Geriatric and Nongeriatric Patients Using Functional Status Information: An Example of Classification Tree Analysis Via UniODA

A procedure is described that involves iterative use of univariable optimal discriminant analysis (UniODA) in order to construct a classification tree model for discriminating observations from different groups, say A and B. In the first step of the procedure, which involves the entire sample, the single best discriminator is selected. On the basis of this step, some portion of the total sample is predicted to be from group A (some of whom may actually be from group B, and are thus misclassified), and the remainder are predicted to be from group B (some of whom may actually be from group A, and are thus misclassified); these are partitions of the total sample. In the second and subsequent steps of the procedure, UniODA is employed to identify attributes that improve classification accuracy for successively smaller subsets (partitions) of the total sample. This procedure yields hierarchically optimal classification performance that may surpass the classification performance achieved using linear methodologies. The procedure is illustrated using an application involving discriminating between geriatric and nongeriatric adult ambulatory medical patients on the basis of their self-reported functional status.