A statistical DataMining technique used when ClusteringPatients.
"Step-wise multiple discriminate analysis" is a statistical technique that gives weights to test results used to best differentiate or discriminate between positive or negative patient prognoses. For example, suppose 22 test results were tested on 66 patients, then 33 got worse and 33 got better. A formula to use these test results in a prognosis would be a linear function with 22 parameters.
Drop the test result that displayed the least discrimination, and run either the calculation or the entire experiment again. Repeat over and over each time dropping the test subset which least contributed to discriminating between the failed and successful patients.