Statistical approach to immunosuppression classification using lymphocyte surface markers and functional assays.

We analyzed results of 22 in vitro parameters of immunocompetence in 72 cancer patients and 73 healthy controls. We then applied three statistical methodologies (discriminant analysis, logistic regression analysis, and recursive partitioning) in an effort to select the best predictors of immunosuppression. Using either of two definitions of immunosuppression (deviation by more than 1 standard deviation from the control mean on any assay, or having a diagnosis of advanced cancer), the same variables were selected. The best predictors were percentage of lymphocytes, percentage of suppressor cells, pokeweed mitogen stimulation, percentage of Ia+ cells, and number of helper cells. By all three methods, immunosuppressed and immunocompetent individuals were selected with 95 to 97% accuracy using a decision tree with these five tests as variables. In a cohort of individuals with incomplete data, the three methods still accurately classified the two groups with 70 to 83% accuracy. We conclude that a much smaller battery of tests can be used to identify immunosuppressed individuals for purposes of evaluation of responses to immune modulating agents.