Predictive tests for the response to cancer chemotherapy: limitations related to the prediction of rare events.

The reliability of a predictive test may be characterized by its sensitivity and specificity. The probability that a predicted event will actually occur depends on these indexes as well as the frequency of the predicted event. If such a test is used to predict for the therapeutic effect of single-agent chemotherapy in cancer patients, its predictive value is markedly affected by a generally low therapeutic efficacy in an unselected population.