Individual outcome prediction models for intensive care units.

Prognostic criteria based on static analysis of group statistics do not help much in decisions to withhold or withdraw therapy from intensive care unit (ICU) patients too ill to benefit, since they do not provide adequate information on the features that distinguish non-survivors from survivors. A predictive model which uses dynamic analysis of severity scores based on physiological variables is presented here along with the results of tests of the model in 831 ICU patients. 109 patients were correctly predicted to die by the model. Of 722 whose prediction was outcome unknown, 181 died. Thus, the odds for prediction of death among non-survivors were 0.376. Since there were no false predictions of death, the estimated chance of a false prediction was 0.0055.