Outcome prediction models on admission in a medical intensive care unit: do they predict individual outcome?

Prospectively acquired data from 941 patients staying greater than 24 h in a medical ICU were analyzed to determine the relevance of scoring on ICU admission by the following methods of outcome prediction: Acute Physiology and Chronic Health Evaluation (APACHE II), Simplified Acute Physiology Score (SAPS), and Mortality Prediction Model (MPM). Analysis was performed separately for all patients (group A) and for a subsample (group B), obtained by excluding coronary care patients. Calculation of risk and classification of patients were carried out as recommended in the literature for MPM, APACHE II, and SAPS. In group A, sensitivities (correct prediction of hospital mortality) were 44.7%, 51.1%, and 21.2% and specificities (correct prediction of survival) were 84.5%, 85.4%, and 96.8%, respectively; overall correct classification rates were 73.3%, 75.8%, and 75.6%. In group B, sensitivities were slightly higher, but total correct classification rates did not reach group A levels. Goodness-of-fit testing showed low levels of fit for all methods in both groups. Application of APACHE II to diagnostic subgroups, using disease-adapted risk calculations, revealed marked inconsistencies between the estimated risk and the observed mortality. We conclude that the estimation of risk on admission by the three methods investigated might be helpful for global comparisons of ICU populations, although the lack of disease specificity reduces their applicability for severity grading of a given illness. The inaccuracy of these methods makes them ineffective for predicting individual outcome; thus, they provide little advantage in clinical decision-making.