Outcome prediction in critical care: physicians' prognoses vs. scoring systems

Abstract

Background and objective: To compare the accuracy of prognoses made by intensive care physicians with the performance of two indicators, the original Simplified Acute Physiology Score (SAPS) II and a modified version optimized to the patient sample.

Methods: Data from 412 patients consecutively admitted to intensive care units of Göttingen University Hospital, Germany, were collected according to the original score criteria. Information necessary for the computation of SAPS II and the vital status on hospital discharge was recorded. To customize the original SAPS II in our cohort, the database was randomly divided into two subgroups. Logistic regression analysis with physiological values as explanatory variables was used. A bootstrap procedure completed the process. Furthermore, physicians were asked to indicate their prognostic judgement concerning the patients' hospital mortality.

Results: Discrimination analysis showed the following areas under receiver operating characteristic curves: physicians' prognoses 0.84 (confidence interval (CI): 0.79-89), SAPS II 0.75 (CI: 0.69-0.80) and customized SAPS 0.72 (CI: 0.66-0.78). The physician's forecast was significantly better, while the customized and the original SAPS were not substantially different as regards their accuracy.

Conclusions: Prognoses made by physicians are superior to objective models. This may result from more extensive knowledge and other kinds of information available to clinicians. A clinician's action also depends on his/her prognosis at the beginning of the treatment, giving raise to a possible correlation between medical outcome and the clinician's prognosis. Our findings indicate that physicians do not limit their prognosis to the objective factors at their disposal, but indicate that they base their decisions on experience and individual observations.