Objective: To compare the performance of three severity of illness scoring systems used commonly for intensive care unit (ICU) patients in a large international data set. The systems analyzed were versions II and III of the Acute Physiology and Chronic Health Evaluation (APACHE) system, versions I and II of the Simplified Acute Physiology Score (SAPS), and versions I and II of the Mortality Probability Model (MPM), computed at admission and after 24 hrs in the ICU.
Design: A multicenter, multinational cohort study.
Setting: One hundred thirty-seven ICUs in 12 European and North American countries.
Patients: During a 3-month period, 14,745 patients were consecutively admitted to 137 ICUs enrolled in the study.
Interventions: Collection of information necessary to compute the APACHE II and APACHE III scores, SAPS I and SAPS II, and MPM I and MPM II scores. Patients were followed until hospital discharges. Statistical comparison, including indices of calibration (goodness-of-fit) and discrimination (area under the receiver operating characteristic curve).
Measurements and main results: Despite having acceptable receiver operating characteristic areas, the older versions of the systems analyzed (APACHE II, SAPS, and MPM I computed at admission-MPM I computed after 24 hrs in the ICU) demonstrated poor calibration for the whole database. The new versions of the systems (SAPS II and MPM II) were superior to their older counterparts. This superiority is reflected by larger receiver operating characteristic areas and better fit. The APACHE III system improved its receiver operating characteristic area compared with the APACHE II system, which showed the best fit of the old systems analyzed.
Conclusions: The new versions of the severity systems analyzed (APACHE III, SAPS II, MPM II) perform better than their older counterparts (APACHE II, SAPS I, and MPM I). APACHE II, SAPS II, and MPM II show good discrimination and calibration in this international database.