Predictive value of severity scoring systems: Comparison of four models in Tunisian adult intensive care units

Objectives: To compare the performance of four severity scoring systems: the Acute Physiology end Chronic Health Evaluation (APACHE) II, the new versions of the Mortality Prediction Model (MPM0 and MPM24), and the Simplified Acute Physiology Score (SAPS) II. Design: A prospective cohort study. Setting: Three Tunisian intensive care units (ICUs). Patients: Consecutive, unselected adult patients (n = 1325). Interventions: None. Measurements and Main Results: Overall, observed death rates were higher than predicted by all models except MPM0. All the evaluated scoring systems had good discrimination power as expressed by area under the receiver operating characteristics curve, but their calibration was less perfect when compared with original validation reports. There were no major differences between the models with regard either to discrimination or calibration performance. Conclusion: Despite an overall good discrimination, APACHE II, MPM0, MPM24, and SAPS II showed a less satisfactory calibration in our Tunisian sample of ICU patients. Part of the models inaccuracy could be related to quality of care problems in our ICUs, but this issue needs further analysis.