Comparison of physician prediction with 2 prognostic scoring systems in predicting 2-year mortality after intensive care admission: A linked-data cohort study

Abstract

PurposePatients who survive an episode of critical illness continue to experience significant mortality after hospital discharge. This study assessed the accuracy of physician prediction of 2-year mortality and compared it with 2 objective prognostic models. MethodsSensitivity (probability of a prediction of death in patients who died within 2 years) and specificity (probability of a prediction of survival in patients who survived at least 2 years) of physicians' 2-year prediction were compared with those from 2 objective prognostic models, Acute Physiology and Chronic Health Evaluation (APACHE) II and Predicted Risk Existing Disease Intensive Care Therapy (PREDICT). ResultsPhysician prediction of 2-year mortality was available for 2497 (94.8%) intensive care unit admissions. Specificity was high (85.2%; 95% confidence interval [CI], 83.7-86.4), but sensitivity (65.0%; 95% CI, 61.1-68.8) and positive predictive value (57.4%; 95% CI, 53.6-61.2) were relatively low, suggesting overpessimistic prediction of 2-year mortality. Age, Charlson comorbidity index, and APACHE score were independent risk factors for an inaccurate physician prediction. The diagnostic odds ratio for the physician predictions was at least comparable with the APACHE and PREDICT models, which both had very good discrimination of mortality at 2-year follow-up. ConclusionsPhysicians tended to overpredict the risk of 2-year mortality of critically ill patients, but accuracy was comparable with 2 objective prognostic models.