Predicting mortality after surgery in a middle eastern intensive care unit

Background: Predicting postoperative mortality could optimize the use of resources including intensive care unit (ICU) admission. Aim: We analyzed postoperative mortality predictors in the largest Middle Eastern ICU and tailored a prediction model accordingly. Patients and Methods: All pertinent data of adult postoperative patients who were admitted to our ICU from 01-01-2016 to 31-12-2017 were retrospectively analyzed by means of logistic regression to identify mortality predictors. Using the probabilities recorded for the latter and a cut-off value of 0.5 for death, a prediction model was tailored and its accuracy tested by receiver operator characteristics (ROC) analysis with reporting of area under the curve (AUC). Results: The study included 751 patients [aged 53.12 ± 18.1years old, 464 males, with Acute Physiology and Chronic Health Evaluation (APACHE) 4 score of 100.8 ± 26.5 and American Society of Anesthesiology Physical Status score (ASA-PS) of 2.8 ± 1] baring a postoperative mortality rate of 5.1%. Significant mortality predictors depicted were (all p < 0.05): Age [odds ratio (OR) 1.09, 95% confidence intervals (CI): 1.02 - 1.2], longer ICU length of stay (OR 1.1, 95% CI: 1.04 - 1.17), ASA-PS (OR 33.2, 95% CI: 4.97 - 222.02), lactate levels upon ICU admission (OR 1.73, 95% CI: 1.08 - 2.76), and intraoperative use of inotropes/vasopressors (OR 1.9, 95% CI: 52.6 - 367.5). The featured prediction model integrating the aforementioned variables showed that 98.8% of cases were appropriately classified with AUC of 0.997 (95% CI: 0.989 - 0.999). Conclusion: Despite limitations, our postoperative mortality prediction model proved to be an effective tool in depicting patients’ outcome.