Development and Validation of an Intraoperative Predictive Model for Unplanned Postoperative Intensive Care

Background: The allocation of intensive care unit (ICU) beds for postoperative patients is a challenging daily task that could be assisted by the real-time detection of ICU needs. The goal of this study was to develop and validate an intraoperative predictive model for unplanned postoperative ICU use. Methods: With the use of anesthesia information management system, postanesthesia care unit, and scheduling data, a data set was derived from adult in-patient noncardiac surgeries. Unplanned ICU admissions were identified (4,847 of 71,996; 6.7%), and a logistic regression model was developed for predicting unplanned ICU admission. The model performance was tested using bootstrap validation and compared with the Surgical Apgar Score using area under the curve for the receiver operating characteristic. Results: The logistic regression model included 16 variables: age, American Society of Anesthesiologists physical status, emergency case, surgical service, and 12 intraoperative variables. The area under the curve was 0.905 (95% CI, 0.900-0.909). The bootstrap validation model area under the curves were 0.513 at booking, 0.688 at 3h before case end, 0.738 at 2h, 0.791 at 1h, and 0.809 at case end. The Surgical Apgar Score area under the curve was 0.692. Unplanned ICU admissions had more ICU-free days than planned ICU admissions (5 vs. 4; P < 0.001) and similar mortality (5.6 vs. 6.0%; P = 0.248). Conclusions: The authors have developed and internally validated an intraoperative predictive model for unplanned postoperative ICU use. Incorporation of this model into a real-time data sniffer may improve the process of allocating ICU beds for postoperative patients.