Background: Postoperative nausea and vomiting (PONV) is a frequently observed complication in patients undergoing surgery under general anesthesia. Moreover, it is a frequent cause of distress and dissatisfaction during the early postoperative period. The tools used for predicting PONV at present have not yielded satisfactory results. Therefore, prognostic tools for the prediction of early and delayed PONV were developed in this study with the aim of achieving satisfactory predictive performance. Methods: The retrospective data of adult patients admitted to the post-anesthesia care unit after undergoing surgical procedures under general anesthesia at the Sheba Medical Center, Israel, between September 1, 2018, and September 1, 2023, were used in this study. An ensemble model of machine learning algorithms trained on the data of 54848 patients was developed. The k-fold cross-validation method was used followed by splitting the data to train and test sets that optimally preserve the sociodemographic features of the patients, such as age, sex, and smoking habits, using the Bee Colony algorithm. Findings: Among the 54848 patients, early and delayed PONV were observed in 2706 (4.93%) and 8218 (14.98%) patients, respectively. The proposed PONV prediction tools could correctly predict early and delayed PONV in 84.0% and 77.3% of cases, respectively, outperforming the second-best PONV prediction tool (Koivuranta score) by 13.4% and 12.9%, respectively. Feature importance analysis revealed that the performance of the proposed prediction tools aligned with previous clinical knowledge, indicating their utility. Interpretation: The machine learning-based tools developed in this study enabled improved PONV prediction, thereby facilitating personalized care and improved patient outcomes.

After a machine learning (ML)-based system is deployed in clinical practice, performance monitoring is important to ensure the safety and effectiveness of the algorithm over time. The goal of this work is to highlight the complexity of designing a monitoring strategy and the need for a systematic framework that compares the multitude of monitoring options. One of the main decisions is choosing between using real-world (observational) versus interventional data. Although the former is the most convenient source of monitoring data, it exhibits well-known biases, such as confounding, selection, and missingness. In fact, when the ML algorithm interacts with its environment, the algorithm itself may be a primary source of bias. On the other hand, a carefully designed interventional study that randomizes individuals can explicitly eliminate such biases, but the ethics, feasibility, and cost of such an approach must be carefully considered. Beyond the decision of the data source, monitoring strategies vary in the performance criteria they track, the interpretability of the test statistics, the strength of their assumptions, and their speed at detecting performance decay. As a first step towards developing a framework that compares the various monitoring options, we consider a case study of an ML-based risk prediction algorithm for postoperative nausea and vomiting (PONV). Bringing together tools from causal inference and statistical process control, we walk through the basic steps of defining candidate monitoring criteria, describing potential sources of bias and the causal model, and specifying and comparing candidate monitoring procedures. We hypothesize that these steps can be applied more generally, as causal inference can address other sources of biases as well.