A machine learning system, Prescience, was developed to predict hypoxemia during surgery and explain the risk factors. The system uses real-time data from electronic medical records of over 50,000 surgeries to predict hypoxemia risk and provide explanations. It improves the performance of anaesthesiologists by helping them anticipate hypoxemia events, which could benefit from early intervention. The system provides insights into the exact changes in risk induced by patient and procedure characteristics, enhancing clinical understanding of hypoxemia risk during anaesthesia care. Prescience was trained on data from the Anaesthesia Information Management System (AIMS), including static and real-time features such as patient demographics, surgical procedures, and physiological measurements. It uses a gradient boosting machine model to predict hypoxemia risk and provides interpretable explanations of the risk factors. The system outperforms previous prediction models in accuracy and provides explanations that are consistent with clinical knowledge. Prescience was tested by comparing anaesthesiologists' predictions with and without the system's assistance. The results showed that anaesthesiologists using Prescience were able to anticipate more hypoxemia events, with the system's explanations improving their ability to make accurate predictions. The system's predictions were validated against real-time data and showed high accuracy in predicting hypoxemia events. Prescience's explanations help anaesthesiologists understand the factors contributing to hypoxemia risk, enabling them to make informed decisions about patient care. The system's ability to provide simple explanations of predictions from complex models helps eliminate the typical accuracy vs. interpretability tradeoff, allowing broader applicability of machine learning to medicine. Prescience's results suggest that with the system's assistance, anaesthesiologists could anticipate up to 30% of hypoxemia events, compared to the current 15%. The system's explanations are based on features such as patient BMI, tidal volume, and pulse rate, which are modifiable and can be used to mitigate hypoxemia risk. The system's ability to provide interpretable explanations of predictions is crucial for clinical decision-making and improving patient outcomes.A machine learning system, Prescience, was developed to predict hypoxemia during surgery and explain the risk factors. The system uses real-time data from electronic medical records of over 50,000 surgeries to predict hypoxemia risk and provide explanations. It improves the performance of anaesthesiologists by helping them anticipate hypoxemia events, which could benefit from early intervention. The system provides insights into the exact changes in risk induced by patient and procedure characteristics, enhancing clinical understanding of hypoxemia risk during anaesthesia care. Prescience was trained on data from the Anaesthesia Information Management System (AIMS), including static and real-time features such as patient demographics, surgical procedures, and physiological measurements. It uses a gradient boosting machine model to predict hypoxemia risk and provides interpretable explanations of the risk factors. The system outperforms previous prediction models in accuracy and provides explanations that are consistent with clinical knowledge. Prescience was tested by comparing anaesthesiologists' predictions with and without the system's assistance. The results showed that anaesthesiologists using Prescience were able to anticipate more hypoxemia events, with the system's explanations improving their ability to make accurate predictions. The system's predictions were validated against real-time data and showed high accuracy in predicting hypoxemia events. Prescience's explanations help anaesthesiologists understand the factors contributing to hypoxemia risk, enabling them to make informed decisions about patient care. The system's ability to provide simple explanations of predictions from complex models helps eliminate the typical accuracy vs. interpretability tradeoff, allowing broader applicability of machine learning to medicine. Prescience's results suggest that with the system's assistance, anaesthesiologists could anticipate up to 30% of hypoxemia events, compared to the current 15%. The system's explanations are based on features such as patient BMI, tidal volume, and pulse rate, which are modifiable and can be used to mitigate hypoxemia risk. The system's ability to provide interpretable explanations of predictions is crucial for clinical decision-making and improving patient outcomes.