Preoperative assessment of patients at risk of postoperative respiratory depression

Abstract

Respiratory depression during sleep is a major health challenge after surgery. The main cause is reduction in breathing due to opioids, which are commonly used for management of postoperative pain. The consequences are hypoxemia and hypercapnia, which may increase the risk of cardiovascular complications, mortality, and healthcare utilization. Identifying individuals who are at risk of postoperative respiratory depression prior to the surgery can help guide the perioperative care to reduce adverse outcomes. In this project, we developed a risk assessment model to identify individuals at risk of postoperative respiratory depression prior to the surgery, based on the demographics and changes in preoperative overnight oxyhemoglobin saturation (SpO₂) levels. To achieve this, we retrospectively analyzed SpO₂ signals of 159 patients, which were recorded continuously preoperatively and on the third night after surgery. Respiratory depression was defined as postoperative episodes where SpO₂ was ≤85% for more than 3 minutes. From preoperative SpO₂ signals, we extracted features to characterize overnight SpO₂ and desaturation episodes. We streamlined a systematic process for feature selection and model development using a nested cross-validation pipeline. Our results indicated that random forest, XGBoost, and Naïve bayes demonstrated the highest predictive performance, consistently surpassing the recent available PRODIGY model. These findings suggest that demographics and preoperative SpO₂ characteristics can preoperatively identify individuals at high-risk of postoperative respiratory depression, which offers a non-invasive and cost-effective method of monitoring respiratory health.