EXPLAINABLE MACHINE LEARNING FOR PERSONALIZED POSTOPERATIVE PAIN PREDICTION FOLLOWING MAJOR CANCER SURGERY UNDER GENERAL ANESTHESIA

Abstract

This study aims to explore the use of explainable machine learning algorithms that can predict individual postoperative pain outcomes for patients undergoing major cancer surgery under general anesthesia. Perioperative care has become more complex, and the problem of opioids is ongoing, making the need for accurate and transparent predictive systems that enable individualized care of the analgesics more important. Many traditional statistical methods are unable to represent the complex non-linear relationships between perioperative variables and many advanced artificial intelligence models are too hard to interpret in the clinic. To overcome this challenge, this study utilized a retrospective cross sectional study design with clinical, demographic, laboratory and intraoperative data from 13,700 surgical patients. To determine the most important predictors related to postoperative pain intensity, several machine learning methods such as Gradient Boosting Machines and Least Absolute Shrinkage and Selection Operator regression were applied. To avoid overfitting and enhance the robustness of the model, stratified 10-fold cross validation was used. The explainability was further strengthened by Shapley Additive explanations (SHAP), which allow the interpretation of the contributions of the features and the increase of the trust of the clinicians in the model predictions. This study revealed that intra-operative factors like blood transfusion and tourniquet use were significant factors affecting the pain trajectories after surgery. SHAP analysis demonstrated that these factors played a significant role in the increase in pain scores, underscoring the potential of inflammation, ischemia-reperfusion injury, and surgical trauma as factors in postoperative recovery. Moreover, the study illustrated the value of AI that can be understood by and interpreted by humans, and how that could help anesthesiologists and peri-operative care teams make actionable clinical decisions based on black-box predictions. Explainable machine learning could enhance patient-centered outcomes and facilitate precision medicine approaches in oncological surgery by allowing the development of personalized opioid-sparing strategies and the proactive identification of risk. The results highlight how transparent artificial intelligence systems are increasingly becoming a key tool for optimizing peri-operative decision making, provided they are integrated in an ethical, reliable, and clinically meaningful way in a contemporary healthcare context.