Machine Learning to Predict Surgery Duration: Towards Implementing AI and Digital Twin for Effective Scheduling

Traditional elective patients' surgical scheduling relies on plan-makers' subjective estimates or historical averages, leading to inefficiencies such as surgery cancellations or underutilisation of resources. Machine learning (M/L)-based predictive algorithms offer a promising solution with data-driven models to forecast surgical times, however, their application in NHS hospital settings remains limited. This study explores the implementation of multiple M/L algorithms for surgical time estimation for Trauma and Orthopaedics related procedures in an NHS Trust hospital. Results indicate that Neural Networks, along with ElasticNet regression, Gradient Boosting, and Bayesian Ridge regression models, demonstrate robust performance. Additionally, expansion to procedure specific models, built separately for each procedure shows promising results. This study contributes insights into the integration of M/L algorithms into healthcare digital resources, paving the way for enhanced surgical planning strategies. Future research will focus on integrating the predictive models into a comprehensive AI driven Digital Twin framework for simulation and optimisation-driven automated decision-making.