A machine learning-enabled digital twin for an orthopaedic clinic: a proof of concept

Abstract

The emergence of Digital Twins (DTs) creates opportunities for ad-hoc solutions in complex systems. This paper proposes a DT for the Luton & Dunstable Orthopaedic Clinic in the UK to support decision making. In this study, the interactive operations learn from actual patient data and offer predicted feedback over time, using machine learning algorithms such as Random Forest, Light GBM, and XG Boost. The system simulation model is built in Any Logic using a Discrete Event Simulation (DES) technique and tested with data from the hospital's DASH clinic app. The best-performing model, XG Boost, for predicting patient waiting times, achieved a mean absolute error (MAE) of 13.67 minutes. Based on predicted wait times, a feedback mechanism decides whether to notify patients about their expected waiting duration. The simulation records actual waiting times for upcoming patients, allowing prediction validation. Real-world validation yielded an MAE of 31.40 minutes, demonstrating promising accuracy for a clinical setting. The DT's added value lies in providing individual feedback to patients, enabling a real-time feedback loop with potential for optimisation of appointment scheduling.