A data-driven priority assessment and deployment framework for medical equipment maintenance in a tertiary hospital

Background: Effective maintenance management of medical equipment is essential to ensure patient safety, operational continuity, and cost control in hospitals. Traditional experience-based maintenance strategies often fail to capture the dynamic risk profiles of heterogeneous equipment, particularly in large healthcare institutions. Data-driven approaches may improve maintenance prioritization, yet evidence from real-world hospital deployment remains limited. Methods: We developed and implemented a machine learning-assisted priority evaluation system for medical equipment maintenance in a tertiary hospital. Separate priority assessment frameworks were established for preventive maintenance (PM) and corrective maintenance (CM), each incorporating domain-specific features and weighted scoring schemes. Multiple machine learning models, including logistic regression, decision tree, support vector machine, naïve Bayes, and XGBoost, were trained and evaluated using a stratified training-testing split. Model performance was assessed using accuracy, precision, recall, F1-score, receiver operating characteristic (ROC) curves, and confusion matrices. The optimal model was deployed into the hospital maintenance workflow and evaluated in a parallel controlled implementation. Results: A total of 9,924 medical devices were included, comprising 8,967 devices with preventive maintenance (PM) records and 957 devices with corrective maintenance (CM) records. Devices were stratified into low-, medium-, and high-urgency groups using clustering-derived labels. Among the five machine learning algorithms evaluated, XGBoost achieved the best performance, with a testing accuracy of 0.9379 in the PM dataset and 0.8646 in the CM dataset. In the real-world deployment phase (2025.1.2-2025.12.25), 830 devices in the intervention campus and 849 devices in the control campus were compared. The intervention campus showed lower proportions of failures, recurrence, and unplanned maintenance events, and a lower overall maintenance cost ratio than the control campus (4.8% vs. 7.3%). Conclusion: This study demonstrates the feasibility and practical value of deploying a machine learning-assisted priority evaluation system for medical equipment maintenance in a real hospital environment. By distinguishing preventive and corrective maintenance scenarios and integrating model outputs into routine workflows, the proposed framework supports more efficient, consistent, and cost-effective maintenance decision-making.