Integrating Predictive Analytics into Hospital Automation: Comparative Evaluation of Prediction Models for Patient Monitor Demand in Operating Rooms

Predictive analytics has emerged as an important component of hospital automation by enabling proactive resource management and data-driven decision-making. Efficient allocation of patient monitors in operating rooms represents a practical application where forecasting can support perioperative safety and resource management. This study compared five forecasting models for predicting patient monitor availability using integrated clinical, operational, and equipment management data to identify appropriate forecasting approaches for operating room resource planning. We conducted a retrospective longitudinal study using monthly surgical operational data from the anesthesia information system and equipment-related data from the medical equipment management system of a tertiary referral hospital. The outcome was the monthly number of available patient monitors recorded in the equipment management system, which served as the reference value for evaluating five prediction models, including naïve persistence, autoregressive integrated moving average (ARIMA), multivariable linear regression, LSTM, and hybrid LSTM–regression. Model performance was evaluated using root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Across a 36-month study period, the mean monthly surgical volume was 1960 (SD 182) procedures, with a mean operative duration of 105.0 (SD 3.5) minutes. Weighted multivariable regression showed that service age (standardized β = 0.72, 95% CI 0.66–0.78), maintenance frequency (standardized β = 0.09, 95% CI 0.03–0.15), and surgical volume (standardized β = 0.15, 95% CI 0.09–0.21) were positively associated with available patient monitor, whereas mean operative duration was inversely associated (standardized β = −0.19, 95% CI −0.26 to −0.12). The naïve persistence (RMSE 0.17, MAE 0.03, MAPE 0.16%) and ARIMA (RMSE 0.17, MAE 0.04, MAPE 0.21%) showed higher predictive performance, while the multivariable linear regression, LSTM-only, and LSTM–regression models showed relatively higher prediction errors. Beyond predictive accuracy, the model provides an operational framework for transforming routinely collected clinical and equipment data into actionable information for automated resource planning. This study showed that predictive analytics can support hospital automation by enabling proactive patient monitoring and resource planning. Classical statistical models remain robust alternatives for hospital resource forecasting in small-sample settings, while regression-based approaches provide interpretability for operational decision-making.