AI-Driven Forecasting of Operating Room Quality Indicators Using LSTM Neural Networks

Operating Room (OR) efficiency directly influences patient safety, resource allocation, and overall hospital performance. Despite the routine monitoring of operating room quality indicators, most analyses remain retrospective and provide limited support for forward-looking management. This study develops a forecasting framework based on Long Short-Term Memory (LSTM) neural networks and integrates the prediction results into a Business Intelligence (BI) visualization environment. Monthly specialty-level operational data from a medical center in Taiwan covering 2021 to 2024 were analyzed, including seven key operating room quality indicators. Different historical input windows ranging from 3 to 24 months were examined to evaluate their effects on predictive performance. To assess whether deep learning was necessary for this task, model performance was compared with traditional statistical approaches, including Moving Average, automated Autoregressive Integrated Moving Average (auto-ARIMA), and Naïve forecasting models, using a strictly chronological training and testing split. Indicators characterized by short-term operational variability, such as cancellation and scheduling delay rates, were more accurately predicted using shorter historical windows, whereas structurally stable metrics including utilization and occupancy rates showed improved performance with longer input sequences. While conventional statistical models performed competitively for some stable indicators, LSTM provided more consistent performance across heterogeneous metrics. The forecasting outputs were incorporated into an interactive dashboard to facilitate real-time monitoring and threshold-based management. Although the analysis was conducted using data from a single medical center, the results demonstrate the feasibility of integrating predictive analytics into routine surgical quality management.