Evaluating the Radiology Exam Turnaround Time Using the RIS Workflow Logs: An AI-Based Predictive Modeling Approach

This study developed and evaluated machine learning models to predict radiology examination turnaround time (TAT) using Radiology Information System (RIS) workflow logs while identifying operational factors associated with prolonged turnaround time. A retrospective quantitative design was employed using 21,870 de-identified radiology examination records obtained from a Level III private hospital. Three machine learning algorithms—Linear Regression, Random Forest, and XGBoost—were developed and evaluated using a 70:30 training-testing split with five-fold cross-validation. Among the models, Random Forest achieved the lowest Mean Absolute Error (154.30 minutes), whereas XGBoost demonstrated the highest explanatory power (R² = 0.59). The most influential predictors of TAT were the day of the week, request hour, daily examination volume, time of day, and hourly examination volume. Workflow analysis further identified the interval between resident interpretation and report transmission as the primary operational bottleneck contributing to delays. Although the study was limited to a single institution and relied on retrospective RIS data without incorporating variables such as radiologist experience or equipment downtime, the findings demonstrate the potential of predictive analytics to support staffing optimization, workload forecasting, scheduling, and workflow improvement. Overall, this study provides local evidence supporting the application of artificial intelligence (AI) to enhance radiology workflow management and operational efficiency in Philippine hospitals.