Identifying Key Predictors of Nursing Workload in Emergency Infusion Rooms: A Decision Tree Approach

Purpose: Accurate assessment of nursing workload is essential for staffing allocation and operational management in emergency infusion rooms. However, workload generation is influenced by complex and potentially nonlinear interactions among patient volume, treatment duration, and care activities, which may not be adequately captured by conventional statistical approaches. This study aimed to identify key predictors associated with nursing workload intensity and develop an interpretable workload stratification framework using a Classification and Regression Tree (CRT) model. Methods: Daily operational data were collected from an emergency infusion room between July 2023 and August 2025. Daily chair utilization rate was used as a proxy indicator of workload intensity. Candidate predictors included total infusion duration, direct care encounters, number of patients receiving infusions, medication dispensing time, severe dependency, fall-risk patients, and triage-level patient volume. A CRT model was developed to identify hierarchical predictor relationships and threshold-based workload classification rules. Model robustness was evaluated using 10-fold cross-validation, comparative analyses with multiple linear regression, random forest, and gradient boosting models, and sensitivity analyses excluding total infusion duration. Results: The analysis included 761 valid observation days. Total infusion duration emerged as the most influential predictor, followed by direct care encounters and the number of patients receiving infusions. The CRT model identified clinically interpretable workload thresholds and generated a parsimonious decision structure for workload stratification. Re-substitution and cross-validation risk estimates were 0.045 (SE = 0.005) and 0.046 (SE = 0.005), respectively, indicating stable model performance. Although random forest and gradient boosting achieved higher predictive accuracy, the CRT model provided greater interpretability through transparent decision rules. Sensitivity analyses demonstrated that the overall workload stratification pattern remained largely unchanged after excluding total infusion duration. Conclusions: The CRT model identified total infusion duration, direct care encounters, and patient volume as key predictors associated with workload intensity in emergency infusion rooms. Although alternative models achieved higher predictive performance, the CRT approach provided interpretable workload stratification rules that may support staffing allocation and operational decision-making. The findings offer a practical data-driven framework for workload assessment in infusion care settings.