Development of a Data-Based Method for Predicting Nursing Workload in an Acute Care Hospital: Methodological Study.

Area of Science:

• Healthcare Operations Research
• Applied Machine Learning
• Nursing Informatics

Background:

• Effective nurse staffing is critical for patient care quality and hospital efficiency.
• Traditional workload prediction methods lack accuracy.
• Machine learning (ML) offers data-driven solutions for precise workload forecasting.

Purpose of the Study:

• To utilize nursing activity data (LEP) for predicting future workload requirements.
• To apply machine learning techniques for enhanced workload prediction accuracy.

Main Methods:

• Retrospective observational study using inpatient nursing workload data (2017-2021).
• Trained and tested ML models (Lasso regression, Random Forest, XGBoost) to predict workload 72 hours in advance.
• Assessed performance using Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) against a baseline.

Main Results:

• ML models consistently outperformed the baseline, with Lasso regression showing a 25.0% average accuracy improvement.
• The Lasso model achieved strong classification performance (AUC 0.79, precision 66.2%-75.3%) for predicting workload changes.
• Critical misclassifications (increase as decrease, vice versa) occurred in only 0.17% of cases.

Conclusions:

• Machine learning shows significant potential for improving nurse workload prediction accuracy.
• Further research is needed to address data quality and explore advanced ML architectures.
• Real-world evaluation through controlled trials is essential for practical implementation.