Evaluation of Predictive Models of Inpatient Nursing Workloads

Area of Science:

• Health Informatics
• Nursing Resource Management
• Machine Learning in Healthcare

Background:

• Efficient medical personnel allocation is crucial for adapting to fluctuating healthcare demands.
• Nursing workload is influenced by patient volume and individual patient characteristics documented in Electronic Health Records (EHRs).
• EHR data offers potential for automated prediction of patient-specific nursing requirements to aid human resource management.

Purpose of the Study:

• To evaluate the effectiveness of machine learning models trained on EHR data for predicting patient-specific nursing workload.
• To compare the performance of Gradient Boosting (GB) and Random Forest (RF) algorithms in this predictive task.

Main Methods:

• Trained Random Forest (RF) and Gradient Boosting (GB) models using an anonymized dataset of 27,178 inpatient cases and 53 attributes from an acute-care hospital.
• Evaluated model performance using statistical metrics.
• Assessed the impact of reducing the number of input attributes on predictive accuracy.

Main Results:

• Gradient Boosting models demonstrated superior predictive performance compared to Random Forest models.
• Random Forest models exhibited signs of overfitting.
• Reducing the dataset to 25 relevant attributes had a minimal impact on the predictive performance of the models.

Conclusions:

• Machine learning models, particularly Gradient Boosting, can effectively predict patient-specific nursing workload using EHR data.
• EHR-driven workload prediction holds promise for optimizing nursing resource allocation.
• Further research should explore model generalizability across different settings and investigate alternative algorithms like neural networks.