A Systematic Review of Features Forecasting Patient Arrival Numbers

Area of Science:

• Healthcare Operations Research
• Applied Machine Learning
• Data Science

Background:

• Adequate nurse staffing is essential for quality healthcare delivery.
• Accurate prediction of patient arrival rates is crucial for effective nurse staffing.
• Current predictive models primarily use temporal data, with limited exploration of other features.

Purpose of the Study:

• To systematically review studies using supervised machine learning for patient arrival prediction.
• To identify nontemporal features that can reduce prediction error.
• To explore potential new data sources for improving prediction accuracy.

Main Methods:

• Systematic literature review of studies predicting patient arrival numbers using supervised machine learning.
• Focused on nontemporal features (not based on time or dates).
• Screened 26,284 studies, with 27 selected for analysis.

Main Results:

• Identified three main groups of nontemporal features: weather data, internet search/usage data, and social interaction data.
• Internet and social interaction data showed promise, reducing errors by up to 33% in some studies.
• Weather data's utility was less clear, and sources like smartphone and social media data remain largely unexplored.

Conclusions:

• Nontemporal features, particularly internet and social interaction data, can significantly enhance patient arrival prediction accuracy.
• Further research is needed to explore novel data sources and address potential data privacy challenges.
• Improving prediction models is vital for optimizing healthcare resource allocation and nurse staffing.