Nonlinear Modelling for Predicting Patient Presentation Rates for Mass Gatherings.

Area of Science:

• Emergency Medicine
• Public Health
• Data Science

Background:

• Mass gatherings in Australia present unique health challenges due to factors like crowd density, behavior, and substance use.
• On-site health services are crucial for managing attendee well-being and reducing community health facility strain.
• Accurate prediction of patient presentations is essential for efficient planning and management of mass gathering health services.

Purpose of the Study:

• To develop reliable tools for predicting patient presentations at mass gatherings.
• To understand the influence of various factors on patient load.
• To address the nonlinear nature of patient presentation prediction.

Main Methods:

• Collected data from 216 Australian mass gatherings, including event type, crowd demographics, and weather.
• Constructed nonlinear models using regression trees.
• Employed nine-fold cross-validation to estimate model error.

Main Results:

• Nonlinear models demonstrated high mean training errors but a skewed error distribution, with small errors for most events.
• The error was 15 or less for 85% of events, with a median error of 6.9 presentations.
• Models were less accurate for very large events but generally useful for smaller, community-based gatherings.

Conclusions:

• Nonlinear modeling offers a more realistic approach to understanding mass gathering health event variables.
• Developed models are effective for typical, smaller-scale community events.
• Further research is needed to refine models for predicting patient presentations at very large international events.