Forecasting severe respiratory disease hospitalizations using machine learning algorithms.

Area of Science:

• Epidemiology
• Health Informatics
• Machine Learning

Background:

• Forecasting hospitalization rates aids hospital management during seasonal epidemics.
• Predicting severe respiratory illness admissions can optimize elective surgery scheduling.
• Forecasting models can guide interventions to prevent health system overload.

Purpose of the Study:

• To evaluate forecasting models for predicting hospital admissions in Auckland, New Zealand, within a three-week horizon.
• To assess the performance of probabilistic forecasts.
• To determine the impact of integrating laboratory data on forecasting accuracy.

Main Methods:

• Utilized data from active hospital surveillance using the World Health Organization Severe Acute Respiratory Infection (SARI) case definition.
• Employed machine learning, generative pre-trained transformers, and artificial neural networks for forecasting.
• Systematically tested SARI patients for nine respiratory viruses, including influenza and RSV.

Main Results:

• Machine learning models outperformed naive seasonal models in forecast accuracy.
• Reducing forecast temporal resolution improved point forecast accuracy and probabilistic forecast reliability.
• Season-to-season variations in virus incidence correlated with hospitalizations, but integrating this data did not improve forecasts.

Conclusions:

• Active SARI surveillance data supports prediction of hospital bed utilization.
• Machine learning shows potential for proactive hospital management systems.
• Consistent data collection is crucial for effective predictive modeling in healthcare.