Machine learning-based forecasting of daily acute ischemic stroke admissions using weather data

Area of Science:

• Environmental Health
• Medical Informatics
• Epidemiology

Background:

• Weather phenomena are significant contributors to global stroke burden, accounting for approximately 11% of cases.
• The climate crisis necessitates advanced weather-based predictive analytics for healthcare systems.
• Accurate forecasting of disease incidence is crucial for effective public health management.

Purpose of the Study:

• To develop and evaluate machine learning models for forecasting daily acute ischemic stroke (AIS) admissions.
• To identify key weather parameters influencing AIS incidence.
• To establish a generalizable framework for weather-related disease burden prediction.

Main Methods:

• Development of predictive models using locoregional weather data and AIS patient admissions (2015-2021).
• Geospatial matching of a 7914-patient AIS cohort from University Medical Center Mannheim, Germany, to German Weather Service data.
• Evaluation of multiple machine learning algorithms including Poisson regression, GAMs, SVM, Random Forest, and XGBoost within a nested cross-validation framework.

Main Results:

• Extreme Gradient Boosting (XGB) demonstrated the highest performance with a mean absolute error of 1.21 cases/day.
• Maximum air pressure was identified as the primary predictor, while temperature showed a bimodal relationship with AIS admissions.
• Increased AIS admissions were associated with cold stress (Tmin_lag3 < -2°C), heat stress (Tperceived < -1.4°C; Tmin_lag7 > 15°C), and stormy conditions (wind gusts > 14 m/s).

Conclusions:

• Machine learning models effectively forecast daily acute ischemic stroke admissions based on weather data.
• Specific weather conditions, including extreme temperatures and high winds, are linked to increased stroke incidence.
• The developed framework offers a valuable tool for real-time hospital resource planning and predicting weather-related health impacts.