Machine learning prediction of acute myocardial infarction incidence using regional meteorological data

Acute myocardial infarction (AMI) is significantly influenced by meteorological conditions; however, leveraging meteorological data to predict AMI incidence remains challenging. This study aimed to analyze weather-AMI associations and construct a predictive model using machine learning.

We conducted a retrospective analysis of AMI patients from a regional chest pain center, coupled with local daily weather data spanning 10 years. The relationship between weather variables and daily AMI case counts was analyzed. A Random Forest model was employed to capture potential non-linear relationships. Model performance was validated using real-world data.

Over 4,197 days (January 2013-June 2024), 11,527 AMI patients were included. Days with higher AMI incidence were characterized by lower temperatures, greater daily temperature differences (ΔT), and reduced air speed, while exhibiting lower humidity and precipitation compared to days with fewer cases. After multivariable adjustment, daily ΔT and air speed showed statistically significant associations with increased AMI incidence (p < 0.05), whereas other weather elements did not. In the predictive model, daily ΔT emerged as the most important factor. Validation demonstrated moderate discriminative ability, with an AUC-ROC of 0.67 (95% CI: 0.60-0.74). Sensitivity analysis using the 95th percentile as a threshold further confirmed the model's effectiveness.

This study identifies daily temperature variation (ΔT) and air speed as significant meteorological predictors of AMI incidence. The Random Forest model effectively captured non-linear weather-AMI relationships, supports integrating weather data-particularly temperature variability-into AMI risk stratification systems. Future research should enhance predictive power by incorporating clinical and demographic variables alongside environmental factors.