Use of artificial intelligence and environmental data to estimate respiratory hospitalizations in support of SUS management

Scritto il 01/07/2026
da Gabriel Fuscald Scursone

Cien Saude Colet. 2026 May;31(5):e23522025. doi: 10.1590/1413-81232026315.23522025. Epub 2025 Dec 12.

ABSTRACT

This study developed a predictive model of hospital admissions due to respiratory diseases using environmental and meteorological data from São Paulo (2017-2022). It analyzed hospitalizations classified under ICD-10 codes J00-J99. Weekly public data on air pollutants (PM2.5, PM10, O3, NO2, SO2, CO) and climate variables (temperature, humidity, precipitation, among others) were used. The methodology included feature engineering (lags, moving averages, interactions, trend, seasonality), Lasso regression for variable selection, and application of the CatBoost algorithm optimized via GridSearchCV. The model showed strong performance (R²≈0.895), with good accuracy in estimating healthcare demand. The Shapley Additive Explanations (SHAP) technique ensured model explainability, identifying the most influential predictors of respiratory admissions. The results highlight the potential of AI as a strategic Digital Health tool, especially for early outbreak detection and resource allocation within Brazil's Unified Health System (SUS).

PMID:42385076 | DOI:10.1590/1413-81232026315.23522025