Artificial intelligence platform to predict children's hospital care for respiratory disease using clinical, pollution, and climatic factors.

BACKGROUND

Hospitals and health care systems may benefit from artificial intelligence (AI) and big data to analyse clinical information combined with external sources. Machine learning, a subset of AI, uses algorithms trained on data to generate predictive models. Air pollution is a known risk factor for various health outcomes, with children being a particularly vulnerable group.

METHODS

This study developed and validated an AI-based platform to forecast paediatric emergency visits and hospital admissions for respiratory diseases, using clinical and environmental data in the Metropolitan Area of São Paulo, Brazil. We applied XGBoost, a tree-based machine learning algorithm, to predict hospital use at Sabará Children's Hospital, incorporating clinical, pollution, and climatic variables.

RESULTS

We analysed 24 366 emergency department visits and 2973 hospital admissions for respiratory diseases International Classification of Diseases, 10th Revision, Chapter J (ICD-10 J), excluding COVID-19, from January to December 2022. Only geocoded cases within the spatial accuracy thresholds of the study were included. Logistic regression revealed that outpatient visits were associated with higher particulate matter with a diameter of 10 µm or less (PM) concentrations near children's residences on the day of hospital arrival. In contrast, admissions were linked to lower relative humidity, particularly on drier days. Additional associations were found between admissions and the spring season, as well as male sex.

CONCLUSIONS

We developed a platform that integrates clinical and environmental databases within a big data framework to process and analyse information using AI techniques. This tool predicts daily emergency department and hospital admission flows related to paediatric respiratory diseases. The algorithms can distinguish whether a child arriving at the emergency department is likely to be treated and discharged or will require hospital admission. This predictive capability may support hospital planning and resource allocation, particularly in contexts of environmental vulnerability.