Artificial intelligence risk prediction model for common respiratory pathogens in China based on heterogeneous multi-source clinical and geographic data: A modeling study.

Abstract

Most respiratory pathogens exhibit distinct seasonal and periodic outbreak patterns driven by climatic factors. However, predictive models that jointly consider climate, air quality index (AQI), and socioeconomic variables are lacking. We retrospectively analyzed targeted or metagenomic next-generation sequencing data from 153,544 respiratory samples collected from 1,880 centers across 30 provinces in China between September 2022 and September 2024. Monthly positivity rates were matched with geographic, climatic, AQI, and GDP data. CO(0.098 ± 0.016), HCHO(0.096 ± 0.021), O3(0.102 ± 0.019), sunshine hours(0.103 ± 0.028), wind speed(0.114 ± 0.024), and GDP(0.095 ± 0.019). were identified as the key geographical factors for the positivity across most respiratory pathogens via mean Gini index reduction, and a gradient boosting decision tree(GBDT) model was trained and benchmarked against other AI methods using the DISO metric. This model accurately simulated the epidemiological trends from September 2022 to September 2024 and outperformed alternative models with the lowest DISO metric of 0.12 in influenza A, 0.21 in SARS-CoV-2, 0.25 in RSV. The GBDT model was used to predict the short-term epidemic of 10 respiratory pathogens between October and December 2024. The predictions showed consistent trends with the external validation cohort for RNA viruses including SARS-CoV-2 and influenza A virus, but differed for bacterial pathogens. Integrating air quality, climatic, and socioeconomic data yields robust predictions of respiratory infection dynamics in the short-term by the GBDT model, bolstering public health surveillance and offering a framework potentially applicable to other infectious diseases.