Predicting Atherogenic index of plasma risk from long-term spatiotemporal air pollution exposure using an attention-based deep learning

• Air pollution, especially PM2.5, increases cardiovascular disease risk. • An AGRU model predicts AIP risk and its link to air pollution exposure. • Data combine 86 EPA stations (2005–2024) with the Taiwan Biobank cohort. • AGRU outperforms six ML models and one DL model in multiple metrics. • AGRU achieves 98.17 % accuracy, aiding precision and preventive medicine. Cardiovascular disease remains a leading cause of mortality worldwide. Accumulating epidemiological and toxicological evidence indicates that exposure to air pollution, particularly fine particulate matter (PM 2.5 ), significantly elevates the risk of cardiovascular events through multiple biological mechanisms, including autonomic dysregulation, endothelial dysfunction, systemic inflammation, and activation of coagulation pathways. In this study, we developed an Attention-Gated Recurrent Unit (AGRU) model to predict the risk of the atherogenic index of plasma (AIP) and to investigate its association with ambient air pollutant exposures. Air quality data were obtained from 86 Environmental Protection Administration (EPA) monitoring stations across Taiwan from 2005 to 2024 and were linked with individual-level data from the Taiwan Biobank. The predictive performance of the proposed AGRU model was benchmarked against six machine learning algorithms: Random Forest, Gradient Boosted Decision Tree (GBDT), CatBoost, eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), and HistGradientBoosting, as well as a conventional deep learning model, the Gated Recurrent Unit (GRU). Model performance was evaluated using accuracy, precision, recall, and F1-score. The AGRU model achieved the highest overall accuracy (98.2%), outperforming all comparator models, particularly tree-based approaches and SVM. Importantly, the attention-based deep learning framework demonstrated clear advantages in capturing long-term exposure dynamics and in identifying influential features. Collectively, these findings enhance the mechanistic understanding of cardiovascular risk associated with environmental exposures and provide a robust scientific basis for precision medicine strategies aimed at early prevention and targeted health management in high-risk populations.