Huang, A

Abstract Rationale Asthma exacerbations remain a leading cause of emergency visits and impaired quality of life. The 2025 Global Initiative for Asthma (GINA) report emphasizes early identification of Type 2 inflammation, treatable traits, and comorbidities to prevent attacks. Machine learning (ML) was applied to population level data to uncover complex, nonlinear relationships between biomarkers, lifestyle factors, and exacerbation risk. This study aimed to identify key predictors of asthma exacerbations using explainable ML techniques and to interpret these predictors through the lens of GINA 2025 guidelines. Methods A cross-sectional analysis was performed using the National Health and Nutrition Examination Surveys (NHANES) 2017-2020. Adults aged ≥18 years with available asthma history and laboratory data (n = 7,922) were included. The outcome was self-reported asthma exacerbation within the past 12 months. Univariable logistic regression was used to preselect covariates from 680 demographic, dietary, laboratory, and anthropometric variables using a significance threshold of p < 0.0001. Selected covariates were then entered into an XGBoost classifier. Model discrimination was assessed by the area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Feature importance was ranked by the Gain metric, and Shapley Additive Explanations (SHAP) were applied to improve interpretability and align results with GINA 2025 clinical domains. Results The final model included 55 covariates and achieved an AUROC of 0.737, sensitivity of 0.960, and NPV of 0.967 for predicting exacerbations. The most influential features were octanoic acid intake, eosinophil percentage, hip circumference, standing height, and high-sensitivity C-reactive protein (hs-CRP). Individuals reporting exacerbations had higher eosinophil percentages (3.36% vs. 2.76%), larger hip circumference (115.2 cm vs. 107.2 cm), and higher hs-CRP (6.08 vs. 4.00 mg/L). These findings correspond to GINA 2025’s emphasis on eosinophils as biomarkers for anti-inflammatory therapy, systemic inflammation as a treatable trait, and obesity as a modifiable risk factor for symptom control and exacerbation prevention. Conclusion Explainable ML applied to NHANES data accurately identified physiologic and inflammatory predictors of asthma exacerbations that mirror the GINA 2025 framework. Eosinophil levels, systemic inflammation, and adiposity emerged as key treatable traits linked to exacerbation risk. These findings demonstrate the utility of ML in translating population-level data into actionable insights consistent with GINA 2025’s strategy for individualized, inflammation-guided asthma management. This abstract is funded by: None