Machine learning-based identification of multifactorial asthma phenotypes in Korean adults: insights from the Korea National Health and Nutrition Examination Survey

Asthma is a heterogeneous respiratory disorder influenced by genetic, environmental, and lifestyle factors. The present study aimed to identify distinct asthma phenotypes among Korean adults using a multifactorial analytic framework that integrates machine learning with epidemiological modeling based on nationwide health data. Data from 198,964 participants (3,529 with physician-diagnosed asthma) in the 8th Korea National Health and Nutrition Examination Survey (KNHANES VIII, 2019) were analyzed. Cluster analysis and machine learning algorithms (Gradient Boosting and Random Forest) were applied to classify asthma phenotypes using clinical, lifestyle, and nutritional variables including BMI, FEV₁%, smoking, stress level, and nutrient intake. Odds ratios (ORs) were computed to determine associations between demographic, environmental, and nutritional factors and asthma risk within each cluster. Four distinct asthma phenotypes were identified: (1) higher BMI with smoking and higher atopy prevalence, (2) higher smoking prevalence with relatively low atopy prevalence, (3) higher FEV₁% with higher allergy prevalence, and (4) lower FEV₁% with lower smoking, atopy, and allergy prevalence. Machine learning models achieved over 95% accuracy in asthma prediction. High stress and urban residence were associated with higher odds of asthma in specific clusters, while associations with n-3 fatty acids, thiamine, vitamin D, and dietary fiber varied across clusters. This study provides evidence for the multifactorial heterogeneity of asthma in the Korean adult population and highlights the potential of machine learning–based phenotyping for personalized asthma management. Integration of clinical, environmental, and nutritional factors may improve targeted prevention and treatment strategies in precision respiratory medicine.