채우리 (서울대학교 의과대학)

조주연 (서울대학교 의과대학)

채우리 (서울대학교 의과대학)

김태범 (서울아산병원)

조주연 (서울대학교 의과대학)

Biologic therapies for asthma have emerged, but their high cost necessitates identifying responsive patients before use. Reliable biomarkers for predicting treatment response, however, remain limited. This study aimed to identify and validate predictive biomarkers of biologic response in asthma patients enrolled in the prospective PRISM cohort in Korea using an untargeted metabolomics approach. A total of 275 adult asthma patients were assigned to a discovery set, and baseline plasma samples were analyzed. For candidate biomarkers, quantified levels were further evaluated in the dupilumab group in both the discovery (n=150) and validation (n=65) sets. Logistic regression and four machine learning-based models for response prediction were developed and validated. In the dupilumab group, responders (n=121) had significantly higher levels of 1-methylnicotinamide (1-MNA) compared to non-responders (n=22), while no significant differences were observed in other treatment groups. Both logistic regression and random forest models outperformed single-variable models based on conventional clinical parameters for dupilumab treatment. The elevated baseline 1-MNA levels in dupilumab responders suggest a pre-existing anti-inflammatory state favorable for Th2 inflammation regulation. Further studies are needed to clarify the direct relationship between 1-MNA and Th2 immune pathways. These findings may help advance precision medicine and reduce medical costs for asthma patients.​