Background: Asthma is associated with an increased risk of acute respiratory infections (ARI). Little is known about whether natural language processing (NLP)-powered digital biomarkers can identify a high-risk asthma subgroup for ARI during early childhood.
Objective: We assessed whether a digital biomarker could identify a high-risk subgroup of childhood asthma for ARI.
Methods: We applied validated NLP algorithms for Predetermined Asthma Criteria (NLP-PAC) and Asthma Predictive Index (NLP-API) to electronic health records of the 1997-2016 Mayo Clinic Birth Cohort. We categorized the cohort into 4 subgroups: both criteria positive (NLP-PAC+/NLP-API+), PAC positive only (NLP-PAC+), API positive only (NLP-API+), and both criteria negative (NLP-PAC-/NLP-API-). We assessed the risk of 5 medically attended ARI (pneumonia, frequent group A streptococcal pharyngeal infection, Bordetella pertussis, influenza A/B, and respiratory syncytial virus infection) and asthma exacerbation defined by NLP algorithms at 3 years of age among the 4 subgroups. We also examined whether such associations emerged during the first 3 years of life.
Results: There were 22,370 eligible subjects (51% male and 81% White). The NLP-PAC+/NLP-API+ subgroup had the highest risk of pneumonia, influenza A/B, and asthma exacerbation compared to other groups. No significant differences were found in other ARI. The same subgroup had the highest occurrence of pneumonia, influenza A/B, and respiratory syncytial virus infection, compared to other groups, during the first 3 years of life.
Conclusion: NLP-PAC+/NLP-API+ can be a novel digital biomarker for a high-risk subgroup of childhood asthma for pneumonia, influenza A/B, and asthma exacerbation. This phenotype may emerge early in life.
Keywords: Asthma; NLP; artificial intelligence; cohort; epidemiology; infection; pediatric; phenotyping; precision; risk; subgroup.
Copyright © 2025. Published by Elsevier Inc.