Asthma can be predicted using non-biological measurements from the age of 3 years, according to one study.
Machine learning models can predict physician-diagnosed asthma in early childhood, a recent study found. Wheezing, atopy, antibiotic exposure, lower respiratory tract infections, and the child’s mother having asthma were the strongest markers of an asthma diagnosis.
This population-based cohort study is published in Pediatric Research.
“Childhood asthma is a chronic respiratory disease that often persists throughout an individual’s lifetime, imposing a significant burden on both patients and health care systems,” wrote the researchers of the study. “Despite numerous treatment options, there are currently no curative therapies available for asthma, and patients often require ongoing treatment to manage their symptoms and prevent exacerbations.”
The study aimed to better understand the relative timing and importance of early markers of asthma using machine learning models.
The researchers utilized data from the Canadian Health infant Longitudinal Development (CHILD) birth cohort, one of the largest ongoing cohort studies in Canada that gathered data on asthma and allergy research from mid-pregnancy through childhood. The study included pregnant women from multiple sites across 4 Canadian provinces between 2008 and 2012.
In the current study, only children with complete CHILD questionnaires and physician-diagnosed asthma from clinical visits at age 5 years were included. The dataset was stratified by asthma status and randomly split into a training dataset (85%) and a holdout dataset (15%) to assess the training and tuning machine learning models’ performance.
The training dataset included 1484 children, of which 1395 did not have asthma and 90 had asthma. The holdout dataset included 270 children, in which 250 did not have asthma and 20 had asthma. The test set was put aside, while the training set was used for model tuning and feature selection.
Additionally, the researchers compared the predictive performance of 5 different models—Logistic Regression, Random Forest, eXtreme Gradient Boost, Decision tree, and Support Vector Machine—to identify the best machine learning algorithms for predicting childhood asthma. Furthermore, 132 variables were used from 6 time points—birth, 6 months, 1 year, 2 years, 3 years, and 4 years—for each machine learning algorithm to predict physician-diagnosed asthma at the 5-year clinic visit.
The predictive models showed that early-life data at 1 year and younger had limited predictive ability for physician-diagnosed asthma at 5 years (area under the precision-recall curve [AUPRC] < 0.35). The earliest reliable prediction of asthma was achieved at 3 years (area under the receiver-operator curve [AUROC] > 0.90; AUPRC > 0.80).
Additionally, maternal asthma, antibiotic exposure, and lower respiratory tract infections remained highly predictive of asthma throughout childhood, while wheezing status and atopy were the most important predictors of early childhood asthma.
However, the researchers identified some limitations to their study. Because a large portion of the data used to develop the machine learning models were derived from questionnaires and clinical assessments completed by parents and clinicians, their findings may have been susceptible to bias. Therefore, the researchers believe that future studies should incorporate objective measures, such as biological and genetic markers. However, the researchers also acknowledged that these types of measurements are often costly, have time constraints, and have specialized equipment requirements, which may limit the generalizability of the model to a wider study population.
Despite limitations, the researchers believe this study showed the potential of using machine learning models to predict asthma diagnosis in young children, to identify early-life risk factors in the development of asthma.
“Importantly, our findings suggest that physician-diagnosed asthma at age 5 years could be reliably predicted using non-biological and non-genetic data by the age of 3 years, whereas accurate prediction before 1 year of age using existing clinical dataset was challenging,” wrote the researchers. “These results have significant implications for early detection and intervention strategies for asthma.”
Reference
He P, Moraes TJ, Dai D, et al. Early prediction of pediatric asthma in the Canadian Healthy Infant Longitudinal Development (CHILD) birth cohort using machine learning. Pediatr Res. Published online January 11, 2024. https://doi.org/10.1038/s41390-023-02988-2
Could On-Body Delivery of Isatuximab Bring More Competition to Anti-CD38 Myeloma Treatment?
June 6th 2025Results for IRAKLIA show noninferiority for Sanofi's on-body delivery system for isatuximab, compared with IV administration. Patients overwhelmingly preferred the hands-free delivery option.
Read More
ICS Use Tied to Fewer Exacerbations in Patients With Bronchiectasis and Elevated Blood Eosinophils
June 6th 2025Inhaled corticosteroid (ICS) use was common among patients with bronchiectasis and was associated with reduced exacerbations and hospitalizations in those with elevated blood eosinophil counts.
Read More
Real-World Data Support Luspatercept vs ESAs for Anemia in Lower-Risk MDS
June 5th 2025Patients with myelodysplastic syndrome (MDS) who received luspatercept showed greater hemoglobin gains and transfusion independence compared with erythropoiesis-stimulating agents (ESAs) in a real-world analysis.
Read More
At EHA 2025, Hematology Discussions Will Stretch Across Lifespans and Locations
June 5th 2025The 2025 European Hematology Association (EHA) Congress, convening virtually and in Milan, Italy, from June 12 to June 15, 2025, will feature a revamped program structure for the meeting’s 30th anniversary while maintaining ample opportunities to network, debate, and absorb practice-changing findings in hematology and oncology.
Read More