Machine Learning Predicts Childhood Asthma Risk from Eczema in New EMJ Study

Machine learning models can predict which children with early-onset eczema are most likely to develop persistent asthma and allergic rhinitis by school age, according to a study published in the Journal of Allergy and Clinical Immunology.

The research, led by Wansu Chen, Ph.D., from Kaiser Permanente Southern California, analyzed electronic health record data from 10,688 children diagnosed with atopic dermatitis before age 3. The study developed and validated machine learning models to forecast individualized risk for moderate-to-severe persistent asthma and allergic rhinitis in children aged 5 to 11 years.

For asthma prediction, the comprehensive model demonstrated strong discrimination with an area under the curve (AUC) of 0.893, while a simplified version achieved an AUC of 0.892. At a 95 percent specificity threshold, the comprehensive model reached 40.4 percent sensitivity and 39.3 percent positive predictive value (PPV), and the simplified model showed 36.2 percent sensitivity and 33.8 percent PPV.

Rhinitis prediction models showed moderate performance, with the comprehensive model achieving an AUC of 0.793 and the simplified model 0.773. At 90 percent specificity, the comprehensive model delivered 35.5 percent sensitivity and 72.7 percent PPV, while the simplified model yielded 34.0 percent sensitivity and 69.2 percent PPV.

The researchers noted acceptable calibration across models, with strong agreement among the highest-risk groups. They emphasized that integrating such prediction tools into clinical workflows could help providers identify children at elevated risk and prioritize them for interventions such as environmental control, allergist evaluation, or early initiation of preventative therapy.

Atopic dermatitis, a common form of eczema, is recognized as an early step in the “atopic march,” a sequence where allergic conditions often progress from skin inflammation to respiratory diseases like asthma and allergic rhinitis. This study supports the use of data-driven approaches to interrupt that trajectory through timely, targeted care.