Machine learning models showed roughly 40% improvement in diagnostic accuracy compared to
standard oral food challenges, skin prick tests, and allergen-specific IgE measurements
MILWAUKEE, Feb. 14, 2026 /PRNewswire/ — Both machine learning and deep learning AI models show significant improvements over existing clinical criteria of food allergy diagnostics, according to new research being presented at the 2026 AAAAI Annual Meeting.
“The current standard of care for food allergy diagnosis relies on skin prick testing, allergen-specific IgE and oral food challenges in the case of inconclusive results. Artificial intelligence machine learning (ML) models showed 40% improvement in diagnostic accuracy over existing clinical criteria, and more advanced deep learning (DL) models further improved diagnostic performance over ML methods, with a 10-15% improvement in area under the curve. Diagnostic methods for food allergy are enhanced by ML/DL and have the potential to outperform current strategies and improve standard of care” said lead author McKenzie J. Williams, Howard University Karsh STEM Scholar.
In the study, researchers trained machine learning (ML) and deep learning (DL) convolutional neural networks (CNN) on skin prick test (SPT) measurements, allergen-specific IgE (sIgE) and serum component proteins including peanut (PN)-IgE rAra h 1,2,3,6; PN-IgG4 rAra h 1,2,3, 6, that were collected as part of the IMPACT trial at the time of the baseline 146 peanut oral food challenges (OFCs). The trial included children aged 1-to-4 years old.
Within the study, algorithmic performance showed the strong predictive value of PN-sIgE Ara h2 and PN-IgE/IgG4 (sensitivity: 88.9; specificity: 84.5; positive predictive value (PPV): 89). ML models showed notable improvement over existing clinical criteria, with about 40% improvement in diagnostic accuracy, according to the researchers. The use of more advanced DL models showed improved diagnostic performance over ML methods, with a 10-15% improvement in the area under the curve. As a result, DL models trained with tests for standards of care and were able to greatly improve sensitivity and PPV while being non-inferior to diagnostic methods used in practice.
The researchers suggest that this improvement in diagnostic performance for OFC biomarker discovery can be used to develop a diagnostic alternative for food allergy that is scalable and more efficient than the standard OFC, skin prick tests and allergen-specific IgE (sIgE) measurements.
Visit aaaai.org to learn more about food allergy. Research presented at the 2026 AAAAI Annual Meeting, February 27 – March 2 in Philadelphia, PA, is published in an online supplement to The Journal of Allergy and Clinical Immunology (JACI).
The American Academy of Allergy, Asthma & Immunology (AAAAI) is the leading membership organization of more than 7,100 allergists, asthma specialists, clinical immunologists and other professionals with a special interest in the research and treatment of allergic and immunologic diseases. Established in 1943, the AAAAI is the go-to resource for patients living with allergies, asthma and immune deficiency disorders.
View original content to download multimedia:https://www.prnewswire.com/news-releases/food-allergy-diagnostics-are-enhanced-by-machine-learning-and-deep-learning-ai-models-302687311.html
SOURCE American Academy of Allergy, Asthma & Immunology (AAAAI)
