Machine Learning-Based Prediction of Food Allergy in Children Aged <2 Years with Atopic Dermatitis

Area of Science:

• Pediatric Allergy and Immunology
• Computational Biology and Bioinformatics
• Dermatology

Background:

• Children with atopic dermatitis (AD) face a higher risk of developing food allergy (FA).
• Early identification of FA in infants and toddlers with AD is crucial for timely intervention.
• Predictive tools can aid clinicians in managing this high-risk population.

Purpose of the Study:

• To develop and assess machine learning (ML) models for predicting FA in children under 2 years old with AD.
• To evaluate the performance of different ML algorithms in this prediction task.
• To identify key clinical and laboratory features contributing to FA prediction.

Main Methods:

• Retrospective analysis of electronic medical records for 435 children (1 month to 2 years) diagnosed with AD.
• Extraction of demographic data, clinical features, and laboratory findings.
• Evaluation of four ML algorithms: CatBoost, XGBoost, LightGBM, and logistic regression.

Main Results:

• The CatBoost model, using clinical features alone, achieved the highest AUC of 0.91, with 85% sensitivity and 91% specificity.
• XGBoost, using combined clinical and laboratory features, yielded an AUC of 0.88.
• Other models also showed promising predictive capabilities, highlighting the potential of ML in this domain.

Conclusions:

• Machine learning, particularly the CatBoost algorithm with clinical data, effectively predicts food allergy in young children with atopic dermatitis.
• These ML models can assist in identifying at-risk children who warrant further diagnostic workup for FA.
• The findings support the integration of ML tools into clinical practice for managing pediatric AD and FA.