Trajectory Analysis of Pediatric Longitudinal Allergen Sensitization Patterns From Real World Data

Area of Science:

• Allergy and Immunology
• Computational Biology
• Pediatric Medicine

Background:

• Allergen sensitization patterns are crucial for understanding allergic disease development.
• Longitudinal tracking of sensitization is key to predicting disease trajectories.
• Machine learning offers novel approaches to analyze complex sensitization data.

Purpose of the Study:

• To investigate the link between allergen sensitization patterns and allergic disease diagnosis.
• To identify distinct sensitization clusters using unsupervised learning.
• To develop a predictive model for allergic disease based on early sensitization profiles.

Main Methods:

• Retrospective cohort study using electronic health records from a common data model.
• Latent class analysis and Gaussian mixture models for cross-sectional and longitudinal sensitization clusters.
• Categorical Boosting multilabel classification with SHapley Additive exPlanations for predictive modeling.

Main Results:

• Four cross-sectional sensitization clusters and nine longitudinal trajectories were identified.
• Early house dust mite (HDM) sensitization strongly predicted asthma.
• Sensitization trajectories showed associations with eczema, food allergies, and rhinitis, with varying predictive power.

Conclusions:

• Longitudinal sensitization patterns, especially early HDM dominance, are predictive of asthma.
• Machine learning models can identify children at high risk for allergic diseases.
• Findings support personalized strategies for allergy prevention based on sensitization history.