D3.62 - Application of a logistic regression model to predict the risk of allergic reactions in children with salmon sensitization (Sal s 1)

Oral food challenges (OFC) for salmon allergy carry risks of severe systemic reactions. This study aimed to clinically validate our previously developed logistic regresion mathematical model that predicts clinical reactivity to salmon based on anamnesis and sensitization profiles, thereby optimizing the diagnostic pathway.

A clinical risk score (0-100 points) was calculated using a custom software application based on a 10-variable mathematical model. The model incorporates clinical and anamnestic predictors: age, sex, infant feeding type, parental allergy, atopic dermatitis, respiratory allergy, sensitization to other foods, as well as laboratory data (number of β-parvalbumin sensitizations, and Class 1 or 4 sensitization levels). A score ≥50 corresponds to a >95% probability of clinical symptoms. We retrospectively evaluated the model's performance on 65 pediatric patients: 47 with a confirmed history of severe systemic reactions to salmon and 18 with asymptomatic sensitization.

In the confirmed allergy group, 45 of 47 patients achieved a score >50 (sensitivity 95.7%; 95% CI: 85.5-99.5%), identifying them as high-risk and retrospectively avoiding the need for OFC. The remaining 2 patients (4.3%) fell into a diagnostic "grey zone" (<50 points) requiring OFC. In the asymptomatic group, 17 of 18 patients correctly scored <50 (specificity 94.4%; 95% CI: 70.7-99.8%). The model demonstrated higher discriminatory ability for severe phenotypes in this validation cohort compared to the initial training sample (sensitivity 86%, specificity 72%), though the confidence intervals overlap.

The implemented algorithmic tool demonstrates high diagnostic accuracy in real-world clinical settings. It safely identifies high-risk patients, significantly reducing the need for potentially dangerous OFCs. However, further validation in larger cohorts with a broader spectrum of clinical symptoms is warranted.