Component-resolved diagnosis in childhood and prediction of asthma in early adolescence: A birth cohort study

IntroductionComponent-resolved diagnosis (CRD) has been decisive in exploring the mechanisms of IgE sensitization, but the predictive ability to detect asthma has not been addressed. We aim to develop and evaluate the performance of a personalized predictive algorithm for asthma that integrates information on allergic sensitization using CRD.MethodsOne thousand one hundred one twenty-five children from the Generation XXI birth cohort were randomly selected to perform a screening test for allergic sensitization and a subsample was characterized using CRD against 112 allergen components. Allergen components were analyzed using volcano plots and partial least squares (PLS) analysis. Logistic regression was performed to assess the associations between the obtained latent components (LC) and allergic outcomes (asthma, rhinitis, eczema) including other potential predictors used in previous asthma risk scores. The accuracy of the model in predicting asthma was assessed using Receiver Operating Characteristic (ROC) curve statistics.ResultsIn the PLS, the first LC was positively associated with asthma, rhinitis, and eczema. This LC was mainly driven by positive weights for Der p 1/2/23, Der f 1/2, and Fel d 1. The main components in the second LC were pollen and food allergens. History of early wheezing and parental allergy were included in the predictive model and the area under the curve improved to 0.82.ConclusionsThis is the first approach to improve the clinical applicability of CRD by combining CRD and clinical data to predict asthma at 13 years. Sensitization to distinct allergen molecules seems relevant to improve the accuracy of asthma prediction models.

Automated summary

This study aims to develop and evaluate a personalized predictive algorithm for asthma that combines component-resolved diagnosis (CRD) with clinical data, and uses information on allergic sensitization to improve the accuracy of asthma prediction models.