BACKGROUND: Allergic diseases often develop jointly during early childhood. Potential disease trajectories and relevant early-life factors have been described, yet existing prediction approaches mostly focus on single allergic diseases cross-sectionally. Models addressing allergic multimorbidity and disease trajectories are lacking. We aim to predict allergic disease trajectories from birth up to adolescence using early-life factors. METHODS: Preceding research using data from 4646 adolescents of the German birth cohorts GINIplus and LISA identified seven allergic disease trajectories up to the age of 15 years. A set of predictors comprising parental and perinatal factors, early allergic or respiratory symptoms, lifestyle and environmental factors was used with an XGBoost machine learning approach to perform multiclass classification. In a subsample (N = 2109), polygenic risk scores (PRS) for asthma, allergic rhinitis, atopic dermatitis, and any allergy were added to the predictor set. RESULTS: Our approach revealed moderate classification success (multiclass area under the curve (AUC) = 0.69). A macro-averaged sensitivity of 0.26 and specificity of 0.89 were obtained. The most important predictors were early-life skin rash, respiratory symptoms, and air pollution. In the sub-analysis, the PRS were among the factors with high importance, but the prediction performance in external test data was not improved. CONCLUSIONS: Our prediction success was comparable to established prediction scores while accounting for multiple allergic disease trajectories and using solely early-life factors. This study cannot yet provide reliable individual-level prediction in a clinical setting but can inform development of future work on this.
Keywords
