Exacerbation risk prediction in pediatric wheeze or asthma patients using machine-learning-based integrative omics analyses.

<bold>Introduction:</bold> Omics data has augmented the molecular understanding of phenotypes in asthma; however, it is unclear if integrating omics with clinical data can improve the accurate prediction of future exacerbation risks in asthma. <bold>Objective:</bold> To assess if integrating gene expression and nasopharyngeal microbiome data to clinical data improves the prediction of future exacerbation risk in children with wheeze or asthma. <bold>Methods:</bold> In the U-BIOPRED pediatric prospective cohort study, we integrated omics and clinical data from children with wheeze or asthma. We included preschoolers aged 1-5 years (n=103) and children aged 6-17 years (n=111) with mild-to-moderate wheeze (n=40), severe wheeze (n=63), mild-to-moderate asthma (n=37) or severe asthma (n=74) with clinical, gene expression, and microbiome data. Exacerbation was defined as an oral corticosteroid burst, urgent visit, or hospitalization for asthma during the 12-18 months after enrolment. Multiple imputation was used to address missing data for exacerbations and clinical variables. We compared the area under the curve (AUC) metrics for predicting exacerbation with vs. without omics data using Logistic Regression (LR), XGBoost (XGB), and Feedforward Neural Network (FFNN) models. <bold>Results:</bold> LR (AUC 0.82), XGB (AUC 0.79), and FFNN (AUC 0.81) achieved similar performance using clinical data only. Incorporating gene expression and microbiome data significantly boosted FFNN (AUC 0.94) and XGB (AUC 0.86) performance but not LR (AUC 0.81). <bold>Conclusion:</bold> Omics integration using gene expression and nasopharyngeal microbiome data added predictive value for future exacerbation risk in children with wheeze and/or asthma.