Development of Childhood Asthma Prediction Models using Machine Learning Approaches

ABSTRACT Background Wheeze is common in early life and often transient. It is difficult to identify which children will experience persistent symptoms and subsequently develop asthma. Machine learning approaches have the potential for better predictive performance and generalisability over existing childhood asthma prediction models. Objective To apply machine learning approaches for predicting school-age asthma (age 10) in early life (Childhood Asthma Prediction in Early life, CAPE model) and at preschool age (Childhood Asthma Prediction at Preschool age, CAPP model). Methods Data on clinical symptoms and environmental exposures were collected from children enrolled in the Isle of Wight Birth Cohort (N=1368, ∼15% asthma prevalence). Recursive Feature Elimination (RFE) identified the optimal subset of features predictive of school-age asthma for each model. Seven state-of-the-art machine learning classification algorithms were used to develop the models and the results were compared. To optimize the models, training was performed by applying 5-fold cross-validation, imputation and resampling. Predictive performances were evaluated on the test set and externally validated in the Manchester Asthma and Allergy Study (MAAS) cohort. Results RFE identified eight and 12 predictors for the CAPE and CAPP models, respectively. The best predictive performance was demonstrated by a Support Vector Machine (SVM) algorithm for both the CAPE model (area under the receiver operating curve, AUC=0.71) and CAPP model (AUC=0.82). Both models demonstrated good generalisability in MAAS (CAPE 8YR=0.71, 11YR=0.71, CAPP 8YR=0.83, 11YR=0.79). Conclusion Using machine learning approaches improved upon the predictive performance of existing regression-based models, with good generalisability and ability to rule in asthma.