Machine learning algorithms for predicting the risk of developing non-protective antibodies in pediatric co-immunization following hepatitis B exposure

OBJECTIVE The present study utilized data from the Information Management System of Prevention of Mother-to-Child Transmission of HIV, Syphilis, and Hepatitis B in Guangdong Province to construct and compare a machine learning (ML) algorithm for predicting the risk of non-protective antibody response in hepatitis B-exposed children following combined immunization. METHODS The information reported by all midwifery institutions in 13 prefectures and cities was selected. Data reported in the Information Management System for Prevention of Mother-to-Child Transmission of HIV, Syphilis, and Hepatitis B in Guangdong Province between January 1, 2020, and December 31, 2021, were divided at random into a training cohort and a validation cohort (7:3). We employed six machine learning (ML) techniques to construct predictive models for the development of non-protective antibodies following combined immunization in children exposed to HBV,including multilayer perceptron model (MLP), support vector machine (SVM), K-nearest neighbor algorithm (KNN), random forest (RF), decision tree (DT), and naive bayes (NBC). The hyperparameters that best fit the model were determined through internal five-fold cross validation, while external five-fold cross validation was employed to identify machine learning models with superior average performance. Subsequently, the machine learning model exhibiting the highest average performance was selected as the final model for external validation. The comprehensive evaluation of the machine learning model encompassed ROC curves, AUC size comparisons, accuracy, and other relevant metrics. CONCLUSION It was discovered that the RandomForest model was a more powerful algorithm in terms of prediction. The objective of this model is to enhance future risk decision making in relation to the development of non-protective antibodies following combined immunization in children exposed to HBV,The reporting region emerged as the most significant predictor.