Predicting parental Human papillomavirus vaccine hesitancy: development and internal validation of machine learning models

The study aimed to develop machine learning models to predict parental hesitancy toward Human papillomavirus (HPV) vaccination in Cameroon, where cervical cancer is the second most common cancer among women. The research included data from 1,156 parents of children aged 9-18 years from Buea Health District. Three models were tested: logistic regression, random forest, and XGBoost. All models achieved comparable discrimination with AUC-ROC values of 0.841-0.870. Logistic regression showed the highest sensitivity of 73.1% and the lowest Brier score of 0.1435. The main factors predicting hesitancy were insufficient vaccine information, perceived vaccine safety concerns, and distrust in the Ministry of Health. The results indicate that hesitancy is more influenced by access to information and institutional trust than by sociodemographic disadvantage.