The prediction model of incident dental caries in 5-year-old children in Beijing
10.12016/j.issn.2096-1456.202660145
- Author:
REN Wen
1
;
ZHAO Mei
1
;
ZHANG Hui
1
;
CHEN Wei
1
;
SHEN Anqi
1
;
LIU Min
1
Author Information
1. Department of Preventive Dentistry, School of Stomatology, Capital Medical University
- Publication Type:Journal Article
- Keywords:
preschool children;
caries risk;
risk prediction;
stratified prevention and control;
prediction model;
regression analysis;
machine learning;
extreme gradient boosting;
oral health behavior
- From:
Journal of Prevention and Treatment for Stomatological Diseases
2026;34(10):982-992
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the influencing factors of incident caries at age 5 among caries-free 3-year-old children in Beijing, construct a machine learning prediction model based on oral health behavioral data, and provide a scientific basis and practical tool for early risk identification, stratified prevention and control, and home-kindergarten-medical collaborative intervention of incident dental caries in preschool children.
Methods:This study has been reviewed and approved by the Medical Ethics Committee, and written informed consent was obtained from the guardians of the study participants and their guardians. Children with complete baseline data at age 3 and follow-up data at age 5 from Beijing public oral health programs were included. Oral examinations were conducted according to World Health Organization diagnostic criteria, and questionnaires were completed by guardians. Risk factors were screened by Logistic regression. The total sample was randomly divided into a test set (20%) and a validation set (80%). Five-fold cross-validation was performed on the validation set for model training and tuning, and the final model performance was evaluated on the test set. Four machine learning models—decision tree, random forest, Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM)—were constructed. Model performance was evaluated using the area under the curve (AUC), accuracy, and specificity. Shapley Additive Explanations (SHAP) was used for model interpretation.
Results:The incidence of incident caries for 3 372 5-year-old children in Beijing was 50.96%. Multivariate logistic regression showed that parental caries, sleeping with milk, frequent feeding at night, no postprandial cleaning, high sugar intake, insufficient bedtime tooth brushing, and non-fluoride toothpaste use were significant risk factors (P < 0.05). Although all four machine learning models performed well, the XGBoost model performed the best (AUC = 0.806). Feature importance and SHAP analysis indicated that the frequency of bedtime tooth brushing, the frequency of candy consumption, and parental caries status were the top three predictors.
Conclusion:The XGBoost model based on oral health behaviors at age 3 can accurately predict the risk of incident caries in 5-year-old children in Beijing. Bedtime tooth brushing, sugar control, and parental oral health management are key points for caries prevention in preschool children.
- Full text:2026093010312308252北京市5岁儿童新发龋预测模型.pdf
