Predicting plaque-gingivitis risk in schoolchildren using an interpretable machine learning model: a cross-sectional...

Area of Science:

• Pediatric Dentistry
• Machine Learning in Healthcare
• Public Health

Background:

• Plaque-induced gingivitis is a common oral health issue in schoolchildren.
• Accurate risk prediction is crucial for effective prevention strategies.
• Interpretable machine learning (ML) offers a novel approach to analyze risk factors.

Purpose of the Study:

• To develop an interpretable ML model for predicting gingivitis risk in schoolchildren using questionnaire data.
• To identify and explain key risk factors associated with plaque-induced gingivitis.
• To assess the generalizability and scalability of the developed ML model.

Main Methods:

• Multi-stage cluster random sampling of 1755 children (aged 6-12) in Lanzhou.
• Data collection via questionnaires and clinical dental examinations.
• Feature selection using LASSO regression and model development with six ML algorithms (RF, LightGBM, LR, XGBoost, DT, KNN).
• Model performance evaluation using AUC, sensitivity, specificity, accuracy, precision, F1-score, and decision curve analysis.
• Risk factor interpretation using SHapley Additive exPlanations (SHAP).

Main Results:

• 51.3% of children were diagnosed with plaque-induced gingivitis.
• The Random Forest (RF) model demonstrated the highest performance (training AUC: 0.991; testing AUC: 0.909; external validation AUC: 0.824).
• SHAP analysis identified brushing frequency, age, regular dental checkups, brushing time, gingival bleeding, and annual income as key predictors.

Conclusions:

• An interpretable RF model accurately predicts gingivitis risk based on self-reported factors.
• This ML-driven approach can reduce reliance on clinical examinations for pediatric gingivitis prevention.
• The model supports scalable and resource-efficient gingivitis prevention in various settings.