Related Experiment Video
Updated: Jun 6, 2025
Digital Hybrid Model Preparation for Virtual Planning of Reconstructive Dentoalveolar Surgical Procedures
Published on: August 5, 2021
A personalized periodontitis risk based on nonimage electronic dental records by machine learning
Laura Swinckels1, Ander de Keijzer2, Bruno G Loos3
1Academic Centre for Dentistry Amsterdam (ACTA), University of Amsterdam and Vrije Universiteit Amsterdam, The Netherlands; Department Oral Hygiene, Faculty of Health, Sports and Welfare, Inholland University of Applied Sciences, Amsterdam, The Netherlands; Centre of Expertise Prevention in Health and Social Care, Inholland University of Applied Sciences, Haarlem, The Netherlands; Medical Technology Research Group, Inholland University of Applied Sciences, The Netherlands; Data Driven Smart Society Research Group, Inholland University of Applied Sciences, Alkmaar, The Netherlands.
Machine learning models can predict Periodontal Disease (PD) risk using electronic dental records (EDRs). This approach aids early detection and personalized prevention strategies for better patient outcomes.
Area of Science:
- Dental Informatics
- Machine Learning in Healthcare
- Periodontal Disease Research
Background:
- Periodontal Disease (PD) poses a significant public health challenge.
- Early detection and prevention are crucial for managing PD and improving patient outcomes.
- Non-image electronic dental records (EDRs) contain valuable data for risk prediction.
Purpose of the Study:
- To develop and evaluate a machine learning (ML) model for predicting Periodontal Disease (PD) risk.
- To utilize non-image EDRs for PD risk assessment.
- To identify key predictors of PD from EDR data.
Main Methods:
- A Random Forest ML model was trained using EDRs from 43,331 US dental patients.
- Patients were classified as cases or controls based on PD diagnosis, treatment, and pocketing.
- Model performance was assessed using accuracy, sensitivity, specificity, and AUROC; feature importance was determined.
Main Results:
- The Random Forest model achieved high sensitivity (81%) and an excellent AUROC (94%) on the development set.
- Key predictors included bleeding proportion, age, visit frequency, prior preventive care, smoking, and drug use.
- The model demonstrated strong case detection (91%) on the validation set, though specificity was lower (0.54).
Conclusions:
- ML models show excellent potential for early PD detection and prevention using consistent EDRs.
- Clinical application can provide personalized risk predictions and guide targeted interventions.
- Further research is needed for model validation and improved EDR documentation.
