Systemic Periodontal Risk Score Using an Innovative Machine Learning Strategy: An Observational Study.

Area of Science:

• Oral Health and Periodontology
• Machine Learning in Healthcare
• Predictive Diagnostics

Background:

• Early diagnosis of periodontal diseases is vital to prevent tooth loss and systemic health complications.
• Current diagnostic methods may not fully capture individual risk profiles for developing periodontal diseases.
• Non-invasive predictors are desirable for routine clinical assessment of periodontal health.

Purpose of the Study:

• To develop a personalized, explainable machine learning (ML) algorithm for predicting periodontal disease risk.
• To utilize solely non-invasive, easily collectible clinical predictors for risk assessment.
• To identify individuals susceptible to periodontitis and gingival inflammation for timely intervention.

Main Methods:

• A cohort of 532 subjects had their individual data and periodontal health assessed.
• A machine learning pipeline incorporating feature selection, multilayer perceptron, and SHapley Additive exPlanations (SHAP) was employed.
• The algorithm was trained to predict healthy periodontium, periodontitis, and gingival inflammation.

Main Results:

• The ML algorithm achieved F1-scores of 0.74 for healthy periodontium and 0.68 for periodontitis.
• Gingival inflammation prediction yielded a lower F1-score of 0.32.
• Key predictors identified include age, body mass index, smoking, systemic diseases, diet, alcohol, education, and hormonal status.

Conclusions:

• The developed ML algorithm effectively identifies individuals at risk for periodontal diseases using non-invasive data.
• The model reveals distinct risk profiles based on age, particularly around 35 years, indicating transition periods for disease susceptibility.
• This approach offers a novel strategy for periodontal health prediction, integrating ML and explainability for personalized risk assessment.