Machine Learning-Based Risk Stratification Models Incorporating Systemic Health Factors for Identification of Clinically Significant Periodontitis

Abstract

To develop and internally validate machine learning–based risk stratification models integrating periodontal and systemic health variables to identify patients with clinically significant (moderate-severe) periodontitis, and to evaluate their discrimination, calibration, interpretability, and potential adjunctive clinical utility.  An observational analytical study was conducted using clinical, radiographic, and systemic health data from 268 systemically compromised patients. Multiple supervised machine learning algorithms, including logistic regression, random forest, and gradient boosting, were developed to classify periodontal disease status as a binary diagnostic outcome. Internal validation was performed using repeated cross-validation and an independent test dataset. Diagnostic performance was assessed using area under the receiver operating characteristic curve (AUC-ROC), accuracy, sensitivity, specificity, and Brier score. Calibration plots and decision curve analysis were used to evaluate reliability and potential clinical utility. Model interpretability was assessed using Shapley Additive Explanations (SHAP). Moderate to severe periodontitis was identified in 49.3% of patients. Among the evaluated models, the gradient boosting classifier demonstrated the highest diagnostic performance, with an AUC-ROC of 0.89, accuracy of 83.5%, sensitivity of 82.4%, specificity of 84.7%, and a Brier score of 0.13. Incorporation of systemic health variables improved diagnostic discrimination compared with models using periodontal variables alone. Explainable AI analysis identified clinical attachment loss, probing pocket depth, and duration of diabetes mellitus as key diagnostic contributors. Artificial intelligence-based diagnostic classification models integrating periodontal and systemic health data demonstrated high diagnostic accuracy, reliable calibration, and favourable clinical utility in identifying moderate to severe periodontitis among systemically compromised patients. These models may serve as adjunctive diagnostic decision support tools in periodontal care; however, external validation in independent populations is required before clinical implementation.