Development and Validation of a Machine Learning Model for Early Prediction of Peri-Implantitis: Toward Personalized Risk Stratification in Implant Dentistry

Peri-implantitis is one of the most common complications after dental implant surgery, as approximately 20-47 percent of the recipients of the implants develop peri-implantitis complications. Machine learning (ML) would provide an opportunity to predict early and allow timely intervention and individual care. Purpose: The purpose of the study was to create and test the predictive model of peri-implantitis based on clinical, demographical, and radiographic data of 40 patients during a 12-month follow-up period with the use of ML. Methods and Materials: The retrospective cohort included 40 patients that were implant surgically operated. The extracted parameters were clinical (probing depth, bleeding on probing, plaque index, bone loss), patient demographics, systemic conditions, and radiographic features. When 5-fold cross-validation was used, several ML algorithms were trained using Logistic Regression, random forests, Support Vector Machine (SVM), gradient boosting, and Artificial Neural Network (ANN). The area under the receiver operating characteristic curve (AUC), sensitivity, specificity and accuracy have been used to evaluate model performance. Findings: Gradient Boosting model resulted in the best AUC of 0.91 (95% CI: 0.84 -0.97), then the Random Forest (AUC = 0.88) and ANN (AUC = 0.86). Marginal bone loss at 6 months, smoking, probing pocket depth, and diabetes mellitus were the major predictive characteristics. Conclusion: MLs, especially Gradient Boosting, are highly predictive of peri-implantitis and can be incorporated into the clinical decision-support systems to stratify risks individually in implant dentistry.