Related Experiment Videos
A Commentary on AI-based Prediction Models in Periodontitis Progression
Fahad Umer1,2, Ayesha Nooruddin3, Aamna Khalid3
1Section of Dentistry, Department of Surgery, The Aga Khan University Hospital, Karachi, Pakistan. dr.fahadumer@gmail.com.
A Commentary On:
Furquim CP, Caruth L, Chandrasekaran G, et al. "Developing Predictive Models for Periodontitis Progression Using Artificial Intelligence: A Longitudinal Cohort Study" Journal of Clinical Periodontology, 2025;52:1478-1490 https://doi.org/10.1111/jcpe.14194 .
Study Design:
A prospective multi-center cohort study utilized data from previous work that evaluated periodontitis progression over a 12-month follow-up. Baseline demographic, clinical, and immunological variables were used to develop predictive models employing Logistic Regression (LR), Multi-Layer Perceptron (MLP), and Probabilistic Graphical Modelling (PGM) approaches.
Case Selection:
A total of 415 participants aged 25 years or older with at least 20 natural teeth were initially enrolled as part of the original study. Smokers, diabetic patients, pregnant or lactating women, individuals who had recently received periodontal therapy, antibiotics, or chronic anti-inflammatory medications, and those with incomplete data were excluded. Only participants who completed the 12-month follow-up and all bi-monthly assessments were included in the final analysis. Consequently, 273 participants remained, comprising 72 healthy/gingivitis individuals, 24 with Stage II periodontitis, and 177 with Stage III periodontitis.
Data Analysis:
Model performance was assessed using accuracy, Area Under the Receiver Operating Characteristic curve (AUROC), Area Under the Precision Recall Curve (AUPRC), sensitivity, specificity, and Brier scores. Feature importance and model interpretability were evaluated using SHAP (Shapley Additive Explanations) analysis.
Results:
PGM approach demonstrated the strongest overall performance. The best-performing model incorporated clinical variables, salivary IL-1β, age, and sex. LR showed moderate discrimination but poor classification balance, while the MLP demonstrated limited clinical usefulness. Among all predictors, the number of periodontal pockets ≥5 mm emerged as the most influential clinical feature, while IL-1β was the most important biological marker.
Conclusions:
This study provides encouraging evidence that Machine Learning (ML) approaches can improve the prediction of periodontitis progression. It also suggests the value of integrating biological markers with conventional clinical assessment.