Machine learning for periodontitis classification using non-imaging multidimensional data: a systematic review and meta-analysis

To systematically review machine learning (ML) models for periodontitis classification using non-imaging multidimensional predictors and evaluate their performance and explainability. A thorough search was conducted across Ovid MEDLINE, Web of Science, PubMed, and Scopus until December 2025. Studies were screened according to the inclusion and exclusion criteria, and the quality of all eligible studies was evaluated using the Prediction Model Risk of Bias Assessment Tool with the AI extension (PROBAST + AI) tool. An exploratory meta-analysis of a classification test was performed using hierarchical modelling frameworks. Fourteen studies were included in the systematic review, of which seven studies were included in the meta-analysis. The hierarchical summary receiver operating characteristic (HSROC) model was used as the primary synthesis approach because of a substantial threshold effect. The area under the SROC curve was 0.81 (95% CI : 0.78–0.84), indicating moderate overall discriminative performance. ML models using non-imaging multidimensional data suggested moderate overall discriminatory performance in HSROC analysis for periodontitis classification. However, this exploratory finding should be interpreted with caution, given the limited number of studies, heterogeneous reference standards and high or unclear risk of bias in the analysis domain. Therefore, current evidence remains insufficient to support clinical implementation.