Machine learning models for dental caries risk prediction: a systematic review with risk-of-bias appraisal

Introduction: Dental caries is preventable, but its risk is multifactorial and hard to quantify in the clinic. We synthesized the performance and risk of bias of machine-learning models that predict caries risk or incidence from clinical, behavioral, and social data, appraised with PROBAST. Methods: We searched PubMed, OpenAlex, and Crossref for studies published from 2013 to 2025, screened 295 records after deduplication, and included prediction-model studies with a quantitative performance measure. Discrimination was the primary outcome, calibration and comparative performance secondary. Risk of bias was appraised with PROBAST. Results: Eleven studies were included, most in children and most using tree-based models. Discrimination was inconsistently reported; where the area under the curve was given, it ranged from 0.77 to 0.98, with the highest value coming from a model trained on simulated data and the lowest from a large national survey. Calibration was reported by only a minority of studies. PROBAST risk of bias was high or unclear in most studies, driven mainly by small samples, internal validation, and incomplete reporting of predictors and outcomes. Conclusions: Machine-learning models predict caries risk with reported discrimination that is highly variable and frequently overestimated by design, and the evidence is weakened by a high risk of bias, so these models are not yet ready to replace conventional risk assessment in routine practice.