Predictive Accuracy of Caries Risk Assessment: A Systematic Review.

OBJECTIVES: To evaluate the predictive accuracy of single-factor, multi-factor, and machine learning-based caries risk assessment (CRA) methods in predicting caries risk among children and adults, updating the 2015 review. DATA: The review was reported in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020, supplemented by relevant items adapted for prognostic accuracy reviews. Paired sensitivity and specificity data were extracted, and 2 × 2 contingency tables were reconstructed where necessary. SOURCE: PubMed, Web of Science, Cochrane Library, Embase, and Scopus were searched from February 2015 to July 2025. STUDY SELECTION: Studies evaluating CRA methods with a minimum 1-year follow-up and sufficient data to reconstruct a 2 × 2 table were included. Of the 7,332 records identified, 30 studies met the inclusion criteria; 16 contributed to the meta-analysis. RESULTS: The full Cariogram in children demonstrated pooled sensitivity 72.10% and specificity 69.96%. Caries Management by Risk Assessment (CAMBRA) yielded a sensitivity of 67.65% and a specificity of 58.96% with substantial heterogeneity. The American Academy of Paediatric Dentistry (AAPD) CRA showed high sensitivity (95.56%) but very low specificity (5.12%), suggesting it may be more suitable for initial screening than for definitive risk stratification. Past caries experience demonstrated moderate predictive accuracy. Machine learning showed potential but lacked sufficient data for pooling. CONCLUSIONS: CRA methods demonstrate moderate predictive accuracy, with substantial variability across populations, tools, and study designs. CLINICAL SIGNIFICANCE: CRA methods should be used alongside clinical judgement. Further prospective validation and incorporation of objective biomarkers may improve future caries risk prediction.