Asian Journal of Dental Sciences

Caries risk assessment (CRA) is intended to estimate an individual's likelihood of developing new caries lesions or experiencing lesion progression over a defined period so that preventive, diagnostic and recall strategies can be tailored. Contemporary CRA is changing rapidly. Established approaches such as Cariogram, Caries Management by Risk Assessment (CAMBRA), professional checklists and risk-based care pathways are increasingly being supplemented by statistical prediction models, machine learning, digital self-assessment, salivary and microbiome biomarkers, and population-specific tools. This critical narrative review evaluates whether these developments have improved prognosis rather than merely increased technical complexity. Literature published principally from 2015 to 8 August 2026 was examined, while earlier seminal studies were retained when necessary to establish concepts and validation history. The synthesis distinguishes future-event prediction from cross-sectional detection or classification, because this distinction is frequently blurred in the recent literature. The strongest and most reproducible conventional predictor remains previous or current caries experience, while social conditions, fluoride exposure, plaque, diet, salivary function and life-course factors modify risk. Established multifactorial tools can structure clinical reasoning and risk-linked care, but external validation remains uneven and evidence that their use itself improves patient outcomes is limited. Machine-learning studies often report promising discrimination, yet many rely on cross-sectional outcomes, internal validation, small samples or incomplete calibration and clinical-utility assessment. Prospective cohorts demonstrate that simple, readily collected predictors can perform competitively, challenging assumptions that algorithmic complexity necessarily improves care. Salivary, microbiome and genetic approaches remain biologically informative but are not yet sufficiently standardised or externally validated for routine prognostic use. The most important recent trend is therefore methodological rather than technological: CRA is moving towards explicit prediction horizons, transparent validation, calibration, decision-linked utility, population transportability and repeated reassessment. Future progress will depend on prospective multicentre validation and impact studies that compare complex models against parsimonious clinical baselines and test whether risk-guided decisions improve outcomes without widening inequalities.