Introduction: The use of dental caries indices in oral health surveys, such as the decayed, missing and filled teeth (DMFT) index, may mask heterogeneity in both disease severity and its distribution. This study aimed to identify age-specific clusters of dental caries across a 20-year period in Brazil, assess their temporal trends and examine their associations with perceived oral health. Methods: Secondary data from three Brazilian National Oral Health Survey (SB Brasil) conducted in 2003 (N = 70,119), 2010 (N = 29,681), and 2023 (N = 33,229) were analysed across four age groups: children (12 years), adolescents (15–19 years), adults (35–44 years), and older adults (65–74 years). Latent Profile Analysis (LPA) was applied to DMFT components to identify dental caries clusters. Multivariable ordinal and logistic regression models examined associations between caries cluster, self-rated oral health (SROH) and oral health-related quality of life (OHRQoL). Results: LPA identified five caries clusters in children, seven in adolescents, seven in adults, and four in older adults. The low-caries-experience cluster increased consistently across children, adolescents and adults over time. Severe untreated caries clusters declined in children and adolescents, and severe tooth loss clusters decreased in adults and older adults. Worse caries clusters were consistently associated with poorer SROH and OHRQoL across all age groups and survey years. These associations were generally stronger than those observed for DMFT and DMFT components. In older adults, dental caries clusters showed greater discriminatory capacity than DMFT components. Conclusion: The relationship between dental caries and perceived oral health is dynamic, age-dependent, and cluster-specific. LPA of DMFT components captured clinically meaningful heterogeneity beyond DMFT index and its components, offering a more nuanced understanding of caries severity and its impact on subjective oral health. These findings offer preliminary evidence supporting the complementary use of cluster-based approaches within dental caries indices for oral health surveillance.
Introduction: Although preventable, dental caries remains a global public health challenge due to its multifactorial nature and the deep influence of socio-structural determinants. This study evaluates the performance of artificial intelligence (AI) models as decision-support frameworks for population-level risk stratification, screening, and public health planning. Methods: We analyzed a representative sample (N = 23,726) from the SB Brasil 2023 national survey, stratified into three age cohorts: children (12 years), adolescents (15–19 years), and adults (35–44 years). Eleven predictors, encompassing socioeconomic (wealth score via PCA, household density, race, water access), psychosocial (self-rated oral health status), behavioral, and clinical domains, were used to predict caries presence (≥1 untreated decayed tooth) and severity (DMFT index regression). To account for overdispersion and the high frequency of zero counts, we utilized advanced count-based regressions, including Zero-Inflated Negative Binomial (ZINB) and Negative Binomial models. Performance was assessed using AUC, mean absolute error (MAE), and decision curve analysis (DCA) via 5-fold cross-validation. Results: Caries prevalence increased progressively with age (36.8–52.7%). The non-clinical triage models demonstrated significant net benefit in DCA across all groups, effectively identifying high-risk individuals for prioritization without requiring clinical examination. For severity prediction, the ZINB model identified structural signals of inequity, particularly in late adolescence (15–19 years), where socio-structural determinants explained a meaningful proportion of clinical variance (R2 = 0.1010; MAE = 2.79). Across all stages, SHAP analysis reinforced that wealth score and household density are the primary structural predictors of caries risk. Conclusion: AI provides a feasible framework for population-level risk stratification and public health planning. The findings suggest that AI-driven frameworks should be utilized for managerial triage and resource allocation, enabling proportional universalism by identifying high-risk clusters through structural determinants rather than individual-level clinical diagnostics.
Introduction: Evidence regarding the performance of visual test for root caries severity assessment is still limited. This in vitro diagnostic performance study evaluated visual test with ICDAS (International Caries Detection and Assessment System) root criteria against three histological reference standards —stereomicroscopy (SM), polarised-light microscopy (PLM), and micro-computed tomography (micro-CT). Methods: A total of 125 extracted permanent teeth were classified with the visual test using ICDAS root criteria. Histological evaluation was performed using SM and PLM. In addition, in a subsample of 22 teeth was assessed further using micro-CT. Discriminative power, global and categorised performance (confidence interval with a 95% confidence level) and agreement (Cohen’s Kappa and percentage), and indicators of validity and diagnostic utility (sensitivity, specificity, positive and negative predictive values - PPV and NPV, likelihood ratios - LR+ and LR-) were stablished. Results: Compared with histological reference standards, the visual-tactile test showed moderate to high diagnostic performance for root caries for sensitivity (≥ 0.73), specificity (≥ 0.90), and predictive values (PPV ≥ 0.71, NPV ≥ 0.91). Overall likelihood ratios supported moderate confirmatory value (LR+ = 7.43–9.39) and rule-out ability (LR− = 0.26–0.30) after visual–tactile classification. Agreement between the visual test and the histological references was substantial (weighted kappa ≥ 0.88), with moderate discriminative power (0.63-0.68). Complete observed agreement was found for the sound category. Performance was high for initial caries (≥90% for sensitivity, specificity, and predictive values; LR+ ≥10; discriminative power ≥0.82), whereas extensive lesions showed good rule-out ability but lower confirmatory value. Conclusion: The visual-tactile test showed substantial agreement against histological reference standards, moderate to high diagnostic performance, and moderate discriminative power. The test showed complete observed agreement for sound surfaces and high performance for identifying initial caries, whereas extensive lesions showed good rule-out ability but lower confirmatory value. These findings support the visual-tactile ICDAS root caries criteria as useful for the detection and assessment of root caries lesions.
INTRODUCTION:Diffuse reflectance spectroscopy (DRS) is a proposed optical adjunct for caries detection, but its diagnostic accuracy against histology has not been validated alongside its ability to discriminate restorative materials. We aimed to validate DRS in vitro for both tasks. METHODS:This in vitro study comprised two arms. In Part 1, 200 occlusal regions of interest from 100 extracted teeth were assessed by DRS (350-1730 nm) and by two independent examiners using the International Caries Detection and Assessment System (ICDAS), with histological scoring as the reference standard. In Part 2, four freshly prepared restorative materials (45 sites each: a self-adhesive resin cement, two resin composites, a resin-modified glass-ionomer) were compared with 37 sound-enamel sites from a separate cohort (one site per tooth). Spectra were classified by partial least squares discriminant analysis (PLS-DA) with specimen-grouped 10-fold cross-validation (sites from a specimen kept in one fold, preventing leakage). RESULTS:DRS reached 80.5% exact agreement with histology (quadratically weighted κ = 0.916, 95% CI 0.890-0.939) versus 54.0-57.5% for the visual examiners. At the any-caries threshold (histological score ≥ D1), DRS achieved sensitivity 90.8% (95% CI 84.3-94.8) and specificity 95.0% (87.8-98.0), with four false-positive classifications of sound surfaces against 25 and 23 for ICDAS examiners. The DRS AUC (0.946) exceeded both examiners (0.908 and 0.914), reaching significance against one (DeLong test, p = 0.027). Sensitivity for dentin-level caries was low (37.5%). PLS-DA discriminated the five classes (four materials plus sound enamel) with overall accuracy 99.5% (95% CI 97.4-99.9). CONCLUSION:In vitro, DRS showed markedly higher specificity than visual ICDAS scoring for occlusal caries and discriminated four freshly prepared restorative materials from sound enamel. Its low sensitivity for dentin-level lesions and the controlled laboratory conditions warrant cautious interpretation pending in vivo validation.
Introduction Early-stage proximal caries detection on bitewing radiographs is challenging and hampered by inter-observer variability. This study aimed to assess how artificial intelligence (AI) support influences radiographic detection and staging of proximal early caries lesions, confidence, and assessment time across observer experience levels. Methods Eighteen observers (dental students, general dentists, and dental radiologists) evaluated 369 proximal surfaces on 25 bitewing radiographs in two sessions: first unaided, then aided with AI-annotated caries lesion regions. Observers assigned lesion stages and provided self-reported (SR) confidence ratings. Radiographic detection and staging accuracy relative to an expert consensus reference standard was assessed using cumulative link mixed models (CLMM). Ordinal agreement and binary detection metrics were computed for each observer group. Results The AI-based tool had divergent effects across the different groups of observers. Dental students showed a statistically significant improvement in radiographic detection and staging accuracy relative to the reference standard (p = 0.039). General dentists and dental radiologists showed no statistically significant change in radiographic detection and staging accuracy when using AI support (p = 0.28 and p = 0.15, respectively). SR-confidence increased when the AI-based tool was used for all groups. Conclusion AI support induced a shift in observers' detections toward the AI-based tool's performance pattern, with its effect on radiographic detection and staging accuracy varying across observer groups and no statistically significant change observed in more experienced observers. AI generally increased the SR-confidence of the clinicians.
