A Predictive Framework for Early Detection of Dental Caries Risk

Thesis Abstract

Early detection of dental caries risk remains a critical challenge in preventive dentistry, given the multifactorial etiology involving biological, behavioral, and environmental determinants that interact over time to initiate demineralization processes. This study addresses the gap between irregular risk assessment in routine care and the need for a robust predictive framework that integrates clinical, microbiological, and psychosocial factors to forecast caries development in high-risk populations. The aim is to develop and validate a predictive framework for early detection of dental caries risk that informs targeted prevention strategies. Specific objectives are (i) to identify and quantify demographic, clinical, microbiological (e.g., mutans streptococci load), salivary, dietary, and psychosocial predictors of incident caries over 24 months; (ii) to construct a multivariate risk score using regression-based modeling and machine learning techniques; (iii) to validate the framework in an independent cohort and evaluate its predictive accuracy against conventional caries risk assessments; and (iv) to assess the framework’s feasibility for integration into routine dental practice through a cost–effectiveness analysis. A longitudinal, mixed-methods design was employed. The quantitative component recruited 1,200 participants aged 6–12 years from five urban dental clinics with elevated caries prevalence, and a 12-month holdout validation cohort of 400 participants. Data collection combined electronic medical records, calibrated clinical examinations with ICDAS scoring, unstimulated saliva collection for flow rate and buffering capacity, and microbiological assays (quantitative PCR for mutans streptococci and Lactobacillus spp.). Dietary intake was assessed using a validated 24-hour recall across three non-consecutive days, and psychosocial factors were measured with standardized instruments evaluating oral health behaviors, caregiver education, and perceived dental anxiety. The qualitative strand involved semi-structured interviews with 30 clinicians to examine feasibility, acceptability, and potential integration barriers of the framework in routine care. Analytical approaches included descriptive statistics to characterize the cohort, multivariable logistic regression to identify independent predictors of caries incidence, and penalized regression (LASSO) to derive a parsimonious risk model. Advanced machine learning techniques—random forests and gradient boosting—were employed to enhance predictive performance and to explore non-linear associations and interactions among predictors. Model performance was evaluated by area under the receiver operating characteristic curve (AUC), calibration plots, Brier score, and decision-curve analysis to determine clinical usefulness. Internal validation used bootstrap resampling, while external validation assessed transportability to the independent cohort. A cost–effectiveness analysis compared the predictive framework against standard caries risk assessment tools, using quality-adjusted life years (QALYs) and incremental cost-effectiveness ratios (ICERs). The theoretical underpinning combined the Health Belief Model to interpret psychosocial predictors and the Ecological System Theory to contextualize multi-level determinants across biological, behavioral, and environmental domains. Expected findings indicate that a composite framework integrating microbial load, salivary parameters, dietary patterns, and caregiver-related factors will produce superior predictive accuracy (AUC > 0.85) compared with conventional risk assessment tools (AUC ~0.70). The framework is anticipated to demonstrate good calibration, robust discrimination in the validation cohort, and favorable cost-effectiveness in identifying high-risk children for preventive interventions. The study is expected to illuminate interaction effects, such as the amplification of microbial risk by frequent exposure to fermentable carbohydrates and low fluoride exposure, moderated by parental health beliefs and health-seeking behaviors. The contribution to knowledge includes (i) a validated, multi-domain predictive framework for early caries risk detection operable in primary care and community clinics; (ii) empirical elucidation of how microbiological, salivary, dietary, and psychosocial factors interact to modulate risk trajectories in children; (iii) methodological advancement through the integration of traditional statistical modeling with machine learning in a clinically feasible risk score; and (iv) practical guidance on implementation and cost-effectiveness to support policy and practice. Overall, the study concludes that early, accurate prediction of caries risk is achievable through a transdisciplinary framework that informs targeted preventive strategies, reduces incidence, and optimizes resource allocation. Recommendations include pilot integration of the framework into electronic dental records, clinician training on interpretation of risk scores, and policy support for risk-guided prevention programs in high-prevalence settings.