Development of a Predictive Framework for Minimally Invasive Restorative Dentistry Outcomes

Thesis Abstract

Minimally invasive restorative dentistry aims to preserve tooth structure while achieving durable clinical outcomes, yet there is limited predictive capacity tying patient-, lesion-, and technique-related factors to treatment success. This study develops a predictive framework to estimate outcomes of minimally invasive restorative interventions by integrating biological, material, and procedural determinants into a theoretical model that informs clinical decision-making and patient-specific prognostication. The aim is to construct and validate a robust, generalizable framework capable of forecasting restoration longevity, marginal integrity, and post-treatment sensitivity, thereby guiding evidence-based minimally invasive choices. Specific objectives are (1) to identify a comprehensive set of predictors from patient demographics, caries risk, lesion characteristics, adhesive systems, and operative protocols; (2) to operationalize outcome measures encompassing restoration survival, marginal adaptation, secondary caries incidence, and patient-reported outcomes; (3) to develop a predictive model using multivariate regression and machine learning approaches; (4) to test theoretical underpinnings drawn from the Self-Regulation and Theory of Planned Behavior to explain adherence to minimally invasive protocols; (5) to validate the model across diverse clinical settings and populations. A multicenter prospective cohort design will be employed, enrolling 1,200 patients presenting with eligible proximal or occlusal lesions across five tertiary dental centers over 18 months. Data collection will utilize standardized case report forms and calibrated calibration sessions to ensure inter-operator reliability. Instruments include a validated Caries Risk Assessment tool, lesion activity and size metrics by standardized photography and intraoral scans, adhesive protocol documentation, and a patient-reported outcome measure for postoperative sensitivity and satisfaction. Outcome data will be collected at baseline, 6, 12, and 24 months post-treatment. Predictors will cover patient-level variables (age, sex, systemic health, caries risk, oral hygiene index), lesion-level variables (location, depth, activity, lesion size), material-level variables (adhesive system, composite type, glass ionomer cement usage), and procedure-level variables (cavity preparation technique, operator experience, chairside time). The analytical strategy comprises a two-stage approach. First, descriptive statistics will summarize baseline characteristics and outcome distributions. Second, multivariate Cox proportional hazards models and logistic regression will assess predictors of restoration survival and dichotomous outcomes (marginal integrity, secondary caries), while linear mixed-effects models will analyze continuous measures of patient-reported outcomes across time. To enhance predictive accuracy, machine learning techniques, including random forest and gradient boosting, will be applied, with nested cross-validation to prevent overfitting. Model performance will be evaluated using calibration plots, receiver operating characteristic curves, Brier scores, and net reclassification improvement. The framework will be theoretically anchored by integrating two pertinent theories the Self-Regulation Theory to account for patient adherence to post-treatment maintenance and the Theory of Planned Behavior to elucidate clinician decision-making regarding adhesive strategies. A conceptual framework diagram will illustrate relationships among predictors, theoretical constructs, and outcomes. Expected findings include (i) identification of a parsimonious set of high-impact predictors (e.g., initial lesion size, bond failure risk factors, operator experience, and adhesive system) that substantially influence restoration survival and marginal integrity; (ii) demonstration that integrative models combining traditional regression with machine learning yield superior predictive discrimination and calibration; (iii) empirical support for the relevance of Self-Regulation and Planned Behavior constructs in predicting adherence to minimally invasive protocols and maintenance behaviors; (iv) development of a user-friendly nomogram and an executable decision-support algorithm for clinicians to forecast outcomes and tailor interventions. The study contributes to knowledge by operationalizing a comprehensive, theory-informed predictive framework that bridges material science, clinical technique, and patient behavior, facilitating personalized minimally invasive dentistry. It provides empirical benchmarks for outcome expectations, informs training curricula, and supports policy-oriented recommendations for standardizing minimally invasive practices. Limitations include potential residual confounding, loss to follow-up, and center-specific practice variability, which will be mitigated through sensitivity analyses and site-adjusted models. The main conclusion anticipates that a validated, integrative predictive framework can substantially improve prognosis estimation for minimally invasive restorations and optimize material and procedural choices, prompting routine use of data-driven decision support in everyday practice. Recommendations emphasize routine collection of standardized predictor data, integration of the framework into electronic health records, ongoing model recalibration with new evidence, and exploration of cost-effectiveness implications for broader adoption.