Innovative Methods for Next-Generation Evidence Synthesis in Oral Health Research - PROJECT SUMMARY Evidence synthesis is the foundation for guideline development and clinical decision-making in oral health. Yet current practice faces several persistent barriers. First, screening and data extraction are slow and error-prone; while semi-automation tools exist, none are trained or validated specifically for oral health, so performance on dental topics is largely unknown. Second, many oral health syntheses draw on few studies and lack empirically grounded parameter inputs for planning and analysis; there is no domain-specific, comprehensive resource to supply calibrated assumptions for future studies and syntheses. Third, clustered designs that are common in dentistry, such as split-mouth trials and teeth within patients, are often handled with ad hoc adjustments that understate uncertainty. These limitations reduce the precision, reproducibility, and timeliness of conclusions from oral health evidence synthesis. The proposed project extends established methodological expertise into a new domain by developing oral-health-specific methods, resources, and tools for evidence synthesis. This project has four aims that provide coordinated, cutting-edge solutions to these gaps. Aim 1 will validate and adapt AI tools for study-eligibility assessment and structured data extraction using oral health gold-standard corpora, priority screening, and human-in-the-loop safeguards; outputs will include high-recall screening models, field-level extraction with source-span provenance, and auditable operating procedures. Aim 2 will build a har- monized, living evidence resource from published oral health syntheses and derive predictive distributions for key parameters such as between-study heterogeneity and within-patient correlations; these calibrated priors will support both principled analyses and realistic trial planning. Aim 3 will develop Bayesian hierarchical models that explicitly represent within-patient and hierarchical correlations, and propagate this uncertainty into pooled effects. These models will improve interval calibration and decision stability in settings with few studies or small samples, while remaining compatible with standard effect measures. Aim 4 will deliver an integrated, open-source platform that puts these advances into practice through reproducible AI workflows and training modules, alongside an interactive evidence resource and correlation-aware analysis modules, with accompanying documentation and worked examples in oral health for immediate use. Expected outcomes include faster and more reliable screen- ing and extraction with preserved high recall, empirically calibrated priors that stabilize estimation when data are sparse, models that properly handle clustering, and an accessible software suite that enables exact replication. Collectively, the work will enhance the rigor, efficiency, and transparency of oral health evidence synthesis, sup- porting better-informed guidelines, study design, and patient care.
