Study-Level Data and Reproducible R Code for Artificial Intelligence in Dental Caries Detection: A Systematic Review and Diagnostic Test Accuracy Meta-Analysis

This repository contains the study-level data, analytical code, methodological audit files, and derived results supporting the systematic review and diagnostic test accuracy meta-analysis entitled “Clinical Utility and Translational Readiness of Artificial Intelligence for Dental Caries Detection: A Systematic Review and Diagnostic Test Accuracy Meta-Analysis.” The literature search was updated through 10 June 2026. The primary quantitative synthesis evaluates standalone artificial intelligence systems with complete 2 × 2 diagnostic contingency data. Studies assessing clinicians working with artificial intelligence assistance are retained as a separate secondary analysis. Diagnostic performance is estimated using a bivariate Reitsma random-effects model, hierarchical summary receiver operating characteristic analysis, likelihood ratios, diagnostic odds ratios, and prevalence-dependent positive and negative predictive values. Uncertainty in derived measures is propagated using 200,000 Monte Carlo draws with a prespecified random seed. Additional analyses examine the unit of analysis, alternative analytical specifications, and the influence of studies identified through the updated search. The repository also contains study characteristics, estimate-selection decisions, QUADAS-3 assessments, PRISMA flow data, complete search strategies, excluded-study records, protocol amendments, the PRISMA-DTA checklist, and an auditable comparison with the review by Lam et al. Implementation barriers are examined through exploratory TF-IDF vectorisation, cosine distance and average-linkage hierarchical clustering, followed by an auditable rule-based assignment of substantive statements to seven predefined domains. All information consists of study-level secondary data extracted or derived from published reports. No individual participant data, identifiable clinical information, clinical images, or copyrighted full-text articles are included. The review protocol was prospectively registered in PROSPERO (CRD420251232014).