Assessment profiles in undergraduate dental anatomy education: a five-year retrospective analysis with implications for future AI-supported question Bank Development

Longitudinal examination data can provide valuable evidence for monitoring assessment quality in health professions education. In dental anatomy education, such evidence is particularly relevant because assessment should support both foundational anatomical knowledge and later clinical application. The growing interest in artificial intelligence-supported item generation further increases the need to understand existing local assessment profiles before developing AI-assisted question banks. Archived individual-level final examination records from first- and second-year undergraduate dental anatomy courses were retrospectively analysed across five academic years, from 2021–2022 to 2025–2026. Final examination scores were evaluated as continuous variables and categorised into predefined achievement bands. Class-level differences, year-to-year variation and achievement band distributions were examined using non-parametric and categorical statistical analyses. AI-generated questions were not administered to students and did not contribute to grades. A total of 1,216 eligible final examination records were analysed, comprising 666 first-year and 550 s-year records. The overall mean final examination score was 63.91 ± 20.00, with a median score of 66.0. Final examination scores differed significantly across academic years ( p < 0.001). Although second-year scores were numerically higher overall, the class-level difference was not statistically significant ( p = 0.072). Five-year final examination data demonstrated meaningful longitudinal variation in undergraduate dental anatomy assessment performance. The elevated 2021–2022 performance coincided with pandemic-era online assessment conditions; however, the available data do not permit causal attribution of the higher scores to the online examination format. The findings provide a local retrospective assessment baseline that may inform future AI-assisted question bank development, while also indicating the need for expert review and prospective item-level psychometric validation before AI-assisted items are used in student assessment.