Shahroud University of Medical siences Shahroud Journal of Medical Sciences 2423-6594 12 5 2026 08 29 Predictive Models for Performance in High-Stakes Medical Assessments: A Systematic Review within Health Professions Education 29 49 eng Haniye Mastour Department of Medical Education, School of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran.. Haniye.Mastour@gmail.com Toktam Dehghani School of Medical Education and Learning Technologies, Shahid Beheshti University of Medical Sciences, Tehran, Iran.. dehghanit@mums.ac.ir Mahdie Jajroudi Department of Medical Informatics, School of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran; Pharmaceutical Sciences Research Center, Institute of Pharmaceutical Technology, Mashhad University of Medical Sciences, Mashhad, Iran.. Jajroudimh991@mums.ac.ir Mohammadreza Farrokhnia Department of Learning, Data Analytics and Technology, Instructional Technology Section, University of Twente, Enschede, the Netherlands.. M.farrokhnia@utwente.nl Mitra Zarei Medical Librarian and Information Counselor, School of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran.. ZareiM3@mums.ac.ir Saeid Eslami Department of Medical Informatics, School of Medicine, Mashhad University of Medical Sciences, Mashhad, Iran; Pharmaceutical Sciences Research Center, Institute of Pharmaceutical Technology, Mashhad University of Medical Sciences, Mashhad, Iran; Department of Medical Informatics, UMC-Location AMC, University of Amsterdam, Amsterdam, Netherlands.. EslamiS@mums.ac.ir 2025 12 04 2026 02 10 Background: High-stakes medical assessments play a critical role in certifying future healthcare professionals. Enhancing these assessments through predictive models offers a promising way to identify at-risk students and predict exam outcomes. The current study aims to address this need by providing a comprehensive overview of the application and effectiveness of predictive models, particularly in high-stakes medical assessments. Methods: This systematic review searched four major databases (MEDLINE, PubMed, Scopus, and Web of Science) for studies using predictive models to evaluate performance in high-stakes medical assessments, following PRISMA guidelines and the Best Evidence in Medical Education framework. Results: The review included 14 studies across various stages of professional training, including medical students, residents, chiropractors, and physician assistants. Sample sizes ranged from 147 to 5,825 participants, and most studies used local datasets. Regression models were the most commonly employed (78%), while random forests and artificial neural networks were frequently used for classification tasks. Although deep learning models were not observed, 14% of studies incorporated explainable AI (XAI) to improve interpretability. To provide a broader context, four systematic reviews of predictive modeling in nonmedical high-stakes assessments were also analyzed. Conclusion: While predictive methods show significant promise for enhancing the predictive ability of high-stakes medical assessments, there is still ample room for improvement. Integrating comprehensive feature sets and longitudinal tracking is crucial for increasing accuracy. Expanding studies to more extensive, diverse populations can improve generalizability and pave the way for more precise evaluations and better educational outcomes.