MedEdMentor is an innovative, AI-powered platform designed to support health professions education (HPE) scholars in navigating the complexities of theory-driven research. Launched in September 2023 by co-founders Dr. Geoffrey Stetson and Dr. Gregory Ow, the platform aims to democratize access to mentorship and scholarly resources, particularly benefiting early-career researchers and those in under-resourced settings
MedEdMentor offers the first dedicated search tool for medical education literature, indexing over 100,000 articles from 151 journals. This tool enhances the relevance and efficiency of literature searches, outperforming traditional databases like PubMed and Google Scholar in specificity.
The platform hosts an extensive database of over 250 theories and frameworks pertinent to medical education. Each entry includes concise summaries and examples of practical application, assisting researchers in selecting appropriate theoretical underpinnings for their studies.
MedEdMentor AI, built upon GPT-4, serves as a virtual mentor, guiding users through research design, theory selection, and methodological considerations. In evaluations, it successfully recommended the same theoretical constructs used in 55% of analyzed qualitative studies, demonstrating its potential to augment traditional mentorship.
The platform provides a suite of lessons and glossaries that demystify complex concepts in medical education research. These resources are tailored to assist both novices and experienced scholars in understanding paradigms, theoretical frameworks, and research methodologies.
The MedEdMentor Paper Database is built upon the Medical Education Corpus (MEC), a curated collection of approximately 119,000 medical education papers. These papers were identified from an analysis of 2.3 million articles across 151 journals, using a machine learning model known as the MEC Classifier. This classifier evaluates titles, abstracts, and journal sources to determine the relevance of each paper to medical education.
Traditional search methods, such as relying on PubMed's Medical Subject Headings (MeSH), often fall short in accurately capturing the breadth of medical education literature. The MEC Classifier demonstrates a higher sensitivity (90%) compared to MeSH terms (66%), meaning it more effectively identifies relevant medical education papers. This enhanced accuracy ensures that researchers have access to a more comprehensive and precise set of resources.
The development of the MEC involved several key steps:
Journal Selection: Journals were categorized into three groups:
MEJ-Core: Journals consistently publishing medical education papers.
MEJ-Adjacent: Journals occasionally publishing relevant content.
MEJ-Peripheral: Journals rarely publishing medical education papers.
Data Collection: A total of 2.3 million papers from these journals were collected using Semantic Scholar.
Model Training: A sample of 2,532 papers was manually labeled to train the MEC Classifier, which then analyzed the entire dataset to identify medical education papers.
The MedEdMentor Paper Database is freely accessible and can be explored at mededmentor.org. Researchers interested in testing new features or contributing feedback are encouraged to reach out to the MedEdMentor team.
