Reporting Quality of Mendelian Randomization Studies Based on STROBE-MR: An Investigation Powered by Large Language Models

The research team of Zhou, Wang, Xu et al. built an automated assessment pipeline based on OpenAI's GPT-5-mini model, tested its agreement with human consensus on 46 manually annotated Mendelian randomization (MR) papers, obtaining an agreement rate of 89.0%, and then applied the pipeline to more than 13,000 MR papers published between 2005 and 2025 to assess their adherence to the STROBE-MR reporting guideline. The study reports that, on average, 22 of 43 items (51.2%) met the full-compliance criterion, with marked differences in reporting rates across items and sections; the least-reported items were concentrated in the "Methods", "Results" and "Other information" sections; the number of reported items remained essentially stable over time, with a slight decline after 2021. Sensitivity analyses showed that the temporal trend patterns differed under the full-compliance and partial-compliance criteria, especially in the Methods, Discussion and Other information sections. The authors state that the results suggest the reporting quality of MR studies still needs improvement, and recommend further refinement of the STROBE-MR guideline.