(Vienna, 07-10-2025) An international study led by Nikola Vladic and Cihan Ay from the Department of Medicine I (Division of Hematology and Hemostaseology, MedUni Vienna) shows that artificial intelligence (AI) language models sometimes provide better answers than experts when it comes to questions of patient education on venous thromboembolism (VTE). In complex clinical decision-making situations, the recommendations of the AI models were in many cases equivalent to those of thrombosis specialists. The results were published in the Journal of Thrombosis and Haemostasis (JTH), the leading journal of the International Society on Thrombosis and Haemostasis (ISTH).
The study compared three large language models (ChatGPT, DeepSeek, Le Chat Mistral) with internationally renowned thrombosis experts. In a blinded assessment, doctors from 18 countries evaluated both answers to frequently asked questions in patient education and recommendations for action on three complex clinical case vignettes.
"Particularly in the area of patient education, AI was able to impress with clear, understandable and structured answers and was rated higher than the experts' answers by the independent reviewers," explains lead author Nikola Vladic.
In the clinical case studies – including the treatment of deep vein thrombosis in a young woman, pulmonary embolism in a young man with no risk factors, and a tumour patient with bleeding complications – the language models delivered results comparable to those of the experts.
"Our results show that AI language models can meaningfully complement medical practice – provided there are clear guidelines for their use. Data protection, up-to-date evidence and responsibility in decision-making remain key challenges," emphasises study co-author Cihan Ay.
The authors point out that uncritical use carries risks, for example in terms of knowledge loss among young doctors. Artificial intelligence cannot replace doctors , but it can support them in their decision-making and reduce their workload.
Publication: Journal of Thrombosis and Haemostasis
Large language models versus thrombosis experts: A comparative study on patient education and clinical decision-making in venous thromboembolism.
Vladic N, Nopp S, Pabinger I, Ageno W, Connors JM, Eichinger S, Ay C, on behalf of the ClotGPT study group.
https://doi.org/10.1016/j.jtha.2025.09.004