Reliability of generative artificial intelligence in identifying the major risk factors for venous thrombosis

We accessed five of the most widely used, freely available generative artificial intelligence tools available on the Web on July 29, 2024, formulating the specific question: ‘Which are the leading risk factors for venous thrombosis?’ (the term ‘venous thrombosis’ was preferred to ‘venous thromboembolism’ as it is likely more familiar to patients). The responses from these five generative artificial intelligence tools were tabulated and then directly compared against each other. Five major risk factors were listed by all five generative artificial intelligence tools (cancer, immobilization, obesity, smoking, surgery, and trauma), four were listed by four generative artificial intelligence tools (advanced age, hormonal replacement therapy, oral contraceptives, pregnancy/puerperium), while the remaining risk factors were listed in three or fewer generative artificial intelligence tools. Of note, one recognized risk factor (lupus anticoagulant/antiphospholipid antibodies) was only listed by a single generative artificial intelligence tool, while genetic abnormalities of natural inhibitors (i.e., protein c, protein S, antithrombin) or other prothrombotic polymorphisms (i.e., Factor V Leiden, Prothrombin gene mutation) were reported by three of five generative artificial intelligence tools. Smoking and diabetes, whose role in the pathogenesis of venous thrombosis is still largely controversial, were instead, respectively, listed as major risk factors for venous thrombosis by five and two of the generative artificial intelligence tools that were queried. This analysis suggests that the reliability of generative artificial intelligence in identifying the leading risk factors for venous thrombosis is still, at best, modest.