Rare heart diseases present a major global challenge: expanding lifesaving access to subspecialist expertise to millions with complex conditions. AI could help. Learn more about this early stage research 👇
Happy to share our new work “"Towards the Democratization of Subspecialty Medical Expertise”. Rare heart disease exemplifies a major global healthcare challenge: improving access to subspecialist medical expertise for millions who have complex conditions - https://lnkd.in/ei58hjGV We explored how AMIE, a research LLM optimized for diagnosis, might help diagnosis and treatment decisions in suspected inherited cardiomyopathies. Using de-ID case information, world-experts at Stanford performed a blinded evaluation of responses by AMIE or board-certified general cardiologists. AMIE’s diagnosis ,triage and management were preferred in 5/10 domains, and equivalent in the remainder. AMIE and cardiologists were complementary, revealing great potential for AI as an assistive tool. Despite a slightly higher rate of errors, AMIE was preferred by subspecialists due to its comprehensiveness. Access to AMIE outputs improved general-cardiologist’s responses in 64% of cases across all 10 domains of evaluation. While promising, this is very early-stage research. Significant further work is needed before such systems might be safe for clinical use and to catalyze this progress we are open-sourcing the de-ID evaluation cases here: https://lnkd.in/eu4_ebsE It was such a privilege to work with the great experts in subspecialty medicine at Stanford - Euan Ashley, and Jack O'Sullivan MD, PhD with my outstanding teammates Google DeepMind - Khaled Saab, Anil Palepu, Tao Tu, Vivek Natarajan and Wei-Hung Weng. And… just for fun… we made a podcast discussing the paper thanks to #NotebookLM! It's a quirky and quick listen that we hope helps explain what we studied and why: https://lnkd.in/eQq9NEkG