Dr. Isaac Kohane explores AI's rapid medical evolution and the challenges of accuracy, updating, and education in infectious diseases at IDWeek 2023.
(A version of this article was originally published on our sister brand, Contemporary Pediatrics®.)
At IDWeek 2023, during the opening plenary session, Isaac "Zak" Kohane, MD, PhD, an esteemed figure known for his pioneering work in the convergence of artificial intelligence and precision medicine, engaged the audience of infectious diseases experts with a discussion on the vast potential and inherent challenges of medical AI. Throughout his presentation, he described the AI’s capabilities, underlining its rapid evolution and impact.
Kohane, also the inaugural chair of Harvard Medical School’s Department of Biomedical Informatics, began discussing the capabilities of AI, drawing upon a case study from the Undiagnosed Diseases Network (UDN). This groundbreaking program is designed to offer answers to individuals and families dealing with undiagnosed diseases. He shared an instance where AI, leveraging what he dubbed "old-school AI" techniques, successfully identified a rare condition known as guanosine triphosphate (GTP) cyclohydrolase I deficiency in a 4-year-old boy who had developed dystonia in 2021. The child's motor function was fully restored after implementing an appropriate treatment regimen involving folinic acid, L-DOPA, and 5-hydroxytryptophan. Notably, AI played a pivotal role in this diagnosis.
Kohane showcased the AI’s evolution with another example. By providing OpenAI's Generative Pre-Trained Transformer-4 (GPT-4) with a summary of a UDN case and a list of whole exome sequencing variants, the model identified POLR3A as an "interesting" potential cause. This finding was verified through cellular models.
The progress of AI in medical practice was underscored by its ability to tackle US Medical License Exam (USMLE)-)-style questions. Kohane explained how earlier AI models often struggled to pass such medical examinations, but by late 2022, Google's MED-PaLM achieved passing scores. Its successor, MED-PaLM 2, achieved even higher scores, indicating AI's increasingly complex medical problem-solving ability.
However, Kohane also shed light on the challenges and limitations of current AI tools. These challenges contain potential inaccuracies, reliance on outdated information, and the risks associated with the propagation of fabricated data. He cited a well-known instance where an AI model, trained initially on one patient population, failed to predict the severity of subsequent SARS-CoV-2 variants during the COVID-19 pandemic. These examples were a stark reminder of AI's imperfections and the need for constant vigilance.
Kohane emphasized the inevitability of AI integration into health care. He foresaw a global "sea change in health care with AI" that would surpass the rate of regulatory adaptations and educational advancements. To effectively navigate this ever-changing landscape, institutions must devote significant resources to data and computational infrastructure while fostering expertise in quantitative data analysis.
This approach, he said, would empower health care providers to play a pivotal role in harnessing the potential of AI while maintaining patient care at its core. The future of health care, as Kohane suggested, is undeniably intertwined with the vast capabilities of medical AI.
Click here for more coverage of IDWeek 2023, held in Boston, Massachusetts, from October 11 to 15, 2023.
Kohane I. AI and the future of ID. IDWeek 2023. October 11, 2023. Boston, Massachusetts.