NEW YORK–(BUSINESS WIRE)–K Health, a leading platform for clinical AI in Primary Care, has announced groundbreaking study findings that indicate its AI Physician Mode aligns with the clinical decisions of doctors in two-thirds of actual patient cases, while demonstrating superior quality in the remaining third.
The research, carried out by teams from Cedars-Sinai, Tel Aviv University, and K Health, involved seasoned primary care physicians who evaluated and rated AI-driven recommendations against traditional physician advice for acute health issues in a virtual urgent care context. The results revealed that AI suggestions were more often deemed optimal than the decisions made by physicians.
The findings were shared today during a scientific plenary session at the American College of Physicians Internal Medicine Meeting and were also published in the peer-reviewed journal Annals of Internal Medicine.
“We rigorously tested AI in practical scenarios rather than mere textbook examples. Primary care involves navigating complex, multifaceted patient cases, where AI encounters incomplete information and diverse patient profiles. Our results indicate that by training the AI on a wealth of de-identified clinical notes and utilizing everyday provider care as a constant reinforcement mechanism, we can achieve accuracy comparable to human doctors, sometimes even surpassing it,” stated Ran Shaul, Co-founder & Chief Product Officer of K Health.
The U.S. healthcare system stands to gain from improved diagnostic efficiencies, enabling it to better address more intricate patient care needs. K Health aims to democratize medicine by providing AI-powered solutions that offer smarter, faster, and more effective patient care around the clock. The AI Physician Mode analyzes a patient’s complete medical history prior to their primary care appointment, supplying healthcare providers with AI-informed clinical recommendations. This technology is already being implemented by leading health systems nationally, such as Cedars-Sinai, Hackensack Meridian Health, and Hartford HealthCare.
“This study reinforces what we at K Health have long asserted—that AI trained with high-quality clinical data has the potential to revolutionize healthcare delivery. We have utilized billions of data points from patient records to improve medical decision-making at the point of care in ways that were previously unimaginable,” remarked Allon Bloch, Co-Founder and CEO of K Health. “By merging AI’s analytical capabilities with physician expertise, we are creating a new paradigm for patient care that emphasizes accessibility, accuracy, and clinical excellence.”
The retrospective study utilized data from Cedars-Sinai Connect, a digital primary and urgent care platform launched in 2023. Extending Cedars-Sinai’s in-person services, Cedars-Sinai Connect aims to broaden virtual healthcare access for patients in California, through a mobile application that facilitates quick consultations with Cedars-Sinai specialists for various healthcare needs.
The study analyzed 461 physician-led visits where AI recommendations were provided to the treating physicians between June 12 and July 14, 2024. Key medical concerns addressed during these virtual urgent care sessions included respiratory, urinary, vaginal, vision, and dental symptoms among adults.
The expert review showed that in roughly two-thirds of the cases, physicians made clinical decisions that were nearly identical to the AI recommendations. Furthermore, in the remaining third of cases, AI recommendations were rated as superior almost twice as often as those made by human doctors.
“The outcomes of this study emphasize the synergistic advantages of AI and physician expertise,” said Caroline Goldzweig, MD, Chief Medical Officer of Cedars-Sinai Medical Network and co-senior author of the study. “This research highlights the potential for the judicious integration of AI, allowing physicians to dedicate more time to the essential aspects of patient care that thrive on human interaction.”
The study also revealed that potentially harmful recommendations were recorded in only 2.8% of cases with AI compared to 4.6% with physicians. This discrepancy may be attributed to the AI Physician Mode’s strict adherence to medical guidelines and its ability to identify subtle nuances that doctors might overlook. In approximately 20% of instances, the AI mimicked prudent physician behavior by stating ‘I don’t know’ when it lacked sufficient confidence to provide a recommendation, reflecting the uncertainty management strategies of experienced clinicians.
