Clinically Trained AI Achieved High Level of Parity with Physicians in Generating Differential Diagnosis, Large-Scale Study Finds

The peer-reviewed study published in Mayo Clinic Proceedings: Digital Health analyzed more than 102,000 primary care clinical cases, demonstrates how K Health’s clinical AI performs in real-world clinical practice

K Health, the clinical Artificial Intelligence (AI), 24/7 national primary care provider, announces the publication of peer-reviewed research titled “Diagnostic Accuracy of Artificial Intelligence in Virtual Primary Care”. The large-scale research compared K’s clinical AI diagnostic capability with the actual diagnosis clinicians gave to patients on K virtual primary care platform. This comprehensive study of over 102,000 real-world primary care clinical cases was co-authored by Dan Zeltzer, Princeton University-trained Health Economist from Tel Aviv University, and Jon O Ebbert, M.D., Professor of Medicine at Mayo Clinic.

 Latest AiThority Interview Insights : AiThority Interview with Rebecca Clyde, Co-founder and CEO of Botco.ai

The study’s key findings:

• Clinicians providing clinical care to real-life patients agreed with the AI-generated diagnoses 84.2% of the time.
• In 57 of the most common primary care diagnoses, more than 90% agreement existed between AI-generated and clinician differential diagnoses.
• Independent clinicians adjudicators blind to the source of the diagnosis, chose K’s AI differential diagnosis as often as the clinician’s diagnosis.
• Adjudicator consensus diagnosis was always in the AI differential diagnosis.
• The study also demonstrates how the AI can improve from the clinician’s feedback in a unique reinforcement learning loop.
• The study observed no biases across gender, age and ethnicity in the AI-generated differential diagnoses.

AI holds the potential to revolutionize decision-making in healthcare, but has been understudied in primary care, until now. The study results show AI’s strong accuracy in primary care, with the possibility for AI to assist in pre-visit medical intake and differential diagnosis.

“When clinical cases were independently reviewed by three experts, the AI-generated list of possible diagnoses included the diagnosis most experts agreed on in every case,” said Dr. Ebbert. “We also observed that retraining can improve the performance of these AI models.”

K created a primary care platform that uses proprietary AI models trained on clinical data, such as data from the Mayo Clinic Platform_Discover, combined with doctor oversight, enabling widespread access to quality clinical care. K’s clinical AI enables K’s clinicians to make high-quality medical decisions. K’s close-looped reinforcement model allows K’s AI to learn from the data and under clinical supervision, improve its predictive algorithms.

Read More about AiThority Interview : AIThority Interview with David Lambert, VP & GM, Strategy & Growth, APAC, Medallia

“We have built a clinical-grade AI system that works with a 24/7 primary care clinic,” said Ran Shaul, Co-Founder and Chief Product Officer. “As our AI learns from data and becomes smarter, it can make a huge difference in terms of saving doctors’ time and providing doctors with AI superpowers at their fingertips. As a next step, we are in the process of integrating our platform and service into points of care at leading health systems.”

“This pioneering study marks the first comprehensive examination of AI’s performance versus physicians in real-life cases as opposed to text book vignettes. The outcomes are incredibly promising, showcasing K Health’s AI potential to be on par with a doctor’s capabilities. Beyond its high accuracy, this AI stands out for its absence of hallucinations, errors and impartiality across demographics. Similar to a physician, K’s AI evolves through data assimilation, ensuring constant enhancements, and raising the accuracy bar across diverse conditions,” shared Zeltzer.

The average medical chat between K’s AI and patients takes 5 minutes and consists of 25 medical questions, 6 positive symptoms, and 2.5 potential diagnoses. K has integrated this into a clinical platform that covers the 48 continental states around the clock and assists doctors in diagnosing and treating patients based on this information and any additional information they collect. This significantly decreases the time doctors need to make high-quality medical decisions and allows the millions of patients who have used K to get high-quality diagnoses and treatment, including prescriptions, specialty referrals, labs and tests.

 Latest AiThority Interview Insights : AiThority Interview with Rebecca Clyde, Co-founder and CEO of Botco.ai

 [To share your insights with us, please write to sghosh@martechseries.com]