Becker's Hospital Review

20 INNOVATION 4 hospitals, health systems testing out ChatGPT By Naomi Diaz G enerative AI such as ChatGPT, the artificial intelligence-based chatbot developed by OpenAI, is being touted as a tool that can revolutionize healthcare, and although it is pretty new, hospitals and health systems are working on piloting this technology to see if it can be applied to the clinical setting. Here are four hospitals and health systems piloting, testing or deploying the new technology: • Palo Alto, Calif.-based Stanford Health Care said it plans to roll out its ChatGPT-like feature for physicians in May. e organization is rolling out Microso's generative artificial intelligence platform from OpenAI that is integrated into Epic's MyChart patient portal so that it can dra messages for clinicians. • UC San Diego Health and Madison, Wis.-based UW Health are among the first healthcare organizations to pilot a new integration from Epic Systems and Microso that uses Azure OpenAI to dra messages within the EHR to patients. • Boston Children's told Becker's that the organization is hiring a prompt engineer to work on large language models such as ChatGPT. e person will help Boston Children's design and develop prompts to effectively gather data from generative AI programs and refine the models for healthcare-specific applications. n ChatGPT outperforms physicians when answering patient questions By Naomi Diaz C hatGPT may be better at providing more empathetic answers to patient questions, according to an April 28 study published in JAMA Internal Medicine. In the study, researchers compared two sets of written responses to 195 real-world patient questions. The physician responses were sourced from patient questions asked on Reddit, while the other responses were written by ChatGPT. Both sets of responses were evaluated by a three-member panel of licensed healthcare professionals, with 79 percent preferring the answers of ChatGPT. More than a quarter of the physicians' responses were marked as less acceptable in quality by the panel, while ChatGPT's responses were deemed less acceptable only 3 percent of the time. The panel also found that ChatGPT's responses to patient inquiries were more empathetic. Only 5 percent of physicians' responses were marked as empathetic. The authors noted one limitation of the study is that the physician responses came from a public online forum, and physicians in a clinical setting may form their answers differently based on their established relationships with patients. n Why healthcare data privacy is an 'illusion,' according to Yale professor By Giles Bruce B ecause so much patient information is digital nowadays, healthcare data privacy is just an "illusion" in the U.S., a Yale School of Medicine professor wrote June 1 in BMJ. Harlan Krumholz, MD, professor of medicine at the New Haven, Conn.-based university, said healthcare data is now shared "among healthcare providers to whom you may not have originally disclosed the information, and private companies that work with health systems, or others to whom data are sold." "Although no one would relish a return to paper records, that cumbersome system at least made it difficult for patients' data to be made into a commodity," he wrote. "The digital transformation of healthcare data has enabled wonderous breakthroughs — but at the cost of our privacy." He recommended using tech that protects private healthcare conversations, seeking explicit permission for data-sharing, and educating patients about the risk of disclosing sensitive information. n

• Cover