Munjal Shah, CEO of Hippocratic AI, recently participated in the HLTH conference, where he spoke of alleviating healthcare staffing troubles through AI-based “superstaffing”.
The 2023 HLTH conference was held Oct. 8-11 in Las Vegas. Billed as health care’s “No. 1 Innovation Event,” it featured panel discussions and keynotes from executives, startup founders, investors, and medical professionals. Naturally, much of the discussion at the HLTH conference centered on how to make the best use of the recent rapid advancements in generative artificial intelligence in the health care space. At the “There’s No ‘AI’ in Team” panel, Munjal Shah, co-founder and CEO of Hippocratic AI, took the opportunity to raise an issue that he sees as a fundamental starting point for utilizing these new technologies: a shortage in nondiagnostic medical staff.
Shah’s point was simple. There’s still good reason to be concerned about using large language models for diagnosis because, even as they improve, there remains the chance that they will “hallucinate” and give a patient potentially life-threatening misinformation about their condition and treatment options. On the other hand, health care isn’t just about diagnosis. It’s also about things like chronic care nursing, patient navigation for scheduling appointments and follow-ups, and dietitian services.
Hippocratic AI was founded to use LLMs to help address shortages and lower costs for these nondiagnostic services, with the ultimate goal of reaching more people who need this sort of care.
“Right now, health care is facing a massive worldwide staffing crisis. The World Health Organization is projecting a shortfall of 10 million health workers by 2030, but overstretched health care systems and underserved patients around the globe are already feeling the effects. Simply put, this industry is in desperate need of change,” wrote Shah in a LinkedIn News wrap-up of the conference.
“Enter generative AI: the key to closing this staffing gap and ensuring that more people can receive a level of high-quality, comprehensive care that’s never existed before — what I like to call ‘super-staffing.’ At Hippocratic AI, we envision generative AI and large language models (LLMs) working alongside health care workers of all kinds to fill the staffing gap and dramatically improve health care access, equity, and outcomes.”
The panel, which also included Lifepoint chief innovation officer Jessica Beegle, Johnson & Johnson’s global medtech head Shan Jegatheeswaran, and Overjet CEO Wardah Inam, agreed that there’s no AI-based cure-all for the problems of the health care system. But they sought to find manageable problems where AI could make a real difference in combination with human oversight. Shah argued that staffing shortages are a prime example of this type of problem.
Bridging AI and the Human Touch
According to its program, the premise of the panel was that “a combined human/machine i.e. ‘centaur’ approach that leverages the power of AI, augmented by an expert human touch, will shape the majority of decisions [in health care].”
For Munjal Shah, this is the key to training LLMs in the health care space. Hippocratic AI has hired thousands of nurses and other health care professionals to test and train the LLMs that could be used for nondiagnostic applications. Because LLMs work by generating appropriate responses based on a particular dataset of similar responses, the ideal training strategy in health care is to devise a training set of evidence-based, expert responses and to have real human experts judge whether the LLM is able to mirror the sort of response they would give or that they would expect from a human colleague.
Shah emphasized the importance of reinforcement learning with human feedback when designing LLMs. This is the process by which humans inform an LLM when it is wrong or off topic. Over time, the system learns not to make the same mistakes. This process of human feedback, coupled with what Shah calls “overtraining” on expert health care sources, is crucial to building trustworthy LLMs.
“We said, ‘Hey, you can’t build generative AI safely without first partnering with health systems and payers,’” he explained during the panel. “You really have to create a safety governance council. You’ve got to work with folks, you’ve got to get their elements. But what we came back to is the realization that there are a lot of use cases that actually make a lot of sense to create [AI] staffing around.”
Generative AI and ‘Super-Staffing’
Shah refers to this virtual staffing as “super-staffing” because of its potential exponential scale. AI can provide nondiagnostic services that could cost close to $100 per hour for a human nurse to cover. Given this cost, and the fact that humans are unable to reach all of the people who need this care, AI that costs around $1 per hour and isn’t susceptible to burnout or time constraints could be a viable solution. The goal is not to replace human beings with AI, but rather to augment humans with AI to cover the ground that needs to be covered — ground that’s impossible for humans to cover alone.
“You can’t call every patient two days after they start every new medication,” said Shah. “But at this cost structure, maybe you can, maybe you can give every single person with two or more chronic diseases in the country, which is 68 million people, their own chronic care nurse.”
Some of the services Hippocratic AI is focusing on beyond chronic care nursing include explaining benefits and billing, dietetics, genetic counseling, pre-op and post-op instructions and questions, and delivering negative test results.
The key is that the LLMs used for these applications are capable of conversing in a very human-like style, replicating the human touch that we expect when interacting with health care workers.
In fact, a recent study published in the Journal of the American Medical Association found that participants preferred responses to patient questions drafted by ChatGPT over those prepared by physicians on measures of both quality and empathy.
“Its main strengths are conversational AI, which I think in a health care setting means we finally have the technology for patient-facing conversations,” Munjal Shah said of generative AI.
“And the second thing is it can reason across lots of documents, but it’s not typically better at one specific task than classifier AI,” he continued. “But if you’re focused on something that needs to converse much better, that needs to talk to patients that needs to interact, that needs to even use general analysis, you should really focus on generative AI.”
The Editorial Team at Healthcare Business Today is made up of skilled healthcare writers and experts, led by our managing editor, Daniel Casciato, who has over 25 years of experience in healthcare writing. Since 1998, we have produced compelling and informative content for numerous publications, establishing ourselves as a trusted resource for health and wellness information. We offer readers access to fresh health, medicine, science, and technology developments and the latest in patient news, emphasizing how these developments affect our lives.
