Mayo Clinic clinicians are starting to explore how they might use healthcare-specific large language models – accessed through a generative artificial intelligence chat application – to enhance patient care and improve clinical decisions.
Where ChatGPT and Google Gemini may only produce relevant, evidence-based answers in healthcare a fraction of the time, California-based Atropos Health said its federated healthcare data network can offer healthcare users a detailed, accurate consultation to even the most obscure medical questions because it’s based only on peer-reviewed, real-world data.
For example, when clinicians consider how to treat a patient with an unusual genetic condition that predisposes them to a particular cardiac disease, drawing on data from millions of patients to identify similar patients and learning about their outcomes could help inform treatment, according to Dr. Peter Noseworthy, chair of cardiac electrophysiology at Mayo Clinic.
Last year Atropos launched a generative AI-enhanced platform that allows users to query its trove of clinical data. It claimed that it had become the largest healthcare data network in the U.S. in June. Recently, the company released its chat-based interface, ChatRWD.
“This is a way to interact with real-world data in real-time and then surface those insights at the point of care,” Noseworthy, who leads control trials and research using national datasets has begun testing ChatRWD, told Healthcare IT News on Friday.
Scoring healthcare-specific data reliability
Atropos’ platform provides a Real World Data Score – a moment-in-time snapshot of the data quality measured by size, completeness, patient timelines and more – and a Real World Fitness Score – based on a proprietary algorithm that accounts for how well question criteria are represented within the dataset – for each dataset.
These ratings can help users select the dataset that is most suitable to answering each of their questions on the platform, the company said in June.
Saurabh Gombar, adjunct faculty at Stanford Healthcare and chief medical officer at Atropos, said he led a study that analyzed the accuracy and efficacy of five LLMs, including healthcare-specific OpenEvidence and ChatRWD, to examine the accuracy and efficacy of model outputs.
Looking at reliability, the general-purpose LLMs fell far short of answering physicians’ questions, he said.
“Whereas OpenEvidence and ChatRWD were able to produce actionable, reliable evidence either 42% or 60% of the time – an entire magnitude greater than the general purpose LLMs,” Gombar told Healthcare IT News in July.
Since 2022, Atropos has been working with Mayo Clinic to pilot and develop data-driven methods that can improve healthcare – both techniques and improving care delivery to historically under-represented patients – by serving up real-world evidence de novo through automated reports called Prognostograms.
The collaboration allowed physicians and researchers on Atropos’ digital consult platform to access Mayo Clinic’s deidentified data repository and its analytical tools.
“This is a way to interact with real-world data in real-time and then surface those insights at the point of care.”
Dr. Peter Noseworthy, Mayo Clinic
For patients in critical care, the ability of their care teams to find answers to questions locked in research through the platform could save time. While it could take many weeks to determine treatment through traditional means, an AI-driven Prognostogram can be completed in a matter of days, according to Atropos.
Noseworthy noted that observational clinical researchers have access to large real-world data, “but the timeline to generate insights from that is months.” They not only need to pull data, but clean it and analyze it.
“You need to have a statistician to work with,” he said.
“With a tool like this that essentially can set up those studies in real-time and pull those data, you can get pretty close to what is research grade or publication-grade information just through a chat interface.”
Atropos said it is projecting more than 200% growth in additional dataset availability over the next year.
The power of patient data emerges with AI
Where experienced clinicians can recognize patterns in patient outcomes and responses to treatments based on experience, capturing the totality of experience with a medication or treatment – “the gestalt of patient outcomes” – has been constrained by traditional modalities of medical research, Noseworthy explained.
“We could get at that with clinical trials, but that’s a slow process and patients are highly selected.”
LLMs, however, can offer physicians faster answers to medical questions, and that may help to improve treatments for patients that are historically outside of the reach of clinical trials.
“Rare or unusual presentations of disease or rare conditions or rare confluence of conditions are not well characterized in clinical trials, but they’re present in a large data sample,” Noseworthy said.
Mayo Clinic has been working to extend the limits of clinical trials beyond the walls of major academic medical centers, launching a decentralized clinical trial program last year.
Access to clinical trials has exacerbated health disparities, according to Dr. Tufia Haddad, a medical oncologist, chair of faculty development for the Mayo Clinic Department of Oncology and coleader of the clinic’s Comprehensive Cancer Center Office of Platform and Digital Innovation.
“We have an underrepresentation of racial-ethnic minority populations, patient populations, in our trials, as well as underrepresentation of those in underserved rural-based communities,” she said.
The overarching goal of improving clinical trial access is to “bring more cures to more people,” she told Healthcare IT News after the program launched.
While some practices within Mayo Clinic have been piloting ChatRWD, Noseworthy said he became interested because colleagues in his cardiac group have experience using real-world data from other data engines.
“It was attractive to me that we could essentially generate the data in real-time and at the point of care,” he said.
Using real-world clinical data – “it’s vastly different than just using ChatGBT” or other LLMs.
While ChatRWD has yet to be deployed at scale at Mayo Clinic, “it’s been able to give us some interesting insight,” Noseworthy said.
Andrea Fox is senior editor of Healthcare IT News.
Email: afox@himss.org
Healthcare IT News is a HIMSS Media publication.
Source : Healthcare IT News