Since ChatGPT entered the Zeitgeist, it has reshaped how patients and clinicians get their medical information and interact with the healthcare system.
Mass General Brigham’s Marc Succi, MD, and his research group, have approached the use of generative artificial intelligence (AI) and large language models (LLMs) for medical innovation with enthusiasm — while still exercising caution.
Marc Succi, MD
Their research has shown that while AI and LLMs can fill some current gaps in the healthcare system, there are still limitations that will prevent them from taking over for your doctor anytime soon.
“AI is like a bicycle — just like bicycles enabled humans to own the most efficient mode of transportation among animals, the hope is that AI will significantly augment the modern-day clinician," says Succi, founder and executive director of the MESH Incubator at Mass General Brigham, a center that enables healthcare researchers of all levels of experience to translate their technological solutions into clinically impactful products.
He discusses the MESH Incubator’s recent studies and underlying problem-solving approach to medical innovation.
The Underlying Ethos of Medical Innovation at MESH
It always starts with a question. That is the underlying ethos that guides any project that researchers at the MESH Incubator pursue.
“We always start off by trying to figure out a really meaningful question or unmet need that exists in medicine and seeing whether we can apply the right technology or tool to address it,” says Succi. “And in a lot of cases these days, we can.”
And when they do pursue a health care problem they can potentially solve with technology, they move quickly.
In the past two years alone, the MESH Incubator has published more than 50 high-impact studies investigating AI and LLMs, and examining these how new technologies can impact clinical operations and value-based care across all medical specialties.
This is all in addition to spinning out numerous startups in the fields of digital health and devices based on ideas that have emerged from MESH's biodesign division.
The traditional method of conducting research, which includes applying for and securing grants, conducting the investigation, writing up the results and submitting a study for publication, can often take months or years.
In the fast-paced world of AI, that often means there’s a newer model on the market by the time the study publishes.
The MESH Incubator, however, brings an entrepreneurial approach to research. It serves as a self-funded entity that is equipped to bring a project from a question to a rapid solution.
Succi attributes much of this success to his commercial-first model, as well as a creative culture within the new generation of students and researchers that power his 30-person research division, “MESH IO.”
Here, Succi encourages (and even demands) that creative early-career students develop their own ideas for medical innovation, and with his mentorship, truly lead their own research projects.
“I always felt if you create an environment where curiosity, collaboration, and problem-based execution are top priorities and encouraged that students will be empowered to participate and actively develop their own conceptual basis for new studies and technologies,” he says.
Paging Dr. Chatbot
As his team’s pioneering research investigating the use of LLMs in clinical care grew, Succi expanded his AI subteam to multidisciplinary researchers — data scientists, engineers, business leads, and of course, clinicians — who are advancing studies on the integration of LLMs and AI in clinical care.
The team has produced high-impact papers demonstrating that LLMs such as ChatGPT could be leveraged to recommend the best physical exams to administer to a patient, de-prescribe unnecessary or potentially harmful medications in seniors, and accurately recommend imaging services for patients with breast cancer and breast pain.
In one study that became one of the most cited research papers in medicine in 2023, the researchers found ChatGPT showed about 72% accuracy in diagnosing a patient from a clinical workup, roughly the equivalent of a recently graduated medical student.
LLMs improve the more they are used and fed new data, and the researchers are currently studying whether their diagnostic abilities will improve over time (Spoiler alert: they have).
While many patients and doctors fear the prospect of AI taking over the clinic, Succi feels strongly that AI will “augment” intelligence for doctors, not replace them altogether.
“Managing the medical care of a patient, getting them to buy into treatment plans and gaining trust – that’s what the art of medicine is,” he says. “AI might augment some aspects of that care, but it can’t replace the doctor-patient relationship.”
AI-Enabled Medical Education
The latest question raised in MESH came from members' own experiences in medical school. They wondered if there was a better way to assess the diagnostic and clinical skills of medical trainees in real-world scenarios other than hiring actors to pose as patients.
Standardized patients (SPs), or actors who are trained to simulate medical conditions, have been essential to clinical exams like Objective Structured Clinical Examinations (OSCEs), helping students practice medicine in a realistic setting without risk to actual patients.
SPs enable evaluations of a resident’s cultural dexterity, which ensures a physician can deliver equitable care regardless of a patient’s background.
But there is a limit to the number of actors available and the clinical scenarios they can portray, which poses challenges across U.S. training programs. MESH researchers believed LLMs could offer a scalable alternative.
In a pilot study recently published in the Journal of the American College of Surgeons, the team showed that a Standardized Patient LLM (SP-LLM) they developed provided realistic, clinically accurate simulations that supported culturally sensitive communications and exponentially increased the number of scenario.
Arya Rao
“A residency program needing 15 actors to test 15 different scenarios is rather primitive,”says Harvard MD-PhD student, Arya Rao, lead author of the study and Chair of the MESH IO AI subteam. “Instead of training on one standardized glioblastoma patient at a time during a clinical rotation, you can now train on 500 at home on your own time.”
SP-LLMs are now being deployed across Harvard-affiliated training programs with the help of grants from HMS and pharmaceutical industry, with several other studies testing the technology currently under peer review. The team is also commercializing this technology under a new startup spinout.
Medical Innovation with an Entrepreneurial Spirit
Succi has always had an entrepreneurial drive. The Toronto native went from high school robotics competitions in his native Canada to volunteering at one of the largest biomedical engineering labs in the world at MIT to inventing technology to regenerate biological cells that would be licensed to industry and spinning out his own ventures — even landing on a Forbes 30 under 30 list at only 25 years old.
His university education emphasized problem-based learning and small-group tutorials over traditional lectures, which has inspired the MESH Incubator’s approach.
Since 2016, the incubator has supported more than 3,000 clinicians and researchers, helping to facilitate over 25 new startups that have gained more than 22x follow-on funding, and over 100 publications — earning a global reach and reputation.
A Bootcamp to Break Down Barriers
Frustrated by the barriers to his own pursuit of translating discoveries to clinic, Succi created the MESH Core Healthcare Innovation Bootcamp, which brings together clinicians, researchers, investors, startups and entrepreneurs from over 20 countries to explore new health care technologies and learn hands-on skills for success from investors and inventors.
This year’s hybrid conference attracted more than 500 attendees. Succi and colleagues followed this event with a two-day intensive MESH boot camp in Riyadh, Saudia Arabia, attracting over 1,000 registrants from the region, with support from top hospitals and ministers.
In February 2026, they’ll be expanding to World Health Expo Dubai (formerly Arab Health), the largest medical conference in the world, bringing together more than 270,000 healthcare professionals.
Like the members of MESH, bootcamp participants will be encouraged to begin the innovation process by asking a question and working towards a solution. While innovative new technologies such as AIs and LLMs could be part of the solution, they are not the only answer.
“I try to assess new technologies not by their inherent complexity or public hype, but by asking how they can function as tools to augment our mission to provide cutting-edge healthcare, says Succi. "AI just happens to be a leading disruptive technology right now in terms of quality and access to medical care."
"In five years, AI might mature in its implementation and plateau, and there will be a newer technology available," he says. "When that happens, we’ll be studying it.”
Subscribe to BenchMarks!
Subscribe to our monthly research newsletter and be the first to know what’s shaking in science at Mass General Brigham. From groundbreaking discoveries to the latest “you-heard-it-here-first” breakthroughs, we’ve got the updates that’ll keep your curiosity grooving.
Leave a Comment