Artificial intelligence technology is changing health care in a real way
Imagine if physicians could predict a patient's diagnosis, ideal treatment regimen, or health outcome. This is the promise of artificial intelligence (AI), and it's not fictional—this branch of computer science is being used today at BIDMC and across the Beth Israel Lahey Health (BILH) system. "There is great value in delivering the right care at the right time in the right setting," says John D. Halamka, M.D., M.S., Harvard Medical School's International Healthcare Innovation Professor and executive director of BILH's Health Technology Exploration Center (HTEC). "That's exactly what AI allows us to do."
AI—which enables computers to learn without being programmed through algorithms and patterns—can be applied to virtually any aspect of medicine. Case in point: a sophisticated new machine learning model integrating AI technology with hospital data can effectively detect where simple operating room schedule changes would improve efficiency and balance BIDMC's load during busier times. "Because our services are in such high demand, we are always looking for innovative ideas to help maximize our capacity," says BIDMC Chief Information Officer Manu Tandon, M.B.A., M.P.A. Using information about surgical procedures, physician schedules, and patient length of stay, this machine learning model performs calculations and makes recommendations. As a result, two simple changes were made to orthopedic surgeons' schedules leading to an astonishing 18 percent reduction in beds needed early in the week. "This significant decrease in demand—when the hospital is busiest—has a positive effect on operations and enhances our ability to care for patients who would otherwise be waiting for beds," adds Tandon.
"We have reached a point where medicine's complexity cannot be solved with the human mind," says Halamka, a globally renowned thought leader in health care information technology. "Even the most experienced physicians misdiagnose patients. Human error is inevitable." The National Academy of Medicine confirms diagnostic errors contribute to 10 percent of patient deaths. But according to Halamka, this situation will soon be remedied as data analytics, forecasting, and machine learning—a subset of AI and the powerful technology behind web searches and speech recognition software—transform the clinical decision process. "Big data allows us to look at a million patients from the past to influence the care we provide the patients of the future," says Halamka.
This kind of innovation is not new to BIDMC, one of the first hospitals in the world to apply computers to patient care. Halamka—who with his team developed one of the first electronic health records in 1985 and launched the first patient portal in 1998—has a long history of collaboration with organizations like Google, Amazon, the Bill & Melinda Gates Foundation, and the Massachusetts Institute of Technology (MIT), as well as governments around the world. Today, BILH continues to expand this legacy by leveraging AI to improve diagnosis, personalize care, and enhance quality and safety. Essential to HTEC's success is the generosity of visionary supporters. Notably, BIDMC received $6 million from the global private equity firm HAVY International to support health care information technology efforts system-wide. Among other gifts is an academic research sponsorship grant from Amazon. "We are thrilled to be embarking on this pioneering multi-year research program to evaluate machine learning," says Tandon. "This work is scalable across the health care industry, with the goal of advancing outcomes for patients everywhere."
Thanks to this support, Halamka, Tandon, and their teams are leading a range of AI initiatives that positively impact BIDMC and BILH's bottom line—and, most importantly, enhance patient-centered care—through everything from reducing hospital length of stay to maximizing hospital capacity to managing chronic diseases to preventing missed appointments. "Patients no-show for appointments for many reasons," says Halamka. "Maybe English isn't their first language; maybe they don't have a car. So we use machine learning to predict who is less likely to make it, and then we reach out proactively to help them."
Halamka also serves as a mentor to HTEC physician–engineers who are leveraging AI to disrupt health care delivery. As a trauma surgeon, Gabriel Brat, M.D., M.P.H., is addressing one of the nation's greatest public health crises: the opioid epidemic. With data showing that addiction often originates after surgery, Brat has launched an unprecedented effort to optimize post-surgical opioid prescribing. He is using crowd-sourced national data and AI to personalize opioid prescribing based on patient risk profiles during hospitalization, which BIDMC surgeons are using in clinical practice. As this project expands, it has the potential to become a global prescription resource—with the aim of ultimately reducing both addiction and the number of unused opioid pills.
Emergency medicine physician Steven Horng, M.D., M.M.Sc., is working with collaborators at MIT to bring machine learning to the bedside by automating clinical workflows and amplifying human cognition. Using next-generation user interfaces powered by machine learning, Horng and his colleagues are reimagining how clinicians interact with computers. They have created a redesigned, interactive, user-friendly electronic health record interface that acts as a personal digital assistant, collecting and synthesizing relevant patient information, as well as providing diagnostic and therapeutic suggestions. Horng believes this technology will result in decreased diagnostic error and physician burnout.
Innovative work in AI has piqued the interest of leaders, physicians, and staff across the medical center and its system. "Emerging AI technologies offer great promise for BIDMC," says BIDMC Chief Medical Officer Anthony Weiss, M.D., M.B.A. "Beyond advancing our clinical care and physician–patient communication, this technology also has vast potential to help us continually improve our workflow." And with early research demonstrating that AI improves patient outcomes in a wide range of conditions from diabetic retinopathy and sepsis to melanoma and colorectal cancer, there is more to come. "We are just scratching the surface of what AI can do," says Halamka. "It will help us profoundly impact patient lives."
We have reached a point where medicine's complexity cannot be solved with the human mind
John Halamka, M.D.
International Healthcare Innovation Professor, Harvard Medical School
Executive Director, Health Technology Exploration Center, Beth Israel Lahey Health
Computer-aided polyp detection is the future of GI care
As a world-renowned academic medical center, BIDMC serves as a major referral hospital for patients from across New England—and even nationwide—who have been diagnosed with advanced polyps or colon cancer. Early on in his career, gastroenterologist Tyler Berzin, M.D., noticed a recurring trend that troubled him. "When patients are referred to us with large polyps or colon cancers, even after having undergone recent screening colonoscopies, it's a clue that some polyps may be getting missed," says Berzin.
With literature showing that up to 20 to 30 percent of polyps are not identified by physicians during colonoscopies—which in turn allows the polyps to become larger and more advanced—Berzin became determined to address the problem. Larger polyps may be more difficult to remove endoscopically or surgically and can sometimes already harbor areas of cancer. "My question was, 'How can we reduce the likelihood of polyps being missed in the first place?'" says Berzin.
He knew that one answer was likely AI, but at the time the software didn't exist. "The key was training a computer to detect polyps," Berzin says, explaining that small, flat, or flesh-colored polyps can easily be camouflaged within human tissue. "We just needed a team to create the AI software to do that." In 2015, Berzin published a paper describing this need, which was followed by the creation of several AI-aided polyp detection systems. This software enables a computer to learn from images of subtle polyps and then to recognize new ones. "The explosion of AI truly aligned with our need," says Jeremy Glissen-Brown, M.D., a fellow on Berzin's team.
Today, Berzin is principal investigator of a first-of-its-kind trial in the U.S. to evaluate this technology in patients at three sites: BIDMC (the primary site), New York University, and The University of Kansas. "We believe AI-assisted colonoscopy will become the standard of care," says Berzin, whose work has already yielded promising findings in two key international trials based in China. "The physician will always be essential to a colonoscopy, but we only have two eyes, whereas the computer has the ability to be attentive to all areas of the screen at the same time. This technology will help us save lives."
Robots and AI help clinicians and enhance safety
The advancement of AI begs the question: When will we start to see robot health care workers collaborating alongside humans? As a physician dedicated to improving the experience of women throughout the often-challenging pregnancy, labor, and delivery process, Neel Shah, M.D., wondered that himself. "I was curious whether the use of robots might be a plausible way to enhance the quality and safety of care we deliver," says the BIDMC obstetrician and gynecologist.
Together with his wife, roboticist Julie Shah, Ph.D., who leads the Interactive Robot Group in the Computer Science and Artificial Intelligence Laboratory at the Massachusetts Institute of Technology, Neel conducted a study to explore AI's potential to help physicians perform complex clinical tasks. "In almost every job, some part of a person's work can be easily accomplished by robots," says Julie. "In the hospital setting, if we are able to offload even the most simple decisions and tasks and free up nurses' cognitive capacity to handle the most challenging decisions, we might significantly improve the safety of care."
The couple and their team conducted a pilot study where they placed a robot named Ginger on the labor floor and programmed it with AI technology. This program enabled Ginger to help nurses decide how to assign rooms to patients and which personnel to staff for certain procedures. "Our robot used computer vision to read clinicians' handwriting," explains Julie. "Then, through AI, she learned patterns and was able to make these decisions on her own."
In the end, the Shahs showed that Ginger was able to make recommendations with a high degree of accuracy, and they published their findings in The International Journal of Robotics Research. This work demonstrated AI's vast promise to supplement, rather than replace, clinicians in performing complicated jobs. "What is clear is that this technology can augment clinicians' ability to make complex decisions and, by extension, their performance—but it does not replace them," says Julie. "Designing robots like Ginger as collaborators really opens up the many possibilities of what AI can offer." Adds Neel: "We're better together."