‘A continuous cycle of learning’
Department of Health Outcomes and Biomedical Informatics builds collaborative, AI-enabled learning health system
March 9, 2023 — For any health care organization to provide the best patient care long term, continuous learning and improvement must be a priority. There is something to be learned from each patient visit, diagnosis and treatment that can help produce a better experience for others in the future.
Under the direction of the National Academy of Medicine, academic health centers across the U.S. are striving to create this ideal by developing learning health systems. In these systems, science, informatics, incentives and culture combine for ongoing improvement, best practices are smoothly implemented and data are collected to continue the cycle.
Led by Chair Elizabeth Shenkman, Ph.D., the University of Florida College of Medicine Department of Health Outcomes and Biomedical Informatics, or HOBI, has been building a learning health system with UF Health and the University of Florida Clinical and Translational Science Institute, or CTSI, for years. Thanks to successful team science and UF’s artificial intelligence initiative, progress has developed rapidly, with promising impacts for students, scientists, practitioners and patients across the university, state and country.
Since the learning health system initiative began at UF, UF Health and CTSI leadership, clinicians, scientists and community representatives known as citizen scientists have collaborated to fund pilot programs to improve the quality and outcomes of patient care. Several of these programs have since received support from the National Institutes of Health and led to practice changes at UF, Shenkman said.
Data infrastructure has also grown to support the initiative and improve medical research. The OneFlorida+ Clinical Research Network, a group of united health care institutions around Florida and portions of Georgia and Alabama, is one such example. Launched in 2013 by the CTSI, Shenkman and David Nelson, M.D., senior vice president for health affairs at UF and president of UF Health, this resource includes over 20 million patients and their records. It is also part of a national clinical data research network known as PCORnet, which merges health data, research expertise and patient insights to answer important health questions and advance health outcomes nationwide.
William “Bill” Hogan, M.D., M.S., serves as director of biomedical informatics within the CTSI and HOBI, as well as director of informatics for OneFlorida+. He said the network has been an essential engine in allowing UF researchers and collaborators to conduct larger, more equitable studies on leading health challenges, like hypertension and diabetes.
“OneFlorida+ is like a magnet,” Hogan said. “It’s having a huge impact.”
Shenkman agreed, adding, “We are very excited about the science that we lead and our team science approaches. We have had a wonderful trajectory of sustained success and growth. We look forward to being a hub for exciting work around artificial intelligence and continuing to focus on biomedical informatics and implementation science to ensure our novel discoveries are translated into practice.”
Navigating big data
To deidentify and analyze millions of patient data records, HOBI’s experts have developed leading-edge AI and natural language processing models with the help of UFs supercomputer, HiPerGator AI — the fastest artificial intelligence supercomputer in higher education.
Recently, Hogan, Shenkman, UF Health Chief Data Scientist Jiang Bian, Ph.D., and Yonghui Wu, Ph.D., the natural language processing director at CTSI and OneFlorida+, published a paper introducing GatorTron™, the largest clinical language model in the world. GatorTron™ was developed from scratch at UF and includes over 82 billion words of de-identified clinical text.
It is a game-changer for researchers, Wu said, because it allows easier exploration of unstructured patient data, like physicians’ narrative notes, rather than just coded information, like disease diagnoses and medications administered. Much research has historically been conducted using only the medical code data, he said, but a huge amount of patient information is stored in the notes.
“Being able to utilize the unstructured patient information, such as social determinants of health, in a large scale allows for more accurate, fair, inclusive and equitable research,” Wu said. “The intelligence that emerges from GatorTron™ is amazing. The next generation of GatorTron™ is even smarter; it can help autocomplete sentences and paragraphs of a clinical document, like a progress report, to reduce the documentation burden, assist in differential diagnosis and composing treatment plans and contribute to next-generation intelligent electronic health record systems with its conversational AI ability. This will culturally revolutionize health care.”
AI analysis has skyrocketed in popularity as the computer science and biomedical informatics fields expanded over the past 15 years. Many trivial daily tasks use AI, like checking the weather on your phone app in the morning or mapping which route home is fastest using your GPS. AI’s efficiency and real-time information gathering abilities are powerful tools for health care, according to Rui Yin, Ph.D., an assistant professor in the division of biomedical informatics.
Yin is working to merge precision medicine and public health through AI, ultimately aiming to solve real-world problems, like predicting the antigenicity and potential variants of RNA viruses such as influenza and SARS-CoV-2 or identifying markers for rare diseases and suggesting diagnoses and treatment options.
At UF, he is leading the effort to build a comprehensive genomic platform that could help clinicians better understand and diagnose inherited genetic diseases. Currently, genomic sequencing and analysis is a costly, laborious and time-consuming process. But with well-built AI models, diverse patient data from UF Health and OneFlorida+ and a supercomputer like HiPerGator AI, Yin is working to speed up the process and compile a diverse, searchable database for researchers at and beyond UF. He hopes to complete the computational pipeline within a year and the integrative database in the next five years.
“Precision public health is the natural evolution of precision medicine and public health, and AI holds the promise to implement it,” Yin said. “I hope that my research will benefit not only the majority of groups but people with rare or undiagnosed diseases. It will also further reduce health disparities and advance health equities. I think it’s a job worth doing.”
Another AI example is the Real-time Online Assessment and Mobility Monitor, or ROAMM, developed at UF in part by HOBI researcher Todd Manini, Ph.D., and his team. ROAMM is a smartwatch app that collects and monitors health data with the goal of reducing patients’ repeat follow-up visits and providing objective information for health events like hospitalizations.
While UF’s models and resources are unique in the U.S., the university’s experts are devoted to increasing accessibility, education and research collaboration opportunities. GatorTron™’s models are publicly available, and a universitywide initiative is bringing AI training to all colleges and interested students. Faculty can also access free or reduced-cost AI classes for continuing education credit, and UF aims to expand training opportunities to other Florida schools and universities.
Within the College of Medicine, Hogan has been introducing medical students to biomedical informatics since the 2015-16 academic year through a fourth-year elective class. The knowledge and interests they gain from this exposure sets them up for potential clinical informatics fellowships and careers as medical information officers or clinician scientists.
“We’re really poised to be leaders in putting AI into clinical practice,” Hogan said. “As we continue to up our game, we’ll be out there blazing more trails and delivering nation-leading training.”
Implementation science is imperative
In addition to promising data and infrastructure, HOBI has the rare benefit of interdepartmental collaboration with implementation scientists like Ramzi Salloum, Ph.D., an associate professor and director of the learning health system initiative at CTSI.
Implementation science is a crucial field, Salloum said. He and his colleagues collaborate with the biomedical informatics division and the broader community to disseminate evidence-based practices throughout the college and UF Health, speeding up science and discoveries into practice to create a more equitable and efficient health care system.
“Every patient who receives care in the health system is an opportunity for us to learn and improve health care delivery,” he said. “With the data systems, AI and capabilities we have now, we want to make this a continuous cycle of learning that maximizes the benefits of science to improve population health.”