N.YU Langone Health’s MCIT Department of Medical Informatics, Institute for Medical Education Innovation, and Institute for Health Excellence will host the first Generative AI Prompt-A-Thon in Healthcare on August 18th. A team of clinicians and educators participated during the event. and researchers will work together to find artificial intelligence (AI)-powered solutions to healthcare challenges using real-world, anonymized patient data.
The event provides possible choices for the next word in a sentence, paragraph, or essay based on how real people have used the word in context billions of times in documents on the Internet. We cover large-scale language models (LLMs) that predict. Also called generative AI, such systems input random combinations of possible next words, giving variety and creativity. As a side effect of this next word prediction, the model is “adept” at summarizing long texts, extracting key information from databases, and generating human-like conversations as a chatbot.
Despite these advances, event organizers say such AI programs don’t think and can produce conclusions and references that don’t exist. Therefore, especially in the medical field, where technology has the potential to increase safety and improve care, it requires close supervision by human users.
Anticipating the boom in the generative AI space, in March, NYU Langone requested access to its latest generative AI tool from Microsoft (OpenAI), a partner of the company that created ChatGPT. Azure is Microsoft’s cloud computing platform that offers private instances of GPT4, a new relative of the famous LLM ChatGPT, to clients such as his NYU Langone. This allowed the application to provide teams in the health system with secure access to software and servers, allowing the tool to meet federal privacy standards.
“We have introduced the nation’s first privately managed, secure, HIPAA compliant GPT4 ecosystem to a healthcare facility,” said Nader Merabi, executive vice president, vice dean and chief digital information officer at New York University Langone. ‘ said. “This has enabled us to launch a large-scale effort to test potential medical applications of large language models like GPT4 in a safe and responsible manner.”
Physicians, nurses and administrators at New York University Langone agreed on strict terms to write prompts to a private instance of GPT4 to see how well it produced patient-friendly explanations and improve care plans. to suggest or warn of potential safety issues.
“Equally important is the employee’s ability to identify the limitations of these models,” added Jonathan S. Austrian, M.D., Deputy Chief Medical Information Officer, Inpatient Informatics. Even if it finds new uses for GPT4, he says any AI effort will only help improve the work of healthcare providers.
At the August 18th event, a team of clinicians, educators and researchers will collaborate to test GPT4-based solutions to healthcare challenges using real-world, anonymized patient data To do. Held at the Science Building of NYU Langone, the day of the event will kick off with a “Lightning Round” talk by the following NYU Langone experts:
- Mark M. Triola, M.D., Associate Dean of Educational Informatics, on the New World of AI in Healthcare
- Dr. Tim Requarth, Science and Writing Lecturer, Vilcek Institute of Graduate Biomedical Sciences, on Using AI Tools for Scientific Writing
- New York University Data Science Center PhD student Lavender Jean, BA, participating in the NYUTron “AI Doctor” project
- Dr. Kelly Owens, Assistant Professor, Department of Population and Health, On the Ethics of Generative AI in Medicine
Indaron Afignanaphonse, M.D., Director of Operational Data Science and Machine Learning at NYU Langone University and leader of the Predictive Analytics Unit in the School of Health Informatics, previews the day’s activities.
At lunchtime (11:30am – 12:30pm), participants (80 registrants selected from over 400 entries) will join pre-assigned teams of four to participate in the event. decide which problems to solve during the workshop portion of the . Participants were previously grouped into teams based on their interest and background in generative AI. Common interests included patient education, diagnosis and treatment, diversity and equity. Research-themed groups focus on grant application support and research literature summaries. Working groups are organized across interests, with a mix of levels of comfort in GPT4, from inexperienced to experienced experimenters. After lunch, participants head to the meeting room to start working on the project (12:30-14:30). An AI Mentor, along with other participants, will support each team’s “Prompt Her Journey”.
AI efforts at scale
Prompt-a-Thon is part of a larger effort at NYU Langone to empower employees about the potential benefits of generative AI. The MCIT Department of Medical Informatics has provided over 200 doctors, nurses, researchers and educators with exploratory access to her private instance of GPT4 in the healthcare system to conduct experiments. Additionally, over 100 people have submitted formal project requests, and approved efforts are mentored by AI experts to prepare solutions for real-world use.
Paul A. Testa, M.D., Ph.D., Forensic Medicine, MPH, chief medical information officer at New York University, said, “Given the recent advent of this tool and the level of interest, there are many people with ideas for generative AI. has not yet held an individual session with the AI leadership team.” Langone. “Thus, our goal at Prompt-A-Song is to give more employees the opportunity to explore their ideas and connect with other members of the generative AI community at New York University Langone Health. is to do.”
One example of a promising generative AI solution under development at NYU Langone University includes a project led by Fritz François, M.D., Director of Hospital Operations. He studies the benefits of generative AI models that can review clinical notes to find examples of his two drugs being used. Anticoagulants and immunosuppressants were on the care plan but not on the patient’s active medication list. In such cases, the AI will send a prompt to the physician warning of this potential discrepancy. This project recently went live across the healthcare system, making it the first generative AI intervention introduced into clinical care.
“Discrepancies like this are extremely rare, but generative AI is expected to help eliminate them once and for all,” added Dr. Testa. “Importantly, while NYU Langone is already one of the safest healthcare systems, we are committed to becoming responsible stewards of new technologies that can further advance safety.”
NYU Langone’s rapid adoption of generative AI builds on years of experience applying more traditional machine learning AI models to clinical care. This includes one that addresses natural language processing like his LLM specific to New York University called NYUTron. Nature In July 2023, read the doctor’s note to accurately estimate the patient’s length of stay and other factors important to their care. Other rule-based pattern recognition machine learning projects scan image results and ECG readings to flag potentially unidentified disease in patients (using the single-lead ECG on the Apple watch). (e.g., pre-diabetes). Some older AI models can be “supercharged” by combining with generative AI.
Additionally, LLM is online at New York University Langone as part of a healthcare system that has spent 10 years building 83 data informatics dashboards. These dashboards are about 750 safety and care effectiveness monitoring dashboards (for example, flagging an infection spike in a particular hospital). floor).
Follow the latest on Generative AI Prompt-A-Thon in Healthcare on Twitter/X with the hashtags #NYUAI and #PromptAThon.