Smart Tech in Medicine: How AI is Changing Doctor Training

Doctors often face a mountain of patient data that can be messy and incomplete. This makes it tough for them to stay updated with the latest research and developments. NYU Langone Health in New York City has found a clever way to help the next generation of doctors tackle these issues. They've created a large language model (LLM) that acts as a smart research buddy and medical advisor. Every night, this model digs through electronic health records (EHR) and matches them with relevant research, diagnosis tips, and important background info. It then sends this information in a neat, personalized email to residents the next morning. This is part of NYU Langone's unique approach to medical education, which they call "precision medical education. " It uses AI and data to create highly customized learning paths for students. The model they use is based on the latest version of Llama-3. 1-8B-instruct and an open-source database called Chroma. It doesn't just access documents—it actively searches for the latest research. Every night, it connects to the facility's EHR database, pulls out medical data for patients seen the previous day, and searches for background information on diagnoses and medical conditions. It also looks through medical literature in PubMed, which has millions of papers. The LLM then picks out a few of the most relevant papers and puts everything together in an email. Early the next morning, medical students and residents receive a personalized email with detailed patient summaries. For example, if a patient with congestive heart failure had a checkup the previous day, the email will provide a refresher on heart conditions and the latest treatments. It also includes self-study questions and AI-curated medical literature, as well as tips on what steps the residents could take next. This AI system is a big part of NYU Langone's precision medical education model. The institution has collected lots of data over the past decade about students' performance, the environments they're in, the EHR notes they're writing, and the clinical decisions they're making. They also have a vast catalog of resources available to medical students, like videos, self-study materials, and online learning modules.

The success of the project is also thanks to the medical facility's streamlined architecture. It has centralized IT, a single data warehouse for healthcare, and a single data warehouse for education. This allows Langone to combine its various data resources effectively. The main goal of this project is to link the diagnosis, the context of the individual student, and all of the learning materials. Generative AI has enabled the school to move away from a "one-size-fits-all" model. It's important that students get tailored education throughout their schooling, as well as "educational nudges" that adapt to their needs. There have been challenges along the way, such as model "immaturity. " Sometimes the LLM makes unexpected choices, but the team has been working on refining the prompts and grounding the model. The result has been remarkable. The team is confident that their approach can serve as a great example for other medical institutions. They believe that other medical schools can follow their lead, even those with limited resources. The system is designed to be reproducible and easily disseminated across healthcare. There are concerns about biases in AI systems, but in this case, it's mostly about searching, choosing from a list, and summarizing. The bigger concern is about "unskilling" or "deskilling. " Some people are resistant to giving up certain tasks to AI or digital systems. However, the amount of medical knowledge available today and the fast pace of clinical medicine demand a different way of doing things. AI can handle researching and retrieving information better, and that's an uncomfortable truth that many people are hesitant to believe. Instead, it's important to figure out the co-pilot relationship between humans and AI, not a competitive one. AI can give doctors superpowers and help them make better decisions.

Actions