What drives you to investigate the intersection of cardiology and AI?
Cardiovascular disease is the top cause of death globally, and the data used to assess cardiovascular disease is well suited to make prognostications with AI. Currently, in healthcare, we tend to react to problems. Using AI helps us discover heart issues early and be proactive instead of reactive.
Can you share some examples of how you envision AI positively influencing the future of cardiovascular medicine?
AI opens up so many possibilities. Heart failure can be really different from one person to the next—but tailoring care to each patient is possible with AI. A physician reaching out and receiving a response that can guide their care in real time is possible with AI. Automated notes generated while a doctor is discussing the patient’s condition with them is possible with AI. AI can take details from remote management systems to help physicians. It can help physicians obtain clinical insights much faster. We are actually building an electronic health recordbased workflow to be able to bring AI predictions to the heart failure team so they can identify patients and triage them. This solution will help us capture people with undetected disease before it’s too late.
How is NYP’s approach to combining cardiac care with AI different from what’s happening at other top hospital systems?
There are many modalities to gain insights that physicians wouldn’t otherwise be able to predict. It’s part of the strategy to help accelerate the growth at the intersection of data science. We have 10 million electrocardiograms (ECGs) in a format available for use in AI and machine learning models. We’ve been able to aggregate our echocardiogram data across all our hospital sites in a structured form for use in data science. Now we’re creating an architecture to deploy those models, looking beyond the model performance and looking at solution performance. ValveNet, an AI deep learning tool that’s outperformed radiologists in detecting heart failure on chest X-rays, is what it looks like when we’ve brought a model through its life cycle.