AI Initiative at VP&S Aims to Advance Health

Artificial intelligence (AI) has the potential to transform clinical care by helping providers make faster, safer, and more informed decisions. At Columbia, that promise is already being realized through a range of clinical tools, such as an early-warning system that decreases risk of in-hospital mortality and a tool for guiding safer infant spinal taps. Building on this momentum, the Columbia University Vagelos College of Physicians and Surgeons (VP&S) has launched the AI at VP&S Initiative, bringing together an interdisciplinary team of experts who are positioned to be a global leader in AI-driven innovation within healthcare.

"There is a real buzz around AI right now. It is on everyone's radar and the momentum is building," said Noémie Elhadad, chair of the Department of Biomedical Informatics and the leader of the initiative. "There's a real sense of urgency to rapidly integrate AI into research and clinical care. But at the same time, AI still feels nebulous for a lot of people. So we felt the need to to take a step back, reflect on how best to harness AI, and how VP&S, CUIMC, and Columbia can lead in translating its potential into tangible impact for patients and health care systems."

A comprehensive, school-wide initiative

The AI at VP&S Initiative aims to advance patient health, support health care workers, and push forward biomedical sciences through AI innovation. While the initiative officially launched this year, six working groups-clinical care, research, oversight and governance, operations, training and education, and community health-have been meeting for months to develop guidelines on the responsible use of AI at VP&S and exploring ways to use AI across the school's missions.

"This is a call to action, and an invitation to the broader community to collaborate with us as we think about how to scale this work and bring AI into everyday practice at VP&S," Elhadad said during a kick-off event held earlier this year. "It is all part of Columbia's ambitious push to integrate AI throughout its research and training programs."

AI and clinical care

Clinicians at Columbia want access to accurate and safe AI tools to speed up administrative work and improve patient care, said Timothy Crimmins, associate professor of medicine, chief medical information officer at ColumbiaDoctors, and lead of the AI at VP&S Clinical Workgroup.

Teams from across multiple University departments-including radiology, neurology, and ophthalmology-demonstrated tools that can improve clinical decision-making. Columbia's portal to ChatGPT Education is HIPAA-compliant, and with permission, Columbia researchers and clinicians can enter protected health information. Other generative AI tools are available inside the EPIC electronic health record system.

"In our focus groups, we also heard a need to provide guidance in using these tools, so we are offering best practices in AI in health care delivery that focus on some of the pitfalls," Crimmins said. "There are many ways to use AI tools in clinical care, but it's important that we don't become over-reliant on AI or let AI overrule or undermine the decisions we make with our patients."

The working group is now in the process of writing user guides for clinicians to prepare them for using AI in the clinic and the best way to interact with the tools to achieve the best results.

AI in research

The Research Workgroup, led by Noémie Elhadad, is focused on advancing the scientific foundations of artificial intelligence in health care and exploring novel methodologies, to push the boundaries of what AI can do to accelerate discoveries. It also aims to facilitate and foster collaborations within the Columbia community and beyond.

Nurses and AI Collaborate to Save Lives, Reduce Hospital Stays

The CONCERN Early Warning System, an AI tool created by Sarah Rossetti and colleagues, harnesses this expertise, uses machine learning to analyze nursing documentation patterns and predict patient deterioration in the hospital up to 42 hours earlier.

New AI Predicts Inner Workings of Cells

In the same way that ChatGPT understands human language, a new AI model developed by Columbia computational biologists captures the language of cells to accurately predict their activities.

Working Toward a Virtual Cell

Columbia's Mohammed AlQuraishi says an AI-powered cell simulation might be less than 15 years away, and he is working to make it a reality.

AI in medical education

The Training and Education workgroup is working to reimagine the medical school curriculum and provide medical students with the skills they need to leverage AI tools.

For example, medical students can enter symptoms into ChatGPT and ask for a differential diagnosis. "There is a process of coming up with a differential diagnosis that we teach in medical school, starting with broad range of possibilities and then narrowing it down," said Lauren Richter, assistant professor of biomedical informatics and pediatrics, who leads the Training and Education working group with Rosa Lee, associate professor of medicine and senior associate dean for curricular affairs at VP&S. "Using AI tools like this requires helping students evaluate what chatGPT does: Is this a valid differential or not? How does it compare with a clinical expert? There's a lot of value in going through that with students."

The group also works with residents, faculty, and educators. "We want to use AI to help educators improve the medical school experience, to assist with medical simulations, for example," said Richter. The group is looking to create a consult service where educators can get help integrating AI into existing workflows or curricula.

Business processes

The VP&S AI Initiative is also working to introduce AI to speed up business processes across the campus and make routine work more efficient.

"Unlike other working groups, our users probably use AI less frequently," says Namita Azad, vice dean of administration and leader of the business processes working group, "but with great success. Some members of the groups have used AI to slash the time it takes to complete some processes from hours to minutes."

Azad said the group is beginning to provide training to more staff. "We want to ensure that everyone understands what we mean when we say generative AI, or prompt engineering, and we've given them access to CU-GPT to start creating ideas about how we can use AI in our different business units."

Success stories and case examples will be provided in the future on the AI at VP&S website.

Future plans

The initiative is planning several events in the coming months to help researchers, clinicians, staff and students connect with each other and learn how they can use AI.

Past events created by the initiative have helped clinicians understand how they can develop their own AI tools (Clinical AI: From Idea to Deployment to Making a Difference), explore AI-driven approaches to social determinants of health that could optimize patient outcomes and shape future health policy, and explored how federated learning enables collaborative AI development across institutions.

"To me, one of the exciting things about AI at VP&S is there's so much happening already," Elhadad said.