Latest / Becker’s Healthcare Podcast / Sanjiv Shah, MD, Director of Research at Northwestern Medicine Bluhm Cardiovascular Institute

Transcript

Sanjiv Shah, MD, Director of Research at Northwestern Medicine Bluhm Cardiovascular Institute · Becker’s Healthcare Podcast

• 0:02
  - There's a version of AI in health care
• 0:04
  - that lives mostly in press releases,
• 0:06
  - pilot programs, partnership announcements,
• 0:09
  - theoretical
• 0:10
  - use cases.
• 0:11
  - And then there's the version that lives in
• 0:13
  - the work, the CIO trying to get a
• 0:15
  - model into production without breaking a dozen workflows,
• 0:19
  - the revenue cycle leader weighing automation against accuracy,
• 0:22
  - the CISO who knows that every new integration
• 0:25
  - is a new attack surface.
• 0:28
  - Becker's eleventh annual IT and revenue cycle conference
• 0:31
  - is built for the people doing that work.
• 0:33
  - More than 2,500
• 0:35
  - executive level attendees,
• 0:37
  - 600 plus speakers,
• 0:39
  - and four days of sessions designed around the
• 0:41
  - real operational questions behind
• 0:44
  - AI, cybersecurity,
• 0:45
  - revenue cycle performance, and digital health.
• 0:48
  - Join us September fourteenth through the seventeenth.
• 0:51
  - For the agenda and event details, visit beckershospitalreview.com
• 0:55
  - and click on the events tab in the
• 0:57
  - upper right.
• 1:00
  - Welcome to the Becker's Healthcare Cardiology and Heart
• 1:03
  - Surgery podcast. I'm Mariah Taylor, assistant editor, and
• 1:07
  - I'm thrilled to be speaking with Sanjeev Shah,
• 1:10
  - director of research at Northwestern Medicine, Bloom Cardiovascular
• 1:13
  - Institute.
• 1:14
  - Thank you so much for joining me, doctor
• 1:16
  - Shah. I'm really excited to learn more about
• 1:18
  - your strategic thinking about human intelligence versus artificial
• 1:22
  - intelligence.
• 1:23
  - But before we dive into that, I'd love
• 1:24
  - for you to introduce yourself and give us
• 1:26
  - a little bit of your background.
• 1:28
  - Sure. Yeah. It's a pleasure to be here
• 1:30
  - today, and I am a heart failure cardiologist.
• 1:34
  - I take care of patients with heart failure,
• 1:36
  - pulmonary hypertension,
• 1:37
  - and I have expertise in echocardiography.
• 1:39
  - And I've been at Northwestern for the past
• 1:41
  - nearly twenty years. When I came here in
• 1:44
  - 2007, I started the world's first heart failure
• 1:46
  - with preserved ejection fraction program, and a lot
• 1:49
  - of my work is in that space and
• 1:51
  - in cardiac amyloidosis.
• 1:53
  - But, over the last six years, I've been
• 1:56
  - the director of research
• 1:57
  - for our Bloom Cardiovascular
• 1:59
  - Institute at Northwestern,
• 2:01
  - And I'm also
• 2:02
  - the director for the Center for Deep Phenotyping
• 2:05
  - and Precision Therapeutics
• 2:06
  - for our Institute for Artificial Intelligence in Medicine.
• 2:11
  - Excellent.
• 2:12
  - Now I'm really excited to dive more into
• 2:14
  - what you're thinking and strategizing
• 2:16
  - around human intelligence versus AI. I think this
• 2:19
  - is something everyone's been thinking about,
• 2:21
  - but I haven't heard much about how this
• 2:24
  - is applying in research. So can you give
• 2:26
  - us a top down view of how research
• 2:28
  - leaders are currently thinking about AI's use?
• 2:32
  - Sure. And I don't think it's necessarily that
• 2:34
  - much different than how clinical
• 2:37
  - leaders are thinking about it or should be
• 2:39
  - thinking about it. You know, I think that
• 2:42
  - in the research realm
• 2:43
  - for a long time, we have been really
• 2:46
  - on the side of development,
• 2:48
  - developing machine learning models, developing AI models,
• 2:52
  - and how we actually do the computing.
• 2:54
  - I was a biomedical engineering major, and, you
• 2:57
  - know, it's very engineering heavy, and it seems
• 3:00
  - like that has been the research realm, but
• 3:02
  - now it's very different.
• 3:04
  - Now we're seeing that it's not just developing
• 3:07
  - the tools, but it's really implementing the tools.
• 3:09
  - How do humans, how do physicians, how do
• 3:12
  - health care providers actually use these tools? How
• 3:14
  - do we get them to use the tools?
• 3:16
  - And, you know, there's more than just the
• 3:19
  - engineering. There's the automation.
• 3:21
  - What task do we need automated? What still
• 3:24
  - needs the human's touch?
• 3:25
  - We're learning a lot in research about how
• 3:27
  - to better classify
• 3:29
  - clinical syndromes that are very diverse and heterogeneous
• 3:32
  - as many of our cardiovascular
• 3:34
  - clinical syndromes are. But I think the most
• 3:36
  - exciting areas, I would say, are
• 3:39
  - treatment and innovation.
• 3:40
  - In that sense, the treatment
• 3:43
  - is really something that, you know, we haven't
• 3:45
  - really been able to personalize.
• 3:47
  - You know, I think because we don't have
• 3:48
  - the bandwidth. We can't call patients every day,
• 3:50
  - for example.
• 3:51
  - We can't sort of
• 3:53
  - understand their entire medical history all at once.
• 3:56
  - AI is giving us ways to do that,
• 3:58
  - you know, and I think that we have
• 4:00
  - AI agents,
• 4:01
  - in you know, going through large language models,
• 4:03
  - going through the medical record, being able to
• 4:05
  - read every single
• 4:07
  - telephone encounter, ER
• 4:09
  - note, physician's note, etcetera, every test, and really
• 4:12
  - understand what is the best way to treat
• 4:15
  - that individual patient.
• 4:17
  - And further than that, I think on the
• 4:19
  - innovation front, what we've seen over the last
• 4:21
  - two years with tools like Claude Code and
• 4:23
  - Codex and others is that now it's all
• 4:26
  - about ideas.
• 4:27
  - You know, before I needed to assemble a
• 4:30
  - team of engineers
• 4:32
  - and data scientists
• 4:34
  - and information architects,
• 4:36
  - etcetera,
• 4:37
  - to really get anything done. And all of
• 4:39
  - those people are still really necessary.
• 4:41
  - But now the health care providers,
• 4:44
  - the clinicians,
• 4:46
  - the administrators,
• 4:47
  - It's like we all have the tools at
• 4:50
  - our fingertips. We all have AI tools that
• 4:52
  - can do a lot of this for us.
• 4:54
  - And so I think it's really democratizing
• 4:57
  - the development
• 4:58
  - of new tools and how we're even thinking
• 5:00
  - about the things that we could never have
• 5:02
  - done before being able to scale
• 5:05
  - across the health system and across health systems
• 5:08
  - in general.
• 5:09
  - Excellent.
• 5:10
  - Now I feel like you just went ahead
• 5:12
  - and answered the next question I have, but
• 5:14
  - I still wanna dive just a little bit
• 5:15
  - deeper to everything you were just saying. For
• 5:18
  - those health system executives that are weighing AI
• 5:21
  - investments in their research programs, what does the
• 5:23
  - business case actually look like, and where are
• 5:25
  - you seeing the fastest returns?
• 5:28
  - Great question. I mean, I think that for
• 5:30
  - most health systems, where I would
• 5:33
  - really invest in the research is in implementation
• 5:37
  - because we have tons of AI tools,
• 5:40
  - tons of things that, you know, will help
• 5:42
  - us diagnose diseases earlier, will help us diagnose
• 5:45
  - when a patient needs a certain treatment.
• 5:48
  - For example, in heart failure, does this patient
• 5:50
  - need the mechanical circulatory support? Do they need
• 5:53
  - a transplant? Do they need to be admitted?
• 5:55
  - You know, can we get this patient,
• 5:58
  - out of the hospital and keep them from
• 5:59
  - being readmitted? So there's these are all diagnostics,
• 6:02
  - so to speak, in a way. And we
• 6:05
  - have tons of those tools out there. I
• 6:06
  - mean, think of all the ECG, electrocardiography,
• 6:09
  - AI tools, the echocardiography.
• 6:11
  - The list goes on and on, but none
• 6:12
  - of them are being used.
• 6:14
  - And so with implementation
• 6:16
  - science, implementation studies, it may not sound very
• 6:19
  - glamorous, but it's really, really important.
• 6:21
  - Because what we found is that,
• 6:23
  - you know, you can have all these tools.
• 6:25
  - You can even have them on autopilot in
• 6:27
  - the sense that we don't even have the
• 6:29
  - the clinicians order them. We have them opt
• 6:31
  - out if they want. And so they're getting
• 6:33
  - the results, but are they doing anything with
• 6:35
  - it?
• 6:36
  - And so where the research is, and I
• 6:38
  - think for most,
• 6:40
  - clinic clinical institutions
• 6:42
  - and places, it's sort of like AB testing
• 6:44
  - that's done in business all the time, but
• 6:47
  - we don't really do it in health care,
• 6:48
  - but we should. So we have two different
• 6:50
  - ways of presenting the information to the physician
• 6:53
  - or the clinician
• 6:54
  - or the patient, etcetera. So, you know, okay.
• 6:57
  - If the if the clinician sees it in
• 6:58
  - EPIC as a pop up versus
• 7:00
  - we have a nurse coordinator or a program
• 7:03
  - coordinator getting all the information and then them
• 7:05
  - contacting the clinician or the patient, what works?
• 7:09
  - And so these different implementation
• 7:11
  - testing, focus groups,
• 7:12
  - you know, these are the kinds of things
• 7:14
  - that we're leaning into now, and I think
• 7:16
  - most places will get the fastest return out
• 7:19
  - of that.
• 7:20
  - I think that the other area of research
• 7:23
  - really has to do with
• 7:24
  - using the data
• 7:26
  - more smartly. So, you know, we in health
• 7:28
  - care
• 7:29
  - generate tons and tons and tons of data.
• 7:32
  - And so I think for some health systems,
• 7:35
  - you know, thinking about what's the best way
• 7:38
  - to leverage that data. And that's really where
• 7:40
  - r and d comes in, and I think
• 7:42
  - what we'll see it and maybe not everywhere,
• 7:44
  - but in a few places is where the
• 7:46
  - health care,
• 7:47
  - AI
• 7:48
  - and the health care delivery type of research
• 7:51
  - will go really into research and development
• 7:54
  - because the clinicians are at the front line.
• 7:56
  - They really know what tools they need, and
• 7:58
  - now they can build them themselves. So I
• 8:00
  - think investing in that r and d for
• 8:02
  - some of the some of the larger health
• 8:04
  - systems
• 8:04
  - to create their own products is still, I
• 8:07
  - think, a really viable solution.
• 8:09
  - Excellent.
• 8:10
  - And you kind of hinted at this already,
• 8:12
  - but we've been seeing that AI is transforming
• 8:15
  - medicine, it seems like, primarily in diagnosis,
• 8:18
  - automation,
• 8:19
  - and patient classifications.
• 8:21
  - So when you're looking at those three categories,
• 8:23
  - which of those do you think delivers the
• 8:25
  - most immediate impact for operational and patient care
• 8:29
  - results?
• 8:32
  - Yeah. I think, you know, what what is
• 8:34
  - really important is thinking about
• 8:36
  - what is the ROI for the patient, number
• 8:40
  - one,
• 8:41
  - for the health care system and provider, number
• 8:44
  - two, and can we marry those two desires?
• 8:48
  - And and there's many areas where we can.
• 8:50
  - You know, there's a lot of testing that
• 8:53
  - gets done routinely
• 8:55
  - for other conditions, noncardiovascular conditions, let's say, but
• 8:59
  - there's still a lot of information in there.
• 9:01
  - So that can help
• 9:04
  - us on a population scale
• 9:06
  - really
• 9:07
  - start screening for disease. So a real simple
• 9:10
  - one is that there are plenty of patients
• 9:12
  - who get chest CTs for all kinds of
• 9:14
  - things. You know, they might be getting a
• 9:16
  - chest CT
• 9:17
  - to rule out lung cancer or to evaluate
• 9:19
  - a pneumonia or lung disease or, you know,
• 9:23
  - pulmonary embolism.
• 9:24
  - Well, in that chest CT, we can get
• 9:26
  - coronary artery calcium. We can get aortic valve
• 9:29
  - calcium. There's a lot that we can do,
• 9:31
  - and, you know, we see products like that
• 9:33
  - on the market. Now why I like that
• 9:36
  - as one of the first automation areas and
• 9:38
  - the first sort of screening
• 9:41
  - is that it's very clear
• 9:43
  - what the ROI is for the patient
• 9:45
  - and the provider.
• 9:47
  - You take the example of severe aortic stenosis.
• 9:50
  - If you can find those, you're gonna be
• 9:52
  - able to intervene in those patients' lives with
• 9:54
  - a TAVR, and the health system's gonna benefit
• 9:57
  - from that, and overall outcomes are gonna improve.
• 10:00
  - You know, there's there's for coronary artery calcium,
• 10:02
  - that is gonna be a lot more widespread.
• 10:05
  - And so there, the treatment, of course, like
• 10:07
  - statins,
• 10:08
  - PCSK nine inhibitors, etcetera, patients are getting better
• 10:12
  - treatment, but also,
• 10:13
  - downstream, the health care system is gonna reap
• 10:16
  - the benefits as well because you're getting more
• 10:18
  - patients who actually need care,
• 10:21
  - who need these procedures.
• 10:23
  - And,
• 10:24
  - I think that is gonna be really helpful.
• 10:26
  - So that's where I see
• 10:28
  - a lot of it. You know, with,
• 10:30
  - automation
• 10:31
  - and classification,
• 10:33
  - to me, automation
• 10:35
  - you know, there's many areas where we could
• 10:37
  - automate and do a lot better, but that's
• 10:38
  - gonna be harder to,
• 10:40
  - monetize or I shouldn't say monetize.
• 10:43
  - It's harder to prove the ROI. I think
• 10:46
  - for sure nobody's trying to monetize on it,
• 10:48
  - but it's more like, how can I prove
• 10:50
  - that this efficiency is actually helping?
• 10:53
  - I think in the long term, you're gonna
• 10:55
  - see
• 10:57
  - the provider satisfaction
• 10:59
  - improve, and that's overall gonna improve efficiency. You
• 11:02
  - know, the sort of quality of life, less
• 11:03
  - burnout, etcetera, but it's just really hard to
• 11:06
  - quantify.
• 11:07
  - And so, you know, I think the early
• 11:10
  - diagnosis and getting people into the right place
• 11:12
  - for the procedures, that's gonna be first. Automation,
• 11:15
  - second. The other thing is automation when it
• 11:18
  - comes to our testing, echocardiography,
• 11:21
  - you know, routine radiology and stuff like that.
• 11:24
  - I I know in cardiology,
• 11:25
  - the way I see it going is that
• 11:27
  - the automation
• 11:28
  - is really just gonna be part of the
• 11:30
  - systems we purchase. So when we buy a
• 11:32
  - new echocardiography
• 11:34
  - reading software, new echocardiography
• 11:35
  - machine, we're gonna expect that everything's automated, all
• 11:38
  - the measurements,
• 11:39
  - even part of the scan. You know? So
• 11:42
  - I think that in terms of where can
• 11:44
  - you transform medicine
• 11:46
  - the most, it's that sort of,
• 11:48
  - finding the patients with disease early, getting them
• 11:51
  - treated early.
• 11:53
  - Excellent.
• 11:54
  - And last I have for you is what
• 11:57
  - does a health system need to have in
• 11:58
  - place before it can realistically leverage AI to
• 12:01
  - accelerate its clinical research mission?
• 12:05
  - Yeah. I think that that the most important
• 12:08
  - part is data infrastructure.
• 12:09
  - If you don't have the data infrastructure,
• 12:12
  - then
• 12:13
  - you can't really do anything else
• 12:15
  - because everything falls from that. And these days,
• 12:18
  - AI can really help with the data infrastructure.
• 12:20
  - You know? So what I tell
• 12:23
  - health care executives and leaders in health care
• 12:26
  - today is that we need to stop thinking
• 12:28
  - of ourselves
• 12:30
  - as health care delivery organizations
• 12:33
  - only. You know? Of course, we're always going
• 12:35
  - to be that, but we are data centers.
• 12:38
  - We are high volume, high quality data centers,
• 12:41
  - and we need to be able to liberate
• 12:43
  - that data. We need to be able to
• 12:45
  - get that data,
• 12:46
  - AI, and machine learning ready. So every time
• 12:49
  - an echocardiography
• 12:50
  - exam gets done, every time you see an
• 12:53
  - angiogram getting done, every time an MRI, EP
• 12:56
  - study, etcetera, etcetera,
• 12:57
  - that needs to go straight into
• 13:00
  - a server. It needs to be processed so
• 13:03
  - that it is machine learning ready. We call
• 13:05
  - that, you know, that first phase embeddings,
• 13:08
  - and we get basically that imaging data into
• 13:10
  - numeric format. All of our notes are getting
• 13:13
  - abstracted
• 13:14
  - by automated algorithms that the LLMs came up
• 13:17
  - with, and all of that information is getting
• 13:19
  - joined. So then you can really easily use
• 13:21
  - it downstream, and all the images are being
• 13:24
  - saved in one repository.
• 13:26
  - That's really going to
• 13:28
  - help
• 13:29
  - health systems and cardiology divisions and cardiovascular programs
• 13:33
  - really take things further. And so when a
• 13:35
  - company or a partner comes to them, number
• 13:36
  - one, they're gonna wanna partner with them because
• 13:38
  - the data is beautiful and they can use
• 13:40
  - it and they can validate their stuff, but
• 13:42
  - also they can implement their solutions much
• 13:45
  - quicker. You know, and and the other thing
• 13:47
  - I think is,
• 13:48
  - you know, you mentioned talent and partnerships.
• 13:52
  - I think partnerships are abound, and they'll come
• 13:54
  - after the data infrastructure if people see that.
• 13:56
  - Talent is, you know, at
• 14:00
  - it I think everyone's looking for the talent,
• 14:02
  - so the competition is very high. But I
• 14:04
  - personally don't know that you need
• 14:06
  - AI talent,
• 14:07
  - because we have so many resources with AI
• 14:10
  - tools now. So what I would recommend instead
• 14:13
  - is,
• 14:14
  - yes, you need a few people that understand
• 14:16
  - the clinical and the AI, deeply understand both
• 14:19
  - sides,
• 14:20
  - and I call those people AI ambassadors.
• 14:22
  - So from a, you know, clinical research, clinical
• 14:25
  - implementation
• 14:26
  - type of strategy,
• 14:28
  - what you can do is you can have
• 14:30
  - these AI ambassadors
• 14:32
  - sit with people that who are not AI
• 14:34
  - savvy and say, hey. What are your pain
• 14:36
  - points? You know, study coordinator, what's your pain
• 14:39
  - point?
• 14:40
  - Regulatory
• 14:41
  - finance for clinical trials, clinical trialist,
• 14:44
  - all across there, and do the same on
• 14:46
  - the clinical side as well. What's the pain
• 14:48
  - point? If you could solve five things today
• 14:51
  - that are just really
• 14:52
  - holding you back from an efficiency standpoint from
• 14:55
  - being the team player and, you know,
• 15:00
  - have the role that you have be successful?
• 15:02
  - What is that? And that is really where
• 15:04
  - the AI ambassador can take those notes, those
• 15:07
  - pain points, and translate them into solutions. We've
• 15:10
  - never been able to do that better than
• 15:12
  - we do today with all of the AI
• 15:14
  - tools.
• 15:15
  - Excellent.
• 15:16
  - Well, this has been such an informative discussion.
• 15:19
  - Thank you for joining me today and sharing
• 15:21
  - all of your thoughts.
• 15:22
  - Thank you. It's a pleasure.