Latest / Becker’s Healthcare Podcast / Building the Foundation for Clinical AI: Bedside Data, ROI and Governance

Transcript

Building the Foundation for Clinical AI: Bedside Data, ROI and Governance · Becker’s Healthcare Podcast

• 0:00
  - Hello, everyone. This is Erica Spicer Mason with
• 0:02
  - Becker's Healthcare. Thank you so much for tuning
• 0:05
  - in to the Becker's Healthcare podcast series.
• 0:08
  - Today, we're going to talk about how health
• 0:10
  - systems can build the right foundation for clinical
• 0:12
  - AI
• 0:13
  - from harnessing bedside data and identifying high value
• 0:16
  - use cases
• 0:17
  - to measuring ROI and establishing effective governance.
• 0:21
  - And joining me for this conversation,
• 0:22
  - we have Matt Davis, the zone leader of
• 0:25
  - hospital patient monitoring at Philips North America,
• 0:28
  - and Adam Alcado, the senior director of biomedical
• 0:31
  - engineering and life science technology at Stanford Health
• 0:34
  - Care.
• 0:35
  - Matt, Adam, great to have you with us
• 0:37
  - today. Thanks so much for joining. Yeah. Thanks
• 0:39
  - for having us. Good. Thanks for having us.
• 0:41
  - Great to be here.
• 0:43
  - Yeah. So great to have you both. And
• 0:45
  - I wanted to get our conversation
• 0:47
  - started off on a note really around some
• 0:50
  - survey data that we saw.
• 0:53
  - Matt, I know recently, Beckers and Phillips conducted
• 0:56
  - a survey of a 151
• 0:58
  - hospital and health system leaders,
• 1:00
  - and that survey found some pretty strong interest
• 1:03
  - in AI.
• 1:04
  - However, only 7% of those survey respondents described
• 1:08
  - having a highly advanced
• 1:09
  - AI driven monitoring environment.
• 1:12
  - So fragmented systems and nonstandard data were the
• 1:15
  - leading barriers.
• 1:16
  - What does this tell you about the foundation
• 1:18
  - that health systems really need to build here?
• 1:21
  - Yeah. Health systems
• 1:23
  - should really take a step back before adopting
• 1:26
  - AI tools and pilots,
• 1:29
  - and really start with the structure, the workflow
• 1:31
  - gaps, and the business problems that they're trying
• 1:33
  - to fix.
• 1:35
  - The foundation must be consistent
• 1:37
  - from the patient connection
• 1:39
  - all the way through the monitor, the software,
• 1:41
  - the interfaces, the workflow, the data,
• 1:44
  - all the way to the EMR.
• 1:45
  - That's something that at Philips, we're we're doing,
• 1:49
  - let's say based off of industry feedback and
• 1:52
  - and some of the things that we've seen
• 1:53
  - from other organizations is this enterprise posture and
• 1:57
  - being able to scale,
• 1:59
  - in parallel with the EMR so that you
• 2:01
  - can have that consistent
• 2:03
  - data from the patient
• 2:05
  - all the way through to the EMR.
• 2:08
  - And so what this really tells us is
• 2:10
  - that, you know, interest in AI is clearly
• 2:14
  - there, but readiness
• 2:16
  - is where the gap exists.
• 2:18
  - With only 7% of leaders describing
• 2:21
  - an advanced AI driven monitoring environment,
• 2:24
  - most organizations need to focus on that foundation
• 2:27
  - building phase and create that clinical consistency
• 2:31
  - from the patient
• 2:32
  - all the way up into the EMR.
• 2:36
  - So getting the foundation
• 2:38
  - right first sounds like a hard thing,
• 2:41
  - but an absolutely essential piece of this process,
• 2:43
  - and I appreciate how you outlined that for
• 2:45
  - us, Matt.
• 2:47
  - And, Adam, I wanna learn a little bit
• 2:48
  - more about this and what it's looking like
• 2:50
  - for you on the ground at a system
• 2:52
  - like Stanford.
• 2:53
  - What can AI learn from bedside data that
• 2:56
  - it can't get from the EHR alone, and
• 2:58
  - why does it matter to capture signals across
• 3:00
  - the room, even those that never make it
• 3:02
  - into the EHR?
• 3:04
  - Yeah. Absolutely. I mean, it's a it's a
• 3:05
  - very interesting time for us in health care
• 3:08
  - today. There's a there's a lot of convergence
• 3:10
  - from
• 3:11
  - from an era that took an analog environment
• 3:13
  - and and digitized it, through point solutions and
• 3:16
  - spot solutions.
• 3:18
  - In my mind, largely, the EHR is essential
• 3:21
  - as a central record. I mean, the presence
• 3:24
  - of the EHR and the ability for organizations
• 3:27
  - to create integrations and interoperabilities
• 3:30
  - from systems within the patient ecosystem to be
• 3:33
  - able to be,
• 3:35
  - the enriching factor to to a patient's record
• 3:38
  - is something that has been happening for quite
• 3:41
  - a... Some... Quite some time. And,
• 3:44
  - the the difference in the parallel that we're
• 3:45
  - seeing today is really continuous monitoring,
• 3:49
  - and physiological data offers a more immediate and
• 3:52
  - prospective
• 3:53
  - view of of the patient where the EHR
• 3:56
  - really is is capturing events
• 3:59
  - and and being that catch net for
• 4:02
  - the,
• 4:03
  - the the events that are occurring around inside
• 4:06
  - the patient ecosystem. And so
• 4:08
  - multimodal data is really where our focus is
• 4:10
  - right now. And so what that means is
• 4:12
  - really
• 4:13
  - the catch net of all signals
• 4:16
  - within the patient patient ecosystem. And that's where,
• 4:19
  - the physiological
• 4:20
  - bedside data really comes to play,
• 4:23
  - with the goal that a more complete picture
• 4:26
  - of the patient ecosystem
• 4:28
  - can currently,
• 4:30
  - enrich and enhance
• 4:32
  - both the data that comes out of the
• 4:33
  - EHR and is captured there.
• 4:36
  - From our perspective at Stanford, the work to
• 4:38
  - capture additional bedside data is really at the
• 4:40
  - central point of our approach to,
• 4:44
  - solidifying
• 4:45
  - our perspective on the patient ecosystem,
• 4:47
  - both prospective data as well as retrospective data
• 4:51
  - through the,
• 4:53
  - approach of multimodal data capture.
• 4:55
  - So, Matt, I'll look to you for insights
• 4:58
  - on this first.
• 4:59
  - But when health systems evaluate a clinical AI
• 5:02
  - use case, how should they think about economic
• 5:04
  - value and ROI?
• 5:06
  - And what should they measure to know that
• 5:08
  - the investment is truly paying off?
• 5:11
  - I don't know. That's a great question. I
• 5:13
  - think it's something that every health system should
• 5:15
  - start with. And
• 5:17
  - I look at the Facebook or or Meta
• 5:20
  - tokenization
• 5:21
  - cost and,
• 5:22
  - and what they saw as they scale. They
• 5:24
  - had a leaderboard,
• 5:26
  - they were, you know, not looking at utilization,
• 5:29
  - And it was all unmetered, and their cost
• 5:32
  - went through the roof. And so the productivity
• 5:34
  - gains didn't balance the utilization
• 5:37
  - of AI.
• 5:38
  - So for health systems, now we we pull
• 5:41
  - that, I think, parallel back in.
• 5:43
  - They really need to focus on
• 5:45
  - defining the clinical,
• 5:47
  - operational, and financial measures upfront,
• 5:50
  - and what they're trying to achieve.
• 5:52
  - Are they going after a staffing problem?
• 5:55
  - Are they expecting to see some type of
• 5:57
  - efficiency through coding?
• 5:59
  - Or whatever that is, we need to ask
• 6:02
  - whether that application is changing the metrics it
• 6:05
  - was meant to change, and whether that value
• 6:07
  - is worth the ongoing cost.
• 6:12
  - Yeah. Really useful, Matt. And, Adam, I'm curious
• 6:15
  - if that tracks with with what you're looking
• 6:18
  - at at Stanford when you, look to have
• 6:20
  - some of these innovations at the bedside. Is
• 6:22
  - that along the lines of what your team
• 6:24
  - was looking at?
• 6:25
  - Yeah. It's it's an interesting challenge we face
• 6:27
  - today. This frontier of of AI and all
• 6:30
  - the technology that's kinda being modeled in the
• 6:33
  - market is is is new to everyone. And
• 6:35
  - so
• 6:36
  - the the business decision component of it is
• 6:38
  - very paramount. I think to add to what
• 6:40
  - Matt was saying is that
• 6:42
  - I think organizations
• 6:43
  - really need to take a pause from from
• 6:46
  - what we've all been conditioned,
• 6:48
  - with over the last twenty years about buying
• 6:51
  - the next solution or assessing it or doing
• 6:54
  - an RFP
• 6:55
  - and really take the time to look at,
• 6:57
  - are we getting the most value out of
• 7:00
  - our
• 7:00
  - existing investments in technology? Because that technology investment
• 7:05
  - really portrays the tapestry for
• 7:07
  - the the goal of the connected signal and
• 7:10
  - the patient ecosystem to drive
• 7:13
  - both
• 7:14
  - retrospective,
• 7:14
  - you know, actionable event captures through the EHR
• 7:17
  - integration as well as prospective data view through
• 7:21
  - continuous bedside monitoring
• 7:23
  - and using that data. And so, firstly, you
• 7:25
  - know, organizations really need to look at their
• 7:27
  - current investment and the architecture that has put
• 7:29
  - in place into their infrastructure.
• 7:32
  - How is it being used? Is it being
• 7:33
  - maximized through
• 7:35
  - interoperability, through integration?
• 7:37
  - And then ultimately start with an outcome. Right?
• 7:40
  - I think
• 7:41
  - while
• 7:42
  - AI and medicine poses all of this immense,
• 7:45
  - you know,
• 7:47
  - opportunity to solve all of these,
• 7:50
  - difficult complexities within our, health.
• 7:54
  - Really, at the end of the day, we
• 7:55
  - need to look at low hanging fruit opportunities,
• 7:58
  - those that are tied to direct
• 8:00
  - health organization outcomes, whether that's, you know, improving
• 8:03
  - efficiencies,
• 8:04
  - operational or related goals. And and the reason
• 8:06
  - that's important because
• 8:08
  - through through
• 8:09
  - it... Approaching
• 8:10
  - AI in that lens and grabbing operational wins
• 8:14
  - through the deployment and leveraging of the environment
• 8:17
  - to drive more operational efficiencies, you build the
• 8:20
  - case to go into more complex,
• 8:23
  - problem solving through AI and the datasets that
• 8:26
  - they present. But from a business perspective,
• 8:29
  - the ROI has to start with something near
• 8:31
  - and dear to the organization's goals and outcomes,
• 8:33
  - And that's where we think concrete value for
• 8:35
  - AI,
• 8:36
  - really starts to lay its foundation.
• 8:39
  - Mhmm. Yeah. Thanks, Adam. I know such a...
• 8:42
  - Such an important piece to all of this
• 8:44
  - too, of course, is ensuring that that investment
• 8:47
  - that is delivering value is also
• 8:50
  - operating in a way that's safe
• 8:52
  - and governed, especially when we're talking about clinical
• 8:54
  - AI.
• 8:55
  - So I I wanna shift a little bit
• 8:57
  - here to talk to to talk about governance.
• 9:00
  - Matt, what should a health system put in
• 9:02
  - place to govern clinical AI before it goes
• 9:04
  - live?
• 9:07
  - Yeah. I, you know, my perspective comes from
• 9:09
  - the vendor,
• 9:10
  - you know, side of things. But, you know,
• 9:13
  - when we look at it, we wanna we
• 9:14
  - wanna come in and partner with health organizations.
• 9:17
  - This is complex and we want
• 9:20
  - to have alignment
• 9:21
  - with their strategies and priorities.
• 9:24
  - And then look
• 9:26
  - at everything that that encompasses
• 9:28
  - a governance.
• 9:30
  - The data, the privacy, the protections, the regulatory
• 9:33
  - environment,
• 9:34
  - the security and liability and, you know, what
• 9:38
  - other partners do we need to be working
• 9:40
  - with outside of, you know, the vendor I'm
• 9:43
  - working for?
• 9:44
  - You know, do we need a BAA? What
• 9:46
  - other kind of contractual obligations need to be
• 9:49
  - present and true?
• 9:51
  - And so all of that should really be
• 9:53
  - thought through
• 9:54
  - prior to
• 9:56
  - even moving down that path in my opinion.
• 9:59
  - So we need to build this into the
• 10:01
  - AI strategy upfront,
• 10:03
  - you know, before the technology is selected.
• 10:06
  - Yeah.
• 10:08
  - That sounds like a a proactive approach, Matt.
• 10:11
  - And, Adam, I'm curious at Stanford,
• 10:13
  - once AI has been in use, how has
• 10:16
  - your team governed
• 10:18
  - how has your team governed it continuously over
• 10:21
  - time, and do you have any best practices
• 10:22
  - there?
• 10:23
  - Yeah. Absolutely. I I think governance really starts
• 10:26
  - prior to the deployment of AI, and and
• 10:29
  - its intent should be not one that's static
• 10:32
  - or single in dimension, but one that's iterative
• 10:35
  - for the purpose of monitoring,
• 10:38
  - AI tools. And so for us, you know,
• 10:39
  - at Stanford, governance is something that,
• 10:42
  - has multiple dimensions. It's... It elicits
• 10:45
  - interorganizational
• 10:47
  - partnerships from privacy, protection,
• 10:50
  - quality,
• 10:51
  - all the way through to medic... Our medical
• 10:54
  - chairs and diad representation
• 10:56
  - in different aspects of our governance committees as
• 10:58
  - well as,
• 11:00
  - the foundation of our
• 11:02
  - computer science and AI principles
• 11:04
  - around how we assess models,
• 11:07
  - both
• 11:08
  - prior to even considering building them. And then
• 11:11
  - once... And if they're built,
• 11:13
  - you know, how do we monitor them? And
• 11:15
  - so the governance structure is really important for
• 11:17
  - organizations
• 11:18
  - to think about prior to even considering a
• 11:21
  - solution. So, like, the conversation should really start
• 11:25
  - with how do we set up the framework
• 11:27
  - to be able to,
• 11:29
  - facilitate
• 11:30
  - the thought around bringing in or,
• 11:34
  - you know, approaching problems within our environment through
• 11:36
  - AI. And that's
• 11:38
  - largely why our governance is is one that's
• 11:40
  - not static but quite dynamic.
• 11:43
  - It's it's led by our,
• 11:45
  - technology leaders as well as our,
• 11:48
  - business leaders and has an immense amount of
• 11:51
  - involvement,
• 11:52
  - from from everyone in the organization. And I
• 11:54
  - think that's the human capital investment that organizations
• 11:57
  - need to realize is just as important as
• 12:00
  - it is partnering with someone in industry.
• 12:03
  - And then, ultimately, industry partnership has an immense
• 12:05
  - amount of value in this governance structure too
• 12:08
  - because we're we're sharing and essentially building
• 12:11
  - out a new catalog of technology,
• 12:15
  - value and and functions,
• 12:17
  - largely based upon, you know, a a percentage...
• 12:20
  - A large percentage of what we already have
• 12:22
  - from said industry partners. And so it's it's
• 12:25
  - very important to address governance
• 12:27
  - first
• 12:28
  - and then selecting solutions
• 12:30
  - afterwards.
• 12:31
  - Mhmm.
• 12:32
  - Well, Adam, thank you for those insights and
• 12:35
  - really,
• 12:36
  - you know, threading the needle from what Matt
• 12:38
  - said about this this proactive
• 12:40
  - stance for governance.
• 12:42
  - And I know in just a few minutes
• 12:43
  - together, we've covered a lot of ground. We've
• 12:45
  - talked about foundational
• 12:47
  - data and infrastructural needs for for clinical AI.
• 12:50
  - We've talked about what's possible from bedside data
• 12:53
  - all the way to governance.
• 12:55
  - But I wanna close with a few final
• 12:56
  - takeaways for our listeners.
• 12:58
  - When health systems deciding where to start with
• 13:00
  - clinical AI, what makes a use case worth
• 13:03
  - pursuing?
• 13:04
  - Matt, do you wanna give that one a
• 13:05
  - shot first?
• 13:07
  - Well,
• 13:08
  - for me, I think we... We're starting backwards.
• 13:12
  - What not to do,
• 13:14
  - don't just add another AI pilot into your
• 13:16
  - organization.
• 13:18
  - Really
• 13:19
  - align and set those
• 13:21
  - business outcomes and priorities upfront.
• 13:24
  - Make sure that you have the governance, the
• 13:26
  - infrastructure
• 13:27
  - architecture to support it, because,
• 13:30
  - as these things are realized,
• 13:32
  - you want to be able to scale appropriately.
• 13:35
  - I think asking simple questions like, is it
• 13:37
  - useful?
• 13:38
  - Or am I achieving what I intended to,
• 13:41
  - to see out of deploying this
• 13:44
  - AI model, AI solution?
• 13:47
  - And define what success looks like before implementation.
• 13:52
  - Organizations can job... Can judge the adoption
• 13:55
  - and results afterwards,
• 13:56
  - but you need to really go in thinking
• 13:59
  - what is it supposed to to do and,
• 14:02
  - what is the ROI ahead of time?
• 14:05
  - Yeah. Thanks, Matt. And, Adam, anything that you'd
• 14:09
  - add here too?
• 14:11
  - Yeah. I think I think it's it's really
• 14:12
  - important to start with the problem rather than
• 14:14
  - trying to find the the best solution first.
• 14:17
  - Right? I think,
• 14:18
  - from our perspective at Stanford, our our goal
• 14:22
  - is is heavily focused on, you know, restoring
• 14:25
  - and enriching the time
• 14:27
  - between the patient and the provider in a
• 14:29
  - highly digitized
• 14:31
  - ecosystem.
• 14:32
  - And so that's a very unique
• 14:34
  - way to,
• 14:36
  - think through the challenges of health care or
• 14:38
  - technology enabled health care today.
• 14:41
  - And so starting with the problem
• 14:43
  - is immensely important because it invokes the type
• 14:45
  - of thought that a governance model such as
• 14:48
  - we talked about would really,
• 14:50
  - you know, dig its roots into to base
• 14:53
  - its, approach for how they
• 14:56
  - bring in AI into their environments. And so
• 14:58
  - for us at Stanford,
• 15:00
  - we really look at understanding what the problems
• 15:03
  - are today
• 15:05
  - that are potentially preventing or presenting
• 15:08
  - opportunity for us to restore time back between
• 15:10
  - the patient and the provider
• 15:12
  - while leveraging the most value out of our
• 15:15
  - technology investments
• 15:16
  - that we've put into our environments and we
• 15:19
  - plan to put into our environments. And so
• 15:20
  - it's largely an approach change rather than
• 15:24
  - a overall implementation or selection of technologies. I
• 15:27
  - think it's, it's largely around the evolution of
• 15:31
  - how we view the relationship between medicine and
• 15:33
  - technology. And so, you know, I would I
• 15:35
  - would really challenge organizations to think through the
• 15:38
  - problem sets that they face today,
• 15:40
  - identify opportunities
• 15:42
  - where,
• 15:43
  - existing technology
• 15:45
  - is providing data that potentially could be useful,
• 15:48
  - to put in investment in and and do
• 15:50
  - something with.
• 15:52
  - And and then, ultimately,
• 15:53
  - you know, always circle back through,
• 15:56
  - what the clinician and the patient and the
• 15:58
  - patient's family experience is both today and moving
• 16:01
  - forward because technology is only going to advance,
• 16:04
  - but we really want to make sure we're
• 16:06
  - upholding that
• 16:07
  - time and integrity between the patient and the
• 16:10
  - provider. Well, Adam, it's been great getting your
• 16:12
  - perspective on what this has looked like at
• 16:14
  - Stanford,
• 16:15
  - that approach change, as you mentioned.
• 16:17
  - And, Matt, hearing, you know, what you're seeing
• 16:20
  - at a high level across the industry at
• 16:22
  - Philips,
• 16:23
  - also immensely valuable. So I just wanna thank
• 16:26
  - you both again for the time for this
• 16:27
  - conversation today.
• 16:28
  - Thank you, Eric. I appreciate it.
• 16:30
  - And we also wanna thank our podcast sponsor,
• 16:33
  - Phillips Hospital patient monitoring.
• 16:35
  - Listeners, be sure to tune in to more
• 16:37
  - podcasts from Becker's Healthcare by visiting our podcast
• 16:40
  - page at beckershospitalreview.com.