Latest / Becker’s Healthcare Podcast / Dr. Hannah Galvin on AI at Scale: Governance, Trust and Equity in Healthcare

Transcript

Dr. Hannah Galvin on AI at Scale: Governance, Trust and Equity in Healthcare · Becker’s Healthcare Podcast

• 0:02
  - The strategic agenda facing hospital and health system
• 0:04
  - leaders right now is not a short list.
• 0:07
  - Growth in a consolidating market, sustainable margins in
• 0:09
  - a reimbursement environment that rarely moves in your
• 0:12
  - favor,
• 0:12
  - a tech investment cycle that demands both urgency
• 0:15
  - and discipline,
• 0:16
  - and a workforce story that is still being
• 0:19
  - written.
• 0:20
  - Eker's fourteenth annual CEO CFO roundtable brings together
• 0:23
  - more than 1,500 executive level attendees and nearly
• 0:27
  - 400 speakers across four days of sessions that
• 0:30
  - are designed to help c suite leaders benchmark,
• 0:32
  - pressure test, and sharpen their priorities.
• 0:35
  - Not in the abstract, but against the real
• 0:37
  - decisions they'll face heading into the next year.
• 0:39
  - It's the kind of event you leave with
• 0:40
  - a clear sense of where you stand and
• 0:42
  - what comes next.
• 0:44
  - So join us November second through the fifth
• 0:46
  - at the Hyatt Regency in Chicago,
• 0:48
  - and come ready to do the hard thinking.
• 0:50
  - Register on our events page at beckershospitalreview.com
• 0:54
  - by clicking on the events tab in the
• 0:56
  - upper right.
• 0:58
  - Hello, and welcome to the Becker's Healthcare podcast.
• 1:00
  - My name is Chanel Banger. And today, I
• 1:02
  - have the pleasure of speaking with doctor Huney
• 1:04
  - Galvin,
• 1:05
  - the chief health information officer at Cambridge Health
• 1:08
  - Alliance,
• 1:09
  - who joins podcast today to share insights in
• 1:11
  - neuro background,
• 1:12
  - health care trends keeping an eye on, and
• 1:14
  - a bit more. Doctor Galvin, thank you so
• 1:16
  - much for joining me today. Thanks so much
• 1:18
  - for having me. Perfect. Well, to get us
• 1:20
  - started, could you please introduce yourself and tell
• 1:22
  - us a bit about your career in health
• 1:24
  - care? Yeah. Absolutely. So as you, mentioned, I'm
• 1:28
  - Hannah Galvin. I'm a pediatrician
• 1:29
  - by training, and I serve as the chief
• 1:31
  - health information officer at Cambridge Health Alliance,
• 1:34
  - which is a public academic
• 1:37
  - mission driven,
• 1:38
  - health system, a safety net health system that
• 1:40
  - serves
• 1:41
  - everyone in need across the Metro North Region
• 1:44
  - in Boston. That's Cambridge, Somerville,
• 1:46
  - Revere, Everett, and the other Metro North communities.
• 1:50
  - We are a three hospital
• 1:53
  - system with a broad ambulatory network,
• 1:56
  - a
• 1:57
  - significant behavioral health footprint.
• 1:59
  - We also operate the Cambridge public health department.
• 2:02
  - We're affiliated with Harvard Medical School and Tufts,
• 2:05
  - University School of Medicine, and I sit on
• 2:08
  - faculty for both of those
• 2:10
  - institutions.
• 2:11
  - What makes
• 2:13
  - CHA,
• 2:14
  - particularly
• 2:15
  - interesting from an informatics
• 2:17
  - perspective is
• 2:18
  - through the combination
• 2:20
  - of complexity and innovation and mission.
• 2:23
  - We care for
• 2:24
  - a very diverse population, including
• 2:27
  - many patients who are medically and socially complex.
• 2:30
  - More than 40% of our patients don't speak
• 2:33
  - English as their primary
• 2:34
  - language.
• 2:36
  - And we operate,
• 2:37
  - with,
• 2:38
  - the resource constraints that are very real in
• 2:41
  - safety net health care. So,
• 2:44
  - that is my
• 2:45
  - day job. I also
• 2:47
  - founded and lead the Shift Collaborative, which is
• 2:50
  - a national multi stakeholder initiative
• 2:53
  - focused on some of the harder policy and
• 2:56
  - implementation
• 2:57
  - questions involving
• 2:58
  - interoperability,
• 2:59
  - patient choice,
• 3:01
  - privacy, and clinical safety,
• 3:03
  - focusing on on how patients can,
• 3:06
  - really choose what data they share,
• 3:09
  - in this increasingly
• 3:10
  - interoperable
• 3:11
  - ecosystem.
• 3:12
  - And we work on advancing
• 3:15
  - standards and implementation
• 3:16
  - guidance
• 3:18
  - related to data,
• 3:20
  - segmentation and computable consent. And that work
• 3:24
  - has given me a broader view of how
• 3:26
  - health systems and vendors and policymakers
• 3:30
  - as well as other members of the health
• 3:31
  - ecosystem like clinicians and patients,
• 3:34
  - are all wrestling with a lot of the
• 3:36
  - same challenges but from different perspectives.
• 3:39
  - So overall, my work
• 3:41
  - sits at this intersection of clinical care and
• 3:43
  - technology,
• 3:44
  - data
• 3:45
  - operations,
• 3:46
  - policy, and and increasingly a lot of that
• 3:49
  - is being influenced by emerging technologies like AI.
• 3:53
  - And when I think about,
• 3:56
  - this work, a lot of it comes down
• 3:59
  - to this question of how do we use
• 4:01
  - technology to make health care work better
• 4:04
  - for all of us and especially patients and
• 4:07
  - the people delivering health care, but without just
• 4:10
  - adding another
• 4:11
  - layer of complexity
• 4:13
  - on top of already really complex work.
• 4:16
  - Got it. Got it. And now that we
• 4:17
  - have a sense of your background,
• 4:19
  - can you walk us through what are the
• 4:21
  - trends that you're currently watching either in,
• 4:24
  - pediatrics,
• 4:25
  - health IT, or just health care overall?
• 4:28
  - Yeah. I think that the the biggest one
• 4:30
  - that we are seeing right now
• 4:33
  - is this transition of AI from
• 4:36
  - experimentation
• 4:37
  - and piloting to
• 4:38
  - being part of actual enterprise infrastructure.
• 4:41
  - Right? Over the last few years, health systems
• 4:43
  - were really running a bunch of very interesting
• 4:46
  - AI pilots,
• 4:47
  - but we're reaching the point where we need
• 4:50
  - to understand how these capabilities should become part
• 4:53
  - of
• 4:54
  - the enterprise
• 4:55
  - workflows
• 4:56
  - and day to day operations.
• 4:58
  - And what data infrastructure is needed,
• 5:02
  - for them to function effectively?
• 5:04
  - How do we govern them, not just in
• 5:06
  - tool selection, but ongoing algorithm of vigilance
• 5:10
  - and monitoring?
• 5:11
  - And how can we demonstrate
• 5:13
  - that,
• 5:14
  - they actually improve,
• 5:16
  - outcomes and that they actually have some return
• 5:19
  - on investment?
• 5:21
  - I'm watching
• 5:22
  - the rapid commoditization
• 5:24
  - of AI.
• 5:26
  - These model capabilities are advancing really quickly, and
• 5:29
  - I think that
• 5:30
  - we need to be careful as we develop
• 5:32
  - our long term strategies around sort of a
• 5:35
  - particular model or a couple of particular models,
• 5:38
  - because,
• 5:40
  - the the the system and, the the industry
• 5:44
  - is just,
• 5:45
  - coming out with,
• 5:47
  - new technology and advancing these model capabilities incredibly
• 5:51
  - rapidly.
• 5:52
  - And increasingly, the more durable assets that we
• 5:55
  - have as as health systems are our data,
• 5:57
  - our workflow integration,
• 6:00
  - our governance, and our ability to scale,
• 6:03
  - and our ability to really evaluate
• 6:07
  - at scale.
• 6:08
  - So so I think that's another piece that
• 6:10
  - we're we're really sort of watching and trying
• 6:12
  - to figure out how how do we really
• 6:14
  - do this,
• 6:15
  - at scale.
• 6:18
  - The other sort of area that I'm watching
• 6:20
  - closely is trust. Right? As AI becomes embedded
• 6:23
  - in these everyday workflows and becomes
• 6:27
  - less visible,
• 6:29
  - there are
• 6:30
  - increasing questions
• 6:31
  - around transparency,
• 6:33
  - privacy,
• 6:34
  - patient choice. Do patients have a choice whether
• 6:37
  - we use these tools or not?
• 6:39
  - How do we really
• 6:41
  - assess for bias in an ongoing manner
• 6:45
  - and provide appropriate human insight, or an oversight?
• 6:48
  - And I think that,
• 6:51
  - becomes harder,
• 6:52
  - not easier as we scale these operations.
• 6:55
  - And,
• 6:57
  - there's this balance between ensuring
• 7:00
  - that we have,
• 7:01
  - the appropriate oversight for trust
• 7:04
  - while still allowing the innovation to move forward,
• 7:07
  - in this rapid manner.
• 7:09
  - Thank you. Yeah. Definitely no shortage of things
• 7:12
  - going on in health care right now. And
• 7:14
  - as we close out 2026,
• 7:16
  - I'm curious to know, what are you most
• 7:18
  - focused on and excited about?
• 7:20
  - Yeah. I mean, I think we are we
• 7:22
  - are right in this. Right? We are
• 7:25
  - excited about AI tools, and we are moving
• 7:28
  - from
• 7:29
  - the just like everyone else, right, piloting AI
• 7:33
  - to,
• 7:34
  - really,
• 7:35
  - incorporating it at scale.
• 7:37
  - We have, we were really early adopters of,
• 7:41
  - mature AI governance, and we have evolved that
• 7:43
  - over time to
• 7:45
  - make sure that we,
• 7:47
  - keep up with the trends in the industry
• 7:49
  - and the continually
• 7:51
  - evolving,
• 7:53
  - need to be able to handle
• 7:55
  - many more tools and opportunities,
• 7:58
  - being,
• 7:59
  - available to us.
• 8:01
  - We have,
• 8:02
  - really tried to create a pathway where people
• 8:04
  - can bring forward good ideas and evaluate them
• 8:07
  - rigorously,
• 8:08
  - and,
• 8:10
  - and try to move useful technology into practice
• 8:13
  - safely while also,
• 8:15
  - saying no to to tools that we do
• 8:18
  - not think will are are ready,
• 8:20
  - for prime time or that would be of
• 8:23
  - high value
• 8:24
  - to us. And we've seen some really tangible
• 8:25
  - operational results. Some of them are not very
• 8:28
  - flashy.
• 8:29
  - For example, we've used AI enabled technology in
• 8:32
  - our call centers
• 8:34
  - to reduce operator call volumes by over a
• 8:37
  - third, by roughly 35%. And in in some
• 8:39
  - specialties,
• 8:40
  - even more than that in in our orthopedic
• 8:42
  - workflow,
• 8:43
  - call abandonment rate that had been running around
• 8:46
  - forty to fifty percent fell to under 10%.
• 8:48
  - So sort of sort of a huge
• 8:50
  - decline there. And and that's not,
• 8:53
  - you know, some of the really kind of
• 8:55
  - glamorous,
• 8:57
  - workflows as some of the the generative AI
• 8:59
  - headlines that we've been seeing, but,
• 9:01
  - that really matters
• 9:03
  - a ton to patients
• 9:04
  - who are waiting on hold trying to to
• 9:07
  - contact someone. And,
• 9:09
  - so,
• 9:10
  - we're really proud of of those results, and
• 9:13
  - we are trying to use AI
• 9:15
  - in ways that really matter
• 9:17
  - to our patients and to our providers and
• 9:20
  - to increasing access.
• 9:22
  - I'm also really excited about the evaluation side.
• 9:26
  - We are conducting
• 9:27
  - a,
• 9:29
  - prospective,
• 9:31
  - study looking at a, specialty
• 9:34
  - medical
• 9:35
  - clinical decision making tool,
• 9:37
  - that many providers are using today,
• 9:41
  - and looking at, the decisions that our residents,
• 9:45
  - are making using that tool and also comparing
• 9:48
  - the,
• 9:49
  - the several several metrics,
• 9:52
  - around the outputs of that tool with foundational
• 9:54
  - models.
• 9:55
  - And I think it's really important to not
• 9:58
  - just turn on these tools and allow their
• 10:00
  - usage
• 10:01
  - because everyone wants to use them and there's
• 10:03
  - a a large clamor to use them, but
• 10:05
  - also
• 10:06
  - to really have rigorous evaluation and understand,
• 10:10
  - what happens when they are in the hands
• 10:12
  - of clinicians.
• 10:13
  - And does it change the decision making? Does
• 10:16
  - it make them better? Does it actually introduce
• 10:18
  - new risks, which is what the literature shows
• 10:20
  - us,
• 10:21
  - to date that that it actually can introduce
• 10:24
  - new risks? And so I think that combination,
• 10:27
  - implementation, thoughtful implementation plus,
• 10:30
  - rigorous evaluation
• 10:32
  - is I think what,
• 10:34
  - we are very focused on and excited about,
• 10:36
  - at this time.
• 10:38
  - Absolutely. That is exciting.
• 10:40
  - And now you've shared what you're looking forward
• 10:42
  - to. Looking back for a moment, what are
• 10:44
  - you most proud of accomplishing over the last
• 10:46
  - six to eighteen months or so?
• 10:49
  - Yeah. I think that that some of those
• 10:51
  - same those same things, I think the call
• 10:53
  - center work has been really, really helpful.
• 10:56
  - We have also,
• 10:58
  - done some work
• 10:59
  - evaluating
• 11:00
  - AI tools and their performer
• 11:02
  - performance
• 11:03
  - with our diverse populations and trying to assess
• 11:07
  - how well they perform across languages. We have
• 11:10
  - some
• 11:11
  - interesting data there. We expected the tools to
• 11:14
  - not perform as well in non English languages,
• 11:18
  - which is what we we saw, but we
• 11:20
  - we we thought that they would not perform
• 11:23
  - particularly
• 11:24
  - well with with,
• 11:26
  - languages of lesser diffusion, like Haitian Creole. And
• 11:28
  - we actually found that the particular tool that
• 11:31
  - we use performed just as well with Haitian
• 11:34
  - Creole speaking patients as it does with,
• 11:38
  - with,
• 11:38
  - patients who speak,
• 11:40
  - languages that are more common, Spanish or Brazilian,
• 11:44
  - Portuguese.
• 11:46
  - And and so that was very interesting and
• 11:48
  - and sort of is new information,
• 11:50
  - compared to the previous literature.
• 11:53
  - We additionally,
• 11:54
  - found that these types of ambient AI tools,
• 11:58
  - are performing just as well with
• 12:01
  - patients
• 12:01
  - who have an interpreter or or in encounters
• 12:05
  - where an interpreter is used and what when
• 12:07
  - providers are speaking that language with the patient
• 12:10
  - and bilingual providers are speaking that language directly
• 12:12
  - with the patient.
• 12:14
  - And that's really interesting because it means that
• 12:16
  - the models underlying some of this ambient AI
• 12:18
  - technology could potentially be
• 12:21
  - used in an interpretation capacity,
• 12:24
  - in the future. And so we're pretty proud
• 12:26
  - of of this study that we're going to
• 12:28
  - be submitting for publication
• 12:29
  - as well, and really that focus,
• 12:32
  - not just how these tools are used with
• 12:35
  - an English speaking population, but that focus on
• 12:37
  - equity, which is really part of, our organization's
• 12:41
  - mission and vision.
• 12:42
  - That's amazing. And now in the short conversation,
• 12:45
  - I could tell that you're very
• 12:47
  - driven and passionate about what you do. And
• 12:50
  - with that, I'd love to hear what advice
• 12:52
  - would you share to emerging leaders looking to
• 12:54
  - have that same impact and drive in their
• 12:56
  - careers as you have?
• 12:57
  - Yeah. When I think about what I would
• 13:00
  - tell emerging leaders, one of the things that
• 13:02
  - I would focus on is
• 13:04
  - not confusing,
• 13:06
  - understanding, and knowing the technology
• 13:09
  - with
• 13:10
  - leading transformation.
• 13:12
  - It's really important to understand and and have
• 13:15
  - that technical fluency,
• 13:16
  - so that you can know what's possible and
• 13:18
  - and to recognize
• 13:20
  - the opportunities and also when something doesn't make
• 13:22
  - sense. But I think
• 13:24
  - the harder skills are understanding the problem,
• 13:28
  - building trust,
• 13:30
  - is is essential
• 13:32
  - and aligning people who have very different incentives.
• 13:35
  - In order to do this, it's really important
• 13:37
  - to stay close to the people doing the
• 13:39
  - work. Some of the best informatics
• 13:41
  - ideas and the prioritization of those ideas
• 13:44
  - come from somebody at the front line saying,
• 13:47
  - hey. I do this, like, ridiculous thing 40
• 13:49
  - times a day.
• 13:50
  - And if you only operate on that strategic
• 13:53
  - level,
• 13:54
  - right, you miss those opportunities. So gemba walks,
• 13:57
  - rounding,
• 13:58
  - talking to the people at the front lines
• 14:01
  - is really essential
• 14:02
  - to
• 14:03
  - leading transformation.
• 14:06
  - The other thing I'd say is it's really
• 14:07
  - important to be comfortable saying yes, but also
• 14:10
  - saying no.
• 14:12
  - Sometimes people think it's it's really important to
• 14:14
  - just champion every new technology. And and,
• 14:17
  - you know, one of the sort of both
• 14:19
  - blessings and curse of being curse of being
• 14:21
  - a safety net organization is we we have
• 14:24
  - a a tight bottom line and and and
• 14:26
  - thin margins.
• 14:27
  - And so we cannot say yes to everything.
• 14:30
  - But but that also means we evaluate them,
• 14:34
  - more closely as well.
• 14:36
  - And it's really,
• 14:38
  - the job of
• 14:39
  - a leader to be able to distinguish between
• 14:42
  - something that's genuinely transformative
• 14:44
  - and something that's simply
• 14:46
  - new and exciting.
• 14:48
  - And we at Cambridge Health Alliance have a
• 14:50
  - lot of practice doing that, and I think
• 14:53
  - we're better off for it because we really
• 14:55
  - try to implement,
• 14:57
  - the truly
• 14:58
  - new exciting innovative tools
• 15:00
  - that that are going to have a long
• 15:02
  - runway
• 15:03
  - and not just those that are shining objects.
• 15:06
  - And finally, I I sort of tell people
• 15:08
  - it's important to build your career around problems,
• 15:11
  - not just titles. Take on the hard problems,
• 15:15
  - especially problems that cross traditional organizational
• 15:18
  - boundaries.
• 15:20
  - If you can become the person
• 15:22
  - who can bring
• 15:24
  - lots of multidisciplinary
• 15:26
  - entities and individuals together,
• 15:28
  - Right? Bring those clinical folks together, operational, technical,
• 15:33
  - financial, and all of those,
• 15:36
  - actors together,
• 15:38
  - and really understand all of those considerations
• 15:42
  - that is going to,
• 15:44
  - really make you indispensable.
• 15:46
  - So those those would be some areas of
• 15:49
  - advice that I would give to those who
• 15:51
  - are listening.
• 15:52
  - I love it. Excellent advice. Well, doctor Galvan,
• 15:54
  - I've enjoyed our time today. But before I
• 15:56
  - let you go, is there anything else that
• 15:58
  - listeners should know?
• 15:59
  - I would say everything is moving very fast
• 16:02
  - right now, and if you can find the
• 16:04
  - area or areas
• 16:06
  - where you can add value
• 16:08
  - and that you are passionate about, you can
• 16:11
  - really
• 16:12
  - add a lot to where this industry is
• 16:14
  - going and it's an exciting time to be
• 16:16
  - in technology.
• 16:17
  - Please check us out, at at Cambridge Health
• 16:21
  - Alliance at, challiance.org,
• 16:23
  - challiance.org,
• 16:24
  - and please check us out, at the Shift
• 16:26
  - Collaborative,
• 16:28
  - at www.shiftinterop.org
• 16:32
  - as well if you want to learn more
• 16:33
  - about our work and join,
• 16:36
  - our collaborative.
• 16:37
  - Absolutely. Well, that's a great note to Anand.
• 16:40
  - Doctor Galvin, I wanna thank you once again
• 16:41
  - for your time today and for sharing your
• 16:43
  - insights on the Becker's HealthCare podcast. Thank you
• 16:45
  - so much. Thanks so much. Have a great
• 16:47
  - day.