Latest / Becker’s Healthcare Digital Health + Health IT / Merage Ghane, Ph.D., Director of Responsible AI in Health at the Coalition for Health AI (CHAI

Transcript

Merage Ghane, Ph.D., Director of Responsible AI in Health at the Coalition for Health AI (CHAI · Becker’s Healthcare Digital Health + Health IT

• 0:00
  Welcome to the Becker's podcast. I'm Naomi Diaz,
• 0:03
  health IT reporter here at Becker's. And today,
• 0:06
  we're joined by doctor Mirage Gaine, director of
• 0:09
  responsible
• 0:10
  AI in health at the Coalition for Health
• 0:13
  AI, also known as CHI.
• 0:15
  In this episode, we'll discuss what hospital leaders
• 0:18
  need to know about AI today, how to
• 0:20
  prepare for the next few years, and where
• 0:22
  AI is really headed in health care. Doctor
• 0:24
  Gain, thank you so much for joining us
• 0:26
  today. I appreciate it. I wanna start off
• 0:29
  with our first question here. Just a simple
• 0:30
  one of just introducing yourself and tell us
• 0:33
  a little bit about your background.
• 0:36
  Great to be here. Thanks so much, Naomi.
• 0:40
  I am.
• 0:41
  I have a PhD in clinical psychology,
• 0:44
  actually,
• 0:45
  and I have a background in computational and
• 0:48
  clinical neuroscience,
• 0:50
  behavioral science, and I used a lot of
• 0:53
  machine learning and AI
• 0:54
  in my research back in the academic days.
• 0:57
  But,
• 0:58
  I'm
• 0:59
  far from those, not super far, but pretty
• 1:02
  far from the academic days.
• 1:05
  Since
• 1:07
  finishing my postdoc, I've been working
• 1:10
  I I started working as a principal behavioral
• 1:12
  designer at ideas forty two, which is a
• 1:15
  mission driven nonprofit,
• 1:17
  and I focus my work in,
• 1:19
  AI and machine learning and health,
• 1:22
  sort of things around, like, identifying bias and
• 1:25
  health AI solutions,
• 1:28
  for organizations and helping organizations do that in
• 1:30
  a behaviorally informed way.
• 1:33
  I actually started working with CHI then.
• 1:36
  So before CHI was ever a nonprofit, it
• 1:39
  was a coalition for the willing,
• 1:41
  coalition of the willing.
• 1:44
  And I led their fairness and bias work
• 1:46
  group,
• 1:48
  in developing sort of responsible AI guidance,
• 1:52
  and then
• 1:53
  I transitioned to this role. So here at
• 1:56
  CHI, I work across many sectors. I bring
• 1:59
  together
• 2:00
  any
• 2:02
  sector you can think of in health care.
• 2:04
  So anybody on the developer side, the clinical
• 2:07
  side,
• 2:09
  sometimes policymakers.
• 2:13
  Who else do we have? Patient advocacy groups,
• 2:15
  community health centers,
• 2:17
  all of them into one space, and we
• 2:20
  talk about how to build shared frameworks,
• 2:23
  tools,
• 2:24
  processes,
• 2:28
  best practice guidance to make AI,
• 2:31
  safer, more transparent,
• 2:32
  and responsible
• 2:33
  in health care.
• 2:36
  Amazing. Great to hear about your background. And,
• 2:38
  again, just responsible,
• 2:40
  safety of AI implementations is just really what
• 2:43
  we need right now. Would love if you
• 2:44
  could just dive in and just tell us
• 2:46
  about your work with consensus driven frameworks for
• 2:49
  AI in health care. What have you learned
• 2:50
  over these past years, and what do you
• 2:52
  feel like is most important today?
• 2:55
  Great question. Consensus work is hard and traditionally,
• 2:59
  very slow, but we don't actually have that
• 3:02
  privilege,
• 3:04
  really in the AI space of being slow
• 3:06
  because things are moving so so fast. And
• 3:09
  so,
• 3:11
  we
• 3:12
  at CHI use sort of modified Delphi methods
• 3:16
  to build these consensus driven frameworks,
• 3:19
  And we convene, like I said, many stakeholders
• 3:23
  across,
• 3:24
  large academic medical centers like the Mayo's, like
• 3:27
  the Dukes, UT Health,
• 3:30
  as well as non academic medical centers, small
• 3:33
  community health centers,
• 3:35
  and, like I said, the developers of these
• 3:37
  solutions as well to kind of think about,
• 3:42
  how can we take what it currently exists
• 3:45
  in many different silos with many different incentive
• 3:48
  structures,
• 3:50
  and bring them together to find
• 3:52
  the shared
• 3:53
  motivations.
• 3:54
  How can we help each other improve the
• 3:57
  health care system,
• 3:59
  without sacrificing innovation, without sacrificing patient safety and
• 4:03
  privacy,
• 4:05
  and keep things moving at a pace that
• 4:08
  that this this whole field really needs.
• 4:11
  I think one of the biggest things that
• 4:14
  I've learned is that everyone is trying to
• 4:16
  do the right thing,
• 4:19
  with AI.
• 4:20
  We
• 4:21
  most people. We have,
• 4:23
  I think different lenses depending on where we're
• 4:26
  coming at this problem from.
• 4:29
  But and there is no map that says
• 4:31
  here is how all these different lenses should
• 4:33
  connect, and that's where we come in,
• 4:36
  to sort of co create
• 4:39
  practical practical,
• 4:41
  sort of flexible guardrails,
• 4:43
  tools, resources
• 4:45
  to help
• 4:46
  everyone in this landscape make better safer decisions.
• 4:50
  And I think it's more important than ever
• 4:52
  now because the consequences
• 4:55
  of AI misuse
• 4:57
  in health care in particular,
• 5:00
  is huge, and
• 5:02
  we have a health care system that is
• 5:04
  already struggling,
• 5:06
  both as a business
• 5:08
  and for patients.
• 5:10
  And so how can we really take this
• 5:12
  opportunity, and I see this as a big
• 5:14
  opportunity
• 5:15
  to do better,
• 5:17
  without
• 5:19
  hurting the most vulnerable people.
• 5:21
  Mhmm.
• 5:22
  Mhmm. I I love what you touched on
• 5:24
  there too is not stifling that innovation, but
• 5:26
  also ensuring that, you know, you have those
• 5:29
  safety guardrails for a technology that's continue
• 5:32
  continuously evolving. I think that's super important there.
• 5:35
  I think for our audience,
• 5:37
  obviously, and our listeners of health care execs,
• 5:39
  I would love to ask you as as
• 5:41
  a follow-up here as well. What do you
• 5:43
  feel like hospital executive teams really need to
• 5:45
  know about health care AI today in this
• 5:48
  such rapidly changing landscape, and what do they
• 5:50
  need to do to prepare their teams for
• 5:52
  the next two to three years?
• 5:54
  It's a great question.
• 5:56
  And they're we're doing a lot of work
• 5:57
  in this space, but I think the first
• 5:59
  thing to realize is AI is already being
• 6:01
  used.
• 6:03
  And and,
• 6:05
  you know,
• 6:06
  sometimes it's chat GPT
• 6:08
  on somebody's phone.
• 6:10
  And and it's also being used
• 6:13
  for real use cases and documentation
• 6:16
  and triage and
• 6:18
  diagnostics.
• 6:19
  Radiology is a big area and and administrative
• 6:22
  tasks. So people think about AI and health
• 6:25
  care like, oh, it must be a clinical
• 6:26
  use,
• 6:27
  but there are so many operational and administrative
• 6:30
  processes
• 6:31
  in health care that are struggling,
• 6:34
  and leading to workforce
• 6:36
  issues,
• 6:37
  and and cost of having to retrain people
• 6:40
  because people are so burnt out.
• 6:42
  So,
• 6:44
  just recognizing
• 6:45
  that it is already being used, and there
• 6:47
  are lots of spaces that you can use
• 6:50
  it within the health care
• 6:52
  landscape.
• 6:54
  I think the second thing is that folks
• 6:56
  need a plan,
• 6:58
  and that means establishing internal policies,
• 7:01
  governance structures,
• 7:03
  processes,
• 7:06
  working together with your teams and and within
• 7:08
  your network. So some organizations
• 7:11
  are much larger and have very robust internal
• 7:14
  teams.
• 7:15
  Some organizations like safety net hospitals, FQHCs,
• 7:19
  community health centers,
• 7:21
  they don't have that
• 7:22
  luxury, so to speak, but they do have
• 7:25
  access to things like
• 7:26
  their
• 7:28
  health center control network, their primary care association
• 7:31
  network. And so how can we get really
• 7:33
  creative about what it means to establish internal
• 7:37
  policies, governance structures,
• 7:39
  and not, you know, burden those that don't
• 7:41
  have resources and and be able to develop
• 7:44
  that? So we're doing that,
• 7:46
  in partnership
• 7:47
  with
• 7:48
  NAC, the National Association for Community Health Centers,
• 7:53
  but also with just our member organizations that
• 7:56
  have come up and said we really care
• 7:57
  about this. Even the larger organizations will come
• 7:59
  up and say, if we're proposing this as
• 8:01
  a process,
• 8:03
  how will the smaller
• 8:07
  entities, the less resource entities, actually make this
• 8:09
  happen? So we're really thinking about this carefully.
• 8:13
  So as part of that plan, folks need
• 8:15
  to understand the risks. They need to determine
• 8:17
  what are some of the responsible use cases.
• 8:20
  Where can they start,
• 8:22
  in this AI journey for their particular organization?
• 8:25
  And, again, context will matters a lot.
• 8:29
  And then thinking about the processes that you're
• 8:31
  developing, the governance structures, the internal policies should
• 8:35
  address things like bias
• 8:37
  or risk of bias, security and privacy issues,
• 8:41
  the usability and usefulness of that solution. You're
• 8:44
  paying so much money not just to procure
• 8:47
  this, but to use it. And if it's
• 8:49
  not helpful, if it's just gonna make your
• 8:51
  staff angry, like,
• 8:53
  shouldn't
• 8:54
  shouldn't be the you know, even if it's
• 8:56
  bright and shiny and new,
• 8:58
  there's a lot of change management that needs
• 9:00
  to go in
• 9:01
  to to bringing new, processes in and making
• 9:05
  sure that they're helpful. And then safety and
• 9:07
  transparency. So those are, I think, areas that
• 9:10
  folks really need to start developing plans around.
• 9:14
  And then finally I think investing in capacity
• 9:17
  and in education
• 9:19
  for staff, so upskilling existing staff,
• 9:24
  training new staff to sort of have that
• 9:27
  come in with the mindset of, like, where
• 9:29
  AI is used, when it's helpful, how it's
• 9:32
  helpful,
• 9:33
  and what is it, for for many folks.
• 9:35
  That's the first question.
• 9:38
  So investing in that capacity now and investing
• 9:40
  in a road map for the future,
• 9:43
  data infrastructures,
• 9:44
  assets, all of that making sure that
• 9:48
  you have a sense of, like, where your
• 9:50
  data's at, how it could be used, how
• 9:52
  it could be leveraged
• 9:53
  to actually improve the business of your organization,
• 9:58
  in this sort of
• 10:01
  in these times where data data is king
• 10:04
  and queen,
• 10:05
  all of the above.
• 10:08
  I I really loved your takeaways, doctor Gaine.
• 10:10
  Really hit the nail on the head when
• 10:11
  it comes to just things I'm even hearing
• 10:13
  from our audience here on just the the
• 10:15
  obstacles they're going through in their own organizations
• 10:18
  there. Really appreciate you diving deep. I wanna
• 10:21
  stay on that track of what you mentioned
• 10:23
  of, you know, those governance, those guardrails, and
• 10:25
  ask you, you know, how should hospitals and
• 10:28
  really health systems just address regular regulatory concerns
• 10:32
  with AI and health care? And I know
• 10:34
  you touched on this a bit in your
• 10:35
  answer as well, but how do you really
• 10:36
  create ethical standards within the organization
• 10:39
  to deploy this safety and ethically as well?
• 10:44
  Yeah. I think
• 10:46
  I think we're in this sort of regulatory
• 10:49
  purgatory, I would call it. And and I
• 10:52
  don't think I think at the end of
• 10:54
  the day, we shouldn't have to wait. Nobody
• 10:57
  should have to wait for full regulation to
• 10:59
  act responsibly.
• 11:01
  If we're gonna only do the right thing
• 11:03
  when somebody tells us to do the right
• 11:04
  thing, I think, you
• 11:06
  know, maybe it loses its value a little.
• 11:09
  So
• 11:10
  for me, ethical standards has to be they
• 11:13
  have to be built into the AI life
• 11:15
  cycle.
• 11:15
  So all the way from procurement
• 11:18
  to development, if that's something you're doing internally
• 11:20
  in your organization or co development with,
• 11:24
  with a vendor or purchasing,
• 11:27
  all the way to monitoring and oversight. So
• 11:30
  when we think about
• 11:33
  ethics in health care, how can we then
• 11:35
  extend that to
• 11:37
  ethics in AI?
• 11:39
  And I think
• 11:41
  there is this
• 11:43
  sense that sometimes people try to reinvent the
• 11:46
  wheel. They're like, oh, AI is so different.
• 11:48
  Everything about it needs
• 11:51
  to be different, but safety still matters. Security
• 11:51
  privacy still matters, and these things existed. You
• 11:54
  know, transparency,
• 11:55
  consent, like, these things have existed in health
• 11:58
  care
• 11:59
  even pre AI. Now the question is, how
• 12:01
  do we identify where AI uniquely,
• 12:05
  needs ethics in these spaces, and how can
• 12:08
  we just add that to the processes that
• 12:10
  already exist without
• 12:12
  fully reinventing the wheel?
• 12:16
  And I think at CHI, we try to
• 12:18
  align our work with some of the regulatory
• 12:21
  momentum that's out there.
• 12:23
  You know, FDA has action plans out there.
• 12:25
  ONC has their HTI one rule,
• 12:28
  but
• 12:29
  I think what we hear from our community,
• 12:31
  and that's we are a community led nonprofit,
• 12:34
  is they want practical
• 12:36
  playbooks. They want practical tools, resources
• 12:40
  to make these things possible. So one of
• 12:43
  the things we're doing,
• 12:45
  in partnership with the joint commission is we're
• 12:47
  developing a set of governance playbooks for organizations
• 12:50
  of all sizes.
• 12:53
  And hopefully down the line, I mean, in
• 12:55
  the near future, accompanying tools, templates, resources that
• 12:58
  folks can leverage.
• 13:00
  And as part of that playbook, I think
• 13:02
  one of the biggest things that we wanna
• 13:04
  prioritize,
• 13:05
  not only getting feedback from our community, which
• 13:07
  we are gonna do. We have a series
• 13:09
  of workshops coming up,
• 13:11
  but also illustrating
• 13:13
  how differently sized organizations are aligning with some
• 13:17
  of these ethical
• 13:19
  guidelines that we're putting out there or some
• 13:21
  of these governance controls that we're putting out
• 13:23
  there in the playbook. So if we say
• 13:26
  everybody should have AI policy, what does that
• 13:29
  look like? How are different organizations
• 13:31
  doing this?
• 13:32
  So that folks have kind of a road
• 13:34
  map or template to look to. Say, oh,
• 13:37
  that worked for this size organization, which is
• 13:39
  similar to me, so maybe I could start
• 13:41
  there.
• 13:43
  And also being very clear about what the
• 13:45
  challenges
• 13:46
  are that people are facing in aligning with
• 13:49
  some of these ethical guidelines,
• 13:52
  and,
• 13:54
  you know, recommended governance controls
• 13:57
  and being able to say, like, okay. Are
• 13:59
  there tools, are there templates, resources that we
• 14:01
  can provide to help reduce some of that
• 14:03
  barrier
• 14:04
  and making sure that the ones who are
• 14:07
  resourced, who have it, and I can develop
• 14:09
  those things, great. For the ones who can't,
• 14:13
  how can we help them get a step
• 14:14
  up
• 14:16
  in that process and not widen that digital
• 14:19
  divide?
• 14:22
  And I think ethical AI
• 14:25
  isn't shouldn't just be within the health care
• 14:28
  system. I think a big part of
• 14:31
  what is now happening if we really want
• 14:34
  to create person centered or human centered
• 14:37
  solutions,
• 14:40
  it is a lot about
• 14:41
  the developers also understanding what ethics means in
• 14:45
  health care and being able to
• 14:47
  align their products with that so that when
• 14:49
  they go to their customers,
• 14:51
  the health care systems, and they say,
• 14:54
  I have this really cool thing that they're
• 14:56
  already speaking the same language. And I think
• 14:58
  that's a lot of what we're trying to
• 15:00
  do is get folks to hear each other's
• 15:03
  different languages and at least understand, like, do
• 15:06
  a little bit of translation there.
• 15:10
  And and it's about the relationships.
• 15:12
  So it's not it's also about how are
• 15:15
  we supporting each other, how are developers supporting
• 15:17
  their customers
• 15:19
  in in aligning, and how
• 15:21
  are the
• 15:22
  the health systems
• 15:23
  supporting the developers and improving the tools to
• 15:26
  better fit their needs.
• 15:28
  And so I think that's another big piece
• 15:30
  of ethical AI is that relationship
• 15:32
  between
• 15:34
  the
• 15:35
  the developer and the implementer.
• 15:39
  Wow. I you just mentioned so many important
• 15:42
  topics here. I think one of just an
• 15:43
  incredible quote you just, you know, said in
• 15:45
  the beginning as well, just nobody should wait
• 15:47
  for regulations to do the right thing. I
• 15:49
  think in essence, that was just amazingly said,
• 15:52
  doctor Gaines. So appreciate that.
• 15:55
  I wanna I I wanna look as as
• 15:57
  well to the future a little bit here
• 15:59
  for for our last question and and ask
• 16:01
  you just where do you really see AI
• 16:04
  headed in health care over the next two
• 16:06
  years, and what do you expect will be
• 16:08
  a little bit different because of AI?
• 16:12
  Yeah.
• 16:14
  So there's a lot
• 16:15
  happening in the space.
• 16:17
  I think we'll see definitely some expanded
• 16:21
  AI use in areas like ambient documentation,
• 16:25
  prior authorization,
• 16:27
  both on the provider and the payer side,
• 16:30
  and patient risk stratification.
• 16:32
  I think that's gonna continue,
• 16:34
  sort of predictive
• 16:35
  prescriptive analytics types pieces aren't it's not gonna
• 16:38
  go away even though GenAI has has come
• 16:40
  up. I think GenAI will help improve some
• 16:43
  of those processes.
• 16:46
  I think
• 16:48
  we have started
• 16:50
  focusing on a lot of back office tools
• 16:52
  as
• 16:54
  a first pass for AI and health care.
• 16:56
  Like, that's been the cautious approach.
• 16:58
  I think this will expand to frontline clinical
• 17:01
  decision support,
• 17:03
  and I think it will also
• 17:06
  move towards
• 17:08
  patients taking agency over their health care differently
• 17:12
  and interacting differently with their providers. Right? So
• 17:15
  with
• 17:17
  tools like ChatGPT,
• 17:19
  people are inevitably going to ask their health
• 17:22
  questions,
• 17:23
  bring
• 17:24
  a different kind of knowledge to their appointments,
• 17:27
  and so I think that will really shift
• 17:28
  the landscape quite a bit.
• 17:32
  I think there will be greater demand for
• 17:34
  transparency
• 17:35
  and evaluation
• 17:37
  over time, and I think we at CHI
• 17:39
  are
• 17:40
  developing sort of recertification
• 17:42
  programs for
• 17:45
  assurance resource providers,
• 17:47
  which might be folks that are validated, like
• 17:49
  third party validation of AI solutions,
• 17:52
  individuals that enable sort of local validation or
• 17:56
  training in a secure,
• 17:57
  privacy enhanced way,
• 18:00
  really making sure that people can actually access
• 18:02
  these tools in a way that is is
• 18:05
  safe.
• 18:07
  And then I think we'll see a lot
• 18:09
  more
• 18:10
  involvement
• 18:11
  from nursing staff, from community health centers, from
• 18:15
  under resourced providers,
• 18:16
  but we need to equip them with the
• 18:18
  tools and the training to really participate in
• 18:21
  this space.
• 18:22
  So one of the things that we are
• 18:24
  doing, I think there is quite a bit
• 18:25
  of curricula out there for,
• 18:28
  for physicians, and we are working on sort
• 18:31
  of expanding that into specialty society type work,
• 18:34
  but we also are partnering with Florida State
• 18:36
  University to bring a responsible AI education to
• 18:39
  nurses.
• 18:41
  And I think
• 18:42
  that will
• 18:45
  be pretty important.
• 18:47
  We have cross functional teams, and if only
• 18:49
  some parts of the team understand what's happening,
• 18:52
  and other parts don't, it can continue to
• 18:55
  increase the tensions, the silos in a space
• 18:57
  where we want actually
• 18:59
  better care continuity and better collaboration.
• 19:03
  So I think that's that's some of it.
• 19:05
  And then agentic
• 19:08
  things, hopefully,
• 19:09
  maybe will come out in health care. Maybe
• 19:12
  not in the next two years,
• 19:14
  but definitely down the line, especially for some
• 19:16
  of that back office stuff. I think for
• 19:18
  the administrative
• 19:20
  and operational work,
• 19:21
  it it makes a little bit more sense.
• 19:24
  But
• 19:25
  there are no solutions that I know of
• 19:27
  currently that are being deployed,
• 19:29
  in the health care system there, but it
• 19:31
  will
• 19:32
  come up,
• 19:33
  down the line for sure.
• 19:37
  Lots to look forward to. Lots evolving so
• 19:39
  quickly. I think, doctor Gain, you just had
• 19:42
  such great light for our audience on what
• 19:44
  this means for them and just the great
• 19:46
  work Chai is doing as well. So it
• 19:48
  doesn't become a place of AI have and
• 19:50
  have nots as I like to describe it
• 19:51
  where, you know, smaller system versus larger systems
• 19:54
  can't get that
• 19:55
  access to really, deploying this technology, like you
• 19:58
  said, ethically and safety. So I wanna just,
• 20:01
  give you a a a huge thank you
• 20:03
  for shedding light on there, sharing your insights
• 20:05
  on this, and as well as to our
• 20:07
  listeners,
• 20:08
  for taking the time out, listening to the
• 20:10
  Becker's podcast today. That's it for our episode
• 20:13
  today. Doctor Gaine, any last words or anything
• 20:15
  you would like to share with our audience
• 20:16
  before, we end today's episode?
• 20:19
  No. I think,
• 20:21
  join our newsletter list.
• 20:23
  Keep up to date even if you're not
• 20:25
  a member. If you want to be a
• 20:26
  member, feel free to reach out to us.
• 20:28
  We would love to chat and see how
• 20:30
  we can be helpful.
• 20:33
  And,
• 20:34
  yeah, I'm just great I'm grateful for the
• 20:36
  community. Without the community, like, the the things
• 20:39
  we put out, the understanding we have of
• 20:41
  the field just, like, really would not be
• 20:43
  possible.
• 20:43
  And so
• 20:45
  I'm grateful for that. We we,
• 20:47
  can only do this together.
• 20:50
  Actualizing that learning health system and learning community.
• 20:52
  Well, thank you again, doctor Gaines, and I
• 20:54
  appreciate your time today. Appreciate you. Thanks so
• 20:57
  much, Naomi.