Latest / Becker’s Healthcare Digital Health + Health IT / Dr. Christopher Kelly, Leader in Medical Informatics and Pediatric Ophthalmology

Transcript

Dr. Christopher Kelly, Leader in Medical Informatics and Pediatric Ophthalmology · Becker’s Healthcare Digital Health + Health IT

• 0:00
  This is Laura Dirdle with the Becker's Healthcare
• 0:02
  podcast. I'm thrilled today to be joined by
• 0:05
  doctor Chris Kelly, associate chief medical information officer
• 0:08
  for data and analytics at MultiCare Health System.
• 0:10
  Doctor Kelly, it's a pleasure to have you
• 0:12
  on the podcast today.
• 0:14
  Hey. Thanks, Laura. I appreciate being on.
• 0:17
  Now I'm excited to speak with you a
• 0:18
  little bit more about just all the transformation
• 0:20
  happening in the health care space and particularly
• 0:23
  some of the cool things you're doing with
• 0:24
  artificial intelligence at MultiCare. But before we dive
• 0:27
  into that discussion, can you tell me a
• 0:28
  little bit more about yourself and your background?
• 0:31
  Yeah. Sure. So I am currently the, ACM
• 0:34
  IO for data analytics at MultiCare.
• 0:37
  We're a 13 hospital health care system in,
• 0:40
  Washington state.
• 0:41
  I I got into this through informatics
• 0:44
  about, 10, 12 years ago, when my hospital
• 0:47
  went on EPIC. And I really didn't have
• 0:50
  a lot of background in this, but I
• 0:51
  found I really enjoyed the work, became a
• 0:54
  physician builder. I went on getting
• 0:56
  training in medical informatics. And,
• 0:59
  ultimately, you know, what I I realized what
• 1:02
  so what we really need to do is
• 1:03
  is make use of the data that we
• 1:05
  spend so much time entering into the EMR.
• 1:08
  So I ended up getting a master's degree
• 1:09
  in data science, and that led to this
• 1:11
  job I have here. And I've been doing
• 1:14
  a lot of the work around,
• 1:16
  implementing artificial intelligence and evaluating,
• 1:20
  the performance of artificial intelligence
• 1:22
  at MultiCare.
• 1:23
  That's really been a major focus on what
• 1:26
  I've been doing over the past year or
• 1:27
  so.
• 1:29
  Oh, that's amazing to hear. You know, what
• 1:30
  a great story in terms of how you've,
• 1:33
  just really navigated getting into the informatics space.
• 1:35
  I I really appreciate that. Now if you
• 1:37
  tell me a little bit about what you're
• 1:39
  most excited on, and, you know, what what's
• 1:42
  top of mind for you right now?
• 1:44
  Well, you know, we've been doing a lot
• 1:46
  of stuff. You know, the the
• 1:47
  there's a lot of angles on artificial intelligence.
• 1:50
  I really think that where AI is gonna
• 1:52
  be going in health care, it's gonna be
• 1:53
  transformative. I think the next decade is going
• 1:56
  to be
• 1:57
  as if not more transformative than any decade
• 1:59
  in the history of health care. I I
• 2:01
  really see a lot of opportunity.
• 2:03
  But one of the things that that I
• 2:05
  really been, you know, that our organization has
• 2:07
  been this has really been trying to do
• 2:09
  is find ways to utilize AI to make
• 2:12
  the experience better for providers.
• 2:15
  So that's a lot of what we've been
• 2:17
  doing so far.
• 2:18
  We've implemented an ambient note generation platform.
• 2:21
  We've implemented
• 2:23
  EPYC's
• 2:24
  automated response technology for in basket messages. We're
• 2:28
  working on EPYC's note summarization
• 2:30
  approach.
• 2:31
  There's a number of different algorithms that we're
• 2:33
  introducing to improve patient care as well, but
• 2:35
  those are really the angles that we're looking
• 2:37
  at to try to improve how providers,
• 2:41
  are able to,
• 2:42
  utilize the the EMR more efficiently
• 2:45
  and how we can take better care of
• 2:46
  our patients.
• 2:47
  That makes a lot of sense. And, you
• 2:49
  know, really, it is a great way to,
• 2:51
  it seems like begin getting into AI,
• 2:54
  and
• 2:55
  having it actually produce those results that you're
• 2:57
  looking for both for the care providers as
• 2:59
  well as patients.
• 3:01
  You know, when you look at those projects
• 3:02
  that you're doing today,
• 3:04
  is there anything that you've learned along the
• 3:06
  way as you were,
• 3:07
  implementing some of these things? What have you,
• 3:10
  leaned into more so? Is there anything that,
• 3:13
  you know, you've had to pivot away from?
• 3:16
  Well, you know, that's a really good question
• 3:18
  because I think that,
• 3:20
  you know, 2 years ago, it was almost
• 3:22
  2 years ago now when,
• 3:24
  ChatGPT
• 3:25
  came out, everybody was really caught by surprise,
• 3:28
  I think. I shouldn't say everybody, but a
• 3:29
  lot of folks were caught by surprise at
• 3:31
  how incredibly powerful it seemed. And I think
• 3:34
  a lot of people
• 3:35
  really gravitated towards it and thought this could
• 3:37
  be transformative
• 3:38
  for
• 3:39
  a human intensive,
• 3:42
  industry
• 3:43
  like health care where we really have to
• 3:45
  do a lot of 1 on 1, can
• 3:47
  we
• 3:48
  improve the quality of our interactions, reduce the,
• 3:50
  you know, delays and, you know, just improve
• 3:54
  patient satisfaction overall with this? But I think
• 3:56
  it's proven to be a little more elusive
• 3:58
  than people realized initially.
• 4:00
  I think we're past that, wow, isn't this
• 4:03
  cool, isn't this magic phase of AI? And
• 4:06
  we really have to be
• 4:07
  thinking about how we critically evaluate these investments
• 4:11
  and understand the ones that are really delivering,
• 4:14
  the value
• 4:16
  that we hope they're going to deliver. And,
• 4:18
  you know, at MultiCare, we're really trying to
• 4:19
  take a
• 4:21
  multidimensional
• 4:23
  approach to value, not just
• 4:25
  financial return on investment. Although, obviously,
• 4:29
  you know, these tools, you know, tend to
• 4:30
  be fairly expensive,
• 4:32
  but we also have to think about it
• 4:33
  is, you know,
• 4:35
  does it improve the quality of patient care?
• 4:37
  Can we demonstrate
• 4:39
  that it actually improves the quality of patient
• 4:41
  care? Does it improve patient experience and does
• 4:44
  it improve the physician experience? Does it reduce
• 4:47
  the time that doctors actually spend in EPIC?
• 4:49
  Does it reduce the time that people spend
• 4:52
  outside of scheduled hours continuing to chart? And
• 4:55
  I think we really have to be fairly
• 4:57
  rigorous about our approach to this in order
• 4:59
  to understand
• 5:01
  what's really effective and what's not effective. I
• 5:04
  think that's going to be our our main
• 5:06
  hurdle going forward is trying to understand
• 5:09
  where our time and resources are best spent.
• 5:14
  That's such an excellent point, you know, and
• 5:16
  really,
• 5:17
  makes a lot of sense as you're, you
• 5:19
  know, trying to be innovative, but then also
• 5:21
  understand exactly what success looks like for some
• 5:24
  of these types of projects,
• 5:25
  you know, in the results that you're looking
• 5:27
  for.
• 5:28
  When looking ahead, how are you thinking about
• 5:30
  growth and in particular with AI,
• 5:33
  continuing to evolve and add value to to
• 5:36
  multicare overall?
• 5:37
  You know, it because that's really the
• 5:41
  that's really the key that that I think
• 5:42
  you hit on is how do we add
• 5:43
  value overall?
• 5:45
  And, you know, fortunately, I we have an
• 5:48
  analytics governance team,
• 5:50
  and and one of my responsibilities
• 5:52
  is, you know, with
• 5:54
  few other folks to evaluate
• 5:57
  the incoming opportunities
• 5:59
  for AI.
• 6:00
  And, you know, how do we
• 6:02
  understand
• 6:03
  what they can offer? How do we understand
• 6:06
  how we're going to measure,
• 6:08
  whether they actually do add value? And, again,
• 6:10
  what we need to think about value in
• 6:12
  in a multidimensional
• 6:14
  way. We're we're we're health care system and,
• 6:17
  you know, part of what we do are
• 6:18
  good.
• 6:19
  You know, part of what we do is
• 6:20
  we need to make sure the finances work,
• 6:22
  but our ultimate mission is to take better
• 6:24
  care of our patients. Well, how do we
• 6:27
  measure that? You know, how do we understand
• 6:29
  whether these tools which seem so incredible
• 6:32
  are actually improving
• 6:33
  the patient care that we deliver?
• 6:35
  And and that's not always easy to do,
• 6:38
  you know, given the limitations that we have.
• 6:40
  It's, you know, it's funny. We've been having
• 6:42
  some conversations around this recently, and,
• 6:46
  you know, I think the conclusion that a
• 6:47
  lot of us have reached is that the
• 6:50
  the AI algorithm
• 6:52
  is actually the easy part.
• 6:55
  And I don't mean to say that it's
• 6:56
  easy, and these your algorithms are incredibly sophisticated.
• 7:00
  But taking that AI algorithm and actually incorporating
• 7:04
  it into clinical workflows,
• 7:06
  making sure that you've standardized those workflows across
• 7:09
  an incredibly
• 7:10
  complex health care environment,
• 7:12
  and then being able to then go back
• 7:15
  and and measure the performance of the model
• 7:18
  and the the impact that the model has
• 7:21
  on the delivery of care,
• 7:23
  that's a real challenge. And,
• 7:25
  I think that's really where,
• 7:27
  we're going to be seeing some,
• 7:30
  real need for investment in coming years. There's
• 7:33
  a lot of opportunities. There's literally
• 7:36
  dozens of these that that are coming up
• 7:37
  every month, but,
• 7:39
  they all seem so promising. But we really
• 7:42
  need to have a good understanding about what
• 7:44
  really is going to work,
• 7:47
  showing what isn't. And and I don't think
• 7:49
  we're there yet. I think that's really where
• 7:50
  we need to put our our energies.
• 7:54
  That makes a lot of sense. And, you
• 7:55
  know, it'll just be fascinating to see how,
• 7:58
  you you're able to parse out those things,
• 8:00
  you know, that really do add value in
• 8:02
  those that work and are important to continue
• 8:05
  investing in, for the future. So, you know,
• 8:08
  I I think that just seems really, really
• 8:10
  critical and helpful.
• 8:12
  Are there any headwinds or roadblocks that you
• 8:14
  see potentially coming down the pipe or or
• 8:16
  something that you and your leadership team is
• 8:18
  just are just keeping an eye on as
• 8:20
  a trend that might unfold in the future?
• 8:23
  Well, so, you know, this is the the
• 8:26
  difficulty that we have,
• 8:28
  I think, in health care
• 8:30
  is
• 8:31
  really understanding
• 8:32
  how these models
• 8:35
  work,
• 8:36
  you know, not on an abstract dataset, but
• 8:38
  in
• 8:40
  a clinical
• 8:41
  or operational context.
• 8:43
  You know, there's a lot of different ways,
• 8:45
  for example, to document
• 8:48
  pieces of information on a patient. You can
• 8:50
  put it in a flow sheet. You can
• 8:51
  store it as discrete data. You can store
• 8:54
  it as
• 8:55
  free text. And there may be, you know,
• 8:57
  different ways to do it across a complex
• 9:00
  health care system where there's multiple different hospitals.
• 9:04
  There may be different ways to do it
• 9:05
  even within a hospital. And and these
• 9:08
  problems or the way that this documentation is
• 9:10
  done isn't just arbitrary. There's usually very good
• 9:13
  reasons for why it's done this way. But
• 9:15
  if we don't have that level of standardization
• 9:18
  across a hospital or across the system, it
• 9:21
  gets very difficult
• 9:23
  to implement
• 9:25
  these algorithms in such a way that you
• 9:27
  can actually make a difference with them. You
• 9:29
  know, the thing that I think it's a
• 9:32
  little tricky
• 9:33
  from a data standpoint is that,
• 9:36
  you know, there's this assumption that we have
• 9:39
  this this, you know, clean set of health
• 9:41
  care data now that we're on electronic medical
• 9:44
  records. But the data that get entered into
• 9:45
  the EMR don't get entered
• 9:48
  to later build algorithms off of. The data
• 9:50
  get entered to take care of patients,
• 9:52
  and we're doing it in the way that
• 9:54
  we think we can best take care of
• 9:56
  the patients given the
• 9:58
  unique environment and the unique problems and the
• 10:00
  unique workflows that we have.
• 10:03
  But
• 10:04
  if it's all done in different ways, if
• 10:06
  there's not a lot of standardization, it becomes
• 10:08
  increasingly difficult to make use of the data,
• 10:11
  to normalize the data, to make sure we
• 10:13
  have a good understanding
• 10:14
  of how,
• 10:15
  we're going to be able to make sense
• 10:17
  of this going forward. So I think the
• 10:20
  first big challenge that we have to do
• 10:21
  is,
• 10:22
  operationalize
• 10:24
  these models in such a way that we
• 10:26
  can really,
• 10:28
  deliver the impact that we think they're going
• 10:30
  to have,
• 10:31
  in health care.
• 10:32
  As you know, a fascinating point you brought
• 10:34
  up there and especially just looking at data,
• 10:36
  it seems like a very much colossal challenge
• 10:40
  to get that organized in a way that,
• 10:43
  makes sense for those algorithms and and for
• 10:45
  an AI informed,
• 10:47
  healthcare system. And so I I can imagine
• 10:49
  that, you know, we'll take a huge mind
• 10:51
  share, in the future.
• 10:53
  I'm wondering, you know, are there any other
• 10:55
  opportunities that you see for growth or development,
• 10:57
  as you really move forward?
• 10:59
  So, you know, my training has been in
• 11:02
  data science and,
• 11:04
  you know, addition to to the informatics side
• 11:06
  of things. And I really think that what
• 11:09
  we're going to
• 11:11
  see is an increased demand
• 11:14
  for the use of people, not just to
• 11:18
  report on things, but to really
• 11:20
  make sure we understand
• 11:22
  from a,
• 11:24
  causal or hypothesis testing
• 11:28
  way of, actually trying to understand,
• 11:31
  you know, what the
• 11:32
  reasons for various outcomes are.
• 11:35
  There's some
• 11:37
  health care is is so multifaceted and so
• 11:39
  complicated,
• 11:41
  that it gets really tricky to try to
• 11:44
  actually make sense of
• 11:47
  how a specific workflow
• 11:49
  is,
• 11:50
  working well in one environment and not working
• 11:52
  well in another environment. The number of different
• 11:55
  hypotheses that you can throw out, but you
• 11:57
  need to be able to test them, and
• 11:58
  you need to be able to have people
• 12:01
  on the data side of things who can
• 12:03
  abstract,
• 12:04
  you know, the data from the electronic medical
• 12:06
  record, from the various databases that we have,
• 12:10
  and then be able to
• 12:12
  use data science techniques to,
• 12:14
  do some inference testing to to try to
• 12:16
  understand
• 12:17
  this
• 12:19
  relationship. Is there a correlation here? Can we,
• 12:21
  you know, try to figure out some way
• 12:23
  to infer causation?
• 12:25
  But it's challenging, and I I think there's
• 12:28
  a lot of folks who understand data science.
• 12:30
  But understanding data science is only part of
• 12:32
  the problem. You really need to understand that
• 12:34
  health care environment.
• 12:35
  And I really think that the next, you
• 12:38
  know, few years, we're going to see increasing
• 12:40
  need for people who who have this
• 12:42
  ability to have one foot in the clinical
• 12:45
  world and one foot in the the data
• 12:47
  science world to help us move to that
• 12:50
  next level of of understanding and and appreciation
• 12:53
  for how
• 12:54
  nuanced with so much of the work we
• 12:56
  do is.
• 12:57
  Absolutely. I I think that really, you know,
• 13:00
  is a helpful reminder of just the complexity
• 13:03
  within the health care system and having as
• 13:05
  you do, you know, such a a great
• 13:07
  background,
• 13:08
  within health care delivery as a physician as
• 13:11
  well as, you know, your data science training
• 13:13
  and the technology side of it. You can
• 13:15
  really have a unique
• 13:17
  position there to see where,
• 13:19
  there are huge challenges, but also I can
• 13:21
  imagine great ways to solve problems and try
• 13:23
  to really
• 13:25
  get to
• 13:26
  a point where you can create that value
• 13:29
  in the future as the technology evolves as,
• 13:32
  you know, get to a point where you've
• 13:34
  got more and more data in a great
• 13:36
  space. And then 2, from the health care
• 13:38
  side of it, just really can see what
• 13:40
  is emerging as most
• 13:42
  beneficial,
• 13:43
  for patients as well as caregivers.
• 13:44
  Of course, that's my bias. Yeah. That's my
• 13:46
  bias.
• 13:48
  But, of course, that's that's my perspective. I
• 13:50
  won't say that's necessarily universal, but I I
• 13:53
  you know, that's the way I think we
• 13:55
  can really approach these types of problems.
• 13:58
  Absolutely. Well, I think that's just very helpful.
• 14:01
  Doctor Kelly, is there anything else you wanted
• 14:03
  to leave our our audience with? Anything that,
• 14:05
  you know, would be helpful for our folks,
• 14:08
  you know, other health care leaders and hospital
• 14:10
  health system leaders as they're thinking about and
• 14:11
  trying to figure out out what their AI
• 14:14
  strategy is gonna be, how, you know, they
• 14:16
  need to prepare in order to really, you
• 14:18
  know, create robust programs in the future?
• 14:22
  No. I think the I guess the way
• 14:24
  I end this is to say that
• 14:27
  the real challenge is going to be there's
• 14:30
  gonna be so much opportunity. There's so many
• 14:32
  amazing tools and amazing technologies coming down the
• 14:36
  line very quickly. What we really have to
• 14:39
  be able to do is to understand
• 14:42
  how these different tools can
• 14:45
  be implemented and then how the effectiveness of
• 14:47
  the tools can be measured. That's not an
• 14:50
  easy thing to do. And and right now,
• 14:52
  I think we're we're still in that that
• 14:54
  early phase of adoption. It's going to be
• 14:56
  that next level where we can really say,
• 14:59
  this is the tool that makes a difference
• 15:00
  and these 6 tools don't. This is where
• 15:03
  we need to put our priorities. You know,
• 15:04
  those are the kinds of things that we
• 15:06
  need to be able to work with with
• 15:07
  3rd party vendors on and be able to
• 15:09
  say, listen. This is a nice product, but
• 15:11
  it's not delivering what we need it to.
• 15:13
  And and those are hard things to do.
• 15:15
  But I think if we can get that
• 15:17
  in place, it's going to take that that
• 15:19
  level of integration to really move,
• 15:22
  health care artificial intelligence forward.
• 15:25
  That's amazing. Doctor Kelly, thank you so much
• 15:27
  for joining us on the podcast today. This
• 15:28
  has been such a a wonderful discussion. I
• 15:31
  feel like I I've learned a lot in
• 15:32
  the last few minutes here, and I look
• 15:34
  forward to connecting with you again soon.
• 15:36
  I appreciate it, Laura. Thank you.