Latest / Becker’s Healthcare Podcast / Using AI to Turn Healthcare Data Into Actionable Insights with Dr. Cheng-Kai Kao

Transcript

Using AI to Turn Healthcare Data Into Actionable Insights with Dr. Cheng-Kai Kao · Becker’s Healthcare Podcast

• 0:00
  - What we wanna do with AI is help
• 0:01
  - clinicians go from being reactive
• 0:04
  - to proactive.
• 0:05
  - At Philips, we can recognize patterns in monitoring
• 0:08
  - data across populations
• 0:10
  - and apply those insights to specific patients.
• 0:14
  - Safe and responsible AI can support sharper insights,
• 0:18
  - pave the way for faster diagnosis,
• 0:21
  - and help streamline care.
• 0:23
  - That can benefit clinicians, it can benefit health
• 0:26
  - systems. And most importantly, it can benefit patients
• 0:29
  - and their families.
• 0:33
  - Phillips.
• 0:34
  - Hello, everybody. This is Gracelyn Keller with Becker's
• 0:37
  - Healthcare. Thank you so much for tuning in
• 0:38
  - to the Becker's Healthcare podcast series.
• 0:41
  - Joining me for today's discussion is doctor Kai
• 0:44
  - Ko, who is
• 0:45
  - the chief medical information officer at UChicago Medicine.
• 0:48
  - So, Kai, thanks for being here. We'd love
• 0:50
  - to have you start off by introducing yourself
• 0:52
  - quickly before we get started. Thanks so much,
• 0:54
  - Grace. It's really my pleasure to be here,
• 0:56
  - with you today. So, yes, my name is
• 0:58
  - Chiang Kai Kao. Usually, I just go by
• 1:00
  - Kai. I'm the chief medical information officer at
• 1:03
  - UChicago Medicine.
• 1:04
  - Awesome.
• 1:05
  - So jumping right in, health systems are receiving
• 1:08
  - a growing stream of patient data from bed
• 1:10
  - side monitoring,
• 1:11
  - ambulatory devices, and care delivered at home. Where
• 1:14
  - is that data creating actionable insights for clinicians,
• 1:17
  - and where does it still create more work
• 1:19
  - than insight?
• 1:20
  - Yeah. Thank you for the question.
• 1:22
  - There's no mistake that we are literally,
• 1:26
  - engaging with numerous information every single day.
• 1:29
  - The amount of journal, for example, we are
• 1:31
  - reading is essentially,
• 1:33
  - growing outpace the the speed that we can
• 1:36
  - read all the journals. And not to mention
• 1:38
  - all the patient generated care data from, you
• 1:40
  - know, home, from the outpatient, from inpatient, and
• 1:43
  - from health system is outside of ours. It's
• 1:45
  - really exceeding what everybody can go through, quickly.
• 1:49
  - So to your question, I think the real
• 1:51
  - actionable
• 1:52
  - insights, you know, from these data is depending
• 1:55
  - on whether, they can be,
• 1:57
  - summarized and and sort of pieced together in
• 2:00
  - a thoughtful way in terms of where the
• 2:02
  - signal,
• 2:03
  - you know, against all the
• 2:05
  - noise that you could have.
• 2:07
  - The non actionable insights will be, for example,
• 2:10
  - the raw data. So make example, if a
• 2:13
  - diabetic patient was, you know, using home glucose,
• 2:17
  - meter to measure the blood glucose,
• 2:20
  - You know, for three months, they can be,
• 2:21
  - you know,
• 2:22
  - four times 30 times three. They can be,
• 2:25
  - almost 360
• 2:27
  - ratings, and that's way beyond anyone who can
• 2:29
  - read through, you know, busy clinic section. So
• 2:31
  - being able to have a system where intelligently,
• 2:34
  - you know, tease out where are the anomaly
• 2:36
  - data, what's the average, what's the outliers,
• 2:38
  - really help, re understanding the data itself. And
• 2:41
  - that's just a small, you know, example.
• 2:44
  - When we resemble and admit the patient in
• 2:46
  - the hospital, there's just a ton of information
• 2:48
  - from outside of the health care system we
• 2:49
  - need to digest through as well. So I
• 2:51
  - think, you know, having
• 2:53
  - AI or having some sort of system to
• 2:56
  - mark, you know, abnormalities
• 2:58
  - will be really key to making sure that
• 2:59
  - these become actionable insights.
• 3:02
  - Absolutely.
• 3:03
  - And when clinicians are trying to reach a
• 3:04
  - diagnosis, they often must piece together information from
• 3:08
  - radiology, cardiology, pathology, urology, and other systems.
• 3:12
  - Where does that fragmentation create the most friction
• 3:14
  - in your organization, and how do you bring
• 3:16
  - together data sources and address this fragmentation or
• 3:19
  - close the gaps?
• 3:21
  - Yeah. No. Thanks for the question again. So
• 3:23
  - I'm a houseless. So this could look like,
• 3:25
  - for example, a patient admitted at 2AM,
• 3:27
  - you know, with a complex medical diagnosis with
• 3:30
  - lots of report from the, specialties that you
• 3:32
  - just mentioned and trying to piece out what's
• 3:34
  - the right diagnosis, what's the next step of
• 3:36
  - the treatment plan. So this, again, goes back
• 3:39
  - to, you know, having actionable insights, not just
• 3:42
  - a raw data is very important. So,
• 3:44
  - for example, in our institution, we deploy a
• 3:46
  - tool that's used AI to summarize the previous
• 3:49
  - charts,
• 3:50
  - including, you know, all of their, obviously, treatment
• 3:52
  - records in our institution, but also some of
• 3:54
  - the care everywhere and they'll see from other
• 3:56
  - institution too. So clinician at a quick glance
• 3:59
  - can sort of see through what's been done
• 4:01
  - in the past, what's the current diagnosis, and
• 4:03
  - why the patient's here, and also what are
• 4:05
  - some of the most recent lab results or
• 4:07
  - study results so they can make a decision.
• 4:09
  - And all of these insights are by by
• 4:10
  - specialties,
• 4:11
  - so they can see, alright. This is what
• 4:13
  - recommendation from cardiology that says yesterday.
• 4:16
  - So hopefully, they'll help them make, quicker decisions,
• 4:19
  - and reduce the fragmentation that you talk about.
• 4:22
  - And patients increasingly move among
• 4:26
  - outpatient settings, and home. Where does clinical context
• 4:28
  - most often get lost during these transitions, and
• 4:31
  - what would it take to give care teams
• 4:33
  - a more continuous view of the patient?
• 4:35
  - Yeah. So we see a lot of these,
• 4:38
  - when patient transition from different care settings, you
• 4:41
  - know, whether that's a inpatient setting,
• 4:43
  - being discharged to the outpatient, you know, where
• 4:45
  - the primary care and other specialists would take
• 4:47
  - over or whole health
• 4:49
  - agencies, rehab,
• 4:51
  - whenever there's a care settings, which there's almost,
• 4:55
  - always some gap, in terms of either the
• 4:57
  - information flow. So the next caregiver knows what
• 5:01
  - happened in the previous encounter
• 5:03
  - and also any change that happened, you know,
• 5:05
  - in the process.
• 5:06
  - There's also gap in scheduling.
• 5:09
  - You know, outpatient provider prescribe, you know, or
• 5:12
  - or place a referral order to make sure
• 5:14
  - cardiology sees the patient for chest pain. Is
• 5:17
  - the appointment actually scheduled, to make sure the
• 5:19
  - patient is properly followed? There's also a lot
• 5:22
  - of care coordination,
• 5:23
  - issues like you mentioned.
• 5:25
  - When patient went through care setting to care
• 5:27
  - settings,
• 5:29
  - there's no, like,
• 5:30
  - overseeing type of coordinator sort of sees through
• 5:33
  - all these encounter, you know, information and figure
• 5:35
  - out if anything get lost. Did you ever
• 5:37
  - get a appointment that you need? Did you
• 5:38
  - get a medication filled when there's a medication
• 5:40
  - change? So I think,
• 5:43
  - you know, having some sort of mechanism,
• 5:45
  - by processes, by the use of a technology
• 5:48
  - to really improve the flow,
• 5:50
  - and also capturing all the missed opportunity and
• 5:53
  - gaps, in a process is something that we've
• 5:56
  - been working on very closely at UChicago Medicine.
• 5:59
  - And with so many AI use cases competing
• 6:01
  - for attention, how do you distinguish a compelling
• 6:04
  - pilot from something ready to become part of
• 6:06
  - everyday care, and what must be in place
• 6:09
  - to scale it responsibly?
• 6:11
  - Yeah. Honestly, there's AI products everywhere. And even
• 6:14
  - some of the older products, they call themselves
• 6:16
  - smart or call themselves AI enabled, and there's
• 6:18
  - just a ton of AI products we're dealing
• 6:21
  - with or,
• 6:22
  - evaluating every day. So we create a new
• 6:24
  - innovation intake process where we sort of streamline
• 6:28
  - everything from cybersecurity review, the use case, our
• 6:31
  - eyes, sort of the, finance review, everything into
• 6:34
  - one, process with a single entry point. So
• 6:37
  - the vendor or the internal submitter will start
• 6:39
  - with the process by submitting,
• 6:41
  - you know, the AI products or use case
• 6:43
  - they want. And then we have, you know,
• 6:46
  - a a team sort of triage this on
• 6:48
  - a weekly basis and making sure all the
• 6:49
  - right part of the party are engaged. And
• 6:52
  - that also includes risk assessments about how risky
• 6:55
  - the solution is, whether we need a pilot,
• 6:57
  - like you mentioned. So for example, something that's
• 7:00
  - more,
• 7:01
  - administrative and there's always a human in the
• 7:03
  - loop, probably lower risk, you can probably turn
• 7:05
  - it on and just have, you know, close
• 7:07
  - feedback with
• 7:09
  - the the frontline staff to to understand whether
• 7:11
  - it works or not. For something that directly
• 7:13
  - to the patients without any human in the
• 7:15
  - loop and patient has seen this firsthand, you
• 7:17
  - obviously want to have a lot of, validation
• 7:20
  - before it goes live. So this is a
• 7:21
  - type of triage and deciding whether there's a
• 7:24
  - pilot and how the pilot's been done in...
• 7:26
  - To what greater detail,
• 7:27
  - that we have decide in the beginning of
• 7:29
  - the intake process.
• 7:31
  - And as a technology relationship moves beyond one
• 7:34
  - time purchase,
• 7:35
  - what should a partner remain accountable for after
• 7:38
  - a go live to make sure the relationship
• 7:40
  - genuinely stays valuable over time?
• 7:43
  - Yeah. I mean, I think there's numerous health
• 7:46
  - care problems we are solving every day and
• 7:48
  - really appreciate
• 7:49
  - the vendors, not just as a seller of
• 7:52
  - a technology solution, like you said, but really
• 7:55
  - more of our partner to help us, solve
• 7:57
  - the problem together. So if you claim your
• 8:00
  - solution, can reduce, for example, readmission rate by
• 8:02
  - 30%,
• 8:03
  - well, let's work together to make it happen.
• 8:06
  - So... Because, you know, we we have... I
• 8:09
  - think we have deployed a lot of the
• 8:11
  - point solution in the past where, you know,
• 8:14
  - it's...
• 8:14
  - The relationship is more like, well, here's a
• 8:17
  - solution... Here's a little technology,
• 8:19
  - deploy this, and, you know, offer you go.
• 8:21
  - Best luck. But I think what we really
• 8:23
  - are thriving for, especially in this new world
• 8:26
  - where AI become more pervasive,
• 8:28
  - the infrastructure work, it become more important. We
• 8:31
  - would like to look for vendors that can
• 8:32
  - really sit down and think through, with us.
• 8:35
  - You know, what infrastructure we have today? What
• 8:37
  - are the problem we're trying to solve? And
• 8:39
  - this designer processes
• 8:40
  - utilizing the technology that it provide to actually
• 8:43
  - solve real world problems that get a result
• 8:45
  - we all want and and really share the
• 8:47
  - benefits together.
• 8:48
  - And finally, as care increasingly moves beyond hospital
• 8:51
  - walls, how are you using connected technologies and
• 8:54
  - AI to identify patients who need earlier follow-up,
• 8:58
  - support care coordination teams, and close gaps in
• 9:00
  - post discharge care? Yeah. Absolutely. And I happen
• 9:03
  - to be a medical director of our our
• 9:04
  - house at home team. So, you know, in
• 9:07
  - which where we try to go beyond the
• 9:09
  - four walls of a hospital and try to
• 9:10
  - see the patient where they are at home,
• 9:12
  - you know, for in inpatient encounter. So there's
• 9:15
  - really a lot, we are seeing through. So
• 9:18
  - changing the order in the hospital and says,
• 9:19
  - let's change your, you know, diuretic dose to
• 9:22
  - half. Doesn't necessarily mean the patient will automatically
• 9:25
  - adjust the drug bottles at home to making
• 9:27
  - sure they take the new dose rather than
• 9:28
  - the the older one. So there's really a
• 9:31
  - lot of, you know, process to work with
• 9:33
  - the patients,
• 9:34
  - to figure out how we can make sure
• 9:36
  - these changes were implemented in the processes.
• 9:39
  - Like I mentioned right in the beginning, I
• 9:41
  - think, you know, using technology is gonna be
• 9:44
  - very key. We don't really have a lot
• 9:47
  - of
• 9:48
  - personnel that can sort of oversee every single
• 9:51
  - patient's care,
• 9:52
  - processes and making sure there's nothing get lost.
• 9:55
  - But this is where I think technology,
• 9:57
  - like, AI can really help us look at
• 9:59
  - what's the missed gap, after something happened.
• 10:02
  - So we pay some referral to a cardiology.
• 10:04
  - Did did a patient actually get a cardiology
• 10:06
  - appointment fulfilled?
• 10:08
  - You know? We we make a medication change.
• 10:10
  - Do we actually see the patient pick up
• 10:12
  - the medication at a pharmacy? Do we actually
• 10:14
  - making sure it validate with the patient they
• 10:15
  - are taking a new dose? So I think
• 10:17
  - a lot of these can really be, facilitated
• 10:20
  - through, AI to help us,
• 10:22
  - you know, orchestrate the whole care process more
• 10:24
  - smoothly so that health care can move away
• 10:26
  - from a more episodic
• 10:28
  - care model to be a more seamless continuous
• 10:30
  - model that we all strive for.
• 10:33
  - Wonderful. Well, Kai, thank you so much for
• 10:35
  - your time and insights today. We also wanna
• 10:37
  - thank our podcast sponsor, Philips. You can tune
• 10:39
  - in to more podcasts from the Becker's Healthcare
• 10:42
  - podcast by visiting our podcast page at beckershospitalreview.com.
• 10:46
  - Thank you so much for having me.