EHR Precision Medicine

April 30, 2021

ID6548

To CiteDCA Citation Guide

00:00It's my pleasure to introduce Doctor
00:03Wade Schultz, who is an assistant
00:05professor of laboratory medicine
00:06and a computational healthcare
00:08researcher at Yale School of Medicine.
00:11He received his pH.
00:12D in microbiology immunology
00:14and cancer biology,
00:15as well as his medical degree
00:17from the University of Minnesota.
00:20He is the Director of Informatics for
00:22the Department of Laboratory Medicine,
00:24director of the Core Center for
00:27Computational Health and Medical
00:29Director of Data Science for Yale,
00:31New Haven Health System.
00:32So thank you so much for
00:34joining us Doctor Scholl's.
00:37Thank you for the introduction and
00:39for people who might be on Gallery
00:41View and zoom if you switch to either
00:43the speaker view or pin my video,
00:44that will make it easiest to see
00:46the presentation if any issues.
00:48Just let me know before I get started.
00:50I just really want to acknowledge one.
00:52All of the support we've gotten from
00:54so many people across the School
00:55of Medicine as well as the old New
00:57Haven Dr Churchwell being one of them
00:59along with rich Lisitano and Lisa
01:01Stump and the fantastic collaboration
01:02that we have with Charlie Tori to
01:04really make this platform a reality.
01:06I do have a couple of potential
01:08conflicts of interest.
01:08Consultant for Hugo Health and
01:11a founder for refactor Health.
01:13To start off the talk,
01:15you know I really want to highlight
01:17one what is computational health
01:18research and this is really defined.
01:21Now is the interdisciplinary application
01:23of computer science tools to address
01:25health related questions and problems
01:26and what this means is really that
01:29we want to leverage different
01:30components of technology to really
01:32advance science and what we do with
01:34this within the computational health
01:35platform is really leverage high
01:37performance computing capabilities
01:38layer on top of that data acquisition
01:41and data modeling strategies,
01:42and then begin to build tools applications.
01:45An analytic insights on top of it.
01:47This system is really aimed to enable
01:49the next generation of healthcare,
01:51analytics and integrated insights
01:52and something that does Doctor
01:54Church will mention during the time.
01:56During Covid,
01:56we rapidly pivoted to really provide
01:58clinical and operational insights
01:59into how are we treating patients?
02:01What are the outcomes in different
02:03populations and to really also
02:05accelerate data driven research
02:06across the TO2T4 spectrum,
02:07which I'll talk about a little bit later.
02:11The last piece of this is that we
02:13really need to integrate large evolving
02:15and heterogeneous datasets to really
02:17make use of these data moving forward,
02:19being able to integrate data from the
02:21HR genomic data sources are real time
02:24sensors and devices across the health
02:26systems provides a really rich data set,
02:28but until we can get those into a
02:30single environment where we can link
02:32patients across the different data
02:33silos and make that available to
02:35investigators limits the potential up
02:37until we have the back capacity available.
02:41The primary goals of this or that
02:42you know one when we started this,
02:44we were really trying to drive
02:46that multi modal data analysis.
02:48How can we integrate the clinical,
02:49the EHR with genomics or radiology
02:51data and so on?
02:53We wanted to also make sure that
02:54we have the capacity available
02:56to allow for AI driven analytics,
02:58so making sure that we have the newest
03:01computing technology available both
03:02from the resources GPU's to really
03:04make sure that our investigators
03:05could leverage and accelerate the
03:07research that they wanted to do,
03:09as well as with some focus on
03:11precision medicine initiatives
03:12such as the generations project,
03:14along with all of that we've
03:15been trying to integrate
03:17real time Insights so that with
03:18some of our collaborators say
03:20in the emergency Department,
03:21making sure that we can.
03:23Access data in real time because if
03:25you start to build predictive models,
03:27say for something like sepsis,
03:28it's not useful if you don't can't deliver
03:31that until a day after observations are made.
03:34Anthem is noting early on, you know,
03:36there's a few different areas that we work
03:38on with the computational health platform.
03:40One of them is the infrastructure and
03:42what we've been developing out in our
03:44newest version of this platform is
03:45making sure that we have a composable
03:48architecture that can allow us to scale
03:49both our compute as well as our storage
03:52as we move between different datasets.
03:54Initially, as I noted,
03:55you know we're really focused on
03:56how can we support generations and
03:58genomic data with covid that's changed.
04:00How can we really accelerate our access
04:01to real time datasets and moving forward?
04:04We really need to be able to scale
04:06that independently as new use cases.
04:08And new datasets come online within
04:10the health system from a data
04:12management perspective,
04:13we really have this need to again,
04:15not only capture E HR data,
04:17but how can we get data from
04:19different multimodal datasets,
04:20including our clinical imaging
04:22datasets and genetics,
04:23and get that into a data model that
04:25can really be understood by analysts
04:27with a variety of backgrounds.
04:29For those of you who are a little
04:32more familiar with our data models
04:34within the health system,
04:35our main data repository or
04:37data warehouse is.
04:38Clarity that has somewhere
04:39around 22,000 tables.
04:40It takes a lot of time and experience
04:42to really become comfortable with.
04:44How do you even extract data from
04:46that database and what we're moving
04:48into our some more simplified data
04:50models that align with many national
04:52initiatives to make that much lower
04:54barrier and burden for investigators.
04:55With some of these clinical
04:58common data models.
04:59And then finally making sure that again
05:01we have that analysis capacity to one
05:03support the the hardware and infrastructure,
05:05but also the tools and libraries on top
05:07of it that individuals might need to use.
05:11The data sources that we acquire
05:12from are quite diverse,
05:14so we do get data from the HR.
05:16Some of that directly from the
05:18main HR database,
05:18as well as the data warehouses
05:20that are on the back end of it.
05:22Digital pathology is something that
05:24we are looking forward to as we
05:26bring that hopefully online within
05:27the health system digital image
05:29and we are fully integrated with
05:31our clinical imaging platform so
05:32we can acquire chest X Rays.
05:34CT scans,
05:34Mris and so on and then are physiologic
05:36monitoring data which come from all of
05:38our attached sensors across the enterprise,
05:40so are.
05:41EKG's pulse oximetry.
05:42All of that streams in real
05:44time into this platform as well.
05:46Genetics and molecular.
05:47We've been acquiring for some
05:48time now on the semantic side,
05:50so for oncology for both our
05:52hematologic malignancy and solid
05:54tumor panels and with the generations
05:55project with both Mike and now having
05:58our onboard starting to move in
05:59germline sequencing results as well.
06:01Digital and Home Health is
06:03something that is again need a
06:05little bit more forward looking.
06:06We do have some ways of acquiring
06:08that within the EHR and are continuing
06:11to look at how we integrate and
06:13expand those datasets moving forward.
06:16For people who haven't heard it
06:17talk about the computational
06:19health platform in the past,
06:20this might be a figure that you've seen.
06:23Really, what this is demonstrating
06:24is the different layers that
06:26we've really put into the system.
06:28On the bottom are data ingestion and storage,
06:30which we are now migrating to our
06:32newest version to allow us a little
06:34bit more elasticity as well as more
06:36efficiency and how we can scale and
06:39acquire both storage and compute resources.
06:41Key to this, you know,
06:42really working those for investigators
06:44is how we integrate those data and
06:46can layer different tools on top of.
06:48The platform that we are building one of
06:50those is a so much comap common data model,
06:53which I'll talk about a little
06:55bit more on the next slide,
06:56as well as some tooling that we're
06:58starting to build to be able to more
07:00efficiently extract data and build what
07:02we call computed or computable phenotypes.
07:04So one challenge with real world
07:05and EHR data is how do we even
07:08identify patients to begin with?
07:09Often this is done in research
07:11studies based off of diagnostic code,
07:13but research by our group and others
07:14you know it has really shown that
07:16diagnostic codes alone are often
07:18not sufficient to really identify.
07:19The breadth and complexity of clinical
07:22diagnosis within the electronic health
07:24record and I can give the use case
07:27example on covid for that in a little bit.
07:29We are also layering on from software to
07:31try to integrate some of our biobanking
07:34capacity both from a physical and
07:36infrastructure capacity as well as
07:38from a tooling and informatics capacity.
07:40On top of that,
07:41use leveraging tool called Freezer Works.
07:44We are also building in some additional
07:46entry points using different applications.
07:48C bio portal,
07:49which I'll speak about a little bit
07:51as well as some API based access to
07:53be able to get slightly more direct
07:55access to data through tools such as
07:58R and Python for investigators and
08:00researchers who have that experience.
08:02So sorry,
08:03but you know talked about on a
08:05couple of these slides.
08:06OMAP is the common data model that we
08:09really leverage to try to push forward
08:11a more accessible data model for our
08:14clinical an EHR data OMAP is managed
08:16by an organization called the Odyssey.
08:18The observation,
08:18ULL health,
08:19data Sciences and informatics organization
08:21and what the script really formed
08:22as several years ago with a private
08:24public partnership between the number
08:26of different pharmaceutical companies
08:28as well as some academic partners,
08:29this is now primarily led
08:31by Columbia University.
08:32But forms and more really simplified data
08:35model to access the clinical information.
08:37So rather than having the 22,000 tables
08:40of clarity simplifies a majority of
08:42clinical data down into about 12:00
08:44or so clinical domains all centered
08:47around the person or the individual.
08:49So the person table maintains the
08:51demographics and then around that
08:53we have observations, providers,
08:54cohorts, drugs, observations,
08:56measurements and by linking those
08:58we can much more rapidly on board
09:01analysts and researchers to really
09:02work with these real world datasets.
09:05In addition to this more simplified
09:06aspect of the data model,
09:08this is the same data model that's being
09:10used by many genomics initiatives,
09:12including emerged the UK Biobank
09:13in the all of US initiative and
09:15then more recently the National
09:16Covid Cohort Collaborative.
09:18Because of this,
09:18individuals who learn the data model
09:20for any of those initiatives will
09:22already know some of the intricacies
09:24of working with our EHR data.
09:26How do you find an encounter?
09:27How do you find a pulse rate?
09:29How do you find a creatinine value?
09:31All of that is a transferable skill skill
09:34between all of those different databases.
09:38And now we did, you know, really,
09:40start to build this computational health
09:42platform with a focus on generations.
09:44When we started what we've shifted
09:45to over the last year is really
09:47supporting covid analytics and
09:49we had to do this quite rapidly.
09:51This has been a tremendous collaboration
09:53between the number of groups you know,
09:55especially our joint data and analytics team,
09:57which Charlie Tori also leads,
09:58and being able to integrate this with
10:01our other clinical and operational
10:02data assets has really allowed us to
10:05one drive how we can acquire data in
10:07real time for operational planning.
10:08But to build out really integrated
10:11research platforms that can go from
10:13bench to bedside and back again,
10:15one of these integrated informatics
10:16pipelines that we've used as help
10:18drive research with a lot of our
10:20fantastic immuno biologist across
10:22the organization Icoi, Wisocky,
10:24Ehrenring and others.
10:25The way that we've implemented this is
10:27really by using chip to help monitor
10:30our real-world data feeds for COVID-19
10:32testing as well As for visits across
10:34the health system to potentially to
10:35or rather to identify potentially
10:37eligible patients in clinical trials.
10:39And then integrate that with tools
10:41like Red Cap where we can do more
10:44protocol driven data capture.
10:46So what this lets us do is balance
10:47out and acquire real-world outcomes
10:49from the HR where people hospitalized
10:51where they intimated what medications
10:53did they receive, receive,
10:55and what were the results of
10:57laboratory testing.
10:58And then if patients are consented
11:00with the appropriate IRB's and all
11:02of the regulatory pieces in place,
11:04we can combine those clinical data
11:05with protocol driven data that we
11:07acquire through Redcap and provide
11:09back integrated data resources.
11:11For investigators,
11:11we've been pretty successful in terms of
11:14being able to advance science with this.
11:16Throughout the course of the pandemic.
11:18Again,
11:18a lot of this fantastic work that's
11:20really being driven by our immuno
11:22biologist looking at things from
11:23the immune response to developing
11:25out predictive models.
11:26With clinical informatics in our
11:28emergency Department and looking at
11:30outcomes in our cohort and now changes
11:32in mortality over time and the other area,
11:34then that I'm just going to hop back
11:36to really quickly since we're a little
11:38bit short on time is how we've also been.
11:41Leveraging Chip to advance genomic research
11:43primarily focused around Cymatic mutation.
11:45To date,
11:46hematology is really one of the first
11:48areas that we're working on this with.
11:51With Stephanie Haleine and others
11:52in our hematology,
11:53and he he monk groups and what we use
11:56chip for the computational Hellpop
11:57platform is to build out these
12:00analysis pipelines and integrate
12:01data from all of these different
12:04clinical sources so the EHR, the LIS,
12:06the imaging system and others.
12:09We can then also integrate our
12:10biospecimen data so we can actually
12:12track our populations in real time
12:14and for patients who have consented
12:17or for excess specimens acquire
12:19clinical specimens through the
12:20clinical laboratory and then integrate
12:22those with freezer works.
12:23Our BIOSPECIMEN data management software.
12:26From our sequencing laboratories.
12:27So right now that's primarily our molecular
12:30diagnostics laboratory and lab medicine,
12:31but also with YCGA&R tumor profiling groups.
12:34We can integrate our genomic data as well
12:36as registry data from Redcap and others,
12:38and push those back to different data
12:41analysis tools so it believes Doctor
12:43Murray was mentioning a little bit
12:45about Nomad is one of the options that
12:47we have and are working to deploy
12:49or others are working to deploy
12:51for germ line for tumor sequencing.
12:53We're really looking at C bio portal
12:55as being one of the entry points for
12:58investigators to access these data.
13:00C bio portal is actually open source
13:02software that was developed at Memorial
13:04Sloan Kettering as well as a couple
13:07of different consulting groups,
13:09but provides the investigator
13:10driven ability to have really used
13:12user interface to start to look at
13:15outcomes across these populations.
13:16So using our computable phenotypes,
13:18we can identify patients with AML,
13:20AML, CML,
13:21integrate the genomic data that
13:23we've done for clinical sequencing,
13:25and provide this back in these drivable
13:27interfaces to start investigation
13:28and hypothesis testing while still
13:30providing more advanced analytics.
13:32And integration capacity through data
13:34driven API's to extract the underlying
13:37genomic data for investigators who need it.
13:40As we move forward and continue to
13:41expand our work with generations,
13:43just the one area I wanted to highlight
13:46which might get talked about a little bit.
13:48Is the pharmacogenetic genetics
13:50aspect of this program.
13:52For this which is really being
13:54led by Rebecca Vulcan Pharmacy,
13:55is that we've got an EHR order that
13:57we can use to order pharmaco genomic
13:59testing directly on individuals.
14:01Or we can get it as part of
14:04the generations panel.
14:05Once the specimens are obtained,
14:07those specimens go to why CGA for sequencing.
14:12Within the EHR, we've got a third
14:14party vendor called Act X that can
14:17provide integrated decision support.
14:18So at the time of medication order,
14:20if we enter in order into the EHR,
14:23we can actually provide back in real
14:25time a BPA that provides individuals
14:27with some feedback of whether a
14:29person has a genetic interaction with
14:31the drug that is being prescribed.
14:33And this is for the handful of drugs
14:36that we have on that panel today.
14:38How we're leveraging Chip and the
14:40computational health platform.
14:41Is that once why SGA is done with sequencing?
14:44We pulled those genetic variants
14:46across and integrate that within
14:47the Actex database so that those
14:49pharmaco genomic interactions can be
14:51immediately available to providers
14:52while also on the other side,
14:54providing researchers with access to
14:55not only those clinical variants,
14:57but also every other variant that
14:59we have available on the chip.
15:01So this provides us with the very
15:03rich environment to be able to
15:05capture data integrated and really
15:06provide real time feedback to the EHR,
15:09not just in a research capacity.
15:12So in summary,
15:13you know Chip is really focused
15:15on how can we link all of these
15:18different datasets together,
15:19but with that ultimate goal of really
15:21enabling the next generation of healthcare
15:23analytics and integrated insights,
15:24we're trying to move this just,
15:26you know,
15:27beyond thinking about just genomics
15:29or just the HR data and really
15:31providing this as an integrated
15:32repository to advance science,
15:34clinical care and clinical operations.
15:36Ship has been integral to providing real
15:37time data for clinical and operational
15:40dashboards related to COVID-19.
15:41But now as we're moving more
15:43into the chronic phases,
15:44as we are starting to recruit people
15:47back into generations and opening
15:48up more of our research again,
15:50we do have our ongoing plans to
15:52continue to integrate these molecular
15:53data with clinical outcomes in
15:55some targeted collaborations,
15:56and would are also looking to expand
15:58that out to several more general
16:00populations as we move forward.
16:02These plans,
16:02software deployments including C,
16:03Bio Portal no Bad,
16:04and others will continue to increase
16:06Accessibility to both germline
16:07Anthem Attic sequencing data across
16:09the organization and with the
16:11goal of integrating that.
16:12With this oh map data model so that
16:14we can also provide clinical outcomes
16:16and clinical data to pair with this
16:19generated research information,
16:20or rather, genomic information.
16:22So thank you for the time and if
16:24anybody has questions happy to follow
16:26up by email afterwards as well.