In this episode of the Scope of Things, host Deborah Borfitz brings you the latest news on AI-recommended precision dosing, organoid drug testing aiding treatment selection for bowel cancer, an AI tool for stratifying lung cancer patients, using HIV drugs to treat Alzheimer’s disease, and the potential value of magic mushrooms to remedy the mood symptoms of Parkinson’s. Blythe Adamson, international head of outcomes research and evidence generation at Flatiron Health, also joins in to discuss groundbreaking work harmonizing patient-level real-world data across four countries to enable multinational oncology research.
News Roundup
CURATE.AI platform
FORECAST-2 clinical trial
- News on the Walter and Eliza Hall Institute website
AI tool for sorting cancer patients
- Study in Nature Communications
HIV drugs for Alzheimer’s protection
- Study in Alzheimer’s & Dementia
“Magic mushrooms” for Parkinson’s disease
- Study in Neuropsychopharmacology
Guest
Blythe Adamson, Ph.D., international head of outcomes research and evidence generation at Flatiron Health
- Flatiron Health enabling multinational oncology researh – article in Bio-IT World
Subscribe to the podcast on Apple Podcasts and leave a review to support the show. Join us at Scope Europe on October 14-15 in Barcelona - use code SOT10 for an additional 10% discount.
GUEST BIO
Blythe Adamson, PhD, MPH, Head of Outcomes Research and Evidence Generation, International at Flatiron Health & Founder of Infectious Economics
Dr. Blythe Adamson is the head of international outcomes research and evidence generation at Flatiron Health.
As a visionary senior leader at Flatiron Health, her team pioneered deep learning language models for extraction of clinical details from EHR documents, breaking the limits of what was possible for humans to do alone. Learning from the experience of millions of patients with cancer, they generate evidence of treatment effectiveness and value used by governments around the world. Dr. Adamson co-invented a patented clinical decision-support tool, enabled by machine learning, that is used by cancer clinics to benefit patients.
She holds degrees in microbiology, epidemiology, and pharmaceutical economics with a focus on infectious disease prevention. Dr. Adamson has held roles at the Bill and Melinda Gates Foundation Institute for Disease Modeling, the NIAID HIV Vaccine Trials Network, and Flatiron Health.
The Scope of Things podcast explores clinical research and its possibilities, promise, and pitfalls. Clinical Research News senior writer, Deborah Borfitz, welcomes guests who are visionaries closest to the topics, but who can still see past their piece of the puzzle. Focusing on game-changing trends and out-of-the-box operational approaches in the clinical research field, the Scope of Things podcast is your no-nonsense, insider’s look at clinical research today.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Deborah Borfitz (00:02):
Hello and welcome to the Scope of Things
podcast, a no-nonsense look atthe promise and problems of
clinical research, based on asweep of the latest news and
emerging trends in the field andwhat I think is worthy of your
30 or so minutes of time.
I'm Deborah Borfitz, seniorScience Writer for Clinical
Research News, which means Ispend a lot of time with my ear
(00:23):
to the ground on your behalf anda lot of hours every week
speaking to top experts fromaround the world.
Please consider making thisyour trusted go-to channel for
staying current on things thatmatter, whether they give us
hope or cause for pause.
In another five minutes or so,I'll be talking with Blythe
Adamson, international Head ofOutcomes Research and Evidence
(00:44):
Generation at Flatiron Health,about its groundbreaking work
harmonizing patient-levelreal-world data across four
countries to enablemultinational oncology research.
But first the latest news,including AI-recommended
precision dosing, organoid drugtesting, aiding treatment
selection for bowel cancer,organoid drug testing, aiding
(01:07):
treatment selection for bowelcancer, an AI tool for
stratifying lung cancer patients.
Using HIV drugs to treatAlzheimer's disease and the
potential value of magicmushrooms to remedy the mood
symptoms of Parkinson's.
Researchers in Singapore havebuilt a platform known as
Curateai that creates clinicallyactionable patient profiles, or
digital twins when given dataon only two blood biomarkers
(01:30):
together with drug type and dose.
This could aid precision dosingfor multiple conditions,
including patients with advancedsolid tumors.
Per a small interventionalclinical trial, finding a high
rate of user adherence to thetreatment recommendations.
For some patients, theAI-proposed doses were lower on
average than those suggested bystandard of care guidelines.
(01:50):
Beyond cancer, this dynamicdose adjustment approach has now
been used in trials forhypertension, immunotherapy and
cognitive training.
The latest feasibility studywill expand into larger
randomized controlled trials tovalidate the effectiveness of
the Curateai platform againsttraditional treatment regimens
and factor in savings to thepatient and healthcare system,
(02:12):
in addition to the usualefficacy and patient outcome
measures.
In Australia, meanwhile,researchers have launched a
game-changing Forecast 2clinical trial where tumor
organoids many cancers grown inthe lab from a patient's own
tissue samples are being used topredict what drugs will work
for newly diagnosed cases ofbowel cancer before treatment
(02:33):
begins.
The findings could help replacecurrent trial and error
treatment practices with a moretailored and personalized
approach to improve survivalrates and quality of life for
patients.
Researchers previouslyvalidated organoid drug testing
as an accurate tool in thetreatment selection process for
patients with advanced bowelcancer.
The new trial will look atwhether those results can be
(02:55):
replicated in people who haverecently been diagnosed.
In collaboration with RegeneronPharmaceuticals, investigators
at Weill Cornell Medicine reporton a new AI-based method for
sorting lung cancer patientsinto groups that have similar
characteristics before treatmentand similar outcomes after
treatment.
The approach predicted outcomesfrom health record data on
(03:16):
patients treated with immunecheckpoint inhibitors and
outperformed other methodspublished to date.
The machine learning algorithmwas trained on the de-identified
health records of over 3,000patients with lung cancer in a
commercial database and sortedthem into three groups based on
survival time from the start oftreatment.
When applied to the new,smaller data set of lung cancer
(03:37):
patients, it yielded almostidentical groupings.
The AI tool could find utilityfor patient stratification and
clinical trials of newpharmaceuticals, as well as
individual treatment selection.
New research indicates HIV drugscalled nucleoside reverse
transcriptase inhibitors, orNRTIs, could offer protection
(03:58):
against Alzheimer's disease,based on the discovery that
patients taking the medicationsare substantially less likely to
develop the memory-robbingcondition.
An analysis of two large healthinsurance databases found the
risk of developing Alzheimer'sdecreased every year by 6% and
13% respectively in patientstaking NRTIs, even after
(04:19):
adjusting for factors that mightcloud the results by that
yardstick.
Taking the drugs could possiblyprevent about 1 million new
cases of Alzheimer's diseaseworldwide every year.
The drug appears to prevent theactivation of important immune
system agents known asinflammasomes.
The research team now plans totest its own
(04:39):
inflammasome-blocking drug inupcoming clinical trials to
assess its preventive poweragainst Alzheimer's disease.
And finally, in a small pilotstudy, researchers in California
have found that magic mushrooms, previously showing promise in
treating depression and anxiety,could also help Parkinson's
disease patients.
(05:00):
Psilocybin therapy meaningfullyimproved mood, cognition and
motor symptoms and was welltolerated by study participants
and, remarkably, the beneficialeffects lasted for weeks after
the drug was out of their system.
It is the first time apsychedelic has been tested on
patients with anyneurodegenerative disease.
Researchers point out that moodsymptoms and Parkinson's are
(05:22):
linked to faster physicaldecline and are a stronger
predictor of patients' qualityof life with the disease than
the better-known motor symptoms.
A larger randomized controlledtrial, now underway at UC San
Francisco as well as Yale, aimsto enroll 100 participants and
incorporate non-invasive brainstimulation, neuroimaging and
other tools to understand howpsilocybin impacts inflammation
(05:45):
and neuroplasticity.
As a reminder, links to thevarious articles and studies
informing this news segment canbe found in the show notes.
It is now time for our chatwith Blythe Adamson about the
harmonized cancer-specificdatasets created by Flatiron
Health, which researchersanywhere in the world can now
(06:06):
analyze in a secure, cloud-based, trusted research environment.
Welcome to the show, Blythe.
Blythe Adamson (06:11):
Thank you for having me, Deborah.
It's great to be here.
Deborah Borfitz (06:14):
The cross-border patient-level data
sharing exercise we're talkingabout here was, by all accounts,
a monumental feat because itinvolved four countries the US,
of course, but also the UK,germany and Japan all with
different regulations andprivacy standards and all having
different places for, andpractices around data collection
(06:35):
.
I know it took a while toconvince the scientific
community that this was evenpossible.
So take us back to the earlydays, if you would, when you
were on the conference circuitsharing the news.
What was the initial receptionto the data sets and what did it
take to flip the conversationfrom disbelief to the
potentially enormous benefits tocancer researchers and patients
(06:57):
?
Blythe Adamson (06:58):
Well, when I first started presenting
research studies about, forexample, colorectal cancer in
Japan having earlier age atonset than we'd ever seen
historically, or looking atbreast cancer in the UK and
Germany, I thought that thepeople coming up to ask
(07:18):
questions about the poster orour oral presentations would be
wanting to know about theepidemiology of these diseases
or the statistical methods thatwe used.
But everyone first came up andsaid I can't even believe it.
I didn't think data like thisexisted.
I didn't think that it was thatthe regulations and laws within
(07:43):
Europe and Japan would allowthis to be possible, and so it
was just so interesting to mebecause I had really prepared to
talk about deep methods and epi.
But really what people wantedto understand was how did you
build a global healthcare datainfrastructure in a way that was
(08:04):
compliant with, we know, ourstricter data privacy laws
within these different marketsand, honestly, it took Flatiron
more than five years to navigatethrough this.
Yes, exactly, to find the wayto do this so that patient
(08:24):
privacy is completely respected,that the processes of
de-identification of theelectronic medical record data
or the anonymization of therecords.
It's now done locally withinthe country and then uploaded
into a secure, compliantenvironment.
(08:44):
So this has just been a keyturning point, because
previously, what people had beenmost familiar with was these
federated data models, which iswhere Data is not allowed to
leave the country, and so youhave to, for example, do one
analysis in Germany, oneanalysis in Spain, one analysis
in France, and then you justprint out the results and try to
(09:05):
smush them all up together andthey often just don't completely
make sense, and this is nowwhat technology has been able to
overcome.
So it's a really excitingmoment and the way that I then
had to step back and say okay,if you're not ready for me to
show you research results yet,how can I figure out a way to
(09:25):
more effectively communicatethis?
And it really came through theISPOR EHR-derived data
suitability checklist that'savailable and published by a
task force from thisprofessional society, and it
really gives you a checklist togo through to ensure whether or
not EHR data is fit for purposefor different research questions
(09:48):
, regulatory decisions and itgoes through how to characterize
the data relevance, reliability, governance, provenance,
relevance, reliability,governance, provenance.
And so this is an extensivepeer-reviewed paper that I wrote
with my team and have nowpublished to be able to go
through every single item in thechecklist to transparently
(10:11):
communicate.
Where does the CHR data comefrom?
How is it governed?
How is it kept secure?
How is it curated so thatpeople can see?
Oh wow, we've never hadanything like this before
outside the United States.
Deborah Borfitz (10:25):
And you just mentioned data curation.
I want to deep dive on thatjust a little bit, because I
think that's part of thedistinctiveness of this approach
here.
Could you talk a little bitabout that, how that was handled
?
I think our readers might, orour listeners might, appreciate
hearing more about that.
Blythe Adamson (10:40):
Right.
So in contrast to other typesof real world data that
sometimes they send physicianssurveys that say, okay, tell me
about a patient you've had withmultiple myeloma, when did they
get diagnosed, what treatmentsdid you give them we are
integrating at the source of theEHR.
(11:01):
So imagine, you know,installing servers and basements
of hospitals and clinics acrossGermany where we really are
getting real-time access andintegrations to all of the
clinical notes, the pathologyreports, all this
non-standardized documents.
This is really where a lot ofthe treasures are in oncology.
(11:23):
It's not in the structuredbilling data.
The richness comes from thesepatients' stories that are often
described in paragraphs ofnotes that these doctors are
writing about them.
So we do this longitudinalcuration.
So when we integrate, we curateand understand the whole
(11:45):
retrospective history of thesepatients' experiences and
prospectively follow themforward to start continuing to
capture their experiences andfollowing them over time.
So, with regular refreshes, atleast once a quarter, we're
updating these data sets so thatwe can have real-time
(12:08):
information, nuanced analysesthat have the clinical details
with provenance, where you cantrace all the way back to the
source of the document, the linenumber.
You know, the place where wecan verify this information is
found.
Speaker 3 (12:25):
Are you enjoying the conversation?
We'd love to hear from you.
Please subscribe to the podcastand give us a rating.
It helps other people find andjoin the conversation.
If you've got speaker or topicideas, we'd love to hear those
too.
Deborah Borfitz (12:45):
You can send them in a podcast review.
So what are the clinicalresearch opportunities you're
most hoping to see pursued withthe integrated data sets, and
have you seen any movement inthose directions?
Absolutely.
Blythe Adamson (12:54):
You know, interestingly it was the FDA who
started asking for some of thisinformation, rather than what
you might expect is the EMA, andit came because, out of the
increasing use of externalcontrol arms, so often for a
single arm clinical trial thatmight be global, they may do a
(13:16):
control arm that comes fromreal-world patient experiences,
and previously many of thosecontrol arms were comparing
real-world patients in theUnited States to a global
clinical trial.
But we know that there's a lotof variability in patient
characteristics and thetreatments used and healthcare
systems between countries.
(13:38):
And so the FDA had said wouldn'tit be better if the real world
external control arms had a morediverse and representative
composition that's morecomparable to the global single
arm trial?
So I'm already seeing theadoption and use by many life
science companies in designingglobal external control arms to
(14:02):
compare them to their globalsingle arm trials, and we're
also seeing this in comparativeeffectiveness studies within
these different countries.
So one of the new initiativesis the joint clinical assessment
in Europe where regulatory andHTA decisions are happening
together.
They're wanting to enumeratewhat people call PICOs the
(14:26):
population interventioncomparator outcomes that are
relevant for all the differentcountries in the European Union,
and the dilemma they often haveis the local standard of care
in Latvia might not be somethingthat's even used in the United
States, and so we're using thesecurated data sets in UK,
(14:47):
germany and Japan to surfacemore patients that have unique
experiences and treatmentpathways that we can learn from
that may be very different fromthe United States, so it's
really exciting to see the quickadoption and range of use cases
that life science companies arenow taking advantage of with
(15:09):
global EHR-derived data.
Deborah Borfitz (15:11):
How do researchers go about accessing
these data sets and getting intothis trusted research
environment to do their analysiswork, and are there any you
know caveats we need to mentionaround that?
Blythe Adamson (15:24):
Right, as I mentioned before, many are
familiar with that federatedmodel where you have to do
separate analyses in everycountry, and it's such a relief
to me now that we've technicallyovercome that limitation.
So trusted research environmentsnow are the way where, as you
(15:45):
mentioned, it's cloud-based,compliant, secure, and this is
where we upload thede-identified or anonymized EHR
oncology data sets so that theycan be pooled together at a
patient level.
So if you imagine a tablethat's describing the
demographic characteristics withone row per patient, you can
(16:08):
now pool together and combineall these different countries
into one table, and that enableslots of different types of more
rigorous methods for analysis.
And it allows you, within theFlatiron Trusted Research
environment for researchers, touse familiar tools like coding
(16:31):
in R or Python within theseenvironments, really being able
to touch and visualize thepatient level data within the
trusted research environment.
And then, once you have theresults from your studies where
you've pooled all the countriestogether, then you can export
the results, so download thoseto your local computer to be
(16:55):
able to use in publications.
Regulatory submissions and thisis the step that we use to
ensure the data security andcompliance with regulations like
GDPR in Europe and AAPI inJapan.
Deborah Borfitz (17:13):
Now are all cancers represented, or will
they be at some point?
Blythe Adamson (17:17):
Yes, we're increasing the number very
rapidly.
So I think right now we haveeight different cancer types
that have been curated and areresearch ready sitting in this
tier trusted researchenvironment where researchers
are now asking questions aboutthe effectiveness of different
treatments.
And the team at Flatiron isquickly adding more and more
(17:42):
cancer types into thisenvironment to do analyses and
then also increasing the cohortsizes of them very quickly.
So we've got a network ofFlatiron sites within each of
the countries.
So, for example, in the UK itis a collection of NHS trusts,
(18:05):
which is where cancer care isdelivered in the UK.
In Germany, for example, wehave a huge network of lots of
small community clinics, whichis more of how cancer care is
delivered in Germany.
So the number of sites areincreasing, patients are
increasing and the cancer typesthat are research-ready to use
(18:27):
is increasing as well.
Deborah Borfitz (18:29):
a great segue to my next question, because
we're talking aboutpatient-level data from four
countries.
So let's get to the topic oftransportability of real-world
evidence across country borders.
That's an area where FlatironHealth has been actively working
, I know, with sponsor companiesto determine when it is
appropriate for one country tokind of borrow information from
(18:51):
another.
Is this the intention, then,with the data sets that you're
developing here, that?
Are they geographically diverseenough, or is the plan to
repeat the harmonizationexercise in yet more nations?
Blythe Adamson (19:05):
Well, when I speak with regulators and HTA
bodies, you know governmentdecision makers all of them have
a preference for local data.
They would rather havereal-world data from their own
country to make a decision, butthe dilemma is it's often not
available, it's too sparse orinsufficient for some reason.
So they do have this dilemma ofwould we rather have nothing or
(19:28):
do we make the assumption thatit's okay to use data from
another country to make adecision about cancer care in
our country?
And it's an assumption that hasnever been tested before.
And so we really had a lot ofbenefit from doing a formal
research collaboration with NICEin the UK, and all these other
(19:51):
countries had so much interestin joining us, and many life
science companies wanted to worktogether that we established a
research consortium called theFORUM, which stands for
Fostering Oncology, rwe Uses andMethods, where together we are
doing a huge set of validationstudies testing whether or not
(20:14):
differences between twocountries' healthcare systems is
so big and impactful that itactually might change how long
people survive with cancer thesame patient.
You know, if you live inGermany versus Denmark, versus
Spain or the US, if you haveidentical patient
characteristics, if you receivethe exact same drug.
(20:36):
Do you survive the same amountof time?
Or are differences in the waynutrition, the environment, how
people are cared for in thehospital do those things
actually make a difference?
And this is what the forum isstudying.
So we are going country bycountry for breast cancer, lung
(20:57):
cancer, heme and looking atdifferent combinations of
countries to test whether or not.
Once you control for thepopulation characteristics being
different, once you select fora specific type of cancer and
drug that people are using andyou control for everything, we
are learning from the experienceof patients in the US,
(21:20):
predicting the outcomes ofpatients in a different country
and then comparing that to theirtrue overall survival in that
country to see whether or not itactually was modified by
differences in the healthcaresystem.
So it's been pretty exciting.
I mean, we've learned that inmany cases that the real world
evidence is transportablebetween countries.
(21:42):
But there are some times whereit's really not appropriate,
like where we found where, whenclinical guidelines between two
countries are wildly different,like everyone in the US starts
with drug A, then B, then C, butin Europe maybe they start with
B, then D, then F.
(22:03):
You know, you can't reallyadjust away when people receive
completely different drugs, andso we're learning about what are
the situations where you reallydo need local data to be able
to inform decisions in thatcountry referenced when life
science companies are submittingto these different countries to
support whether or not thatassumption is appropriate that
(22:38):
it's okay for them to borrowevidence from another country.
Deborah Borfitz (22:41):
Wow, fascinating, interesting and
what a work in progress.
Right, we've covered a lot ofground here, blythe, so let me
end with this when is this allgoing?
Just how big and robust do youimagine the global data
infrastructure will be in, say,another five years?
Blythe Adamson (23:00):
I am really, really optimistic about this,
and when I think about globalhealthcare data infrastructure,
I feel really proud to beworking at a private company
that's solving this, because itreally is bigger than what an
individual country can do bythemselves, and one of the keys
is having harmonized, commondata models across these
(23:23):
populations and also havingrespect for true differences
between these countries and howcancer care is delivered, how
patients are different, howdocumentation styles are
different.
But I anticipate, you know,within five years there's going
to be a much larger, more robustglobal health data ecosystem,
(23:47):
and so much of that, I think, isgetting super powered by large
language models being one of thetools we use now for curation.
In the past, so much of thiswas done manually by cancer
nurses opening up charts,pulling out information.
Now a lot of that is being usedto validate LLMs doing some of
(24:10):
those same tasks for us.
So I think that it's a reallyexciting moment in history to be
a researcher in this field,because the number of questions
that we can answer that werenever possible to understand
before is just exponentiallygrowing.
Deborah Borfitz (24:28):
I hear your enthusiasm.
This has been a trulyenlightening and encouraging
conversation, and you havesucceeded in making it sound
very cool to be a data geek.
So I thank you for being ontoday's show and ask only that
you leave us.
As I told you in advance, Iwould do to leave us with your
favorite data geek quote.
Blythe Adamson (24:49):
Garbage in, garbage out.
I work at Flatiron because it'sthe most beautiful data in the
world that I've ever touched,and so I am so thrilled to be in
a place with high quality datato be able to put out high
quality findings.
Deborah Borfitz (25:07):
You heard it here.
A big thank you to everyone outthere for listening, and if you
are not subscribed to thispodcast yet, please consider
going to Apple Podcasts anddoing so right now, so you don't
miss your monthly dose of newsand perspectives.
You'll be hard-pressed to findanywhere else and, if you're up
for it, I'd also be so verygrateful if you leave a rating
(25:29):
and review while you are there.
One more thing before we go.
If you liked today'sconversation, it is only a
glimpse of what you can expectfrom Scope Europe.
Presenters and panelists,please plan to join us October
14th and 15th in Barcelona whenclinical operations executives
will be exploring the latesttrends in clinical trial
innovation, planning andoperations.
(25:51):
Save an additional 10% off anycurrent rate by using the code
SOT10.
For more information, visitscopesummiteuropecom.
Bye for now.
Hello and welcome to the Scope of Things
podcast, a no-nonsense look atthe promise and problems of
clinical research, based on asweep of the latest news and
emerging trends in the field andwhat I think is worthy of your
30 or so minutes of time.
I'm Deborah Borfitz, seniorScience Writer for Clinical
Research News, which means Ispend a lot of time with my ear
(00:23):
to the ground on your behalf anda lot of hours every week
speaking to top experts fromaround the world.
Please consider making thisyour trusted go-to channel for
staying current on things thatmatter, whether they give us
hope or cause for pause.
In another five minutes or so,I'll be talking with Blythe
Adamson, international Head ofOutcomes Research and Evidence
(00:44):
Generation at Flatiron Health,about its groundbreaking work
harmonizing patient-levelreal-world data across four
countries to enablemultinational oncology research.
But first the latest news,including AI-recommended
precision dosing, organoid drugtesting, aiding treatment
selection for bowel cancer,organoid drug testing, aiding
(01:07):
treatment selection for bowelcancer, an AI tool for
stratifying lung cancer patients.
Using HIV drugs to treatAlzheimer's disease and the
potential value of magicmushrooms to remedy the mood
symptoms of Parkinson's.
Researchers in Singapore havebuilt a platform known as
Curateai that creates clinicallyactionable patient profiles, or
digital twins when given dataon only two blood biomarkers
(01:30):
together with drug type and dose.
This could aid precision dosingfor multiple conditions,
including patients with advancedsolid tumors.
Per a small interventionalclinical trial, finding a high
rate of user adherence to thetreatment recommendations.
For some patients, theAI-proposed doses were lower on
average than those suggested bystandard of care guidelines.
(01:50):
Beyond cancer, this dynamicdose adjustment approach has now
been used in trials forhypertension, immunotherapy and
cognitive training.
The latest feasibility studywill expand into larger
randomized controlled trials tovalidate the effectiveness of
the Curateai platform againsttraditional treatment regimens
and factor in savings to thepatient and healthcare system,
(02:12):
in addition to the usualefficacy and patient outcome
measures.
In Australia, meanwhile,researchers have launched a
game-changing Forecast 2clinical trial where tumor
organoids many cancers grown inthe lab from a patient's own
tissue samples are being used topredict what drugs will work
for newly diagnosed cases ofbowel cancer before treatment
(02:33):
begins.
The findings could help replacecurrent trial and error
treatment practices with a moretailored and personalized
approach to improve survivalrates and quality of life for
patients.
Researchers previouslyvalidated organoid drug testing
as an accurate tool in thetreatment selection process for
patients with advanced bowelcancer.
The new trial will look atwhether those results can be
(02:55):
replicated in people who haverecently been diagnosed.
In collaboration with RegeneronPharmaceuticals, investigators
at Weill Cornell Medicine reporton a new AI-based method for
sorting lung cancer patientsinto groups that have similar
characteristics before treatmentand similar outcomes after
treatment.
The approach predicted outcomesfrom health record data on
(03:16):
patients treated with immunecheckpoint inhibitors and
outperformed other methodspublished to date.
The machine learning algorithmwas trained on the de-identified
health records of over 3,000patients with lung cancer in a
commercial database and sortedthem into three groups based on
survival time from the start oftreatment.
When applied to the new,smaller data set of lung cancer
(03:37):
patients, it yielded almostidentical groupings.
The AI tool could find utilityfor patient stratification and
clinical trials of newpharmaceuticals, as well as
individual treatment selection.
New research indicates HIV drugscalled nucleoside reverse
transcriptase inhibitors, orNRTIs, could offer protection
(03:58):
against Alzheimer's disease,based on the discovery that
patients taking the medicationsare substantially less likely to
develop the memory-robbingcondition.
An analysis of two large healthinsurance databases found the
risk of developing Alzheimer'sdecreased every year by 6% and
13% respectively in patientstaking NRTIs, even after
(04:19):
adjusting for factors that mightcloud the results by that
yardstick.
Taking the drugs could possiblyprevent about 1 million new
cases of Alzheimer's diseaseworldwide every year.
The drug appears to prevent theactivation of important immune
system agents known asinflammasomes.
The research team now plans totest its own
(04:39):
inflammasome-blocking drug inupcoming clinical trials to
assess its preventive poweragainst Alzheimer's disease.
And finally, in a small pilotstudy, researchers in California
have found that magic mushrooms, previously showing promise in
treating depression and anxiety,could also help Parkinson's
disease patients.
(05:00):
Psilocybin therapy meaningfullyimproved mood, cognition and
motor symptoms and was welltolerated by study participants
and, remarkably, the beneficialeffects lasted for weeks after
the drug was out of their system.
It is the first time apsychedelic has been tested on
patients with anyneurodegenerative disease.
Researchers point out that moodsymptoms and Parkinson's are
(05:22):
linked to faster physicaldecline and are a stronger
predictor of patients' qualityof life with the disease than
the better-known motor symptoms.
A larger randomized controlledtrial, now underway at UC San
Francisco as well as Yale, aimsto enroll 100 participants and
incorporate non-invasive brainstimulation, neuroimaging and
other tools to understand howpsilocybin impacts inflammation
(05:45):
and neuroplasticity.
As a reminder, links to thevarious articles and studies
informing this news segment canbe found in the show notes.
It is now time for our chatwith Blythe Adamson about the
harmonized cancer-specificdatasets created by Flatiron
Health, which researchersanywhere in the world can now
(06:06):
analyze in a secure, cloud-based, trusted research environment.
Welcome to the show, Blythe.
Blythe Adamson (06:11):
Thank you for having me, Deborah.
It's great to be here.
Deborah Borfitz (06:14):
The cross-border patient-level data
sharing exercise we're talkingabout here was, by all accounts,
a monumental feat because itinvolved four countries the US,
of course, but also the UK,germany and Japan all with
different regulations andprivacy standards and all having
different places for, andpractices around data collection
(06:35):
.
I know it took a while toconvince the scientific
community that this was evenpossible.
So take us back to the earlydays, if you would, when you
were on the conference circuitsharing the news.
What was the initial receptionto the data sets and what did it
take to flip the conversationfrom disbelief to the
potentially enormous benefits tocancer researchers and patients
(06:57):
?
Blythe Adamson (06:58):
Well, when I first started presenting
research studies about, forexample, colorectal cancer in
Japan having earlier age atonset than we'd ever seen
historically, or looking atbreast cancer in the UK and
Germany, I thought that thepeople coming up to ask
(07:18):
questions about the poster orour oral presentations would be
wanting to know about theepidemiology of these diseases
or the statistical methods thatwe used.
But everyone first came up andsaid I can't even believe it.
I didn't think data like thisexisted.
I didn't think that it was thatthe regulations and laws within
(07:43):
Europe and Japan would allowthis to be possible, and so it
was just so interesting to mebecause I had really prepared to
talk about deep methods and epi.
But really what people wantedto understand was how did you
build a global healthcare datainfrastructure in a way that was
(08:04):
compliant with, we know, ourstricter data privacy laws
within these different marketsand, honestly, it took Flatiron
more than five years to navigatethrough this.
Yes, exactly, to find the wayto do this so that patient
(08:24):
privacy is completely respected,that the processes of
de-identification of theelectronic medical record data
or the anonymization of therecords.
It's now done locally withinthe country and then uploaded
into a secure, compliantenvironment.
(08:44):
So this has just been a keyturning point, because
previously, what people had beenmost familiar with was these
federated data models, which iswhere Data is not allowed to
leave the country, and so youhave to, for example, do one
analysis in Germany, oneanalysis in Spain, one analysis
in France, and then you justprint out the results and try to
(09:05):
smush them all up together andthey often just don't completely
make sense, and this is nowwhat technology has been able to
overcome.
So it's a really excitingmoment and the way that I then
had to step back and say okay,if you're not ready for me to
show you research results yet,how can I figure out a way to
(09:25):
more effectively communicatethis?
And it really came through theISPOR EHR-derived data
suitability checklist that'savailable and published by a
task force from thisprofessional society, and it
really gives you a checklist togo through to ensure whether or
not EHR data is fit for purposefor different research questions
(09:48):
, regulatory decisions and itgoes through how to characterize
the data relevance, reliability, governance, provenance,
relevance, reliability,governance, provenance.
And so this is an extensivepeer-reviewed paper that I wrote
with my team and have nowpublished to be able to go
through every single item in thechecklist to transparently
(10:11):
communicate.
Where does the CHR data comefrom?
How is it governed?
How is it kept secure?
How is it curated so thatpeople can see?
Oh wow, we've never hadanything like this before
outside the United States.
Deborah Borfitz (10:25):
And you just mentioned data curation.
I want to deep dive on thatjust a little bit, because I
think that's part of thedistinctiveness of this approach
here.
Could you talk a little bitabout that, how that was handled
?
I think our readers might, orour listeners might, appreciate
hearing more about that.
Blythe Adamson (10:40):
Right.
So in contrast to other typesof real world data that
sometimes they send physicianssurveys that say, okay, tell me
about a patient you've had withmultiple myeloma, when did they
get diagnosed, what treatmentsdid you give them we are
integrating at the source of theEHR.
(11:01):
So imagine, you know,installing servers and basements
of hospitals and clinics acrossGermany where we really are
getting real-time access andintegrations to all of the
clinical notes, the pathologyreports, all this
non-standardized documents.
This is really where a lot ofthe treasures are in oncology.
(11:23):
It's not in the structuredbilling data.
The richness comes from thesepatients' stories that are often
described in paragraphs ofnotes that these doctors are
writing about them.
So we do this longitudinalcuration.
So when we integrate, we curateand understand the whole
(11:45):
retrospective history of thesepatients' experiences and
prospectively follow themforward to start continuing to
capture their experiences andfollowing them over time.
So, with regular refreshes, atleast once a quarter, we're
updating these data sets so thatwe can have real-time
(12:08):
information, nuanced analysesthat have the clinical details
with provenance, where you cantrace all the way back to the
source of the document, the linenumber.
You know, the place where wecan verify this information is
found.
Speaker 3 (12:25):
Are you enjoying the conversation?
We'd love to hear from you.
Please subscribe to the podcastand give us a rating.
It helps other people find andjoin the conversation.
If you've got speaker or topicideas, we'd love to hear those
too.
Deborah Borfitz (12:45):
You can send them in a podcast review.
So what are the clinicalresearch opportunities you're
most hoping to see pursued withthe integrated data sets, and
have you seen any movement inthose directions?
Absolutely.
Blythe Adamson (12:54):
You know, interestingly it was the FDA who
started asking for some of thisinformation, rather than what
you might expect is the EMA, andit came because, out of the
increasing use of externalcontrol arms, so often for a
single arm clinical trial thatmight be global, they may do a
(13:16):
control arm that comes fromreal-world patient experiences,
and previously many of thosecontrol arms were comparing
real-world patients in theUnited States to a global
clinical trial.
But we know that there's a lotof variability in patient
characteristics and thetreatments used and healthcare
systems between countries.
(13:38):
And so the FDA had said wouldn'tit be better if the real world
external control arms had a morediverse and representative
composition that's morecomparable to the global single
arm trial?
So I'm already seeing theadoption and use by many life
science companies in designingglobal external control arms to
(14:02):
compare them to their globalsingle arm trials, and we're
also seeing this in comparativeeffectiveness studies within
these different countries.
So one of the new initiativesis the joint clinical assessment
in Europe where regulatory andHTA decisions are happening
together.
They're wanting to enumeratewhat people call PICOs the
(14:26):
population interventioncomparator outcomes that are
relevant for all the differentcountries in the European Union,
and the dilemma they often haveis the local standard of care
in Latvia might not be somethingthat's even used in the United
States, and so we're using thesecurated data sets in UK,
(14:47):
germany and Japan to surfacemore patients that have unique
experiences and treatmentpathways that we can learn from
that may be very different fromthe United States, so it's
really exciting to see the quickadoption and range of use cases
that life science companies arenow taking advantage of with
(15:09):
global EHR-derived data.
Deborah Borfitz (15:11):
How do researchers go about accessing
these data sets and getting intothis trusted research
environment to do their analysiswork, and are there any you
know caveats we need to mentionaround that?
Blythe Adamson (15:24):
Right, as I mentioned before, many are
familiar with that federatedmodel where you have to do
separate analyses in everycountry, and it's such a relief
to me now that we've technicallyovercome that limitation.
So trusted research environmentsnow are the way where, as you
(15:45):
mentioned, it's cloud-based,compliant, secure, and this is
where we upload thede-identified or anonymized EHR
oncology data sets so that theycan be pooled together at a
patient level.
So if you imagine a tablethat's describing the
demographic characteristics withone row per patient, you can
(16:08):
now pool together and combineall these different countries
into one table, and that enableslots of different types of more
rigorous methods for analysis.
And it allows you, within theFlatiron Trusted Research
environment for researchers, touse familiar tools like coding
(16:31):
in R or Python within theseenvironments, really being able
to touch and visualize thepatient level data within the
trusted research environment.
And then, once you have theresults from your studies where
you've pooled all the countriestogether, then you can export
the results, so download thoseto your local computer to be
(16:55):
able to use in publications.
Regulatory submissions and thisis the step that we use to
ensure the data security andcompliance with regulations like
GDPR in Europe and AAPI inJapan.
Deborah Borfitz (17:13):
Now are all cancers represented, or will
they be at some point?
Blythe Adamson (17:17):
Yes, we're increasing the number very
rapidly.
So I think right now we haveeight different cancer types
that have been curated and areresearch ready sitting in this
tier trusted researchenvironment where researchers
are now asking questions aboutthe effectiveness of different
treatments.
And the team at Flatiron isquickly adding more and more
(17:42):
cancer types into thisenvironment to do analyses and
then also increasing the cohortsizes of them very quickly.
So we've got a network ofFlatiron sites within each of
the countries.
So, for example, in the UK itis a collection of NHS trusts,
(18:05):
which is where cancer care isdelivered in the UK.
In Germany, for example, wehave a huge network of lots of
small community clinics, whichis more of how cancer care is
delivered in Germany.
So the number of sites areincreasing, patients are
increasing and the cancer typesthat are research-ready to use
(18:27):
is increasing as well.
Deborah Borfitz (18:29):
a great segue to my next question, because
we're talking aboutpatient-level data from four
countries.
So let's get to the topic oftransportability of real-world
evidence across country borders.
That's an area where FlatironHealth has been actively working
, I know, with sponsor companiesto determine when it is
appropriate for one country tokind of borrow information from
(18:51):
another.
Is this the intention, then,with the data sets that you're
developing here, that?
Are they geographically diverseenough, or is the plan to
repeat the harmonizationexercise in yet more nations?
Blythe Adamson (19:05):
Well, when I speak with regulators and HTA
bodies, you know governmentdecision makers all of them have
a preference for local data.
They would rather havereal-world data from their own
country to make a decision, butthe dilemma is it's often not
available, it's too sparse orinsufficient for some reason.
So they do have this dilemma ofwould we rather have nothing or
(19:28):
do we make the assumption thatit's okay to use data from
another country to make adecision about cancer care in
our country?
And it's an assumption that hasnever been tested before.
And so we really had a lot ofbenefit from doing a formal
research collaboration with NICEin the UK, and all these other
(19:51):
countries had so much interestin joining us, and many life
science companies wanted to worktogether that we established a
research consortium called theFORUM, which stands for
Fostering Oncology, rwe Uses andMethods, where together we are
doing a huge set of validationstudies testing whether or not
(20:14):
differences between twocountries' healthcare systems is
so big and impactful that itactually might change how long
people survive with cancer thesame patient.
You know, if you live inGermany versus Denmark, versus
Spain or the US, if you haveidentical patient
characteristics, if you receivethe exact same drug.
(20:36):
Do you survive the same amountof time?
Or are differences in the waynutrition, the environment, how
people are cared for in thehospital do those things
actually make a difference?
And this is what the forum isstudying.
So we are going country bycountry for breast cancer, lung
(20:57):
cancer, heme and looking atdifferent combinations of
countries to test whether or not.
Once you control for thepopulation characteristics being
different, once you select fora specific type of cancer and
drug that people are using andyou control for everything, we
are learning from the experienceof patients in the US,
(21:20):
predicting the outcomes ofpatients in a different country
and then comparing that to theirtrue overall survival in that
country to see whether or not itactually was modified by
differences in the healthcaresystem.
So it's been pretty exciting.
I mean, we've learned that inmany cases that the real world
evidence is transportablebetween countries.
(21:42):
But there are some times whereit's really not appropriate,
like where we found where, whenclinical guidelines between two
countries are wildly different,like everyone in the US starts
with drug A, then B, then C, butin Europe maybe they start with
B, then D, then F.
(22:03):
You know, you can't reallyadjust away when people receive
completely different drugs, andso we're learning about what are
the situations where you reallydo need local data to be able
to inform decisions in thatcountry referenced when life
science companies are submittingto these different countries to
support whether or not thatassumption is appropriate that
(22:38):
it's okay for them to borrowevidence from another country.
Deborah Borfitz (22:41):
Wow, fascinating, interesting and
what a work in progress.
Right, we've covered a lot ofground here, blythe, so let me
end with this when is this allgoing?
Just how big and robust do youimagine the global data
infrastructure will be in, say,another five years?
Blythe Adamson (23:00):
I am really, really optimistic about this,
and when I think about globalhealthcare data infrastructure,
I feel really proud to beworking at a private company
that's solving this, because itreally is bigger than what an
individual country can do bythemselves, and one of the keys
is having harmonized, commondata models across these
(23:23):
populations and also havingrespect for true differences
between these countries and howcancer care is delivered, how
patients are different, howdocumentation styles are
different.
But I anticipate, you know,within five years there's going
to be a much larger, more robustglobal health data ecosystem,
(23:47):
and so much of that, I think, isgetting super powered by large
language models being one of thetools we use now for curation.
In the past, so much of thiswas done manually by cancer
nurses opening up charts,pulling out information.
Now a lot of that is being usedto validate LLMs doing some of
(24:10):
those same tasks for us.
So I think that it's a reallyexciting moment in history to be
a researcher in this field,because the number of questions
that we can answer that werenever possible to understand
before is just exponentiallygrowing.
Deborah Borfitz (24:28):
I hear your enthusiasm.
This has been a trulyenlightening and encouraging
conversation, and you havesucceeded in making it sound
very cool to be a data geek.
So I thank you for being ontoday's show and ask only that
you leave us.
As I told you in advance, Iwould do to leave us with your
favorite data geek quote.
Blythe Adamson (24:49):
Garbage in, garbage out.
I work at Flatiron because it'sthe most beautiful data in the
world that I've ever touched,and so I am so thrilled to be in
a place with high quality datato be able to put out high
quality findings.
Deborah Borfitz (25:07):
You heard it here.
A big thank you to everyone outthere for listening, and if you
are not subscribed to thispodcast yet, please consider
going to Apple Podcasts anddoing so right now, so you don't
miss your monthly dose of newsand perspectives.
You'll be hard-pressed to findanywhere else and, if you're up
for it, I'd also be so verygrateful if you leave a rating
(25:29):
and review while you are there.
One more thing before we go.
If you liked today'sconversation, it is only a
glimpse of what you can expectfrom Scope Europe.
Presenters and panelists,please plan to join us October
14th and 15th in Barcelona whenclinical operations executives
will be exploring the latesttrends in clinical trial
innovation, planning andoperations.
(25:51):
Save an additional 10% off anycurrent rate by using the code
SOT10.
For more information, visitscopesummiteuropecom.
Bye for now.
Popular Podcasts
Fudd Around And Find Out
UConn basketball star Azzi Fudd brings her championship swag to iHeart Women’s Sports with Fudd Around and Find Out, a weekly podcast that takes fans along for the ride as Azzi spends her final year of college trying to reclaim the National Championship and prepare to be a first round WNBA draft pick. Ever wonder what it’s like to be a world-class athlete in the public spotlight while still managing schoolwork, friendships and family time? It’s time to Fudd Around and Find Out!
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy, Jess Hilarious, And Charlamagne Tha God!