April 18, 2024 • 21 mins
While a lot of the news around AI is doom and gloom, the potential for positive innovation in health care offers a hopeful perspective. Hosts Beth Coleman and Rahul Krishnan are joined by University of Toronto experts Christine Allen and Andrew Pinto to talk about the transformative power of AI in health care, from revolutionizing primary care to advancing drug development.
About the hosts:
Beth Coleman is an associate professor at U of T Mississauga’s Institute of Communication, Culture, Information and Technology and the Faculty of Information. She is also a research lead on AI policy and praxis at the Schwartz Reisman Institute for Technology and Society. Coleman authored Reality Was Whatever Happened: Octavia Butler AI and Other Possible Worlds using art and generative AI.
Rahul Krishnan is an assistant professor in U of T’s department of computer science in the Faculty of Arts & Science and department of laboratory medicine and pathobiology in the Temerty Faculty of Medicine. He is a Canada CIFAR Chair at the Vector Institute, a faculty affiliate at the Schwartz Reisman Institute for Technology and Society and a faculty member at the Temerty Centre for AI Research and Education in Medicine (T-CAIREM).
About the guests:
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
- How was your weekend?
(00:01):
- My weekend was good.I was mostly doing a bit
of heavy gardening,so to speak,
- Heavy gardening.
[Laughter]
- And so I'm nursing someaches and bruises all over,
but still recovering
- From the University ofToronto.
I'm Beth Coleman.
- I'm Rahul Krishnan.- This is What Now? AI
(00:25):
- It's one of the thingsthat we don't think about
until we have to,which is our health.
- Yeah. So Rahul, this isreally your wheelhouse in terms
of research, and I'mvery excited that we get
to talk about this today.
A lot of the news aroundAI is very doom and gloom.
The prospects, the things thatare possible in terms of AI
(00:48):
medicine and kind offuture of health.
Really the question we'reasking, is innovation for good?
- So health care is avery multifaceted field,
and in today's episodewe're going
to be diving into talkabout both the human aspects
of health care delivery as wellas AI and drug development.
(01:08):
I spoke with Dr. Andrew Pinto,who's the founder
and the directorof the Upstream Lab,
a research team focused onaddressing social determinants
of health,population health management,
and utilizing data sciencefor proactive care.
Dr. Pinto is a family physicianat St. Mike's Hospital,
an associate professor atU of T's Temerty Faculty of Medicine,
(01:29):
and the Dalla LanaSchool of Public Health.
- And I spoke withChristine Allen,
who's a full professorat U of T's Leslie Dan
Faculty of Pharmacy.
She's a member of thescientific leadership team
of the Acceleration Consortiumat U of T.
Allen is a co-founder andCEO of Intrepid Labs Inc.
A company that is acceleratingpharmacy drug development
(01:50):
through the integrationof AI, automation
and advanced computing.
We caught up with Allenat Intrepid Labs.
So let's get into it.
- Hi, Christine,
calling in from New York City.
It's such a pleasureto get to talk to you.
- Great to see you even if onlyvirtually. Great to see you.
- Okay, so I've got, I don't know,
(02:11):
a million questions for you.
Are you ready? I'm,
- I think I'm ready.
- Good. All right.So we're gonna start
with only the easy ones.
What is drug discovery using AI?
- So I would say to youthat you'll,
you will probably have seen a lot of work
around the integration of AIor use of AI in drug discovery.
(02:33):
So that's identificationof new drug candidates,
new molecules that couldpotentially be used
as therapeutics, right?
Using AI.
And there are other areas
where AI is being quite usefulactually in the entire drug
development pipeline. Iwork in drug formulation,
so it's immediately adjacentto drug discovery in
(02:55):
that drug development pipeline.
So I would say that throughdrug discovery, you identify
this molecule with potential, you know,
therapeutic applications, Iconsider that the passenger,
the formulation isessentially the plane, right?
So we're designing the planethat will take that drug,
take that passenger towhere it needs to go,
so it can exert itstherapeutic effect
(03:17):
in a safe manner.
- So can you talk about
where AI fits into theformulation process?
- When I think of the waythat AI has been used in drug
development to date,
for the most part it'sbeen in drug discovery
and clinical development.
And where I think we really need it,
and certainly that's wherethis new company is focused,
is around drug formulation.
(03:38):
So just to give you asense of the daunting task
that formulation of a drugcan be, for a drug that is
to be administered orally,
there are more than 10billion possible formulations
that could be used fordelivery of that drug.
10 billion possible formulations.
This is not a number
that the human mindcan contemplate, right?
(03:59):
This is a very complicated,
tremendous design space.
And so when welook at the drugs
that have beenformulated orally
and you know thecompositions of matter
that have been used, we'veonly scratched the surface.
We estimate then lessthan 0.01% of
that total design space has been explored.
(04:19):
And so it's thisconcept of using AI
to explore the unexplored.
And what if you know, thatformulation that could
transform the properties
and performance of your drug is one
of those unexplored formulations, right?
And that will then, youknow, take that drug
through clinical development smoothly
and get it to patients faster,
(04:41):
which is really the goal.
- Christine, that's pretty incredible.
I mean, I know you're modest,
but are you one of the firstgroups to bring this process
to drug formulation?
- You know, I have to say we are,
and it's really thanksto great discussions
with Alán Aspuru-Guzik aboutfive years ago, in fact,
I was sitting with himchatting about something else
(05:02):
and I said to him, canyou help me find someone
that would be willing towork with us in this area?
And he said, well, whywouldn't you work with me?
- Intrepid Labs Incorporatedis the first startup
to emerge from theAcceleration Consortium
A U of T institutionalstrategic nitiative led
by Alán Aspuru-Guzik thatuses self-driving labs
to expedite the discovery of materials
(05:23):
and molecules crucialfor a sustainable future.
Allen and Aspuru-Guzik arecombining their expertise in
pharmaceutical sciences, AI
and machine learning to develop new drug
formulations faster.
- You know, he and Istarted chatting about this
and started involvinggraduate students from each
of our labs, which is always fun, right?
(05:43):
Different, you know, studentsfrom different labs working in
very different disciplinesnow working together.
- Yeah.- So very powerful
and they just moved the research forward.
- Can you talk aboutthe Intrepid Lab motto,
"better medicine, fast"?
I mean, how does this kindof work impact a public?
And you know, when wehear "fast," it's hard not
(06:07):
to think about Meta
and other groups who havehistorically been moving
fast and breaking things.
So I'd like to hear
what your confidence isin "fast" in this case.
- It's a great question.So I often say better medicine,
faster and cheaper.
And, you know, just to sharea few stats with you,
(06:28):
it's a little hard to believe in this day
and age with all of thetechnology that we have,
that it's about one in 10 drugs
that actually succeedsin clinical development.
So about 90% of the drugs
that enter clinical development fail.
And it's, you know, those,no, those numbers to me are,
(06:48):
are really hard to believeand, and frankly unacceptable.
And in particular, ifyou've been a patient,
let's say you're a patient,you've been on different lines
of therapy and you'rewatching to see a molecule
or a drug moving through the clinic, and,
and this may be your last best hope,
and that drug fails in a phase two
or a phase three clinical trial.
And when that drug fails, it will vary.
(07:10):
It's very unlikely that it's going
to get another chancein the clinic, right?
Oftentimes the drug gets one chance.
The other thing is, if it's afirst mover, if it's a, a drug
that's first in class, itmay kill an entire class
of drugs for many years.
So that's the first thing thatI would say is that we need
to do better.
The pharmaceuticalindustry, the community,
we need to do better, you know,
(07:31):
across the entire drugdevelopment pipeline in terms of,
you know, addressing thesuccess, the low success rate
or high failure rate,however you wanna put it.
I believe that we don't alwaysmove the best formulations
of drugs into the clinic.
So I mentioned before that drug discovery
or identification of that molecule,
that active pharmaceuticalingredient is that first step.
(07:54):
And many people look atthat and, and think, oh, one
and done, this is the most important step,
the only step that matters.
And in fact, there are about 99 steps
after that to getthat drug across the line such
that it's approved for use in humans.
And one of them that's grosslyunderrated under-resourced
is formulations.
And so, you know, this is
where our technology really comes in.
(08:16):
It's really about identifyingthe best formulation,
a fully optimized formulationthat, of that drug such
that we put our best foot forward when
that drug goes into the clinic.
- So what is your sense about
drug formulation for special populations
(08:36):
where you have very tiny numbers of people
and I don't know if paediatricpatients are relevant in
terms of this question.
So I'll start with the first one.
In terms of special populations,
- That's a great question and,
and we have thought about that.
And, and paediatric formulationsis actually a big challenge.
(08:58):
I would say that becausewe can design formulations
so quickly according to a,a target product profile,
that this is something thatwould be easy for us to do.
If you have an adultformulation for a specific drug,
we could, you know,take from that as a bit
of a starting point, butwith a good understanding of
what we want the target product
profile to be for children.
(09:20):
We could certainly designformulations for children.
In fact, one example thatwe worked on recently in,
in the lab was, was justthat here at Intrepid,
we did have a company, a potential partner
that approached us interestedin formulating a drug
for a paediatric population.
And so we just did a casestudy, which is often something
that we do just to demonstrateour potential to prospective,
(09:43):
you know, partner and, youknow, so we had some target
performance metricsthat we had established,
and just in a few days we wereable to design a formulation
that met that profile.
And that's what I'm
so excited about here isthis is not a daunting task
for us anymore, which it was.
- So, Christine, if I'munderstanding you right,
(10:05):
customization is not a
science fiction future tense.
This is something thatis almost in hand now.
- I mean, I really hope so Beth.
I think that's where weneed to go. Certainly.
I mean, we are, we are all
so different when we talkabout diverse patient
populations, right?
We are also different
and we have to recognize those differences
(10:28):
and understand the impactthat it has on outcomes,
patient outcomes.
You know, this comes back tothough it's not the topic,
you know, it comes back tothe importance of ensuring
that every single modelthat we use in research,
that we understand whatthat model is, who
that model is representative of.
I am as a scientist, I'm accountable.
(10:49):
I think it does comedown to the scientists
or the humans that are inthe loop being accountable.
You know, this is thiswhere it's so important
to have humans, evenwhen we talk about AI in
the loop, right?
To ensure that the research
that we are doing isresponsible, is ethical,
and that we're genuinely, youknow, developing drugs that,
that are going to treat everyone.
(11:11):
You know, it's justso critically important.
- Scale is not the samething as customization.
It's kind of the opposite.
- I don't see them as being separate
because, you know, you can change the look
and feel of your Apple MacBook Pro in
many, many different ways.
(11:31):
And at the end of the day,
it's still a general purposedevice that is capable
of some form of customization, right?
With, you know, machinesthat build these drugs.
I think the trick is toget to a device that,
given an individual'sprofile can then customize
the molecules for them.
And that's the trick. And Ihave no idea how to solve it,
but I think if it existed...- It's a pretty good trick.
(11:52):
- It's a pretty good trick.
- Andrew, as a clinician,
as well as a scientist,you work both in terms
of delivering care as well as thinking of
how we can improve the qualityof care,
improve equity in care
using a varietyof computational tools.
I was wondering if you couldsort of give us an overview
of some of the work that your lab does.
- The team that I lead iscalled the Upstream Lab.
(12:13):
We divide our work into threemain areas that all intersect,
tackling the socialdeterminants of health
using artificial intelligence,
and then also conductingclinical trials of interventions.
We're also doing work on things like
respiratory surveillance.
So can we automate detection
(12:33):
of rising in respiratory illnesscases or unusual patterns?
And then within the AI workthat we're doing,
it's quite a few different areas.
One is kind of tacklingthe administrative burden.
When we started to do thiswork around AI in primary care,
we really knew that wewanted to be directed
(12:55):
by primary care providers
and patients rather than just kind of
what cool things could we do with data.
So we, we actually did a wenational deliberative dialogue
with different groupsfrom across Canada to say,
if we have these AI tools,where should we put them?
And the things that peoplecame up with were things like,
(13:15):
help us with theadministrative burden.
Help us serve more people,
given that weare facing this crisis.
And that's what's kind
of directed our work.
So I think some of the key areas will be,
and some of the work that we're doing is
how do we automate theseadmin tasks that doctors
(13:35):
and other members of the primarycare team are spending way
too much time doing.
Everything from documentation of a visit.
As a family doctor, I spend a lot
of time just putting datain which could be collected
by an intelligent programand actually put in for me.
- Yeah, I mean, I've gotfriends who've already told me
about going to the doctor
(13:56):
and having an AI scribe in the room.
Not everybody is comfortable with kind
of a machine assistant in the room,
but a lot of people areperfectly comfortable with it,
and it seems like a great way
to reduce certain types of labour.
- I agree. I think the realemphasis on AI in health care is
really being able to put, you know, more
(14:17):
of the care back in health care.
I think clinicians over thelast few decades have spent
so much more of their time doing things
that aren't patientfacing, which is the thing
that they were trained to dothat I'm hopeful that tools
and AI can give them backsome of that autonomy
to engage better with patients
and have more meaningfulconnections with their patients
(14:40):
as they guide them through theprocess of delivering care.
You know, from yourinteractions with patients,
what do you find that they mightbe interested in doing
with their own health records?
Like imagine if your patientscould open up an app on their
phone and be able to interact
with the health care data thesame way you open up emails on
the Gmail app, for example.
An interesting thing is thatconsistently patients want
(15:03):
to have access to their data,
but they also need anoverlay of interpretation.
Right now, the way these portalsare working is people just
get access to the raw data.
And it's just notthat helpful sometimes.
And we have had a lot of people
who get worried suddenly they're like,
I saw these lymph nodes grew,
or I saw this red number or something.
(15:24):
So I think with some work between patients
and clinicians, we could help interpret
and help people see this, you don't need
to worry about this, butalso this is the trend.
This is kind of where you're going.
We have a projectright now that's under review
for funding that would proposeintegrating family history
(15:45):
and social data with genomic data
to provide much more tailored suggestions
around prevention of disease.
Let's say a large languagemodel like gets deployed
as a scribe, how can we ensure
that it's performing at parityor across different groups?
Because in some sensethese models are designed
to always produce an output,which means they run the risk
(16:07):
of perhaps failing silently.
Yeah. It's, it's where oneis right now in Canada,
we don't have a greatprocess of regulation
of these types of devices.
They exist in this kindof grey space between
a full medical device
and then just some randomapp that someone's created.
So we need to have, you know,
(16:29):
very urgently we need some sort
of regulatory approach to these.
I think the part of thatregulation would be mandating
and just building withinthe workflow is this type
of assessment of bias,so that any type of app
that's being deployedwithin health settings,
there is this assessment
(16:49):
of is it working the sameacross different groups?
And starting just with, forexample, people who identify
as men, women,
and then moving to more, like,
does it work across different age groups
and different racial groups
and people who are newcomers, people who,
where English isn't thefirst language,
this is all really important because
(17:09):
otherwise there's a risk that these new,
this new technology is actuallygonna worsen disparities
because it could work really,really well for one group
and then leave others kind of behind.
I think in terms of equity considerations,
the most obvious one is that all
of these technologies are driven by data.
And what do you do for the people
(17:31):
where there isn't that data?
So I think a lot about my patients who,
because of social challenges,because of homelessness,
because they'reexperiencing incarceration,
because of other challenges,they are not coming in
and they certainlydon't have a cell phone.
They don't have that data being collected.
So what happens to them
(17:51):
and is there a risk thatthey get left further
and further behind whileyou have a group of people
who are affluent,
well-educated, can affordsmartphones
and have data plans
and their data is beinglinked right into their chart.
So that's like an obviousarea where we have
to always think about, are weleaving certain groups behind?
I think the other piece isaround we always want
(18:15):
to be really transparent
with patients about what'shappening with their data,
who has access to it,what's the use of it for?
We're already at a stagewhere there's lack of trust
because there have beenall these instances
of data breaches and things like that.
So engaging community membersand patients at the start
(18:36):
and forefronting equity considerations,
I think will be a big part of this.
And a lot of patient partners I work with,
they really can see that
and they really like thatthese tools could actually help
to create equity.
They could actually help us so
that we're not just focusingon our energy on the people
who are loud and canadvocate for themselves,
(18:58):
but we're actually usingthese tools to look at all
of the patients in a population
and then focus our attention on the people
who need it the most.
- I think the other piece ofthis is also
in the contextof education at least,
is minimising the tendency to,
you know, trust the machine on everything.
Like you stillwant to teach student's how to
retain their ability to be objective
(19:19):
even when the machine is giventhem, you know, a hundred,
or 500, a thousand correct answers.
- We need to be training ourmedical students and residents
and other trainees, not forthe system we have right now,
but the system that we're gonna have five
or 10 years from now.
- Star Trek when Bones wouldrun the little rod over people
(19:41):
and make different sounds.
So that's kind of like, that'sthe ideal one instrument,
no matter what it is.
Broken bones, broken heart,you know, Vulcan mind meld.
You just run it acrossand it makes its sounds
and then you can diagnose what's going on.
- I think that there have beena lot of attempts to start
(20:02):
to miniaturize a lot of the big machines
that we have in hospitals,ultrasounds, for example.
We now have ultrasounddevices that can be hooked up
to an iPhone that are about assmall as an electric shaver.
And so that's a huge reductionin size that, you know,
eventually hope takes us towards
an all-purpose tricorder device
that tells us what'swrong with the human body.
(20:24):
But to be honest, even if itdoesn't tell us everything
that's wrong, if it getsus to 80% of everything
that could go wrong with a human,
that's already incredibly useful.
- How could artificialintelligence impact health care?
- I'm Alexis Kruntovski
and I'm adouble major in human biology
and book and media studies.
(20:45):
I would say the most promising
or exciting developments inAI in health care would have
to be, I'm personally excitedabout precision probiotic
development in regards
to AI can speed up the formulation process
that pharmaceutical companiesuse towards creating these
probiotics for specificpatient profiles
(21:05):
and kind of giving an alternativemedicine source for people
who haven't been able tofind things that help them
with their disease.
- From the University ofToronto, this is What Now? AI.
- Listen to us whereveryou get your podcasts
and watch us on YouTube.
- How was your weekend?
(00:01):
- My weekend was good.I was mostly doing a bit
of heavy gardening,so to speak,
- Heavy gardening.
[Laughter]
- And so I'm nursing someaches and bruises all over,
but still recovering
- From the University ofToronto.
I'm Beth Coleman.
- I'm Rahul Krishnan.- This is What Now? AI
(00:25):
- It's one of the thingsthat we don't think about
until we have to,which is our health.
- Yeah. So Rahul, this isreally your wheelhouse in terms
of research, and I'mvery excited that we get
to talk about this today.
A lot of the news aroundAI is very doom and gloom.
The prospects, the things thatare possible in terms of AI
(00:48):
medicine and kind offuture of health.
Really the question we'reasking, is innovation for good?
- So health care is avery multifaceted field,
and in today's episodewe're going
to be diving into talkabout both the human aspects
of health care delivery as wellas AI and drug development.
(01:08):
I spoke with Dr. Andrew Pinto,who's the founder
and the directorof the Upstream Lab,
a research team focused onaddressing social determinants
of health,population health management,
and utilizing data sciencefor proactive care.
Dr. Pinto is a family physicianat St. Mike's Hospital,
an associate professor atU of T's Temerty Faculty of Medicine,
(01:29):
and the Dalla LanaSchool of Public Health.
- And I spoke withChristine Allen,
who's a full professorat U of T's Leslie Dan
Faculty of Pharmacy.
She's a member of thescientific leadership team
of the Acceleration Consortiumat U of T.
Allen is a co-founder andCEO of Intrepid Labs Inc.
A company that is acceleratingpharmacy drug development
(01:50):
through the integrationof AI, automation
and advanced computing.
We caught up with Allenat Intrepid Labs.
So let's get into it.
- Hi, Christine,
calling in from New York City.
It's such a pleasureto get to talk to you.
- Great to see you even if onlyvirtually. Great to see you.
- Okay, so I've got, I don't know,
(02:11):
a million questions for you.
Are you ready? I'm,
- I think I'm ready.
- Good. All right.So we're gonna start
with only the easy ones.
What is drug discovery using AI?
- So I would say to youthat you'll,
you will probably have seen a lot of work
around the integration of AIor use of AI in drug discovery.
(02:33):
So that's identificationof new drug candidates,
new molecules that couldpotentially be used
as therapeutics, right?
Using AI.
And there are other areas
where AI is being quite usefulactually in the entire drug
development pipeline. Iwork in drug formulation,
so it's immediately adjacentto drug discovery in
(02:55):
that drug development pipeline.
So I would say that throughdrug discovery, you identify
this molecule with potential, you know,
therapeutic applications, Iconsider that the passenger,
the formulation isessentially the plane, right?
So we're designing the planethat will take that drug,
take that passenger towhere it needs to go,
so it can exert itstherapeutic effect
(03:17):
in a safe manner.
- So can you talk about
where AI fits into theformulation process?
- When I think of the waythat AI has been used in drug
development to date,
for the most part it'sbeen in drug discovery
and clinical development.
And where I think we really need it,
and certainly that's wherethis new company is focused,
is around drug formulation.
(03:38):
So just to give you asense of the daunting task
that formulation of a drugcan be, for a drug that is
to be administered orally,
there are more than 10billion possible formulations
that could be used fordelivery of that drug.
10 billion possible formulations.
This is not a number
that the human mindcan contemplate, right?
(03:59):
This is a very complicated,
tremendous design space.
And so when welook at the drugs
that have beenformulated orally
and you know thecompositions of matter
that have been used, we'veonly scratched the surface.
We estimate then lessthan 0.01% of
that total design space has been explored.
(04:19):
And so it's thisconcept of using AI
to explore the unexplored.
And what if you know, thatformulation that could
transform the properties
and performance of your drug is one
of those unexplored formulations, right?
And that will then, youknow, take that drug
through clinical development smoothly
and get it to patients faster,
(04:41):
which is really the goal.
- Christine, that's pretty incredible.
I mean, I know you're modest,
but are you one of the firstgroups to bring this process
to drug formulation?
- You know, I have to say we are,
and it's really thanksto great discussions
with Alán Aspuru-Guzik aboutfive years ago, in fact,
I was sitting with himchatting about something else
(05:02):
and I said to him, canyou help me find someone
that would be willing towork with us in this area?
And he said, well, whywouldn't you work with me?
- Intrepid Labs Incorporatedis the first startup
to emerge from theAcceleration Consortium
A U of T institutionalstrategic nitiative led
by Alán Aspuru-Guzik thatuses self-driving labs
to expedite the discovery of materials
(05:23):
and molecules crucialfor a sustainable future.
Allen and Aspuru-Guzik arecombining their expertise in
pharmaceutical sciences, AI
and machine learning to develop new drug
formulations faster.
- You know, he and Istarted chatting about this
and started involvinggraduate students from each
of our labs, which is always fun, right?
(05:43):
Different, you know, studentsfrom different labs working in
very different disciplinesnow working together.
- Yeah.- So very powerful
and they just moved the research forward.
- Can you talk aboutthe Intrepid Lab motto,
"better medicine, fast"?
I mean, how does this kindof work impact a public?
And you know, when wehear "fast," it's hard not
(06:07):
to think about Meta
and other groups who havehistorically been moving
fast and breaking things.
So I'd like to hear
what your confidence isin "fast" in this case.
- It's a great question.So I often say better medicine,
faster and cheaper.
And, you know, just to sharea few stats with you,
(06:28):
it's a little hard to believe in this day
and age with all of thetechnology that we have,
that it's about one in 10 drugs
that actually succeedsin clinical development.
So about 90% of the drugs
that enter clinical development fail.
And it's, you know, those,no, those numbers to me are,
(06:48):
are really hard to believeand, and frankly unacceptable.
And in particular, ifyou've been a patient,
let's say you're a patient,you've been on different lines
of therapy and you'rewatching to see a molecule
or a drug moving through the clinic, and,
and this may be your last best hope,
and that drug fails in a phase two
or a phase three clinical trial.
And when that drug fails, it will vary.
(07:10):
It's very unlikely that it's going
to get another chancein the clinic, right?
Oftentimes the drug gets one chance.
The other thing is, if it's afirst mover, if it's a, a drug
that's first in class, itmay kill an entire class
of drugs for many years.
So that's the first thing thatI would say is that we need
to do better.
The pharmaceuticalindustry, the community,
we need to do better, you know,
(07:31):
across the entire drugdevelopment pipeline in terms of,
you know, addressing thesuccess, the low success rate
or high failure rate,however you wanna put it.
I believe that we don't alwaysmove the best formulations
of drugs into the clinic.
So I mentioned before that drug discovery
or identification of that molecule,
that active pharmaceuticalingredient is that first step.
(07:54):
And many people look atthat and, and think, oh, one
and done, this is the most important step,
the only step that matters.
And in fact, there are about 99 steps
after that to getthat drug across the line such
that it's approved for use in humans.
And one of them that's grosslyunderrated under-resourced
is formulations.
And so, you know, this is
where our technology really comes in.
(08:16):
It's really about identifyingthe best formulation,
a fully optimized formulationthat, of that drug such
that we put our best foot forward when
that drug goes into the clinic.
- So what is your sense about
drug formulation for special populations
(08:36):
where you have very tiny numbers of people
and I don't know if paediatricpatients are relevant in
terms of this question.
So I'll start with the first one.
In terms of special populations,
- That's a great question and,
and we have thought about that.
And, and paediatric formulationsis actually a big challenge.
(08:58):
I would say that becausewe can design formulations
so quickly according to a,a target product profile,
that this is something thatwould be easy for us to do.
If you have an adultformulation for a specific drug,
we could, you know,take from that as a bit
of a starting point, butwith a good understanding of
what we want the target product
profile to be for children.
(09:20):
We could certainly designformulations for children.
In fact, one example thatwe worked on recently in,
in the lab was, was justthat here at Intrepid,
we did have a company, a potential partner
that approached us interestedin formulating a drug
for a paediatric population.
And so we just did a casestudy, which is often something
that we do just to demonstrateour potential to prospective,
(09:43):
you know, partner and, youknow, so we had some target
performance metricsthat we had established,
and just in a few days we wereable to design a formulation
that met that profile.
And that's what I'm
so excited about here isthis is not a daunting task
for us anymore, which it was.
- So, Christine, if I'munderstanding you right,
(10:05):
customization is not a
science fiction future tense.
This is something thatis almost in hand now.
- I mean, I really hope so Beth.
I think that's where weneed to go. Certainly.
I mean, we are, we are all
so different when we talkabout diverse patient
populations, right?
We are also different
and we have to recognize those differences
(10:28):
and understand the impactthat it has on outcomes,
patient outcomes.
You know, this comes back tothough it's not the topic,
you know, it comes back tothe importance of ensuring
that every single modelthat we use in research,
that we understand whatthat model is, who
that model is representative of.
I am as a scientist, I'm accountable.
(10:49):
I think it does comedown to the scientists
or the humans that are inthe loop being accountable.
You know, this is thiswhere it's so important
to have humans, evenwhen we talk about AI in
the loop, right?
To ensure that the research
that we are doing isresponsible, is ethical,
and that we're genuinely, youknow, developing drugs that,
that are going to treat everyone.
(11:11):
You know, it's justso critically important.
- Scale is not the samething as customization.
It's kind of the opposite.
- I don't see them as being separate
because, you know, you can change the look
and feel of your Apple MacBook Pro in
many, many different ways.
(11:31):
And at the end of the day,
it's still a general purposedevice that is capable
of some form of customization, right?
With, you know, machinesthat build these drugs.
I think the trick is toget to a device that,
given an individual'sprofile can then customize
the molecules for them.
And that's the trick. And Ihave no idea how to solve it,
but I think if it existed...- It's a pretty good trick.
(11:52):
- It's a pretty good trick.
- Andrew, as a clinician,
as well as a scientist,you work both in terms
of delivering care as well as thinking of
how we can improve the qualityof care,
improve equity in care
using a varietyof computational tools.
I was wondering if you couldsort of give us an overview
of some of the work that your lab does.
- The team that I lead iscalled the Upstream Lab.
(12:13):
We divide our work into threemain areas that all intersect,
tackling the socialdeterminants of health
using artificial intelligence,
and then also conductingclinical trials of interventions.
We're also doing work on things like
respiratory surveillance.
So can we automate detection
(12:33):
of rising in respiratory illnesscases or unusual patterns?
And then within the AI workthat we're doing,
it's quite a few different areas.
One is kind of tacklingthe administrative burden.
When we started to do thiswork around AI in primary care,
we really knew that wewanted to be directed
(12:55):
by primary care providers
and patients rather than just kind of
what cool things could we do with data.
So we, we actually did a wenational deliberative dialogue
with different groupsfrom across Canada to say,
if we have these AI tools,where should we put them?
And the things that peoplecame up with were things like,
(13:15):
help us with theadministrative burden.
Help us serve more people,
given that weare facing this crisis.
And that's what's kind
of directed our work.
So I think some of the key areas will be,
and some of the work that we're doing is
how do we automate theseadmin tasks that doctors
(13:35):
and other members of the primarycare team are spending way
too much time doing.
Everything from documentation of a visit.
As a family doctor, I spend a lot
of time just putting datain which could be collected
by an intelligent programand actually put in for me.
- Yeah, I mean, I've gotfriends who've already told me
about going to the doctor
(13:56):
and having an AI scribe in the room.
Not everybody is comfortable with kind
of a machine assistant in the room,
but a lot of people areperfectly comfortable with it,
and it seems like a great way
to reduce certain types of labour.
- I agree. I think the realemphasis on AI in health care is
really being able to put, you know, more
(14:17):
of the care back in health care.
I think clinicians over thelast few decades have spent
so much more of their time doing things
that aren't patientfacing, which is the thing
that they were trained to dothat I'm hopeful that tools
and AI can give them backsome of that autonomy
to engage better with patients
and have more meaningfulconnections with their patients
(14:40):
as they guide them through theprocess of delivering care.
You know, from yourinteractions with patients,
what do you find that they mightbe interested in doing
with their own health records?
Like imagine if your patientscould open up an app on their
phone and be able to interact
with the health care data thesame way you open up emails on
the Gmail app, for example.
An interesting thing is thatconsistently patients want
(15:03):
to have access to their data,
but they also need anoverlay of interpretation.
Right now, the way these portalsare working is people just
get access to the raw data.
And it's just notthat helpful sometimes.
And we have had a lot of people
who get worried suddenly they're like,
I saw these lymph nodes grew,
or I saw this red number or something.
(15:24):
So I think with some work between patients
and clinicians, we could help interpret
and help people see this, you don't need
to worry about this, butalso this is the trend.
This is kind of where you're going.
We have a projectright now that's under review
for funding that would proposeintegrating family history
(15:45):
and social data with genomic data
to provide much more tailored suggestions
around prevention of disease.
Let's say a large languagemodel like gets deployed
as a scribe, how can we ensure
that it's performing at parityor across different groups?
Because in some sensethese models are designed
to always produce an output,which means they run the risk
(16:07):
of perhaps failing silently.
Yeah. It's, it's where oneis right now in Canada,
we don't have a greatprocess of regulation
of these types of devices.
They exist in this kindof grey space between
a full medical device
and then just some randomapp that someone's created.
So we need to have, you know,
(16:29):
very urgently we need some sort
of regulatory approach to these.
I think the part of thatregulation would be mandating
and just building withinthe workflow is this type
of assessment of bias,so that any type of app
that's being deployedwithin health settings,
there is this assessment
(16:49):
of is it working the sameacross different groups?
And starting just with, forexample, people who identify
as men, women,
and then moving to more, like,
does it work across different age groups
and different racial groups
and people who are newcomers, people who,
where English isn't thefirst language,
this is all really important because
(17:09):
otherwise there's a risk that these new,
this new technology is actuallygonna worsen disparities
because it could work really,really well for one group
and then leave others kind of behind.
I think in terms of equity considerations,
the most obvious one is that all
of these technologies are driven by data.
And what do you do for the people
(17:31):
where there isn't that data?
So I think a lot about my patients who,
because of social challenges,because of homelessness,
because they'reexperiencing incarceration,
because of other challenges,they are not coming in
and they certainlydon't have a cell phone.
They don't have that data being collected.
So what happens to them
(17:51):
and is there a risk thatthey get left further
and further behind whileyou have a group of people
who are affluent,
well-educated, can affordsmartphones
and have data plans
and their data is beinglinked right into their chart.
So that's like an obviousarea where we have
to always think about, are weleaving certain groups behind?
I think the other piece isaround we always want
(18:15):
to be really transparent
with patients about what'shappening with their data,
who has access to it,what's the use of it for?
We're already at a stagewhere there's lack of trust
because there have beenall these instances
of data breaches and things like that.
So engaging community membersand patients at the start
(18:36):
and forefronting equity considerations,
I think will be a big part of this.
And a lot of patient partners I work with,
they really can see that
and they really like thatthese tools could actually help
to create equity.
They could actually help us so
that we're not just focusingon our energy on the people
who are loud and canadvocate for themselves,
(18:58):
but we're actually usingthese tools to look at all
of the patients in a population
and then focus our attention on the people
who need it the most.
- I think the other piece ofthis is also
in the contextof education at least,
is minimising the tendency to,
you know, trust the machine on everything.
Like you stillwant to teach student's how to
retain their ability to be objective
(19:19):
even when the machine is giventhem, you know, a hundred,
or 500, a thousand correct answers.
- We need to be training ourmedical students and residents
and other trainees, not forthe system we have right now,
but the system that we're gonna have five
or 10 years from now.
- Star Trek when Bones wouldrun the little rod over people
(19:41):
and make different sounds.
So that's kind of like, that'sthe ideal one instrument,
no matter what it is.
Broken bones, broken heart,you know, Vulcan mind meld.
You just run it acrossand it makes its sounds
and then you can diagnose what's going on.
- I think that there have beena lot of attempts to start
(20:02):
to miniaturize a lot of the big machines
that we have in hospitals,ultrasounds, for example.
We now have ultrasounddevices that can be hooked up
to an iPhone that are about assmall as an electric shaver.
And so that's a huge reductionin size that, you know,
eventually hope takes us towards
an all-purpose tricorder device
that tells us what'swrong with the human body.
(20:24):
But to be honest, even if itdoesn't tell us everything
that's wrong, if it getsus to 80% of everything
that could go wrong with a human,
that's already incredibly useful.
- How could artificialintelligence impact health care?
- I'm Alexis Kruntovski
and I'm adouble major in human biology
and book and media studies.
(20:45):
I would say the most promising
or exciting developments inAI in health care would have
to be, I'm personally excitedabout precision probiotic
development in regards
to AI can speed up the formulation process
that pharmaceutical companiesuse towards creating these
probiotics for specificpatient profiles
(21:05):
and kind of giving an alternativemedicine source for people
who haven't been able tofind things that help them
with their disease.
- From the University ofToronto, this is What Now? AI.
- Listen to us whereveryou get your podcasts
and watch us on YouTube.
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com