0:13

Hello and welcome to the Data

0:13

Science Innovation Driving

0:18

Alzheimer's Disease Insights.

0:18

I am Jeana Konstantakopolous,

0:22

the senior Director of Partner

0:22

Engagement here at Matter and Matter is a

0:26

healthcare technology

0:26

incubator and innovation hub.

0:29

Built on the belief that collaboration

0:29

between entrepreneurs and industry

0:33

leaders is the best way to

0:33

develop healthcare solutions.

0:36

Our mission is to accelerate the

0:36

pace of change of healthcare,

0:39

and we do three things in service of this

0:39

mission. First, we incubate startups.

0:44

Since we launched eight years ago, we've worked with more than 800 companies

0:46

that range from very early growth

0:50

stage startups to larger companies,

0:53

and we offer them a suite of services

0:53

to help at every stage of their

0:56

development. Our member companies have raised more

0:57

than $5 billion to fuel their growth.

1:02

Second, we work with larger

1:02

organizations such as health systems,

1:05

life science companies, payers, foundations to strengthen

1:07

their innovation capacity,

1:11

and we help them find value in emerging

1:11

technology solutions by unlocking full

1:15

potential of both their internal

1:15

innovators and then creating more

1:19

human-centered healthcare experiences

1:19

through system level collaboration. And

1:24

third, we're a nexus for people who are

1:24

passionate about healthcare innovation.

1:28

We bring people together to be inspired

1:28

to learn and to connect with each other.

1:33

We produce a lot of programs

1:33

like this one today,

1:35

including large scale events for the

1:35

broader community and small forums

1:39

exclusively for our members and partners. Today's event is being conducted

1:42

alongside our Brain Health Innovation

1:47

Challenge, which you'll hear a little bit more about

1:47

at the end of this program that we're

1:51

producing with support from the

1:51

Lundbeck US Charitable Fund.

1:56

It is an independently managed

1:56

nonprofit 5 0 1 C three that is

2:01

committed to the responsibility

2:01

and appropriate support of programs

2:06

about restoring brain health. The Lundbeck US Charitable Fund

2:08

is wholly owned by Lundbeck,

2:12

a global pharmaceutical company

2:12

specializing in brain disease.

2:16

But for more than 70 years, Lundbeck has been at the forefront

2:18

of neuroscience research tirelessly

2:22

dedicated to restoring brain health so

2:22

that every person can be their best.

2:27

So for today, for years,

2:29

data analytics has been used

2:29

in healthcare to fuel faster,

2:33

accurate diagnose to

2:33

inform decision-making,

2:37

personalized treatment, improve

2:37

patient care and outcomes,

2:41

lower costs and more. But with the recent advances that we're

2:43

seeing with big data and generative

2:48

artificial intelligence, more organizations are exploring

2:50

how these new ways can take

2:55

modern data science tools to address

2:55

persistent healthcare challenges.

3:00

So talking about challenges, one of the key challenges right now

3:02

in advancing care for our growing

3:07

population of older people is living

3:07

with Alzheimer's disease and the related

3:12

dementias. And we're seeing that there

3:13

are a wide range of disparate

3:18

sources of raw data, including

3:18

electronic health records,

3:21

personal health records, patient

3:21

portals, health related smartphones,

3:25

wearables, and lots of

3:25

unstructured data out there.

3:29

And the question is how can we

3:29

gain meaningful insights? Well,

3:33

hopefully our panel today

3:33

will help us to dig into that.

3:37

Today we're joined by Mary Furlong,

3:37

a leader in the longevity market.

3:42

Elizabeth Powers the Vice President and

3:42

general manager of US Regulatory Science

3:46

and Study Innovation, and I

3:46

Q V I A and Ryan Urbanowicz,

3:51

research scientist, computational

3:51

biomedicine at Cedar,

3:54

Cedar-Sinai Medical Center, and the co-lead of Tech ID and Training

3:56

Corps at Penn AI Tech in the A two

4:01

collective. And our conversation today

4:01

is going to just dig into this topic.

4:05

So with that, hello everyone.

4:11

Hey, how's it going? Hi, Ryan. Hi, Elizabeth.

4:16

I'm good. Great to have you guys join us today.

4:20

Before we really kind of

4:20

dig into the topic at hand,

4:23

I'm going to go around the table and

4:23

have you tell me a little bit more about

4:27

yourselves and how you've

4:27

come to look both at

4:32

the growing older population in the

4:32

US brain health as it relates to

4:37

Alzheimer's disease and dementia. And

4:37

then this question of data. All right.

4:41

So I'm going to start with you, Mary, if you can tell me a little

4:42

bit about you and your role.

4:46

Well, I've been at this a long

4:46

time. I'm a serial entrepreneur,

4:50

have started three companies

4:50

and I think have raised

4:55

about 250 million in corporate

4:55

sponsorships and venture

4:59

financing for startups that are

4:59

building companies in the longevity

5:04

market. I produced the

5:04

Longevity Venture Summit,

5:09

which I've done for about 20 years,

5:09

and the Washington Innovation Summit.

5:14

And I have a podcast called

5:14

Longevity Deal Talk in terms of brain

5:18

science. I'm an advisor to the

5:18

Canadian Brain Health CBI Group.

5:24

I've been part of Poit

5:24

Science since the beginning,

5:27

and I'm recently judging

5:27

the UK competition for

5:32

business plans related

5:32

to research and dementia.

5:37

Thank you. I'm sure your a wide array

5:38

of experience here will be of

5:43

great use today in our

5:43

conversation. Elizabeth,

5:45

can you tell me a little bit about yourself? Yes. So as you said,

5:49

I'm vice president and general manager

5:49

of a group within I Q V I A called

5:53

Regulatory Science and Study Innovation.

5:56

Our primary mission is to find ways

6:01

of unlocking access to

6:01

clinically rich data,

6:06

including new data sources, some traditional data sources

6:07

like electronic medical records,

6:11

but also new data sources

6:11

like wearables and so forth,

6:15

big data sources, small data sources,

6:19

and really figuring out how to use

6:19

those data sources in a way that has

6:24

scientific credibility and rigor.

6:28

And so we are in the early stages of

6:33

building out some new research

6:33

networks that focus on C N Ss,

6:38

including Alzheimer's, and obviously there are a

6:40

lot of exciting treatments

6:45

coming to market for

6:45

Alzheimer's and dementia.

6:48

And with that comes a bolus of research sponsored

6:53

by pharma companies,

6:56

and we are heavily involved in

6:56

various efforts around that.

7:01

So really happy to be here with you today. Thank you.

7:04

We're excited to have you and certainly

7:04

have someone who's riding the wave,

7:08

so to speak, of what's happening

7:08

out there in the ecosystem. Last,

7:12

but certainly not least, Ryan, we have you and I think that

7:14

you are our resident data guru,

7:18

so if you would please share a little

7:18

bit about yourself, that would be great.

7:22

Sure. I'm Ryan Urbanowicz. I'm currently an assistant professor at

7:24

Cedar Sign and Medical Center as well as

7:28

an adjunct at UPenn. I run the Herbs Lab.

7:31

We do research in the development

7:31

of machine learning and artificial

7:34

intelligence methods as well as our

7:34

application to a variety of biomedical

7:39

target data points or objectives.

7:42

And our lab specializes in development

7:42

of automated machine learning tools as

7:47

well as interpretable

7:47

rule-based machine learning.

7:49

So I'm very much coming at this

7:49

from the computer science side,

7:52

data analytics side, and I've gotten involved in Alzheimer's

7:54

research in particular over the last

7:57

couple of years through the Penn AI Tech

7:57

and a two collaborative that's funding

8:03

research grants for technologies and ai,

8:06

especially applied to Alzheimer's

8:06

and dementia and other aging

8:10

issues. And I approach data kind of agnostically

8:12

because I'm involved in a lot of

8:16

domains and also because many of the

8:16

challenges and questions and data

8:21

science are kind of

8:21

universal and generalizable.

8:26

But there's a lot of things that I

8:26

think about in terms of Alzheimer's

8:31

that are unique in terms of data

8:31

quality, what features to collect,

8:36

things like that. So anyway. Yep.

8:39

Great. I think we have the perfect set

8:39

of perspectives to get our conversation

8:44

underway here. So let's kind

8:44

of pivot here, which is,

8:47

I'm going to start with you, Elizabeth,

8:50

which is Alzheimer's disease and related

8:50

dementias are on the rise. As we know,

8:55

our older population is growing and

8:55

because Alzheimer's disease and dementia

8:59

tend to have a later

8:59

onset for most people,

9:02

it kind of goes in hand with the

9:02

prevalence of these cognitive disorders

9:07

increasing. It means that we're looking at more

9:08

people encountering these diseases.

9:12

And what are you hearing about some of

9:12

the challenges in the marketplace as it

9:16

relates to this? I think

9:21

I'll kind of organize this along

9:21

the patient journey, if you will.

9:27

I think first of all, there is patient and caregiver fear

9:33

and uncertainty about what is

9:38

happening with what is happening to

9:38

me, what is happening to my parents,

9:44

my aunt, my grandparent,

9:44

my brother, my sister,

9:49

and deep fear about knowing

9:49

an answer because of the

9:54

implications of that from

9:54

a caregiver perspective.

9:57

Then once someone is

9:57

within seeing a clinician

10:02

about this, we're not.

10:06

So I was just sitting here thinking, I've been working in Alzheimer's

10:08

for almost 20 years now,

10:12

and I think we're still

10:12

seeing very inconsistent

10:17

practices. It's not like frankly,

10:20

oncology where they're

10:20

relatively clear lines of

10:25

care. There's not even clear diagnoses.

10:30

And that is still the case. Now,

10:33

my own personal hope is that over the

10:33

next five to 10 years with new therapies

10:37

coming, it will overcome both

10:42

physician, patient and caregiver

10:44

resistance to a diagnosis

10:49

because there are treatments oftentimes

10:49

when you're in therapies where there

10:53

aren't really meaningful treatments,

10:56

it can be very difficult

10:56

to get to a diagnosis.

11:02

And then there's just record keeping.

11:04

Actually what gets put into

11:04

the EMRs is very different

11:09

from physician to physician. And then the last thing

11:13

I'll say is then there's the

11:18

burden on the caregiver,

11:20

whether that is someone in a nursing

11:25

facility, step down,

11:25

step up care facility,

11:30

assisted living facility, or just in a home with family,

11:37

there has to be a limit on the

11:37

burden that's put on the caregiver,

11:41

and that's something that really has to

11:41

be taken into strong consideration both

11:46

in terms of treatment and care and in

11:46

terms of data collection for research.

11:51

So I'll just hit pause there. I think that's a rich ground for

11:54

us, I think to play in. But Mary,

11:58

I'm going to pivot to you next. I know that longevity is something

11:59

that you think a lot about and kind of

12:04

this growing population of

12:04

concern around brain health,

12:09

not just by the way, do I

12:09

have OID disease or dementia,

12:13

but preventatively, what can I

12:13

be thinking about to do that?

12:16

And kind of having a

12:16

community of peers focused as

12:21

well. What's kind of your take on

12:21

what's happening in the marketplace?

12:25

I thought I might size market. So the longevity market's

12:27

an 8.3 trillion market,

12:32

and then there's a lot

12:32

of riches in the niches.

12:34

So if you look at the boomers,

12:39

they are at the top end 77,

12:42

so in three years they're going to be 80,

12:46

and then you've got 20 years

12:50

of older adults coming behind

12:50

them. So it's a huge market. Now,

12:56

the opportunity for innovation

12:56

in the home, in the care setting,

13:01

in the adult day setting, in senior housing communities and

13:03

in places with dementia wings,

13:09

that's really important to look at.

13:09

But we're just at the very beginning.

13:13

I mean, more people watch Wheel of

13:13

Fortune then that's their

13:18

cognitive fitness. And so if

13:18

you take an issue like driving,

13:22

which I'm very concerned about

13:22

right now because you look at

13:27

the number of people who are not going

13:27

to be able to drive in the next 10 years,

13:32

and we're not prepared for that in

13:32

terms of accessing resources in the

13:36

home. So some of the analysts think Uber

13:37

Health is one of the most important new

13:42

brands or brands out there

13:42

that could play a role,

13:46

but lighting can play a

13:46

role, pharma can play a role,

13:51

and there's a huge staff shortage.

13:54

So there's really got to be

13:54

brand new models for how we

13:59

find train and retain caregivers.

14:07

It certainly, no, there's not.

14:10

I think that maybe that's part of this,

14:10

right, which is that there's so much,

14:13

not just from a standpoint of talking

14:13

about the number of people that are

14:18

potentially impacted by this, but the myriad of kind

14:19

of concerns that they're

14:24

starting to have to consider not just

14:29

healthcare and data, but these larger access issues

14:30

that you pointed to that all

14:35

have a role to play in things

14:35

like diagnosis or care.

14:39

So really interesting. I think I want to take some of

14:41

these things and maybe Ryan,

14:44

I'll come to you next, which is this notion of data

14:46

we heard from Elizabeth about

14:50

data and relating to this kind of

14:50

group being in lots of different

14:55

places. As someone who spends their days

14:57

looking at these troves of data,

15:02

what do you think the current

15:02

state of certainly Alzheimer's

15:07

data, but maybe broad more broadly,

15:09

this older adult and population level

15:09

data that could start to have a play

15:14

into things, what does that look like to you? I think like a lot of biomedical domains,

15:20

the data is distributed,

15:20

it's siloed, it's messy.

15:26

We're still figuring out in many

15:26

cases, what are the right variables?

15:29

What is the right information

15:29

to collect on patients?

15:32

What do we need to be

15:32

collecting in order to target

15:36

care or to monitor for care?

15:40

So there's a lot of unanswered questions

15:40

in terms of just knowing what to gather

15:45

correctly, let alone how to do it well,

15:45

and thinking ahead is really important.

15:50

I think that's already been brought up. We need to think about what the future

15:52

needs are going to be and make our data

15:56

collection systems adaptable so that

15:56

as our understanding of these issues

16:01

changes, so can be the way that we collect our

16:01

data and the way we utilize our data

16:06

to translate it back into patient

16:06

care or help or whatever it is that we

16:11

want to focus on. And maybe as a way to

16:14

take another kind of step

16:19

back here, when you're looking

16:19

day-to-day at these datasets,

16:23

are you only thinking about things like

16:23

the older adult and Alzheimer's disease,

16:27

or are you looking at other kind of

16:27

patterns and approaches that you're

16:32

taking elsewhere? Yeah, no, I wish I could say I'm

16:34

entirely focused on Alzheimer's disease,

16:37

but no, I think very broadly about a lot

16:37

of biomedical outcomes and work

16:42

with a number of data

16:42

types. One small example,

16:47

I was involved in a clinical trial for,

16:54

I forget, I'm getting my biomedical

16:54

outcomes mixed up. But anyway,

16:58

in this project they had clinical

16:58

trial data from multiple sites and just

17:02

harmonizing the data from

17:02

these pretty well structured

17:07

programs was an absolute

17:07

nightmare. It took two,

17:11

three years just to bring this data

17:11

together before we can even really analyze

17:15

it, and just this as a reflection of

17:15

what the current state tends to

17:20

be in the medical field in terms of

17:20

being on the same page from the get

17:25

go and how we're going to

17:25

collect information or how

17:29

that's already out there, how do we bring it together and leverage

17:30

it in a way that is reliable and

17:34

trustworthy. Ryan, if I can just

17:36

dovetail on that. I mean,

17:40

you just said that you were involved

17:40

in a clinical trial and there was an

17:43

enormous amount of effort

17:43

to harmonize the data across

17:48

sites. That's under the best of circumstances

17:49

where sites are entering things into a

17:54

pre-designed E C R F case

17:54

report form to feed into

17:59

an electronic data capture system. When you're working with

18:01

real world data where

18:07

even if the E M R is epic, every

18:07

system is configured differently,

18:12

much less than adding in data

18:18

that is occurring from outside

18:18

the actual side of care,

18:24

but getting a sense of is a person's activity level,

18:32

what's happening with certain biometric,

18:36

biometric data, heart rate, sweat anxiety,

18:41

these are all things that are

18:41

relevant to patients, people,

18:46

people with dementia, and

18:50

that data gets very hard to

18:50

integrate and is incredibly messy.

18:55

Absolutely. And one other

18:55

quick sort of side note,

19:00

in addition to when we're thinking

19:00

about collecting new data,

19:04

another big point that might be worth

19:04

discussing more is thinking about patient

19:08

privacy concerns. We want to be collecting this data

19:10

so that we can make best use of it,

19:14

but also how do we do that

19:14

without patients feeling

19:19

or giving away their personal freedoms? There's actually a question

19:24

in the chat about passive data

19:29

collection. So in I Q v's business, we're really just

19:35

beginning to see large scale studies,

19:40

real world studies come through

19:40

where sponsors are hoping

19:45

to have passive data collection

19:45

through a wearable. I mean,

19:50

that's among what it really

19:50

has to be in a certain way,

19:55

but actually having a

19:55

validated tool for that.

20:02

In my experience, it's still very much early days and

20:12

in our view, it is critical for research and

20:13

dementia and Alzheimer's to have this

20:18

for a whole range of reasons

20:18

tied to things I've already said.

20:24

But I think we're still some years

20:29

at least two to three, if not five to 10,

20:31

from really having the sophistication of

20:36

tools and sensors to be able to do this

20:36

easily and without burdening the patient

20:41

or the caregiver. It's interesting because I think we

20:43

do this in healthcare lot, Elizabeth,

20:47

which is one of the first

20:47

questions I asked you.

20:50

You started with patient experience

20:50

and caregiver experience,

20:53

which is that individual

20:53

level of healthcare,

20:56

which is so intimate and so personable,

20:59

but at the same point to make

20:59

that a really meaningful kind of

21:04

evidence-based experience for them.

21:04

We depend on population level data,

21:09

we depend on the rollup of all

21:09

of that to enable that personal

21:13

experience. So maybe

21:13

Ryan, a question for you.

21:18

As we talk about these

21:18

disparate realms of data

21:22

development, production

21:22

living, little living spots,

21:27

how do you think about population

21:27

level data and unlocking some of those

21:31

insights where either for this older

21:31

population or you thinking that we

21:36

need to look and what's your

21:36

experience as someone who

21:41

is looking at things

21:41

like AI to uncover some

21:46

of those things? So there's already an incredible

21:48

wealth of tools out there for

21:53

machine learning and ai.

21:53

There's some great advancements.

21:56

Obviously that's still an

21:56

evolving field as well.

21:59

There's a billion unanswered questions, but in terms of thinking about

22:01

analyzing this kind of data,

22:04

the first thing that I worry about that

22:04

sometimes can get overlooked I think is

22:10

the data quality and collection. And that is absolutely essential. You

22:13

might've heard the phrase garbage in,

22:17

garbage out, right? It is tempting to have this magical

22:19

thinking about machine learning and ai.

22:23

It's like, I'll just pass it to the

22:23

tools and then I'll get something good.

22:27

And the real meat of all of

22:27

this is always going to be

22:32

what variables do I collect? What is the quality of the data I'm

22:34

collecting so that I can leverage these

22:38

awesome new tools to really make

22:38

the boat most out of the data,

22:42

but at the same time not

22:42

fall into pitfalls like

22:47

bias, right?

22:47

Bias is a huge one.

22:49

Making sure predictive

22:49

algorithms are fair, that

22:55

they allow fairness in

22:55

their decision-making

23:00

information we glean from

23:00

these tools and models. One,

23:04

in terms of methodologies and making

23:04

use of the data. I mentioned earlier,

23:08

one of the areas of research I'm

23:08

in is automated machine learning,

23:11

and I'm actually just writing on a paper

23:11

on it right now and surveyed a large

23:16

number of auto mail tools, which is making it a lot easier for

23:17

people to use machine learning and

23:22

hopefully do a better job than

23:27

the problem with machine learning analysis

23:27

pipelines is that there's a billion

23:31

ways to make one, right? Everyone

23:31

has an opinion, everyone has belief,

23:36

and there's a lot of right ways to do it,

23:38

but there's also a lot of wrong ways to do it. So paying attention to the

23:40

evolving machine learning AI in

23:45

terms of how data analysis

23:45

conducted I think is important.

23:49

We all taking a role and

23:49

being critical of how we do

23:54

that, I think is going to be

23:54

really valuable in the future.

23:59

Well, maybe kind of

23:59

taking a page from that,

24:03

this notion of garbage in garbage out, there are a lot of efforts currently

24:05

underway such as the National

24:10

Institute on Aging is creating

24:10

some data repositories.

24:14

There are a number of other organizations

24:14

that are looking to kind of create

24:18

these data pools of information that has

24:22

been well collected so that

24:27

train their solutions, train their ai,

24:29

have data that is relevant and meaningful.

24:35

I'd love to hear maybe Elizabeth and Mary,

24:37

if you know about some of these

24:37

kinds of efforts underway,

24:40

what your thoughts are as far as these

24:40

data collection and repository efforts

24:44

that are happening. Well, I know it from the N I A funding,

24:51

so there's $160 million I

24:51

think every year that they are

24:56

funding. And I think about 60%

24:56

of that is going into innovations

25:01

related to brain health.

25:03

And so we can look to the work of

25:03

some of those entrepreneurs and see

25:09

what they see. But maybe I should turn it

25:10

to Elizabeth to say more.

25:16

So there are some data sets that

25:24

are large enough, but with the evolution of treatment,

25:32

I think what we're seeing

25:32

is that those dataset,

25:36

what needs to be in those

25:36

data sets is evolving.

25:41

And so they don't always fit the bill

25:41

depending on what you're wanting to

25:46

research. And I do want to address a couple of

25:54

points. Ryan made a point earlier,

25:57

and there are a couple of things in

25:57

going on in the q and a chat here about,

26:04

I'll call it broadly social

26:04

determinants of health.

26:07

Part of what we're seeing

26:07

with the coming surge

26:12

of the existing and

26:12

tidal wave of Alzheimer's

26:17

research that we're seeing is the need for more diverse data.

26:23

And what that means is

26:23

that you can't just go.

26:27

So typically large pools of data have been

26:32

driven through academic research centers.

26:36

Those research centers skew wealthier,

26:40

wider, I'm sorry to make

26:40

mass generalizations,

26:44

but this is what we see again and

26:44

again regardless of therapeutic area.

26:49

And so we are seeing a push

26:49

on the part of pharmaceutical

26:54

companies and specialty

26:54

organizations to really try and get

26:58

attached to community

27:03

community points of care. And what that means is

27:06

that you're tapping into

27:10

physicians who are not

27:10

accustomed to research.

27:15

They want to be part of research,

27:18

but they don't have the practices, they

27:18

don't have the staff to support it.

27:24

And even then getting,

27:30

there's also in the chat

27:30

a little stream of, yes,

27:33

so much valuable healthcare information

27:33

exists outside of a point of care

27:39

and getting access to that

27:39

also means that you're

27:43

skewing richer, wider.

27:49

And so it's really, I think there's one challenge in

27:51

getting to the community physicians.

27:56

There's another challenge in getting

27:56

at the data that is happening

28:01

outside of a care setting and

28:01

just part of activities of

28:06

daily living that is really important. And

28:12

again, I think we're probably five

28:12

to 10 years from really having

28:17

good validated ways of

28:17

collecting that data.

28:21

I hope it's faster, but I think that the reality

28:23

of that term validated

28:29

means that it's a medium term

28:29

thing, not a near term thing.

28:34

I think this is an important

28:34

thread to talk about,

28:37

which is the notion of social

28:37

determinants of health. I think, Mary,

28:41

you had that really interesting

28:41

comment earlier about transportation,

28:46

and I know in some of our

28:46

earlier conversations that

28:50

you talked to me a little bit about

28:50

banking and financial records,

28:54

and you've talked to me a little bit

28:54

also about the role of that secretary of

28:58

state and the driver's license and what

28:58

this means for older adults is to kind

29:02

of, I'll say outside the healthcare

29:03

point of data that still have

29:08

incredible relevancy. And I'd love for you to just maybe

29:09

talk about a few of those things. Yeah.

29:12

I have a really great example. So

29:17

I am renewing my license,

29:17

and so I'm going to be 75.

29:21

A lot of my friends are doing the same, and so they all have to take and

29:23

prepare for the driver's test.

29:27

And so I took the a r P

29:27

driver's class and I had

29:32

found, my husband found for me a really

29:32

good program with AI built into it.

29:37

So it doesn't just teach

29:37

you the rules of the road,

29:39

it helps you understand rules of

29:39

the road. So in this class I was in,

29:44

which was live, I said, oh,

29:47

this program will really help you

29:47

understand the signs and everything. Well,

29:52

people said, I don't have a computer. So when you realize that digital

29:54

access to your point about data

29:59

sets is not there for everyone.

30:02

So the notion of just what Covid

30:02

taught us is we first have to help

30:07

people get digitally literate, and then that's another

30:09

way to gather information.

30:14

Otherwise, you could just have someone

30:14

check the box and say unfit to drive.

30:19

And so I think that's a big place that you

30:24

see it. Banking is another,

30:26

so older people are

30:26

looking for things to do,

30:30

and so they want to go and talk to the

30:30

banker, the savings and loan person.

30:34

They don't want the automated

30:34

necessarily want the automated

30:40

a T m because it's part

30:40

of their socialization.

30:43

And so it's the book

30:43

clubs, it's the pharmacies,

30:47

it's these local places where older adults

30:47

appear that that's where you begin to

30:52

see that they might just be losing it. So when their boyfriend tells them to

30:54

withdraw some of money, someone they met,

31:01

then the banker now sees that

31:01

maybe something's not okay

31:06

with mom or dad. So it's looking at maybe some of

31:08

these unconventional sources of

31:13

data, and not just data,

31:16

but potentially thinking about this

31:16

a little bit more humanistically

31:21

contact points that we

31:21

have in our community.

31:24

But for a lot of those contacts,

31:24

there is kind of a record.

31:28

There's something that exists out there. I think the surgeon general's report is

31:31

really important about loneliness and

31:36

even the fewer hours that

31:36

families are connecting,

31:39

fewer places that people can

31:39

go and gather. And as they

31:43

they have often fewer friends

31:43

because they lose their friends,

31:48

they lose some of their friends. So we have to think locally and we have

31:49

to think very broadly about how do we

31:54

reach these caregivers and how

31:54

do we train the caregivers,

31:58

many of whom do not understand the

31:58

nuances of the research reports

32:03

on things like dementia, but they're

32:03

in the front lines day to day.

32:09

I think it's a really interesting

32:09

point. Maybe Ryan, I'll ask you,

32:14

we've talked about some of

32:14

these kind of unstructured data

32:18

things that we see, for instance,

32:21

like notes in a health record or

32:21

maybe a banking flag or things

32:25

like that. How do you, as someone who's looking at data

32:27

sets start to consider some of this

32:31

unstructured data that's out

32:31

there to help weave a more

32:36

rich and thorough story about

32:36

the that we're trying to analyze?

32:42

Sure. So first off, I guess I should acknowledge that

32:43

I'm not really an expert on analyzing

32:46

unstructured data. I certainly

32:46

collaborate with those that do.

32:50

And this is an exciting time

32:50

to be in that area of research.

32:56

There have been over

32:56

the last 5, 6, 10 years,

32:59

pretty incredible advances in natural

32:59

language programming use of large

33:04

language models, deep learning methods in general to

33:06

work directly with unstructured data to

33:10

analyze that type of data. And I think it's important to keep an

33:12

eye on these emerging methodologies.

33:17

It's hard to keep track of because

33:17

there's so many people working in this

33:22

domain, it's hard to decide what is the

33:22

most valuable research to focus

33:27

on. And part of that is

33:27

that from a machine learning

33:32

there aren't really established

33:32

rigorous benchmarking

33:36

approaches. Every problem is different.

33:36

It is challenging to say, well,

33:41

this method works better than this method

33:41

for these reasons. It actually turns

33:45

out to be a whole can of worms. And so there is currently a

33:47

little bit of faith in saying,

33:50

I'm going to use this method

33:50

and run with it. But in general,

33:54

trying to understand the advantages and

33:54

disadvantages to any given approach,

33:58

typically with these new methodologies

34:01

that rely on deep learning.

34:04

The biggest trade off for me is that

34:04

you're most of the time giving up

34:09

interpretability, which in medicine is often a huge

34:10

selling point for methodology.

34:14

We want to be able to trust our models

34:14

and understand the predictions that are

34:19

being made by 'em. And that's always been a big disadvantage

34:20

of deep learning despite the hype

34:25

and the attention that

34:25

deep learning has received.

34:30

Not to say anything deep

34:30

learning is also incredible.

34:32

It does some pretty amazing stuff,

34:32

but it's just one tool in the toolkit,

34:37

right? We've got to remember to use

34:37

the right tool for the right job.

34:40

And often that's not going to be deep

34:40

learning for some of those reasons that I

34:44

just mentioned. With that, maybe Elizabeth,

34:49

you're sort of in the throes of where

34:49

the rubber meets the road between these

34:53

things, which is maybe traditional

34:53

and conventional data thoughts,

34:56

and then the understanding that they're

34:56

not capturing necessarily the full

35:00

picture. How do you bridge that chasm and what

35:02

are some of those things that you think

35:05

about day to day? Well,

35:13

one is let's start with counts.

35:18

Patient counts like how

35:18

many to answer a certain

35:23

question with

35:27

the required degree of

35:27

rigor for the purpose.

35:33

How many patients do you need

35:33

in your analysis and how are you

35:38

going to get there? On some

35:38

level you would say, oh,

35:43

it should be easy. There are

35:43

lots of Alzheimer's patients,

35:48

but actually finding the patients.

35:52

And then someone mentioned federated data

35:57

in the chat. I would add

36:02

this idea of federated data to Ryan's

36:07

list of wishful thinking. It's just not that easy

36:10

because everybody's structured

36:15

differently. Similar terms, even among very,

36:24

the leading experts or maybe even

36:24

especially amongst the leading

36:29

experts are used somewhat

36:29

differently and means somewhat

36:33

different things depending

36:33

on who entered the data.

36:38

How you bring all that together

36:38

in the end means that actually

36:42

getting sufficient patient

36:42

numbers can be tough.

36:49

I think that, I dunno, I feel like we've covered

36:53

a fair amount of this ground, but

36:58

we do think a lot about

37:02

are there big data

37:02

records that we can go to?

37:06

So we have another study,

37:09

another in another therapeutic area

37:09

where we actually are using driving

37:14

records as an indicator of safety events.

37:22

I've heard of research

37:22

around gun ownership and

37:26

Alzheimer's. As someone said, we now

37:32

doesn't take much to get a gun. A lot of people are going

37:33

to have Alzheimer's and

37:37

And that is when someone mentioned this

37:42

research to me, I was like, oh my

37:42

God, I'd never thought about that.

37:45

That's really, really scary. So we think about things like that.

37:53

Someone has mentioned prisons, we've been exploring work with prisons.

37:59

It is very hard for a

37:59

whole range of reasons,

38:05

but we do try and push

38:05

ourselves to think about

38:09

how big can you get with

38:09

some degree of reliability

38:14

depending on the research

38:14

purpose. Maybe I'll pause there.

38:20

Well, it's interesting. I want to maybe pivot the conversation

38:20

just a little bit as we are getting

38:25

towards the tail of our conversation here.

38:28

The first thing I guess I want

38:28

to ask a little bit about is

38:34

you mentioned how many people do

38:34

I need to have a successful study?

38:38

What is that number? And I would ask because of the varied

38:40

nature of Alzheimer's disease and

38:44

dementia, which is that it

38:44

involves a person who is not

38:48

cognitively at normal function

38:48

and sometimes is depending on

38:53

other people to help guide

38:53

them to appointments or create

38:58

continuity in their day-to-day life

38:58

and structure and medical records.

39:02

Is that part of the reason that you all

39:02

think that we have difficulty getting

39:07

access to this is because of the

39:07

nature of the disease itself and the

39:12

necessity for that? I'll say caregiver role

39:13

in its acquisition and

39:19

regular study. So I would say that a

39:21

lot of the challenges in

39:26

accessing data for

39:26

Alzheimer's are in many,

39:31

many therapeutic areas. It's just,

39:35

it's the reason even real world

39:35

research is supposed to be

39:40

more efficient and cheaper than

39:40

clinical trials. And it isn't always.

39:47

But I think for all the other

39:47

societal reasons that we've been

39:51

talking about, Alzheimer's

39:51

presents and a special challenge,

39:56

Alzheimer's psychiatry,

40:02

dare we even say the word pain, where

40:06

you begin. So we begin to ask ourselves,

40:10

is it that actually

40:10

there is that the science

40:15

is nascent because there's a lack of data

40:19

or is there a lack of data because

40:19

the science is really nascent?

40:23

Because when you, I mean, look, I know there's so much research

40:25

going on in Alzheimer's, I get that.

40:29

But when you look at

40:33

the way research has evolved relative

40:33

to how research has evolved in

40:37

oncology, inflammation, rheumatology,

40:40

those sorts of things,

40:46

the science isn't as mature

40:46

as those disease areas.

40:52

We simply don't know as much

40:52

about the evolution of the

40:56

disease, the methods of research,

40:59

the methods of clinical practice

40:59

as we do about these other areas.

41:04

And I do think that the lack of

41:04

data, it becomes an iterative thing.

41:10

Well, I think that's actually really

41:10

interesting and where we want to end our

41:13

conversation, which is

41:13

talking about insights,

41:17

because the hope is

41:17

that through using data

41:22

more constructively, that we are able to glean some

41:23

of these insights that help us

41:28

create better care pathways that empower

41:28

caregivers to have more meaningful

41:32

interactions. I'll just say

41:32

all the good things that we

41:37

hope for here in healthcare. So let's talk a little bit about

41:39

that interpretation part of it.

41:44

We're talking about all these

41:44

things, these massive data sets,

41:48

these social determinants of health, lots of points of M for data

41:50

to enter into the picture.

41:55

How do we make that meaningful? And

41:55

going back to this patient experience,

42:00

what does that maybe look

42:00

like for this disease state?

42:05

And I know this is asking us

42:05

to be a little bit of that half

42:09

full kind of perspective, but I think

42:09

that's a good place to sometimes go.

42:15

Maybe I'll start with you, Mary.

42:15

Thinking about a lot of people,

42:18

where do you think people who are going

42:18

through some of this would want to see

42:24

this data that they're

42:24

participating in their data?

42:27

How would they make it meaningful or what

42:27

is maybe some of the hope around that?

42:32

I don't know if this is the

42:32

exact answer to that question,

42:35

but what I see entrepreneurs

42:35

doing is focusing on

42:40

the care managers. And the problem with being a caregiver,

42:46

it's an average of 49 year old

42:46

woman who's already got a full-time

42:51

job in a family. She's

42:51

very club sandwiched.

42:54

And so to add anything else

42:54

into that role is hard.

42:58

And a lot of the caregiving

42:58

is done by the family,

43:01

but recently I've been seeing

43:01

a kind of new category.

43:05

And companies like the Key, I think do a really good

43:07

job of finding great

43:12

caregivers and educating

43:12

them about dementia.

43:15

So I think you have to look

43:15

at the corporate role of

43:20

who is playing a role in the

43:20

caregiving economy. In fact,

43:26

we're having a conference

43:26

on that in December.

43:30

And I also think AI can play a role. So

43:35

I think AI can play a

43:35

role in the training.

43:38

So I think that we have

43:38

to look at the people who

43:42

are touching the patient,

43:42

not the patient themselves,

43:47

to build data sets around

43:47

the caregiver economy

43:51

because that's the one that's going

43:51

to be on the front lines pointing to

43:56

this person has a problem or this

43:56

person doesn't have a problem.

44:00

There still will be a need to

44:00

have research with the end user.

44:04

And groups like Cabi

44:04

and the group in the uk,

44:08

they're beginning to create panels

44:08

where you can aggregate and find some of

44:13

those data samples. Thanks, Mary. I think that's

44:17

food for thought. Ryan,

44:21

when you think about the

44:21

synthesis, so what of it all,

44:25

what are things that you're thinking about? Sorry.

44:33

Yeah, I mean, we have,

44:33

all data is wonderful,

44:35

but if we can't distill it to have

44:35

a meaningful insight that can change

44:40

either a care pathway or

44:40

inform how people should

44:45

be approaching their disease management,

44:49

just data numbers. So how do we do that?

44:52

How do we look for those insights?

44:55

Gotcha. I mean, so first off I mentioned transparent

44:56

or interpretable machine learning.

45:01

That's one element that I always sort

45:01

of fall back to as being important.

45:05

And like I said, there's a lot of hype around

45:07

AI technologies that are

45:12

They're opaque. And so I'd like to see more research

45:14

and researchers developing and using

45:19

methods that are directly interpretable.

45:19

There's a small subset of us out there,

45:22

but we're largely overwhelmed. So

45:22

that's one thing to pay attention to.

45:27

And focusing on understanding what

45:32

are the variables that are important in

45:32

our data sets and what else we should be

45:37

collecting. So when we're analyzing data,

45:40

I feel like one question we should always

45:40

have in the back of our heads is if I

45:43

was to do this study again in the future,

45:43

what would I want to collect instead?

45:48

What would be better variables or

45:48

a better way to collect the data?

45:53

Another thing that's important

45:53

on the topic of data size and

45:58

fairness, most people when

45:58

they think of data collection,

46:02

they think more is better. And

46:02

that is in most cases, true.

46:07

We need more data to have more

46:07

power to more have more confidence.

46:10

But with larger data sets, typically they're going

46:12

to end up being messier.

46:16

They're going to be more heterogeneous,

46:16

which is both good and bad.

46:20

It's good because we're representing

46:20

a greater diversity of people,

46:23

hopefully if we're doing a good

46:23

job gathering a broader dataset.

46:29

But the downside is that a lot of

46:29

methodologies are not really set up

46:34

in analyzing data to

46:34

consider this heterogeneity.

46:37

And what I mean heterogeneity outs, I

46:37

don't just mean different backgrounds,

46:40

but I mean heterogeneous

46:40

associations where if we're trying to

46:45

predict an outcome, the factors that contribute to the

46:46

occurrence of that outcome can be very

46:50

different for different groups of people.

46:50

So this group of people over here,

46:54

they get the disease due to these genes.

46:56

And over here it's this combination

46:56

environment and some maybe other gene or

47:01

and beyond. And a lot of

47:01

methodologies that we have,

47:05

they're just trying to put together the

47:05

one best holistic model that's going to

47:09

make a decision for everyone. And that's a problem in itself

47:11

from a methodological perspective.

47:15

This is an area that we're

47:15

interested in, we study, and again,

47:18

I'd like to see more people just take

47:18

this into consideration and think about

47:23

what methodologies could we develop or

47:23

could we improve to tackle those kinds of

47:28

problems. Might be a little

47:28

bit in the weeds, but.

47:32

Oh, you know what? This is all about creating

47:34

these data insights.

47:37

So I don't think it's in the weeds at all. And hopefully we have people sitting on

47:39

the line that are thinking very much in

47:43

the same way that you are about

47:43

this. And maybe Elizabeth,

47:47

I'll give you kind of the

47:47

last call on this one,

47:50

which is what are those insights? What's

47:50

the things that we're hoping to glean?

47:55

What are some of the things that you'd

47:55

like to glean as you're thinking about

47:59

this? So when I look at what

48:03

one can look at the pharma

48:03

pipeline for treatment,

48:08

and it seems to me that there is a shift

48:13

to earlier and earlier

48:13

treatment. This is harder.

48:18

This is hard to do because people don't,

48:21

diagnosis doesn't always happen early.

48:24

And someone, clearly,

48:29

I've been monitoring

48:29

the chat all along here,

48:32

but someone mentioned something

48:32

about could we do something

48:36

like what was done during

48:36

Covid, where many people,

48:41

unfortunately I think not enough, but many people shared

48:43

their data in some way.

48:47

And I think that being

48:47

able to get access to

48:52

data in a way that allows us to understand

48:57

what early diagnosis really looks

48:57

like and begin to develop some more

49:02

definitive early diagnosis,

49:05

there's a lot to be overcome in that.

49:08

We've talked about a lot of

49:08

that here just in the last hour.

49:11

But I would really like

49:11

to begin to see more rigor

49:16

around early diagnosis, more concrete understanding of

49:17

what early disease looks like

49:25

and what both patients

49:25

and provider and payer

49:30

systemically, what does

49:30

that need to look like?

49:33

Because that is not where our

49:33

system is early diagnosis is

49:38

not where our system is geared to. I think.

49:43

It's going to be essential

49:43

for good treatment.

49:46

Mary and Ryan, nodding

49:46

aggressively. Go ahead.

49:50

I just want to say I really

49:50

agree with you about the

49:55

identification of a

49:55

patient population that can

49:59

participate in that early diagnosis,

50:01

and I think that's where you're going

50:01

to get the motivation of the end

50:06

user to participate. Absolutely.

50:12

Yeah. I think this is a good example for an

50:12

opportunity for maybe citizen science.

50:17

How do you get people engaged? How do you give them incentives

50:19

to be engaged in either

50:23

providing data or helping us

50:23

to understand early detection

50:28

and at the same time link that

50:28

to a direct benefit to them?

50:35

Well. Wild idea, which is

50:39

why not make it fun? So ais Innovation has partnered with a R

50:46

P to stimulate new people to play games,

50:52

and they launched new games like

50:58

Monopoly and Trivial Pursuit

50:58

with many generations.

51:01

But one of the things older people

51:01

talk about is using gaming to

51:07

keep their brains active,

51:07

like the crossword puzzle.

51:10

So if you kind of begin

51:10

with what they're doing,

51:12

150 million people watch Wheel of

51:12

Fortune talk about an audience and a

51:17

population. And I was at a memorial

51:18

for a woman who was 102,

51:22

and the secret was Wheel of

51:22

Fortune, keeping that brain

51:27

So I kind of think we need

51:27

a big initiative to get

51:32

people to get motivated and to

51:32

make brain health as important

51:37

as physical health and cardiac health. Well, I love this. I think we've heard

51:41

everything from citizen scientists to

51:47

extended multi-generational

51:47

gameplay from Mary and Elizabeth's

51:52

concerns about keeping the caregiver

51:52

involved and staying patient-centric.

51:57

I think ultimately what we've heard

51:57

is there's lots of opportunity here.

52:02

So I'm actually going

52:02

to say at this point,

52:05

thank you to all of my

52:05

guests today for your

52:10

really poignant insights for your

52:14

experience in your relative fields. I

52:14

think there's a lot to be done here.

52:19

So with that, thank you for joining us

52:19

today. We're thrilled to have had you,

52:24

and we hope that you'll visit us for

52:24

more information. Have a great day.