Speech Changes in Early Alzheimer’s

Speech Changes in Early Alzheimer’s

Record date: Aug 02, 2017

Kimberly Mueller, Associate Researcher at the Wisconsin Alzheimer's Institute at UW-Madison, discusses how analyzing everyday speech may provide early detection of cognitive decline.

University Place Campus: 

University Place Lecture Series: 

University Place Subjects: 

Episode Transcript

- Welcome, everyone, to

Wednesday Nite @ the Lab.

I'm Tom Zinnen.

I work here at the UW-Madison

Biotechnology Center.

I also work for UW-Extension

Cooperative Extension,

and on behalf of those folks

and our other co-organizers,

Wisconsin Public Television, the

Wisconsin Alumni Association,

and the UW-Madison

Science Alliance.

Thanks again for coming to Wednesday Nite @ the Lab.

We do this every Wednesday

night, 50 times a year.

Tonight it's my pleasure to

introduce to you Kim Mueller.

She's with the Wisconsin

Alzheimer's Institute.

She was born in

Ridgewood, New Jersey,

and went to River Dell

High School, right there.

Then she went 20 miles south?

- 35.

- 35, I stand corrected,

to Rutgers University

for her undergraduate

in psychology.

And then she got a master's in

speech pathology at

Columbia University, across

the river in Manhattan.

And then she came here to

UW-Madison in 2005.

She also did a PhD program,

which she just finished in May.

Way to go.

- Thank you.


- Tonight, the topic

is preclinical

Alzheimer's in speech.

She has a longer title,

which she'll give to you.

This is something that

anybody who has parents,

you're watching out for them.

I'm rolling the

odometer over to 60.

I'm watching myself not be able

to pull things up, words.


Like I used to.


So, this is going to be

interesting, and I think

that's why we have such a

great crowd here tonight.

Please join me in

welcoming Kim Mueller

to Wednesday Nite @ the Lab.


- What a great introduction. [laughter] Thank you.

Can you hear me okay?

All right. Thank you.

So, the longer title of this

presentation is

"Connected Speech in Early Mild Cognitive Impairment:

"Results from the

Wisconsin Registry

"for Alzheimer's Prevention."

So before I go any further,

I want to tell you what

is connected speech.

And connected speech basically

refers to everyday talking.

It's a kind of technical term.

I was just at a conference and

somebody said, "Well, what would

"my speech sound like

if it was unconnected?"

And the answer is that typically

when we test language,

we do so in a single word way.

We have people look at

pictures and name them.

That's a cat, that's a dog.

But connected speech is

putting words together,

everyday talking, just

like I'm doing now.

So, keep that in mind as we

go forward with this talk.

And, also, I just want to

thank everyone here at

Wednesday Nite @ the Lab,

Tom Zinnen, for having me.

Thank you to Sandy Rotter

for getting me here.

So I really want to thank--

I'm very grateful to

be here tonight.

So, thank you again.

All right, so this is Iris

Murdoch, award-winning novelist,

former president Ronald Reagan,

and novelist Agatha Christie.

So they all had

something in common.

They all had Alzheimer's

disease or, in the case of

Agatha Christie, suspected

Alzheimer's disease.

But they also had

something else in common.

They had a very personal

biomarker or cognitive marker

that was publicly and

is publicly available:

their speech and their language.

And we have this on record

for decades and decades prior

to the diagnosis of

Alzheimer's disease.

So in the case of Iris Murdoch,

she was an

award-winning novelist.

She wrote over 26 novels

in her lifetime.

She died of Alzheimer's disease

at the age of 79 in 1999.

And this was Alzheimer's disease

confirmed by autopsy.

And the last novel that

she wrote was in 1994,

and it was called

"Jackson's Dilemma."

And this novel was very

interesting because it got

really bad reviews,

which was unusual for her.

So in the New York Times book

review, in 1996, this is.

The review goes on to say that

the writing is a mess,

strewn with imprecisions

and blatant redundancies.

The author of this review also

mentioned how she would often

repeat the same phrase,

like now suddenly,

several times within

the same paragraph.

So some researchers then

took Iris Murdoch's novels

and looked at them over time.

So this researcher named

Peter Garrard in London.

And so he looked at her novels

at the beginning of her career,

at the mid-height of her career,

and then this last novel,

"Jackson's Dilemma," and

analyzed the content of her

writing and found that, over

time, by the end of this novel,

her content words decreased, her

lexical, her vocabulary became

more limited, and she did in

fact repeat words and phrases

much more often

by the last novel.

And so the authors argued

that this is evidence

of Alzheimer's disease at

play before the diagnosis,

before symptoms set in.

President Ronald Reagan was

diagnosed with Alzheimer's

disease six years after

he left office.

And some researchers in Arizona,

Berisha and colleagues,

they analyzed Ronald Reagan's

unscripted speeches from

press conferences from

1981 through 1988.

And they found that Ronald

Reagan's speech,

over time, became less fluent.

And that means that he

started to pause more often,

he used more filler words,

like "um," "well," "uh,"

and nonspecific words,

like pronouns,

and words like "thing"

and "that" over time.

And this was a significant

change over time.

So I'm going to play you two

examples from Ronald Reagan's

speeches, and see if you can

hear the difference.

Now, this is extremely

subtle differences,

so it may be difficult

to hear them.

But I'll try to point

them out as we go.

So this first one is his first

press conference in 1981.

- ...detente is possible?

- "Well, so far detente has

been a one-way street

"the Soviet Union has used

to pursue its own aims.

"I don't have to

think of an answer

"as to what I think

their intentions are.

"They have repeated it.

"I know of no leader of the

Soviet Union since the

"revolution and including the

present leadership

"that has not more

than once repeated

"in the various

communist congresses

"they hold their determination

"that their goal must

be the promotion

"of world revolution and a

one-world socialist or

"communist state, whichever

words you want to use."

- Okay, so aside from that

being the longest run-on

sentence in the

history of sentences-- [laughter]

And a lot of embedded clauses, it was a very complex sentence,

it was pretty fluent,

and he kind of knew

where he was going with

what he was saying.

So now we look at

press conference,

an excerpt from six years later.

And this is in 1987.

And this is his 41st

press conference.

- "Secretary of State Schultz

says that he's a good man

"and he can keep his job.

"Is Schultz right?

"Can Elliott Abrams keep

that job as long as he wants?

- "I know the statement made by

the Secretary of State,

"and that is the

administration's position.

"And I know the reference

that you're making to the

particular point

"in which he himself

volunteered that we had

made a misstatement,

"but I accept the Secretary's

statement in this."

- Okay, so this second sample,

very subtly you can

hear some hesitations.

Some "uh" filler

words like that,

which we all use by the way.

We all use these.

You can hear me using them

now throughout this talk.

But from a change, this

is a changed standpoint.

From earlier to later,

the researchers found a change.

Now, they did this in

a very controlled way.

So each press conference they

took a specified number of words

per conference and analyzed it

according to those 100 words

or 200 words, whatever they

used as their standard.

So, relatively speaking,

an increase in filled pauses

and pauses over time.

So, this is the nun study, and

this happened right here in

Wisconsin, in Milwaukee and

other areas of Wisconsin.

These researchers studied

nuns throughout their lives.

And one study of this nun study

found that by examining the

nuns' free-writing samples when

they were in their 20s, they

looked at idea density, so this

means, how many ideas relative

to the total numbers of

words did they express?

How rich were their

writing samples?

So they looked at that, and

nuns with lower idea density

in their 20s predicted

who of those nuns

would go on to develop Alzheimer's disease.

Not only on cognitive tests did

they perform worse 58 years

later, but also it predicted who

would have the brain pathology

of Alzheimer's on autopsy.

So something about linguistic

ability in early life

predicted later life

cognitive function

and Alzheimer's

disease pathology.

So here is a picture

of a hypothetical

mother and daughter.

And the mother in this case

has Alzheimer's disease,

and the daughter

is caring for her.

So my research question is, can

we use that same speech analysis

that was use retrospectively

with Ronald Reagan, Iris

Murdoch, can we use that same

analysis with the daughter while

she is still healthy and

follow that speech over time

to see if it is a

marker for change?

And that's the basis

of my research.

And the Wisconsin Registry

for Alzheimer's Prevention,

or the WRAP study, is going

on here in UW-Madison.

It's been going on since 2001,

and it is a cohort

of individuals who are at

risk for Alzheimer's disease.

So they're in their

late middle age.

They have no symptoms of

Alzheimer's disease.

We don't know who of them is

going to go on to develop the

disease, but they're enriched

for parental family history.

So we are following this group

over time to determine who is

going to get the disease and

what about them is possibly

increasing or decreasing risk?

So, pausing, and to talk about

what is Alzheimer's disease.

So Alzheimer's disease is

defined by the brain pathology.

So the amyloid plaques, which

are hard, sticky substances

which you can see up there in

the yellow areas, and the

neurofibrillary tangles, which

clumps of tau proteins that

eventually cause the neurons

in the brain to die.

And this is the definition

of Alzheimer's disease.

It's the presence of these

plaques and tangles.

Now, Alzheimer's disease

is a cause of dementia.

Often people wonder, well,

what's the difference between

dementia and

Alzheimer's disease?

Dementia is the

set of syndromes,

and Alzheimer's disease is

a cause of that dementia.

Alzheimer's disease is

the most common cause.

There are other causes, like

changes due to vascular changes

from stroke or Lewy

body dementia.

Alzheimer's disease

is the most common.

Currently, there are over 5.5

million Americans living

with Alzheimer's disease

in the US today.

And this number is

expected to triple by 2050

if a cure or prevention

is not found.

Alzheimer's disease is the sixth

leading cause of death in the

United States and is the only

cause of death for which there

is no prevention or known

cure for this disease.

Alzheimer's disease is not

a part of normal aging.

It is not inevitable that people

will get Alzheimer's disease

if they live long enough.

It is an abnormal disease,

and it needs to be understood.

And that's why there is a huge

push to try to understand what's

causing this disease and how

to prevent it or cure it.

So the symptoms of Alzheimer's

dementia are memory loss

that disrupts daily life.

So you can no longer live

independently when you're

diagnosed with

Alzheimer's dementia.

There's a confusion

with time or place,

withdrawal from work

or social activities,

severe challenges in

solving everyday problems.

Things that used to be automatic

are no longer easy to do.

And there's a problem

with words and speaking,

which is my area of interest.

It's generally accepted now that

Alzheimer's disease is on a

continuum and that there is an

asymptomatic phase where the

plaques and tangles are forming

but there are no symptoms.

And this asymptomatic phase

could be occurring decades

and decades prior to

the onset of symptoms.

Mild cognitive impairment is the

precursor to Alzheimer's disease

dementia, or the

intermediary stage between

asymptomatic and dementia.

And the criteria for being

diagnosed in a clinic for mild

cognitive impairment, or MCI, is

that there are cognitive changes

that are concerning to either

oneself or one's family, one or

more cognitive domains are

affected from previous function.

There's a decline from

previous ability,

but a person is

still functional.

A person can still

live independently,

and that's the difference between MCI and AD dementia

is that they're

still functional.

And it's possible that

compensation strategies

can be taught and used

effectively at this MCI phase.

So it's really important that

we diagnose this disease

and this process as

early as possible.

So, what do we know about

speech and language

in Alzheimer's disease and MCI?

Basically, as Tom was

referring to, as we age

problems with word retrieval

are a normal part of aging.

Word retrieval becomes slower

and more difficult as we age.

However, this word retrieval

problem is not, it's a nuisance

but it's not a major

disruption to communication.

You might not be able to

retrieve the word, but it will

come back to you at the most

inopportune time hours later

when it's no longer relevant

but it does come back to you.

Also, it doesn't cause a person

to completely withdraw and feel

so frustrated that they no

longer want to communicate.

This is what it looks like

in mild cognitive impairment

and Alzheimer's disease.

The word retrieval problems are

so intense that they will occur

multiple times within a

sentence or a conversation.

And they disrupt fluency,

like we talked about

with Ronald Reagan.

Also, the word retrieval

problems became so severe that

sentence structure becomes less

complex and disorganized.

And on this slide, that bottom

piece about pragmatics and

discourse, this refers to one's

ability to communicate back

and forth with other people.

And this part becomes affected

in mild Alzheimer's disease,

possibly even in MCI,

because of that frustration

and kind of a quitting of

what you were going to say.

Losing the thread of thought in

conversation and getting to the

point where you change the

topic or you withdraw.

So these are the

things that we see.

We know that in severe

Alzheimer's dementia people can

become completely nonverbal and

lack any means of communication.

So, again, just like I said

earlier, the declines in

language are on a continuum

as well with this disease.

So, connected speech, I told

you what that definition is.

And this is just

an illustration.

Currently in research

studies or in the clinic,

we evaluate language by typically asking people

to name objects or pictures.

So, on the right-hand side, tell

me the name of what you see.

That's a cat. A ladder.

That's not really going to get

at what people are feeling

frustrated with and

are complaining about

in their daily lives.

How can we do that

in a controlled way?

One way is by picture

description tasks.

So, on the left-hand side, this

is a line drawing, a picture,

and some things are

going on there.

And the question is,

tell me everything

you see going on in the picture.

And then that will then elicit

more typical everyday speech

and language than another

kind of task would.

And so that's what I'm

interested in, in my work.

I alluded to this already, but

why would we study connected

speech in this

preclinical phase?

Why is that important?

And this article was

written in 2014.

And it asks the question, if we

have, we have better methods

of identifying the plaques

and tangles now.

We can see plaques and tangles

on PET scans and MRIs.

We can see a lot of information,

more than we ever

could in the past.

So we've made great advances

there, but we haven't made as

much advance in detecting

subtle change in cognition

that actually affects

everyday life.

So if we have a drug trial,

we want to know. "Is the drug

working in everyday life?"

That's the question.

It's not, okay, the

person's staying stable

on this neuropsychological test.

It's more, is it actually having

effect on everyday life?

And so that's my question Can

speech be a possible tool

for that, a tool for screening

or for disease monitoring?

That's one aspect of this

that I think is important.

So, again, this research

question that I'm asking,

can everyday language,

just talking, be a tool

for assessing cognition

and a tool for assessing

a functional activity of

daily living?

That's my question.

So I gave you a definition

of connected speech.

You'll hear me interchanging

language and speech together.

And just as a

speech pathologist,

the word speech refers

to the automatic process

of producing sounds

to form words.

And that's different from

language, which has to do with

the rules of vocabulary

and the rules of grammar.

So these two words are not

quite interchangeable.

They have different meanings.

But in this task I'm actually

looking at both aspects,

mechanical aspects, timing,

as well as the content

of the language.

It's also important that you

keep in mind as I talk about

my study that mild cognitive

impairment, what I talked about

earlier, is a clinical condition

that diagnosed by a doctor.

And that's different from

what I'm going to talk about,

which is early mild

cognitive impairment.

So, something that's not yet

diagnosed by a physician

but something that we see

in a longitudinal study.

So it's almost a pre-mild

cognitive impairment state.

And the last thing is

ecological validity.

This is something that we're

looking for in our tools.

Ecological validity gets at

whether or not a test measures

what it's designed to measure.

So, is speech, is a single

word task going to measure

how someone actually converses?

Or my argument is that connected

speech tasks might actually be

better at measuring

an everyday activity.

So that's ecological validity.

So the first step in doing this

work was to review connected

language tasks in MCI and AD.

What have people done so far,

other than the few studies that

I talked about at the beginning?

And what I found by doing this

review was that most of the

studies of connected speech

were done with people

with Alzheimer's disease

dementia later on.

Very few were done in the

mild cognitive impairment

stage or earlier.

I also found that across studies

there were multiple

measures used.

People looked at percent nouns,

percent pronouns,

number of pronouns divided

by nouns plus pronouns.

I mean, the list went

on and on and on.

And sometimes it was

similar constructs

but different names

for these constructs.

So there was not a

lot of continuity

among all of these studies.

I also found that picture

description was the most common

method of eliciting

connected language.

There are other ways

to elicit language,

such as tell me about

where you grew up.

That's a confounding way of

eliciting that in a memory

impairment kind of disorder

because we don't know then

is it a memory of events

that's getting in the way.

And that's why picture

description is a confined task

with the stimuli right there

in front of the person.

And so memory for events and

things is less of an issue

with a picture description task.

So I want to tell you about one

other study in particular.

This is a study of people

with the inherited form

of Alzheimer's disease.

It's a very rare genetic form.

It's involved with

the presenilin gene,

and it's a very rare form.

Usually the onset of

Alzheimer's disease

happens in a person's 40s

or younger, so very rare.

But there is a

family in Columbia

who have this

genetic form.

It's a rich family for

this genetic form of

Alzheimer's disease.

So what these researchers did

was they studied people who had

the gene and then family members

who did not have the gene

in their early 30s and 40s

before any symptoms of

Alzheimer's disease.

And the people with the gene,

the carriers of the gene,

showed reduced content in

their picture descriptions

of connected speech versus

those who were not carriers.

So this is an asymptomatic

cohort, and already

the researchers were seeing

change in connected speech

prior to the onset of symptoms.

So this cohort that I'm looking

at is the Wisconsin Registry

for Alzheimer's Prevention,

as I mentioned earlier.

I'm going to tell you a little

bit more about this group.

The goals of the WRAP study are

to identify Alzheimer's disease

in its preclinical phase, before

symptoms set in, and to identify

the biological health and

lifestyle factors that might

increase or decrease risk

of Alzheimer's disease.

So, participants, this is

the design of the study.

And the idea, if you look at

this graph, on the bottom is age

and on the Y axis is

cognitive function.

And so the idea is that in

normal aging, you see that black

line, cognitive function will

decrease slightly but not a lot.

But then this other

trajectory in red is the

trajectory of

Alzheimer's disease.

And so we're following

these individuals

when they start at a young age.

The mean age at baseline

was 54 years old.

So they are cognitively healthy,

no problems, they had a parent

with the disease, but that's

about all that we know.

And so we follow them over time.

After the first visit,

they come in four years later

and then two years

every visit thereafter.

Currently, we're on

our sixth wave visit

for many of our participants.

Currently, we have over

1500 people enrolled

in this Wisconsin study.

The last visit the

mean age was 62.5.

There are 70% women

in this group,

and the mean age of

education is 16 years.

So they have at least a

college degree, on average.

A very highly-educated group.

Their average IQ is also high

for average. Above average.

There are 90%

non-Hispanic white.

We are working on increasing

our diversity of this sample.

We have a cohort in Milwaukee. We have a site in La Crosse.

So we're trying to sample

as much diversity

in terms of rural and urban communities as possible.

40% of these people have an

gene that is a risk factor gene

for Alzheimer's disease,

the later onset version.

And this gene is called APOE E4.

And so it's enriched for that,

probably because of this parental family history.

But we don't know whether or not if you had that gene,

you could still go on to not

develop Alzheimer's disease.

So it's a risk factor gene.

Your risk is increased,

but it's not a definitive

like that other study I

was talking about earlier.

72% have a family history

of Alzheimer's disease.

So these participants

are extremely giving.

They come in and spend

about six hours with us.

And they undergo a

fasting blood draw.

So we look at all kinds of

aspects of risk that we know

about in their blood, including

the genetic risk factor.

Their glucose levels,

their insulin levels,

vitamin B, vitamin D.

We're looking at all

kinds of things

that we know about that

may increase risk.

We also get other

physical measures,

like blood pressure and

resting heart rate.

We ask lots and lots

of questions.

They fill out questionnaires

prior to coming in.

It takes them a couple

of hours to fill out

all these questionnaires

about their lifestyle,

their medical history,

their physical activities.

What are they doing?

What are they doing?

If they're retired,

are they volunteering?

Are they doing computer games?

Are they, what are they

doing for their cognition?

Are they in groups?

What is their social support?

So we're looking at all of the

things that you may have heard

about in the news about

increased or decreased

risk of Alzheimer's disease.

And they undergo

two-and-a-half hours

of extremely detailed

cognitive testing.

And these are neural

psychological tests that have

been validated and used

across other studies, and

they look at memory, language,

speed, executive function,

and a global measure

of cognition.

Also, in this WRAP study, I've

cited two articles from the WRAP

study that detail how we go

about determining if someone is

showing signs of this early mild

cognitive impairment, or EMCI.

So this is a construct

that was developed

by establishing robust norms

within the WRAP study.

So that means that we're able

to look at people relative to

themselves and relative to

people in their cohort

and determine if they're

declining over time.

They may not feel like they're

declining, they may not feel

like they have symptoms

of Alzheimer's disease

or of cognitive decline,

but we are seeing this

according to our construct.

And so the people that I'm

interested in are the people

that are circled there in red.

So, people who are cognitively

healthy over several visits.

They're not declining.

In fact, some of them

might even be getting

better on the tests over time.

And the people who are

declining, this EMCI group.

I just want to tell you a

little bit more about WRAP.

This slide is extremely busy

because the WRAP study

is extremely busy.

So there's the main WRAP

study but off-shoots of that

are other studies

that are going on.

And so, because of these other

studies, we're able to collect

biomarkers, like PET scans,

MRIs, cerebral spinal fluid,

where we can really

detect the plaques

and the tangles in

these controlled ways.

And so there's all kinds of

other studies that are going on

that are off-shoots of

the main WRAP study.

Okay, so back to the

connected speech task.

So I'm going to show

you this picture.

This isn't the exact picture

that I'm using in the WRAP

study, but it's

extremely similar.

It's a black and

white line drawing.

It's very simple.

The prompt is tell me

everything you see

going on in this picture.

And we have a very simple $50

digital recorder that we use.

And the participants talk

into this digital recorder.

It's actually a very sensitive

digital recorder, so we can hear

very, very subtle nuances

in their speech.

So a participant might

give a sample like this:

"Well, I see a man stuck up in a tree," and then they laugh,

"while a daughter, um,

I mean a girl," and so on.

And so this is what

we're recording.

After we record that, we come

back and then we transcribe it.

And this is done by a

speech pathology student

or a speech pathologist.

And the transcription is

basically writing verbatim

everything that they say,

including these hesitations and

these filler words, clearing

their throat, laughing, all of

these things are coded in this

Codes for the Human Analysis

of Transcripts, or

the CHAT program.

So it's a specialized coding

method that we then feed

into a natural

language processor.

And this is called Computerized

Language Analysis,

or the CLAN program.

It was developed at

Carnegie Mellon University

by Dr. Brian MacWhinney

and his colleagues.

And what this does is it allows

us to feed these transcripts

in and quickly, automatically,

extract all kinds of measures

about speech and language.

So it's a very quick analysis.

The transcription part, after

training, it takes approximately

15 minutes to transcribe

about a minute of speech.

So 15 minutes to transcribe

about a minute of speech.

And then this part, the

analysis, takes seconds,

basically, to do

multiple data points.

So the first study I did

was kind of a pilot,

a cross-sectional design to see

can we just see anything

with this connected speech task.

And this was done

kind of early on

when we didn't have

a lot collected.

Presently, there are over 2,500

connected speech

samples that we have.

But at the time we didn't

have a lot transcribed.

So we looked at 78 people, and

we used a matched pairs design.

So this means that 39 of them

were determined to cognitively

healthy over time, and 39 of

them were determined to have

early mild cognitive impairment.

And this paper was published in

the Journal of Alzheimer's Disease in 2016.

And what we found was that, in

fact, people with early mild

cognitive impairment in these

one-minute picture description

tasks were producing

less content words

and they showed

lower idea density,

liked I talked about

with that nun study.

So, fewer ideas relative to

the total number of words

that they expressed.

So this is what I found at that

point, that cross-sectional

point, and this was at

their first speech sample.

So I wanted to next address this

question, since we found that

this is working or it's

possible, I wanted to address

that next problem about so many

measures that we can extract.

How do we choose what to look at

when there's so many

measures we can look at?

And so the way that I

approached this problem was by

doing a factor analysis.

And a factor analysis basically

takes all those measures

and distills them down into

these underlying constructs.

And we had a really

impressive number

of transcribed speech samples: 399.

This is a large amount

of transcriptions.

And they were 399 people who

were cognitively stable.

So we wanted to do

that on purpose.

We wanted to look at

these measures in people

who were not declining.

So we separated these 399 people

into two different groups.

One was the exploratory group

and one was the

confirmatory group.

So the exploratory

group had 250 people,

and the other one had 150.

And the two groups were

very similar except that

the confirmatory group had more

people with a family history.

I addressed that problem,

and I'll tell you about

that in a minute.

Basically, we found a

confirmed factor structure,

first in the exploratory group

and then we found it again

in the confirmatory group.

So that says that this

structure is relatively stable.

And, basically, we're

taking all these measures

and we found that they kind of load on four different factors.

And this is what they look like.

There's a lexical factor,

which has to do with the

diversity of vocabulary used.

The content or semantic factor,

which has to do with this

number of content words,

but also this pronoun ratio.

So, how many pronouns

are they using

relative to the number of nouns?

A syntax factor looks at

how complex the

sentence structure is.

And then this fluency factor,

which looks at all of those repetitions and filler words

and pauses in their

speech in an indexed way.

So, now that I confirmed

that factor structure,

I also confirmed that it was

invariant across groups.

So it didn't matter if they

were family history positive,

family history negative,

men versus women.

The factor structure was the

same across those groups.

And I looked at the factor

structure in stable people over

time at two time points

and found that people

who were stable on other

neural psycho measures

were stable on these

factor scores.

So that shows that the factor

structure can possibly be a

reliable way of looking

at speech over time.

And that was my goal

for that study.

So the final study then looked

at those factor scores,

those constructs,

in participants

with language samples

at two time points.

And we had 264 people.

And the question here was, are

declines in memory associated

with declines in

connected language?

So, the results.

First we looked at the

demographics of this group.

And so we just took 264 people

who had language samples

at two time points.

That's all we knew about

them at the outset.

And when we looked further,

we found that 200 of them were

cognitively healthy at several

time points, and 64 of them had

this early MCI construct

at the most recent visit.

So what else we know about

these groups is that

the early MCI group had a

higher percentage of men,

and they were slightly

older in the EMCI group.

Otherwise, and this

is important,

the two groups were the

same in terms of age,

level of education,

their literacy ability

as measured by neural

psychological testing,

and their family history of Alzheimer's disease,

as well as their ethnicity.

So the two groups were similar

except on age and sex.

So I did a linear

mixed effects model,

and so that basically allows

you to look at, over time,

how a person, how a group,

is performing.

And so if you look here,

the dotted line is the

group with the EMCI status.

And the first, the top graph,

the bottom, the X axis is age,

so as they get older, and the Y

axis is that fluency factor.

And you see that the dotted

line, the group with EMCI,

is declining faster on

these measures of fluency.

So they're becoming more and

more disfluent in a faster way

as compared to the

cognitively stable group.

The other piece was that the

semantic factor was also

significantly different in that

the EMCI group in the dotted

line declined faster

over time in content.

And they produced more pronouns

relative to nouns over time.

It's interesting, if you look

at that graph on the bottom,

the EMCI group appears

to start higher

than the cognitively stable group in measures of content

but then declines faster

and more steeply.

I don't have a full explanation

for that at this point in time.

I also did a logistic

regression, and what that says

is can we predict who's going

to have that EMCI status

at the most recent visit.

And that's done by a

logistic regression.

And so what we found was that

that fluency factor, again,

was a significant predictor of

who would go on to be diagnosed

with this EMCI construct

at the most recent visit.

So on the right-hand

side is the EMCI group,

and that's the amount of negative change in fluency

compared to the cognitively healthy group.

Also interestingly found that at

baseline syntax or sentence

structure complexity was a

predictor at baseline of who

would then be diagnosed with

EMCI at the most recent visit.

So this is similar to the

finding of the nun study

where early life

complexity seemed to be

a predictor of later life cognitive function.

I don't know that this

is the same finding,

but it's interesting

anyway at this point.

Other metrics of

the speech sample.

I found that at both visits the

group with EMCI produced shorter

sentences overall, and they

were slower to produce the same

amount of words and ideas as

the cognitively stable group.

So they took longer.

So what does this sound like?

I'm going to show you or play

for you an example of this.

The first is a 62-year-old

participant in 2013, his first

sample, and then that same

participant two years later.

So, this participant,

at time two,

declined significantly

in fluency.

So he becomes less

fluent over time.

He also produces less

content-rich speech over time.

And his lexical diversity,

his word choice,

becomes narrower over time.

So I'm going to

play these samples

and see if you can

hear the difference.

And, again, it's going

to be very subtle,

but you may be able to

detect the difference.

So this is the first time point.

The speech is slightly altered

to protect the

person's identity.

- Tell me everything you see

going on in this picture.

- Okay, I see a family, um,

together in the kitchen

with some possible

potential accidents.

Um, mother is

washing the dishes.

She's daydreaming and the

water is running over

and it's spilling in a way

that might cause her to fall.

And the siblings

are not supervised

because mother's daydreaming.

And they are, the brother

is on a high stool,

which he probably shouldn't

be on, going for the cookies,

and he's just about

to take a tumble.

And sister might be

laughing about that.

There may be some

sibling rivalry there.

- Okay, so that's the

first speech sample.

You do hear some

disfluencies there.

He says "um" a few times,

but now, this next sample,

there are significantly

more filler words

like that and hesitations

and things like that.

So you'll be able

to hear that now.

- I'd like you to

tell me everything

that you see going

on in this picture.

- Okay, so I see a kitchen

scene, um, that's chaotic.

Um, mother has two children who

are not, not given, um, um,

line-of-sight supervision.

And so the children are

reaching for cookies

on the top shelf

and eating away.

Probably overeating on sugar.

And then the boy is about to be

launched from a three-legged

stool that is tipping over.

And then mother appears to be

daydreaming about something.

That looks like a nice daydream

because she's smiling.

Um, so she's kind of ritualistically

cleaning a plate.

Simultaneous, the um, the sink

is overflowing to the floor.

So it looks to me

like it's a scene

that will soon be complete

chaos when the chair tips over.

- If you notice, he

does speak a lot more

the second time around,

but, relative to that

increase in words,

the number of ideas

actually went down.

So he's speaking more, but

conveying less over time.

So, what does this mean?

Well, first, it's really

important to note

that these individuals

have this EMCI construct,

early mild cognitive impairment.

We don't know a lot about that.

We don't know if that is going

to then go on to develop mild

cognitive impairment, go on to

develop Alzheimer's disease.

We don't know that

about this group.

And so that's why it's really

important for my next step

to look at the biomarkers

that we know about.

So the amyloid plaques

and the tangles.

If we have participants who have

evidence of Alzheimer's disease

pathology in their brain and

then these findings stay

and hold, then it will be more

convincing about what that means

for this tool in the future.

The other finding is that this

fluency factor, at least at this

stage, was the most robust

in terms of findings here.

And that, to me, was encouraging

because the other factor

that has been typically

shown, the content,

typically shown to be affected early on in the disease process

is dependent upon the

stimulus that's used.

And if we can detect changes in

everyday speech where we don't

need a particular stimulus but,

rather, it could be a

conversation at the doctor's

office, for example,

or a conversation at home

that we tape record.

This would be even closer to an

ecologically valid measure,

an everyday activity.

This measure, you know, showing

a picture in a confined setting

is closer to everyday speech,

but it's still not quite.

And so if this fluency factor

can be translated to an everyday

activity that's truly an

everyday activity, then it will

be even stronger as a measure

to use in the clinical setting.

Finally, what I found here,

which to my knowledge this is

the largest study of a

preclinical cohort of connected

speech at two time points.

Usually the sample sizes for

this kind of work range

between one and 30 people.

And so this is a large sample,

and we have many more to use.

And it was feasible

to do it this way,

to have manual transcription from a student.

It took 15 minutes per sample.

But now, with technology improving, the next step would

be if we could find an automatic

speech recognition tool

that would eliminate that step

for manual transcription,

that would make this

tool even better.

And so that's another

future direction.

Things here at the UW-Madison

are going on, looking at these

automatic speech recognition

tools and other machine-learning

techniques of doing

this in a faster way.

And so that's also a future

direction for this work.

I want to bring it back and

end with the participants.

So, they are at an

increased risk.

They have seen this

disease firsthand.

They've cared for someone

with Alzheimer's disease.

We have some control subjects

who have no family history,

but they're often spouses of a

person who cared for a parent.

So they have seen the

disease firsthand.

I would bet that almost everyone

in this room has had some

experience with dementia

or Alzheimer's disease.

And so it is not easy to come

in and give up a day and put

yourself on the line and do

these difficult memory tests.

It is not an easy thing to do.

It's an emotional thing to do.

It's an extremely generous

thing to do because,

as you'll hear this sample,

this participant will tell you,

they don't expect to benefit

by their participation.

They do this for

future generations.

They know that they're being

watched over years and that

their contributions are going to

go far, far into the future.

But they don't know

that they'll benefit.

So, again, that makes it even

more heroic what they're doing.

So I just want to play another

part of a speech sample.

So we also collect other

parts for the speech sample.

In addition to that picture

description, we ask open-ended

questions as well, which I plan

to analyze in the future.

But one of them is, "Tell by

about why being involved

"in the WRAP study is

important to you."

And here's the words from

a participant about that.

- You know, when you have a

parent with the disease and you

feel pretty helpless and I

learned about the study, it was

a way that I thought I could do

something by giving my time.

I mean, I don't really know what

I was getting into signing up

for it, quite frankly, but I

think those of us who have the

experience of having a parent

with the disease feel so

helpless in many respects that

being in a study is a way to

contribute, and it helps to kind

of offset that helplessness.

And it also feels like

you're being part of

a potential solution.

Thank you.

- So, my first thank you, then,

goes to our WRAP participants

here in Wisconsin.

They also come from

all over the US.

So I really want to thank

the participants for what

they do for us and our studies.

On this slide are

all kinds of people

who are involved in the

WRAP study and in my work.

And so, special thanks to

Dr. Sterling Johnson,

who is the PI of the WRAP study,

Dr. Lyn Turkstra, who was my

mentor for my dissertation work,

and all of the

faculty and staff

at the Wisconsin

Alzheimer's Institute

and the Wisconsin Alzheimer's Disease Research Center.

Finally, the WRAP study is

funded by the National

Institutes on Aging of the

National Institutes of Health.

So, thank you very much for you

time, and I look forward to

answering your questions.

Thank you.


Share this page