How technology can help people affected by Dementia
[Music]
dementia
is a near degenerative condition in
which
the brain cells die on the
on the right hand side you can see a
healthy brain tissue
and on on the left hand side you can see
a brain
tissue affected by dementia
currently there aren’t any
pharmaceutical or medical intervention
to cure or stop dementia
struggling to remember the current
events and
memory loss are the most common symptoms
of dementia other symptoms include
changing mood behavior and
being lost in familiar places
there are around 50 million people
around the world
affected with dementia it’s estimated by
2050 we will have over 130 million
people
with dementia in fact in the next
15 minutes over 200 people around the
globe will be
diagnosed with dementia that’s in
average something around
one every four seconds the
hospital admissions in people with
dementia is also
often higher in the uk for example
at any given time one in four
hospital beds are taken with by
someone with dementia
the alzheimer’s society in the uk did a
study and they reported
close to 20 of these hospital admissions
are due to preventable causes the most
common
uh reasons of people with dementia being
admitted to hospitals
are falls hip fractures
breathing problems stroke and urinary
tract infections
providing care for dementia is also a
very
resource intensive task my own
grandmother had dementia
and i’ve i’ve seen and noticed the
family members
usually shoulder a large part of this
burden
there are also disparities in how people
access to care depending on where they
live
in this map you see parts of the united
kingdom
and the darker shades of blue
shows more people had their care being
with dementia
had their care been reviewed over the
past 12 months the lighter you see the
shades are
less people their care had been reviewed
over the past 12 months
you can see in parts of the country
there were close to
50 percent of the of people affected
with dementia uh their care having been
reviewed uh over the past 12 months
often when people get diagnosed with
dementia they did they get
a diagnosis they meet the doctor and
they are sent home
and they may occasionally visit they see
their doctor
but often they the case is that
something happens they
decay their health deteriorates and
sometimes becomes serious and they’re
admitted to hospital or they need to go
and see their gp
also there is this assumption dementia
is a condition which affects people in
industrial countries but that’s not
necessarily true
actually there has been an increased
number of people being diagnosed with
dementia in lower income
countries as well especially in south
asia
and parts of the pacific
three years ago my colleagues in our
national
health services in the nhs and a group
of
clinicians we’re working with our
technical team we started a project
called
tim for dementia the idea was can we use
low-cost and connected technologies
these are devices which provide
environmental monitoring physiological
monitoring can be used as technology
and power of ai and machine learning to
provide
better care and support to people
affected with
dementia devices we work with several
companies and the devices we use
are in two categories devices that
monitor
environmental data like movement around
the house
if you open a fridge door cabinet tour
people go a number of times people go to
bathroom
sleep and use of home appliances like in
the uk people usually wake up and switch
on a kettle make a cup of tea probably
here they make a
coffee and the second category of
devices uh are
technologies which they use uh to
monitor
they basically use them to monitor
physiological symptoms like
uh heart rate blood pressure body
temperature
weight and hydration the
idea is if you use connected devices
user devices which are off the shelf
available and you can integrate if you
can integrate all this data you will
have more continuous information about
people and their day-to-day activities
first we needed to create a system to
which is
safe and secure because we are
collecting highly personalized
information we work with different
groups we have created a system which
allows to integrate
data from those devices and then we use
machine learning and ai
to analyze this information most of this
data are
numerical measurements they on the
single pieces of data information
usually they don’t make much sense
unless you combine them with other
information and you analyze them
over time one of the algorithms we have
analyzes people’s day-to-day activities
and their routine in this graph
the x-axis shows days of the week and
the y-axis shows time of the day and
each colored block shows one type of
activity for example you can imagine
uh red black shoes the sleeve green
shoes
having breakfast let’s say blue is
watching tv
and often if it’s someone it does the
same thing every day
the same time obviously you can see i’ve
made up that figure
uh on the right the horizontal line the
colors will look the same but
in reality no one will do the exactly
the same activity
exactly the same time of the day there
will be some randomness in people’s
activities
and you can see the other figures shows
a picture from a real
home of someone affected with dementia
but what we would
wanted to do uh we wanted to see how
much randomness is in this activities at
whether
within this randomness we can find some
patterns we have created an algorithm
which looks at the activities and looks
at the transitional probabilities
between them for example
if i wake up in the morning and i go and
make my cup of tea what is the
likelihood
i go and back to the bedroom i go and
sit on a chair
or go and watch tv we let the machines
to observe these activities and they’ll
learn this probabilities for example
let’s say if i make my cup of tea
eighty percent of the time i go back to
the bedroom ten percent i go and sit on
the sofa
ten percent of the time i go and switch
on the tv but
it’s highly unlikely i leave the house
now we let the machines to look at this
data over two months machines can
are good if you program them to learn
from experience then for two weeks we
let the machines to look at how much
people deviate from these activities
once we have learned that now we have a
personalized
model of each person’s activity routines
and what we are interested in
is level of surprise for example if one
day i woke up
had my cup of tea and i left the house
that’s something machine hadn’t seen
before
this level of surprise will increase and
what we do basically we measure this
level of surprise per day
and if we notice this passes over the
threshold we create
for example an alert that could be
related to someone’s health
declining they become less active they
become socially isolated depressed
or can be related to hyperactivity which
sometimes is related to mood changes
agitation and irritability one of the
other top reasons in people
with dementia being admitted to
hospitals is urinary tract infection and
urinary tract infection a bacteria gets
into the
bladder and that if that get detected
early
is very treatable you can treat them
with antibiotics
but in people with dementia because this
some of the symptoms gets also mixed up
with symptoms of dementia
it’s very difficult to sometimes to
detect that and if you don’t detect it
then can
the infection can spread in the blood
can become a really a serious health
problem
and often people are admitted to
hospital
uh the standard test the the medical
test is is using a dip test or blood
tests which happen
in the clinic but you do that only if
you know someone has a higher risk of
uti without that information how can we
actually start looking for the risk
factors
we worked with our clinical team and we
have created an algorithm it turns out
if someone has uti the number of times
they go to bathroom will increase
we put sensors and we count number of
times they go to bathroom
and what we are interested in the
increase someone having infection
could be possibly have a slight
temperature we ask people to
measure their body temperature twice uh
the sleep patterns change their movement
pattern change
because uti can combat like delirium and
then we have created a machine learning
model which has learned from examples we
have given
and basically detects the risk of uti
once we detect the risk
now the important part is how do we
communicate this information to
clinical teams we didn’t want to remove
human
from the clinical interactions we wanted
to help our clinicians to have better
information
to make more informed and efficient
decisions and a prioritized late task
to do that we have created a system
which we call it integrated view each
home will appear like a card
and these cards are dynamic depending on
the priority of the events we have if
something
serious happens they move always on top
left hand side of your screen you will
have the
highest priorities related to the cases
now our clinical team when they see a
system generates
the system generates an alert for
example say someone has
uti or hypertension they will click
and they will see a screen like this
this screen now will give them
uh all the information which we have
been collected for example
uh uh all the information we have been
collecting
uh for example they can click and see
the all the blood pressure for the past
two
two months plus two weeks depending on
the what they want they can look at all
the environmental information
but in reality what we want to do we
want to help machines to simplify this
task of decision making and make it much
more efficient
what happens is the machines when they
for example they detect someone has a
uti
or someone has hypertension they also
will give some explanation
like an algorithm will tell our
clinicians why i think this is a case of
uti and in that case they will have a
basically we have worked with our
clinicians to design clinical pass
pathways how to respond to these alerts
and in those cases sometimes a clinician
might need to
go and look at the the basically
background data the provenance
information to make uh
decisions when we started this project
uh most of the our work was like looking
for
identifying patterns and detecting the
cases but
more and more information we collect we
can now become more protective
because over time we have created
algorithms which we have seen the cases
before we are collecting more data
and we can train new ai and machine
learning models to extract his patterns
and to learn from the experiences
and examples we had before
also we can start personalizing this
model the model of activity detection
was an example that showed how you can
use
machines to learn something and
personalize it to
an individual we have for example models
looking at people’s vital signals
and then they learned what are the
people what are the
norms for an individual for example i
can be an upper boundary of blood
pressure
someone can be an uh a lower boundary uh
my doctor might think where i am is it
seems okay now for me
what we are interested in are changes if
my blood pressure
keeps changing enough from where i am
these are type of patterns we are
interested
in and training machines to be able to
pick up
part of the workouts that we have done
now we are creating a daily wellness
score because we are monitoring people’s
activities we are monitoring people’s
vital signals
if we create combine all this
information we can create like a daily
wellness
score and what happens is if you have
someone’s daily wellness
score over time you can now create also
models can become
protective you can see whether that that
trend is going towards
decline or towards improvement
uh in the next short video clip uh some
of my colleagues who
from our national health service have
been involved in designing and leading
this research they will talk about
their experience and my colleagues from
our technical team
and some of our user groups they will
speak about their
experience about this project it’s been
a fantastic opportunity it just
demonstrates
what can be achieved when sectors come
together in partnership
i think we will see a fully digitally
enabled nhs in the future
not only will it help us to understand
patients but it will move us towards
earlier diagnosis
and precision medicine techniques my
grandparents have dementia
although my my grandpa and father is not
involved in this project in a way i’m
thinking
and maybe what i do right now as a
research in future come up as a solution
that would also
help him it would be very nice to think
that this
is the norm for the future this is how
people
will be looked after in their own home
for as long as and for as much
time as possible
thank you
you