[Music]
so I’m an emergency doctor and I’m gonna
tell you about a patient that I saw in
the emergency room a few years ago it
was towards the end of a busy shift and
I’d made the mistake of thinking that I
might be able to get home early when an
ambulance arrived with a new patient and
it was an older gentleman who was having
trouble breathing
now you can learn a lot from just
watching someone breathe and so as the
ambulance guys were rattling off his
numbers I just watched and what I saw
made me very nervous
when most people breathe we use our
diaphragm and we use the muscles of our
of our chest wall when you have a hard
time breathing you also start using the
muscles up here that are called the
accessory muscles of breathing and your
shoulders and your neck when you get
tired of breathing it’s not like other
muscles you can’t just kind of give him
a break breathing doesn’t work like that
and when I watched my patient Bree and I
could see him getting tired I could see
every single muscle of his chest
standing out trying to get in there and
I knew that even though I didn’t know
what he had whether it was a pneumonia
or heart failure or something else I
knew he didn’t have long there’s no good
way to tell someone that they’re about
to die and so I just took a deep breath
and I said it his daughter had just
walked into the room and both he and his
daughter were devastated they were
heartbroken and who wouldn’t be but
there was something else as well which
is that they were surprised surprise has
a number of problems in this situation
the first and most obvious is that it
takes a tragic situation and turns it
into a terrifying situation for patients
and for families but but that’s not the
worst part the worst part is that
surprise means people are not prepared
and when people aren’t prepared they
haven’t had time to think surprised
people do not make good decisions
especially not about life or death
so I want you guys to imagine for a
moment that you’re living with a serious
illness maybe it’s heart failure maybe
it’s cancer and someone asks you when
you look ahead when the time comes what
does a good death look like to you if
you’re like most other people in this
country we have a lot of survey data and
so I already know what you’re gonna say
the data are very clear everyone wants
to die at home surrounded by family no
one wants to be hooked up to a machine
everyone wants to have had time to put
their affairs in order to make peace
with themselves and often with God and
that’s what people say when they have
time to look ahead but now let’s go back
to the ER imagine you’re my patient or
maybe it’s your dad
imagine someone’s just told you that
you’re a few heartbeats away from death
now quick decide do you want a breathing
tube and a medically induced coma do you
want to be hooked up to a ventilator
should I get you a bed in the ICU what
about a tight-fitting pressurised mask
as an alternative to the breathing tube
that might eventually diminish our
chances of putting in the breathing tube
effectively that’s a lot to think about
when you’re hyperventilating and
terrified and in shock and people in
shock do not make very good decisions so
I’ll tell you what usually happens in
these situations which is someone very
much like me will say you know what this
is really hard let’s just put in the
breathing tube let’s see how it goes
let’s put him in the ICU just for now
and that’s like you’re drowning and
someone throws you a life raft and it
sounds so reasonable that that’s what we
do so the patient goes up to the ICU and
one decision leads inexorably to another
decision let’s put in a permanent
surgical tube in the trachea to support
is breathing for a little longer while
we’re in the operating room let’s just
put in a feeding tube to support his
nutrition too let’s add on some
medications for pain because
looks uncomfortable I mean they’ll make
him confused but we can’t leave him like
this
after the ICU let’s plan on getting him
to a long-term hospital because that’s
really the only place where they can
manage all these tubes and he’s too weak
to go home one intervention leads to
more interventions and months later
patients are still cycling between
hospitals and nursing homes they’re
subjected to uncomfortable and invasive
procedures and along the way they
generate enormous costs costs that
threatened the entire viability of our
health system it’s not an outcome that
anyone wants but it happens all the time
one decision after another so what if we
could see further what if we had more
time to think more time to plan ahead
for the kinds of decisions we want to
make rather than the decisions we have
to make when it’s a split second and
we’re terrified when I was younger when
I was a teenage boy I read a lot of
science fiction and fantasy novels there
were some role-playing games I am a huge
nerd which you could probably tell and
so it was always very natural for me and
my research to start thinking about a
world of crystal balls a world where we
could trace someone’s illness all the
way from today to the day of their death
and I’ll tell you something about my
patient that makes this sound maybe more
like science and less like science
fiction so I remember when I was reading
through his medical record shaking my
head because I was reading about all of
the admissions to the hospital he had
over the year before all the visits to
his primary care doctor to the many
specialists every time a little bit
worse and I also remember and this is
actually a little embarrassing I
remember feeling angry or at least
frustrated because how could anyone be
surprised that this man was about to die
the signs were all there had the family
not been there for all the
hospitalizations and the new medications
and the worsening tests and what about
my colleagues surely they had seen this
coming had no one talked to him about
this before how is it my job to do this
in the ER
so all this is embarrassing in
retrospect because it is such an
unproductive way to think about this
problem from my perspective knowing how
the story ends it’s very easy to look
back at this gentleman’s medical record
and say oh the signs were all there we
should have seen it coming that’s a lot
harder in the other direction the man
that I saw in the emergency room was
very different from the man that my
patients had seen in their clinics one
day he’s a man who’s living with
multiple chronic illnesses spending time
with his family going to his doctor’s
appointments and the next day he’s
gasping for air in the ER the body is
incredibly complex and change happens
unpredictably and quickly it turns out
we’re just very bad at predicting the
future but you know who’s really good at
predicting the future have you ever
really liked the movie that’s been
suggested to you by Netflix
have you ever ended up buying a product
that you just happened to see on Amazon
so think about what Netflix and Amazon
know about you they get to know you
through a little collection of mouse
clicks and time you spend on screen and
now think about what the health care
system knows about you what happens
every time you walk into a doctor’s
office get a laboratory test get an MRI
it turns out that the same algorithms
that predict how much money you’re gonna
spend online or how much traffic you’re
gonna sit in later tonight on the way
home those algorithms are also really
good at predicting the day that you’re
going to die so how good are they I’ll
give you some numbers our algorithms can
trawl through a number of electronic
health record data sources from my
hospital system and we can identify a
pretty exact probability of death for a
number of different patients and we can
identify a group of high-risk patients
what does high risk mean when we follow
those patients up for six months after
we make our judgement 75% of the
have died same for low-risk when we
follow them up for the same time period
it’s about two tenths of one percent so
those are the kinds of numbers that can
really start changing the conversation
and how we think about the end of life
and how we spend those last month’s and
we can do that for cancer patients that
are starting chemotherapy or we can do
it for elderly patients seeing their
primary care doctors for routine visits
it’s all just data and the computer
doesn’t really care it’s just another
outcome to predict one question that I
that I often get from friends and
colleagues is sort of how does this work
what’s the what’s the secret and and I
thought what I what I do this afternoon
is I just tell you some secrets you
might be expecting me to tell you
something about the fancy master
algorithm we’re using or about how many
floating-point operations our computers
can do in a second but instead I’m going
to tell you about something much less
glamorous I’m going to tell you about
the nature of the data that we use so
here’s the first secret the first secret
is about the nature of death the great
thing about death and there aren’t that
many great things about death but sort
of you know if you bear with me
medically the great thing about death is
that it is very easy to measure
there’s there’s not a lot of miss
diagnosis of death I did have one
harvard medical student that i can tell
you guys about later who struggle a
little bit making the Psygnosis but for
the rest of us we don’t need a lot of
tests it’s not a it’s not a diagnostic
mystery and that’s actually quite
important because it makes death a very
different from other things that we
might also like to predict using
electronic health record data like
Alzheimer’s disease or appendicitis
there’s no judgment or diagnosis
involved in death it’s very easy to
measure and that’s what makes it very
easy for an algorithm to predict so
here’s the second secret it’s about the
variables that we use to predict another
question that I often get from from
friends and colleagues is what are the
variables that the model uses I mean
what are the variables that end up being
important in doing most of the work in
the model and that’s actually a very
human question to ask the human mind is
very good at focusing in on a couple of
things that really matter for predicting
and we’ve actually gotten a lot of
mileage out of that in medicine it’s how
we know that cigarettes lead to cancer
or how we know that high blood pressure
leads to stroke but algorithms don’t
need to focus in on a small number of
variables they don’t get distracted they
don’t they don’t forget and so which
variables does the algorithm use it just
uses all of them more variables than
could ever fit into a human mind let
alone variables that a human mind could
use effectively and if you think about
it that’s actually a very deep point one
of the things that makes me most excited
about artificial intelligence as
medicine is the ability to discover
truths that are too complex to fit into
our own heads the complexity of the
human body exceeds the capacity of the
human mind and artificial intelligence
will become a tool not just for
predicting outcomes and improving our
healthcare but for actually making
fundamental discoveries in health itself
so there are a lot of people who are
are much less optimistic than I am about
artificial intelligence and medicine and
those people asked a number of very good
questions well doctors ever use these
algorithms will they come to terms with
something that they don’t even
understand are they too proud or maybe
government’s the problem all these
complex regulations stifle change or
maybe health care organizations or it’s
you fossilised to really adapt to this
new environment and those are all good
questions and to answer them I’ve
actually got some good news and some bad
news and as tradition would have it I’ll
start with the good news
here’s the good news medicine has been
adopting new technology for a long time
first we adopted leeches then we adopted
penicillin and this year we adopted
immunotherapy and gene therapy
algorithms are just another new
technology for us to adopt it’s not our
first rodeo and there’s a playbook for
change all we need to do is follow the
PlayBook and that brings me to the bad
news we need to follow the PlayBook we
need to convince hospitals insurers
doctors and especially patients that
algorithms are more like penicillin and
less like leeches if anything our
problem in medicine is that we adopt too
many new technologies not not enough
technologies and this is going to
involve publishing in medical journals
it’s going to involve running randomized
trials it’s going to involve getting
regulatory approval it is a ton of work
but that’s the playbook so I should tell
you what happened to my patient I told
him that if you were my dad I wouldn’t
put in the breathing tube and that I’d
give him medication that he needed to
make him more comfortable but I’d keep
him as alive as I could until his family
came in and that’s what we did
his family arrived they held his hand we
turned off all the alarms and he died in
the ER and while that might be better
than a long drawn-out process
cycling in and out of hospitals I don’t
think anyone would call that a good
death but you know that was a few years
ago and so it’s not so hard to imagine a
few years from now things going very
differently if over that time we managed
to get intelligent predictions into the
hands of more patients and more doctors
so here’s what could have happened six
months before his doctors could have had
the conversation with him he could have
used that time to get his affairs in
order to reconcile with his strange
family members to arrange for care at
home so that when he got short of breath
he didn’t call 9-1-1 he called a nurse
who came over and helped with his
symptoms his family came over to and he
and his family could spend his last few
moments together focusing on being human
in one of life’s most human experiences
and picturing that it’s really hard to
understand why so many people are so
worried about algorithms dehumanizing
medicine in some way those people have
never seen the inside of the healthcare
system we do a perfectly good job of
dehumanizing people without algorithms
and so I believe that algorithms are
going to do the opposite they’re going
to let us refocus on what’s most
important and what’s most true about our
human nature
they won’t dehumanize us they will
recommand as us thank you very much
[Applause]
so I’m an emergency doctor and I’m gonna
tell you about a patient that I saw in
the emergency room a few years ago it
was towards the end of a busy shift and
I’d made the mistake of thinking that I
might be able to get home early when an
ambulance arrived with a new patient and
it was an older gentleman who was having
trouble breathing
now you can learn a lot from just
watching someone breathe and so as the
ambulance guys were rattling off his
numbers I just watched and what I saw
made me very nervous
when most people breathe we use our
diaphragm and we use the muscles of our
of our chest wall when you have a hard
time breathing you also start using the
muscles up here that are called the
accessory muscles of breathing and your
shoulders and your neck when you get
tired of breathing it’s not like other
muscles you can’t just kind of give him
a break breathing doesn’t work like that
and when I watched my patient Bree and I
could see him getting tired I could see
every single muscle of his chest
standing out trying to get in there and
I knew that even though I didn’t know
what he had whether it was a pneumonia
or heart failure or something else I
knew he didn’t have long there’s no good
way to tell someone that they’re about
to die and so I just took a deep breath
and I said it his daughter had just
walked into the room and both he and his
daughter were devastated they were
heartbroken and who wouldn’t be but
there was something else as well which
is that they were surprised surprise has
a number of problems in this situation
the first and most obvious is that it
takes a tragic situation and turns it
into a terrifying situation for patients
and for families but but that’s not the
worst part the worst part is that
surprise means people are not prepared
and when people aren’t prepared they
haven’t had time to think surprised
people do not make good decisions
especially not about life or death
so I want you guys to imagine for a
moment that you’re living with a serious
illness maybe it’s heart failure maybe
it’s cancer and someone asks you when
you look ahead when the time comes what
does a good death look like to you if
you’re like most other people in this
country we have a lot of survey data and
so I already know what you’re gonna say
the data are very clear everyone wants
to die at home surrounded by family no
one wants to be hooked up to a machine
everyone wants to have had time to put
their affairs in order to make peace
with themselves and often with God and
that’s what people say when they have
time to look ahead but now let’s go back
to the ER imagine you’re my patient or
maybe it’s your dad
imagine someone’s just told you that
you’re a few heartbeats away from death
now quick decide do you want a breathing
tube and a medically induced coma do you
want to be hooked up to a ventilator
should I get you a bed in the ICU what
about a tight-fitting pressurised mask
as an alternative to the breathing tube
that might eventually diminish our
chances of putting in the breathing tube
effectively that’s a lot to think about
when you’re hyperventilating and
terrified and in shock and people in
shock do not make very good decisions so
I’ll tell you what usually happens in
these situations which is someone very
much like me will say you know what this
is really hard let’s just put in the
breathing tube let’s see how it goes
let’s put him in the ICU just for now
and that’s like you’re drowning and
someone throws you a life raft and it
sounds so reasonable that that’s what we
do so the patient goes up to the ICU and
one decision leads inexorably to another
decision let’s put in a permanent
surgical tube in the trachea to support
is breathing for a little longer while
we’re in the operating room let’s just
put in a feeding tube to support his
nutrition too let’s add on some
medications for pain because
looks uncomfortable I mean they’ll make
him confused but we can’t leave him like
this
after the ICU let’s plan on getting him
to a long-term hospital because that’s
really the only place where they can
manage all these tubes and he’s too weak
to go home one intervention leads to
more interventions and months later
patients are still cycling between
hospitals and nursing homes they’re
subjected to uncomfortable and invasive
procedures and along the way they
generate enormous costs costs that
threatened the entire viability of our
health system it’s not an outcome that
anyone wants but it happens all the time
one decision after another so what if we
could see further what if we had more
time to think more time to plan ahead
for the kinds of decisions we want to
make rather than the decisions we have
to make when it’s a split second and
we’re terrified when I was younger when
I was a teenage boy I read a lot of
science fiction and fantasy novels there
were some role-playing games I am a huge
nerd which you could probably tell and
so it was always very natural for me and
my research to start thinking about a
world of crystal balls a world where we
could trace someone’s illness all the
way from today to the day of their death
and I’ll tell you something about my
patient that makes this sound maybe more
like science and less like science
fiction so I remember when I was reading
through his medical record shaking my
head because I was reading about all of
the admissions to the hospital he had
over the year before all the visits to
his primary care doctor to the many
specialists every time a little bit
worse and I also remember and this is
actually a little embarrassing I
remember feeling angry or at least
frustrated because how could anyone be
surprised that this man was about to die
the signs were all there had the family
not been there for all the
hospitalizations and the new medications
and the worsening tests and what about
my colleagues surely they had seen this
coming had no one talked to him about
this before how is it my job to do this
in the ER
so all this is embarrassing in
retrospect because it is such an
unproductive way to think about this
problem from my perspective knowing how
the story ends it’s very easy to look
back at this gentleman’s medical record
and say oh the signs were all there we
should have seen it coming that’s a lot
harder in the other direction the man
that I saw in the emergency room was
very different from the man that my
patients had seen in their clinics one
day he’s a man who’s living with
multiple chronic illnesses spending time
with his family going to his doctor’s
appointments and the next day he’s
gasping for air in the ER the body is
incredibly complex and change happens
unpredictably and quickly it turns out
we’re just very bad at predicting the
future but you know who’s really good at
predicting the future have you ever
really liked the movie that’s been
suggested to you by Netflix
have you ever ended up buying a product
that you just happened to see on Amazon
so think about what Netflix and Amazon
know about you they get to know you
through a little collection of mouse
clicks and time you spend on screen and
now think about what the health care
system knows about you what happens
every time you walk into a doctor’s
office get a laboratory test get an MRI
it turns out that the same algorithms
that predict how much money you’re gonna
spend online or how much traffic you’re
gonna sit in later tonight on the way
home those algorithms are also really
good at predicting the day that you’re
going to die so how good are they I’ll
give you some numbers our algorithms can
trawl through a number of electronic
health record data sources from my
hospital system and we can identify a
pretty exact probability of death for a
number of different patients and we can
identify a group of high-risk patients
what does high risk mean when we follow
those patients up for six months after
we make our judgement 75% of the
have died same for low-risk when we
follow them up for the same time period
it’s about two tenths of one percent so
those are the kinds of numbers that can
really start changing the conversation
and how we think about the end of life
and how we spend those last month’s and
we can do that for cancer patients that
are starting chemotherapy or we can do
it for elderly patients seeing their
primary care doctors for routine visits
it’s all just data and the computer
doesn’t really care it’s just another
outcome to predict one question that I
that I often get from friends and
colleagues is sort of how does this work
what’s the what’s the secret and and I
thought what I what I do this afternoon
is I just tell you some secrets you
might be expecting me to tell you
something about the fancy master
algorithm we’re using or about how many
floating-point operations our computers
can do in a second but instead I’m going
to tell you about something much less
glamorous I’m going to tell you about
the nature of the data that we use so
here’s the first secret the first secret
is about the nature of death the great
thing about death and there aren’t that
many great things about death but sort
of you know if you bear with me
medically the great thing about death is
that it is very easy to measure
there’s there’s not a lot of miss
diagnosis of death I did have one
harvard medical student that i can tell
you guys about later who struggle a
little bit making the Psygnosis but for
the rest of us we don’t need a lot of
tests it’s not a it’s not a diagnostic
mystery and that’s actually quite
important because it makes death a very
different from other things that we
might also like to predict using
electronic health record data like
Alzheimer’s disease or appendicitis
there’s no judgment or diagnosis
involved in death it’s very easy to
measure and that’s what makes it very
easy for an algorithm to predict so
here’s the second secret it’s about the
variables that we use to predict another
question that I often get from from
friends and colleagues is what are the
variables that the model uses I mean
what are the variables that end up being
important in doing most of the work in
the model and that’s actually a very
human question to ask the human mind is
very good at focusing in on a couple of
things that really matter for predicting
and we’ve actually gotten a lot of
mileage out of that in medicine it’s how
we know that cigarettes lead to cancer
or how we know that high blood pressure
leads to stroke but algorithms don’t
need to focus in on a small number of
variables they don’t get distracted they
don’t they don’t forget and so which
variables does the algorithm use it just
uses all of them more variables than
could ever fit into a human mind let
alone variables that a human mind could
use effectively and if you think about
it that’s actually a very deep point one
of the things that makes me most excited
about artificial intelligence as
medicine is the ability to discover
truths that are too complex to fit into
our own heads the complexity of the
human body exceeds the capacity of the
human mind and artificial intelligence
will become a tool not just for
predicting outcomes and improving our
healthcare but for actually making
fundamental discoveries in health itself
so there are a lot of people who are
are much less optimistic than I am about
artificial intelligence and medicine and
those people asked a number of very good
questions well doctors ever use these
algorithms will they come to terms with
something that they don’t even
understand are they too proud or maybe
government’s the problem all these
complex regulations stifle change or
maybe health care organizations or it’s
you fossilised to really adapt to this
new environment and those are all good
questions and to answer them I’ve
actually got some good news and some bad
news and as tradition would have it I’ll
start with the good news
here’s the good news medicine has been
adopting new technology for a long time
first we adopted leeches then we adopted
penicillin and this year we adopted
immunotherapy and gene therapy
algorithms are just another new
technology for us to adopt it’s not our
first rodeo and there’s a playbook for
change all we need to do is follow the
PlayBook and that brings me to the bad
news we need to follow the PlayBook we
need to convince hospitals insurers
doctors and especially patients that
algorithms are more like penicillin and
less like leeches if anything our
problem in medicine is that we adopt too
many new technologies not not enough
technologies and this is going to
involve publishing in medical journals
it’s going to involve running randomized
trials it’s going to involve getting
regulatory approval it is a ton of work
but that’s the playbook so I should tell
you what happened to my patient I told
him that if you were my dad I wouldn’t
put in the breathing tube and that I’d
give him medication that he needed to
make him more comfortable but I’d keep
him as alive as I could until his family
came in and that’s what we did
his family arrived they held his hand we
turned off all the alarms and he died in
the ER and while that might be better
than a long drawn-out process
cycling in and out of hospitals I don’t
think anyone would call that a good
death but you know that was a few years
ago and so it’s not so hard to imagine a
few years from now things going very
differently if over that time we managed
to get intelligent predictions into the
hands of more patients and more doctors
so here’s what could have happened six
months before his doctors could have had
the conversation with him he could have
used that time to get his affairs in
order to reconcile with his strange
family members to arrange for care at
home so that when he got short of breath
he didn’t call 9-1-1 he called a nurse
who came over and helped with his
symptoms his family came over to and he
and his family could spend his last few
moments together focusing on being human
in one of life’s most human experiences
and picturing that it’s really hard to
understand why so many people are so
worried about algorithms dehumanizing
medicine in some way those people have
never seen the inside of the healthcare
system we do a perfectly good job of
dehumanizing people without algorithms
and so I believe that algorithms are
going to do the opposite they’re going
to let us refocus on what’s most
important and what’s most true about our
human nature
they won’t dehumanize us they will
recommand as us thank you very much
[Applause]
Please follow and like us: