Florence Nightingale, Artificial Intelligence, and the Future of Health Care – James Scott, Statistics and Data Sciences

This is so session that everyone has been looking forward to, because who knows what

the connection is between Florence NIGHTINGALE and artificial intelligence.
It takes almost intellectual acrobats to connect the
two much starboard. It’s going to introduce our speaker.
I’ve once introduced Mike himself as being the most famous
mathematician at the University of Texas. And then someone said that’s an insult to the mathematics
department. Mike? Oh, I’ve found
that pretty insulting right away. OK. Well,
in any case, my job not only to defend myself, but to introduce James Scott here.
So James James is a wonderful scholar. Bayesian
statistician who was started at the University of Texas at Austin
as an undergraduate. And I had him and I had the pleasure of having him into my classes at least
two to two classes, and as did Roger. And
and he has gone on to have a wonderful career at a Marshall Scholarship, went to England on a Marshall Scholarship.
He has came back out of PHC at Duke and has been a professor here in the business school
and in the Department of Statistics and Data Science in the College of Natural
Sciences. He’s won many awards, including international awards and
as well as an NSF career award, the Savage Award, which is a by the one
given a year by the International Society of Bayesian Analysis for his doctoral
work and by Ari Award for Early Career Research Achievements in Bayesian
statistics. He’s really accomplished. But what I like, among many things I like about him
is that he’s also accomplished in other areas, such as teaching. He’s won the UC System Regents
Outstanding Teaching Award. How much did you get for that? Got twenty five thousand dollars.
Was that not? Not quite. How much? How much? Something like that. Something like that. Yeah.
I just wanted him to share. But no. No. Do I hear anything? But. But, but I
wanted to say one one, not one less. You know, form one, which is
one of the things I admire, among many things I admire about James, as is his.
His strategy of taking advantage of things and then taking them to another
level. So while he was an undergraduate student here at u._t, he was in my among other classes,
my number theory class. And for the only time in the history of my teaching at
U.T., he organized a collection of students in the class and he said,
we’d like to learn more. No theory than you’re offering. Could we come in on Tuesday and Thursday
and have an other sessions? And so a group of have’nt led by him and three or four
others came in every Tuesday and Thursday for the last half of the semester and learned things beyond the
course. There’s no credit. You know, just good. So it was a
and I think he’s done this throughout his his career. And it’s and
it’s great. The other thing that is great is that he is taking his work and taking it beyond the academy.
I think that our academia is guilty of inward
looking. And and the fact that he has written for people outside
the academy is really important. He’s written a new book called a-I. Q How People and Machines are Smarter,
together with coauthor Polson, which is coming out
next month, next month by a Macmillan press. And I wanted to read a blurb
by Stephen D. Levitt, who is the coauthor of Freakonomics, you know, Freakonomics.
And he said of this coming book, There comes a time in the life of a subject when someone steps up
and writes the book about it. Q That’s the book that he just wrote, explores
the fascinating history of the ideas that drive this technology of the future and demystify demystifies
the core concepts behind it. The result is a positive and entertaining look at the great potential
unlocked by marrying human creativity with powerful machines. So I
think we’re all looking forward to reading his work when it comes out next month. And so it’s my great pleasure
to introduce James Scott.
Thank you so much, Roger, for inviting me here, it is always such a pleasure to be among friends and colleagues
back in British studies to see many familiar old faces. Thank you very much, Mike, for that kind introduction.
Mike got out. I’ll say a little bit about the book before I kind of launch into the subject of today’s talk. I
asked my coauthor, Nick Paulson, and I thought of this book first as a way to answer all of the great questions
that our students had about artificial intelligence. Things like how does a self-driving car work? How does an Amazon
echo understand what I’m saying? That sort of thing. We noticed that there was a lot of writing about
A.I. out there that was very technical. A lot that was kind of fizzy pop, sociology
and then a lot that was borderline science fiction of the Elon Musk. The robots
are coming for you variety. But if you wanted the non-technical version of how this stuff actually works, then
you were stuck. But then along the way we realized that the public narrative surrounding
artificial intelligence were broken. On the one hand, you have nothing but hype about A.I.
coming from the business world. You know, if you believed all of those IBM Watson ads doing during the
Super Bowl, you would come away believing that A.I. is going to fix the health care system and sell more
Cheerios and make your toilet smell like roses and and so on. But then on the other side,
you have folks claiming that A.I. is going to destroy everything we care about from our jobs toward democracy
to our privacy. With A.I., we’ve clearly reached the point where a non-expert can’t
tell the difference between hype and the reality. So as educators, Nick and I just came back to a simple bedrock
that if you want to participate in the great debates of the 21st century, it’s really important. Three people have a sense of
what’s hot air and what’s genuine promise when it comes to A.I. And they can’t do that without understanding how
the underlying technologies actually work. In particular, they can’t do that without understanding
the role of data in A.I., which is why we wrote a i._q. I now tell you a quick story here that will
that will give you some sense of how naive I was in dealing with a mainstream nonacademic press
for the first time. When we talk with our American publishers about cover designs, we told
them we wanted something simple and elegant. Academic. Right. And they came up with this, which I really
like to write. So Nick and I thought it was great. We then asked a British publisher if they could do
something similar and they said, yeah, sure, sure. We’ll do something very much along the same lines. And then finally in
mid-March, I got an email from the British publisher saying that this was their idea of something
similar. Right. So so I told them that I hated it.
In fact, when I saw the email, I was in an airport at the time. And my reaction to this design
was to point out that I could find exactly one book in the airport bookstore that was had
a cover that was in such a lurid shade of yellow. And I didn’t think it offered a very flattering comparison.
So my wife was laughing me, by the way, she was in the airport, too, as all this happened. And she said, you’re color
blind, what do you care? And I said, I’m not that color blind. So, you know, in the
end, the publisher was pretty firm about the need for a design that pops or some nonsense like that.
So if you see a i._q in America, you’ll get the simple, elegant cover. And if you see it in a Commonwealth country somewhere,
it will be despite my protests in that very fluorescent shade of yellow. OK, so I
will I will stop complaining about yellow now, because I did come here to talk about a serious topic,
which is health care. Now, if you read the stories about health care these days, you will encounter
two very different narratives. First, there’s the bad news, which is that health care systems
across the rich world are an awful shape. Obesity and heart disease are up. Costs are spiraling
out of control. In 2016, two thirds of all British NHS trusts ran a deficit.
Americans, meanwhile, spend far more of their GDP on health care than anyone else and aren’t
any healthier to show for it. Doctors tend to spend their days, at least in America, sweating lawsuits,
fighting insurance companies and typing data into an electronic health records system. Compared
to non-doctors, they are 40 percent more likely to abuse alcohol or drugs and twice as likely
to commit suicide. But then, perhaps as an antidote to all of these depressing stories,
we are also told that artificial intelligence is set to transform health care. a-I evangelist’s
describe a futuristic world where your surgeon is assisted by a laser guided robot, just like the Google
car, where your vital signs are algorithmically monitored for anomalies, just like your
credit card and where your treatments are personalized. Jeff, just like your Netflix account,
it’s a world where your smartwatch can tell you whether you’re going into labor and where you can snap a picture of
a skin lesion with your smartphone and get an instant diagnosis. In this world of the future,
doctors no longer spend a third of their time doing manual data entry. Instead, they
tell everything to a sort of Amazon echo on steroids, which immediately updates your medical record.
It is a future where a technology update accessible through smartphone brings
better health care to underserved communities. Probably first here in. Rich world. And then eventually in the developing
world, it is a future where childbirth becomes safer, where diseases
are caught earlier and where oceans of human potential reach full tide.
So here’s the question I’d like to address today. Why aren’t we there already? After
all, each of the technologies I just listed already exists in some form or another.
And it is dead obvious what’s needed in order to prompt their widespread adoption. We need better data.
We need deeper collaboration between health care professionals and data scientists. And we need smarter laws
that can foster innovation and yet still safeguard patients and their privacy. But as I’ll argue
today. Just because something good can be done with data doesn’t mean it will be done.
If you look across the spectrum of human activity, I claim that health care is the one area where artificial
intelligence could probably do more good than anywhere else. And yet the grim reality today, at least,
is that we are still likely years away from seeing our most advanced A.I. technologies used to help real
patients in substantial numbers. And I’m not talking about speculative future technologies. I’m talking
about stuff that exists right now. For example, here is a smartphone
app designed by researchers at Stanford. You can snap a picture of a skin lesion and it will using
something called a deep neural network, classify it into one of over two thousand different types
of skin lesions. And we’ll do so. Moreover, with accuracy comparable to a panel of 23 board certified
dermatologists. I’m talking about something developed here at the University of Texas
by a chemistry professor named Livia Evelin that made the news. This is the BBC. They call it the mass spec
pen. It’s a pen that you can insert into a tissue during cancer surgery. And within 10 seconds
it will run mass spectrometry and tell you whether that tissue is cancerous or healthy, which can really help
tell you which parts of the tissue to resect during cancer surgery. I’m talking about epidermal electronics.
It’s a little tattoo, no thicker than the width of human skin. You can see for scale right here.
Here’s a person’s wrist. Here’s the little epidermal electronic with EKG and EEG sensors,
with temperature and hydration sensors with wireless technology that you could hook it up to your cell phone
program with algorithms that can monitor your vital signs for anomalies. This is stuff that exists right
now. And the reasons why we aren’t seeing widespread adoption of these technologies that have
nothing to do with science or computing power, statistics and everything to do with culture incentives and bureaucracy.
Healthcare systems in America, Europe and Asia different important ways, but they all share some similarities
in terms of how I could help and why it isn’t helping already.
As I like to put it, cancer and kidney disease have no nationality. But there is a word for bureaucracy
in every language. Now I know that historians tend to be suspicious of analogies, but
I am not in the story and I’m a data scientist and I spend my days using health care data
to help doctors and patients in that role. It really helps me at least to seek an historical
example of someone who faced a similar problem and overcame it. Someone who possess the knowledge,
the stature and the moral authority to stand up to the powerful people who run health care systems
and say basically, get your act together. In my example for that is Florence NIGHTINGALE.
You all surely know NIGHTINGALE as the most famous nurse of all time. The lady with the lamp
who tended the wounded British soldiers of the crimeand war. But when she wasn’t caring for soldiers,
NIGHTINGALE was also a skilled data scientist who successfully convinced hospitals
that they could improve health care using statistics. In fact, there’s no other data science scientist
in history who can claim to have saved as many lives as NIGHTINGALE. And as a result of her achievements in 1859,
she became the first woman ever inducted into the u.k.’s Royal Statistical Society. Nightingale’s
path to unlocking the power of health care data offers some really good lessons for today in her
quest to bring data analysis to health care in the 1850s. She fought entrenched interests
that defended the status quo against reforms that could help patients. And the fight to do the same thing today
is playing out in a shockingly similar way. Now, I know that some of you in the room are Victorian ists
and even some of you who aren’t. We’ll still know a lot more about Florence Nightingale’s biography than I do.
But for those of you who are not experts in this area like me, I want to give a brief bit of background
on Nightingale’s life here. And I hope I won’t give you any calls to tell me I’ve gotten something wrong. So
Knightdale became famous primarily as a result of her experience as a nurse during the crimeand war.
Britain first sent troops to Crimea in the spring of 1854 to lay siege to Sebastopol, the main harbor
for Russia’s Black Sea fleet. And people back in London had assumed that the war would be over quickly,
which sounds awfully familiar, but there would be no quick victory. And it soon became clear
that the British Army, which was a generation removed from its last major war against Napoleon
in 1815, was not at all prepared to face the Russians. And nowhere
was this more obvious than in the Army’s decaying medical system, where. Basic matters
of supply chains and sanitation were thought to be beneath the dignity of the medical man in charge.
And the result of all this poor planning was predictably a logistical and humanitarian catastrophe.
A soldier wounded in the Crimea would find himself packed onto a grimy ship. Here’s the Crimea.
Here’s the field of battle packed onto a grimy ship and sent 300 miles away to the Barrack
Hospital at Scutari, which was on the Anatolian side of the Bosphorus, opposite Constantinople.
And there the soldier might wait as long as three days on the ship to be taken ashore, where he would be
loaded on a stretcher or maybe strapped to a mule for this kind of jarring climb up a steep
hill to the hospital right there in the hospital’s filthy and I mean filthy.
Rats crawled over the injured soldiers who lay sprawled on thin mats. Cholera
and dysentery were rampant. The sewers were clogged. The toilets leaked excrement into the main courtyard,
and a water main was blocked by a decomposing carcass of a dead horse. Who knows how it got
in there? The hospital was badly short of medical supplies, clean clothes, healthy
food, you name it. Many amputations were even done without chloroform. By the autumn of
The September 30th editorial in the Times channeled the public’s growing outrage. It went as
follows. Not only are the men left to expire in agony, unheeded and shaken off the
catchin desperately at the surgeon whenever he makes his rounds through the fetid ship. But now, when they
are placed in the hospital where we were led to believe that everything was ready, which could ease their pain or facilitate
their recovery. It was found that the communist appliances of a workhouse psych ward are wanting.
Well, as you can imagine, Sidney Herbert, who is secretary of war at the time, came under enormous public
pressure. He was a family friend of the Nightingale’s, and he had seen her rapid rise in the field
of nursing. And so he asked her to lead a government sponsored group of nurses to Scutari to assist
the doctors and to tend to the suffering soldiers. Florence agreed, and she steeled herself
for the worst. But really, nothing could have prepared her for the condition she found upon her arrival
for miles of corridors with the conditions I described to you men in sleeping 18
inches apart and their lives made miserable by what NIGHTINGALE in her diary called foul air
and preventable mischiefs, the hospital supply chain had broken down completely.
NIGHTINGALE could find no linen to make bandages. She couldn’t find fresh shirts to replace
those soaked with blood. There are plenty of gangrene, lice, bugs and fleas. Yet, as NIGHTINGALE
wrote to a friend back home, no mops, no plates, no wooden trays, no slippers, no knives
and forks, no scissors for cutting the men’s hair, which is literally alive, no basin’s,
no toweling, no chloride of lime. She soon learned that her requests for provisions
had to pass through no less than eight different government departments back in London. And when those requests
were finally processed, sometimes the wrong supplies were sent or the right supplies were sent to the wrong place
at Squitieri itself. NIGHTINGALE encountered nothing but dawdling and obstruction from the chief purveyor.
Matters got so bad that she asked the times to entrust her with the donations it had collected for Soldiers Fund.
That way, she could bypass the chief purveyor and go shopping for necessities herself and the Grand Bazaar of Constantinople.
And after that, she effectively became the shadow purveyor at school thirty. as the conduit for the
enormous variety of goods that civilians sent to Scutari. Things like food,
cash, slippers, linens, a drying cupboard even. I noted in reading about this raspberry
preserves and ginger biscuits from one Mrs Gallop, Buckinghamshire. God bless her.
She soon found herself charged with reorganizing virtually every non-medical function at the
hospital. She described her role as cook, housekeeper, scavenger washer, woman, general
dealers’ storekeeper, the effort tireder to the bone. She worked 20 hour days. She took meals
on her feet. She was exhausted by, as she put it in a letter home. The quantity of writing, the quantity
of talking, the dealing with the selfish, the mean. I feel like Prometheus bound to the
rock of ignorance and incompetency. Yet all the while, she was making a difference.
Only two months after her arrival, the hospital chaplain noted in a letter, a surprising
air of comfort and enjoyment. There were Stow’s on every ward. There were tin baths in every corner.
Every man had a bed, a clean mattress and a change of shirt twice a week. And mortality was dropping,
having peaked at a shocking 52 percent of admissions in the winter of 1855.
It had fallen to 20 percent by March and thereafter continued downward to the following winter.
By which point it had reached the level of mortality at no higher than the rate among civilians in a major
city. Now, Florence NIGHTINGALE could hardly take all of the credit for this herself, and she never tried
to. Still, for more than a year, the hospital at school three had been a ship,
barely surviving the gale. And in the words of an army colonel who’d. First things firsthand,
Miss NIGHTINGALE was its only anchor in their letters home, her colleagues noted
her energy, her example, her way of cutting through red tape with a machete. They recall the darkest
days of winter when wounded troops arrived by the hundreds. And when winter, as one fellow noted nurse put
it, the officials lost their heads, crying out to flow for this and that. And they also
recalled the chaos that reigned during Nightingale’s brief absences, like the one day in 1854
when she took a brief rest from her duties as shadow purveyor. And when the man of sea corridor therefore all ended
up drunk because they had guzzled their wine straight from the bottle, as no one had given them any cups from
the store cupboard. Back in Britain, the famous journalist of the Times
conveyed the image of NIGHTINGALE that would endure forever. And it was this when all the medical men
he wrote have retired for the night, and silence and darkness have settled upon those miles of prostrate
sick. She may be observed alone with a little lamp in her hands, making her solitary
rounds and with time. Of course, the NIGHTINGALE legend only grew up. Poems and sentimental
songs were written about her soldiers, private diaries of the day recorded daydreams
of leaping to her aid in the face of some imaginary danger. Ships, race horses, babies
of every social class were named in her honor. But NIGHTINGALE herself called this reputation,
and I quote, nothing but a false popularity based on ignorance. She actually
believed that her work back in London far after the war was over ultimately made a much bigger difference.
And modern historians largely agree with her.
Now, much of the historical work on NIGHTINGALE concerns her legacy in the field of nursing, specifically her
role in the decades long period of reform in the training and certification of nurses that took place
the middle of the 19th century. But here I want to focus on a different part of Nightingale’s legacy, which is her legacy
as a data scientist, not as a nurse. A big part of that is her
personal analysis of medical statistics from the crimeand war. NIGHTINGALE It’s fair to say
it was really into math and statistics. A lot has been written about how she aspired to be
a nurse from a very young age, about how she treated injured dogs, about how she nursed a cow
with a bad cough in the field next to her house, how she visited the sick and the dying of the village
nearly every evening as a teenager, but also from a very young age. She was precociously
talented at math. As a child, she played mathy word games. I took breath
and I made forty words, she wrote in her diary. At age seven, her parents letters talked
of how and how NIGHTINGALE positively threw herself into her math book as a child solving
world problems from a vanished age. I’ll give you an example of one of Florence’s Victorian era word problems.
If there are six hundred millions of heathens in the world, how many missionaries are needed to supply one to every
twenty thousand. But as a teenager,
she she learned geometry by reading Euclid herself. She learned logarithms from her cousin
Henry, who studied mathematics at Trinity College, Cambridge, and she once begged her parents to give a visit to
her uncle Octavius for the simple reason that he had a fantastic math library. Moreover, all
of this mathematical ability was married to an incredibly strong power of will. Will,
in fact, that her sister pathetic be called the most resolute and iron thing I ever knew.
As a young adult, Florence would awake as early as 3:00 a.m. to read anything statistical she could find
on social welfare. Minutes from Parliament. Data from the census. A report on the sanitary
conditions of the laboring classes of Great Britain. So, as you might imagine,
Florence came home from the scandal of Scutari, full of righteous indignation.
In her diary, she wrote I stand at the altar of the murdered men, and while I live, I fight
their cause. And it absolutely was a fight against those in the army and
the medical establishment who stood in the way of change and defended the status quo, like Army Doctor
John Hall, for example, who dismissed NIGHTINGALE as and I quote, a petticoat
in pure use. And NIGHTINGALE brought all of her weapons to bear in that
fight, her intellect, her network of friends, her acid pen. I’ll give you some examples of that.
But above all, math and statistics, which she clearly viewed as the mightiest arrows
in her quiver. Nightingale’s first biographer called it?i Cook, nicknamed her the passionate
statistician, which really didn’t stick in the public’s imagination the way the lady with the lamp did for obvious
reasons, but did provide a far better description of how she changed the world for the better.
NIGHTINGALE was especially adept at using graphical representations of data data visualization
in modern parlance to draw the nation’s attention to the conditions that had prevailed in military hospitals.
As one of her colleagues put it, Nightingale’s pictures of data could affect through the eyes what we may fail
to convey to the brains of the public through their word proof. She even invented
a new kind of statistical figure, which I’ll show you here. The Polar Area or coxcomb diagram,
which here shows changes in mortality over time using a series of colored wedges. So it begins
here in March of 1850, rather April of 1854. And as you go
clockwise around the the the rows here, the size of each colored wedge
represents deaths due to a particular cause. And by far the largest pie slice
here is deaths due to preventable disease. And
as you can see, it peaks in the winter of eighteen fifty five down here in January 1855
into February, and then falls starting here in the spring of eighteen fifty five
and all the way around until eight feet to fifty six until the deaths due to disease are no different than
in a major city. So her analysis and these figures revealed that in the first seven
months of the crimeand campaign, British soldiers suffered a 60 percent mortality rate
from disease alone that was higher than Londoners had experienced during the great plague of sixteen sixty
five. And it was higher even than the rate of death among a population of civilians who have cholera.
It was literally safer to have cholera at home than to take your chances in the Crimea. And that was before you faced
a single enemy bullet. A nightingale referred to this as the finest experiment modern history
has seen. As to what given number may be put to death, it will by the sole agency of bad food
and bad air, an experiment she reckoned, that had sent sixteen thousand men to death. And
NIGHTINGALE did more than anything else in the wake of the crime war to bring these facts to the public’s attention.
A second very important data science legacy of NIGHTINGALE was her contribution to evidence. Based hospital
design, together with English statistician William Farr, she analyzed data from army hospitals
during peacetime, and she discovered that because of poor sanitation, the army’s rate of mortality
at home was twice that of a comparable civilian population. She called this situation criminal.
Remember her acid pan. I referred to earlier? No different than to take eleven hundred men out upon Salisbury
Plain and shoot them on account of her report. The Army Sanitary Committee visit
every army, barracks and hospital in England between 1858 and 1861, and they recommended concrete
steps that were taken to retrofit barracks and redesign hospitals, which produced an immediate drop
in disease related mortality across the British Armed Services. Her recommendation soon caught on in
the civilian world as well. Hospitals with long corridors and stuffy rooms came to be seen as
infection incubators. Her preferred model of hospital construction became the norm, which was
the pavilion style hospital, which had wings of light and abundant ventilation. These nightingale
wards, my sister in law, who’s a doctor in England tells me, are still popular to the extent that
NHS hospitals built in the 40s and 50s are almost all Knightdale words.
Finally, perhaps the least known of Nightingale’s data science contributions was her role in creating
a new standard of professionalism in the collection and analysis of medical data. Now,
I’m sure, sure, you’ve heard it said of generals that they are always fighting the last war, but a doctor
trying to learn from the experience of the crimeand war couldn’t have even done that. The medical
staff at School Three had collected no statistics. They had preserved very few case histories.
They had done almost no postmortem examinations in many cases. Sick men were loaded on
one side of the boat in the Crimea, shipped 300 miles and dumped off the other side of the boat dead. When they reach
Scuderi. NIGHTINGALE despaired at the fate of the soldiers. But she also found it
deeply discouraging that this scientific treasury had been lost to mismanagement. And upon returning to England
after the war, she also found that these failings were mirrored in the civilian system, that the country had no system
for the collection of even the most basic medical statistics. Recoveries, lengths of stay at a hospital.
Deaths due to different disease and so on. And even if there had been such a system, there would have been no way to compare
these statistics across hospitals because every hospital used its own idiosyncratic classification system
for disease. NIGHTINGALE saw this lack of attention to good health care data as
an emergency. She saw how the new discipline of statistics was transforming other fields
like astronomy and earth science. She also noted how Continental statisticians, most notably
the eminent Belgian adult Lay, who is one of her idols. We’re using these new statistical
tools to look at complex social science questions and crime demographic change. NIGHTINGALE
saw incredible potential in these new mathematical and statistical tools, but in her view, that required
much better health care data. And so to that end, she drew up a standard set of medical forms.
She obtained the endorsement of many of the world’s leading statisticians, and she urged the big hospitals in London
to begin using these standardised forms. She also lobby the government to begin collecting data on illness and
housing quality as part of the census. From top to bottom, her work on evidence based health care
clearly foreshadowed the coming a hundred and sixty years. Her ideas formed a clear model for the international
system of disease classification use today, which really is the bedrock for all of modern epidemiology
and medical data science. So finally, we come back to the
modern day and the questions about today’s health care system that I raised in the beginning.
I think the Nightingale’s three data science legacy’s all have very clear parallels today. And for me at
least, they also raise some very sharp questions. NIGHTINGALE spoke of the foul air and preventable mischiefs.
And while the air in hospitals today may be a lot less foul, there are I sure you mischiefs aplenty.
One big question left for me at least is this how should we staff and train a modern health care
team? After NIGHTINGALE, no hospital could function without nurses. And my question
is, when will the same be true of data scientists and experts in artificial intelligence who today
play almost zero day to day role in the practice of health care? A second question
is how evidence based hospital design should work today. NIGHTINGALE helped to establish new
sanitary standards hospitals re-engineered from the ground up in response to her findings.
But when will hospitals undergo another era of redesign to accomplish what is now possible using data science
and A.I.? When, in other words, will data hygiene be taken just as seriously as
patient hygiene? Finally, the most important question of all is how medical
statistics should be collected, shared, analyzed and used. We’ve improved a lot
in that department in the last hundred sixty years, thanks in no small part to Nightingale’s efforts, but as
I’ll argue here, we’ve improved only in certain ways, and we could be doing so very much
more in light of what’s happening outside health care. This is starting to look like a moral embarrassment.
We live in an age when Formula One race cars are monitored in real time by algorithms
and teams of engineers. When you’re movie watching, preferences are the subject of multi-billion
dollar A.I. operations like Netflix, and when your propensity to click on an ad for dog food
is analyzed on supercomputers using millions of variables and billions of data points.
Yet for the most part, we still rely on numbers that Florence NIGHTINGALE could have crunched with pen and paper to quantify
the risk that your kidneys will fail. And in some ways we haven’t improved at all. A 2017 paper
in the Journal of the Royal Statistical Society referred to Nightingale’s 1860 protocol for hospital
data collection as conceptually more complete than many systems today. Which leaves
me, and I think it should leave us all wondering when will medical data science move into the 21st
century? So I want to be clear upfront that this is not the fault of individual doctors
and nurses. It is the fault of the whole health care system. The plain fact is that in health care today,
most data simply goes to waste. It’s no different than those soldiers on the boat in Scutari.
It comes on one side of the boat is used to send you a bill and then basically gets dumped over the other
side of the boat dead. Now, to illustrate this point, I want to tell you a story. It is a story
of a man from somewhere on the East Coast, the United States. We’ll call him Joads, a real patient, but Joe’s not his real name,
who died at age 62 with chronic kidney disease. Joe’s story tells you a lot
about how the contemporary approach to medical data science is failing patients and why a combination of better
data curation and artificial intelligence could prevent so very much suffering.
Now, by his mid-forties, our patient Joe was already suffering from Type 2 diabetes and congestive
heart failure. Maybe his job was stressful. Maybe his diet and exercise habits were poor. We don’t know.
But whatever the mix of causes, they finally caught up with him. And a few weeks shy of his forty seventh birthday,
Joe had a stroke and was rushed to the emergency room. I just survived the stroke, and although
his blood pressure and diabetes at that point marked him as having a higher risk for kidney disease at some point the
future. For now, his kidneys tested fine. The standard measure of kidney function is something
called the Jaffar on, as you can pronounce this right. It is the glue Malula filtration rate
to the Jaffar. Joh’s Jaffar was estimated at ninety nine, which is well above the danger
zone. For some reference here at GFI 60 or below indicates mild to moderate loss
of kidney function and GFI 30 or below means severe loss. Joe is at ninety
nine. Over the next year, he made nine more trips to the emergency room for problems
of one kind or another. None of them explicitly related to his kidneys. On two of those occasions he was admitted
to the hospital, and in his kidney function was measured. His Jaffar was first ninety six and then ninety
five. A few months later, this decline was a little bit steeper than the expected 1 to 2
percent decline per year in a healthy patient. But still, each individual reading was way above the
threshold that doctors used to think about danger here. About a year after a stroke,
Joe started making regular trips to an outpatient clinic eight visits over 14 months.
Each time, the doctor ordered a routine series of tests and the lab techs entered the data on his kidney function
into a electronic health records database. The same database, I should add that his hospital
used his GFA our numbers Yo-Yo to bet between 60 and seventy five,
which was still above the threshold of 64 moderate loss of kidney function, but still quite a bit down
from the prior year reading of ninety nine. And on an unmistakable downward trend
at age forty nine. Two years after his first stroke, Joe was readmitted to the hospital, and his Jaffar was
measured at 54. Over the next several months, he made 10 more visits to the E.R.,
as well as a dozen more visits to the outpatient clinic. Joe was very sick at
this point, as you might imagine. A month before his 50th birthday, his GFI was measured at 40. Well
into the danger zone. Yet he received no treatment that might have prevented his slide toward kidney failure.
We can only speculate why, but one reason might be that test results sometimes take a couple of days to
come back from the lab, and by which point the patient might be already home, checked out of the emergency room, no longer
under the direct care of the doctor who ordered the test in the first place. Who knows? Bottom line is,
over the next three years, Joe had 20 further encounters with the doctor and his kidney function was dropping
at a scary rate below 30 by age 50, one below 20 by age 52. At
which point Joe was finally referred to a kidney specialist more than a year after his kidney function had fallen
below the level of 30. That typically triggers that kind of referral. But for Joe. Kidney function,
kidney failure, I should say it was by now inevitable. Three months after his first appointment
with the specialist his kidneys gave out, he was rushed to the emergency room. His 25th
such visits since his first stroke, his GFA was measured at twelve. His kidney
function had declined 34 percent per year each of the last five years from his initial reading
of ninety nine after the stroke. The doctors in the E.R. put him on emergency dialysis, which is
one of the most traumatic and expensive procedures on the books in medicine for the next
decade. Joe became what the insurance industry refers to as a super analyzer, which is just
management speak for an appallingly sick human being. One of the 5 percent of patients who account
for 50 percent of all health care spending in United States. And in Joe’s case, that meant severe
diabetes, stage five kidney disease, angina, vascular disease, inflammatory connective tissue disease,
along with a series of several heart attacks. His kidneys were tested one hundred and twenty four times over
that decade long period, which included twenty six more visits to the E.R. and nine to a kidney
specialist. His Jaffar bounced around, but it never again rose above 20. And
Joe died a week shy of the 60 third birthday, roughly 10 years after beginning dialysis.
So my question is, what did Jo die of? And in one sense, the answer is clear. His
kidneys failed. But in order for that to happen, something else had to fail first in a manner so complete
that to me at least, it beggars belief. For if you take all of Joe’s GFI readings the eight years
after his stroke and you plot them over time and a scatterplot, that’s the trend right there
in those three years of steep decline between ages forty seven to fifty. When he falls
below that level four specialist referral, not a single one of Joe’s
health care providers had looked at a simple scatterplot of his GFI readings over time. The issue was
quite literally one of failing to connect the dots. Doing so would have yield a simple
and I think to all of us obvious prediction. This guy’s kidney function is declining
so rapidly that it is almost surely going to keep declining. And if it does, the result is going to be very painful and
very expensive. So my conclusion is that, yeah, Joe certainly died for wanting a kidney. But more
fundamentally, he died for want of a scatterplot. My question is, how could this have happened?
I actually had a discussion with about this topic with a friend and colleague of mine named Dr. Katherine
Heller, who is a professor of statistics and machine learning at Duke University. It was Katherine who brought Joe’s
attention to a case to my attention in the first place. And I will
quote for you what she told me. In retrospect, she said that steep decline
between ages 47 and 50 represents such an obvious missed opportunity.
All you have to do is draw a straight line through the cloud of data points and you can see where things are going.
So why did no one, whether human or machine, draw that straight line?
This is the essential question in modern health care. To understand the answer, we have to revisit
two earlier questions that Florence NIGHTINGALE asked 160 years ago when she pondered how the new mathematical
tools of the 1850s might be used in hospitals of that age. First, how does the health
care system use data today? And second, in light of new data analysis technologies, what
could it be doing instead? So today, the main way the health care system uses data is
to create checklists. These checklists and code the standards of care of medical bodies, things
like the American Medical Association, the u.k.’s General Medical Council. And these standards of care
in turn are driven by data from published research findings about what warning signs to look for,
about what treatments actually work, about what diagnostic protocols help the most people. That kind of thing.
Now, look, I think medical checklists are great. To me, the way they’re created and updated represents a triumph
of data over anecdote, which is something that a Florence NIGHTINGALE were around today. She could take immense pride in
checklists, save lives by helping doctors catch subtle clues when making complex decisions.
Some of you may have even read the checklist manifesto by the surgeon to go on day about how checklists can
help make complex decisions everywhere, not just in medicine. And I could recommend the book go on. They makes
a fantastic case, but I will also point out that checklists can fail, especially when
they rely on what Katherine Heller in our conversation about Joe’s case called threshold
thinking. So to see that, let’s return to the trend that’s really obvious from the scatterplot of Joe’s
kidney readings. So Heller to me in our conversation surmise that every doctor along
this tragic trail of dots right here was thinking about Joe’s case in
terms of a binary threshold on a checklist. Is the patient’s GM far above 30?
Check our blood levels of potassium above five point five million miles per liter check.
Does he have normal levels of albumin in his urine check? Now, all of those checks can tell you something
about Joe’s kidney function on that one isolated visit. And they’re really important for delivering good
health care. But those checks don’t tell you anything about the long term trend.
So even though Joe had been hurtling towards that terrible threshold for years,
he hadn’t yet crossed it. And nobody raised an alarm until it was basically too late.
So in retrospect, to me at least, this shouldn’t be surprising. Checklists,
as useful as they are supposed to help doctors understand and respond to what’s happening right now.
What what’s likely to happen in the future? That’s an inherent design feature of checklists. They focus
the doctor’s mind on the details of the present. But in a world where the biggest and most expensive medical conditions
are chronic diseases that unfold over a timescale of years or decades, that feature
of checklists is starting to look like a bug. You might ask why not just fix the bug
with a longer checklist by adding an item that encourages doctors to look at the long term trend? So
what did it have even been possible for a doctor at Joe’s bedside to call up a scatterplot
like this of his historical electronic health records and plotted over time to look for a trend?
The short answer is no. Maybe you could do it if you were a database experts.
But it certainly would not be a simple and obvious way for a doctor to use a modern electronic health records
system to see the trend in Joe’s GFI readings. You really would have needed to go back through the
electronic health records. One. Meeting at a time. And ironically, this almost surely would have been easier back
in the days of paper charts. Just recall the power of Nightingale’s figures
and how clearly the underlying trend jumped out of a figure like this. And I reflect
on the fact that in the subsequent hundred and sixty years, we haven’t managed to bring those figures to patients bedsides.
To me, that’s a tragedy. Of course, this is also where we begin to see the power of
artificial intelligence, because in modern medicine, it’s not just one set of readings to look at
either in time or over. Over time, it’s hundreds or even thousands
of readings, blood tests, urine tests, EEG, EKG, heart rate, blood pressure,
clinical symptoms, social factors, and soon real time data on a patient’s genetics or epigenetic profile.
There is just so much data. You’ve never seen this much data in your life. It is hard
for a human to comprehend it all. Even as a single snapshot, much less is a story that unfolds
over time. And then finally, there’s the issue of how this kind of hypothetical look for trends.
Item on a checklist would fit into a doctor’s usual workflow. When you show up to the emergency
room, your doctor’s main concern is how bad is your case right now? Should you be treated and sent home? Or are you sick
enough to be admitted to the hospital? Doctors face high stakes and enormous pressure in making those decisions.
And even outside of an emergency room and back in a normal clinic setting, they have to make them fast because
there are dozens of other patients in the waiting room need their help, too. How reasonable is it to expect those doctors
to stop what they’re doing, fire up a statistical software package, download some data from a database and
mind through a vast collection of electronic health records? All define the one or two historical
trends out of thousands or millions of possible trends that might predict something bad
months or years in the future. Doctors might do that kind of thing on a TV show like House, but I can assure you that
they don’t do it in real hospitals. In researching our book, I actually chatted with the medical data
science expert and a physician named Mark sendek who thinks about data science in hospitals for
living specifically these workflow issues. Here’s what he told me. Physicians always say they want
the data. Sendak said what the problem is that there’s no workflow for them to access or use
the data. The way that the records are structured, it takes time and skill. You have to write a query. You have to download the
data into a spreadsheet and then you actually have to do things with it. But physicians are under so much stress already.
They have 15 minute clinic visits. When exactly are they going to be playing with the data for their clinic
and figuring out what it tells them that they need to do with their patients? So to me, that brings us to
the deepest issue of all, which is the fact that the entire system of medical data science was designed
only to address questions at the level of a population, not at the level of an individual patient.
For example, how many lives could we save if we used Jeff our threshold a rather than GFI threshold
be for detecting kidney disease? There must be hundreds of research articles that bear on that question
or any such question you could find in medicine. But medical data science is nearly silent
at the level of basic statistical questions at the level of an individual patient. How are
Joh’s individual GFI readings changing over the long term? Where are they likely to go from here? What does that
predict about his health next month or next year? These questions would have been straightforward for either a human
or an algorithm to answer using Joh’s historical medical record. Yet all of those data points were never given
a chance to speak. There was no routine in place to sift his health record for signs of an underlying chronic
condition. There was no team of data scientists who algorithm no doctor with interdisciplinary training and statistics
and with some exceptions here and there. The same is true at most modern hospitals and clinics today. In speaking
with friends and colleagues about this question, I’ve noticed that many people seem to have this impression that there’s like
some kind of medical robot car behind the scenes of a modern hospital, like some fancy suite of
algorithms that’s analyzing individual patient data and helping doctors make personalized
treatment and diagnostic decisions. Maybe they get that impression from seeing how A.I. has transformed
other industries. Maybe they get it from seeing just how much damn data entry doctors do with their own appointments.
But whatever the reason, they’re usually shocked when I tell them the reality, which is not only is there no rub. Robot car
behind the scenes when it comes to individual patient level data analysis. There is literally nobody at the
steering wheel. When I spoke with Catherine Heller about this topic, her frustration
was clear, as she put it. It turns out that it’s not enough to just collect all that data.
You actually have to do something with it. And here she unknowingly was channeling NIGHTINGALE,
who wrote of St. Thomas’s Hospital in London in 1859 that it appears to keep
its statistics more for the sake of checking obstreperous patients, which is an object, certainly,
but not a scientific one. So the story of Joe. It turns out, is much more
than the story of a man with kidney disease. It is the story of the vast canyon between what data
could be doing for us and what the health care system lets it do.
So if it if it seems to you that health care professionals are drowning in data and could really use a life preserver,
that a combination of human and machine intelligence could radically improve healthcare. You’re not
alone in that thought. Companies and researchers are hard at work already on a new generation of A.I.
based technologies that stand waiting in the wings, ready to help doctors and nurses do their jobs more effectively.
Kathryn Heller The reason she was looking at this data on kidney disease is because she has invented an app that
will do this right, look at historical readings on kidney disease and predict their lung function. It’s the kind of
thing that a doctor could call up on an iPad at the bedside of patients. So they really could bring that hypothetical
look for trends item into their usual workflow. But for me, I’ll close here with just a few
thoughts about the barriers, both cultural and legal. One big issue is incentives. I
mean, look, American hospitals buy into something like this. The question that every hospital will be asking
itself is what does it mean for buying my bottom line? If you can better predict kidney disease,
you do not have to be much of a cynic to observe that health care systems make money on advancing chronic disease.
The legal system is another set of incentives or disincentives in this case. Imagine being in Katherine Heller’s position
and pondering the wisdom of selling or even just giving away an app that could predict the chronic kidney disease.
Imagine the legal peril that would await such an app designer when you think of the first inevitable
case of missed kidney disease. For the simple reason that policymakers and
lawmakers have not gotten off their backsides to address a simple question who ultimately is responsible
for an algorithm’s medical advice? How can we answer that question in a way that fosters innovation
while simultaneously protects patients and their privacy? Another big issue is whether data science
teams will get access to the data they need to improve existing systems and build new ones.
The key thing that makes A.I. work is data at scale. That’s why Google has great
on A.I. That’s why Amazon has great A.I. And the more I ponder this topic,
the more I started to believe that this problem will not be solved until Google and Amazon own
all of the hospitals because thousands of patient records at one
hospital are almost useless for this. You need millions or hundreds of millions of patient
records, and there is no reason in principle why that cannot be done. But the
practical and legal challenges of making that happen. Well, it’s just back to Nightingale’s
concerns about data standardization in the 1860s all over again
and the sources of social and financial value that go unrealized from failing
to pool and failing to analyze the trends in those data sets to me are mind boggling.
Now, even if there were a common data standard, let’s say you could solve that problem. Hospitals, I found, are very
hesitant to team up with data scientists, even under terms that guarantee patient privacy. In fact, I found
to be downright paranoid about it and nobody will really tell you why. I’ve read other
researchers, by the way, say the same thing. This isn’t just me. Nobody will tell you why. But I’ve
always thought it’s for a pretty craven reason that hospitals in America don’t want their competitors to
reverse engineer their business team pricing models. And so their default position is as a corporation
is just lock up all the hard drives. Whatever the reason, all those electronic health records are
used to generate very detailed bills, but almost never to help people require fewer hospital services
in the first place. So I find this mind boggling and I am not alone.
Can you imagine if we allowed hospitals to treat your kidneys in the same way that they treated the data
about your kidneys? I mean, shouldn’t be there a form that you could sign to overrule them donating your data
to save someone else’s life? For me, if hospitals are the only ones with this information
and you were paying them to take care of you. How are they not using it? So
as as I close up here, as you now appreciate, the barriers to adoption of A.I.
in the health care system have nothing to do with technology or science. But there are enormous
barriers of culture, law and incentives. Some of these barriers are specific to America,
but many others affect health care systems of the rich world. And the upshot is that the next data science,
revolution and health care won’t just take one person like Florence NIGHTINGALE. It will take people who
keep working on cool projects that keep convincing their colleagues that this stuff actually works and
keep generating good evidence for policymakers and other stakeholders. It’ll take doctors, nurses,
software engineers, lawyers, database managers, privacy experts, venture capitalists, insurers,
hospital administrators and patients, all of whom must come together to make this thing work.
It is a daunting task. And so I’ll leave you here today with a simple prayer. May Florence Nightingale’s
power of Will. What her sister pathetically called that most resolute and iron thing live on
in us all. Thank you.