1
00:00:04,000 --> 00:00:05,000
[CLICK]

2
00:00:16,000 --> 00:00:18,680
DAVID SONTAG: So welcome
to spring 2019 Machine

3
00:00:18,680 --> 00:00:20,200
Learning for Healthcare.

4
00:00:20,200 --> 00:00:21,280
My name is David Sontag.

5
00:00:21,280 --> 00:00:22,988
I'm a professor in
computer science.

6
00:00:22,988 --> 00:00:25,030
Also I'm in the Institute
for Medical Engineering

7
00:00:25,030 --> 00:00:26,563
and Science.

8
00:00:26,563 --> 00:00:28,480
My co-instructor today
will be Pete Szolovits,

9
00:00:28,480 --> 00:00:31,360
who I'll introduce more towards
the end of today's lecture,

10
00:00:31,360 --> 00:00:33,430
along with the rest
of the course staff.

11
00:00:33,430 --> 00:00:38,660
So the problem.

12
00:00:38,660 --> 00:00:41,050
The problem is that healthcare
in the United States

13
00:00:41,050 --> 00:00:42,820
costs too much.

14
00:00:42,820 --> 00:00:46,660
Currently, we're spending
$3 trillion a year,

15
00:00:46,660 --> 00:00:50,800
and we're not even necessarily
doing a very good job.

16
00:00:50,800 --> 00:00:53,650
Patients who have
chronic disease

17
00:00:53,650 --> 00:00:56,910
often find that these chronic
diseases are diagnosed late.

18
00:00:56,910 --> 00:01:00,460
They're often not managed well.

19
00:01:00,460 --> 00:01:03,490
And that happens even
in a country with some

20
00:01:03,490 --> 00:01:06,250
of the world's best clinicians.

21
00:01:06,250 --> 00:01:09,640
Moreover, medical
errors are happening

22
00:01:09,640 --> 00:01:13,240
all of the time,
errors that if caught,

23
00:01:13,240 --> 00:01:15,800
would have prevented
needless deaths,

24
00:01:15,800 --> 00:01:22,022
needless worsening
of disease, and more.

25
00:01:22,022 --> 00:01:24,780
And healthcare
impacts all of us.

26
00:01:24,780 --> 00:01:27,150
So I imagine that almost
everyone here in this room

27
00:01:27,150 --> 00:01:32,970
have had a family member,
a loved one, a dear friend,

28
00:01:32,970 --> 00:01:38,160
or even themselves suffer
from a health condition which

29
00:01:38,160 --> 00:01:42,210
impacts your quality of
life, which has affected

30
00:01:42,210 --> 00:01:46,800
your work, your
studies, and possibly

31
00:01:46,800 --> 00:01:50,410
has led to a needless death.

32
00:01:50,410 --> 00:01:53,470
And so the question that we're
asking in this course today

33
00:01:53,470 --> 00:01:55,340
is how can we use
machine learning,

34
00:01:55,340 --> 00:01:58,360
artificial intelligence, as
one piece of a bigger puzzle

35
00:01:58,360 --> 00:02:02,570
to try to transform healthcare.

36
00:02:02,570 --> 00:02:04,532
So all of us have
some personal stories.

37
00:02:04,532 --> 00:02:05,990
I myself have
personal stories that

38
00:02:05,990 --> 00:02:08,430
have led me to be
interested in this area.

39
00:02:08,430 --> 00:02:11,210
My grandfather, who had
Alzheimer's disease,

40
00:02:11,210 --> 00:02:14,763
was diagnosed quite late
in his Alzheimer's disease.

41
00:02:14,763 --> 00:02:17,180
There aren't good treatments
today for Alzheimer's, and so

42
00:02:17,180 --> 00:02:19,490
it's not that I would
have expected the outcome

43
00:02:19,490 --> 00:02:20,480
to be different.

44
00:02:20,480 --> 00:02:22,790
But had he been
diagnosed earlier,

45
00:02:22,790 --> 00:02:24,980
our family would have
recognized that many

46
00:02:24,980 --> 00:02:26,840
of the erratic things
that he was doing

47
00:02:26,840 --> 00:02:29,340
towards the later years of his
life were due to this disease

48
00:02:29,340 --> 00:02:32,580
and not due to
some other reason.

49
00:02:32,580 --> 00:02:36,150
My mother, who had multiple
myeloma, a blood cancer, who

50
00:02:36,150 --> 00:02:40,930
was diagnosed five
years ago now,

51
00:02:40,930 --> 00:02:43,240
never started treatment
for her cancer

52
00:02:43,240 --> 00:02:47,570
before she died one year ago.

53
00:02:47,570 --> 00:02:48,640
Now, why did she die?

54
00:02:48,640 --> 00:02:53,140
Well, it was believed
that her cancer was still

55
00:02:53,140 --> 00:02:56,250
in its very early stages.

56
00:02:56,250 --> 00:02:58,320
Her blood markers that
were used to track

57
00:02:58,320 --> 00:03:01,950
the progress of the cancer put
her in a low risk category.

58
00:03:01,950 --> 00:03:04,560
She didn't yet have
visible complications

59
00:03:04,560 --> 00:03:06,795
of the disease that
would, according

60
00:03:06,795 --> 00:03:09,750
to today's standard
guidelines, require treatment

61
00:03:09,750 --> 00:03:11,310
to be initiated.

62
00:03:11,310 --> 00:03:14,700
And as a result, the belief
was the best strategy

63
00:03:14,700 --> 00:03:16,980
was to wait and see.

64
00:03:16,980 --> 00:03:21,480
But unbeknownst to her and
to my family, her blood

65
00:03:21,480 --> 00:03:25,350
cancer, which was caused
by light chains which

66
00:03:25,350 --> 00:03:30,270
were accumulating, ended
up leading to organ damage.

67
00:03:30,270 --> 00:03:32,940
In this case, the light chains
were accumulating in her heart,

68
00:03:32,940 --> 00:03:35,540
and she died of heart failure.

69
00:03:35,540 --> 00:03:39,400
Had we recognized that her
disease was further along,

70
00:03:39,400 --> 00:03:40,880
she might have
initiated treatment.

71
00:03:40,880 --> 00:03:44,150
And there are now over 20
treatments for multiple myeloma

72
00:03:44,150 --> 00:03:49,640
which are believed to have
life-lengthening effect.

73
00:03:49,640 --> 00:03:51,670
And I can give you four
or five other stories

74
00:03:51,670 --> 00:03:54,592
from my own personal
family and my friends,

75
00:03:54,592 --> 00:03:56,050
where similar things
have happened.

76
00:03:56,050 --> 00:03:59,180
And I have no doubt that
all of you have as well.

77
00:03:59,180 --> 00:04:00,940
So what can we do about
it is the question

78
00:04:00,940 --> 00:04:04,180
that we want to try to
understand in today's course.

79
00:04:04,180 --> 00:04:05,490
And don't get me wrong.

80
00:04:05,490 --> 00:04:07,240
Machine learning,
artificial intelligence,

81
00:04:07,240 --> 00:04:09,430
will only be one
piece of the puzzle.

82
00:04:09,430 --> 00:04:11,680
There's so many other
systematic changes

83
00:04:11,680 --> 00:04:14,450
that we're going to have to
make into our healthcare system.

84
00:04:14,450 --> 00:04:18,555
But let's try to understand
what those AI elements might be.

85
00:04:18,555 --> 00:04:19,930
So let's start in
today's lecture

86
00:04:19,930 --> 00:04:23,030
by giving a bit of a background
on artificial intelligence

87
00:04:23,030 --> 00:04:25,020
and machine learning
in healthcare.

88
00:04:25,020 --> 00:04:28,960
And I'll tell you why I think
the time is right now, in 2019,

89
00:04:28,960 --> 00:04:32,575
to really start to make a
big dent at this problem.

90
00:04:32,575 --> 00:04:34,160
And then I'll tell you about--

91
00:04:34,160 --> 00:04:35,260
I'll give you a
few examples of how

92
00:04:35,260 --> 00:04:37,343
machine learning is likely
to transform healthcare

93
00:04:37,343 --> 00:04:38,500
over the next decade.

94
00:04:38,500 --> 00:04:40,130
And of course we're
just guessing,

95
00:04:40,130 --> 00:04:41,925
but this is really
guided by the latest

96
00:04:41,925 --> 00:04:45,370
and greatest in research, a lot
of it happening here at MIT.

97
00:04:45,370 --> 00:04:47,110
And then we'll close
today's lecture

98
00:04:47,110 --> 00:04:48,970
with an overview of
what's different,

99
00:04:48,970 --> 00:04:51,850
what's unique about machine
learning healthcare.

100
00:04:51,850 --> 00:04:53,980
All of you have taken some
machine learning course

101
00:04:53,980 --> 00:04:55,930
in the past, and so
you know the basics

102
00:04:55,930 --> 00:04:58,000
of supervised prediction.

103
00:04:58,000 --> 00:05:00,760
Many of you have studied
things like clustering.

104
00:05:00,760 --> 00:05:02,830
And you're certainly
paying attention

105
00:05:02,830 --> 00:05:06,070
to the news, where you
see news every single day

106
00:05:06,070 --> 00:05:08,848
about Google,
Facebook, Microsoft's

107
00:05:08,848 --> 00:05:11,140
latest advances in speech
recognition, computer vision,

108
00:05:11,140 --> 00:05:11,945
and so on.

109
00:05:11,945 --> 00:05:13,570
So what's really
different about trying

110
00:05:13,570 --> 00:05:16,840
to apply these techniques
in the healthcare domain?

111
00:05:16,840 --> 00:05:20,160
The answer is that there's
a huge amount of difference,

112
00:05:20,160 --> 00:05:21,910
and there are a lot
of subtleties to doing

113
00:05:21,910 --> 00:05:22,990
machine learning right here.

114
00:05:22,990 --> 00:05:23,980
And we'll talk about
that throughout

115
00:05:23,980 --> 00:05:25,063
the whole entire semester.

116
00:05:28,920 --> 00:05:32,050
So to begin, this
isn't a new field.

117
00:05:32,050 --> 00:05:35,350
Artificial intelligence in
medicine goes back to the 1970s

118
00:05:35,350 --> 00:05:38,600
or sometime even in the '60s.

119
00:05:38,600 --> 00:05:40,690
One of the earliest
examples of trying

120
00:05:40,690 --> 00:05:42,880
to use artificial
intelligence for diagnosis

121
00:05:42,880 --> 00:05:50,860
was this MYCIN system developed
at Stanford, where the goal was

122
00:05:50,860 --> 00:05:54,540
try to identify bacteria
that might cause infection

123
00:05:54,540 --> 00:05:57,690
and then to try to
guide what would

124
00:05:57,690 --> 00:05:59,970
be the appropriate
therapy for that bacteria.

125
00:05:59,970 --> 00:06:03,120
Now, it was found that this
algorithm, this machine

126
00:06:03,120 --> 00:06:07,350
learning, this
simple AI algorithm

127
00:06:07,350 --> 00:06:09,870
was able to propose a good
therapy in 69% of cases, which

128
00:06:09,870 --> 00:06:12,945
at the time was better
than the best or very good

129
00:06:12,945 --> 00:06:14,070
infectious disease experts.

130
00:06:17,160 --> 00:06:19,470
Now, it also had some
other elements of it

131
00:06:19,470 --> 00:06:22,360
which were quite interesting.

132
00:06:22,360 --> 00:06:24,060
So those of you
who've been following

133
00:06:24,060 --> 00:06:28,510
the latest in chat bots might
find this a little prescient.

134
00:06:28,510 --> 00:06:31,030
So there's a dialog interface
where the computer says,

135
00:06:31,030 --> 00:06:31,890
I am ready.

136
00:06:31,890 --> 00:06:35,610
The computer clinician
responds, this

137
00:06:35,610 --> 00:06:37,590
is a 26-year-old male patient.

138
00:06:37,590 --> 00:06:39,598
The computer tests
its understanding

139
00:06:39,598 --> 00:06:40,390
of what's going on.

140
00:06:40,390 --> 00:06:42,730
It says, the age of
the patient is 26.

141
00:06:42,730 --> 00:06:45,160
The sex of the patient is male.

142
00:06:45,160 --> 00:06:47,700
The clinician writes,
five days ago,

143
00:06:47,700 --> 00:06:50,290
he had respiratory
tract infections.

144
00:06:50,290 --> 00:06:51,960
The computer asks
what is his name.

145
00:06:51,960 --> 00:06:53,480
Joe.

146
00:06:53,480 --> 00:06:55,403
The computer responds,
my understanding

147
00:06:55,403 --> 00:06:56,820
is the name of the
patient is Joe.

148
00:06:56,820 --> 00:07:00,580
Respiratory tract is one of
the symptoms the patient had.

149
00:07:00,580 --> 00:07:02,440
Then the clinician
writes, a couple of days

150
00:07:02,440 --> 00:07:04,750
before the admission,
he had malaise,

151
00:07:04,750 --> 00:07:07,495
which is general tiredness.

152
00:07:07,495 --> 00:07:08,960
And the computer
responds, please

153
00:07:08,960 --> 00:07:10,550
give me a date of admission.

154
00:07:10,550 --> 00:07:13,820
The clinician responds, March
12, 1979, and the computer

155
00:07:13,820 --> 00:07:16,200
again confirms that it's
understood appropriately.

156
00:07:16,200 --> 00:07:21,720
And this is the preface to
the later diagnostic stages.

157
00:07:21,720 --> 00:07:25,040
So the ideas of how AI
can really impact medicine

158
00:07:25,040 --> 00:07:26,780
have been around a long time.

159
00:07:26,780 --> 00:07:28,310
Yet these algorithms
which have been

160
00:07:28,310 --> 00:07:32,450
shown to be very effective,
even going back to the 1970s,

161
00:07:32,450 --> 00:07:35,450
didn't translate
into clinical care.

162
00:07:35,450 --> 00:07:38,930
A second example, oh so equally
impressive in its nature,

163
00:07:38,930 --> 00:07:42,900
was work from the
1980s in Pittsburgh,

164
00:07:42,900 --> 00:07:46,010
developing what is known as the
INTERNIST-1 or Quick Medical

165
00:07:46,010 --> 00:07:47,850
Reference system.

166
00:07:47,850 --> 00:07:50,460
This was now used not
for infectious diseases,

167
00:07:50,460 --> 00:07:53,240
but for primary care.

168
00:07:53,240 --> 00:07:56,730
Here one might ask, how
can we try to do diagnosis

169
00:07:56,730 --> 00:08:00,180
at a much larger scale, where
patients might come in with one

170
00:08:00,180 --> 00:08:02,280
of hundreds of
different diseases

171
00:08:02,280 --> 00:08:05,010
and could report thousands of
different symptoms, each one

172
00:08:05,010 --> 00:08:07,140
giving you some
view, noisy view,

173
00:08:07,140 --> 00:08:11,030
into what may be going on
with a patient's health.

174
00:08:11,030 --> 00:08:13,220
And at a high level, they
modeled this as something

175
00:08:13,220 --> 00:08:14,360
like a Bayesian network.

176
00:08:14,360 --> 00:08:16,820
It wasn't strictly
a Bayesian network.

177
00:08:16,820 --> 00:08:18,650
It was a bit more
heuristic at the time.

178
00:08:18,650 --> 00:08:21,350
It was later developed to be so.

179
00:08:21,350 --> 00:08:24,205
But at a high level, there were
a number of latent variables

180
00:08:24,205 --> 00:08:25,580
or hidden variables
corresponding

181
00:08:25,580 --> 00:08:27,710
to different diseases
the patient might have,

182
00:08:27,710 --> 00:08:30,670
like flu or pneumonia
or diabetes.

183
00:08:30,670 --> 00:08:32,539
And then there were
a number of variables

184
00:08:32,539 --> 00:08:35,735
on the very bottom, which were
symptoms, which are all binary,

185
00:08:35,735 --> 00:08:37,850
so the diseases are
either on or off.

186
00:08:37,850 --> 00:08:40,470
And here the symptoms are
either present or not.

187
00:08:40,470 --> 00:08:43,919
And these symptoms can include
things like fatigue or cough.

188
00:08:43,919 --> 00:08:47,870
They could also be things that
result from laboratory test

189
00:08:47,870 --> 00:08:53,280
results, like a high
value of hemoglobin A1C.

190
00:08:53,280 --> 00:08:57,712
And this algorithm would
then take this model,

191
00:08:57,712 --> 00:08:59,920
take the symptoms that were
reported for the patient,

192
00:08:59,920 --> 00:09:02,410
and try to do reasoning over
what action might be going on

193
00:09:02,410 --> 00:09:04,660
with that patient, to figure
out what the differential

194
00:09:04,660 --> 00:09:06,550
diagnosis is.

195
00:09:06,550 --> 00:09:10,030
There are over 40,000 edges
connecting diseases to symptoms

196
00:09:10,030 --> 00:09:12,650
that those diseases were
believed to have caused.

197
00:09:12,650 --> 00:09:18,010
And this knowledge base, which
was probabilistic in nature,

198
00:09:18,010 --> 00:09:21,340
because it captured the idea
that some symptoms would only

199
00:09:21,340 --> 00:09:25,420
occur with some
probability for a disease,

200
00:09:25,420 --> 00:09:28,000
took over 15 person
years to elicit

201
00:09:28,000 --> 00:09:30,770
from a large medical team.

202
00:09:30,770 --> 00:09:33,180
And so it was a lot of effort.

203
00:09:33,180 --> 00:09:37,020
And even in going
forward to today's time,

204
00:09:37,020 --> 00:09:41,250
there have been few similar
efforts at a scale as

205
00:09:41,250 --> 00:09:43,440
impressive as this one.

206
00:09:43,440 --> 00:09:45,240
But again, what happened?

207
00:09:45,240 --> 00:09:48,180
These algorithms are not
being used anywhere today

208
00:09:48,180 --> 00:09:50,790
in our clinical workflows.

209
00:09:50,790 --> 00:09:55,440
And the challenges that
have prevented them

210
00:09:55,440 --> 00:09:59,260
from being used
today are numerous.

211
00:09:59,260 --> 00:10:01,108
But I used a word
in my explanation

212
00:10:01,108 --> 00:10:02,400
which should really hint at it.

213
00:10:02,400 --> 00:10:04,490
I used the word
clinical workflow.

214
00:10:04,490 --> 00:10:07,560
And this, I think, is one
of the biggest challenges.

215
00:10:07,560 --> 00:10:09,510
Which is that the
algorithms were designed

216
00:10:09,510 --> 00:10:11,370
to solve narrow problems.

217
00:10:11,370 --> 00:10:13,800
They weren't necessarily even
the most important problems,

218
00:10:13,800 --> 00:10:17,920
because clinicians generally do
a very good job at diagnosis.

219
00:10:17,920 --> 00:10:22,350
And there was a big gap between
the input that they expected

220
00:10:22,350 --> 00:10:24,840
and the current
clinical workflow.

221
00:10:24,840 --> 00:10:30,330
So imagine that you have
now a mainframe computer.

222
00:10:30,330 --> 00:10:33,510
I mean, this was the '80s.

223
00:10:33,510 --> 00:10:38,910
And you have a clinician who
has to talk to the patient

224
00:10:38,910 --> 00:10:40,230
and get some information.

225
00:10:40,230 --> 00:10:41,610
Go back to the computer.

226
00:10:41,610 --> 00:10:44,460
Type in a structured
data, the symptoms

227
00:10:44,460 --> 00:10:46,320
that the patient's reporting.

228
00:10:46,320 --> 00:10:49,110
Get information back from
the computer and iterate.

229
00:10:49,110 --> 00:10:54,700
As you can imagine, that takes a
lot of time, and time is money.

230
00:10:54,700 --> 00:10:59,100
And unfortunately, it
prevents it from being used.

231
00:10:59,100 --> 00:11:02,430
Moreover, despite the fact
that it took a lot of effort

232
00:11:02,430 --> 00:11:05,290
to use it when outside of
existing clinical workflows,

233
00:11:05,290 --> 00:11:08,360
these systems were also
really difficult to maintain.

234
00:11:08,360 --> 00:11:10,440
So I talked about
how this was elicited

235
00:11:10,440 --> 00:11:12,180
from 15 person years of work.

236
00:11:12,180 --> 00:11:13,920
There was no machine
learning here.

237
00:11:13,920 --> 00:11:15,540
It was called
artificial intelligence

238
00:11:15,540 --> 00:11:19,140
because one tries to reason
in an artificial way,

239
00:11:19,140 --> 00:11:20,910
like humans might.

240
00:11:20,910 --> 00:11:23,737
But there was no learning
from data in this.

241
00:11:23,737 --> 00:11:26,070
And so what that means is if
you then go to a new place,

242
00:11:26,070 --> 00:11:28,410
let's say this was
developed in Pittsburgh,

243
00:11:28,410 --> 00:11:33,270
and now you go to Los Angeles
or to Beijing or to London,

244
00:11:33,270 --> 00:11:35,310
and you want to apply
the same algorithms,

245
00:11:35,310 --> 00:11:38,400
you suddenly have to
re-derive parts of this model

246
00:11:38,400 --> 00:11:39,240
from scratch.

247
00:11:39,240 --> 00:11:42,032
For example, the prior
probability of the diseases

248
00:11:42,032 --> 00:11:43,740
are going to be very
different, depending

249
00:11:43,740 --> 00:11:46,882
on where you are in the world.

250
00:11:46,882 --> 00:11:48,840
Now, you might want to
go to a different domain

251
00:11:48,840 --> 00:11:50,555
outside of primary care.

252
00:11:50,555 --> 00:11:52,680
And again, one has to spend
a huge amount of effort

253
00:11:52,680 --> 00:11:54,990
to derive such models.

254
00:11:54,990 --> 00:11:57,810
As new medicine
discoveries are made,

255
00:11:57,810 --> 00:11:59,550
one has to, again,
update these models.

256
00:11:59,550 --> 00:12:05,570
And this has been a huge
blocker to deployment.

257
00:12:05,570 --> 00:12:08,520
I'll move forward to
one more example now,

258
00:12:08,520 --> 00:12:10,330
also from the 1980s.

259
00:12:10,330 --> 00:12:13,530
And this is now for a
different type of question.

260
00:12:13,530 --> 00:12:16,110
Not one of how do you do
diagnosis, but how do you

261
00:12:16,110 --> 00:12:18,450
actually do discovery.

262
00:12:18,450 --> 00:12:22,060
So this is an example
from Stanford.

263
00:12:22,060 --> 00:12:23,730
And it was a really
interesting case

264
00:12:23,730 --> 00:12:27,420
where one took a
data-driven approach to try

265
00:12:27,420 --> 00:12:29,550
to make medical discoveries.

266
00:12:29,550 --> 00:12:34,350
There was a database of what's
called a disease registry

267
00:12:34,350 --> 00:12:37,050
from patients with
rheumatoid arthritis, which

268
00:12:37,050 --> 00:12:37,980
is a chronic disease.

269
00:12:37,980 --> 00:12:42,750
It's an autoimmune
condition, where

270
00:12:42,750 --> 00:12:46,830
for each patient, over a
series of different visits,

271
00:12:46,830 --> 00:12:48,690
one would record,
for example, here

272
00:12:48,690 --> 00:12:50,130
it shows this is
visit number one.

273
00:12:50,130 --> 00:12:54,880
The date was January 17, 1979.

274
00:12:54,880 --> 00:12:57,750
The knee pain, patient's knee
pain, was reported as severe.

275
00:12:57,750 --> 00:12:59,070
Their fatigue was moderate.

276
00:12:59,070 --> 00:13:02,070
Temperatures was 38.5 Celsius.

277
00:13:02,070 --> 00:13:05,430
The diagnosis for this
patient was actually

278
00:13:05,430 --> 00:13:09,780
a different autoimmune
condition called systemic lupus.

279
00:13:09,780 --> 00:13:13,050
We have some laboratory test
values for their creatinine

280
00:13:13,050 --> 00:13:16,020
and blood nitrogen,
and we know something

281
00:13:16,020 --> 00:13:17,410
about their medication.

282
00:13:17,410 --> 00:13:21,550
In this case, they were
on prednisone, a steroid.

283
00:13:21,550 --> 00:13:24,650
And one has this data
at every point in time.

284
00:13:24,650 --> 00:13:28,670
This almost certainly
was recorded on paper

285
00:13:28,670 --> 00:13:30,890
and then later,
these were collected

286
00:13:30,890 --> 00:13:33,050
into a computer format.

287
00:13:33,050 --> 00:13:34,790
But then it provides
the possibility

288
00:13:34,790 --> 00:13:36,630
to ask questions and
make new discoveries.

289
00:13:36,630 --> 00:13:38,540
So for example, in
this work, there

290
00:13:38,540 --> 00:13:40,580
was a discovery
module which would

291
00:13:40,580 --> 00:13:44,630
make causal hypotheses
about what aspects

292
00:13:44,630 --> 00:13:47,240
might cause other aspects.

293
00:13:47,240 --> 00:13:49,520
It would then do
some basic statistics

294
00:13:49,520 --> 00:13:51,620
to check about the
statistical validity

295
00:13:51,620 --> 00:13:53,390
of those causal hypotheses.

296
00:13:53,390 --> 00:13:56,260
It would then present
those to a domain expert

297
00:13:56,260 --> 00:13:59,630
to try to check off does
this make sense or not.

298
00:13:59,630 --> 00:14:02,750
For those that are accepted,
it then uses that knowledge

299
00:14:02,750 --> 00:14:04,400
that was just
learned to iterate,

300
00:14:04,400 --> 00:14:06,470
to try to make new discoveries.

301
00:14:06,470 --> 00:14:08,600
And one of the main
findings from this paper

302
00:14:08,600 --> 00:14:11,090
was that prednisone
elevates cholesterol.

303
00:14:11,090 --> 00:14:15,240
That was published in the Annals
of Internal Medicine in 1986.

304
00:14:15,240 --> 00:14:17,520
So these are all
very early examples

305
00:14:17,520 --> 00:14:21,150
of data-driven approaches
to improve both medicine

306
00:14:21,150 --> 00:14:23,580
and healthcare.

307
00:14:23,580 --> 00:14:25,880
Now flip forward to the 1990s.

308
00:14:25,880 --> 00:14:28,453
Neural networks started
to become popular.

309
00:14:28,453 --> 00:14:30,120
Not quite the neural
networks that we're

310
00:14:30,120 --> 00:14:32,040
familiar with in
today's day and age,

311
00:14:32,040 --> 00:14:36,240
but nonetheless, they shared
very much of the same elements.

312
00:14:36,240 --> 00:14:38,910
So just in 1990, there
were 88 published studies

313
00:14:38,910 --> 00:14:41,340
using neural networks for
various different medical

314
00:14:41,340 --> 00:14:42,998
problems.

315
00:14:42,998 --> 00:14:45,540
One of the things that really
differentiated those approaches

316
00:14:45,540 --> 00:14:48,510
to what we see in
today's landscape

317
00:14:48,510 --> 00:14:50,640
is that the number of
features were very small.

318
00:14:50,640 --> 00:14:54,030
So usually features which were
similar to what I showed you

319
00:14:54,030 --> 00:14:55,420
in the previous slide.

320
00:14:55,420 --> 00:14:58,140
So structured data
that was manually

321
00:14:58,140 --> 00:15:02,610
curated for the purpose of
using in machine learning.

322
00:15:02,610 --> 00:15:07,380
And there was nothing
automatic about this.

323
00:15:07,380 --> 00:15:12,720
So one would have to have
assistants gather the data.

324
00:15:12,720 --> 00:15:14,430
And because of that,
typically, there

325
00:15:14,430 --> 00:15:18,330
were very small number of
samples for each study that

326
00:15:18,330 --> 00:15:20,980
were used in machine learning.

327
00:15:20,980 --> 00:15:23,590
Now, these models,
although very effective,

328
00:15:23,590 --> 00:15:26,050
and I'll show you some examples
in the next slide, also

329
00:15:26,050 --> 00:15:28,720
suffered from the same
challenges I mentioned earlier.

330
00:15:28,720 --> 00:15:30,940
They didn't fit well
into clinical workflows.

331
00:15:30,940 --> 00:15:32,920
It was hard to get enough
training data because

332
00:15:32,920 --> 00:15:35,440
of the manual efforts involved.

333
00:15:35,440 --> 00:15:39,250
And what the community found,
even in the early 1990s,

334
00:15:39,250 --> 00:15:42,192
is that these algorithms
did not generalize well.

335
00:15:42,192 --> 00:15:44,650
If you went through this huge
effort of collecting training

336
00:15:44,650 --> 00:15:47,980
data, learning your model, and
validating your model at one

337
00:15:47,980 --> 00:15:50,470
institution, and you then
take it to a different one,

338
00:15:50,470 --> 00:15:51,980
it just works much worse.

339
00:15:51,980 --> 00:15:52,480
OK?

340
00:15:52,480 --> 00:15:54,280
And that really
prevented translation

341
00:15:54,280 --> 00:15:57,200
of these technologies
into clinical practice.

342
00:15:57,200 --> 00:15:59,560
So what were these different
domains that were studied?

343
00:15:59,560 --> 00:16:00,810
Well, here are a few examples.

344
00:16:00,810 --> 00:16:04,160
It's a bit small, so
I'll read it out to you.

345
00:16:04,160 --> 00:16:08,830
It was studied in breast cancer,
myocardial infarction, which

346
00:16:08,830 --> 00:16:12,370
is heart attack,
lower back pain,

347
00:16:12,370 --> 00:16:14,620
used to predict
psychiatric length of stay

348
00:16:14,620 --> 00:16:19,450
for inpatient, skin tumors,
head injuries, prediction

349
00:16:19,450 --> 00:16:26,680
of dementia, understanding
progression of diabetes,

350
00:16:26,680 --> 00:16:30,430
and a variety of other problems,
which again are of the nature

351
00:16:30,430 --> 00:16:32,770
that we see about, we
read about in the news

352
00:16:32,770 --> 00:16:34,870
today in modern
attempts to apply

353
00:16:34,870 --> 00:16:36,970
machine learning in healthcare.

354
00:16:36,970 --> 00:16:38,980
The number of training
examples, as mentioned,

355
00:16:38,980 --> 00:16:42,460
were very few, ranging from
39 to, in some cases, 3,000.

356
00:16:42,460 --> 00:16:45,250
Those are individuals, humans.

357
00:16:45,250 --> 00:16:48,580
And the networks,
the neural networks,

358
00:16:48,580 --> 00:16:50,320
they weren't completely
shallow, but they

359
00:16:50,320 --> 00:16:51,550
weren't very deep either.

360
00:16:51,550 --> 00:16:53,830
So these were the
architectures they

361
00:16:53,830 --> 00:16:59,070
might be 60 neurons, then
7, and then 6, for example,

362
00:16:59,070 --> 00:17:02,580
in terms of each of the
layers of the neural network.

363
00:17:02,580 --> 00:17:04,140
By the way, that
sort of makes, sense

364
00:17:04,140 --> 00:17:08,170
given the type of data
that was fed into it.

365
00:17:08,170 --> 00:17:12,040
So none of this is new,
in terms of the goals.

366
00:17:12,040 --> 00:17:13,270
So what's changed?

367
00:17:13,270 --> 00:17:15,280
Why do I think that
despite the fact

368
00:17:15,280 --> 00:17:17,079
that we've had
what could arguably

369
00:17:17,079 --> 00:17:20,260
be called a failure for
the last 30 or 40 years,

370
00:17:20,260 --> 00:17:23,130
that we might actually have
some chance of succeeding now.

371
00:17:23,130 --> 00:17:25,839
And the big differentiator, what
I'll call now the opportunity,

372
00:17:25,839 --> 00:17:27,369
is data.

373
00:17:27,369 --> 00:17:29,490
So whereas in the
past, much of the work

374
00:17:29,490 --> 00:17:34,220
in artificial intelligence in
medicine was not data driven.

375
00:17:34,220 --> 00:17:37,510
It was based on trying to elicit
as much domain knowledge as one

376
00:17:37,510 --> 00:17:41,020
can from clinical
domain experts.

377
00:17:41,020 --> 00:17:44,080
In some cases, gathering
a little bit of data.

378
00:17:44,080 --> 00:17:46,210
Today, we have an
amazing opportunity

379
00:17:46,210 --> 00:17:50,710
because of the prevalence of
electronic medical records,

380
00:17:50,710 --> 00:17:52,960
both in the United
States and elsewhere.

381
00:17:52,960 --> 00:17:56,080
Now, here the United States,
for example, the story

382
00:17:56,080 --> 00:17:58,630
wasn't that way,
even back in 2008,

383
00:17:58,630 --> 00:18:00,640
when the adoption of
electronic medical records

384
00:18:00,640 --> 00:18:04,630
was under 10% across the US.

385
00:18:04,630 --> 00:18:08,110
But then there wasn't an
economic disaster in the US.

386
00:18:08,110 --> 00:18:11,710
And as part of the economic
stimulus package, which

387
00:18:11,710 --> 00:18:17,410
President Obama initiated, there
was something like $30 billion

388
00:18:17,410 --> 00:18:21,040
allocated to
hospitals purchasing

389
00:18:21,040 --> 00:18:23,062
electronic medical records.

390
00:18:23,062 --> 00:18:24,520
And this is already
a first example

391
00:18:24,520 --> 00:18:27,520
that we see of
policy being really

392
00:18:27,520 --> 00:18:29,880
influential to create the--

393
00:18:29,880 --> 00:18:32,830
to open the stage
to the types of work

394
00:18:32,830 --> 00:18:35,560
that we're going to be able
to do in this course today.

395
00:18:35,560 --> 00:18:38,410
So money was then made available
as incentives for hospitals

396
00:18:38,410 --> 00:18:40,480
to purchase electronic
medical records.

397
00:18:40,480 --> 00:18:43,780
And as a result, the adoption
increased dramatically.

398
00:18:43,780 --> 00:18:47,620
This is a really old number
from 2015 of 84% of hospitals,

399
00:18:47,620 --> 00:18:50,330
and now today, it's
actually much larger.

400
00:18:50,330 --> 00:18:53,638
So data is being collected
in an electronic form,

401
00:18:53,638 --> 00:18:56,180
and that presents an opportunity
to try to do research on it.

402
00:18:56,180 --> 00:18:58,180
It presents an opportunity
to do machine learning on it,

403
00:18:58,180 --> 00:19:00,700
and it presents an opportunity
to start to deploy machine

404
00:19:00,700 --> 00:19:02,230
learning algorithms,
where rather

405
00:19:02,230 --> 00:19:04,960
than having to manually
input data for a patient,

406
00:19:04,960 --> 00:19:06,850
we can just draw
it automatically

407
00:19:06,850 --> 00:19:11,207
from data that's already
available in electronic form.

408
00:19:11,207 --> 00:19:12,790
And so there are a
number of data sets

409
00:19:12,790 --> 00:19:16,480
that have been made available
for research and development

410
00:19:16,480 --> 00:19:17,860
in this space.

411
00:19:17,860 --> 00:19:21,670
Here at MIT, there has
been a major effort

412
00:19:21,670 --> 00:19:27,410
pioneered by Professor Roger
Mark, in the ECS and Institute

413
00:19:27,410 --> 00:19:29,950
for Medical
Engineering department,

414
00:19:29,950 --> 00:19:34,480
to create what's known as the
PhysioNet or Mimic databases.

415
00:19:34,480 --> 00:19:37,600
Mimic contains data from
over 40,000 patients

416
00:19:37,600 --> 00:19:39,680
and intensive care units.

417
00:19:39,680 --> 00:19:41,140
And it's very rich data.

418
00:19:41,140 --> 00:19:42,790
It contains basically
everything that's

419
00:19:42,790 --> 00:19:44,800
being collected in the
intensive care unit.

420
00:19:44,800 --> 00:19:47,950
Everything from notes that
are written by both nurses

421
00:19:47,950 --> 00:19:52,390
and by attendings, to vital
signs that are being collected

422
00:19:52,390 --> 00:19:55,690
by monitors that are
attached to patients,

423
00:19:55,690 --> 00:20:01,300
collecting their blood pressure,
oxygen saturation, heart

424
00:20:01,300 --> 00:20:06,100
rate, and so on,
to imaging data,

425
00:20:06,100 --> 00:20:13,745
to blood test results as they're
made available, and outcomes.

426
00:20:13,745 --> 00:20:15,370
And of course also
medications that are

427
00:20:15,370 --> 00:20:17,590
being prescribed as it goes.

428
00:20:17,590 --> 00:20:20,470
And so this is a wealth
of data that now one

429
00:20:20,470 --> 00:20:22,140
could use to try
to study, at least

430
00:20:22,140 --> 00:20:24,340
study in a very narrow
setting of an intensive care

431
00:20:24,340 --> 00:20:28,780
unit, how machine learning
could be used in that location.

432
00:20:28,780 --> 00:20:31,780
And I don't want to
under-emphasize the importance

433
00:20:31,780 --> 00:20:34,150
of this database, both
through this course

434
00:20:34,150 --> 00:20:35,440
and through the broader field.

435
00:20:35,440 --> 00:20:37,690
This is really the
only publicly available

436
00:20:37,690 --> 00:20:39,280
electronic medical
record data set

437
00:20:39,280 --> 00:20:41,660
of any reasonable size
in the whole world,

438
00:20:41,660 --> 00:20:44,235
and it was created here at MIT.

439
00:20:44,235 --> 00:20:45,610
And we'll be using
it extensively

440
00:20:45,610 --> 00:20:47,875
in our homework
assignments as a result.

441
00:20:47,875 --> 00:20:50,250
There are other data sets that
aren't publicly available,

442
00:20:50,250 --> 00:20:54,390
but which have been
gathered by industry.

443
00:20:54,390 --> 00:21:01,350
And one prime example is the
Truven Market Scan database,

444
00:21:01,350 --> 00:21:03,660
which was created by a
company called Truven, which

445
00:21:03,660 --> 00:21:05,640
was later acquired by
IBM, as I'll tell you

446
00:21:05,640 --> 00:21:08,200
about more in a few minutes.

447
00:21:08,200 --> 00:21:11,430
Now, this data-- and there are
many competing companies that

448
00:21:11,430 --> 00:21:13,470
have similar data sets--

449
00:21:13,470 --> 00:21:17,350
is created not from
electronic medical records,

450
00:21:17,350 --> 00:21:18,780
but rather from--

451
00:21:18,780 --> 00:21:21,120
typically, it's created
from insurance claims.

452
00:21:21,120 --> 00:21:24,132
So every time you
go to see a doctor,

453
00:21:24,132 --> 00:21:25,590
there's usually
some record of that

454
00:21:25,590 --> 00:21:27,810
that is associated to the
billing of that visit.

455
00:21:27,810 --> 00:21:33,930
So your provider will send a
bill to your health insurance

456
00:21:33,930 --> 00:21:36,870
saying basically what
happened, so what procedures

457
00:21:36,870 --> 00:21:40,740
were performed,
providing diagnoses

458
00:21:40,740 --> 00:21:44,490
that are used to justify
the cost of those procedures

459
00:21:44,490 --> 00:21:46,740
and tests.

460
00:21:46,740 --> 00:21:50,540
And from that data, you
now get a holistic view,

461
00:21:50,540 --> 00:21:52,620
a longitudinal view,
of what's happened

462
00:21:52,620 --> 00:21:54,450
to that patient's health.

463
00:21:54,450 --> 00:21:56,250
And then there is
a lot of money that

464
00:21:56,250 --> 00:22:01,020
passes behind the scenes
between insurers and hospitals

465
00:22:01,020 --> 00:22:03,270
to corporate companies,
such as Truven,

466
00:22:03,270 --> 00:22:05,520
which collect that
data and then resell it

467
00:22:05,520 --> 00:22:07,655
for research purposes.

468
00:22:07,655 --> 00:22:09,780
And one of the biggest
purchasers of data like this

469
00:22:09,780 --> 00:22:13,080
is the pharmaceutical industry.

470
00:22:13,080 --> 00:22:17,220
So this data, unfortunately, is
not usually publicly available,

471
00:22:17,220 --> 00:22:19,460
and that's actually a big
problem, both in the US

472
00:22:19,460 --> 00:22:20,300
and elsewhere.

473
00:22:20,300 --> 00:22:22,310
It's a big obstacle to
research in this field,

474
00:22:22,310 --> 00:22:24,530
that only people who
have millions of dollars

475
00:22:24,530 --> 00:22:26,120
to pay for it really
get access to it,

476
00:22:26,120 --> 00:22:27,320
and it's something that
I'm going to return to

477
00:22:27,320 --> 00:22:28,320
throughout the semester.

478
00:22:28,320 --> 00:22:29,870
It's something
where I think policy

479
00:22:29,870 --> 00:22:31,680
can make a big difference.

480
00:22:31,680 --> 00:22:34,250
But luckily, here
at MIT, the story's

481
00:22:34,250 --> 00:22:35,600
going to be a bit different.

482
00:22:35,600 --> 00:22:40,430
So thanks to the MIT
IBM Watson AI Lab,

483
00:22:40,430 --> 00:22:42,920
MIT has a close
relationship with IBM.

484
00:22:42,920 --> 00:22:45,980
And fingers crossed, it
looks like we'll get access

485
00:22:45,980 --> 00:22:48,270
to this database for our
homework and projects

486
00:22:48,270 --> 00:22:49,020
for this semester.

487
00:22:53,232 --> 00:22:54,940
Now, there are a lot
of other initiatives

488
00:22:54,940 --> 00:22:57,460
that are creating
large data sets.

489
00:22:57,460 --> 00:22:59,860
A really important
example here in the US

490
00:22:59,860 --> 00:23:02,650
is President Obama's
Precision Medicine Initiative,

491
00:23:02,650 --> 00:23:07,150
which has since been renamed
to the All of Us Initiative.

492
00:23:07,150 --> 00:23:09,250
And this initiative
is creating a data set

493
00:23:09,250 --> 00:23:13,302
of one million patients, drawn
in a representative manner,

494
00:23:13,302 --> 00:23:15,010
from across the United
States, to capture

495
00:23:15,010 --> 00:23:19,210
patients both poor and rich,
patients who are healthy

496
00:23:19,210 --> 00:23:21,880
and have chronic disease,
with the goal of trying

497
00:23:21,880 --> 00:23:23,440
to create a research
database where

498
00:23:23,440 --> 00:23:27,370
all of us and other people,
both inside and outside the US,

499
00:23:27,370 --> 00:23:29,640
could do research to
make medical discoveries.

500
00:23:29,640 --> 00:23:31,660
And this will include
data such as data

501
00:23:31,660 --> 00:23:34,720
from a baseline health exam,
where the typical vitals are

502
00:23:34,720 --> 00:23:37,240
taken, blood is drawn.

503
00:23:37,240 --> 00:23:39,340
It'll combine data of
the previous two types

504
00:23:39,340 --> 00:23:40,900
I've mentioned,
including both data

505
00:23:40,900 --> 00:23:44,470
from electronic medical records
and health insurance claims.

506
00:23:44,470 --> 00:23:47,620
And a lot of this work is
also happening here in Boston.

507
00:23:47,620 --> 00:23:50,145
So right across the street
at the Broad Institute,

508
00:23:50,145 --> 00:23:51,520
there is a team
which is creating

509
00:23:51,520 --> 00:23:52,937
all of the software
infrastructure

510
00:23:52,937 --> 00:23:55,160
to accommodate this data.

511
00:23:55,160 --> 00:23:58,270
And there are a large
number of recruitment sites

512
00:23:58,270 --> 00:24:00,190
here in the broader
Boston area where

513
00:24:00,190 --> 00:24:02,890
patients or any one of you,
really, could go and volunteer

514
00:24:02,890 --> 00:24:04,270
to be part of this study.

515
00:24:04,270 --> 00:24:07,720
I just got a letter in the mail
last week inviting me to go,

516
00:24:07,720 --> 00:24:10,780
and I was really
excited to see that.

517
00:24:10,780 --> 00:24:12,430
So all sorts of
different data is

518
00:24:12,430 --> 00:24:16,210
being created as a
result of these trends

519
00:24:16,210 --> 00:24:17,350
that I've been mentioning.

520
00:24:17,350 --> 00:24:20,050
And it ranges from unstructured
data, like clinical notes,

521
00:24:20,050 --> 00:24:23,860
to imaging, lab
tests, vital signs.

522
00:24:23,860 --> 00:24:27,580
Nowadays, what we used to think
about just as clinical data

523
00:24:27,580 --> 00:24:29,320
now has started to
really come to have

524
00:24:29,320 --> 00:24:33,010
a very tight tie to what we
think about as biological data.

525
00:24:33,010 --> 00:24:35,110
So data from genomics
and proteomics

526
00:24:35,110 --> 00:24:38,410
is starting to play a major
role in both clinical research

527
00:24:38,410 --> 00:24:40,750
and clinical practice.

528
00:24:40,750 --> 00:24:43,150
Of course, not everything
that we traditionally

529
00:24:43,150 --> 00:24:46,750
think about healthcare data--

530
00:24:46,750 --> 00:24:49,180
there are also some
non-traditional views

531
00:24:49,180 --> 00:24:49,910
on health.

532
00:24:49,910 --> 00:24:53,380
So for example, social
media is an interesting way

533
00:24:53,380 --> 00:24:59,440
of thinking through both
psychiatric disorders, where

534
00:24:59,440 --> 00:25:02,140
many of us will post things
on Facebook and other places

535
00:25:02,140 --> 00:25:04,060
about our mental
health, which give

536
00:25:04,060 --> 00:25:06,490
a lens on our mental health.

537
00:25:06,490 --> 00:25:09,630
Your phone, which is
tracking your activity,

538
00:25:09,630 --> 00:25:12,250
will give us a view
on how active we are.

539
00:25:12,250 --> 00:25:15,850
It might help us diagnose
early the variety of conditions

540
00:25:15,850 --> 00:25:18,560
as well that I'll mention later.

541
00:25:18,560 --> 00:25:22,100
So we have-- to this
whole theme right

542
00:25:22,100 --> 00:25:24,910
now is about what's changed
since the previous approaches

543
00:25:24,910 --> 00:25:26,388
at AI medicine.

544
00:25:26,388 --> 00:25:28,180
I've just talked about
data, but data alone

545
00:25:28,180 --> 00:25:29,960
is not nearly enough.

546
00:25:29,960 --> 00:25:31,910
The other major
change is that there

547
00:25:31,910 --> 00:25:36,830
has been decades' worth of work
on standardizing health data.

548
00:25:36,830 --> 00:25:40,510
So for example, when I
mentioned to you that when

549
00:25:40,510 --> 00:25:43,332
you go to a doctor's office,
and they send a bill,

550
00:25:43,332 --> 00:25:45,040
that bill is associated
with a diagnosis.

551
00:25:45,040 --> 00:25:49,610
And that diagnosis is coded in
a system called ICD-9 or ICD-10,

552
00:25:49,610 --> 00:25:51,350
which is a standardized
system where,

553
00:25:51,350 --> 00:25:54,400
for many, not all,
but many diseases,

554
00:25:54,400 --> 00:25:58,680
there is a corresponding
code associated with it.

555
00:25:58,680 --> 00:26:01,680
ICD-10, which was
recently rolled

556
00:26:01,680 --> 00:26:07,200
out nationwide about a year
ago is much more detailed

557
00:26:07,200 --> 00:26:09,210
than the previous
coding system, includes

558
00:26:09,210 --> 00:26:10,450
some interesting categories.

559
00:26:10,450 --> 00:26:14,370
For example, bitten by a
turtle has a code for it.

560
00:26:14,370 --> 00:26:17,640
Bitten by sea lion,
struck by [INAUDIBLE]..

561
00:26:17,640 --> 00:26:20,610
So it's starting to get
really detailed here,

562
00:26:20,610 --> 00:26:24,210
which has its benefits and its
disadvantages when it comes

563
00:26:24,210 --> 00:26:25,860
to research using that data.

564
00:26:25,860 --> 00:26:28,920
But certainly, we can do
more with detailed data

565
00:26:28,920 --> 00:26:32,580
than we could with
less detailed data.

566
00:26:32,580 --> 00:26:35,220
Laboratory test results are
standardized using a system

567
00:26:35,220 --> 00:26:38,880
called LOINC, here
in the United States.

568
00:26:38,880 --> 00:26:43,740
Every lab test order has
an associated code for it.

569
00:26:43,740 --> 00:26:46,470
I just want to point out briefly
that the values associated

570
00:26:46,470 --> 00:26:48,840
with those lab tests
are less standardized.

571
00:26:51,700 --> 00:26:55,700
Pharmacy, national drug codes
should be very familiar to you.

572
00:26:55,700 --> 00:26:59,860
If you take any medication
that you've been prescribed,

573
00:26:59,860 --> 00:27:02,740
and you look carefully,
you'll see a number on it,

574
00:27:02,740 --> 00:27:08,470
and you see 0015347911, that
number is unique to that

575
00:27:08,470 --> 00:27:09,110
medication.

576
00:27:09,110 --> 00:27:13,210
In fact, it's even unique to
the brand of that medication.

577
00:27:13,210 --> 00:27:15,260
And there's an associated
taxonomy with it.

578
00:27:15,260 --> 00:27:19,170
And so one can really understand
in a very structured way what

579
00:27:19,170 --> 00:27:21,670
medications a patient is on and
how those medications relate

580
00:27:21,670 --> 00:27:24,140
to one another.

581
00:27:24,140 --> 00:27:28,830
A lot of medical data is found
not in the structured form,

582
00:27:28,830 --> 00:27:32,790
but in free text, in
notes written by doctors.

583
00:27:32,790 --> 00:27:36,090
And these notes have,
often, lots of mentions

584
00:27:36,090 --> 00:27:39,180
of symptoms and
conditions in them.

585
00:27:39,180 --> 00:27:41,822
And one can try to
standardize those

586
00:27:41,822 --> 00:27:44,280
by mapping them to what's called
a unified medical language

587
00:27:44,280 --> 00:27:48,450
system, which is an
ontology with millions

588
00:27:48,450 --> 00:27:51,610
of different medical
concepts in them.

589
00:27:51,610 --> 00:27:54,400
So I'm not going to go
too much more into these.

590
00:27:54,400 --> 00:27:58,150
They'll be the subject of much
discussion in this semester,

591
00:27:58,150 --> 00:28:01,750
but particularly in the
next two lectures by Pete.

592
00:28:01,750 --> 00:28:05,800
But I want to talk very briefly
about what you can do once you

593
00:28:05,800 --> 00:28:09,460
have a standardized vocabulary.

594
00:28:09,460 --> 00:28:10,990
So one thing you
can do is you could

595
00:28:10,990 --> 00:28:14,710
build APIs, or Application
Programming Interfaces,

596
00:28:14,710 --> 00:28:18,520
for now sending that
data from place to place.

597
00:28:18,520 --> 00:28:23,770
And FHIR, F-H-I-R,
is a new standard,

598
00:28:23,770 --> 00:28:28,380
which has widespread adoption
now here in the United States

599
00:28:28,380 --> 00:28:36,610
for hospitals to provide data
both for downstream clinical

600
00:28:36,610 --> 00:28:38,990
purposes but also
directly to patients.

601
00:28:38,990 --> 00:28:41,470
And in this
standard, it will use

602
00:28:41,470 --> 00:28:43,240
many of the
vocabularies I mentioned

603
00:28:43,240 --> 00:28:51,220
to you in the previous slides to
encode diagnoses, medications,

604
00:28:51,220 --> 00:28:54,640
allergies, problems, and
even financial aspects

605
00:28:54,640 --> 00:28:57,160
that are relevant to the
care of this patient.

606
00:28:57,160 --> 00:28:59,970
And for those of you who have
an Apple phone, for example,

607
00:28:59,970 --> 00:29:03,010
and if you open up a
Apple Health Records,

608
00:29:03,010 --> 00:29:05,680
it makes use of this
standard to receive data

609
00:29:05,680 --> 00:29:07,975
from over 50
different hospitals.

610
00:29:07,975 --> 00:29:09,850
And you should expect
to see many competitors

611
00:29:09,850 --> 00:29:11,600
to them in the future,
because of the fact

612
00:29:11,600 --> 00:29:13,997
that it's now an open standard.

613
00:29:13,997 --> 00:29:16,080
Now other types of data,
like the health insurance

614
00:29:16,080 --> 00:29:17,940
claims I mentioned
earlier, is often

615
00:29:17,940 --> 00:29:20,970
encoded in a slightly
different data model.

616
00:29:20,970 --> 00:29:24,930
One which my lab works quite
a bit with is called OMOP,

617
00:29:24,930 --> 00:29:28,860
and it's being maintained by a
nonprofit organization called

618
00:29:28,860 --> 00:29:32,940
the Observational Health Data
Sciences Initiative Odyssey.

619
00:29:32,940 --> 00:29:36,360
And this common data
model gives a standard way

620
00:29:36,360 --> 00:29:38,970
of taking data from
an institution which

621
00:29:38,970 --> 00:29:41,520
might have its own intricacies
and really mapping it

622
00:29:41,520 --> 00:29:44,490
to this common language, so that
if you write a machine learning

623
00:29:44,490 --> 00:29:46,260
algorithm once, then
that machine learning

624
00:29:46,260 --> 00:29:48,420
algorithm reads in
data in this format,

625
00:29:48,420 --> 00:29:50,850
you can then apply it
somewhere else very easily.

626
00:29:50,850 --> 00:29:52,470
And the portions
of these standards

627
00:29:52,470 --> 00:29:55,410
really can't be
understated, the importance

628
00:29:55,410 --> 00:29:57,600
for translating what
we're doing in this class

629
00:29:57,600 --> 00:29:58,690
into clinical practice.

630
00:29:58,690 --> 00:30:00,398
And so we'll be
returning to these things

631
00:30:00,398 --> 00:30:03,130
throughout the semester.

632
00:30:03,130 --> 00:30:04,390
So we've talked about data.

633
00:30:04,390 --> 00:30:05,940
We've talked about standards.

634
00:30:05,940 --> 00:30:09,937
And the third wheel
is breakthroughs

635
00:30:09,937 --> 00:30:10,770
in machine learning.

636
00:30:10,770 --> 00:30:13,258
And this should be no surprise
to anyone in this room.

637
00:30:13,258 --> 00:30:15,300
All right, we've been
seeing time and time again,

638
00:30:15,300 --> 00:30:19,920
over the last five years,
benchmark after benchmark being

639
00:30:19,920 --> 00:30:23,040
improved upon and human
performance beaten

640
00:30:23,040 --> 00:30:25,350
by state-of-the-art machine
learning algorithms.

641
00:30:25,350 --> 00:30:28,020
Here I'm just
showing you a figure

642
00:30:28,020 --> 00:30:29,850
that I imagine many
of you have seen,

643
00:30:29,850 --> 00:30:33,210
on the error rates on the image
net competition for object

644
00:30:33,210 --> 00:30:34,200
recognition.

645
00:30:34,200 --> 00:30:37,290
The error rates
in 2011 were 25%.

646
00:30:37,290 --> 00:30:39,330
And even just a few
years ago, it already

647
00:30:39,330 --> 00:30:44,430
surpassed human
level to under 5%.

648
00:30:44,430 --> 00:30:48,930
Now, the changes that have led
to those advances in object

649
00:30:48,930 --> 00:30:52,860
recognition are going to have
some parallels in healthcare,

650
00:30:52,860 --> 00:30:54,850
but only up to some point.

651
00:30:54,850 --> 00:30:59,070
For example, there was big
data, large training sets

652
00:30:59,070 --> 00:31:01,230
that were critical for this.

653
00:31:01,230 --> 00:31:03,510
There were algorithmic
advances, in

654
00:31:03,510 --> 00:31:05,610
particular convolutional
neural networks,

655
00:31:05,610 --> 00:31:07,650
that played a huge role.

656
00:31:07,650 --> 00:31:10,770
And there was open source
software that was created,

657
00:31:10,770 --> 00:31:14,010
things like TensorFlow
and PyTorch,

658
00:31:14,010 --> 00:31:19,650
which allow a researcher or
industry worker in one place

659
00:31:19,650 --> 00:31:23,400
to very, very quickly
build upon successes

660
00:31:23,400 --> 00:31:25,530
from other researchers
in other places

661
00:31:25,530 --> 00:31:29,310
and then release the code,
so that one can really

662
00:31:29,310 --> 00:31:33,400
accelerate the rate of
progress in this field.

663
00:31:33,400 --> 00:31:35,530
Now, in terms of those
algorithmic advances that

664
00:31:35,530 --> 00:31:37,030
have made a big
difference, the ones

665
00:31:37,030 --> 00:31:39,250
that I would really
like to point out

666
00:31:39,250 --> 00:31:41,620
because of their
relevance to this course

667
00:31:41,620 --> 00:31:43,690
are learning with high
dimensional features.

668
00:31:43,690 --> 00:31:46,570
So this was really the
advances in the early 2000s,

669
00:31:46,570 --> 00:31:47,410
for example.

670
00:31:47,410 --> 00:31:51,040
And support vector machines and
learning with L1 regularization

671
00:31:51,040 --> 00:31:53,140
as a type of sparsity.

672
00:31:53,140 --> 00:31:56,350
And then more recently,
in the last six years,

673
00:31:56,350 --> 00:32:00,790
on stochastic gradient
descent, like methods

674
00:32:00,790 --> 00:32:04,083
for very rapidly solving these
convex optimization problems,

675
00:32:04,083 --> 00:32:05,500
that will play a
huge role in what

676
00:32:05,500 --> 00:32:07,445
we'll be doing in this course.

677
00:32:07,445 --> 00:32:08,820
In the last few
years, there have

678
00:32:08,820 --> 00:32:10,300
been a huge amount
of progress in

679
00:32:10,300 --> 00:32:12,640
unsupervised and semi-supervised
learning algorithms.

680
00:32:12,640 --> 00:32:14,410
And as I'll tell you
about much later,

681
00:32:14,410 --> 00:32:16,360
one of the major
challenges in healthcare

682
00:32:16,360 --> 00:32:20,050
is that despite the fact that
we have a large amount of data,

683
00:32:20,050 --> 00:32:22,480
we have very little
labeled data.

684
00:32:22,480 --> 00:32:24,845
And so these semi-supervised
learning algorithms

685
00:32:24,845 --> 00:32:26,470
are going to play a
major role in being

686
00:32:26,470 --> 00:32:29,650
able to really take advantage
of the data that we do have.

687
00:32:29,650 --> 00:32:32,485
And then of course the modern
deep learning algorithms.

688
00:32:32,485 --> 00:32:34,860
Convolutional neural networks,
recurrent neural networks,

689
00:32:34,860 --> 00:32:37,180
and ways of trying
to train them.

690
00:32:37,180 --> 00:32:40,600
So those played a major
role in the advances

691
00:32:40,600 --> 00:32:41,832
in the tech industry.

692
00:32:41,832 --> 00:32:44,290
And to some extent, they'll
play a major role in healthcare

693
00:32:44,290 --> 00:32:45,520
as well.

694
00:32:45,520 --> 00:32:47,230
And I'll point out a
few examples of that

695
00:32:47,230 --> 00:32:50,720
in the rest of today's lecture.

696
00:32:50,720 --> 00:32:56,470
So all of this coming together,
the data availability,

697
00:32:56,470 --> 00:32:59,530
the advances in other
fields of machine learning,

698
00:32:59,530 --> 00:33:04,276
and the huge amount of potential
financial gain in healthcare

699
00:33:04,276 --> 00:33:07,300
and the potential social
impact it could have

700
00:33:07,300 --> 00:33:08,668
has not gone unnoticed.

701
00:33:08,668 --> 00:33:10,210
And there's a huge
amount of industry

702
00:33:10,210 --> 00:33:11,800
interested in this field.

703
00:33:11,800 --> 00:33:15,260
These are just some examples
from names I think many of you

704
00:33:15,260 --> 00:33:18,150
are familiar with, like
DeepMind Health and IBM Watson

705
00:33:18,150 --> 00:33:20,250
to startup companies
like Bay Labs

706
00:33:20,250 --> 00:33:23,650
and PathAI, which is here
in Boston, all of which

707
00:33:23,650 --> 00:33:27,190
are really trying to build
the next generation of tools

708
00:33:27,190 --> 00:33:30,550
for healthcare, now based on
machine learning algorithms.

709
00:33:30,550 --> 00:33:35,610
There's been billions
of dollars of funding

710
00:33:35,610 --> 00:33:39,130
in the recent quarters towards
digital health efforts,

711
00:33:39,130 --> 00:33:40,990
with hundreds of
different startups

712
00:33:40,990 --> 00:33:44,590
that are focused specifically
on using artificial intelligence

713
00:33:44,590 --> 00:33:46,870
and healthcare.

714
00:33:46,870 --> 00:33:49,990
And there's the
recognition that data

715
00:33:49,990 --> 00:33:52,030
is so essential to
this process has

716
00:33:52,030 --> 00:33:55,330
led to an all-out purchasing
effort to try to get as much

717
00:33:55,330 --> 00:33:56,710
of that data as you can.

718
00:33:56,710 --> 00:34:00,850
So for example, IBM
purchased a company

719
00:34:00,850 --> 00:34:04,870
called Merge, which made
medical imaging software

720
00:34:04,870 --> 00:34:07,990
and thus had accumulated a large
amount of medical imaging data

721
00:34:07,990 --> 00:34:10,480
for $1 billion in 2015.

722
00:34:10,480 --> 00:34:14,480
They purchased Truven
for $2.6 billion in 2016.

723
00:34:14,480 --> 00:34:16,900
Flatiron Health, which
is a company in New York

724
00:34:16,900 --> 00:34:20,230
City focused on oncology, was
purchased for almost $2 billion

725
00:34:20,230 --> 00:34:23,230
by Roche, a pharmaceutical
company, just last year.

726
00:34:23,230 --> 00:34:25,627
And there's several more
of these industry moves.

727
00:34:25,627 --> 00:34:28,210
Again, I'm just tying to get you
thinking about what it really

728
00:34:28,210 --> 00:34:30,969
takes in this field, and
getting access to data

729
00:34:30,969 --> 00:34:35,280
is actually a really
important one, obviously.

730
00:34:35,280 --> 00:34:39,570
So let's now move
on to some examples

731
00:34:39,570 --> 00:34:44,340
of how machine learning
will transform healthcare.

732
00:34:44,340 --> 00:34:47,170
To begin with, I want to really
lay out the landscape here

733
00:34:47,170 --> 00:34:49,260
and define some language.

734
00:34:49,260 --> 00:34:51,120
There are a number
of different players

735
00:34:51,120 --> 00:34:53,340
when it comes to the
healthcare space.

736
00:34:53,340 --> 00:34:56,790
They're us, patients, consumers.

737
00:34:56,790 --> 00:34:59,012
They are the doctors
that we go to,

738
00:34:59,012 --> 00:35:00,720
which you could think
about as providers.

739
00:35:00,720 --> 00:35:02,345
But of course they're
not just doctors,

740
00:35:02,345 --> 00:35:08,250
they're also nurses and
community health workers

741
00:35:08,250 --> 00:35:10,200
and so on.

742
00:35:10,200 --> 00:35:14,457
There are payers, which
provide the-- where there is--

743
00:35:14,457 --> 00:35:16,290
these edges are really
showing relationships

744
00:35:16,290 --> 00:35:19,770
between the different
players, so our consumers, we

745
00:35:19,770 --> 00:35:21,990
often, either from our
job or directly from us,

746
00:35:21,990 --> 00:35:24,900
we will pay premiums for a
health insurance company,

747
00:35:24,900 --> 00:35:26,785
to a health insurance
company, and then

748
00:35:26,785 --> 00:35:29,160
that health insurance company
is responsible for payments

749
00:35:29,160 --> 00:35:33,720
to the providers to provide
services to us patients.

750
00:35:33,720 --> 00:35:38,130
Now, here in the US, the
payers are both commercial

751
00:35:38,130 --> 00:35:39,150
and governmental.

752
00:35:39,150 --> 00:35:44,760
So many of you will know
companies like Cigna or Aetna

753
00:35:44,760 --> 00:35:47,010
or Blue Cross, which
are commercial providers

754
00:35:47,010 --> 00:35:49,830
of healthcare, of
health insurance,

755
00:35:49,830 --> 00:35:52,170
but there are also
governmental ones.

756
00:35:52,170 --> 00:35:56,250
For example, the Veterans
Health Administration

757
00:35:56,250 --> 00:35:59,700
runs one of the biggest health
organizations in the United

758
00:35:59,700 --> 00:36:03,210
States, servicing our
veterans from the department,

759
00:36:03,210 --> 00:36:06,240
people who have retired from
the Department of Defense, which

760
00:36:06,240 --> 00:36:08,580
has the one of
the second biggest

761
00:36:08,580 --> 00:36:12,030
health systems, the
Defense Health Agency.

762
00:36:12,030 --> 00:36:13,763
And that is an
organization where--

763
00:36:13,763 --> 00:36:15,180
both of those
organizations, where

764
00:36:15,180 --> 00:36:18,840
both the payer and the
provider are really one.

765
00:36:18,840 --> 00:36:21,930
The Center for Medicare
and Medicaid Services

766
00:36:21,930 --> 00:36:26,070
here in the US provides
health insurance

767
00:36:26,070 --> 00:36:31,560
for all retirees in
the United States.

768
00:36:31,560 --> 00:36:34,710
And also Medicaid, which
is run at a state level,

769
00:36:34,710 --> 00:36:39,852
provides health insurance
to a variety of individuals

770
00:36:39,852 --> 00:36:41,310
who would otherwise
have difficulty

771
00:36:41,310 --> 00:36:44,460
purchasing or obtaining
their own health insurance.

772
00:36:44,460 --> 00:36:47,547
And those are examples of
state-run or federally run

773
00:36:47,547 --> 00:36:48,630
health insurance agencies.

774
00:36:51,170 --> 00:36:53,390
And then internationally,
sometimes the lines

775
00:36:53,390 --> 00:36:54,380
are even more blurred.

776
00:36:54,380 --> 00:36:57,290
So of course in places
like the United Kingdom,

777
00:36:57,290 --> 00:37:02,210
where you have a government-run
health system, the National

778
00:37:02,210 --> 00:37:07,940
Health Service, you have the
same system both paying for

779
00:37:07,940 --> 00:37:10,410
and providing the services.

780
00:37:10,410 --> 00:37:12,560
Now, the reason why this
is really important for us

781
00:37:12,560 --> 00:37:14,540
to think about
already in lecture one

782
00:37:14,540 --> 00:37:18,500
is because what's so
essential about this field

783
00:37:18,500 --> 00:37:21,200
is figuring out where the
knob is that you can turn

784
00:37:21,200 --> 00:37:22,580
to try to improve healthcare.

785
00:37:22,580 --> 00:37:24,770
Where can we deploy
machine learning algorithms

786
00:37:24,770 --> 00:37:26,420
within healthcare?

787
00:37:26,420 --> 00:37:29,360
So some algorithms are going
to be better run by providers,

788
00:37:29,360 --> 00:37:31,730
others are going to be
better run by payers,

789
00:37:31,730 --> 00:37:34,730
others are going to be
directly provided to patients,

790
00:37:34,730 --> 00:37:36,870
and some all of the above.

791
00:37:36,870 --> 00:37:39,410
We also have to think
about industrial questions,

792
00:37:39,410 --> 00:37:42,950
in terms of what is it going to
take to develop a new product.

793
00:37:42,950 --> 00:37:45,170
Who will pay for this product?

794
00:37:45,170 --> 00:37:46,910
Which is again an
important question

795
00:37:46,910 --> 00:37:50,110
when it comes to
deploying algorithms here.

796
00:37:50,110 --> 00:37:55,300
So I'll run through a couple of
very high-level examples driven

797
00:37:55,300 --> 00:37:59,620
from my own work, focused
on the provider space,

798
00:37:59,620 --> 00:38:03,770
and then I'll bump up to
talk a bit more broadly.

799
00:38:03,770 --> 00:38:05,740
So for the last
seven or eight years,

800
00:38:05,740 --> 00:38:07,833
I've been doing a lot
of work in collaboration

801
00:38:07,833 --> 00:38:09,250
with Beth Israel
Deaconess Medical

802
00:38:09,250 --> 00:38:12,633
Center, across the river, with
their emergency department.

803
00:38:12,633 --> 00:38:14,800
And the emergency department
is a really interesting

804
00:38:14,800 --> 00:38:17,410
clinical setting, because
you have a very short period

805
00:38:17,410 --> 00:38:19,750
of time from when a patient
comes into the hospital

806
00:38:19,750 --> 00:38:23,762
to diagnose what's going on
with them, to initiate therapy,

807
00:38:23,762 --> 00:38:25,220
and then to decide
what to do next.

808
00:38:25,220 --> 00:38:26,650
Do you keep them
in the hospital?

809
00:38:26,650 --> 00:38:28,810
Do you send them home?

810
00:38:28,810 --> 00:38:30,550
If you-- for each
one of those things,

811
00:38:30,550 --> 00:38:32,670
what should the most
immediate actions be?

812
00:38:32,670 --> 00:38:36,730
And at least here in the US,
we're always understaffed.

813
00:38:36,730 --> 00:38:40,900
So we've got limited resources
and very critical decisions

814
00:38:40,900 --> 00:38:42,560
to make.

815
00:38:42,560 --> 00:38:45,295
So this is one example
of a setting where

816
00:38:45,295 --> 00:38:47,170
algorithms that are
running behind the scenes

817
00:38:47,170 --> 00:38:49,628
could potentially really help
with some of the challenges I

818
00:38:49,628 --> 00:38:51,460
mentioned earlier.

819
00:38:51,460 --> 00:38:54,340
So for example, one could
imagine an algorithm which

820
00:38:54,340 --> 00:38:56,140
builds on techniques
like what I mentioned

821
00:38:56,140 --> 00:38:59,620
to you for an internist one
or quick medical reference,

822
00:38:59,620 --> 00:39:03,130
try to reason about what's
going on with the patient based

823
00:39:03,130 --> 00:39:07,628
on the data that's available
for the patient, the symptoms.

824
00:39:07,628 --> 00:39:09,670
But the modern view of
this shouldn't, of course,

825
00:39:09,670 --> 00:39:12,670
use binary indicators
of each symptom, which

826
00:39:12,670 --> 00:39:15,460
have to be entered in manually,
but rather all of these things

827
00:39:15,460 --> 00:39:16,960
should be
automatically extracted

828
00:39:16,960 --> 00:39:21,282
from the electronic medical
record or listed as necessary.

829
00:39:21,282 --> 00:39:22,990
And then if one could
reason about what's

830
00:39:22,990 --> 00:39:25,498
going on with a patient,
we wouldn't necessarily

831
00:39:25,498 --> 00:39:27,790
want to use it for a diagnosis,
although in some cases,

832
00:39:27,790 --> 00:39:29,832
you might use it for
an earlier diagnosis.

833
00:39:29,832 --> 00:39:32,290
But it could also be used for
a number of other more subtle

834
00:39:32,290 --> 00:39:34,900
interventions, for
example, better triage

835
00:39:34,900 --> 00:39:38,050
to figure out which patients
need to be seen first.

836
00:39:38,050 --> 00:39:40,840
Early detection of adverse
events or recognition

837
00:39:40,840 --> 00:39:44,230
that there might be some unusual
actions which might actually

838
00:39:44,230 --> 00:39:47,980
be medical errors that you
want to surface now and draw

839
00:39:47,980 --> 00:39:49,360
attention to.

840
00:39:49,360 --> 00:39:53,710
Now, you could also
use this understanding

841
00:39:53,710 --> 00:39:55,090
of what's going
on with a patient

842
00:39:55,090 --> 00:39:56,590
to change the way
that clinicians

843
00:39:56,590 --> 00:39:59,390
interact with patient data.

844
00:39:59,390 --> 00:40:05,560
So for example, one can try
to propagate best practices

845
00:40:05,560 --> 00:40:08,912
by surfacing clinical
decision support,

846
00:40:08,912 --> 00:40:10,870
automatically triggering
this clinical decision

847
00:40:10,870 --> 00:40:14,150
support for patients that you
think it might be relevant for.

848
00:40:14,150 --> 00:40:15,940
And here's one
example, where it says,

849
00:40:15,940 --> 00:40:19,210
the ED Dashboard, the Emergency
Department Dashboard decision

850
00:40:19,210 --> 00:40:20,860
support algorithms
have determined

851
00:40:20,860 --> 00:40:25,180
this patient may be eligible for
the atria cellulitis pathway.

852
00:40:25,180 --> 00:40:29,300
Cellulitis is often
caused by infections.

853
00:40:29,300 --> 00:40:30,890
Please choose from
one of the options.

854
00:40:30,890 --> 00:40:33,360
Enroll in the pathway, decline--

855
00:40:33,360 --> 00:40:35,390
and if you decline,
you must include

856
00:40:35,390 --> 00:40:38,850
a comment for the reviewers.

857
00:40:38,850 --> 00:40:43,410
Now, if you clicked on enroll
in the pathway, at that moment,

858
00:40:43,410 --> 00:40:45,240
machine learning disappears.

859
00:40:45,240 --> 00:40:48,030
Rather, there is a
standardized process.

860
00:40:48,030 --> 00:40:51,240
It's an algorithm, but it's
a deterministic algorithm,

861
00:40:51,240 --> 00:40:55,290
for how patients with cellulitis
should be properly managed,

862
00:40:55,290 --> 00:40:57,660
diagnosed, and treated.

863
00:40:57,660 --> 00:41:00,930
That algorithm comes
from best practices,

864
00:41:00,930 --> 00:41:07,140
comes from clinicians coming
together, analyzing past data,

865
00:41:07,140 --> 00:41:09,120
understanding what
would be good ways

866
00:41:09,120 --> 00:41:12,300
to treat patients of this
type, and then formalizing

867
00:41:12,300 --> 00:41:14,310
that in a document.

868
00:41:14,310 --> 00:41:16,530
The challenge is that there
might be hundreds or even

869
00:41:16,530 --> 00:41:19,070
thousands of these
best practices.

870
00:41:19,070 --> 00:41:21,330
And in an academic
medical center, where

871
00:41:21,330 --> 00:41:22,770
you have patients coming--

872
00:41:22,770 --> 00:41:26,970
where you have medical
students or residents who

873
00:41:26,970 --> 00:41:30,503
are very quickly rotating
through the system and thus

874
00:41:30,503 --> 00:41:31,920
may not be familiar
with which are

875
00:41:31,920 --> 00:41:35,700
the most appropriate clinical
guidelines to use for any one

876
00:41:35,700 --> 00:41:37,530
patient in this institution.

877
00:41:37,530 --> 00:41:42,000
Or if you go to a
rural site, where

878
00:41:42,000 --> 00:41:45,000
this academic nature of
thinking through what

879
00:41:45,000 --> 00:41:47,250
the right clinical guidelines
are is a little bit less

880
00:41:47,250 --> 00:41:51,480
of the mainstream, everyday
activity, the question of which

881
00:41:51,480 --> 00:41:53,255
one to use when is
very challenging.

882
00:41:53,255 --> 00:41:55,380
And so that's where the
machine learning algorithms

883
00:41:55,380 --> 00:41:56,097
can come in.

884
00:41:56,097 --> 00:41:58,180
By reasoning about what's
going on with a patient,

885
00:41:58,180 --> 00:41:59,340
you might have a
good guess of what

886
00:41:59,340 --> 00:42:00,570
might be appropriate
for this patient,

887
00:42:00,570 --> 00:42:02,278
and you use that to
automatically surface

888
00:42:02,278 --> 00:42:05,210
the right clinical
decisions for a trigger.

889
00:42:05,210 --> 00:42:07,130
Another example
is by just trying

890
00:42:07,130 --> 00:42:08,630
to anticipate clinician needs.

891
00:42:08,630 --> 00:42:12,890
So for example, if you think
that this patient might

892
00:42:12,890 --> 00:42:16,790
be coming in for a psychiatric
condition, or maybe

893
00:42:16,790 --> 00:42:19,505
you recognize that the
patient came in that triage

894
00:42:19,505 --> 00:42:21,230
and was complaining
of chest pain,

895
00:42:21,230 --> 00:42:23,090
then there might
be a psych order

896
00:42:23,090 --> 00:42:26,660
set, which includes
laboratory test results that

897
00:42:26,660 --> 00:42:29,000
are relevant for
psychiatric patients,

898
00:42:29,000 --> 00:42:34,070
or a chest pain order set, which
includes both laboratory tests

899
00:42:34,070 --> 00:42:38,610
and interventions, like aspirin,
that might be suggested.

900
00:42:38,610 --> 00:42:41,325
Now, these are also examples
where these order sets are not

901
00:42:41,325 --> 00:42:42,950
created by machine
learning algorithms.

902
00:42:42,950 --> 00:42:44,950
Although that's something
we could discuss later

903
00:42:44,950 --> 00:42:45,890
in the semester.

904
00:42:45,890 --> 00:42:47,243
Rather, they're standardized.

905
00:42:47,243 --> 00:42:49,160
But the goal of the
machine learning algorithm

906
00:42:49,160 --> 00:42:51,950
is just to figure out
which ones to show when

907
00:42:51,950 --> 00:42:54,158
directly to the clinicians.

908
00:42:54,158 --> 00:42:55,700
I'm showing you
these examples to try

909
00:42:55,700 --> 00:42:59,270
to point out that diagnosis
isn't the whole story.

910
00:42:59,270 --> 00:43:01,730
Thinking through what are the
more subtle interventions we

911
00:43:01,730 --> 00:43:05,180
can do with machine learning
and AI and healthcare

912
00:43:05,180 --> 00:43:07,890
is going to be really
important to having the impact

913
00:43:07,890 --> 00:43:10,650
that it could have.

914
00:43:10,650 --> 00:43:13,770
So other examples, now a bit
more on the diagnosis style,

915
00:43:13,770 --> 00:43:16,050
are reducing the need
for specialist consults.

916
00:43:16,050 --> 00:43:17,880
So you might have
a patient come in,

917
00:43:17,880 --> 00:43:21,570
and it might be really quick
to get the patient in front

918
00:43:21,570 --> 00:43:24,570
of an X-ray to do a
chest X-ray, but then

919
00:43:24,570 --> 00:43:26,910
finding the radiologist
to review that X-ray

920
00:43:26,910 --> 00:43:29,220
could take a lot of time.

921
00:43:29,220 --> 00:43:32,910
And in some places,
radiologist consults

922
00:43:32,910 --> 00:43:36,760
could take days, depending on
the urgency of the condition.

923
00:43:36,760 --> 00:43:39,720
So this is an area where
data is quite standardized.

924
00:43:39,720 --> 00:43:42,960
In fact, MIT just
released last week

925
00:43:42,960 --> 00:43:45,720
a data set of
300,000 chest x-rays

926
00:43:45,720 --> 00:43:48,240
with associated labels on them.

927
00:43:48,240 --> 00:43:50,490
And one could try to ask
the question of could we

928
00:43:50,490 --> 00:43:51,960
build machine
learning algorithms

929
00:43:51,960 --> 00:43:54,210
using the convolutional
neural network type

930
00:43:54,210 --> 00:43:56,820
techniques that we've seen
play a big role in object

931
00:43:56,820 --> 00:43:58,500
recognition to try
to understand what's

932
00:43:58,500 --> 00:43:59,340
going on with this patient.

933
00:43:59,340 --> 00:44:01,048
For example, in this
case, the prediction

934
00:44:01,048 --> 00:44:04,540
is the patient has pneumonia,
from this chest X-ray.

935
00:44:04,540 --> 00:44:06,340
And using those
systems, it could

936
00:44:06,340 --> 00:44:10,120
help both reduce the load
of radiology consults,

937
00:44:10,120 --> 00:44:13,180
and it could allow us to really
translate these algorithms

938
00:44:13,180 --> 00:44:14,680
to settings which
might be much more

939
00:44:14,680 --> 00:44:19,280
resource poor, for example,
in developing nations.

940
00:44:19,280 --> 00:44:20,810
Now, the same
sorts of techniques

941
00:44:20,810 --> 00:44:22,590
can be used for other
data modalities.

942
00:44:22,590 --> 00:44:27,500
So this is an
example of data that

943
00:44:27,500 --> 00:44:30,230
could be obtained from an EKG.

944
00:44:30,230 --> 00:44:33,560
And from looking
at this EKG, one

945
00:44:33,560 --> 00:44:36,140
can try to predict, does the
patient have a heart condition,

946
00:44:36,140 --> 00:44:38,060
such as an arrhythmia.

947
00:44:38,060 --> 00:44:40,880
Now, these types of
data used to just

948
00:44:40,880 --> 00:44:43,280
be obtained when you go
to a doctor's office.

949
00:44:43,280 --> 00:44:45,470
But today, they're
available to all of us.

950
00:44:45,470 --> 00:44:49,550
For example, in Apple's most
recent watch that was released,

951
00:44:49,550 --> 00:44:53,662
it has a single-lead
EKG built into it,

952
00:44:53,662 --> 00:44:56,120
which can attempt to predict
if a patient has an arrhythmia

953
00:44:56,120 --> 00:44:56,790
or not.

954
00:44:56,790 --> 00:44:58,207
And there are a
lot of subtleties,

955
00:44:58,207 --> 00:45:01,192
of course, around what it took
to get regulatory approval

956
00:45:01,192 --> 00:45:02,900
for that, which we'll
be discussing later

957
00:45:02,900 --> 00:45:06,080
in the semester, and how one
safely deploys such algorithms

958
00:45:06,080 --> 00:45:08,060
directly to consumers.

959
00:45:08,060 --> 00:45:10,370
And there, there are a
variety of techniques

960
00:45:10,370 --> 00:45:11,405
that could be used.

961
00:45:11,405 --> 00:45:13,160
And in a few lectures,
I'll talk to you

962
00:45:13,160 --> 00:45:16,790
about techniques from
the '80s and '90s

963
00:45:16,790 --> 00:45:19,430
which were based on trying
to signal processing,

964
00:45:19,430 --> 00:45:22,220
trying to detect where are
the peaks of the signal,

965
00:45:22,220 --> 00:45:24,260
look at a distance
between peaks.

966
00:45:24,260 --> 00:45:26,600
And more recently, because
of the large wealth

967
00:45:26,600 --> 00:45:28,160
of data that is
available, we've been

968
00:45:28,160 --> 00:45:30,230
using convolutional neural
network-based approaches

969
00:45:30,230 --> 00:45:32,360
to try to understand this
data and predict from it.

970
00:45:35,330 --> 00:45:38,180
Yet another example
from the ER really

971
00:45:38,180 --> 00:45:41,420
has to do with not how do we
care for the patient today,

972
00:45:41,420 --> 00:45:43,940
but how do we get
better data, which

973
00:45:43,940 --> 00:45:45,590
will then result
in taking better

974
00:45:45,590 --> 00:45:47,960
care of the patient tomorrow.

975
00:45:47,960 --> 00:45:50,690
And so one example
of that, which

976
00:45:50,690 --> 00:45:52,580
my group deployed at
Beth Israel Deaconess,

977
00:45:52,580 --> 00:45:55,220
and it's still running there
in the emergency department,

978
00:45:55,220 --> 00:45:58,460
has to do with getting higher
quality chief complaints.

979
00:45:58,460 --> 00:46:02,390
The chief complaint is
usually a very short,

980
00:46:02,390 --> 00:46:06,950
two or three word quantity, like
left knee pain, rectal pain,

981
00:46:06,950 --> 00:46:11,390
right upper quadrant,
RUQ, abdominal pain.

982
00:46:11,390 --> 00:46:13,220
And it's just a
very quick summary

983
00:46:13,220 --> 00:46:16,160
of why did the patient
come into the ER today.

984
00:46:16,160 --> 00:46:20,160
And despite the fact
that it's so few words,

985
00:46:20,160 --> 00:46:23,452
it plays a huge role in
the care of a patient.

986
00:46:23,452 --> 00:46:25,160
If you look at the
big screens in the ER,

987
00:46:25,160 --> 00:46:27,650
which summarize who are the
patients and on what beds,

988
00:46:27,650 --> 00:46:30,050
they have the chief
complaint next to it.

989
00:46:30,050 --> 00:46:34,130
Chief complaints are used as
criteria for enrolling patients

990
00:46:34,130 --> 00:46:35,690
in clinical trials.

991
00:46:35,690 --> 00:46:39,500
It's used as criteria for doing
retrospective quality research

992
00:46:39,500 --> 00:46:42,120
to see how do we care for
patients in a particular type.

993
00:46:42,120 --> 00:46:43,790
So it plays a very big role.

994
00:46:43,790 --> 00:46:46,100
But unfortunately, the data
that we've been getting

995
00:46:46,100 --> 00:46:47,780
has been crap.

996
00:46:47,780 --> 00:46:50,180
And that's because
it was free text,

997
00:46:50,180 --> 00:46:52,070
and it was sufficiently
high dimensional

998
00:46:52,070 --> 00:46:54,140
that just attempting
to standardize it

999
00:46:54,140 --> 00:46:57,433
with a big dropdown list,
like you see over here,

1000
00:46:57,433 --> 00:46:59,100
would have killed the
clinical workflow.

1001
00:46:59,100 --> 00:47:01,010
It would've taken
way too much time

1002
00:47:01,010 --> 00:47:04,070
for clinicians to try to
find the relevant one.

1003
00:47:04,070 --> 00:47:06,037
And so it just wouldn't
have been used.

1004
00:47:06,037 --> 00:47:08,120
And that's where some very
simple machine learning

1005
00:47:08,120 --> 00:47:10,740
algorithms turned out
to be really valuable.

1006
00:47:10,740 --> 00:47:13,922
So for example, we changed
the workflow altogether.

1007
00:47:13,922 --> 00:47:16,130
Rather than the chief
complaint being the first thing

1008
00:47:16,130 --> 00:47:19,790
that the triage nurse assigns
when the patient comes in,

1009
00:47:19,790 --> 00:47:21,140
it's the last thing.

1010
00:47:21,140 --> 00:47:24,030
First, the nurse takes the vital
signs, patient's temperature,

1011
00:47:24,030 --> 00:47:26,250
heart rate, blood
pressure, respiratory rate,

1012
00:47:26,250 --> 00:47:27,538
and oxygen saturation.

1013
00:47:27,538 --> 00:47:28,580
They talk to the patient.

1014
00:47:28,580 --> 00:47:31,397
They write up a 10-word or
30-word note about what's

1015
00:47:31,397 --> 00:47:32,480
going on with the patient.

1016
00:47:32,480 --> 00:47:37,940
Here it says, "69-year-old
male patient with severe

1017
00:47:37,940 --> 00:47:40,910
intermittent right
upper quadrant pain.

1018
00:47:40,910 --> 00:47:42,840
Began soon after eating.

1019
00:47:42,840 --> 00:47:44,390
Also is a heavy drinker."

1020
00:47:44,390 --> 00:47:46,730
So quite a bit of
information in that.

1021
00:47:46,730 --> 00:47:47,330
We take that.

1022
00:47:47,330 --> 00:47:49,705
We use a machine learning
algorithm, a supervised machine

1023
00:47:49,705 --> 00:47:52,130
learning algorithm in
this case, to predict

1024
00:47:52,130 --> 00:47:54,170
a set of chief
complaints now drawn

1025
00:47:54,170 --> 00:47:56,420
from a standardized ontology.

1026
00:47:56,420 --> 00:47:59,113
We show the five most likely
ones, and the clinician,

1027
00:47:59,113 --> 00:48:01,280
in this case, a nurse, could
just click one of them,

1028
00:48:01,280 --> 00:48:03,950
and it would enter
it into there.

1029
00:48:03,950 --> 00:48:10,340
We also allow the nurse to type
in part of a chief complaint.

1030
00:48:10,340 --> 00:48:13,160
But rather than just
doing a text matching

1031
00:48:13,160 --> 00:48:16,940
to find words that match
what's being typed in,

1032
00:48:16,940 --> 00:48:19,190
we do a contextual autocomplete.

1033
00:48:19,190 --> 00:48:21,290
So we use our
predictions to prioritize

1034
00:48:21,290 --> 00:48:23,900
what's the most likely chief
complaint that contains

1035
00:48:23,900 --> 00:48:25,610
that sequence of characters.

1036
00:48:25,610 --> 00:48:28,240
And that way it's
way faster to enter

1037
00:48:28,240 --> 00:48:29,660
in the relevant information.

1038
00:48:29,660 --> 00:48:31,580
And what we found
is that over time,

1039
00:48:31,580 --> 00:48:34,123
we got much higher
quality data out.

1040
00:48:34,123 --> 00:48:35,540
And again, this
is something we'll

1041
00:48:35,540 --> 00:48:39,860
be talking about in one of
our lectures in this course.

1042
00:48:39,860 --> 00:48:42,860
So I just gave you an
example, a few examples,

1043
00:48:42,860 --> 00:48:45,020
of how machine learning
and artificial tolerance

1044
00:48:45,020 --> 00:48:47,690
will transform the
provider space,

1045
00:48:47,690 --> 00:48:49,430
but now I want to
jump up a level

1046
00:48:49,430 --> 00:48:52,490
and think through not how
do we treat a patient today,

1047
00:48:52,490 --> 00:48:55,790
but how do we think about the
progression of a patient's

1048
00:48:55,790 --> 00:48:58,550
chronic disease over
a period of years.

1049
00:48:58,550 --> 00:49:00,920
It could be 10 years, 20 years.

1050
00:49:00,920 --> 00:49:04,760
And this question of how do
we manage chronic disease

1051
00:49:04,760 --> 00:49:08,390
is something which affects
all aspects of the healthcare

1052
00:49:08,390 --> 00:49:09,440
ecosystem.

1053
00:49:09,440 --> 00:49:11,510
It'll be used by
providers, payers,

1054
00:49:11,510 --> 00:49:15,590
and also by patients themselves.

1055
00:49:15,590 --> 00:49:18,440
So consider a patient with
chronic kidney disease.

1056
00:49:18,440 --> 00:49:23,390
Chronic kidney disease, it
typically only gets worse.

1057
00:49:23,390 --> 00:49:26,900
So you might start with
the patient being healthy

1058
00:49:26,900 --> 00:49:28,790
and then have some
increased risk.

1059
00:49:28,790 --> 00:49:31,700
Eventually, they have
some kidney damage.

1060
00:49:31,700 --> 00:49:35,270
Over time, they
reach kidney failure.

1061
00:49:35,270 --> 00:49:37,640
And once they reach
kidney failure,

1062
00:49:37,640 --> 00:49:45,368
typically, they need dialysis
or a kidney transplant.

1063
00:49:45,368 --> 00:49:47,160
But understanding when
each of these things

1064
00:49:47,160 --> 00:49:48,850
is going to happen for
patients is actually really,

1065
00:49:48,850 --> 00:49:50,070
really challenging.

1066
00:49:50,070 --> 00:49:53,300
Right now, we have one way
of trying to stage patients.

1067
00:49:53,300 --> 00:49:56,820
The standard approach
is known as the EGFR.

1068
00:49:56,820 --> 00:50:00,030
It's derived predominantly from
the patient's creatinine, which

1069
00:50:00,030 --> 00:50:03,150
is a blood test
result, and their age.

1070
00:50:03,150 --> 00:50:04,470
And it gives you a number out.

1071
00:50:04,470 --> 00:50:05,640
And from that
number, you can get

1072
00:50:05,640 --> 00:50:07,740
some sense of where the
patient is in this trajectory.

1073
00:50:07,740 --> 00:50:09,330
But it's really coarse
grained, and it's not

1074
00:50:09,330 --> 00:50:11,122
at all predictive about
when the patient is

1075
00:50:11,122 --> 00:50:14,920
going to progress to the
next stage of the disease.

1076
00:50:14,920 --> 00:50:17,400
Now, other conditions,
for example,

1077
00:50:17,400 --> 00:50:19,990
some cancers, like I'll
tell you about next,

1078
00:50:19,990 --> 00:50:22,480
don't follow that
linear trajectory.

1079
00:50:22,480 --> 00:50:25,068
Rather, patients' conditions
and the disease burden,

1080
00:50:25,068 --> 00:50:26,860
which is what I'm
showing you in the y-axis

1081
00:50:26,860 --> 00:50:31,300
here, might get worse,
better, worse again, better

1082
00:50:31,300 --> 00:50:33,460
again, worse again, and
so on, and of course

1083
00:50:33,460 --> 00:50:35,560
is a function of the
treatment for the patient

1084
00:50:35,560 --> 00:50:37,485
and other things that
are going on with them.

1085
00:50:37,485 --> 00:50:40,150
And understanding
what influences,

1086
00:50:40,150 --> 00:50:42,670
how a patient's disease
is going to progress,

1087
00:50:42,670 --> 00:50:46,120
and when is that
progression going to happen,

1088
00:50:46,120 --> 00:50:48,790
could be enormously valuable for
many of those different parts

1089
00:50:48,790 --> 00:50:50,020
of the healthcare ecosystem.

1090
00:50:54,490 --> 00:50:57,010
So one concrete example of
how that type of prediction

1091
00:50:57,010 --> 00:51:00,410
could be used would be in a
type of precision medicine.

1092
00:51:00,410 --> 00:51:04,510
So returning back to the
example that I mentioned

1093
00:51:04,510 --> 00:51:06,370
in the very beginning
of today's lecture

1094
00:51:06,370 --> 00:51:09,850
of multiple myeloma, which
I said my mother died of,

1095
00:51:09,850 --> 00:51:12,730
there are a large number
of existing treatments

1096
00:51:12,730 --> 00:51:15,280
for multiple myeloma.

1097
00:51:15,280 --> 00:51:19,440
And we don't really know which
treatments work best for whom.

1098
00:51:19,440 --> 00:51:21,473
But imagine a day where
we have algorithms

1099
00:51:21,473 --> 00:51:23,640
that could take what you
know about a patient at one

1100
00:51:23,640 --> 00:51:24,840
point in time.

1101
00:51:24,840 --> 00:51:27,510
That might include, for
example, blood test results.

1102
00:51:27,510 --> 00:51:30,000
It might include RNA
seq, which gives you

1103
00:51:30,000 --> 00:51:32,250
some sense of the
gene expression

1104
00:51:32,250 --> 00:51:33,990
for the patient,
that in this case

1105
00:51:33,990 --> 00:51:37,260
would be derived from a sample
taken from the patient's bone

1106
00:51:37,260 --> 00:51:39,040
marrow.

1107
00:51:39,040 --> 00:51:40,980
You could take that
data and try to predict

1108
00:51:40,980 --> 00:51:44,790
what would happen to a patient
under two different scenarios.

1109
00:51:44,790 --> 00:51:46,770
The blue scenario that
I'm showing you here,

1110
00:51:46,770 --> 00:51:50,580
if you give them treatment
A, or this red scenario here,

1111
00:51:50,580 --> 00:51:53,090
where you give them treatment
B. And of course, treatment A

1112
00:51:53,090 --> 00:51:55,090
and treatment B aren't
just one-time treatments,

1113
00:51:55,090 --> 00:51:56,320
but they're strategies.

1114
00:51:56,320 --> 00:51:58,530
So they're repeated
treatments across time,

1115
00:51:58,530 --> 00:52:01,260
with some intervals.

1116
00:52:01,260 --> 00:52:05,910
And if your algorithm says
that under treatment B,

1117
00:52:05,910 --> 00:52:08,420
this is what's going to
happen, then you might--

1118
00:52:08,420 --> 00:52:10,340
the clinician might think, OK.

1119
00:52:10,340 --> 00:52:12,160
Treatment B is probably
the way to go here.

1120
00:52:12,160 --> 00:52:15,570
It's going to long-term
control the patient's disease

1121
00:52:15,570 --> 00:52:17,640
burden the best.

1122
00:52:17,640 --> 00:52:20,040
And this is an example
of a causal question.

1123
00:52:20,040 --> 00:52:22,380
Because we want
to know how do we

1124
00:52:22,380 --> 00:52:27,060
cause a change in the
patient's disease trajectory.

1125
00:52:27,060 --> 00:52:29,020
And we can try to answer
this now using data.

1126
00:52:29,020 --> 00:52:32,160
So for example, one of the data
sets that's available for you

1127
00:52:32,160 --> 00:52:34,710
to use in your course projects
is from the Multiple Myeloma

1128
00:52:34,710 --> 00:52:36,480
Research Foundation.

1129
00:52:36,480 --> 00:52:38,220
It's an example of
a disease registry,

1130
00:52:38,220 --> 00:52:41,040
just like the disease registry
I talked to you about earlier

1131
00:52:41,040 --> 00:52:42,540
for rheumatoid arthritis.

1132
00:52:42,540 --> 00:52:44,220
And it follows
about 1,000 patients

1133
00:52:44,220 --> 00:52:46,980
across time, patients who
have multiple myeloma.

1134
00:52:46,980 --> 00:52:49,860
What treatments they're
getting, what their symptoms

1135
00:52:49,860 --> 00:52:52,890
are, and at a couple
of different stages,

1136
00:52:52,890 --> 00:52:56,390
very detailed biological
data about their cancer,

1137
00:52:56,390 --> 00:52:59,170
in this case, RNA seq.

1138
00:52:59,170 --> 00:53:01,420
And one could attempt to use
that data to learn models

1139
00:53:01,420 --> 00:53:03,340
to make predictions like this.

1140
00:53:03,340 --> 00:53:06,420
But such predictions
are fraught with errors.

1141
00:53:06,420 --> 00:53:08,170
And one of the things
that Pete and I will

1142
00:53:08,170 --> 00:53:10,360
be teaching in this
course is that there's

1143
00:53:10,360 --> 00:53:13,090
a very big difference between
prediction and prediction

1144
00:53:13,090 --> 00:53:15,880
for the purpose of
making causal statements.

1145
00:53:15,880 --> 00:53:18,160
And the way that you interpret
the data that you have,

1146
00:53:18,160 --> 00:53:20,147
when your goal is to
do treatment suggestion

1147
00:53:20,147 --> 00:53:22,480
or optimization, is going to
be very different from what

1148
00:53:22,480 --> 00:53:24,313
you were taught in your
introductory machine

1149
00:53:24,313 --> 00:53:27,190
learning algorithms class.

1150
00:53:27,190 --> 00:53:29,690
So other ways that we could try
to treat and manage patients

1151
00:53:29,690 --> 00:53:32,240
with chronic disease
include early diagnosis.

1152
00:53:32,240 --> 00:53:35,178
For example, patients
with Alzheimer's disease,

1153
00:53:35,178 --> 00:53:37,220
there's been some really
interesting results just

1154
00:53:37,220 --> 00:53:39,380
in the last few years, here.

1155
00:53:39,380 --> 00:53:41,100
Or new modalities altogether.

1156
00:53:41,100 --> 00:53:42,800
For example, liquid
biopsies that

1157
00:53:42,800 --> 00:53:47,000
are able to do early
diagnosis of cancer,

1158
00:53:47,000 --> 00:53:51,980
even without having to do a
biopsy of the cancer tumor

1159
00:53:51,980 --> 00:53:54,040
itself.

1160
00:53:54,040 --> 00:53:58,220
We can also think about how
do we better track and measure

1161
00:53:58,220 --> 00:54:00,320
chronic disease.

1162
00:54:00,320 --> 00:54:03,250
So one example shown
on the left here

1163
00:54:03,250 --> 00:54:07,180
is from Dina Katabi's lab
here at MIT and CSAIL,

1164
00:54:07,180 --> 00:54:09,880
where they've developed a
system called Emerald, which

1165
00:54:09,880 --> 00:54:12,730
is using wireless signals,
the same wireless signals

1166
00:54:12,730 --> 00:54:17,645
that we have in this room
today, to try to track patients.

1167
00:54:17,645 --> 00:54:19,270
And they can actually
see behind walls,

1168
00:54:19,270 --> 00:54:20,990
which is quite impressive.

1169
00:54:20,990 --> 00:54:23,530
So using this for
the signal, you

1170
00:54:23,530 --> 00:54:26,200
could install what looks
like just a regular wireless

1171
00:54:26,200 --> 00:54:29,320
router in an elderly
person's home,

1172
00:54:29,320 --> 00:54:33,280
and you could detect if
that elderly patient falls.

1173
00:54:33,280 --> 00:54:36,133
And of course if the patient
has fallen, and they're elderly,

1174
00:54:36,133 --> 00:54:38,050
it might be very hard
for them to get back up.

1175
00:54:38,050 --> 00:54:40,180
They might have broken
a hip, for example.

1176
00:54:40,180 --> 00:54:43,070
And one could then alert the
caregivers, maybe if necessary,

1177
00:54:43,070 --> 00:54:44,440
bring in emergency support.

1178
00:54:44,440 --> 00:54:48,110
And that could have a long-term
outcome for this patient which

1179
00:54:48,110 --> 00:54:50,380
would really help them.

1180
00:54:50,380 --> 00:54:54,340
So this is an example of what
I mean by better tracking

1181
00:54:54,340 --> 00:54:55,940
patients with chronic disease.

1182
00:54:55,940 --> 00:54:57,580
Another example
comes from patients

1183
00:54:57,580 --> 00:55:00,160
who have type 1 diabetes.

1184
00:55:00,160 --> 00:55:03,190
Type 1 diabetes, as
opposed to type 2 diabetes,

1185
00:55:03,190 --> 00:55:07,810
generally develops in
patients at a very early age.

1186
00:55:07,810 --> 00:55:09,880
Usually as children
it's diagnosed.

1187
00:55:09,880 --> 00:55:14,920
And one is typically managed by
having an insulin pump, which

1188
00:55:14,920 --> 00:55:22,030
is attached to a patient and
can give injections of insulin

1189
00:55:22,030 --> 00:55:24,010
on the fly, as necessary.

1190
00:55:24,010 --> 00:55:27,460
But there's a really challenging
control problem there.

1191
00:55:27,460 --> 00:55:30,445
If you give a patient too much
insulin, you could kill them.

1192
00:55:30,445 --> 00:55:33,790
If you give them
too little insulin,

1193
00:55:33,790 --> 00:55:35,223
you could really hurt them.

1194
00:55:35,223 --> 00:55:36,640
And how much insulin
you give them

1195
00:55:36,640 --> 00:55:38,220
is going to be a function
of their activity.

1196
00:55:38,220 --> 00:55:40,512
It's going to be a function
of what food they're eating

1197
00:55:40,512 --> 00:55:42,920
and various other factors.

1198
00:55:42,920 --> 00:55:46,237
So this is a question which the
control theory community has

1199
00:55:46,237 --> 00:55:48,070
been thinking through
for a number of years,

1200
00:55:48,070 --> 00:55:50,680
and there are a number of
sophisticated algorithms

1201
00:55:50,680 --> 00:55:52,680
that are present in
today's products,

1202
00:55:52,680 --> 00:55:55,180
and I wouldn't be surprised if
one or two people in the room

1203
00:55:55,180 --> 00:55:57,202
today have one of these.

1204
00:55:57,202 --> 00:55:59,410
But it also presents a really
interesting opportunity

1205
00:55:59,410 --> 00:56:00,490
for machine learning.

1206
00:56:00,490 --> 00:56:02,800
Because right now, we're
not doing a very good job

1207
00:56:02,800 --> 00:56:05,350
at predicting future
glucose levels, which

1208
00:56:05,350 --> 00:56:08,140
is essential to figure out
how to regulate insulin.

1209
00:56:08,140 --> 00:56:10,450
And if we had algorithms
that could, for example, take

1210
00:56:10,450 --> 00:56:13,480
a patient's phone, take
a picture of the food

1211
00:56:13,480 --> 00:56:16,900
that a patient is eating,
have that automatically

1212
00:56:16,900 --> 00:56:19,390
feed into an algorithm that
predicts its caloric content

1213
00:56:19,390 --> 00:56:22,270
and how quickly that'll
be processed by the body.

1214
00:56:22,270 --> 00:56:24,190
And then as a result
of that, think about

1215
00:56:24,190 --> 00:56:28,810
when, based on this patient's
metabolic system, when should

1216
00:56:28,810 --> 00:56:31,600
you start increasing insulin
levels and by how much.

1217
00:56:31,600 --> 00:56:33,600
That could have a huge
impact in quality of life

1218
00:56:33,600 --> 00:56:36,610
of these types of patients.

1219
00:56:36,610 --> 00:56:38,290
So finally, we've
talked a lot about how

1220
00:56:38,290 --> 00:56:41,290
do we manage healthcare,
but equally important

1221
00:56:41,290 --> 00:56:43,520
is about discovery.

1222
00:56:43,520 --> 00:56:45,040
So the same data
that we could use

1223
00:56:45,040 --> 00:56:48,040
to try to change the way that
algorithms are implemented

1224
00:56:48,040 --> 00:56:51,730
could be used to think through
what would be new treatments

1225
00:56:51,730 --> 00:56:55,160
and make new discoveries
about disease subtypes.

1226
00:56:55,160 --> 00:56:56,770
So at one point later
in the semester,

1227
00:56:56,770 --> 00:56:59,050
we'll be talking about
disease progression modeling,

1228
00:56:59,050 --> 00:57:01,450
and we'll talk about how to
use data-driven approaches

1229
00:57:01,450 --> 00:57:06,017
to discover different
subtypes of disease.

1230
00:57:06,017 --> 00:57:07,600
And on the left,
here, I'm showing you

1231
00:57:07,600 --> 00:57:11,280
an example of a really nice
study from back in 2008

1232
00:57:11,280 --> 00:57:13,510
that used a k-means
clustering algorithm

1233
00:57:13,510 --> 00:57:16,552
to discover subtypes of asthma.

1234
00:57:16,552 --> 00:57:18,010
One could also use
machine learning

1235
00:57:18,010 --> 00:57:26,120
to try to make discoveries about
what proteins, for example,

1236
00:57:26,120 --> 00:57:29,830
are important in
regulating disease.

1237
00:57:29,830 --> 00:57:33,250
How can we differentiate at a
biological level which patients

1238
00:57:33,250 --> 00:57:35,350
will progress quickly,
which patients

1239
00:57:35,350 --> 00:57:37,360
will respond to treatment.

1240
00:57:37,360 --> 00:57:41,080
And that of course will
then suggest new ways of--

1241
00:57:41,080 --> 00:57:46,670
new drug targets for new
pharmaceutical efforts.

1242
00:57:46,670 --> 00:57:49,240
Another direction also
studied here at MIT,

1243
00:57:49,240 --> 00:57:52,900
by quite a few labs, actually,
has to do with drug creation

1244
00:57:52,900 --> 00:57:54,280
or discovery.

1245
00:57:54,280 --> 00:57:56,230
So one could use machine
learning algorithms

1246
00:57:56,230 --> 00:58:01,117
to try to predict what
would a good antibody be

1247
00:58:01,117 --> 00:58:02,950
for trying to bind with
a particular target.

1248
00:58:06,170 --> 00:58:09,940
So that's all for my overview.

1249
00:58:09,940 --> 00:58:12,100
And in the remaining
20 minutes, I'm

1250
00:58:12,100 --> 00:58:13,930
going to tell you a
little bit about what's

1251
00:58:13,930 --> 00:58:15,970
unique about machine
learning in healthcare,

1252
00:58:15,970 --> 00:58:18,610
and then an overview
of the class syllabus.

1253
00:58:18,610 --> 00:58:22,180
And I do see that it says,
replace lamp in six minutes,

1254
00:58:22,180 --> 00:58:24,290
or power will turn off
and go into standby mode.

1255
00:58:24,290 --> 00:58:26,036
AUDIENCE: We have
that one [INAUDIBLE]..

1256
00:58:26,036 --> 00:58:27,500
DAVID SONTAG: Ah, OK.

1257
00:58:27,500 --> 00:58:29,530
Good.

1258
00:58:29,530 --> 00:58:31,210
You're hired.

1259
00:58:31,210 --> 00:58:34,990
If you didn't get into the
class, talk to me afterwards.

1260
00:58:34,990 --> 00:58:35,490
All right.

1261
00:58:35,490 --> 00:58:36,933
AUDIENCE: [INAUDIBLE].

1262
00:58:36,933 --> 00:58:39,825
DAVID SONTAG: [LAUGHS] We hope.

1263
00:58:39,825 --> 00:58:41,950
So what's unique about
machine learning healthcare?

1264
00:58:41,950 --> 00:58:45,160
I gave you already
some hints at this.

1265
00:58:45,160 --> 00:58:50,410
So first, healthcare is
ultimately, unfortunately,

1266
00:58:50,410 --> 00:58:53,440
about life or death decisions.

1267
00:58:53,440 --> 00:58:59,550
So we need robust algorithms
that don't screw up.

1268
00:58:59,550 --> 00:59:02,220
A prime example of this, which
I'll tell you a little bit more

1269
00:59:02,220 --> 00:59:06,930
about towards the
end of the semester

1270
00:59:06,930 --> 00:59:12,210
is from a major software
error that occurred something

1271
00:59:12,210 --> 00:59:17,430
like 20, 30 years ago in a--

1272
00:59:17,430 --> 00:59:21,030
in an X-ray type
of device, where

1273
00:59:21,030 --> 00:59:23,790
an overwhelming
amount of radiation

1274
00:59:23,790 --> 00:59:27,390
was exposed to a patient just
because of a software overflow

1275
00:59:27,390 --> 00:59:29,370
problem, a bug.

1276
00:59:29,370 --> 00:59:33,890
And of course that resulted
in a number of patients dying.

1277
00:59:33,890 --> 00:59:39,020
So that was a software
error from decades ago,

1278
00:59:39,020 --> 00:59:41,120
where there was no machine
learning in the loop.

1279
00:59:41,120 --> 00:59:46,200
And as a result of that and
similar types of disasters,

1280
00:59:46,200 --> 00:59:50,030
including in the space industry
and airplanes and so on,

1281
00:59:50,030 --> 00:59:54,050
led to a whole area of
research in computer science

1282
00:59:54,050 --> 00:59:57,840
in formal methods and how do we
design computer algorithms that

1283
00:59:57,840 --> 01:00:01,440
can check that a
piece of software

1284
01:00:01,440 --> 01:00:03,690
would do what it's supposed
to do and would not make--

1285
01:00:03,690 --> 01:00:05,940
and that there
are no bugs in it.

1286
01:00:05,940 --> 01:00:08,610
But now that we're going to
start to bring data and machine

1287
01:00:08,610 --> 01:00:11,290
learning algorithms
into the picture,

1288
01:00:11,290 --> 01:00:15,630
we are really suffering
for lack of good tools

1289
01:00:15,630 --> 01:00:18,510
for doing similar formal
checking of our algorithms

1290
01:00:18,510 --> 01:00:20,240
and their behavior.

1291
01:00:20,240 --> 01:00:22,260
And so this is going
to be really important

1292
01:00:22,260 --> 01:00:26,220
in the future decade, as
machine learning gets deployed

1293
01:00:26,220 --> 01:00:27,970
not just in settings
like healthcare,

1294
01:00:27,970 --> 01:00:30,430
but also in other settings
of life and death,

1295
01:00:30,430 --> 01:00:32,690
such as in autonomous driving.

1296
01:00:32,690 --> 01:00:34,770
And it's something
that we'll touch on

1297
01:00:34,770 --> 01:00:36,558
throughout the semester.

1298
01:00:36,558 --> 01:00:39,100
So for example, when one deploys
machine learning algorithms,

1299
01:00:39,100 --> 01:00:41,910
we need to be thinking about
are they safe, but also

1300
01:00:41,910 --> 01:00:44,310
how do we check for
safety long-term?

1301
01:00:44,310 --> 01:00:45,810
What are checks and
balances that we

1302
01:00:45,810 --> 01:00:48,540
should put into the deployment
of the algorithm to make

1303
01:00:48,540 --> 01:00:52,200
sure that it's still
working as it was intended?

1304
01:00:52,200 --> 01:00:54,270
We also need fair and
accountable algorithms.

1305
01:00:54,270 --> 01:00:57,120
Because increasingly,
machine learning results

1306
01:00:57,120 --> 01:00:59,430
are being used to
drive resources

1307
01:00:59,430 --> 01:01:02,190
in a healthcare setting.

1308
01:01:02,190 --> 01:01:04,740
An example that I'll discuss
in about a week and a half,

1309
01:01:04,740 --> 01:01:06,840
when we talk about
risk stratification,

1310
01:01:06,840 --> 01:01:10,560
is that algorithms are
being used by payers

1311
01:01:10,560 --> 01:01:12,420
to risk stratify patients.

1312
01:01:12,420 --> 01:01:14,310
For example, to figure
out which patients

1313
01:01:14,310 --> 01:01:17,610
are likely to be readmitted
to the hospital in the next 30

1314
01:01:17,610 --> 01:01:20,340
days, or are likely
to have undiagnosed

1315
01:01:20,340 --> 01:01:22,440
diabetes, or are likely
to progress quickly

1316
01:01:22,440 --> 01:01:23,545
in their diabetes.

1317
01:01:23,545 --> 01:01:25,170
And based on those
predictions, they're

1318
01:01:25,170 --> 01:01:26,503
doing a number of interventions.

1319
01:01:26,503 --> 01:01:29,820
For example, they might send
nurses to the patient's home.

1320
01:01:29,820 --> 01:01:35,790
They might offer
their members access

1321
01:01:35,790 --> 01:01:39,368
to a weight loss program.

1322
01:01:39,368 --> 01:01:41,910
And each of these interventions
has money associated to them.

1323
01:01:41,910 --> 01:01:42,630
They have a cost.

1324
01:01:42,630 --> 01:01:44,288
And so you can't do
them for everyone.

1325
01:01:44,288 --> 01:01:46,080
And so one uses machine
learning algorithms

1326
01:01:46,080 --> 01:01:50,070
to prioritize who do you
give those interventions to.

1327
01:01:50,070 --> 01:01:53,460
But because health
is so intimately tied

1328
01:01:53,460 --> 01:01:57,570
to socioeconomic
status, one can think

1329
01:01:57,570 --> 01:02:00,660
about what happens if these
algorithms are not fair.

1330
01:02:00,660 --> 01:02:04,098
It could have really long-term
implications for our society,

1331
01:02:04,098 --> 01:02:06,390
and it's something that we're
going to talk about later

1332
01:02:06,390 --> 01:02:09,460
in the semester as well.

1333
01:02:09,460 --> 01:02:10,920
Now, I mentioned
earlier that many

1334
01:02:10,920 --> 01:02:14,820
of the questions that we
need to study in the field

1335
01:02:14,820 --> 01:02:18,190
don't have good label data.

1336
01:02:18,190 --> 01:02:19,940
In cases where we know
we want to predict,

1337
01:02:19,940 --> 01:02:21,750
there's a supervised
prediction problem,

1338
01:02:21,750 --> 01:02:23,667
often we just don't have
labels for that thing

1339
01:02:23,667 --> 01:02:24,762
we want to predict.

1340
01:02:24,762 --> 01:02:26,220
But also, in many
situations, we're

1341
01:02:26,220 --> 01:02:28,053
not interested in just
predicting something.

1342
01:02:28,053 --> 01:02:29,410
We're interested in discovery.

1343
01:02:29,410 --> 01:02:31,770
So for example, when I talk
about disease subtyping

1344
01:02:31,770 --> 01:02:34,560
or disease progression,
it's much harder

1345
01:02:34,560 --> 01:02:36,415
to quantify what
you're looking for.

1346
01:02:36,415 --> 01:02:38,040
And so unsupervised
learning algorithms

1347
01:02:38,040 --> 01:02:40,030
are going to be really
important for what we do.

1348
01:02:40,030 --> 01:02:42,447
And finally, I already mentioned
how many of the questions

1349
01:02:42,447 --> 01:02:44,440
we want to answer
are causal in nature,

1350
01:02:44,440 --> 01:02:45,898
particularly when
you want to think

1351
01:02:45,898 --> 01:02:47,430
about treatment strategies.

1352
01:02:47,430 --> 01:02:51,110
And so we'll have two
lectures on causal inference,

1353
01:02:51,110 --> 01:02:53,610
and we'll have two lectures on
reinforcement learning, which

1354
01:02:53,610 --> 01:02:55,890
is increasingly being
used to learn treatment

1355
01:02:55,890 --> 01:02:57,511
policies in healthcare.

1356
01:03:02,810 --> 01:03:06,830
So all of these
different problems

1357
01:03:06,830 --> 01:03:11,150
that we've talked about result
in our having to rethink how

1358
01:03:11,150 --> 01:03:14,550
do we do machine
learning in this setting.

1359
01:03:14,550 --> 01:03:19,050
For example, because
driving labels

1360
01:03:19,050 --> 01:03:23,890
for supervised
prediction is very hard,

1361
01:03:23,890 --> 01:03:27,422
one has to think through how
could we automatically build

1362
01:03:27,422 --> 01:03:29,630
algorithms to do what's
called electronic phenotyping

1363
01:03:29,630 --> 01:03:32,148
to discover, to figure
out automatically,

1364
01:03:32,148 --> 01:03:34,190
what is the relevant labels
for a set of patients

1365
01:03:34,190 --> 01:03:37,830
that one could then attempt
to predict in the future.

1366
01:03:37,830 --> 01:03:40,080
Because we often have
very little data,

1367
01:03:40,080 --> 01:03:42,810
for example, some
rare diseases, there

1368
01:03:42,810 --> 01:03:44,880
might only be a few
hundred or a few thousand

1369
01:03:44,880 --> 01:03:48,000
people in the nation
that have that disease.

1370
01:03:48,000 --> 01:03:51,270
Some common diseases
present in very diverse ways

1371
01:03:51,270 --> 01:03:54,390
and [INAUDIBLE] are very rare.

1372
01:03:54,390 --> 01:03:57,435
Because of that, you have just a
small number of patient samples

1373
01:03:57,435 --> 01:03:59,060
that you could get,
even if you had all

1374
01:03:59,060 --> 01:04:00,850
of the data in the right place.

1375
01:04:00,850 --> 01:04:03,990
And so we need to think through
how can we bring through--

1376
01:04:03,990 --> 01:04:08,010
how can we bring together
domain knowledge.

1377
01:04:08,010 --> 01:04:10,740
How can we bring together
data from other areas--

1378
01:04:10,740 --> 01:04:14,280
will everyone look
over here now--

1379
01:04:14,280 --> 01:04:15,978
from other areas,
other diseases,

1380
01:04:15,978 --> 01:04:17,520
in order to learn
something that then

1381
01:04:17,520 --> 01:04:20,640
we could refine for
the foreground question

1382
01:04:20,640 --> 01:04:21,150
of interest.

1383
01:04:23,890 --> 01:04:27,950
Finally, there is a ton of
missing data in healthcare.

1384
01:04:27,950 --> 01:04:32,880
So raise your hand
if you've only

1385
01:04:32,880 --> 01:04:35,310
been seeing your current
primary care physician

1386
01:04:35,310 --> 01:04:36,450
for less than four years.

1387
01:04:40,370 --> 01:04:40,870
OK.

1388
01:04:40,870 --> 01:04:42,828
Now, this was an easy
guess, because all of you

1389
01:04:42,828 --> 01:04:46,170
are students, and you
probably don't live in Boston.

1390
01:04:46,170 --> 01:04:51,420
But here in the US, even
after you graduate, you go out

1391
01:04:51,420 --> 01:04:54,345
into the world, you
have a job, and that job

1392
01:04:54,345 --> 01:04:55,470
pays your health insurance.

1393
01:04:55,470 --> 01:04:56,110
And you know what?

1394
01:04:56,110 --> 01:04:57,430
Most of you are going to
go into the tech industry,

1395
01:04:57,430 --> 01:04:59,620
and most of you are going to
switch jobs every four years.

1396
01:04:59,620 --> 01:05:00,670
And so your health
insurance is going

1397
01:05:00,670 --> 01:05:01,690
to change every four years.

1398
01:05:01,690 --> 01:05:03,190
And unfortunately,
data doesn't tend

1399
01:05:03,190 --> 01:05:07,180
to follow people when you
change providers or payers.

1400
01:05:07,180 --> 01:05:09,478
And so what that
means is for any one

1401
01:05:09,478 --> 01:05:11,020
thing we might want
to study, we tend

1402
01:05:11,020 --> 01:05:13,150
to not have very good
longitudinal data

1403
01:05:13,150 --> 01:05:15,880
on those individuals, at least
not here in the United States.

1404
01:05:15,880 --> 01:05:18,088
That story is a little bit
different in other places,

1405
01:05:18,088 --> 01:05:21,060
like the UK or
Israel, for example.

1406
01:05:21,060 --> 01:05:26,560
Moreover, we also
have a very bad lens

1407
01:05:26,560 --> 01:05:28,060
on that healthcare data.

1408
01:05:28,060 --> 01:05:31,390
So even if you've been going
to the same doctor for a while,

1409
01:05:31,390 --> 01:05:34,835
we tend to only have data on you
when something's been recorded.

1410
01:05:34,835 --> 01:05:37,210
So if you went to a doctor,
you had a lab test performed,

1411
01:05:37,210 --> 01:05:38,620
we know the results of it.

1412
01:05:38,620 --> 01:05:40,810
If you've never gotten
your glucose tested,

1413
01:05:40,810 --> 01:05:43,420
it's very hard,
though not impossible,

1414
01:05:43,420 --> 01:05:46,330
to figure out if you
might be diabetic.

1415
01:05:46,330 --> 01:05:48,837
So thinking about how
do we deal with the fact

1416
01:05:48,837 --> 01:05:50,670
that there's a large
amount of missing data,

1417
01:05:50,670 --> 01:05:53,830
where that missing data
has very different patterns

1418
01:05:53,830 --> 01:05:56,560
across patients, and
where there might

1419
01:05:56,560 --> 01:06:00,340
be a big difference between
train and test distributions

1420
01:06:00,340 --> 01:06:04,030
is going to be a major part of
what we discuss in this course.

1421
01:06:04,030 --> 01:06:06,160
And finally, the last
example is censoring.

1422
01:06:06,160 --> 01:06:08,310
I think I've said
finally a few times.

1423
01:06:08,310 --> 01:06:11,680
So censoring, which we'll
talk about in two weeks,

1424
01:06:11,680 --> 01:06:14,140
is what happens
when you have data

1425
01:06:14,140 --> 01:06:15,800
only for small windows of time.

1426
01:06:15,800 --> 01:06:19,520
So for example, you have a
data set where your goal is

1427
01:06:19,520 --> 01:06:21,640
to predict survival.

1428
01:06:21,640 --> 01:06:24,130
You want to know how
long until a person dies.

1429
01:06:24,130 --> 01:06:27,700
But a person-- you
only have data on them

1430
01:06:27,700 --> 01:06:30,190
up to January 2009,
and they haven't yet

1431
01:06:30,190 --> 01:06:31,485
died by January 2009.

1432
01:06:31,485 --> 01:06:32,860
Then that individual
is censored.

1433
01:06:32,860 --> 01:06:34,527
You don't know what
would have happened,

1434
01:06:34,527 --> 01:06:36,322
you don't know when they died.

1435
01:06:36,322 --> 01:06:38,780
So that doesn't mean you should
throw away that data point.

1436
01:06:38,780 --> 01:06:40,197
In fact, we'll
talk about learning

1437
01:06:40,197 --> 01:06:45,540
algorithms that can learn from
censored data very effectively.

1438
01:06:45,540 --> 01:06:48,740
So there are a number of also
logistical challenges to doing

1439
01:06:48,740 --> 01:06:50,450
machine learning in healthcare.

1440
01:06:50,450 --> 01:06:53,130
I talked about how having
access to data is so important,

1441
01:06:53,130 --> 01:06:54,950
but one of the reasons--
there are others--

1442
01:06:54,950 --> 01:06:58,070
for why getting large amounts
of data in the public domain

1443
01:06:58,070 --> 01:07:00,230
is challenging is because
it's so sensitive.

1444
01:07:00,230 --> 01:07:04,250
And removing identifiers,
like name and social,

1445
01:07:04,250 --> 01:07:06,980
from data which
includes free text notes

1446
01:07:06,980 --> 01:07:09,230
can be very challenging.

1447
01:07:09,230 --> 01:07:12,260
And as a result, when we
do research here at MIT,

1448
01:07:12,260 --> 01:07:15,710
typically, it takes us
anywhere from a few months--

1449
01:07:15,710 --> 01:07:18,140
which has never happened--
to two years, which

1450
01:07:18,140 --> 01:07:21,140
is the usual situation, to
negotiate a data sharing

1451
01:07:21,140 --> 01:07:24,860
agreement to get the health
data to MIT to do research on.

1452
01:07:24,860 --> 01:07:27,260
And of course then
my students write

1453
01:07:27,260 --> 01:07:30,140
code, which we're very happy to
open source under MIT license,

1454
01:07:30,140 --> 01:07:31,640
but that code is
completely useless,

1455
01:07:31,640 --> 01:07:34,268
because no one can reproduce
their results on the same data

1456
01:07:34,268 --> 01:07:35,810
because they don't
have access to it.

1457
01:07:35,810 --> 01:07:39,682
So that's a major
challenge to this field.

1458
01:07:39,682 --> 01:07:41,390
Another challenge is
about the difficulty

1459
01:07:41,390 --> 01:07:46,250
in deploying machine learning
algorithms due to the challenge

1460
01:07:46,250 --> 01:07:47,610
of integration.

1461
01:07:47,610 --> 01:07:48,860
So you build a good algorithm.

1462
01:07:48,860 --> 01:07:52,370
You want to deploy it at
your favorite hospital,

1463
01:07:52,370 --> 01:07:53,330
but guess what?

1464
01:07:53,330 --> 01:07:58,460
That hospital has Epic
or Cerner or Athena

1465
01:07:58,460 --> 01:08:00,700
or some other commercial
electronic medical records

1466
01:08:00,700 --> 01:08:03,500
system, and that electronic
medical records system

1467
01:08:03,500 --> 01:08:07,070
is not built for your
algorithm to plug into.

1468
01:08:07,070 --> 01:08:11,120
So there is a big
gap, a large amount

1469
01:08:11,120 --> 01:08:15,050
of difficulty to getting your
algorithms into production

1470
01:08:15,050 --> 01:08:19,520
systems, which we'll talk about
as well during the semester.

1471
01:08:22,970 --> 01:08:27,649
So the goals that Pete and I
have for you are as follows.

1472
01:08:27,649 --> 01:08:30,640
We want you to get intuition for
working with healthcare data.

1473
01:08:30,640 --> 01:08:33,130
And so the next two
lectures after today

1474
01:08:33,130 --> 01:08:38,420
are going to focus on what
healthcare is really like,

1475
01:08:38,420 --> 01:08:40,819
and what is the
healthcare data that's

1476
01:08:40,819 --> 01:08:45,229
created by the practice
of healthcare like.

1477
01:08:45,229 --> 01:08:46,910
We want you to get
intuition for how

1478
01:08:46,910 --> 01:08:49,850
to formalize machine learning
challenges as healthcare

1479
01:08:49,850 --> 01:08:51,279
problems.

1480
01:08:51,279 --> 01:08:53,567
And that formalization step
is often the most tricky

1481
01:08:53,567 --> 01:08:55,609
and something you'll spend
a lot of time thinking

1482
01:08:55,609 --> 01:08:58,620
through as part of
your problem sets.

1483
01:08:58,620 --> 01:09:01,470
Not all machine learning
algorithms are equally useful.

1484
01:09:01,470 --> 01:09:03,479
And so one theme
that I'll return to

1485
01:09:03,479 --> 01:09:06,720
throughout the semester
is that despite the fact

1486
01:09:06,720 --> 01:09:09,660
that deep learning is good
for many speech recognition

1487
01:09:09,660 --> 01:09:12,720
and computer vision problems,
it actually isn't the best match

1488
01:09:12,720 --> 01:09:14,293
to many problems in healthcare.

1489
01:09:14,293 --> 01:09:16,710
And you'll explore that also
as part of your problem sets,

1490
01:09:16,710 --> 01:09:18,910
or at least one of them.

1491
01:09:18,910 --> 01:09:21,750
And we want you to understand
also the subtleties in robustly

1492
01:09:21,750 --> 01:09:24,520
and safely deploying
machine learning algorithms.

1493
01:09:24,520 --> 01:09:27,590
Now, more broadly,
this is a young field.

1494
01:09:27,590 --> 01:09:32,680
So for example, just recently,
just about three years ago,

1495
01:09:32,680 --> 01:09:35,830
was created the first
conference on Machine Learning

1496
01:09:35,830 --> 01:09:38,689
in Healthcare, by that name.

1497
01:09:38,689 --> 01:09:43,210
And new publication venues are
being created every single day

1498
01:09:43,210 --> 01:09:48,670
by Nature, Lancet, and also
machine learning journals,

1499
01:09:48,670 --> 01:09:51,808
for publishing research on
machine learning healthcare.

1500
01:09:51,808 --> 01:09:53,850
Because it's one of those
issues we talked about,

1501
01:09:53,850 --> 01:09:59,610
like access to data, not
very good benchmarks,

1502
01:09:59,610 --> 01:10:01,515
reproducibility has
been a major challenge.

1503
01:10:01,515 --> 01:10:04,140
And this is again something that
the field is only now starting

1504
01:10:04,140 --> 01:10:05,930
to really grapple with.

1505
01:10:05,930 --> 01:10:08,400
And so as part of this
course, oh so many of you

1506
01:10:08,400 --> 01:10:11,285
are currently PhD students
or will soon be PhD students,

1507
01:10:11,285 --> 01:10:12,660
we're going to
think through what

1508
01:10:12,660 --> 01:10:16,163
are some of the challenges
for the research field.

1509
01:10:16,163 --> 01:10:17,580
What are some of
the open problems

1510
01:10:17,580 --> 01:10:19,830
that you might want to work
on, either during your PhD

1511
01:10:19,830 --> 01:10:21,770
or during your future career.