1
00:00:00,240 --> 00:00:02,399
Hello, everyone. This is Jacob Emerson with the

2
00:00:02,399 --> 00:00:04,179
Becker's Health IT podcast.

3
00:00:04,639 --> 00:00:07,200
Thrilled today to be joined by doctor Nigam

4
00:00:07,200 --> 00:00:09,679
Shah, who is the chief data scientist at

5
00:00:09,679 --> 00:00:12,000
Stanford Health Care. Doctor Shah, thank you so

6
00:00:12,000 --> 00:00:13,759
much for taking the time to be with

7
00:00:13,759 --> 00:00:15,059
me on the podcast today.

8
00:00:15,414 --> 00:00:16,234
Oh, great to

9
00:00:16,614 --> 00:00:18,214
be here. Yeah. And before we dive into

10
00:00:18,214 --> 00:00:20,074
everything we wanna talk with you about, Nigam,

11
00:00:20,214 --> 00:00:21,734
can you tell us a little bit more

12
00:00:21,734 --> 00:00:24,375
about yourself, your background in health care, and

13
00:00:24,375 --> 00:00:25,814
what it is that you do today at

14
00:00:25,814 --> 00:00:26,314
Stanford?

15
00:00:27,350 --> 00:00:29,670
Sure. Sure. So I often joke I've never

16
00:00:29,670 --> 00:00:32,070
had a real job. I got my PhD

17
00:00:32,070 --> 00:00:34,310
from Penn State, came to Stanford as a

18
00:00:34,310 --> 00:00:35,829
postdoc in 02/2005,

19
00:00:35,829 --> 00:00:36,890
and never left.

20
00:00:37,750 --> 00:00:40,344
Been on the faculty since 02/2011,

21
00:00:41,284 --> 00:00:42,664
tenured in 2015.

22
00:00:43,445 --> 00:00:44,804
And then I was doing the usual, you

23
00:00:44,804 --> 00:00:47,604
know, writing papers, real world evidence, machine learning

24
00:00:47,604 --> 00:00:50,564
on EHRs and so on. And, got sucked

25
00:00:50,564 --> 00:00:52,804
into working on things during the pandemic in

26
00:00:52,804 --> 00:00:53,704
the health system

27
00:00:54,179 --> 00:00:56,659
and found it very rewarding to do things

28
00:00:56,659 --> 00:00:59,560
in an operational environment. So in 2022,

29
00:01:00,420 --> 00:01:02,340
we set up this new role of chief

30
00:01:02,340 --> 00:01:04,739
data scientist for the health care system where

31
00:01:04,739 --> 00:01:07,239
the job basically is to translate things

32
00:01:07,585 --> 00:01:10,564
into practice such that patient care directly benefits.

33
00:01:13,105 --> 00:01:15,585
Fantastic. So, Nigam, we're here to talk with

34
00:01:15,585 --> 00:01:18,885
you today about an announcement that Stanford HealthCare

35
00:01:18,944 --> 00:01:21,520
made earlier this year saying that the system

36
00:01:21,520 --> 00:01:23,680
was going to be launching an Ambient AI

37
00:01:23,680 --> 00:01:26,959
pilot with Atropos Health to integrate real world

38
00:01:26,959 --> 00:01:29,140
evidence within the electronic health record

39
00:01:29,439 --> 00:01:31,060
via Ambient AI listening,

40
00:01:31,599 --> 00:01:34,795
from from Microsoft decks. So talk to us

41
00:01:34,795 --> 00:01:35,915
a little bit about this.

42
00:01:36,715 --> 00:01:38,814
How is this gonna be working, this pilot,

43
00:01:39,114 --> 00:01:40,635
and and how is this gonna help your

44
00:01:40,635 --> 00:01:41,135
physicians

45
00:01:41,594 --> 00:01:44,094
access real world evidence to make better decisions

46
00:01:44,394 --> 00:01:45,295
for your patients?

47
00:01:45,950 --> 00:01:46,450
Mhmm.

48
00:01:46,909 --> 00:01:48,750
So real world evidence has been part of

49
00:01:48,750 --> 00:01:49,409
my work

50
00:01:50,270 --> 00:01:51,870
almost since 2011,

51
00:01:51,870 --> 00:01:52,689
2012.

52
00:01:53,549 --> 00:01:55,790
And as you probably know, I mean, most

53
00:01:55,790 --> 00:01:57,549
of the times, real world evidence is an

54
00:01:57,549 --> 00:02:00,430
observational study that takes anywhere from six to

55
00:02:00,430 --> 00:02:02,894
nine months or maybe one year. It gets

56
00:02:02,894 --> 00:02:04,734
published in a journal like JAMA or New

57
00:02:04,734 --> 00:02:06,914
England Journal or Lancet or what have you.

58
00:02:06,974 --> 00:02:08,974
And maybe a physician reads it or doesn't

59
00:02:08,974 --> 00:02:11,314
read it. And the chance that it affects

60
00:02:11,455 --> 00:02:12,834
patient care is

61
00:02:13,610 --> 00:02:14,269
very low.

62
00:02:15,129 --> 00:02:17,930
There's some very shocking numbers out there saying

63
00:02:17,930 --> 00:02:20,569
where physicians tracked the number of decisions they

64
00:02:20,569 --> 00:02:23,150
make, they go back and ask how often

65
00:02:23,530 --> 00:02:25,229
was some evidence available

66
00:02:26,055 --> 00:02:28,775
particularly relevant to that case at hand. And

67
00:02:28,775 --> 00:02:30,775
often the answer is, you know, under five

68
00:02:30,775 --> 00:02:32,955
percent, under three percent, things like that.

69
00:02:33,335 --> 00:02:34,794
So that's what I've been researching.

70
00:02:35,974 --> 00:02:38,215
And we did a project here called the

71
00:02:38,215 --> 00:02:40,074
green button project that

72
00:02:40,860 --> 00:02:43,340
took that nine month time frame down and

73
00:02:43,340 --> 00:02:45,659
got it to under two days. So at

74
00:02:45,659 --> 00:02:48,000
the bedside, a physician can make a request

75
00:02:48,540 --> 00:02:50,800
and a team produces a written report,

76
00:02:51,180 --> 00:02:52,080
in two days.

77
00:02:52,794 --> 00:02:53,995
Once that was done,

78
00:02:54,314 --> 00:02:56,094
we spun out Atropos Health,

79
00:02:56,395 --> 00:02:58,555
which took those forty eight hours, turned around,

80
00:02:58,555 --> 00:03:00,395
and got it down to under three to

81
00:03:00,395 --> 00:03:01,135
four hours.

82
00:03:02,555 --> 00:03:05,260
So at the bedside, you have a question,

83
00:03:05,479 --> 00:03:07,879
you ask something, what happened to similar patients,

84
00:03:07,879 --> 00:03:09,960
and you get a written report before the

85
00:03:09,960 --> 00:03:10,780
end of the day.

86
00:03:11,639 --> 00:03:13,740
So then Ambient AI comes around,

87
00:03:15,240 --> 00:03:16,860
and number of physicians

88
00:03:17,159 --> 00:03:19,895
are super excited about how it's now saving

89
00:03:19,895 --> 00:03:22,055
them time and freeing up their sort of

90
00:03:22,055 --> 00:03:22,555
cognitive

91
00:03:22,935 --> 00:03:25,435
bandwidth to think more about the patient.

92
00:03:25,895 --> 00:03:26,875
On the other hand,

93
00:03:27,735 --> 00:03:30,615
Atropos adoption is going fine. About a thousand

94
00:03:30,615 --> 00:03:32,074
or 1,200 consultations

95
00:03:32,375 --> 00:03:35,430
are requested and reports being produced within within

96
00:03:35,430 --> 00:03:36,090
a day.

97
00:03:36,789 --> 00:03:37,689
Now separately,

98
00:03:38,229 --> 00:03:39,530
not at Stanford,

99
00:03:39,990 --> 00:03:42,870
Atropos had been developing this solution, which they

100
00:03:42,870 --> 00:03:44,009
call CHAT RWD,

101
00:03:44,789 --> 00:03:47,129
which essentially automates this

102
00:03:47,865 --> 00:03:49,245
procedure of producing

103
00:03:49,544 --> 00:03:51,324
observational studies on demand,

104
00:03:51,784 --> 00:03:54,284
takes the human mostly out of the loop,

105
00:03:54,504 --> 00:03:55,965
and can get an answer,

106
00:03:56,425 --> 00:03:58,264
for a lot of cases, in under five

107
00:03:58,264 --> 00:03:58,764
minutes.

108
00:03:59,669 --> 00:04:01,509
So now you look at these two things

109
00:04:01,509 --> 00:04:02,810
together and you ask, alright.

110
00:04:03,590 --> 00:04:06,150
We have a patient visit that's fifteen minutes

111
00:04:06,150 --> 00:04:08,550
now getting transcribed, and in about a minute

112
00:04:08,550 --> 00:04:10,009
or two, you get the full transcript.

113
00:04:10,550 --> 00:04:12,250
We have this separate ability

114
00:04:12,784 --> 00:04:14,485
that Stanford has already purchased

115
00:04:14,944 --> 00:04:17,024
where the physician, instead of a day, can

116
00:04:17,024 --> 00:04:18,725
get an answer in five minutes.

117
00:04:19,264 --> 00:04:21,024
What would happen if we put the two

118
00:04:21,024 --> 00:04:21,524
together?

119
00:04:22,064 --> 00:04:23,904
So that led to the current project you're

120
00:04:23,904 --> 00:04:24,644
talking about.

121
00:04:25,050 --> 00:04:28,089
Wow. That's that's some incredible numbers you decided

122
00:04:28,089 --> 00:04:29,050
there in terms of,

123
00:04:29,930 --> 00:04:32,490
five minutes and getting an answer back. That's

124
00:04:32,490 --> 00:04:32,990
incredible.

125
00:04:34,410 --> 00:04:36,750
I I wonder, have you heard from clinicians

126
00:04:37,370 --> 00:04:40,175
across the Stanford Enterprise in terms of how

127
00:04:40,175 --> 00:04:42,495
this is being received, what their thoughts are?

128
00:04:42,495 --> 00:04:44,514
Is this, being received enthusiastically?

129
00:04:45,615 --> 00:04:46,835
So lots of enthusiasm

130
00:04:47,535 --> 00:04:50,095
to pilot it and be, you know, the

131
00:04:50,095 --> 00:04:52,435
first to get access to it, 100%.

132
00:04:53,970 --> 00:04:56,949
The Atropos service, the one that produces this

133
00:04:57,009 --> 00:04:57,509
report

134
00:04:57,810 --> 00:05:00,389
has been live on our campus since almost

135
00:05:00,610 --> 00:05:01,810
2022

136
00:05:01,810 --> 00:05:03,410
actually or 2021.

137
00:05:03,410 --> 00:05:03,910
Okay.

138
00:05:05,889 --> 00:05:08,264
And that typically is used in the inpatient

139
00:05:08,264 --> 00:05:10,264
setting because there you can tolerate a one

140
00:05:10,264 --> 00:05:12,125
day or a four hour turnaround. Right?

141
00:05:13,464 --> 00:05:16,044
Ambient scribes are used in the outpatient setting.

142
00:05:16,264 --> 00:05:17,944
And so this will be the first time

143
00:05:17,944 --> 00:05:19,944
that real world evidence will be offered in

144
00:05:19,944 --> 00:05:21,964
the outpatient setting on demand

145
00:05:22,360 --> 00:05:24,360
at the point of care. So everybody's super

146
00:05:24,360 --> 00:05:25,420
excited about it.

147
00:05:26,520 --> 00:05:28,520
Wow. No. It's incredible. And is it my

148
00:05:28,520 --> 00:05:30,600
understanding that this is going to be continued

149
00:05:30,600 --> 00:05:32,840
to be rolled out this year? Is that

150
00:05:32,840 --> 00:05:34,860
the general timeline for this pilot?

151
00:05:35,384 --> 00:05:38,504
Absa absolutely. Definitely before the 2025

152
00:05:38,504 --> 00:05:41,164
ends. You know, most likely, the first actual

153
00:05:41,785 --> 00:05:43,625
user will get it in the next sixty

154
00:05:43,625 --> 00:05:46,185
days or ninety days. Wow. And and this

155
00:05:46,185 --> 00:05:48,284
is all your physicians across Stanford?

156
00:05:48,970 --> 00:05:51,050
No. We'll be actually piloting this with what

157
00:05:51,050 --> 00:05:54,990
we call our, community practice providers or UMP,

158
00:05:55,290 --> 00:05:58,810
university medical partners, or sometimes called Stanford medical

159
00:05:58,810 --> 00:06:02,269
partners. Sure. So these are physicians who are,

160
00:06:02,814 --> 00:06:05,375
like, not at the university. They're in the

161
00:06:05,375 --> 00:06:08,254
community practices all over the Bay Area. And

162
00:06:08,254 --> 00:06:10,415
they're the ones who are super excited that

163
00:06:10,415 --> 00:06:10,915
finally,

164
00:06:11,454 --> 00:06:13,615
technology is at a at a stage at

165
00:06:13,615 --> 00:06:16,254
ease of use that they can incorporate into

166
00:06:16,254 --> 00:06:18,870
their care. Wow. That's amazing. And and I

167
00:06:18,870 --> 00:06:20,629
know we did briefly just touch on this,

168
00:06:20,629 --> 00:06:22,810
but if from the patient's perspective,

169
00:06:23,270 --> 00:06:24,810
how would how would this,

170
00:06:25,189 --> 00:06:26,870
on the ground? What does this look like

171
00:06:26,870 --> 00:06:29,350
to me? I come into to the office

172
00:06:29,350 --> 00:06:31,430
and I receive those answers within five minutes,

173
00:06:31,430 --> 00:06:33,064
or can you can you detail that for

174
00:06:33,064 --> 00:06:35,545
us a little bit? Mhmm. So imagine before

175
00:06:35,545 --> 00:06:37,144
all of this technology, you go to a

176
00:06:37,144 --> 00:06:37,644
physician.

177
00:06:38,264 --> 00:06:40,824
You're speaking. They're looking at a computer screen,

178
00:06:40,824 --> 00:06:42,985
typing away furiously so as to not to

179
00:06:42,985 --> 00:06:43,725
miss anything.

180
00:06:44,060 --> 00:06:46,240
They maybe glance at you once or twice.

181
00:06:46,460 --> 00:06:48,460
And then, you know, once their note is

182
00:06:48,460 --> 00:06:50,139
done, they look at you, tell you three

183
00:06:50,139 --> 00:06:51,660
things you should be doing, and off you

184
00:06:51,660 --> 00:06:52,400
go. Right?

185
00:06:53,900 --> 00:06:56,779
Ambient scribes come around, and now the doctor's

186
00:06:56,779 --> 00:06:58,240
actually looking at you.

187
00:06:58,875 --> 00:06:59,375
And

188
00:06:59,834 --> 00:07:02,495
there's this thing in the ether that's transcribing,

189
00:07:03,115 --> 00:07:06,014
and by the time the the, the conversation

190
00:07:06,314 --> 00:07:08,314
ends in about a minute, a note shows

191
00:07:08,314 --> 00:07:10,154
up, which they can turn the screen around

192
00:07:10,154 --> 00:07:12,154
and and show it to the patient directly.

193
00:07:12,154 --> 00:07:15,339
In fact, our CMIO, doctor Topher Sharp, has

194
00:07:15,339 --> 00:07:18,300
done even video interviews with reporters showing them,

195
00:07:18,300 --> 00:07:20,399
you know, how powerful this is

196
00:07:20,779 --> 00:07:21,759
in terms of

197
00:07:22,220 --> 00:07:24,139
the physician being able to pay attention to

198
00:07:24,139 --> 00:07:27,105
you. Right? So in normal encounter, everything is

199
00:07:27,585 --> 00:07:29,685
fine. This automation describing

200
00:07:29,985 --> 00:07:31,665
sort of does the job. But then if

201
00:07:31,665 --> 00:07:32,645
there's a question,

202
00:07:33,025 --> 00:07:34,725
what does a physician do today?

203
00:07:35,745 --> 00:07:38,305
They will go to UpToDate or some other

204
00:07:38,305 --> 00:07:39,444
literature summary.

205
00:07:39,904 --> 00:07:40,404
Or

206
00:07:41,139 --> 00:07:43,139
if they had a question that can't be

207
00:07:43,139 --> 00:07:46,020
answered by literature, they will trigger this real

208
00:07:46,020 --> 00:07:48,279
world evidence report from from Atropos.

209
00:07:48,980 --> 00:07:51,240
But that's a manual step, both of them.

210
00:07:51,779 --> 00:07:53,699
So now you can imagine a situation where

211
00:07:53,699 --> 00:07:56,040
we fully have the transcript and it's electronic.

212
00:07:57,134 --> 00:07:59,954
Can I distill a question out of it,

213
00:08:00,974 --> 00:08:02,595
send it to literature evidence,

214
00:08:03,055 --> 00:08:05,474
send it to Atropos for data driven evidence,

215
00:08:06,014 --> 00:08:08,495
produce both of them back if something exists

216
00:08:08,495 --> 00:08:10,779
and is relevant, and put that as two

217
00:08:10,779 --> 00:08:12,699
links in the note that is being shown

218
00:08:12,699 --> 00:08:14,699
to the physician? And then they can choose

219
00:08:14,699 --> 00:08:16,220
to say, alright. Yeah. This looks good enough.

220
00:08:16,220 --> 00:08:17,660
I should click on it and read it

221
00:08:17,660 --> 00:08:19,279
and talk to the patient about it.

222
00:08:20,699 --> 00:08:23,264
Understood. No. This is incredible. And I it

223
00:08:23,264 --> 00:08:24,785
it really reflects what we're hearing from a

224
00:08:24,785 --> 00:08:26,725
lot of other health systems around the country,

225
00:08:27,264 --> 00:08:30,064
piloting and introducing this type of technology as

226
00:08:30,064 --> 00:08:30,564
well.

227
00:08:31,345 --> 00:08:31,845
So

228
00:08:32,625 --> 00:08:34,945
in that vein, Nigam, what what would you

229
00:08:34,945 --> 00:08:36,485
say have been some of the key challenges

230
00:08:36,544 --> 00:08:37,284
that systems,

231
00:08:37,860 --> 00:08:40,100
maybe Stanford alone, but from other systems as

232
00:08:40,100 --> 00:08:41,240
well that you're hearing,

233
00:08:41,700 --> 00:08:44,360
challenges they're facing when trying to integrate ambient

234
00:08:44,500 --> 00:08:46,759
AI tech into their existing workflows?

235
00:08:47,379 --> 00:08:49,539
Yeah. So there's sort of two broad buckets

236
00:08:49,539 --> 00:08:50,200
of challenges.

237
00:08:50,740 --> 00:08:53,024
The first one are sort of the the

238
00:08:53,024 --> 00:08:55,745
the seemingly daunting ones where or the easy

239
00:08:55,745 --> 00:08:58,225
ones, I would call them technical challenges. You

240
00:08:58,225 --> 00:09:00,544
know, what exact API are you gonna call?

241
00:09:00,544 --> 00:09:01,445
What's your latency?

242
00:09:01,985 --> 00:09:03,825
How fast can you get the answer? How

243
00:09:03,825 --> 00:09:07,024
will you maintain security, privacy, compliance? You know,

244
00:09:07,024 --> 00:09:07,845
all of that,

245
00:09:08,870 --> 00:09:11,269
excuse the word, junk that goes in with

246
00:09:11,269 --> 00:09:13,990
health IT in general in in in doing

247
00:09:13,990 --> 00:09:15,750
things with duct tape and bubble gum and

248
00:09:15,750 --> 00:09:16,570
string. Right?

249
00:09:17,750 --> 00:09:18,490
I'm sure.

250
00:09:19,350 --> 00:09:21,769
And then the second bucket are the challenges

251
00:09:21,830 --> 00:09:24,195
we haven't yet thought of, which is how

252
00:09:24,195 --> 00:09:26,595
are we gonna evaluate these things? Because, you

253
00:09:26,595 --> 00:09:28,595
know, given a Google query, for example, we

254
00:09:28,595 --> 00:09:30,834
always get an answer. Given an up to

255
00:09:30,834 --> 00:09:32,375
date search, you get an answer.

256
00:09:32,995 --> 00:09:35,235
Given an RW real world data search, you'll

257
00:09:35,235 --> 00:09:36,995
get an answer. And there's some safeguards built

258
00:09:36,995 --> 00:09:39,000
in that if the cord's too small or

259
00:09:39,000 --> 00:09:41,559
the phenotypes are not clearly defined, the system

260
00:09:41,559 --> 00:09:42,940
will refrain from answering.

261
00:09:43,639 --> 00:09:45,580
But when you put all this together,

262
00:09:46,840 --> 00:09:47,820
how do you evaluate?

263
00:09:48,919 --> 00:09:51,259
How do you know it's adding value?

264
00:09:51,815 --> 00:09:54,294
Those are two different kinds of evaluations. Those

265
00:09:54,294 --> 00:09:56,054
are the things I think in the next

266
00:09:56,054 --> 00:09:59,495
year, everybody is gonna be, thinking about. Yeah.

267
00:09:59,495 --> 00:10:00,934
No. Absolutely. It's a great point. And I

268
00:10:01,014 --> 00:10:02,615
that leads me to my next question for

269
00:10:02,615 --> 00:10:05,115
you is, how do you think that Ambient

270
00:10:05,174 --> 00:10:07,899
AI tech within the hospital, how's that gonna

271
00:10:07,899 --> 00:10:10,460
continue to evolve over the next few years?

272
00:10:10,460 --> 00:10:12,700
And and what's the broader impact that you

273
00:10:12,700 --> 00:10:15,040
foresee this having on on the industry?

274
00:10:16,059 --> 00:10:18,779
So, I mean, again, I'm speculating just like

275
00:10:18,779 --> 00:10:20,720
I'm guessing like everybody else. Sure.

276
00:10:21,644 --> 00:10:22,144
Ambient

277
00:10:22,524 --> 00:10:23,904
AI entered the scene

278
00:10:24,605 --> 00:10:25,665
as an automation,

279
00:10:27,004 --> 00:10:30,045
as in something that a human was doing

280
00:10:30,045 --> 00:10:31,504
typing in a suboptimal

281
00:10:31,805 --> 00:10:33,805
manner, or you might have had a human

282
00:10:33,805 --> 00:10:34,625
Skype present.

283
00:10:34,925 --> 00:10:36,305
And that thing is now

284
00:10:36,899 --> 00:10:37,399
automated,

285
00:10:38,339 --> 00:10:38,839
which

286
00:10:39,379 --> 00:10:42,500
immediate return and ROI is the physician now

287
00:10:42,500 --> 00:10:44,740
has the attention to spend on the patient.

288
00:10:44,740 --> 00:10:45,879
So that's great check.

289
00:10:46,659 --> 00:10:49,379
But now we're collecting a new kind of

290
00:10:49,379 --> 00:10:49,879
data

291
00:10:50,514 --> 00:10:53,414
in electronic form that we never had before,

292
00:10:54,514 --> 00:10:55,975
the real time conversation

293
00:10:56,434 --> 00:10:57,654
of what is happening.

294
00:10:58,674 --> 00:11:00,774
Now many things can come from that.

295
00:11:01,475 --> 00:11:02,375
This kind of

296
00:11:02,700 --> 00:11:06,059
evidence generation on demand is sort of near

297
00:11:06,059 --> 00:11:07,279
and dear to my heart,

298
00:11:07,740 --> 00:11:09,519
but you can imagine other things

299
00:11:11,019 --> 00:11:14,240
like documentation for billing could be completely automated.

300
00:11:15,605 --> 00:11:17,605
This whole idea that a biller and a

301
00:11:17,605 --> 00:11:19,924
coder has to now review everything and then

302
00:11:19,924 --> 00:11:22,324
send a query back to the physician saying,

303
00:11:22,324 --> 00:11:24,485
hey, doc. Do you remember? Was this the

304
00:11:24,485 --> 00:11:26,565
case with the patient or not? That whole

305
00:11:26,565 --> 00:11:27,625
thing can go away.

306
00:11:29,839 --> 00:11:31,459
We might be able to produce

307
00:11:32,480 --> 00:11:32,980
summaries

308
00:11:33,360 --> 00:11:35,839
that are directed to the patient. So the

309
00:11:35,839 --> 00:11:38,799
same note can be pushed through another kind

310
00:11:38,799 --> 00:11:39,459
of summarizer,

311
00:11:40,000 --> 00:11:43,375
AI that translate it to whatever language you

312
00:11:43,375 --> 00:11:46,575
need and outlines the three things the the

313
00:11:46,575 --> 00:11:48,595
patient needs to watch out for. So

314
00:11:48,975 --> 00:11:51,454
I think it's gonna enable a whole new

315
00:11:51,454 --> 00:11:53,695
class of care activities that we did not

316
00:11:53,695 --> 00:11:54,914
think were even possible.

317
00:11:56,539 --> 00:11:58,220
Absolutely. So it sounds like I mean, you're

318
00:11:58,220 --> 00:12:00,779
predicting not even just, an improvement in patient

319
00:12:00,779 --> 00:12:03,919
and and quality outcomes, but really massive repercussions

320
00:12:03,980 --> 00:12:07,259
for the compliance and reimbursement perspective within the

321
00:12:07,259 --> 00:12:09,980
health systems as well. So it's absolutely fascinating

322
00:12:09,980 --> 00:12:10,879
to think about.

323
00:12:11,345 --> 00:12:13,665
And even patient care. So, like, imagine if,

324
00:12:13,665 --> 00:12:14,945
you know, you went to you went to

325
00:12:14,945 --> 00:12:15,524
a physician

326
00:12:15,985 --> 00:12:18,085
and they said, you know, watch out for

327
00:12:18,144 --> 00:12:18,884
these symptoms

328
00:12:20,705 --> 00:12:21,764
once every week.

329
00:12:22,545 --> 00:12:25,445
If we recorded that conversation, we can trigger

330
00:12:25,750 --> 00:12:28,149
an agent that checks in with you by

331
00:12:28,149 --> 00:12:31,110
text, phone, or email or calling you saying,

332
00:12:31,110 --> 00:12:33,110
hey, Jacob. Are you having this in the

333
00:12:33,110 --> 00:12:34,170
past seven days?

334
00:12:35,670 --> 00:12:37,370
We're never gonna do that manually.

335
00:12:37,910 --> 00:12:38,410
Sure.

336
00:12:39,855 --> 00:12:41,315
No. It's it's really exciting.

337
00:12:41,695 --> 00:12:42,595
And and certainly,

338
00:12:43,054 --> 00:12:44,894
I know here in in the Chicago area,

339
00:12:44,894 --> 00:12:47,855
we've we've been seeing health systems, personally as

340
00:12:47,855 --> 00:12:49,634
well, rolling out this technology.

341
00:12:50,654 --> 00:12:52,975
And and it's, very clearly from the physician

342
00:12:52,975 --> 00:12:53,475
perspective

343
00:12:53,970 --> 00:12:56,230
already and in the data as well, reducing

344
00:12:56,289 --> 00:12:58,850
that administrative burnout that that so many of

345
00:12:58,850 --> 00:12:59,509
them face.

346
00:13:00,370 --> 00:13:02,230
So really exciting things on the horizon

347
00:13:02,690 --> 00:13:04,870
at Stanford as well in this atmosphere.

348
00:13:06,049 --> 00:13:08,414
But before we go, Nigam, what else are

349
00:13:08,414 --> 00:13:09,934
we missing? You know, you've got the ears

350
00:13:09,934 --> 00:13:12,335
of a lot of other CIOs and and

351
00:13:12,335 --> 00:13:14,975
health IT executives in health systems from around

352
00:13:14,975 --> 00:13:16,735
the country. What else do you wanna share

353
00:13:16,735 --> 00:13:18,815
with them about this this new pilot with

354
00:13:18,815 --> 00:13:22,095
Atropos or or about, this technology that we've

355
00:13:22,095 --> 00:13:23,394
discussed in general?

356
00:13:24,149 --> 00:13:26,309
So I would like to sort of plant

357
00:13:26,309 --> 00:13:28,629
this idea that one of our faculty members

358
00:13:28,629 --> 00:13:29,690
here, Eric Brynjolfsson,

359
00:13:30,549 --> 00:13:32,329
calls the Turing trap.

360
00:13:32,870 --> 00:13:34,950
I'm sure every all the listeners, the everyone's

361
00:13:34,950 --> 00:13:36,970
part of the Turing test where a computer

362
00:13:37,429 --> 00:13:39,209
passes off as being a human.

363
00:13:39,725 --> 00:13:41,884
The tutoring trap is the opposite of that,

364
00:13:41,884 --> 00:13:43,345
where we as humans

365
00:13:44,125 --> 00:13:46,764
only imagine doing with the computer what we

366
00:13:46,764 --> 00:13:48,225
already know how to do.

367
00:13:49,165 --> 00:13:51,085
So that's why we think about compliance and

368
00:13:51,085 --> 00:13:51,585
billing.

369
00:13:52,889 --> 00:13:55,230
But we don't follow-up with our patients

370
00:13:56,009 --> 00:13:57,929
routinely as much as we do. We just

371
00:13:57,929 --> 00:13:58,990
don't have the time.

372
00:13:59,529 --> 00:14:01,710
But what if technology enables that?

373
00:14:02,090 --> 00:14:03,769
And so the way I would encourage people

374
00:14:03,769 --> 00:14:05,725
to think of it is that, you know,

375
00:14:05,945 --> 00:14:07,384
AI writ large, or I like to think

376
00:14:07,384 --> 00:14:09,325
of it more as a data plus algorithm

377
00:14:09,785 --> 00:14:10,285
configuration,

378
00:14:11,384 --> 00:14:13,625
it started out doing simple things, you know,

379
00:14:13,625 --> 00:14:15,965
giving alerts and nudges and classifications

380
00:14:16,345 --> 00:14:17,004
and predictions.

381
00:14:17,850 --> 00:14:20,110
And then it matured into giving recommendations,

382
00:14:21,690 --> 00:14:25,069
giving observational studies and bedside evidence.

383
00:14:25,850 --> 00:14:26,909
Now it's getting

384
00:14:27,529 --> 00:14:31,690
further advanced into actively listening and summarizing and

385
00:14:31,690 --> 00:14:32,190
translating.

386
00:14:33,715 --> 00:14:36,514
But what's next? What other actions can it

387
00:14:36,514 --> 00:14:38,915
take? Can it do follow-up? Can it check-in

388
00:14:38,915 --> 00:14:41,154
on your meds? Can it check-in on your,

389
00:14:41,394 --> 00:14:43,735
glucose levels? Can the conversation

390
00:14:44,274 --> 00:14:47,315
trigger an agent that checks in on your

391
00:14:47,315 --> 00:14:50,250
CGM monitor and its values automatically.

392
00:14:51,350 --> 00:14:52,710
So I think what we need to be

393
00:14:52,710 --> 00:14:54,730
thinking about is what are the actions

394
00:14:55,509 --> 00:14:57,990
that we should be taking but are not

395
00:14:57,990 --> 00:14:58,730
yet taking

396
00:14:59,269 --> 00:15:02,389
that can be done for cheap and fast

397
00:15:02,389 --> 00:15:03,129
with AI.

398
00:15:05,164 --> 00:15:05,664
Fantastic.

399
00:15:06,204 --> 00:15:09,644
Great parting words for our listeners. So, doctor

400
00:15:09,644 --> 00:15:11,725
Shah, I wanna thank you for taking the

401
00:15:11,725 --> 00:15:13,644
time to sit down with us and for

402
00:15:13,644 --> 00:15:15,725
sharing about all the impactful work going on

403
00:15:15,725 --> 00:15:18,365
under your leadership at Stanford HealthCare. We really

404
00:15:18,365 --> 00:15:20,569
appreciate it. Thank you for having me.

405
00:15:20,870 --> 00:15:22,549
And to our listeners, if you'd like to

406
00:15:22,549 --> 00:15:25,029
listen to more podcasts from Becker's Health Care,

407
00:15:25,029 --> 00:15:27,929
you can visit beckershospitalreview.com.