1
00:00:00,719 --> 00:00:02,879
This is Laura Dirdle with the Becker's Healthcare

2
00:00:02,879 --> 00:00:05,120
podcast. I'm thrilled today to be joined by

3
00:00:05,120 --> 00:00:08,080
doctor Chris Kelly, associate chief medical information officer

4
00:00:08,080 --> 00:00:10,580
for data and analytics at MultiCare Health System.

5
00:00:10,719 --> 00:00:12,160
Doctor Kelly, it's a pleasure to have you

6
00:00:12,160 --> 00:00:13,299
on the podcast today.

7
00:00:14,804 --> 00:00:16,905
Hey. Thanks, Laura. I appreciate being on.

8
00:00:17,285 --> 00:00:18,725
Now I'm excited to speak with you a

9
00:00:18,725 --> 00:00:20,505
little bit more about just all the transformation

10
00:00:20,725 --> 00:00:22,824
happening in the health care space and particularly

11
00:00:23,125 --> 00:00:24,725
some of the cool things you're doing with

12
00:00:24,725 --> 00:00:27,204
artificial intelligence at MultiCare. But before we dive

13
00:00:27,204 --> 00:00:28,890
into that discussion, can you tell me a

14
00:00:28,890 --> 00:00:30,829
little bit more about yourself and your background?

15
00:00:31,850 --> 00:00:34,729
Yeah. Sure. So I am currently the, ACM

16
00:00:34,729 --> 00:00:36,829
IO for data analytics at MultiCare.

17
00:00:37,450 --> 00:00:39,850
We're a 13 hospital health care system in,

18
00:00:40,090 --> 00:00:41,149
Washington state.

19
00:00:41,755 --> 00:00:44,094
I I got into this through informatics

20
00:00:44,475 --> 00:00:47,354
about, 10, 12 years ago, when my hospital

21
00:00:47,354 --> 00:00:50,075
went on EPIC. And I really didn't have

22
00:00:50,075 --> 00:00:51,835
a lot of background in this, but I

23
00:00:51,835 --> 00:00:54,155
found I really enjoyed the work, became a

24
00:00:54,155 --> 00:00:56,094
physician builder. I went on getting

25
00:00:56,670 --> 00:00:58,909
training in medical informatics. And,

26
00:00:59,630 --> 00:01:01,829
ultimately, you know, what I I realized what

27
00:01:02,030 --> 00:01:03,789
so what we really need to do is

28
00:01:03,789 --> 00:01:05,709
is make use of the data that we

29
00:01:05,709 --> 00:01:07,969
spend so much time entering into the EMR.

30
00:01:08,270 --> 00:01:09,790
So I ended up getting a master's degree

31
00:01:09,790 --> 00:01:11,915
in data science, and that led to this

32
00:01:11,915 --> 00:01:14,155
job I have here. And I've been doing

33
00:01:14,155 --> 00:01:15,454
a lot of the work around,

34
00:01:16,715 --> 00:01:19,295
implementing artificial intelligence and evaluating,

35
00:01:20,234 --> 00:01:22,334
the performance of artificial intelligence

36
00:01:22,715 --> 00:01:23,454
at MultiCare.

37
00:01:23,859 --> 00:01:26,019
That's really been a major focus on what

38
00:01:26,019 --> 00:01:27,459
I've been doing over the past year or

39
00:01:27,459 --> 00:01:27,959
so.

40
00:01:29,060 --> 00:01:30,899
Oh, that's amazing to hear. You know, what

41
00:01:30,899 --> 00:01:33,060
a great story in terms of how you've,

42
00:01:33,219 --> 00:01:35,939
just really navigated getting into the informatics space.

43
00:01:35,939 --> 00:01:37,875
I I really appreciate that. Now if you

44
00:01:37,875 --> 00:01:39,155
tell me a little bit about what you're

45
00:01:39,155 --> 00:01:42,275
most excited on, and, you know, what what's

46
00:01:42,275 --> 00:01:43,895
top of mind for you right now?

47
00:01:44,594 --> 00:01:46,115
Well, you know, we've been doing a lot

48
00:01:46,115 --> 00:01:47,495
of stuff. You know, the the

49
00:01:47,954 --> 00:01:50,215
there's a lot of angles on artificial intelligence.

50
00:01:50,275 --> 00:01:52,594
I really think that where AI is gonna

51
00:01:52,594 --> 00:01:53,979
be going in health care, it's gonna be

52
00:01:53,979 --> 00:01:56,219
transformative. I think the next decade is going

53
00:01:56,219 --> 00:01:56,879
to be

54
00:01:57,420 --> 00:01:59,979
as if not more transformative than any decade

55
00:01:59,979 --> 00:02:01,739
in the history of health care. I I

56
00:02:01,739 --> 00:02:03,119
really see a lot of opportunity.

57
00:02:03,659 --> 00:02:05,099
But one of the things that that I

58
00:02:05,099 --> 00:02:07,325
really been, you know, that our organization has

59
00:02:07,325 --> 00:02:09,585
been this has really been trying to do

60
00:02:09,884 --> 00:02:12,925
is find ways to utilize AI to make

61
00:02:12,925 --> 00:02:15,185
the experience better for providers.

62
00:02:15,724 --> 00:02:17,004
So that's a lot of what we've been

63
00:02:17,004 --> 00:02:17,905
doing so far.

64
00:02:18,219 --> 00:02:21,280
We've implemented an ambient note generation platform.

65
00:02:21,659 --> 00:02:22,800
We've implemented

66
00:02:23,580 --> 00:02:24,080
EPYC's

67
00:02:24,780 --> 00:02:28,060
automated response technology for in basket messages. We're

68
00:02:28,060 --> 00:02:29,759
working on EPYC's note summarization

69
00:02:30,060 --> 00:02:30,560
approach.

70
00:02:31,254 --> 00:02:33,175
There's a number of different algorithms that we're

71
00:02:33,175 --> 00:02:35,814
introducing to improve patient care as well, but

72
00:02:35,814 --> 00:02:37,415
those are really the angles that we're looking

73
00:02:37,415 --> 00:02:39,754
at to try to improve how providers,

74
00:02:41,014 --> 00:02:41,974
are able to,

75
00:02:42,455 --> 00:02:44,794
utilize the the EMR more efficiently

76
00:02:45,110 --> 00:02:46,310
and how we can take better care of

77
00:02:46,310 --> 00:02:47,050
our patients.

78
00:02:47,750 --> 00:02:49,270
That makes a lot of sense. And, you

79
00:02:49,270 --> 00:02:51,189
know, really, it is a great way to,

80
00:02:51,430 --> 00:02:53,610
it seems like begin getting into AI,

81
00:02:54,310 --> 00:02:54,810
and

82
00:02:55,110 --> 00:02:57,909
having it actually produce those results that you're

83
00:02:57,909 --> 00:02:59,909
looking for both for the care providers as

84
00:02:59,909 --> 00:03:00,889
well as patients.

85
00:03:01,245 --> 00:03:02,605
You know, when you look at those projects

86
00:03:02,605 --> 00:03:03,745
that you're doing today,

87
00:03:04,125 --> 00:03:06,044
is there anything that you've learned along the

88
00:03:06,044 --> 00:03:07,105
way as you were,

89
00:03:07,965 --> 00:03:10,525
implementing some of these things? What have you,

90
00:03:10,764 --> 00:03:13,724
leaned into more so? Is there anything that,

91
00:03:13,724 --> 00:03:15,665
you know, you've had to pivot away from?

92
00:03:16,830 --> 00:03:18,610
Well, you know, that's a really good question

93
00:03:18,750 --> 00:03:20,610
because I think that,

94
00:03:20,990 --> 00:03:22,510
you know, 2 years ago, it was almost

95
00:03:22,510 --> 00:03:24,050
2 years ago now when,

96
00:03:24,430 --> 00:03:24,930
ChatGPT

97
00:03:25,310 --> 00:03:28,189
came out, everybody was really caught by surprise,

98
00:03:28,189 --> 00:03:29,710
I think. I shouldn't say everybody, but a

99
00:03:29,710 --> 00:03:31,745
lot of folks were caught by surprise at

100
00:03:31,745 --> 00:03:34,064
how incredibly powerful it seemed. And I think

101
00:03:34,064 --> 00:03:34,965
a lot of people

102
00:03:35,585 --> 00:03:37,824
really gravitated towards it and thought this could

103
00:03:37,824 --> 00:03:38,564
be transformative

104
00:03:38,865 --> 00:03:39,365
for

105
00:03:39,905 --> 00:03:41,284
a human intensive,

106
00:03:42,385 --> 00:03:42,885
industry

107
00:03:43,665 --> 00:03:45,849
like health care where we really have to

108
00:03:45,849 --> 00:03:47,209
do a lot of 1 on 1, can

109
00:03:47,209 --> 00:03:47,709
we

110
00:03:48,009 --> 00:03:50,889
improve the quality of our interactions, reduce the,

111
00:03:50,889 --> 00:03:53,789
you know, delays and, you know, just improve

112
00:03:54,169 --> 00:03:56,724
patient satisfaction overall with this? But I think

113
00:03:56,724 --> 00:03:58,485
it's proven to be a little more elusive

114
00:03:58,485 --> 00:04:00,105
than people realized initially.

115
00:04:00,805 --> 00:04:03,365
I think we're past that, wow, isn't this

116
00:04:03,365 --> 00:04:06,004
cool, isn't this magic phase of AI? And

117
00:04:06,004 --> 00:04:07,305
we really have to be

118
00:04:07,605 --> 00:04:10,745
thinking about how we critically evaluate these investments

119
00:04:11,599 --> 00:04:14,019
and understand the ones that are really delivering,

120
00:04:14,560 --> 00:04:15,379
the value

121
00:04:16,079 --> 00:04:18,160
that we hope they're going to deliver. And,

122
00:04:18,160 --> 00:04:19,919
you know, at MultiCare, we're really trying to

123
00:04:19,919 --> 00:04:20,660
take a

124
00:04:21,519 --> 00:04:22,019
multidimensional

125
00:04:23,199 --> 00:04:25,139
approach to value, not just

126
00:04:25,724 --> 00:04:28,464
financial return on investment. Although, obviously,

127
00:04:29,324 --> 00:04:30,925
you know, these tools, you know, tend to

128
00:04:30,925 --> 00:04:31,904
be fairly expensive,

129
00:04:32,284 --> 00:04:33,805
but we also have to think about it

130
00:04:33,805 --> 00:04:35,024
is, you know,

131
00:04:35,564 --> 00:04:37,500
does it improve the quality of patient care?

132
00:04:37,579 --> 00:04:38,480
Can we demonstrate

133
00:04:39,100 --> 00:04:41,199
that it actually improves the quality of patient

134
00:04:41,579 --> 00:04:44,300
care? Does it improve patient experience and does

135
00:04:44,300 --> 00:04:47,439
it improve the physician experience? Does it reduce

136
00:04:47,500 --> 00:04:49,980
the time that doctors actually spend in EPIC?

137
00:04:49,980 --> 00:04:52,415
Does it reduce the time that people spend

138
00:04:52,475 --> 00:04:55,675
outside of scheduled hours continuing to chart? And

139
00:04:55,675 --> 00:04:57,214
I think we really have to be fairly

140
00:04:57,274 --> 00:04:59,675
rigorous about our approach to this in order

141
00:04:59,675 --> 00:05:00,334
to understand

142
00:05:01,115 --> 00:05:04,060
what's really effective and what's not effective. I

143
00:05:04,060 --> 00:05:06,160
think that's going to be our our main

144
00:05:06,459 --> 00:05:08,800
hurdle going forward is trying to understand

145
00:05:09,740 --> 00:05:13,120
where our time and resources are best spent.

146
00:05:14,539 --> 00:05:16,459
That's such an excellent point, you know, and

147
00:05:16,459 --> 00:05:16,959
really,

148
00:05:17,425 --> 00:05:19,345
makes a lot of sense as you're, you

149
00:05:19,345 --> 00:05:21,745
know, trying to be innovative, but then also

150
00:05:21,745 --> 00:05:24,144
understand exactly what success looks like for some

151
00:05:24,144 --> 00:05:25,524
of these types of projects,

152
00:05:25,904 --> 00:05:27,664
you know, in the results that you're looking

153
00:05:27,664 --> 00:05:28,164
for.

154
00:05:28,544 --> 00:05:30,464
When looking ahead, how are you thinking about

155
00:05:30,464 --> 00:05:32,960
growth and in particular with AI,

156
00:05:33,420 --> 00:05:36,060
continuing to evolve and add value to to

157
00:05:36,060 --> 00:05:37,040
multicare overall?

158
00:05:37,980 --> 00:05:40,400
You know, it because that's really the

159
00:05:41,180 --> 00:05:42,699
that's really the key that that I think

160
00:05:42,699 --> 00:05:43,900
you hit on is how do we add

161
00:05:43,900 --> 00:05:44,879
value overall?

162
00:05:45,805 --> 00:05:48,604
And, you know, fortunately, I we have an

163
00:05:48,604 --> 00:05:50,384
analytics governance team,

164
00:05:50,764 --> 00:05:52,224
and and one of my responsibilities

165
00:05:52,845 --> 00:05:54,384
is, you know, with

166
00:05:54,845 --> 00:05:56,544
few other folks to evaluate

167
00:05:57,245 --> 00:05:58,704
the incoming opportunities

168
00:05:59,084 --> 00:05:59,985
for AI.

169
00:06:00,329 --> 00:06:01,790
And, you know, how do we

170
00:06:02,250 --> 00:06:02,750
understand

171
00:06:03,529 --> 00:06:06,410
what they can offer? How do we understand

172
00:06:06,410 --> 00:06:07,629
how we're going to measure,

173
00:06:08,250 --> 00:06:10,729
whether they actually do add value? And, again,

174
00:06:10,729 --> 00:06:12,810
what we need to think about value in

175
00:06:12,810 --> 00:06:13,629
in a multidimensional

176
00:06:14,089 --> 00:06:16,764
way. We're we're we're health care system and,

177
00:06:17,064 --> 00:06:18,745
you know, part of what we do are

178
00:06:18,745 --> 00:06:19,245
good.

179
00:06:19,625 --> 00:06:20,664
You know, part of what we do is

180
00:06:20,664 --> 00:06:22,425
we need to make sure the finances work,

181
00:06:22,425 --> 00:06:24,745
but our ultimate mission is to take better

182
00:06:24,745 --> 00:06:27,144
care of our patients. Well, how do we

183
00:06:27,144 --> 00:06:29,259
measure that? You know, how do we understand

184
00:06:29,259 --> 00:06:31,759
whether these tools which seem so incredible

185
00:06:32,139 --> 00:06:33,199
are actually improving

186
00:06:33,500 --> 00:06:35,040
the patient care that we deliver?

187
00:06:35,419 --> 00:06:37,600
And and that's not always easy to do,

188
00:06:38,379 --> 00:06:40,639
you know, given the limitations that we have.

189
00:06:40,814 --> 00:06:42,814
It's, you know, it's funny. We've been having

190
00:06:42,814 --> 00:06:45,795
some conversations around this recently, and,

191
00:06:46,175 --> 00:06:47,855
you know, I think the conclusion that a

192
00:06:47,855 --> 00:06:50,514
lot of us have reached is that the

193
00:06:50,894 --> 00:06:52,194
the AI algorithm

194
00:06:52,975 --> 00:06:54,449
is actually the easy part.

195
00:06:55,170 --> 00:06:56,290
And I don't mean to say that it's

196
00:06:56,290 --> 00:06:59,430
easy, and these your algorithms are incredibly sophisticated.

197
00:07:00,850 --> 00:07:04,389
But taking that AI algorithm and actually incorporating

198
00:07:04,610 --> 00:07:05,990
it into clinical workflows,

199
00:07:06,654 --> 00:07:09,454
making sure that you've standardized those workflows across

200
00:07:09,454 --> 00:07:10,194
an incredibly

201
00:07:10,654 --> 00:07:12,514
complex health care environment,

202
00:07:12,894 --> 00:07:15,134
and then being able to then go back

203
00:07:15,134 --> 00:07:18,115
and and measure the performance of the model

204
00:07:18,334 --> 00:07:20,754
and the the impact that the model has

205
00:07:21,629 --> 00:07:23,089
on the delivery of care,

206
00:07:23,470 --> 00:07:25,470
that's a real challenge. And,

207
00:07:25,870 --> 00:07:27,470
I think that's really where,

208
00:07:27,790 --> 00:07:29,410
we're going to be seeing some,

209
00:07:30,189 --> 00:07:33,470
real need for investment in coming years. There's

210
00:07:33,470 --> 00:07:35,569
a lot of opportunities. There's literally

211
00:07:36,095 --> 00:07:37,935
dozens of these that that are coming up

212
00:07:37,935 --> 00:07:39,394
every month, but,

213
00:07:39,935 --> 00:07:42,415
they all seem so promising. But we really

214
00:07:42,415 --> 00:07:44,254
need to have a good understanding about what

215
00:07:44,254 --> 00:07:45,555
really is going to work,

216
00:07:47,055 --> 00:07:49,214
showing what isn't. And and I don't think

217
00:07:49,214 --> 00:07:50,975
we're there yet. I think that's really where

218
00:07:50,975 --> 00:07:52,649
we need to put our our energies.

219
00:07:54,149 --> 00:07:55,750
That makes a lot of sense. And, you

220
00:07:55,750 --> 00:07:58,329
know, it'll just be fascinating to see how,

221
00:07:58,949 --> 00:08:00,870
you you're able to parse out those things,

222
00:08:00,870 --> 00:08:02,709
you know, that really do add value in

223
00:08:02,709 --> 00:08:05,654
those that work and are important to continue

224
00:08:05,654 --> 00:08:08,294
investing in, for the future. So, you know,

225
00:08:08,294 --> 00:08:10,055
I I think that just seems really, really

226
00:08:10,055 --> 00:08:11,514
critical and helpful.

227
00:08:12,055 --> 00:08:14,615
Are there any headwinds or roadblocks that you

228
00:08:14,615 --> 00:08:16,615
see potentially coming down the pipe or or

229
00:08:16,615 --> 00:08:18,669
something that you and your leadership team is

230
00:08:18,669 --> 00:08:20,509
just are just keeping an eye on as

231
00:08:20,509 --> 00:08:22,529
a trend that might unfold in the future?

232
00:08:23,550 --> 00:08:26,189
Well, so, you know, this is the the

233
00:08:26,189 --> 00:08:27,730
difficulty that we have,

234
00:08:28,110 --> 00:08:29,569
I think, in health care

235
00:08:30,029 --> 00:08:30,529
is

236
00:08:31,375 --> 00:08:32,195
really understanding

237
00:08:32,975 --> 00:08:34,514
how these models

238
00:08:35,054 --> 00:08:35,554
work,

239
00:08:36,335 --> 00:08:38,894
you know, not on an abstract dataset, but

240
00:08:38,894 --> 00:08:39,394
in

241
00:08:40,095 --> 00:08:40,754
a clinical

242
00:08:41,455 --> 00:08:43,075
or operational context.

243
00:08:43,860 --> 00:08:45,799
You know, there's a lot of different ways,

244
00:08:45,940 --> 00:08:47,559
for example, to document

245
00:08:48,100 --> 00:08:50,179
pieces of information on a patient. You can

246
00:08:50,179 --> 00:08:51,620
put it in a flow sheet. You can

247
00:08:51,620 --> 00:08:54,019
store it as discrete data. You can store

248
00:08:54,019 --> 00:08:54,759
it as

249
00:08:55,220 --> 00:08:57,700
free text. And there may be, you know,

250
00:08:57,700 --> 00:09:00,575
different ways to do it across a complex

251
00:09:00,575 --> 00:09:03,554
health care system where there's multiple different hospitals.

252
00:09:04,174 --> 00:09:05,615
There may be different ways to do it

253
00:09:05,615 --> 00:09:07,794
even within a hospital. And and these

254
00:09:08,335 --> 00:09:10,894
problems or the way that this documentation is

255
00:09:10,894 --> 00:09:13,820
done isn't just arbitrary. There's usually very good

256
00:09:13,820 --> 00:09:15,899
reasons for why it's done this way. But

257
00:09:15,899 --> 00:09:17,840
if we don't have that level of standardization

258
00:09:18,700 --> 00:09:21,580
across a hospital or across the system, it

259
00:09:21,580 --> 00:09:22,800
gets very difficult

260
00:09:23,180 --> 00:09:23,759
to implement

261
00:09:25,804 --> 00:09:27,964
these algorithms in such a way that you

262
00:09:27,964 --> 00:09:29,964
can actually make a difference with them. You

263
00:09:29,964 --> 00:09:32,044
know, the thing that I think it's a

264
00:09:32,044 --> 00:09:33,024
little tricky

265
00:09:33,325 --> 00:09:35,345
from a data standpoint is that,

266
00:09:36,924 --> 00:09:39,325
you know, there's this assumption that we have

267
00:09:39,325 --> 00:09:41,470
this this, you know, clean set of health

268
00:09:41,470 --> 00:09:43,649
care data now that we're on electronic medical

269
00:09:44,029 --> 00:09:45,870
records. But the data that get entered into

270
00:09:45,870 --> 00:09:47,490
the EMR don't get entered

271
00:09:48,029 --> 00:09:50,829
to later build algorithms off of. The data

272
00:09:50,829 --> 00:09:52,370
get entered to take care of patients,

273
00:09:52,745 --> 00:09:54,904
and we're doing it in the way that

274
00:09:54,904 --> 00:09:56,904
we think we can best take care of

275
00:09:56,904 --> 00:09:58,445
the patients given the

276
00:09:58,825 --> 00:10:00,985
unique environment and the unique problems and the

277
00:10:00,985 --> 00:10:03,004
unique workflows that we have.

278
00:10:03,625 --> 00:10:04,125
But

279
00:10:04,889 --> 00:10:06,809
if it's all done in different ways, if

280
00:10:06,809 --> 00:10:08,889
there's not a lot of standardization, it becomes

281
00:10:08,889 --> 00:10:11,050
increasingly difficult to make use of the data,

282
00:10:11,050 --> 00:10:13,210
to normalize the data, to make sure we

283
00:10:13,210 --> 00:10:14,269
have a good understanding

284
00:10:14,730 --> 00:10:15,470
of how,

285
00:10:15,930 --> 00:10:17,370
we're going to be able to make sense

286
00:10:17,370 --> 00:10:20,024
of this going forward. So I think the

287
00:10:20,024 --> 00:10:21,865
first big challenge that we have to do

288
00:10:21,865 --> 00:10:22,365
is,

289
00:10:22,985 --> 00:10:23,485
operationalize

290
00:10:24,264 --> 00:10:26,825
these models in such a way that we

291
00:10:26,825 --> 00:10:27,544
can really,

292
00:10:28,184 --> 00:10:30,024
deliver the impact that we think they're going

293
00:10:30,024 --> 00:10:30,684
to have,

294
00:10:31,065 --> 00:10:32,044
in health care.

295
00:10:32,350 --> 00:10:34,509
As you know, a fascinating point you brought

296
00:10:34,509 --> 00:10:36,769
up there and especially just looking at data,

297
00:10:36,830 --> 00:10:40,370
it seems like a very much colossal challenge

298
00:10:40,429 --> 00:10:42,610
to get that organized in a way that,

299
00:10:43,070 --> 00:10:45,470
makes sense for those algorithms and and for

300
00:10:45,470 --> 00:10:46,610
an AI informed,

301
00:10:47,230 --> 00:10:49,764
healthcare system. And so I I can imagine

302
00:10:49,764 --> 00:10:51,365
that, you know, we'll take a huge mind

303
00:10:51,365 --> 00:10:52,664
share, in the future.

304
00:10:53,125 --> 00:10:55,044
I'm wondering, you know, are there any other

305
00:10:55,044 --> 00:10:57,225
opportunities that you see for growth or development,

306
00:10:57,764 --> 00:10:59,065
as you really move forward?

307
00:10:59,764 --> 00:11:02,084
So, you know, my training has been in

308
00:11:02,084 --> 00:11:03,204
data science and,

309
00:11:04,540 --> 00:11:06,460
you know, addition to to the informatics side

310
00:11:06,460 --> 00:11:09,580
of things. And I really think that what

311
00:11:09,580 --> 00:11:10,720
we're going to

312
00:11:11,179 --> 00:11:13,200
see is an increased demand

313
00:11:14,460 --> 00:11:17,279
for the use of people, not just to

314
00:11:18,054 --> 00:11:20,315
report on things, but to really

315
00:11:20,934 --> 00:11:22,154
make sure we understand

316
00:11:22,855 --> 00:11:23,674
from a,

317
00:11:24,534 --> 00:11:26,475
causal or hypothesis testing

318
00:11:28,054 --> 00:11:30,714
way of, actually trying to understand,

319
00:11:31,254 --> 00:11:32,279
you know, what the

320
00:11:32,680 --> 00:11:35,100
reasons for various outcomes are.

321
00:11:35,800 --> 00:11:36,700
There's some

322
00:11:37,160 --> 00:11:39,800
health care is is so multifaceted and so

323
00:11:39,800 --> 00:11:40,300
complicated,

324
00:11:41,800 --> 00:11:44,540
that it gets really tricky to try to

325
00:11:44,759 --> 00:11:46,620
actually make sense of

326
00:11:47,095 --> 00:11:48,634
how a specific workflow

327
00:11:49,014 --> 00:11:49,514
is,

328
00:11:50,054 --> 00:11:52,294
working well in one environment and not working

329
00:11:52,294 --> 00:11:55,034
well in another environment. The number of different

330
00:11:55,495 --> 00:11:57,575
hypotheses that you can throw out, but you

331
00:11:57,575 --> 00:11:58,855
need to be able to test them, and

332
00:11:58,855 --> 00:12:00,794
you need to be able to have people

333
00:12:01,299 --> 00:12:03,559
on the data side of things who can

334
00:12:03,860 --> 00:12:04,360
abstract,

335
00:12:04,980 --> 00:12:06,980
you know, the data from the electronic medical

336
00:12:06,980 --> 00:12:09,720
record, from the various databases that we have,

337
00:12:10,259 --> 00:12:11,559
and then be able to

338
00:12:12,100 --> 00:12:14,259
use data science techniques to,

339
00:12:14,875 --> 00:12:16,955
do some inference testing to to try to

340
00:12:16,955 --> 00:12:17,455
understand

341
00:12:17,995 --> 00:12:18,495
this

342
00:12:19,035 --> 00:12:21,835
relationship. Is there a correlation here? Can we,

343
00:12:21,835 --> 00:12:23,595
you know, try to figure out some way

344
00:12:23,595 --> 00:12:24,575
to infer causation?

345
00:12:25,595 --> 00:12:28,075
But it's challenging, and I I think there's

346
00:12:28,075 --> 00:12:30,269
a lot of folks who understand data science.

347
00:12:30,990 --> 00:12:32,590
But understanding data science is only part of

348
00:12:32,590 --> 00:12:34,350
the problem. You really need to understand that

349
00:12:34,350 --> 00:12:35,330
health care environment.

350
00:12:35,950 --> 00:12:38,110
And I really think that the next, you

351
00:12:38,110 --> 00:12:40,370
know, few years, we're going to see increasing

352
00:12:40,429 --> 00:12:42,284
need for people who who have this

353
00:12:42,924 --> 00:12:45,245
ability to have one foot in the clinical

354
00:12:45,245 --> 00:12:47,584
world and one foot in the the data

355
00:12:47,725 --> 00:12:50,204
science world to help us move to that

356
00:12:50,204 --> 00:12:52,944
next level of of understanding and and appreciation

357
00:12:53,084 --> 00:12:53,824
for how

358
00:12:54,204 --> 00:12:56,044
nuanced with so much of the work we

359
00:12:56,044 --> 00:12:56,784
do is.

360
00:12:57,690 --> 00:13:00,330
Absolutely. I I think that really, you know,

361
00:13:00,330 --> 00:13:02,990
is a helpful reminder of just the complexity

362
00:13:03,210 --> 00:13:05,529
within the health care system and having as

363
00:13:05,529 --> 00:13:07,290
you do, you know, such a a great

364
00:13:07,290 --> 00:13:07,790
background,

365
00:13:08,250 --> 00:13:11,370
within health care delivery as a physician as

366
00:13:11,370 --> 00:13:13,725
well as, you know, your data science training

367
00:13:13,725 --> 00:13:15,485
and the technology side of it. You can

368
00:13:15,485 --> 00:13:16,704
really have a unique

369
00:13:17,004 --> 00:13:18,845
position there to see where,

370
00:13:19,245 --> 00:13:21,245
there are huge challenges, but also I can

371
00:13:21,245 --> 00:13:23,964
imagine great ways to solve problems and try

372
00:13:23,964 --> 00:13:25,184
to really

373
00:13:25,870 --> 00:13:26,529
get to

374
00:13:26,909 --> 00:13:29,149
a point where you can create that value

375
00:13:29,149 --> 00:13:31,490
in the future as the technology evolves as,

376
00:13:32,309 --> 00:13:34,350
you know, get to a point where you've

377
00:13:34,350 --> 00:13:36,029
got more and more data in a great

378
00:13:36,029 --> 00:13:38,110
space. And then 2, from the health care

379
00:13:38,110 --> 00:13:40,189
side of it, just really can see what

380
00:13:40,189 --> 00:13:41,445
is emerging as most

381
00:13:42,004 --> 00:13:42,504
beneficial,

382
00:13:43,044 --> 00:13:44,424
for patients as well as caregivers.

383
00:13:44,804 --> 00:13:46,804
Of course, that's my bias. Yeah. That's my

384
00:13:46,804 --> 00:13:47,304
bias.

385
00:13:48,804 --> 00:13:50,965
But, of course, that's that's my perspective. I

386
00:13:50,965 --> 00:13:53,445
won't say that's necessarily universal, but I I

387
00:13:53,764 --> 00:13:55,044
you know, that's the way I think we

388
00:13:55,044 --> 00:13:56,904
can really approach these types of problems.

389
00:13:58,379 --> 00:14:00,559
Absolutely. Well, I think that's just very helpful.

390
00:14:01,179 --> 00:14:03,019
Doctor Kelly, is there anything else you wanted

391
00:14:03,019 --> 00:14:05,500
to leave our our audience with? Anything that,

392
00:14:05,500 --> 00:14:07,980
you know, would be helpful for our folks,

393
00:14:08,220 --> 00:14:10,059
you know, other health care leaders and hospital

394
00:14:10,059 --> 00:14:11,980
health system leaders as they're thinking about and

395
00:14:11,980 --> 00:14:14,245
trying to figure out out what their AI

396
00:14:14,245 --> 00:14:16,004
strategy is gonna be, how, you know, they

397
00:14:16,004 --> 00:14:18,644
need to prepare in order to really, you

398
00:14:18,644 --> 00:14:20,825
know, create robust programs in the future?

399
00:14:22,404 --> 00:14:24,325
No. I think the I guess the way

400
00:14:24,325 --> 00:14:26,504
I end this is to say that

401
00:14:27,590 --> 00:14:30,149
the real challenge is going to be there's

402
00:14:30,149 --> 00:14:32,629
gonna be so much opportunity. There's so many

403
00:14:32,629 --> 00:14:36,389
amazing tools and amazing technologies coming down the

404
00:14:36,389 --> 00:14:39,110
line very quickly. What we really have to

405
00:14:39,110 --> 00:14:41,049
be able to do is to understand

406
00:14:42,065 --> 00:14:44,725
how these different tools can

407
00:14:45,105 --> 00:14:47,985
be implemented and then how the effectiveness of

408
00:14:47,985 --> 00:14:50,304
the tools can be measured. That's not an

409
00:14:50,304 --> 00:14:52,544
easy thing to do. And and right now,

410
00:14:52,544 --> 00:14:54,790
I think we're we're still in that that

411
00:14:54,790 --> 00:14:56,710
early phase of adoption. It's going to be

412
00:14:56,710 --> 00:14:58,809
that next level where we can really say,

413
00:14:59,269 --> 00:15:00,710
this is the tool that makes a difference

414
00:15:00,710 --> 00:15:03,110
and these 6 tools don't. This is where

415
00:15:03,110 --> 00:15:04,950
we need to put our priorities. You know,

416
00:15:04,950 --> 00:15:06,309
those are the kinds of things that we

417
00:15:06,309 --> 00:15:07,875
need to be able to work with with

418
00:15:07,955 --> 00:15:09,634
3rd party vendors on and be able to

419
00:15:09,634 --> 00:15:11,394
say, listen. This is a nice product, but

420
00:15:11,394 --> 00:15:13,254
it's not delivering what we need it to.

421
00:15:13,315 --> 00:15:15,335
And and those are hard things to do.

422
00:15:15,875 --> 00:15:17,715
But I think if we can get that

423
00:15:17,715 --> 00:15:19,634
in place, it's going to take that that

424
00:15:19,634 --> 00:15:21,815
level of integration to really move,

425
00:15:22,970 --> 00:15:24,910
health care artificial intelligence forward.

426
00:15:25,290 --> 00:15:27,370
That's amazing. Doctor Kelly, thank you so much

427
00:15:27,370 --> 00:15:28,970
for joining us on the podcast today. This

428
00:15:28,970 --> 00:15:31,450
has been such a a wonderful discussion. I

429
00:15:31,450 --> 00:15:32,649
feel like I I've learned a lot in

430
00:15:32,649 --> 00:15:34,170
the last few minutes here, and I look

431
00:15:34,170 --> 00:15:35,950
forward to connecting with you again soon.

432
00:15:36,809 --> 00:15:38,458
I appreciate it, Laura. Thank you.