WEBVTT
1
00:00:00.080 --> 00:00:04.519
Welcome to Farmer Talk Radio. I'm Danny McCarthy. Today I'm
2
00:00:04.559 --> 00:00:08.880
speaking with Mike Eckrot, SVP of Strategic Solutions at Edvara,
3
00:00:09.119 --> 00:00:12.880
about how they're deploying artificial intelligence to pull insights out
4
00:00:12.880 --> 00:00:16.399
of structured and unstructured data sources in order to help
5
00:00:16.440 --> 00:00:21.399
sponsors avoid protocol amendments and delays. Mike, thanks for being here,
6
00:00:21.879 --> 00:00:24.239
and i'd like for you to start by talking about
7
00:00:24.280 --> 00:00:27.359
the technology that Avarra is deploying and the problems that
8
00:00:27.399 --> 00:00:28.320
it's aiming to solve.
9
00:00:29.160 --> 00:00:29.879
I think at the.
10
00:00:29.800 --> 00:00:34.240
Starting at the highest level, our product and technology leadership
11
00:00:34.320 --> 00:00:38.560
here at Atvara have used the model around build once,
12
00:00:38.679 --> 00:00:42.079
deploy everywhere, right, which is if we can build a
13
00:00:42.119 --> 00:00:46.920
really strong foundation, which is what we call PARADE, which is,
14
00:00:47.439 --> 00:00:51.399
in simplest terms, it's a data backbone, governance structure AI
15
00:00:51.920 --> 00:00:55.719
strategy that will allow us to put AI in many
16
00:00:55.719 --> 00:00:57.840
different parts of our business. And where we sit in
17
00:00:57.840 --> 00:01:01.920
the ecosystem as not only the largest central IRB but
18
00:01:02.000 --> 00:01:05.000
also the largest provider of site technologies. We see from
19
00:01:05.040 --> 00:01:09.359
a data perspective, obviously there's a lot of different artifacts
20
00:01:09.400 --> 00:01:12.120
and data fragments in our orbit. And I think if
21
00:01:12.159 --> 00:01:15.439
you think about the protocols themselves as an example, there
22
00:01:15.480 --> 00:01:19.040
are tons of lessons to be learned and derivative insights
23
00:01:19.079 --> 00:01:23.120
to be gained from those trials, not only on day
24
00:01:23.159 --> 00:01:25.560
one when they show up via the central IRB, but
25
00:01:25.599 --> 00:01:29.519
the different permutations of the trial, the different evolution of
26
00:01:29.519 --> 00:01:33.760
those protocols in those studies over time, which often go unnoticed.
27
00:01:34.159 --> 00:01:34.439
Right.
28
00:01:34.560 --> 00:01:38.239
So, if you think about your coupling the baseline design
29
00:01:38.480 --> 00:01:41.200
with what was the actual outcome that happened in the
30
00:01:41.239 --> 00:01:44.599
real world, whether it be the evolution of the protocol
31
00:01:44.640 --> 00:01:47.480
as I mentioned, but also what happened at the site level.
32
00:01:47.599 --> 00:01:50.480
Right, So, again the view of the world we have.
33
00:01:50.519 --> 00:01:54.359
By being the leading site technology offering, we see actually
34
00:01:54.519 --> 00:01:58.159
how poorly designed trials impact research sites too as well directly.
35
00:01:58.239 --> 00:02:01.480
So I think it is the sort of dual thread
36
00:02:01.519 --> 00:02:05.439
of being able to see what happens on day one,
37
00:02:05.599 --> 00:02:08.919
the evolution of those things over time via why trials
38
00:02:08.960 --> 00:02:12.520
get amended, right, the rationale behind those amendments, how those
39
00:02:12.919 --> 00:02:17.039
amendments potentially impact patient burden, have the impact site burden,
40
00:02:17.360 --> 00:02:19.080
how the impact recruitment.
41
00:02:18.639 --> 00:02:22.039
Velocity, and then ultimately outcomes at the site.
42
00:02:22.080 --> 00:02:24.439
That's where we see sort of a complete picture or
43
00:02:24.479 --> 00:02:25.560
a complete view of the world.
44
00:02:27.599 --> 00:02:32.479
How does that quantify and what metrics or numbers are
45
00:02:32.520 --> 00:02:38.280
you using to contextualize this for sponsors, it's not an.
46
00:02:38.159 --> 00:02:40.719
Easy thing to contextualize, to be very honest, And the
47
00:02:40.759 --> 00:02:42.520
analogy I often use this is a little bit of
48
00:02:42.599 --> 00:02:46.159
the chicken of the egg, right, So there are reasons
49
00:02:46.199 --> 00:02:49.919
that we've captured why trials have gone off kilter, right,
50
00:02:49.960 --> 00:02:52.560
and why they've amended. And actually, if you think about
51
00:02:52.960 --> 00:02:55.360
the cost of an amendment, we've done a ton of
52
00:02:55.400 --> 00:02:59.319
research on this, there is no single best proxy for
53
00:02:59.439 --> 00:03:03.080
what that is because there is obviously some of the
54
00:03:03.120 --> 00:03:06.280
direct costs to amend a protocol, but then there are
55
00:03:06.319 --> 00:03:09.199
also all the different downstream ratifications, whether it be re
56
00:03:09.319 --> 00:03:13.080
engaging the koos, time spent and opportunity costs, not getting
57
00:03:13.080 --> 00:03:17.080
to market, the reconfiguration of all the different twenty e
58
00:03:17.159 --> 00:03:20.039
clinical solutions and tools that now need to be remodified
59
00:03:20.039 --> 00:03:22.159
because the protocols changed, right, and the schematic of the
60
00:03:22.159 --> 00:03:25.639
trials change, etc. So the costs are I think I
61
00:03:25.680 --> 00:03:28.319
could confidently tell you the costs are exorbitant, right in
62
00:03:28.360 --> 00:03:31.479
Some of these technologies and data science approaches quickly pay
63
00:03:31.520 --> 00:03:33.840
for themselves. But I don't think it is a one
64
00:03:33.879 --> 00:03:37.360
size fits all is there's definitely nuance based on the
65
00:03:37.400 --> 00:03:40.960
trial type. The second thing is that when we think
66
00:03:41.000 --> 00:03:43.280
about how we can textualize this, when I talk about
67
00:03:43.280 --> 00:03:44.039
the chicken or the egg.
68
00:03:44.360 --> 00:03:48.360
There was a reason that the trials got caught off balance.
69
00:03:48.439 --> 00:03:52.439
But then there were also additional reasons that somebody even
70
00:03:52.520 --> 00:03:54.520
raised their hand and said, hey, we should amend this protocol.
71
00:03:54.560 --> 00:03:55.759
And that's further upstream.
72
00:03:56.120 --> 00:03:58.879
They noticed that recruitment was lagging, They noticed that there
73
00:03:58.919 --> 00:04:03.039
were sites that were potentially non compliant or protocol deviations, right,
74
00:04:03.080 --> 00:04:05.960
and maybe the protocol was too complex. We saw patients
75
00:04:06.039 --> 00:04:09.719
dropping out right, Maybe the trial was to cumbersome in
76
00:04:09.800 --> 00:04:13.280
terms of the requirements that put on patients. So there's
77
00:04:13.319 --> 00:04:15.599
a number of areas where I think you look at
78
00:04:15.919 --> 00:04:17.959
why people go down the path of amending the protocol
79
00:04:18.000 --> 00:04:20.879
to begin with, where the protocol went off kilter, and
80
00:04:20.879 --> 00:04:24.319
then using good data sciences potentially predict those things before
81
00:04:24.319 --> 00:04:26.040
they even go off off the tracks.
82
00:04:26.639 --> 00:04:29.639
And what has Edvara learned already from working with sponsors
83
00:04:29.639 --> 00:04:30.600
on this technology.
84
00:04:32.240 --> 00:04:34.959
I think the short answer is we're learning every day, right.
85
00:04:35.040 --> 00:04:37.680
So obviously we brought this offering to market over the
86
00:04:37.800 --> 00:04:40.639
last probably six to eight months. And obviously as the
87
00:04:40.720 --> 00:04:42.600
again the leading central iod be and the leading product
88
00:04:42.600 --> 00:04:45.560
our site check, we operate at the center of clind
89
00:04:45.600 --> 00:04:49.000
of operations and study startup, and I think what.
90
00:04:48.879 --> 00:04:50.560
You're seeing is that you're seeing a.
91
00:04:50.480 --> 00:04:54.399
Lot of those study startup and clinical operations stakeholders are
92
00:04:54.439 --> 00:04:56.319
waving their hand and they're saying, hey, I want to
93
00:04:56.319 --> 00:04:57.759
optimize my protocol. Right.
94
00:04:57.839 --> 00:04:58.759
But as we've kind.
95
00:04:58.600 --> 00:05:00.560
Of peeled back the onions, I think we found is
96
00:05:00.560 --> 00:05:03.000
that you have stakeholders in clinops that are interested in
97
00:05:03.040 --> 00:05:07.199
the solution, but the authorities internally are really sitting within
98
00:05:07.240 --> 00:05:10.639
clinical development, right and they are the ones who are
99
00:05:10.720 --> 00:05:14.720
ultimately the internal resident experts on how the protocols are written,
100
00:05:15.199 --> 00:05:18.160
what internal and external guidance they use, whether that be
101
00:05:18.560 --> 00:05:21.600
archatic approaches or whether it be the most cutting edge
102
00:05:21.680 --> 00:05:25.519
data science, real world data, using data potentially from organizations
103
00:05:25.560 --> 00:05:27.800
like Atvara to help drive those decisions.
104
00:05:28.360 --> 00:05:30.439
So that has required us to.
105
00:05:30.319 --> 00:05:32.680
Forge relationships on the clinical development side of the house
106
00:05:32.720 --> 00:05:35.920
where we traditionally have an operated, or rather leveraging the
107
00:05:35.959 --> 00:05:38.879
relationships we have in clinical operations to make those connections
108
00:05:38.959 --> 00:05:41.839
to their counterparts in clin dev but it is really
109
00:05:41.839 --> 00:05:43.720
a there are two sides of the coin in terms
110
00:05:43.720 --> 00:05:46.319
of how protocols come to life as well as how
111
00:05:46.439 --> 00:05:49.040
sort of a medical writing functions factor into all this.
112
00:05:49.639 --> 00:05:52.000
And yeah, again we're learning that on a daily basis,
113
00:05:52.040 --> 00:05:53.720
but I think it's been interesting so far.
114
00:05:55.079 --> 00:05:57.600
How are you doing the data science and what are
115
00:05:57.680 --> 00:06:00.199
the data sources that you're using.
116
00:06:01.480 --> 00:06:03.680
One of those are just common denominators in terms of
117
00:06:03.680 --> 00:06:06.480
how we're doing the data scientist is obviously the protocol itself.
118
00:06:06.839 --> 00:06:09.279
But when you think about a protocol, it is in
119
00:06:09.360 --> 00:06:13.040
one part structured and other parts unstructured, particularly as it
120
00:06:13.360 --> 00:06:18.560
changes and evolves over time. So our product and technology
121
00:06:18.639 --> 00:06:23.160
organizations have done a good job of effectively building models
122
00:06:23.160 --> 00:06:27.560
that allow us to extract structured insights from the right
123
00:06:27.639 --> 00:06:30.959
parts of the protocol versus going down to path that
124
00:06:31.000 --> 00:06:35.839
would not be as reliable or supportable. So when you're
125
00:06:35.879 --> 00:06:38.560
looking at the scheduled assessments, we can look at complexity.
126
00:06:38.600 --> 00:06:40.800
When you're looking at the IE criteria of the trial,
127
00:06:41.199 --> 00:06:43.600
you can look at how stringent is that criteria and.
128
00:06:43.560 --> 00:06:44.879
How it might impact recruitment.
129
00:06:45.480 --> 00:06:48.160
You look at the scheduled assessments, you could also look
130
00:06:48.199 --> 00:06:52.439
at potential areas where quality and eventual protocol.
131
00:06:52.040 --> 00:06:53.240
Deviations could happen.
132
00:06:53.839 --> 00:06:55.639
So I think it is obviously the protocol, and then
133
00:06:55.680 --> 00:06:58.040
I think when we think about downstream the connectivity to
134
00:06:58.120 --> 00:07:02.240
the site native system that they're using to operationalize these trials,
135
00:07:02.759 --> 00:07:05.279
you see a range of different functions, whether it be
136
00:07:05.800 --> 00:07:09.560
potentially the impact of those protocols on cost, on budget,
137
00:07:10.120 --> 00:07:12.759
you think about from a forecasting perspective, how long these
138
00:07:12.759 --> 00:07:15.439
trials take to come to fruition and potentially read out.
139
00:07:15.839 --> 00:07:19.199
So I think again, it's synthesizing insights across sort of
140
00:07:19.199 --> 00:07:22.000
the paradigm of different systems that we touch and being
141
00:07:22.000 --> 00:07:25.759
able to make very clear, evidence based decisions based on
142
00:07:25.839 --> 00:07:28.399
what we see and what do you.
143
00:07:28.319 --> 00:07:31.839
Feel like is something that is only achievable or capable
144
00:07:31.839 --> 00:07:35.439
of being done now with the data sources that you're
145
00:07:35.519 --> 00:07:39.439
using with the introduction of AI to create insights that
146
00:07:39.480 --> 00:07:42.600
we wouldn't have been able to create otherwise or before AI.
147
00:07:44.759 --> 00:07:46.879
I think the way that we look at the amended
148
00:07:47.000 --> 00:07:49.600
protocol or the evolution of the protocol over time, right,
149
00:07:49.680 --> 00:07:52.319
so when you look at an evolved protocol and how
150
00:07:52.360 --> 00:07:55.879
in the real world we can learn from history and
151
00:07:56.120 --> 00:07:58.399
a protocol goes and takes it's bumps and bruises and
152
00:07:58.439 --> 00:07:59.680
gets amended ten.
153
00:07:59.519 --> 00:08:00.480
Different ways ways.
154
00:08:01.360 --> 00:08:04.680
The sequencing of those amendments in terms of the time
155
00:08:04.720 --> 00:08:07.160
horizon in which they happen as well as the way
156
00:08:07.199 --> 00:08:10.639
that those things are categorized, is all over the place, right.
157
00:08:10.680 --> 00:08:13.360
So in some cases when you approach the IRB, you're
158
00:08:13.480 --> 00:08:16.240
very clear with your intent on why you're amending the protocol.
159
00:08:16.680 --> 00:08:20.560
In other cases you see structured but also unstructure changes
160
00:08:20.600 --> 00:08:25.600
within the documents themselves. And it's something, as you said,
161
00:08:25.759 --> 00:08:28.000
five to ten years ago, maybe you can get to
162
00:08:28.040 --> 00:08:30.160
the bottom of it if you had hundreds and hundreds
163
00:08:30.199 --> 00:08:34.559
of manual full timers who could go through these documents
164
00:08:34.600 --> 00:08:36.919
and sort of do deep analysis. But here in the
165
00:08:36.960 --> 00:08:39.519
AI era you have the ability to obviously to structure
166
00:08:39.519 --> 00:08:41.679
that data in a way that it's analyzable very quickly
167
00:08:42.000 --> 00:08:44.200
and then glean insights from it very very quickly.
168
00:08:44.399 --> 00:08:46.639
Right. So again, I think that's something that one it.
169
00:08:46.759 --> 00:08:49.720
Is the most probably the most important feature of what
170
00:08:49.759 --> 00:08:52.200
we're bringing to market, but also how we improve trials
171
00:08:52.240 --> 00:08:54.679
is learning from previous history.
172
00:08:55.120 --> 00:08:57.120
But also particularly.
173
00:08:56.480 --> 00:08:59.240
Something that we probably couldn't solve for have solve for
174
00:08:59.360 --> 00:09:01.039
five years ago, but a heck of a lot more.
175
00:09:00.960 --> 00:09:02.000
Expensive to do so.
176
00:09:02.399 --> 00:09:04.840
Right, But with now a cost of compute be coming
177
00:09:04.840 --> 00:09:07.679
down potentially a little bit, the ability to have access
178
00:09:07.679 --> 00:09:10.840
to tools and AI infrastructure like we do now, it's
179
00:09:10.840 --> 00:09:12.120
more solvable now than ever.
180
00:09:14.360 --> 00:09:19.279
What's emerged as either a common cause for study delay
181
00:09:19.480 --> 00:09:23.759
or a particularly oorous process for sites that you think
182
00:09:23.799 --> 00:09:25.120
sponsors should be aware of.
183
00:09:26.440 --> 00:09:29.120
Honestly, one of the things that I've found very interesting
184
00:09:29.159 --> 00:09:31.639
is obviously we have shared dialogue with sponsors who are
185
00:09:32.039 --> 00:09:35.600
developing protocols, but also the sites who are obviously contracted
186
00:09:35.600 --> 00:09:39.440
by those sponsors to operationalize those protocols, and so many
187
00:09:39.440 --> 00:09:41.159
of the sites are like I didn't need AI to
188
00:09:41.200 --> 00:09:44.080
tell you that, right, I have human level intelligence, and
189
00:09:44.159 --> 00:09:47.720
I saw the same protocol from five different sponsors in
190
00:09:47.759 --> 00:09:51.240
the last two months, and I took again the bumps
191
00:09:51.240 --> 00:09:54.759
and bruises on those early trials, and I could almost
192
00:09:54.879 --> 00:09:57.639
predict and project same thing was going to happen on
193
00:09:57.679 --> 00:10:00.320
a very similar study design that came shortly thereafter. Right.
194
00:10:00.480 --> 00:10:03.440
So I think in one part, obviously this is fueled
195
00:10:03.440 --> 00:10:07.960
by industry leading data, it's fueled by an industry leading technology.
196
00:10:08.039 --> 00:10:10.039
But I think the other part is just being good
197
00:10:10.200 --> 00:10:15.000
stewards of connecting sites and sponsors to make research more
198
00:10:15.000 --> 00:10:18.159
feasible for everybody. And some of these things I think
199
00:10:18.200 --> 00:10:23.039
are unanimous, right, and they're obvious, and particularly things around
200
00:10:23.559 --> 00:10:26.720
the schedule of assessments in the way that trials are designed,
201
00:10:26.799 --> 00:10:28.639
where a site would raise iron and blag.
202
00:10:29.679 --> 00:10:30.360
Not only a site.
203
00:10:30.399 --> 00:10:33.639
Actually, I think sometimes the clinical research stakeholders on the
204
00:10:33.639 --> 00:10:36.159
operations side of the house, within the sponsors that are
205
00:10:36.240 --> 00:10:38.840
often delegated here is the protocol, right, go run it,
206
00:10:38.879 --> 00:10:41.440
go figure it out. They're like, this isn't going to
207
00:10:41.519 --> 00:10:44.840
work in the real world. Right, I've run these trials countless,
208
00:10:45.080 --> 00:10:47.200
countless times. So I think what we're also trying to
209
00:10:47.240 --> 00:10:50.480
do is give them a data backbone that gives them
210
00:10:51.039 --> 00:10:55.720
some evidence to say, Okay, well, if clinical development effectively
211
00:10:55.720 --> 00:10:59.240
designed the protocol based on real world data, right, which
212
00:10:59.279 --> 00:11:03.639
is real world medical data outside of the clinical trial context,
213
00:11:03.919 --> 00:11:06.440
we can give clinical operations a view of here is
214
00:11:06.480 --> 00:11:09.600
actually the operational life cycle data of what happens in
215
00:11:09.639 --> 00:11:12.279
clinical trials in the real world, to be a little
216
00:11:12.279 --> 00:11:14.960
bit bit of a backbone to allow them to say, no,
217
00:11:15.120 --> 00:11:17.799
here's are his view on why this design might not
218
00:11:17.879 --> 00:11:18.480
actually work.
219
00:11:20.159 --> 00:11:23.279
What do you see as driving this new change or
220
00:11:23.320 --> 00:11:28.000
this new appetite for adopting AI, deploying it for protocol
221
00:11:28.039 --> 00:11:32.080
development and reducing amendments. What stands out to you about
222
00:11:32.120 --> 00:11:32.840
that change?
223
00:11:35.039 --> 00:11:38.519
I think right now, in our world, aiyes an unstoppable force,
224
00:11:38.960 --> 00:11:43.919
and in our consumer lives it's taken over, right, And
225
00:11:43.960 --> 00:11:45.639
I think then people will come to their day to
226
00:11:45.679 --> 00:11:48.159
day lives, their work lives, and they're like, why should
227
00:11:48.159 --> 00:11:49.480
I have any different expectation?
228
00:11:49.799 --> 00:11:49.960
Right?
229
00:11:50.000 --> 00:11:53.279
How can I harness these capabilities and these technologies to
230
00:11:53.320 --> 00:11:56.159
make my day to day life better, right, and obviously
231
00:11:56.600 --> 00:12:01.080
improve the propensity for success within my organization, right, my job,
232
00:12:01.200 --> 00:12:02.200
my professional career.
233
00:12:02.639 --> 00:12:03.759
So I think we're seeing a bit of that.
234
00:12:03.919 --> 00:12:07.120
I think also in a competitive marketplace, we work in
235
00:12:07.120 --> 00:12:11.279
an industry that's predicated on two things. Novelty, right, so
236
00:12:11.399 --> 00:12:13.600
everything is new and you don't bring things to market
237
00:12:13.639 --> 00:12:17.559
that are old and archaic. The second thing is volatility, right,
238
00:12:17.600 --> 00:12:19.799
So so many of these trials have our high risk,
239
00:12:19.840 --> 00:12:22.120
high reward and they can go off the rails very quickly,
240
00:12:22.200 --> 00:12:25.200
and they're expensive screw ups, right if.
241
00:12:25.080 --> 00:12:26.480
The trial goes off kilter.
242
00:12:27.320 --> 00:12:31.120
So I think it is again necessitating the need to
243
00:12:31.240 --> 00:12:35.919
use different techniques and different approaches, again fueled by competitiveness,
244
00:12:35.919 --> 00:12:37.600
which is like if one organization is.
245
00:12:37.559 --> 00:12:39.480
Doing it and you're stuck in the past.
246
00:12:39.600 --> 00:12:42.919
Right, if you're driving a Model T and your competitors
247
00:12:42.960 --> 00:12:45.600
are driving a flying car, and it's going to be
248
00:12:45.679 --> 00:12:48.399
very obvious shortly thereafter who took a different approach to
249
00:12:48.440 --> 00:12:50.720
get to market. So I'd say that's probably the biggest
250
00:12:50.759 --> 00:12:51.320
drive effect.
251
00:12:53.519 --> 00:12:56.320
How are you making sure that the protocol insights that
252
00:12:56.360 --> 00:13:00.399
you're generating based on a previous protocols are applicable to
253
00:13:00.840 --> 00:13:03.200
the protocol that's in front of you that you're analyzing.
254
00:13:04.399 --> 00:13:07.519
Yeah, so what you describe actually can be challenging because
255
00:13:07.840 --> 00:13:10.440
obviously our ability to do this is all predicated based
256
00:13:10.480 --> 00:13:14.679
on data anonymization criteria and data to privacy techniques that
257
00:13:14.720 --> 00:13:17.679
we use. So when you're looking to find a needle
258
00:13:17.679 --> 00:13:20.600
in a haystack, there's a high probability that var has
259
00:13:20.639 --> 00:13:22.440
it right because.
260
00:13:22.039 --> 00:13:25.600
We are arguably the largest common denominator from a trial perspective.
261
00:13:25.960 --> 00:13:30.159
But if we don't hit certain anonymization criteria, it becomes
262
00:13:30.159 --> 00:13:32.320
a moot point, Right, So I think it is a
263
00:13:32.360 --> 00:13:34.519
fine line between being able to filter their trials at
264
00:13:34.519 --> 00:13:37.879
the most granular level and make them as applicable and
265
00:13:37.960 --> 00:13:41.759
pragmatic as possible to the data science experiment that we're running,
266
00:13:42.120 --> 00:13:46.559
but also being realistic to say, again, as I mentioned earlier,
267
00:13:46.720 --> 00:13:49.600
it's an industry fueled by novelty. So the folks that
268
00:13:49.639 --> 00:13:53.279
are going after solving very complex things for very novel
269
00:13:53.600 --> 00:13:57.399
study types, there's probably not a whole lot of precedents
270
00:13:57.840 --> 00:14:00.360
that said, we do have the ability to filter data
271
00:14:00.440 --> 00:14:04.879
at really the therapeutic area level, the type indication level.
272
00:14:05.200 --> 00:14:07.039
In some cases, we're looking now at the mechanism of
273
00:14:07.080 --> 00:14:09.360
actions to try to make them as applicable as possible.
274
00:14:09.759 --> 00:14:12.799
And I think the reason that that is become another
275
00:14:12.919 --> 00:14:16.440
necessity is because you use an analogy earlier, which is
276
00:14:17.200 --> 00:14:19.279
upload a protocol and tell me how my protocol could
277
00:14:19.279 --> 00:14:22.519
be better. I think what we're seeing is at that
278
00:14:22.639 --> 00:14:25.600
stage in the game, the train has alread left the
279
00:14:25.600 --> 00:14:28.320
station right. You need to be able to understand these
280
00:14:28.399 --> 00:14:32.200
insights at sort of the conceptual level, before clinical development
281
00:14:32.320 --> 00:14:35.000
has put their hands on keys and written the protocol,
282
00:14:35.639 --> 00:14:39.039
and you don't have a structured document to feed into AI.
283
00:14:39.200 --> 00:14:42.360
At that point, you just know what phase, what therapeutic area,
284
00:14:42.600 --> 00:14:45.240
what indication, you might know what mechanism of action obviously,
285
00:14:45.320 --> 00:14:46.840
what study schematic, et cetera.
286
00:14:47.320 --> 00:14:49.399
So the ability to filter the insights at that level
287
00:14:49.519 --> 00:14:50.559
is also important.
288
00:14:52.639 --> 00:14:55.960
Ideally, what point of the process would you want to
289
00:14:56.000 --> 00:14:59.279
be working with sponsors and what would have the biggest
290
00:14:59.480 --> 00:15:01.600
downstream impact for sponsors.
291
00:15:03.600 --> 00:15:05.639
Yeah, I mean, the further we can get upstream in
292
00:15:05.679 --> 00:15:07.679
the value cycle, the better. I think.
293
00:15:07.720 --> 00:15:11.519
It all is really predicated on where in the trial
294
00:15:11.559 --> 00:15:15.440
design process can change management happen most successfully. And like
295
00:15:15.519 --> 00:15:18.600
I said, once the train has already left the station
296
00:15:19.039 --> 00:15:22.720
in the trials on a glide path to start, it
297
00:15:22.879 --> 00:15:26.120
is very hard to impact change at that level, Whereas
298
00:15:26.360 --> 00:15:29.080
if you can get involved, it's sort of the origins
299
00:15:29.120 --> 00:15:31.240
of the trial before the protocol is written that is
300
00:15:31.960 --> 00:15:34.679
more ideal. Now that said, we can also use these
301
00:15:34.759 --> 00:15:37.559
techniques to be applied to trials that are already out
302
00:15:37.600 --> 00:15:40.159
there in the real world where somebody says, hey, we're
303
00:15:40.240 --> 00:15:43.840
having a low recruitment rate, a high patient burden hydrop out,
304
00:15:43.879 --> 00:15:47.279
et cetera. Let's do a benchmark against where we stands
305
00:15:48.000 --> 00:15:51.279
versus the broader industry versus your data asset to understand
306
00:15:51.759 --> 00:15:54.440
how we're going to amend the protocol right, because I
307
00:15:54.480 --> 00:15:58.799
think in most cases these amendments are so common.
308
00:15:58.639 --> 00:16:00.000
And in some cases unavoidable.
309
00:16:00.440 --> 00:16:01.799
So if you are going to go through the arc
310
00:16:01.840 --> 00:16:05.320
of an amendment, maybe make some adjustments right then in
311
00:16:05.360 --> 00:16:08.360
there and call some audibles to make it the protocol
312
00:16:08.360 --> 00:16:09.360
a little bit more turnkey.
313
00:16:11.240 --> 00:16:15.159
You've mentioned change management in this conversation. I wonder if
314
00:16:15.200 --> 00:16:16.799
you can dive a little bit deeper into what you
315
00:16:16.919 --> 00:16:20.600
mean by that. Is it, you know, working earlier with
316
00:16:20.960 --> 00:16:24.279
companies like Edbara, Is it maintaining more flexibility to make
317
00:16:24.440 --> 00:16:27.799
changes throughout the process. What does change management look like
318
00:16:27.879 --> 00:16:28.600
in this scenario?
319
00:16:29.960 --> 00:16:34.080
Change management it means, in one part, being willing to
320
00:16:34.200 --> 00:16:41.240
look at different approaches that maybe aren't perfectly applied, but
321
00:16:41.440 --> 00:16:45.679
there's some glaring insight that could be gleaned from it
322
00:16:46.000 --> 00:16:48.240
that is going to potentially take the trial off the rails.
323
00:16:48.240 --> 00:16:49.759
And you want to say, hey, I'm going to raise
324
00:16:49.799 --> 00:16:53.000
my hand and we're going to hit the pause button
325
00:16:53.080 --> 00:16:55.080
here and we're going to add two weeks under our
326
00:16:55.159 --> 00:16:57.639
total timelines here, or we're going to go and evaluate
327
00:16:57.720 --> 00:16:58.399
this and we're going to.
328
00:16:58.399 --> 00:16:59.919
Make the decision that's best for the trial.
329
00:17:00.000 --> 00:17:02.200
Well, I think there needs to be more of that, right,
330
00:17:02.320 --> 00:17:05.200
because ultimately you end up paying for a downstream anyway,
331
00:17:05.880 --> 00:17:07.599
sometimes in multitudes.
332
00:17:08.359 --> 00:17:09.480
Whereas I think the other part.
333
00:17:09.319 --> 00:17:12.240
Of change management is just being able to balance innovation internally.
334
00:17:12.319 --> 00:17:16.920
And some sponsors and sites and clinical research organizations actually
335
00:17:17.000 --> 00:17:20.400
for that reason have developed innovation groups who are the
336
00:17:20.519 --> 00:17:23.880
leading internal resident experts on keeping the pulse in the
337
00:17:23.960 --> 00:17:26.920
market of what's new and worthy of being evaluated and
338
00:17:27.000 --> 00:17:30.599
having sort of a structured process for evaluating those tools.
339
00:17:31.200 --> 00:17:36.640
But in some organizations, those innovation organizations and departments.
340
00:17:36.920 --> 00:17:39.759
Are have a lot more prowess, right, so they have
341
00:17:39.799 --> 00:17:42.960
the ability to say, hey, that the insights that I've
342
00:17:43.160 --> 00:17:45.759
leaned are a mandate to incorporate in the trial, where
343
00:17:45.799 --> 00:17:48.599
in other organizations they don't have the same authority and
344
00:17:48.599 --> 00:17:50.720
they're just nice to have. I think that's another piece,
345
00:17:50.759 --> 00:17:53.839
which is like really being able to lean into the
346
00:17:53.920 --> 00:17:57.440
groups that are generating those innovative insights. And again, I
347
00:17:57.480 --> 00:17:59.680
think in a place where this starts to become highly
348
00:17:59.720 --> 00:18:04.240
competitive industry that's again based on novelty that'll start to
349
00:18:04.319 --> 00:18:06.519
just become commonplace, you'll have to lean into some of
350
00:18:06.559 --> 00:18:09.079
these techniques because otherwise you're going to be left in
351
00:18:09.160 --> 00:18:09.480
the dust.
352
00:18:12.319 --> 00:18:15.960
Obviously, artificial intelligence is something that everybody is interested in
353
00:18:16.400 --> 00:18:19.599
adopting in some way. Thinking about what would be your
354
00:18:19.720 --> 00:18:24.720
advice for sponsors about how to evaluate when to use AI,
355
00:18:25.160 --> 00:18:28.279
what to look for in AI partners to make sure
356
00:18:28.319 --> 00:18:30.119
that it's actually useful for them.
357
00:18:32.400 --> 00:18:38.240
The reality is nobody has been doing these cutting edge
358
00:18:38.559 --> 00:18:42.799
techniques and approaches for twenty years, right, So if you
359
00:18:42.960 --> 00:18:45.640
want to go to an organization that's going to be
360
00:18:45.720 --> 00:18:48.480
doing something on the bleeding edge, but they have ten
361
00:18:48.559 --> 00:18:51.839
years of case studies and anecdotes, we're just not at
362
00:18:51.880 --> 00:18:55.680
that state of affairship. So I think entering your AI
363
00:18:55.799 --> 00:18:59.440
evaluation with the right level of expectation, which is we
364
00:18:59.519 --> 00:19:02.000
need to be we need to be contemplative about how
365
00:19:02.039 --> 00:19:04.119
we do this. We probably need a phased approach, which
366
00:19:04.200 --> 00:19:06.559
is to pilot and do this in sort of different
367
00:19:06.599 --> 00:19:09.039
cohorts and pockets and roll it out before it sort
368
00:19:09.079 --> 00:19:13.279
of hit max scale and velocity is probably the right
369
00:19:13.319 --> 00:19:15.400
way to look at it. That said, I think what
370
00:19:15.480 --> 00:19:17.400
you're seeing is you're gonna see a lot of vendors
371
00:19:17.400 --> 00:19:20.640
who are earlier in their days of AI. We're based
372
00:19:20.680 --> 00:19:22.960
on where the industry is heading, and based on the
373
00:19:23.039 --> 00:19:25.559
fact that this is all becoming table stakes, it's becoming
374
00:19:25.599 --> 00:19:27.559
an even playing field, it's becoming ubiquitous.
375
00:19:28.559 --> 00:19:31.440
They will probably catch up in short order.
376
00:19:32.119 --> 00:19:34.480
So you have to really evaluate what are some of
377
00:19:34.519 --> 00:19:37.880
the underlying fundamentals that you have to have right, And
378
00:19:37.920 --> 00:19:39.519
I think it really boils down to one.
379
00:19:39.960 --> 00:19:41.960
You have to have strong data science leadership.
380
00:19:42.200 --> 00:19:45.920
Right, so while the technology is ubiquitous, you can screw
381
00:19:46.000 --> 00:19:48.000
it up, and the devils and the details, and as
382
00:19:48.039 --> 00:19:50.559
we know, these models can hallucinate, they need to be
383
00:19:50.640 --> 00:19:51.480
governed over time.
384
00:19:51.680 --> 00:19:53.680
You have to have strong data governance techniques.
385
00:19:54.039 --> 00:19:55.680
So I would say one is just make sure the
386
00:19:55.799 --> 00:19:58.960
organization their technology may be evolving, but just make sure
387
00:19:58.960 --> 00:20:01.680
that they have the right level of thought leadership and
388
00:20:02.119 --> 00:20:04.279
product and technology level leadership because that will be the
389
00:20:04.279 --> 00:20:08.799
difference maker. And then the second piece is the underlying
390
00:20:08.839 --> 00:20:13.240
data assets themselves. So again, as I said, even at
391
00:20:13.279 --> 00:20:17.319
the sponsor level, sponsors are building their own proprietary tools
392
00:20:17.400 --> 00:20:20.759
because the access to these tools is heavily is very
393
00:20:20.799 --> 00:20:24.240
easy to access. The underlying data assets that need to
394
00:20:24.759 --> 00:20:27.880
be fed into these models to generate deep insights, those
395
00:20:27.960 --> 00:20:31.240
are foreign few between. So I think it's evaluating partners
396
00:20:31.279 --> 00:20:33.119
that can give you a different view of the world
397
00:20:33.160 --> 00:20:35.519
that you can get from the public domain. The public
398
00:20:35.599 --> 00:20:38.000
domain is easier to scrape than ever, and I think
399
00:20:38.039 --> 00:20:42.680
we're seeing sponsors are synthesizing high volumes of public data
400
00:20:42.839 --> 00:20:46.200
very quickly. So again, I think it's partnering with folks
401
00:20:46.200 --> 00:20:49.920
who have a roadmap or a pipeline of new novel
402
00:20:49.960 --> 00:20:53.160
innovation that's predicated off some competitive mode.
403
00:20:56.359 --> 00:20:59.079
Is there anything you want to say about the utility
404
00:20:59.200 --> 00:21:03.599
of what this means for sponsors already sponsored response, anything
405
00:21:03.640 --> 00:21:06.160
it's changed on how it bar is approaching this work.
406
00:21:07.400 --> 00:21:10.119
I think AI is obviously being used in the broad
407
00:21:10.160 --> 00:21:13.480
based capacity. Finally here in clinical research, which is fantastic.
408
00:21:14.000 --> 00:21:16.440
I think where we finally have transcended as a place
409
00:21:16.480 --> 00:21:19.720
where people are attacking the low hanging fruit, the stuff
410
00:21:19.759 --> 00:21:23.720
that's safe. It doesn't put patient patient data at risk,
411
00:21:24.119 --> 00:21:26.880
it doesn't put patient equality at risk, it doesn't put
412
00:21:27.039 --> 00:21:29.480
trial quality at risk. So there are a lot of
413
00:21:29.640 --> 00:21:33.000
low hanging for use cases that are more operational in nature,
414
00:21:33.119 --> 00:21:36.359
more project management based in nature, workflow in nature that
415
00:21:36.799 --> 00:21:39.720
a get low risk high reward or maybe even low
416
00:21:39.839 --> 00:21:41.920
risk medium levels of a reward, but you have the
417
00:21:41.960 --> 00:21:45.640
ability to decrease cost you have, in some cases transparently
418
00:21:45.720 --> 00:21:49.000
have the ability to decrease human capital or repurpose human
419
00:21:49.079 --> 00:21:52.799
capital and other more meaningful uses of their time. So
420
00:21:52.839 --> 00:21:55.640
I think those are probably the big ticket items or
421
00:21:56.039 --> 00:21:59.079
things that are that we're traditionally done in an archaic
422
00:21:59.200 --> 00:22:03.480
way that we're just rudimentary.
423
00:22:03.000 --> 00:22:04.559
Level tasks that can now be automated.
424
00:22:04.920 --> 00:22:07.000
Again, that's low hanging fruit, and I think we're finally
425
00:22:07.039 --> 00:22:09.319
getting to a place in the industry where that is
426
00:22:09.440 --> 00:22:10.559
becoming more commonplace.
427
00:22:12.880 --> 00:22:15.119
If you had a crystal ball and could look one
428
00:22:15.200 --> 00:22:17.799
year from now, five years from now, what would you
429
00:22:18.119 --> 00:22:21.200
like to see with a product like braid or with
430
00:22:22.240 --> 00:22:28.480
the industry moving towards AI utilization More broadly, if I had.
431
00:22:28.400 --> 00:22:31.960
A crystal ball, I think it's probably just getting our peers,
432
00:22:32.640 --> 00:22:37.759
our partners, clinical resource organizations at large comfortable with the
433
00:22:37.839 --> 00:22:40.799
change management that has to happen to take advantage of
434
00:22:41.480 --> 00:22:44.640
some of these leading technologies. There are no shortage of
435
00:22:45.440 --> 00:22:48.160
leading technologies or insights speak leaning from a bar or
436
00:22:48.200 --> 00:22:50.240
from any company in this space. Right there's a lot
437
00:22:50.319 --> 00:22:53.160
of good technology out there. It is really who can
438
00:22:53.960 --> 00:22:57.000
again do this in a way that's contemplative. It's moving
439
00:22:57.039 --> 00:23:01.200
at the right rate and pace, it's trusted and ultimately,
440
00:23:01.319 --> 00:23:03.880
rising tides raise all ships. So I think we encourage
441
00:23:03.920 --> 00:23:06.039
our peers and different partners to also think about it
442
00:23:06.119 --> 00:23:09.519
this way and building sort of an open ecosystem of
443
00:23:09.519 --> 00:23:11.880
folks who can collaborate on these topics. And we launched
444
00:23:12.359 --> 00:23:18.240
earlier this year the EDVARA Council for FCOAI and Responsible
445
00:23:18.319 --> 00:23:20.640
Use of AI and Clinical Research, And really what that
446
00:23:20.880 --> 00:23:25.440
was was a cross industry, pre competitive space where folks
447
00:23:25.519 --> 00:23:30.039
could collaborate on what good data science looked like, all
448
00:23:30.119 --> 00:23:31.880
of the different things that keep them up that night
449
00:23:31.920 --> 00:23:34.519
in terms of scaling these technologies and approaches. And I
450
00:23:34.559 --> 00:23:37.680
think that's probably if I fast forward five years, I
451
00:23:37.720 --> 00:23:41.480
would say hopefully things like that are things that really
452
00:23:41.759 --> 00:23:43.759
move the needle for or not just our organization, but
453
00:23:43.839 --> 00:23:44.680
our industry is forward.
454
00:23:46.559 --> 00:23:50.279
That was Mike Eckro, SVP of Strategic Solutions at Vara,
455
00:23:50.599 --> 00:23:53.640
talking about how their new AI product, Braid, is mining
456
00:23:53.759 --> 00:23:57.519
data sources to provide sponsors with insights on protocol development.
457
00:23:58.240 --> 00:24:00.759
I'm Dannie McCarthy, and thank you for listening to Pharma
458
00:24:00.799 --> 00:24:01.359
Talk Radio.
1
00:00:00.080 --> 00:00:04.519
Welcome to Farmer Talk Radio. I'm Danny McCarthy. Today I'm
2
00:00:04.559 --> 00:00:08.880
speaking with Mike Eckrot, SVP of Strategic Solutions at Edvara,
3
00:00:09.119 --> 00:00:12.880
about how they're deploying artificial intelligence to pull insights out
4
00:00:12.880 --> 00:00:16.399
of structured and unstructured data sources in order to help
5
00:00:16.440 --> 00:00:21.399
sponsors avoid protocol amendments and delays. Mike, thanks for being here,
6
00:00:21.879 --> 00:00:24.239
and i'd like for you to start by talking about
7
00:00:24.280 --> 00:00:27.359
the technology that Avarra is deploying and the problems that
8
00:00:27.399 --> 00:00:28.320
it's aiming to solve.
9
00:00:29.160 --> 00:00:29.879
I think at the.
10
00:00:29.800 --> 00:00:34.240
Starting at the highest level, our product and technology leadership
11
00:00:34.320 --> 00:00:38.560
here at Atvara have used the model around build once,
12
00:00:38.679 --> 00:00:42.079
deploy everywhere, right, which is if we can build a
13
00:00:42.119 --> 00:00:46.920
really strong foundation, which is what we call PARADE, which is,
14
00:00:47.439 --> 00:00:51.399
in simplest terms, it's a data backbone, governance structure AI
15
00:00:51.920 --> 00:00:55.719
strategy that will allow us to put AI in many
16
00:00:55.719 --> 00:00:57.840
different parts of our business. And where we sit in
17
00:00:57.840 --> 00:01:01.920
the ecosystem as not only the largest central IRB but
18
00:01:02.000 --> 00:01:05.000
also the largest provider of site technologies. We see from
19
00:01:05.040 --> 00:01:09.359
a data perspective, obviously there's a lot of different artifacts
20
00:01:09.400 --> 00:01:12.120
and data fragments in our orbit. And I think if
21
00:01:12.159 --> 00:01:15.439
you think about the protocols themselves as an example, there
22
00:01:15.480 --> 00:01:19.040
are tons of lessons to be learned and derivative insights
23
00:01:19.079 --> 00:01:23.120
to be gained from those trials, not only on day
24
00:01:23.159 --> 00:01:25.560
one when they show up via the central IRB, but
25
00:01:25.599 --> 00:01:29.519
the different permutations of the trial, the different evolution of
26
00:01:29.519 --> 00:01:33.760
those protocols in those studies over time, which often go unnoticed.
27
00:01:34.159 --> 00:01:34.439
Right.
28
00:01:34.560 --> 00:01:38.239
So, if you think about your coupling the baseline design
29
00:01:38.480 --> 00:01:41.200
with what was the actual outcome that happened in the
30
00:01:41.239 --> 00:01:44.599
real world, whether it be the evolution of the protocol
31
00:01:44.640 --> 00:01:47.480
as I mentioned, but also what happened at the site level.
32
00:01:47.599 --> 00:01:50.480
Right, So, again the view of the world we have.
33
00:01:50.519 --> 00:01:54.359
By being the leading site technology offering, we see actually
34
00:01:54.519 --> 00:01:58.159
how poorly designed trials impact research sites too as well directly.
35
00:01:58.239 --> 00:02:01.480
So I think it is the sort of dual thread
36
00:02:01.519 --> 00:02:05.439
of being able to see what happens on day one,
37
00:02:05.599 --> 00:02:08.919
the evolution of those things over time via why trials
38
00:02:08.960 --> 00:02:12.520
get amended, right, the rationale behind those amendments, how those
39
00:02:12.919 --> 00:02:17.039
amendments potentially impact patient burden, have the impact site burden,
40
00:02:17.360 --> 00:02:19.080
how the impact recruitment.
41
00:02:18.639 --> 00:02:22.039
Velocity, and then ultimately outcomes at the site.
42
00:02:22.080 --> 00:02:24.439
That's where we see sort of a complete picture or
43
00:02:24.479 --> 00:02:25.560
a complete view of the world.
44
00:02:27.599 --> 00:02:32.479
How does that quantify and what metrics or numbers are
45
00:02:32.520 --> 00:02:38.280
you using to contextualize this for sponsors, it's not an.
46
00:02:38.159 --> 00:02:40.719
Easy thing to contextualize, to be very honest, And the
47
00:02:40.759 --> 00:02:42.520
analogy I often use this is a little bit of
48
00:02:42.599 --> 00:02:46.159
the chicken of the egg, right, So there are reasons
49
00:02:46.199 --> 00:02:49.919
that we've captured why trials have gone off kilter, right,
50
00:02:49.960 --> 00:02:52.560
and why they've amended. And actually, if you think about
51
00:02:52.960 --> 00:02:55.360
the cost of an amendment, we've done a ton of
52
00:02:55.400 --> 00:02:59.319
research on this, there is no single best proxy for
53
00:02:59.439 --> 00:03:03.080
what that is because there is obviously some of the
54
00:03:03.120 --> 00:03:06.280
direct costs to amend a protocol, but then there are
55
00:03:06.319 --> 00:03:09.199
also all the different downstream ratifications, whether it be re
56
00:03:09.319 --> 00:03:13.080
engaging the koos, time spent and opportunity costs, not getting
57
00:03:13.080 --> 00:03:17.080
to market, the reconfiguration of all the different twenty e
58
00:03:17.159 --> 00:03:20.039
clinical solutions and tools that now need to be remodified
59
00:03:20.039 --> 00:03:22.159
because the protocols changed, right, and the schematic of the
60
00:03:22.159 --> 00:03:25.639
trials change, etc. So the costs are I think I
61
00:03:25.680 --> 00:03:28.319
could confidently tell you the costs are exorbitant, right in
62
00:03:28.360 --> 00:03:31.479
Some of these technologies and data science approaches quickly pay
63
00:03:31.520 --> 00:03:33.840
for themselves. But I don't think it is a one
64
00:03:33.879 --> 00:03:37.360
size fits all is there's definitely nuance based on the
65
00:03:37.400 --> 00:03:40.960
trial type. The second thing is that when we think
66
00:03:41.000 --> 00:03:43.280
about how we can textualize this, when I talk about
67
00:03:43.280 --> 00:03:44.039
the chicken or the egg.
68
00:03:44.360 --> 00:03:48.360
There was a reason that the trials got caught off balance.
69
00:03:48.439 --> 00:03:52.439
But then there were also additional reasons that somebody even
70
00:03:52.520 --> 00:03:54.520
raised their hand and said, hey, we should amend this protocol.
71
00:03:54.560 --> 00:03:55.759
And that's further upstream.
72
00:03:56.120 --> 00:03:58.879
They noticed that recruitment was lagging, They noticed that there
73
00:03:58.919 --> 00:04:03.039
were sites that were potentially non compliant or protocol deviations, right,
74
00:04:03.080 --> 00:04:05.960
and maybe the protocol was too complex. We saw patients
75
00:04:06.039 --> 00:04:09.719
dropping out right, Maybe the trial was to cumbersome in
76
00:04:09.800 --> 00:04:13.280
terms of the requirements that put on patients. So there's
77
00:04:13.319 --> 00:04:15.599
a number of areas where I think you look at
78
00:04:15.919 --> 00:04:17.959
why people go down the path of amending the protocol
79
00:04:18.000 --> 00:04:20.879
to begin with, where the protocol went off kilter, and
80
00:04:20.879 --> 00:04:24.319
then using good data sciences potentially predict those things before
81
00:04:24.319 --> 00:04:26.040
they even go off off the tracks.
82
00:04:26.639 --> 00:04:29.639
And what has Edvara learned already from working with sponsors
83
00:04:29.639 --> 00:04:30.600
on this technology.
84
00:04:32.240 --> 00:04:34.959
I think the short answer is we're learning every day, right.
85
00:04:35.040 --> 00:04:37.680
So obviously we brought this offering to market over the
86
00:04:37.800 --> 00:04:40.639
last probably six to eight months. And obviously as the
87
00:04:40.720 --> 00:04:42.600
again the leading central iod be and the leading product
88
00:04:42.600 --> 00:04:45.560
our site check, we operate at the center of clind
89
00:04:45.600 --> 00:04:49.000
of operations and study startup, and I think what.
90
00:04:48.879 --> 00:04:50.560
You're seeing is that you're seeing a.
91
00:04:50.480 --> 00:04:54.399
Lot of those study startup and clinical operations stakeholders are
92
00:04:54.439 --> 00:04:56.319
waving their hand and they're saying, hey, I want to
93
00:04:56.319 --> 00:04:57.759
optimize my protocol. Right.
94
00:04:57.839 --> 00:04:58.759
But as we've kind.
95
00:04:58.600 --> 00:05:00.560
Of peeled back the onions, I think we found is
96
00:05:00.560 --> 00:05:03.000
that you have stakeholders in clinops that are interested in
97
00:05:03.040 --> 00:05:07.199
the solution, but the authorities internally are really sitting within
98
00:05:07.240 --> 00:05:10.639
clinical development, right and they are the ones who are
99
00:05:10.720 --> 00:05:14.720
ultimately the internal resident experts on how the protocols are written,
100
00:05:15.199 --> 00:05:18.160
what internal and external guidance they use, whether that be
101
00:05:18.560 --> 00:05:21.600
archatic approaches or whether it be the most cutting edge
102
00:05:21.680 --> 00:05:25.519
data science, real world data, using data potentially from organizations
103
00:05:25.560 --> 00:05:27.800
like Atvara to help drive those decisions.
104
00:05:28.360 --> 00:05:30.439
So that has required us to.
105
00:05:30.319 --> 00:05:32.680
Forge relationships on the clinical development side of the house
106
00:05:32.720 --> 00:05:35.920
where we traditionally have an operated, or rather leveraging the
107
00:05:35.959 --> 00:05:38.879
relationships we have in clinical operations to make those connections
108
00:05:38.959 --> 00:05:41.839
to their counterparts in clin dev but it is really
109
00:05:41.839 --> 00:05:43.720
a there are two sides of the coin in terms
110
00:05:43.720 --> 00:05:46.319
of how protocols come to life as well as how
111
00:05:46.439 --> 00:05:49.040
sort of a medical writing functions factor into all this.
112
00:05:49.639 --> 00:05:52.000
And yeah, again we're learning that on a daily basis,
113
00:05:52.040 --> 00:05:53.720
but I think it's been interesting so far.
114
00:05:55.079 --> 00:05:57.600
How are you doing the data science and what are
115
00:05:57.680 --> 00:06:00.199
the data sources that you're using.
116
00:06:01.480 --> 00:06:03.680
One of those are just common denominators in terms of
117
00:06:03.680 --> 00:06:06.480
how we're doing the data scientist is obviously the protocol itself.
118
00:06:06.839 --> 00:06:09.279
But when you think about a protocol, it is in
119
00:06:09.360 --> 00:06:13.040
one part structured and other parts unstructured, particularly as it
120
00:06:13.360 --> 00:06:18.560
changes and evolves over time. So our product and technology
121
00:06:18.639 --> 00:06:23.160
organizations have done a good job of effectively building models
122
00:06:23.160 --> 00:06:27.560
that allow us to extract structured insights from the right
123
00:06:27.639 --> 00:06:30.959
parts of the protocol versus going down to path that
124
00:06:31.000 --> 00:06:35.839
would not be as reliable or supportable. So when you're
125
00:06:35.879 --> 00:06:38.560
looking at the scheduled assessments, we can look at complexity.
126
00:06:38.600 --> 00:06:40.800
When you're looking at the IE criteria of the trial,
127
00:06:41.199 --> 00:06:43.600
you can look at how stringent is that criteria and.
128
00:06:43.560 --> 00:06:44.879
How it might impact recruitment.
129
00:06:45.480 --> 00:06:48.160
You look at the scheduled assessments, you could also look
130
00:06:48.199 --> 00:06:52.439
at potential areas where quality and eventual protocol.
131
00:06:52.040 --> 00:06:53.240
Deviations could happen.
132
00:06:53.839 --> 00:06:55.639
So I think it is obviously the protocol, and then
133
00:06:55.680 --> 00:06:58.040
I think when we think about downstream the connectivity to
134
00:06:58.120 --> 00:07:02.240
the site native system that they're using to operationalize these trials,
135
00:07:02.759 --> 00:07:05.279
you see a range of different functions, whether it be
136
00:07:05.800 --> 00:07:09.560
potentially the impact of those protocols on cost, on budget,
137
00:07:10.120 --> 00:07:12.759
you think about from a forecasting perspective, how long these
138
00:07:12.759 --> 00:07:15.439
trials take to come to fruition and potentially read out.
139
00:07:15.839 --> 00:07:19.199
So I think again, it's synthesizing insights across sort of
140
00:07:19.199 --> 00:07:22.000
the paradigm of different systems that we touch and being
141
00:07:22.000 --> 00:07:25.759
able to make very clear, evidence based decisions based on
142
00:07:25.839 --> 00:07:28.399
what we see and what do you.
143
00:07:28.319 --> 00:07:31.839
Feel like is something that is only achievable or capable
144
00:07:31.839 --> 00:07:35.439
of being done now with the data sources that you're
145
00:07:35.519 --> 00:07:39.439
using with the introduction of AI to create insights that
146
00:07:39.480 --> 00:07:42.600
we wouldn't have been able to create otherwise or before AI.
147
00:07:44.759 --> 00:07:46.879
I think the way that we look at the amended
148
00:07:47.000 --> 00:07:49.600
protocol or the evolution of the protocol over time, right,
149
00:07:49.680 --> 00:07:52.319
so when you look at an evolved protocol and how
150
00:07:52.360 --> 00:07:55.879
in the real world we can learn from history and
151
00:07:56.120 --> 00:07:58.399
a protocol goes and takes it's bumps and bruises and
152
00:07:58.439 --> 00:07:59.680
gets amended ten.
153
00:07:59.519 --> 00:08:00.480
Different ways ways.
154
00:08:01.360 --> 00:08:04.680
The sequencing of those amendments in terms of the time
155
00:08:04.720 --> 00:08:07.160
horizon in which they happen as well as the way
156
00:08:07.199 --> 00:08:10.639
that those things are categorized, is all over the place, right.
157
00:08:10.680 --> 00:08:13.360
So in some cases when you approach the IRB, you're
158
00:08:13.480 --> 00:08:16.240
very clear with your intent on why you're amending the protocol.
159
00:08:16.680 --> 00:08:20.560
In other cases you see structured but also unstructure changes
160
00:08:20.600 --> 00:08:25.600
within the documents themselves. And it's something, as you said,
161
00:08:25.759 --> 00:08:28.000
five to ten years ago, maybe you can get to
162
00:08:28.040 --> 00:08:30.160
the bottom of it if you had hundreds and hundreds
163
00:08:30.199 --> 00:08:34.559
of manual full timers who could go through these documents
164
00:08:34.600 --> 00:08:36.919
and sort of do deep analysis. But here in the
165
00:08:36.960 --> 00:08:39.519
AI era you have the ability to obviously to structure
166
00:08:39.519 --> 00:08:41.679
that data in a way that it's analyzable very quickly
167
00:08:42.000 --> 00:08:44.200
and then glean insights from it very very quickly.
168
00:08:44.399 --> 00:08:46.639
Right. So again, I think that's something that one it.
169
00:08:46.759 --> 00:08:49.720
Is the most probably the most important feature of what
170
00:08:49.759 --> 00:08:52.200
we're bringing to market, but also how we improve trials
171
00:08:52.240 --> 00:08:54.679
is learning from previous history.
172
00:08:55.120 --> 00:08:57.120
But also particularly.
173
00:08:56.480 --> 00:08:59.240
Something that we probably couldn't solve for have solve for
174
00:08:59.360 --> 00:09:01.039
five years ago, but a heck of a lot more.
175
00:09:00.960 --> 00:09:02.000
Expensive to do so.
176
00:09:02.399 --> 00:09:04.840
Right, But with now a cost of compute be coming
177
00:09:04.840 --> 00:09:07.679
down potentially a little bit, the ability to have access
178
00:09:07.679 --> 00:09:10.840
to tools and AI infrastructure like we do now, it's
179
00:09:10.840 --> 00:09:12.120
more solvable now than ever.
180
00:09:14.360 --> 00:09:19.279
What's emerged as either a common cause for study delay
181
00:09:19.480 --> 00:09:23.759
or a particularly oorous process for sites that you think
182
00:09:23.799 --> 00:09:25.120
sponsors should be aware of.
183
00:09:26.440 --> 00:09:29.120
Honestly, one of the things that I've found very interesting
184
00:09:29.159 --> 00:09:31.639
is obviously we have shared dialogue with sponsors who are
185
00:09:32.039 --> 00:09:35.600
developing protocols, but also the sites who are obviously contracted
186
00:09:35.600 --> 00:09:39.440
by those sponsors to operationalize those protocols, and so many
187
00:09:39.440 --> 00:09:41.159
of the sites are like I didn't need AI to
188
00:09:41.200 --> 00:09:44.080
tell you that, right, I have human level intelligence, and
189
00:09:44.159 --> 00:09:47.720
I saw the same protocol from five different sponsors in
190
00:09:47.759 --> 00:09:51.240
the last two months, and I took again the bumps
191
00:09:51.240 --> 00:09:54.759
and bruises on those early trials, and I could almost
192
00:09:54.879 --> 00:09:57.639
predict and project same thing was going to happen on
193
00:09:57.679 --> 00:10:00.320
a very similar study design that came shortly thereafter. Right.
194
00:10:00.480 --> 00:10:03.440
So I think in one part, obviously this is fueled
195
00:10:03.440 --> 00:10:07.960
by industry leading data, it's fueled by an industry leading technology.
196
00:10:08.039 --> 00:10:10.039
But I think the other part is just being good
197
00:10:10.200 --> 00:10:15.000
stewards of connecting sites and sponsors to make research more
198
00:10:15.000 --> 00:10:18.159
feasible for everybody. And some of these things I think
199
00:10:18.200 --> 00:10:23.039
are unanimous, right, and they're obvious, and particularly things around
200
00:10:23.559 --> 00:10:26.720
the schedule of assessments in the way that trials are designed,
201
00:10:26.799 --> 00:10:28.639
where a site would raise iron and blag.
202
00:10:29.679 --> 00:10:30.360
Not only a site.
203
00:10:30.399 --> 00:10:33.639
Actually, I think sometimes the clinical research stakeholders on the
204
00:10:33.639 --> 00:10:36.159
operations side of the house, within the sponsors that are
205
00:10:36.240 --> 00:10:38.840
often delegated here is the protocol, right, go run it,
206
00:10:38.879 --> 00:10:41.440
go figure it out. They're like, this isn't going to
207
00:10:41.519 --> 00:10:44.840
work in the real world. Right, I've run these trials countless,
208
00:10:45.080 --> 00:10:47.200
countless times. So I think what we're also trying to
209
00:10:47.240 --> 00:10:50.480
do is give them a data backbone that gives them
210
00:10:51.039 --> 00:10:55.720
some evidence to say, Okay, well, if clinical development effectively
211
00:10:55.720 --> 00:10:59.240
designed the protocol based on real world data, right, which
212
00:10:59.279 --> 00:11:03.639
is real world medical data outside of the clinical trial context,
213
00:11:03.919 --> 00:11:06.440
we can give clinical operations a view of here is
214
00:11:06.480 --> 00:11:09.600
actually the operational life cycle data of what happens in
215
00:11:09.639 --> 00:11:12.279
clinical trials in the real world, to be a little
216
00:11:12.279 --> 00:11:14.960
bit bit of a backbone to allow them to say, no,
217
00:11:15.120 --> 00:11:17.799
here's are his view on why this design might not
218
00:11:17.879 --> 00:11:18.480
actually work.
219
00:11:20.159 --> 00:11:23.279
What do you see as driving this new change or
220
00:11:23.320 --> 00:11:28.000
this new appetite for adopting AI, deploying it for protocol
221
00:11:28.039 --> 00:11:32.080
development and reducing amendments. What stands out to you about
222
00:11:32.120 --> 00:11:32.840
that change?
223
00:11:35.039 --> 00:11:38.519
I think right now, in our world, aiyes an unstoppable force,
224
00:11:38.960 --> 00:11:43.919
and in our consumer lives it's taken over, right, And
225
00:11:43.960 --> 00:11:45.639
I think then people will come to their day to
226
00:11:45.679 --> 00:11:48.159
day lives, their work lives, and they're like, why should
227
00:11:48.159 --> 00:11:49.480
I have any different expectation?
228
00:11:49.799 --> 00:11:49.960
Right?
229
00:11:50.000 --> 00:11:53.279
How can I harness these capabilities and these technologies to
230
00:11:53.320 --> 00:11:56.159
make my day to day life better, right, and obviously
231
00:11:56.600 --> 00:12:01.080
improve the propensity for success within my organization, right, my job,
232
00:12:01.200 --> 00:12:02.200
my professional career.
233
00:12:02.639 --> 00:12:03.759
So I think we're seeing a bit of that.
234
00:12:03.919 --> 00:12:07.120
I think also in a competitive marketplace, we work in
235
00:12:07.120 --> 00:12:11.279
an industry that's predicated on two things. Novelty, right, so
236
00:12:11.399 --> 00:12:13.600
everything is new and you don't bring things to market
237
00:12:13.639 --> 00:12:17.559
that are old and archaic. The second thing is volatility, right,
238
00:12:17.600 --> 00:12:19.799
So so many of these trials have our high risk,
239
00:12:19.840 --> 00:12:22.120
high reward and they can go off the rails very quickly,
240
00:12:22.200 --> 00:12:25.200
and they're expensive screw ups, right if.
241
00:12:25.080 --> 00:12:26.480
The trial goes off kilter.
242
00:12:27.320 --> 00:12:31.120
So I think it is again necessitating the need to
243
00:12:31.240 --> 00:12:35.919
use different techniques and different approaches, again fueled by competitiveness,
244
00:12:35.919 --> 00:12:37.600
which is like if one organization is.
245
00:12:37.559 --> 00:12:39.480
Doing it and you're stuck in the past.
246
00:12:39.600 --> 00:12:42.919
Right, if you're driving a Model T and your competitors
247
00:12:42.960 --> 00:12:45.600
are driving a flying car, and it's going to be
248
00:12:45.679 --> 00:12:48.399
very obvious shortly thereafter who took a different approach to
249
00:12:48.440 --> 00:12:50.720
get to market. So I'd say that's probably the biggest
250
00:12:50.759 --> 00:12:51.320
drive effect.
251
00:12:53.519 --> 00:12:56.320
How are you making sure that the protocol insights that
252
00:12:56.360 --> 00:13:00.399
you're generating based on a previous protocols are applicable to
253
00:13:00.840 --> 00:13:03.200
the protocol that's in front of you that you're analyzing.
254
00:13:04.399 --> 00:13:07.519
Yeah, so what you describe actually can be challenging because
255
00:13:07.840 --> 00:13:10.440
obviously our ability to do this is all predicated based
256
00:13:10.480 --> 00:13:14.679
on data anonymization criteria and data to privacy techniques that
257
00:13:14.720 --> 00:13:17.679
we use. So when you're looking to find a needle
258
00:13:17.679 --> 00:13:20.600
in a haystack, there's a high probability that var has
259
00:13:20.639 --> 00:13:22.440
it right because.
260
00:13:22.039 --> 00:13:25.600
We are arguably the largest common denominator from a trial perspective.
261
00:13:25.960 --> 00:13:30.159
But if we don't hit certain anonymization criteria, it becomes
262
00:13:30.159 --> 00:13:32.320
a moot point, Right, So I think it is a
263
00:13:32.360 --> 00:13:34.519
fine line between being able to filter their trials at
264
00:13:34.519 --> 00:13:37.879
the most granular level and make them as applicable and
265
00:13:37.960 --> 00:13:41.759
pragmatic as possible to the data science experiment that we're running,
266
00:13:42.120 --> 00:13:46.559
but also being realistic to say, again, as I mentioned earlier,
267
00:13:46.720 --> 00:13:49.600
it's an industry fueled by novelty. So the folks that
268
00:13:49.639 --> 00:13:53.279
are going after solving very complex things for very novel
269
00:13:53.600 --> 00:13:57.399
study types, there's probably not a whole lot of precedents
270
00:13:57.840 --> 00:14:00.360
that said, we do have the ability to filter data
271
00:14:00.440 --> 00:14:04.879
at really the therapeutic area level, the type indication level.
272
00:14:05.200 --> 00:14:07.039
In some cases, we're looking now at the mechanism of
273
00:14:07.080 --> 00:14:09.360
actions to try to make them as applicable as possible.
274
00:14:09.759 --> 00:14:12.799
And I think the reason that that is become another
275
00:14:12.919 --> 00:14:16.440
necessity is because you use an analogy earlier, which is
276
00:14:17.200 --> 00:14:19.279
upload a protocol and tell me how my protocol could
277
00:14:19.279 --> 00:14:22.519
be better. I think what we're seeing is at that
278
00:14:22.639 --> 00:14:25.600
stage in the game, the train has alread left the
279
00:14:25.600 --> 00:14:28.320
station right. You need to be able to understand these
280
00:14:28.399 --> 00:14:32.200
insights at sort of the conceptual level, before clinical development
281
00:14:32.320 --> 00:14:35.000
has put their hands on keys and written the protocol,
282
00:14:35.639 --> 00:14:39.039
and you don't have a structured document to feed into AI.
283
00:14:39.200 --> 00:14:42.360
At that point, you just know what phase, what therapeutic area,
284
00:14:42.600 --> 00:14:45.240
what indication, you might know what mechanism of action obviously,
285
00:14:45.320 --> 00:14:46.840
what study schematic, et cetera.
286
00:14:47.320 --> 00:14:49.399
So the ability to filter the insights at that level
287
00:14:49.519 --> 00:14:50.559
is also important.
288
00:14:52.639 --> 00:14:55.960
Ideally, what point of the process would you want to
289
00:14:56.000 --> 00:14:59.279
be working with sponsors and what would have the biggest
290
00:14:59.480 --> 00:15:01.600
downstream impact for sponsors.
291
00:15:03.600 --> 00:15:05.639
Yeah, I mean, the further we can get upstream in
292
00:15:05.679 --> 00:15:07.679
the value cycle, the better. I think.
293
00:15:07.720 --> 00:15:11.519
It all is really predicated on where in the trial
294
00:15:11.559 --> 00:15:15.440
design process can change management happen most successfully. And like
295
00:15:15.519 --> 00:15:18.600
I said, once the train has already left the station
296
00:15:19.039 --> 00:15:22.720
in the trials on a glide path to start, it
297
00:15:22.879 --> 00:15:26.120
is very hard to impact change at that level, Whereas
298
00:15:26.360 --> 00:15:29.080
if you can get involved, it's sort of the origins
299
00:15:29.120 --> 00:15:31.240
of the trial before the protocol is written that is
300
00:15:31.960 --> 00:15:34.679
more ideal. Now that said, we can also use these
301
00:15:34.759 --> 00:15:37.559
techniques to be applied to trials that are already out
302
00:15:37.600 --> 00:15:40.159
there in the real world where somebody says, hey, we're
303
00:15:40.240 --> 00:15:43.840
having a low recruitment rate, a high patient burden hydrop out,
304
00:15:43.879 --> 00:15:47.279
et cetera. Let's do a benchmark against where we stands
305
00:15:48.000 --> 00:15:51.279
versus the broader industry versus your data asset to understand
306
00:15:51.759 --> 00:15:54.440
how we're going to amend the protocol right, because I
307
00:15:54.480 --> 00:15:58.799
think in most cases these amendments are so common.
308
00:15:58.639 --> 00:16:00.000
And in some cases unavoidable.
309
00:16:00.440 --> 00:16:01.799
So if you are going to go through the arc
310
00:16:01.840 --> 00:16:05.320
of an amendment, maybe make some adjustments right then in
311
00:16:05.360 --> 00:16:08.360
there and call some audibles to make it the protocol
312
00:16:08.360 --> 00:16:09.360
a little bit more turnkey.
313
00:16:11.240 --> 00:16:15.159
You've mentioned change management in this conversation. I wonder if
314
00:16:15.200 --> 00:16:16.799
you can dive a little bit deeper into what you
315
00:16:16.919 --> 00:16:20.600
mean by that. Is it, you know, working earlier with
316
00:16:20.960 --> 00:16:24.279
companies like Edbara, Is it maintaining more flexibility to make
317
00:16:24.440 --> 00:16:27.799
changes throughout the process. What does change management look like
318
00:16:27.879 --> 00:16:28.600
in this scenario?
319
00:16:29.960 --> 00:16:34.080
Change management it means, in one part, being willing to
320
00:16:34.200 --> 00:16:41.240
look at different approaches that maybe aren't perfectly applied, but
321
00:16:41.440 --> 00:16:45.679
there's some glaring insight that could be gleaned from it
322
00:16:46.000 --> 00:16:48.240
that is going to potentially take the trial off the rails.
323
00:16:48.240 --> 00:16:49.759
And you want to say, hey, I'm going to raise
324
00:16:49.799 --> 00:16:53.000
my hand and we're going to hit the pause button
325
00:16:53.080 --> 00:16:55.080
here and we're going to add two weeks under our
326
00:16:55.159 --> 00:16:57.639
total timelines here, or we're going to go and evaluate
327
00:16:57.720 --> 00:16:58.399
this and we're going to.
328
00:16:58.399 --> 00:16:59.919
Make the decision that's best for the trial.
329
00:17:00.000 --> 00:17:02.200
Well, I think there needs to be more of that, right,
330
00:17:02.320 --> 00:17:05.200
because ultimately you end up paying for a downstream anyway,
331
00:17:05.880 --> 00:17:07.599
sometimes in multitudes.
332
00:17:08.359 --> 00:17:09.480
Whereas I think the other part.
333
00:17:09.319 --> 00:17:12.240
Of change management is just being able to balance innovation internally.
334
00:17:12.319 --> 00:17:16.920
And some sponsors and sites and clinical research organizations actually
335
00:17:17.000 --> 00:17:20.400
for that reason have developed innovation groups who are the
336
00:17:20.519 --> 00:17:23.880
leading internal resident experts on keeping the pulse in the
337
00:17:23.960 --> 00:17:26.920
market of what's new and worthy of being evaluated and
338
00:17:27.000 --> 00:17:30.599
having sort of a structured process for evaluating those tools.
339
00:17:31.200 --> 00:17:36.640
But in some organizations, those innovation organizations and departments.
340
00:17:36.920 --> 00:17:39.759
Are have a lot more prowess, right, so they have
341
00:17:39.799 --> 00:17:42.960
the ability to say, hey, that the insights that I've
342
00:17:43.160 --> 00:17:45.759
leaned are a mandate to incorporate in the trial, where
343
00:17:45.799 --> 00:17:48.599
in other organizations they don't have the same authority and
344
00:17:48.599 --> 00:17:50.720
they're just nice to have. I think that's another piece,
345
00:17:50.759 --> 00:17:53.839
which is like really being able to lean into the
346
00:17:53.920 --> 00:17:57.440
groups that are generating those innovative insights. And again, I
347
00:17:57.480 --> 00:17:59.680
think in a place where this starts to become highly
348
00:17:59.720 --> 00:18:04.240
competitive industry that's again based on novelty that'll start to
349
00:18:04.319 --> 00:18:06.519
just become commonplace, you'll have to lean into some of
350
00:18:06.559 --> 00:18:09.079
these techniques because otherwise you're going to be left in
351
00:18:09.160 --> 00:18:09.480
the dust.
352
00:18:12.319 --> 00:18:15.960
Obviously, artificial intelligence is something that everybody is interested in
353
00:18:16.400 --> 00:18:19.599
adopting in some way. Thinking about what would be your
354
00:18:19.720 --> 00:18:24.720
advice for sponsors about how to evaluate when to use AI,
355
00:18:25.160 --> 00:18:28.279
what to look for in AI partners to make sure
356
00:18:28.319 --> 00:18:30.119
that it's actually useful for them.
357
00:18:32.400 --> 00:18:38.240
The reality is nobody has been doing these cutting edge
358
00:18:38.559 --> 00:18:42.799
techniques and approaches for twenty years, right, So if you
359
00:18:42.960 --> 00:18:45.640
want to go to an organization that's going to be
360
00:18:45.720 --> 00:18:48.480
doing something on the bleeding edge, but they have ten
361
00:18:48.559 --> 00:18:51.839
years of case studies and anecdotes, we're just not at
362
00:18:51.880 --> 00:18:55.680
that state of affairship. So I think entering your AI
363
00:18:55.799 --> 00:18:59.440
evaluation with the right level of expectation, which is we
364
00:18:59.519 --> 00:19:02.000
need to be we need to be contemplative about how
365
00:19:02.039 --> 00:19:04.119
we do this. We probably need a phased approach, which
366
00:19:04.200 --> 00:19:06.559
is to pilot and do this in sort of different
367
00:19:06.599 --> 00:19:09.039
cohorts and pockets and roll it out before it sort
368
00:19:09.079 --> 00:19:13.279
of hit max scale and velocity is probably the right
369
00:19:13.319 --> 00:19:15.400
way to look at it. That said, I think what
370
00:19:15.480 --> 00:19:17.400
you're seeing is you're gonna see a lot of vendors
371
00:19:17.400 --> 00:19:20.640
who are earlier in their days of AI. We're based
372
00:19:20.680 --> 00:19:22.960
on where the industry is heading, and based on the
373
00:19:23.039 --> 00:19:25.559
fact that this is all becoming table stakes, it's becoming
374
00:19:25.599 --> 00:19:27.559
an even playing field, it's becoming ubiquitous.
375
00:19:28.559 --> 00:19:31.440
They will probably catch up in short order.
376
00:19:32.119 --> 00:19:34.480
So you have to really evaluate what are some of
377
00:19:34.519 --> 00:19:37.880
the underlying fundamentals that you have to have right, And
378
00:19:37.920 --> 00:19:39.519
I think it really boils down to one.
379
00:19:39.960 --> 00:19:41.960
You have to have strong data science leadership.
380
00:19:42.200 --> 00:19:45.920
Right, so while the technology is ubiquitous, you can screw
381
00:19:46.000 --> 00:19:48.000
it up, and the devils and the details, and as
382
00:19:48.039 --> 00:19:50.559
we know, these models can hallucinate, they need to be
383
00:19:50.640 --> 00:19:51.480
governed over time.
384
00:19:51.680 --> 00:19:53.680
You have to have strong data governance techniques.
385
00:19:54.039 --> 00:19:55.680
So I would say one is just make sure the
386
00:19:55.799 --> 00:19:58.960
organization their technology may be evolving, but just make sure
387
00:19:58.960 --> 00:20:01.680
that they have the right level of thought leadership and
388
00:20:02.119 --> 00:20:04.279
product and technology level leadership because that will be the
389
00:20:04.279 --> 00:20:08.799
difference maker. And then the second piece is the underlying
390
00:20:08.839 --> 00:20:13.240
data assets themselves. So again, as I said, even at
391
00:20:13.279 --> 00:20:17.319
the sponsor level, sponsors are building their own proprietary tools
392
00:20:17.400 --> 00:20:20.759
because the access to these tools is heavily is very
393
00:20:20.799 --> 00:20:24.240
easy to access. The underlying data assets that need to
394
00:20:24.759 --> 00:20:27.880
be fed into these models to generate deep insights, those
395
00:20:27.960 --> 00:20:31.240
are foreign few between. So I think it's evaluating partners
396
00:20:31.279 --> 00:20:33.119
that can give you a different view of the world
397
00:20:33.160 --> 00:20:35.519
that you can get from the public domain. The public
398
00:20:35.599 --> 00:20:38.000
domain is easier to scrape than ever, and I think
399
00:20:38.039 --> 00:20:42.680
we're seeing sponsors are synthesizing high volumes of public data
400
00:20:42.839 --> 00:20:46.200
very quickly. So again, I think it's partnering with folks
401
00:20:46.200 --> 00:20:49.920
who have a roadmap or a pipeline of new novel
402
00:20:49.960 --> 00:20:53.160
innovation that's predicated off some competitive mode.
403
00:20:56.359 --> 00:20:59.079
Is there anything you want to say about the utility
404
00:20:59.200 --> 00:21:03.599
of what this means for sponsors already sponsored response, anything
405
00:21:03.640 --> 00:21:06.160
it's changed on how it bar is approaching this work.
406
00:21:07.400 --> 00:21:10.119
I think AI is obviously being used in the broad
407
00:21:10.160 --> 00:21:13.480
based capacity. Finally here in clinical research, which is fantastic.
408
00:21:14.000 --> 00:21:16.440
I think where we finally have transcended as a place
409
00:21:16.480 --> 00:21:19.720
where people are attacking the low hanging fruit, the stuff
410
00:21:19.759 --> 00:21:23.720
that's safe. It doesn't put patient patient data at risk,
411
00:21:24.119 --> 00:21:26.880
it doesn't put patient equality at risk, it doesn't put
412
00:21:27.039 --> 00:21:29.480
trial quality at risk. So there are a lot of
413
00:21:29.640 --> 00:21:33.000
low hanging for use cases that are more operational in nature,
414
00:21:33.119 --> 00:21:36.359
more project management based in nature, workflow in nature that
415
00:21:36.799 --> 00:21:39.720
a get low risk high reward or maybe even low
416
00:21:39.839 --> 00:21:41.920
risk medium levels of a reward, but you have the
417
00:21:41.960 --> 00:21:45.640
ability to decrease cost you have, in some cases transparently
418
00:21:45.720 --> 00:21:49.000
have the ability to decrease human capital or repurpose human
419
00:21:49.079 --> 00:21:52.799
capital and other more meaningful uses of their time. So
420
00:21:52.839 --> 00:21:55.640
I think those are probably the big ticket items or
421
00:21:56.039 --> 00:21:59.079
things that are that we're traditionally done in an archaic
422
00:21:59.200 --> 00:22:03.480
way that we're just rudimentary.
423
00:22:03.000 --> 00:22:04.559
Level tasks that can now be automated.
424
00:22:04.920 --> 00:22:07.000
Again, that's low hanging fruit, and I think we're finally
425
00:22:07.039 --> 00:22:09.319
getting to a place in the industry where that is
426
00:22:09.440 --> 00:22:10.559
becoming more commonplace.
427
00:22:12.880 --> 00:22:15.119
If you had a crystal ball and could look one
428
00:22:15.200 --> 00:22:17.799
year from now, five years from now, what would you
429
00:22:18.119 --> 00:22:21.200
like to see with a product like braid or with
430
00:22:22.240 --> 00:22:28.480
the industry moving towards AI utilization More broadly, if I had.
431
00:22:28.400 --> 00:22:31.960
A crystal ball, I think it's probably just getting our peers,
432
00:22:32.640 --> 00:22:37.759
our partners, clinical resource organizations at large comfortable with the
433
00:22:37.839 --> 00:22:40.799
change management that has to happen to take advantage of
434
00:22:41.480 --> 00:22:44.640
some of these leading technologies. There are no shortage of
435
00:22:45.440 --> 00:22:48.160
leading technologies or insights speak leaning from a bar or
436
00:22:48.200 --> 00:22:50.240
from any company in this space. Right there's a lot
437
00:22:50.319 --> 00:22:53.160
of good technology out there. It is really who can
438
00:22:53.960 --> 00:22:57.000
again do this in a way that's contemplative. It's moving
439
00:22:57.039 --> 00:23:01.200
at the right rate and pace, it's trusted and ultimately,
440
00:23:01.319 --> 00:23:03.880
rising tides raise all ships. So I think we encourage
441
00:23:03.920 --> 00:23:06.039
our peers and different partners to also think about it
442
00:23:06.119 --> 00:23:09.519
this way and building sort of an open ecosystem of
443
00:23:09.519 --> 00:23:11.880
folks who can collaborate on these topics. And we launched
444
00:23:12.359 --> 00:23:18.240
earlier this year the EDVARA Council for FCOAI and Responsible
445
00:23:18.319 --> 00:23:20.640
Use of AI and Clinical Research, And really what that
446
00:23:20.880 --> 00:23:25.440
was was a cross industry, pre competitive space where folks
447
00:23:25.519 --> 00:23:30.039
could collaborate on what good data science looked like, all
448
00:23:30.119 --> 00:23:31.880
of the different things that keep them up that night
449
00:23:31.920 --> 00:23:34.519
in terms of scaling these technologies and approaches. And I
450
00:23:34.559 --> 00:23:37.680
think that's probably if I fast forward five years, I
451
00:23:37.720 --> 00:23:41.480
would say hopefully things like that are things that really
452
00:23:41.759 --> 00:23:43.759
move the needle for or not just our organization, but
453
00:23:43.839 --> 00:23:44.680
our industry is forward.
454
00:23:46.559 --> 00:23:50.279
That was Mike Eckro, SVP of Strategic Solutions at Vara,
455
00:23:50.599 --> 00:23:53.640
talking about how their new AI product, Braid, is mining
456
00:23:53.759 --> 00:23:57.519
data sources to provide sponsors with insights on protocol development.
457
00:23:58.240 --> 00:24:00.759
I'm Dannie McCarthy, and thank you for listening to Pharma
458
00:24:00.799 --> 00:24:01.359
Talk Radio.
