1
00:00:00,000 --> 00:00:02,159
Hi, everyone. This is Brian Zerman with Becker's

2
00:00:02,159 --> 00:00:04,000
Healthcare. Thank you so much for tuning into

3
00:00:04,000 --> 00:00:05,379
the Becker's Healthcare podcast.

4
00:00:06,000 --> 00:00:08,000
Today, we're going to talk about how AI

5
00:00:08,000 --> 00:00:09,779
is transforming the payer strategy.

6
00:00:10,320 --> 00:00:12,799
Joining me for today's discussion live in Chicago

7
00:00:12,799 --> 00:00:15,359
at Becker's Payer Issues Roundtable is doctor Mira

8
00:00:15,359 --> 00:00:18,475
Atkins, chief medical officer at Lyric. Doctor Erica

9
00:00:18,475 --> 00:00:20,094
Atkins, thank you so much for being here.

10
00:00:20,394 --> 00:00:22,475
Thanks for having me. So to get us

11
00:00:22,475 --> 00:00:24,154
started here, can you just share a bit

12
00:00:24,154 --> 00:00:26,074
more about yourself, the work you're doing at

13
00:00:26,074 --> 00:00:28,474
Lyric, just so listeners can sort of appreciate

14
00:00:28,474 --> 00:00:29,755
where you're coming from as we move through

15
00:00:29,755 --> 00:00:33,030
the conversation today? Sure. So I'm actually a

16
00:00:33,030 --> 00:00:36,070
board certified OB GYN, and I practiced for

17
00:00:36,070 --> 00:00:39,049
several years prior to transitioning out of practice.

18
00:00:39,590 --> 00:00:41,429
And I've been in the payer space for

19
00:00:41,429 --> 00:00:44,390
the last ten years, but recently joined Lyric

20
00:00:44,390 --> 00:00:45,929
as their chief medical officer,

21
00:00:46,265 --> 00:00:48,585
where my focus is really to scale clinically

22
00:00:48,585 --> 00:00:50,844
informed AI for payment accuracy.

23
00:00:51,304 --> 00:00:54,524
And we're really moving towards prepayment intelligent that

24
00:00:54,825 --> 00:00:57,565
just helps to preserve the provider trust.

25
00:00:58,024 --> 00:01:00,929
Excellent. I appreciate that background. So here at

26
00:01:00,929 --> 00:01:02,549
at at Becker's Parish's roundtable,

27
00:01:03,090 --> 00:01:04,930
you've discussed with other leaders how AI and

28
00:01:04,930 --> 00:01:07,409
advanced analytics are are helping payers move from

29
00:01:07,409 --> 00:01:09,829
reactive to proactive care and payment models.

30
00:01:10,290 --> 00:01:12,129
So from your perspective, how is the shift

31
00:01:12,129 --> 00:01:14,474
showing up in areas like payment integrity and

32
00:01:14,474 --> 00:01:16,795
clinical validation, which you just flagged as really

33
00:01:16,795 --> 00:01:18,495
important to the to the work you're doing?

34
00:01:18,795 --> 00:01:21,034
And, really, I guess, in that context, if

35
00:01:21,034 --> 00:01:21,614
you could

36
00:01:22,075 --> 00:01:24,075
unpack what the word proactive actually means. I

37
00:01:24,075 --> 00:01:26,075
think we hear that word a lot, but

38
00:01:26,075 --> 00:01:26,715
put some,

39
00:01:27,275 --> 00:01:29,880
a concrete definition for folks out there. Sure.

40
00:01:29,880 --> 00:01:32,680
So for decades, you know, claims review has

41
00:01:32,680 --> 00:01:33,980
been really a reactive

42
00:01:34,439 --> 00:01:34,939
adversarial

43
00:01:35,319 --> 00:01:35,819
exercise

44
00:01:36,599 --> 00:01:37,099
where

45
00:01:37,480 --> 00:01:40,280
we are looking at either post payment and

46
00:01:40,280 --> 00:01:42,620
audits, and it's just been a really

47
00:01:43,155 --> 00:01:44,215
blunt force.

48
00:01:44,754 --> 00:01:47,234
I think the future is really we're moving

49
00:01:47,234 --> 00:01:50,515
towards the left where clinical AI provides a

50
00:01:50,515 --> 00:01:51,655
lot more transparency,

51
00:01:51,954 --> 00:01:54,534
and it drives more appropriate payments,

52
00:01:55,010 --> 00:01:58,049
where it reduces provider friction, and it really

53
00:01:58,049 --> 00:01:58,549
focuses

54
00:01:59,569 --> 00:02:03,189
people on the outliers. Those cases that really

55
00:02:03,329 --> 00:02:05,569
matter where you're getting the the high hit

56
00:02:05,569 --> 00:02:06,469
rates, I guess.

57
00:02:07,025 --> 00:02:09,525
Yeah. And that moving away from that adversarial

58
00:02:09,824 --> 00:02:10,645
sort of nature,

59
00:02:11,264 --> 00:02:12,564
do you think that AI

60
00:02:13,344 --> 00:02:15,425
is really an opportunity for that finally to

61
00:02:15,425 --> 00:02:17,525
happen? Do you really think that this technology

62
00:02:17,584 --> 00:02:19,584
is kind of might be the the the

63
00:02:19,584 --> 00:02:21,264
solution to some of the that sort of

64
00:02:21,264 --> 00:02:21,764
administrative

65
00:02:22,719 --> 00:02:25,439
gridlock, I guess. Yes. Absolutely. And I think

66
00:02:25,439 --> 00:02:27,840
it takes, like, true partnership and trust in

67
00:02:27,840 --> 00:02:30,020
order for to make that happen.

68
00:02:30,480 --> 00:02:32,180
So I think that the technology

69
00:02:32,800 --> 00:02:35,040
is out there. I think we just the

70
00:02:35,040 --> 00:02:35,540
people

71
00:02:36,240 --> 00:02:37,540
have to have the trust

72
00:02:37,840 --> 00:02:38,294
and

73
00:02:39,814 --> 00:02:40,555
the willingness

74
00:02:41,495 --> 00:02:43,655
to do something a little bit different. Yeah.

75
00:02:43,655 --> 00:02:45,574
And and coming to that trust, you really

76
00:02:45,574 --> 00:02:46,634
need sort of that,

77
00:02:47,655 --> 00:02:50,709
clinical validation. It's gotta be clinically sound. So

78
00:02:50,709 --> 00:02:51,909
how do you do that? How how can

79
00:02:51,909 --> 00:02:54,310
peer organizations really integrate these tools in a

80
00:02:54,310 --> 00:02:54,969
way that

81
00:02:55,509 --> 00:02:56,889
is equitable, transparent?

82
00:02:57,750 --> 00:02:59,349
And thinking from your experience as a as

83
00:02:59,349 --> 00:03:01,349
a clinician and also someone who who leads

84
00:03:01,349 --> 00:03:02,250
digital strategy,

85
00:03:02,834 --> 00:03:04,914
what are sort of your guiding principles or

86
00:03:04,914 --> 00:03:05,414
safeguards

87
00:03:05,715 --> 00:03:08,114
that are are essential to achieving that balance

88
00:03:08,114 --> 00:03:09,174
to making sure that

89
00:03:09,715 --> 00:03:11,794
AI is clinically sound and then you're coming

90
00:03:11,794 --> 00:03:13,655
to to that place of trust together?

91
00:03:13,955 --> 00:03:15,014
I think the first

92
00:03:15,330 --> 00:03:15,830
step

93
00:03:16,209 --> 00:03:17,430
is that intentionality.

94
00:03:18,050 --> 00:03:19,989
I think you have to be really thoughtful

95
00:03:20,129 --> 00:03:21,269
and intentional

96
00:03:22,050 --> 00:03:24,150
to ensure that we're embedding

97
00:03:24,530 --> 00:03:25,909
our human values

98
00:03:27,009 --> 00:03:30,150
of being patient centric. Right? Member centric.

99
00:03:30,764 --> 00:03:31,264
So

100
00:03:31,884 --> 00:03:32,704
at Lyric,

101
00:03:33,084 --> 00:03:36,305
our governance model is really a clinical governance

102
00:03:36,525 --> 00:03:37,025
first

103
00:03:37,324 --> 00:03:37,824
model.

104
00:03:38,764 --> 00:03:40,064
So we have multidisciplinary

105
00:03:40,685 --> 00:03:42,844
reviews. So we have clinicians. We have data

106
00:03:42,844 --> 00:03:45,719
scientists. We have legal. We have compliance. We

107
00:03:45,960 --> 00:03:49,180
really have just the whole gamut of people

108
00:03:49,639 --> 00:03:51,020
to give their input

109
00:03:51,400 --> 00:03:51,900
on

110
00:03:52,199 --> 00:03:52,699
what

111
00:03:53,159 --> 00:03:53,659
AI

112
00:03:54,680 --> 00:03:56,060
should be deployable

113
00:03:56,520 --> 00:03:59,659
before it's actually deployed and during deployment

114
00:03:59,960 --> 00:04:00,620
as well.

115
00:04:01,294 --> 00:04:03,474
And then we have continuous monitoring

116
00:04:03,854 --> 00:04:04,834
for equity

117
00:04:05,134 --> 00:04:07,614
and safety. Right? We hold our AI to

118
00:04:07,614 --> 00:04:09,474
really rigorous safety standards,

119
00:04:10,175 --> 00:04:10,675
and

120
00:04:11,215 --> 00:04:11,955
we require

121
00:04:12,495 --> 00:04:14,034
that we have this, like,

122
00:04:14,569 --> 00:04:16,430
subgroup performance checks,

123
00:04:16,889 --> 00:04:19,389
in real time governance to detect

124
00:04:19,769 --> 00:04:22,990
drift because the models can actually shift based

125
00:04:23,209 --> 00:04:24,889
on the data that's being,

126
00:04:25,290 --> 00:04:25,790
inputted.

127
00:04:26,410 --> 00:04:29,230
So and then finally, it's like that transparency.

128
00:04:30,705 --> 00:04:33,365
You can't build trust without the transparency.

129
00:04:34,464 --> 00:04:37,185
So the automated decisions, they have to be

130
00:04:37,185 --> 00:04:37,685
explainable,

131
00:04:38,305 --> 00:04:39,845
to clinicians and providers,

132
00:04:40,384 --> 00:04:41,285
to the people.

133
00:04:41,745 --> 00:04:42,064
So,

134
00:04:43,459 --> 00:04:44,819
you have to be able to back them

135
00:04:44,819 --> 00:04:47,699
up. Yeah. And that that transparency is a

136
00:04:47,699 --> 00:04:50,180
continuous act too. Just like as you're tweaking

137
00:04:50,180 --> 00:04:51,860
the models, there's no, like, set it and

138
00:04:51,860 --> 00:04:53,379
it's done. You know what I mean? Like,

139
00:04:53,379 --> 00:04:56,144
it's it's it's always constant work. It is

140
00:04:56,384 --> 00:04:59,604
constant work. Right? Because things are constantly changing,

141
00:04:59,664 --> 00:05:00,724
constantly evolving.

142
00:05:01,264 --> 00:05:02,944
I mean, look at medical evidence. When we

143
00:05:02,944 --> 00:05:06,004
talk about clinical, the medical evidence is constantly

144
00:05:06,144 --> 00:05:08,224
changing and evolving, so there is no set

145
00:05:08,224 --> 00:05:09,365
it and forget it.

146
00:05:09,740 --> 00:05:11,580
Yeah. And you could make the argument too

147
00:05:11,580 --> 00:05:13,660
that clinical evidence and medical evidence is just

148
00:05:13,740 --> 00:05:15,980
is gonna start evolving faster than what we've

149
00:05:15,980 --> 00:05:17,519
seen in the past. Absolutely.

150
00:05:17,819 --> 00:05:19,279
I mean, I think there's,

151
00:05:20,139 --> 00:05:22,459
a lot of evidence out there that says

152
00:05:22,540 --> 00:05:23,199
and I

153
00:05:23,754 --> 00:05:26,334
think someone quoted to me in 1950,

154
00:05:26,714 --> 00:05:29,194
the rate of doubling of medical literature is

155
00:05:29,194 --> 00:05:33,214
fifty years. And I think in 2024,

156
00:05:33,355 --> 00:05:35,295
it's something like sixty two days.

157
00:05:35,754 --> 00:05:37,295
So it's astonishing.

158
00:05:37,915 --> 00:05:38,879
So we have

159
00:05:39,920 --> 00:05:42,580
to be conscious of that and really intentional,

160
00:05:43,600 --> 00:05:45,680
about our models. Yeah. I I heard it

161
00:05:45,680 --> 00:05:46,720
put to me too. That's a I mean,

162
00:05:46,720 --> 00:05:49,520
that's that's a wild stat. Yeah. I also

163
00:05:49,520 --> 00:05:51,040
heard it put to me once that it's,

164
00:05:51,704 --> 00:05:54,824
you know, technology advancement innovation has never moved

165
00:05:54,824 --> 00:05:56,584
as fast as it's moving now, and it'll

166
00:05:56,584 --> 00:05:59,224
never move this slow again. Right. Yeah. I

167
00:05:59,224 --> 00:06:00,985
I have heard the same thing. Right? Like,

168
00:06:00,985 --> 00:06:02,824
this is the worst that it's gonna be.

169
00:06:02,824 --> 00:06:03,964
Yeah. So

170
00:06:04,584 --> 00:06:05,865
I don't know. That could be a little

171
00:06:05,865 --> 00:06:08,210
scary. Yeah. Yeah. That's a lot. It's a

172
00:06:08,210 --> 00:06:08,610
lot.

173
00:06:09,090 --> 00:06:11,090
But but let's let's come back to sort

174
00:06:11,090 --> 00:06:13,569
of thinking about your expertise and and what

175
00:06:13,569 --> 00:06:14,629
you're doing at Lyric.

176
00:06:15,090 --> 00:06:17,410
How does all this sort of fit into,

177
00:06:17,410 --> 00:06:19,970
I guess, helping health plans accelerate value based

178
00:06:19,970 --> 00:06:20,710
care goals,

179
00:06:21,330 --> 00:06:23,785
which are crucial to to getting to that

180
00:06:23,785 --> 00:06:24,605
payer provider,

181
00:06:25,865 --> 00:06:28,185
equanimity, I guess, to where things are sort

182
00:06:28,185 --> 00:06:29,865
of in harmony. Value based care is gonna

183
00:06:29,865 --> 00:06:31,625
have to flourish for that to happen. Right?

184
00:06:31,625 --> 00:06:34,024
Yeah. You know, I mean, that partnership is

185
00:06:34,024 --> 00:06:34,764
so important.

186
00:06:35,680 --> 00:06:36,740
I think that

187
00:06:37,120 --> 00:06:39,620
you can use AI, like, from that predictive

188
00:06:39,759 --> 00:06:40,899
modeling perspective,

189
00:06:41,599 --> 00:06:44,419
and you'll get better risk stratification or identification

190
00:06:44,720 --> 00:06:46,659
of at risk patients early.

191
00:06:47,039 --> 00:06:49,779
So you can really direct your clinical efforts

192
00:06:50,024 --> 00:06:52,125
to members who are gonna benefit the most

193
00:06:52,345 --> 00:06:54,205
and drive those better outcomes.

194
00:06:55,305 --> 00:06:56,764
I think from an administrative

195
00:06:57,225 --> 00:06:57,725
perspective,

196
00:06:58,105 --> 00:07:01,324
it can really decrease the administrative burden,

197
00:07:01,865 --> 00:07:03,564
which ultimately means

198
00:07:04,560 --> 00:07:06,339
providers, doctors, clinicians

199
00:07:06,639 --> 00:07:08,480
are gonna be spending more time with the

200
00:07:08,480 --> 00:07:10,720
patients. Right? And that's really what I think

201
00:07:10,720 --> 00:07:11,460
we want.

202
00:07:11,839 --> 00:07:12,339
So,

203
00:07:13,199 --> 00:07:15,779
automating those activities, it really frees,

204
00:07:16,160 --> 00:07:18,339
clinicians up to work top of license.

205
00:07:19,040 --> 00:07:19,699
And then

206
00:07:20,305 --> 00:07:24,004
shared measure measurement. Right? There's always been

207
00:07:24,865 --> 00:07:28,324
just that source of truth kind of friction,

208
00:07:28,464 --> 00:07:30,944
who has the right source of truth. And

209
00:07:30,944 --> 00:07:33,264
I think AI can really help align all

210
00:07:33,264 --> 00:07:34,004
the stakeholders,

211
00:07:34,579 --> 00:07:37,540
payers, providers, patients, by really integrating the data

212
00:07:37,540 --> 00:07:40,680
sources so you have one source of truth

213
00:07:40,819 --> 00:07:41,720
in a comprehensive

214
00:07:42,180 --> 00:07:43,319
view of outcomes.

215
00:07:44,019 --> 00:07:46,579
So data has always been really difficult to

216
00:07:46,579 --> 00:07:48,105
do with value based care, but I think

217
00:07:48,105 --> 00:07:50,985
there's a real opportunity here with AI to

218
00:07:50,985 --> 00:07:53,545
make the data far more accessible than it

219
00:07:53,545 --> 00:07:55,004
has been in the past,

220
00:07:55,785 --> 00:07:56,605
and transparent.

221
00:07:57,384 --> 00:08:00,129
So and then trust. Again, trust is at

222
00:08:00,129 --> 00:08:01,329
the center of all of it. That that

223
00:08:01,410 --> 00:08:03,089
that's where where my head went to. It

224
00:08:03,089 --> 00:08:04,949
comes back to, like, trust and transparency.

225
00:08:05,569 --> 00:08:07,730
You have the best technology in the world,

226
00:08:07,730 --> 00:08:09,670
but if you don't have trust and transparency

227
00:08:09,889 --> 00:08:11,730
sort of around it, it's not gonna not

228
00:08:11,730 --> 00:08:13,245
gonna do what it's supposed to do. Right?

229
00:08:13,324 --> 00:08:15,485
Right. Nobody's gonna use it. Right. Right. So

230
00:08:16,125 --> 00:08:17,964
So for folks listening out there who want

231
00:08:17,964 --> 00:08:19,904
to sort of, I guess, perhaps

232
00:08:20,444 --> 00:08:23,004
accelerate their their value based care goals and

233
00:08:23,004 --> 00:08:25,245
and put AI sort of at the center

234
00:08:25,245 --> 00:08:26,845
of that strategy, what's one step they could

235
00:08:26,845 --> 00:08:29,149
take? What advice would you give? You

236
00:08:29,529 --> 00:08:31,310
know, that's a tough one.

237
00:08:31,850 --> 00:08:32,350
But

238
00:08:32,649 --> 00:08:33,549
I would say,

239
00:08:34,009 --> 00:08:37,470
look at what's actually actionable. Right? Take something

240
00:08:38,009 --> 00:08:40,509
do a pilot on it. Partner with someone

241
00:08:40,570 --> 00:08:43,144
just on, you know, one aspect.

242
00:08:43,605 --> 00:08:44,745
You know, for example,

243
00:08:45,524 --> 00:08:47,625
here at Lyric, you know, we look at

244
00:08:47,684 --> 00:08:49,705
all sorts of different types of claims,

245
00:08:50,245 --> 00:08:52,904
look at a high cost inpatient DRG

246
00:08:53,605 --> 00:08:55,384
or coordination of benefits,

247
00:08:56,009 --> 00:08:58,730
Measure the hit rate. Look at provider appeals.

248
00:08:58,730 --> 00:09:00,670
Look at time to, you know, resolution.

249
00:09:02,009 --> 00:09:04,410
Grab some KPIs with just a short pilot

250
00:09:04,410 --> 00:09:06,350
and see if it works, and then iterate

251
00:09:06,410 --> 00:09:08,250
on that. Right? It's not gonna work the

252
00:09:08,250 --> 00:09:10,735
first time. It's just those multiple iterations and

253
00:09:10,735 --> 00:09:11,954
learning off the iterations,

254
00:09:12,815 --> 00:09:15,294
and I think that you'll find it really

255
00:09:15,294 --> 00:09:15,794
successful.

256
00:09:16,254 --> 00:09:18,334
That's, I think wonderful advice.

257
00:09:18,735 --> 00:09:20,334
You started at you said it was a

258
00:09:20,334 --> 00:09:21,774
tough question, but I think you landed the

259
00:09:21,774 --> 00:09:22,914
plane pretty good there.

260
00:09:23,419 --> 00:09:25,519
Mara, is there anything we didn't touch on

261
00:09:25,579 --> 00:09:27,500
or final thoughts you'd like to share? Perhaps

262
00:09:27,500 --> 00:09:29,579
it's something you wanna reemphasize for listeners. I'll

263
00:09:29,579 --> 00:09:31,019
leave it to you how how you wanna

264
00:09:31,019 --> 00:09:32,000
close out the podcast.

265
00:09:32,539 --> 00:09:35,339
Yeah. I think when we build AI with

266
00:09:35,339 --> 00:09:36,319
clinical intelligence,

267
00:09:36,779 --> 00:09:38,399
the governance, and the transparency,

268
00:09:39,235 --> 00:09:40,995
we move from a system that has a

269
00:09:40,995 --> 00:09:43,315
lot of friction in it to much better

270
00:09:43,315 --> 00:09:45,335
alignment across the system.

271
00:09:45,794 --> 00:09:46,934
It can help clinicians

272
00:09:48,115 --> 00:09:48,934
be clinicians.

273
00:09:49,634 --> 00:09:52,035
So it's a tool that we can use

274
00:09:52,035 --> 00:09:54,615
to improve outcomes for patients while

275
00:09:54,990 --> 00:09:55,490
really

276
00:09:55,950 --> 00:09:57,970
decreasing that administrative burden.

277
00:09:58,269 --> 00:10:00,029
Wonderful. Mara, thank you so much for being

278
00:10:00,029 --> 00:10:02,512
on the podcast. Thanks for having me. We

279
00:10:02,512 --> 00:10:05,392
also wanna thank our podcast sponsor, Lyric. Listeners,

280
00:10:05,392 --> 00:10:07,232
you can tune to more podcasts from Becker's

281
00:10:07,232 --> 00:10:11,332
Healthcare by visiting our podcast page at beckerspodcast.com.