Latest / Becker’s Payer Issues Podcast / Building Trustworthy AI in Healthcare Through Transparency

Transcript

Building Trustworthy AI in Healthcare Through Transparency · Becker’s Payer Issues Podcast

• 0:00
  - Hi, everyone. This is Brian Zimmerman with Becker's
• 0:02
  - Healthcare. Thank you so much for tuning into
• 0:04
  - the Becker's Healthcare podcast series. I'm excited to
• 0:07
  - welcome Halsey Wise to the podcast.
• 0:09
  - Halsey is the CEO of Lyric, a role
• 0:12
  - he stepped into after serving on the company's
• 0:15
  - board since 2022.
• 0:16
  - Over nearly three decades, he's led six companies
• 0:19
  - as a CEO or president and served on
• 0:21
  - corporate boards across health tech software and financial
• 0:25
  - services,
• 0:26
  - including board seats at Cerner and Cotivity.
• 0:29
  - At Lyric, he's focused on scaling the company's
• 0:31
  - investments in AI and health care decision intelligence
• 0:34
  - to bring greater trust, truth, and transparency to
• 0:37
  - the broader health care ecosystem.
• 0:39
  - Halsey, welcome to the podcast.
• 0:41
  - Hey, Brian. Thanks for having me. Appreciate being
• 0:44
  - here. Great to have you. So you've said
• 0:46
  - health care cannot improve without trust, truth, and
• 0:48
  - transparency as I as I noted in our
• 0:50
  - the the intro there.
• 0:52
  - After thirty years leading companies across the provider
• 0:54
  - EHR and payer sectors,
• 0:56
  - what convinced you that transparency is an operating
• 0:59
  - requirement rather than a nice to have?
• 1:02
  - Yeah. Good question. I think transparency may be,
• 1:05
  - mean something different to a lot of folks.
• 1:07
  - But for us, definitely, transparency is an operating
• 1:10
  - requirement. And after thirty years of being in
• 1:12
  - health care
• 1:13
  - as a human being, as a patient, and
• 1:15
  - and being in the provider world now helping
• 1:17
  - health plans and seeing a a wide aperture
• 1:20
  - in health care, I think a lot of
• 1:22
  - health care decisions
• 1:23
  - depend on logic that others need to understand
• 1:26
  - and defend.
• 1:27
  - And across the whole value chain, decisions about
• 1:29
  - eligibility,
• 1:30
  - coverage, payment, care, risk, other things like that
• 1:33
  - require a a very clear line
• 1:36
  - back to the underlying
• 1:38
  - policy data and reasoning.
• 1:40
  - I call that the decision lineage. The decision
• 1:43
  - lineage has to be,
• 1:45
  - available and trusted. And we're in a regulated
• 1:48
  - industry, and health care, as you know, is
• 1:50
  - a lattice work of rules, regulations,
• 1:54
  - complexity,
• 1:55
  - and there's a high cost of failure in
• 1:57
  - our industry. So when a decision
• 1:59
  - affects a claim,
• 2:01
  - a provider,
• 2:02
  - or a member experience, or a patient,
• 2:05
  - there's a super high cost of failure here.
• 2:08
  - And regulators,
• 2:09
  - clients, partners, and providers, and all expect that
• 2:12
  - they have a logic that's dependable and reliable
• 2:14
  - that they can follow it. So transparency is
• 2:17
  - not a feature layered on top of AI.
• 2:19
  - It's a prerequisite condition
• 2:21
  - for the adoption of AI.
• 2:23
  - And
• 2:24
  - at Lyric, we believe every recommendation needs to
• 2:26
  - be,
• 2:27
  - traceable,
• 2:28
  - auditable,
• 2:30
  - and attached to the policy data
• 2:32
  - and the source data from which it comes.
• 2:34
  - And that gives organizations not only transparency,
• 2:38
  - but the confidence on which to act upon
• 2:40
  - it. Yeah. The the decision lineage as you
• 2:42
  - as you put it, you've gotta be able
• 2:44
  - to to trace that back. And and can
• 2:46
  - we go even further here, which is you
• 2:49
  - really I guess, in practical terms, what separates
• 2:51
  - an AI system that is able to do
• 2:53
  - that, that that clinicians and executives can trust
• 2:56
  - from one that simply produces a very confident
• 2:59
  - answer?
• 3:00
  - And and why does that distinction become even
• 3:02
  - more important as AI advances? And it's gonna
• 3:05
  - continue to move closer to clinical and and
• 3:07
  - operational and financial decisions as as it becomes
• 3:10
  - more
• 3:11
  - more advanced. Why why does this distinction really
• 3:13
  - matter between an AI system
• 3:16
  - that folks can trust and one that, you
• 3:18
  - know, just looks nice?
• 3:19
  - Yeah. And that's something that's worth spending a
• 3:21
  - little time on.
• 3:23
  - Confidence is not the same as trust. You
• 3:25
  - know, a confidence answer and a trustworthy answer
• 3:28
  - can look identical or similar from the start,
• 3:31
  - but that's part of the problem. The difference
• 3:33
  - is really whether you can open a recommendation
• 3:35
  - and see the policy data and logic behind
• 3:38
  - it. You know, I would I would posit
• 3:40
  - that a confident sounding answer that turns out
• 3:44
  - to be wrong
• 3:45
  - is more dangerous
• 3:47
  - than having some uncertainty acknowledged at the beginning.
• 3:50
  - And so, you know, the health plan clients
• 3:52
  - that we work with and others across the
• 3:54
  - ecosystem,
• 3:55
  - trustworthiness,
• 3:57
  - you know, needs governance.
• 3:58
  - And so, you know, at Lyric, we call
• 4:00
  - it governed AI. It operates in a defined
• 4:03
  - architecture that combines
• 4:04
  - policy data
• 4:06
  - that deterministic
• 4:07
  - rules, a workflow,
• 4:09
  - intelligence, and clinical and payment data that's trusted
• 4:13
  - and proven at the source. So it's proven,
• 4:15
  - verifiable,
• 4:16
  - and trusted.
• 4:17
  - And so AI is not generating
• 4:19
  - an answer in any situation in isolation.
• 4:22
  - It's really structuring evidence and servicing the highest
• 4:26
  - value action
• 4:27
  - for,
• 4:28
  - a human in the loop to review, and
• 4:30
  - we call it augmenting human decisions. We don't
• 4:32
  - make any decisions on our own. We have
• 4:35
  - health care decision intelligence that pushes information
• 4:38
  - to our,
• 4:40
  - health plan clients for them to make the
• 4:42
  - most informed decision at their given their circumstances.
• 4:46
  - That distinction really moves AI closer to decisions
• 4:49
  - that affect payment and patient care, and that's
• 4:51
  - super important because,
• 4:53
  - you know, health care is now almost 20%
• 4:55
  - of GDP or $5,000,000,000,000
• 4:57
  - and 20% of that's waste. Every dollar that
• 5:00
  - waste is dollars that have a big opportunity
• 5:02
  - cost that could be redirected back to patient
• 5:05
  - care, and that's what we're trying to be
• 5:07
  - a part of. So a confident answer that
• 5:09
  - can't be traced is a liability.
• 5:11
  - A trustworthy answer is something a health plan
• 5:14
  - can rely on, act on, defend
• 5:16
  - to a patient, provider, or a regulator.
• 5:20
  - Yeah.
• 5:21
  - The and to your point too earlier, the
• 5:23
  - stakes here are high. Right? The stakes are
• 5:25
  - high in clinical care, but the stakes are
• 5:26
  - high financially too. Right? This is a these
• 5:28
  - are these big swings of money moving, and
• 5:30
  - those have implications for direct patient care and
• 5:33
  - the health of these organizations.
• 5:34
  - Correct?
• 5:36
  - A 100%.
• 5:37
  - And, you know, it just a lot of
• 5:38
  - times health care has been opaque in the
• 5:40
  - past.
• 5:41
  - Patients,
• 5:42
  - you know, you,
• 5:43
  - our family, my family, no one accepts a
• 5:46
  - black box answer. Physicians don't accept a black
• 5:48
  - box answer.
• 5:49
  - Health plans don't accept black box answers. You
• 5:52
  - have to have the traceability, the auditability,
• 5:55
  - the grounding in the context of the decision
• 5:58
  - so people can understand
• 6:00
  - where the model or AI
• 6:03
  - source the data that is used to augment
• 6:05
  - human decisioning.
• 6:07
  - Yeah. Yeah. I wanna get to something too
• 6:09
  - that's been
• 6:10
  - sort of a a a truth in the
• 6:12
  - industry for for a long time, which is
• 6:14
  - that
• 6:14
  - providers, payers, even technology companies as well are
• 6:18
  - really trying to attack many of the same
• 6:20
  - problems.
• 6:21
  - But but the challenges there is often the
• 6:23
  - incentives don't align, and and, of course, that
• 6:26
  - will create an environment of limited trust.
• 6:29
  - Where can AI help, I guess, here, sort
• 6:31
  - of help solve this perennial challenge in the
• 6:33
  - industry and create greater transparency between those groups?
• 6:37
  - And and I guess also the flip side
• 6:39
  - of that coin, where could it do the
• 6:40
  - opposite? Where could it if if leveraged inappropriately
• 6:44
  - perhaps, where could it have deepened that this
• 6:46
  - challenge worsen it?
• 6:48
  - Yeah. Brian, AI can make decisions easier to
• 6:51
  - understand.
• 6:52
  - The biggest opportunity some of the biggest opportunity
• 6:54
  - we see in workflows where providers, payers, and
• 6:57
  - other firms already interact, but they don't see
• 7:00
  - each other's reasoning or logic.
• 7:02
  - So payment and coverage decisions are maybe one
• 7:04
  - of the clearest examples.
• 7:06
  - You have AI showing the policy data and
• 7:09
  - logic behind a recommendation.
• 7:11
  - A provider then has something to respond to
• 7:14
  - instead of a black box denial. So we're
• 7:17
  - very overt. And, again, that traceability and the
• 7:19
  - logic and that I've used the word decisioning
• 7:22
  - decision lineage.
• 7:24
  - So it's
• 7:25
  - all transparent, available. There's sunlight on it. There's
• 7:28
  - no attempt to,
• 7:31
  - to hide or have any opacity in in
• 7:33
  - the process. It's all, you know, fully straightforward.
• 7:36
  - But an AI makes it helps, you know,
• 7:38
  - pick up pattern recognition,
• 7:40
  - anomaly detection,
• 7:42
  - and signals
• 7:43
  - that are otherwise
• 7:45
  - harder to
• 7:46
  - determine. But we
• 7:48
  - use that again to augment human decision making
• 7:50
  - and never making a decision only in an
• 7:52
  - autonomous manner. We just simply don't do that.
• 7:55
  - But the speed that AI
• 7:57
  - delivers
• 7:58
  - is really not as helpful without the true
• 8:00
  - governance because without the governance, the mistrust exists.
• 8:04
  - So,
• 8:05
  - AI has to be moved faster, but it
• 8:07
  - has to be explainable
• 8:08
  - and at scale. And so a human in
• 8:10
  - the loop design is critical,
• 8:12
  - and technology should inform judgment, not replace it.
• 8:16
  - And the people that are affected by a
• 8:18
  - decision, you, me,
• 8:19
  - all of us as patients need to understand
• 8:21
  - why a particular decision was made, and that's
• 8:24
  - what we do on behalf of, our health
• 8:26
  - plan clients and providers.
• 8:28
  - Yeah. And
• 8:29
  - showing how that decision was made is Right.
• 8:31
  - Is what is what can bring back that
• 8:34
  - trust and perhaps even do you do you
• 8:36
  - believe it's possible align incentives,
• 8:39
  - even more so?
• 8:41
  - Yeah. I mean, I think I think to
• 8:42
  - have it everyone wants to reduce waste in
• 8:44
  - the system. I don't think there's any entity
• 8:47
  - that that, you know, is pleased about the
• 8:49
  - amount of waste. And so I think everyone's
• 8:51
  - aligned to try to get the waste out
• 8:52
  - of the system and redirect it back at
• 8:54
  - patient care.
• 8:56
  - And and I think AI can be a
• 8:58
  - tool in that. And I think that the
• 9:01
  - there's such a significant amount of money that
• 9:03
  - could be re redirected to patient care given
• 9:06
  - the size, scale, and complexity of the problem.
• 9:09
  - Excellent. Well, closing here by maybe
• 9:11
  - reasserting some of the stakes involved in in
• 9:13
  - this transformation. So what is at stake if
• 9:16
  - health care gets trust and transparency and AI
• 9:19
  - wrong?
• 9:20
  - And then,
• 9:21
  - again, another flip side question, what could getting
• 9:24
  - it right make possible that's never been possible
• 9:26
  - in health care?
• 9:28
  - Yeah. I mean, there's a tremendous amount at
• 9:30
  - stake, you know. Right? Health care, as I
• 9:32
  - say, is an enormous
• 9:34
  - market. It's 20% of GDP and 25%
• 9:37
  - of that's waste. It's incredibly important
• 9:40
  - to get it right. If it gets it
• 9:41
  - wrong,
• 9:42
  - AI adoption stalls, regulators, and providers, and patients
• 9:46
  - don't accept decisions that they can't question
• 9:48
  - or they can't audit and have traceable data.
• 9:51
  - And worse, really, the industry risk entrenching the
• 9:54
  - estimated 300,000,000,000 in annual administrative waste that we're
• 9:58
  - already fighting.
• 9:59
  - And so AI can help us
• 10:01
  - address that, but if AI gets it wrong,
• 10:04
  - that deadweight loss in the system and the
• 10:07
  - lack of care is certainly a penalty that
• 10:10
  - our society shouldn't be willing to accept.
• 10:13
  - If health care really gets AI AI right,
• 10:14
  - we can improve the speed of decisions and
• 10:16
  - the accuracy of decisions at scale.
• 10:19
  - Decisions that once took days to review
• 10:22
  - and appeal can happen correctly on the first
• 10:24
  - time. We at Lyric are processing
• 10:26
  - over 10,000,000,000
• 10:28
  - annual claims a year, and we're doing it
• 10:30
  - in milliseconds.
• 10:32
  - So we provide deterministic
• 10:34
  - information that's traceable, auditable, and grounded in context.
• 10:38
  - And so AI can really provide the efficiency.
• 10:41
  - It's a foundation for an industry that spends
• 10:43
  - more resources
• 10:44
  - on care and and fewer litigating its own
• 10:47
  - decisions of the past. It's a super important
• 10:49
  - tool. It can help propel
• 10:52
  - degrees of freedom in patient care that were
• 10:54
  - never possible before.
• 10:57
  - Yeah. And and and to your point, getting
• 10:59
  - it wrong means a lot of that transformation,
• 11:02
  - the positive transformation stalls out. Right?
• 11:05
  - 100%.
• 11:06
  - 100%. And,
• 11:07
  - we use AI aggressively and confidently at the
• 11:10
  - request of our health plan clients, but we
• 11:13
  - also
• 11:14
  - use it, you know, as I said, in
• 11:16
  - a very responsible, ethical manner,
• 11:19
  - very focused on, again, providing transparency and traceability
• 11:23
  - so that there's this concept of trust, truth,
• 11:26
  - and transparency where we started is is earned
• 11:29
  - every day.
• 11:31
  - Well, Halsey, I think that's a perfect place
• 11:32
  - for us to end. Thank you so much
• 11:34
  - for coming on the podcast.
• 11:35
  - Hey, Brian. Thanks for having me. Appreciate it.
• 11:38
  - We also wanna thank our podcast sponsor, Lyric.
• 11:41
  - You can tune in to more podcasts from
• 11:43
  - Becker's Healthcare by visiting our podcast page at
• 11:46
  - beckershospitalreview.com.