Latest / Becker’s Healthcare Podcast / How Specialty Pharmacy Leaders Can Build Stronger Governance, Accountability, and Trust Around the Use of AI

Transcript

How Specialty Pharmacy Leaders Can Build Stronger Governance, Accountability, and Trust Around the Use of AI · Becker’s Healthcare Podcast

• 0:00
  - Hello, everyone. This is Erica Spicer Mason with
• 0:02
  - Becker's Healthcare. Thank you so much for tuning
• 0:05
  - in to the Becker's Healthcare podcast series.
• 0:08
  - So today, we're going to talk about how
• 0:10
  - specialty pharmacy leaders can build stronger governance,
• 0:13
  - accountability,
• 0:14
  - and trust around the use of AI.
• 0:16
  - And joining me for this conversation is Steven
• 0:19
  - McGee, the executive vice president and chief compliance
• 0:22
  - officer at SandsRx.
• 0:24
  - Steven, great to have you with us. Thank
• 0:26
  - you so much for joining. Thank Thank you
• 0:27
  - so much for having me. Happy to be
• 0:29
  - here. Oh, really glad to have you with
• 0:31
  - us. Before we get into it, can you
• 0:33
  - just share a little bit more about yourself
• 0:35
  - with our listeners and also how your organization
• 0:37
  - fits into the specialty pharmacy landscape?
• 0:41
  - Sure. So,
• 0:42
  - Steven, like she said, executive vice president,
• 0:46
  - been with that organization for almost ten years.
• 0:48
  - I've been working in the health care space
• 0:50
  - for over fifteen years in compliance, regulatory.
• 0:54
  - And as far as our organization, we're specialty
• 0:56
  - pharmacy, so
• 0:57
  - our world is centered around helping patients get
• 1:00
  - access to complex medications
• 1:02
  - and supporting them throughout therapy.
• 1:05
  - And specialty pharmacy is not just filling a
• 1:08
  - prescription. A lot of the work is in
• 1:10
  - the middle with benefits verifications,
• 1:13
  - prior authorizations,
• 1:14
  - co pay support,
• 1:15
  - refill coordination,
• 1:17
  - and most importantly, answering questions when,
• 1:20
  - answering the patients when they have questions or
• 1:22
  - they run into a problem.
• 1:24
  - Those patients are often dealing with serious conditions,
• 1:27
  - expensive therapies, and sometimes a lot of frustration
• 1:29
  - just trying to get started on a treatment.
• 1:32
  - And that is where I think for us
• 1:34
  - and for me, when it comes
• 1:37
  - to AI, we really started to see a
• 1:39
  - real opportunity here.
• 1:41
  - Yeah. Steven, really appreciate the overview. And sounds
• 1:44
  - like your organization is really helping folks navigate
• 1:46
  - that, I don't know, for lack of a
• 1:48
  - better term, the messy middle, in the specialty
• 1:51
  - pharmacy continuum. So
• 1:53
  - great to learn about that. And you mentioned
• 1:55
  - AI, so that's kind of where I wanna
• 1:57
  - go next. I know that SandsRx holds your
• 2:00
  - accreditation
• 2:01
  - for both specialty pharmacy and health care AI.
• 2:05
  - So once you begin the accreditation process,
• 2:08
  - what did it force you to change or
• 2:10
  - even rebuild
• 2:11
  - rather than just document? And I'm curious how
• 2:14
  - that surfaced anything that maybe you would have
• 2:16
  - missed otherwise.
• 2:19
  - Yeah. So,
• 2:20
  - one of the things you'll probably learn, is
• 2:22
  - that I I love a good analogy.
• 2:25
  - So,
• 2:26
  - if you were to think of the pharmacy
• 2:27
  - like a castle,
• 2:29
  - we already had the walls and the gates
• 2:31
  - and the rules for who comes in and
• 2:33
  - how things are handled.
• 2:35
  - That is what our existing specialty pharmacy accreditation
• 2:38
  - and our compliance structure gave us. But bringing
• 2:41
  - AI in was like bringing in a very
• 2:44
  - capable new guard. Now this guard can move
• 2:47
  - fast, handle a lot of traffic, and help
• 2:49
  - direct people where they need to go. The
• 2:51
  - problem is
• 2:53
  - before the accreditation process,
• 2:55
  - we had not fully thought through every part
• 2:57
  - of what that guard was allowed to do,
• 2:59
  - which doors can it open, which rooms can
• 3:01
  - it enter,
• 3:03
  - When does it have to stop and call
• 3:04
  - a pharmacist?
• 3:05
  - Who's watching the guard? What happens if it
• 3:07
  - gets confused and starts doing something differently
• 3:10
  - after an update?
• 3:11
  - So the process forced us to build out
• 3:14
  - inside of the castle, not just the outer
• 3:16
  - walls. So we had to
• 3:18
  - set clear rules for the AI, name who's
• 3:21
  - responsible for it, create a way to test
• 3:23
  - it before it goes live, monitor it after
• 3:25
  - it goes live, and then train our people
• 3:28
  - on how to use it and have a
• 3:29
  - plan for when something doesn't work the way
• 3:31
  - we expected it to. It also made us
• 3:33
  - put better rules around our vendors.
• 3:36
  - A vendor may tell you they have a
• 3:38
  - secure castle of their own. Maybe they have
• 3:40
  - a SOC two report and good data security
• 3:43
  - controls. But
• 3:44
  - once their AI is operating inside our castle,
• 3:48
  - we still have to know exactly what is
• 3:50
  - what it's doing with our patients, our workflows,
• 3:53
  - and our information.
• 3:54
  - And the accreditation helped us build a guardrail
• 3:57
  - so that AI could be helpful inside the
• 3:59
  - castle without being handled these master keys to
• 4:02
  - every single room.
• 4:04
  - Yeah. Steven, I so appreciate the castle metaphor.
• 4:08
  - It really helps to, I don't know, just
• 4:10
  - paint this visual picture of
• 4:12
  - what this looks like for you and how
• 4:14
  - you're keeping everything in check. And,
• 4:17
  - you know, it sounds like a very governance
• 4:19
  - minded kind of framework that you're going by
• 4:21
  - here.
• 4:22
  - And I noticed that in your your response,
• 4:24
  - you had also mentioned about kind of planning
• 4:26
  - for the unexpected. I know a lot of
• 4:28
  - organizations right now are are piloting AI,
• 4:32
  - looking at how to expand it, but, of
• 4:34
  - course, unexpected things happen with with new tools.
• 4:37
  - So when it comes to AI, can you
• 4:39
  - walk us through a time that maybe
• 4:41
  - when maybe something didn't perform as you
• 4:45
  - curious how you caught it and maybe how
• 4:47
  - quickly you're able to change course as a
• 4:49
  - result.
• 4:50
  - In our day to day operation, we haven't
• 4:52
  - had an issue where that's occurred just yet,
• 4:54
  - but
• 4:55
  - I think the more honest version or answer
• 4:57
  - is in early testing and implementation, you really
• 5:00
  - do learn quickly that a system can be
• 5:03
  - technically functional
• 5:04
  - but still miss the real life context of
• 5:06
  - a pharmacy call. For example,
• 5:09
  - a a caller may start with a simple
• 5:11
  - refill or prescription status question. But once you
• 5:15
  - listen closely,
• 5:16
  - the issue may really be a payment barrier
• 5:19
  - or delivery problem or clinical question or something
• 5:23
  - that needs a staff member immediately.
• 5:25
  - The system may have correctly identified the first
• 5:28
  - thing the caller said, but not the actual
• 5:30
  - need behind the call. And that is why
• 5:32
  - you
• 5:33
  - you you have to build and test escalation
• 5:35
  - pathways,
• 5:36
  - review call flows, keep the staff feedback in
• 5:39
  - the loop. And if we see a pattern
• 5:41
  - that says a type of call is not
• 5:42
  - being handled the way we intended, we don't
• 5:44
  - just leave it there. We review the calls,
• 5:47
  - figure out where the workflow broke down, update
• 5:49
  - the routing of the language, test the change,
• 5:51
  - and then monitor it again.
• 5:53
  - The goal is to catch the problem early
• 5:56
  - and make a course correction before it becomes
• 5:59
  - a bigger issue.
• 6:00
  - In our experience, those adjustments can be made
• 6:03
  - quickly once the issue is identified because we
• 6:05
  - have direct communication with the partner and a
• 6:08
  - documented process for evaluating and updating the workflow.
• 6:12
  - This is why I truly believe having a
• 6:14
  - strong, trusted relationship with your vendor is imperative.
• 6:17
  - And our relationship and work order includes that
• 6:20
  - workflow mapping, the established escalation criteria,
• 6:24
  - and pathways and testing and feedback before going
• 6:27
  - live with ongoing support and regular meetings with
• 6:30
  - the staff.
• 6:32
  - Yeah. I I'm glad that you touched on
• 6:34
  - this concept of trust, Steven, because that was
• 6:37
  - something I had also wanted to ask about
• 6:39
  - in the same vein of okay. So perhaps
• 6:42
  - there's a mistake made. You said that your
• 6:44
  - goal is to catch the problem early and
• 6:46
  - be able to correct,
• 6:48
  - which is great.
• 6:49
  - And at the same time, I know for
• 6:51
  - clinicians and patient facing staff, trust in AI
• 6:54
  - tools can definitely be fragile, especially if they
• 6:56
  - see a tool get something wrong.
• 6:59
  - So how do you build confidence among your
• 7:02
  - clinical staff, especially pharmacists?
• 7:05
  - Yeah. I can tell you how you don't
• 7:07
  - do it. You certainly don't rebuild trust by
• 7:09
  - telling the people that the AI is fine.
• 7:11
  - You rebuild it by showing them that they
• 7:12
  - have a voice, that the issue was taken
• 7:14
  - seriously, and that the system actually changed because
• 7:18
  - of what they saw.
• 7:19
  - If a pharmacist or a team member flagged
• 7:21
  - something,
• 7:22
  - we want them to know that there's a
• 7:23
  - cliff clear path to report it, have it
• 7:26
  - reviewed, and see what happened next.
• 7:28
  - We review the issue. We determine,
• 7:31
  - whether it was a training issue or workflow
• 7:33
  - issue or vendor issue or system limitation, and
• 7:36
  - then we put the right correction in place.
• 7:39
  - I also think it's important to be honest
• 7:40
  - about limitations. The the goal isn't to convince
• 7:43
  - the staff that the AI is perfect. It
• 7:45
  - isn't. It never is. It will never be
• 7:47
  - perfect. But the goal is to show them
• 7:49
  - that the AI is a tool that they
• 7:50
  - can trust within its defined role
• 7:53
  - and that when it's outside the role, the
• 7:55
  - human being still has the authority to take
• 7:57
  - over.
• 7:58
  - That is what makes people comfortable using it.
• 8:01
  - Mhmm. Yeah. And especially one that I'm sure
• 8:04
  - is critical as we're seeing AI capabilities
• 8:08
  - advance so quickly.
• 8:10
  - I would say arguably much more quickly than
• 8:12
  - any other technology we've seen evolve before.
• 8:16
  - And capabilities are moving faster than most governance
• 8:19
  - cycles can keep up with. So
• 8:21
  - how are you deciding what's worth adopting and
• 8:24
  - and what you let pass?
• 8:26
  - I'd say we try to start with the
• 8:28
  - problem, not the technology.
• 8:30
  - We don't ask where can we use AI?
• 8:33
  - We say,
• 8:34
  - where do patients or staff consistently
• 8:36
  - experience friction, delay,
• 8:38
  - or repetitive work that doesn't require a pharmacist
• 8:41
  - judgment?
• 8:42
  - I kinda go back to Uber. Uber didn't
• 8:44
  - start because taxi cab service was perfect. It
• 8:47
  - happened because a guy could not catch a
• 8:49
  - cab in San Francisco and got tired of
• 8:51
  - it. So then we look at four things.
• 8:53
  - First, is there a real operational or patient
• 8:55
  - experience problem to solve?
• 8:57
  - Two, is that AI actually the right tool,
• 8:59
  - or are we trying to use technology where
• 9:01
  - a simpler process
• 9:03
  - fix would work better?
• 9:04
  - Three, what is the risk level, especially for
• 9:07
  - patients, clinical staff, privacy, compliance, and the business?
• 9:11
  - And then finally, can we put the right
• 9:13
  - guardrails around it, including the testing and the
• 9:15
  - human escalation training and vendor accountability?
• 9:19
  - Now if the use case is high risk
• 9:20
  - or unclear or requires the AI to make
• 9:23
  - a final clinical decision, we're probably gonna go
• 9:25
  - ahead and let that pass. We're not gonna
• 9:27
  - move forward with it. But if it takes
• 9:28
  - repetitive work off the team and improves access,
• 9:31
  - stays within a clearly defined lane, and lets
• 9:34
  - humans step in when necessary,
• 9:37
  - that's where it becomes interesting.
• 9:40
  - And before implementation
• 9:41
  - or any material change,
• 9:43
  - the accreditation requires an impact analysis addressing potential
• 9:47
  - benefits and risks to the patients or the
• 9:49
  - practitioners
• 9:50
  - and finances
• 9:52
  - with strategies to reduce or eliminate any of
• 9:54
  - those risks.
• 9:55
  - So we look for AI that acts like
• 9:58
  - a good assistant,
• 9:59
  - not a replacement pharmacist. If it helps remove
• 10:02
  - that friction and it gets the patients to
• 10:03
  - the right person faster and it stays inside
• 10:05
  - the clear guardrails,
• 10:07
  - we will consider it. If it's trying to
• 10:08
  - make those clinical decisions on its own or
• 10:10
  - creating more confusion than it removes or it
• 10:13
  - can't be monitored or governed,
• 10:15
  - we're gonna go ahead and move on.
• 10:17
  - Yeah. Really pragmatic approach that you've outlined there.
• 10:21
  - And, you know, I you mentioned accreditation again,
• 10:24
  - Steven, and I do wanna circle our conversation
• 10:27
  - back to that
• 10:28
  - because we know that accreditation
• 10:30
  - means that you also have to hold vendors
• 10:32
  - to the same standards that you're setting internally.
• 10:35
  - So when a vendor updates a tool or
• 10:38
  - changes the underlying model, what do you expect
• 10:41
  - that vendor to disclose, and what should other
• 10:43
  - specialty pharmacy leaders require from their AI partners?
• 10:47
  - Yeah. My view is simple. A vendor if
• 10:50
  - a vendor changes anything that could affect patient
• 10:53
  - interaction, workflow, performance, privacy, or risk, We need
• 10:57
  - to know about it before or at least
• 10:59
  - as early as possible after
• 11:01
  - it affects our operation.
• 11:03
  - Disclosure to me is very similar to prenuptial
• 11:05
  - agreement. We are putting all of our cards
• 11:07
  - out on the table and hoping that our
• 11:08
  - partnership doesn't end. But if it does, we
• 11:11
  - need to have a clear understanding of what's
• 11:12
  - expected and what's not. And we expect the
• 11:15
  - vendor to explain what's changed, why it changed,
• 11:17
  - which workflows or patient populations may be affected,
• 11:20
  - whether the intended use or limitations have changed,
• 11:23
  - what
• 11:24
  - testing was done, what are the known risks,
• 11:27
  - and and what we need to do on
• 11:30
  - our side before the update goes live. If
• 11:33
  - the change is material, we should have the
• 11:34
  - ability to test it in our environment and
• 11:37
  - make sure our staff scripts and escalation paths
• 11:39
  - and communications are still appropriate.
• 11:41
  - We also expect transparency
• 11:43
  - around incidents and uptime and errors,
• 11:46
  - data handling, and any security or privacy event.
• 11:50
  - A vendor relationship cannot be, hey. Here's the
• 11:53
  - technology. Good luck.
• 11:55
  - If they are part of a patient facing
• 11:56
  - a pharmacy workflow, they are part of our
• 11:59
  - risk environment now.
• 12:01
  - And as far as what I would say,
• 12:02
  - especially pharmacy leadership require, I would tell
• 12:06
  - them to acquire more than a sales deck
• 12:07
  - or a SOC two and a statement that
• 12:10
  - the vendor is HIPAA compliant. Those are important,
• 12:12
  - but they are not the whole story.
• 12:15
  - Ask the vendor what exactly will the AI
• 12:18
  - do and what will it never do? Where
• 12:20
  - is the human handoff? How do you test
• 12:22
  - it in workflows like ours? How do you
• 12:24
  - monitor for wrong answers, missed calls, drift, bias,
• 12:28
  - changes in performance?
• 12:30
  - And even then, I didn't something I didn't
• 12:32
  - know about drift. I was like, because I
• 12:34
  - saw it. I'm like, what in the world
• 12:35
  - is drift? And I learned that it means
• 12:37
  - that that's when AI gradually starts performing differently
• 12:41
  - from how it performed when you first tested
• 12:43
  - and you approved it.
• 12:45
  - And
• 12:45
  - yet another analogy, and I apologize. But
• 12:48
  - imagine it's 2003
• 12:50
  - and you bought a Magellan GPS and it's
• 12:53
  - sitting in your car, and it works perfectly.
• 12:55
  - But
• 12:56
  - over time, roads close, new roads are built,
• 12:59
  - traffic patterns change, and the directions become less
• 13:02
  - reliable unless the maps are updated and checked.
• 13:05
  - So the GPS isn't not necessarily broken. The
• 13:07
  - world around it has changed. And if you
• 13:09
  - convert that into the pharmacy,
• 13:12
  - okay, it happens with drift where patients begin
• 13:15
  - asking different types of questions than the system
• 13:17
  - was trained around or a new medication or
• 13:19
  - a payroll or workflow or a program requirement
• 13:23
  - changes what the actual correct answer should be.
• 13:26
  - So that's just one example. And I would
• 13:30
  - make sure that whatever we're putting in that
• 13:32
  - contract or whatever we're saying is inside the
• 13:35
  - contract. Governance can't just be something that the
• 13:38
  - vendor says in a meeting. It needs to
• 13:40
  - be documented, measurable,
• 13:42
  - reviewed, and enforceable.
• 13:43
  - That's how you protect the patient, the pharmacy,
• 13:46
  - and the partnership. So I would say the,
• 13:48
  - like, the agreement that we have with our
• 13:50
  - great partner,
• 13:51
  - it assigns us as the final authority over
• 13:53
  - clinical determinations and the content of patient communications.
• 13:56
  - It requires a structured performance review and documented
• 13:59
  - actions. It includes vendor obligations around HIPAA HIPAA
• 14:03
  - safeguards,
• 14:04
  - minimum necessary,
• 14:05
  - PHI,
• 14:06
  - subcontractor restrictions, breach notification.
• 14:10
  - So for us, the value of the accreditation
• 14:12
  - wasn't just
• 14:13
  - being able to say we have a credential.
• 14:16
  - It made us better at asking questions, better
• 14:18
  - at defining risks, and better at making sure
• 14:20
  - AI serves the pharmacy team and the patient,
• 14:23
  - not the other way around.
• 14:25
  - So well said, Steven. And, again, I appreciate
• 14:28
  - how clearly you've outlined what governance, accountability, and
• 14:32
  - trust is looking like for you at Sanza
• 14:34
  - SandsRx
• 14:35
  - and how your organization is using AI. It's
• 14:37
  - been great learning about, you know, how that
• 14:39
  - all bolts back to some of these accountability
• 14:42
  - and accreditation frameworks that you've built and,
• 14:45
  - and, of course, appreciate the many metaphors, whether
• 14:48
  - it was castles or prenuptials and GPS,
• 14:50
  - all helpful to get our minds around this
• 14:53
  - concept. So, Steven, thank you again for your
• 14:55
  - time.
• 14:56
  - Thanks you thanks so much for having me.
• 14:57
  - I really appreciate it.
• 14:59
  - Oh, we appreciated having you. And we also
• 15:01
  - wanna thank our podcast sponsor for today, Yurak.
• 15:05
  - Listeners, be sure to tune in to more
• 15:06
  - podcasts from Becker's by visiting our podcast page
• 15:09
  - at beckershospitalreview.com.