Latest / Pharmacy Podcast Network / Talking AI and the Evolution of Personalized Medicine | NASP Specialty Pharmacy Podcast
Transcript
- 0:06Speaker 1: You are listening to the Pharmacy Podcast Network.
- 0:22Speaker 2: You're listening to the NASP podcast. This Specialty Pharmacy podcast
- 0:27Speaker 2: is a collaboration with the National Association of Specialty Pharmacy
- 0:31Speaker 2: and the Pharmacy Podcast Network. The mission of the National
- 0:34Speaker 2: Association of Specialty Pharmacy is to improve specialty pharmacy practice
- 0:39Speaker 2: by promoting continuing professional education and certification of specialty pharmacists
- 0:44Speaker 2: while advocating for public policies that ensure patient access to
- 0:48Speaker 2: specialty medications. As the healthcare industry's leading podcast dedicated to
- 0:53Speaker 2: the pharmacy profession, the Pharmacy Podcast Network is proud to
- 0:56Speaker 2: bring our listeners the NASP podcast collaboration with the National
- 1:01Speaker 2: Association of Specialty Pharmacy.
- 1:03Speaker 3: Hello, my name is Sheila Arquette and I'm the President
- 1:06Speaker 3: and CEO of the National Association of Specialty Pharmacy. I
- 1:09Speaker 3: am delighted to welcome a very special guest, Stuart Caman,
- 1:12Speaker 3: to this episode of the NSP podcast. Stuart is a
- 1:16Speaker 3: healthcare analytics specialist. As SVP of Analytics and Innovation, he
- 1:21Speaker 3: manages the strategic direction and deployment of the claratas RX
- 1:25Speaker 3: technical platform to support launches for innovative specialty medicines. He
- 1:29Speaker 3: built his expertise in specialty channel data through his previous
- 1:33Speaker 3: work in life cycle pricing and contract strategy at genen
- 1:37Speaker 3: Tech and in global consulting with IMS Health. Stuart holds
- 1:42Speaker 3: dual MBA and MPH degrees from UC Berkeley and a
- 1:46Speaker 3: BS in psycho biology from UCLA. Stuart, thanks so much
- 1:50Speaker 3: for joining me today.
- 1:51Speaker 4: It's great to be here, Sheila.
- 1:53Speaker 1: So let's just jump right in.
- 1:54Speaker 5: So, Stuart, let's talk a little bit about this evolution
- 1:59Speaker 5: of personal eyed medicine and the impact that it's had
- 2:02Speaker 5: on market and patient access teams right their.
- 2:04Speaker 1: Ability to catch up.
- 2:05Speaker 3: How has that been an opportunity for Clara TASRX to
- 2:10Speaker 3: use AI and.
- 2:11Speaker 1: Machine learning to assist these folks.
- 2:13Speaker 4: Well, that is a big question, Sheila, thank you for
- 2:17Speaker 4: asking it. I when I think about this, I'd like
- 2:22Speaker 4: to come back to the beginning of that question, which
- 2:25Speaker 4: is to look at the evolution of personalized medicine. So
- 2:30Speaker 4: a few decades ago, it was commonplace to have broadband,
- 2:36Speaker 4: small molecule retail based therapy pills essentially for millions of patients,
- 2:46Speaker 4: and everybody accepted that there would be this bell curve
- 2:51Speaker 4: of response, some patients would do pretty well, many patients
- 2:57Speaker 4: would do okay, then some patients would do very well.
- 3:02Speaker 4: And as we come into the twenty first century and
- 3:07Speaker 4: the genome has been sequenced and significant developments have occurred
- 3:14Speaker 4: around protein synthesis and development, we've seen the advent of
- 3:21Speaker 4: atologus cell therapy and gene therapy come into the fold.
- 3:26Speaker 4: We've watched clinical practice become more and more and more
- 3:30Speaker 4: precision and in some cases truly personalized. And so ten
- 3:36Speaker 4: or fifteen years ago, teams of biopharmaceutical manufacturers and partners
- 3:45Speaker 4: like specialty pharmacies and other patient support partners would think
- 3:51Speaker 4: about patient populations of several tens of thousands of patients
- 3:56Speaker 4: who might need a monoclonal antibody or an inflammatory condition,
- 4:02Speaker 4: or on polytics for particular types of cancers. And we've
- 4:09Speaker 4: continued to watch that journey of personalization progress to the
- 4:16Speaker 4: point where now it is truly best practice to be
- 4:19Speaker 4: sequencing patients' genomes, to be taking labs, to be developing
- 4:24Speaker 4: a targeted approach for them, where eligible patient populations may
- 4:30Speaker 4: now be five thousand people instead of fifty thousand or
- 4:35Speaker 4: five million people. And as the patient populations for these
- 4:42Speaker 4: different therapies have grown more and more and more targeted
- 4:46Speaker 4: life sciences companies have needed to support each patient with
- 4:53Speaker 4: more and more and more services to make sure that
- 4:55Speaker 4: those patients have exactly the right experience on their medicine. Now,
- 5:03Speaker 4: for us at claritaser X, you know, we see an
- 5:08Speaker 4: opportunity there to support that broader mission of enabling the
- 5:13Speaker 4: patient to have the best experience by using data and
- 5:17Speaker 4: by using analytics, and by using increasingly AI. Because it's
- 5:23Speaker 4: one thing to say, well, of course i want a
- 5:28Speaker 4: patient to have the best experience. I'm going to higher hub,
- 5:33Speaker 4: I'm going to deploy patient services. I'm going to have
- 5:35Speaker 4: a limited distribution of specialty pharmacies. They are all going
- 5:38Speaker 4: to be experts in their field. I'm going to bring
- 5:40Speaker 4: on a copay affordability solution and a digital patient engagement solution,
- 5:46Speaker 4: and I'm going to have a field team of access
- 5:49Speaker 4: professionals and nurses who are all going to help the
- 5:54Speaker 4: practice understand how to navigate the prior authorization landscape. That's
- 5:59Speaker 4: a lot of moving pieces now that have to be
- 6:03Speaker 4: coordinated around every single patient, and not every patient needs
- 6:09Speaker 4: the same combination of all of those services, and so
- 6:14Speaker 4: Our belief is that patient services is like anything else.
- 6:18Speaker 4: You can approach it in a data driven way, in
- 6:20Speaker 4: a scientific way, measure what works, and use information to
- 6:29Speaker 4: better guide the deployment of these services so that patients
- 6:34Speaker 4: can can truly get the best experience from their medicine.
- 6:40Speaker 4: That's where we come in. So I'm happy to talk
- 6:42Speaker 4: more about sort of what we're seeing around the use
- 6:46Speaker 4: of technology and machine learning and AI in the space,
- 6:48Speaker 4: but it does ultimately come from this evolution of medicine
- 6:53Speaker 4: itself towards a more personalized approach, right, And.
- 6:57Speaker 3: As you're talking through that shirt, I am kind of
- 6:59Speaker 3: thinking about everything that we learned right when we went
- 7:03Speaker 3: down this road of population health, right based like that
- 7:09Speaker 3: type of management approach. And so now as you're talking,
- 7:12Speaker 3: I'm thinking, you know, everything that we learned we don't
- 7:15Speaker 3: have to kind of toss aside. But now the definition
- 7:18Speaker 3: of population has changed, right Like where you said used
- 7:21Speaker 3: to be you know, large large numbers of patients. Now
- 7:23Speaker 3: a population could be one hundred patients.
- 7:26Speaker 4: Perhaps that's right, And in fact, each one of those
- 7:32Speaker 4: patients now has one hundred times or ten thousand times
- 7:37Speaker 4: as much data associated with them as every patient did
- 7:42Speaker 4: thirty years ago in the broadband scenario, and of course
- 7:46Speaker 4: that information is sensitive. There are important privacy considerations to
- 7:52Speaker 4: be very mindful of when working with that data. But
- 7:56Speaker 4: it also creates an opportunity to truly unders stand the
- 8:01Speaker 4: patient experience in a way that just wasn't possible twenty
- 8:04Speaker 4: years ago.
- 8:06Speaker 3: Right, And as you better understand, like you had kind
- 8:08Speaker 3: of talked in the beginning about, right, there was that
- 8:11Speaker 3: subset of patients that didn't respond to what we did
- 8:14Speaker 3: in the past.
- 8:15Speaker 1: So now as you know more about them.
- 8:17Speaker 3: As an individual and their individual needs and their responses
- 8:21Speaker 3: to therapy, is it possible not to switch to make
- 8:25Speaker 3: sure that they do have a positive income. Right, You're
- 8:28Speaker 3: able to make those changes and pivot when necessary so
- 8:31Speaker 3: that there isn't a treatment failure.
- 8:33Speaker 4: That's exactly right. And not only is there that opportunity
- 8:37Speaker 4: from clinical perspective to understand the patient's genetic makeup to
- 8:43Speaker 4: configure their treatment protocol to be aligned with best practices
- 8:48Speaker 4: for their genotype their sub phenotype, but you can do
- 8:54Speaker 4: the same thing on the commercial and patient services side
- 8:59Speaker 4: as well. Can look at the degree to which a
- 9:03Speaker 4: patient is responding to pharmacy programs or responding to a
- 9:08Speaker 4: cadence of phone calls, or responding to a digital patient
- 9:14Speaker 4: engagement app or the content that is served to them,
- 9:18Speaker 4: and by measuring ideally in a sort of robust, controlled way,
- 9:26Speaker 4: the patient's engagement with all of those various methods of outreach,
- 9:32Speaker 4: you can tune those various levers and optimize them and
- 9:38Speaker 4: watch as a patient's experience gets better. So you don't
- 9:41Speaker 4: even need to wait for the patient to have a
- 9:46Speaker 4: negative reaction to a services approach. You can constantly optimize
- 9:57Speaker 4: and reconfigure what they're getting in real time to really
- 10:02Speaker 4: do the best by every patient in a personalized way.
- 10:07Speaker 1: All right, So let's talk more about that.
- 10:08Speaker 3: So how how does Claritas use right aim machine learning
- 10:13Speaker 3: to positively impact that journey?
- 10:18Speaker 1: Right in that outcome?
- 10:19Speaker 4: It starts with data and pulling data together because just
- 10:25Speaker 4: the same way that the patient journey is a perhaps nonlinear,
- 10:31Speaker 4: but continuous and longitudinal process, we need to pull together
- 10:36Speaker 4: many different types of data in order to mirror that
- 10:40Speaker 4: longitudinal journey. So it's working with specialty pharmacies and patient
- 10:45Speaker 4: call centers and copay card solutions and digital patient engagement
- 10:48Speaker 4: apps and fuel access teams. It's all of the various
- 10:51Speaker 4: services that are wrapped around the patient produce information, Connecting
- 10:57Speaker 4: all of that at a patient level in a compliant way,
- 11:01Speaker 4: cleansing the data, shaping it is a big part of
- 11:06Speaker 4: creating something that can be fed into machine learning models
- 11:10Speaker 4: AI models. So that's the first step, and we have
- 11:16Speaker 4: some just delightful stories of where we have done that
- 11:22Speaker 4: for programs where we're consuming tens of thousands of data
- 11:27Speaker 4: elements about every single patient every fifteen minutes from the
- 11:32Speaker 4: whole ecosystem of partners and then using that to drive
- 11:37Speaker 4: real time visibility into what the patient is going through.
- 11:41Speaker 4: So once we have all that data pull together, then
- 11:45Speaker 4: the question is, well, how do we not only understand
- 11:49Speaker 4: where the patient is and what barriers are they running into?
- 11:54Speaker 4: Now that's certainly the first step, but also can wedict
- 12:01Speaker 4: what barriers the patient will run into in the near future.
- 12:05Speaker 3: Not just talking about clinical barriers, right, We're talking about
- 12:09Speaker 3: anything that could impact the patient's right ability to be
- 12:13Speaker 3: compliant and adherent, take their medication on time, manage side effects.
- 12:17Speaker 4: Right.
- 12:18Speaker 1: It's not just the medication itself.
- 12:20Speaker 4: That's absolutely right, and for us, it starts even upstream
- 12:25Speaker 4: of compliance and persistency. We in the data, see a
- 12:29Speaker 4: lot of patients who are prescribed these therapies who never
- 12:32Speaker 4: even start. They don't start because their practice runs into
- 12:37Speaker 4: complexity around prior authorizations, or they don't start because they
- 12:41Speaker 4: don't realize they have access to a commercial affordability solution,
- 12:46Speaker 4: or they don't start because somebody keeps trying to call
- 12:49Speaker 4: their cell phone and they see an eight hundred number
- 12:51Speaker 4: come up and they don't know who it is, so
- 12:53Speaker 4: they don't answer it. So there's lots of reasons why
- 12:58Speaker 4: patients never even start therapy. Is that their physicians prescribe
- 13:01Speaker 4: to them. And so that's an important piece of where
- 13:07Speaker 4: we can not only detect what's happening in the moment,
- 13:11Speaker 4: but predict where those barriers will show up that might
- 13:14Speaker 4: even prevent them from starting in the first place. And
- 13:16Speaker 4: then to your point, we want them to stay on
- 13:18Speaker 4: treatment as long as it is appropriate, and so we
- 13:20Speaker 4: can also forecast at a patient level which patients are
- 13:26Speaker 4: likely to run into challenges. It may disproportionately cause them
- 13:31Speaker 4: to drop off of therapy prematurely. And you know, sometimes
- 13:38Speaker 4: these events are pretty obvious. If you have a prior
- 13:45Speaker 4: authorization that's denied, that creates complexity. If you have a
- 13:50Speaker 4: very high copay, that creates complexity for you as a patient.
- 13:54Speaker 4: But sometimes the risks for patients initiating treatment and persisting
- 14:02Speaker 4: are actually a confluence of lots of little factors that
- 14:06Speaker 4: any and of themselves wouldn't be a cause for concern,
- 14:10Speaker 4: but when you put them all together, they create this confluence,
- 14:15Speaker 4: this headwind that does create material risk for the patient.
- 14:20Speaker 4: And it's that web of weak barriers that are creating
- 14:27Speaker 4: hardship for patients, where machine learning and AI can be
- 14:31Speaker 4: a great tool for figuring that out before they combine
- 14:37Speaker 4: and cause a patient too to stop their journey.
- 14:42Speaker 3: So I could see tremendous utility here for physicians, right,
- 14:46Speaker 3: or prescribers as well, especially those in shared risk arrangements.
- 14:49Speaker 4: Right.
- 14:49Speaker 1: If you could, if you could bring.
- 14:51Speaker 3: Back some of this information and say, you know, for
- 14:54Speaker 3: this type of you know, diagnosis, or perhaps a patient
- 14:57Speaker 3: with you know, these.
- 15:00Speaker 1: Like confounding variables, right.
- 15:02Speaker 3: This is the information that you need to know right
- 15:06Speaker 3: right from the get go, right when you're selecting a drug,
- 15:08Speaker 3: or you're talking to them about their therapy, or you're
- 15:11Speaker 3: you know, you're you're putting together a treatment plan.
- 15:13Speaker 1: I would think that this would be incredibly helpful.
- 15:16Speaker 4: I think that that's really a stute Sheila. You know,
- 15:19Speaker 4: it's funny, it's something we probably don't talk about as
- 15:23Speaker 4: much as we could, But the experience of the physician
- 15:28Speaker 4: and the practice is really correlated with the outcome of
- 15:33Speaker 4: the patient. And you know that's intuitive at some level
- 15:36Speaker 4: because many times the practice is the patient's gateway into
- 15:39Speaker 4: the healthcare system. But you see it in the numbers
- 15:44Speaker 4: all the time. You know, a physician that is more
- 15:48Speaker 4: experienced with a brand, that understands the brand's clinical profile,
- 15:52Speaker 4: understands the support services that are available for the brand,
- 15:57Speaker 4: that doctor's patients are going to be more likely to
- 16:02Speaker 4: initiate therapy and stay on as appropriate. A physician that
- 16:08Speaker 4: maybe hasn't figured out as much how to incorporate that
- 16:11Speaker 4: therapy into their practice, even if it's right for the
- 16:15Speaker 4: next patient. That second physicians patients are going to be
- 16:19Speaker 4: less likely to start treatment and persis even if their
- 16:23Speaker 4: coverage and affordability is relatively decent. And so, you know,
- 16:29Speaker 4: distilling that the risk factors here, you know, things like
- 16:33Speaker 4: physician experience that can influence the propensity of a patient
- 16:37Speaker 4: to have a good experience is really helpful to your
- 16:40Speaker 4: point because it's actionable. You can enable, whether it's a
- 16:46Speaker 4: life sciences company or a health system that's taking risk
- 16:49Speaker 4: to educate the right clinical stakeholders practice management stakeholders. You
- 16:57Speaker 4: can see patients are likely to be at risk much
- 17:02Speaker 4: earlier in the process. And basically what we do is
- 17:07Speaker 4: in order to do that is create what folks have
- 17:10Speaker 4: called digital phenotypes. So we see from all this data
- 17:16Speaker 4: that we're integrating, this next patient looks a lot like
- 17:21Speaker 4: many other patients that we've seen before, and that allows
- 17:26Speaker 4: us to mine through various scenarios that might impact the
- 17:32Speaker 4: patient different ways and recommend on a faster clip what
- 17:38Speaker 4: the manufacturer and the partners and others in the stakeholder
- 17:42Speaker 4: ecosystem might best do to support the patient.
- 17:46Speaker 1: So tell me more about that.
- 17:47Speaker 3: So how do you tailor write the predictive capabilities of
- 17:52Speaker 3: AI or machine learning to service a unique needs of
- 17:56Speaker 3: patients in these patient access teams who are trying best
- 17:59Speaker 3: to support I.
- 18:01Speaker 4: Think the first thing we do when we're working with
- 18:05Speaker 4: a life sciences company to deploy AI to help their
- 18:09Speaker 4: patients is listen. We really appreciate the opportunity to co
- 18:14Speaker 4: create with our life science as partners with others in
- 18:19Speaker 4: the ecosystem. Because AI that predicts what's going to happen
- 18:24Speaker 4: with patients is great, it's best if it's actually used.
- 18:31Speaker 4: It doesn't do any good to have AI predictions or
- 18:35Speaker 4: recommendations that case management teams aren't using as a general practice.
- 18:41Speaker 4: So the first thing that I think is important is
- 18:43Speaker 4: to understand where those teams are coming from with their
- 18:47Speaker 4: objectives are, what their service portfolios look like, their policy
- 18:53Speaker 4: landscape looks like, and then figure out, okay, well, what
- 18:57Speaker 4: is the right combination of models to be deploying. We
- 19:01Speaker 4: have a library of different models that will predict endpoints
- 19:06Speaker 4: from prescription abandonment and free drug conversion to prior authorization denial,
- 19:13Speaker 4: premature discontinuation. So a piece of it is understanding what
- 19:18Speaker 4: is what are the objectives, what models are helpful, and
- 19:22Speaker 4: then we tune the models. We actually there are ways
- 19:26Speaker 4: of really sort of tweaking levers in these machine learning
- 19:33Speaker 4: pipelines so that you can say, hey, I want this
- 19:36Speaker 4: model to be super precise when it says there's a
- 19:41Speaker 4: patient that is at risk. I want there to be
- 19:44Speaker 4: a very high likelihood that that's actually true. And so
- 19:49Speaker 4: if you say to a case manager, hey, I think
- 19:52Speaker 4: that this patient's at very high risk and the precision
- 19:56Speaker 4: on my model is very high. That case manager has
- 19:59Speaker 4: a lot of confidence that they're talking to a patient
- 20:02Speaker 4: who really needs help. And conversely, other teams might say, well,
- 20:09Speaker 4: I like the precision when there's a lot of risk,
- 20:11Speaker 4: but I also want a surveillance net. I want a
- 20:15Speaker 4: safety net that is going to detect all the patients
- 20:19Speaker 4: that might be at some risk for let's say, premature
- 20:25Speaker 4: drop off, you know, lack of adherence. So we say, okay,
- 20:28Speaker 4: we can tune the model a different way and then
- 20:32Speaker 4: characterize all the patients that are at some risk by
- 20:39Speaker 4: by solving for what they call recall, you know, finding
- 20:42Speaker 4: all of the patients at risk. And we put these
- 20:45Speaker 4: different techniques together and can sub segment the patient population
- 20:51Speaker 4: in such a way that these case manager teams understand
- 20:54Speaker 4: what they're looking at and they feel compelled to take
- 21:01Speaker 4: action when they see an alert. It's driven by these
- 21:05Speaker 4: AI systems, and the stories that come back to us about,
- 21:10Speaker 4: you know, places where a patient would have fallen through
- 21:13Speaker 4: the cracks had it not been for something like this,
- 21:15Speaker 4: are are pretty amazing.
- 21:17Speaker 1: So Stuart, as you're talking.
- 21:18Speaker 3: I'm envisioning, right future state there being like some type
- 21:22Speaker 3: of either computer programmer or an app where you know,
- 21:25Speaker 3: before medication is even prescribed, you could you know, enter
- 21:29Speaker 3: into you know this this is Stuart and these are
- 21:32Speaker 3: you know, this is what's going on with Stuart. Helped
- 21:34Speaker 3: me develop a treatment plan for Stuart. It's going to work, right, Well,
- 21:38Speaker 3: what do I where.
- 21:40Speaker 1: Are the you know, the.
- 21:42Speaker 3: Land minds that I have to navigate around, right, And
- 21:45Speaker 3: what potential issues could we have? And how do I
- 21:47Speaker 3: get Stuart from point A to point B in you know,
- 21:50Speaker 3: the straightest possible line.
- 21:52Speaker 4: I love that vision. Yeah, and you know, it's funny
- 21:57Speaker 4: just thinking about that as you talk about it in
- 22:00Speaker 4: that way. I do think there's a fair amount of
- 22:03Speaker 4: planning that practices try to do around what is the
- 22:08Speaker 4: best treatment for a patient, particularly where there are multiple options.
- 22:13Speaker 4: But of course they only have the information that they have.
- 22:16Speaker 4: There's a lot of information floating out there that often
- 22:19Speaker 4: is not present at the point of care or the
- 22:22Speaker 4: point of decision making about a patient's social determinants of health,
- 22:28Speaker 4: or their coverage policy, or the journey of patients that
- 22:40Speaker 4: have looked like them in the past. And so I
- 22:43Speaker 4: think there's utility in these techniques for physicians, for health plans,
- 22:49Speaker 4: for manufacturers, and I think it's interesting. I think you
- 22:52Speaker 4: are starting to see the adoption of AI by these
- 22:57Speaker 4: different groups. You see physician groups adopting AI to go
- 23:03Speaker 4: work through prior authorizations, for example, in a more efficient
- 23:07Speaker 4: and effective way. You see the adoption of AI by
- 23:12Speaker 4: the payer community to better stratify risk. And you see
- 23:16Speaker 4: the adoption of AI by a lot of those sort
- 23:20Speaker 4: of commercial leaning folks on the manufacturing side to go
- 23:23Speaker 4: figure out which patients are likely to be eligible for
- 23:27Speaker 4: a new prescription. Where I think we see an opportunity,
- 23:33Speaker 4: you know, where this technology isn't being leveraged nearly as
- 23:36Speaker 4: much is actually once that prescription is written on the
- 23:41Speaker 4: manufacturing side, to help those teams better figure out how
- 23:47Speaker 4: to take care of patients that are you know, clinically
- 23:50Speaker 4: appropriate for therapy. Yeah.
- 23:53Speaker 1: Absolutely, So I've talked a lot about these models. How
- 23:58Speaker 1: accurate are they?
- 24:00Speaker 4: They're you know, they're pretty accurate. And I will say,
- 24:02Speaker 4: you know, just to sort of take a step back,
- 24:05Speaker 4: there's a lot I think that can be done without AI,
- 24:10Speaker 4: just in getting the data together and doing this reporting
- 24:14Speaker 4: and structuring the inside of the patient experience. All of
- 24:18Speaker 4: that is really valuable for teams on the ground. But
- 24:23Speaker 4: once folks have you know, crawled and then walked and
- 24:27Speaker 4: they're ready to run, this can be a great way
- 24:31Speaker 4: to begin that process. You know, we routinely see that
- 24:38Speaker 4: when we're deployees moles, they'll they'll correctly predict for high
- 24:45Speaker 4: risk groups eighty percent of patients that are likely to
- 24:51Speaker 4: experience a negative event. And when we're configuring that surveillance
- 24:56Speaker 4: safety net, we're seeing again and again, you know, the
- 25:03Speaker 4: detection of more than eighty percent of possible cases out there.
- 25:09Speaker 4: So there is a lot of heavy lifting and a
- 25:12Speaker 4: lot of sort of empirical support for the use of
- 25:16Speaker 4: machine learning in patient services.
- 25:19Speaker 3: So let's say we start with the patient, new patient
- 25:21Speaker 3: to therapy, right, we put them into one of these
- 25:23Speaker 3: models or into these groups. Then do you take that
- 25:27Speaker 3: patient's individual response and also then use it right to
- 25:31Speaker 3: further drill down for that particular person, right, like future
- 25:36Speaker 3: actions or what we could anticipate with this one patient. Right,
- 25:39Speaker 3: Because I always tell people patients do what they do. Right.
- 25:43Speaker 3: Being a practicing pharmacists from a long time ago, I
- 25:46Speaker 3: was always surprised at some of the stuff that people
- 25:48Speaker 3: would do, right despite you trying to counsel them, and
- 25:50Speaker 3: you're trying to cover everything that you can, right, cover
- 25:53Speaker 3: all your bases, but still sometimes they just do what
- 25:55Speaker 3: they do. So can you take some of those behaviors, right,
- 26:01Speaker 3: are those actions, and then use that then to to
- 26:05Speaker 3: further tailor your approach? You know these you know these
- 26:09Speaker 3: life sciences patient access teams to further tailor the approach
- 26:13Speaker 3: for that person.
- 26:15Speaker 4: Absolutely, and I think that's where a lot of the
- 26:16Speaker 4: frontier is on this and recommendation engines will when sort
- 26:22Speaker 4: of like properly built and configured, we'll say, well, hey,
- 26:26Speaker 4: you know, this type of patient tends to prefer to
- 26:32Speaker 4: get a text in the evening with this sort of content,
- 26:37Speaker 4: and you know when they click on the link, they
- 26:40Speaker 4: prefer to go to this sort of view and to
- 26:44Speaker 4: see their benefits. That's what they like. You know, it's
- 26:47Speaker 4: this other segment actually would rather get a phone call
- 26:51Speaker 4: or maybe even have the doctor's office call them. There
- 26:55Speaker 4: they respond to telephone better. I mean, we all know
- 27:00Speaker 4: folks in our lives who you know, you try to
- 27:04Speaker 4: send them an email that's complicated, but you send them
- 27:06Speaker 4: text to get right back to you, well, I mean,
- 27:09Speaker 4: patients are just people, right, So to your point, the
- 27:13Speaker 4: same sorts of techniques can be used to optimize the
- 27:19Speaker 4: outreach to those patients. The question is, at some level
- 27:26Speaker 4: are we getting the data necessary to drive these models
- 27:32Speaker 4: into and optimize outreach, And so so you sort of
- 27:35Speaker 4: come back to thinking carefully about in order to set
- 27:38Speaker 4: the system up that really personalizes patient support for each individual,
- 27:48Speaker 4: it pays to be thoughtful about how information is being collected,
- 27:53Speaker 4: how it's being pulled together, what are what is the
- 27:57Speaker 4: framework with which the patient journey is being looked at,
- 28:01Speaker 4: you know, and who on the care team is being
- 28:05Speaker 4: pulled into that because the other thing that I think
- 28:09Speaker 4: we often see is, you know, just the same way
- 28:12Speaker 4: that some patients prefer a text and some patients prefer
- 28:16Speaker 4: a phone call. Sometimes where a patient is in the
- 28:19Speaker 4: journey is can you know, can be perfectly well supported
- 28:24Speaker 4: by the hub or patient call center. Sometimes you really
- 28:28Speaker 4: need the specialty pharmacy to be the one to reach out.
- 28:33Speaker 4: And knowing the difference, even for the same patient can
- 28:37Speaker 4: make it and it can have an impact on the
- 28:42Speaker 4: likelihood of them getting on and staying on therapy as appropriate,
- 28:46Speaker 4: and so so it does require a continuous feedback loop,
- 28:52Speaker 4: a continuous management of information about the patient as quickly
- 28:56Speaker 4: as possible, a sort of robust patient framework to look
- 29:00Speaker 4: at what are they likely to run into next, and
- 29:03Speaker 4: who is the best stakeholder on the care team to intervene,
- 29:08Speaker 4: whose best position to make.
- 29:10Speaker 3: A difference, right, And like you said that that can
- 29:13Speaker 3: change for the same patient and the same journey, right,
- 29:16Speaker 3: but they're just at a different point in the journey.
- 29:17Speaker 3: If everything's going okay, hey, maybe it's fine. We can
- 29:20Speaker 3: communicate by a text. Oh geez, I you know, I
- 29:23Speaker 3: I went to the doctor and I have cancer and
- 29:25Speaker 3: I'm progressing. I want to talk to somebody, right, you know,
- 29:29Speaker 3: I need some reassurance or what are we going to do?
- 29:31Speaker 3: I want to I want to talk to somebody. So
- 29:33Speaker 3: we have to be flexible enough right to pivot to
- 29:35Speaker 3: meet their needs.
- 29:37Speaker 4: That's absolutely right, And I think that sort of flexibility
- 29:42Speaker 4: and the the productive communication across the care team is
- 29:51Speaker 4: really sort of important in order to serve the patient well.
- 29:55Speaker 4: And in fact, you know, when we see these sorts
- 29:58Speaker 4: of aim machine learning techniques be most impactful, again, they're
- 30:02Speaker 4: not using a vacuum. They're They're just another tool in
- 30:05Speaker 4: the toolkit that teams can use to focus their efforts.
- 30:10Speaker 4: Many times it stimulates a dialogue between a HUB and
- 30:18Speaker 4: an SP or an SP in a practice that otherwise
- 30:22Speaker 4: might not have occurred. And it's often that dialogue that
- 30:27Speaker 4: is what is the real intervention point that really sort
- 30:33Speaker 4: of like is the moment where the patient journey changes? Yeah,
- 30:39Speaker 4: and you know, if nothing else, hopefully a lot of
- 30:42Speaker 4: this sort of machine learning and AI work and patient
- 30:45Speaker 4: services helps us create more of the right conversations between
- 30:51Speaker 4: the right folks at the right time.
- 30:54Speaker 1: All right, So explain to me how we collect this data?
- 30:57Speaker 3: Right, And you keep saying machine learning and AI, what's
- 31:00Speaker 3: the difference, right, I'm assuming that we take the data
- 31:03Speaker 3: and we do something with it and then that's into
- 31:04Speaker 3: the machine who's learning.
- 31:05Speaker 1: But I don't know.
- 31:07Speaker 4: Yeah, that's a great question. So machine learns just one
- 31:09Speaker 4: type of AI. I think one way that we like
- 31:13Speaker 4: to think about things is you sort of have AI broadly,
- 31:18Speaker 4: which is about having computer systems, synthetics systems, you know,
- 31:24Speaker 4: facilitate decision making or recommendations or do things that we
- 31:31Speaker 4: associate with people with the way our minds work. Now
- 31:36Speaker 4: there's no computer system today that works the way our
- 31:39Speaker 4: minds work. There are many scholars who doubt that that
- 31:43Speaker 4: will ever happen. If it does, it's far in the future.
- 31:47Speaker 4: So today when we talk about AI, when anybody talks
- 31:51Speaker 4: about AI, they're generally talking about one of a few
- 31:55Speaker 4: different buckets of capabilities. I think a lot lot of
- 32:00Speaker 4: what people see in the press today and what we're
- 32:06Speaker 4: sort of constantly surrounded with is generative AI. So if
- 32:10Speaker 4: you have sort of the chat GPTs of the world
- 32:15Speaker 4: mining through big libraries of text and video and other
- 32:20Speaker 4: content and producing something novel. Now that that has many
- 32:27Speaker 4: use cases, it's not as measurable as some other use
- 32:32Speaker 4: cases for AI, and it doesn't really sort of predict
- 32:35Speaker 4: the future. It's used to create something new. If you
- 32:39Speaker 4: want to predict an outcome, then what a lot of
- 32:44Speaker 4: people will will look to is another subset of AI
- 32:50Speaker 4: that you know, folks call machine learning because it is,
- 32:53Speaker 4: as you alluded to, methods for helping computers learn, and
- 33:00Speaker 4: without getting into too many of the gory details, basically
- 33:03Speaker 4: what a lot of these models do is they create
- 33:08Speaker 4: a proxy for a neural net. There are many different
- 33:12Speaker 4: nodes of information and the nodes of information, look at
- 33:18Speaker 4: many different scenarios of possibility and the likelihood of those possibilities,
- 33:26Speaker 4: and solve for the possibilities that have the least amount
- 33:32Speaker 4: of error, if you will, and then and then generate
- 33:37Speaker 4: an output based on that solution, and then you can
- 33:42Speaker 4: measure that right, because if I if I put all
- 33:45Speaker 4: this information through a neural net and it predicts that
- 33:50Speaker 4: a patient will discontinue in the next ninety days, well
- 33:54Speaker 4: that event is either going to happen or it's not
- 33:56Speaker 4: going to happen. And so we can measure the accuracy
- 34:02Speaker 4: of these machine learning models. And that's that's how we're
- 34:05Speaker 4: able to say, oh, yeah, you know, we have models
- 34:07Speaker 4: that are eighty percent precise or more for example.
- 34:12Speaker 3: So are there any compliance concerns with working with all
- 34:15Speaker 3: of this data right and running it through all these
- 34:18Speaker 3: different algorithms or machinations.
- 34:23Speaker 4: Yes, certainly. I mean it's patient level data. So we
- 34:29Speaker 4: have to be really mindful and compliant with hip high tech.
- 34:36Speaker 4: We at Clairetowseracts have a hip high tech Sock two
- 34:41Speaker 4: and now GDPR compliant platform, and so we spend a
- 34:46Speaker 4: lot of time thinking about where we're getting data from,
- 34:50Speaker 4: how we're combining with other sets of data what we're producing,
- 34:56Speaker 4: how that information that's being produced is being you in
- 35:01Speaker 4: support of patients. So I do think that this goes
- 35:05Speaker 4: hand in hand with conversations about de identification and patient
- 35:11Speaker 4: consent and the the the management of of of privacy considerations.
- 35:20Speaker 4: A great example of this is, you know, one thing
- 35:24Speaker 4: that we like to do to really tune our models
- 35:30Speaker 4: is mine through free text. Many times the data we get,
- 35:35Speaker 4: the sort of structured data we get is really helpful.
- 35:38Speaker 4: But if we can mine through pharmacy notes or hubnotes,
- 35:43Speaker 4: that can tell us a lot about the for example,
- 35:46Speaker 4: the emotional state of a patient over time, and that
- 35:49Speaker 4: can correlate with their outcomes too. Well. Forget about this
- 35:54Speaker 4: sort of all the technical complexity around minding free text.
- 35:57Speaker 4: There's privacy complexity around doing that because it's phi I
- 36:01Speaker 4: mean or PII. If the patient's consented, you might have
- 36:05Speaker 4: lots of information about who that patient is in that text,
- 36:08Speaker 4: and so managing that appropriately is a foundational part of
- 36:13Speaker 4: making this work well. And that is maybe one reason
- 36:16Speaker 4: why this is a bit of a different space than
- 36:20Speaker 4: using AI for a video game, for example, because you
- 36:23Speaker 4: have to be really careful about protecting patients privacy in
- 36:28Speaker 4: the course of doing this work.
- 36:30Speaker 3: Absolutely, and you had said earlier, right, that a computer
- 36:33Speaker 3: doesn't think the way that the human mind does. So
- 36:36Speaker 3: this mining seems to be interesting, right, trying to get
- 36:39Speaker 3: more than just objective data, but there may be subjective
- 36:42Speaker 3: or some emotional, right, some insights that could be used.
- 36:48Speaker 4: Yeah, it is really it's fascinating. I've been really impressed
- 36:53Speaker 4: by our data science team as they've sort of creatively
- 36:59Speaker 4: looked at ways is to use methods of mining free
- 37:04Speaker 4: text and then translating that into metrics that can be
- 37:10Speaker 4: used to measure a patient's emotional state. I mean, to
- 37:14Speaker 4: your point, a computer is not going to understand what
- 37:20Speaker 4: it is for a patient to feel hopeless or in
- 37:25Speaker 4: need of support or sort of questioning where they are
- 37:30Speaker 4: on their journey. And there's been a lot of really
- 37:33Speaker 4: good literature written about the importance of social emotional support
- 37:38Speaker 4: for patients, particularly with going through you know, complex treatment journeys.
- 37:47Speaker 4: But by that same token, if we can translate that
- 37:52Speaker 4: emotional state into data that can be fed into these
- 37:57Speaker 4: am models, then we can take better care of those patients.
- 38:00Speaker 4: Maybe because we can detect when patients are going through
- 38:06Speaker 4: that emotional rollercoaster and configure our outreach to them more effectively.
- 38:12Speaker 4: We can have a better understanding of their likelihood to
- 38:18Speaker 4: run into particular types of problems. So, while may be
- 38:23Speaker 4: a little counterintuitive, there's actually a lot of value in
- 38:28Speaker 4: appropriately mining text and using it to better understand the
- 38:35Speaker 4: needs of each patient.
- 38:37Speaker 3: It's just fascinating, sture, what are there other opportunities right
- 38:41Speaker 3: to enhance these models or the predictions that they're making
- 38:44Speaker 3: that you know, folks who are listening may not be
- 38:47Speaker 3: as familiar with.
- 38:49Speaker 4: Sure there are there are a lot, and I'll give
- 38:51Speaker 4: you a couple sort of that have to do with data,
- 38:55Speaker 4: but then I think there's sort of a broader call
- 38:57Speaker 4: to action. So a couple that are interesting. You know,
- 39:01Speaker 4: we're doing a lot of work right now around social context.
- 39:08Speaker 4: There's been a lot that's been written about social determinants
- 39:11Speaker 4: of health and how that can correlate with the patient's outcomes. Separately,
- 39:18Speaker 4: I think you know a lot of the conversation we've
- 39:22Speaker 4: had about the experience of a doctor predicting the outcome
- 39:26Speaker 4: of the patient, Well, that's true of a physician's experience
- 39:30Speaker 4: on a particular therapy. What if that were also true
- 39:33Speaker 4: of the physician's experience across a class right across many
- 39:38Speaker 4: similar types of therapy. The data on that has been
- 39:44Speaker 4: harder to get at, but we've done some work in
- 39:47Speaker 4: the last few years to create data sets that sort
- 39:51Speaker 4: of supplement more traditional market level data so we can understand,
- 39:58Speaker 4: you know, how aarticular prescriber, for example, has behaved across
- 40:05Speaker 4: different specialty therapy. So I think there's some interesting opportunity
- 40:09Speaker 4: to be leveraging that looks sort of beyond the brand,
- 40:12Speaker 4: if you will, at the broader market. But ultimately, I
- 40:16Speaker 4: think the biggest opportunity for a lot of this is
- 40:20Speaker 4: for market access and patient services teams and their partners
- 40:25Speaker 4: to look at this as a very practical tool. By this,
- 40:29Speaker 4: I mean machine learning and artificial intelligence methods, you know,
- 40:33Speaker 4: advanced analytical methods. Look at this as a very practical
- 40:37Speaker 4: tool that they can be bringing into their operations the
- 40:43Speaker 4: way many other teams within and beyond the biofarmat industry have.
- 40:50Speaker 4: There's a real opportunity to help patients more by leveraging
- 40:55Speaker 4: the technology that's right in front of us.
- 40:58Speaker 3: It's fascinating, fascinating, fascinating. So if I heard you correctly,
- 41:03Speaker 3: we can look at the broader market right to predict
- 41:08Speaker 3: how a patient may react or their compliance or adherence
- 41:13Speaker 3: to one particular brand.
- 41:14Speaker 1: Is that true?
- 41:16Speaker 4: That's a lot of where our existing, our current research
- 41:19Speaker 4: is focused on. Yeah, is sort of like, what are
- 41:22Speaker 4: the other sources of data we could be pulling in
- 41:25Speaker 4: here to make our models even more accurate? And then
- 41:29Speaker 4: what else can we be doing with our partners to
- 41:33Speaker 4: make them more accessible and help even more people? Sort
- 41:37Speaker 4: of get the benefit of letting technology do the heavy lifting.
- 41:44Speaker 1: Any closing thoughts short, Well.
- 41:46Speaker 4: You know, we started this conversation about the personalization of
- 41:51Speaker 4: medicine from a clinical perspective, and we've talked a lot
- 41:55Speaker 4: about the opportunity to leverage AI and types of AI
- 42:02Speaker 4: like machine learning to personalize it from a support perspective.
- 42:08Speaker 4: I think that it's an exciting time in the journey
- 42:13Speaker 4: of medicine broadly, as we think about the movement from
- 42:22Speaker 4: the old days of broad based retail small molecule therapies
- 42:27Speaker 4: through the monoclonal antibodies that have been launched over the
- 42:33Speaker 4: last couple of decades, and the combinations of those targeted
- 42:38Speaker 4: therapies and response to different patient subpopulations and we think
- 42:44Speaker 4: about the the foray into selling gene therapy and even
- 42:52Speaker 4: perhaps what's beyond that. One thing is really clear clinically,
- 42:58Speaker 4: medicine is becoming increasingly personalized. Right. So the exciting thing
- 43:06Speaker 4: for me, and I think for you know, lots of
- 43:08Speaker 4: folks that we partner with, is the opportunity to match
- 43:13Speaker 4: that degree of personalization when we're supporting patients by really
- 43:19Speaker 4: understanding that person, and not just on a human level,
- 43:26Speaker 4: though that is certainly central and will remain central to
- 43:29Speaker 4: the provision of care, but to assist the case manager,
- 43:34Speaker 4: to assist the pharmacists, to assist the field access team
- 43:38Speaker 4: member with better AI that can help them be the
- 43:44Speaker 4: best version of themselves in the moment when caring for patients,
- 43:49Speaker 4: and that takes a village. So I would invite everybody
- 43:53Speaker 4: listening to the podcast to reach out if this is
- 43:59Speaker 4: of interest, and explore ways that we can sort of
- 44:04Speaker 4: co create this future of AI assistant patient services together.
- 44:09Speaker 3: Well that's awesome, yes, and I second that please reach out, right,
- 44:13Speaker 3: it's only through working together, right and being collaborative, that
- 44:19Speaker 3: we're going to continue to advance this and keep up
- 44:22Speaker 3: right with the train that's already left the station exactly.
- 44:27Speaker 1: Oh my gosh.
- 44:27Speaker 3: Sure. Thanks so much for chatting with me and sharing
- 44:30Speaker 3: this incredibly fascinating information.
- 44:33Speaker 4: Anytime, Sheila, It's great to be talking with you as always.
- 44:37Speaker 3: As a reminder, NISP corporate members are invited to be
- 44:40Speaker 3: my guest on one complimentary podcast per year. For more
- 44:44Speaker 3: information about this or other na sp membership benefits and events,
- 44:48Speaker 3: please feel free to reach out to membership at NA
- 44:51Speaker 3: sp n E t dot org or visit our website
- 44:55Speaker 3: at www dot na s p n e t dot org.
- 45:00Speaker 4: Eat.
- 45:00Speaker 3: Until next time, have a great day.