Automation, innovation, and the future of drug safety

Research in Action - A podcast by Oracle Corporation

Categories:

International Data Corporation reports safety caseloads are increasing by 30% to 50% each year, and emerging technology will be the only way to keep up. But how are powerful technologies like generative AI advancing safety and pharmacovigilance? Is touchless case processing a good or bad thing? And how do we balance AI, automation, and the human touch? We will get answers to those questions and more in this episode with Bruce Palsulich, Vice President of Safety Solutions at Oracle Life Sciences. His portfolio includes Argus Safety, the industry-leading adverse event case processing and analytics solution, and Empirica Signal, the standard for signal detection and risk management. He has more than 30 years of experience in the healthcare and life sciences industry, including 25 in pharmacovigilance.    -------------------------------------------------------- Episode Transcript: 00;00;00;00 - 00;00;13;22 What is pharmacovigilance? How can technology best handle the tracking of adverse drug events? And is touchless case processing a good or a bad idea? We'll get those answers and more on this episode of Research in Action.   00;00;15;01 - 00;00;18;28 The lead, the Building.   00;00;20;10 - 00;00;48;22 Hello, welcome to Research in Action, brought to you by Oracle Life Sciences. I'm Mike Stiles. Today we are talking with Bruce Palsulich, vice president of Safety Solutions at Oracle Life Sciences. Bruce's portfolio includes Argus Safety, the industry leading adverse event, case processing and analytics solution, and empirical signal, the standard for signal detection and risk management. He's got more than 30 years of experience in the healthcare and life sciences industry, including 25 and pharmacovigilance.   00;00;49;02 - 00;01;03;25 Now, why is that important? Well, International Data Corporation reports safety caseloads are increasing 30 to 50% each year. Bruce is intimately involved in tackling that volume. So, Bruce, thanks for thanks for being with us today.   00;01;04;05 - 00;01;06;00 Yeah, thanks, Mike. Happy to be here.   00;01;06;16 - 00;01;17;04 Yeah. Let's get acquainted with you first. How did Life's path bring you into life sciences technology? How did you kind of wind up at Oracle and what are you tasked with getting done there?   00;01;17;29 - 00;01;50;08 You know, back back when I was still in university, I actually started off doing software development and consulting with a medical device company. And so early in my career, it was working on the actual embedded software that controlled medical devices. And early on ended up joining a consulting firm that started off doing engineering, consulting on medical devices, and eventually working towards quality software and regulatory submissions.   00;01;50;24 - 00;02;17;04 And so came to Oracle in 2009. So we had acquired a company that was that small engineering startup that I mentioned. And this is the company that originally developed Argus Safety, so I managed the team that developed Argus safety originally and through my time at Oracle, I jumped out of a safety for a little while.   00;02;17;04 - 00;02;42;24 For about four years I was running our healthcare strategy. That was when we had a much smaller healthcare footprint than we now have with our acquisition of Cerner. But at the time we did a lot of things in sort of what was called health-information exchange, sort of the foundation for national platforms under Australia and Singapore and multiple provinces in Canada.   00;02;43;09 - 00;02;51;18 And after doing that for about four years and then I came back to the safety side of the business about ten years ago or so.   00;02;52;03 - 00;03;02;25 Well, did you always see yourself doing something in medicine and life sciences, like when you were younger, or did this was this a life path that kind of surprised you?   00;03;03;08 - 00;03;29;12 You know, I ommitted the part where for four years I actually worked in aerospace. So I even though when I was still at university, I started off in medical devices. I did take a job in aerospace for four years. But that's sort of left a hollow feeling and not the same sort of mission driven purpose. When you do have a role that's within the broader health care or clinical development.   00;03;29;12 - 00;03;55;04 So, you know, I think many people like myself that, you know, whether you're on the vendor side or whether you're on the the pharma side of drug safety or pharmacovigilance or even broader clinical development, I think you do appreciate that there's there's a calling and you feel more purpose driven life. I suppose working in a field that's helping individuals, helping patients.   00;03;55;26 - 00;04;13;27 Well, for our audience, and I'm deflecting because our audience is smart, this is mostly for me. Let's just level set. What's what's the main goal of safety and pharmacovigilance? And I imagine safety standards would apply across every step in that drug development process.   00;04;14;10 - 00;04;46;07 Yeah. So drug safety and pharmacovigilance is really trying to understand the safety of drugs that are under both clinical development as well as once they complete their clinical development and are approved for broad market use. And so clinical trials really focus on safety and efficacy, but that's done under controlled conditions with a limited number of patients and and sort of restricted patients as well.   00;04;46;07 - 00;05;27;27 And once a marketed drug is approved, it's going to be exposed to significantly more patients. And so during a clinical development, a clinical trial, if you had an adverse event that occurs in one out of 10,000 people, that's that's sort of defined as a rare adverse event or adverse reaction. You can imagine if you gave that to a billion people, maybe, for instance, in the example of the COVID vaccines, Now that rare adverse event that's only occurring in one out of 10,000 people is actually occurring 10,000 times in a billion people.   00;05;27;27 - 00;05;42;04 And so so really, you know, pharmacovigilance is looking at and trying to understand that benefit risk and manage that risk when it's being exposed under real world conditions to to actual patients.   00;05;42;24 - 00;06;11;20 So the study of a drug is hardly done after it's approved by the FDA and goes out into the public, the public market, that monitoring is still happening while safety is paramount, It can't be easy. I mean, for whatever reason, the public does seem to expect perfection without risk when it comes to their drugs. So, I mean, what are the biggest challenges that Pharmacovigilance and the industry has to deal with currently?   00;06;12;04 - 00;06;50;12 So, you know, getting back to sort of those controlled conditions that are under clinical trials, for instance, typically you're not looking at pediatric or children exposure. Quite often you're not dealing with elderly patients or immune compromised patients or patients taking multiple medications. You know, do you have the diversity within your clinical trials such that you're getting genetic differences that might exist within different populations and such?   00;06;50;12 - 00;07;21;16 And so so all of those are exposures that are going to occur during broad use of those products once they get approved. And so so pharmacovigilance is really trying to, you know, track that, trying to collect as many adverse reactions that occur. It's trying to evaluate whether or not those events truly are a reaction that's related to the drug that's being studied and the drug of interest.   00;07;21;16 - 00;07;46;15 Or is it just occurring, for instance, within the general background rate that you would expect within within a patient population? And so all of that analysis is to try and understand, is it more than correlation that just, you know, we have an adverse event that occurred with a drug? Is that coincidence or is that related to other drugs you're taking?   00;07;46;15 - 00;08;13;16 Is that a progression of the disease that the patient is taking a medication for, or is it something that is actually induced by by the drug of interest? And how serious is that reaction? And is that something that should be, you know, updated on the prescribing information that's tracked along with a drug and the, you know, communication and education that's done to the health care community.   00;08;13;16 - 00;08;16;08 So they understand the risks associated with the drug.   00;08;16;28 - 00;08;46;17 So I get the challenge, which is that in a clinical trial to get a drug approved and on the market, there's no way to cover every possible circumstance and every type of person and every type of situation where, like you said, there are other actions with other drugs. And I already get the enormity of the challenge of keeping track of all of those people, all of those interactions, all of those adverse effects.   00;08;46;20 - 00;08;59;13 I imagine technology is tackling those challenges, right, Or at least helping to tackle them. For instance, like how can we better efficiently do data management? How does that play a big role in tackling these problems?   00;08;59;28 - 00;09;24;29 Yeah, So the you know, we talked about the increasing volumes somewhat. It's still generally estimated that somewhere on the order of between five and 10% of the actual adverse events that occur are actually reported. And so many people might just say, well, I felt dizzy when I took that and so I stopped taking it. And, you know, did you ever tell your doctor, Well, no, I just manage that on my own.   00;09;24;29 - 00;09;56;21 So so really part of the challenge is how can you make it easier to collect a higher number of of these adverse reactions that actually occur? How can you reduce the burden on both the patient and on a health care professional to report those? The other is that, you know, we want to move beyond the handling and the workflow of processing these individual adverse event reports and get to a more of the emphasis being placed on driving or deriving insights from the data itself.   00;09;56;21 - 00;10;18;20 So so we want to make, as we deliver our own solutions, we want to make the pharma companies more efficient at being able to handle these sort of transactions. But with the real value out of that of then more, more effort and more value can be derived from the insights. From the data itself.   00;10;19;10 - 00;10;37;03 Yeah. I mean, there's a need to track adverse events that are happening all the time. The volume and the sources of that data increases exponentially. So you kind of touched on it there. How do you go about not just effectively managing the data flow but actually making it actionable?   00;10;37;14 - 00;11;18;12 So I think part part of this is, is within an ecosystem where perceptions are changing. And I'll say when I entered the field, you know, back in the mid nineties and such, the perception was sort of like an ostrich putting their head in the sand or something. And, and I don't want to know about what hasn't specifically been reported and, and Pharmacovigilance and drug safety was really looked at as sort of a a tax on the business a cost of doing business and wasn't appreciated as a valuable information asset that can be leveraged, you know, within a biopharma organization.   00;11;18;12 - 00;12;00;26 And so now I think PV data being an expensively curated data set, is now looked as a valuable information asset within organizations. It can be used to identify new indications, it can be used to inform drug discovery and portfolio prioritization. I think more and more we're seeing safety used as a competitive differentiator and certainly we saw that with the COVID vaccines and those that were commercially successful versus those that perhaps were perceived as having a more risks associated with those.   00;12;00;26 - 00;12;26;19 And towards this, I think, you know, we're looking at, you know, how can advances in data science, technology, things like machine learning, predictive models, generative AI, how can they be leveraged in order to process and be able to make use of these increasing volumes of information as well as diverse sources of adverse event information as well?   00;12;27;07 - 00;12;42;22 Yeah, that's where I want to go next. Are you seeing cloud based platforms and AI transforming pharmacovigilance? I mean kind of balance the hope and the hype for me. How do you see those technologies changing, how we approach drug safety and in like, say, the next decade or so?   00;12;43;05 - 00;13;18;23 So I really think and not not even just in this field, but in all fields, if you look at sort of the proliferation and the scaling of accumulation of data and information, it really requires new methods to approach that. So I do think that things like the large language models like Generative AI, are really going to be transformational into how we leverage this data and information specifically within health care and life science, but but also broader, I think, as a global population.   00;13;18;23 - 00;13;50;04 But so you can imagine even things like, you know, querying the data versus the natural language conversation, you know, perhaps you could ask how rare is this actual event or how does the rate of this adverse event compare for my drug versus other drugs within the same therapeutic class or given the volume of adverse events for this drug in 2023, how might how many reports might we expect to receive in in 2024?   00;13;50;04 - 00;14;26;08 Or are there clusters of patients that appear to be more likely to have this adverse event than other patients? And could you describe those differences? And so those I think, are all sort of examples that we're going to move from strictly having skills of of a data science list or query builder, a developer and such accessing data to sort of expose those questions of the data closer to the the individuals that are forming the question.   00;14;26;08 - 00;15;06;24 And so I think right now, you know, we really don't know what sort of insights or what sort of interactions are going to exist between these diverse data sources that are going to lead towards improved insights, improve patient safety. You know, we really want to, you know, identify what drugs work for, what patients and inversely know which patients shouldn't be exposed to certain drugs and and what characteristics, what scientific information is out there already, both broadly, you know, basic chemistry, genomics, pharmacokinetics, things like that.   00;15;07;12 - 00;15;10;29 But then bring that down to the experience of an individual patient.   00;15;11;18 - 00;15;24;29 Well, you've talked before about touchless case processing and what that could look like in the future. Tell us what that is and what companies should be doing now to start transitioning to that kind of model.   00;15;25;17 - 00;15;55;05 So I think sometimes the the phrase touchless case processing can sound a little scary, you know, that humans are going to be completely out of the loop and such. And I think the industry is generally looking for something a little bit more incremental. So we're not looking to say all cases should now be touchless. We're looking at things like, well, perhaps non-serious cases that don't provide a lot of new scientific information.   00;15;55;05 - 00;16;37;28 Perhaps those should be handled automatically by the system, perhaps for drugs that are well understood or have been on the market for a long time. Perhaps those would be better candidates for having automated case processing then things that are going to be a new a new drug on the market with less experience and exposure, perhaps cases that are received electronically and, you know, or cases from partners, you know, quite often they'll be global relationships between one pharma who partners with another pharma to to market that product in another region of the world.   00;16;37;28 - 00;17;03;22 And so you're receiving adverse event cases from this partner who who is originating those from patients or health care professionals. But if you're receiving that from a partner, you probably trust that they're sending it to you and maybe you can process that item automatically. The other is, is I think again, people get get a bit concerned if you say, well, this is going to be end to end and no human ever touched it.   00;17;03;22 - 00;17;35;06 And now we're going to be reporting this. You know, it doesn't necessarily have to be end to end. It can be the high volume of effort activities like doing the actual data entry. It can be decision support to support perhaps the causal assessments or to assess whether or not this is team serious or to look at is this an adverse event that's already listed on the the product label or prescribing information So it can be, you know, specific work steps are workflow steps.   00;17;35;06 - 00;18;15;27 Could be touchless, but overall, you know where it is appropriate. I think we still want humans in the loop to to oversee the process overall. So I think there are tremendous opportunities again, to take repetitive non value added processes out of and automate those from from requiring human effort to process those and allow the humans to focus on, you know, insights and focus on more value rather than these repetitive steps that that computers are well suited to be able to process as well.   00;18;15;27 - 00;18;39;08 You said something earlier, and that's very legitimate that, you know, a lot of patients will start taking a drug and experience some kind of adverse reaction to it and then just stop and not even tell their doctor about it. No one's ever going to know about the adverse reaction that they had. So there's even a reliability factor on the part of the patients and their willingness to report.   00;18;39;27 - 00;19;05;15 How far away are we from being able to have essentially a digital model of patients that drugs can be tested on? I mean, am I going way far ahead in the world of science fiction where in Silico gets kicked up a notch and safety procedures are tested on not real people, but essentially digital versions of patients?   00;19;05;15 - 00;19;35;22 Yeah, I think this whole concept and people may have heard the term digital twin and such is is obviously very interesting and I think we'll have certain benefit. I think, you know, certainly, you know, establishing toxicity and such would much better be supported through some of these models than than experimenting on on animals or on humans in order to establish toxicities and such.   00;19;35;22 - 00;20;12;08 And so so I think, you know, it's going to start from sort of the bottoms up that way when you're looking at those types of exposures. And I think as we get again, as we sort of stitch together these diverse data sources and have tools to be able to look for correlations and linkages that that are there, that would be difficult for humans to ascertain, then I think, you know, that will allow us to sort of advance these digital models that that represent a human response to medications and such.   00;20;12;08 - 00;20;44;18 So I think that's something that is definitely being advanced and we have pockets of that, and those pockets will ultimately end up being combined into a larger simulation of, you know, humans. So yeah, it's certainly an interesting area. And even myself, you know, it took me a while to sort of get my head around what that concept of digital twin and how that's going to benefit clinical development as well as is health care overall.   00;20;45;16 - 00;21;06;07 Well, we touched on the balance of hope and hype, but there's another balance here that you also touched on a bit. It feels like we want every advantage that technologies and automation and machines can bring us, but then we only trust those things up to a point. We do want human experience, human judgment and expertise to kind of have the final word.   00;21;06;07 - 00;21;13;22 So how do you view where that balance is now between tech and human? What gets us to the lowest error rates?   00;21;14;07 - 00;21;47;22 So I think, you know, one of the perception challenges that exists right now is that people think the humans are probably doing a better job than they really are right now. So if you gave the same health care record source document to five different people and said, you know, take from this piece of paper and enter it into the system, you would probably end up you would not end up with five identical versions of data entry from abstraction from that source medical record.   00;21;47;22 - 00;22;15;14 And so, you know, which one of those five is right. And what's the error rate there? And so I think you would normally say that humans are going to be somewhere on the order of six or 7% error rate for that type of work. And so even in manual processing is adverse event cases, typically there's going to be some sort of QC sampling that's trying to keep a handle on detecting errors and keep a handle on the overall process and such.   00;22;15;14 - 00;22;43;04 And so looking at how, you know, automation or machine learning is going to apply similar things are going to occur. You still want some checks and balances in order to know that you still have control of the automated process and things that are getting into medical judgment. I still think we we want to stick within what we would say is sort of augmented processing or decision support.   00;22;43;04 - 00;23;27;27 Speaker 3 So you want to provide assistance to the person making those judgments and say the system has determined that we think this might be related to this drug and based on these factors, why we think that might lead to that decision. Again, it would be up to the health care professional to make the final judgment there. So I think we are you're trying to bring the facts, bring the the right parameters and such into view so that the human can make the best decision, given the data points and the assessments that are being being suggested by by the system.   00;23;28;04 - 00;24;04;15 So I think we're still, you know, I was listening to NPR yesterday and they had a discussion on self-driving cars and there are self-driving cars ever going to get to the same accuracy and insights of of a human. And I think, you know, this is similar here, although probably, you know, certainly a different problem than looking at real time sensors in forming a automated self-driving car, but trying to look at human experience, human judgment, you know, how do we model some of those?   00;24;04;26 - 00;24;16;25 I think right now we'll stay in this augmented decision support mode for many of these, you know, clinical medical decisions and certainly leave the final judgment up to a clinician.   00;24;16;25 - 00;24;34;24 So, yeah, I remain terrified of human drivers. So in your role at Oracle Life Sciences, how is Oracle specifically leveraging these emerging technologies that we talked about like AI and big data to enhance drug safety and pharmacovigilance?   00;24;35;10 - 00;25;15;20 So there's a number of technologies and that's that's one of the benefits of being part of the broader Oracle, is that, you know, you kind of have all of these other areas and big areas of investment in AI and data science and high capacity compute and large language models and generative AI. And so so we get to it's like going to the toy store or something and decide which which things already have been built that you get to pull off the shelf and decide how we could apply those into our area of drug safety and pharmacovigilance.   00;25;15;20 - 00;25;45;13 And so, for instance, we just added the translation facility and, you know, out of the box in our Argus Cloud, you now have a translate button and it doesn't sound like a big deal, but if you were using an external tool before and then had to cut and paste and you were doing that 20 or 30 times within an adverse event report case to report it to local regions, just taking out that cut and paste and making it as a button straight in the system.   00;25;45;13 - 00;26;07;27 And by default we'll hook it up to the Oracle Cloud Translation Service. But if you wanted to hook it up to Google or you wanted to get up to a life science translation service, you could do that as well. Again, we're trying to look for where there are bottlenecks and we're trying to go out and look at where can we leverage an investment that Oracle's already making and then apply that into our specific field.   00;26;07;27 - 00;26;52;00 And, and part of that's what's exciting about our acquisition of Cerner is that, you know, I may have had a use case that sounded interesting in Pharmacovigilance. Maybe it's a case narrative generation or a case narrative is not all that different than a discharge summary for a health care record, or if you're doing a health care referral letter for referring the patient to a specialist and giving a summary of their their specific case and such, that's not that different than perhaps auto generating a letter that is a follow up request for collecting additional information on an adverse event case and so on.   00;26;52;00 - 00;27;17;28 Many of these there's there's overlap and we're able to team up with the teams that are focused on the health care use cases and add on our life science use cases and, you know, really benefit both teams or sometimes health care is leading the charge and sometimes life science is leading the charge. But ultimately that power together is like a multiplier, not not addition.   00;27;17;28 - 00;27;31;14 And and I think is a big benefit. And one of the big benefits of our of our acquisition of Cerner and the fact that we now are a leading health care company, in addition to, you know, what we've traditionally done in life science.   00;27;32;06 - 00;27;48;28 Yeah, there are a lot of industry players in life science. So is is what you describe what makes Oracle a real differentiator in the space when it comes to safety and pharmacovigilance? So things like combined assets and the Cerner acquisition.   00;27;49;11 - 00;28;25;29 Yeah I think there's there's a couple of things. One is sort of foundational with our cloud infrastructure and capacity there. For instance, we have high capacity compute and GPUs and just within our drug safety solution area, you know, we have two GP2 cloud instances available, dedicated 100% to our use and that's multimillion dollar worth of compute that we have dedicated to to our team of data scientists working to NPV.   00;28;25;29 - 00;28;58;05 And that would be difficult, not impossible, but difficult for a lot of other vendors to sort of dedicate that sort of compute capacity in such just to their life science use cases. Now the other I think is is around, you know, the acquisition of Cerner. So we talked about we now have a point of care footprint. So where, you know, clinicians are using Cerner software as the electronic health record when they're interacting with patients.   00;28;58;05 - 00;29;28;19 And so if we want to collect information as part of that point of care relationship, we can do that if we want to leverage, You know, we have something that's called the Learning Health Network that has a electronic health record, real world data asset. And so companies that our health systems sign on to use this because they they want a few benefits, they want access to clinical trials.   00;29;28;19 - 00;29;55;07 So they want their their patients and such to be able to be included within cohort selection and recruitment, site selection and recruitment for clinical trials. They also want to understand how they're delivery of care matches against other health systems across the country and eventually across the globe. So that they can sort of benchmark and compare how they're doing.   00;29;55;16 - 00;30;23;05 So that ends up creating this research data asset That, for instance, is very important for drug safety and pharmacovigilance, so that if you have a particular risk or an adverse event that's been reported against your drug or therapy, that you can then go out and say, well, is that just a correlation? Is there enough information within these individual cases to establish causality to the drug, actually cause that adverse reaction?   00;30;23;18 - 00;31;11;09 Or do I really need to go investigate that and understand its usage within the, you know, electronic health care record or claims data? And so so that's one of the areas that we are really focused on right now of sort of benefiting this better together with with the combined assets of and expertise between Oracle and Cerner is how can we leverage that real world data to understand and investigate risks that have been reported in adverse event reports to be able to go out and and understand real world usage there and and look at and understand how many patients are taking this drug, how many patients potentially had this reaction?   00;31;11;25 - 00;31;31;17 How many patients generally have this reaction not taking our drug, you know, understand those background rates and such. And so it's another level of understanding of the benefit risk once you have not only the adverse event reports, but the ability to research these within a real world dataset also.   00;31;32;03 - 00;31;38;03 Okay. I've got one more question for you. The all those warnings at the end of the pharma TV ads, is that because of you?   00;31;38;24 - 00;32;07;22 Well, ultimately, you know, I feel like sometimes we're plane name that tune or something. So a commercial comes on and I'll say, Oh, that's a pharma access to pharma y company. And you know, I'm usually right on naming the drug to that company. But, but it is, it is vitally important, you know, what is being done and where traditionally pharmacovigilance has sort of been a retrospective.   00;32;07;22 - 00;33;02;11 What can we learn after it has occurred? We're really trying to move towards what or labeling as precision pharmacovigilance, which is better understand that safety profile, better understand that risk benefit profile, not at these broad population levels that might be by by gender and age group, but getting down to smaller and smaller subpopulations and ultimately ideally to be able to go back and impact proactively the care of an individual patient where we might be able to identify based on a certain patient characteristics, a patient history, genomic marker, current labs, other concomitant medications they may be on presently, that maybe there is a higher risk to that individual patient of therapy versus therapy and provide that   00;33;02;11 - 00;33;39;18 information to the clinician that's treating the patient at that point of care. So so we intend to continue to drive towards that advances in drug safety that can improve overall population level help, but want to drive that down to to the ability to inform care around an individual patient. And thus, you know, when we see and hear those commercials and we hear the list of adverse events that are potentially associated with that drug, to give us better context, to say, well, what does that mean for me as bruise versus what does that mean for Mike?   00;33;39;18 - 00;33;51;15 And maybe one of us needs to be concerned and maybe one of us doesn't, and wouldn't that be great rather than just hear the list and and know that randomly that might be meaningful or not so obvious?   00;33;51;15 - 00;34;08;19 It's a vital part of drug development. And it's been interesting to hear what approaches are being taken and who's leading them. We appreciate you being on the show. For those who are interested in Pharmacovigilance and their interest has been tweaked, is there any way they can connect with you or get more information on what's going on?   00;34;09;09 - 00;34;51;14 So for me, I can be reached at [email protected]. If you're on any one of your web search engines, you could just search on Oracle pharmacovigilance. The other is that we do have a community that we call the Oracle Safety Consortium. So if you search on Oracle Safety Consortium, you'll come up with and that's sort of our end user community where we have regular monthly events and such that are discussing industry, but as well as Oracle Solutions and how we're addressing the needs of industry through this sort of peer consortium group as well.   00;34;51;14 - 00;35;00;09 So those are sort of three ways that you could either follow up with me individually or learn more what we're doing here in Oracle for drug safety and Pharmacovigilance.   00;35;00;24 - 00;35;29;25 All right, we've got it. And if you want to see if Oracle can accelerate your life sciences research, just head over to Oracle dot com slash life dash sciences and you'll probably find out what you need to know. Don't forget to subscribe to this show and join us next time for Research in Action.