Deep 6 AI CEO discusses transforming clinical trial recruitment on AWS’s Podcast

Deep 6 AI CEO discusses transforming clinical trial recruitment on AWS’s Podcast

Wout Brusselaers, CEO & Founder of Deep 6 AI, is on episode #72 of Amazon Web Services (AWS)’s Health Innovation Podcast. He spoke with host Alex Merwin, Head of Growth for Healthcare & Life Science Startups at AWS, about how Deep 6 AI emerged from The Cedars-Sinai Accelerator and became the industry leader in accelerating clinical trial recruitment by using artificial intelligence (AI) and natural language processing (NLP) to mine structured and unstructured electronic medical record (EMR) data, precision matching patients to trials.

Alex and Wout discuss topics including Wout’s founder journey, revolutionizing patient recruitment, and the site-sponsor paradigm. Listen to the episode for more about how Deep 6 AI is transforming the field of clinical research with the ultimate goal of getting novel therapies to patients faster.

Video transcript: 

Wout Brusselaers (00:01):

It easily takes an hour per patient to review a chart for a study to see if patient’s there. We can bring it down to a couple of minutes. That is all the more time they can spend with actual patients for recruitment, and that is the most important thing.

Alex Merwin (00:13):

Welcome to the AWS Health Innovation Podcast where you can learn from entrepreneurs and investors who are driving progress in healthcare and life science around the globe. Hello and welcome back. I’m Alex Merwin from AWS, and in this episode, we welcome Wout Brusselaers, the CEO and founder of Deep 6 AI. To the show, we’ll learn about their innovative and transformative approach to recruiting patients to clinical trials and while’s founder journey, including the important role Cedars-Sinai Accelerator had in developing their initial business thesis. AI is having a big impact across health innovation, and this conversation really brought to light how the technology is accelerating clinical research, not potentially, but right now, ultimately getting new medicines to patients faster. Let’s go. Wout Brusselaers, the founder and CEO of Deep 6 AI. Thank you so much for joining me today on the AWS Health Innovation Podcast. How are you doing?

Wout Brusselaers (01:09):

I’m good. Thanks for having me, Alex. It’s a pleasure to be here.

Alex Merwin (01:12):

So where are you joining us from?

Wout Brusselaers (01:14):

I’m calling in from Pasadena, California, which is sometimes referred to as a suburb of Los Angeles that’s probably not doing it justice.

Alex Merwin (01:23):

Yeah, fair. And where are you from originally? When did you move there?

Wout Brusselaers (01:26):

I’m originally from Belgium in Europe and I moved here how close to 20 years ago, so it’s been a long time. I lived in Asia and Middle East before, so I kept moving east until I found myself on the west coast of the US.

Alex Merwin (01:39):

A true globe trotter. Yeah, I’m from Colorado and I moved to the UK in 2015 and I’ve always found it’s useful to live in another country just to know that the way that you did it growing up isn’t the only way to do things. You can’t see all the world, but just knowing that there’s another way is quite helpful I find.

Wout Brusselaers (01:57):

Exactly everything is contingent.

Alex Merwin (01:59):

Yes, that’s right. Well, Wout. Can you introduce us to Deep 6 AI, who your customers are and what you all do?

Wout Brusselaers (02:05):

Yeah, sure. So Deep 6 AI in a nutshell de-risks and accelerates clinical research for hospitals or sites, life sciences companies, often referred to sponsors and CROs. So basically we de-risk and accelerate clinical research for all the major stakeholders in that process. And we do this by connecting all of those study stakeholders that I just mentioned on a real-time clinical trials acceleration software platform, which uses AI to precision match patients to studies. So what it does to those clients or for those clients is it brings much needed visibility and access to all the eligible patients for a study across an entire health system. And that is, as you can imagine, very different from how clinical trials traditionally operated, where it was much more a relationship game, right? Physicians reach out to patients that they happen to remember or they know they’re fresh in their mind. Sponsors reach out to sites where they have good relationships or they worked with in the past who’s very much driven on a past history, again, contingency and access to relationships rather than a true data-driven process.

(03:11):

So by our ability to mine real patient data, which means all of the EMR data, including all of the physician notes, pathology reports, genomic reports, right? That’s where our AI comes in or the nlp. We offer a real-time view into how many eligible patients there are today at any given site. And that allows you then once you have that view, once you have the visibility, it de-risks the centers and the studies that you work on, and it allows the research teams and the physicians at those sites to use those lists of screen ready patients to start recruiting.

Alex Merwin (03:45):

Great. I can’t wait to get into the Go-to-market a little later and to learn more about where you sell into and how you get distribution with the sites and the integrations to the EMRs. But just to start, can you give us an idea of where you have the most market segment penetration, whether you think about that in terms of geography or therapeutic area? Where are most of your studies conducted?

Wout Brusselaers (04:05):

Yeah, no, that’s a good question. And lemme start. Geography we’re operating in the US early on we had a couple of clients in Australia, which was purely opportunistic. I wish I could say there was a big strategy to go to Asia Pacific, et cetera. We just ran into a maverick decision maker, a visionary actually at A CRO in Australia who helped brought us in there at a couple of sites during covid. They shut down research. We haven’t really picked it up again, it was just small. But we’ve been growing very rapidly in the US I think we can now say that we probably have one of the largest, one of the largest ecosystems of leading academic medical centers with access to unstructured data. Can talk later about some of the names there. In terms of therapeutic area, we are by design agnostic, our AI or platform mines all of the data regardless of the disease type or therapeutic area.

(04:55):

We can find patients for any trials, but we reflect the distribution of clinical research in the wild. Most of the money goes to oncology clinical trials. So you see many of our clients are leading cancer centers and we do a lot of studies there. Then there’s cardio, there’s neuro, there’s rare diseases we play everywhere. We’re representative of it at this point of how those studies are distributed in the wild and that in terms of the different stakeholders, because in a way we’re connecting all of these stakeholders, you could almost treat that as a marketplace where you have supply and demand and in a way the supply for us is those health systems. They’re the ones who are executing on the clinical trials. It is their patients that are going to studies and it’s their data that we give to them or we help ’em combine their data so they can actually find those patients.

(05:46):

So we spend the first couple of years of our existence in selling to those major health systems was quite an arduous journey. I think it’s not easy selling to health systems, very decentralized decision making and tight budget, but I believe it’s become a strength of ours and we’ve been very successful selling into them. And that’s again, that’s why I think we have a leading role there. And now we’re at that point in our journey where we have a big enough ecosystem that we can start selling into life sciences and that’s been more of our focus for this past year with success there as well.

Alex Merwin (06:15):

That’s great. We have a lot of different types of companies and founders on the show, but we have had a few who are focusing on driving innovation in the way that new therapeutics and diagnostics are clinically validated and brought to market. And I just think it’s so important because we’re really living in this golden age of innovation within biology. We’re truly gaining a very nuanced and detailed understanding of the many different s as these foundation models come online and more and more companies are using them to identify novel drugs. Unfortunately it doesn’t matter how tight we get with our patient stratification and how personalized we get with our therapeutics if we can’t clinically validate it. And today trials are just so grossly inefficient and expensive. It’s a massive bottleneck and this amazing innovation we’re seeing in the lab and driven by tech bio and computational biology.

(07:07):

It won’t get into the clinic, it won’t get into the clinic unless we really rethink the way that clinical research is done. Hat’s off to you and what you and the team are doing, but maybe we can learn a little bit more about you as a person. So you’ve really had a fascinating career. You’ve worked as a diplomat, a consultant, a finance leader, a founder, and a broad range of sectors. So intelligence, social commerce, now healthcare. Can you tell us more about your professional journey, especially the aha moment you had during your time at the Cedar-Sinai Accelerator and really what drove you and inspired you to focus on clinical trials?

Wout Brusselaers (07:43):

Yeah, I wish I could say it’s a very linear trajectory, but it was not at all actually probably to the despair of my parents early on. What are you going to do? I think I grew up in Europe and in what I call a very humanist societal background, right? We’re very idealistic about the pursuit of knowledge and doing something meaningful, et cetera. And I read a lot and I think I was influenced by many things. You have a lot of opportunities. There’s many things you could be doing with your life. So picking the right thing to do is important. And as I said by reading Octa Paaz and Jge pu and many other writers that happen to be diplomats as well, I thought this is an amazing environment, right? For people who have an interest in the world, they want to be surrounded by smart people.

(08:23):

You’ve got the center of politics and all of these things that are happening. So this is what I want to do. In reality, when I became a diplomat and I actually went to the Middle East, I greatly enjoyed it at first and there were many very interesting things about it, but it was a bit more bureaucratic than I had actually counted on. And maybe that was part of a self discovery. I found out that I’m not the most patient person or that I’m not that good in operating in bigger structures where you don’t have as much control and you’re just a cog in the wheel. And as I was actually diplomat, I met some people from consulting companies that basically asked me for some feedback on doing business in Middle East and I say, Hey, you should be consultant if you’re interested in working with smart people.

(09:04):

That’s where today smart people are, join us and work there. So I did that, joined one of the big consulting firms, learned a lot. I thought it was fascinating. Met a lot of indeed very smart people, but also felt like it was still a big structure and the pure pursuit of shareholder value creation was not really enough for me. I wanted to do something more and have more of an impact. I also, I wanted to have my hands basically on the steering wheel and be able to not just do planning and consulting, but also to execute. That led me to my next couple of steps was helping turning around a physical security company that operated in Latin America and Africa and other parts of the world, which was very interesting. And through those connections with state departments and other things, we were able to participate in a contests for the US government, which was basically unnamed but open to a handful of select smaller companies, larger companies.

(09:57):

We happened to win that contest and it was really based on our ability to mine massive amounts of unstructured data. They gave us basically a large corpus and they identified and so can you figure out what the topics are there? So some of that topic modeling and assign a set of priority and ownership, who should look at what when? So we did pretty well there beating out some much bigger names in what is now AI or machine learning and other sectors. We had some contracts with the government that came from that we’re for a while bootstrapped and we could have grown there. But I also realized that we were more a west coast tech company than an East coast government contractor. So decided to look for a different market and taking the capabilities and skillsets that we had built up in dealing with massive amounts of unstructured data to a different industry.

(10:43):

And because we wanted to do something meaningful and build a mission-driven company that meant for us either healthcare or education, which both I think are meaningful pursuits and they both deal with a lot of unstructured data, which I think you can use to provide a better service in each of them. Healthcare had the easier path and the greater market, and also was even a little bit closer to my heart still. So we ended up participating in the Techstars Cedar-Sinai Healthcare Accelerator program where we were suddenly exposed to over 300 physicians, nurses, administrators, payers, pharmas researchers, basically everybody. And we had an ability to mine unstructured data and we tried to quickly make sense of what is the best problem for us to solve? What do we do with this ability, right? And we got requests from can you design a better EMR with a hundred million dollars in three years of sailboat?

(11:35):

Yes, but that’s probably not us, right? Can you find an algorithm to detect patients with sepsis? Which yeah, maybe, but it’s a pretty small and narrow use case and we would be like a small plug in a better system. So we developed a framework that we applied to all these opportunities to say where do we have a truly large total addressable? Whereas there are big markets of opportunity that is still largely fragmented where there’s no clear winner where our ability to mine that unstructured data allows us to have a 10 x better outcome or product than others. And where we have a clear ROI story for our stakeholder or multiple stakeholders. And as we did that to the stacking, the opportunities became clear that clinical trials was that big unsolved problem that had a tremendously big market and there was no clear winner. Most companies trying to address it were services companies. There was very little products trying to deal with it. And we did believe because there were so many stakeholders, as I mentioned before, that there was a strong ROI for all of them, if we could find that one solution that would connect those stakeholders and offer them more efficiency, faster recruitment with fewer dedicated resources, more de-risking of all the efforts they put into that research. And that’s where we got started.

Alex Merwin (12:50):

So Wout, we have a number of people who listen to the show that are founding or thinking about founding new healthcare or life sciences startups. And there’s a really interesting lesson and an important one in your story, which is focusing on something that has enough leverage to have, so you mentioned how you saw many different opportunities, but you really wanted to find something that could 10 x the status quo. And I think that’s really important because there’s so much cost and change in healthcare, there’s a lot of liability. It’s very risk averse. Change carries a lot of risk with it, so it needs to be worth it. You really need to have a strong compelling value proposition in order to make it worth driving the changes within the organization. So maybe we can talk a little bit more about how patient recruitment was done before. Can you help draw a baseline of what patient recruitment looks like, how it works in a world without Deep 6 AI’s technology? And then we can talk a little bit more specifically about how your technology works and feel free to go into however much technical detail you’d like. Yeah, it’d be good to learn about the baseline, how patient recruitment worked and how it worked with your technology.

Wout Brusselaers (13:54):

Yeah, I know that’s a great question to start from Indeed. So traditional clinical trial recruitment, right, hadn’t changed very much from say the days of aspirin where you develop one drug or one pill, you give it to anybody or everybody who has a headache or pain or a fever, and they all respond pretty well. Just find any patient, put ’em on a study. And the way to do that is you ask physicians, which of your patients has complained about a headache? Can we put ’em on a study, reach out ’em the next time they come in? Or as we started digitizing more information, you try to find some of those patients through data-driven queries typically in relational databases, which can only find the structured data in healthcare, which is a very limited dataset with very little granularity or nuance. So it at best gives you a very big haystack of potential patients, but you still have to do manual chart review to see which of those patients meet all of the inclusion and exclusion criteria.

(14:48):

And in a way that kind of sums up what most solutions up until some fairly recent ones have done, they try to build bigger haystacks. So if I’m trying to find patients for a study, let me try to find as many patients first and they’ll see which of those patients are actually eligible. Social media campaigns can say, okay, if you have prostate cancer, go to this website and click on this button and then we will contact you, we’ll ask you some questions. We’ll see if you’re eligible for the study or you work with more traditional media campaigns or you do those query searches right in your database, you find any patient would say prostate cancer and you find out which one has all of the other components of the study. What we try to do that’s different is we really try instead of building bigger haystacks to only focus on the needles from the start.

(15:34):

And the reason why is that over the years, clinical trials have become a lot more complex. Finding patients for aspirin is fairly easy because any patient regardless of age of gender or whatever that as a headache has a complaint. You can put ’em on a study that is very different for some of these precision medicine drugs that require a very specific phenotype and genotype of a patient for them to work. You do not want to put your study at risk, but you have to submit data for the FDA to show what the efficacy and the safety was of your study if you actually targeted the wrong patient population. And that has negatively impacted the success of your drug on those patients. So you want to be laser sharp in which patient makes it onto your study because they’re the ones that are going to respond to the drug and they’re the ones that the FD is going to look at to see can we actually expand this to a wider market.

(16:19):

So the way that we try to bring that precision to clinical trial recruitment is, as I said, by looking at all of the patient data, not just the structured data, but also including all the much more nuanced data that is typically in physician notes, in pathology reports, in genomic reports, and in so many other places that traditional software cannot really mine. And what we do is we go and we work with those health systems that have all of that data that we spoke about before. We ingest all of that data actually onto our AWS cloud platform where we run all of those. We first go in for historical data, then we update that nightly or in some cases we can update it in real time depending on the data protocol that they use, like fire HL seven. And we turn all of that data for each patient basically into a vector that represents a patient.

(17:12):

So we read the notes, we use NLP to find all the relevant clinical concepts in those notes, right? This is a symptom, this is a disease, right? This was a test, this was an outcome. And then we basically labeled those and reduce all of those texts, all of those notes, all of the fields in those relational databases into that patient vector. And they’re done a sum, a multi-meter representation of all of those concepts. And those are now an ideal sofa or subject of analysis. You can search against that. And then you can start from a protocol for clinical trial with inclusion and exclusion criteria that says, fine, we all the patients prostate cancer without neuroendocrine differentiation who’ve shown up in the last six months for this test and who’ve had a radical prostatectomy or have this failed line of therapy and who do not have metastatic cancer and are not on this drug.

(18:02):

So you can then actually use all those criteria to mine those vectors and reduce ’em to dose patients that are kind of precision matched to that protocol. Once you have that, and we can do that very quickly, we ingest a lot of the data in real time. It’s searchable immediately in real time. You can now start from that precision cohort of needles or eligible patients and you can share that cohort with the research team so they can actually go and reach out, which you can also alert treating physicians who may not even be researchers who may not even be at the same site, but they’re part of your network that maybe a couple of miles away from you and you don’t know them personally and you didn’t know their patients. But now you can ask those treating physicians if they can suggest your study to their patients.

(18:42):

And we we’re actually doing it now inside of the EMR for a couple of clients. We’re rolling belt inside of the epic EMR, so a physician can get a notification, Hey Alex, the patient that you’re seeing now about is eligible for the study, here’s information about the study. Are you okay referring them? So without trying to take up too much time from these valuable consult that you have with the patient, we want to quickly bring a study to your attention, allow you to click and move on. That then allows the research team to reach out to the patient, plus the patient can receive some information about the study themselves and now they’re interested as well. So again, it’s that visibility into the needles, right? The patients are truly a match and then access to those patients beyond just the PI themselves.

Alex Merwin (19:19):

At what point does your team get engaged to identify the needles? Are you getting involved in providing feedback to inform the protocol design or is the team coming to you with the trial ready to go and you’re just driving recruitment? Because I imagine that some of this, the vector database would be very valuable for even getting feedback on the protocol design of the study. Does that ever happen?

Wout Brusselaers (19:40):

Yeah, absolutely. At some of our sites, the PIs or the IRB is even using our app to see whether they should even take on a study or start a study when they say, we have an idea that maybe we should look into patients before they’re diagnosed with Parkinson’s and maybe we can do something there. So they can actually say, do we have those patients? Do enough of those patients come in on a regular basis so we can build a trial around them and before the IRB will even approve or greenlight a new study, they have to show that those patients are there and they’re basically addressable and you can basically serve as them on the sponsor side. So the pharma companies and biotech companies, et cetera, it’s the same thing when we started last year working with pharma companies. Many of them basically brought us rescue studies like a study that has already failed in so many cases and it’s a Hail Mary, can you help with this?

(20:28):

And what we learned there is that too much damage has been done. In many cases the trial design wasn’t great or it had certain inclusion exclusion criteria that stops eligible patients from participating or the protocol itself was so hard, it required patients to drive too many miles and show up for visits too often. Or sometimes there is a competing drug already in the markets and treating physicians are not ready to try a new non-woven drug on their patients when there’s something else that seems to be working. So we took on a lot of that risk when it was too late. And indeed we learned from that. Well, rather than only working on studies when they’ve already been launched and started, yes, we can help you there if the protocol is good. Let’s also take a look at those studies that you still have in design and see if the patient population that you’re comfort on is really there.

(21:16):

And where I think we can play a role there is in that during the planning phase of many clinical trials today, state of the world, and going back to your previous question, what is the racial way of doing things today, much of the planning and study design is done by using real world data, and that’s already a step up from the relationships. The problem with real world data in a way is that it’s normalized, aggregates, flattened data, right? You’ve lost so many degrees of freedom, so many insights, you’ve lost many of the nodes. You try to normalize it to the most common denominator and those patients are de-identified. So you can design a trial on that theoretical population of patients, but once you start selecting patients and sites or sites to recruit patients, you don’t know which real life patients are matching those criteria and how many of them are there.

(22:01):

So what we are saying is well use one and the same database for your observational study design and planning and for your patient recruitments. We can tell you that using our patient vectors, how many patients are there today across the site system that meets your current study design, we can allow you to make some tweaks. If you change your third I exclusion criteria or your second exclusion criteria, you broaden your patient population, you make it more accessible, or you do not disproportionately exclude patients from a certain demographic. If you actually do not insist on not having patients who had a recent say, a viral load in some cases, then you can actually allow for access to more patients across a certain area or whatever. So tying the planning back to the execution on the same dataset is definitely helpful to do that and to ensure the success going forward of being able to recruit those patients that you actually design your trial around.

Alex Merwin (22:56):

That’s fascinating. And the example of the PI or an IRB will use your tool in order to get early feedback on a proposed study design and then decide to either change that study protocol or to abort it altogether because of information they get. That is groundbreaking in and of itself because you’re saving so much cost in inefficiency by you’re not launching a trial that’s going to fail and most trials fail. And it’s not just the sponsor’s money, it’s the PI’s time and it’s the ClinOps team, it’s the delivery nurses not to mention the patients. That’s really incredible. Can you tell me a little bit more about the dynamic between the sites and the sponsors and how that relationship usually works? And given this is such a new technology and you’re really changing the way that the trial was planned and conducted, are there any meaningful changes to that relationship now? And this might be a good time to talk a little bit about your business model and how you monetize your offering at Deep 6 AI

Wout Brusselaers (23:49):

Ai. Yeah, no, again, great questions Alex, and you’re getting at the heart of some of the change that is required, right? As you mentioned before, healthcare actually carries a premium risk to innovation. If you want to innovate in financial services, maybe you ruin the economy. That’s one thing. If you want to innovate in healthcare and you do something wrong, patient lives are at stake, right? So the barriers are greater. That is why clinical trials exists to make sure that any innovation in treatments, any new device or treatment or procedure is tested to make sure it’s safe and that it works. But because of that, there is an apprehension around how quickly and how far you can innovate. Also, many healthcare organizations and pharma companies are big companies with tens or hundreds of thousands of people. So affecting meaningful change at a major level there is not that easy.

(24:41):

And what we have to do is not just build a great product and hoping that people will come and use it. We have to make sure that we actually help people change and adapt and adopt the change. And the most important thing there is we have to show immediate ROI to all the stakeholders and show like, look, this change is going to make your life easier. And back to the point that you made is when you take on a study, it is not just about the money, it’s also you have very limited resources. You only have so many people and so many hours and so many patients that you can spend time on and you can spend the time on a study that will fail or you can spend the time on a study that does not failed. So selecting the right study to spend those scars resources out is super important.

(25:23):

And what we’ve seen at many of the sites is that the sites have a very quick innate sense of which study has legs and which doesn’t. They’ll quickly come to a conclusion like, yeah, this study, they’ll give it a try for a couple of weeks If it doesn’t work, if it’s hard to find patients, they’ll just give up on it and find another study to work on because they have to be ruthlessly rigorous in where they spend time. They cannot afford to spend time somewhere that’s not going to work. What we can do is, again, de-risk, that whole proposition for both of them, we can help de-risk the studies that sites accept. So they do not have to come to that conclusion later while they’re already committed to something and work on that and have upfront feasibility and visibility to screen. We can also make sure that they actually have more time and resources available by taking some of the most cumbersome effort they to do, which is manually screening patients and their charts to find those patients and map.

(26:15):

If we can actually take care of that side of the business, it easily takes an hour per patient to review a chart for a study to see if patient’s there. We can bring it down to a couple of minutes. That is all the more time they can spend with actual patients for recruitment. And that is the most important thing. So we have to make sure that as we deal with those two different stakeholders decide and the sponsors that one, we derisk for both of them, and two, that we mitigate and set the right expectations, involve some delicacy. We have to act in the interest of both of them. We have to understand the limitations that sites have in their very busy schedules and how much time they can spend on any given study. We also understand that sponsors expect to see results. Traditionally, there is a veil between those two stakeholders, right?

(27:01):

Sites don’t like to have sponsors see what they’re doing because they feel, ah, that’s just intrusion. Let just do the work. I’m too busy to do all of this. And they don’t necessarily want to know which studies they’re going give up on which studies are going to continue on immediately because maybe they’ll revisit that later. Maybe we’ll find more patients that. So part of that change that we’re bringing is more visibility and we have to show that the visibility is worth it to both stakeholders, not just to the sponsors who see what the sites are doing or not doing, but also to the sites themselves, like having that greater visibility or report on it being rigorous in that also puts a little bit more on the emphasis back on the sponsor to make sure that their trial design is correct and that they’re doing the best they can to help you find patients for. So it’s a combination of change management and delicate balancing of the incentives and interests of both of the stakeholders.

Alex Merwin (27:50):

And ultimately you’re enabling a feedback loop that’s tighter than it was before. And so ultimately the sponsor and the site, they need to work effectively together. And the more shots on goal, the easier that’s going to get. I might’ve missed it, but did you talk about your business model?

Wout Brusselaers (28:04):

Yeah, no, sorry, I was going to get to that next two. Our business model in a way is that we sell our software to the sites. They are a stakeholder and we want to make sure that we meet their incentives and we offer our I to ’em. A typical site or healthcare organization will do from a third to more right of other studies for industry. Another third will typically come from government, right? NIH sponsored studies or other agency sponsored studies. And another third will be investigator initiated. That’s typical for most academic medical centers with some changes. Some smaller sites may pick one over the other may focus mostly on industry sponsored studies because that may just how they’re set up or they don’t have the capabilities to do their own investigator studies. That’s the mix. So we sell our software to them and we give or we have our support teams that help them be successful, we help ’em build the queries and quickly get to a list of screen ready patients for their studies.

(28:58):

Now on those industry sponsored studies, we are more and more also working with the sponsor where we’re trying to lift some of the burden or push it back to the sponsor, help them with their study design. So the studies that they actually bring to the sites are vetted and are better and help them select the sites that are known to have the right amount or sufficient amount of eligible patients for their study. So do risk that as well. And then what we try to do is assist the study teams on the site to be able to start recruiting from that list of screen ready patients as quickly as possible with minimum times spent on manually doing chart review or others. And what we do as well is we make sure that site study team can access patients beyond the patient being treated merely by the pi.

(29:42):

If the PI was selected by a pharma company, that PI typically takes care of about a thousand patients. Let’s say 20 of those patients may be eligible, but if there’s five other treating physicians in the same therapeutic area and they have another 100 patients, we’ll make sure to now access those over 100 patients as well. Because like you said before, it’s a numbers game. If you can increase the end, increase the pool, then the number of patients that may actually accept a study and that may feel that it fits in their life and in all the other obligations that they have will grow. And that’s how we connect the relationship between both of those stakeholders, the sites and the sponsors, and try to align their incentives to bring value to both of them.

Alex Merwin (30:19):

Wout, you’re seven years into this, so now you’ve got a good amount of liquidity on the patient side and the site side, and you probably have a great working motion with many prominent sponsors. So you’ve earned a lot of trust. And I’m never saying it’s easy go to market’s, always tough, but I’m always interested in zero to one stories and with any marketplace product, and I hate breaking it down and being reductive, but if you don’t have patience, there’s a limited value that you can add at sponsor. So can you share the story about how Deep 6 AI secured its first customers? So rewinding the clock a little bit, what did those first customer engagements look like and how has your go-to-market changed over time?

Wout Brusselaers (30:56):

Yeah, I know that’s a great question. I think it’s an important one as well. You’re right, we had to build up the supply side. We had to have product on our shelves so we could open the doors to our clients, and that’s why we had to build up that patient database and that side database. And our first client was Cedar-Sinai, thanks to the strong relationship we had built up through the accelerator and to the fact that they are very much an innovation first kind of organization. And I think you’ll probably hear that from any of your podcast guests when they talk about it zero to one. Your first clients are typically the mavericks, right? They’re the people that have a vision, that have willpower and that are willing to take a risk to realize their vision. They’re willing to do things that many other people in their positions are not ready to do.

(31:39):

So I think it’s really important that you find those early mavericks, you build a relationship with ’em, you understand their vision and you try to help ’em realize it. So at Cedars, we had met a couple of researchers that were trying to do studies, and we actually took their studies into our very first Alpha product that was very limited and we let ’em play with it, and they very quickly said, oh, I just build a list, or I asked it for a list of patients for my non-small cell lung cancer study, and they gave me list of 3000 patients back and I had to manually go through 3000 patients to whittle it down to maybe a hundred patients that are truly eligible that is going to take me months because I have to spend an hour per patient. I cannot get to that. And when they played around with our tool, said the tool isn’t perfect and in terms of ui, but within 10 minutes I got to my list of a hundred patients here.

(32:26):

So I see, and I recognize some of those patients doesn’t make sense. They were willing to put their name and their drive and reputation behind it. They were going to start lobbying for us. So the first deals were really us selling vision and capabilities to some of those mavericks, them running with it and basically doing a lot of the legwork themselves, pulling a budget together, getting the decision makers together and getting us to that first implementation. And we learned a lot from that. We learned a lot from the product use itself, but also how to get access to the data, how to build the pipeline up, how to get telemetry involved and everything, how to get our cloud environments to be responsive enough and to be elastic enough to fire up. We had more users, so there was a lot of learning that we had there once we started going to the next generation of clients, so maybe the early adopters beyond those mavericks, right?

(33:15):

We could take that and we felt like a lot more variability and we had to deal with that and build a program around that. Where we are today, which is the next set, we are in a much more RFP kind of process where there are other companies as well. It’s a much more structured process and what it means is in many cases now it’s us dealing with the bureaucracy rather than those mavericks. So it’s first shielded us from IT and they guided us and they took it all again through tremendous vision of willpower. Now that is our team was to deal with that, we have to go through the IT audits, we have to go through the RFP processes, we have to talk to multiple decision makers, et cetera. And it’s a whole different mindset. It’s a whole different skillset and it’s a different organization. We have to build around them.

Alex Merwin (33:56):

We’ll come back later to how you think about culture and scaling and hiring. There’s important stuff there, but I want to talk about your customers a little bit. So at Amazon, we’re customer obsessed and we’ve talked a lot about the technology, but we really haven’t quantified the impact. So maybe you’re going to share one or two customer stories, maybe somebody who able to get a treatment to market faster using Deep 6 AI, bring some numbers to it, what kind of impact have you had?

Wout Brusselaers (34:18):

Yeah, we track numbers right across a variety of metrics. One is indeed your ability to quickly say yes or no to a study and save time there. The next is then how does that translate into the number of failed studies that you have? And we have evidence that is coming from individual studies, right? And individual departments. I haven’t had a single failed study since I started using Deep 6 AI, et cetera. One or some of the organizations that have a more centralized organization for clinical trials, they’ve actually done some of that work for us where they actually started looking at organization-wide impact, whereas some of our health centers are more decentralized and we work for instance with the ology department or with different departments within the ecology department or even with PIs within the department. And the impact cases that we get are typical for a specific study or the startup of a study or specific departments or specific pi.

(35:10):

This one use case of one, the leading research centers in the world. Basically they had looked at the studies that they set up on our DB six platform within the first 10 months of our go live, and they basically at triggered about 2,600 patients back to our platform. So they said we found 2,600 patients on the platform and when comparing that with some of the other ways that they were looking for patients, they found about 600 of those patients were unique to us. So they would not have found those patients without us, which was really good. They also found that they did some time analysis of one team there that the amount they spent every week in looking at patients’ charts and doing a chart review dropped by about 80%. They were able to actually get through their entire patient population of patients coming in that week and prioritizing who to reach out to for their study in less than two hours where that used to take them more than a day.

(36:02):

So that was very significant. The other thing that they saw, and that is going back more specifically to just one specific study, is that sometimes when you take on a study, you do feasibility, say, we’re going to recruit X number of patients for this. That is based on the number of patients that you think are realistic. Sometimes you’re right and sometimes you’re too conservative, sometimes you’re too aggressive, et cetera. In this case, the number that they had set for this study was five. We were able to find many more patients and they went back to the sponsor and they negotiated a competitive enrollments target so they could say, if you can find more patients, you don’t have to stop at five. You can keep enrolling for that study. And what they were able to do is fairly quickly enroll 13 patients for that study, and this was a rare disease study, which was a hard study to recruit for, and as a result, they were able to bring significant management fees back to that team for that study because of how hard those patients were and just the ability to take it from five to 13.

(37:02):

That alone, I think paid for a couple of years of the license fee of our platform. So really good metric when it comes down to shaving time of a study, that is our ultimate goal. What we would like to do is indeed being able to say we are working on the study with a certain number of sites and we were able to shave off six months or a year or more. Unfortunately we’re not there yet. What we would need to do there is start at the right point in the study. We’ve only been doing these studies with life sciences company since last year. That doesn’t give us enough time to be involved from the start where we can impact both study design and planning and site selection and recruitment planning alone can take two years, right? Site selection can take a year or more. Recruitment can take multiple years. So we’re too early into the journey to do all of that. We do believe, however, that’s where eventually we’ll be the strongest. Having that focusing on the needles upfront from the study designed to the site selection, PI selection, and to the patient recruitment even down to the data collection and submission. Because we are so integrated in all of the data, we can just get better access to data faster. That I think adding all of that up, that’s what we’re really aiming for is shaving down study times by at least a year or more.

Alex Merwin (38:10):

So let’s talk a little bit about your technology. So you mentioned vector databases and you mentioned running AI to create a representation of patients that is more nuanced than you would get through a traditional structured database. And now vector databases are very popular because they’re commonly being used to develop knowledge graphs to generate information to pass into an LLM through retrieval, augmented generation, and it’s a way to ground truth and ringfence the wild stuff. You get out of these models. And so it’s a very exciting technology, but you’ve been doing this for a while, so it’d be great to just hear you explain the difference between natural language processing, different types of ai. We might as well include generative AI within that and how each plays a role in accelerating clinical research.

Wout Brusselaers (38:52):

Yeah, absolutely. You’re right. The large language models that are basically supporting generative AI are initially NLP models, right? And the bird models, et cetera. The way that we use generative ai, the way that they’re being used to complete sentences and then paragraphs and stuff like it lend itself really well to basically creating new text from new content. That’s true, that generation there, but you can also use them in a more traditional NLP sense where you’re using this more for classification and annotation to get a deeper understanding of natural language and retrieve information from it with greater accuracy, so with better precision and recall. So it’s almost closer to search to a search problem that you’re trying to solve for the generating new text. We’ve been using different models actually for that kind of NLP thing, SDM, other deep learning originally, even SVMs and other things to help augment our precision recall.

(39:47):

And LMS are another way that you can basically structure raw embeddings around, like you said, right? We’re doing that as well. LLMs are of course pretty expensive. There are cheaper models. You have to use ’em judiciously in small focal areas where you think you can get a significant delta over your other methods to push your precision A recall. And we’re doing both of those. What we’re doing as well, and I think that’s a more unique attribute of the generative AI interface itself, is using that to create better summaries of the data. So we use the embeddings and the other use cases to get a better understanding of the data and a better precision or recall to help us with better search. And then an LLM in its more generative interface can then summarize that data for easier UI and easier absorption and adoption by users. The other interface that it can also simplify is assisting in building your search query in this first place. So basically when you take a protocol, a clinical trial rights, and you take that through an LLM, you can actually use that to help derive or summarize the right clinical concept from that to power your search or your query right, and return your results. So reduce the amount of user interface that is required. That is something that we’re working right now as well.

Alex Merwin (41:05):

I almost think we should UB generative AI to participatory AI because this part about the interactivity is I enough people pay attention to this, and that’s the big unlock it. Is it cool that I can generate all this texture images? Sure, but you really begin to play with these models and you realize that it’s much more useful to say, ask me 10 questions to prepare for this project I’m working on. So it helps you actually think through a problem, and I can appreciate how in the usability of your product and helping people think of the right questions to ask, it could be very useful.

Wout Brusselaers (41:38):

Absolutely. I love that term. Exactly. I love how you say that indeed, the participatory AI is that it’s almost like it’s a synthesis between the human expertise that you have and the AI’s capability to quickly apply it to a massive dataset and return it. And by having that feedback loop as you formulate the questions or the problem that you’re trying to solve for and having that feedback loop may be very intuitive. So you don’t need to be a coder. You don’t need to actually write algorithms to do it, but you can have actually an AI recognize what you’re trying to go for and feed your results and allow that feedback. That is exactly how you’re going to get wide adoption and how you’re going to make the complex much more intuitive for more users.

Alex Merwin (42:19):

So Wout, you and I are blasting through this interview. I’m personally having a lot of fun, but we do have to start coming to an end, and I want to end with advice, and I want to end it through the lens of leadership, development and culture. So can you share a little bit about your path as a leader through your career and thinking about the journey of Deep 6 AI? What advice do you have for other startup founders and how they should be deliberate and intentional about culture and just any learnings you have?

Wout Brusselaers (42:45):

Well, that’s a one hour question to end on. I’ll try to take your time. Yeah, I’ll try to be somewhat concise there. Let me maybe start with the first thing about the leadership journey, right? What I have learned and what I have seen change in myself is that originally when we were a small team, I was very much in the weeds and doing a lot of things myself. I was very heavily involved in products, in selling, in customer success, in everything, and everything I care about was performance. And again, me being not the most patient person, wanting things to happen quickly, I was very much, we can do this better, we can do this faster, et cetera. As you grow a bigger team, you cannot be as much in the front lines anymore all the time. You have to build a team that can drive their own performance, and I had to become much more focused on motivation rather than pure performance and somehow turn that motivation into performance, right?

(43:39):

Motivate to have other people drive their performance rather than say, this is good, but I’ll do it myself or I’ll do this or I can do this faster. I had to take myself out of it, and that is definitely something that I feel has changed and an insight that has unlocked a lot of things for me. I’m not sure how actionable that is for other founders because some people may already have figured it out on day one and may not have had that as an issue. What I do believe is important are two things that we touched upon before. One is that mission, making sure that you build a team that cares about the mission, making sure that you have those missionaries around you rather than mercenaries, because those are the people that are going to go the extra mile. They’re going to actually put in the extra effort when times are harder, when things are tough because they want to do well for your users, for your clients, for those patients that you can actually help get access to better drugs faster, so they care little bit extra. The second thing is to really think about your values, both as a leader but also as a team. What are the things you really care about?

(44:40):

If you try to organize an organization around rules and processes and everything that’s really hard. It takes up a lot of effort and a lot of trial and error, and it may actually stifle innovation on all things. If you can organize your company around a handful of rules that help you make decisions in complex situations where you are dealt with two less than perfect choices, you pick the one that adheres to closer to your value. It empowers people on your team. It gives ’em the autonomy to actually make decisions guided by a smaller set of principles rather than having to go back to authority or to processes or guidelines, and you get more autonomy, you get more purpose, and I think you get more out of your people. I’m not sure how helpful that is. It’s a small piece of the bigger iceberg, right? It’s a small tip.

Alex Merwin (45:29):

It’s very helpful. Wout. What’s your ambition for the team? What do you hope to accomplish over the next few years?

Wout Brusselaers (45:36):

What I really hope is again, comes back to the mission. We want to help bring lifesaving cures to patient faster. So we want to indeed be able to shave off the time years of the time that it takes to bring new drugs to market, and for us, we hope that our platform will become a must have tool for all pharmacists. It’s like, why would I not have access to a platform that gives me upfront visibility into real patient-centric data? When I plan my study, when I select my sites and when I execute my studies, so I know that the people that are now going to go and recruit patients based on my study protocol, they or their data was involved in designing a protocol, and they already know every patient. They know every patient within their network by name on day one of the study. All they need to do is click on the list and they see a list of patients that all they need to start recruiting and not just recruit the patients that they as a PI have access to today. Every patient in the organization, whether they’re miles away treated by another physician or patients that don’t even know about the study, but we can send information about the study to through their patient portal. It’s really making clinical trials much more accessible to every patient, physician, and researcher in the system.

Alex Merwin (46:45):

I got to say, Wout, thank you so much for all the work you’re doing. We really need this innovation right here to what we were talking about earlier with all the amazing innovation coming down the pipe, and we’ve got it clinically validated, so keep up the great work. Really inspired to hear more about your story. Thanks for joining me on the show.

Wout Brusselaers (47:00):

Thank you, Alex. This was great.

Alex Merwin (47:03):

Thanks for joining us today for the AWS Health Innovation Podcast. If you want to get in touch with AWS, please check out our show notes where you can find a link. If you enjoy the podcast, the best way to support us is to share it with your colleagues and friends. We also really appreciate your reviews and ratings wherever you listen to podcasts. We love hearing feedback from our listeners, so please don’t hesitate to get in touch. Again, you’ll find all the details in our show notes. See you next week.

 

Media Contact

Laurie Thackeray
laurie.thackeray@deep6.ai
(919) 619-3105 

About The Author

Nour Malki