In this podcast Mitchell Goldburgh, SR. Director of Solutions – Enterprise Imaging & Advocate AI Services from NTT DATA, and Rob Hauser, the Chief Business & Analytics Officer from Vidence, discuss applications of datasets that extend oncology data from a large national cancer center with radiology studies and reports such as chest CT scans of lung cancer patients. The discussion explains why synthetic control arms and label expansion studies are important and explores how advanced oncology data can reduce the need to recruit patients into clinical trials. The panel also provides insight into the process to build new digital diagnostic tools based on radiomics data that are being developed to predict which patients will respond to personalized therapies. You will also learn about the multiple approaches NTT DATA has put in place to make the most of unstructured data within the NTT DATA environment including feasibility studies, abstraction, machine learning/AI, and annotation by radiologists.
Hello everyone, this is Dan Housman. I’m here again with the Novel Cohorts podcast with Mitchell Goldburgh, we talked recently with him about the work we’re doing together, NTT DATA and Graticule, regarding radiology data coming from the VNA (Vendor Neutral Archive) and beyond. We are also today talking with Rob Hauser., Mitchell and Rob can help introduce the work they’re doing together between NTT DATA and Vidence.
Mitchell Goldburgh 0:34
Thanks Dan. Yes, Rob Hauser from Vidence and NTT DATA are very excited to have this partnership in place. We’ve been working together for the past year on processes, tools, and projects around cancer oncological data. We’ve developed a platform that integrates the Vidence data with third party data and medical images. We are very excited to have both the clinical access and expertise that Vidence brings. Rob, with that, why don’t you tell us a little bit about Vidence.
Rob Hauser 1:17
Hi everyone, Rob Hauser. I am the Chief Business Development and Analytics Officer at Vidence. We’re excited to be working with NTT DATA and Graticule; this has been a great partnership in the ability to add value into our clients, as well as, more importantly, impact the way patients are cared for and in really driving better outcomes in the oncology setting. Vidence is a relatively new organization, built on a foundation of lots of experience in RWE (Real-World Evidence) and HEOR (Health Economics and Outcomes Research) in oncology. The organization grew out of Cancer Treatment Centers of America, an organization that has five hospitals throughout the United States; very innovative cancer care, going back over 30 years, being very patient centric, which has allowed for the capture of a lot of very good robust structured patient centric data, including everything from patient reported outcomes to all of their lines of therapy, as well as genomic information. Mitch, as you mentioned, the images and being able to really tie all of that information together gives us a very deep robust dataset to do lots of different things with.
Dan Housman 2:48
So Mitchell, I’m sure people are wondering, how can we figure out feasibility for these kinds of projects? What’s possible, What’s not?
Mitchell Goldburgh 2:55
Well, I think that this team represents a great opportunity for our audience. Because we have a data cooperative, and we have integrated our services together to come up with challenging questions and understanding what type of data is available. More importantly, what insights can be derived in a simple feasibility study is relatively short-term response, rather than having to recruit all the sites and clean up all the data. Our tools and processes allow us to provide summaries in a relatively short time, that can lead to further questions for us to answer as well as full-blown synthetic studies.
Dan Housman 3:55
So let’s maybe discuss some of the applications of the data. I think we’ll start with looking at what we can do with regulatory filings, because we’ve seen at Graticule an increasing interest in being able to both label expansion studies and synthetic control arms in oncology. Maybe we just discuss what are some of the areas that you think are interesting there.
Rob Hauser 4:24
Yeah, so I think there’s a lot of interesting opportunity with using Real-World Evidence for label expansion and synthetic control arms. There are obviously challenges around it. It’s a little bit new in terms of what the FDA (Food & Drug Administration) is looking for and how that opportunity continues to grow. I think there are a lot of very good robust datasets out there that need to go through sort of the regulatory paces to make sure that the data being utilized is the right data to drive those and protect patients and get the outcomes that we’re looking for.
Dan Housman 5:08
So Rob, I’m not sure everyone’s familiar with what a synthetic control arm is so maybe you could provide a little more background on it.
Rob Hauser 5:15
Yeah, sure, be happy to. So, you know, again, the ability to speed up the knowledge and learning is really something that data has the ability to do. Historically, clinical trials in oncology take a very long time to accrue to. It’s somewhere in the neighborhood of 2- 4%, depending on which literature you read, the number of patients that participate in oncology clinical trials, there’s only like I said about 2- 4%. So the ability to utilize synthetic control arms, as the comparison group to a new therapy is really important because it means that we have the ability to utilize years’ worth of knowledge of the standard of care that’s been out there in these particular patient populations, and accrue thousands of those patients into a data set that can then be compared to the treatment ordered, and the treatment arm being the new drug; and it allows for a significant savings on the clinical trial side, plus it means that we don’t have to recruit a bunch of patients into a standard of care, which allows again more patients to get access to novel, and hopefully more effective drugs going forward.
Dan Housman 6:50
I guess some of the things that I get excited about with the Vidence data is that from a regulatory standpoint they’re trying to match the level of depth that they have in a clinical trial. It’s always hard to do that with Real-World Data, but when we look at sources that come from an aggregator that may have just access to the EMR, but not strong access back into the system it seems insufficient, in terms of the data we can pull out, so if they want to confirm, , they want to go all the way back to the radiology study and the radiology reports to get resist scores and information that just isn’t available with these other aggregate sources. So I feel like, without a solution like Vidence, combined with what NTT DATA can do, we just can’t do what the FDA wants.
Rob Hauser 7:46
I would agree. It’s definitely challenging if you don’t have either highly curated data, beginning with highly curated data, with the ability to go back into historic documents, and it could be everything from PDF files from, you know, previous providers of care, to be able to get to those sources and get to those images, yeah it does, it creates a very big gap, and a challenge. At Vidence we have that information. We have the technology to be able to pull those key data elements out in structure, as well as using human abstractors to fill in those gaps where those gaps exist as well, and then, enriching it with the image data to be able to go back and actually look at changes through the images, to get at the resist and things along those lines. Yeah, it is a challenge and while the dataset may not be, in the millions, it’s the ability to have that rich, robust data to do the comparison to those in clinical trials,
Dan Housman 9:02
I guess, you know, Mitchell, maybe you can talk a little bit about Advocate AI. But certainly having these frameworks for doing annotation and extraction, where we can combine information not just from Vidence, but from other sources, you know, that seems to be a pretty big piece of this too because we’re not going to see ready baked RWE (Real-World Evidence) data for regulatory purposes. So maybe, most people don’t even know what Advocate AI is doing, and what kind of tools your team’s added to look at improving the data.
Mitchell Goldburgh 9:34
That’s true and I think there’s always a lot of hidden information in the descriptive areas of impressions and diagnostic reports and pathology reports, where we’re able to go in and using NLP (NATURAL LANGUAGE PROCESSING) extract that unstructured information into something that’s quantifiable and highlight the changes in geometry, density, or other characteristic texture in the images, especially when you’re dealing with oncology, to understand the impact of the clinical treatment. So, it is definitely an area where what Advocate AI does is expand the data dictionary to help with some of this synthetic and Real-World Data.
Dan Housman 10:32
And maybe we’ll also talk about that we’ve gotten multiple requests on our side of Graticule, I’m sure you’ve seen them as well, from AI (artificial intelligence) diagnostic companies, I’m gonna call them AI companies, but really they want to make a medical exam or test or faster tools for supporting decision support in oncology care. Maybe we talk a little bit about those applications and how Vidence can help support them. Rob, have you seen some of these requests to translate in AI, into helping cancer patients, and what does Vidence think about it given the close connectivity to the health system itself?
Rob Hauser 11:15
Yeah, so we have been doing some of that work, specifically in prostate cancer. During COVID we created some AI and ML (machine learning) algorithms utilizing that rich data to identify patients who may be at risk for not coming back for their next round of chemotherapy because of their fear of the virus and others, and allowing them to then be identified to go through telehealth process. It was a really successful project using that AI and ML; and we’re seeing more and more of those requests coming in. And again, it goes back to, at least in my mind, the difference between big data and high-quality data. I have a personal belief that if you run AI and ML on very large holey data that looks like Swiss cheese, you’re going to get very poor algorithms that could potentially cause harm. Whereas if you’re running AI and ML on high-quality data, that’s been highly curated, and is complete, while it might not have the huge numbers, but the completeness of the data approaches 100% and it’s accurate, you’re going to get much better algorithms and machine learning; which can then be applied to larger populations. So, we are seeing more of those requests, Dan, because of the high quality and completeness of the datasets.
Mitchell Goldburgh 12:49
And just to echo that we, you know, as our data cooperative grows for Advocate AI, our dataset is expanding, but the requests we’re getting with Graticule and from AI vendors is for deeper per patient, rather than broader patient populations, because they want to take into account more factors when coming up with predictive algorithms. So I echo what Rob said. There’s certainly a lot of discussion about big data, and the representation of patient populations to prevent biases. But as we have a larger and larger data cooperative, inclusive of the Vidence family of sites that they’re building, I think we will begin to see the value of high-quality annotations, extractions of descriptive terms, as well as the completeness, from a lung screening test through the end of the treatment, showing the depth of what we can provide to our customers.
Dan Housman 14:11
And I’m really hopeful that these new diagnostic solutions will work, because what we’ve seen is, with all of the investment in oncology, oncology I think 50% of all Real-World Data, but it’s a huge space in the life sciences world. You know, we still don’t know whether immuno-oncology treatment will work for a patient or not, so we’re using some markers like PDL1 (a protein found on T cells) to determine if IO (immuno-oncology) therapies will work, and that’s just one of many novel approaches to treating cancer. And we don’t have good predictions, and you don’t have an infinite amount of time to find the right treatment. And so if we can go dive deep into, you know pathology images, radiology images, tumor genetics, which are still pretty hard to do, I would say this is, unfortunately, bleeding edge, not the normal process by which we evaluate whether we can predict response to treatment, you know, we’ll be serving patients so much better. And so, at least I look at this group as one piece in hopefully a machine that can generate better outcomes for people using these diagnostics and it should benefit both the diagnostic company. The patient first and foremost, the pharma companies that have the treatments, as well as the groups that are treating them being able to be better able to satisfy the need that they’re in their mission. So it’s great that you have so much breadth of data and Vidence and I look forward to you guys growing it, however you do, on both sides, from entity data environments, it’s really important in the end to the patient.
Rob Hauser 15:57
Dan, I was just going to echo that it’s a really good and very, very important point, right, for these populations, you know, the earlier we can diagnose, the better. We know that patients that are diagnosed in earlier stages do well. If we can predict better based on genomics and other, markers that we find through AI and ML, that a patient’s going to respond to a certain therapy better than another, we can avoid giving products that may not work as well and may cause toxicity. And in the end that, you know, is incredibly powerful, not only for the patient but for the healthcare system in general, but the savings that can be seen across the spectrum is huge; and like you said, most importantly, the better outcome for the patient.
Dan Housman 16:54
So I want to close, to some degree, and I know we’ve talked a bit about it, you know, do you have a long term vision for where Vidence is going and how it can really help, you know, beyond what you’ve already done, but left to do and what we need people listening to have visionary projects to drive.
Rob Hauser 17:15
Yeah, so I like I said, I’ve been in this oncology and data space for 25 plus years. I wish I saw an end tomorrow. I wish we could use data to cure cancer, but I think many of us know that is not necessarily going to happen, like I said, tomorrow. So where do we go? I think it’s continuing to break down silos, it’s continuing to bring data of high quality and quantity together to be able to get to our insights, faster, and that’s what it really boils down to is having a big enough, robust enough, high quality enough dataset to be generalizable across the oncology population, so we can get to better outcomes for those patients as quickly as possible. That’s our goal. That’s been my personal goal for 25 plus years. And I think we’re on the cusp of really seeing data, being able to do that, and maybe not relying so much on clinical trials that take 10 plus years, and then take another five years to get to the bedside. If we can cut that in half, think about what we’ve done for society, just by using data, and I really believe we can get there.
Dan Housman 18:55
Thanks Rob. Thanks Mitchell, I think we’re reaching out to the world, you know, the pharma world that has oncology drugs, other health systems that want to participate in these kinds of projects, groups in these diagnostic companies or even within pharma companies that want companion diagnostics. I think the best call to action for folks listening is to reach back out to us to give us an idea of what kind of problems you’re trying to solve, so that we can come back and propose what’s possible with Vidence and the whole team support. If you want to access us you can go send an email to AdvocateAI@NTTDATA.com . You can go to the Graticule.life website and fill out the response forms. You’re welcome to email me firstname.lastname@example.org . or talk to Rob or Mitchell. I’m sure you can find them through LinkedIn and other means, but we’re really looking forward to hearing from groups that have problems that may be considered hard to impossible. and hopefully they’re on the possible side, you know, in the near term future and we can, you know, quickly get to motion to helping to bring the data towards the groups that have the best need for it. So, thanks for your time today Rob and Mitchell, and we’re really looking forward to having group conversations with anyone out there listening, so that we can tackle some big problems in oncology.
Mitchell Goldburgh 20:25
Rob Hauser 20:26
Thanks Dan. Appreciate it.
Dan Housman 17:35
Alright, catch you soon.