I'd like to welcome everyone to the second day of the SSO 2020 virtual meeting. Today's meeting starts with a John Wayne Clinical Research Lecture. In 1993, the John Wayne Cancer Foundation provided the Society of Surgical Oncology with an endowment to establish and fund the John Wayne Clinical Research Lecture at the Society's annual meeting. The lectureship features a prominent clinical researcher whose work has contributed to the scientific advancement of the treatment of cancer. SSO expresses its appreciation to the John Wayne Cancer Foundation for providing support in the form of an independent educational grant for this lecture. In addition, SSO expresses its appreciation to the National Cancer Institute of the National Institutes of Health and Pfizer for providing partial support in the form of independent educational grants for the SSO 2020 virtual meeting. The John Wayne Clinical Research Lecture, Building Collaborations to Create a Learning Health System, is presented by Monica Bertagnoli. Dr. Bertagnoli is the Richard E. Wilson Professor of Surgery in the field of surgical oncology at Harvard Medical School and an associate surgeon at Dana-Farber Brigham and Women's Cancer Center. She previously served as the chief of the Division of Surgical Oncology from 2007 to 2018, and she's the past president of the American Society of Clinical Oncology from 2018 to 2019. Dr. Bertagnoli graduated from Princeton University and attended medical school at the University of Utah. She trained in surgery at the Brigham and Women's Hospital and was a research fellow at the Dana-Farber Cancer Institute. Her laboratory work focuses on understanding the role of the inflammatory response in epithelial tumor formation. From 1994 to 2011, she led gastrointestinal correlative science initiatives within the NCI-funded Cancer Cooperative Groups, where she facilitated integration of tumor-specific molecular markers of treatment outcome into nationwide clinical cancer treatment protocols. Dr. Bertagnoli has held numerous leadership roles in multi-institutional cancer clinical research consortia and currently serves as the group chair of the Alliance for Clinical Trials in Oncology. She is also the chief executive officer of the Alliance Foundation Trials, a not-for-profit corporation that conducts international cancer clinical trials. Dr. Bertagnoli is the ideal participant and lecturer for the John Wayne Award this year, and I'm extremely honored and proud to have her giving this lecture. Please join me in welcoming our John Wayne lecturer, Dr. Bertagnoli. I am delighted and honored to be asked to deliver the 2020 John Wayne Clinical Research Lecture for the Society of Surgical Oncology. The topic today is the learning health system, and I'd like to start just by defining what many of us mean by the learning health system. This is my favorite definition, comes from a manuscript in Science Translational Medicine in 2010, and it says, an integrated health system which harnesses the power of data and analytics to learn from every patient and feed the knowledge of what works best back to clinicians, health professionals, patients, and other stakeholders to create cycles of continuous improvement. What would a learning health system do for us? Well, I'm going to give you an example to start with. This is a theoretical patient referred to me from a colleague, an elderly woman showed up in an emergency room with GI bleeding. The colleague worked her up, found she had a high-risk gastrointestinal stromal tumor by a pretty severe GI bleed, and she was referred to me for evaluation. Technology-guided clinical care allowed her to come before her first visit. Her entire electronic health record, including all the images, were already available to me for my review, formatted in a manner identical to the one that our team uses, even though the referring doctor was coming from a completely different health system. After seeing Mrs. Smith and confirming, yes, she had a non-metastatic gastrointestinal stromal tumor, it was high-risk because it was bulky and right at the GE junction, some decision support present within our electronic health system showed me that the National Comprehensive Cancer Network guidelines, based on level 1A evidence, the highest evidence possible, suggested how this tumor is to be treated and actually gave me the references for my own review without me having to search them out. But I wanted something a little different for this elderly patient, and knowing that clinical trials results, level 1A, don't always give us the answers we seek, I also looked in the same database and looked at what happened from large longitudinal outcomes databases. This gave me information from a large dataset, prospectively collected, that had over 12,000 patients with gastrointestinal stromal tumors, here indicated as being captured by ASCO's Cancer Link, and it let me look at 914 patients who were age-matched to Mary Smith, the patient in front of me in clinic, and gave me the references for these particular data. So looking at that subset that I thought might be more relevant to the patient in front of me maybe than even clinical trials, I found that of that dataset, patients who received surgery only had the outcome you see here on the screen, and with the addition of a mat nib for a high-risk tumor in the gastric location, they did somewhat better. But this decision support also gave me two little clues before I got ready to treat this patient. Number one, tumor mutational testing may change the recommendation, and the exact likelihood of this is about 15% with a gastric location of the primary tumor, and in big, bold letters, clinical trials may be available, alerting me to this before a treatment decision was ever made. And then finally, the decision support indicated preoperative functional assessment was recommended because she was elderly, had a fairly high BMI, and was perhaps a bit deconditioned. So based on this information, this wealth of information that we don't usually have at our fingertips, we decided to do a genotyping, initiate preoperative mat nib for this patient, but also to place her on a home-based fitness program, anticipating that she would have surgery in from six to nine months once the response to her neoadjuvant therapy was achieved. Gave her a wearable fitness tracker with the data from that wearable delivered to her electronic health record so that the team at the hospital could monitor her progress. We'd know if her performance status was falling off. And also know if she was succeeding and becoming more active and fit as the program succeeded. So to keep going in this theoretical model, let's just say that we got a rather nasty surprise about a week and a half later that the tumor genomic characterization showed that her tumor was one of those less than 3% of gastrointestinal stromal tumors with the resistance mutation, D842V, and it's unlikely, alerted us that she was going to be unlikely to respond to a mat nib and actually gave us the data that was behind this recommendation. So we're faced or now alerted by this result with a treatment change, either surgery or the possibility of a clinical trial for this very rare, but very important resistance mutation. So we offered her these two options and she decided to be treated on a clinical trial for patients with this rare resistance mutation. She had successful reduction in her tumor volume with that clinical trial. She went on to have successful surgery by a minimally invasive gastric resection made possible because of the substantial shrinkage in the tumor from the GE junction. And she had a very smooth postoperative recovery. And now perhaps the most important part of this new learning health system, we continued to gather information on this patient. In fact, on every patient who comes to our clinic. She originally was treated in Boston, decided that it's too cold there in the winter, moved to Florida, but instead, in spite of her change in location, data continued to be gathered for this patient. So that five years later, now using an international cohort encompassing 90% of patients with gastrointestinal stromal tumor, we now had a data set of 390 cases with this rare mutation and also had prospectively collected overall survival data from both clinical trials as well as non-clinical trials patients in an EHR based longitudinal data cohort. So as a result, if we use the technology that we actually have today already, nothing about this really amazing story I've just told you is not already possible if we just had the coordination. If we use our technology available today effectively, what would happen? Clinical care would be more efficient. Unnecessary costs due to tracking inadequate treatment approaches would be reduced. Home-based evaluations would be facilitated. Patients and their families would be more informed. Communication errors would be reduced. Research would be greatly facilitated. And the most important aspect of the learning health system, learning would be integrated into everyday practice in a way that's just not possible, even among the best of us today. Why is this so difficult to do in our current environment when we have such amazing infrastructure and such highly capable information system resources? I don't need to tell you that oncology is incredibly complicated. We are making decisions all the time, juggling our best talents and abilities and judgment as surgeons with information from medical oncology and its wealth of new therapies, both cytotoxic and targeted, layering on radiation therapy and its many manifestations from curative to palliative. And now an entirely new field of immunotherapy, which has really changed the way we think about so many of the solid tumors that we treat. We're juggling all of these complexities. We're also, for each individual patient, trying to manage the benefits, duration of survival and quality of life with the costs, treatment associated morbidity, and even the economic costs of what we are delivering. And the problem is, for all of these complexities, we have vastly insufficient data resources and vastly insufficient, at the current time, ability to collect the data that can inform our progress. So, let me just give you an example of this, rectal cancer, such a common treatment. This is a paper that came from a group in Belgium that did a recent analysis from cancer treatment reviews. And they just looked at evidence gap and gray areas in the management of local or oligometastatic rectal cancer by comparing recommendations from clinical guidelines, and they picked 26 relevant clinical topics, clinical decision points that need to be made, and looked at five really prominent guidelines, the ESMO guidelines, NCCN, the Japanese Society of Cancer of Colon and Rectum, the Australian guidelines, and Cancer Care Ontario, and simply compared them, both the level of evidence and what management choices the guidelines recommended for this very surgically intensive stage of rectal cancer. And here's what they found. If you look at the dark green, that showed substantial or complete agreement between the five different guidelines. If you look at the dark red, substantial or complete disagreement among the five guidelines. And remember, these are all highly qualified guidelines committees using the best evidence they can to come up with important policies. And the other very interesting aspect of this analysis showed that level one evidence was actually pretty prominent. There were 50% of the recommendations overall that were examined had level one evidence available. But what was really interesting is some of the decision points that had the highest level of evidence were the ones that were most in disagreement among the recommenders for clinical management. You can see the top 13 of the recommendations listed here, and as is pretty obvious, there's about a third that are agreed, a third that are disagreed, and a few in the middle. So what is this telling us? It's not telling us that our clinicians are not capable and wise. It's telling us that the data underlying even some of the most solid of our recommendations are vastly insufficient. I'm sure I'm not telling any of you anything you don't already deal with every single day. So what do we do? What do we do? Bad data. We have a lot of insufficient data. This is just a little outline from actually a business school approach. This is 10 years old, this manuscript, and it's not directed toward medicine. This is directed toward industry 10 years ago, which was the Stone Age compared to what we have in information technology today. But 10 years ago, the industry said, we don't have good data, we need to fix our data. And the four steps to fixing the data, I think, are relevant to where medicine is today, 10 years behind what our colleagues in industry, where our colleagues in industry have gone. First on the list, admit you have a data quality problem. Number two, focus on the data that you're actually using. Here, it's business. They're talking about customers, regulators, and others outside your organization. For us, it could be our patients, our colleagues, and certainly regulators as well. Number three, define and implement an advanced data quality program. How have we done that as a profession? And then number four, take a hard look at the way you treat your data more generally. I propose that this, which has been quite successful for the business world, is not a bad roadmap for what we need to do in medicine. So next, thinking about that, what is the vision, the hard look we need to take about where our data come from, how we collect them? We know that clinical trials are just not sufficient to fill our data needs. So how are we going to make up for that? We know our registries are contributing a lot, but again, vastly insufficient. We need to think about a data environment, a data infrastructure, that in the very middle of this diagram are providers and the health systems and researchers side by side. We need to make the focus of data collection on those who are really using the data and have the most to benefit from it, the patients and the providers. And then around the edges of this diagram, you see many of the other users that are absolutely essential to our progress in medicine. The patients themselves, pharma, medical product companies, the payers, and larger aggregated research teams. So the vision overall of data should be patient data, patient data collected once by the end users, the clinician, to support many different uses. How can we achieve that? So you can think about our data system as two different aspects that we really need to manage and think differently and innovate on, the act of acquiring data and the act of using data. And we really need to start at the very, very basic level of what we're doing. Acquiring data, meaning build a common data language, something I'm going to talk about in the last part of this talk. Number two, enable data collection across multiple treating institutions. You're never going to know what to do with an incredibly rare gastrointestinal stromal tumor resistance mutation until we all work together to find that result for our patients. Enable access to your data. And then finally, engage patients themselves in clinical care and research. Some of the things we think about all the time, we hear about all the time, are the using data aspect of this. Data analytics, implementation science initiatives, artificial intelligence and machine learning. I would propose to you that the using data part of this really needs to take a back seat until we have all the nuts and bolts properly worked out on the data quality and the acquiring data end of this. So next, we think about the first, I think, what is a major advance in what we're trying to do to create a learning health system. And that was the development and launch for the oncology world of something called mCode. First launched in June 2019. The purpose of this was to develop and maintain standard, this is a really important word, computable data formats known as Minimal Common Oncology Data Elements, mCode, to achieve data interoperability and enable progress by having a standard data language in clinical care quality, research and healthcare policy development. And I am so very proud to have the Society of Surgical Oncology as one of the lead members of the executive code that has developed and is maintaining mCode. The little diagram in the bottom, HL7 FHIR, indicates that mCode is grounded in the electronic health record development space and in the interoperability standard world. So what's happening with mCode? We now have a common data language, and what this is doing is putting tools for the growth of this language in the hands of the users. This is just a little map of mCode, and you can tell by looking at this map that it is very practically focused on the data that is required to make clinical decisions. This is a research and development data language, but it is also primarily a patient care data language. That's really key if it's going to be useful to us in a learning health system. When you look at what's happening with mCode, a wide range of organizations are now developing mCode-enabled tools for analysis, data capture, and clinical application. Clinical pathways, decision support, clinical trials data management, registries, outcome research, quality initiatives, and then at the very bottom, but a really important one, the development and implementation of machine learning approaches. How absolutely essential it is that if we're going to apply tools like machine learning, it has to be done on valid, accurate, prospective clinical data. So having mCode be able to deliver that is incredibly important. Next, mCode is also greatly facilitating data sharing. I've listed on this slide some of the major data sharing initiatives, mostly from the clinical trials world, but also you see on there CancerLink, Flatiron, SOAR, some of the other electronic health record-gathered real-world data sets that are now being converted to mCode formats so that they can become interoperable, so that we can gather the kind of data at scale that will finally allow us to address some of these critical clinical questions like what do you do with a 75-year-old woman who's got a very rare GE junction gastrointestinal stromal tumor? Without this kind of scale and ability, we will never answer those millions of questions that we have that we see in the clinic every single day. So mCode is greatly facilitating research. This is another effort called the iCare Data Project that intends to perform an integration between clinical trials and real-world data elements. I put this on here because I wanted to remind all of you that it's not just the research groups, the National Cancer Institute, and some not-for-profit corporations, but also the electronic health record vendors such as Epic participating in iCare Data that have really stepped up to help facilitate this kind of standards-driven research. Finally, there is a growing community of users of mCode. The first version was really a nascent one. It is now out there for trial and testing, and there are tools and systems being developed for clinical trials matching, for cancer registries, for radiation oncology, clinical care pathways, and with the payers to do prior authorization, all using this standard data format. You can think about how each of these can also bring about individual aspects of the learning health system that will help to grow this community that we will so depend on to try to achieve the vision that I described in that very brief clinical case I gave you in the beginning. With that, this is my final slide. We have a data language. Now we have to tackle the next big hurdle, which is data sharing and collaboration. To realize that none of us, no institution, no research group, no company can achieve what our patients need on our own. This is a quote that I found from an individual named John Spencer, who it turns out is a middle school teacher, reminding us something that we learned when we were teenagers. Collaboration isn't a 21st century skill. It's a timeless skill. It's critical in the past, and it will be critical in the future. We had to collaborate in the age of hunter-gathering, and we especially need to collaborate in the age of artificial intelligence. At the end, I just want to thank, again, the organizations that have been the founding members of MCODE and really starting us off on this journey to create the learning health system. Then issue a welcome and a challenge to all of you to engage with us in building this important new world. Our researchers, our clinicians, and our patients desperately need us to do this. Thank you. Thank you, Dr. Bertagnoli. Thank you again to the John Wayne Cancer Foundation for providing support for this lecture. Please join us for the ACS SSO basic science lecture, starting next by navigating to that session. ♪♪

Dr. Monica Bertagnoli

Building Collaborations

Learning Health System

John Wayne Clinical Research Lecture

Data and Analytics

mCode