okay we're gonna get started and and kick things off um so I know we're still having people come in but we're just gonna get going as people find seats okay um so good morning it is my pleasure to welcome you to the first session of our 2022 women's health cancer conference um so my name is Jessica santolo Gonzalez I'm a surgical oncologist and a co-chair for this year's event before we begin I do want to just give out a few housekeeping notes so we are encouraging people to wear masks when not eating please silence your electronic devices and we will be circulating a microphone for questions at the end of each session this is a hybrid event and we're thrilled that over 300 people are also joining us virtually so for Zoom attendees you can chat your questions in as you're listening to the sessions and a volunteer will read them out loud during the Q a lastly if you have signed up for continuing education credits and have any questions you can check in with the cmie team at the registration desk or look for an email on Monday all right so now with our first session it is my honor to introduce to you Dr Randall Randall Holcomb who is the host of today's panel new technologies to reduce the burden of cancer Dr Holcomb is a medical oncologist he's the director of the University of Vermont cancer center and has really been a visionarian where he wants to see that cancer center go and he is a health services researcher we know you're interested in learning what's new in the field and this panel is all about where the science is going in the future so thank you so much for joining us Dr Holcomb uh thank you so much uh Jessica so um so this session uh is entitled new technologies to reduce the burden of cancer I think it's going to be really exciting I want to tell you who our panelists are up here um first James Gerson who is an assistant professor of medicine in the division of Hematology Oncology um and uh he received his medical degree from Case Western Reserve University School of Medicine did some postgraduate training at Stanford and Fox Chase and uh is coming to us from the University of Pennsylvania actually started here at UVM about two weeks ago he's going to be serving as director of our car T program which you're going to hear about uh in just a few minutes we've also got Dr Nicoletta sideropolis who is an associate professor of pathology and laboratory medicine here at the larner College of Medicine she's the inaugural medical director of the genomic medicine program here at UVM Medical Center she received her medical degree from the University of Connecticut school of medicine completed residency at Dartmouth Hitchcock Medical Center with some additional fellowships in cytopathology and molecular genetics her expertise is on molecular genetic pathology and she's going to talk today about new Innovations in genetic testing for cancer and we've got Dr John Shepard who's a professor at the University of Hawaii Cancer Center he's visiting us today uh he's a Fulbright scholar a fellow of the American Institute for medical and biological engineering former president of the international Society for clinical densitometry he received his PhD in engineering physics at the University of Virginia did some postdoctoral fellowship in biophysics at Princeton University he studies body composition bone density algorithms for Women's Health and he spent much of his professional career in the radiology department at the University of California at San Francisco before he moved to Hawaii he's interested in biomarkers for breast cancer and he's going to talk today about artificial intelligence and how that can be utilized to help in breast cancer screening and to reduce the burden of cancer so I'm going to have each speaker talk for about 10 minutes we're going to just go rapid fire through the three of them so save your questions or write them down on the paper uh that's on the table in front of you or at home and then um we'll have plenty of time for questions after that I'm sure that uh the talks will stimulate lots of questions if you're at home uh looking at following this on Zoom put your questions into chat we will monitor them and we'll try to get to all the questions at the end we're going to start with James Gerson thank you Dr Holcomb on really such an honor and pleasure to be here um I'm I'll be honest I'm mostly just uh relieved that I founded the Davis Center this morning because as Dr hoga mentioned this is my the end of my second week and I've never never been in this building before so glad I made it um so I'll be talking to you about uh chimeric antigen receptor T cells car t as it's uh commonly called which is a very exciting treatment currently for the uh for hematological malignancies such as lymphoma and Leukemia but there's a lot of hope that this may uh someday be a treatment option for patients with a lot of other kinds of cancers um these are some disclosures so because the audience I think is uh consists of a wide variety of uh some patients some caregivers some lay public and also providers I thought I'd just give a very general overview of what we're talking about here and so what I tell my patients is that car T Cell Therapy involves harnessing the immune system to Target cancer and it's taking a normal kind of cell in the body called A T Cell whose job it is is to survey the body for infected cells usually with viruses and to kill them off and so shown here is a normal style that might get infected by any virus such as covid-19 but you know lots and lots of other viruses even before uh the pandemic that cell would be infected with a virus and the cytotoxic T cell at the bottom would come in and identify that that cell had been infected and engage with it and then kill it and uh there is a roll of T cells in surveying the body for cancer as well and it turns out that the immune system is able to kill off uh lots of different types of cancer that might develop in the body but the cancers that grow to become clinically relevant are those that are able to evade the the immune system and so the immune system has a much more limited role once you're diagnosed with cancer and so what we do in this car T Cell Therapy is we take those T cells that normally are targeting viruses and we engineer them through a very high level process to Target and kill off the uh the patient's cancer cells so shown here this is a very fancy animation put together by by colleague at Penn David Porter where using a lentaviral vector a T cell is engineered to produce a novel T Cell receptor and the T Cell receptor is what a normal T cell uses to engage with those infected uh the infected normal cells With viral with virus that novel uh chimeric antigen receptor which is where the name car t comes from goes to the surface and allows this new engineered T Cell to engage with and kill off a cancer cell this is a very involved process that involved take involves taking immune cells out of a patient sending them to a generally a company although there are some institutions that are doing some investigational products in their own in their own walls and doing the Engineering Process which takes somewhere between two and four weeks although there's some new technologies coming out maybe bringing that down to just about two days and then putting it back into the patient hoping that will then Target and kill off the cancer and currently car T Cell Therapy which was first FDA approved in 2018 is approved for a wide variety of hematologic malignancies mostly lymphomas some leukemias and also multiple myeloma very recently and as I mentioned before there's a lot of interest in bringing this to uh solid solid tumors as well so breast cancer prostate cancer so there's an unbelievable number of uh of clinical trials that are currently open across the country in solid tumor as well as other blood cancers not listed uh on the slide before I'm hoping that we can really expand this and and bring it to uh to patients a wide variety of cancer types and what's really novel about this is you know there's lots of therapies that um are very Advanced and scientifically very interesting but what's really amazing about this one is this is really a living breathing treatment that goes back into the patient and lives with them for years the longest Survivor after a car T Cell Therapy uh was still alive 10 years after he received this therapy and still had those car T cells in his body and that's unlike any other therapy that's ever ever been given before um so very exciting and uh really uh such a pleasure to be starting a new program here at UVM to offer this therapy for patients who need it in the area foreign I'm sure that stimulated questions so write them down because we're going to get to them shortly next up is uh Dr sideropolis Who's going to move from T cells uh to genetics hi it's been a long time I think the last time I was up here it was uh before the pandemic keynote so it feels like brand new again speaking in front of people so we have a sunny day right a little crisp in the air for fall um so my name is Nikki sitaropoulos and today um I'll be talking about liquid biopsy so what does that mean instead of tissue based biopsies that a lot of us are familiar with instead we want to leverage fluids things that are not um tissue if you will from the body and try and look for biomarkers I'm a molecular genetic pathologist so I'll be specifically talking about genomic detection of tumor DNA so liquid biopsy specifically using the blood today and to look for tumor shedding it's DNA and and what can we detect in the blood instead of using tissue so um just to give a little bit of background not everybody is familiar right with genomics and DNA and testing for mutations in the DNA and so um nucleic acids were first described in the blood actually um in the 1940s um by in and it was in the French literature so I haven't read that primary literature because I I don't understand French um and and then in about 30 years later um uh Leon at all is a group was actually able to detect that group was able to detect DNA in the serum so that's part of the blood in cancer patients and so um what they were able to determine by the 1970s was that um a dynamic that that the the DNA in the blood was actually a dynamic biomarker that would shift in concentration depending on the patient's clinical state with their cancer and then fast forward to the 1990s right where molecular Diagnostics and that's my field so using DNA RNA things of that nature and studying them to generate a clinical report that could be used for clinical care molecular Diagnostics actually emerged with a critical role in molecular biomarker detection for clinical care and so in 1992 in 1993 even up to 1994 right they started to detect researchers started to be able to detect these cancer-specific mutations in genes in things such as the stool for colorectal cancer in things such as sputum so right right something you spit up if you have lung cancer and so um moving forward from the 1990s and all those discoveries that were being made it really paved the way forward for emerging clinical applications of of cell-free DNA that's that's been seen over the past several decades and So currently um you know from the 1990s moving forward the FDA actually made some approvals for molecular biomarker tests and when the FDA I think steps in and makes an approval it's almost sort of a Mark that something's here to stay it's recognized at that level and so in 2016 the first PCR test for egfr mutations that's a gene that's commonly mutated in non-small cell lung cancer that can be leveraged for therapy those mutations in plasma from non-small cell lung cancer patients that type of testing was approved by the FDA and then in 2020 a technology called Next Generation sequencing which we use here it's a much more robust genomic testing technology that can test for a lot of different biomarkers across a lot of different genes in one shot right one stop shopping um yeah so so NGS assays were approved for multiple therapeutic targets so lots of different genes in the plasma so using liquid biopsy from solid tumor patients and this is not just one cancer type anymore so really what we're seeing is that this technology is reaching an Ever broadening population of cancer patients with every passing year so what's the specimen Journey right um I like to think of the pathologist as really having a role in tissue stewardship a lot of us get you know a sample taken from us and you know you just trust you're going to hear back right what's the answer so what happens to your tissue right when it leaves your side with cancer um you know you have a lump or a bump something that's suspect and typically um you undergo a tissue biopsy where a chunk of that tissue or cells from that tissue or that lump or bump are taken given to the Pathology laboratory anatomic pathology fixes them processes them and looks at them under a microscope and so let's see here can you see my little arrow no you can't what's the pointer do we know this ah so so this is the tissue biopsy right and so that's what the tissue looks like under the microscope um but instead now we're moving towards liquid biopsy and so what does that mean tumors just like tumor cells just like normal cells they have to have access to the blood right that's where they're getting their nutrients it's where they're getting rid of their their uh byproducts and so um tumors are constantly turning over right just like a lot of other cells in our body and so when they die um they they basically dump out parts of their DNA into the systemic circulation um sometimes even when they're alive cells will package up pieces of their DNA and that also can make its way out into the systemic circulation and so with a blood draw we can process that blood and separate um uh the the DNA from the tumor into a part of the blood and use that for testing and so considerations for each specimen type the key piece of information here is that the classic way of doing this and the way we mostly do it now with tissue-based biopsy is that you get to see what it is you're putting into your assay right you don't want to you want to know what's going into your assay making sure that the tumor is represented because if the results come out negative and you don't know if you had enough tumor then you risk a false negative right the thing you were trying to test wasn't there so the negative doesn't mean too much so with tissue-based biopsy we feel very secure in doing that all of us in the molecular lab are trained to also look at tissue under the microscope and make sure we qualify what's going in because you don't want garbage and garbage out cell free DNA the technology is evolving but really what you're getting is an invisible fluid right that you can't see the pieces in there unless you use these very sophisticated detection mechanisms and there are ways to say okay I feel confident that this is tumor DNA that we're going to test it's just those those Technologies are maturing as we speak um so the other key piece of information to walk away with today if you're hearing about liquid biopsy and you're curious about it is that it's not a one-size-fits-all model there are different two you know there's obviously many to all of us in this room have a a wide variety of tumors that we're interested in that are represented here today and so different types of tumors are known to shed differently so we have high shutters and low shutters so this technology may work actually really well for for some Cancers and not so well for others but being able to again the technology is emerging to enrich so that it can be a more Equitable technology across all tumor types um another another thing so we get now into this space where the the life cycle of a cancer um what's the event and what's the treatment strategy there's certainly screening early interventions right you're healthy but you're getting your screenings are there any early interventions there's localized cancer where you you detect it and you want to know what's my risk of dissemination and um you know what's the risk of recurrence um then you get into a metastatic state where uh there's treatment selection and monitoring response to that treatment and these targeted therapies we talk about um and then refractory cancer where now you've broken through that therapy what is the mechanism of resistance how did this tumor mutate in a way to create a new mutation so it could skirt whatever treatment you're throwing at it what's that mutation and now what's another targeted therapy we can go after it with right so you keep chasing it down so these um we we really have implemented liquid biopsy more so in that at later stage right with refractory cancers we could monitor them in that way you know what type of cancer you have you've already been in the treatment protocols and so now you can just sort of monitor metastatic cancer similar situation liquid biopsy does have a pretty um robust role currently clinically in that space but really I think where this is becoming an emerging technology is the multi-cancer early detection assays um and and so what these assays are as is their screening assays for relatively otherwise healthy people and and some of the claims there's one company in particular in the United States that's gaining traction and some of the claims that they make are you know fighting cancer starts with knowing it's there right and approximately 70 percent of cancer deaths are caused by cancers without recommended screening another thing they like to talk about is you know you can detect more cancers early when used alongside other mechanisms of screening and then it's recommended for use in adults with an elevated risk for cancer such as those 50 or older so an age is a factor and they like to tout that one blood draw can detect the cancer signal across 50 plus types of cancer so that's pretty attractive right because all of us would like to detect a cancer that's why we do our screening right we all want to detect it early and and take care of something if something's wrong so what are the counter points to these multi-cancer early detection assays right they are intended to be used as screening tests and and in layman's terms what these tests do is they tell your doctor that you may have cancer right the patient may have cancer it's probably cancer X you know whatever site in the body they determine but we don't know where it may be right so um you know in a study with about 6 600 patients healthy participants there were 92 positive screens and 57 of them were false the false positive results can lead to morbidity and sometimes mortality and certainly psychological harms and significant costs um something to be aware of is that not all insurance covers tests like this right so you you might pay out of pocket and that's fine but if you get a positive result you will be left trying to navigate the Health Care system with insurance that will not cover you because you don't necessarily have a diagnosis of cancer and so Imaging and trying to chase this thing down can be quite a costly thing and it's Crossing that divide right now that I think people need to have eyes wide open um and then I think lastly early detection doesn't necessarily lead to improved survival all the time and so I think to be right now they're claiming well we'll detect all these early cancers across healthy populations he will do it early but I think um you know even assuming and I think it's fair to assume um honestly in my opinion that this technology is sound and could be considered a major Advance right for how we use genomic testing in our cancer population clinically but instead I think initially focusing on at-risk populations right so for specific cancers where we don't necessarily have wonderful screening right now pancreatic cancers ovarian cancers right focus on those populations it might be a more efficient and effective way to identify potential high-value applications that could allow for Integrations into public health initiatives and into medical practice so just my concluding remarks and again these are are my opinions uh professional opinions but I do liquid biopsy is definitely here to stay we're trying to move it into our our laboratory here the technology is not necessarily out of place right now where where it completely replaces tissue biopsy and then Laboratories working on these assays for these multi-cancer early detection tests are utilizing remarkably deep sequencing these are very highly sensitive testing technologies that are complex to detect cancer that's otherwise clinically not detectable undetectable and so lastly clinical trial data will continue to emerge while preliminary data you know shows promise of clinical utility certainly additional data from large-scale well-designed trials will determine clinical utility and this type of testing is now incorporated into what Joe Biden had announced with the cancer moonshot so if you're interested in reading a little bit more about that it is it is worked into that with a massive clinical trial coming down the pike so um thank you for your attention and enjoy the next talk thank you Dr sidaropoulos we're moving on to Dr Shepherd good morning it's great being with you in Vermont it's a long ways from my current home and uh it's nice to have a go to the place where the weather's different occasionally so um yes I'm John Shepard I'm uh from the University of Hawaii Cancer Center and I specialize in uh using and creating biomarkers for disease using uh Advanced methods that for the most part utilize artificial intelligence so we might ask what what is artificial intelligence and that could be a talk into itself and all I want to say is that artificial intelligence is effectively a learning algorithm and if you compare that to a normal like computer algorithm uh normal computer algorithms do exactly what you tell them to do they're deterministic so they'll get the same outcome every time from their deterministic nature uh learning algorithms like artificial intelligence will look at data and we'll learn patterns that are there that you couldn't have even guessed so that's why we call it artificial intelligence it has this in intelligence that it's bringing to the problem solving the the second aspect I want to emphasize about artificial intelligence is that it can utilize more information than than than we can as humans and often their solutions to problems we call superhuman because they're beyond what we're able to do and to give you an example Our Eyes under the best circumstances can distinguish about 256 different Shades of Gray and if you tried to see those on just a small you know a place card you could probably only see about eight but in the right condition she says you could see 256. and it's really unlimited what an algorithm could see but we commonly used 65 000 levels of Gray so if a algorithm is a artificial intelligence program is looking at a a mammogram or a medical image it's looking at it with 65 different 65 000 levels of discernment while when we visually look at it we have about 256 the most the third thing I want to say is that uh it's very complicated and it takes a long time and it takes a lot of data to train these artificial intelligence algorithms in the same way as it takes a lot of data and time to train like a radiologist uh it might take eight years to train a radiologist of medical school then four years after that to specialize in Reading images but if you want a second radiologist how long does it take it takes eight years and if you want but if you want a second AI algorithm to accomplish not all the tasks that the radiologist does but let's say one task how long does it take to get a copy of that artificial intelligence program seconds so that's why when you finally accomplish the goal and I'm going to I'm going to talk about several goals here if you accomplish any of your goals you can replicate that and spread that solution around the world immediately and do good things but there are some downsides to artificial intelligence and I'll I'll talk about that at the end so I'm going to talk about artificial intelligence with the within the realm of screening mammography and the challenges really are finding the five cancers in the Thousand women that are screened so just to run through kind of the the nominal numbers if you screen a thousand women for breast cancer you'll have about a hundred 900 will be effectively ruled as negative right off the bat Radiologists can discern that fairly quickly in fact in the old days when it was just uh films it would take five to five to ten seconds to to determine a negative uh mammogram very accurately but a hundred of those uh uh women might there might be questions and old films may be recalled and the women might even be uh recalled back in for further Imaging of those hundred on average about 25 might receive a biopsy but only five of those biopsies will be invasive cancer and only one of them will be Advanced stage cancer so the challenge is is how can we better uh narrow that down to serve the women that are at highest risk and to avoid these unnecessary biopsies especially when we have a national shortage in Radiologists the time it takes to do all these exams is is very expensive and so you could ask the question is AI a kind of a Panacea solution and let's look at the the roles that AI has been proposed in screen mammography there's actually several uh the first one is when uh the mammogram is available for reading there can be a pre-read of those mammograms to select which mammograms are the most likely to have cancer and versus the ones that are the least likely likely and then in many centers you have a dual read setup where you have two Radiologists read it to get a um a consensus view that's more accurate than one radiologist reading it and often the AI program can act as one of the Radiologists in that process and uh and to the to the right over here instead of acting as a as a second reader it can act as a reading tool that actually this helped the radiologist when they're doing the reading and then the final part of this process is after all the readings are done and you have all the images in your archive you can revisit those images later if you have something more sensitive or maybe a an additional marker that you want to look at the images with AI you can recall all those images and AI program can do that so the first example here is the pre-read here's an AI algorithm that can pre-read the algorithm pre-read the mammograms and it can discern what's which mammograms they think it has the highest probability of having cancer so if you take the score that it gives on the left hand side over here it gives a score from 1 to 10 and you divide the population and put 10 percent of the screening population across the board here you would find that over 78 percent of the cancers would be in in the in the highest score of the ten so the 10 is what the algorithm says this is the you know suspicious cat most suspicious category and it puts virtually all the mammograms there now when the radiologist uh in this study used this algorithm uh to read those mammograms first when they're you know they've had their cup of coffee and they're all cherry in the morning those are those are the mammograms you want them to read first right not flipping through the 900 that are negatives and when you do that you find that every one of the Radiologists uh improved and their accuracy so the orange bar here is with the AI and the blue is without so they improve their accuracy but most importantly if you look at this plot you can see that to read these films that were had the number 10 score the high the the most suspicious it took them about five percent longer to read those they spent more time on them but on the on the ones at the opposite end of the scale that was the least suspicious they could spend a lot less time on them so they're focusing their time on the mammograms that uh give you the the most benefit because those are the ones that are the most suspicious the second category is have the of the dual read and uh this is a study where this AI program was used to highlight what the AI thought was the mammogram and uh digital breast homosynthesis images and if you've had what's called 3D mammography it creates a whole stack of images for the radiologist to read and they have to scroll through those and it takes a lot longer when I said it took five ten seconds before it'll take 40 or 50 seconds for them to read these uh these 3D mammograms so that's that's a lot of time and it can be fatiguing to go through case after case when uh the radiologist didn't have the AI their sensitivity and specificity uh is plotted here and just to give you an idea if you're not used to looking at these uh these sensitivity specificity curves this is perfect right here on the top left that would be perfect no cancers missed all cancers found and so when the radiologist had access to the AI program you can see that it brought this the low performing radiologist here way up to the top here and this curve is the uh is the uh just the raw performance of the algorithm by itself because you can set its sensitivity at different levels so you see on the on the whole all the Radiologists did better when they had this dual read and they and they operated faster the last thing here is that there can be biases in any algorithm and in this case we have the breast cancer surveillance Consortium algorithm that's used for determining risk and it's it's one of our most accurate algorithms to use clinical risk factors and it can rate women as high risk or low risk and if we looked at just the women that that had invasive cancer uh within this study uh it found that about 45 percent of white women uh were that had invasive cancer were thought of to be high risk before the cancer was found and that sounds like it's low but if you look at Filipino Japanese and Chinese only 10 percent of the uh women that ultimately got invasive cancer were considered high risk so that means the model is not well calibrated for these subgroups of Asian populations because they weren't focused on uh in the model but if you look at how AI Works across the board compared to our clinical risk models so you can see that five examples here all give better performance numbers than the clinical uh model and the only if again if you're not used to looking at these AUC numbers the higher the number the better so in all cases of either looking at cancers that were at the time of diagnosis the risk at the time of diagnosis or even cancers that would five years after the mammogram was taken the AI algorithms did better across the board but there is this fairness and bias and explainability issue if you just present a number to somebody and say there's a high probability of cancer and they go well what's that based on I mean how do I know that's accurate um explainability in AI means you provide more information than just the um just the probability of malignancy you say it's it's we think it's that way because of these features or we'll show you exactly where we're looking to uh think that that it's uh malignant and and that helps the the acceptance of the AI quite a bit and the last part is fairness if you use only data that's from White populations or from uh wealthy populations often there can be this built-in bias and in this example the ground through training here is the as bride and so when you feed it through the AI algorithm it predicts bride dress ceremony woman and wedding but when you feed in another bride from a East Asian woman and that's dressed differently that wasn't part of the training data it doesn't even come up that identifies her as a woman it says clothing event costume red and performance art so it would it got the wrong answer it's a bride that the answer is what you wanted but it was never had that as an example so this cultural bias can be burned into AI if you're not careful so we're running a large registry in Hawaii to try to bring this fairness to AI training to collect data in Hawaii on women that are of our region 60 percent of the women in Hawaii are Asian and 20 percent are native Hawaiian and if we don't have data from them in our AI models we could potentially be misdiagnosing them and lastly we're trying to bring these large data sets together with what are with what's called federations so once we have our data we want to join it with other other sets of data in such a way where we don't actually give our data to anybody but we build this Federation where the models are shared as they learn across the ocean so that we build Fair explainable models that will work on a wide you know broad population so that's it uh thank you very much for letting me talk a little bit about what we do and I hope that gives you some idea of how AI is used in uh and in breast cancer research thank you Dr Shepard so uh we have we have someone walking around with a microphone um and uh so if you have questions uh raise your hand and uh we'll get to those we've got one right up here in the front uh I want to thank you if you took off your mask to speak because then I can see what your lips are saying otherwise I just admire your pretty face number I have a question for each of you but I will just uh ask a quick one of the cell therapy does the therapy need refreshing yeah no it does not um there are some cases where the cells will not persist in a person and um unfortunately that can be a reason for the cancer to come back in that case generally retreatment is not necessarily going to work so uh the the goal is that you would not need to refresh it thank you next is a quick question it appears you can amplify and delete how this is done without damaging or how can you amplify and delete without damaging the or eliminating your desired sample parts and can you repeat the question well you had part of your one of the cells said amplify delete that's just part of what you're looking at in your circle of maybe I'll ask you separate so that it's not okay the last thing is when did you start using bias or being concerned with bias do you have a year when that was a concern um I would say it it started happening or sort of being thought of when uh the algorithm started working and then they noticed uh that it didn't work in some examples um and that's only within the last like the the paper that I think of that's the kind of the definitive paper on um explainability was 2017. yeah uh so it started at what year and it wasn't until 2017 that you noticed the the very first examples of what we call Deep learning now because ai's been around for I mean a long time but but deep learning where it's uh a lot a lot of success uh I would say the very first success was shown in papers in 20 uh 2012. but it took till um 2017 2018 to get into medical research raise your hand for other questions we have a question in the back thank you hi I'm glad you're all here today how do you prevent um with artificial intelligence the data getting um manipulated like if someone we all have our hidden prejudices and some don't want to hide them but anyway um how do you prevent that from from happening I'm sorry I missed the manipulation example well when you're entering the data like I'm just using Hawaiian as an example and how does it get it not become manipulated in a in a racist way that oh um well so sometimes you don't you don't really know uh it's it's very interesting that you don't necessarily know how it's getting its answered at the at the start for example we trained a model for melanoma and it was working fabulously it was working so well we thought we had done such a great job and then we go let's look at the images on which one it's getting right and which one it's getting wrong and we found out that in the training data we've gotten from dermatologists that when they had a suspicious lesion they thought was melanoma they would always put a little a little ruler next to it so that it could measure the diameter and so the algorithm only learned that if there was a ruler there uh that meant melanoma it didn't learn the signal of the melanoma at all and so as soon as we figured that out we we slapped it and you know told it not to do that and took the ruler data out and it forced it to learn the signal that it was of Interest so the only way you can really understand that type of biases that I think you described would be to test it you have to you have to create test data to run the algorithm on and to um to understand make sure it's not biased question for Dr sitaropoulos so uh uh President Biden suggested that in in part because of the clinical trial that's going on with liquid biopsies we might decrease cancer mortality by 50 in the next 25 years so we want to get your thoughts on that as to whether that's an achievable goal I think people if um one of the things that I mentioned as a Counterpoint to this type of testing is really there's this Chasm right where um and and maybe some of you have dealt with this where you might go to a direct to Consumer genetic company like 23andMe one of them and you don't necessarily know exactly the ins and outs of what you're going to be getting and next thing you know you're told you have let's say a brca one mutation and what does that mean and and there's certainly um you know there have been uh uh women and men who have spoken about you know opening their test result because oh they were cuis curious like how a story for instance that was published how Jewish am I right but instead now she got all this information and and she has a serious breast cancer risk so there's that moment of um psychological despair if you will and trying to access the Health Care System um and then determine and track it down you know where where are you going to get your Imaging and and how are you going to work this up so if you go with these multi-cancer early detection assays you know you may be told okay you're at risk of having this particular cancer but I think the problem right now is there's a disconnect between what these potential assays are telling you but then how insurance companies then manage how you would how they would help you pay for your care to determine where is that cancer right they're not unless you have an official diagnosis classically how we make it you're up against it so I think there has to be we have to connect the dots there and that's that Missing Link that's that Chasm um that's so stressful for people that you really don't know is going to happen um until you're there unless maybe you're listening to a talk like this so um you know it's possible so we need to make sure that the uh the moonshot initiative covers or Forces Insurance to cover people who may be identified as high risk yes which is not currently the plan so correct all right we've got uh we've got three three questions going over here um am I next no go ahead thank you for this um I love the way that you um each described your Specialties and language that we could pretty much understand so it's very encouraging to hear about um better or more advanced detection methods as well as treatment um my question is it's uh sitting here thinking of Star Trek and um just spitting into something and learning everything about us um but my question is do you foresee a time when this could become more of a screening thing I've known more women in their 30s who are um being diagnosed with 3C breast cancer more advanced breast cancer whether their lives are so busy they don't have time to do self-breast exams or whatever so um do you foresee a time when instead of going in for mammograms or bone scans or whatnot that you're testing modalities may be more of a screening um detection that early um younger women and men could take advantage of and again thank you for being here yeah so I think uh the genetic testing is is one uh possibility using AI for uh for screening where you don't have to go through the the process of going to the hospital getting Radiologists to read you know if we perfect that better that maybe make it much easier for patients maybe even home home screening and using home AI algorithms and of course if you do get that uh that cancer you want your immune system to uh to fight it so we're going to go to Dr Gerson for that for that last step um we got a question over here yeah so um we have a question over Zoom from Betsy Morse uh with respect to many people having cancers that will not be clinically significant what are the estimated percentages of people who really do not have any cancer cells in their blood and how is this affected by age wow that's a tough one does anybody want to uh so so what what is the percentage of of patients that do not have cancer cells in their blood you know um I think that's a hard question to answer because if we if we really think about how many mutations occur across ourselves every day and how our different immune systems are able to fight that off it's we're not we're not screening people across just a wide population of people to be able to answer that question um you know accurately I think that's a fair statement yep there's two questions here quick one for Dr Gerson uh before you get the microphone there the um you talked about the car T cell for patients with lymphoma and Leukemia what what is the response rate of car T cells for those patients because they've got usually gotten lots of treatment before compared to the standard therapies that we have at the moment um so uh unfortunately Cartier does not work for everyone um it's the most of the literature suggests that it only works in about 40 to 50 percent of patients um so we obviously have a lot of work to do however that those numbers are in patients just as you said who've gotten you know every standard treatment and have really nothing else left for them and so the hope is that as we move this therapy earlier and earlier in the treatment Paradigm where patients may even someday get it right when they're diagnosed that the response rates will be higher and then more people will will benefit from it okay question over here hi thank you for being here today um in the arena of surveillance and patients that already had established cancers How likely um in the class of Highly shedding cancer types will it be um for liquid biopsy to replace scannings MRIs pet scans for ongoing surveillance right that no one wants to do that that's not fun so um The Hope is right is that we could implement this and I think as a monitoring uh process um again I think it's a matter of there's there's still a lot of clinical trial data coming out around that um and there has to be cost Effectiveness work that's done so then again the insurance companies it's hard enough to get a lot of insurance companies to pay for genomic testing when it's necessary medically necessary never mind moving it into the sphere of them paying for it on an ongoing basis so I think that work that's needed to be done to move it into that space is well underway it's within the foreseeable future I believe that that will come come to be um and I will be thrilled for us all when it happens obvious benefits in terms of exposure to radiation Etc and so forth but do you have any concept at this point of cost of doing a liquid biopsy versus say a nose to knee CT scan or a pet scan is any of that work been done um I think honestly I do believe that the cost is probably lower um to do it the way we're talking about uh in in the foreseeable future um but I don't have the the numbers off the top of my head but I would imagine it's it's with the cost of uh sequencing decreasing right it's just like when you think about CDs or you know any technology cell phones um that that cost is going down and the technology is only getting more sensitive so um it doesn't take as much overhead necessarily to do that um and certainly it's less burden on the patient to come in and go through the rigmarole so um I I would hypothesize it's less so we're we're out of time we had one question here I'm going to ask you to come up and ask the uh the panelists in a minute because we have to turn over for the next uh um uh concurrent uh session that's going on let's give a big round of applause for Dr scarcity to the Rockies and Shepherd thank you thank you all for being here just the note that there are exhibitors as you if you're shifting your space and going down to the levec ballroom on the far side of the hallway um take a peek at our exhibitors uh throughout the day thank you for being here
2022-10-23