Myriad Live episodes are recordings of an open-forum webinar hosted by Dr. Thomas Slavin. The opinions and views expressed in this recording do not necessarily represent those of Myriad Genetics or its affiliates. To participate in a future recording, visit https://myriad.com/live/ for a list of dates, times, and subjects.
References for this episode:
Kerlilowske reference: JCO 2010. https://pubmed.ncbi.nlm.nih.gov/20644098/
Other suggested articles from Dr. Pederson: Automated Quantitative Measures of Terminal Duct Lobular Unit Involution and Breast Cancer Risk https://pubmed.ncbi.nlm.nih.gov/32917665/
0:00:13.4 Speaker 1: Welcome this episode of inside the Genome is a recent recording of Myriad Oncology live a webinar hosted by me, Dr. Thomas Slavin, chief medical officer for Myriad genetics. The opinions and views expressed in this recording do not necessarily represent those of Myriad genetics or its affiliates to participate in a future recording. Please visit Myriad live for a list of dates, times and subjects, I look forward to exploring the world of genetics with you all.
0:00:38.5 Speaker 1: Hello everyone how's everyone doing? It's Friday we don't tend to do Myriad lives on Friday a lot, so excited. Especially a lot of us have been at ASCO so we've been working for about 14 days straight [laughter] so great to be here we're all about education here at Myriad Live. We have... We will not disappoint today, so we have Alicia Hughes, Dr. Hughes will be on and we're talking breast density is a risk factor for breast cancer. We'll be showing some of the recent research that we showed at ASCO. And I'll kick that off in a second, we have I see Holly, thanks Holly for coming on. Speaking of Holly, Holly is gonna help us July 22nd so mark that on your calendars and we're gonna be going through chemo prevention for breast cancer.
0:01:37.7 S1: So if you want to hear Dr. Peterson talk about chemo prevention, you will be in the right place if you are there at 2:00 PM Eastern on Friday, the 22nd of July. I'm trying to also slot in at least one more probably in early July, if not late June that's being finalized right now. That's gonna be around mammography sweet risk assessment and, you know, kind of the journey that women go through as they're getting their mammograms. Shelly, thank you, I saw your email this morning is also working on another one that will be slotting in somewhere and that will be on LGBTQ plus and hereditary assessment we did one last year on that topic and it was very largely attended it is on our podcast and it's a fantastic podcast.
0:02:32.9 S1: That one in particular, we had a lot of external guests and really covered a lot of ground. So bear with us, we're building on some schedule and then we'll be... We're working on some August and September topics even so yeah, lots to build out. If you have a topic that you would love to hear about, please let us know because we're always looking for, just to make this very educational and the purpose, if this is your first time on Myriad live is just, you know, this is just a safe platform, everybody can come on, talk about whatever they want we do keep them theme focused as you see by the titles here and then these are recorded, we put them up on the inside the Genome podcast just for people to listen to, and this is available really anywhere, Apple, Spotify, etcetera.
0:03:20.4 S1: Oh, the last one's already up. So we just did this one so I'll have to put, we'll have to get Myriad live. I'll make a note, anything that says Myriad live in front of it is from this podcast or sorry from this webinar. And if it doesn't say Myriad live like this one here, Olympia, this is just me sitting down with Dr. Garber, Judy Garber. And we just spoke about the Olympia study. And this, yeah I think I mentioned this last time, this 911 podcast is pretty, pretty intense. That's an interesting one as well, that was me sitting down with a few people that used to work at, or that, that work at Myriad that used to, that were here during the 911 tragedy. And just how you know, gen X companies can kind of help with victim identification and things.
0:04:18.4 S1: So it was just very interesting webinar. That's with myself, John Ryan, and Benoit or Ben Wallace sorry so without further do, let me introduce Alicia a little bit more, so I'm gonna stop sharing this really quick. So Alicia is statistician extraordinaire. So she... We are very fortunate to have her at Myriad I would say she is one of the leading authors now on polygenic risk score based publications since we've just been pumping them out over the years and today we're gonna be talking about breast density before we get into that or, or before I'd transfer over to Dr. Hughes, I wanted to give a little bit of background, so everyone's up to speed on, you know, breast density in risk assessment so let me share, share this PowerPoint really quick.
0:05:22.8 S1: I'm gonna, I'm gonna flip it over alright so can people see this PowerPoint look good? Not sharing speaker notes and everything I don't know if we have any speaker notes, but okay. [chuckle] so breast sensei and risk assessment yeah, for those that you know, haven't really seen a lot of this over the last, I would say, you know, few years in particular, the field's really been advancing particularly around artificial intelligence ways to look at breast density how that can be incorporated into risk. I have here just a little bit from 2006 paper showing just relative risk of breast cancer as you go up in some of these different BI-RADS density. So BI-RADS is you know, a commonly used way to look at breast density. There's other ways to also look at it, which I'll show in a second, but you know, those ways also are somewhat captured here in the sense of, a gradient moving from fatty breast to breast with some scattered glandular density to more heterogeneously dense, and then extremely dense, on the far right.
0:06:37.2 S1: And, you know, it shows here your relative risk actually substantially goes up for breast cancer as and this was from a meta-analysis, it goes up substantially as you go across this continuum.
0:06:50.4 S1: And this is interesting and I wanted to put this together for people because I've also, I've, I've often wondered myself, why is density, you know, really the main thing here because it's not breast size, you would think to some extent having larger breasts could potentially be at higher risk for breast cancer, but it's really at the end of the day, the density and this is kind of the, the, the main take home as I did, did a little bit more digging and research and if people wanna reference some of this, it's not the most scientifically [chuckle] but it is, I didn't pull from like a review or anything. I was actually, I found honestly, some of the most consolidated best information on densebreast-info.org, which is a well curated and well known site for breast cancer information.
0:07:39.9 S1: And they just had, I thought one of the better descriptions of how this interacts so, density is, really just means there's more glandular breast tissue around and more glandular breast tissue is means more glandular tissue. That's susceptible for cancers because that's the tissue that cancers arise from. If you think of like ductal breast cancer, for instance it does seem like there's other, you know in addition to just that fact alone it does seem like there's other hormonal signalling events from the glandular tissue in the environment. So again, more glandular tissue equals more hormonal signalling which may influence all this. But I would say that this is still not completely understood down to a molecular level. I mean, I think people are, are understanding these main points here.
0:08:34.6 S1: And then the last is just, if you have really dense breast, it makes it harder sometimes to find lesions. So if you're getting imaging and you have extremely dense breasts, I mean, it's, it's very hard to find a small calcification for instance, or like a, like you know, starting breast cancer versus someone that has very fatty breast tissue. There's a few main ways that density is evaluated. So I already spoke about you know, the BI-RADS method. And what I like about this figure here is that this shows that the rough population buckets, so most people are kind of in these two middle categories, they either have some scattered areas of fiber ULNAR density, or they're pretty heterogeneously dense. So that's kind of the average population. So it's a nice bell curve in that sense.
0:09:24.8 S1: And then there's, not many people that have entirely fatty breasts, but there definitely some folks. And then similarly, there's not many people that have extremely dense breasts but there's some folks there as well. So let me, let me pause there because I think that's just a good background. I wanna make sure that people don't have questions before we kind of jump into, then some of the recent research that we were showing. I haven't looked at the chat and I should say, Shelly is on if you have chat questions please send to her, you can also unmute yourself and just ask away if you have any questions.
0:10:01.8 Speaker 2: I have a question.
0:10:05.2 Speaker 3: Go ahead, Shelly.
0:10:06.2 S1: Sure.
0:10:06.4 Speaker 2: So I was going to put a reference in the chat on this paper that showed it was a Gemma open network paper that showed breast density in women over 65. So I'd be curious. So conversation about the over 65, I'm just magically picking that because of this paper and under 65. 'Cause, so I hesitated to put that reference in here because I don't wanna miss the other younger population with a reference.
0:10:38.2 S1: Yeah. I mean, breast density goes down with age. You know, that's a big part of it. I don't know Holly, if you wanna, I mean, a lot with this.
0:10:49.5 Speaker 4: In following these women you, it seems that they don't change a lot with age. They change a little bit with age, they change quite a bit with Tamoxifen, but Mayor has done a lot of work with lobular evolution and sort of the the decrease in probable density with lobular evolution over time and L older women who lack lobular involution are definitely at an increased risk. And so, you know, if you have older patients who have retained extreme breast density in particular, they really are at increased risk.
0:11:32.6 S2: Question for you, Holly, this is Edie. If a patient is on hormone replacement therapy, would that diminish the ability of lobular Involution to take place?
0:11:51.0 Speaker 4: Oh, that's beyond my understanding of the biology of it, but that's an interesting question. Hormone replacement therapy and particular combined therapy does increase your breast density somewhat. And the risk associated with hormone therapy is definitely more pronounced in patients with baseline heterogeneously dense or extremely dense tissue. Kolikowski had a great paper that I can send to Shelly about the interaction between density and hormone, the effects of hormone replacement. I'll dig that up.
0:12:35.2 S2: Yeah. Thank you.
0:12:36.7 S1: Yeah. And I, I... You skipped over really quick too. I should mention that you know, what I'm showing on the left here is from Tyrer-Cuzick version eight. So people are familiar with Tyrer-Cuzick version eight. This is literally just a snippet of that. So you know, version seven did not include breast density. It was identified that, breast density is clearly a large risk factor for disease. We'll talk in a second about, some of the influence there on the actual percentages, but as, as part of that in the Tyrer-Cuzick eight model really largely the, the difference from seven to eight is the addition of breast density. So if you're using version eight, you will see these three metrics for adding breast density. So you have the BI-RADS, which is, shown here and then Volpara, and then the VAs percentage density.
0:13:36.5 S1: Alright, so I'm gonna pass it over to Alicia... Thank you, Dr. Hughes for coming on. If you wanna yeah, just talk through a little bit of some of the research that you've been working on.
0:13:49.2 Speaker 5: Absolutely. Thank you so much for having me and please, if anyone wants to jump in at any time with questions or comments please don't hesitate. So here is just a snapshot of the poster that we presented at ASCO. Gosh, what was that like, four days ago? And I will... Let's see, I guess just next slide, I'll go through this in a bit more detail.
0:14:18.3 S1: I see Brian. Thank you, Brian. In the chat, I saw you wrote perhaps relevant to the earlier question. "When using density, Tyrer-Cuzick accounts for age and BMI associations of both density and cancer." Brian, feel free to unmute yourself if you wanna talk through that a little bit more or just introduce yourself 'cause you're clearly an expert in this field. [chuckle]
0:14:40.5 S6: Well, thank you and I love this series. I caught up with the one with Judy Garber where you were teasing her earlier, that was tons of fun. I work for Volpara Health, which is a company that builds technology for assessing breast cancer, the breast cancer risk as well as breast density measurement, and one of the things that we know in the risk models is they account for the interaction... When you put these things in a quantitative model, it's important to account for the interactions, which is a lot of what Alicia is sort of speaking to here. And so they look for the association between BMI and breast density, and age and breast density in order to make it the most accurate prediction that they can.
0:15:24.8 S1: Yeah, no, thank you. Yeah, and Brian has lived a long time in the risk assessment world for breast cancer, I don't... [chuckle] Feel free to [chuckle] tease what... Thank you for the compliments on the podcast yeah, but... If you wanna tell people a little bit about your history of research, that'd be pretty impressive.
0:15:43.9 S6: Well, sure, so I was lucky enough to work at Mass General Hospital back in 2015 or so, when the other Dr. Hughes that [chuckle] we worked with... Kevin Hughes is this breast surgeon, recognized that the models were emerging as a tool for identifying patients who might benefit from high risk interventions like genetic testing or additional screening, but were underutilised in the clinic because of the disconnect in collecting family history, and running the quantitative models, so we set about to build a technology, it was a tool for the clinical folks to help patients provide information for risk assessment, help the clinicians interpret that, and help them match patients against guidelines for additional high-risk interventions.
0:16:38.6 S6: So we were super excited to have that be a research collaborative with a couple of government institutions as well as partnerships with industry like Myriad, and then to see that develop into a company called Hughes Risk Apps, which then became CRA Health, which is now flourishing under the Volpara Health Technologies family where our mission to prevent late stage breast cancer specifically, personally, through this mission of combining breast density and hereditary cancer risk messaging to patients to help them really take that next step in their personalized care plan and really help the genomic revolution make a difference in public health. So that's our story which couldn't be more synergistic with what we're here talking today, so a great privilege to be...
0:17:35.1 S1: Yeah, no, thank you so much. Yeah, thank you. And the field is really exploding. Just based on imaging, there's a lot of now... And I mentioned at the beginning, just even one time, short-term risk estimates for... Off a mammogram for instance, of when... What is the risk over the next few years, one to two-year risk for a woman to get breast cancer. So the field is really looking at ways to use this data, these large data sets to inform more of a near-term risk, and so some of those do not incorporate any sort of family history or genomics or anything, so I think there's a really...
0:18:17.9 S1: It's kind of like, you could almost call it a debate, but there's just some... There's different ways to look at breast cancer risk, both near-term and long-term and people are trying to make the best models. We're gonna talk now... Transition into a model that incorporates multiple things, so data from Tyrer-Cuzick which is clinical and family history variables, as well as breast density, and then also brings in some genomic information. So you...
0:18:49.0 S2: Before we get to that, Shelley [0:18:50.2] ____ had a question about why Tyrer-Cuzick does not include breast density in the calculations in women under 40, that she's encountering patients who are under 40 with an over-estimation due to the calculation when they have less breast tissue.
0:19:08.6 S1: That's a good question.
0:19:11.7 Speaker 3: Yeah, we have a... We're doing them in mammography and I'm an advanced genetic nurse and I've run a high risk clinic, so I'm seeing patients under 40 who are referred and some of the primaries they're seeing in SMI that depending on their age, if they're under 40, they qualify and then once the... For MRI, then once the Tyrer-Cuzick takes into account their breast density, they don't. So when I see them, I actually show them... If they're 38, I say. "Well, when you're 40," because since they have fatty or scelofibrograndular densities, I always do the calculation as if they're 40 to show them with this model, this version eight, we do use breast density so anyway, just we would love some clarity from you if you have any of... If Tyrer-Cuzick will include under 40 or is it not included because based on data, it's not applicable, or is it more of a cut-off? So we're just finding some issues with the under 40.
0:20:18.8 S1: Yeah. Does anyone know? My first thought is data just wasn't there to train the model but...
0:20:26.9 S2: Yeah.
0:20:27.5 S1: That's right.
0:20:29.2 S2: Okay.
0:20:29.8 S3: So then you recommend... What do you recommend for those of us who are using an in-breast imaging centers to do the initial screening? Do you find people aren't doing them for anyone under 40 or do we continue and then I see them and do the calculation as if they're 40 if they have less dense breasts, heterogeneously dense and extremely dense of course they'll continue to qualify but we've had quite a few patients who have been getting breast MRIs based on the calculations say 36, 37, 38 and then they turn 40 and it's the, you know, they have scattered fibro glandular densities and it drops down to 16%. I saw someone yesterday that was referred, so if you have any advice for what we should do practically.
0:21:19.4 S1: Yeah, others feel free to chime in. I have some thoughts but curious if...
0:21:23.0 S2: First I would... This is Idey. I would...
0:21:24.7 S4: I have some thoughts too.
0:21:27.1 S2: Yeah, the... And Holly, I can't wait to you add to this. The one thing I would say is it is correct, the creators of the Tyrer-Cuzick model when they created Tyrer-Cuzick version eight, the data that they used with the density was in a cohort that started at age 40. So that's why that age 40 is the baseline age that they're using. I'd love to hear others thoughts on under 40. I know that's why they chose 40 but I don't necessarily think that you can't use it in someone under 40. I just think that the validation in the training was on a group that started at age 40. Holly, what can you add to that?
0:22:17.2 S4: Oh you know, just a couple of thoughts. The under 40 about 70% of women have dense breast tissue and so that's... I'm not sure how helpful it is in terms of risk distribution in that age group. I don't know whether Alicia's done additional investigation into that area but with regard to the practical question of, what does one do when faced with a patient who you really feel should have MRI screening and if you enter scattered fibro glandular densities and suddenly when they're 40 it comes down. You have to ask yourself, why you're using the risk models? And ideally ultimately we'll use them to predict risk but I don't think we're quite there yet and so if one is using the risk model in order to obtain an MRI pre-authorization from an insurance company you may just not put in breast density in that patient's risk assessment would be, it would be my advice practically.
0:23:32.9 S1: The only thing I'll add to this is, there's been a... At least I have felt there's been confusion. I'll speak on my own about how to use risk models like Tyrer-Cuzick in particular on lifetime screening recommendations for women. So for instance if you have that 36 year old that's at a 22% do you screen that woman the rest of her life if at some point you rerun the model in 10 years and she's 18% or you add in breast density at 41 and goes to 16%? Fortunately I think recently the NCCN breast cancer risk reduction committee added a little clarity to this to use residual risk at least, lifetime residual risk when you're thinking about that. Most people would use that 20% threshold although, yes, I mean if based on certain factors some people even think about the 15-20% range is a little gray area and up for debate but at least it gives some clarity that you should at least be thinking about this as updated information and ways to manage your patients.
0:24:43.0 S4: Yeah, and you were saying, I don't remember what you were saying at the very beginning of that discussion. I can't recall what I was gonna say.
0:24:58.4 S3: I'm mostly curious if there are pushes or if we anticipate changes moving toward a shorter term. I personally feel like maybe five or 10 year risk thresholds would make more sense than remaining lifetime but.
0:25:12.8 S4: I know what it was. I agree with you Alicia and the lifetime risk is really just most useful for insurance pre-authorization. I'll recalculate it clinically every three years, that's, you know, we we don't leave like one risk estimate in there. I don't know what I would do but...
0:25:31.2 S1: Yeah, I was talking about at the beginning, the residual risk.
0:25:33.8 S4: Right.
0:25:34.9 S1: Comment, yeah. Yeah. Well good, Alicia...
0:25:39.1 S2: Well, thank you so much. I appreciate your input.
0:25:42.0 S1: Oh yeah I know, let's keep the questions coming.
0:25:47.9 S3: Okay, great. Let's see next slide. Okay shoot, I was a little worried about this, sorry. The mac and pc don't always get along but I think after this first line we'll be okay. So I think this first line if we could read it would say that Tyrer-Cuzick... The Tyrer-Cuzick model is used to estimate breast cancer risk. Seems like everyone's probably pretty familiar with this. I think it's pretty well established now work that I've been involved with work from independent groups has shown that the accuracy of the Tyrer-Cuzick model version seven, version eight can be improved by incorporating a polygenic score. We recently developed and clinically validated a polygenic score. I really believe the first and only polygenic score that's clinically validated for diverse ancestries. Dr Holly Peterson presented this work in a podium presentation at Esko last year.
0:26:56.2 S4: On behalf of Alicia Hayes. [laughter]
0:27:00.0 S3: Thanks. On behalf of a lot of people and I'm not planning to go into details about that but certainly this has been a major area of focus for a lot of us so please don't hesitate to reach out if there are any questions about that.
0:27:13.5 S3: And maybe I'll just suffice to say I really think this methodology we developed is an important scientific advancement. And I think it will be the key to making polygenic risk assessment equitable across all disease types. There's a big problem right now with polygenic scores not performing as well for non-European ancestries. And we've developed and validated a solution to that. Okay. But our topic today...
0:27:45.4 S4: Elisa, I wonder if you had any comments as to comparing TC version eight and the CanRisk model. I've had experience with that recently. You can't really use it in clinic. It's unwieldy. But in the research setting, in combination with the PRS, it's very, very interesting. And I just wondered about the validity of that model in comparison, in your opinion.
0:28:17.2 S3: Yeah, great question. I don't think I have any major objections from what I've seen in literature. I've played around with CanRisk a little bit. My goal was really to understand how we might be able to combine CanRisk with a PRS without going through their rather complicated methodology to have it built in. One thing that I think is a little unfortunate is that they are very attached to working the PRS through the segregation model, which would be great, except it's so computational. And they also have to do a lot of simplification. Instead of the continuous PRS, they just break it into... I think they might be using five categories now. So maybe not too bad. But just treat it as high, low sort of PRS in order to be able to make that, I think, computationally feasible. So yeah, I don't have any major objections. I know the risk estimates come out quite a bit lower. And I'd love if other people have thoughts about this. But my impression is this is mainly due to, for one thing, predicting risk to age 80 rather than 85. Also, I believe... Correct me if I'm wrong here. I believe the default in the CanRisk is to use the competing mortality adjustment. And often in Tyrer-Cuzick, you have that option. But I think the default generally does not adjust for computing mortality, which...
0:29:54.6 S1: Yeah, you'd have to... You have to click...
0:29:55.8 S4: Then it should, yeah, yeah. I've been getting similar risk estimations with the TC version eight to the CanRisk without the PRS. And so it doesn't seem that far off. But I typically will use competing mortality unless I'm un-checking it in a patient that I feel needs to have an MRI screen.
0:30:17.7 S3: Oh, that's great. It's great to know. That was my impression as well. So, really glad to hear that. Okay. So yeah, thank you. Next slide. Okay, so we had some really important questions to answer. First of all, we needed to understand if the polygenic score was correlated with breast density or more importantly, in statistical terms confounded. Like is it capturing the same risk information? If so, we need to be able to measure that really well, so that we're not effectively double counting that information. And there were good reasons to expect correlation. There are published UVA studies, obviously predicting breast cancer. There are also UVA studies predicting breast density. And they have come up with some of the same genetic markers. So this was an important question for us. And then if they are correlated, we have to have enough data to be able to measure really the risk that's independent to the polygenic score after accounting for breast density. Here, I'm showing a figure that was part of a presentation I gave at the BASTRA conference a few years ago. We had a similar problem with our initial development of a score, combining the polygenic score with Tyrer-Cuzick version seven, because we knew the polygenic score should be correlated or more specifically, confounded with family history.
0:31:52.2 S3: And normally, like the normal thing according to the statistical best practices that you would do when you have risk factors that are at risk of double counting risk information, is you throw all the individual factors into a multi-variable analysis and build your combined risk model in that way. But we don't have the luxury of doing that because Tyrer-Cuzick is already built. We can't really rebuild it from scratch using the independent risk factors. So this work that we presented is a methodology to really achieve the same thing as the multi-variable analysis, that would be recommended by statistical best practices. But when you don't have the luxury of actually doing that analysis, when you have an existing model and you wanna add one thing in, and it's correlated with some of those factors in the existing model. And we call this the fixed stratified method. Here you can see... Sorry it's a little hard to see. But often in literature, you'll see people building risk models where they just assume everything's independent or just treat it that way, because of the complexity of handling the correlation. And here you can see an example of over-estimation. The green dots are overestimated, if you ignore the correlation. The coloured lines are for multi-variable O analysis, so the correct estimates.
0:33:23.9 S3: And then it's probably hard to see, but the red dots from our fixed stratified method are lining up perfectly on those multi-variable estimates. So anyway, this was family history. But my point here is just we have a way to account for this. It's just that we needed data to be able to measure it really well.
0:33:43.4 S1: Yeah, thank you. And I see Michelle Weaver knows the question, how do you explain to a patient what PRS is? Others can chime in. I have my thoughts.
0:33:56.4 S2: Please go ahead, TJ.
0:34:00.3 S1: It's just background genetic factors that if added together, an aggregate can add to breast cancer risk stratification, I guess if you're specifically talking about breast cancer, but it's their common to any... They're seen in any common disease like diabetes, cardiovascular disease, colorectal cancer, prostate cancer, etcetera. Here we're talking about breast cancer in particular, and the way I also think about it is things like CHEK2 and an ATM, we're getting pretty comfortable with at those mutations as is... You know where... How they affect risk for breast cancer, so we tend to think of CHEK2 and ATM just for instance, of them having odds ratios for breast cancer of 2 to 2.5 or something like that, for argument's sake, that would be the simple way I would think of correlating that would be if you have a woman with the, a lifetime risk for breast cancer of the average population, we tend to think of that risk for about 10%, so that would be like an Odd ratio of one, and then if you go to two, that would be like a risk of 20% somewhere in there.
0:35:19.7 S1: So that's an example of one gene making a pretty large effect on a woman's risk for breast cancer, but there's a lot of variance even within those genes, let's now take in CHEK2, so that adds ratio of 2 that I kinda brought up, or two to two and a half. That is kinda for most of the loss of function variance, strong Miss experience etcetera. If you look at... Something that a lot of people on this call are probably familiar with. I157T, for instance, so a CHEK2 variant, it's common in populations about 1 in 200 people have that variant, it's well known to not be a strong breast cancer risk factor. We have a paper under review right now that they're not under review. Sorry, under review internally, I guess we could get it out under review externally, that shows that the odds ratio for that particular variant is around 1.2, 1.3, meaning that if you have a risk for breast cancer around 10%, it might take you it to 13% or 12%. So in a sense, if you just look at that one variant, it's a bit meaningless.
0:36:36.1 S1: Because you're probably not gonna drastically change the way you're evaluating or managing someone based on a difference of their lifetime risk for breast cancer going from like 10 to 12% or 13% for instance, but if you take a bunch of those variants that have those low odds ratios and flip them together, now you can actually start getting some meaningful risk stratification because kind of people spread out, and there are not some people that really do have... Still like a 40% lifetime risk for breast cancer based on just having many unfavorable risk snips. So that's a framework that I think of. For instance, in our polygenic risk score, the CHEK2 example that I use, the I157T, that's our first snip, so not all of them are just random snips in the middle of genes, some of them are actually in genes themselves, but this is the way to really better inform on the risk of someone that has that variant for instance, because then you can pull in to other genomic factors beyond it, and then what we have learned is exactly then the getting back to what Alicia is talking about here, which is... It even adds more, if you can bring in other factors like clinical and family history variables and for instance that we're about to show.
0:37:53.0 S2: I think practically too, you know, I don't know whether you were looking for sort of a short practical, you know, patient oriented discussion, but, you know, I'll explain that there are common genetic variants that individually confer very small levels of risk, but an aggregate it can affect one's risk, significantly explaining, up to 20% of familial clustering and also subs stratifying risk in gene carriers, you know, affecting the penetrance or the likelihood that you'll get cancer within that genetic category. And then I'll go on to explain that, you know, we estimate your risk with mathematical models, but that of course doesn't include any of your own genetic information. And combining the two is much more powerful.
0:38:53.9 S1: Yeah, that's a good way to say it.
0:38:53.9 S2: Little patient spiel.
0:38:55.3 S1: And I see, Allison, thanks for coming on. Allison, you brought up yes, specifically then getting at what I said, that I157 CHEK2, would it be worthwhile to send to a lab, like yes said, does a PRS, etcetera? Yeah, I think that's just provide a preference. I mean, you know, I would argue that to make sense of that particular variant, you definitely need more information, so you need to pull it into, you know, in my mind, a polygenic risk score I would argue and you need to pull in clinical and family history variables, or else you just can't really make, a lot of sense of it on its own because you know, it just becomes too generic, so, well, great. I'll go to the next slide, Alicia.
0:39:36.8 S3: Okay, Maybe I'll just kind of skim over this. So we were able to do development work with a cohort of about 12,000 women who had, BI-RADS best breast density measurements and the PRS and all the other Tyrer-Cuzick factors. We'll see some results presented by categories like sub cohorts defined by self-reported ancestry, but I just wanna clarify we do not look at self-reported ancestry at all when we do these calculations, it's entirely based on, you know, just the Tyrer-Cuzick version eight and genetic markers. And yeah, then we evaluated this combined score in an independent cohort of 6,000 women with BI-RADS and polygenic data. We looked at the relative contributions of the individual factors, family history, breast density, the PRS, and we will look at reclassification. So we'll see how much adding the polygenic score to Tyrer-Cuzick version eight, can change classification with respect to a 20% threshold on remaining lifetime risk. Okay. Next slide. Oh dear. Sorry I'm so sorry shared from my screen. That should be African American. But yeah.
0:41:12.1 S1: It's interesting. Oh yeah there are some other ones. Look that.
0:41:14.6 S3: Yeah yeah, the jolly Roger at the top. I don't... Anyway. I'm sorry, probably sharing from my macs and PCs didn't really get along. Okay. So pretty much what I just mentioned 12,000 women in the development cohort, some of the analysis had to be restricted to unaffected women and some analysis were actually predicting breast cancer. So using all 12,000, the vast majority here were unaffected. And then 6,000 in an independent sort of test cohort. This looks kind of typical for our patient population, mostly women of European ancestry, but we tend to have, actually this looks a little lower than usual. I think it's usually closer to like nine or 10% Hispanic and African American. Yeah. Okay. But roughly comparable to our expectation. Okay. So did we skip one I'm so sorry. No. Okay.
0:42:18.6 S1: I don't think so.
0:42:20.3 S3: Maybe go forward one more. Sorry. Maybe I skip, I think I accidentally deleted one. Okay. So as we mentioned earlier, I was planning to show kind of similar-ish figures from literature. We already mentioned that, breast density is associated with age and with BMI and the Tyrer-Cuzick developers did a lot of work to take that into account. I think of this, like they made their own breast density score. That is a function of not just breast density, but breast, breast density, age and BMI. So as a first step, we had to really understand that score because what really mattered for us was not correlation of PRS with breast density on its own, but rather with the score, the breast density score in Tyrer-Cuzick, and here this vertical axis apologies, it's not well labeled, but you can think of it as, really the score inside the Tyrer-Cuzick model that takes into account that's, it's like the, the log of the relative risk associated with breast density, according to the Tyrer-Cuzick model.
0:43:31.1 S3: And the colors here are by BI-RADS category, purple at the top is extremely dense. The kind of green at the bottom is, fatty. So quite clearly there's a relationship with BMI in terms of how this was incorporated. The age association is not as obvious, but that's basically the width of each of these bands are due to the age of the patient. So anyway, I thought it was kind of interesting to see, you know, looking at the vertical access. Anyone at the same height is getting the same risk due to breast density. And there are certainly places where like the purple blue and reddish are kind of all at the same level, despite being, you know, extremely dense on one hand or all the way down the scattered fibro glandular, depending on age and BMI.
0:44:31.3 S1: Yeah, this is a good kind of reality check to make sure the data looks good.
0:44:37.2 S3: And right. Kind of a challenge for us, to like reverse engineer based on literature and then based on, you know, actual trial and error with the calculator to make sure we really understood how this went in. And yeah. Great. Thank you. So I was, surprised, but I guess kind of happy to see there's very little correlation between the polygenic score and this breast density score, if you like, that's incorporated in Tyrer-Cuzick. There is a significant P value. It's very easy to get significant P values with large data sets. So I think we're looking at, the 11,000, each dot in this figure is one of the 11,000 unaffected women from the development cohort. We do have a positive correlation, but it's like 1% it's super tiny. So arguably it's pretty okay to just incorporate these as independent, but since we already have well established methodology, we just worked it through our typical methodology that does account for that 1%, although that's pretty minor.
0:45:46.1 S3: Okay. Here. We're looking at an ANOVA analysis to describe how much each factor contributes to the model. Something important here. If you're familiar with ANOVA, here, we're really answering the question, how much of the variability of the combined score PRS plus Tyrer-Cuzick version eight is due to each factor. It really matters the order in which you enter the factors into the ANOVA analysis. So we gave, we decided to go kind of historically we put family history in first and it gets an advantage. I can tell you if you put PRS in first, it gets to like 40%, and breast density gets close if you put breast density in first, but we thought let's go kind of historically. So first family history, then the other, we kind of just combined all the other factors that are in Tyrer-Cuzick version seven. So, you know lots of factors, and then breast density, which is kind of the newer, a newer addition to version eight. We put that in third and finally at the end the PRS and the way to interpret this is that after accounting for everything else, PRS explains 20% of the variability or after accounting for everything in version seven breast density, explains, you know, also a substantial fraction of the variability in the scores.
0:47:19.0 S1: Yeah, that's interesting. And I didn't... You know, quite take a step back and appreciate that, since that was the last put into the analysis of variance. A model that... Yeah, that it still showing this strong of effect is pretty impressive.
0:47:34.2 S3: Yeah. I feel they're all actually really pretty equal, family history, PRS and breast density and contributing, there is some correlation between them, but they're close to independent and for accurate risk assessment, each one adds a lot.
0:47:51.5 S6: So Alicia, this is layering in your population, right? And accounting for the prevalence of these risk factors in that group, right? So, this is maybe a more informed view than just looking at the relative risks associated with these individual factors. Is that fair to say?
0:48:07.5 S3: I think so. Yeah. And maybe along those lines, I think family history looks a little stronger here than it would in a general population sample. Our patients can have anywhere from like known family history to really extreme family histories. We're definitely enriched for the strong family histories. So I think we have more... I think that's a stronger contributor in our population where if you were just to look at relative risk, you wouldn't capture that. I think I agree with that. Yeah. Great. Thank you. Okay. Here, we're looking at reclassification with respect to a 20% threshold on remaining lifetime risk. And what we see is that overall 16%. So what is that maybe roughly one in six or so women are classified differently with respect to this threshold by Tyrer-Cuzick version eight versus this combined score that adds the polygenic score.
0:49:18.0 S1: Which is really, yeah. So for clarity this is Tyrer-Cuzick version eight, plus the polygenic score. Polygenic risk score.
0:49:23.4 S3: It might... For today's talk. I'm just really wishing I had the same thing with breast density. Breast density is also really powerful. I think CRS does a little more reclassification or the polygenic is a little more, but right. Both are super important. Overall you should have roughly the same number of women called high risk. It tends to be just slightly lower with the combined risk score. But a lot of different women called high risk. And you can see...
0:49:57.5 S1: Yeah, this is more than just a reclassification when you add in the polygenic risk score. Yeah. I see what you're saying. We have that other figure with breast density.
0:50:04.0 S3: Yeah. But yeah, so I guess the point of this slide is the polygenic score increases accuracy, and it actually gives a really different result for some women. Okay.
0:50:18.4 S1: Do you remember the reclassification from the other figure with just breast density alone? I don't remember it.
0:50:24.5 S3: Yeah. We never made a fancy graphic like this, apologies I don't have it handy. It's similar, but a little lower.
0:50:28.8 S1: Yeah. I cant remember if we had it in this abstract.
0:50:33.1 S3: I don't think so.
0:50:34.0 S1: In the results. Yeah. Maybe not.
0:50:37.4 S3: I think we ran out of space.
0:50:38.7 S1: Yeah.
0:50:38.9 S3: But it was a little lower for memory, I would guess like the 12 or 13% reclassified similar to the PRS but just slightly lower.
0:50:49.9 S1: Yeah. So we can go quickly through this too, because very similarly then this is looking at Tyrer-Cuzick version eight too the combined score that includes the polygenic risk score. So it's not really independently looking at the effects of breast density, but you know, it kind of largely gets at the same. I do wanna leave some time at the end for question. So yeah, so then this gets back to kind of like the synopsis of the abstract looking at polygenic risk score, the addition of that on top of Tyrer-Cuzick eight alone. So clearly it increases your ability to do better risk discrimination.
0:51:31.8 S3: Yeah.
0:51:32.8 S1: Good. So, thank you Alicia. That was great. I'll stop there. Let's... 'Cause I know there's some other questions coming in and we only have seven minutes. I wanna make sure we get to them.
0:51:45.8 S4: I have a question, couple questions from the chat.
0:51:50.5 S1: Yes.
0:51:51.3 S4: Are you okay with me, Jerry? So there's a question as to why Tyrer-Cuzick eight... Is there a scientific reason why it's partnered with Volpara versus other third parties? I think it's also with MagView. I didn't know if there was other reasons for that other than a business reason or...
0:52:15.4 S1: Yeah. I mean, I guess there's different definitions of the word partnership, one is. Yeah, I mean Tyrer-Cuzick can be licensed and used across different software. Like we have a license for instance, for Tyrer-Cuzick which allows us to use it in our fashion for risk work. Many other companies have that kind of license and then, you know, getting at the breast density itself. Yes. Why... You know you have BI-RADS, Volpara, VAS. I would think that's a bit research mixed with some potential business. I mean, I know at least we have some people from Volpara and if they want to comment there.
0:52:50.9 S7: Yeah. Hey DJ, it's Dave Mesocrady from Volpara, how are you?
0:52:54.5 S1: Hey Dave?
0:52:55.2 S7: This was great. Thanks for having us.
0:52:57.1 S1: Yeah.
0:52:58.2 S7: I think I can probably explain it is that the... We started working with Dr. Cuzick with volumetric breast density assessment many years ago when he was starting with the version eight. So, we did it over at a number of locations in Europe and here in the United States and validated the volumetric breast density percentage into Tyrer-Cuzick eight.
0:53:22.7 S7: So that's the Volpara density portion of it, that's how it got into Tyrer-Cuzick eight. Many places can use Tyrer-Cuzick eight, you can get it for free off of the web or it's integrated in with Epic or Cerner or some of the big EMRs. Or you can do it in a Pen Rad or MagView. So they can report Tyrer-Cuzick eight, but it's the Volpara volumetric breast density percentage that is embedded inside of Tyrer-Cuzick eight. I hope that makes sense.
0:53:56.7 S1: Yeah, yeah, I see your side here.
0:54:00.1 S7: Yeah.
0:54:00.8 S2: That's clear.
0:54:00.9 S7: I don't know who shared that. Brian? Somebody did... [chuckle]
0:54:03.0 S6: No, no, that wasn't me.
0:54:04.5 S7: Miraculously I speak...
0:54:05.9 S1: Rogue sharing had gone on, that's fine.
0:54:07.2 S6: Yeah. It comes down to the data, right? So, Volpara really is unique in that the technology has a volumetric measurement as opposed to a traditional area-based, where the use of medical physics to interpret the attenuation across a known breast thickness derives a density map. So there's an attempt to accommodate for a three-dimensional view, and that's a different kind of measurement than the 2D stuff. So Tyrer-Cuzick put all three, right, into the model. There's the 2D visual assessment, there's the 3D volumetric, and then there's the BI-RADS. And those three paradigms became part of that model.
0:54:50.2 S1: Yeah, no, thank you. Very informative.
0:54:53.2 S6: And I had one other one that snuck in. This was something we did with the EWEC, where we talked about reclassification due to breast density. And this is something we see across our site to be variable to the population, so that distribution of breast density, right? So, although, TJ, I think that opening distribution of BI-RADS categories is a legitimate sort of high-level view, we do see breast density vary quite a bit across the country, depending on where you live, and those populations are gonna impact this reclassification, for sure.
0:55:30.5 S1: Yeah, good point.
0:55:31.7 S2: And then can you speak to the clinical utility of using these risk-based models for women under the age of 40?
0:55:40.0 S6: Well, certainly the... I don't know if that's a question for me, but I'm happy to start it off. But the guidelines certainly recognize, and all the major medical societies on the topic are talking about doing a risk assessment for women prior to the typical age in which screening mammography would start. I think likely to recognize the significance that hereditary cancer has on younger women, right?
0:56:07.2 S1: Yeah, trying to get to that 25 to 30-year-old woman as the ideal time to do some screening, so that's popping up in many guidelines now.
0:56:16.8 S6: Yeah, absolutely. And we know those women are most likely to have the denser breasts in the population, right? So certainly we believe that by studying that group of women more, and by studying populations with more extreme dense breasts, we're gonna learn more about the associated risks. But I think the notion that younger women prior to the age of screening mammography should have a risk assessment where at a bare minimum we collect their family history and identify who needs to have additional conversations about the risks of hereditary cancer for sure, seems a no-brainer.
0:56:54.7 S1: Yeah. Well said.
0:56:56.2 Speaker 5: I would say in those groups too, the 25 to 30, or really the 30 to 40 even, there are a variety of family history weighted tools that can be used to assess risk. So, a Claus is one of those, as well as BOADICEA. So there are options out there outside of the breast density piece that can provide that family history weighted model information that can guide management decisions.
0:57:30.5 S1: Great. Any other questions? We have about 30 seconds.
0:57:33.9 S2: I have an important one.
0:57:36.0 S1: Go for it. Thank you, Holly.
0:57:37.6 S2: When will the PRS for Myriad be commercially independently available? I ask this every three months or so.
0:57:46.7 S1: Yeah. The PRS itself?
0:57:50.9 S2: Yes.
0:57:51.6 S1: Yeah, so that's still being evaluated. It's definitely not anybody's decision on this call. [chuckle] So I'll put that out there. The breast density...
0:58:04.1 S2: I'll keep clamouring.
0:58:05.5 S1: Yeah, no, definitely, please do. The breast density, that is being evaluated right now, as when we can bring that in. There's great research here. Clearly it shows that you wanna be able to incorporate that, people have wanted our model to go to Tyrer-Cuzick eight at least for some time. So we're working hard there, for sure. But yeah, good question.
0:58:29.4 S3: I do feel fine about adding the PRS Audrey show from our report into the Version eight calculator. Yeah, the only issue there is if the family history changes from what was reported, it'll be slightly off, but that's a pretty small difference.
0:58:49.2 S1: Yeah. No, good point, yeah, yeah. Well, great, well thanks everyone. We are at time, wanna be respectful. I appreciate everyone spending their Friday. Please join us on the 22nd of July. We'll have Dr. Peterson, so you just got a little segue of maybe what is to come there, so that's fantastic. I'll try to sprinkle in also something in the middle, so please keep apprised of, at least I'll share it on LinkedIn or it'll definitely go up on the website. I also just opened a Twitter account that I am horrible at using, so my goal is to get better over time. [chuckle] So feel free to tweet me, whatever that means, and I will sort it out. [chuckle] Alright, well, thanks everyone for coming on.
0:59:35.2 S2: Thank you.
0:59:35.4 S1: Oh my Twitter handle, Jeez, I think it's like Slavin TJ or something. [chuckle] Well, thanks.