Dr. Strom discusses the value of imaging and how to use outcomes research with imaging data to better inform our clinical practice and clinical management.
Well, thank you so much for that uh really kind introduction and uh and, and thank you again for the invitation. This is uh I I I think a great opportunity uh for me to be able to, to, to talk with you this morning about um uh a little bit about the value of imaging and how we can potentially use outcomes research and connect outcomes research uh with, with imaging uh data to really be able to better inform our, our clinical practice and, and clinical management. So these are um kind of closure. So, uh today we'll, we'll first talk about the current landscape of imaging use and, and value uh for imaging and then we'll talk a little bit more about uh the state of outcomes research and imaging and existing registry data that might uh be tapped for, for really looking at some of the central questions. And then uh we'll, we'll, we'll talk about two case examples of use of outcomes data and imaging that uh exemplify how we can use these data to be able to better inform our our clinical management. So I'd like to start with this uh pyramid. This is uh the Thornberry Andre Beck pyramid and, and some of you may have already seen this, but uh Thornberry and Frey Beck were both um um uh radiologists uh who theorized back in the, in the 19 seventies that, that evidence and within imaging really started in, in the sort of form of a pyramid that is to, to show that something was uh I I improved the, the therapeutic efficacy or the, the efficacy of the technology at a patient level, you really had to establish that it technically worked well. First, you have to make sure that um that it diagnostically was, was accurate, et cetera. And so those were really the base of the pyramid. But at the very top of the pyramid was this concept that if we use this technology or di di diagnostic modality in on a broader scale, then it has AAA broader reach. Um and, and may be sort of efficacious on a societal level. And so they theorized that the the clinical impact of this uh pyramid was, was really an upside down pyramid. That is while we want to uh we, we really want to get to the top of the pyramid because we can maximize the impact that we have on a given uh framework. And this has been done for particularly well for, for example, C T um in, in evaluating outcomes after C T scans. Um uh but, but it, but in fact, actually, I think echo and I would stand to reason that echo is really towards the top of the pyramid at this point in time. And we have a lot to do in terms of this uh area of really exploring how, how do we better utilize this technology? Um And when we think about do domains of, of diagnostic evidence, we can really think about sort of three main uh domains. One is test attributes. Can we can we actually can the test detect the target condition? Is it accurate? Is it reproducible? Is it available? And the second is, does it really alter our behavior? Does it alter the diagnosis or our management? And then ultimately, at the end of the day, we're interested, does it change our patient outcomes? Does it improve resource utilization or is it cost effective? And and why, why does this matter? So this is uh data from uh now a little bit uh time ago but, but still relevant. Uh but this is back in 2006 and looking at the growth in volume for uh physician services for be Medicare beneficiary uh 1999 to 2004. And in the purple boxes, you can see it's imaging and so imaging outpaced the growth in E N M or, or other major procedures uh during that time period. And, and so as a result, the crosshairs are really on imaging to demonstrate value because it is such a costly technology. And So, uh well, where do things stand with regard to sort of imaging overall use in, in the US? Well, we don't really, I'll come back to that, but we're going to start with, with Canada because we have a little bit better data from Canada. And here you can see uh growth in the rate of echocardiography per, per 1000 persons from 2001 to 2009. And what's really interesting is, is, you know, obviously there's an increased trend here. But, but that only tells you part of the story. Um If you look at the rate of repeat echocardiograms, you can see that starting in 2003, there really has been an escalating increase in repeat echocardiograms compared to all echocardiograms as a whole. And, and when you drill down into that a little further, you can see this is from the same article from Ontario. And if you look at um the where it says the number of persons, uh as I said, the percentage of uh the repeats performed by the same person, you could see that overall and this hasn't changed consistently over time. More than half of all epoch cardiograms were performed by a different physician. So we have a, we have a trust issue um in the US and that we um we often re perform echocardiograms. Um uh If we don't, for example, have the images from those echocardiograms. So how many times have you have you accepted somebody from another site and, and decided to repeat the echocardiogram as opposed to um looking at the, the images. And so that really fundamentally that get that, that information that we, we see from the echo. Uh We, we don't necessarily trust. And so, um well, is it just the modality, are there other types of imaging that might be uh better for this purpose? And um uh I, I can say here that, that there's been a, a large growth in, for example, cardiaca. This is also from Canada, looking at um age, age and sex, standardized rates of use of, of technologies in in those with heart failure. And you can see that since 2006, there's really been a huge rise in cardiac, this lighter blue color but ultimately, uh aroar gray is still the the lion's share of, of what we do here um in the US and um or in or in Canada. And if we look at costs, we can see that the total costs have been uh rising in Canadian dollars. Um But, and at the rate of, of, of cost for, for advanced imaging technology such as cardi or otherwise, um it seems fairly flat and it's really this brown uh here which is the resting ear that's driving the increase in cost. And so we really have a problem and that we need to justify the cost of, of imaging. So well, you say that that's Canada but what about the US? Um and the US is it we, we have far less data on this actually. But back uh if we look at uh 2011, there was uh a hr Q publishes these data points which um with the agency for health care research and quality and they, they uh published data points which are sort of synopsis of uh the data. And, and they did want an echocardiography in 2011. And, and then here you can see that 80% of all Medicare beneficiaries in a given year got an echocardiogram and 15% got two. Um And if you look at where those were done, you can see there's a, a vast range of across states uh in terms of those who are potentially eligible, receiving an echocardiogram. So there's, there's lots of variability in this area. But, but I think this gets at the sense that we're using a, doing a lot of echoes. Um 25% of all Medicare beneficiaries get an an echo in a given year, 8% receive 1 15% received two. So, uh what about the current state of affairs? So this is um uh again, that was just Medicare data. Um We don't really have a national health care system here in the United States, but we, we did look um there's a, a commercial analytics company called Ever Sana that has aggregated about 90% claims on uh uh 90% of the in the US. And you can see you, you're here there. Um It's about 280 million patients that they have overall. Um and we've looked at this, we have AAA paper I looking at contract utilization as part of this. Um but um they, they really collect all payer claims. So we looked at 2019 to 2022 there were about 17.6 million patients with at least one echocardiogram. And, and so we would estimate um that about five point, you know, 5.3 million patients receive an echo annually, which is about 1.6% of the US population. So even if outside of Medicare, there's a, there's a, a massive use, you know, well, you say, well, echoes are obviously very useful and why, why is uh why does anybody doubt that? And if you look at the data behind this, you can see that this is the at least one paper that was published back in, in Jack in 2016 where you can see this is an analysis of the national inpatient sample. And if you look, you can see that echo volume has been increasing. And concurrently, you can see that ho in hospital mortality has been decreasing and, and uh hospitalization length of stay has been decreasing. Um And uh and now that that's well and good. Um and uh they showed that Echo was only performed in about 8% of uh heart failure hospitalizations but was associated with a significant reduction in, in death and in all these uh populations. But what's the problem? Um Well, uh first of all, they don't actually need to be the same hospitalization. This is a secular trend. You could see, for example, that, that an echo was performed in somebody. Um But it, it just happened to be, uh but it wasn't the same person who died. So, you know, there's, there's that um the, the aspect of sort of a um ecological fallacy as we sometimes talk about. But then there's also the, the issue is that you have to survive long enough to receive an echo. And so the people who receive echoes are different than those who don't. And, and so there's a, there's a uh a survival bias inherent in all of this and that if somebody comes in the door crashing and burning, then ultimately, they may not have time to receive a full echo. And so we sort of penned um uh some of these concerns and, and what we think is the next direction heading forward in an article in Jase. Um and, and it really fundamentally, you know, that there's a lot of focus on overutilization. And then in part is because it's, it's easy to measure overutilization, it's harder to measure underutilization. Um And there are a number of other challenges that are, are um uh inherent in, in demonstrating value, for example, it's harder to separate the proper uh value of the diagnosis versus the treatment. Uh And an example of this is IC DS and, and, and uh and uh ejection fraction. Ideally, if we get our ejection fraction right, um we're gonna have an effect, um a preventative effect from IC DS that's similar to those that are seen in the trials that being said or even potentially even greater than that uh that being said, um It's hard to necessarily uh separate the value of that, that diagnosis of getting the E F right versus the actual treatment of uh the uh the IC DS, which the IC DS are ultimately making the difference. Um The impact of mistakes is harder to measure. We don't really have a ground truth, we can compare it to another modality. But ultimately, that, that modality still has a um a a ground truth that they have to compare against as well. Um So it's hard to measure if somebody makes a mistake um and it's acted upon or not acted upon what, what ultimately ends up happening at them. And nevertheless, we we really do need to study outcomes to justify the cost and inconvenience of, of testing. So how do we use outcomes or how could we use outcomes? Well, one is uh using imaging itself as, as a sort of data source. And so using imaging as a biomarker, I often say to our fellows that imaging is the best biomarker of the heart. It's better than any component or B MP. It's really amazing how much information you can get out of it. And so we can use it for preop risk stratification to guide shared decision making, to guide the use of treatments or other diagnostic tests and to understand cardiac structure and function. And then secondarily, we can use images ultimately, at the end of the day, images are fundamentally high resolution data arrays of ones and zeros. And we can use those uh to be able to actually uh for computer vision to be able to actually make predictions and, and uh assessments and things. Um And then ultimately, we can use imaging as a surrogate outcome. For example, for trials provide outcomes for trials like L B mass alen enhancement. And it's already starting to be to use in, in some cases. Um But we can also use outcomes to understand imaging and that we can use outcomes to understand variability in imaging to, to understand where why there are uh why there's variation. And if there that variation translates to differences in outcomes, we can use the outcomes to understand areas of underutilization potentially um where uh and we can use outcomes to define ultimately normality. Uh And an example of this is aging and diastolic function. Um As many of you know, the uh transmitral E N A patterns switch from a predominantly early diastolic to a predominately late diastolic pattern at around age 60. But does that mean everybody in the population develops uh impaired relaxation or, and, and what constitutes uh a 60 or 70 year old who ultimately is at risk for heart failure um compared to another 60 or seven year old. So we can use the, the those outcomes and, and heart failure hospitalizations. For example, in this case, to be able to help, better define what actually constitutes normality. And this is a big data talk. So why, why big data? Um and, and thankfully, our outcomes of interest in cardiology are generally still uh uncommon. Um And there's other, also this sort of more um theoretical reason, which is that uh as we uh if, if we're collecting quantitative measures, unless we collect them in a biased manner, there, imaging measurements in large numbers are ultimate in app app app proximate the population means. So I I I see a lot of people say, oh well, you, you measure it in a core lab. Well, you know, core labs in of themselves are really critical if, if you have to have a very precise measurement and precise measurements are really critical if you're dealing with small numbers, if you're dealing with huge numbers really, really. But at the end of the day, what we're interested in is do those uh is, is the population inferences we're able to make from those uh numbers. And so collecting them in large numbers is gonna get us closer to making population inferences. Um The other is a practical reason which is that the technology advances have, have really made large data analysis feasible. And if back in the seventies uh regression uh was was technologically challenging to, to do from a computing standpoint. Um because it um it requires an iterative um computing to be able to, to, to solve a maximum likelihood estimation. And so this is now available uh on my laptop in about less than uh a millisecond. Um I can, I can do a logistic regression and, and, and it, and it's out an answer. So technology advances have have made uh this large data ingestion really much more feasible. And then ultimately, the the the availability of large multi center registries um can improve the generalize of these results and ultimately understand uh particular subgroups of interest. Um And, and, and then, you know, Encouragingly there have been a sort of increasingly large data repositories that are being built through our interactions with the health care system. The problem is that few of them have relatively few of them have actually been linked together to be able to develop these sort of newer immer. And there are a number of challenges I think to using imaging data. Um uh One of which I've already mentioned, which is that the data are really only actionable if the diagnosis or misdiagnosis is recognized. So if you make a mistake and, and call something a um uh an M I um uh but, but somebody ignores it. Um Then what's the impact of that mistake? It's literally um minimal potentially. Um uh But, but the, but the converse is if you act on a, on a, on a mistake, then it, then it, of course has uh ramifications for outcomes. Um There's large amounts of missing data. So I don't know how it, it works at Centra, but, you know, if you're reading an an echo, um and we don't have the ability to be able to see a given structure, um we will often leave that uh second blink um or leave the measurement blink. And so there's large amounts of missing data uh in, in all sort of imaging um uh technology that we've seen um they're hier critical data structures. So uh an example of this is uh you know, a given person may have multiple echoes and may have multiple types of echoes. Um There are multiple modalities with within echo itself. Um And then a given person exists within the clinic which is exists within the hospital, which exists within um a state, for example, which exists within the country. And so we have these hierarchical data structures and that actually influences how we analyze the data from the uh statistical standpoint. Um And so it would influence the standard errors and, and the accuracy of our um uh of our uh calculations if unless we use specialized means to, to evaluate it, there's large amounts of col linearity. So for example, L V internal uh systolic diameter is very correlated to a diastolic di uh diameter. Um There's uh differences in, in variable names, in, in conventions and that's uh an ongoing issue. And I'm, I'm part of a group through the AC to, to um put together uh a guideline standard for uh adult echo reporting um which is in part motivated because of the work that we've been doing um to try to um extract data from uh from imaging reports. And so uh difference in variable names and conventions is a big one you centimeters versus me millimeters, et cetera. Um And then ultimately, there are data entry errors. Uh and those can be sometimes hard to, to know, you know what it's an outlier versus what's a true data entry error. Um You know, obviously some things are non phaser logic. If you have an a negative value for a, a true volume, that's, that's not uh you know, that's gotta be an error but, but the uh the border line at the borderline states, it could be someone hard to distinguish what's an outlier versus what's what's true. Um And then uh I already mentioned this already but different study types within the modality. So with even within an echo, we have T E trans and stress echo as as well, um there's large amounts of unstructured data. So a lot of people um have uh a text block uh or edit a text block as part of their conclusion statement or it's part of their indications. Um And it's, it's a little bit otherwise we can use natural language processing to go through this. But otherwise it's hard to get structured uh information from this and then the images themselves are, are large and, and uh complex the videos as well as still frames. They have patient information attached to them um that, that makes them hard to, to analyze. And then fundamentally, at the end of the day, there's, there's a real bias in, in who we see. Um Because ultimately, at the end of the day, there's a, uh somebody thought that there's an uh suspicion of cardiac disease. And so they referred them for, for an image. And so therefore, there's some referral bias that makes those people different than the general population. Uh And the last is what I like to call imaging leakage. You know, I don't, I don't know how uh this works in, in Virginia. But um up in Boston, we have multiple uh medical centers all around the same town. And if somebody doesn't like what they uh you know, see on, I don't know, B I, they'll go, go across town to M G H or to Brigham and, and get their uh echo done there and uh and vice versa. And, and so uh we're not capturing all of the data on our patients in the observation. So if we're observing a disease, for example, disease progression, um they uh we, we may be missing certain peach features of that. So what do we have so far? Um So we have a data set called Encore. It was actually um constructed by Pam Douglas as part of a Y2K project. And it ended up being a reporting software for, for 18 years and it's collected structured echo data on 100 and 35,000 individuals. Uh 271,000 echocardiograms. We've linked those to the uh Social Security death master file uh data as well as uh linked it to 100% Medicare fee for service claims that we have from 2003 to 2017. And that independent includes about 100 and 33,000 echo reports on 64,000 individuals. And we've used uh claims algorithms to generate uh 23 clinical covariant hypertension smoking, diabetes, and then ultimately, outcomes that had things that happened after the, the, the echo. So we know what happened before and afterwards. And I should say that um we would help them in capture. Uh We, we have another about 100 and um uh and 3000 um echocardiogram reports that we are adding to this about 9000 variables for each echo. So we have quite a lot of information and we're also in the process of linking that to Medicare fee for service claims again, the other is that we uh at, at B I have a data set called Mimic and, and, and that's a publicly available data set. And those of you may have work with it if you, if you work within the machine learning space, but it contains clinical and lab data from over 60,000 IC U admissions at the B I. And there's a joint partnership that we, we, we have a from the Massachusetts Life Sciences Center um to, to really sort of in between us and M I T and um uh and Mass Mass Life Sciences Center to really build out this mimic data set. So mimic three, which is the most recent version, uh features 350 deidentified chest x-ray dico images linked to uh patient information and clinical data from 260,000 E D visits. And, and we're adding now mimic four which will add about 100 and 45,000 transthoracic echoes. Um the raw, the raw data from those echo uh cardiograms as well as 980,000 EKG dico images. And so all of that will be there and you'll be able to do some really interesting novel cross comparative work and see what the added value of that of echo is compared to EKG alone, etcetera. Um There's also a number of foundations and non for profit companies that are are uh have uh imaging registries. The S AM R registry is the is the biggest one so far. Uh But mostly has uh image reports of 62,000 card from a across the country. But um uh in the image Guide Echo registry, I'll talk in in depth about um uh a little bit more. Um So image guide a registry is an attempt uh from the AC as well as a to be able to build um image registries, not only with images but also with the imaging data themselves. The reports um UK Bio Bank has a lot of information. And then we've had a lot of collaborations with NATA, which is the National Echo Database of Australia, which is the largest database Echo Database in the world. It's over 40 million echo reports across 14 labs across Australia and it's linked to 60,000 deaths and they really are sort of very forward thinking uh when it comes to this. Now, if you don't like giving up the um your own data, um there, there's um there's a Federated data networks which is uh the Sentinel network, the net and N H collaborator. And, and, and all three of these are set up in a way that allows you to sort of participate in individual projects. So rather than pulling the data together data together at a central source, this uh Federated data network is essentially a allow it's an, an alliance of uh of institutions that are, are, are looking to do data analysis. And if there's a given project, you can send that project request out to individual sites and they can choose to participate and send just a selected part of their data to be able to complete that project. And so it's a way for, for the images to, to live locally uh to avoid some of the sort of legal and, and and otherwise is sort of um uh issues behind some of this kind of collaborative work. Um Well, what about registries overall? So why, what are the challenges with registries? Well, the National Quality Registry Network surveyed 100 and 52 societies or associations and you, the response rate was about 52% and about a third spent up to $10 million per year. Um On, on registries, it's just an enormous amount. The average registry had three full time employees, 88% Uh use manual data entry and only 18% were actually linked to external sources and mostly were used for QI benchmarking clinical decision support. But, but there's a lot of um interest ultimately, you can see in this planning to use down the road. A lot of them did ultimately want to look at patient reported outcomes and outcomes in general. But cost interoperability and vendor management were really some of the main barriers to continued development. And so I like to show this, this is now the been updated. Um So this 327 should be 420 thou 1000. But we've, we've continued to grow the Image Guide Echo Registry. The uh to which I, I share the uh the committee for the AC on the Image Guide Echo Registry. And, and we, we're now expanding, uh We have about 70 physicians submitting data. Um I should say seven participating institutions, eight pending enrollments. We really have a number of sites that have been interested in, in joining us across the country. And as is usual with, with sort of registries, it's a slow start and then it really takes off. And so we're really re reaching that miss that critical point where we're really getting a lot of studies into it. And you can see just a, a sense of um where we stand here, this is um Houston Methodist and V I have already rolled as, as well as a number of other sites. Uh But there are a number of large sites including Universe of Kentucky, uh Universe of Pennsylvania, University of Washington. We health records, Dartmouth and Sony Brook that are right now in the process of, of, of contract review. And we anticipate that with the future annual volume that we have about 600,000 already print or attic echoes. And and anticipated we would, we would estimate that we have about 305,000 echocardiograms entering the data set uh annually. And that make means for very quickly, we become the largest data set in the world. Um The echo data set in the world, uh We're also developing a pediatric module. So sort of look at uh pediatric congenital heart diseases. Um And, and Alina NACOA is uh from Duke is working on a, a Perative module um for that predominantly captures T E E. Um We already have a system for DICOM up upload and export uh configured. And we are starting in the process of uh developing an anonyms platform with the aid of uh uh David Yang from Cedar Sinai um to uh to develop a, a system. And really the issues with anonymity are not the the meta data that can be easily sort of scrubbed from the data, but it's actually these burnt in images that you know where it says Salva or um um et cetera. Those kinds of words can are very hard to remove from the actual film itself because they become part of this. Um We're working with the IC. So we have a partnership ongoing with the uh IC where we hope that ultimately, we're gonna be able to create a um a, a designation of a continuous uh a lab that's a con interested in continuous quality improvement. And uh and if that becomes the case, then we're hoping that we will extend the timeline for accreditation uh potentially up to every seven years for those who are participating, we can also use it to use sort of um uh disease specific research. So this is uh an ongoing study. People may may know about ways one and two um ways one and two looked at normative values in echo uh as well as normative values in, in, in in COVID phase three is planned as a uh disease specific study and and it can be done through the registry. So in this case, we have about 600 some odd um patients across the globe um with amyloidosis. Um The clinical data are all collected from that, the images have been deidentified. And in the image that echo register can act as a clearing house to be able to, to take all those images. And and ultimately, we can, we can use that to develop, to keep learning mo uh models to identify amy doses to really understand it and understand unique features of um of it as well as to, to really do substu uh on, on this. Um So, what I've ultimately talked to you about is, is really sort of a combination of, of adverts that really is a, is a new field, so to speak, it's a outcomes research within cardiac imaging is really a multidisciplinary field that seeks to establish or evaluate the relationship between cardiac structure and function to health outcomes, to evaluate the use of imaging, to, to guide medical decision making and prognostication, as well as to understand the use cost and sources of variation of cardiac imaging in practice and the optimal imaging for and cost effectiveness for diagnostic uh strategies related to imaging and ultimately hopefully down the road conduct trials for diagnostic imaging strategies. And I can tell you a couple of those are, are are on the in the pipeline. Um And, and it really the methods that we need uh need are are unique in that. It's a combination of things, both epi and bios stas methods but also cost effectiveness, data science techniques like data scrubbing, machine learning and database management. Um that that really uh are needed to be able to combine a number of different data sources um and analyze them. So uh let's talk, we've talked a lot, let's talk about a case. Um and, and bring this home. So it's a 75 year old male with COVID-19 with hypoxic respiratory failure. This is something that's uh I think very um common to all of our experiences recently um intubated in the IC U unable to give a history, a history of multiple cardiac risk factors and was found to have a elevated cardiac component. And this is the image that you get. And this is not uncommon to, to see a uh an echo image where uh this is what my uh former boss likes to call nocardia. Um You can't really see much of anything here and of course, you can see that burn an image that says intimated. Uh But then we give, we give a, a contrast agent. In this case, we use loom the sun. But in any case, you could use regardless of the agent. And you can see that there's an inferior uh inferior hypokinesis. And then for motion, everybody and this person had an M M I and that changed their uh management. And so um uh ultras enhancing agents or contrast echo are, are safe and effective for less than psychopath vacation. Um Some data from Mike Maine um uh looking at 4.3 million individuals undergoing inpatient echo in the national inpatient sample. 2002 to 2017, only 1.4% receive uh ultrasound contrast and controlling for case mix, they were actually less likely to die uh one day. Now, I don't think that that's a necessary causal relationship, but it certainly suggests that they're not at increased risk. Um they uh are effective. So, um this is a study by Kurt at all from um Houston Methodist. Looked at 632 individuals receiving contrast for L V opacification, uninterpretable rate of an uninterpretable studies dropped to 6320.3% and technically difficult studies dropped from 87 to 10%. And ultimately, that resulted in a change in management um from you use and about 30 and, and about a third um and about two thirds of all those uh in the in the SI U. So those sort of higher acuity patients seem to, to benefit more. Um they're cost effective. So if you think about that, the downstream change in management from that current study that ultimately resulted in the net savings about 100 and $22 per patient uh or 76 or 77,000 overall. Um There's been some other um evaluations of this. So if you uh use a MOOC cardio contrast guided approach towards um uh patients with E D or patients with chest pain in the E D that um prevented about 207 admissions for chest pain and, and, and decreased the observation saved for about 316 others. With the net saving is about $900 per patient. And then Costello uh looked at this in university hospitals and, and, and a sonographer driven contrast protocol and, and of course, it worked as expected, but within every 10 minutes of time saved the net benefit of of contrast uh increased dramatically more. And so, what's the problem? Well, we've looked at this as well. Um And they're really underutilized relative to, to need, we, we estimate from now from national data that about 8% of institutions use contrast for trans echo. Uh but there's a vast uh variability and about 22% for stress echo, it's used a lot more for, for stresses, but regardless. So we looked at our own data. This is uh from this encore data set across 18 years. And you can see in the red here, the probability of sub optimal image quality which has been on the rise and the probability of contrast received was was also on the rise but significantly less than, than the rate of increase in sub optimal image quality. So overall, we had about 2.6% of echoes that received contrast, 17% were were suboptimal. We we'd estimate about 10 to 20% based on studies of of images are suboptimal. So that's right on par. Um And uh the rate of contrast uh increased less than that of the rate of sub optimal image quality, the use was proportionate greater amongst inpatient. And so why do we see these trends? So one is we have a growing understanding of safety. So if you look before and after 2008, which is when the black box warning was sort of lessened on on uh uh there was an inflection in in rate of use about threefold increase in in rate they're increasingly available uh agents. Um And there's a growing list of indications both for for contrast as well as for echo as a whole, sub optimal image quality itself is also increasing. And why is that the case? Well, if we look at the overall um weight of the patients that we see, we can see that the patients um uh mean weight has increased dramatically over time. We're also using echo in a broader population and we have a change in perception of sub image quality. And so we've published this uh as well. Um But four overriding practices heavily influenced the use of contrast. One is the the presence of the standing order for Contrast administration. Two is sonographer enabled to make the decision to administer contrast at the point of care. Three is, is a big one which is sonographer trained and enabled to place I V and only about 18% of uh sonographer get any kind of training in I V placement and as part of their schooling. So most do not. And, and um whereas other professions, other technologists uh do have a lot of training in in I V placement. Um sonograph do not. And then the presence of a physician sonographer advocate and, and so there's um we're now up to about a 22% FBI. Um And we've, we've done it um through enabling our sonographer to give contrast at the point of care. Um But, but we, we also have a workflow solution that we've developed from this. And so this is from the same paper looking at um dividing our, our data set into a a derivation validation set. We used sort of three variables that were only available prior to the image acquisition that we use to create a prediction model. And, and it worked pretty well and we use the prediction model, not just to predict contrast use, but also the potential need for contrast in in uh that is suboptimal image quality amongst outpatients. So whether or not they got the contrast or whether they potentially needed the contrast, but didn't get it. We, we, we, we were looking for both of those and, and these were the really the three variables that seemed to matter the most, I think, age and age make a lot of sense to everybody. Heart rate was a little bit of a surprise. But we think it's maybe sort of a proxy for acuity um uh the community of the patients. But these three variables you can contact, you can determine ahead of time and, and essentially this can give you a strategy. So if, if, if you have an ma or, or have somebody who's, who's not necessarily medically skilled, um they can go uh and calculate this for all your patients ahead of uh schedule and then can have them arrive a little early in the lab to get their I V S placed. Um so that you can make the decision to give contrast, not based on a score, but based on the actual images themselves and not based on waiting for 45 minutes for an I V nurse to come and place an I V for this patient. Um And so, uh we've externally validated this algorithm um at Mid America Heart Institute, which is um where um Michael Main is. And they use contrast in about 23% of all trans echoes. And they performed very similarly to similarly um to the, the chat two vasco to predict contrast use alone, they didn't actually have information that the image quality. And so even just contrast alone, it pre did pretty well. Um And now we've um we've now credentialed sonograph at FBI to administer Contrast. Um We have the IC Connect app which you can download for free I I S is the International Contrast Ultrasound Society. And um so this is the app that we've created that has lots of good, great content on this. And there's now an online calculator available. So you can go to their website IC society dot org, you can go and see this calculator and um and this will uh or you can screenshot this. Um and all you need to do is plug in the weight, the age and the heart rate and it will spit out the probability of contrast use. And it says if it's probability is greater than 15% you should consider the use of an I V uh to support contrast and, and we uh are in the process of, of studying this perspective. Now, um to see if this changes patient management, uh if, if uh it changes the use of I V S, the use in contrast and patient confidence. Um So the next case is really somebody with a 72 year old with hypertension hyperlipidemia diabetes and moderate A s who presents with dym on exertion for a year and, and the echo shows no significant progression of, of their A S but the E F is low and the corne angiogram, it shows no obstructive coronary disease. And they have a limiting dym on a, on a stress test with, uh with really no significant changes on EKG or echo otherwise. And they do a cardio pulmonary exercise test and that ultimately shows a cardiac mutation and, and it really gets at the question, this is supposed to be a somebody with, you know, potentially symptomatic moderate A s and really, what's the attributable risk to that moderate ask? And should we be doing anything about this? Um And I, and I bring it back to this graph, uh which was really sort of incredible for the times in 1968 Ross and brawd did, of course, this, um this study, this was done with, with the sort of assessment of um The severity of uh aosis and, and really for its time, it was really one of the first studies to really look at um a assessment of severity. And of course, everybody knows there's this long waiting period and then you have uh um death upon onset symptoms within two uh 2 to 5 years of, of the onset of symptoms. Well, what I draw attention to is the average age of death here, which is around 63. And you can see that's a really relatively young cohort for those with uh A s and that's because this was AAA mix of bicuspid rheumatic as well as the calcific A s and it was really a small group of individuals and it was incredible for the, but is it, is it the full story, I guess, is the question at the end of the day? And so really get that, what's the risk of, of, of other forms or less severe forms of moderate ads? Now, I don't think anybody really feels that there would be a benefit to, for example, doing an an aortic valve replacement in, in individuals with mild A S because the um the uh gradients you get after uh A V R are very similar to that of mild A ad. So you're sort of trading uh mild A S or mild A S but what about moderate A S and, and so 25 studies, uh 12.1000 moderate A S patients, this was published in Jack Inter uh interventions recently. And you can see that the um what, what we see on here is is that there's a gradation in, in increased risk uh uh depending on the severity of the A S and, and it, it really runs the spectrum um from one uh perspective to the other that, that it seems to be that the moderate A S uh individuals are, are at higher risk and, and not necessarily the same as severe. But um but, but that it's a gradation and, and risk, we look at the um economic impact. This is something we published in the lancet healthy longevity uh recently. But this is looking at the economic impact of different forms of A s and of course, as you get outside the severe spectrum, there's um quite a bit more individuals. And so the impact of doing an intervention on a potential intervention on those individuals is is potentially greater. Um the overall cost uh were about $400 million for men and 467 million in, in women. So really a huge uh cost to our society. And so, you know, we, we don't really the best way to study this is ultimately a trial and I'll get to that in a, in a, in a bit. But, but we, we realized that in order to, to do the trials, we really need to understand what is the baseline rate of outcomes that we should expect from, from the trials. And so we look, we ran a large uh prospective or sorry, a large, large retrospective cohort study and this was uh a, a combination study between us and, and A um and N A being the site that I mentioned earlier, they have their large multi center academic community sites. Uh They, they have cause of death information, uh cardiovascular or non cardiovascular death and predominantly outpatients where what they don't have is really the granular phenotyping of those individuals and understanding if they have coronary disease, which we of course thought or, or, or for example, cancer or dementia, which may be driving some risk in this particular setting. Uh But we have predominantly inpatient echoes and um set up a single center. And so linking our, our data sets together, we, we sort of realized that we, we potentially had AAA really sort of good opportunity to be able to study this. And so we've looked at this and we published on this um uh both uh in, in Jace recently as well as in uh plus one. And we looked at this as uh about 30,000 us cases and 217,000 Australian cases and across the board adjusted for age sex echo characteristics in over 30 clinical lab and treatment variables including uh coronary disease, revascularization, dementia, and cancer. And you can see here that these are the adjusted hazard rats and really that uh across the increments of peak aval velocity, there seems to be a fairly linear association. So it's not this cliff that people sort of drop off of, but rather that there's an increased risk uh as and now the question is, is this associated with the underlying valve? And that's, I think the, the, the interesting part of this, but, but there does seem to be uh a gradation and increased risk. And this was consistent if we looked at the first versus last echo, we adjusted for the time in, in, in each stage. If we looked at cardiovascular specific death in the Australian Cohort, if we looked at those who were under age 65 in the Australian cohort, um it turns out that if you have uh any heart failure or coronary heart disease, that any amount of aosis was, was prognostically significant and severity didn't seem to matter much. Um So if you had existing heart failure or coronary disease, it doesn't matter even mild a acid was, was significantly associated with risk. Um but um and this is sort of taking this apart and these are some of the things that we adjusted for. You can see that um the the aortic sonos is on par with, with that uh the risk of that of, of dementia or cancer. And so, you know, this is a significant increased risk, but the question is really um uh the benefit and the risk benefit ratio. So risk is, is distinguished from benefit. Now, that being said, should we readdress this risk benefit ratio? And, and this is an article that was um put together by Bron Wald um in circulation. And a couple of years ago looked at historical operative mortality for a AD R at the time when these recommendations were made was about 15%. And now when a partner um uh low risk trials, the current upper mortality for the lowest have population is around 1%. So it's really dropped dramatically. And does that risk benefit? Uh now favor early intervention and, and so we've, we, we published on this beforehand as well. Um, with that, um, this is not, um, an area that we should jump right into and we think we have to be cautious that risk is, is not the same thing and can't be completed with benefit of procedure. You can have risk for, for the alternative treatment as well. Um, and so it's, um, uh, a chicken or the egg phenomenon. Are we really seeing um a lot of people potentially having sort of a adverse outcomes on, on account of the their moderate aosis or is this AAA bystander? And, and when I get to the scene, I think about uh M R M R as a disease, we've increasingly now recognized that there are sub clinical changes before the onset of symptoms that drive outcomes in micro regurgitation even then it's, it's somewhat controversial to um to, to, to do um a procedure on somebody with severe M R and, and no symptoms whatsoever. But, but regardless, you know, the, the question is, are there subclinical changes? Um and, and, and we, we know there's an increasing fibrosis and uh dial function that occurs with increasingly severe eros. Uh are we simply by waiting too long, are we replacing a disease of the valve with a disease of the ventricle and, and ending up with, with diastatic dysfunction? And, and potentially there may be benefit from upfront um intervention versus this is just an epi phenomenon and a bystander and people with moderate A s uh have simply have increased risk from, from other means. And um and I think that that the jury is a little out on that. And so, and I like to show this, this is um a correlation is not equal causation. So this is the letters and the winning word for the scripts National spelling bee and the number of people who are killed by spiders. And this, this is Tyler given is a great website. They have all these kind of correlations like this and really fun to look at. But you can see it's a very high, highly correlated but that of course, is correlation is not equals causation and does not equal benefit from a valve replacement. So we have to be really careful about how we address this, that being said the right tests are on the way. So uh the progress trial is is in uh in progress, so to speak. And so 22 main trials looking at moderate A s right now, the Taber un mode trial um which is looking at SAPIEN three uh trans heart and in those with, with heart fail and mo A um and there is also as well as the progress trial looking at um intervention for those with moderate and some kind of uh biomarkers of injury of the of the heart. Um And then there's also a number of different studies ongoing. Uh for the, the uh asymptomatic severe aosis study, we've already had two trials published with, with really positive results suggesting a benefit from early intervention in asymptomatic severe individuals. Um, early tar evolved avatar estimate and recover trial. Um Well, there's another piece of this, which is that even when we have severe symptomatic A I, there's really um a number of people who don't get interventions and, and if you look at the heart survey about a third, didn't get intervention. Um and that's consistent across the board depending how you look at it. Um In the European Heart Journal, about a third of severe symptomatic A S not referred for a valve replacement. Uh Ben Fried published an article in, in uh American Journal uh cardiology. About 31% were with severe A S were referred and, and they, they, they looked at the reasons for non referral and it was really interesting that the top reason is that the symptoms were not attributable to, to the A S. And so, um you know, we have an issue with identifying, you know, that the symptoms truly are related to A S and, and, and, and importantly, rec recommending um uh treatment. Um 77% of, of those with a guideline, uh only 77% got guideline recommended transthoracic echo intervals and about one in five patients overall have a three greater and three months delay from, from uh recognition to treatment. Uh 80% of those had had symptoms and, and 52% had severe symptoms. And what's the problem? And this is um this is the M G H experience, which is those of you who uh may have seen, this paper was recently published in within uh with Jack. And this is looking to say that it's about a third of patients don't get intervened upon. And um despite an increasing trend uh in a replacement at a, you know, a very, very good hospital and, and uh the, the low flow low gradient A S uh population is the most commonly non referred population, about two thirds are not preferred for A V R in that case. Um What, what's the problem? Well, there's, there's risk in waiting. Um there's a mortality on the waiting list is about 3.7% um at one month and about 12% at six months higher for um and median wait times about uh 6, 60 days and mortality being two point 2.5% over that time interval. And with a, with an estimate that, you know, the relative increase in one year mortality increased by about 2% per week after a referral for ta for ta after the, the the symptoms were recognized. And so, uh you know, we're talking about an enormous problem and we really need to get people into care faster and better. And so we've, we've developed a uh a system uh to recognize and identify high risk A S. Um And so this is using echo report data. Um um And this is uh hoping to become a commercially sort of available. Um We're putting it forth in front of the FDA to, to evaluate this as a technology, but this uses echo report data and um there's no need for any B O T diameter, but you would just feed in uh AAA uh the variables um and it will identify whether or not you have severe uh disease by guidelines, uh severe, not meaning guidelines, um moderate A A S or, or low risk A S. And it's really very, very good at identifying severe. As you can see. The R O C curve here on the right hand corner, the A O C area under the curve is about 00.99 for severe aosis. And it really does track with incident uh A valve replacement as well as mortality and five year mortality. And so we're hoping that this tool can be useful to identify those, both those who have severe A S but also those who have less severe A s who might be still behaving like they have severe a ass that have sort of increased risk moderate because we don't know if that whole moderate is, is the entire group of them or if there's moderate, low risk, moderates and high risk moderates and, and who those high risk moderates are. Um, but, you know, I think anybody who's read echoes, there are cases where you're not entirely sure whether or not something is moderate or severe or they're right on the cost and you call it moderate. But recognizing that it could be, um, it could be severe very soon. Um, and those are high risk patients and so identifying those, those individuals and, and, and triaging them appropriately is, is important. Um, and then there's, is there something else that we can do um from the perspective of preventing the A S um And this is a study that we, we've submitted. Uh that's a collaboration between ERIC, the chronic renal and sufficiency cohort study, which is a, like the Framingham of, of C K D. Um And uh and the N H O B I and, and this is essentially looking at incident um aortic stenosis and, and trying to understand what it is about renal disease that preempts uh and um causes such rapid progression of aortic stenosis and a calcification. There's been lots of work done and this, of course, renal disease has altered mineral metabolism, altered libid metabolism and altered human dynamics. Um And, and ultimately, those all contribute to to that of incident avios but really hasn't been looked at from the perspective of does it uh relate people look at aortic valve cal that's, that's very different than a aortic stenosis. And what really we're interested in is, is looking at the progression of a of A S over time and saying what factors really are conserved across renal, different levels of renal disease. And what factors are unique to renal disease and different levels of renal disease that really predisposed to something developing faster progression of, of their aosis. With the idea that potentially we identifying markers of uh of of rapid calcification, then we can treat those upfront not only in patients with renal disease, but potentially in people without renal disease as well to prevent or slow progression of A S. And so there's a number of benefits from, for using this cohort. Um the quick cohort, it's a large number of serial echoes. They did an echo at uh 2725 and seven years out out after um enrollment, um the anos incidence is greater and a lot of the population cohorts, the er incidence is very small but in the uh C K D, it's much a lot higher. There's no referral bias uh getting to that issue earlier because all the echoes were done for uh for research purposes, there's diversity of demographics and C K D stage and really granular phenotyping of blood and urine as well as adjudicated echo data and outcomes really, really high quality modern echo um information. And so uh at least it will leave you with one sort of provocative thought, which is these are the prior studies um looking at the progression of C K D or A S within C K D. And it's estimated that it's, it's relative to double that of, of a normal progression. But, and, but you can see that most of these are single center dialysis patients really retrospective or, or um small studies. And based on Cato's work, uh previously published rates of progression are are so about 0.1 centimeters per year for the A R valve area and then increase it in peak velocity about 0.3 m per second per year. If you just look at the preliminary data that we have from the quick participants, uh we don't have a um yeah, that's what we're working on uh reviewing the echoes and, and ultimately sort of uh identifying. But if you look at just uh those who have serial trans echoes, we do have peak velocity and the, the mean change is about 0.4 m per second. So it's, it's higher than that was, was observed in the general population, but varied widely um up to 0.9 m per second per year. And if you look at a uh where there was a serial echo done at um at two different time points already, the mean change in area area was substantially less than what's been reported in the, in the literature, but so far about point oh four centimeters per second per year, but varied quite a bit um to up to a uh a decrease of about 40.4 centimeters per second per year. So there's, there's quite a bit of variability in these measures and it just using a, a single point estimate, it's not going to give us that information. And, and I think it is intriguing, there was an article by Roberto Lang and, and, and uh European Heart Journal Cardiovascular Imaging that looked at their own data set in um University of Chicago and found similar results in terms of the progression of the valve area. Um And so I think it's, it's intriguing, but we we we can't co too much on sort of the historical data that we have and we really need to look at this a new. So um in in summary, rising costs of cardiac imaging have forced our hands to justify the value of imaging imaging registries, particularly when the linked to outcomes could be used to demonstrate value. But we uh we need to uh to link the those data sets together and to better understand the the risk stratification and prognostication for patients, insights into diseases as well as understanding gaps in variation in care and outcomes research. And imaging is really to your own unique discipline with the distinct set of methodologies, challenges and questions. And A s is a really important use case to use imaging data to better inform clinical management and mechanistic insights. And for that, I like to thank you for having me
