High Cholesterol and Plaque Regression?! Updates on KETO-CTA with Dave Feldman

Hey, this doesn't make sense to me.

This is the hardest I've ever had to work at not running my mouth.

And it's his part that we're sharing officially now, but I wasn't sharing before.

What is a lean mass hyper-responder? And what is this keto CTA study that has been the center of drama?

There was no association between this plaque progression and LDL cholesterol or ApoB.

You can visibly see the regression without a shadow of a doubt. What predicts plaque progression is is not LDL, and baseline plaque predicts plaque progression.

And that's kind of the part of the story that I've not been able to share until now.

All right, everybody, I'm here with another interview. I hope this person needs no introduction. He's one of my dearest friends and closest colleagues, Dave Feldman. Um, you should be forewarned, we have the potential to go on forever. We have indeed sat for a 6-hour podcast with no pee break before, and I'm going to really try hard to make sure that doesn't happen in this one. We want to keep things tight. Really focused on the new keto-CTA data that Dave has just released. And believe it or not, I've actually been in the dark about a lot of this despite being associated with this research. So Dave, thanks for joining me for this particularly exciting interview. Um, let's start just a high level for people who aren't aware of lean mass hyperresponders and what keto-CTA is. Can you briefly tell us What is a lean mass hyperresponder? Why is it important? And what is this keto CTA study that has been the center of drama in my academic life for the past several months?

Yeah, so first of all, a lean mass hyperresponder I defined in 2017 with the cut points of high LDL, high HDL, low triglycerides. I won't go into those cut points. Rather, I'd like to focus real quick on the cut points for our study, which is a slightly more relaxed version of that for the Keto-CTA study. So the eligibility was an LDL cholesterol of 190 mg/dL or higher, HDL of 60 or higher, and triglycerides of 80 or lower. So that triad is something we tend to find in people who observationally are more likely to be lean and also typically tend to be a bit more fit. And what we've been wanting to do since— I mean, literally since I wrote that article in 2017— is to specifically get them into a study like Keto-CTA, where what we're doing is we're recruiting a number of people, which this has now already transpired. 100 have been recruited. They were flown to the Lundquist Institute, which is out of UCLA, to get blood work and also what's known as a CT angiogram. It's a high-resolution heart scan that looks at both soft and calcified plaque, and they would get one scan at baseline and then another 1 year later for each of the participants. And so with 100 scans for a set of both, we're able to not only look individually and look at how the data fits there, but also to compare the sets to each other to see at a population level of what they look like and what the different subsets of the population look like as far as that, those cut points, and especially how it associates with plaque development.

Forgive me, my audience, for editorializing, and I'm going to suck up a little bit of air rather than doing what I should do and giving it to my guest. But I just, for those of you who don't know how important this research is, like, this is actually, I have a personal connection, which I think a lot of you will know, which is that, you know, when I went low carb for ulcerative colitis, which really, you know, ultimately saved my life, at least that's my narrative, I saw this the triad Dave just described. So the high LDL, the high HDL, and the low triglycerides. And no one could tell me what was happening. No doctor I went to knew what was going on. And that is typical of the state of knowledge around this phenomenon where people go low carb and their LDL goes through the roof. And just to be clear, the cut points Dave mentioned are like basement level cut points. They're astronomically high at a population level, but they're still basement for the phenotype. So My LDL runs in the 500s, and, and that is scary. And the reason this is important for people who, you know, not just lean mass hyperresponders, but the entire population, is I feel very strongly, and I would debate anybody on this, that ketosis and ketogenic diets have a broad suite of potential clinical applications, including in people without obesity and diabetes, for mental health disorders and neurological disorders, autoimmune disorders. But the implementation of keto clinically is hindered by what is perceived as kind of like a, a cholesterol jack-in-the-box phenomenon where some people have their LDL go through the roof. And so in order to better serve the general public who can use this metabolic tool, we need to understand what's going on and what's the risk. And what Dave stumbled upon through his own journey, and then I stumbled upon, and we kind of had this convergence, is this bizarre pattern where it's the lean insulin-sensitive people who have their LDL go through the roof. And Dave, I'm telling his story for him now, but I do know it pretty well as I've gotten to know him. Like, he got intensely curious, and because this question wasn't being answered, he went about crowdsourcing, which isn't done, a clinical trial. It isn't done in general. It's definitely not done by an outsider. The gauntlet was dropped at his feet. He picked it up, crowdsourced a study where he then took this population of people and organized to have them flown out to a lipid center to have scans done on their hearts, state-of-the-art coronary CT angiography scans prospectively a year apart to see how much plaque progresses. Is it what we'd expect? Sorry for all that, Dave. Um, I just, I needed to—

I like the expanded answer. It's, uh, and, and it's even more fun when you get a chance to have someone else kind of expand it out.

Yeah.

In that way. So, all right. That's good.

Well, we're gonna build up to the controversial part, but we did publish a paper in 2024 that was well received. So it was on this population of lean mass hyper-responder and lean mass hyper-responder-like people from this keto CTA cohort compared to a matched control. Can you tell us what the main findings were of that initial paper setting the stage for the second paper, which we wanna talk about.

Yeah, so certainly we should emphasize this is with the cohort. We had 100 people, as I just mentioned, 80 of those that were comparable to the age range of this other study outside of our study called Miami Heart with principal investigator Karam Nasir. And that we were able from a population they had of 2,400 there was able to be extracted a matched control. So our 80 versus another 80 in Miami Heart, they were matched for, uh, ethnicity, age, sex, and risk factors very closely. And so the neat thing was, is basically we got all of these matches on both sides, except that our average for those 80 for their LDL cholesterol was 272 milligrams per deciliter, for the Miami Heart Group, it was 123, so less than half the LDL. But for our baseline scans, this is the one from last year, we only just had the baseline completed by that point. Uh, the baseline scans, the plaque levels, in spite of there being an average of 4.7 years on the diet for those 80, their plaque levels were, were very comparable, were statistic— that there was no statistically significant difference to that of the Miami Heart Group. In fact, it trended to slightly better than the Miami Heart Group. But bottom line is there was no statistically significant difference between both groups in spite of the Miami Heart Group for almost half a decade having less than half of the LDL. So that was very exciting for us and we were very interested to see, you know, where the research would come next once we had the longitudinal data, right?

So just to reiterate, when you took this population, the keto CTA population, and looked at them at baseline compared to a matched population, so similar in age and other variables, matched as best you could, the plaque level was trending lower, not statistically significant, but actually trending lower despite much higher LDL and much higher ApoB levels. Ironically, they had the populations very similar, uh, pre-keto baseline. So before these participants started keto, we we had some LDL statistics which are basically the same as Miami Heart. I think it was like 122 versus 123 LDL. So that kind of set the stage and expectation. But then the prospective portion of the study dropped. So again, these participants were going to be followed individually, comparing the person to their own baseline, their baseline being 4.7 years or 4.6 years after they started keto. And then that paper, which has been the center of controversy since April 7th, 2025, when it was published in JACC Advances, had some interesting findings. Tell us about the findings, Dave.

Well, it was in the title, which, by the way, has since— there's since been a correction with the title under JACC Advances. But basically the two major findings are Plaque Begets Plaque, which has been retitled to Plaque Predicts Plaque. It's a little bit more accurate to put it that way. The, uh, and ApoB does not. So bottom line was if our participants had plaque at baseline, it was more likely that they would have future plaque progression and is more likely that it would be a higher amount of future plaque progression. And this is all kinds of plaque, whether non-calcified plaque, whether calcified, and so forth. And the second major, uh, finding of interest was that there was no association between this plaque progression in LDL cholesterol or ApoB, that the lipid levels didn't seem to have that association you might expect would be a driver of plaque progression.

So has that changed? So the emphasis of the initial paper was that, um, whereas one might expect that LDL levels and ApoB levels predict plaque progression, and that was the emphasis as I understand it, having also written the paper, of that initial paper, has that changed? Because we're going to talk about what might have changed, but has that main emphasis changed? Has new data come to light that refutes that in any way, shape, or form?

No, you're kind of jumping a little ahead in that we haven't yet introduced the audience to— there's now two new data sets including the pre-specified methodology finally coming in, which is the Q-ANGIO, and that's for the final paper that gets matched with the pre-specified endpoint of percent change in non-calcified plaque volume. That we just did the preliminary, um, I got to, to announce the preliminary data, literally just fresh out of Lundqvist from last week. And it just now posted as a video literally yesterday. So that's just now landed.

Okay. Yeah, you're right. I was jumping the gun a little bit. So let me back up and say on this April 7th, 2025 paper, what was the nucleus of controversy?

Well, there was— I want to say that there were two things. One is that the Figure 1A that we had did not have the numeric— it wasn't stated plainly that the percent change in non-calcified plaque volume was in Figure 1A, which for this analysis I think was a fair critique by many of our critics that that should have been something that was in there. Granted, many people took it to the place of that, like there's just no mention at all of non-calcified plaque volume. In our paper, which that's plainly not true. If you read our paper, we discuss it all over the place. It's just I do think that that was something that, you know, to the extent to where we could quickly correct it, we sought to do so. And I actually discussed this ad nauseam in a couple of different interviews, so we won't spend too much time discussing it now. But that was an area for which, you know, we should have been able to anticipate our critics would be a lot more interested in any kind of misstep or mistake was something worth highlighting. And again, I appreciate the fact that we owned it, we both owned it, and then pushed for the journal to make this correction as soon as possible. And so we'd gotten that draft together as fast as we could. But that said, you know, any, any way in which something could be looked at in a negative light is something that you need to anticipate. The other one is that there's the CLEARLY dataset is a little there's, there's a separation between it and the semi-quantitative dataset. So we had these two different analyses, and while they agreed on the two major findings that we were reporting on, there was a bit more cohort-wide plaque on the CLEARLY side of the fence. And that became another focus of interest, especially for a lot of people who are bigger critics of our work. I would say.

So the CLEARLY analysis showed rapid plaque progression in non-calcified plaque volume. Is that a fair thing to say?

I think that that's what they would say. I think that if our third colleague, and who's also a co-first author with you, Adrian Sotomayor, were here, he would point out that it's a very heterogeneous group. Um, I don't want to put words in his mouth. It's literally what he says, he— that the majority, did not have what would be considered the current definition, a percent atheroma volume of 1.0 or higher. Uh, but that said, I do think that there's a lot of interest in whether or not you could make the cohort-wide assessment. And it is part of why I became a bit more interested, uh, when we were first chatting about this a year ago, I did express some concern. I did feel like I, I couldn't say a lot more than that because I didn't actually have access to the raw data. But I did express some concern myself that there did seem to be this difference between the semi-quantitative and the quantitative, which you normally don't see in other imaging studies. And so I was looking forward to the point where I would get access to the raw data myself, but that could only happen once the April 7th paper was published.

So you're anticipating my big question, um, which was basically the first paper we published. We always knew there'd be other papers, and we placed emphasis on what we thought was the novel finding— that ApoB predicts plaque progression. It's a heterogeneous group. Most people don't meet the threshold for being called rapid progressors, which some people might have anticipated, um, but baseline plaque does predict future plaque, which is very consistent with the existing literature. I just want to emphasize that. However, the primary pre-specified endpoint was non-calcified plaque volume.

And when you look at the percent change—

percent change. So either way, there wasn't an emphasis on how much the non-calcified plaque volume actually changed. And then, you know, more concerns have been raised about the CLEARLY analysis. So a question that I would have if I were in the audience, I'd have two questions. One, why didn't you quote— you being us— why didn't we do our due diligence to have this checked before publication? And then two, well, if you then noted something back in April Why didn't you go public about it? So Dave, you have been at the center of all of this, and I think those are the two most important questions to answer. So can you reveal to people, like, when did we have concerns? Why did we not manage this before the publication? And why have we been quiet until now?

Okay, so I think I, I think I want to separate out two things because I think one is when I'm discussing how we were chatting a year ago and I was expressing my concerns then. Is that, that's, that's what you're asking about in particular, right? Like the, if, if I'm— Yeah.

So, so you had some concerns, right? Why did that not hold up the paper until all the I's were dotted and T's were crossed? Yeah.

Oh, that's such an easy question to answer. And it's fascinating because I feel like people aren't playing this out themselves if they were pretending they're in my shoes. I'm the funder. I am the president of the Citizen Science Foundation. I'm the one that's driving this research forward, and understandably, I care a lot about acting in the most ethical manner possible, but also recognizing that there's some degree with which this is, you know, literally my first study, and I'm finding out what can and can't happen at every stage. Now, a year ago when we were first chatting about this, and it was me in my capacity as the funder discussing this with the research team, I'm blinded from the data. You were blinded from the data as well. So I get, like any funder in any study, I get checkpoints of where data is amassed, and then they're giving a preview of what it looks like. But it's not, it's not as though I can say, Hey, this doesn't make sense to me. Please keep reworking it. There's, there's the initial questions that I can of course ask that I can say, it doesn't make sense to me. Um, can you better explain to me why this makes sense to you? And there's a degree with which I can do that, but I can't keep reaching until it's basically they're unblinding the data back to me. Right. Recognizing that the increase was higher than what I would've expected given the previous longitudinal data that has been shown with the semi-quantitative data. That was something I wanted to understand better. Like, it's that, that does seem strange that it's this increase higher than the semi-quantitative in a way that doesn't make sense to me. And I was looking it up, other studies, and I was also bringing it back. And so I was probably a bit more, I guess, vocal at that time. But there was appropriately the team saying back, well, wait, what do you want to do next? You know, did like, where do we go from here? Are you wanting to take further action beyond just the questions? And again, put yourself in my shoes. If the one dataset that's coming back suggests that it's higher than what the original dataset is, never mind how high, just that that might be a reason with which I could question it. If I were to suddenly pause the study, or if I were to say, hey, hold, don't publish this data, of course I would be rightly criticized for it. And I think it actually is the wrong thing to do. I think, of course, we want to put out the data. Every time I'm getting a new dataset, there better be a very good reason why I would put the brakes on it. And when I'm not the imaging expert or the biostatistician, I'm fully recognizing that if those experts are the ones who seem to be in agreement with each other as to there not being likely a problem, or that at least even if it is weird, these two datasets agree on the two major findings that we had in the April 7th paper, then I can, I can agree to that because that makes sense. And besides, we should put that out. We should put out every dataset that comes our way as best as we can. Now, I said then, and I'm, I'm glad I had that presence of mind at the time. Okay, well, eventually when the April 7th paper is published as per the contract between the Citizen Science Foundation and Lundquist, I can take a look myself because whereas I was blinded before, I wouldn't be blinded now. And if there were things that I could find that would lead me to want to ask further questions, further follow-up questions. I could have a more informed, um, response to find out if indeed there were some issues. And that's kind of the part of the story that I have not been able to share until now, which is that indeed, after bringing what I was able to find to the attention of Lundquist and to Clearly, that did lead to further meetings. Now, I, I I think you're probably going to ask me a follow-up question to this. I think I'm just going to go ahead and state this now. What were the results of those follow-up meetings, those discussions? That part I can't discuss now because those are still ongoing. But I will say it's of enormous interest to the Citizen Science Foundation that we have the highest integrity of all the datasets that we're presenting for all of our papers.

Thank you for that. And I really want to reinforce the, the dynamics and the chronology here, because they're, they're not intuitive. A word you focused on is integrity. Now people can choose to believe me or not, but you are one of the highest integrity individuals I know. And the way this is all set up, like you see names on a paper, my name, Dave's name, Professor Budoff's name, you just kind of maybe assume, oh, we're all in on all the information. It's not how it works. So when Dave spent his blood, sweat, and tears to get this funded, he consented to be blinded to certain information because he knew, look, I'm the funder, I'm the money. Therefore, while we can, like, you know, once the data is collected and the data are the data, the output of the output, like we can help write the paper, we can be coauthors in the paper to have a check where we can't have our tentacles, so to speak, on the raw analysis that actually was established to support the integrity of the project. And then when results came out that Dave, as he explained, had questions about, and I had similar questions but was related to the study in a similar way, it was inappropriate. And I maintain, given the information we had at the time, it would have been inappropriate to really push back. So as our third colleague, Adrian Sotomayor, likes to say, you know, you just, you publish the data you have. This is always going to be part of an evolving science, and we always knew we'd get to look at the raw data, the raw scans, at some point in time. So given that there were some really interesting novel findings, which as I mentioned earlier have not been overturned, like, um, that ApoB and L-ZL do not predict plaque progression and that baseline plaque does, which was the emphasis of the first paper, we went ahead with the publication. Then what happened, Dave? What happened with CLEARLY? And Um, just to make sure to hammer the point, once it became clear that there were discrepancies, strong discrepancies, why has it taken between April and now August for us to be vocal? Because I'll just be frank, this is the hardest I've ever had to work at not running my mouth in my life.

Yeah.

So people are talking, I'm just like, uh. So, and Dave has me on a little leash.

Yeah.

Actually, just as a quick aside, we have a thing. Dave would just send like a little puppy emoji, which meant Nick, you're on a leash. Don't talk for a while. Now we're talking. Please explain what happened with Clearly and why haven't we been able to talk?

Yeah. And I, I want to say this in your defense, especially because we're, we're in this awkward position where naturally we want to be absolutely as transparent as we can about all stages of this at all times. But it's not as simple as you open the door to what's going on behind doors, behind closed doors, as far as negotiating to get into the direction that you want to get into with whoever's involved. I can understand why there might be some initial reluctance, again, coming from an engineer before more of the data were put forward and before it was presented. But when it was presented, and this I do have clearance to discuss, and this is a part that we're sharing officially now, but I wasn't sharing before, is that on April 24th, uh, fully 6 days after I'd had access to the raw data and I could finally get a meeting, I had a meeting with their then CMO, who's also a co-author on our paper, uh, uh, the CMO clearly he agreed with the presentation of the anomalies that I found, that there clearly needs to do a, um, a quality control pass, basically a fully blinded quality control pass to then confirm or disconfirm these findings. And this was before any new data had come in yet. And so I also got clearance in that meeting to discuss it in the interview I had that evening, which was with Darius Sharp. That was a very important moment for me because I felt really good about, okay, great. You know, if they're— if the CLEARLY analysis is going to get confirmed or disconfirmed, we'll know either way. However, following that meeting, again, this is the piece that I've not been cleared to discuss. There's many parties involved, including some major donors, so they want to maintain everything past that as confidential. But I'll just say that What you can certainly tell is that it's still Clearly's position that unfortunately since that meeting, they are not going to run the quality control, the fully blinded quality control pass, which did surprise us. Once it became pretty clear that that was a position they were not going to move from, that's when we decided to engage HeartFlow, which is a competitor to Clearly, although they actually objected to me even describing them as a competitor to Clearly, because I didn't actually know this. They have a much larger footprint than Clearly does. It's just as, as many of their colleagues would say, that Clearly gets more press than HeartFlow does. But HeartFlow actually is also an AI-guided quantitative analysis service, and we were able to get their agreement to do this, and we got a fully blinded process. It's actually cross-entity blinded on top of their agreement that they'll do quality control passes as needed. So if they came back with— basically it was one of those things where after what had happened with this last analysis, past that point, I was going to be like, okay, everybody needs to absolutely guarantee a quality control pass in all contracts moving forward, full stop. And HeartFlow, to our appreciation, had done that.

I'm going to ask some— what are going to sound like provocative questions. I'm going to do so based on what I understand is public information. And with the framework, I want people to know, again, different parties know different things. A lot of this is brand new information to me. Despite being an author on the paper, there are things that I am blinded to for particular reasons. So to clarify, and this is what I understand from your presentation at— in San Diego a few days ago, Clearly, the CMO of Clearly, who is a co-author on the paper— I actually didn't know he was the CMO when we published the paper. I'm not even sure if he disclosed that in the paper. But that aside, said he agrees with some of your concerns and they do a quality control pass on their scans. The company then reversed that position and said they would not do a quality control pass on the scans. Is that correct?

That's their current position, is the way that I would state that now. Like, it— we— I I'm still hopeful that they change their mind and, and determine with the continual data that's coming in that it's— it would be good to do— to go ahead and do a quality control pass on that dataset just to confirm or disconfirm the results.

And I'm going to ask this as well, because this is public information, but again, I have no idea what's going on behind the scenes. Dave might, I'm not sure.

Let me interject this real quick. Nick, you and I have worked together a long time now. We've worked together several years. I can attest this may be the only time that I'm aware of where I said, Nick, in my capacity as president of the Citizen Science Foundation, I apologize, but I'm not only going to need to conduct this entity to entity behind the scenes as my— in this capacity, but I'm actually going to have to ask you to do something I know is against your DNA. Which is to remain quiet about this until we get to all the way to the point where we have either the QAnon data out or until what is getting worked out with Clearly to be resolved versus not being resolved. And I just want to thank you publicly right now because I know how difficult that is for you. It's, it's certainly in our, in our blood to share and to be as transparent as we can, but there's, it's not just, you know, there's still the interest at the end of the day of just trying to get to the right answers. And this was a delicate situation. So yes, to defend you to everybody out there, this was an ask. You were asked to, in the strongest terms, it may be the most I've ever asked of you on a successive level, to hold tight, wait until all the data is in, do your best to get us to the other side of this thing. And I just want to thank you for accomplishing that. So yes, back to what you were saying, you were in the dark. You were in the dark. Adrian was in the dark while we were trying to work this out entity to entity.

All right. I'm going to ask one more probing question and then got into the questions I really want to ask. And I think people are most interested in about the HeartFlow and QAnon. But you said, and I'm, again, I have no idea what's going on behind the scenes. And you can say, I can't confirm or deny, but you said CMO of Clearly at the time. Is that person still the CMO of Clearly?

I believe that's not the case. I believe that the, the CMO at the time that we were having that meeting is no longer the CMO.

I brought that up because even since you released your presentation, you can imagine how many DMs I get. So people uncover things that is public knowledge. I'm not going to say more on that right now. I do want to get into the results of HeartFlow and the preliminary results of Q-Angio. I want to reinforce HeartFlow served as an additional lens on the scans and then Q-Angio was the pre-specified methodology. It's just more labor intensive. So my understanding is the HeartFlow scans are final, the Q-Angio scans are preliminary. I also want to say I'm going to echo you. The scans are the scans are the scans, correct? Please tell me what that means and why it's so important.

Because I think if you're new to the world of imaging, if you're hearing this, you know, for the first time, when I'm saying there's 4 datasets, you might think that there were technically 4 different sets of scans. That's not the case. These are all analyses. Of the same 200 scans, 2 scans per participant for 100 participants, a baseline scan and a follow-up scan. And this is one of the other challenges we were running into is at the time that there was a lot of activity and interest in, in the wake of the initial paper that came out on April 7th, we were very upfront about the fact that, yeah, we And I, I can see it myself. I thought the pre-specified endpoint was in Table 3, for example, and was happy that we corrected it in a much broader sense across the whole paper in the corrected version. But there was no— it wasn't the pre-specified methodology of Q-ANGIO. That's already in the protocol paper. That's already in the methods paper. But our critics were not necessarily as forthcoming with that aspect of it, at least if they had read the paper. I, I would assume that they've at least read the methods paper when they were quoting from it several times over. So there was a lot of discussion of the pre-specified endpoint, but not the pre-specified methodology. Now that's important because yes, this is, it's all the more reason why it was incumbent on us to just stay quiet further until we got to the pre-specified methodology completion, because I was starting to realize just how attached our critics were to the CLEARLY analysis. So why not wait until we had everything in to feel confident about it? Now, to, to expand on the point that you just made, technically all of these are preliminary until they are in a paper, at least until we can put them out in a preprint and then ultimately they're getting peer-reviewed. So the HeartFlow data, I would say, is still preliminary because it's not published yet. I would say that the Q-Angio is still preliminary. It's not published yet.

But I guess the point is the scans are the source of truth.

Yes. Sorry.

And so you have different lens on the source of truth. The question you'd want to ask is, all right, do they agree with each other? Because that gives you a sense that the picture is actually accurate to reality.

And that's so—

and that's how they clearly compare to HeartFlow compared to QNGI, right? And that's on progression of non-calcified plaque volume. Let's talk absolute numbers because that's most pertinent. Yeah. Can you give us some numbers?

So I did say this earlier. But as we're saying, again, the point that we came up with two different analyses that were different from each other enough that it sparked some interest, I think is a relevant piece to this and why it is that we wanted to add HeartFlow on top of not just having Q-Angio. And I'm glad we did, we did so that we would have four analyses as opposed to three with potentially two of them agreeing with each other. I wanted especially quantitative analyses for that reason.

Okay.

So with all of that said, do, do these analyses agree with each other? 3 of them do. Generally speaking, um, the semi-quantitative, which was the one we also were filming the movie around at the time, cause it was the first to come. Clearly was the next to come. HeartFlow was the next to come after that and could get done much sooner than Q-Angio because like Clearly, it's AI guided and algorithmic., and then finally Q-Angio came. Semi-Quantitative, HeartFlow, and Q-Angio all broadly agree with each other, and they agree with each other in key metrics that I was anticipating and looking at clearly with some concern. For example, and I discussed this in the presentation that got released yesterday, one of the first signs that had me concerned about the Cleerly analysis was that all of the metrics for plaque progression, for total plaque volume, they were all positive. There was only plaque progression. Now, if you had a disease population, that could potentially make sense if they all, all had a lot of plaque at baseline. But for those people especially who had like a CAC of 0 and very small amounts of plaque, just barely detectable amounts of plaque, I would've expected because it's below the noise floor that there would be a wobble that you would see both progression and regression in tiny amounts. Just because it's, it's standard error. It's something that you would expect once you get down to those levels because it's more and more difficult to measure. And is that what I saw when finally we had HeartFlow and Q-Angio come in? Yes. That was one example. You're about to say something. Go ahead.

I want to get on to the whole idea of regression that I see is like the next chapter. I do want to emphasize, like, at a population level, because people have been very focused on this non-calcified plaque volume progression. And now we've said, okay, there are 4 different lenses we have looking at the source of truth, which is the scans. There's the semi-quantitative, which is basically blind experts reading the scans, right? There's the HeartFlow and the Cleerly, which are both AI-guided. And then there's the Q-Angio, which was the pre-specified methodology and considered the gold standard, as I understand it. Correct?

It's, it's certainly the one for which there is the most data that backs it and the one that, uh, Dr. Budoff would consider the gold standard for sure.

And so what you're saying is 3 of them agree with each other, the semi-quantitative, the heart flow, and the Q-Angio, and one stands out, which is CLEARLY. Can you give us some numbers, say comparing on, on the quantitative end? So heart flow, Q-Angio, and CLEARLY what the non-calcified plaque volume numbers were for each scan. Sure, I know the answer, but yeah, the non-calcified plaque volume—

and I'm kind of going a little off memory because I don't have them accessible— but the non-calcified plaque volume in absolute terms, millimeters cubed, I believe that for CLEARLY it was 88.8. However, for HeartFlow and Q-Angio, one was 5.5 5 and one was 5.6, if my memory serves.

80, I think you meant 18.9. So the non, the noncalcified plaque volume for Clear was 18.9, correct?

Yes.

Yeah, I said 88.8.

Oh, sorry. 18. Yeah. 18 point.

Either way. So 18.9, one here versus 5.5 and 5.6. So clearly two are very closely in agreement and one stands out. And the semi-quantitative was standing with the heart Flow and the QAngeo.

So, and remember, remember this, do not lose sight of this. One of those, one of those metrics is also algorithmic AI-guided reading, an entirely different business, entirely different entity, and entirely blinded, right? So you're, we're talking 3 different analyses of talking just quantitative that are separate entities, separate methods. Right. And that becomes important when two of them are in such tight agreement with each other. Not even, and, and I'm not saying all these datasets are in perfect agreement with each other, but are they broadly agreeing? I think there's, there's a lot of, yeah, a lot of broad agreement. And those two, especially for non-calcified plaque volume absolute levels, I was, I was stunned that it was that close.

So now I want to get onto the question of regression. What, was weird about the, the presence or absence of regression in the CLEARLY analysis. And why did that make you suspicious?

Right. This is what I was saying from earlier. There was no regression at all for total plaque volume in the CLEARLY analysis. There was one regressor in non-calcified plaque volume. But in both those cases, it's unlike any dataset that exists in imaging outside of our study.. And that to me was something that immediately was hitting my radar once I was actually like into the data myself and started running some different analyses against other studies that were out there. And it's when I was, I was sure that we, we didn't want just Q-ANGIO. We, we, again, my preference would be if clearly just ran the quality control pass, but assuming that they don't, if we could get at least one more to see if indeed there was confirmed regression because the semi-quantitative had 6 out of the 100. And the semi-quantitative is by the expert-guided readers, which is a much more of a horseshoes and hand grenades. So I speculated then, and it seems to be that I've turned out to be right, at least with regard to HeartFlow and Q-Angio, that we'd see more than 6 once you could get to a more granular level. And indeed we do. On the Q-Angio side, I wanna say there were 12 identified regressors. And I want to say on the HeartFlow side, there was 32 or 33, something to that effect. And I want to be intellectually honest. The thing I was just talking about a little bit earlier, I want to state again, at the very, very, very, very, very low levels of plaque, it may in fact not be true regression. It might be part of statistical noise. By the same token, a certain amount of the progressors in that range might be part of statistical noise. And I want to link this back to why. The percent change in non-calcified plaque volume that was chosen 5 years ago when the study was getting kicked off and we didn't— we were a bit new to this process, why it was falling out of favor and what was coming into favor was percent atheroma volume, which is why you and Adrian were focusing on that a bit more in addition to the non-calcified plaque volume, because that's where the whole imaging industry has gone. Is percent atheroma volume doesn't suffer from what we've been dealing with in our study because so many people had low plaque volume levels at the start. It made percentages and especially pooling the percentages more problematic. And that shows up in my presentation because it is the one place where all three quantitative analyses wildly disagree with each other. Because the, the relative changes can have huge outsized effects and they're very inconsistent from one analysis to the next. But the absolute levels of plaque, which is what we're really ultimately the most interested in, that's where the consistency started falling into place.

So a few things to emphasize. One, and this is more a sidebar, but you were kind of alluding to the fact that maybe the, the, the pre-specified endpoint irrespective of methodology wasn't that well chosen. In retrospect, that's my position. I mean, I came on later after this was all decided. I'm not saying I would have decided differently, but in retrospect, maybe we could have chosen a better pre-specified endpoint. Do you agree?

Yes. And Adrian made the case in like 2 sentences what I just now stated. If you anticipate you're going to have a number of people with low levels of plaque volume at baseline, doing a pooled percent change. So for laypeople, which just emphasizes, let's say that you go and you get a scan and you have a half a millimeter cubed of plaque, and I go and get a scan and I have 100 millimeters cubed of plaque. You go back, you get a second scan, you have 2 millimeters cubed of plaque. Well, your percent change in non-calcified plaque volume could have increased 300%. I have another 20 millimeters cubed of plaque in my second scan. It's only increased 20%.

Right?

Which, which of us is in a more concerning position? And this was part of the problem is the reason I think it's been falling out of favor in imaging is especially if you get some mix of metabolically healthy people, they create an outsized effect potentially at small margins of difference. Yeah.

So it's like if you have a fraction, if the denominators vary hugely, so you have one denominator that is 10, the other one is 0.001. The change in the numerator is going to be like— it creates so much noise. But that aside, another thing you were mentioning was statistical noise in progression regression. And you did a great thought puzzle demonstration at your Low Carb San Diego talk where you had a bathroom scale and then you stood on the scale, you got off the scale, then you picked up— I think it was a quarter-pound weight.

No, it was my— I know it didn't show up in the presentation, but it was my AirPods Pro, which I knew was 5.3 grams.

Okay, so you objectively weigh more. And the proposition was, if I step on the scale 100 times, will the reading always be more than my initial weight? And the answer is obviously no, because that addition is small enough that there's a little bit of noise. You would expect to see regression. Exactly. And one of the interesting things that you were highlighting is in the CLEARLY analysis that just didn't show up. Now, playing devil's advocate, you know, okay, that's statistically unlikely. However, this is also an unprecedented group with super high LDL. So I could see the critics saying, Well, of course you had all like, you know, progressors. That just proves that the high LDL is dangerous. First of all, there was no dose-response effect. We've already gone over that. We'll go more into that in actually part 2 because there are some pretty interesting critiques that were not well thought out. But really, and this was the thing that floored me in your presentation, as I mentioned, we had not seen the scans before the April 7th paper. You show slides, the scans, and remember, the scans are the scans of the scans. The scans are the source of truth. You can visibly see— you don't need to be a cardiac imaging expert— you can visibly see the regression. It is there without a shadow of a doubt. Clearly didn't pick it up. That seems odd, just statistically. Also, when you can see it, and then both HeartFlow and Q-Angio are picking up a decent number of regressors. Did I misstate anything there?

No, I, to be, you know, I want to be fair that I'm not an imaging expert, but yes, that was just one of several slides Dr. Budoff prepared that was pretty exciting because he's literally, it's the same patient. It's their baseline scan. It's their follow-up scan. And while I'm not an imaging expert, those highlights that are on the scan are part of the software. I believe it's the software that actually interjects so that you can highlight not only fibrofatty plaque but low attenuation plaque, which is the more dangerous version. You can see the volume changes and it, it was pretty exciting. You know, I, I try my best when bringing around how exciting it is that we have regression. To remind everyone, to be fair, there's a subset of our cohort that are higher progressors. It's a small subset, but it is worth emphasizing anytime we start talking in excited levels about this regression. But by golly, with that said, I'm very excited, Nick, that we have some regressors, verified regressors, that we have 3 of the data, the analyses confirming that we have regression, and And then we have Dr. Budoff and his team visually verifying themselves that there was actual regression for a number of our participants. That's exciting. That's really exciting. And certainly not anything anyone was predicting.

All right. I want to summarize some high-level points that I think people need to walk away with, and then I'm going to leave them on a cliffhanger. So point one, there's information we didn't have when this paper was published. On the backend, we've been trying to get to the truth and make that as transparent as possible at the time points we can. For both me and David, it's been very, very hard not to speak up about particular things, but there are reasons for that. I hope some of those have been clarified now. I understand there are complicated dynamics and we're now trying to provide more clarity, um, so hopefully we've progressed on that aim. Um, certainly nothing was hidden. Secondly, with respect to the CLEARLY analysis, which has been a point of controversy. The CLEARLY team said they do a quality control check when issues were raised. They then reversed that decision, and subsequent analyses, including HeartFlow and Q-Angio, disagree with CLEARLY but agree with each other. Finally, there is confirmed regression in a subset of participants, and the initial emphasis that yes, this is a heterogeneous risk population. We are not saying lean mass hyperresponders are immune to heart disease. They're heterogeneous. But what predicts plaque progression is not LDL, it is not ApoB, and baseline plaque predicts plaque progression. So from like a functional standpoint, a practical standpoint for people, probably should know your coronary artery calcium scan that predicts progression. And quite honestly, that is entirely consistent with pre-existing literature. So really what we're doing here is not saying LDL, ApoB don't matter in the general sense. It's broadening the aperture and providing people, particularly this population, with the information one needs to make an informed choice. I want to wrap it up there. I'm going to give you the last word, but I do want to say I'm even more excited for part 2 of this interview. We're going to respond to some critics and do some spicy true and false. So I've been waiting for a while to be let off the leash. I'm going to let myself off this leash. Definitely come back for part 2. Dave, any final thoughts?

Yeah, 2 things. One, as Nick already knows, but that I'll let the audience know when they watch this, I still kind of have everybody, uh, who is an author on this on a little bit of a leash with regards to the Clearly situation. We're still in a process. There's still a dialogue going on behind the scenes, and so I've asked them to not comment on it, uh, for the sake of what we're trying to accomplish. At the end of the day, anything that we're not sharing for the purpose ultimately of getting our way, hopefully in getting the best data out to everybody long term, right? Like, that's ultimately the, the goal. The other, the other piece I really do want to emphasize, I hope because this really has kind of come up a lot. I hope that everybody's taking the time to listen to the people who are having it all on the table, who aren't leaving, you know, pieces out for which it's not about getting to the truth. If it's, if the, for example, I think it is appropriate for us to acknowledge if we had some things we could have done better. And I'm proud of the fact that in the wake of April 7th, We did have some things we could have done better, for example, with the social media rollout, some corrections I'm glad we immediately acknowledged and wanted to put in the paper. And I think that that speaks a lot to us both as individuals and as a team. I will say that it has frustrated me that a number of critics have left that piece out of it. There's kind of a larger narrative that I think Either they're getting to the critics in part 2.

Yeah, I have a lot to say.

Yeah, I'm just— all I'm saying is I think I would just suggest for anybody who's following this, just be sure that they have all the information, you know, both from supporters and detractors alike. That's just where I guess I'll wrap it.