It’s been a little while since the release of our April 7th paper for the Keto-CTA study and news on
our documentary around it, The Cholesterol Code. I wanted to wait until all four analyses were in before doing a full update here on the blog.

INITIAL CONTROVERSY

Interestingly, since April, critics have asserted our study wasn’t detecting an association
between LDL cholesterol and plaque progression because “everyone had high LDL.” As
one prominent critic put it, “All of these people had high LDL and high ApoB.”

This criticism assumes we had a narrow range of LDL levels—something like a
standard deviation of, say, 10 mg/dL clustered tightly around our mean of 253 mg/dL. If
that were true, I’d agree the criticism would be valid.

But our actual spread was already published in Table 1 of our paper. That table points
out that we had a standard deviation of 84.7 mg/dL and an interquartile range of 202 to
308 mg/dL. Our participants’ LDL cholesterol levels ranged from 49 mg/dL at the lowest
to 591 mg/dL at the highest—a spread of 542 mg/dL.

I’m going to make what some might consider an audacious statement: I’m confident we
have the largest spread of LDL cholesterol of any prospective imaging study ever
conducted. I wasn’t able to find anything even close to comparable.

This matters because the lipid hypothesis itself is built on dose-dependent relationships.
As the 2017 European Atherosclerosis Society consensus statement puts it: the
probability of plaque development “increases in a dose dependent manner” and is
“proportional to both the absolute magnitude and duration of exposure to elevated LDL
cholesterol.”

The dose-dependent, log-linear relationship between LDL and atherosclerosis isn’t a
movable goalpost—it IS the lipid hypothesis. We had the dose variation. We had the
range. The association simply wasn’t there.

THE CLEERLY SITUATION

So why weren’t we more vocal about defending our work? The answer lies in what was
happening behind the scenes with our data analyses.

Remember: the scans are the scans are the scans. All our analyses—semi-quantitative,
Cleerly, HeartFlow, and QAngio—looked at the exact same 200 CT scans from our 100
participants.

When I received access to the raw Cleerly data on April 18th, eleven days after
publication, I found several concerning issues. I brought these to the attention of both
Lundquist and Cleerly immediately. In an April 24th meeting, Cleerly’s then-CMO, who was also a co-author on our paper, agreed these were important concerns and
committed to running a blinded quality control pass to confirm or disconfirm the results.
However, what followed surprised us. The previous commitment was disregarded, and Cleerly asserted they would not follow through on the blinded quality control pass.

At that point, we knew we needed to engage another independent, AI-guided CTA analysis service. This is what brought us to look into Heartflow.

MULTIPLE INDEPENDENT CONFIRMATIONS

Under normal circumstances, when two analyses disagree, adding a third to break the tie would be fairly standard. But these weren’t normal circumstances. Many long time critics had become quite attached to the Cleerly analysis, and thus we wanted to add an additional blinded, independent confirmation to help us all get a stronger redundancy for confirmation.

Now, here are the results. For absolute change in non-calcified plaque volume—the
numbers that matter most—Cleerly showed 18.8 mm³, while HeartFlow showed 5.5
mm³ and QAngio showed 5.6 mm³. HeartFlow and QAngio were in close agreement
with each other, but both disagreed substantially with Cleerly.

VERIFIED REGRESSION

Here’s the exciting part. Both HeartFlow and QAngio showed substantial numbers of
participants with non-calcified plaque regression—33 regressors for HeartFlow and 15
for QAngio in non-calcified plaque volume. More importantly, Dr. Budoff and his team at
Lundquist have independently verified this regression in a subset of these participants
by direct visual inspection of the scans.

This slides was taken directly from my presentation:

FINAL THOUGHTS

Needless to say, it’s been exciting to see all these data come together.

Across multiple independent analyses—and with direct confirmation from one of the world’s leading cardiology imaging teams—the results have remained consistent: the predicted dose-dependent relationship between LDL cholesterol and plaque progression simply did not appear.

It’s worth underscoring that, yes, there is a subset of rapid progressors, as you’d expect in any middle-aged population. But even here, progression shows no association to LDL cholesterol; in fact, some progressors had levels lower than the cohort average.

Conversely, we observed clear evidence of regression in another subset of participants, including some with much higher than average LDL cholesterol—indeed, some with extremely high levels. (We’ll expand more on this in our upcoming paper.)

There’s still much more work ahead, and I look forward to sharing future announcements. For now, I want to thank each and every one of you for your tremendous support—it’s been essential to moving this research forward.

The following preliminary data from the Keto-CTA / LMHR Study were presented on August 18th at Symposium of Metabolic Health in San Diego.

IMPORTANT – These Data are Preliminary. This has not undergone peer review and preparation for publication, but we’ll likely have more to share this in the coming months.

As always, please continue to work with your doctor. This research is an ongoing effort to inform decision making with care providers, but not to replace it.

Match Analysis Presentation from CoSci

Before reading this article, if you haven’t already, please watch Dr. Matt Budoff’s keynote presentation at CoSci this year on the Keto-CTA / LMHR Study match analysis with the Miami Heart study (MiHeart).

These Data are Presented With Limited Analysis

Consider the following data more a capture-and-report

Quick Review of Total Plaque Score (TPS)

The total plaque score (TPS) utilizes the 15-segment American Heart Association model of the coronary arteries. These 15 segments used in a total plaque score are chosen because they represent areas in the arteries that are more susceptible to plaque formation, due to the mechanical forces that contribute to endothelial dysfunction and inflammation. Each plaque rated with a score of 0 – 3 based on plaque volume. TPS is a summation across the 15 segments, yielding a TPS score range of 0 – 45.

Total Plaque Score (TPS) vs Lipid Metrics

Keto CTA (LMHR Study) vs MiHeart

Spearman correlation of general lipid values vs TPS

Note from Dave: As I discussed from the SMH presentation, these data are not too surprising given the context. See presentation when it’s released for a more in-depth discussion.

Total Plaque Score (TPS) vs Lipid Particle Counts

Total Low Density Lipoprotein Particles (LDL-P) vs TPS

Keto-CTA Only

R² 0.0015 – No correlation between Total LDL Particles (LDL-P) and Total Plaque Score (TPS)

Note from Dave: This was very exciting to see this born out on our data as it has been long speculated on. I’ve had discussions with Peter Attia, Layne Norton, Howard Luks, and many others speculating on high LDL-P and plaque with LMHR at a population level.

Total Plaque Score (TPS) vs Small Dense LDL Particle (sdLDL-P)

R² 0.001 – No correlation between Small LDL Particles (sdLDL-P) and Total Plaque Score (TPS)

Note from Dave: This was likewise exciting to see with the context of the participants of our study as I had speculated on this outcome as well [ie here, here, here. See the talk when it is released for a deeper dive.

Total Plaque Score (TPS) vs Lp(a) & OxPL-ApoB

Total Plaque Score (TPS) vs Lp(a)

R² 0.007 – No correlation between Lp(a) and Total Plaque Score (TPS)

Note from Dave: I’ve had many great discussions with Sam Tsimikas on this given his unique expertise in this area. Indeed, the relevance if Lp(a) in this context outside other acute phase reactants (ie C-Reactive Protein) does appear to be relevant in here. I’ll be interested in seeing our longitudinal data on this as well for comparison with progression too.

Total Plaque Score (TPS) vs OxPL-ApoB

R² 0.0002 – No correlation between OxPL-ApoB and Total Plaque Score (TPS)

Note from Dave: Happy to see this hypothesis get some testing as well. “I think we’ll have many #LMHRs with higher Lp(a) yet lower than expected oxPL-ApoB given both those levels and their ApoB.” While I haven’t seen the aggregates yet, I suspect this may bear out when we are completing the final paper.

Quantitative Analysis

Quantitative analysis using plaque volume and AI-guided reading is a more objective and detailed method compared to the semi-quantitative approach of total plaque score (TPS). Here’s how they compare directly:

• Measurement Precision: While total plaque score offers a rough estimate based on visual inspection, plaque volume provides exact measurements of plaque size in cubic millimeters, allowing for more precise monitoring of plaque progression or regression.
• Observer Variability: TPS is subjective and can vary significantly depending on the clinician’s interpretation, whereas quantitative analysis with AI-guided reading reduces this variability by standardizing measurements across different cases, leading to more consistent results.
• Detail of Plaque Characteristics: Semi-quantitative TPS generally assesses the presence and extent of plaque but lacks detailed information on the specific composition and characteristics. In contrast, AI-guided plaque volume analysis can capture intricate details like the plaque’s composition (e.g., calcified or non-calcified), providing deeper insights into potential risk factors for cardiovascular events.
• Resource and Time Requirements: TPS is faster and easier to perform, requiring fewer resources and no advanced software. On the other hand, quantitative plaque volume analysis is more resource-intensive and time-consuming, as it relies on advanced imaging techniques and AI-powered algorithms to deliver accurate and comprehensive data.

Note from Dave: An interesting aside, early into the study there was a concern at one point we might actually have too few patients with baseline plaque to capture adequate progression data. We discussed possible contingencies, even the possibility of splitting the study into two studies. However, within the year there were enormous advancements in AI-guided analyses of CCTA scans which resolved the issue entirely (See Cleerly). These analyses identify plaque volume in every scan.

Preliminary Quantitative Data for MiHeart Match

The following are the Median PV (Plaque Volume) for the Keto-CTA and MiHeart cohorts. (Note: one scan from each could not be processed). For previous Table 1 & 2 values, see Dr. Budoff’s presentation.

Note from Dave: Understandably, these are the data I was most interested, particularly Non-calcified Plaque Volume. There’s quite a bit to say on this, but for now, I’d rather just emphasize we’ll be doing a deeper analysis on this in the coming paper.

Match Analysis Excluding Cholesterol Lowering Medication

Preliminary Quantitative Data for MiHeart Match

Easily the most requested reanalysis since the match was reported by Dr. Budoff was an exclusion of the 26 participants of MiHeart who were on cholesterol lowering medication.

Note from Dave: As with the PV of the original match analysis above, these values are extremely close, particularly our major endpoint of Non-calcified Plaque Volume with each group. Again – and with emphasis – these data are preliminary. Our final analysis will be published soon.

The KETO-CTA (#LMHRstudy) vs Matched Control (#MiHeart) analysis is now published in Metabolism.

https://doi.org/10.1016/j.metabol.2024.155854

METHODS

80 Participants of #LMHRstudy fell within #MiHeart age range and were then matched 1:1 for age, gender, race, diabetes mellitus, hyperlipidemia, hypertension, and past smoking to asymptomatic subjects from the #MiHeart cohort.

PRIMARY ANALYSIS

High resolution heart scans (#CCTA) allowing for primary analysis of Total Plaque Score (TPS), Total Stenosis Score (TSS) and Segment Involvement Score (SIS)

RESULTS

The matched mean age was 55.5 years, with mean #LDL cholesterol of 272 mg/dL (max LDL-C 591) mg/dl and mean 4.7 years duration on a ketogenic diet.

• There was no significant difference in coronary plaque burden of #LMHRstudy (mean LDL-C 272) cohort as compared to #MiHeart controls (mean LDL 123 mg/dl); nb: pre-KETO LDL-C in KETO group was 122 mg/dl
• There was no significant difference in CAC (median and IQR) [0 (0,56)] versus [1 (0, 49)], p = 0.520
• No relationship of LDL-C elevations and plaque

Note 1 – This analysis is on baseline scans, we will have further data on the Keto-CTA longitudinal analysis in the coming months. And — as always — please continue to work with your doctor.

Note 2 – this is a published abstract, but is not open access. A full paper for this match analysis will be published soon in a different journal and will be completely open access.

On November 27th, 2015, a number on a piece of paper changed my life forever. The massive increase I saw in my LDL cholesterol after adopting a low carb diet would ultimately send me into an entirely different life path, one I’d chronicle in real time on this blog. Now, eight and a half years later, I find myself completely entrenched spearheading research to unravel this mystery and whether it will demonstrate risk for folks like me.

But I have a confession… I’ve had many low points throughout this journey. Obstacles large and small have emerged to slow me down or even outright stop this research.

However, I’ve also seen just how far folks will go to support this effort. Beyond letters and DMs, many have contributed directly to our charity, the Citizen Science Foundation, or have networked us with the right researchers and personnel

Yet nothing compares to what just happened. We decided to take a chance and hold a charity event – the Collaborative Science Conference – or as we call it, “CoSci”. It was over March 15 and 16th and… well… it was amazing. People from all over the world came to donate, with that donation being their ticket to the event.

We had a mix of both featured speakers that were very popular, and citizen science speakers, many of which had never spoken at any conference before this one. And in spite of how short notice we were in announcing it just two and a half months ago, we had over 300 people come and join us.

The part I can’t express in words is how much our event had heart. It felt as though the “spirit of citizen science” was truly in the air. An optimism for what could be possible and how attendees were helping to make it real, just like the rest of us.

Given both the staggering generosity in donations and the considerably positive feedback, we will likely hold another CoSci next year as well. But more importantly, I’m thankful for the level of optimism this inspires for me. We’re proving this model of crowdfunded, self-directed science is quite real and can be well supported with events like these.

For everyone who donated and came to enjoy our event – thank you. You aren’t just helping us fund the next study, you’re reinforcing the drive for us to keep pressing forward.

A few days ago, Mario Kratz, released a video around lipids and ASCVD that also featured our work with Lean Mass Hyper-Responders and the Lipid Energy Model.

To be sure, while I did got a chance to listen to it at 2x speed on a drive between two meetings, it was more of a skimming, in a sense. But today I’m going to listen live with members & patrons at 7am PST (see companion post to this one – and if not a member, you can register here).

I’ve give my live reactions, but as always, I want to keep it both respectful and productive. I can say in advance my interest in this video is in large part due to my existing respect for Mario and his diligent work on his content. But moreover, he and I are already of comparable opinions with regard to metabolic health and its enormous importance in reducing risk.

I’ll then hand off my video to Mario directly, and if he feels it is helpful in advancing the conversation, I may release an edited version of the video for my channel.