I Was Ready to Pay to Watch My Plaque Shrink. Then I Learned the Number Moves on Its Own.

The Bottom Line There is software now that takes a coronary CT scan you already had and measures the plaque in your arteries: how much, what kind, and where. It reports the answer in cubic millimeters, sometimes to a decimal place. Several companies sell it, Medicare gave it a billing code in January, and the pitch that has taken hold in the longevity world is that you can scan yourself every year or two and watch your plaque shrink.

If you are deciding whether to get one, here is the short version. The measurement is real, and if you have real risk and your coronary arteries have never been properly characterized, one of these analyses is worth having, once. The trend line is the part that is oversold. When researchers scanned the same 101 patients twice, one hour apart, with identical protocols, the number that best predicts heart attacks came back different. Nothing in those arteries changed in an hour. The number changed anyway.

This software is precise enough to detect a treatment effect across hundreds of people and not precise enough to tell you whether your own plaque moved. Those are two different jobs, and only one of them is being sold to you.

Vocabulary that matters

• Quantitative plaque analysis (QPA), also sold as AI-QCT: software that takes an existing coronary CT angiogram and measures plaque volume and composition, instead of a radiologist describing it in words like "mild" or "moderate."
• Low-density noncalcified plaque, also called low-attenuation plaque: the soft, lipid-rich, inflamed fraction of a plaque. It shows up dark on CT because fat blocks fewer X-rays than fibrous tissue or calcium. This is the measurement that carries most of the risk signal.
• Scan-rescan reproducibility: scan somebody, then scan them again before anything could possibly have changed. How close are the two answers? That gap is your measurement error, and it sets the floor on what you can detect later.
• Repeatability coefficient: the plain version of that floor. The amount a single person's measurement has to move before you can call the change real instead of measurement noise.

This post started as research for a purchase I was considering, one that likely would have partially come out of pocket vs full insurance coverage.

I have three coronary CT scans in my medical record: one from 2016 at age 36 which showed zero plaque and fully patent arteries, the one I argued my way into at 44, which sent me to the cath lab, where the angiogram measured an 80 percent blockage in my distal right coronary artery, and the follow-up a year later. A radiologist read each one, answered the question in front of him, and moved on. The raw data contains more than either report: how much plaque is in the rest of my arteries, what it is made of, and whether a year of aggressive treatment changed it. My ApoB, the count of cholesterol-carrying particles that actually lodge in an artery wall, is 38. This is rock bottom, newborn baby levels and the subgroups that were studied that led to development of PCSK9 inhibitors.¹³ So, I would very much like to know what that aggressive lowering has bought me.

So I asked whether insurance would pay to have the scans read this way. For one of them, I am still waiting on the answer. For the second, it is already no: a repeat analysis of a follow-up scan is classified as disease surveillance, which is excluded. That left the question of whether to pay cash. Before I did, I read the measurement science. This post is what I found.

The experiment that should end the yearly-scan idea

A group in Denmark took 101 patients with known coronary artery disease and scanned them twice, one hour apart, using identical acquisition protocols, then ran plaque quantification software on both scans.¹

An hour is the point. Atherosclerosis takes years to decades in the absence of stepwise progression from plaque rupture. Whatever was in those arteries at 9 a.m. was still there at 10 a.m., in the same amount, in the same form. Any difference between the two readings has exactly one source: the scanner, the reconstruction, and the software.

For total plaque, calcified plaque, and noncalcified plaque, the two scans correlated at 0.93 to 0.99. For low-density noncalcified plaque, the correlation was 0.74 to 0.77.

That is the measurement that predicts heart attacks. It is the reason anybody wants this software instead of a calcium score. And it is the one the machine is worst at reproducing when nothing has changed. The authors did not spin it. Their own word for the overall result was "modest," and reproducibility got worse with lower image quality, with certain reconstruction settings, and depending on which artery the lesion sat in.

Two readings of the same heart. The gap between them is the measurement error.

Two pushbacks a vendor would make, both fair. The study used semi-automated software, a trained human contouring alongside the algorithm, so this is not any company's fully automated product, and the low-density number carries reader variability stacked on top of scanner physics. And a correlation of 0.93 across 101 very different patients means the software ranks people consistently, not that any one person's number held still.

Why the most important number is the hardest to measure

This is physics. Better software does not fix it.

Low-density plaque is not something the software recognizes the way you recognize a face. It is a brightness cutoff: on the CT brightness scale, everything below roughly 30 Hounsfield units gets called low density. So the measurement is a threshold applied to a continuous, noisy signal, and everything that shifts the image brightness shifts how much tissue falls under the line: the contrast dye concentration at the moment of the scan, the heart rate, the reconstruction settings, the X-ray tube voltage. Underneath all of it, a coronary artery is two to four millimeters across and moves with every heartbeat, and the lipid core you are measuring is a small volume inside that small moving structure. The compartment that matters most biologically is the smallest, dimmest, and most protocol-sensitive one. Total plaque volume is big, bright, easy, and the least interesting of the three.

A separate group scanned patients twice, two heartbeats apart, changing only the tube voltage in half of them. The voltage change alone shifted measured noncalcified plaque by a mean of 25 cubic millimeters, where the matched-protocol repeat shifted it by zero on average.² Worse, it moved the detection floor: with matched settings, a change had to exceed 9 cubic millimeters to count as real biology. With mismatched voltage, it had to exceed 89. One dial on the scanner inflated the threshold tenfold.

Their conclusion on low-attenuation plaque matters because of who wrote it:

Two of the paper's authors disclose ties to Cleerly, one on its scientific advisory board and one as a consultant. People inside the field, some of them paid by the industry, saying the low-density measurement is not ready. They are more generous about total and noncalcified plaque, but only for imaging sites that have measured their own scan-rescan variability and can show it. As far as I can find, almost nobody selling you a scan can.

The strongest case for the other side

A Cedars-Sinai group ran their own scan-rescan study, 30 participants from two centers, repeat scans a median of six days apart, and got excellent results: intraclass correlation coefficients of 0.96 or better across every plaque type, and a repeatability coefficient of only 5.1 cubic millimeters for low-density plaque.³ If that is the real performance, serial tracking works.

Read how they got it. The scans were compared side by side in one systematic session, by two assessors working to a consensus reading, using brightness thresholds tuned to each individual scan rather than a fixed cutoff, a choice they tested head to head and which won. That is a fair description of what careful measurement takes, and an unfair description of what you will get at a scanning center. Three of the authors receive software royalties and hold equity in a company in this space, which does not make the result wrong. It means the people who validated the software have a stake in it.

Population signal, individual noise

There is a good randomized trial of serial plaque imaging. It enrolled 365 asymptomatic, intermediate-risk people with a family history of premature coronary disease, randomized them to calcium-score-informed prevention or usual care, rescanned everyone at three years, and had an independent core laboratory measure the plaque.⁴ It worked. Total plaque grew 24.9 cubic millimeters under usual care and 15.4 under informed care. Noncalcified plaque grew 15.7 versus 5.6. Real differences, statistically significant, on plaque volume. Not on heart attacks; every serial plaque trial so far has measured the surrogate.

Two more numbers from the same results table. The standard deviations on those total-plaque changes were 37.7 and 30.9. The variation between individuals was larger than the average change itself. And the best published error bar on a single person's total plaque measurement, from the scan-rescan study above, is 25.5 cubic millimeters. Three years of average progression in the untreated group was 24.9. One person's measurement noise and three years of disease are the same size. That error bar comes from different software on different scanners, so treat it as a magnitude rather than an exact figure, but nothing published puts the two orders of magnitude apart, and orders of magnitude apart is what personal serial scanning would need.

This is not a contradiction, and the trial is not wrong. It is how measurement works. Average across roughly 180 people per arm and the noise shrinks by the square root of the headcount, making the group mean about thirteen times more precise than any single reading inside it. The trial can see the effect precisely because it is a trial. You, scanned twice, are an n of one, and the same instrument that resolves the population signal cannot resolve yours.

Think of a bathroom scale that reads a few pounds high or low at random while your diet moved you two. Weigh a thousand people and you prove the diet worked, because the errors cancel across the group. Weigh yourself twice and you have learned something about the scale.

What the field's own documents say

The American College of Cardiology published a scientific statement on this at the end of 2025.⁵ It restricts the analysis to patients with visible plaque on their scan. For repeat scans it recommends a minimum interval of at least two to five years depending on baseline risk, the same scanner, tube voltage and reconstruction settings where possible, and only in situations where the result would meaningfully change management. It also says vendor reports should carry a statement of the software's reproducibility, and that each vendor should define what amount of plaque is clinically meaningful at all. That means that at the end of 2025, the field's leading imagers were still asking the companies for their error bars, and the products were already shipping.

Meanwhile a separate international consensus group went the other direction, recommending that any detected plaque lead to a recommendation of drug treatment, with the 70th percentile of plaque volume triggering high-intensity therapy.¹² That came from a Delphi process, a structured poll of experts, not a trial. So we now have percentile treatment thresholds built on a measurement whose reproducibility the ACC is still asking vendors to publish. And scale matters here: this software finds some plaque in up to 96 to 97 percent of the people it looks at, which is why one group, several of whose authors take Cleerly money, went and derived a floor below which risk is low, specifically to head off overdiagnosis.¹⁰ A test that can be positive in nearly everyone is not telling you much until somebody defines the number that means something.

Reading the sales material like somebody who used to write it

I spent fifteen years in the supplement industry. That is where I learned how a health claim gets built, and I want to be careful here, because these are FDA-cleared devices with real validation studies, not somebody's powder. The evidence is far better. The moves are the same moves.

Report the numbers that look good. HeartFlow's validation study is the strongest in the field: 237 patients, 432 lesions, checked against intravascular ultrasound, correlations of 0.91 for total plaque, 0.91 for calcified, 0.87 for noncalcified.⁶ Good numbers, no quarrel. The study also measured low-attenuation plaque, and the abstract, which lists it among the volumes quantified, never comes back to it. The correlation is in the body of the paper: 0.28. And one number further in: the average gap between software and ultrasound for low-attenuation plaque was 3.89 cubic millimeters. The software's own median measurement was 3.3; ultrasound, in the lesions where it found any low-attenuation plaque, put the median at 8.1. The measurement error is the same size as the quantity being measured. The authors offer a fair defense, that ultrasound is itself a poor reference for lipid buried deep in the wall. That defends the 0.28. It does not defend leaving it out of the abstract, the only part most readers and every press release will see. Seven of the authors are full-time HeartFlow employees with salary and equity, which they disclose.

Validate where it is easy, apply where it is hard. Elucid's software carries the strongest possible validation label: checked against real tissue under a microscope.⁷ That tissue was carotid plaque, cut out of necks during surgery, because you cannot cut a plaque out of a living person's coronary artery. Carotids are larger than coronaries and do not move with the heartbeat; the FDA's own clearance document for an earlier version notes the specimens were carotid and warns about small vessels and motion. Then the label travels: a study of 209 psoriasis patients measured lipid-rich necrotic core in coronary arteries using what it called "histopathologically validated" software from the same company, and an Elucid scientist from the carotid work is a co-author.⁸ Nobody hid anything. The validation was done in one kind of artery and applied in another.

Borrow a number from a different question. Cleerly's serial-comparison product is aimed at exactly what I wanted, one scan against the next. Its product page carries a 41 percent reduction in death and non-fatal heart attack, attributed to SCOT-HEART. Real trial, real result, but it is the result for getting a coronary CT at all versus standard care, not for comparing two scans. The same page carries no measurement-error statement, no minimum interval, and no same-scanner requirement, the three things the ACC statement recommends for serial scanning. It does say, twice, that the AI report is "reviewed, refined, and confirmed by a human reviewer," worth knowing when you are picturing an algorithm.

Conflict of interest is not the same as bias, and some of the sharpest warnings in this post came from authors who take vendor money. I am not telling you these companies lie. I am telling you which claims the products actually rest on and which are decoration, a skill I learned from the other side of the table.

The inflammation layer

One newer idea here is more interesting than plaque volume. Inflamed coronary arteries change the character of the fat around them, and that change is visible on the CT you already had. In a study of 40,091 consecutive patients across eight UK hospitals, the patients with no obstructive narrowing were 81.1 percent of the group and accounted for 66.3 percent of the major cardiac events and 63.7 percent of the cardiac deaths.⁹ Four out of five people are told their arteries are not significantly narrowed, and that group produces two thirds of the major cardiac events and nearly two thirds of the cardiac deaths. Reading coronary disease as plumbing misses most of the people it kills, which regular readers know is the drum I keep beating.

The inflammation signal predicted cardiac death independently of risk factors and of how much disease was present, and the FDA authorized the product built on it two weeks ago. The caveats are real. The headline hazard ratio of 29.8, a measure of how much faster one group's risk accumulated, comes from a 3,393-patient subgroup, compares the most extreme contrast the paper reports, and carries a confidence interval from 13.9 to 63.9, meaning the true value could plausibly sit anywhere in that very wide range. Six of the study's authors are employees, founders, or directors of the company selling the test.

Still, I like the direction. Plaque volume is a photograph of damage already done. Inflammation is the process itself, running now, and it sits on the short list of cellular processes biologists think drive aging in every organ of the body, the hallmarks of aging. It is the more interesting measurement. It is also, today, the less proven one.

Should you pay for one of these scans?

If your arteries have never been characterized and you have real risk, one of these analyses is worth having. It measures disease instead of estimating risk, and it sees the soft plaque a calcium score is blind to. This is also the use insurance actually covers: Medicare priced the analysis's new billing code, CPT 75577, in January 2026, at roughly $950 in a hospital outpatient setting and about $1,012 in a physician office. The Medicare contractors that have published coverage rules pay for it only in patients with chest pain, no known coronary disease, and minimal-to-moderate disease on the scan; a dozen states have no published policy and decide case by case.

Read what is not covered. The coverage policies exclude two uses by name: screening people without symptoms, and disease surveillance. Those are the two reasons the longevity world wants this test, and the exclusion was written by the payers' own medical reviewers.

Do not buy the yearly trend line. If a difference between two of your scans has to clear the error bars above before it means anything, a one-year rescan buys you noise. Track ApoB, high-sensitivity CRP, and metabolic markers such as HbA1c and fasting insulin instead: cheap, reproducible, and they already tell you whether treatment is doing its job. If you ever do repeat the scan, wait at least two years, insist on the same scanner and protocol, and ask for the two studies to be read side by side.

The boring instruments are the reproducible ones.

Here is where I landed on my own scans. Insurance may cover the analysis on one of my scans; I am still waiting. The second is a firm no, because a repeat analysis counts as disease surveillance, and after reading everything above I think the payer has the science right on that one. A cash-pay comparison would have handed me a precise-looking number I could not trust: the interval is inside the noise floor, my stented artery is the least measurable segment on the study, since, per the FDA's own authorization document, the companion flow analysis has only been validated in vessels without metal in them, and the two reports were never even read against each other. My follow-up scan came back dramatically better, severe category down to minimal, on ApoB 38 with a statin plus a PCSK9 inhibitor, the injectable drug that pushes cholesterol clearance further than a statin alone. And the report says, in plain text, COMPARISON: None. Two radiologists, a year apart, each answering the question in front of them. My apparent regression might be completely real. I want it to be. What I actually need is a formal side-by-side re-read of the two scans I already have, and that is what I am asking for.

I stopped asking what a number means before asking what its error bar is. That habit is free, and this month it saved me a thousand dollars.

The Final Signal

• What is right about this technology. It measures disease instead of estimating risk, it sees soft plaque a calcium score misses, and the validation for total, calcified, and noncalcified plaque volume is strong. When doctors see this data they change treatment about half the time, mostly by intensifying prevention, though that comes from a registry with no control group.¹¹
• What is oversold. The trend line. Serial scanning to watch your own plaque shrink is being marketed years ahead of the measurement science, and the least reproducible number in the report is the one that carries the risk.
• The counterintuitive part. The same instrument can be good enough for a randomized trial and not good enough for you. Averaging across hundreds of people cancels the noise that dominates a single measurement. That is how measurement works.
• What to do with it. Get it once, on a first real workup, if your arteries have never been characterized. Do not buy it annually. If you repeat it, same scanner, same protocol, at least two years, read side by side.
• What would change my mind. Published, vendor-specific thresholds for how much change exceeds measurement error under ordinary clinical conditions, and an outcome trial showing that managing people by plaque numbers beats managing them by ApoB. The first is being asked for. The second is running: a 7,500-patient trial of AI stage-based care reads out in 2029, and the only randomized trial of serial scanning reads out in 2028.
• What is next. The inflammation thread. If the interesting signal is biology in motion rather than anatomy in a snapshot, the next question is which measurements actually track the machinery of aging while it runs.

References

1. Iraqi N, Mortensen MB, et al. Interscan reproducibility of computed tomography derived coronary plaque volume measurements. J Cardiovasc Comput Tomogr. 2024;18(6):583-592. PMID: 39379201 · doi:10.1016/j.jcct.2024.09.009 [Finding: 101 patients were scanned twice one hour apart using identical acquisition protocols, with semi-automated software. Interscan correlation was 0.93 to 0.99 for total, noncalcified and calcified plaque but only 0.74 to 0.77 for low-density noncalcified plaque; the authors described overall reproducibility as modest.]
2. Calicchio F, Epstein E, et al. Impact of technical, patient-related and measurement variables on serial Hounsfield unit-based quantitative coronary plaque analysis in computed tomography: time for a new chapter. Eur Heart J Imaging Methods Pract. 2025;3(1):qyaf014. PMID: 40066146 · doi:10.1093/ehjimp/qyaf014 [Finding: Changing tube voltage from 140 to 100 kVp shifted measured noncalcified plaque by a mean of 25 mm3 against 0 mm3 for a matched protocol, and inflated the repeatability coefficient from 9 mm3 to 89 mm3. The authors concluded that low low-attenuation plaque volumes with wide scan-rescan limits of agreement will not be statistically meaningful in serial studies of that plaque type.]
3. Lenell J, Park C, Kwiecinski J, et al. High scan-rescan repeatability of AI-enabled coronary plaque quantification from coronary CT angiography. Eur Radiol. 2026. PMID: 42120754 · doi:10.1007/s00330-026-12598-1 [Finding: In 30 participants from two centers rescanned a median of 6 days apart, side-by-side reading gave intraclass correlations of 0.96 or better, with a repeatability coefficient of 25.5 mm3 for total plaque and 5.1 mm3 for low-density plaque. Two assessors read by consensus, and scan-specific attenuation thresholds outperformed fixed ones.]
4. Nerlekar N, et al. Effects of Combining Coronary Calcium Score With Treatment on Plaque Progression in Familial Coronary Artery Disease: A Randomized Clinical Trial. JAMA. 2025;333(16):1403-1412. PMID: 40042839 · doi:10.1001/jama.2025.0584 [Finding: In 365 randomized participants rescanned at 3 years by an independent core laboratory, total plaque progression was 24.9 mm3 (SD 37.7) with usual care versus 15.4 mm3 (SD 30.9) with calcium-score-informed prevention. The primary outcome was plaque volume, a surrogate, not clinical events, and the standard deviations exceed the mean changes.]
5. Chandrashekhar Y, Blankstein R, Shaw LJ, et al. Quantitative Coronary Plaque Analysis in Clinical Practice: 2025 ACC Scientific Statement. JACC Cardiovasc Imaging. 2026;19(5):637-652. PMID: 41405512 · doi:10.1016/j.jcmg.2025.11.008 [Finding: Recommends quantitative plaque analysis only in patients with visible plaque, a minimum repeat interval of at least 2 to 5 years depending on baseline risk using the same scanner and reconstruction parameters, that vendor reports carry a statement of software reproducibility, and that vendors define a clinically meaningful amount of plaque. Published online 2025-12-17; the print issue is dated 2026.]
6. Narula J, et al. Prospective deep learning-based quantitative assessment of coronary plaque by computed tomography angiography compared with intravascular ultrasound: the REVEALPLAQUE study. Eur Heart J Cardiovasc Imaging. 2024;25(9):1287-1295. PMID: 38700097 · doi:10.1093/ehjci/jeae115 [Finding: In 237 patients and 432 lesions, AI plaque quantification correlated with intravascular ultrasound at 0.91 for total plaque, 0.91 for calcified and 0.87 for noncalcified. The abstract reports those three and omits low-attenuation plaque, whose correlation, reported in the body, was 0.28 with a mean disagreement of 3.89 mm3 against a median software-reported volume of 3.3 mm3 (median 8.1 mm3 by ultrasound in lesions with detectable low-attenuation plaque). Seven authors are full-time HeartFlow employees.]
7. Buckler AJ, Sakamoto A, St Pierre S, Virmani R, Budoff MJ. Virtual pathology: Reaching higher standards for noninvasive CTA tissue characterization capability by using histology as a truth standard. Eur J Radiol. 2023;159:110686. PMID: 36603478 · doi:10.1016/j.ejrad.2022.110686 [Finding: Lipid-rich necrotic core measured from CT angiography matched histology with a difference of 0.15 mm2 and a slope of 0.92. The tissue came from endarterectomy specimens, not coronary arteries; the first author is an employee and shareholder of Elucid Bioimaging.]
8. Choi H, Uceda DE, Dey AK, et al. Treatment of Psoriasis With Biologic Therapy Is Associated With Improvement of Coronary Artery Plaque Lipid-Rich Necrotic Core. Circ Cardiovasc Imaging. 2020;13(9):e011199. PMID: 32927971 · doi:10.1161/CIRCIMAGING.120.011199 [Finding: Measured coronary lipid-rich necrotic core in 209 psoriasis patients using software the paper describes as histopathologically validated, an example of carotid-derived validation being applied to coronary arteries. An Elucid scientist is a co-author.]
9. Chan K, Wahome E, Tsiachristas A, et al. Inflammatory risk and cardiovascular events in patients without obstructive coronary artery disease: the ORFAN multicentre, longitudinal cohort study. Lancet. 2024;403(10444):2606-2618. PMID: 38823406 · doi:10.1016/S0140-6736(24)00596-8 [Finding: Among 40,091 consecutive patients having coronary CT angiography, the 81.1% without obstructive disease accounted for 66.3% of major adverse cardiac events and 63.7% of cardiac deaths. In a 3,393-patient prognostic subgroup, perivascular fat attenuation predicted cardiac mortality independently of risk factors, with a hazard ratio of 29.8 for the extreme three-vessel quartile contrast and a wide confidence interval of 13.9 to 63.9. Six authors are employees, founders or directors of the company.]
10. Bar S, Knuuti J, Saraste A, et al. Derivation and validation of an artificial intelligence-based plaque burden safety cut-off for long-term acute coronary syndrome from coronary computed tomography angiography. Eur Heart J Cardiovasc Imaging. 2025;26(7):1163-1173. PMID: 40243706 · doi:10.1093/ehjci/jeaf121 [Finding: AI detects some coronary plaque in up to 96 to 97 percent of patients, so the authors derived a percent atheroma volume safety cut-off of 2.6% to avoid overdiagnosis, validated across Finnish and Dutch cohorts. Several authors disclose Cleerly funding or advisory roles.]
11. Rinehart S, Blankstein R, McCarthy CP, et al. Guiding Preventive Care Strategies for Patients With Atherosclerotic Plaque on Coronary CTA: Primary Outcomes of the DECIDE Registry. JACC Cardiovasc Imaging. 2026;19(6):706-716. PMID: 41817483 · doi:10.1016/j.jcmg.2026.01.014 [Finding: Among 972 symptomatic patients who already had atherosclerotic plaque, management changed in 51.3% after AI plaque results were released, mostly by intensifying medical therapy. LDL fell by a median of 11 mg/dL in patients whose management changed and rose by 1 mg/dL in those whose did not, a non-randomized comparison between self-selected groups in an uncontrolled pre-post registry rather than a measured treatment effect. Three authors are HeartFlow employees, the substudy was supported by HeartFlow site grants, and the authors call for randomized trials.]
12. Schulze K, Dewey M, et al. Coronary CT angiography evaluation with artificial intelligence for individualized medical treatment of atherosclerosis: a Consensus Statement from the QCI Study Group. Nat Rev Cardiol. 2026;23(2):100-115. PMID: 40751112 · doi:10.1038/s41569-025-01191-6 [Finding: A three-step Delphi expert consensus recommending that the presence of any atherosclerotic plaque lead to a recommendation of pharmacological treatment, and that the 70th percentile of total plaque volume warrant high-intensity treatment, with percentile curves derived from the DISCHARGE and SCOT-HEART trials.]
13. Cohen JC, Boerwinkle E, Mosley TH, Hobbs HH. Sequence variations in PCSK9, low LDL, and protection against coronary heart disease. N Engl J Med. 2006;354(12):1264-1272. PMID: 16554528 · doi:10.1056/NEJMoa054013 [Finding: In the ARIC cohort followed 15 years, the 2.6% of Black participants carrying PCSK9 nonsense mutations had 28% lower LDL cholesterol and an 88% reduction in coronary heart disease risk; a milder variant in white participants was associated with 15% lower LDL and a 47% risk reduction. Lifelong genetically low LDL is the observation that made PCSK9 a drug target.]