On Friday, Novo Nordisk announced that ZEUS missed.

Ziltivekimab is a fully human monoclonal antibody against the interleukin-6 ligand. ZEUS randomized 6,376 people with atherosclerotic cardiovascular disease, chronic kidney disease, and a high-sensitivity C-reactive protein of at least 2 mg/L to ziltivekimab 15 mg subcutaneously once a month or matching placebo, both on top of standard care. The primary endpoint was time to first cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke.

The hazard ratio was 0.99. The 95% confidence interval ran from 0.88 to 1.11.

That is not a near miss. That is a flat line.

Within a few hours, the takes had already sorted themselves into two camps. One said the inflammatory hypothesis of atherosclerosis is finished. The other said ZEUS proves nothing because the population was wrong. Both are too fast. I want to argue for something more useful and, I think, more accurate: inflammation is real, it is treatable, and it was never the main event.

Every previous anti-inflammatory cardiovascular trial had an excuse available. CANTOS used a drug that hit interleukin-1 beta, two steps upstream of the cytokine everyone actually cared about. The colchicine trials used a cheap, promiscuous, century-old alkaloid with no clean biomarker readout. CIRT randomized 4,786 patients selected for type 2 diabetes or metabolic syndrome rather than for inflammation, and low-dose methotrexate failed to lower interleukin-1 beta, interleukin-6, or C-reactive protein at all. That null told us almost nothing about the hypothesis, because the drug never engaged the target.

Hold onto CIRT, because it becomes more useful today than it was in 2019. Paired with ZEUS it brackets the question from both ends. In one trial the biomarker never moved and the events did not follow. In the other the biomarker was crushed and the events still did not follow. The first result is uninformative on its own. Together they are not.

ZEUS had none of those problems. Consider what had to go right for this trial to be a fair test, and how much of it did.

The target was chosen by human genetics. In 2012, the IL6R Mendelian Randomisation Consortium looked at a functional variant in the interleukin-6 receptor gene across 133,449 people and found that carriers had the biomarker signature of partial IL-6 blockade and lower coronary heart disease risk. Across 25,458 cases and 100,740 controls, the per-allele odds ratio was 0.95 (95% CI 0.93 to 0.97). This was held up for a decade as the model example of genetics validating a drug target.

The population was enriched for the thing being treated. Entry required hsCRP at or above 2 mg/L. The median at randomization was 4.5 mg/L, and the median interleukin-6 was 4.9 pg/mL. These were not incidentally inflamed patients. They were selected for it.

The lipid competition had already been handled. Mean baseline LDL-C was 77.7 mg/dL. That is a well-treated cohort. Whatever risk remained was, by construction, residual risk, which is precisely the space inflammation was supposed to own.

And the drug worked. Novo Nordisk states that free IL-6 and hsCRP fell as expected. In the phase 2 RESCUE trial, ziltivekimab produced dose-dependent hsCRP reductions of up to 92%, against 4% with placebo. Nobody is arguing this was a dosing problem.

So: the right target, validated by genetics, hit hard and confirmed by biomarkers, in a population selected for inflammation, already optimized on lipids, in an adequately powered event-driven trial. And the answer came back 0.99.

Pay attention to the confidence interval, because it is doing more work than the point estimate. The lower bound is 0.88. The most favorable result still compatible with these data is a 12% relative risk reduction. Canakinumab in CANTOS delivered 15%. ZEUS does not merely fail to find a benefit. It largely excludes a CANTOS-sized one in this population.

There is a cost side to this ledger. Serious infections were more frequent with ziltivekimab than with placebo, which is what you would expect from blockading a central immune signaling pathway. Overall adverse events, serious adverse events, and all-cause mortality were similar between the arms, so this was not a dangerous drug on balance. It was an immunological price paid for no cardiovascular return.

Here is the part I think the field got wrong, and I include myself in that.

Mendelian randomization is the most generous possible setting for a causal exposure. You are not measuring what a drug does for five years in a 70-year-old. You are measuring what a lifetime of altered biology does, starting in utero, compounding across seven decades. If a factor matters, lifelong exposure is where it will show up biggest.

Under those maximally favorable conditions, lifelong partial downregulation of IL-6 signaling bought a per-allele odds ratio of 0.95.

Now put that next to the equivalent exercise for LDL. Ference and colleagues pooled 312,321 participants across nine LDL-lowering polymorphisms and found that naturally randomized lifelong exposure to LDL-C lower by 1 mmol/L was associated with a 54.5% reduction in coronary heart disease (95% CI 48.8% to 59.5%). That is roughly three times the effect statins produce when started in middle age, which is exactly what you would expect if the exposure is causal and cumulative.

The two estimates are on different scales and cannot be compared unit for unit. But the magnitudes are not close, and no amount of methodological care makes them close. One target, under ideal lifelong conditions, moves risk by a few percent per allele. The other cuts it in half.

We treated the IL6R result as a green light. It was better read as a measurement. The genetics were telling us the ceiling, and the ceiling was low.

This is where the "inflammation is dead" reading falls apart, and it is why I am not making that argument.

COLCOT randomized 4,745 patients within 30 days of a myocardial infarction to colchicine 0.5 mg daily or placebo. Hazard ratio 0.77 (95% CI 0.61 to 0.96) over a median 22.6 months. LoDoCo2 randomized patients with chronic coronary disease and reported a hazard ratio of 0.69 (95% CI 0.57 to 0.83) across 451 primary endpoint events over a median 28.6 months. Those are large effects, on top of statins, in well-treated modern populations.

CLEAR SYNERGY then enrolled more than 7,000 post-MI patients treated with PCI and found nothing. The trial's investigators and the COLCOT investigators have been arguing about why ever since, with one camp attributing the null to pandemic-era disruption and inadequate inflammation reduction, and the other pointing out that the prepandemic and pandemic periods showed no significant interaction.

You can take a side or not. What you cannot do is ignore the pooled picture. A six-trial meta-analysis of 21,800 patients, including CLEAR SYNERGY, still found a 25% reduction in the primary efficacy endpoint. The FDA approved colchicine 0.5 mg daily in June 2023 for reducing cardiovascular events in established ASCVD.

And CANTOS itself was positive. Canakinumab 150 mg reduced MACE by 15% against placebo, with event rates of 3.86 versus 4.50 per 100 person-years. In the subgroup that achieved on-treatment hsCRP below 2 mg/L, the adjusted hazard ratio was 0.75 (95% CI 0.66 to 0.85), with a number needed to treat of 16.

So anti-inflammatory therapy has reduced hard cardiovascular events in humans. More than once. In trials that were not small. Anyone reading ZEUS as proof that inflammation is irrelevant owes an explanation for COLCOT and LoDoCo2, and I have not heard a good one.

Three explanations are live. They are not mutually exclusive.

1. Colchicine is not an IL-6 drug. It disrupts microtubule polymerization, impairs neutrophil chemotaxis and adhesion, and interferes with NLRP3 inflammasome assembly. Its cardiovascular benefit has never tracked hsCRP tightly. Neutralizing one circulating cytokine and damping the local cellular machinery inside a plaque are different interventions that happen to share a chapter heading. ZEUS tested the former. Colchicine does the latter.

2. The population may have been wrong for the drug. This is the strongest defense of the hypothesis, and it deserves a fair hearing. ZEUS participants had a mean age of 69.5 years, a mean eGFR of 44.5 mL/min/1.73 m², and 41.3% already had heart failure. In advanced chronic kidney disease, a large share of cardiovascular events are not classic plaque rupture. They are vascular calcification, arterial stiffening, arrhythmia, and pump failure. Blocking IL-6 does not soften a calcified conduit or restore a failing ventricle. It is possible that ZEUS aimed a plaque-inflammation drug at a population whose events were mostly not plaque.

3. hsCRP may not mean the same thing in every patient. A CRP of 4.5 mg/L in a 69-year-old with stage 3b CKD reflects uremic milieu, altered clearance, and chronic volume overload. A CRP of 4.5 mg/L in a 55-year-old two years after a stent reflects something closer to what Ridker described. Same assay, different biology, and we have been treating the number as though it were the disease.

HERMES and ARTEMIS will adjudicate a good deal of this. HERMES is testing ziltivekimab in heart failure, ARTEMIS after acute myocardial infarction, and both are expected in the first half of 2027. If ARTEMIS is positive in a post-MI population without advanced CKD, explanation 2 wins and I will have understated inflammation in this piece.

Atherosclerosis requires the retention of ApoB-containing lipoproteins in the arterial intima. That is not a risk factor relationship. It is a necessary condition. People with lifelong very low LDL do not develop the disease in meaningful numbers, and no amount of inflammation substitutes for the lipoprotein.

What inflammation does is govern what happens after the lipoprotein is there. It shapes how monocytes are recruited, how foam cells accumulate, how the fibrous cap thins, how a plaque destabilizes, and how quickly the whole process proceeds. That is a real and clinically meaningful job. It is also, unavoidably, a secondary one.

The numbers line up with that picture. Lipid lowering delivers roughly 20% relative risk reduction per mmol/L per year of treatment, works in every population tested, scales with dose, and compounds over decades. Anti-inflammatory therapy delivers something in the range of 15% to 30% in some populations with some drugs and nothing at all in others.

One of those is an engine. The other is a governor on the engine.

This matters clinically in a specific and common situation. A patient sits in front of you with an LDL of 95 mg/dL and an hsCRP of 3.2 mg/L. There is a real temptation, encouraged by a decade of enthusiastic inflammation coverage, to treat the CRP as the interesting finding. It is not. Lower the LDL. That is not a close call, and ZEUS is the most expensive demonstration of why.

Incretin therapy lowers hsCRP substantially, and SELECT demonstrated cardiovascular event reduction with semaglutide in people with overweight or obesity and established cardiovascular disease. It has become routine to gesture at the anti-inflammatory effect as part of the explanation.

ZEUS should make us more careful with that story. When you treat obesity effectively, you move weight, visceral adiposity, blood pressure, glycemia, ApoB, triglycerides, and inflammation all at once. Attributing the outcome benefit to the CRP change is exactly the kind of inference that just failed a 6,376-patient test. The honest position is that we do not know the mediation split, and that the components with the strongest independent causal evidence deserve the most credit until proven otherwise.

The same caution applies, more sharply, to the supplement aisle. Anti-inflammatory framing sells a great deal of product on the strength of a mechanism that has now failed its most rigorous pharmacological test. Lowering a marker is not the same as lowering risk. That was the whole lesson.

ZEUS did not falsify the inflammatory hypothesis of atherosclerosis. It measured it, in the most favorable conditions anyone has yet arranged, and the measurement came back at zero.

Colchicine is the reason not to overcorrect. Two large positive trials and a 21,800-patient meta-analysis say there is something real here, worth roughly a quarter of residual risk in the right patient, available for pennies a day.

Hold both. Inflammation is a genuine, modest, treatable modifier of a disease caused by something else. For four practical purposes:

The cholesterol wars ended a while ago. Today the inflammation skirmish got a lot smaller.

REFERENCES