01 The Pattern

01. One Signature, Three Trials

Set the three readouts next to each other and the shape is hard to unsee.

ATTR-CM and AL amyloidosis Phase 3 readouts, 2024–2026
Trial (drug, class) Overall primary Cleaner subgroup Pre-treated subgroup
HELIOS-B
vutrisiran, TTR silencer
Met. ACM + recurrent CV events, HR 0.72 (0.56–0.93), p=0.01 No stabilizer at baseline: HR 0.67 (0.49–0.93), p=0.02 On tafamidis at baseline: HR 0.79 (0.51–1.21), not significant
CARDIO-TTRansform
eplontersen, TTR silencer
Missed. CV mortality + recurrent CV events, no significant benefit No stabilizer at baseline: HR 0.71, nominally significant, pre-specified On stabilizer at baseline: no treatment effect
CARES
anselamimab, AL fibril depleter
Missed. ACM + CV hospitalisation, Mayo IIIa/IIIb, no significant benefit Pre-specified AL-kappa subgroup: clinically meaningful survival + CVH benefit

HELIOS-B differs from the other two in its verdict alone. It won overall, yet the benefit still bunched up in stabilizer-naive patients and thinned out in those already on tafamidis. CARDIO-TTRansform and CARES fell on the wrong side of significance while keeping a clean signal in the patients with the least therapeutic background, or the most uniform disease biology. In both, that signal sat outside the statistics that protect a claim.

One miss can be pinned on a molecule. Two misses and a heavily gradiented win, spanning two mechanisms inside twelve months, point somewhere upstream of any single drug — to how these trials were built. That is what the rest of this analysis works through.

02 The Mechanism

02. What the Stabilizer Ceiling Is, and Why It Exists

The companion analysis to this piece named the general version of the problem the add-on ceiling: a new medicine stays biologically active, but the treatment environment around it has already stripped out enough risk that the incremental effect turns small, slow, or hard to separate statistically. In ATTR-CM that general effect has one specific driver, and the driver deserves its own name. Call it the stabilizer ceiling.

The arithmetic is not subtle. A trial's power to detect a relative risk reduction depends on the number of events it accrues, and the event count depends on how sick the control arm is allowed to be. When most of the placebo group receives an effective stabilizer, the placebo arm stops behaving like an untreated control and starts behaving like an actively-treated one with a lower event rate. Fewer events accrue. The same relative effect now produces a smaller absolute gap, and the confidence interval widens around it. A sponsor can respond by enlarging and lengthening the study, and in ATTR-CM one did: CARDIO-TTRansform enrolled 1,432 patients, more than twice HELIOS-B's 655, and followed them to 140 weeks. It still came up short, because the background had risen faster than the sample size could compensate for.

This is where the two silencer trials diverged, and the usual explanation gets it backwards. The split had nothing to do with one trial restricting stabilizers and the other allowing them. Both let stabilizer-naive patients start tafamidis after enrolment, and in HELIOS-B roughly 22% of the monotherapy group did, at a median of about 18 months. What separated them was the starting point. HELIOS-B opened with around 40% of patients already on a stabilizer. CARDIO-TTRansform opened near 57% and drifted up toward 81% exposure by the end. One trial ran under a lower ceiling and got over it. The other ran under a higher one and did not.

The comparison, like for like
Baseline stabilizer — HELIOS-B~40%
Baseline stabilizer — CARDIO-TTRansform~57%
Any-time exposure — HELIOS-B~53%
Any-time exposure — CARDIO-TTRansform~81%
Enrolment655 vs 1,432
Both the baseline figures (40 vs 57) and the cumulative figures (53 vs 81) point the same way. The gap is not an artifact of how exposure is counted; it holds on either definition.
03 The Winning Trial

03. The Ceiling Is Visible Inside the Trial That Succeeded

The strongest objection to a stabilizer-ceiling reading is that HELIOS-B disproves it. If vutrisiran showed benefit when added to a stabilizer, the ceiling cannot be a hard limit. That objection is worth stating plainly, and then answering with HELIOS-B's own numbers.

Within HELIOS-B, the treatment effect was not flat across stabilizer status. In patients not on tafamidis at baseline, vutrisiran reduced all-cause mortality and cardiovascular events by 33% (HR 0.67, 95% CI 0.49–0.93). In patients already on tafamidis, the reduction fell to 21% and lost significance (HR 0.79, 95% CI 0.51–1.21). The direction of that gradient is identical to CARDIO-TTRansform's: the silencer does its clearest work where a stabilizer is not already doing part of the job.

So the two datasets do not contradict each other. They are the same curve read at two points. HELIOS-B sampled the part of the curve with enough headroom left that even the diluted, higher-exposure segment stayed on the right side of the overall endpoint. CARDIO-TTRansform sampled further along, where the headroom was mostly gone. The ceiling is not a wall that HELIOS-B failed to hit. It is a slope, and HELIOS-B was standing lower on it.

Read this way, the ceiling stops being an argument between two trials and turns into a property of the disease area both of them measured — which is exactly why it carries beyond either one.

04 Across Mechanisms

04. The Pattern Jumps to a Different Mechanism and a Different Disease

If the ceiling were only about silencers stacked on stabilizers, it would be a narrow ATTR-CM lesson. CARES says it is broader.

Anselamimab is not a silencer. It is a fibril depleter, an antibody designed to clear light-chain amyloid already deposited in tissue, and it treats AL amyloidosis rather than ATTR. Different target, different disease, opposite end of the mechanistic logic: where a silencer reduces the protein being made, a depleter removes what has already accumulated. In July 2025, AstraZeneca reported that the CARES Phase 3 programme missed its primary endpoint in the overall Mayo stage IIIa and IIIb population. It also reported a clinically meaningful improvement in survival and cardiovascular hospitalisation in a pre-specified subgroup, patients with AL-kappa disease.

Nothing about anselamimab's molecule connects it to eplontersen. Its structural summary, though, reads almost word for word: overall miss, pre-specified subgroup rescue, effect concentrated in the more tractable slice of patients. That is the connection mainstream coverage has missed. Inside twelve months AstraZeneca has now watched this pattern play out twice in amyloidosis, once through its Ionis-partnered silencer and once through its Alexion depleter. Two programmes, two mechanisms, one signature is past the point where coincidence about a single drug explains much. It looks structural.

Heterogeneity is the through-line. A silencer trial in a stabilizer-saturated market and a depleter trial in an advanced, mixed AL population run into the same trouble: a large share of the enrolled patients have little room for the new mechanism to move their outcome, and that share drags the average toward flat. The signal exists; the patients who were never going to respond much simply average it into invisibility.

05 Architecture

05. Why This Is an Architecture Problem, Not a Molecule Problem

Amyloidosis is an early, vivid case of a pattern that recurs whenever a disease area matures. As background therapy in an indication becomes more complete and more widely adopted, three things happen at once, and they compound.

  • The control arm gets healthier. Effective standard of care lowers the event rate in placebo, which is exactly the number a trial's power depends on.
  • The enrolled population gets more heterogeneous. Early trials recruit untreated patients with room to improve. Later trials recruit a mix of treated and untreated, early and advanced, and the average response flattens.
  • The remaining benefit moves earlier in the disease. Once structural damage is established and background therapy has captured the easy risk, a new mechanism often has to arrive sooner to show a detectable effect at all.

Cardiology has watched a version of this for years. As guideline-directed heart-failure therapy grew more layered, every new agent added onto an already-treated baseline had to earn its place against a smaller residual risk, and add-on effect sizes shrank to match. None of this is new as a principle. Amyloidosis just shows it cleanly and at scale, with the subgroup data laid out so the compression can be read off the page instead of inferred.

So the design decision now carries as much weight as the molecule. A genuinely active drug can still return a flat trial when it is tested as an add-on over a saturated background, in a population chosen for real-world breadth rather than for how likely it is to respond to the mechanism. The drug still has to work. Whether the trial can see it working is a separate question, decided before the first patient is dosed.

06 The Framework

06. Five Design Choices, and What Each One Costs

A structural ceiling can be designed around. Every ATTR-CM programme now has to decide where to point its trial, and each choice trades statistical cleanliness for commercial breadth or the reverse. None of the five comes free. The task is picking the one that suits the asset in hand.

Replace the stabilizer
Head-to-head vs tafamidis / acoramidis
Cleanest read on the molecule, removes the confound entirely. Costs the most: a superiority trial against an approved drug is large, long, and commercially unforgiving if it lands at parity.
Enrich for the naive
Enrol stabilizer-naive patients only
Targets the part of the curve where the effect lives, as both silencer trials showed. Narrows the eventual label and shrinks the addressable population as stabilizer penetration keeps rising.
Treat earlier
Enrol earlier-stage disease
Gives the mechanism room before structural damage dominates. Requires longer follow-up and better early diagnosis, and pushes readouts years further out.
Sequence deliberately
Silencer after stabilizer, by protocol
Matches likely real-world use and supports a combination claim. Demands dedicated sequential evidence rather than an unrestricted add-on, and accepts the ceiling as a known cost.
Add on, unrestricted
The CARDIO-TTRansform design
Maximum real-world relevance and the broadest possible label if it works. Highest exposure to the ceiling, as the readout demonstrated. The default that is now hardest to justify.

CARDIO-TTRansform's error was not the last box itself, but choosing it while leaving the estimand that mattered unprotected. Its statistical hierarchy defended the combination question. The stabilizer-naive patients who actually produced a signal sat outside that protection, so they could not inherit significance once the overall endpoint fell. A programme that expects the ceiling can settle in advance which subgroup its alpha will defend, and that one decision separates a nominal footnote from a claim.

07 The Next Test

07. The Next Readout Enters the Highest Ceiling Yet

The framework makes a falsifiable prediction, and there is a trial coming that will test it. AstraZeneca's Alexion unit, with Neurimmune, is running DepleTTR-CM, the Phase 3 study of cliramitug (ALXN2220, formerly NI006), an antibody that clears deposited transthyretin amyloid from the heart by recruiting phagocytic immune cells. It is the first depleter to reach Phase 3 in ATTR-CM, and its mechanism has no precedent in cardiology.

Consider the market it enters. When HELIOS-B enrolled, one stabilizer dominated. By the time DepleTTR-CM reads out, the ATTR-CM background will hold tafamidis, acoramidis approved in late 2024, and vutrisiran approved for cardiomyopathy in March 2025 — a crowded, well-defended market of the kind that reshapes what a new entrant has to prove. A depleter arriving into it sits on top of the fullest standard of care the disease has had, in patients defined by established cardiac deposits. On the stabilizer-ceiling logic, that is the highest add-on ceiling any ATTR-CM programme has run into.

This does not predict failure. A depleter attacks a different substrate, the deposits already in tissue rather than the protein still being made, and that is exactly the compartment silencers and stabilizers leave alone. That distinction is the best reason DepleTTR-CM could separate from a saturated background where a silencer could not. The ceiling still sets one condition. Run it as an unrestricted add-on across a broad, advanced, heavily-pretreated population and the framework flags an elevated risk of the same signature: a diluted overall number with the real effect tucked into a cleaner subgroup. Enrich instead for patients with the most modifiable deposit burden, or protect a subgroup estimand in the hierarchy, and the ceiling has been designed around. Which of the two the protocol chose is the first thing to read when the data land.

Stabilizer-ceiling exposure — forward ATTR-CM designs
Unrestricted add-on over full standard of careHigh
Three approved background therapies, advanced-disease population. The design that just failed for eplontersen, now over a fuller background.
Add-on with a protected subgroup estimandModerate
Keeps real-world breadth but defends the population most likely to respond, so a signal there survives an overall miss.
Enriched for naive or early-stage patientsLower
Targets the part of the curve with headroom. Trades label breadth for a cleaner read, as both silencer trials suggested.
Head-to-head against a stabilizerLowest
Removes the confound outright. Highest commercial and operational cost, cleanest scientific answer.
08 Systemic View

08. What the Ceiling Does to Payers, Regulators and the Economics of Combination

The stabilizer ceiling is usually framed as a developer's problem. It is also a payer's argument and a regulator's precedent, and both cut against routine combination therapy in rare disease.

For payers, CARDIO-TTRansform is close to an ideal piece of evidence. It is a large, contemporary, real-world trial that asked whether adding an expensive second mechanism on top of a stabilizer improves hard outcomes, and answered no for the treated-at-baseline majority. A payer facing a request to fund silencer-plus-stabilizer combinations now has a well-powered study to point to. The burden shifts onto sponsors to produce direct evidence that a specific patient, at a specific disease stage, benefits from the second drug. Absent that, the default reimbursement answer becomes one drug, not two.

For regulators, the harder question is evidentiary. A pre-specified subgroup that wins inside a failed trial does not, on its own, support a broad label, and the agencies will press on whether a significant interaction test backs the subgroup effect, whether it holds together biologically, and whether multiplicity was controlled — the same tension that has driven other recent evidence-versus-approval decisions. Seeing this exact situation recur across CARDIO-TTRansform and CARES surfaces a question the field has not settled: how should an add-on therapy over a maturing standard of care be judged when the overall population is, by construction, diluted? The ATTR-CM and AL amyloidosis programmes of the next few years may turn less on their molecules than on how regulators choose to weigh subgroup evidence in this structural setting.

For the economics of the field, the ceiling reprices the combination thesis that underwrote much of the silencer and depleter investment case. When a second mechanism cannot easily show incremental outcome benefit over an effective first line, its value in the same patient rides on sequencing, tolerability and disease stage instead of on stacking. That is a narrower proposition than the market had assumed, and it rewards assets that can win first-line or as a replacement over those that can only be added on — a distinction that increasingly shapes where acquisition premiums land.

09 The Countercase

09. The Case Against This Reading

A framework earns its keep by surviving its best objection, so here are the three that bite hardest.

First, three trials is a small base, and two of them share a sponsor. A pattern that runs through AstraZeneca's amyloidosis programmes might reflect that company's specific design choices rather than a law of the disease area. That is a fair caution, and it is why the HELIOS-B gradient matters: the same slope appears in an Alnylam trial, which loosens the pattern from any single sponsor's habits.

Second, the CARES miss can be read as a disease-stage problem rather than a background-saturation problem. Its population was advanced Mayo IIIa and IIIb AL amyloidosis, where mortality is high and any therapy is fighting late-stage biology. On that reading, CARES failed because the patients were too sick, not because the field was too treated. Both explanations may be operating, and they are not mutually exclusive; advanced disease and saturated background are two routes to the same diluted average.

Third, the depleter's different substrate is a genuine reason to expect DepleTTR-CM to behave unlike the silencer trials, and the ceiling framework should not be read to deny that. The prediction here is conditional on design, not on mechanism. A depleter that enriches well could clear a background that defeated an add-on silencer, and that outcome would refine the framework rather than break it: it would show that attacking a fresh substrate can reset the ceiling.

What lets the framework hold up here is that it stays conditional and testable rather than certain. It says what to look for and points to the trial that will confirm or complicate it, which is about as much as any structural reading of three trials has the right to claim.

10 ESC 2026

10. What Will Settle This at ESC

Full CARDIO-TTRansform data arrive at the European Society of Cardiology Congress in August. A handful of numbers will decide whether the stabilizer-ceiling reading holds or has to be revised.

  • The treatment-by-stabilizer interaction p-value. If the interaction is significant, the subgroup gradient is a real modifier and the ceiling reading is on firm ground. If it is not, the monotherapy signal is weaker evidence than it looks.
  • The confidence interval around the 0.71 monotherapy hazard ratio, which sizes how much of that estimate is precision and how much is a small, favourable slice.
  • The timing of stabilizer initiation in the naive subgroup, since patients who started tafamidis mid-trial dilute the very effect the subgroup is meant to isolate.
  • Whether the stabilizer-naive population held its event rate as the design assumed, or whether improving background care flattened it further than expected.

These are the same questions HELIOS-B answered in the affirmative and CARES answered ambiguously. How CARDIO-TTRansform answers them determines whether the three trials are one pattern or a coincidence that happened to rhyme.