When a Failed Primary Endpoint Teaches More Than a Win: What CARDIOTTRansform’s Secondary Analysis Reveals About Stratification in Complex Trials

The CARDIOTTRansform investigators already knew their Phase 3 trial had not met its primary endpoint when they opened the subgroup data. What they found inside that dataset is quietly forcing a rethink of how cardiomyopathy trials handle concomitant therapy from the moment of randomization, not as an afterthought buried in a sensitivity analysis, but as a foundational design variable. Call it the Stratification Debt Problem: the industry’s habit of treating background therapy as background noise, rather than as a treatment modifier that reshapes the entire efficacy surface of the drug being studied.

The signal is specific. This secondary analysis of the CARDIOTTRansform Phase 3 trial, published in Nature Medicine, examined patients receiving eplontersen 45 mg subcutaneously every four weeks with or without background transthyretin stabilizers, primarily tafamidis. Across 1,432 randomized participants spanning 130 sites in 20 countries, the trial was powered to detect a composite outcome. The primary endpoint fell short. But the stratified picture that emerged from this secondary analysis tells a structurally different story about who responds, under what conditions, and why the trial’s aggregate signal was almost certainly diluted by a failure to adequately pre-stratify on background therapy status.

The Concomitant Therapy Blind Spot

Transthyretin amyloid cardiomyopathy has a prevalence of 95.8 per 100,000 individuals aged 65 and older, according to data from the US Merative MarketScan Medicare Database. That patient population is not treatment-naive. By the time a patient reaches enrollment eligibility in a Phase 3 trial like CARDIOTTRansform, a large share will already be on tafamidis, diflunisal, or another stabilizer, because those agents are standard of care. The question a protocol must answer before randomization is not whether those patients will be in the trial. They will be. The question is whether background stabilizer use will be treated as a stratification variable, a covariate, or a footnote.

CARDIOTTRansform treated it as a stratification variable, which is why the secondary analysis is possible at all. But the operational insight buried here is that pre-specifying a subgroup of this kind in the statistical analysis plan is not the same as designing the trial to be powered within that subgroup. The study enrolled across a 1:1 randomization ratio, eplontersen versus placebo, but the resulting subgroup of patients on background stabilizers was not independently powered to detect the treatment effect that the secondary analysis ultimately found. That gap, between pre-specification and statistical powering, is the core design vulnerability this paper exposes.

The FDA’s own guidance on covariate adjustment, “Adjusting for Covariates in Randomized Clinical Trials for Drugs and Biological Products,” is clear that stratification variables should be incorporated into the primary analysis model. What the guidance does not resolve is the harder operational question: when a stratification variable like background therapy status is known to be a strong treatment modifier, at what point does it graduate from covariate to design driver, requiring independent powering of the resulting subgroups? CARDIOTTRansform, with its missed primary endpoint and its informative secondary analysis, sits squarely on that unresolved boundary.

What the Subgroup Data Actually Shows

Read the secondary analysis carefully and you find not a rescue narrative, but a precision signal. Patients receiving eplontersen without background transthyretin stabilizers showed a treatment effect pattern meaningfully different from those on concurrent stabilizer therapy. The mechanistic logic is not obscure: tafamidis stabilizes tetrameric transthyretin, while eplontersen reduces the production of transthyretin protein via an antisense oligonucleotide mechanism. These are not redundant mechanisms. In combination, the disease biology is altered in ways that affect the measurable endpoints, particularly cardiac biomarkers and functional capacity outcomes, differently than either agent alone.

That biological interaction is exactly what a trial design has to anticipate, not discover post hoc.

The Nature Medicine publication does not position itself as a recovery document. It positions itself as a methodological contribution, and that framing is correct. What this analysis demonstrates is that when you sort the CARDIOTTRansform population by background therapy status, the aggregate null result at the primary endpoint level was carrying two heterogeneous treatment effects that cancelled each other’s signal strength in the pooled analysis. Sponsors designing trials in a crowded standard-of-care environment, where patients arrive pre-treated, need to internalize that lesson before they write their first protocol draft.

The broader pattern here extends well beyond ATTR-CM. In oncology, background checkpoint inhibitor use is now routinely used as a stratification variable. In heart failure, background sacubitril-valsartan status is increasingly treated as a design driver. ATTR-CM trials are arriving at the same inflection point later than they should, given that tafamidis received FDA approval in 2019 and has been standard of care for nearly seven years. Every ATTR-CM trial launched after 2020 was enrolling into a landscape where background stabilizer use was the norm, not the exception.

Rebuilding the Protocol from the Stratification Up

For sponsors currently in the design phase of ATTR-CM trials, the CARDIOTTRansform secondary analysis functions as a retroactive protocol audit. The key operational question it raises: are you enrolling a mixed population and planning to adjust statistically, or are you stratifying at randomization, powering within strata, and pre-specifying stratum-specific primary hypotheses? Those are fundamentally different trial designs, and regulators will evaluate them as such.

The FDA’s covariate adjustment guidance explicitly states that covariate adjustment models should include stratification variables, but also that the choice between stratified randomization and covariate-only adjustment carries implications for how the primary analysis is interpreted. In a disease like ATTR-CM, where background therapy is both prevalent and mechanistically interactive, relying on post-hoc adjustment without stratified powering creates exactly the kind of diluted primary endpoint result that CARDIOTTRansform produced. The secondary analysis did not save the trial. It explained why the primary analysis structure was insufficient for the population that actually enrolled.

For CROs managing complex cardiomyopathy protocols, the operational implication is harder to fix than it looks. Stratified randomization on background therapy status requires real-time verification of concomitant medications at the point of randomization, a process that demands tighter integration between the RTSM/IRT system, the site’s electronic medical records, and the data management team’s eligibility verification workflow. A randomization system that cannot lock background therapy status at the moment of allocation creates exactly the kind of stratification error the FDA guidance specifically addresses, noting that if incorrect stratification occurs, the covariate adjustment model should use actual baseline strata variables. That correction is possible analytically, but it introduces uncertainty that reviewers will scrutinize.

Technology vendors selling RTSM platforms to oncology and cardiology sponsors should understand that CARDIOTTRansform’s experience represents a commercial pressure point. The next generation of complex cardiomyopathy trials will require randomization platforms capable of handling multi-variable stratification in near real-time, with background therapy status captured from source documentation, not site-reported fields that may lag by days.

The counterintuitive read of this situation is that a failed primary endpoint in a 1,432-patient Phase 3 trial may ultimately do more to advance ATTR-CM trial methodology than a clean positive result would have. A trial that wins with a diluted design teaches no one anything. A trial that misses its primary endpoint and then produces a rigorously pre-specified secondary analysis revealing differential effects by background therapy status forces the next sponsor into a harder, better protocol conversation before the IND goes in, not after the database locks.

Watch for the next ATTR-CM Phase 3 to either power independently within the stabilizer-naive stratum or to restrict enrollment to that stratum entirely. If it does neither, the CARDIOTTRansform secondary analysis will have been read by the wrong people.

References

1. Nature Medicine — “Eplontersen with and without background transthyretin stabilizers in transthyretin amyloid cardiomyopathy: secondary analysis of a phase 3, randomized controlled trial”
2. AstraZeneca — “Update: CARDIO-TTRansform Phase III Trial” (enrollment and design details, 1,432 participants, 130 sites, 20 countries)
3. FDA — “Adjusting for Covariates in Randomized Clinical Trials for Drugs and Biological Products” (guidance on stratification variable inclusion in primary analysis models)
4. Journal of the American Heart Association — ATTR-CM prevalence: 95.8 per 100,000 individuals aged 65 and older (US Merative MarketScan Medicare Database, 2022)
5. European Society of Cardiology — “Eplontersen trial did not meet its primary endpoint in transthyretin-mediated amyloid cardiomyopathy”