Rapamycin (sirolimus) for subclinical hypertrophic cardiomyopathy in cats: a critical appraisal of the evidence supporting conditional approval

3.1The published report and the regulatory record disagree about the endpoint

The FDA’s corrected Freedom of Information summary states, in its executive summary and again in the effectiveness section: “The sponsor did not select the effectiveness criteria before the study began. After conducting exploratory analyses, the sponsor selected MWT of the LV on Day 180 as the effectiveness endpoint”, and “The effectiveness criteria were not selected a priori.”²

The peer-reviewed report says otherwise. Its methods state that maximum wall thickness change “was assessed as the primary end point”, and its discussion defends that endpoint as “chosen on the basis of the known pathophysiology of HCM”.¹ Nowhere does it disclose that the endpoint was selected after the data had been examined. A reader of the journal article alone cannot know this, because only the regulatory file records it.

The two documents diverge in two further places. The summary reports the low-dose difference as statistically significant “on Day 60 and Day 180”, where the published Table 2 gives p = 0.564 at day 60. The summary also gives within-arm changes at day 180 that do not appear in the paper at all: a fall of 0.17 mm on low dose, rises of 0.94 mm on placebo and 0.50 mm on high dose.

Post hoc endpoint selection is among the strongest known generators of false-positive findings, and it compounds every statistical concern set out below. The failure to disclose it in the published report is a separate problem, and it is the more serious of the two, because the journal record, which is what practitioners read, is missing the single fact that most affects how the result should be weighed.

3.2An exploratory dose-finding study is carrying a marketing approval

The trial’s stated purpose was to find a dose. Its authors describe it three times as nonpivotal, exploratory and dose-determining, and set the alpha threshold accordingly; the FDA summary calls it a “pilot field study” throughout.^{1, 2} Studies of that kind are intended to generate the hypothesis that a pivotal trial then tests. They are not designed to establish that a drug works, and their statistical thresholds are chosen on the understanding that a confirmatory trial will follow.

What followed instead was conditional approval. No pivotal trial had reported at the time of that decision, and the report of the exploratory study remains the only randomised evidence in the public record. The trial itself is not at fault here. The difficulty lies in the distance between what the trial was built to answer and what it is now asked to support. The report’s own conclusion asks for less than the approval grants: that delayed-release rapamycin is “a promising novel drug for continued study”.¹

The remainder of what the approval rests on is thin, in a way that the summary itself does not conceal. Alongside the trial, the FDA lists three published papers: two mouse aortic-banding experiments^{6, 7} and one study of 70 human cardiac transplant recipients switched from a calcineurin inhibitor to sirolimus.⁸ The mouse work establishes that mTOR inhibition attenuates pressure-overload hypertrophy, which is the mechanistic rationale and not evidence in cats. The transplant study is confounded by its own design, because the hypertrophic stimulus, which is the calcineurin inhibitor, is withdrawn at the same moment that sirolimus is introduced, so the regression it reports cannot be attributed to the drug that replaced it.

3.3A relaxed alpha, three uncorrected contrasts, and a null overall test

Alpha was set at 0.10, twice the conventional threshold, and the day-180 comparisons were run unadjusted: low dose against placebo (p = 0.013), high dose against placebo (p = 0.36), and both doses pooled against placebo (p = 0.054).¹ Three overlapping contrasts at a lenient threshold, run without correction, constitute a permissive statistical environment by any standard.

The comparisons were licensed by a treatment-by-visit interaction of p = 0.084, which is significant only because alpha was 0.10; at the conventional threshold the analysis would have stopped at the main effect of treatment, which was p = 0.150. The chain from a null overall test through a marginal interaction to a significant within-visit contrast is not a chain the exploratory design was powered to support, and the authors say as much.

3.4The signal exists in the composite and nowhere else

The difference at day 180 is 0.99 mm, twice the 0.5 mm the trial’s own power calculation nominated as detectable beyond biological variability.¹ The effect is therefore not trivially small, and the difficulty with it lies elsewhere.

The FDA summary reports the same result differently, as within-arm change rather than between-arm difference: a fall of 0.17 mm on low dose against rises of 0.94 mm on placebo and 0.50 mm on high dose.² On those figures the high-dose arm, which received twice the drug, progressed at half the placebo rate and more than the low-dose arm, with no monotonic relationship between dose and effect. In a study whose declared purpose was to determine a dose, that is the finding which most requires explanation, and neither document explains it.

The endpoint takes the largest of four measurements, and the four do not share a convention: two-dimensional septal thickness was measured leading edge to trailing edge, two-dimensional free wall leading edge to leading edge, and both M-mode measurements leading edge to leading edge. Taking a maximum across measurements of differing bias is upward-biased by construction and sensitive to whichever measurement happens to run high on the day.

Examined individually, no wall measurement carried the result. Septal thickness by two-dimensional imaging, septal thickness by M-mode and free-wall thickness by M-mode all showed no treatment effect (p = 0.773, 0.779 and 0.663). Only two-dimensional free-wall thickness differed from placebo, and only when the two visits were pooled (p = 0.032).¹

The effect did not appear at the level of the individual animal either. The trial defined a change worth detecting as 0.5 mm. The proportion of cats whose wall thickness moved by more than that from screening did not differ between groups. A clinician cannot therefore identify a responding cat, and the group-level difference is generated as much by movement in the placebo arm, from 7.52 mm at day 60 to 8.40 mm at day 180, as by stability in the treated one.

3.5No intention-to-treat analysis

Thirty-six of 43 randomised cats were measured at day 180 and the analysis was confined to them. Five of the seven withdrawals were in the high-dose arm, and three of those five left because they developed congestive heart failure or died.¹ Removing the animals with the worst outcomes from the arm in which they occurred works in the drug’s favour, and it did not suffice to produce a high-dose benefit. Whether it distorted the low-dose result cannot be established from the published data: these are missing outcome data that may well be informative, and neither of the standard remedies has been published, meaning an intention-to-treat analysis or a sensitivity analysis conducted under plausible assumptions about the missing cats. The two documents do not even agree on the disposition of the animals, since the FDA summary records six exclusions and 37 cats still enrolled at day 180, whereas the paper records seven withdrawals and 36 cats completing.^{1, 2}

The report states that echocardiographic values did not differ significantly between groups at baseline, and the model adjusted for the screening value, so the wall-thickness comparison is baseline-adjusted. Body weight was not: mean weights were 6.06 kg in placebo, 5.97 kg in the low-dose arm and 4.94 kg in the high-dose arm.¹ Wall thickness scales with body size, and an unadjusted difference of roughly a kilogram between arms in a trial of 43 cats is worth reporting alongside the result.

3.6Every serious cardiac event occurred in the high-dose arm

Three of the fifteen high-dose cats withdrew with congestive heart failure, with congestive heart failure ending in euthanasia, or with death following seizures, against none on placebo and none on low dose.¹ Four of the trial’s five serious adverse events were in cats receiving rapamycin. Total adverse-event counts did not differ between groups (21, 23 and 25 for placebo, low and high dose; p = 0.82), so the signal is in the severity and its distribution rather than in the overall burden.

Attributing that distribution to chance requires assuming randomisation concentrated the sickest animals in one arm. Excess mTOR inhibition impairing myocyte autophagy is a mechanistically plausible alternative that the published record does not evaluate. The trial is far too small to distinguish between those two explanations, which is precisely the point at issue, because the dose selected for marketing was selected in a study that could not have detected harm at the dose it rejected. The report concludes that the drug “was well-tolerated at both doses studied”;¹ that conclusion follows from the absence of a significant difference in adverse-event counts, and a trial of fifteen cats per arm cannot support a claim of that kind.

The FDA reached the same impasse from the other side. Its summary records that the three cats “all had risk factors associated with progression of HCM at enrollment”, two with severe pre-existing structural disease and one with an NT-proBNP of 1344 pmol/L. It then states that “it is unknown if the heart failure and sudden death were associated with treatment”.² That is the correct conclusion to draw, and it differs from the conclusion reached in the paper, which reports the drug as well tolerated at both doses.

The low-dose arm’s death from diabetic ketoacidosis is a recognised sirolimus class effect, and the FDA summary is more specific than the paper about how it happened: the cat developed diabetes during the study, “treatment for diabetes was not initiated, and the cat continued on the study”, before presenting in ketoacidosis and dying.² Baseline glucose and liver enzymes, together with ongoing monitoring, should therefore be treated as routine in any cat receiving this drug.

3.7The liver signal is larger than the label conveys

Three laboratory safety studies were submitted. In the 24-week margin of safety study, 15 of the 24 treated cats had at least one transaminase elevation and none of the eight controls did. The elevations appeared in all three dose groups and, in the FDA’s words, “a dose-response was not present”.² Four male littermates had the most severe values: maximum AST of 110 to 515 U/L against a reference range of 16 to 34, and maximum ALT of 255 to 4552 U/L against a reference range of 41 to 160. In three of the four the elevation was present at the first post-treatment sample, on day 26 or 27, and persisted throughout the 24 weeks.

All four cats remained clinically normal and post-mortem examination found no liver pathology, which is the basis on which the study concluded the drug was well tolerated. Two observations sit uncomfortably with that conclusion. The first is that an ALT of 4552 U/L is roughly twenty-eight times the upper limit of normal, and a value of that magnitude is not treated as a laboratory curiosity in any other context. The second is that the concentration of the severe cases in four littermates points towards heritable susceptibility, which no study has followed up and which no test can currently identify in a client-owned cat.

The shorter dose-tolerance study found milder elevations, up to twice the upper limit of normal, in 8 of 24 cats, and these resolved within four weeks of stopping. That finding is reassuring with respect to reversibility, and it is also the reason the margin of safety study, which included no recovery period, cannot address the question.

The one death in the safety record belongs to a cat that was not in these studies at all. In a pilot field study in chronic kidney disease, a cat enrolled with pre-existing liver disease and a raised alkaline phosphatase developed progressive inappetence, rising ALP, ALT and AST, and icterus, and was euthanised about four months after leaving the study.² The label’s contraindication in pre-existing hepatic disease rests on that single animal.

3.8Non-linear, accumulating, and never studied alongside the drugs it will meet

The pharmacokinetics are more awkward than weekly dosing suggests. At 1.3 times the label dose the half-life was 58.3 hours, with a 95% confidence interval running from 34 to 99 hours. Dose-normalised peak concentrations fell as dose rose, which the FDA reads as absorption becoming saturated at higher doses, and comparison of exposure at day 0 with day 147 “suggests that the pharmacokinetics are non-linear after multiple dosing”, with accumulation ratios of 1.33 for peak concentration and 1.62 for exposure.²

Three consequences follow from this, and none of them is addressed anywhere in the record. First, exposure at six months is not the same as exposure at week one, so a safety profile established over 24 weeks of dosing may not describe a cat that is treated for years. Second, saturable absorption means that the relationship between the 0.3 and 0.6 mg/kg arms of the trial was not a simple doubling, which bears directly on why the higher dose fared worse and was never analysed. Third, a drug that accumulates non-linearly is the kind whose exposure a metabolic inhibitor is most likely to push somewhere unexpected. The label’s warning against CYP3A4 and P-glycoprotein inhibitors anticipates exactly that, and no feline interaction study stands behind it.³

3.9The safety studies did not enrol the patients, the dose, or the individual cat

The three laboratory studies used healthy purpose-bred cats: 32 aged 10 to 11 months, 32 aged 2 to 5 years, and 20 aged about 4 months.² None had hypertrophic cardiomyopathy. The treated population is middle-aged and older cats with cardiac disease, frequently alongside chronic kidney disease or hyperthyroidism, and the safety record does not describe them.

The label regimen itself was not tested either. The lowest dose used in the 24-week study was 1.3 times the label dose, and the dose-tolerance study gave the drug three times weekly rather than once. Neither GLP study administered 0.3 mg/kg once weekly.

The analysis units compound this. In the margin of safety study “the experimental unit was the pair housed room”, and for binary outcomes an event in one or both cats counted as a single event for the room; continuous variables were averaged across the two animals. Sixteen rooms constitute a smaller study than 32 cats, and averaging conceals precisely the kind of individual outlier that the four littermates turned out to be. The analysis was descriptive throughout, with no hypothesis testing.

The vaccine response study has the same shape of limitation. Its endpoint was that every treated cat reach a rabies titre above 0.5 IU/mL, which all twelve did.² That is a threshold rather than a comparison, because a drug could blunt titres substantially and still pass it. The study addressed one killed vaccine in vaccine-naive kittens over 56 days, and says nothing about infection risk in older cats on an mTOR inhibitor for years.

3.10The confirmatory trial is not designed to answer the question that matters

The approximately 300-cat HALT trial began dosing in April 2025 across sites in the United States, United Kingdom and Europe; the 12-month dosing phase runs into late 2026, with results expected in 2027. It is powered for the echocardiographic surrogate, not for clinical outcomes.

At the 2–6% annual event rates observed in subclinical hypertrophic cardiomyopathy cohorts,^{4, 5} 300 cats followed for a year would accrue roughly 6 to 18 congestive heart failure events in total. Detecting even a halving of risk against a 4% control rate requires on the order of 800 cats per arm. A trial of this size can confirm the surrogate; it cannot rule a clinical benefit in or out.

3.11Nothing in the public record requires an outcome study

The Freedom of Information summary sets out what conditional approval is and how long it lasts, one year renewable for up to four further one-year terms. It says nothing about what evidence full approval will require, nor whether any post-approval study is mandatory.² Unlike the FDA’s human accelerated-approval framework, no obligation to measure a clinical outcome has been disclosed. On the public record, two studies confirming the same echocardiographic surrogate would suffice. The eligibility criterion that admitted the drug to this pathway was that an effectiveness study would be unusually difficult to conduct; nothing requires that the difficulty be revisited once it has been shown to be surmountable. Full approval, if granted on that basis, should be read as surrogate confirmed twice, not as clinical benefit demonstrated.