The LEVEL trial (NCT05983250, Tenax Therapeutics) just announced their results and even though I completed the analysis and predictions before the readout, I did not write up the results and so take this with a grain of salt. Nonetheless, there were some interesting issues involved in running the predictions. As before, this analysis is using publicly available information, and I am using Claude to help with medical terminology and interpretation. I also used Claude to quickly iterate on multiple models.
LEVEL was testing whether oral levosimendan improves walking ability in patients with pulmonary hypertension caused by heart failure with preserved ejection fraction (PH-HFpEF), meaning the heart's ejection fraction is preserved (it squeezes normally) but the ventricle is stiff and fills poorly. This is a condition in which high blood pressure in the lung arteries arises because a stiff left heart backs pressure up into the lungs. The trial randomizes 230 patients 1:1 to drug or placebo and measured the change in 6-minute walk distance (6MWD) — how far a patient can walk in six minutes — from the start of treatment to week 12. As with the ACACIA trial, we estimate the probability that LEVEL will meet its primary goal: a statistically significant improvement in 6MWD versus placebo. I wish these trials used a different decision framework (like the probability that the drug is effective exceeding some threshold, for example), but since the statistical significance is the game, that's what we use.
Figure 1 shows the summary of the predictive distribution for the key estimand, with the predictive mean of the treatment effect close to zero.
And here the key event probabilities.
PH-HFpEF is a hard condition to treat and the estimate of the treatment effect is largely a reflection of this fact. As always we have constructed a background set of similar studies. The pooled estimates of the in-scope trials' treatment effects are displayed on Figure 2.
We pool within a class (shrinkage toward the class mean) with a shared global heterogeneity parameter, but the class means themselves are independent as we don't pool across classes. The key source of optimism came from the HELP trial in which levosimendan showed an encouraging walking-distance signal. HELP used an intravenous infusion, whereas in LEVEL the drug was administered orally. Perhaps more importantly, HELP Phase 2 trial enrolled a sicker, more selected population: it required more severe pulmonary hypertension (mPAP ≥35 vs ≥30 mmHg), had a lower baseline 6MWD (292 vs ~333 m), and — most importantly — randomized only acute hemodynamic responders (patients whose exercise wedge pressure fell ≥4 mmHg on a test infusion). The core question, similar to what we wrestled with in ACACIA, is how much of this effect would transfer.
We constructed the model from three pieces and one key assumption.
We then ask how often a trial of LEVEL’s size and design would reach statistical significance, averaged over all of this uncertainty. The one genuinely subjective input is $\pi$. We set it to the base rate: drugs have almost never improved exercise capacity in PH-HFpEF, so we center $\pi$ around 0.20 — the low historical success rate, upweighted modestly because levosimendan’s mechanism is different from the drugs that failed and it has its own positive signal (HELP). As you can see from Figure 3, even under the optimistic assumption of the transfer probability, the success rate does not breach 50%.
The top line results were announced on Monday, August 10 and the trial was a miss.
They reported the mean improvement in 6MWD of 3.5m and a p-value of 0.63, from which we reverse engineer the 95% confidence interval of [-11, 18]. They also reported a subgroup, patients with lower baseline 6MWD, similar in composition to HELP, that did show an improvement passing a nominal statistical significance filter.
The trial was considered a failure and the company stock took a dive on announcement. But does this mean the drug doesn't work? We don't know the answer to that question. If our pre-readout model is even approximately right, the prior (to readout) probability that the true effect was positive at all was about 0.45, but the probability of a clinically meaningful effect (an improvement of at least 25 m) was only about 0.08. The readout makes a large, transferred effect unlikely, so the probability of a meaningful benefit is now lower than before, even as the probability of some positive effect nudges up slightly, since the observed mean came in a hair above zero rather than at our estimate of essentially zero. I have not reviewed the AEs but if they are moderate, a real (if modest) shot at helping in a disease this hard to treat is not nothing. Unfortunately, we may never know if there are people for whom the drug may be beneficial as under the market conditions the company may not have adequate resources to run a follow-up trial in the population similar to their Phase 2 study.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.
Trying to cut corners to make nonlinear PK model fitting faster
.svg)
This is a comment related to the post above. It was submitted in a form, formatted by Make, and then approved by an admin. After getting approved, it was sent to Webflow and stored in a rich text field.