Compass Pathways announced 52-week results from COMP005, one of its two pivotal Phase 3 trials for treatment-resistant depression, on September 9. Patients who got an additional dose of COMP360 in the trial’s final stretch showed an average 13-point drop in depression scores from baseline, sustained out to a full year. Patients who’d been on placebo the whole time and finally got their first active dose showed a 10-point drop. Both numbers come from a stretch of the trial where every patient being measured knew exactly what they were receiving.
What Part C is
COMP005 ran in three stages. Part A, through week six, was blinded and placebo-controlled, the portion that produced the trial’s original, rigorously tested result. Part B kept the blind through week 26, with eligible participants able to receive an additional dose matching their original blinded assignment. Part C, running from week 26 to 52, is open-label: dosing is explicit and no blinding remains. About 70 percent of the original cohort continued into Part C, and of those, roughly 80 percent of prior-25-milligram patients received another dose while roughly 90 percent of prior-placebo patients received their first, each knowing exactly what they were getting. The 13-point and 10-point figures, and the 40-to-45-percent response and roughly 30-percent remission rates measured in the six weeks after the Part C dose, all come from this unblinded stretch.
Why this matters more than a single asterisk
COMP006, the second pivotal trial, handled its reported six-month durability differently: that comparison stayed blinded through week 26, a point in its favor this desk noted at the time, though most patients in its 25 milligram arm also received retreatment along the way, a separate caveat with its own weight. But COMP006’s week-26-to-52 stretch is built exactly like COMP005’s Part C: open-label, with dosing known to every participant. When its one-year data arrives, it will carry the same limitation on display here, since a patient’s knowledge that they’re on the active drug can itself produce improvement that has nothing to do with the drug’s biological effect. The one-year durability evidence behind Compass’s rolling NDA submission will therefore come from unblinded extensions in both pivotal programs: it already does in COMP005, and it will in COMP006 when that data reads out. A reader trying to judge how much of COMP360’s apparent one-year benefit reflects the drug itself, versus the well-documented tendency for symptoms to improve when a patient knows they’re receiving active treatment, won’t have a blinded 52-week dataset to check that against in either trial.
The one comparison that’s more informative than it first looks
The placebo-arm patients who crossed over in Part C are worth reading more carefully than the topline number alone suggests. These are the same people, before and after, who spent months on placebo under blinded conditions, then received their first-ever active dose once the blind lifted. That within-person comparison, blinded placebo response versus open-label active response in the identical patient, is an informative, if imperfect, signal, stronger than a single-arm open-label result alone would be. It’s still not a controlled comparison: the setting changed at the same time the drug did, moving from blinded uncertainty to explicit knowledge of receiving psilocybin, so expectation effects specific to that knowledge remain fully mixed in with whatever the drug itself is doing. It’s a better data point than nothing. It isn’t a substitute for a blinded comparison arm extending through week 52.
What the safety data adds, cleanly
The safety findings carry none of this same caveat. Compass reports COMP360’s profile in Part C was consistent with Parts A and B, with no new safety signals, and adverse event monitoring doesn’t depend on a patient’s expectations the way a subjective mood score does. That’s a useful, independently meaningful piece of this release, separate from the efficacy durability question entirely.
Where this leaves the regulatory picture
Compass says its rolling NDA submission remains on track for completion in the fourth quarter, with commercial launch targeted for the first half of 2027, contingent on FDA approval. Nothing in this release changes that timeline, and nothing here suggests FDA has signaled any concern about the open-label design specifically. FDA’s own published guidance on psychedelic trial design this year treats functional unblinding as a central methodological concern the agency wants sponsors to address directly, not a footnote. Two pivotal trials whose durability claims both rest on unblinded extension data is exactly the kind of pattern that guidance was written with in mind, and how FDA’s eventual review reads it, if it addresses the pattern at all, remains to be seen.
What would resolve this
A blinded comparator arm running through the full 52-week period, in either COMP005 or COMP006, would let a reader see whether the durability gap between active drug and a comparator holds up once expectation effects are controlled for instead of layered on top of the result. Neither trial was built that way. That doesn’t mean COMP360’s one-year benefit isn’t real. It means the current data can’t yet separate how much of it is the drug and how much is what happens to most patients once they know, with certainty, that they’ve received it.