Camlipixant in refractory chronic cough: a Phase 2b effect size Phase 3 could not reproduce

What was tried

The trial that closed here was small. NCT07684586 was an open-label Phase 1 study in healthy adults aged 18 to 55, run by GlaxoSmithKline, comparing the pharmacokinetics and relative bioavailability of camlipixant formulation prototypes (GSK5464714, ChEMBL5095035). Participants were to receive formulations A and B at two dose levels, fasted and fed, with maximum concentration, area under the curve to 24 hours and concentration at 24 hours as primary outcomes. The record was submitted on 29 June 2026, lists a start date of 2 July 2026, and has an enrolment of zero.

It was withdrawn on 31 August 2026, with the reason given as "The study has been withdrawn given the recent decision to cease further development of camlipixant in refractory chronic cough (RCC) based on Phase 3 program results". A formulation bridging study is among the last things a sponsor builds before a filing, so its withdrawal marks a far larger closure. GSK acquired camlipixant with Bellus Health in 2023, reported at about 2 billion dollars. CALM-1 (NCT05599191) enrolled 957 participants and CALM-2 (NCT05600777) enrolled 997.

The biological hypothesis

P2X3 receptors are ATP-gated channels on primary afferent neurons. ATP released from damaged or inflamed airway tissue depolarises those neurons, and the signal is read centrally as an urge to cough. Blocking P2X3 should reverse the sensitised cough reflex that defines refractory chronic cough.

The proof of concept already existed. Gefapixant, a P2X3 antagonist, was approved in 2023 and remains the only marketed agent in the class. Its limitation was taste. Because gefapixant also blocks P2X2/3 heterotrimers, which carry gustatory signalling, dysgeusia occurred in 16.2 percent of COUGH-1 and 21.1 percent of COUGH-2 participants. Camlipixant was built to solve that problem. Garceau and Chauret reported an IC50 of 25 nM at homotrimeric human P2X3 against more than 24 micromolar at P2X2/3, with antitussive activity in guinea pig models far below the concentrations needed to touch the heterotrimer.

Open Targets scores the P2RX3 association with cough at 0.541, of which clinical evidence contributes 0.886 and literature 0.062. That score is almost entirely a record of drugs that reached the clinic, not of human genetics.

What actually happened

Selectivity worked and efficacy did not. GSK reported on 17 July 2026 that CALM-1, a 52-week study, met its primary endpoint, with camlipixant 50 mg twice daily significantly reducing 24-hour cough frequency against placebo at week 12, while 25 mg did not. CALM-2, a 24-week study, missed the same endpoint at both doses. Key secondary endpoints including Chronic Cough Diary measures did not meet target thresholds in either study. GSK stated that treatment-related adverse events were similar in incidence and severity between camlipixant and placebo, and concluded that the limited efficacy shown was unlikely to transform patient care.

The company published no effect sizes with the announcement, so no placebo-adjusted Phase 3 figure can be quoted here. The Phase 2b numbers are public. In SOOTHE (NCT04678206), 310 patients were randomised for four weeks after a 16-day placebo run-in, and placebo-adjusted 24-hour cough frequency fell 34.4 percent on 50 mg twice daily (95% CI -50.5 to -13.3, p = 0.0033) and 34.2 percent on 200 mg (95% CI -50.7 to -12.2, p = 0.0047). openFDA FAERS holds no reports for camlipixant.

Failure mechanism, best guess

The Phase 3 programme was built on the middle of a Phase 2b interval nearly four times wider than the effect it pointed at. SOOTHE at 50 mg was compatible with anything from a 13.3 to a 50.5 percent placebo-adjusted reduction. Independent evidence already argued for the low end. The 2024 network meta-analysis by Yamamoto and colleagues, covering 16 randomised trials and 4,904 participants, placed camlipixant at a median 14.7 percent reduction at its ED50, 95% credible interval 5.4 to 26.0 percent, ranked third of five agents with low certainty. Gefapixant 45 mg twice daily had delivered 18.5 percent in COUGH-1 and 14.6 percent in COUGH-2, a mean of 16.6 percent.

Every external estimate clustered between 14 and 19 percent, and the sponsor's own lower bound of 13.3 percent sat in that band. The 34.4 percent point estimate was the outlier, plausibly inflated by a four-week readout in 310 patients. Duration compounds this. SOOTHE measured effect at day 28, CALM-1 at week 12 and CALM-2 at week 24, and the trial with the longest primary timepoint failed. The archetype is efficacy failure rather than translational mismatch, because the target was engaged and the class-defining side effect was solved.

How to prevent this next time

Endpoint-level Phase 3 data are not public, so the levers are base rate adjustment and design, not a Bayesian recalculation.

Power against the lower confidence bound when the point estimate exceeds the class precedent. SOOTHE's 50 mg interval spanned 13.3 to 50.5 percent. Sizing the pivotals to detect the lower bound would have changed the sample size, the doses carried forward and the price a buyer should pay.

Use the class base rate as an explicit prior. A published network meta-analysis and the only approved competitor's registrational results both predicted an effect near 15 percent before CALM-1 read out. Neither appears to have moved the programme.

Match the Phase 2 readout window to the Phase 3 primary. A four-week endpoint cannot inform a 24-week one in a condition whose placebo response averaged 33.1 percent in the same meta-analysis.

The single highest leverage change would have been running the pivotal dose against a 24-week primary endpoint in Phase 2b, before committing a two-trial, 1,954-patient Phase 3 programme to a four-week estimate.

What this means for similar programs

The Claidex Mechanistic Risk Score for P2RX3 stands at 29 of 100, band yellow, from phase burden 3.53, archetype severity 9.84, recency 4.25, genetic deficit 6.89 and saturation 4.95. The score is held down by the Phase 1 anchor and a strong association score, and it understates the situation. P2X3 antagonism is validated but ceiling-limited. Gefapixant showed that blocking the target reduces cough by roughly 15 percent against placebo, and camlipixant showed that removing the taste liability does not raise that ceiling.

Programmes still pursuing chronic cough should treat 15 percent as the mechanism-level expectation and ask what beyond P2X3 selectivity would exceed it. Sivopixant, eliapixant and filapixant rank lower in the same analysis. GSK is continuing camlipixant in a Phase 2b irritable bowel syndrome study (NCT07519395), testing visceral rather than airway afferents.

Open questions

Will CALM-1 and CALM-2 be published with placebo-adjusted effect sizes, so the shortfall can be measured rather than inferred? Did the two trials differ in placebo response or baseline cough frequency in a way that explains the week 12 and week 24 divergence? Is there a subgroup in whom P2X3 blockade clears a clinically meaningful bar?

Sources

1. ClinicalTrials.gov. NCT07684586, withdrawn 31 August 2026. https://clinicaltrials.gov/study/NCT07684586.

2. GSK. CALM-1 and CALM-2 update, 17 July 2026. https://www.gsk.com/en-gb/media/press-releases/gsk-provides-an-update-on-calm-1-and-calm-2-phase-iii-trials/.

3. Smith JA, et al. Am J Respir Crit Care Med. 2025.

4. Yamamoto S, et al. Chest. 2024.

5. McGarvey LP, et al. Lancet. 2022.

6. Garceau D, Chauret N. Pulm Pharmacol Ther. 2019.

7. Open Targets Platform, P2RX3 and cough, 1 September 2026. https://platform.opentargets.org/evidence/ENSG00000109991/HP_0012735.

8. Fierce Biotech. GSK axes key cough drug candidate, 2026.