Canakinumab plus chemoimmunotherapy in first-line NSCLC: the CANOPY-1 efficacy failure

What was tried

CANOPY-1 (NCT03631199) was a randomized, double-blind, placebo-controlled Phase 3 trial run by Novartis. It tested whether adding canakinumab, a high-affinity monoclonal antibody against interleukin-1 beta (IL1B), to pembrolizumab plus platinum-doublet chemotherapy improved outcomes in previously untreated locally advanced or metastatic non-small cell lung cancer (NSCLC). The study enrolled 673 patients and used a quadruple-masked parallel design with progression-free survival and overall survival as co-primary endpoints (ClinicalTrials.gov, NCT03631199). The sponsor terminated the trial after the primary analysis, recording that canakinumab did not improve tumor response or overall survival and that the decision was not driven by safety (ClinicalTrials.gov, NCT03631199).

The biological hypothesis

The program rested on an exploratory finding from CANTOS, a 10,061-patient cardiovascular outcomes trial of canakinumab in patients with prior myocardial infarction and high-sensitivity CRP of 2 mg/L or greater. In that population, canakinumab produced dose-dependent reductions in incident lung cancer, with a hazard ratio of 0.61 at 150 mg (p=0.034) and 0.33 at 300 mg (p<0.0001), and in lung cancer mortality, with a hazard ratio of 0.23 at 300 mg (Ridker et al., Lancet 2017). The mechanistic rationale was that IL-1 beta, a product of the NLRP3 inflammasome, sustains a tumor-promoting inflammatory microenvironment, so blocking it might slow invasion and metastasis. Open Targets scores the IL1B to NSCLC association at 0.34, and that score is carried almost entirely by literature co-occurrence (0.92) and clinical evidence (0.52) rather than human genetic evidence (Open Targets Platform). The signal that launched four registration-scale trials was biological and epidemiological, not genetic.

What actually happened

CANOPY-1 showed no separation between arms. Median progression-free survival was 6.77 months with canakinumab and 6.77 months with placebo, median overall survival was 20.83 months versus 20.17 months, and overall response rates were 45.6 percent and 45.5 percent (ClinicalTrials.gov posted results, NCT03631199). The companion Phase 3 trial CANOPY-2, in patients who had progressed after chemotherapy and immunotherapy, also missed its primary endpoint, with median overall survival of 10.6 months for canakinumab plus docetaxel versus 11.3 months for placebo plus docetaxel, a hazard ratio of 1.06 (95% CI 0.76 to 1.48) and one-sided p=0.633 (Paz-Ares et al., Lung Cancer 2024). The termination was not attributed to safety, consistent with canakinumab's established profile, and openFDA FAERS holds 1,197 canakinumab case reports, 1,109 serious and 92 with a death outcome, dominated by its approved autoinflammatory indications rather than oncology use (openFDA). Two adequately sized randomized trials returned point estimates sitting on or just past the line of no effect.

Failure mechanism, best guess

This reads as efficacy failure rooted in translational over-extrapolation. The CANTOS lung cancer result was a secondary, exploratory analysis in patients selected for atherosclerosis and elevated CRP, not for tumor biology, and the authors stated that the data were hypothesis-generating and required replication (Ridker et al., Lancet 2017). Moving directly into combinations with chemotherapy and a checkpoint inhibitor added two highly active backbones whose benefit could mask any modest anti-inflammatory contribution. The post-hoc CANOPY-2 analyses fit this picture, because patients with low CRP or undetectable circulating tumor DNA did better, but the advantage tracked baseline prognosis rather than treatment assignment (Paz-Ares et al., Lung Cancer 2024). The most economical explanation is that IL-1 beta blockade shifts a systemic inflammatory biomarker without changing the dominant drivers of tumor growth in unselected first-line NSCLC.

How to prevent this next time

The starting point is an explicit prior. A protective effect seen in one exploratory endpoint inside a trial designed for a different disease deserves a low prior probability of replication. A Bayesian go/no-go framework makes that discipline quantitative by computing the predictive probability of a positive Phase 3 before committing:

PP(success)=∫θ​P(final success∣θ,interim data)⋅π(θ∣interim data)dθ

When the prior is anchored on a single exploratory subgroup rather than confirmatory data, this integral returns a low value, and four parallel registration trials become hard to justify against it.

The second tool is a sober power and sizing calculation tied to a realistic effect. Using the standard log-rank approximation, detecting an overall survival hazard ratio of 0.80 at 80 percent power and two-sided alpha 0.05 requires roughly 630 events, while a more plausible repurposing effect of 0.90 requires more than 2,800 events. Confronting the team with that event budget before launch reframes whether the expected effect is even measurable in the planned design.

The third lever is biomarker enrichment installed prospectively, not recovered after the fact. CANTOS selected on CRP, and the CANOPY-2 subgroups that looked better were defined by CRP and circulating tumor DNA (Ridker et al., Lancet 2017, Paz-Ares et al., Lung Cancer 2024). Building the registration trial around a CRP-high or ctDNA-defined population with a prespecified analysis would have tested the actual hypothesis instead of diluting it across an unselected first-line group.

The single highest leverage change would have been to run one biomarker-enriched, CRP-selected randomized Phase 2 with a prespecified futility boundary before opening any Phase 3.

What this means for similar programs

Programs that originate in an unexpected secondary signal carry a specific risk, because the signal can be real as an association and still be the wrong causal handle. Lythgoe and Prasad reviewed canakinumab's oncology repositioning and reached the same conclusion, that repurposing on the strength of an exploratory readout should be validated in a dedicated study before scaling (Br J Cancer 2022). For the broader IL-1 and inflammasome field, the CANOPY results argue that lowering a systemic inflammation marker is not equivalent to controlling tumor-intrinsic biology, and that CRP behaves as a prognostic marker rather than a predictive one. Anti-inflammatory strategies layered onto already active chemoimmunotherapy backbones face a steep measurement problem that any new program has to address up front.

Open questions

Did a CRP-high or ctDNA-defined subgroup ever receive a fair, prospectively powered test, or does the hypothesis remain formally unfalsified? Would single-agent or earlier-line settings, where chemoimmunotherapy does not dominate the outcome, reveal a measurable IL-1 beta effect? And is the relevant axis IL-1 beta specifically, or the wider inflammasome and IL-6 output that CANTOS also moved?

Sources

1. ClinicalTrials.gov. CANOPY-1, NCT03631199, study record and posted results. https://clinicaltrials.gov/study/NCT03631199.

2. Ridker PM, et al. Effect of interleukin-1 beta inhibition with canakinumab on incident lung cancer in patients with atherosclerosis: exploratory results from CANTOS. Lancet 2017, 390:1833-1842.

3. Paz-Ares L, et al. Canakinumab plus docetaxel versus docetaxel alone in advanced NSCLC (CANOPY-2). Lung Cancer 2024, 189:107451.

4. Lythgoe MP, Prasad V. Repositioning canakinumab for non-small cell lung cancer, important lessons for drug repurposing in oncology. Br J Cancer 2022, 127:785-787.

5. Garrido P, et al. Canakinumab with and without pembrolizumab in resectable NSCLC, CANOPY-N study design. Future Oncol 2021, 17:1459-1472.

6. Open Targets Platform. IL1B (ENSG00000125538), non-small cell lung carcinoma (EFO_0003060) association. https://platform.opentargets.org/.

7. openFDA. Drug adverse event reports, canakinumab. https://open.fda.gov/.

8. ChEMBL. Canakinumab, CHEMBL1201834. https://www.ebi.ac.uk/chembl/.