SPLENDOUR: RANKL blockade in advanced NSCLC did not reproduce its own exploratory signal

What was tried

SPLENDOUR (NCT02129699) was a randomised, open-label Phase 3 trial run by the European Thoracic Oncology Platform with the EORTC, asking whether adding denosumab at 120 mg every three to four weeks to first-line platinum doublet chemotherapy improved overall survival in stage IV non-small cell lung cancer. Randomisation was 1:1, stratified by bone metastases at diagnosis, ECOG performance status, histology and region, with overall survival as the single primary endpoint. The trial ran from 6 January 2015 to 29 February 2020.

The registry status is TERMINATED, with the posted reason: "Completion of recruitment was not feasible as accrual was slower than anticipated. Even if a benefit for denosumab could be shown, these results would be of very limited clinical impact by the time they would be available." The registry enrollment field records 595, while the posted participant flow and the publication both report 514 randomised.

Denosumab (CHEMBL1237023, ChEMBL maximum phase 4) is a fully human monoclonal antibody against RANKL, the TNF superfamily member 11 encoded by TNFSF11 (ENSG00000120659). It was already approved for skeletal-related event prevention when SPLENDOUR opened.

The biological hypothesis

RANKL is the primary signal for osteoclast-mediated bone resorption, which is why denosumab entered oncology as a skeletal agent. The survival hypothesis came from elsewhere. Tan and colleagues showed in 2011 that tumour-infiltrating regulatory T cells drive mammary cancer metastasis through RANKL-RANK signalling. Faget and colleagues later reported that RANKL signalling sustains primary tumour growth in mouse models of lung adenocarcinoma (doi:10.1016/j.jtho.2017.11.121).

The clinical trigger was an exploratory analysis. In the randomised Phase 3 comparison of denosumab against zoledronic acid in bone metastases from solid tumours other than breast or prostate, the lung cancer subset favoured denosumab: 9.5 versus 8.0 months median overall survival in 702 NSCLC patients (HR 0.78, p equal to 0.01) and 8.6 versus 6.4 months in the squamous subset (HR 0.68, p equal to 0.035), labelled exploratory by the authors (doi:10.1097/JTO.0b013e31826aec2b).

Open Targets scores TNFSF11 against non-small cell lung carcinoma at 0.321, from literature at 0.834, clinical evidence at 0.482 and RNA expression at 0.185. No genetic datatype contributes, and the target's strongest associations remain skeletal, led by osteoporosis at 0.729.

What actually happened

SPLENDOUR was powered to detect an increase in median overall survival from 9 to 11.25 months, a hazard ratio of 0.80, requiring 847 overall survival events. It never got close. Accrual slowed, the trial closed early, and conditional power was at or below 10 percent at termination.

The posted results are flat. Median overall survival was 0.73 years (95% CI 0.63 to 0.92) in the control arm and 0.68 years (95% CI 0.62 to 0.87) with denosumab, hazard ratio 0.96 (95% CI 0.78 to 1.19), p equal to 0.355 by pre-specified Cox regression. The publication reports the same comparison as 8.7 versus 8.2 months. Progression-free survival was identical, median 0.39 years in both arms, hazard ratio 0.99 (95% CI 0.82 to 1.19), p equal to 0.459.

The subgroups do not rescue it. Among the 275 patients with bone metastases, the population closest to the one that produced the original signal, the hazard ratio was 1.02 (95% CI 0.77 to 1.35). Without bone metastases it was 0.90 (95% CI 0.66 to 1.23). The authors described denosumab as well tolerated.

Failure mechanism, best guess

The proximate cause of termination was accrual. The scientific cause is that SPLENDOUR did not ask the question the 2012 signal answered.

The exploratory signal came from patients who all had bone metastases, comparing denosumab against zoledronic acid, an active bone agent. SPLENDOUR enrolled an all-comer stage IV population in which 53 percent had bone metastases, and compared denosumab against nothing added. Those are different contrasts. A benefit driven by superior suppression of skeletal disease relative to a bisphosphonate need not appear as survival benefit over no bone agent, and it would be diluted by the 47 percent with no bone disease to modify.

The bone-metastasis subgroup closes the optimistic reading. If the mechanism were skeletal, that subgroup should have moved. It sat at 1.02.

One caution is worth stating rather than glossing. The two estimates are not formally incompatible. Taking the 2012 NSCLC hazard ratio of 0.78 with its reported two-sided p of 0.01, the implied standard error of the log hazard ratio is 0.0965 (0.2485 divided by 1.96), an approximate 95% interval of 0.65 to 0.94. SPLENDOUR's own log hazard ratio standard error is 0.108, computed from its reported interval. The two log hazard ratios differ by 0.208 with a combined standard error of 0.145, roughly 1.4 standard errors. A truncated trial did not refute the earlier estimate so much as fail to reproduce it without the events to separate the two.

How to prevent this next time

Design the replication against the population that generated the signal. The original comparison was denosumab versus zoledronic acid in bone-metastatic disease. A confirmatory trial in unselected stage IV patients with no bone agent in the control arm tests a different hypothesis.

Power against the conservative end of the exploratory interval, not the point estimate. Planning on a hazard ratio of 0.80 when the source interval reaches 0.94 sets an event target a slow-accruing trial will not reach.

Pre-specify a futility rule tied to enrolment velocity. Conditional power was at or below 10 percent by the time the trial closed. A rule converting a projected accrual shortfall into an early stop, rather than a five-year drift toward one, releases patients and money sooner.

The single highest leverage change would have been running the confirmatory trial in bone-metastatic NSCLC against an active bone agent, matching the comparison that generated the exploratory signal, rather than testing denosumab against no bone agent in an unselected stage IV population.

What this means for similar programs

Denosumab remains a skeletal agent in NSCLC. The publication states that the data do not provide evidence of clinical benefit in patients without bone metastases, and the bone-metastasis subgroup adds none.

Programmes reading RANK and RANKL as an immuno-oncology axis should note what SPLENDOUR did not test. It added RANKL blockade to cytotoxic chemotherapy before checkpoint inhibitors became first-line standard here. It did not combine RANKL blockade with PD-1 or PD-L1 inhibition, which is where the regulatory T cell mechanism would predict an effect. That question remains open, and a sponsor pursuing it should say explicitly that SPLENDOUR is not evidence against it.

The wider lesson concerns exploratory survival signals from trials designed for a different endpoint. Treating one as a hypothesis worth an 847-event Phase 3 requires an explicit account of why it should survive a change of comparator and population.

Open questions

Does RANKL blockade add anything to checkpoint inhibition in NSCLC? SPLENDOUR predates that standard of care.

Was the 2012 signal a comparator effect rather than a denosumab effect? Relative harm from zoledronic acid would produce the same subset result, and no trial has separated them.

Would a bone-metastasis-restricted design have shown anything? The subgroup hazard ratio of 1.02 argues against it, on 275 patients with an interval of 0.77 to 1.35.

Is there a biomarker for RANKL dependence in lung adenocarcinoma? The mouse work implies tumour-intrinsic signalling, and no clinical assay has been carried forward.

Sources

1. • ClinicalTrials.gov, NCT02129699, SPLENDOUR, TERMINATED, with posted results. https://clinicaltrials.gov/study/NCT02129699 - Peters S, Danson S, Hasan B, et al. A Randomized Open-Label Phase III Trial Evaluating the Addition of Denosumab to Standard First-Line Treatment in Advanced NSCLC: The ETOP and EORTC SPLENDOUR Trial. Journal of Thoracic Oncology 2020, volume 15, pages 1647-1656.- Scagliotti GV, Hirsh V, Siena S, et al. Overall survival improvement in patients with lung cancer and bone metastases treated with denosumab versus zoledronic acid: subgroup analysis from a randomized phase 3 study. Journal of Thoracic Oncology 2012, volume 7, pages 1823-1829.- Henry DH, Costa L, Goldwasser F, et al. Randomized, double-blind study of denosumab versus zoledronic acid in the treatment of bone metastases in patients with advanced cancer (excluding breast and prostate cancer) or multiple myeloma. Journal of Clinical Oncology 2011.- Tan W, Zhang W, Strasner A, et al. Tumour-infiltrating regulatory T cells stimulate mammary cancer metastasis through RANKL-RANK signalling. Nature 2011.- Faget J, Contat C, Zangger N, et al. RANKL Signaling Sustains Primary Tumor Growth in Genetically Engineered Mouse Models of Lung Adenocarcinoma. Journal of Thoracic Oncology 2018, volume 13, pages 387-398.- Open Targets Platform, TNFSF11 (ENSG00000120659) and non-small cell lung carcinoma (MONDO_0005233), overall association score 0.321. https://platform.opentargets.org/target/ENSG00000120659 - ChEMBL, DENOSUMAB, CHEMBL1237023, maximum phase 4. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL1237023/ - openFDA FAERS, denosumab case counts retrieved 9 September 2026. https://open.fda.gov/apis/drug/event/.