Digoxin in pancreatic cancer: a bar that became unreachable

What was tried

The University of Nebraska, with the National Cancer Institute as collaborator, ran NCT04141995, a Phase 2a pilot adding oral digoxin to neoadjuvant FOLFIRINOX in resectable pancreatic adenocarcinoma. The design was single group and open label, with no randomization. Participants received oxaliplatin 85 mg/m2, irinotecan 150 mg/m2 with folinic acid 400 mg/m2, then 5-fluorouracil 2400 mg/m2 as a 46-hour infusion, repeated every two weeks. Digoxin was added by slow oral digitalization at 0.125 mg daily for participants over 65 and 0.25 mg daily for those 65 or younger, with a target serum concentration of 0.8 to 1.2 ng/mL. Eligibility required pathologically confirmed resectable disease without distant metastasis, ECOG performance status 0 to 1, and normal potassium, magnesium and calcium.

The primary endpoint was the number of patients able to undergo resection surgery at 16 weeks, with a pre-specified rule stating that the regimen would be considered for further investigation if 14 of 20 patients reached resection. The secondary endpoint counted grade 4 thrombocytopenia and grade 3 to 4 diarrhea against a Pocock stopping boundary calibrated to cross with probability at most 0.05 when the true toxicity rate equals 0.182. The study opened on 12 February 2021, reached primary completion on 13 May 2025, and was updated to terminated status on 21 October 2026, with results posted.

The biological hypothesis

The sponsor described digoxin as a modulator of the hypoxia pathway. Cardiac glycosides inhibit the sodium and potassium transporting ATPase, whose alpha-1 subunit is ATP1A1 (ENSG00000163399); digoxin is CHEMBL1751, a small molecule at maximum phase 4. Preclinical work reports that cardiac glycosides suppress HIF-1 alpha and induce Na/K-ATPase-dependent cell death in tumor cells. Pancreatic ductal adenocarcinoma is hypoxic and stroma-dense, so a cheap oral agent that blunts the hypoxic response alongside chemotherapy is an appealing idea.

The evidence base behind that idea is thin where it counts. Open Targets scores the ATP1A1 association with pancreatic adenocarcinoma at 0.0037, and that score is carried entirely by the literature datatype at 0.0304. No genetic association, no somatic mutation evidence and no clinical evidence contribute. Open Targets lists six ATP1A1-directed agents: digoxin, digitoxin, lanatoside C, deslanoside, acetyldigitoxin and istaroxime. All are cardiac agents and none reached approval in oncology. The target was being pursued in cancer on the strength of pharmacology papers rather than human validation of any kind.

What actually happened

Eleven participants enrolled against a plan of 20. Four completed the study and seven did not. Seven participants form the analysis population, and four of those seven underwent resection. The pre-specified bar was 14 of 20.

Safety was unremarkable for this regimen. All seven analyzed participants had an adverse event of any grade, one had a serious adverse event, and there were no deaths. The serious events were lymphocyte count decreased, blood bilirubin increased, sepsis, an obstructed biliary stent and cholangitis, each in one participant. Grade 3 to 4 diarrhea occurred in one participant and grade 4 thrombocytopenia in none, so the Pocock toxicity boundary was not the issue. The most common events were anemia in seven, abdominal pain in five and alkaline phosphatase increased in four. The posted reason for stopping reads: "Due to low accrual and futility."

Failure mechanism, best guess

Two problems compounded, and only one is about digoxin.

The first is arithmetic. The go criterion required 14 resections. Enrollment closed at 11. Once accrual stopped there, the criterion was unreachable even if every enrolled participant had been resected, because 11 is less than 14. A futility call at that point describes the design, not the drug, and any read of this trial as evidence against digoxin in pancreatic cancer takes more than the data carry.

The second is an exposure ceiling, and this is inference rather than measurement. The protocol targeted 0.8 to 1.2 ng/mL, which is the therapeutic range for cardiac indications and is set by digoxin's narrow margin between effect and toxicity. The anti-tumor concentrations reported for cardiac glycosides in the cell biology literature are generally higher than that range delivers in plasma. If the gap holds in tumor tissue, the trial tested a dose chosen by cardiac safety rather than one chosen to engage the target, and a null result would say nothing about the hypothesis. No pharmacodynamic measure of HIF-1 alpha or any hypoxia marker appears among the posted outcome measures, so this cannot be settled from the record. The coded archetype follows the sponsor's stated futility finding, with that caveat attached.

How to prevent this next time

The posted data are a single arm with seven analyzed participants and no comparator, so no Bayesian posterior, power calculation or subgroup analysis can be supported. The useful levers are design-stage and qualitative.

Gate repurposing programs on an exposure-response check before the protocol is written. State the concentration at which the preclinical effect occurs, state the concentration achievable at a tolerated human dose, and require that the second covers the first. When it does not, the program needs a different molecule, route or delivery strategy, not a trial. Adjust for base rates: an association score of 0.0037 with no genetic or clinical evidence, in a class where six agents exist and none is an oncology drug, should raise the bar for what a pilot must show. Build in a pharmacodynamic readout so a negative result distinguishes a wrong target from an unengaged one. Gate on accrual feasibility: a 14-of-20 criterion in resectable pancreatic cancer at a single center needs an accrual rate written into the protocol as a stopping rule.

The single highest leverage change would have been to require, before the first participant was dosed, a measured pharmacodynamic marker of hypoxia-pathway engagement at the cardiac-safe digoxin exposure, with a pre-specified rule to stop if engagement was absent.

What this means for similar programs

ATP1A1 now carries two Claidex failures and a Mechanism Risk Score of 64, in the orange band. The components are phase burden 14.92, archetype severity 15.80, recency 7.15, genetic deficit 14.94 and saturation 11.30. The genetic deficit term sits near its ceiling because the target-disease link has essentially no human evidence, and the earlier Claidex entry for AMAG-423 in severe preeclampsia carried the same structural weakness. ATP1A1 attracts repurposing attempts in diseases where the mechanism is plausible and the human genetics are silent. For context on the agent, openFDA holds 71,339 FAERS reports naming digoxin, of which 59,064 are serious and 6,826 carry a death outcome. Those counts describe decades of cardiac use in an elderly population and say nothing about this trial, but they do describe the therapeutic window any oncology repurposing has to work inside.

Open questions

Were serum digoxin levels in the 0.8 to 1.2 ng/mL target range, and were any published? Of the seven participants who did not complete, how many stopped for progression rather than toxicity or withdrawal? Is there a tumor concentration of any cardiac glycoside that suppresses HIF-1 alpha at a cardiac-tolerable systemic dose?

Sources

1. • ClinicalTrials.gov, NCT04141995, "A Phase IIa (Pilot) Study of Neoadjuvant Chemotherapy With Folinic Acid, 5-FU, Irinotecan and Oxaliplatin (FOLFIRINOX) With Digoxin in Patients With Resectable Pancreatic Cancer", protocol and posted results retrieved 27 October 2026. https://clinicaltrials.gov/study/NCT04141995 - Open Targets Platform v4 GraphQL API, ATP1A1 (ENSG00000163399) and pancreatic adenocarcinoma (MONDO_0006047), association scores and known-drug list, retrieved 27 October 2026. https://platform.opentargets.org/target/ENSG00000163399 - ChEMBL, digoxin, CHEMBL1751, retrieved 27 October 2026. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL1751/ - openFDA drug event endpoint, digoxin report counts (total, serious, death outcome) and reaction term frequencies, queried 27 October 2026. https://api.fda.gov/drug/event.json - Cardiac glycosides inhibit cancer through Na/K-ATPase-dependent cell death induction. Biochem Pharmacol.

2. - Digoxin and its Na+/K+-ATPase-targeted actions on cardiovascular diseases and cancer. Bioorg Med Chem.

3. - Na+/K+-ATPase-Targeted Cytotoxicity of (+)-Digoxin and Several Semisynthetic Derivatives. J Nat Prod.

4. - Claidex failure graph, prior ATP1A1 entry: amag-423-digoxin-immune-fab-severe-preeclampsia-dsmb-futility (24 May 2026).