Camupepimut’s Phase 2 Trial Missed Its Primary Endpoint—What the Encouraging Subgroups Really Mean

The randomized phase 2 AMPLIFY-7P trial did not show a statistically significant disease-free-survival benefit for camupepimut in its prespecified intention-to-treat primary analysis. Among 144 participants randomized 2:1, median disease-free survival was 23.8 months with camupepimut and 21.2 months with observation (hazard ratio 0.85; P=0.5329). The later subgroup, adjusted, ctDNA, and immune-response findings are useful signals for designing another trial, but they do not reverse that primary result.[^paper]

The randomized primary analysis is the central result. Biomarker, subgroup, adjusted, and immunogenicity findings can sharpen the next hypothesis, but they answer different questions. Original conceptual illustration; not a reproduction of trial data.

Quick answer: AMPLIFY-7P produced a negative primary result and several encouraging secondary or exploratory signals. The scientifically defensible reading is “promising enough to test again,” not “the trial was positive after all.”

AMPLIFY-7P at a glance

| Question | What the trial reported | What it means | |---|---|---| | Did the primary endpoint succeed? | ITT median DFS 23.8 vs 21.2 months; HR 0.85; P=0.5329 | No. The prespecified primary endpoint was not met. | | Was there an early 6-month separation? | DFS rate 75.7% vs 61.7%; exploratory 0–6-month HR 0.55; P=0.0557 | A time-limited signal, not a replacement for the full DFS analysis. | | Did adjusted analysis favor camupepimut? | Post-hoc adjusted HR 0.47; P=0.0086 | Hypothesis-generating because the model was selected after the primary result and cannot recreate balance. | | Did prespecified subgroups look favorable? | ctDNA-negative HR 0.51; no-prior-radiation HR 0.59 | Worth confirming, but subgroup analyses were exploratory, unadjusted for multiplicity, and underpowered for interaction effects. | | Was the vaccine immunogenic? | 89/90 evaluable treated participants had an mKRAS-specific T-cell response | Strong evidence of immune engagement, not proof of clinical benefit. | | Is overall survival known? | OS was immature and not reported | No survival conclusion can yet be drawn. Crossover will also complicate later OS interpretation. |

On this page

• What the primary result showed
• Randomized but open-label
• The 6-month signal
• Adjusted and subgroup analyses
• ctDNA and immune findings
• Overall survival and crossover
• Limitations and conflicts
• What would confirm the signal

What did the primary AMPLIFY-7P result show?

In the analysis that mattered most, camupepimut did not significantly improve disease-free survival compared with observation. The trial enrolled adults with stage I–III KRAS-mutated pancreatic ductal adenocarcinoma who had undergone resection and standard therapy and had no radiographic recurrence within six months of completing locoregional treatment.

The 144 randomized participants formed the intention-to-treat population:

• Camupepimut: 96 randomized; 42/96 (43.8%) had a DFS event.
• Observation: 48 randomized; 26/48 (54.2%) had a DFS event.
• Median DFS: 23.8 versus 21.2 months.
• Hazard ratio: 0.85.
• P value: 0.5329.

Disease-free survival counted time from randomization to investigator-assessed recurrence under iRECIST or death from any cause, whichever came first. A hazard ratio below 1 numerically favored camupepimut, but 0.85 was far from the trial’s prespecified success threshold of HR below 0.70 and approximately P<0.10. A numerical direction is not the same thing as a demonstrated effect.

This is also why the median difference should not be translated as “2.6 extra months without disease.” Medians summarize where each estimated survival curve crosses 50%; they are not an individual treatment effect, and the overall time-to-event comparison was not statistically persuasive.

Our guide to statistical significance versus clinical importance explains why the estimate, uncertainty, endpoint hierarchy, and analysis plan must be read together.

How can a trial be randomized and open-label?

Randomization and masking solve different problems. AMPLIFY-7P randomly assigned participants 2:1, with nodal status used as a stratification factor. Site personnel enrolling participants did not have access to the allocation sequence. Randomization helps make the groups comparable on average and supports a causal treatment comparison.

The trial was also open-label, meaning participants and investigators knew who received camupepimut and who was observed. That does not erase randomization. It does mean masking did not protect investigator-assessed recurrence decisions, follow-up behavior, or other judgment-sensitive parts of the trial from knowledge of assignment.

The primary endpoint was investigator-assessed DFS rather than blinded independent central review. Recurrence and death are harder outcomes than a subjective symptom score, but scan timing and interpretation can still involve judgment. “Randomized” therefore remains a major strength; “open-label” remains a limitation. They are not contradictory labels.

Why does the 6-month result not replace the primary result?

The 6-month analysis asks a narrower, time-specific question than the primary analysis. The prespecified 6-month DFS rate was 75.7% with camupepimut and 61.7% with observation, a 14.0-percentage-point difference. An exploratory Cox analysis limited to randomization through six months produced HR 0.55 and P=0.0557.

That early separation is interesting because it coincided with the treatment period. It still does not supersede the full follow-up result for three reasons:

1. Different time window: a 0–6-month comparison does not estimate DFS across the whole observed period.
2. Endpoint hierarchy: the full ITT DFS analysis was the primary test; the 6-month rate was secondary and the reported 0–6-month hazard-ratio analysis was exploratory.
3. Multiplicity: once several time points, endpoints, subgroups, and models are examined, some favorable results can appear by chance. These analyses were not presented as a multiplicity-controlled family capable of rescuing the primary endpoint.

An early effect that weakens later could reflect a real time-varying treatment effect, chance, changing event patterns, or other features of the data. AMPLIFY-7P cannot reliably distinguish those explanations. A future trial can prespecify the time pattern it intends to test.

Exploratory analyses can identify the next questions without replacing the trial’s main answer. Original conceptual illustration; not measured data and not a study figure.

What do the adjusted and subgroup results really mean?

They suggest where a treatment effect might be found, but they do not establish one. The study had meaningful baseline imbalances despite randomization. Nodal status—the stratification factor—was balanced, but the camupepimut group had more baseline ctDNA positivity (20/96, 20.8%, versus 6/48, 12.5%) and more R1 resection margins (25/96, 26.0%, versus 9/48, 18.8%). The observation group had more prior radiation (15/48, 31.3%, versus 23/96, 24.0%).

A post-hoc multivariable Cox model adjusted for treatment and several prognostic factors. It estimated an adjusted HR of 0.47 for camupepimut (P=0.0086). Stage III disease, elevated CA19-9, and baseline ctDNA positivity were associated with worse DFS in the same model.

Adjustment can be informative when a modest randomized trial ends up imbalanced. It cannot guarantee that every relevant prognostic factor was measured, modeled correctly, or chosen independently of the observed results. The covariates may have been clinically motivated, but the model itself was a post-hoc exploratory analysis. It should generate a prospectively testable hypothesis, not overwrite the unadjusted ITT primary analysis.

The subgroup findings require the same discipline:

• ctDNA-negative at baseline: 93/144 participants—62 camupepimut and 31 observation. Events occurred in 17/62 (27.4%) versus 15/31 (48.4%); median DFS was not reached versus 21.2 months; HR 0.51; P=0.0280.
• No prior radiation: 106/144 participants—73 camupepimut and 33 observation. Events occurred in 31/73 (42.5%) versus 20/33 (60.6%); median DFS was 23.8 versus 10.4 months; HR 0.59; P=0.0221.

The subgroup categories were prespecified, but the paper states that subgroup analyses were exploratory, not adjusted for multiplicity, and that the trial was not powered to detect different treatment effects within subgroups. Prespecification reduces cherry-picking; it does not turn a small subgroup into a confirmatory trial. A favorable estimate within one subgroup is especially weak evidence of a predictive marker unless the treatment effect is shown to differ credibly between subgroup levels.

What do the ctDNA results show?

Baseline ctDNA was a strong prognostic marker in AMPLIFY-7P, but the trial did not prove that ctDNA status identifies who benefits from camupepimut. Across both randomized groups, 22/26 ctDNA-positive participants (84.6%) had a DFS event, compared with 32/93 ctDNA-negative participants (34.4%). Median DFS was 4.2 versus 24.0 months (HR 6.56; P<0.000001).

That pooled comparison shows that detectable ctDNA tracked with a much worse prognosis. Prognosis is not the same as treatment prediction. Of 144 randomized participants, 26 were ctDNA-positive and 93 were ctDNA-negative; the remaining 25 were not part of either evaluable baseline-ctDNA subgroup. The ctDNA-positive observation arm contained only 6 participants, compared with 20 in the camupepimut arm. Its nominal subgroup result—median DFS 4.3 versus 1.9 months; HR 0.57; P=0.0489—rests on a very small comparator denominator.

The ctDNA-negative signal is larger and more stable numerically, but it still belongs to an exploratory, multiplicity-unadjusted subgroup analysis. A confirmatory trial would need to define ctDNA assessment, eligibility or stratification, the treatment-by-ctDNA question, and the statistical testing plan in advance.

For the broader distinction between a biomarker and patient benefit, see surrogate endpoints and biomarkers versus clinical outcomes.

What did the immune-response findings prove?

They showed that camupepimut reliably engaged the intended immune target in evaluable treated participants. They did not show that immune response caused longer DFS. Camupepimut combines seven amphiphile-modified mutant-KRAS peptide antigens with an amphiphile-modified immune-stimulatory adjuvant. The lipid-linked design is intended to use albumin trafficking to improve delivery into lymph nodes, where antigen-presenting cells can activate T cells.

Among 90 evaluable camupepimut recipients, 89/90 (98.9%) developed an ex vivo mKRAS-specific T-cell response under the study definition. In addition:

• 74/90 (82.2%) exceeded a 9.17-fold response threshold previously derived from a phase 1 study.
• Among those above that threshold, 26/74 (35.1%) had a DFS event, compared with 13/16 (81.3%) below it; HR 0.22; P<0.0001.
• Antigen spreading was observed in 13/15 selected evaluated participants whose responses exceeded the threshold.

These findings support immunogenicity: the vaccine can induce measurable immune responses. The response-versus-DFS comparison is not randomized. High and low responders may differ in immune fitness, prior therapy, disease burden, time remaining event-free long enough to be measured, or other prognostic features. Immunogenicity was assessed only in the treated arm, so the association cannot establish that the measured response mediated a treatment benefit over observation.

The 13/15 antigen-spreading denominator is narrower still: it was a selected subset of strong responders tested for responses to non-vaccine tumor antigens. It is evidence of a biological phenomenon in that subset, not an 86.7% rate that can be generalized to all 144 randomized participants.

The related PeptideBase appraisal An Immune Response Is Not a Survival Benefit explains why target engagement and clinical efficacy require separate demonstrations.

Why can disease-free survival not be treated as overall survival?

Disease-free survival and overall survival are different outcomes. DFS in AMPLIFY-7P counted recurrence or death. OS counts death. A treatment can alter recurrence timing without producing the same effect on survival, particularly when later therapies and crossover differ.

The paper states that OS data were immature and were not reported. It also says participants assigned to observation could cross over to camupepimut. Crossover may be ethically or scientifically reasonable, but it can make the randomized groups more alike after recurrence and complicate a later intention-to-treat OS comparison.

No current AMPLIFY-7P result therefore establishes that camupepimut helps people live longer. Longer follow-up may inform that question, but crossover means even mature OS data will require careful interpretation rather than a simple headline.

What limitations matter most?

The biggest limitations are the negative primary endpoint, modest sample size, multiple exploratory analyses, short follow-up, and restricted generalizability. The authors identify these issues directly.

• Modest phase 2 sample: 144 participants with 2:1 allocation left only 48 in observation and much smaller control denominators in some subgroups.
• Open-label assessment: investigators knew treatment assignment, and DFS was investigator-assessed.
• Multiplicity: subgroup and exploratory immunogenicity analyses were not adjusted for the many comparisons performed.
• Post-hoc modeling: the adjusted result may address measured imbalance but remains hypothesis-generating.
• Limited follow-up: median follow-up was 19.4 months at the April 20, 2026 cutoff; long-term DFS and OS remained uncertain.
• Crossover: later OS comparisons will be harder to interpret.
• Limited diversity: 81.3% of participants were White, 7.3% Asian, 4.2% Black or African American, and 6.3% did not report race. Black participants were underrepresented relative to the burden of pancreatic cancer.
• Geographic scope: the trial ran at 24 US centers, which limits generalization to other health systems and populations.
• Treatment-arm-only immune correlations: response-versus-outcome findings cannot separate a vaccine-mediated effect from differences between people who mounted stronger or weaker responses.

The study was sponsored, funded, and designed by Elicio Therapeutics in collaboration with academic authors. Several authors were Elicio employees who received salary and stock benefits; some held pending amphiphile-vaccine patents. Multiple academic investigators disclosed research funding, consulting, advisory roles, stock ownership, or other relationships involving Elicio or other companies. These disclosures do not invalidate the trial, but they make independent replication, prespecified analyses, and transparent endpoint hierarchy especially important.[^paper]

What evidence would change the conclusion?

A larger confirmatory phase 3 trial would need to prospectively test the population, endpoint, and biomarker strategy suggested by AMPLIFY-7P. The strongest design would preserve randomization, control multiplicity, predefine how ctDNA and prior radiation are handled, and use an assessment process that reduces open-label outcome bias.

Useful confirmatory features would include:

• a clearly powered primary clinical endpoint with its success threshold fixed in advance;
• prespecified stratification for strong prognostic variables, including baseline ctDNA where justified;
• a formal treatment-by-biomarker interaction plan rather than separate nominal subgroup P values;
• multiplicity control across key secondary endpoints and subgroups;
• independent or blinded recurrence assessment where feasible;
• adequate follow-up for long-term DFS and OS;
• transparent handling of crossover and subsequent therapy;
• broader racial, geographic, and clinical representation;
• immune analyses that test mediation without treating post-randomization response groups as randomized groups.

Before publication, protocol and registry checks also matter. PeptideBase’s guide to auditing a peptide trial registry explains how endpoint history and prespecification should be checked against the final paper.

Bottom line

AMPLIFY-7P did not meet its primary endpoint. Camupepimut generated strong mKRAS-specific immune responses and produced several favorable early, adjusted, and subgroup signals. Those findings justify further research; they do not establish clinical benefit, identify a proven responder population, or show an overall-survival advantage.

The evidence-literate conclusion is narrow but useful: the biological signal is credible, the clinical signal is unresolved, and a prospectively designed phase 3 trial—not reinterpretation of a negative phase 2 primary analysis—is the proper test.

Source and scope note

This article is based on the open full text of George et al., “Lymph node-targeted, mKRAS-specific amphiphile vaccine in pancreatic cancer: the randomized phase 2 AMPLIFY-7P trial,” Nature Medicine (2026), DOI 10.1038/s41591-026-04734-7, and the linked NCT05726864 trial record.[^paper][^registry] It is an evergreen evidence-literacy article, not a news report.

This article is for general education. It does not provide medical advice, treatment recommendations, dosing or administration instructions, or purchasing or sourcing guidance.

[^paper]: George B, et al. Lymph node-targeted, mKRAS-specific amphiphile vaccine in pancreatic cancer: the randomized phase 2 AMPLIFY-7P trial. Nature Medicine. 2026. doi:10.1038/s41591-026-04734-7. [^registry]: ClinicalTrials.gov. NCT05726864: A Study of ELI-002 7P in Subjects With Solid Tumors.