The $35 Billion Press Release

A Company Gained $35 Billion Before Lunch

At 6:45 a.m. Eastern on August 19, 2026, Merck and Moderna put out a press release announcing that their individualized cancer vaccine had succeeded in a Phase 3 trial.¹ By 11 a.m., Moderna's stock was up as much as 159% from the previous close, the largest single-day gain in the company's history. The previous record was 27.8%, set in February 2020.²⁰

More than 74 million shares traded in the first hour. The ten-day average for an entire session was under six million. Merck, a company fourteen times larger, rose about 10.6% — its best day since March 2009. Between them, the two companies added roughly $70 billion in market value in a morning.^18,20

I want to be precise about what caused this, because the gap between the cause and the effect is the most interesting thing in medicine this week.

The drug is intismeran autogene, previously known as mRNA-4157 or V940. It is an individualized neoantigen therapy — a personalized cancer vaccine, though Moderna has quietly stopped using that word.¹⁹ After a patient's melanoma is surgically removed, the tumor is sequenced alongside the patient's healthy DNA. An algorithm identifies mutations unique to that person's cancer. Those mutations are encoded into a single strand of mRNA — up to 34 of them, strung together in one reading frame — wrapped in a lipid nanoparticle, and injected into the patient's arm. The immune system reads the instructions, manufactures the mutant proteins, and learns to hunt anything still carrying them.⁵

It is a genuinely beautiful idea. It has been the field's great hope for fifteen years. And on Wednesday morning it produced, for the first time, a positive Phase 3 trial.

Here is the problem. The press release contains no results.

Eleven Words of Data

I read both the Merck and the Moderna releases line by line, looking for a number. This is the entire efficacy disclosure, verbatim:

That is it. RFS is recurrence-free survival — how long patients go before their cancer comes back. DMFS is distant metastasis-free survival — how long before it spreads to a new organ. Both improved. We are not told by how much.¹

Absent from both releases: the hazard ratio for RFS. The hazard ratio for DMFS. Any p-value. Any confidence interval. Absolute recurrence rates in either arm at any timepoint. Median follow-up. The number of recurrences, metastases, or deaths that had occurred. How many patients were in each arm. Any subgroup result. Any overall survival figure. Any quantitative safety data — no Grade 3+ rate, no discontinuation rate, no rate of injection-site reactions.

The safety disclosure is one sentence: the profiles were "consistent with those observed in previously reported studies," with "no new safety signals observed."¹

The press release contains exactly two numbers about the trial — 1,137 patients enrolled, randomized 2:1 — and neither of them is a result.

To be fair to Merck, this is a house convention, not a cover-up. The company has long insisted that trial detail be held for formal presentation and peer-reviewed publication, and STAT's reporters said as much on the day.¹⁹ There is a defensible argument for it: dribbling out effect sizes in a press release before a journal has looked at them is its own kind of bad practice. NBC News put the situation in its subheadline — the companies "have not published their results in a peer-reviewed journal, nor did they release the new data."

But there is a trap here, and most of the coverage fell into it. Both press releases do contain hazard ratios — 0.51 for recurrence-free survival, 0.411 for distant metastasis-free survival — sitting in the paragraph directly above the Phase 3 description. Those numbers are not from the Phase 3. They are five-year results from a 157-patient open-label Phase 2b that finished years ago.^4,6 If you have seen "a 49% reduction in recurrence risk" attached to this week's news, you have seen the wrong trial's data.

A 157-Patient Trial With a Very Forgiving Bar

Everything anyone actually knows about how well this drug works comes from KEYNOTE-942, published in The Lancet in 2024 with a five-year update presented at ASCO this past June.^4,6 It is a real trial and the results are real. It is also small, open-label, and was powered against a statistical threshold that would not pass muster in most Phase 3 settings.

There is a second problem with KEYNOTE-942 that I have seen almost nobody discuss, and it is in the paper's own methods section. COVID-era manufacturing disruptions meant that nine patients randomized to the vaccine arm were manually reallocated to the control arm, and a further 37 patients were assigned to the combination arm without concurrent randomization. That is 46 of 157 patients — 29% of the trial — not randomized in the normal sense.⁴ The authors ran a sensitivity analysis restricted to patients enrolled before February 2021; the hazard ratio drifted toward the null, to 0.612 (0.310–1.206).

Baseline characteristics also tilted the vaccine's way. Tumor mutational burden of at least 10 mutations per megabase: 74% in the combination arm versus 60% in the control arm. PD-L1 positive: 64% versus 54%.⁴ Both are favorable prognostic features, and both landed disproportionately in the arm that won.

The authors were candid about all of it. Their own concluding limitation reads: "this is a moderately sized phase 2b study designed with one-sided α of 0·1… Both a longer follow-up and a larger phase 3 study are needed to make more definitive conclusions."⁴

That larger Phase 3 study is the one that just read out. It fixes nearly every one of these problems — it is triple-blinded, placebo-controlled, 1,137 patients across 165 sites in 26 countries.² It is a serious piece of work. We simply have not been shown what it found.

Recurrence-Free Is Not the Same as Alive

This is the part that matters most, and it is the part that a 159% stock move is least equipped to think about.

The trial's primary endpoint is recurrence-free survival: the time until the cancer comes back or the patient dies of any cause. It is not overall survival. No overall survival benefit has ever been demonstrated for this drug. In the Phase 2b, overall survival was labeled exploratory throughout and never powered; at five years the hazard ratio was 0.471 with a confidence interval of 0.165 to 1.345, resting on seven deaths in one arm and seven in the other.⁶ Fourteen deaths is not an overall survival result. It is a rounding error with a hazard ratio attached.

Moderna's president Stephen Hoge told analysts in May that "RFS really is survival in this case."¹⁹ I understand why he would want that to be true. The surrogacy literature does not support it.

The case for RFS as a stand-in for survival in adjuvant melanoma rests on a 2018 analysis of 5,826 patients across eleven interferon trials, which reported an R² of 0.91.⁹ That number is quoted constantly. What is quoted less often is the authors' own caveat: one outlier trial meant that "even if rho was high, R² was low and could not reliably be estimated." The 0.91 is what you get after removing the trial that disagreed.

The more recent work is considerably less encouraging. A 2023 analysis of fifteen randomized trials and 13,715 patients found a trial-level R² of 0.63 overall, and 0.67 for melanoma specifically — below the authors' own threshold for a strong association. Their conclusion: the meta-analysis "failed to find a clinically strong association between RFS and OS," and the findings "challenge the use of RFS as the primary efficacy endpoint."⁸

The mechanism behind the decoupling is important, and it is the opposite of bad news. Surrogacy holds up when patients who recur have nowhere to go. It breaks down when they have good salvage options — the same 2023 paper found the correlation strengthens substantially only when post-recurrence immunotherapy use falls below 20%. Metastatic melanoma now has some of the most effective salvage therapy in all of oncology. STAT put it plainly on the day: melanoma drugs "are now so effective that patients who progress in clinical trials often have another treatment option — meaning people in control groups live longer on existing therapies alone."¹⁹

This is not hypothetical. It has happened repeatedly, in this exact disease:

And the sharpest fact of all, which I would like every reader to sit with for a moment: adjuvant pembrolizumab has been FDA-approved for stage III melanoma since February 2019, on an investigator-assessed recurrence-free survival endpoint. As of the seven-year report published in 2024, KEYNOTE-054 has still never performed its overall survival analysis — 286 deaths had occurred against the 380 required to trigger it.¹²

The control arm in INTerpath-001 is a full year of that same pembrolizumab. Which means intismeran is not being tested against nothing. It is being added on top of a drug that is itself very good, in a population that has already had the tumor cut out, most of whom would never have recurred at all.

In fairness — and this is a real counter-argument, not a courtesy — a 2026 analysis of 10,379 patients found that distant metastasis-free survival correlates well with melanoma-specific survival (R² 0.87–0.91), even while correlating weakly with overall survival at every horizon (R² below 0.40).¹³ Those authors argue overall survival is the flawed measure here, diluted by patients dying of other things. That view deserves a hearing. INTerpath-001 hit its DMFS endpoint too.

27,141 Patients and One Approval

Context matters for calibration, and the history of therapeutic cancer vaccines is brutal in a way that is easy to forget on a day when a stock doubles.

A 2015 analysis counted 27,141 patients enrolled in Phase 3 cancer vaccine trials. Oncology vaccines moved from Phase 2 to Phase 3 at a healthy 39.5% rate — and then converted Phase 3 into an approval filing only 8.3% of the time, against 45.2% for oncology overall. The authors' conclusion: "The bottleneck for oncology vaccine progression is late in development."¹⁷

The list of large, well-run, thoroughly negative trials is long. Tecemotide in lung cancer: 1,513 patients, overall survival HR 0.88, p=0.123. MAGE-A3 in non-small-cell lung cancer: 2,312 patients, disease-free survival HR 1.02. MAGE-A3 in melanoma: 1,391 patients, HR 1.01. Rindopepimut in glioblastoma: 745 patients, overall survival HR 1.01, after an earlier study had shown an 85% antibody response rate. In exactly one case — sipuleucel-T in prostate cancer, 512 patients — a therapeutic cancer vaccine produced a significant survival benefit and won approval. Its maker filed for bankruptcy four and a half years later.¹⁷

Two contemporary data points are more directly relevant, and both are recent.

That last point is the one I keep returning to. 84% immunogenicity, and a p-value of 0.31. The immune system did exactly what it was asked to do. The cancer did not care.

And then there is IO Biotech, which is not an individualized therapy — it is an off-the-shelf two-peptide vaccine, a different animal — but whose regulatory experience is instructive. Its Phase 3 in first-line advanced melanoma enrolled 407 patients and produced a median PFS of 19.4 versus 11.0 months, HR 0.77, p=0.0558, against a pre-specified threshold of 0.045. In September 2025, after a pre-BLA meeting, the FDA recommended the company not submit. It filed for Chapter 7 bankruptcy six months later.¹⁶

A 407-patient melanoma trial with an 8.4-month numerical PFS gain and clean safety, and the agency said don't bother filing. That was five months ago.

Worth knowing as well: per reporting from BioSpace and Fierce Biotech, the FDA gave Merck and Moderna discouraging feedback in 2024 when they explored accelerated approval on the strength of the Phase 2b.¹⁶ The regulator has already declined this evidence once at an earlier maturity.

You Cannot Manufacture a Blockbuster One Patient at a Time

Every dose of this drug is a bespoke product. Sequence the tumor, sequence the germline, type the HLA, run the neoantigen selection algorithm, design a concatemer, synthesize the mRNA, encapsulate it, test it, release it, ship it. For one person. Then do it again for the next person.

It works: in the Phase 2b, the median patient received all 34 encoded neoantigens, and 91% of combination-arm patients got the full complement.⁵ Of 224 patients screened, 15 failed for inadequate tumor tissue and 10 for missing the enrollment window.⁵ Merck's Jane Healy told STAT the trial ran about six weeks "needle-to-needle."¹⁹

I want to flag something about that timeline, because it is cleverer than it looks. Both trials require surgery within thirteen weeks of the first pembrolizumab dose, and patients start pembrolizumab immediately while their vaccine is being built. In the Phase 2b, 81% of patients received their first vaccine dose at pembrolizumab cycle three.⁴ The manufacturing delay is absorbed by the checkpoint inhibitor runway. Which is elegant — and also means the platform has never once had to hit a hard deadline.

What has never been publicly disclosed, in any filing, presentation, or release: the manufacturing capacity of the Marlborough facility in patients per year. Nor the cold-chain requirements. Nor the price. Bancel told CNBC that pricing was undecided because the data had only arrived days earlier. For scale, sipuleucel-T launched at roughly $93,000 per course.

And whatever the number turns out to be, it is incremental — intismeran is an add-on to a full year of pembrolizumab, not a replacement for it, given to people who are cancer-free at the time of treatment.

A Real Result, Priced as a Certainty

I want to separate two claims that got welded together on Wednesday.

The first: an individualized neoantigen therapy succeeded in a large, rigorous, placebo-controlled Phase 3 trial. That is true, it is a first, and it matters. Jedd Wolchok at Weill Cornell — an immunologist with no stake in either company — told STAT it was "really an important achievement," something that "might have been considered science fiction 15 or 20 years ago."¹⁹ Catherine Wu at Dana-Farber called it "certainly promising," while explicitly noting that "we need to await the final results."¹⁹ Both reactions seem right to me.

The second claim: we now know this drug works well enough to justify adding $70 billion of combined market value. That claim is not supported, because the quantity in question has not been disclosed. Two days before the readout, Citi told clients that a hazard ratio at or below 0.72 would count as positive and 0.65 as a clear win. Jefferies had set a range of 0.5 to 0.8. We cannot say which of those bars was cleared, because no hazard ratio exists in public.²¹

It is also worth being honest about what else moved the stock. Short interest stood at 13.5% of free float, and at the intraday high, short sellers faced roughly $4.8 billion in mark-to-market losses.¹⁸ Going into the morning, the average analyst price target was $52.95 and the single most bullish target on the entire street was $110. The stock traded above $158. Not one of the six analyst actions that followed was an upgrade to the highest tier; they were target raises chasing a price that had already left. As one market-structure analyst told Reuters, there is "clearly a real catalyst behind a move this size," but "shorts scrambling to cover are adding fuel."¹⁸ A short squeeze is not a clinical finding.