The Number Moderna Didn't Print: mRNA-4157, the Confidence Interval, and the Ledger Problem Personalized Medicine Can't Solve

Last week a headline moved a stock 25%: a personalized mRNA cancer vaccine cut the risk of recurrence or death by 42%. The number that actually matters was never printed. The hazard ratio behind that 42% was 0.561, with a 95% confidence interval running from 0.309 to 1.017. The upper bound crosses one. The nominal p-value sits at roughly 0.05 — the edge, not the middle. And somewhere in the same coverage cycle, a stock that had fallen by half was reported as up 530%. Liquidity is a liar, but attention markets lie faster. intismeran autogene — mRNA-4157, V940 — is the first personalized neoantigen therapy to show a recurrence-free survival benefit in a randomized controlled trial. The protocol: sequence each patient's tumor with whole-exome and RNA sequencing, run an algorithm to select up to 34 neoantigens, encode them into a single mRNA, deliver in lipid nanoparticles, dose every three weeks for up to nine doses, alongside roughly a year of pembrolizumab, in completely resected high-risk stage IIIB/IV melanoma. The trial is KEYNOTE-942 — 157 patients, 2:1 randomization, open label, active comparator rather than placebo. It is Phase IIb, not Phase III. The confirmatory trial, INTerpath-001, is still enrolling and reads out around 2026–2027. FDA granted breakthrough designation in February 2023. Merck and Moderna split costs and profits 50/50, a structure expanded in October 2022 with a $250 million payment from Merck. That is the spec sheet. Now read it as a systems engineer, not an oncologist. Every element of that spec is a systems problem, and most of them are ledger problems wearing clinical costumes. Start with data custody. A per-patient vaccine requires the tumor's genomic sequence to travel from a hospital pathology lab to a sequencing vendor to an algorithm to a manufacturing site, and then the identity of that sequence has to stay bound to the right patient across nine doses spanning months. That is chain-of-custody, not chemistry. It is also the exact problem the DeSci crowd has spent three years promising to solve with on-chain provenance. My read, after sitting through enough of these pitches: the institutions that actually hold sequencing data do not need a public chain to prove custody. They need an auditable private registry and an enforceable contract. RWA-on-chain in biotech has been a three-year storytelling exercise because the buyer was never the problem the sellers imagined. Second, scheduling. The reported "six weeks" is not six weeks per dose — it is turnaround from tumor sampling to first dose available. Read literally, nine doses at six weeks each is 54 weeks, which is clinically absurd and would make the product impossible. Even at six weeks turnaround, a nine-dose course requires months of coordinated cold-chain delivery, appointment-locked administration, and synchronization with pembrolizumab's cycle. Merck's CEO has publicly said he believes the waiting time can be shortened — an admission that it is currently a bottleneck, not a footnote. This is distributed resource allocation under hard deadlines, and it is where the margin leaks. Third, payment. Personalized therapies have no established CPT or J-code. Without a billing code there is no reimbursement rail, and without a rail the product exists but does not clear. An oracle problem in a lab coat. Watch the flow, not the flood. Patient flow, here: roughly 100,000 new melanoma cases a year in the US, 15–20% stage III or higher, of which perhaps 12,000–18,000 are resected high-risk and adjuvant-eligible. Globally, 30,000–40,000. At 30–50% penetration and $120,000–150,000 per course stacked on pembrolizumab, the melanoma indication peaks around $2–3.5 billion. The real number is NSCLC: INTerpath-002 enrolls roughly 1,200 patients, and if it works, resectable high-risk NSCLC adds 100,000-plus patients a year globally — total peak $6–10 billion. Then halve all of it. Moderna keeps 50%. And note the clock: pembrolizumab faces a patent cliff around 2028, which is precisely why Merck is funding a combination that could extend the franchise into an IO-plus-vaccine moat. Here is the angle the coverage missed. Personalized medicine is usually sold as the ultimate DeSci use case — tokenize the data, decentralize the trial, let patients own their sequences. The mRNA-4157 structure argues the opposite. Each batch is n=1. There is no fungible unit to securitize, no pool to underwrite, no secondary market in a vaccine only one person can receive. The more personal the medicine, the less tokenizable it becomes. That is the decoupling thesis nobody wants to hear: the biotech-tokenization narrative and the actual commercial trajectory of personalized therapeutics are moving in opposite directions, and the market is pricing them as if they were the same trade. Meanwhile the sector's graveyard — Canvaxin, Oncophage, Stimuvax, multiple Phase III failures — explains why a 25% drawdown follows a statistically marginal readout. Mechanism-first does not mean clinically confirmed. Code is law until it isn't. There is no regulatory code yet for per-patient CMC — no batch-release logic, no stability standard, no platform-designation precedent for a product where every lot has a sample size of one. FDA has the breakthrough designation; it does not have the framework. Regulation chases shadows, and here the shadow is the manufacturing protocol. The clinical question — does the vaccine work — resolves in 2026 or 2027. The commercial question resolves sooner, and it is not a biology question. If the six-week turnaround becomes four, and a billing code lands, the NSCLC readout is a $6–10 billion option. If not, melanoma stays a $2–3.5 billion proof of concept with a confidence interval that still crosses one. Which of those two clocks is the market actually pricing?