The FDA extended sevabertinib's accelerated approval into untreated HER2 (ERBB2) TKD-mutant non-squamous NSCLC on an open-label, single-arm cohort of 69 patients, ORR 75% (95% CI: 64, 85) (FDA, 9 Sep 2026)

Why it matters (spine). The health lane tracks two frontiers at different speeds: what biology can do, and what a regulator will accept as proof it did. This is the second kind, and it is a small, specific movement. Sevabertinib already held accelerated approval in this mutation for patients who had received prior systemic therapy. What the agency did on 9 September was extend it to patients who have received nothing — and it did so on an open-label, single-arm cohort of 69, measured by response rate rather than survival.

The reason that is worth a card, and the reason the framing is carefully hedged below: single-arm evidence is the normal and largely uncontested currency of accelerated approval in the later lines, where patients have exhausted options and randomising them to a comparator is difficult to justify. The first line is where a comparator exists, is standard, and is ethically available. Accepting a single-arm result there is a different act from accepting one in the third line, even when the statistics look identical. This card's claim is that the agency did it and published the numbers; the claim that it represents a meaningful loosening is this garden's reading, labelled as a reading, and the FDA document asserts nothing of the kind.

What the announcement says, quoted. Published by the Center for Drug Evaluation and Research, headlined "FDA grants accelerated approval to sevabertinib for locally advanced or metastatic non-squamous non-small cell lung cancer," with article:published_time: Wed, 09/09/2026 - 11:34 in the page metadata and "Content current as of: 09/09/2026" at the foot.

The action: "On September 9, 2026, the Food and Drug Administration granted accelerated approval to sevabertinib (Hyrnuo, Bayer Healthcare Pharmaceuticals Inc.), a kinase inhibitor, for adult patients with locally advanced or metastatic non-squamous non-small cell lung cancer (NSCLC) whose tumors have HER2 (ERBB2) tyrosine kinase domain (TKD) activating mutations, as detected by an FDA-authorized test."

What is new about it, in the agency's own words: "This approval expands the existing indication for sevabertinib, which was previously granted accelerated approval for adult patients with locally advanced or metastatic non-squamous NSCLC with HER2 (ERBB2) TKD activating mutations who had received prior systemic therapy."

The evidence: "Efficacy was evaluated in patients with locally advanced or metastatic, non-squamous NSCLC with HER2 (ERBB2) TKD activating mutations who received sevabertinib in SOHO-01 (NCT05099172), an open-label, single-arm, multicenter, multi-cohort clinical trial. HER2 (ERBB2) activating mutations were determined in tumor tissue or plasma by local laboratories prior to enrollment." · "The major efficacy outcome measures were confirmed objective response rate (ORR) and duration of response (DOR) as assessed by blinded independent central review (BICR) using RECIST v1.1." · "Among 69 patients with locally advanced or metastatic NSCLC with HER2 (ERBB2) TKD activating mutations who had not received prior systemic therapy, the ORR was 75% (95% CI: 64, 85); 73% of responding patients had a DOR of ≥6 months and 38% had a DOR of ≥12 months."

Safety, dose and pathway: "The prescribing information includes warnings and precautions for diarrhea, hepatotoxicity, interstitial lung disease (ILD)/pneumonitis, left ventricular dysfunction, ocular toxicity, pancreatic enzyme elevation, and embryo-fetal toxicity." · "The recommended sevabertinib dose is 20 mg orally twice daily with food, until disease progression or unacceptable toxicity." · "This application was granted priority review. Sevabertinib received breakthrough therapy designation and orphan drug designation."

The reading discipline on the duration figures. "73% of responding patients had a DOR of ≥6 months" is a proportion of responders, not of the 69. The card states it that way everywhere, and no combined figure is computed — an arithmetic product of 75% and 73% would be a derived quantity the document does not support as a duration estimate, and none is offered. Note also that the page publishes no median duration of response and no survival figure of any kind, which is the honest limit of what this approval demonstrates: tumours shrank in three quarters of patients, and the page says nothing about how long anyone lived.

What the page does not say, and this card therefore does not. It does not state that a confirmatory trial is a condition of this approval. The cabinet's fda-accelerated-approval-tudriqev-melanoma card records the agency stating that condition explicitly on that drug's page; this page contains no equivalent sentence, and the absence is recorded here rather than filled in from what accelerated approval generally requires. It does not publish the prescribing information — "Full prescribing information for Hyrnuo will be posted on Drugs@FDA" — and the label was not opened, so every safety item above is the announcement's summary of the label and is attributed as such (L66). It does not give the size of the whole SOHO-01 trial, only the 69-patient treatment-naive cohort. It does not name a comparator, because there was none.

The international leg, which is genuinely dated and open. "This review was conducted under Project Orbis... For this review, FDA collaborated with the United Kingdom's Medicines and Healthcare products Regulatory Agency (MHRA). The application reviews are ongoing at the other regulatory agencies." That last clause is a live, checkable state: as of the document's own date, no other regulator had acted. It is one of the falsifier limbs below.

A cabinet-scoped correction, on the record. A draft of this card called it "the cabinet's first Project Orbis card." It is not. A grep for orbis across content/signals/ returns fda-approves-daraxonrasib-pancreatic-adenocarcinoma, which records that approval's "review under Project Orbis in collaboration with Health Canada, with the European Medicines Agency and Japan's medicines agency as 'official observers'". So this is the cabinet's second Orbis card, and the pair is more useful than either alone: two Orbis reviews fourteen days apart, one with three named partner or observer agencies and one with a single collaborating agency (the MHRA), both stating that partner reviews remained open at the time of the US decision.

Derived figures, enumerated. Three, and they are the only arithmetic on this card. The width of the confidence interval on the response rate, 21 percentage points (85 − 64), which is what a 69-patient cohort buys. The drop-off between the two duration thresholds, 35 percentage points of responders (73 − 38) between six and twelve months. And the interval between this approval and the cabinet's other Project Orbis card, fourteen days (26 August to 9 September 2026, each read off its own FDA page). Every other number above is read off the page as printed. No figure is combined across the cohort and the responder subgroup.

Money angle. Narrow and worth stating narrowly. A first-line indication is a substantially larger commercial prize than a second-line one in the same mutation, because every eligible patient passes through the first line and only some reach the second. The agency's own sentence marks exactly that transition — "who had received prior systemic therapy" removed — so the commercial step is visible on the face of a government document rather than inferred from a company forecast. This card prints no market size, no revenue estimate and no epidemiology for HER2 TKD-mutant NSCLC, because the FDA page contains none and no other document was opened. The generalisable point for HKG is the one the cabinet keeps re-learning: the regulatory pathway is often the economically decisive variable, and it is documented in public while the forecasts are not.

Meta-crisis lever. Human health, directly, and with the honest qualifier this cabinet owes: an oral targeted therapy that shrinks tumours in 75% of previously untreated patients with a defined mutation is a real improvement in options for a small, genomically selected population. It is not a survival result, and nothing here speaks to access, price, or the NSCLC that does not carry this mutation — whose share this card does not state, because the FDA page gives no epidemiology and no other document was opened.

Cross-links, quoted before they are leaned on. fda-approves-daraxonrasib-pancreatic-adenocarcinoma (verified, 26 August 2026) records "median overall survival 13.2 months versus 6.7, hazard ratio 0.40 (0.30, 0.53), p under 0.0001" from a 500-patient randomised trial, and its own action line already frames the comparison this card extends: "Any HKG claim citing an FDA approval should carry which of those two it is, because the words 'FDA approved' cover both." fda-accelerated-approval-tudriqev-melanoma (verified, 6 August 2026) is the closer analogue on evidence — "140 enrolled, 91 efficacy-evaluable, objective response rate 24.2%" — and the dissimilar one on setting, since that indication is for patients "who progressed on a PD-1-blocking regimen," i.e. the later line where single-arm evidence is least contested. That contrast is the whole point of this card, and it runs against the new card's novelty in one respect worth stating: Tudriqev's response rate was 24.2% and this one is 75%, so the agency is accepting single-arm evidence here on a far larger effect. A large effect is a defensible reason to accept an uncontrolled design, and this card does not pretend otherwise. fda-approves-zilganersen-alexander-disease (verified, 3 September 2026) is the nearest FDA card in time and the one where the agency wrote its evidentiary reasoning into the announcement itself. Quoted from that card's own body, which quotes the FDA: "For patients under 2 years of age, direct clinical trial data were limited by the rarity of the disease and lack of concurrent control. Pharmacokinetic modeling confirmed that drug levels in this age group are expected to be similar to those seen in older children at the same dose. This observation was also supported by safety data from four patients under 2 years of age treated in the study... This approach supported extending the indication to the youngest patients while maintaining the evidentiary standards required to support approval." That is the counter-example that shows the FDA does sometimes explain itself on the record, which is why this page's silence on a confirmatory requirement is recorded above as an absence rather than assumed away. Note what the comparison does not establish: the agency explained itself where a trial was impossible, and the setting here is one where a comparator exists — so the absence of an explanation is a difference between two documents, not evidence about what was decided internally.

What would prove it wrong (falsifier). Four limbs; the first is a provenance re-check and is labelled as one (L83), and three can fail with the primary unchanged.

• The facts — provenance re-check, not a test of the claim — fail if the FDA page at the cited URL does not carry the 9 September 2026 date, the expansion sentence naming the removed prior-therapy requirement, the SOHO-01 trial identifier, or the 69-patient / 75% / (64, 85) / 73% / 38% figures.
• The framing — that this is a loosening of the evidentiary bar in the first line fails if the approved label or the FDA review documents on Drugs@FDA show a randomised confirmatory trial already reported, already enrolling under a binding postmarketing requirement, or a specific obligation attached to this expansion. Neither document was opened. This is the limb most likely to fail, and it is named first among the substantive limbs deliberately: the garden's reading is the least-supported thing on this card.
• The durability fails if this accelerated approval is withdrawn, or if the confirmatory evidence — whenever it arrives — does not show clinical benefit. Accelerated approval means the benefit is unconfirmed by construction; the cabinet's fda-accelerated-approval-tudriqev-melanoma card makes the same point and the FDA maintains a public "Withdrawn | Cancer Accelerated Approvals" index, linked from this very page, which is the observable.
• The Project Orbis leg fails, in the direction that would make this a smaller story, if the MHRA or another Orbis partner declines or materially narrows the indication. The page states the reviews were "ongoing" on 9 September; an MHRA decision is the check, and a divergence between regulators on identical evidence would be the more interesting finding.

Provenance (what I actually read). One document, fetched from the main thread on 11 September 2026: fda.gov/drugs/resources-information-approved-drugs/fda-grants-accelerated-approval-sevabertinib-locally-advanced-or-metastatic-non-squamous-non-small, attributed on its face to the Center for Drug Evaluation and Research, with article:published_time: Wed, 09/09/2026 - 11:34 and the footer line "Content current as of: 09/09/2026". Every quotation attributed to the primary on this card comes from that page, enumerated: the headline; the approval sentence; the expansion sentence; the SOHO-01 description; the outcome-measures sentence; the 69-patient efficacy sentence; the warnings-and-precautions sentence; the dosing sentence; the Project Orbis and MHRA sentences; the reviews-ongoing sentence; the priority-review / breakthrough / orphan sentence; the "Full prescribing information for Hyrnuo will be posted on Drugs@FDA" sentence; and the "Withdrawn | Cancer Accelerated Approvals" index title, which appears in the page's own left-hand navigation and is where the falsifier's withdrawal limb is checked. Quotations from other cabinet cards — from fda-approves-daraxonrasib-pancreatic-adenocarcinoma, fda-accelerated-approval-tudriqev-melanoma and fda-approves-zilganersen-alexander-disease — are named inline at the point of use and come from those files, not from this page. Not opened and not sourced to: the Hyrnuo prescribing information, Drugs@FDA, the SOHO-01 registration record on ClinicalTrials.gov, any Bayer press release, and any journal publication. No news summary contributed a fact to this card. A search summary is what surfaced the item; the date was read off the document (L71).