A European patent covering one of the best-known cardiovascular results in diabetes medicine has been revoked for telling the reader too little. In T 0251/25, decided on 29 June 2026 and published in writing on 29 September 2026, Technical Board of Appeal 3.3.04 set aside the opposition division’s interlocutory decision of 30 January 2025 and revoked Boehringer Ingelheim’s EP 2 981 255, “Therapeutic uses of empagliflozin” (application 14715274.8), for insufficiency of disclosure under Article 83 EPC. Claim 1 was directed to empagliflozin for use in preventing, reducing the risk of or delaying a cardiovascular event — specifically cardiovascular death and heart failure requiring hospitalisation — in patients with type 2 or type 1 diabetes or pre-diabetes. Ten oppositions had been filed; Dr Reddy’s Laboratories (UK) joined as an intervener, which The IPKat reports followed Boehringer suing it for infringement in the United Kingdom.
The gap the Board found was not a missing experiment but a missing explanation. The application as filed carried biomarker data — empagliflozin counteracting endothelial dysfunction and oxidative stress in a rat model of diabetic cardiomyopathy, and reducing arterial stiffness in type 1 diabetic patients — and no cardiovascular outcome data at all. The Board held that “nothing can be inferred from the examples themselves” about how those biomarkers are linked to the claimed therapeutic use (Reasons 5.14.1), and that the application offered “no evidence for this postulated mechanistic link, such as experimental data or a reference to supporting scientific literature” (Reasons 5.14.4). Two points then closed the door. First, of the documents advanced to show the link was common general knowledge, only two review articles were accepted as representative of common general knowledge at all — original research papers were not — and in those two the connection was found to be speculative and unsupported by data, so the link existed “as an unconfirmed hypothesis only” (Reasons 5.14.6, 5.14.7, 5.14.10). Second, the opponents showed that rosiglitazone had improved the very same markers in rats and humans and was later found to increase cardiovascular risk, and the Board treated the post-2008 FDA and EMA guidance demanding direct cardiovascular safety evidence for new antidiabetics as part of the skilled person’s common general knowledge (Reasons 5.17.3, 7.4). With credibility absent from the application as filed, the later outcome trial and the marketing authorisation could not be relied on, because that “would go beyond corroborating an established finding” (Reasons 5.22). Auxiliary request 17, which deleted the heart-failure limb and left cardiovascular death, failed on the same reasoning (Reasons 7.3–7.7).
What this changes for anyone drafting or defending a second medical use claim
The headline lesson is uncomfortable but precise: the drug works, the effect is now standard of care, and the patent is gone anyway. Sufficiency for a further medical use is judged on what the skilled person would have believed at the filing date, so the question a drafter has to answer is not “will this turn out to be true” but “does my specification, on its own face, make it credible”. T 0251/25 is not a new rule — it sits with G 2/21 and with recent decisions such as T 0883/23 and T 0853/22 — but it is an unusually clean demonstration of where applications fail.
Write the bridge, do not assume it. Every surrogate-marker case contains an unstated inference: improving marker X reduces outcome Y. Boehringer’s description asserted that endothelial dysfunction has prognostic significance for vascular events, and the Board still found the mechanistic link unevidenced, because the assertion was not backed by data or by a citation in the application as filed. If your credibility story runs through a mechanism, the mechanism and its support belong in the specification before filing — added later, they are worth nothing.
Common general knowledge means review literature, and it has to say something definite. The Board’s refusal to treat original research articles as common general knowledge is a practical constraint on how a credibility case can be built: a pile of primary papers is not the skilled person’s baseline knowledge. Equally, a review that frames the link as a hypothesis proves only that it was a hypothesis. Before relying on common general knowledge to carry a claim, read the reviews you intend to cite and ask whether they state the link as established or merely as plausible — and if they hedge, that hedge will be read against you.
Regulatory guidance now cuts both ways. The Board accepted agency guidance as part of the skilled person’s knowledge, which means a regulator’s published scepticism about a surrogate endpoint becomes evidence that the skilled person would not have inferred the clinical effect from that surrogate. Any portfolio with second medical use filings resting on biomarkers deserves an audit against this: where the FDA, EMA or an equivalent has said in guidance that a marker does not establish the outcome, the credibility of a claim to that outcome is already compromised on the day of filing. The same logic applies to near-neighbour compounds — the rosiglitazone analogy did real work here, and a candidate in a class with a failed predecessor starts with a disclosure burden that a first-in-class candidate does not.
Amendment is not a cure for an evidence gap. Auxiliary request 17 narrowed the indication and changed nothing, because the defect was in the evidential foundation of the whole application, not in the breadth of the claim. That is worth factoring into opposition strategy and divisional planning alike: if the credibility deficit is structural, auxiliary requests will burn costs without changing the outcome, and the realistic fallback is a later filing built around the outcome data itself rather than an amended version of the original.
For opponents and generics, the burden finding is the operational prize. The Board confirmed that an opponent under Article 100(b) EPC must raise serious doubts substantiated by verifiable facts (T 19/90), but then tied the weight of that burden to the quality of the patent’s own disclosure: “the type and quality of this disclosure in the patent application determines what type and quality of submissions is required from an opponent”, and in some circumstances an opponent “may discharge its burden of establishing insufficiency merely by submitting convincing arguments alone” (Reasons 5.7, citing T 63/06 and T 1558/11). Where a patent’s credibility rests on a hypothesis rather than data, a well-constructed documentary and argumentative attack can be enough — no counter-experiments required. That changes the cost model of a European opposition, and it makes the quality of the prior-art and literature search behind the attack the decisive variable. It also rewards reading the surrogate-marker story in a target patent early, as part of an invalidity assessment, rather than after the nine-month opposition window has closed.
One practical note on timing: the decision issued at oral proceedings on 29 June 2026 but the written reasons only became available at the end of September, and national proceedings involving the same family — including the UK action that prompted Dr Reddy’s intervention — run on their own clocks. Revocation at the EPO is effective across every designated state, so the ripple through parallel litigation and licence arrangements is immediate, whatever the national courts were separately being asked to decide.