Stopping SGLT2 inhibitors before surgery may be the harm, not the safeguard

1. Perioperative SGLT2 inhibitor discontinuation and cardiac complications

Journal · Published: British Journal of Anaesthesia — first published 11 June 2026, so ~3 months old, not new. It surfaced now because the issue version has appeared. This is the original paper behind the BJA commentary (published 26 August) that I flagged as unreachable in the 31 August brief — the original is now retrievable. Haziri, Durak, Glarner, Bolliger et al. (Basel)

The premise, stated plainly by the authors: international guidelines recommend stopping SGLT2 inhibitors 3–4 days before surgery to prevent euglycaemic ketoacidosis, and that advice rests chiefly on case reports and case series.

Secondary analysis of two prospective cohorts in major noncardiac surgery — Basel-PMI (NCT02573532) and PMI-Vital (NCT05866874). Primary outcome an ordinal composite of acute heart failure hospitalisation and cardiovascular death within 90 days, adjusted for prespecified covariates.

451 participants on chronic SGLT2i therapy; mean age 72, 22% women; 89.6% had diabetes, 36.9% chronic heart failure. SGLT2i were stopped before surgery in 87.1% — 39.2% for 1 day, 34.4% for 2 days, 13.5% for ≥3 days.

Cardiovascular complications, by duration of discontinuation:

	Events	Rate
Continued	1/58	1.7%
Stopped 1 day	10/177	5.7%
Stopped 2 days	13/155	8.4%
Stopped ≥3 days	7/61	11.5%

P=0.011; adjusted odds ratio 1.58 (95% CI 1.08–2.30) per discontinued day.

Euglycaemic diabetic ketoacidosis occurred in 1 of 451 participants — after discontinuation.

Conclusion: discontinuation “was associated with a substantially increased risk of 90-day cardiac complications. This suggests potential harm in current guideline recommendations.”

Interpretation. This is the most practice-relevant item in a fortnight, and it should be read carefully rather than acted on immediately.

What makes it compelling is the monotonic dose-response — 1.7%, 5.7%, 8.4%, 11.5% across increasing days stopped, with an OR of 1.58 per day. Dose-response is the observational finding hardest to explain by confounding alone, and this population is exactly the one that would suffer from SGLT2i withdrawal: 37% have chronic heart failure, in whom these drugs are prognostic therapy, not glycaemic control.

What should restrain you is confounding by indication, which the authors acknowledge by asking for randomised trials. Only 58 patients continued, and they may well be the more minor operations, the better-organised admissions, the patients without the comorbidity that prompts a cautious anaesthetist to stop early. A 1/58 event rate cannot bear much weight.

But note the other half of the ledger. The guideline exists to prevent euglycaemic DKA, and exactly one case occurred in 451 patients — in a patient who had stopped the drug. The harm the recommendation is designed to avert did not appear at any meaningful rate, while a plausible harm from following it did. That asymmetry is what makes this worth raising with your perioperative medicine colleagues now, even before a trial.

Read the paper · PMID 42270529

2. Single-port robotic thoracic surgery — early results and technique

Journal · Published: Annals of Thoracic Surgery, 4 September 2026 — genuinely new, 1 day old Pachos, Cerfolio, Bizekis, Zervos (NYU)

Retrospective series of consecutive extra-thoracic, single-incision operations using the da Vinci SP system, September 2024 – January 2026.

• 82 patients (44 women); median height 165.1 cm, median BMI 26.6
• Thymectomy 28, lobectomy 32, segmentectomy 17, wedge resection 5
• Median operative time 101 min (thymectomy), 137 min (lung resection)
• Median chest tube duration 0 h for thymectomy, 7.3 h in the 54 pulmonary resections
• Median blood loss 20 mL (range 15–40); median length of stay 1 day for all
• Median lymph node count 29 right / 23 left, median 5 N2 and 3 N1 stations; all R0
• The new SP stapler, used in the last 38 patients, had no malfunctions
• No major complications (Clavien-Dindo >IIIa), no 30- or 90-day mortality, 60-day readmission 1%

Described as the largest such experience outside South Korea.

Interpretation. The oncological numbers are the ones that matter and they hold up: 29 nodes and five N2 stations is a proper mediastinal dissection, not a compromise made to fit a single port. A median one-day stay and a zero-hour chest tube for thymectomy are striking.

Read it for what it is, though — a single-centre consecutive series from a high-volume robotic group with no comparator. Cerfolio’s unit is not a representative sample of thoracic surgery, and “feasible and safe in expert hands” is a different claim from “better than multi-port.” Median BMI 26.6 and median height 165 cm also describe a favourable body habitus for single-port access.

Still, this is where the technique is going, and the stapler working reliably across 38 consecutive cases is the practical detail that decides whether a platform is adoptable.

Read the paper · PMID 42697395

3. Nerve blocks for below-knee amputation and phantom limb syndrome

Journal · Published: Regional Anesthesia & Pain Medicine, 4 September 2026 — genuinely new, 1 day old Nguyen, Hoffman, Gotewal, Tsai et al.

Retrospective cohort from Epic Cosmos, 2016–2025. Adults undergoing elective below-knee amputation, categorised by receipt of perioperative peripheral nerve block. Primary outcome: phantom limb syndrome diagnosis within 12 months (ICD-10 G54.6, G54.7). Mantel-Haenszel stratification adjusted for age, sex and race only.

• 38,433 patients — 7,430 (19.3%) received blocks, 31,003 (80.7%) did not
• PLS coding: 18.8% with blocks vs 15.1% without — absolute difference 3.7%
• Adjusted RR 1.24 (95% CI 1.20–1.28)
• Blocks were associated with shorter length of stay

The authors are emphatic that these are exploratory associations, that the effect may reflect residual confounding by indication and differential surveillance rather than causation, and that “PNBs still remain appropriate for improving acute pain and reducing LOS.”

Interpretation. A counterintuitive direction — blocks have long been hypothesised to prevent phantom limb pain by interrupting nociceptive input around amputation — so it is worth being clear about how weak the design is before anyone changes practice.

The adjustment is for age, sex and race only. Not for diabetes, peripheral vascular disease, preoperative opioid use, chronic pain, or severity — every one of which plausibly drives both the decision to block and the later diagnosis of phantom limb. And the outcome is ICD coding, not assessment: a patient who received a block is a patient in a service that thinks about limb pain, follows them up, and codes for it. Differential surveillance alone could produce a 3.7% difference.

So: not evidence that blocks cause phantom limb syndrome, and the authors do not claim it is. What it does show is that the protective effect many of us assume is not visible in 38,000 real patients, which is a fair challenge to a comfortable belief. The randomised evidence for pre-emptive analgesia preventing phantom limb pain has always been thin, and this is a reminder of that.

Read the paper · PMID 42697602

---

Also noted, not a paper: JTCVS carried an obituary this week for Dr David Taggart (1958–2026), titled “Forever the Guardian Angel of CABG” — the ART trial investigator and one of the most forceful advocates for surgical revascularisation of his generation. Given how much of this archive’s recent CABG-versus-PCI material sits downstream of his work, worth knowing.