the stomach that won't empty

the stomach that won't empty

Problem statement

GLP-1 receptor agonists (semaglutide and its relatives) work in part by slowing how fast the stomach empties — that is one of the mechanisms that makes people feel full and eat less. The same mechanism means a patient who has followed every fasting instruction before surgery, endoscopy or sedation may still have solid food sitting in the stomach when the anaesthetist puts them to sleep, which is the classic setup for regurgitation and pulmonary aspiration. In a prospective cross-sectional study of 124 fasted patients, increased residual gastric content was found on ultrasound in 56% of GLP-1 users versus 19% of non-users (adjusted prevalence ratio 2.48, 95% CI 1.23–4.97). The unsolved problem is that there is no practical, validated way to tell which individual patient on these drugs actually has a full stomach on the day of the procedure — so a multi-society guidance document (ASA, AGA, ASMBS, ISPCOP, SAGES) that itself says it "should be considered guidance, and not an evidence-based guideline" is what stands between a very large and fast-growing patient population and an airway emergency.

Why this matters

Tens of millions of people are now on GLP-1 receptor agonists for diabetes and obesity, and they present for surgery, colonoscopy and sedation at ordinary population rates — meaning that a drug class introduced for metabolic disease has quietly become an anaesthetic-risk variable in every operating theatre and endoscopy suite. The cost of not solving it is paid in three currencies: rare but catastrophic aspiration events; a much larger volume of same-day cancellations and delays when a clinician is unsure and stands the case down; and unnecessary rapid-sequence inductions or 24-hour liquid-diet preparations imposed on patients who never had retained gastric contents at all. Because the guidance is consensus rather than evidence, practice varies between hospitals within the same city, so the risk a patient carries depends largely on which institution they walk into.

What’s been tried and why it hasn’t worked

Three approaches are in circulation and each fails in a distinct way. (1) Hold the drug. The 2023 ASA advice to withhold weekly agents for a week was widely adopted, but the multi-society document is explicit that "the duration to hold therapy is unknown"; stopping is also not free, since it destabilises glycaemic control and, for weekly formulations with long half-lives, may not clear the effect anyway. (2) Change the preparation. The guidance recommends a "preoperative liquid diet for at least 24 h, as performed in patients undergoing colonoscopy and bariatric surgery" when delayed emptying is a concern — but this is extrapolated from other populations, is burdensome, and has not been shown to reliably empty the GLP-1 stomach. (3) Measure the stomach. Point-of-care gastric ultrasound is the only bedside test that looks at the actual organ, and the guidance concedes it "may be clinically limited based on institutional resources, inter-user variability, and credentialing requirements" — i.e. it is operator-dependent, unevenly trained, and unavailable at the scale required if it were applied to every GLP-1 patient. Underneath all three is an evidence problem the field cannot brute-force: aspiration is rare, so linking any strategy to the outcome that matters requires enormous cohorts, and the available studies use residual gastric content as a surrogate. The JAMA Surgery authors flag exactly this — increased residual content is a surrogate, not an aspiration event.

What would unlock progress

The bottleneck is a triage measurement, not a drug decision: something that reliably sorts "empty enough" from "not empty" at the bedside, in minutes, without a credentialed sonographer. Two adjacent precedents are promising. Automated interpretation of point-of-care ultrasound has already been shown to make operator-dependent exams (cardiac ejection fraction, lung B-lines) usable by non-experts; the gastric antrum is a comparatively simple target, and a classifier trained on antral images plus patient position could turn a specialist skill into a checkbox. Separately, the anaesthesia community's own risk-prediction tradition — combining drug type, dose, last dose interval, GI symptoms and diet adherence into a stratification rule — has never been built and validated for this population, and would let hospitals reserve scarce ultrasound for the genuinely uncertain middle.

Entry points for student teams

A team could build and evaluate an automated antral-image classifier (or a phone-based capture-and-quality-check aid) against expert gastric ultrasound reads on a public or partner image set, reporting sensitivity for solid contents rather than accuracy alone. A clinical-informatics team could derive a candidate risk-stratification rule from an existing perioperative dataset — which combinations of agent, dose interval, and symptoms predict residual content — and pre-specify how it would be validated. A human-factors team could redesign the pre-procedure instruction pathway itself: what a patient on a weekly GLP-1 is told, when, and by whom, measured by adherence and by cancellation rate. Relevant skills: signal/image processing and machine learning, clinical epidemiology, human factors, and health-service design.