It has been reported, including in people who do not have diabetes. A disproportionality analysis covered 7,349,591 FDA reports [1]. Semaglutide gave a reporting odds ratio of 3.15, 95% CI 2.78 to 3.56, from 261 cases. Tirzepatide gave 1.22, 95% CI 1.06 to 1.39, from 209 cases. Hospitalization was required in 74.3% and 71.3% of them. A reporting ratio is not a rate. Nothing in that database counts how many people took the drug. In a supervised type 1 diabetes trial, two episodes of euglycemic ketosis were recorded [2].
Ketoacidosis is a metabolic emergency normally associated with type 1 diabetes. It is not something expected in a person without diabetes at all. That makes it an unusual thing to find in the adverse event database for a weight drug. This analysis asks how often it is being reported [1]. The wider safety profile these drugs carry in people without diabetes is surveyed in the network meta-analysis of nineteen drugs.
Both drugs produced a signal among non-diabetic users. Semaglutide had a reporting odds ratio of 3.15 (95% CI 2.78–3.56) across 261 cases, and tirzepatide 1.22 (95% CI 1.06–1.39) across 209. Roughly three quarters of the cases in each group required hospitalization.
What lifts this above the usual pharmacovigilance caveats is the population and the severity. Ketoacidosis in someone without diabetes cannot be explained away by the underlying condition. In a diabetes cohort, most adverse events can be. And the hospitalization rate, above 70% in both groups, is a severity marker. Disproportionality analysis does not usually supply one, and these are not minor coded events.
The rising trend is the part most likely to be misread. Tirzepatide reports went from one or two a quarter in 2022 to 28 to 34 a quarter by 2025. That sounds like a worsening problem. It spans precisely the period in which tirzepatide went from newly launched to one of the most prescribed drugs in the country. More users generate more reports. Nothing in this design separates a rising rate from a rising denominator, the same trap set out in the reflux reporting study.
Where it sits in the wider report pile
These reports are a small part of a large file. A multi-source pharmacovigilance mapping took 243,114 GLP-1 records where the drug was the primary suspect, covering 2021 to 2026 [3]. Tirzepatide accounted for 133,100 of them, semaglutide 55,619 and dulaglutide 38,406. Frequently reported terms included incorrect dose administered, off-label use and extra dose administered. The authors call those use-process issues rather than adverse drug reactions. They also state that no source in the study estimates incidence, comparative risk or causality. That is the same limit that applies here, and it is counted in the dose error records.
The one setting where it was measured inside a trial
Type 1 diabetes is the population where ketoacidosis is expected, and it is not an approved indication for any of these drugs. A crossover trial added semaglutide to automated insulin delivery [2]. Time in range rose 4.8 percentage points without more hypoglycemia, and two episodes of euglycemic ketosis were recorded. Two events is a count, not a rate, but it came from supervised randomized follow-up rather than a spontaneous report. The rest of that evidence is in the type 1 diabetes trials.
The practical content is a symptom list rather than a probability. Persistent vomiting, abdominal pain, rapid breathing and confusion during a period of very low food intake are reasons to seek urgent care. Waiting is worse, particularly during dose escalation when intake falls fastest. That is the same window in which the nutritional problems in the thiamine cases arose. It is described from the tolerability side in titration and nausea.