SGLT2 ketoacidosis risk persists through treatment and is predictable

Design

Nationwide cohort with nested case-control analysis, three Scandinavian registers

Population

282,282 adults with type 2 diabetes, 322,597 SGLT2 inhibitor treatment episodes, 2013 to 2022

Primary outcome

Ketoacidosis incidence and associated risk factors

Effect

2.43 per 1000 person-years; HbA1c ≥83 vs ≤52 mmol/mol OR 15.37 (11.50 to 20.53)

This Scandinavian cohort and nested case-control study used national registers from Sweden, Denmark and Norway, covering 322,597 SGLT2 inhibitor treatment episodes in 282,282 adults with type 2 diabetes, from 2013 to 2021–22. It was published in The Lancet Diabetes & Endocrinology in September 2026.

Ketoacidosis occurred at 2.43 per 1000 person-years. Risk was highest soon after starting but did not disappear. The strongest associated factors were HbA1c of 83 mmol/mol or more versus 52 or less (odds ratio 15.4), malnutrition (10.5), previous ketoacidosis (10.4), BMI under 20 (10.0) and recent hypoglycaemia (5.2). Infection co-occurred in about a third of cases against 6% of controls. Exploratory analyses suggested these risk factors were general to type 2 diabetes rather than unique to the drug class. A year after the event, a quarter were still on an SGLT2 inhibitor and insulin use had risen from 32% to 73%.

This shifts the conversation from a one-off warning at initiation to a risk review at every visit. The very poorly controlled, thin or malnourished patient, possibly insulin-deficient, is the one to watch, and a patient with an infection is the one to pause. The study was observational, so these factors identify risk rather than prove cause.

• Reassess ketoacidosis risk at every review, not only when starting an SGLT2 inhibitor.
• Consider insulin deficiency before prescribing to a thin patient with HbA1c above 83 mmol/mol (9.7%).
• Check ketones in any patient on an SGLT2 inhibitor who presents with infection.
• Use caution after previous ketoacidosis or recent hypoglycaemia; both carried high odds.
• Pause the drug during acute illness, surgery or heavy alcohol intake.

Why it matters

Most clinicians warn about ketoacidosis once, at the start; this risk does not keep to that schedule.

Don't overread it

Observational data: the risk factors predict ketoacidosis but do not prove the drug caused each case.

The statistics, in plain English

2.43 per 1000 person-years means about one event a year for every 400 people treated. The odds ratios above 10 are large, but their wide intervals (for example 11.5 to 20.5 for high HbA1c) reflect relatively few cases in each group. The authors note that milder transient illnesses may be under-recorded in controls, which would exaggerate some of those associations.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.