- Design
- Secondary modelling analysis of EASE-2 and EASE-3 trial data (regression and gradient-boosted trees)
- Population
- 1,410 adults with type 1 diabetes
- Primary outcome
- Prediction of DKA or severe ketosis in the following month by ketone testing frequency
- Effect
- AUC once weekly vs twice weekly: 0.692 vs 0.678 (P = 0.19); 0.719 vs 0.711 (P = 0.38)
Using trial data from 1,410 adults with type 1 diabetes in EASE-2 and EASE-3 (empagliflozin added to insulin), a Diabetes Care analysis simulated how often well-day capillary ketone tests need to be done to predict ketoacidosis or severe ketosis in the following month.
Once-weekly testing was the lowest frequency that kept prediction accuracy: the area under the curve was 0.692 versus 0.678 for twice weekly in one model (P = 0.19), and 0.719 versus 0.711 in a machine-learning model (P = 0.38).
Two caveats. Discrimination is modest at either frequency — about 0.7 means the test separates higher- from lower-risk patients, not that it predicts individual episodes. And the data come from trials of adjunctive SGLT2 inhibitor use in type 1 diabetes, a population whose risk differs from most.
Still, it gives a practical routine: a weekly well-day ketone check, using strips patients already hold before they expire, to stratify baseline risk — alongside, not instead of, sick-day testing.
- Consider a once-weekly well-day blood ketone check for people with type 1 diabetes at higher DKA risk
- Use repeatedly raised well-day ketones as a prompt to review insulin doses and adherence
- Keep sick-day ketone testing separate — a weekly check does not replace testing when unwell
- Suggest using strips already at home before they expire, which keeps cost down
- Treat a single normal reading as reassurance about baseline risk, not about the coming week
Why it matters
Ketone strips usually expire unused; a light weekly routine turns them into a risk signal.
Don't overread it
This is a simulation on trial data with modest discrimination; it has not been tested as a strategy that prevents DKA.
The statistics, in plain English
The area under the curve measures how well a test ranks people by risk: 0.5 is a coin toss, 1.0 is perfect. Values around 0.7 are useful for sorting groups but weak for individual prediction. P values of 0.19 and 0.38 mean weekly testing was not detectably worse than twice weekly.
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.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for diabetes & endocrinology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free