Well-day capillary ketone monitoring is known to predict near-term ketoacidosis independently of the usual clinical risk factors. The open question was how little of it you can get away with. Using the EASE 2 and EASE 3 trial repository of 1,410 participants, the authors simulated different testing frequencies over a month and asked which still predicted ketoacidosis or severe ketosis in the month that followed.
Once-weekly testing was the lowest frequency that held. Against a twice-weekly baseline, the area under the ROC curve was 0.692 versus 0.678 in the maximum-ketone models (p=0.19) and 0.719 versus 0.711 in the gradient-boosted tree models (p=0.38). Neither difference was significant, which here is the point: halving the testing did not measurably cost accuracy.
Why it matters: adherence to any monitoring schedule falls as its burden rises, so a schedule that asks for half as much and predicts as well is straightforwardly better. The authors also make the practical suggestion of using strips before they expire, which is what usually happens to them anyway.
What to do about it: for patients you already ask to monitor well-day ketones, weekly is a defensible schedule. This is a simulation on trial data rather than a prospective test of the schedule, so it is a reason to lighten an existing regimen rather than to start one where none was indicated.
- Weekly well-day ketone testing is enough where you already monitor
- Use strips before their expiry rather than letting them lapse unused
- This lightens an existing regimen; it is not a reason to begin monitoring afresh
- Derived from SGLT2-inhibitor trial populations, where ketosis risk is raised
- Sick-day rules are unchanged — this is about well-day baseline risk only
The statistics, in plain English
The area under the ROC curve measures how well a test separates people who go on to have an event from those who do not: 0.5 is a coin toss, 1.0 is perfect. Around 0.69 to 0.72 is modest — useful for stratifying a population, not for deciding about one patient in front of you. The p values of 0.19 and 0.38 mean the small drop from twice-weekly to weekly is well within what chance would produce, which is what licenses halving the frequency. Read this as failing to find a difference rather than proving there is none; with 1,410 participants a small loss of accuracy could hide inside those intervals.
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