Capillary ketone testing on days when a patient feels well, done over a month, predicts their ketoacidosis risk in the following month independently of the usual clinical risk factors. The obvious objection is burden: nobody wants to ask a patient to prick a finger twice a week for a test that is normal almost every time.
Modelling on 1,410 participants from the EASE 2 and EASE 3 trial repository simulated different testing frequencies. Once-weekly well-day testing held its predictive accuracy against twice-weekly: area under the curve 0.692 against 0.678 in the maximum-ketone models (P=0.19), and 0.719 against 0.711 in the machine-learning models (P=0.38). Halving the testing did not cost meaningful discrimination.
Use it where it fits without new spending. Patients on SGLT2 inhibitors, and many with type 1 diabetes, already hold ketone strips that expire unused. A weekly well-day reading turns those strips into a risk signal instead of waste, and it builds the habit of testing before the day when the patient is vomiting and frightened and has forgotten how the meter works.
- Ask patients who hold ketone strips to check once a week on a day they feel well, and to bring the readings.
- Use rising well-day ketones as a prompt to review insulin adequacy, intake and adherence — not as a reason to act on that day alone.
- The weekly habit is also rehearsal: the patient who tests routinely will test correctly when actually unwell.
- Check strip expiry at the same visit; unused strips going out of date is the common failure.
- This stratifies baseline risk. It does not replace testing during acute illness, which stays mandatory.
The statistics, in plain English
An area under the curve of about 0.70 means the test ranks a patient who will develop ketoacidosis above one who will not roughly 70% of the time. That is modest — useful for sorting a clinic list by risk, not for deciding about an individual. The comparison here is not that weekly testing works well, but that it works no worse than twice-weekly: the P values of 0.19 and 0.38 mean the small gap between them is within chance. This is simulation on trial data, so it shows what the numbers support, not what happens when real patients are asked to do it.
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