Weekly well-day ketone testing predicts one-month DKA risk as well as twice weekly

Design

simulation study using regression and gradient-boosted tree models on a randomised trial repository

Population

1,410 participants with type 1 diabetes from the EASE 2 and EASE 3 empagliflozin adjunct trials

Primary outcome

discrimination for diabetic ketoacidosis or severe ketosis in the following month, by ketone testing frequency

Effect

once weekly vs twice weekly: AUROC 0.692 vs 0.678 (P = 0.19) for maximum ketone models and 0.719 vs 0.711 (P = 0.38) for gradient-boosted tree models

Using the EASE 2 and EASE 3 trial repository of 1,410 participants with type 1 diabetes, investigators simulated different frequencies of capillary ketone testing on well days and asked how few tests still predicted diabetic ketoacidosis or severe ketosis in the following month. Once weekly was the lowest frequency that held prediction accuracy against a twice-weekly baseline, in both a regression model using the maximum ketone value and a gradient-boosted tree model.

The idea behind well-day testing is counter-intuitive and worth stating plainly: the point is not to catch ketosis on the day of illness, which is what sick-day rules already cover. It is that a patient's ketone readings when they are well carry information about their baseline risk of ending up in DKA over the next month, independent of the usual clinical risk factors.

What changes is the ask. Twice weekly is a burden most patients will not sustain; once weekly, using strips that are already in the house and heading for their expiry date, is a request that might actually be met. That makes this a deliverable version of a risk-stratification tool rather than a theoretical one.

The population matters: these were trial participants on empagliflozin as an adjunct to insulin, a group in whom ketone surveillance is already advised. Extending weekly well-day testing to every patient with type 1 diabetes is a bigger step than this analysis supports.

• Ask patients on SGLT2 inhibitor adjunct therapy to test ketones once a week on a well day, not only when unwell
• Point them at strips already at home approaching expiry — the test costs nothing extra if the stock would be discarded
• Record the well-day readings; the pattern over a month is the signal, not any single value
• Keep sick-day rules exactly as they are — this sits alongside them, it does not replace them
• Revisit DKA risk at the next review in anyone whose well-day readings have crept up

The statistics, in plain English

The comparison rests on differences that were not statistically significant: AUROC 0.692 vs 0.678 (P = 0.19) for the maximum-ketone model and 0.719 vs 0.711 (P = 0.38) for the machine-learning model. A non-significant difference is not the same as proof that the two frequencies are equivalent — it means this dataset could not distinguish them, and a larger sample might. The AUROC values themselves, around 0.7, describe modest discrimination: useful for ranking who is at higher risk, not accurate enough to rule DKA in or out for an individual.

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.