Weekly ketone testing predicts as well as twice weekly

Design

simulation study using regression and gradient-boosted tree models on pooled randomised trial data

Population

1410 participants with type 1 diabetes from the EASE 2 and EASE 3 trial repository

Primary outcome

accuracy of different well-day capillary ketone testing frequencies over 1 month for predicting ketoacidosis or severe ketosis in the next month

Effect

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

Well-day capillary ketone monitoring - testing when the patient feels fine, not only when they are ill - predicts near-term ketoacidosis risk independently of clinical risk factors. The obvious objection is burden. This analysis asked how little testing is enough.

Using the EASE 2 and EASE 3 trial repository of 1410 participants with type 1 diabetes, the investigators simulated testing frequencies over a one-month window and measured how well each predicted ketoacidosis or severe ketosis in the following month, with both regression and gradient-boosted tree models. Against the twice-weekly baseline, once-weekly testing was the lowest frequency that held accuracy: area under the receiver operating characteristic curve 0.692 against 0.678 in maximum-ketone models (P=0.19), and 0.719 against 0.711 in the machine-learning models (P=0.38).

That is a halving of the testing burden for no measurable loss. The authors make a practical point that is easy to miss - use the strips you already have before they expire, which turns an unused box into a monitoring programme. What the study does not establish is that acting on the result changes outcomes; it establishes that a weekly strip carries the same predictive information as two. The discrimination itself is modest at around 0.7, so this is risk stratification, not a test that tells you what will happen to an individual.

• Ask patients on insulin to test capillary ketones once a week on a well day, not only when unwell.
• Point them at the strips they already have - expiry, not cost, is often what wastes them.
• Use the result to stratify risk and prompt a conversation, not to predict an individual episode.
• Keep sick-day ketone testing entirely separate; this is about baseline risk, not acute illness.
• An area under the curve near 0.70 is modest - treat a reassuring week as reassuring about the month, not about the year.

The statistics, in plain English

The comparison is a non-inferiority argument made with P values, which is worth reading carefully: P=0.19 and P=0.38 mean no significant difference was detected between weekly and twice-weekly testing, not that the two were shown to be equivalent. With 1410 participants a small real difference could hide inside those P values. The absolute discrimination matters more: an area under the curve of about 0.70 means that if you picked one person who went on to have ketoacidosis and one who did not, the model would rank them correctly about 70% of the time - useful for stratifying a clinic list, not for reassuring an individual. This is also a simulation on trial data, where adherence to testing was higher than it will be in your clinic.

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.