PubMed abstract / permitted excerpt

OBJECTIVE: Independent of clinical risk factors, performing well-day capillary ketone monitoring over a month predicts near-term diabetic ketoacidosis (DKA) risk. We aimed to determine the minimum frequency of tests needed to maintain similar accuracy. RESEARCH DESIGN AND METHODS: Using regression and machine-learning gradient-boosted tree (GBT) models on the Empagliflozin as Adjunctive to Insulin Therapy 2 (EASE 2) and EASE 3 trial repository (n = 1,410), we simulated the testing frequencies of capillary ketones over a 1-month interval for predicting DKA or severe ketosis in the next month. RESULTS: Compared with the baseline twice-weekly ketone testing frequency, once-weekly well-day ketone testing was the lowest frequency that maintained prediction accuracy in maximum ketone models (area under the receiver operating characteristic curve 0.692 vs. 0.678; P = 0.19) and GBT models (0.719 vs. 0.711; P = 0.38). CONCLUSIONS: Weekly well-day capillary ketone testing, using existing strips before they expire, may provide a practical approach to stratify baseline DKA risk while reducing testing burden.