A prediction nomogram for sodium-glucose co-transporter 2 inhibitors-induced diabetes ketoacidosis in non-surgical patients: a retrospective multicenter cohort study.

BACKGROUND: Sodium-glucose co-transporter 2 inhibitors (SGLT2i) are associated with a rare but serious risk of diabetic ketoacidosis (S-DKA). This study aimed to identify risk factors for S-DKA, develop a tool for its early detection, and assess the associated in-hospital severity. METHODS: A retrospective, multicenter cohort study was conducted across four hospitals from September 2019 to June 2024. After screening 35,633 SGLT2i-prescribed inpatients, a 1:3 propensity score matching was applied. The study included 432 patients in the development cohort (2019-2023) and 128 in a validation cohort (2023-2024). Lasso regression and logistic regression were used to identify S-DKA risk factors and construct a nomogram. The primary endpoint for severity was the need for intensive care. RESULTS: Six independent predictive factors for S-DKA were identified: female, indications with diabetes, high HbA1c levels, fasting, infection and low BMI. The prediction nomogram demonstrated excellent discrimination, with an area under the curve (AUC) of 0.920(95% CI: 0.888 -0.952) in the development cohort and 0.735 (95% CI: 0.644 -0.826) upon validation. The model was well-calibrated. Furthermore, patients with S-DKA required intensive care at a significantly higher rate compared to those without. CONCLUSION: S-DKA is associated with greater in-hospital severity, necessitating early identification. The developed nomogram provides a clinically valuable tool for predicting S-DKA risk, potentially aiding physicians in proactive patient management and improving outcomes.