Baseline Diabetes Severity Is the Dominant Predictor of Diabetes Medication Re‐Initiation Following Bariatric Surgery

BACKGROUND: Bariatric surgery is highly effective for inducing remission of Type 2 diabetes mellitus (T2DM), yet the frequency and timing of diabetes medication re-initiation after achieving medication-free status are poorly characterized. OBJECTIVE: To estimate diabetes medication re-initiation after a postoperative medication-free interval and determine whether longitudinal postoperative weight and HbA1c trajectories provide incremental prognostic information beyond baseline diabetes severity. METHODS: We performed a retrospective cohort study using electronic health record data mapped to the OMOP Common Data Model. Adults with preoperative T2DM undergoing sleeve gastrectomy or Roux-en-Y gastric bypass between 2013 and 2026 were included. Patients entered the re-initiation risk set after achieving a postoperative medication-free interval of at least 90 days. Time to first subsequent diabetes medication restart was analysed using Cox models incorporating baseline clinical factors and longitudinal postoperative weight and HbA1c trajectories. RESULTS: Among 8012 unique patients undergoing bariatric surgery, 2197 (27.4%) had documented T2DM before the index operation. Of these, 2056 (93.6%) achieved a postoperative medication-free interval of at least 90 days and entered the re-initiation risk set; 882 (42.9%) subsequently re-initiated diabetes medication. Among patients who re-initiated therapy, median time to re-initiation from risk-set entry was 1.04 years (IQR, 0.30-2.84). In the primary Cox model, diabetes duration (HR: 1.219 per SD, 95% CI: 1.152-1.289), preoperative insulin use (HR: 1.627, 95% CI: 1.403-1.886) and baseline HbA1c (HR: 1.089 per SD, 95% CI: 1.030-1.151) were associated with increased medication re-initiation, whereas Roux-en-Y gastric bypass (RYGB) was associated with lower risk (HR: 0.778, 95% CI: 0.684-0.883). Age was not independently associated with re-initiation. Postoperative weight slope had a small association with re-initiation (HR: 1.042 per SD, 95% CI: 1.028-1.056; C-index = 0.6237). In the 1251-patient longitudinal HbA1c cohort, addition of HbA1c slope improved model fit (likelihood-ratio p = 0.0075) but provided little improvement in discrimination. CONCLUSIONS: Diabetes medication re-initiation following bariatric surgery was common and occurred relatively early after achievement of medication-free status. More than 40% of patients re-initiated therapy during follow-up. This frequency exceeds rates reported in much of the previous literature on biochemical diabetes recurrence, emphasizing that medication re-initiation and biochemical relapse represent distinct postoperative outcomes. Baseline diabetes severity was associated with medication re-initiation, whereas longitudinal postoperative weight and HbA1c trajectories provided limited additional prognostic information.