1731-P: Optimising Prescribing in Type 2 Diabetes: Updating a Five–Drug Class Treatment Selection Model to Include Newer GLP-1RAs

Introduction and Objective: Precision medicine approaches can optimise type 2 diabetes management by identifying the best treatment for a specific individual. We previously developed and validated a clinical prediction model using nine routinely available features (such as age, BMI, eGFR and sex) to estimate individual glycaemic responses to five major type 2 diabetes drug classes (n=100,107 initiations). Here, we extend the model to predict glycaemic response to both oral and subcutaneous semaglutide. Methods: We used UK routine primary care data (CPRD Aurum) for non-insulin-treated individuals initiating semaglutide up to June 2024. The primary outcome was the 12-month change in HbA1c. We assessed the model’s calibration by comparing predicted versus observed HbA1c measures against other GLP-1RAs in the original model (liraglutide, dulaglutide, exenatide prolonged-release), informing necessary adjustments. External validation was conducted using Scottish primary care data (Scottish Diabetes Research Network, SDRN). Results: In CPRD, there were 3,056 oral and 6,969 subcutaneous semaglutide initiations (mean [SD] baseline characteristics: HbA1c 74[13] mmol/mol, age 59[11] years, BMI 36[7] kg/m2, eGFR 93[18] ml/min/1.73m2, 55% male) with complete predictor and 12-month HbA1c outcome data on stable therapy. Oral semaglutide responses were consistent with model predictions for other GLP-1RAs, requiring no adjustment. Differential treatment responses with subcutaneous semaglutide were consistent with model predictions, although requiring intercept adjustment to accommodate a mean 7.0 (95% CI 6.2-7.6) mmol/mol greater overall HbA1c response compared to other GLP-1RAs. The updated model displayed excellent calibration in SDRN external validation (n=2,951 initiations). Conclusion: An updated five-drug class model accurately predicts glucose-lowering responses to oral and subcutaneous semaglutide. External validation supports model generalisability and its potential for routine clinical application. Disclosure P. Cardoso: None. K.G. Young: None. L.M. Guedemann: None. A. McGovern: None. A. Jones: None. T. McKinley: None. E. Pearson: None. B. Shields: None. J. Dennis: None. Funding Medical Research Council UK (MR/N00633X/1), Wellcome Trust (227070/Z/23/Z)