Precision prescribing of SGLT2-inhibitors in people with type 2 diabetes for primary prevention of heart failure: model development and validation study

ObjectivesTo develop and validate a model to predict individual-level benefits of SGLT2-inhibitors for heart failure prevention in people with type 2 diabetes who do not have atherosclerotic cardiovascular disease (ASCVD), heart failure (HF) or chronic kidney disease (CKD). DesignPopulation based cohort study. SettingThe SGLT2-inhibitor absolute response (SABRE) model estimated individual-level 5-year heart failure benefits using absolute heart failure risk predictions from the previously validated QDiabetes-Heart Failure model, combined with the hazard ratio of SGLT2i-associated hospitalisation for heart failure benefit (HR: 0.63 [95% CI: 0.50 to 0.80]) from trial meta-analysis. The data source was primary care records with linked hospital admission and death records (Clinical Practice Research Datalink, England) from January 2013 to October 2020. PopulationAdults with T2D without ASCVD, HF or CKD initiating an SGLT2i (n=57,368) or, as a comparator group, DPP4-inhibitors or sulfonylureas (n=111,673). Propensity score weighting and regression adjustment were used to mitigate potential treatment selection bias. Validation of model accuracy was performed by comparing the calibration of predicted versus observed heart failure benefits in those receiving SGLT2i versus comparator therapy. Main outcome measuresNew-onset heart failure recorded in primary care, linked hospital or death records. ResultsAmongst 57,368 SGLT2i and 111,673 DPP4i/SU treatment initiations, estimated risk of new-onset heart failure was 30% lower with SGLT2i versus the comparator arm (HR: 0.70 [95% CI: 0.63 to 0.78]), similar to trial meta-analysis for hospitalisation for heart failure. The relative benefit of SGLT2i on heart failure was consistent across all levels of baseline absolute heart failure risk as estimated from the QDiabetes-Heart Failure model (p=0.82 for treatment arm:baseline heart failure risk interaction). 5-year absolute heart failure benefit predictions with SGLT2i ranged from <0.1% to 14.1% across individuals (median 1.0% [IQR 0.6-1.8%]), and predictions were well-calibrated against observed heart failure outcomes. Model evaluation established that individualised targeting of SGLT2i initiation using SABRE could more efficiently reduce heart failure outcomes compared to current treatment recommendations for those with T2D without ASCVD/HF/CKD. ConclusionsSGLT2i can be targeted to individuals with type 2 diabetes without ASCVD/HF/CKD for primary prevention of heart failure using an easily deployed clinical prediction model integrating evidence from clinical trials.