Abstract P349: Predictive Value of Long-Term Systolic Blood Pressure Variability for the Development of Type 2 Diabetes Mellitus

Background: Better identification of high-risk individuals of type 2 diabetes mellitus (T2DM) for focused delivery of preventive measures requires risk prediction models using novel predictors. We evaluated the predictive value of long-term variability of systolic blood pressure (SBPV), which was recently reported to be associated with T2DM incidence, if added to a model with conventional T2DM predictors in a Japanese cohort study. Methods: A cohort of 3017 Japanese individuals (2446 male, 571 female) ages 36-65 years were followed from 2007 to 2019. Root-mean-square error (RMSE) and slope of systolic blood pressure (SBP) change regressed on year were calculated per individual using SBP values obtained consecutively from 2003 to 2007 to represent SBPV. An initial Cox model included age, sex, smoking status, regular exercise, family history of diabetes, body mass index (BMI), baseline SBP, blood levels of triglycerides (TG), high-density lipoprotein cholesterol (HDLC) and fasting blood glucose (FBG), and backward elimination was used for variable selection. The c-statistics, net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were used to compare the performance of prediction models without (Model 1) and with SBPV (Model 2). Results: During 9.8-year (median) follow-up, 135 developed T2DM. We confirmed that RMSE was significantly associated with T2DM incidence independent of other variables used in a conventional model. Backwards elimination procedure selected BMI, TG, HDLC, FBG, SBP RMSE, and SBP slope for the final model (Table 1). Although the c-indices were not statistically different between Model 1 (0.77) and Model 2 (0.78) as well as the NRI (7.1%), the IDI was statistically significant (0.8%, p<0.05). Conclusions: The present study revealed that long-term variability of SBP slightly improved the predictive value of T2DM if added to a conventional prediction model.