10-year Risk Prediction Models of Stroke in Chinese Patients with Type2 Diabetes Mellitus

Abstract

Group Name: Background And Aims: Stroke is a major complication of T2DM. An accurate stroke prediction model can identify high-risk patients for more intensive medical and psychosocial interventions. Chinese are more prone to stroke than Caucasians but no model for the prediction of 10-year stroke risk in Chinese T2DM patients exist. The aim of this study was to develop a 10-year stroke prediction model among Chinese T2DM patients. Methods: 10-year retrospective cohort study of 141,516 T2DM Chinese patients from 2008 to 2017. The cohort was randomly split to development and validation samples at 2:1 ratio. Potential risk factors identified from systematic review included demographics, disease characteristics, clinical parameters and treatments. Variabilities measured by standard deviation of SBP and HbA1c, non-linear terms and interaction terms were also considered in model development. Cox proportional hazards regression was used to develop the models. Results: The 10-year cumulative incidence of stroke was 11.5% (Male: 11.7%, Female: 11.3%). Age, smoking, DM duration, SBP, variability SBP, TC/HDL-C ratio, HbA1c, variability of HbA1c, Urine ACR and eGFR were statistically significant factors as a main, transformation or interaction term. Age, DM duration, Urine ACR and eGFR were the most important risk factors. The models showed good prediction power (Female: Harrell’s-C ¼ 0.739, AUC ¼ 0.750; Male: Harrell’s-C ¼ 0.728, AUC ¼ 0.741) in the validation sample. Conclusions: This is the first 10-year stroke risk prediction model developed from and for Chinese T2DM patients. The inclusion of variabilities of SBP and HbA1c, transformation and interaction terms of relevant factors improved prediction accuracy significantly. The models can be converted to normograms and charts for clinical use.