Hypertension Risk Prediction among Diabetic Patients Using Unconditional Multivariate Logistic Regression Model

Diabetes is a worldwide prevalent chronic disease, causing many kinds of complications among which hypertension is a common one. Previous studies focused on establishing prediction model on diabetes or essential hypertension with limited studies on predicting hypertension as a complication of diabetes. To fill in a gap of this field, our study targets on predicting hypertension among diabetes with laboratory data from The General Laboratory of the People's Liberation Arm. Continuous values assignment method and categorical values assignment method were used respectively establishing two models. Lasso regression and chi square test were used for variable selection. Unconditional multivariate logistic regression was used for model establishment. There are 15 variables in total identified as prominent predictors of hypertension among diabetes: Age, systolic pressure (BP-HIGH), hemoglobin (HBA1C), triglyceride (TG), high density lipoprotein cholesterol (HDL-C), fibrinogen (FBG), blood urea (BU), serum uric acid (SUA), serum albumin (ALB), lactic dehydrogenase (LDH_L), prothrombin time activity (PTA), globulin (GLO), diastolic pressure (BP_LOW), serum creatinine (SCR), hemoglobin (HB). Hemoglobin (HBA1C) was found as protective factor while it was found as risk factor in previous research. Both methods show high stability. Continuous values assignment method show higher authenticity while categorical values assignment method has better goodness fit. However, more variables on lifestyle, social-demographic features are supposed to be engaged for a more efficient model.