Hypertension risk prediction models for patients with diabetes based on machine learning approaches

To construct effective prediction models for hypertension in diabetic patients based on machine learning. This study used electronic data from 2080 diabetic patients attending the specialized outpatient clinic for metabolic diseases of the Affiliated Hospital of Qingdao University from March 2017 to July 2020. We adopted 5 machine learning algorithms (artificial neural network, decision tree, random forest, support vector machine, Bayesian network) and constructed hypertension risk prediction models based on patients' non-invasive variables. The study showed that artificial neural network (ANN) performed best, accuracy, sensitivity, specificity, and area under the receiver curve in the test set were 92.47%, 92.98%, 92.02%, 0.951, respectively. The prediction model showed that the top three predictors and their weight values were systolic blood pressure (w=0.3346), age (w=0.1437) and diastolic blood pressure (w=0.1236). Factors of daily living such as education level, activity, heart rate and fish intake also showed importance. ANN can apply non-invasive data to well predict the risk of secondary hypertension in diabetic patients. Healthcare providers could use this model to rapidly screen high-risk patients and instruct them to monitor blood pressure regularly and maintain a healthy lifestyle, thereby reducing the risk of hypertension in diabetic patients. The model is more suitable for areas with high morbidity of hypertension and poor socioeconomic conditions. It has important implications for achieving health management in a larger target population, in line with the international mission to improve the global burden of hypertension.