目的 基于决策树和神经网络对高血压病危险因素进行研究。方法 课题组成员从国家人口健康科学数据中心收集高血压患者临床病例资料，对所收集的数据进行预处理，采用卡方自动交互检测法（Chi-squared automatic interaction detector，CHAID）分类回归树（Classification and Regression Tree，C&RT）、二叉树（Quick，Unbiased，Efficient Statistical Tree，QUEST）决策树算法和神经网络模型进行分析。结果 3种决策树算法对高血压病的诊断准确率分别为82.10%、83.00%、83.00%；径向基函数（Radial basis function，RBF）神经网络和多层感知器（Multilayer Perceptron，MLP）的训练准确率和测试准确率分别为77.60%、78.70%和84.70%、84.30%，MLP神经网络的诊断模型优于RBF神经网络。结论 家族病史、高血脂病史、糖尿病病史、心脑血管病病史是高血压病的重要危险因素，且许多高血压病患者有中医舌暗的症状，存在着血管内皮的损伤，易引起心、脑、肾等靶器官的合并症。通过归纳总结其规律，可以为高血压病的预防与治疗提供依据。

Objective To explore the risk factors of hypertension based on decision tree and neural network.Methods Members of the research team collected the clinical data of patients with hypertension from the National Population Health Science Data Center, preprocessed the collected data and analyzed them with CHAID, C&RT, QUEST of decision tree algorithm and neural network model.Results The diagnostic accuracy of the three decision tree algorithms for hypertension was 82.10%, 83.00%, and 83.00%, respectively; the training accuracy and testing accuracy of the radial basis function (RBF) neural network and multilayer perceptron (MLP) were 77.60%, 78.70% and 84.70%, 84.30%, the diagnostic model of MLP neural network is better than RBF neural network.Conclusion The results showed that family history, hyperlipidemia history, diabetes history, cardiovascular and cerebrovascular disease history are important risk factors of hypertension, and many patients with hypertension have the symptom of tongue dark and have vascular endothelial damage, which easily caused complications of heart, brain, kidney and other target organs. By summarizing its patterns, it can provide a basis for the prevention and treatment of hypertension.

R259

国家自然科学基金项目（面上项目，重点项目，重大项目）