MMPC-Tabu混合算法的贝叶斯网络模型在高脂血症相关因素研究中的应用

Using the MMPC-Tabu Hybrid Algorithm based on Bayesian Network Model to Analyze the Related Factors of Hyperlipidemia

目的本研究采用MMPC-Tabu混合算法构建山西省高脂血症的贝叶斯网络模型,研究高脂血症及其相关因素间的网络关系及相关程度,并通过贝叶斯网络对高脂血症进行患病风险推理,为慢病影响因素分析提供更合理的建模方法。方法采用logistic回归对山西省18岁以上居民高脂血症的调查数据进行变量初步筛选,再以MMPC-Tabu混合算法和极大似然估计法构建贝叶斯网络。结果2013年山西省高脂血症患病率为42.6%(95%CI:41.1%~44.1%)。将logistic回归初筛与高脂血症有关的9个变量,采用MMPC-Tabu算法构建高脂血症的贝叶斯网络模型,结果显示:中心性肥胖和BMI与高脂血症直接相关,是高脂血症的父节点,即它们与高脂血症的发生有关;高血压、身体活动、性别、年龄、地区、糖尿病通过影响中心性肥胖和BMI间接影响高脂血症的发生。结论贝叶斯网络可以反映因素与疾病整体联动效应,揭示高脂血症直接和间接相关的因素和关联强度,同时阐明除高脂血症以外的其他影响因素间的关系,可为慢性病与相关因素的研究提供合理的方法。

Objective In this study,a Bayesian networks(BNs)model of hyperlipidemia in Shanxi Province was constructed by the MMPC-Tabu hybrid algorithm,aiming to study the network relationship and correlations between hyperlipidemia and its related factors.Besides,we aimed to achieve the risk reasoning of hyperlipidemia through BNs,thereby,providing a more reasonable modelling for the analysis of chronic disease influencing factors.Methods Logistic regression was used to conduct a variable preliminary screening of survey data on hyperlipidemia in residents over 18 years old in Shanxi Province.Afterwards,the BNs were constructed by MMPC-Tabu hybrid algorithm combined with maximum likelihood estimation.Results The detection rate of hyperlipidemia in Shanxi Province in 2013 was 42.6%(95%CI:41.1%~44.1%).Nine variables related to hyperlipidemia were screened by logistic regression and the BNs model of hyperlipidemia was constructed by the MMPC-Tabu algorithm.The results suggested that central obesity and BMI are directly related to hyperlipidemia and represent the parent nodes of hyperlipidemia,namely,they are directly related to the occurrence of hyperlipidemia:hypertension,physical activity,gender,age,region and diabetes indirectly affect the occurrence of hyperlipidemia by affecting central obesity and BMI.Conclusion BNs could reflect the overall linkage effect between factors and diseases,revealing directly and indirectly related factors and the strength of hyperlipidemia.Besides,it could clarify the relationship between other influencing factors except for hyperlipidemia.Our model can provide a reasonable approach for research of chronic diseases and related factors.