基于贝叶斯网络模型的山西省糖尿病相关因素

Analysis of factors associated with diabetes mellitus in Shanxi Province based on Bayesian network model

目的针对2015年山西省糖尿病调查数据,利用最大最小爬山(max-min hill-climbing,MMHC)算法构建糖尿病相关因素的贝叶斯网络模型,探索糖尿病及其相关因素间的网络关系,通过网络模型推理反映各影响因素对糖尿病的影响程度。方法采用单因素及多因素Logistic回归分析模型对2015年山西省≥18岁居民的糖尿病调查数据进行变量初筛,再以MMHC算法构建贝叶斯网络模型,参数估计采用极大似然估计法。结果2015年山西省糖尿病的检出率是9.5%。经Logistic回归分析模型对变量进行筛选后,年龄、职业、日均摄油量、高血压、高脂血症、BMI和心率被纳入贝叶斯网络模型;贝叶斯网络模型结果显示:年龄、高脂血症、高血压与糖尿病直接相关,BMI通过影响高脂血症与糖尿病间接相关,日均摄油量通过影响BMI和高脂血症与糖尿病间接相关。结论贝叶斯网络模型能很好地揭示糖尿病及其相关因素间复杂的网络关系,在分析疾病相关因素上具有较好的适用性和应用前景。

Objective For the survey data on diabetes in Shanxi Province in 2015,a Bayesian network model of diabetes-related factors was constructed using the max-min hill-climbing(MMHC)algorithm to explore the network relationships between diabetes and its related factors,and the strength of each influencing factor on diabetes was reflected through network model inference.Methods Single-factor analysis and multi-factor logistic regressions were used to initially screen the variables for survey data on diabetes mellitus among residents aged 18 years and above in Shanxi Province.Afterwards,a Bayesian network was constructed with the MMHC algorithm,and the parameters were estimated by great likelihood estimation.Results The detection rate of diabetes mellitus in Shanxi Province in 2015 stood at 9.5%.After logistic regression feature screening,eight variables,namely age,occupation,average daily oil intake,hypertension,hyperlipidaemia,BMI and heart rate,were finally entered into the model.The Bayesian network model demonstrated that age,hyperlipidaemia and hypertension were directly related to diabetes;BMI was indirectly related to diabetes by hyperlipidaemia,and the average daily oil intake indirectly affected diabetes by BMI and hyperlipidaemia.Conclusion Bayesian network models can well reveal the complex network relationships between diabetes and its associated factors and have a good applicability and prospects in the analysis of disease-related factors.