Application Of Bayesian Network Based On MMHC With A Hybrid Algorithm In Studying Influencing Factors Of Type 2 Diabetes

Objective:The purposes of this research were to(1)study on hybrid algorithm for Bayesian network structure learning based on MMHC algorithm;(2)explore the effect of MMHC algorithm in constructing Bayesian network model when compared with tabu search algorithm;(3)establish the Bayesian network to explore influencing factors of diabetes,study the relationship between these factors and diabetes,and reflect the strength of relationship between these influencing factors and diabetes;(4)offer a flexible tool for assessing influencing factors of other chronic diseases.Methods:The standard Bayesian network model was selected to generate simulation data.The models established by tabu search algorithm and MMHC algorithm respectively were evaluated through the number of missing edges and redundant edges and their sum.We established Bayesian network of diabetes with the data of chronic diseases in Shanxi Province in 2013 to explore influencing factors of diabetes.We also compared the model with logistic regression model.Results:(1)First,different sample sizes of datasets were randomly generated through standard Bayesian networks and then constructing a Bayesian network model using tabu search algorithm and MMHC with a hybrid algorithm respectively.The results showed that the larger sample size of the dataset,the higher consistency of the constructed Bayesian network structure with the standard network,regardless of the number of nodes.However,if there were few nodes in the network,the effect of the two algorithms on the Bayesian network construction is not significant,regardless of the sample size.For a large number of nodes,when the sample size is small,the learning effects of the two algorithms are consistent,however,when there is a large sample size Bayesian network based on MMHC with a hybrid algorithm is superior to tabu search algorithm.(2)The univariate analysis was performed on the related factors of diabetes.Multivariate logistic regression analysis was performed on the statistically significant variables(P<0.05),and the Bayesian network structure was established.The logistic regression results showed that factors eventually entered the regression model were age,region,marital status,medical insurance,BMI subgroup,central obesity,hypertension,hyperlipidemia,and passive smoking.The main risk factors of diabetes were hypertension,hyperlipidemia,and central obesity,with risk coefficients as 1.867,1.448 and 1.269 respectively.While constructed Bayesian network model showed that hypertension and hyperlipidemia were directly associated with diabetes;age,central obesity,and BMI were directly associated with hypertension and linked to diabetes through hypertension;There was a different detection rate of urban and rural areas with hyperlipidemia,which will affect the prevalence of diabetes indirectly.There is a correlation between the remaining variables,but it is far away from the network of diabetes.Bayesian reasoning found that it will reduce the prevalence rate of diabetes to 0.104 in those who do not have hypertension and hyperlipidemia.If someone were only suffering from high blood pressure,the probability of developing diabetes is 0.176.If you only have hyperlipidemia,the risk of diabetes is 0.133.The probability of developing diabetes is 0.272 when there are both hypertension and hyperlipidemia.Compared with the logistic regression,various factors in the Bayesian network can connect with diabetes through complex topological structure which can be better reflect the complex relationship between influencing factors and factors and diseases,and express more accurately and intuitively.Conclusions:(1)For a large number of nodes,when there is a large sample size.If there were many nodes,the Bayesian network based on MMHC with a hybrid algorithm is superior to the tabu search algorithm;when the sample size is small,the learning effects of the two algorithms are consistent.(2)The results of Bayesian network model based on MMHC with a hybrid algorithm which was applied to establish the diabetic Bayesian network model suggested that,hypertension and hyperlipidemia may directly affect the prevalence of diabetes,and the remaining variables indirectly affect the development of diabetes.Through the network topology,the complex relationship between the influencing factors and the disease was discovered,and the relationship between the influencing factors was discovered.This offers a flexible tool for assessing influencing factors of diabetes and could be better prevent diabetes.