Study On The Decision Tree Prediction Model For Influence Factors Of Vascular Cognitive Impairment

Objective To collect the demographic,life and clinical factors of patients with cerebrovascular disease,analysis of influence factors leading to vascular cognitive impairment(VCI),non dementia vascular cognitive impairment(VCIND),and vascular dementia(Va D)and set up VCI、VCIND、Va D influence factor models.Methods 505 patients with cerebrovascular disease were selected from our hospital department of health and neurology clinic or hospital treatment during October 2014 to October 2016.The demographics,lifestyle and clinical factors were questionnaire and collected,all patients were randomly divided into training set(n = 421)and test group(n = 84).,the training set were divided into non VCI group(225 cases)and VCI group(196 cases),VCI group was divided into Va D group(98 cases)and VCIND group(98 cases),the decision tree method and Logistic regression analysis were used to analyze the high risk factors of VCI,VCIND and Va D in patients with cerebrovascular.Finally,the corresponding VCI,VCIND,Va D decision tree models and Logistic regression model were established.Results 1.VCI decision tree model was established by training set,cross validation model recognition accuracy of 73.63%,test set prediction accuracy of 73.81%.Alcohol,hobbies,education level,drinking tea,diabetes,hypertension,diet,age,sleep and physical exercise were classification of node variables,drinking as the root node was the most important variable factor in all variables.According to result of Logistic regression analysis,cultural level,drinking,exercise and diabetes were independent risk factors for VCI,the model prediction accuracy of 66.98%,test set prediction accuracy of 53.57%.The decision tree model AUC was 0.737(95%CI 0.688 to 0.786),and the Logistic regression model AUC was 0.664(95%CI 0.612 to 0.717).2.VCIND decision tree model was established by training set,cross validation model recognition accuracy of 81.73%,test set prediction accuracy of 58.49%.Education,hyperlipidemia,hobbies,diabetes,sleep,diet and physical exercise were classification of node variables,education as the root node was the most important variable factor in all variables.According to result of Logistic regression analysis,cultural level,drinking,exercise and diabetes were independent risk factors for VCI,the model prediction accuracy of 72.14%,test set prediction accuracy of 54.72%.The decision tree model AUC was 0.716(95%CI 0.648~0.784),and the Logistic regression model AUC was 0.596(95%CI 0.525~0.666).3.Va D decision tree model was established by training set,cross validation model recognition accuracy of 71.43%,test set prediction accuracy of 69.42%.Drinking tea 、 hobbies 、 age 、 sleep 、 diabetes 、 alcohol were main factors impact VCIND progression to Va D,the root node was drinking tea,and was the most important variable factor in all variables.The probability of Va D had significant difference(p<0.05)in the crowds with different risk factors.According to result of Logistic regression analysis,drinking tea、hobbies、sleep were independent risk factors impact VCIND progression to Va D,the model prediction accuracy of 64.80%,test set prediction accuracy of 53.92%.The decision tree model AUC was 0.714(95%CI 0.641~0.788),and the Logistic regression model AUC was 0.648(95%CI 0.571~0.725).Conclusion 1.It is suggested that the decision tree model might be superior to logistic regression model in the prediction accuracy for VCI happened.2.The alcoholism,diabetes,high blood pressure,high fat diet,insomnia were risk factors for VCI,have hobbies,high level of education,take part in physical exercise,drinking tea could be the protection factors of the VCI.3.Hobbies,high education and physical exercise were the protective factors for VCIND,hyperlipidemia,diabetes,insomnia,and high fat diet were risk factors.4.Insomnia,age,diabetes,drinking were risk factors that patients with Va D,drinking tea and hobbies were the protection factors.