| Cerebrovascular disease seriously endangers people’s health.The latest epidemiological data show that cerebrovascular disease has become the second leading cause of death and disability in the world,and the first death disease in China.There are 70 million patients with cerebrovascular diseases in China,with 2 million new cases annually and 1.65 million deaths annually.Every 12 seconds,one Chinese is onset,and every 21 seconds,one Chinese patient dies from cerebrovascular diseases.Cerebral infarction accounts for 87% of all cerebrovascular diseases.Although in the past 20 years,the treatment level of acute cerebral infarction has been continuously improving and the mortality rate has begun to decline,the disability rate and social burden of patients are still increasing year by year.Therefore,it is very important to discover the high risk population of cerebral infarction earlier,to screen and intervent risk factors,to do a better primary prevention.Carotid artery atherosclerosis(CCA)is an important pathological basis and independent risk factor of ischemic cerebrovascular disease such as cerebral infarction.The detection rate of CAA in patients with cerebral infarction is generally around 80%.CAA is also an important index for predicting ischemic cerebrovascular disease,with sensitivity of 84% and specificity of 79%.Carotid intima thickness and carotid plaque can be used as markers of CAA,and the severity of CAA can be objectively assessed in the population.Hence,early identification and intervention of carotid intima thickness and plaque is of great significance for the prevention of cerebral infarction.Studies mainly focus on the relationship between CAA and the incidence and prognosis of cardiovascular and cerebrovascular diseases,but there are few studies on the risk factors of CAA.Some scholars have studied how to make early prediction of asymptomatic carotid stenosis(ACS).Early identification of ACS can be achieved by rapid intervention with carotid endarterectomy(CEA)or carotid artery stenting(CAS)to reduce the incidence of cerebral infarction.However,ACS is very few in CAA.Therefore,many national guidelines do not recommend ACS screening in large sample populations.CAA is a continuous process.Increased carotid intima-media thickness(CIMT)begins in the early stage,followed by plaque formation,unstable plaque,stenosis of lumen,and eventually occlusion of blood vessels,leading to cerebral infarction.It is also very important for CAA to predict and intervene in the whole process.Few people have studied the risk factors of carotid intimal thickening,plaque or unstable plaque.If these people can be found out early,intervened early and carried out the primary prevention standardized,they may get more benefit.Nevertheless,there is no mature and reliable model to identify CAS as early as possibleMachine learning is a method that can endow machine learning with the ability to accomplish functions that cannot be accomplished by direct programming.In recent years,it has been successfully applied in many fields,such as disease prediction,stock market prediction and so on.The purpose of machine learning is to estimate the dependence between input and output of a system based on a given training sample,so that it can make as accurate a prediction as possible for unknown output.Machine learning prediction model may have special advantages for stroke,which is caused by many risk factors.PART 1 Risk factors of carotid atherosclerosis progression based on logistic regression analysis in high risk group of stroke individualsObjective: To establish the risk factors of carotid atherosclerosis through systematic screening of high-risk population for stroke,early identification of carotid atherosclerosis,and to help for the prevention and treatment of stroke.METHODS: 1.All community residents who participated in the screening of stroke high-risk population by the China National Stroke Screening and Prevention Project(CNSSPP)in Nanjing Brain Hospital from 2012 to 2016.The risk factors associated with carotid plaque were analyzed,including general information: Mean age(years),Sex(male),Marriage(married),Education level(Primary school or below);Past Medical History : Hypertension,Diabetes mellitus,Hypercholesterolemia,Atrial fibrillation;Personal and Family History : overweight,Lack of Physical activity,Current smoker,drinking alcohol,Family history of stroke;Blood Indicators:fasting plasma glucose,Hb A1 c,Homocysteine,Total cholesterol,low-density lipoprotein cholesterol(LDL-C),high-density lipoprotein cholesterol(HDL-C)and Triglyceride.2.Two-thirds of the population was randomly divided into Derivation Set group and one-third into Validation Set group.According to the results of carotid ultrasonography,the patients were divided into Carotid atherosclerosis group and Carotid non-atherosclerosis group.According to the degree of carotid atherosclerosis,the patients were further divided into four subgroups,included: carotid intimal thickness group(CIMT thickness),carotid plaque group,unstable carotid plaque group and moderate to severe carotid stenosis group.3.Measurement data were expressed as Mean ±SD or median.Differences among groups were analyzed by Student t test when the dates meet normal distribution.If the dates meet non-normal distribution,rank sum test were used.Qualitative data were described as percentages and analyzed using Chi-square(χ2)test.In the Derivation Set group,the risk factors of carotid atherosclerosis and its four subgroups were analyzed by univariate analysis.Logistic regression analysis was used to establish the prediction model,and the area under ROC curve was used to verify the effectiveness of the model.When univariate analyses show P≤0.05,multivariate logistical regression was utilized to detect the independent impact factors of carotid atherosclerosis and its four subgroups.When multivariate logistical regression,if the dependent variable was dichotomization classification variable,using Binary Logistic regression.According to the previous study,we generated a scoring system based on the regression coefficients.The lowest coefficient in absolute value was used as denominator.The coefficient of each independent risk factor was divided by the absolute value of the lowest coefficient and then rounded up to the nearest integers.Each subject would have a score according to the model and then scores of all the subjects were used to plot receiver operator characteristic(ROC)curve,Area under curve(AUC)was calculated to determine the prediction power of scoring system.In order to get best prediction cut-off value of the model,youden index was calculated by the ROC curve.The validation set group data was used to validate the model.ROC analysis was used to calculate the AUC value to evaluate the prediction efficiency of the model.In this study,P ≤ 0.05 is considered to have statistical significanceResults: A total of 2841 high-risk stroke patients were enrolled in this study in five years,1500 patients were found carotid atherosclerosis by carotid ultrasonography,the incidence of carotid atherosclerosis was 52.8%.Two-thirds of the population was randomly divided into Derivation Set group(n=1938)and one-third into Validation Set group(n=969).After adjusting for other risk factors,the independent risk factors for carotid atherosclerosis were: male(OR 1.645,95%CI 1.321~2.049)、age(50-59 OR 2.509,95%CI 1.693~3.718;60-69 OR 5.708,95%CI 3.902~8.348;≥70 OR9.800,95%CI 6.203~15.483),low educational level(OR 1.273,95%CI 1.002~1.681)、fasting blood glucose(≥7.0mmol/L OR 1.465,95%CI 1.056~2.034)、LDL-C(OR 1.416,95%CI 1.077~1.862)and high TC(OR 1.320,95%CI 1.004~1.736).Finally,six indicators including male,elder age,low education level,high fasting blood glucose,high LDL-C and high TC were included to the predictive scoring system,with a total score of 16.Using the scoring system,all the samples in the model group were scored,and the ROC curve was drawn with the score,the AUC value was 0.714.The best prediction score was 8.5.The sensitivity and specificity of predicting carotid atherosclerosis were 67.7% and 65.1% respectively.Taking this model into validation set group,the AUC value was 0.698.Subgroup analysis indicated that among the 2841 high-risk stroke patients,836(29.4%)of them were found carotid IMT,1000(35.2%)of them were found carotid plaque,299(9.4%)of them were found carotid unstable plaque,and326(11.5%)of them were found carotid moderate to severe stenosis(> 50%).Establishment of prediction models for four sub-groups by using Derivation Set group: 1.The independent risk factors for carotid IMT thickness were: age(50-59 OR2.010,95%CI 1.260~3.207;60-69 OR 3.069,95%CI 1.966~4.792;≥70 OR 3.685,95%CI 2.244~6.051),diabetes mellitus(OR 1.452,95%CI 1.082~1.948),fasting blood glucose(6.1-6.99mmol/L OR 1.398,95%CI 1.020~ 1.917)and high TC(OR1.604,95%CI 1.202~2.141).Finally,four indicators including elder age,diabetes mellitus,high fasting blood glucose and high TC were included to the predictive scoring system,with a total score of 7.Using the scoring system,all the samples in the model group were scored,and the ROC curve was drawn with the score,the AUC value was 0.621.The best prediction score was 3.5.The sensitivity and specificity of predicting carotid atherosclerosis were 61.8% and 57.5% respectively.Taking this model into validation set group,the AUC value was 0.619.2.The independent risk factors for carotid plaque were: male(OR 1.498,95%CI 1.193~1.880),age(50-59 OR 3.044,95%CI 1.764~5.253;60-69 OR 7.279,95%CI 4.309~12.296;≥70 OR13.371,95%CI 7.574 ~ 23.604),married(OR 1.344,95%CI 1.037 ~ 1.741),high fasting blood glucose(≥7.07mmol/L OR 1.529,95%CI 1.101~2.122),high Hcy(OR1.585,95%CI 1.171~2.145)and high TC(OR 1.343,95%CI 1.007~1.793).Finally,six indicators including male,elder age,married,high fasting blood glucose,high Hcy and high TC were included to the predictive scoring system,with a total score of15.Using the scoring system,all the samples in the model group were scored,and the ROC curve was drawn with the score,the AUC value was 0.719.The best prediction score was 8.5.The sensitivity and specificity of predicting carotid atherosclerosis were 66.1% and 65.5% respectively.Taking this model into validation set group,the AUC value was 0.719.3.The independent risk factors for carotid unstable plaque were: male(OR 2.031,95%CI 1.443-2.858),age(50-59,OR 5.735,95%CI1.343-24.490;60-69,OR 13.113,95%CI 3.180-54.075;≥70,OR 29.008,95%CI6.914-121.705),married(OR 1.702,95%CI 1.129-2.564),high LDL-C(OR 1.672,95%CI 1.076-2.600)and low HDL-C(OR 2.133,95%CI 1.350-3.372).Finally,five indicators including male,elder age,married,high LDL-C and low HDL-C were included to the predictive scoring system,with a total score of 11.Using the scoring system,all the samples in the model group were scored,and the ROC curve was drawn with the score,the AUC value was 0.738.The best prediction score was 6.5.The sensitivity and specificity of predicting carotid atherosclerosis were 73.6% and61.2% respectively.Taking this model into validation set group,the AUC value was0.718.4.The independent risk factors for carotid moderate to severe stenosis(>50%)were: age(60-69 OR 4.127,95%CI 1.962~8.680;≥70 OR 6.103,95%CI 2.783~13.379),married(OR 3.118,95% CI 2.050-4.745),lower overweight or obese(OR0.479 95%CI:0.346-0.662),alcohol dependence(OR 2.089,95%CI 1.424-3.064),fasting blood glucose(≥7.0 mmol/L OR 1.516,95%CI 1.066-2.156)and high LDL-C(OR 1.376,95% CI 1.150-1.646).Finally,six indicators including elder age,married,lower overweight or obese,alcohol dependence,high fasting blood glucose and high LDL-C were included to the predictive scoring system,while Overweight or obesity is a protective factor,with a total score of 11.Using the scoring system,all the samples in the model group were scored,and the ROC curve was drawn with the score,the AUC value was 0.726.The best prediction score was 5.5.The sensitivity and specificity of predicting carotid atherosclerosis were 62.8% and 73.0%respectively.Taking this model into validation set group,the AUC value was 0.694.Conclusion:1.The high-risk population of stroke has a high proportion of carotid atherosclerosis,which needs to be screened as soon as possible;2.Age and gender are the main non intervention risk factors of carotid atherosclerosis.Meanwhile,fasting blood glucose,TC and LDL-C were the main risk factors for intervention.People with low education level deserve more attention;3.Alcohol consumption was an independent risk factor for moderate to severe carotid stenosis in all carotid artery subgroup analyses.Overweight may have a protective effect on moderate and severe carotid stenosis;4.For a married man,who is older than 70 years old with high risk of stroke,especially those with high TC and LDL-C,the incidence of carotid plaque and stenosis is high,check carotid artery Doppler ultrasound immediately,and to start the primary prevention of stroke.PART 2 Risk factors of carotid atherosclerosis progression based on machine learning algorithm in high risk group of stroke individualsObjective: To establish the risk factors of carotid atherosclerosis through systematic screening of high-risk population for stroke,early identification of carotid atherosclerosis,and to help for the prevention and treatment of stroke.METHODS: 1.All community residents who participated in the screening of stroke high-risk population by the China National Stroke Screening and Prevention Project(CNSSPP)in Nanjing Brain Hospital from 2012 to 2016.All the definite risk factors and suspicious risk factors related to carotid atherosclerosis were counted.A total of53 attributes were identified.After removing more than half of the missing attributes and repeated invalid attributes,the following 30 attributes were selected,including general information: age,sex,marriage,education level,job,pay styles and diet habits;physical examination information: height,weight,waist circumference and body mass index.past medical history : hypertension,diabetes mellitus,atrial fibrillation and hypercholesterolemia;personal and family history:overweight,lack of physical activity,current smoker,drinking alcohol,family history of stroke,coronary heart disease,hyperlipidemia and diabetes;blood indicators:fasting plasma glucose,homocysteine,total cholesterol,low-density lipoprotein cholesterol(LDL-C),high-density lipoprotein cholesterol(HDL-C)and triglyceride.The data were input into the edited random forest algorithm,and the task markers were signed whether carotid artery color Doppler ultrasound had atherosclerosis or not.The model was calculated and optimized,and the weights of all risk factors in the model were listed.The model is validated by validation set and evaluated by AUC value.The four subgroups operate in the same way.Results: The general condition and grouping of the patients are the same as the first part.The data were input into the edited random forest algorithm,and the task markers were signed whether carotid artery color Doppler ultrasound had atherosclerosis or not.The model was calculated and optimized,the top five risk factors for carotid atherosclerosis in stroke high-risk population were age,fasting blood glucose,total cholesterol,low density lipoprotein and high density lipoprotein,the weights are 25.1%,8.3%,7.8%,7.3% and 5.9% respectively.The total weight of the top 15 risk factors is 87.4%.The AUC value of the model is 0.738 by validating the model with verification set.In sub-group analysis,data are input into the edited random forest algorithm,and four sub-group classifications are trained as task markers respectively.The model is obtained through calculation and optimization process: 1.the top five risk factors for carotid intima thickeness in stroke high-risk population were age,family history of hyperlipidemia,fasting blood glucose,total cholesterol level and weight,the weights are 25.2%,22.9%,9.3%,7.8% and 6.0% respectively.The total weight of the top 15 risk factors is 97.0%.The AUC value of the model is 0.660 by validating the model with verification set;2.The top five risk factors for carotid plaque in stroke high-risk population were age,fasting blood glucose,Hcy,low density lipoprotein and total cholesterol level,the weights are 43.3%,9.4%,7.0%,6.5% and 4.8%respectively.The total weight of the top 15 risk factors is 95.0%.The AUC value of the model is 0.728 by validating the model with verification set.3.The top five risk factors for carotid instability plaque in stroke high-risk population were age,family history of hyperlipidemia,fasting blood glucose,low density lipoprotein and Hcy,the weights are 34.4%,12.7%,9.0%,8.3% and 7.7% respectively.The total weight of the top 15 risk factors is 96.7%.The AUC value of the model is 0.739 by validating the model with verification set.4.The top five risk factors for carotid moderate to severe stenosis in stroke high-risk population were family history of hyperlipidemia,low density lipoprotein,high density lipoprotein,age and Body Mass Index,the weights are 11.8%,7.6%,7.1%,6.1% and 6.1% respectively.The total weight of the top 15 risk factors is 85.5%.The AUC value of the model is 0.888 by validating the model with verification set.Conclusion: 1.Advanced age remains the most important risk factor for carotid atherosclerosis and its four subgroups.2.Among the controllable risk factors,abnormal cholesterol level and elevated fasting blood gloucose are independent risk factors for carotid atherosclerosis,which should be actively controlled.3.For patients with a family history of hyperlipidemia,especially elderly residents with abnormal cholesterol levels,it is necessary to regularly review the color Doppler ultrasound of the carotid artery,and the possibility of unstable plaque and moderate or severe stenosis of the carotid artery is high.4.The AUC value of machine learning prediction model is higher than that of logistics prediction model,but the practical logistics prediction model is better. |
PDF Full Text Request