Risk Prediction Of Carotid Atherosclerosis Based On Metabolic Related Indicators

In recent years,with the improvement of living standards and quality,the incidence of carotid atherosclerosis increases year by year.Carotid atherosclerosis refers to the gradual development of atherosclerotic lesions in the carotid arteries under the combined influence of aging,abnormal sugar and lipid metabolism,and high blood pressure.The early stage of carotid atherosclerosis is usually asymptomatic,and its progression is an important cause of ischemic stroke.According to relevant surveys,the standard incidence of asymptomatic carotid atherosclerosis in China is as high as 36.2%.Therefore,the establishment of an effective risk prediction model for carotid atherosclerosis,early prediction of the occurrence of carotid atherosclerosis,and active promotion of primary prevention are of great significance to prevent the occurrence of carotid atherosclerosis and cardiovascular and cerebrovascular events.Therefore,in this study,the population without carotid atherosclerosis was selected as the research object using a cohort study design.The integrated learning method in the field of machine learning was used in combination with electronic physical examination information of the hospital to establish a model that can identify carotid atherosclerosis highrisk populations.The aim is to provide a more accurate and convenient method for calculating the risk of carotid atherosclerosis.The main research contents and conclusions of this paper are as follows:(1)The literature related to carotid atherosclerosis was reviewed,and the characteristics affecting carotid atherosclerosis were sorted and classified according to demographic indicators,laboratory indicators and clinical history.A total of 1669 samples were collected from the medical database of a third-class hospital in Shandong Province.The baseline metabolic physical examination data and the later disease outcomes were mainly obtained,and the metabolic index system related to carotid atherosclerosis was constructed by different feature selection methods.(2)Firstly,multi-factor logistic regression,support vector machine,random forest,XGBoost and GBDT algorithms were used to establish a single prediction model for the risk of carotid atherosclerosis,and the AUC value,confusion matrix and other indicators were used to distinguish the model prediction effect.The results showed that the five models all had a certain degree of fitting problems.And the ability to predict the risk of disease is not so good.Then,super learner integration algorithm is used to weight logistic regression,random forest,support vector machine,XGBoost,GBDT.The results show that super learner integration model solves the problem of overfitting and underfitting of a single model,and the prediction effect is better.The ability to predict the risk of carotid atherosclerosis can effectively identify potential high-risk groups.(3)Using SHAP framework to solve the "black box" problem of machine learning models,especially complex models.Each feature of the model was explained,including the feature importance ranking from the perspective of the whole sample,as well as the interpretation of the direction and size of individual features,so as to realize the personalized evaluation of each sample case,so as to give targeted intervention programs,so as to reduce the risk of carotid atherosclerosis in high-risk groups.In conclusion,different from previous studies,which mostly focused on the diagnosis of patients with carotid atherosclerosis,this study took healthy people as the research object,adopted integrated learning algorithm to build a carotid atherosclerosis risk prediction model,and improved the prediction ability of the model.It has enriched the research on the prediction model of carotid atherosclerosis disease.It is of certain practical value to further make personalized interpretation of the model and carry out targeted intervention programs for individuals to reduce the risk of carotid atherosclerosis in high-risk groups.