Study On Influencing Factors And Risk Prediction Model Of Stroke Based On Big Data

The high rate of incidence,prevalence,recurrence,disability and mortality of stroke in China has brought a serious medical and economic burden.For high-risk populations of stroke,the demands of whole-cycle health management and whole-process healthcare services is gradually increasing.Prevention before disease onset is conducive to the implementation and development of active-health concept and risk management of stroke.In addition,comprehensive support of policy and technology for the healthcare services is provided by the national strategy of Healthy China and the context of eHealth.Moreover,the allocation of medical resources,and healthcare management of pre-hospital and in-hospital and posthospital are optimized.However,the feasibility,applicability,and practicality of stroke prediction in existing research are not satisfactory.Therefore,it is necessary to carry out the related research on the risk prediction of stroke,which could promote the risk management of stroke for the whole population and improve the risk prediction system of stroke.Based upon the theory of real world,medical big data,big data analysis and supervised learning,this research explores the crucial influencing factors of stroke.And the risk prediction research of stroke is developed by artificial neural network models with algorithms of Standard and ensemble learning.First,stroke-related influencing factors in three aspects of demographic data,laboratory indicators and clinical history are identified through in-depth literature research.Second,the data set in this paper is selected from clinical patients and physical examination population of hospitals in real world.The source data is then preprocessed for integration,normalization and filling.Then,basic data analysis,univariate analysis and multivariate regression analysis of the data were performed by statistical methods to extract crucial influencing factors profoundly interrelated to stroke,which will be used as predictors of the models.Finally,the Standard models of multi-layer perception and radial basis fundation are constructed.Furthmore,with ensemble algorithms of Bagging and Boosting,the ensemble models of artificial neural network are established respectively,and the performance of the prediction models are compared,analyzed and verified.With the real world data of 5936 patients in hospitals,this paper conducts a risk assessment research of stroke.The following results is proved by this research.First,inview of twenty-two influencing factors of stroke in the literature,twelve predictors of model are obtained through statistical analysis.Second,through comparative analysis,it is found that the multi-layer perception models occupy preferable comprehensive performance than that of the radial basis network models,whether using the algorithms of Standard or Bagging and Boosting.Thirdly,in the models constructed by the three algorithms of Standard,Bagging and Boosting,the ability for prediction and generalization of models is enhanced by adopting ensemble algorithms of Bagging and Boosting.And the models with algorithm of Boosting are better than Bagging.Lastly,seven significant predictors to be total cholesterol,history of stroke,serum creatinine,systolic blood pressure,age,heart disease and white blood cells are considered in the prediction models of stroke.And their degree of importance is decreasing in order.