Disease prevention has promoted the development of prediction models, but present papers have not put forward a comprehensive measure to improve the accuracy of prediction models. Through the cross-sectional and multi-dimensional comparison of a large number of domestic and foreign literature, the measures to improve the accuracy of prediction model for cardiovascular disease（CVD） risk from each module were sought. 395 domestic and foreign literature with high impact in medical journals was retrieved. Statistics analysis and classification of literature were performed. Different data analysis methods as well as mechanisms for chart and table analysis were creatively proposed. Data mining was carried out. In order to ensure the accuracy above 0.8, measures to improve the accuracy of prediction model were obtained from sample size and time interval, data source, feature selection, modeling method and other modules. The accuracy of the prediction model developed through the improvement measures is higher than other models of the same type.