| The problem of drug shortage has always been the focus of many countries.Its causes of formation are complicated and the rules are difficult to follow,which seriously delays the treatment time of patients and even affects the health of the general public.In the 13 th Five-Year Plan for deepening the reform of the medical and health system,it is mentioned that the centralized drug procurement system should be optimized,the reform in the field of drug supply should be deepened,the monitoring and early warning of drug shortages should be strengthened.Therefore,monitoring and early warning has become an important measure to prevent and solve the problem of drug shortage.Based on this background,relevant data of centralized drug procurement was relied on in this paper,in order to find an accurate,effective,practical and applicable drug shortage monitoring and early warning model,which can be applied to solve the actual problem of drug shortage,ensure adequate drug supply and protect people’s health.The main research contents are as follows:First of all,the data of centralized drug procurement were analyzed in charts and graphs to intuitively understand the current situation of drug shortage.Early warning indicators were extracted from massive drug related data.In order to reduce the complexity of model calculation,K-S test and Mann-Whitney U test were used to select early warning indicators.Finally,15 early warning indicators were obtained for the establishment of subsequent monitoring and early warning models.Secondly,the early warning model of drug shortage was established in order to make the prediction effect better,factor analysis was used in this paper to convert 15 possible correlation early warning indicators into 6 unrelated common factor indicators.Meanwhile,Lasso regression was used to compress the original 15 early warning indicators into 8 key early warning indicators,and Logistic regression models were constructed for them respectively.Through comparison,it was found that the classification accuracy and other evaluation indexes of Lasso-logistic regression model were higher than those of factor analysis-Logistic regression model,which could accurately and effectively identify drug shortage in advance.Then,the defect that Logistic regression model can not deal with nonlinear classification problem well was considered,the stochastic gradient descent method was used to construct the factorization machine model to improve the extrapolation and applicability of the drug shortage early warning model.In order to improve the prediction accuracy of the model,the classical random forest model in Bagging algorithm was used,combined with the cross validation method to obtain the best super parameters,and the monitoring and early warning model was constructed.Compared with the factorization model,the stochastic forest model has higher classification accuracy and better performance in the prediction of the most concerned drug shortage,the pre-research and judgment of drug shortage was realized,and the prediction accuracy and applicability of the model were improved.Finally,in view of the good performance of Bagging algorithm,using Boosting algorithm to build a monitoring and early warning model to further improve the prediction accuracy of the model was considered in this paper.The optimal hyperparameters were obtained by grid search and cross validation,and GBDT and XGBoost drug shortage warning models were constructed.The above warning models were comprehensively compared,XGBoost model has the best performance and good performance in application verification.With the update and iteration of the information related to centralized drug procurement,the real-time dynamic warning of drug shortage and promote the solution of the actual drug shortage problem were realized accurately and effectively.The key indicators obtained were consistent with other models,which are helpful for relevant departments to arrange reasonable prevention and resource allocation in advance. |
PDF Full Text Request