Study On The Distribution And Prediction Of Hospitalization Stay Of Workers Suffering From Cancer In Yunnan Province

After the 18 th National Congress of the Communist Party of China,the reform of the medical and health system has been comprehensively deepened,and the medical and health industry has ushered in rapid development.However,there are still many problems to be solved in the management of medical resources and the formulation of insurance compensation policies.The length of hospitalization is an important index to measure the resource consumption of inpatients.Therefore,in order to solve the problems of long hospitalization period and low turnover rate of ward beds,actuaries want to scientifically set the deductible period and maximum subsidy days of inpatient medical insurance,so as to improve the medical level and reduce the burden of patients,in-depth research on the length of patient hospitalization is very necessary.Combined with the fact that the disease and economic burden caused by malignant tumors are both heavy,this thesis studies the distribution and prediction of the duration of hospitalization of employees based on the hospitalization data of employees with malignant tumors extracted from the 17 th period of medical mutual aid activities of employees in Yunnan Province.Based on the right skew and peak characteristics of the actual data,this thesis uses the distribution fitting methods of parameters and non-parameters to construct an effective fitting distribution model for the length of hospitalization from multiple perspectives,so as to better explore the distribution law according to the data characteristics,and help the rational allocation of hospital resources and the scientific design of insurance institutions’ products.In this thesis,parameters were estimated by EM and other algorithms for the total length of hospital stay and the data divided by gender,and distribution fitting models were established under each data set.According to the research results,under the parameter distribution,only the Inverse Gaussian distribution of female workers with malignant tumors passed the goodness of fit test,and in the data group that did not pass the test,the Inverse Gaussian distribution also performed better,while the non-parameter distribution passed the goodness of fit test and performed better.Combined with the fitting performance of each parameter distribution,the inverse Gaussian model was selected to construct the mixed inverse Gaussian distribution models under different dimensions of all employees and male employees respectively.The optimal mixed inverse Gaussian distribution models were obtained by using K-S and other methods to test the two-dimensional length of hospitalization of all employees with malignant tumors and three-dimensional length of hospitalization of male employees.This highlights the good characteristics of the mixed distribution model in the face of complex actual data,and also provides a new idea for the study of the length of stay distribution.In addition,malignant tumor,as a disease with high prevalence and large medical consumption,has a long hospital stay.How to set the maximum subsidy days of inpatient medical insurance scientifically,how to effectively predict the long hospitalization of patients by insurance institutions and hospitals so as to reduce fraud and waste of resources,and other problems need to be urgently solved.Therefore,LightGBM algorithm is also used in this thesis to establish a classification prediction model for the length of extremely long stay,analyze several characteristic variables affecting the length of stay,and compare the prediction effect of different models,so as to obtain a more clinically practical prediction model.In the analysis,the model is compared with the common logistic regression and random forest algorithm.The recall rate,F1 value,Log-loss and other indicators fully demonstrate the good predictive performance of the model built based on LightGBM algorithm,which provides a reference for the prediction of the length of hospital stay.At the same time,at the early stage of establishing the model,there is obvious imbalance in the dependent variables in the data,in view of this article also put forward the treatment methods such as SVM SMOTE,the results prove that SVM SMOTE sampling to deal with the imbalance of the hospital stay data of the staff with malignant tumor,it is feasible and reasonable.In the study of the distribution of hospitalization duration,this thesis explored a variety of distribution methods,such as parametric and non-parametric,and tried to introduce a mixed distribution fitting model,which better adapted to the complexity of actual data distribution and optimized the fitting effect.In terms of the prediction of the length of stay,this thesis uses SVM SMOTE sampling to balance the data,and based on the current popular LightGBM algorithm,establish the classification prediction model,show the characteristic variables affecting the length of stay,the prediction effect is also improved compared with the traditional prediction model.