Research Of Hospitalization Expenses In Acute Appendicitis By Establishing The BP Neural Network

Background With the improvement of medical services and medical needs, the problem of excessive growth of medical exspenses becomes the focus of attention and needs to urgently solve.Controlling the excessive growth of medical exspenses, and analyzing the influencing factors also become the research focus.The previous studies usually use multiple linear regression method,which requires the data to meet many conditions such as normality and homogeneity of variance.But medical exspenses are generally skewed distribution and factors usually associated. Back Propagation neural network is the most widely used neural network model,which has no special requirement of the type and distribution of information.It has the ability of fault tolerance and self-adjustment.So this study uses BP neural network to analyze hospitalization expenses in acute appendicitis.Objective Repeatedly train the data,and set parameters of the network.Finally establish one model that fitting ability and simulation capability of the network reach an appropriate level.Then by sensitivity analysis,observe the main factors that have impaction on hospitalization expenses.By this study,we hope to analysis hospitalization expenses with an optimal modeling method,and provide a reference to control hospitalization expenses.Methods Establish the model of BP neural netwok of acute appendicitis. Take expenses as output variable,and the other 10 variables as inputs.First analyze data by single factor analysis and multiple linear regression analysis,then select different number of neurons in hidden layer(3,5,8,10,15,21,25),different training algorithm(trainlm,trainscg,trainoss,trainbr,trainrp),repeatedly train the data by Matlab7.1.0 and compare, finally establish one appropriate model. Then by sensitivity analysis,observe the main factors that have impaction on hospitalization expenses.Delete factors that has less influence such as gender, marital status and whether the first admission,and establish the model and do sensitivity analysis again.Compare the BP neural netwok model and multiple linear regression model.Results Main parameters of the eventual model: one hidden layer,twenty-one neurons in hidden layer, ten neurons in input layer and one neurons in output layer. The expected error is 0.05 and the maximum training step is set to 10000. The learning speed is set to 0.01.Use relative error to evaluate and compare the network, and training algorithm uses trainlm finally.The network trains 18 times when stopped, and SSE is 1.24879.The relative error of training set is 0.089011, R2 is 0.76544, adjusted R2 is 0.76101, and RMSE is 1610.4. The relative error of test set is 0.072081, R2 is 0.72509, adjusted R2 is 0.71914, and RMSE is 1046.6.Results of sensitivity analysis show that the top three factors are days in hospital, whether surgery and age. Marital status and gender have little influence on expenses.By comparison,the BP neural network has better prediction performance than multiple linear regression model. Conclusion The application of BP network to analyze hospitalization expenses is feasible.The parameters need to train repeatedly and do comparison to select appropriate parameters.Evaluation of the model need to consider both fitting ability and simulation capability of the network.Use MATLAB toolbox to do sensitivity analysis,and analyze influencing factors of hospitalization expenses in acute appendicitis,the results are consistent with the results of previous studies. The first factor influencing hospitalization expenses is the days in hospital. In order to control expenses of acute appendicitis,hospitals should improve the levels of diagnosing and treating ,and decurtate days in hospital.The BP neural network has better prediction performance than multiple linear regression model.