| During recent a few years,the charge of medical care has increased rapidly. the large medical care charge, especially the hospitalization charge,bring heavy economical burden to People. How to control too large health care, optimal allocation of health resources, establishment of a multi-level medical security system has become a hot spot of Social concern in recent years. An effective way to control it is to research about its influence factors. To fit the relationship between hospitalization charge and its influence factors, and establish effective hospitalization fitting model is the key topic in the research. One method which is used frequently for analyzing hospitalization charge is multiple linear regressions, but it has some limitations, such as independent, normal distribution, equal variance, linear and so on. In fact, the relationship between hospitalization charge and its influence factors is very complex, it is perhaps no linear, or there maybe multicollinearity influence factors. So we need to develop another useful model which is suitable for hospitalization charge to express the relationship.Artificial neural network, also known as neural network, is a mathematical model which was developed according to the work of biological neuron. Artificial neural network is widely used for pattern recognition, forecast, numerical approximation, etc. BP neural network is one kind of the multilayer perception; it is named because the adjustment of network weights by back propagation algorithm rules, and is currently the most mature and most widely used network model of all kind of neural network.As BP neural network has no limitation to the type and distribution of the data, and has certain ability to tolerance error, also it can fit complex relationship between the input variables and output variable through self-study, self-adjustment, Therefore, BP neural network can be considered for its impact on the cost of modeling factors, hospitalization expenses and the realization of the relationship between the fitting factors. Objective Set the appropriate parameters established based on BP neural network fitting model hospital fees. BP in the established neural network model, based on various factors impact on the cost of the measure. Using this study method of BP neural network modeling study to provide some of the reference basis, and through the impact on hospital costs and its influence factors 's analysis of health management to help decision-makers and those who make the right medical insurance industry's decision-making and analysis.Method collecting the mortality of Infarction cases who met with hospitalization from 2007 and 2008 in certain hospitals of the Tangshan city . Theer are 2538 cases collected in all. After eliminating imperfect cases,illogical cases and the cases that are not cured,we got 2218 effective cases fanially, which were87.39% of the total case. Result We used BP network to fit function between hospitalization charge and its influence factors which is not visual. During process of e establishing model,ANOVA was used to compare different training arithmetic and different numbers of neurons in hidden layer. Sensitivity analysis was used to analyze those influence factors based On the model .ANOVA was completed by using SAS8.2,and establishing of BP neural network model and sensitivity analysis were both completed by using MATLAB.Results(1)Result of comparing different training arithmetic of model establishing. Four types of BP neural network were established,5,10,15,20 were used as the number of neurons in hidden layer. And each type of network used LM arithmetic,SCG arithmetic,OSS arithmetic and BR arithmetic So there 16 types of neural network were established, and all types of neural network are trained 100 times. By comparing different training arithmetic, we knew that the OSS arithmetic is better than the others.(2) OSS arithmetic was chosen to train the neural network based on the result above, and the number of neurons in hidden layer is near 15.The Parameter of established BP neural network is :one hidden layer, 15 neurons in hidden layer, 8 neurons in input layer , 1 neuron in output layer. The Parameter of neural network training is: the learn rate is 0.01, the performance function of error is set as SSE,10 epochs was got when stopped train the neural network, The fitting result of train net is : R=0.83954,R2=0.70483,Radj=0.64783,RMSE=0.05022;The result of test net is : R=0.85523,R2=0.73142,Radj=0.71649,RMSE=0.04504.(3)Result of sensitivity analysis: Result of sensitivity analysis show that, the order of sensitivity value of all influence factors is: ag(e0.94897),length of hospital stay(0.16101),the condition of get out of hospital(0.15227),the number of rescue(0.14537),the number of hospitalization(0.09421),marital status(0.08733),fee type(0.06751),sex(0.01391),it is obvious that factors influence hospitalization charge the most is age, the least is sex。ConclusionForm our research, some conclusions were got: BP neural network can be considered for its impact on the cost of modeling factors, hospitalization expenses and the realization of the relationship between the fitting factors. Neural network fitting ability and generalization ability are not attained, if BP neural network fitted put variable very well, the generalization ability may be not good. Therefore, modeling should be based on the actual situation on the trade-off between fitting ability and generalization ability. A good neural network model should first have a good generalization ability, Otherwise, even if the capacity of even the best fitting sample can not promote, the model like this is of no significance. From sensitivity analysis by BP neural network model can be displayed on the basis of input variable on the output variable degree of influence. Since this study is limited choice of Medical Records Information, and taking into account the promotion of the network capacity and ability to approximate, therefore, the model prediction accuracy on the cost of limited, only can provide a theoretical reference. In this model, based on sensitivity analysis to measure the degree of influence the size of the input variables is feasible,Because different parameter settings on the network have different effects on the results,such as the number of neurons in different hidden layers, different initial weights and thresholds, still a lack of theoretical support, pending further study... |
PDF Full Text Request