| BackgroundThe guidance on promoting the construction of clinics grading system is putted forward by the General Office of the State Council in2015,it is said that grading treatment policy system gradually becomes perfect,the diagnosis and treatment of primary health care institutions accounted for the proportion is significantly improved in 2017.Irrational allocation of health resources,unbalance of structure and rapid growth of medical expenses are one of the main contradictions in the current medical and health system in China.The grading diagnosis and treatment system at the present stage of our country is not perfect,and the function division of hospitals at all levels needs to be further clarified.In order to solve the current problems,the construction of Medical Union is a medical reform project which has been pushed forward by our country.According to the requirements of Guangdong Province in 13 th Five-Year of the medical system reform,by the end of 2017,grading treatment policy system gradually perfect,the diagnosis and treatment of primary health care institutions accounted for the proportion of the total treatment volume reached more than 65%,but the 2016 data show that grassrootsmedical institutions of Guangzhou City residents first diagnosed the situation needs to be further improved.ObjectiveTo understand the Guangzhou city residents currently attending the first choice of medical institutions and intention,based on a variety of model combined with the analysis of main factors influencing residents first diagnosed medical institutions intention,making the first primary diagnosis,to guide residents to better promote the research based on the background of the relevant departments,the implementation of grading treatment policies and measures to provide theoretical support.MethodsAccording to the actual situation of household survey and the difficulty of household investigation,stratified random sampling method was adopted to conduct random sampling survey to residents in Guangzhou.First of all,according to the cross-sectional study of random sample calculation formula,the sample was expected to survey 945 people,according to the 2016 GDP level,the nine district of Guangzhou city in 2016 GDP level is divided into three grades,a district were randomly selected from each grade,3 community health service centers and 3 hospitals randomly selected from each district,55 people were random surveyed in each medical institutions,925 people were effectively practical investigation.The contents of the questionnaire included the basic demographic characteristics,physical condition,the willingness of choose the first diagnosed hospital and the evaluation of the importance of hospital related factors.During data cleaning and analysis,first,statistical description,followed by the single factor chi square test analysis,attending the first choice of hospital types as the dependent variable,after single factor screening and combining theory with experience,independent variables,the first choice of hospital diagnosis types as the dependent variable,were selected into the model analysis.The other possible influencing factors are independent variables.According to CHAID,CRT decision tree,logistic regression analysis model and neural network model,we compared the advantages and disadvantages of several models,and integrated various models to find out the influencing factors of residents' first choice.Results925 residents were surveyed,including 434 men and 475 women.the average age was 35.9 ± 9.8,and age group was concentrated in18~40 years(64.3%).59%(544)of the people were married.More than a half were(494/53.4%)in the college and above group,and the highest proportion of occupation was the business and service(162/17.6%).80.1% of residents had medical insurance,and the monthly householdincome is concentrated in 2000~5000 yuan,with maximum percentage(380/42%).The results of the influencing factors are as follows :multi-classification Logistic regression analysis showed that education level,marital status,medical insurance and mean family income per month are the factors of residents' selection,CRT decision tree model showed that education level,hospital technology evaluation,age and medical insurance are the selection factors,neural network model displayed that medical insurance,monthly income,education level ranked in the top three,followed by medical technology,medical equipment,marital status.Considering the neural network model,logistic regression model and decision tree model,the important factors affecting the residents' first choice hospitals are the medical insurance and family income per month.The error estimates of the CRT model by the re-substitution method and the cross validation method are 0.577 and 0.596 respectively.The error estimates of CHAID by the re-substitution method and the cross validation method are 0.592 and 0.616 respectively.The fitting effect of CRT model was better than that of CHAID growth tree model,and the error estimate value was relatively small.According to the prediction accuracy,the correct percentage of the CRT model was 42.3% under the same model parameters in the decision tree series model.The prediction accuracy of neural network model was the highest ofall,the prediction correct percentage was 47.7%,followed by a logistic regression analysis model that the prediction correct percentage was 45%,the prediction correct rate of series of decision tree model is lower than the neural network model and logistic regression model,the accuracy range of decision tree models were from 40.8% to 42.3%.The effect of neural network model and logistic regression model on model prediction is relatively better than the decision tree models.ConclusionMulti-classification logistic regression model and neural network model show that the most important factors affecting the choice of the first diagnosed hospitals are the medical insurance and family income per month.The neural network model and the multi-classification logistic regression model are superior to the decision tree model in the prediction and prediction ability of the classification model. |
PDF Full Text Request