To Analyze The Influencing Factors Of Annual Hospitalization Frequency Of Residents Based On Zero Inflation Model

Objective: By fitting the zero-inflation optimal model,this study screened out the main influencing factors of annual hospitalization of residents in Yuncheng City,Shanxi Province under the background of the construction of medical alliances,so as to provide empirical reference for improving the health status of residents at the current stage and improving the health service level of township hospitals.Methods: A multi-stage stratified random sampling method was used in the baseline survey to select Yuncheng city in Shanxi Province,a typical city with innovative primary care service mode as the research site.Software Epidata3.2 was used for data entry and correction,and software Stata MP17.0 was used for multivariate analysis.The optimal model was obtained by constructing four models: negative binomial regression,Possion regression,zero-inflated negative binomial regression and zero-inflated Possion regression.Results: There were 4870 cases(88.18%)of zero hospitalization in the past year.The over-dispersion test statistics showed that the fitting degree of the negative binomial regression model was better than that of the Possion regression model,and the zero inflation test statistics showed that there was zero inflation in the data.The Akaike information criterion(AIC)and Bayesian information criterion(BIC)of zero inflated negative binomial were the smallest among the four models,which were 4269.322 and 4455.089,respectively.The results of model analysis showed that female,all age groups,college and above,4-6family members and living in urban area were the negative influencing factors of annual hospitalization.Chronic disease was a positive influencing factor for annual hospitalization.Age,family members of 4-6 or more than 7,living in urban areas,suffering from chronic diseases and non-smoking were the negative influencing factors for the zero-inflation phenomenon of hospitalization frequency.Non-farmer occupation and health knowledge obtained from non-medical institutions were the influencing factors of zero inflation of residents’ hospitalization times positive influencing factors(P<0.05).Conclusion:The zero-inflated negative binomial regression model had the largest log-likelihood(LL)and the smallest Akaike information criterion(AIC)and Bayesian information criterion(BIC)for annual hospitalization times of residents in Yuncheng City,Shanxi Province.The zero-inflated negative binomial regression model not only increased the goodness of fit when fitting this type of data,but also increased the goodness of fit.At the same time,the role of each factor in the Logit part or negative binomial part was clarified,and the fitting effect was the best.People with average annual household income or chronic diseases had more health seeking behaviors.People who were older,had 4-6 family members,lived in urban areas,suffered from chronic diseases and were regular smokers had higher potential demand for hospitalization.