Hurdle Count Model And Its Medical Application

Hurdle count model is an effective method for zero-inflation count data with covariates.The basic idea is to separate zero observations from positive counts, to fit binary regression model to zero or nonzero events,if nonzero events happen, we consider that has crossed the hurdle, then fit the number of events with zero-truncated count models.Therefore, the Hurdle model can be considered the combination of binary regression model and the zero-truncated count model.The first chapter of the article introduced the principle of zero-truncated count model, parameter estimation and model application. Further confirmed by simulation, zero-truncated count distribution data only fitted the corresponding basic count model, Poisson and negative binomial regression, which may get the biased parameter estimates. Zero-truncated count model could not only solve the issue of zero-truncated count distribution, but also the parameter estimates were more accurate, the fitting results were more reasonable. An analysis of the influence factors on the number of joints of patients with osteoarthritis also showed that the zero-truncated Poisson model was the optimal model to fit zero-truncated count data without overdispersion data, parameter estimates were more reasonable.The second chapter expounded Hurdle model, parameter estimation and model selection methods. Because Hurdle model combined binary model and zero-truncated model, the parameter estimates could be got by the respective maximum likelihood estimations of the two components. When choosing zero-inflation count data models, if the models were nested, using likelihood ratio test; if the non-nested, you need to use Vuong test. In addition, we should consider the goodness of fit, professional knowledge. The influencing factors analysis of the numbers of hospitalization showed that the data not only existed the phenomenon of zero inflation, but also existed overdispersion problems. By comparing logit-Poisson Hurdle (PH) and logit-NB Hurdle (NBH) model, NBH was the best choice for overdispersion and zero inflation count data.The attraction of Hurdle model is that it can find the different effects of explanatory variables between the two processes. It reflects a two-stage decision-making process in human behaviors. In this paper, the influencing factors analysis of the number of hospitalization which screened different covariates through logit process and zero-truncated counting process showed that whether the sick patient chose hospitalization, they considered the personal health and economic situation as well as other personal factors. As the number of hospitalizations, they not only considered the individual situation of the residents, but also taking into accounting the hospital information, the study really reflected the individuals'two-stage decision-making in medical services process. Therefore, for such count data, Hurdle model can provide more realistic and wealthy information.The Hurdle model can not only effectively deal with the zero inflation count data but also can be used to zero deflation. However, no matter how the ratio of zero count, in order to get a good parameter estimation in Hurdle model, sufficient sample size is necessary.