Parameter Estimation Of The Influence Of Recurrent Events On Covariates In Panel Count Data

Panel count data occur in studies that recurrent events as the discontinuous data commonly,and this kind of data has become one of the hotspots in the field of medical statistics in recent years.Therefore,it is of practical significance to explore the influence of recurrent events on covariates in panel count data.In this paper,the corresponding marginal or joint models are constructed for different types of panel count data,the performance of parameter estimation is improved by introducing some algorithms,and the accuracy is compared with other models.The parameter estimation with small deviation and the effective regression model are obtained.Finally,different data sets are used for application practically,and the difference between univariate and multivariate analysis is analyzed.The main conclusions are as follows:1.For the parameter estimation of single panel count data,it is found that the standard error of the model is minimum when the latent variables follow the Dirichlet process.We simulate the situation that latent variables follow different distributions,such as the normal distribution,Student t distribution,Chi-square distribution,Gamma distribution,and Weibull distribution.By comparing the standard error of parameter estimation,we can find that the standard error of latent variables following Dirichlet distribution is the smallest,indicating that the random selectivity of the latent variables is expanded in this case.2.For the solution method of parameter estimation in panel count data,the estimation accuracy can be effectively improved by combining EM algorithm with kernel estimation and MCMC algorithm.With the help of different marginal or joint models,the panel count data with different data backgrounds are analyzed,and the estimation value of the model is solved.It is found that this estimation method can avoid overfitting,and could optimize the estimation process in time and space.3.By exploring the difference between multivariate analysis and univariate analysis of multi-panel count data,it is found that multivariate analysis is more suitable to explore the impact of recurrent events on covariates and reflect the correlation among different types of recurrent events.But univariate analysis is simpler and more efficient for a given set of multipanel count data.Finally,through applying the real data of single panel COVID-19 data and multi-panel COVID-19 vaccination data,some instructive suggestions are obtained,such as the covariates of gender have the least influence on the frequency of hospitalization of COVID-19 patients,and potential diseases have the greatest impact on that.