Test Methods And Confidence Interval Constrcution Of Relative Risk Ratio In A Parallel Study With Missing Data

Missing data often appear due to various reasons in clinical trials.There are two methods: one is filling data according to existing data,and the other is making statistical inference with missing data.Statistical inference has been an attractive research question on parallel control data with missing data since filled data are different from original data.First,we propose several asymptotic methods including Wald-type,log Waldtype,likelihood ratio and Score tests.Those test statistics are compared in terms of type I error rate(TIE)and power.Monte Carlo simulations show that likelihood ratio and Score tests have relatively robust TIEs and satisfactory power.Finally,an example is given to illustrate proposed methods.Then,we propose exact methods including E method,M method,E+M method and C method for relative risk ratios.performance of these methods can be evaluated through TIE and power.Monte Carlo simulation results show that E+M method has more robust TIE and satisfactory power.Finally,a real example is used to illustrate the proposed exact methods.Finally,we construct five confidence intervals(CIs)of relative risk ratio such as Wald-type,log Wald-type,likelihood ratio and score.These methods are evaluated through empirical coverage,interval width,mesial and distal non-coverage rates.Simulation result reveals that Score CI is better than others.A real example is given to illustrate proposed CIs.