Study On Nonparametric Statistical Methods For Comparison Of Two Treatments In The Presence Of Competing Risks

BackgroundCompeting risks data arise frequently in clinical trials.One of the most important problems encountered is the overall heterogeneity between two group,the existing methods lose power when the assumption of proportional SDH is violated,especially when two CIF curves cross each other.However,the existing methods can not be applied to two cases of SDH proportionately and disproportionately.It is suggested that one or a class test method can be applied to these two cases which all have higher test performance is particularly necessary and important.when the CIF curves of the treatments are known to cross,rather than comparison of the entire survival curves,interest is focused on the comparison at a time range.Comparisons at fixed points in time are often of interest when the CIF curves of the treatments are known to cross.ObjectiveIn this paper will proposed some new overall test methods which can be applied to the two cases of proportional and disproportionate SDH.proposed some new test methods to test the long-term homogeneity and test the heterogeneity at a fixed point time.Evaluate the strengths and weaknesses of the tests for a variety of situations and censoring rates and to recommend an appropriate test that will not fail for a wide range of applications.MethodsIn order to test the overall difference,an area test method based on the absolute value of the area difference between two CIF curves is presented In this paper.In order to further improve the adaptability of the area test method,the combination test and the two stage test method are put forward.In order to compare the difference of the effect on a special fixed point time.The navie test and other test methods have been put forward,the other tests were proposed which based on the logar'ithmic,double logarithmic,arcsine square root or logic transformation,their variance base on Gaynor variance.or Aalen variance.Five confidence intervals type were constructed which base on logarithmic,double logarithm,arcsine square root or logical transformation of CIF respectively.The pseudo value test method based on generalized linear models will be introduced.In order to compare the difference of CIFs between groups in a certain area(such as after the point of time).This paper presents some long term test methods to compare the long-term CIF,the statistics were structured by using linear combination which combined the information at a fixed time point and long-term information after a time point.The test to compare the long-term CIF by using Gray test,Pepe and Mori's test or Renyi test.ResultsIn this paper,the Perm.PM.KS and RY test have relatively low power when SDH proportionate assumption is satisfied,but Int,Perm and Comb test have relatively high power.In addition to the late convergence situation,the Int,Perm,Comb test power are low,the four methods are have a relatively high power in other situation.TS method is worth mentioning,it has better performance in any situation and has better adaptability.The tests based on Gaynor variance and Aalen variance at a fixed point time were expanded to competing risk data.The Pointloglog test based on Gaynor variance and the Pointarcsin test based on Aalen have more stable type I error and higher power,they were relatively better.The upper limit of the CIline may exceed the boundary.The simulation results show that the deviation of CIloglog test is closest to the expected constant 0,which is the most robust and reliable.The rest of the transformation methods were relatively conservative or relatively radical.The test method based on the absolute value of the area difference between two CIF curves has higher power,The LongGray test and LongRenyi test have relatively power.ConclusionBefore analyzing the data,we need to draw plot of CIF curve and make a proportional assumption of SDH at the same time.The LG1,LG2,GY,Li1,Li2 and Li3 test will be advised to use when SDH proportionate assumption is satisfied.The TS,Comb and Int proposed in this paper also can be applicabled.When the assumption is violate,the TS and Comb test should be used.In addition to the test for the entire follow-up period,also suggested that should pay attention to the long-term effect or short-term effect,and corresponding long term test methods will be suggest.This paper also recommend some special significance point time,the Pointlohlog test based on Gaynor variance and Pointarcsin test based on Aalen were recommended.