Study On Statistical Methods For Comparison Of Two Treatments With Censored Data

BackgroundDifferences in treatment benefits between two groups is one of common topic in clinical follow-up studies of time-to-event data.For the right censored data,when the proportional hazards assumption is violated,especially when two survival curves cross each other,the exiting tests for overall effects are difficult to intuitively quantify the treatment benefit,and their power are too low to assess the treatment benefit.Therefore,it is very important to propose a proper test without above assumption.And there are similar problems for the interval censored data.ObjectiveTo propose a new method for overall effects without depending on the proportional hazards assumption and extend it to interval time under right censored data,and extend the RMST method to the interval censored data.We assess the effectiveness and robustness by Monte-Carlo simulation to propose the robust methods and analysis strategy for clinical researchers.MethodsFirstly for the right censored data,we introduce an intuitive measure called area between two survival curves to measure the treatment benefit,and propose a permutation test based on it.We assess it statistical performance by Monte-Carlo simulation,and compare it with common Log-rank test and other tests.And two examples are applied with the several measures.Then,we extend the RMST method to the interval censored data,whose effectiveness and robustness is verified by simulation and an example.ResultsThe proposed test is a robust statistical inference method with controlled type I errors and high power in different simulation situations from simulated right censored data,in other words,it has regulated false positive error rates and superior ability to detect differences in treatment effects.Combine with two examples,the several tests perform well under the proportional hazards assumption,and the proposed test can yield better statistical inference results when the proportional hazards assumption is violated and especially when two survival curves cross each other.For interval censored data,the type I errors of extended RMST test fluctuate around the pre-specified nominal level 0.05,and its power are approximately equal to the common Sun model.ConclusionThe ABS can be used to intuitively quantify treatment differences over time for right censored data.The ABS permutation test is a robust statistical inference method with a regulated false positive error rate and superior ability to detect differences in treatment effects,thereby providing reliable conclusions in complicated situations,such as crossing survival curves.In the comparative study of two groups for interval-censored failure time data,the proposed extended RMST test not only can estimate the survival rates precisely for every group,but also can illuminate the size of intergroup difference intuitively by calculating Restricted Mean Survival Time in each group and make statistical inference,which has robust statistical performance and provide evidence for clinical researchers or patients to make decision.