Jointly Consider Treatment Switching And Patient Dropout In Randomized Clinical Trials With Time-to-event Data

When a new treatment is developed,a randomized clinical trial is usually conducted to compare the new treatment with existing ones to determine whether the new treatment is better or not.However,it has been reported widely that usually some patients switched from their original treatments to other groups due to various reasons(e.g.,disease progress,severe side effects,etc.).Based on humane considerations,experiment managers often allow patients to switch from one treatment to another.This is called treatment switching.In addition,some patients will abandon the original treatment plan and the new treatment plan during the trial,which is called patient dropout.Jointly considering treatment switching and patient dropout,this paper compares three different data analyzing methods through numerical study.The first method is to completely eliminate the data of missing patients as invalid data.The second is to keep the data before patient dropout,regardless of the situation of the patient dropout.The third method is to collect as much as possible the actual time of death of the dropout patient and determine whether it is censored.Patient dropout is treated similarly as treatment switching and added to the data analysis process,so that the patient's information in the trial can be saved to the greatest extent.This paper compares three data analyzing methods under two different frameworks.One is the parametric framework,where patient's survival time is assumed to follow the Weibull distribution.According to the actual situation,patients are divided into six groups: Control group,control group to experimental group,control group to dropout group,experimental group,experimental group to control group and experimental group to dropout group.Model parameters(including parameter on the efficacy of the testing therapy)are estimated using the maximum likelihood method.Then three methods are compared by numerical study in terms of the estimates as well as the mean squared error of the estimates they provided.The other is the non-parametric framework,which is more appropriate when the distribution of patient survival time cannot be determined.The KM survival analysis method and Log-rank test are combined to estimate the parameters of the new therapy,and the advantages and disadvantages of three data analyzingmethods are compared again through numerical study.