Study On The Performance Of Restricted And Covariate-adaptive Randomization Methods

Background and objectiveBlock design is most commonly used in restricted randomization methods.In open trials,the risk of selectivity bias is easily caused by the guessability of block design.However,in addition to block design,there are many other restricted randomized designs that have not received much attention.This article hopes to provide practical guidance for the selection of restricted randomization methods through detailed discussion,and propose a new restricted randomization method.Among the covariate-adaptive randomization methods,stratified randomization and minimization methods are widely used,both of which are only applicable to categorical covariates,and stratified randomization is not suitable for situations where the number of categorical covariates is large.This article is expected to propose a new covariate-adaptive randomization method on the basis of exploring the performance of the minimal sufficient balance method that is generally applicable to continuous and categorical covariates.MethodsIn this study,we consulted the literature on restricted randomization methods to determine the randomness and balance index commonly used to evaluate the performance of randomization methods,and explore their relationship.The simple randomization and 13 existing restricted randomization methods are simulated and researched.Combined with the advantages and disadvantages of the existing methods,a new restricted randomization method of biased coin design with double limit is proposed.Consult the literature of the minimization method and the minimal sufficient balance method for simulation research.A calculation method that can measure the balance of baseline covariates between groups was extended to the randomization method,and a new covariate-adaptive randomization method was proposed,which is the standard difference method.ResultsBy exploring the characteristics of randomness and balance index,the conflict relationship between randomness and balance is further clarified.Of the eight restricted randomization methods without the maximal tolerated imbalance(MTI)limit,in terms of randomness and balance,only biased coin design,Ehrenfest urn design,and accelerated biased coin designs are stable under different sample sizes.When the correct guessing probability is close,the balance index of biased coin design is much worse than Ehrenfest urn design,and accelerated biased coin designs.Among the five restricted randomization methods with MTI limit,the degree of imbalance is within a controllable range,and the randomness index is mainly considered.The permuted block design has the worst randomness.Although biased coin design with imbalance tolerance,the larger the probability of biased coin can effectively reduce deterministic assignment probability,but the corresponding correct guessing probability is very large.For the other three methods,judging from the perspective of correct guessing probability,big stick design is the best;judging from the perspective of deterministic assignment probability,block urn design is the best;Considering correct guessing probability and deterministic assignment probability,biased coin design with double limit performs best by choosing appropriate parameters.Both the minimal sufficient balance method and the standard difference method can control the balance of the baseline covariates and the randomness of the assignment by adjusting the control biased coin probability and the baseline covariate imbalance control limit,proper trade-offs are needed to ensure that the baseline covariates are well balanced between groups and that the randomness of the assignment is not too badConclusionsEhrenfest urn design and accelerated biased coin designs show good randomness and balance by selecting appropriate parameters,and perform stable under different sample sizes.The existing big stick design and block urn design are good randomization methods;biased coin design with double limit can flexibly control randomness and balance,it is also a good randomization method.In order to reduce the selective bias of trials,block design should no longer be a recommended method for open clinical trialsBoth the minimal sufficient balance method and the standard difference method can take into account the balance of the baseline covariates and the randomness of the assignment by adjusting the control biased coin probability and the baseline covariate imbalance control limit,it can make up for the shortcomings that stratified randomization and minimization methods are not suitable for continuous baseline covariates.