Bioequivalence Thresholds For Highly Variable Drugs And The Comparison Of Two Evaluation Methods For PK Parameters

Objective:In recent years,the bioequivalence of highly variable drugs?HVD?has attracted more and more attention due to the increasing number of applications for generic drugs.The first issue is the difficulty of bioequivalence assessment for HVD.The simplest and most effective way is to broaden the equivalence threshold.However,existing threshold can result in the inflation of type?error rate because of their mixed strategy.The second issue is that both of the two related primary endpoints,AUC and C_max,should be considered in BE.This is called multivariate BE.The purpose of this study is to explore the best solutions to the above issues.Methods:Firstly,this article proposed a new smoothing equivalence threshold,which based on the existing thresholds of EMA,FDA,and other researchers.Then,a simulation was conducted to investigate the statistical performance of the new threshold in power and type?error rate under various settings.As to the multivariate BE,different statistical methods of the two endpoints were applied.The method of confidence region was extended to the case of HVD,and then the method of joint modeling was applied to analyze the PK parameters.Finally,the merits and demerits of these methods and strategy of separate evaluation are demonstrated through simulation.Results:Firstly,the simulation of the equivalence threshold shows that BELn can maintain power higher than 70%in the case of 2*4 crossover design and GMR=1.1.For type?error rate,if CV=70%,BELn can reach 26.53%and 29.25%respectively in the case of 2*4 and 3*3 crossover design.The type?error rate of BELw in different situations is about 5%.When GMR=1.1 and CV=70%,the power of 2*4 and 3*3crossover design is 32.95%and 13.55%,respectively.When GMR=1.1 in 2*4crossover design,the power of BEL_EMA was above 60%,and the type?error rate was less than 15%in GMR=1.34.The power of the 3*3 crossover was slightly worse,and the error rate control of the first class was slightly better than that of the 2*4crossover at 70%coefficient of variation.In the 2*4 design of BEL_FDA,when CV=50%,the type?error rate reached 17.58%,which was higher than that of BELn under the same conditions;under the 3*3 design,the power is 14.87%when CV=50%.The second is the comparison of methods for simultaneous evaluation and separate evaluation.The trend of the three methods is consistent.The proportion of acceptance of each method decreases with the rise of GMR,and increases with the rise of correlation coefficient.And it decreases first,then increases and then decreases with the rise of CV.When GMR is 1,the power of the separate evaluation and the joint modeling are close to and above 90%,while that of the standard confidence region method is less than 80%.When GMR=1.1,the power of standard confidence region method is lower than that of separate evaluation method and joint modeling method when CV and correlation changes.When CV of the two drugs are not equal,the joint modeling method is slightly worse than the separate evaluation.Conclusions:BELw is too strict to be recommended.BELn shows the best performance,but it performs poor control over type?error rate when CV is high,which needs further improvement.The method of confidence region shows low power and is not recommended for application.Although the joint modeling method has theoretical advantages,it shows slightly lower power than the separate test.At present,the power and type?error rate of the separate test method are acceptable,and the statistical analysis is simpler,so the separate test is recommended.