Study On The Approach For Pooled3-arm Data From Clinical Trials By Propensity Score

It is necessary to evaluate the efficacy, safety and pharmacoeconomics of adrug to verify its prospect of further development. With the development ofmedical technology, the drugs with confirmatory efficacy to a disease areincreasing, which makes it possible to adopt non-inferiority design. Now in China,placebo-controlled trial is adopted for dose-finding in Phase II and theactive-controlled non-inferiority design is considered for efficacy confirmation inthe third phase. But when the two-arm active-controlled non-inferiority design isadopted, a series of problems, such as the determination of non-inferiority region,the trial sensitivity, etc, have to be considered at first. The three-armnon-inferiority design including placebo arm and active arm has its ownadvantage and it is possible and economical to adopt it based on the availabletrial data.Therefore, it is feasible to adopt three-arm non-inferiority design bycombining the data from phase III and the data of placebo group from phase II.The reasonability of non-inferiority region, the efficacy of active drug and the reliability of the conclusion from phase III can be testified and the ethicalconsideration is satisfied. However, the randomization may be undermined if thedata from phase II and III are combined for analysis. The propensity scoremethod, which is one of the post-randomization methods, fits the trial design well.It is to keep the baseline balanced as much as possible after the study iscompleted.This article is to study the method of combining data from phase II and IIIdata for three-arm non-inferiority design by employing propensity score. Themain tasks and results are listed as following.1) A simulation study was conducted for continuous outcomes to explore theproposed method. Balanced covariates and higher test power are achieved whenthe difference of covariates is small. But when the difference becomes large, thecovariates cannot be balance well and the test power is weaken if there is strongrelationship between covariates and outcomes.2) The three-arm non-inferiority and two-arm non-inferiority designs werecompared for binary data by employing another simulation study. It is shown thatthe test power of three-arm non-inferiority design is a little lower. But its resultscan be explained and extrapolated better.3) It is shown by the simulation study for dichotomous variables that thewell balanced baseline variables, reasonable sample size ratio and higher testpower can be achieved in the proposed method.4) The proposed method was also testified by a real example. The result wasconsistent with the conclusion of phase III trial and the baseline covariates werewell balanced. It is worth being applied in practical clinical trials.The innovative points of this study includes:(1) the comparison of three-armand two-arm non-inferiority design for binary data, in which their test powers and sample sizes were compared;(2) the conception of the method of combined datafrom phase II and III trials by employing propensity score for three-armnon-inferiority design, which overcomes the disadvantages of two-armnon-inferiority design, including the poor reliability and more ethicalconsiderations, and enhance the applications of three-arm non-inferiority designand caliper matching of propensity score method;(3) the statistical evaluation ofthe proposed method for both continuous and binary outcomes.