All Times EDT
Keywords: Placebo response, SPCD, EM algorithm, Missing data, Imputation
In clinical trials, placebo response is a beneficial effect arising from the patient’s expectations concerning the treatment. Its presence makes the classical parallel study design suboptimal and can bias the inference. The Sequential Parallel Comparison Design (SPCD), a two-stage design was proposed to address the shortcomings of the classical design. In SPCD, a weighted average of Stage 1 relative treatment difference among overall population and Stage 2 relative treatment difference among placebo nonresponder was used as the efficacy measure. However, regardless of trial designs, studies with potentially high placebo response rates increase the attrition. Moreover, missing data can jeopardize the integrity of such multivariate combined efficacy measure even further: (1) It can affect the placebo response determination at Stage 1; (2) The quality of Stage 2 estimate is in question. Here, under the ignorable missingness and monotone missing data assumption, we propose an imputation strategy for both the latent placebo response status and the missing data using a modified Expectation maximization (EM) algorithm. The estimation of the forementioned SPCD efficacy measure is then based on both the observed and imputed data. An extensive simulation study is conducted to evaluate the proposed method under different missingness mechanisms as well as the impact to the occurrence of missingness in either stage of SPCD. We also apply our method to data from an actual SPCD trial of antidepressant therapy, the prior antidepressant therapy (ADAPT-A) study.