Novel statistical methods for multi-stage designs in clinical trials with high placebo response
Embargo Date
2022-03-17
OA Version
Citation
Abstract
Placebo response occurs when a patient perceives an improvement from the psychological effect of receiving treatment rather than from the therapy itself. High placebo response reduces drug-placebo differences and makes it challenging to demonstrate a statistically significant benefit of an active drug over placebo. Two-way enriched design (TED) and sequential enriched design (SED) are two designs to estimate treatment effect with the existence of placebo response. They are extensions of sequential parallel comparison design (SPCD) and have multiple stages with enrichment strategies. This work aims to propose novel analysis methods and evaluate their performances in the framework of TED and SED. TED is a two-stage, randomized, placebo-controlled design with enrichment in 'placebo non-responders' and 'drug responders'. We first consider the placebo non-response as a measurable binary characteristic, either present or absent in an individual. We then discuss the placebo non-response as a characteristic that exists in every subject to a certain degree. We propose to include it in the model as a weight. In addition, we consider placebo non-response and drug non-response as latent characteristics and introduce stochastic components in the classification of the subjects in the setting of TED. SED, as the only three-stage design, aims to exclude subjects who are 'placebo responders' and those who never respond to either treatment. Considering the complexity of this design, we critically appraise the performance of SED from different perspectives. We first test the robustness of SED by varying values of parameters. We then calculate the actual sample size and the proportion of the target population in the sample. We also apply the first two analysis methods to SED. We evaluate these novel methods on a wide range of simulated data scenarios in terms of type I error, mean squared error, and power. From the appraisals above, SED does not benefit from the additional stage. Therefore, in terms of design, we suggest implementing TED rather than SED when placebo response is a critical issue. In terms of the proposed analysis methods, the approach with stochastic components performs the best based on our evaluations, especially when the definition of response is uncertain.