Adjusting for perception and unmasking effects in longitudinal clinical trials

A blinded clinical trial design requires masking of patients to prevent measurement of their outcome from being influenced by knowledge of treatment assignment. However, during the course of a trial, some patients may be practically unmasked either due to experiencing treatment related side effects in the treatment arm, or lack of efficacy in the placebo arm. In a recent paper, we introduced concepts of perception, unmasking, and placebo effects for point treatment studies. In this paper, we generalize these concepts to longitudinal studies, and use recent advancements in causal inference and semi-parametric efficient estimation to define and estimate perception and unmasking effects. This allows differentiation of the impact on measured outcomes of 'early' versus 'late' unmasking. In particular, two semi-parametric, substitution methods, one based only on the prediction model (G-computation) and an augmented version of that model for targeted bias-reduction (Targeted Maximum Likelihood Estimation; TMLE), are used for estimation of perception and treatment effects. We motivate our discussion by analyzing data from a recent longitudinal study on the effect of gabapentin on pain among diabetic patients experiencing painful neuropathy.