How effective is your blinding?

TL;DR

This paper explores statistical methods to assess progressive unblinding in clinical trials over multiple time points, providing tools to distinguish causes and quantify unblinding extent.

Contribution

It extends the blinding index approach to longitudinal data, introducing multiple statistical tests and a simulation method for assessing unblinding.

Findings

Generalized McNemar test for unblinding detection

Weighted least squares approach for unblinding analysis

Simulation method for unblinding assessment

Abstract

This report answers queries about extending the blinding index approach to a situation with measurements at multiple time points. The key question is how to test if there is progressive unblinding. A related question is how to apportion extent of unblinding between primary cause (trial design) and secondary cause (AE or efficacy). It is indeed possible to answer these questions. Sections 2 to 5 develop the narrative. Section 6 addresses the basic question about testing for progressive unblinding. The strategy is to use various statistical methods available in literature. One method is generalized McNemar test for marginal homogeneity. Second is a weighted least squares approach described by Stokes et al (2000). Third is application of polytomous logistic regression. Fourth and last is a simulation approach. This is the key part of the report. Other questions are answered in Section 7. Numerical illustrations are given based on hypothetical data (since a trial is yet to be conducted). Relevant program code is available if needed.