Improving Statistics in License Applications: Handle with Sensitivity?

Abstract

The statistical quality of license applications would be improved if applicants were to ensure that key aspects of trial design, such as randomization, blindness, and power; and important aspects of analysis, including due accounting for withdrawals, and baseline covariate imbalance, were all adequate.

Special attention needs to be given to demonstrating that the conclusions drawn from the principal analyses are robust to the use of alternative methods of analysis and/or choice of data subsets. The contribution of each study to the argument and evidence presented in the application should be clear, and an integrated summary of the results of the principal studies should be provided.

In Europe, regulatory authorities should encourage better statistical quality of license applications by employing professional statistical assessors and by requiring statistical validation of expert reports.