Should a clinical trial remain a fixed size, or should it adapt to fit a better estimate of outcome?

PSI One Day Meeting: Sample Size Re-estimation – dealing with those known unknowns! 02 Nov, 2016

Adaptive designs present an approach to re-estimate sample size within a confirmatory trial (Phase III), allowing the use of internal pilot data to reassess the sample size required to maintain both power and likelihood of observing a significant effect. The process improves confidence that an adequate sample size has been chosen to achieve study goals.

Standard methodology relies solely upon an estimation of the target difference (‘effect size’) and variance, often derived with little, if any, background knowledge. To address this, adaptive design methodology incorporates an assessment within the trial, at a fixed stage, drawing upon data gathered to calculate sample size using a more accurate estimation of target difference and variance.

The Workshop started off with Prof Chris Jennison of the University of Bath providing an introduction into sample size re-estimation in clinical trial. Rapidly the Workshop moved on to more complex issues encumbering the technique. Simon Day brought forward intriguing concepts of establishing not only the pilot sample size, but number of pilot studies to incorporate within a trial. This was complemented by Nikhil Chauhan’s talk on a variable pilot sample size and the opinion of the FDA in an oncology study. On top of this, the Workshop addressed the concept of Blinded estimation to prevent bias within the study, and difficulties modelling the separate populations.

If you’re interested in learning more about adaptive designs, I would recommend looking into Mehta and Pocock’s (2011, Statistics in Medicine) “promising zone” approach which featured highly across the Workshop. A copy of Prof Chris Jennison’s slides for the meeting is available at http://people.bath.ac.uk/mascj/talks_2016/CJ_PSI_slides.pdf