What to Estimate in Symptom Trials

• Activity Number: 290
• Type: Invited
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Thomas Permutt*
• Companies: FDA
• Keywords: estimand ; causal inference ; effect among the treated ; total direct effect ; missing data ; intent to treat

Abstract

In survival trials the standard practice is to find out which patients are alive at the end of the trial and when the others died, regardless of adherence to treatment. For trials of symptomatic conditions, a similar approach is sometimes recommended. Some practitioners would consider measurements after discontinuation of treatment to be irrelevant, but the question of what is relevant then has rarely been explicitly addressed. We propose to handle symptomatic trials as a distinct problem. If discontinuation has made further observation irrelevant, it may be because discontinuation is itself the relevant outcome. In this case, statistical methods are needed to analyze an outcome that is partly binary (continued or discontinued) and partly numeric (if continued); we propose some. Alternatively, notwithstanding the intent-to-treat principle, the effect in adherent patients may sometimes suffice. Careful definitions of this effect are required but have been given in the literature on causal inference. Unbiased estimation can be difficult but conservative approximations are available.