Thread by @tmorris_mrc

The final ‘ICH E9(R1) Addendum on estimands and senstivity analysis in clinical trials’ was released on 3Dec2019.
database.ich.org/sites/default/…

Work on randomised trials? Talk about trials? Think about trials? This is IMO the most important document of the last two decades.
And, ‘While the main focus is on randomised clinical trials, the principles are also applicable for single arm trials and observational studies.’

No document is perfect and of course you will have quibbles, but if people take this seriously, it will be a really good thing for the quality of trials.

I’m not going to do a tweetorial on what it sayd because YOU REALLY SHOULD READ IT, but I'll mention two things good points and one bad point:

1. The glossary is precise about what they mean by ‘estimand’, which is not just ‘the true value of a parameter’ that some people use. Also about ‘estimate’ and ‘estimator’.

2. Missing data is explicitly listed as *not* being an ‘intercurrent event’ in itself. Intercurrent events do frequently induce missing data for certain estimands (if you wanted to measure outcome on treatment and someone stops).

The lowlight is that it never uses the word causal (the closest it gets is principal strata). I *think* this is the style of writing rather than a deliberate shun.
It would be a shame if this was missed by the causally minded.

BTW, if you are unfamiliar with these documents, they do not cite sources or list authors, which irritates some people. Hopefully knowing this in advance will soothe your irritation.
Anyway, enjoy reading!
/over