,
23 tweets,
5 min read
Read on Twitter
So: ITT estimands are of interest to patients too!
If patients have told you otherwise, you’ve done a bad job of explaining to them (or thinking yourself about) what an ITT estimand is.
1/
If patients have told you otherwise, you’ve done a bad job of explaining to them (or thinking yourself about) what an ITT estimand is.
1/
I've been pondering this for a while on the bike to work. @ProfMattFox has this way of asking great questions that generate discussion (I just discovered I don't even follow him! 🤔 thought I did. Now I do.)
This yesterday:
2/
This yesterday:
2/
I’ve heard people say (paraphrasing)
1) ITT estimands aren’t of any real interest
2) they’re just convenient in respecting randomisation
3) there are adherence-based estimands that respect randomisation and *patients are actually interested in those*
3/
1) ITT estimands aren’t of any real interest
2) they’re just convenient in respecting randomisation
3) there are adherence-based estimands that respect randomisation and *patients are actually interested in those*
3/
ITT effects are frequently touted as being of central interest if you’re making policy decisions. Whatevs. How is ITT of interest to patients?
4/
4/
I’ll give my perspective as a patient.
Adherence-estimand evangelisers can shoot me down if you think my reasoning is off.
(Tangent: ‘Estimand’ = ‘That which is to be estimated’; not ‘the [draft] framework’)
5/
Adherence-estimand evangelisers can shoot me down if you think my reasoning is off.
(Tangent: ‘Estimand’ = ‘That which is to be estimated’; not ‘the [draft] framework’)
5/
Tissues out tweeps:
A few years ago, I snapped my ACL and tore my meniscus in three places playing football (‘soccer’ in one country). I’ve since had surgery twice, which I initially wanted to avoid, and had to stop playing (because now the kids come first).
6/
A few years ago, I snapped my ACL and tore my meniscus in three places playing football (‘soccer’ in one country). I’ve since had surgery twice, which I initially wanted to avoid, and had to stop playing (because now the kids come first).
6/
When the orthopaedic surgeon was discussing surgery, I wanted to know how it might go compared with no surgery. He said ‘If I can reconstruct the ligament and you attend physio classes and do physio at home, you can expect to do great – much better than not having surgery’.
7/
7/
Three ifs.
So I asked how often a surgeon would be unable to fix the ACL and meniscus. He said very rarely.
8/
So I asked how often a surgeon would be unable to fix the ACL and meniscus. He said very rarely.
8/
I asked what physiotherapy would involve and how many people stop. He said it would be quite intense: about an hour-a-week at a class and lots of daily home exercises. A fair proportion of people stop.
9/
9/
He reassured me that I’d recover well as long as I did physio (but otherwise it might not be a great idea to bother with surgery).
10/
10/
That! Didn’t! Reassure! Me!
Do people who don’t adhere just do that for fun? Is it perhaps because physio is too painful? Too hard? Too tiring? Too time-consuming?
11/
Do people who don’t adhere just do that for fun? Is it perhaps because physio is too painful? Too hard? Too tiring? Too time-consuming?
11/
At the time, I didn’t know whether I would be able to adhere to physio. I didn’t want to know the effect of surgery+physio if I adhered to physio; I wanted to know the effect averaged across adherers and non-adherers (an ITT estimand).
12/
12/
(Actually I would have really liked to know *my* predicted adherence so that I could estimate my personal ITT effect averaged over that.)
13/
13/
An adherence-based estimand is good too. I got going with physio, found that I could adhere and then was reassured that things would likely go well (and yes it was painful, tiring and time-consuming, but I needed to be able to look after my kids).
14/
14/
It’s easy to say ‘What we really want is to know how the treatment works’.
Yes! And ITT is the most important question to a patient about how it works.
15/
Yes! And ITT is the most important question to a patient about how it works.
15/
The issue with an ITT effect is of course that it applies to people who were in a trial, but perhaps not to someone like you. If someone like you would always adhere, that’s interesting.
16/
16/
I wouldn't usually say stuff like the above but ITT averages over an interaction (the effect is always expected to be more impressive in adherers).
(@jdwilko and I wonder why we don’t put more energy into predicting adherence?)
17/
(@jdwilko and I wonder why we don’t put more energy into predicting adherence?)
17/
ITT effects are less generalisable than adherence-based effects. If your trial has recruited people who will be good adherers, the ITT effect doesn’t automatically apply to a population who would tend to adhere badly. An adherence-based effect might do.
18/
18/
You might say ‘trials describe what did happen not what would happen’, but I don’t think this argument works when you are intentionally averaging over an interaction. Then, being able to predict adherence becomes important.
19/
19/
People who say ‘ITT effects are the only thing of interest to policy makers’ should probably think about predicting adherence too. They could use predicted adherence to re-weight an ITT estimate.
20/
20/
In summary:
– People tend to say ITT is convenient for statistical reasons but isn’t of interest to patients.
– I think ITT should be of interest to patients but is inconvenient for statistical reasons.
21/
– People tend to say ITT is convenient for statistical reasons but isn’t of interest to patients.
– I think ITT should be of interest to patients but is inconvenient for statistical reasons.
21/
Penny for your thoughts?
22/
22/
