Thread by @lluaces

I think controls needs to be thought of in a more intentional way. It’s not “waiting lists are bad even if you call it care as usual” and “always use a psych placebo.” To be clear, the absolute response or remission rates are not higher in studies using

https://twitter.com/tylerblack32/status/1399395251505217538

2/ weak controls or that the outcomes in psychotherapies behave differently in these trials. It’s that the more inert the control is the bigger the treatment-control difference is. This is not (just) a pedantic point because there are reasons to think carefully about the controls

3/ group depending on what the goals of your study are. Do you want to quantify the relative effects of introducing exposure to the common factors? Then yeah, to make a strong claim about this you would likely need to a bona fide supportive therapy control. But,

4/ let’s say you wanted to study the uptake of internet-based therapy. Then, a waiting list could be an appropriate control. Similarly, you may be interested in whether treating common mental disorder symptoms reduce absenteeism. Why would you need an attention-matched control?

5/ There’s also the question of treatment development. Now, I don’t think it’s good to invest in treatment development in depression but IF you were going to develop a new treatment there’s an honest question about what the control should be. If you don’t have any

6/ comparative outcome data why would you want to compare against an established treatment. People often do this powering trials for medium differences and when they don’t find them claim equivalence. But of course, we know to claim non-inferiority with small

7/ margins you need relatively large samples. Why invest in a study like this with no evidence? Now a problem is that people will conduct a study with a “weak control” at a time in the research base that makes sense BUT then make claims that don’t match the strength

8/ of the evidence. This is on researchers as well as readers to not get all excited about d=2.5 when the control condition is weak (and N is inevitably low). Also, I don’t think the meta- analytics evidence supports the idea that care as usual makes therapy seem more effective

9/ than it is. It tends to be associated with better outcomes than waiting lists although confusingly I’ve seen articles use the term care as usual in setting where no treatment is provided so it’s effectively a waiting control in those scenarios. (But can sometimes be therapy or

10/ medications.) So I don’t disagree with the spirit of this post but I would say that ultimately you need to think about the research question. If you care about isolating the efficacy of specific mechanisms then you wanna control for common and placebo effects

11/ but I don’t like this is required for all of our research questions. some references sciencedirect.com/science/articl… and researchgate.net/profile/David-…