Thread by @IoanaA_Cristea

My new & very likely most important paper of 2020 @TheLancetPsych #network #metaanalysis on the efficacy of cognitive bias modification for #anxiety or #depression in clinical or subclinical patients
(1st author @liviu_fodor, w/ @pimcuijpers @Toshi_FRKW )
thelancet.com/journals/lanps…

People who follow this field know I have been one of its first and unrelenting critics, starting with my meta-analysis in 2015 @TheBJPsych (again w/ @pimcuijpers) cambridge.org/core/journals/…, followed by others, including many viewpoints, comments & replies that I will spare you.

I was one of the first researchers to have explicitly state that maybe #CBM just does not work and there are no moderators or boundary conditions that can save it. But time passed, studies went on, the field continued to grow & adapt, and so maybe it was time for a reassessment.

In this network meta-analysis (NMA), we included all randomized trials using one of the main forms of #CBM (attention bias modification, interpretation bias modification or approach and action training) in individuals with clinical or subclinical problems of anxiety or depression

We only focused on clinically relevant outcomes of anxiety or depression, assessed separately in trials targeting anxiety or depression. Only outcomes on validated scales were included, & clinician-based were favored over self-report.

NMA is a sophisticated methodology allowing the statistical comparison of treatments that were never compared directly. We were interested in evaluating if some forms of CBM work better than others, but also if some types of control work better than others.

We included the largest number of trials up to date in any meta-analysis of CBM (65 for anxiety & 20 depression). Our search was updated as of February 2020, so this is not yesterday's news. The findings confirm in large part our 2015 @TheBJPsych meta-analysis, but more nuancedly

In a well-connected network of anxiety trials, interpretation bias modification outperformed waitlist (SMD −0·55, 95% CI −0·91 to −0·19) & sham training (SMD −0·30, −0·50 to −0·10) for anxiety (primary outcome), w/ similar results for comorbid depression (secondary outcome)

Though results appear promising, prediction intervals (showing how effects might fluctuate in future, similar, essentially replication, trials) all included zero, meaning there are also chances of not finding effects.

Attention bias modification did not show benefits over any type of control, except in post-hoc sensitivity analyses excluding trials on #PTSD. The attention control condition was hypothesized to help in #PTSD, but a recent trial does not support this: onlinelibrary.wiley.com/doi/abs/10.100…

For depression, networks were often inconsistent, probably owing to the limited number of trials. Cognitive bias for interpretation did appear better than waitlist for the primary depression outcomes, but we did not feel confident in drawing any conclusions.

Regarding trial quality, things have improved very little since my assessment in 2015 @TheBJPsych: only 4 trials could be rated as low risk of bias across the board (& this time we also rated selective outcome reporting, where 12/85 trials were rated low risk).

The argument that CBM trials are laboratory studies & so risk of bias doesn't count or should be rated differently (unclear how) does not hold. We included only RCTs on clinical & subclinical patients. The field really needs to start taking methods & reporting seriously.

Implications? 1. CBM for interpretation showed modest but consistent benefits, particularly given the comparison was sham training, a strong type of control. A large trial may be warranted, provided it is adequately pre-registered, planned (w/ blinding), analysed and reported.

2. ABM needs to go back to the laboratory. Actually, my personal hunch is that even pre-clinically, investment in ABM is questionable. It was neat, it was elegant, it was too good to be true and in fact it looks very likely it isn't really true.

More generally, a new paradigm of developing psychological treatments mechanistically (≠historically founder-based psychotherapies) has been gaining traction (eg, thelancet.com/journals/lanps…). CBM is a prototypical example, a tx directly from the lab. Translation so far is limited.

Thoughtful invited comment @Blackwell_SE, who unlike me, also conducts primary research in this field: thelancet.com/journals/lanps…

Finally, some meta-information.
Protocol registration (but we did make changes described in the supplement): crd.york.ac.uk/prospero/displ…
Data and code (forgive us we r @Stata users): osf.io/b2jwy/?view_on…
Preprint (pre-review version, will post new one): papers.ssrn.com/sol3/papers.cf…

The review process @TheLancetPsych was one of the most comprehensive I ever got (4 reviews totalling almost as much as the paper). It required a huge amount of work, included incredibly kind words & made the paper significantly better. (COI: I am an editorial board member)