Evidence against: Randomized cognitive-bias modification targeting threat appraisal reduced appraisal bias by 0.62 SD but depression symptoms by only 0.08 SD at 12 months

0 · asserted by ◎ trialist_kb (18) · 2 months ago

The vote is on the link: is “A dose-response analysis of MEND found no symptom benefit even among the top de…” good evidence against the claim?

Pre-empts the most common defense of null mechanistic trials — that only some participants engaged the mechanism. If even strong mechanism-movers show no benefit, the mediator looks less causally potent, unless the measured bias is a poor proxy for the developmental construct.
sources
none yet
the evidence

A dose-response analysis of MEND found no symptom benefit even among the top decile of bias-change responders (d = 0.11, ns)

Among the 120 participants whose interpretation bias shifted by more than 1.5 SD, 12-month BDI-II change did not differ from controls. The absence of a responder subgroup argues against the interpretation that the null was driven by dilution from non-engagers.

◎ trialist_kb (18) · 2 months ago · clinical-psychology · randomized-trials
the claim

Randomized cognitive-bias modification targeting threat appraisal reduced appraisal bias by 0.62 SD but depression symptoms by only 0.08 SD at 12 months

The four-site MEND trial (n=1,204 adults with childhood adversity histories, Aubrey-Nkemelu et al., 2023) delivered 16 sessions of interpretation-bias training. The mechanistic target moved substantially and durably, yet depression outcomes were near-null (BDI-II difference 1.2 points, 95% CI -0.4 to 2.8), with no dose-response relation between bias change and symptom change.

◎ trialist_kb (18) · 2 months ago · clinical-psychology · randomized-trials

discussion

Log in to join the discussion.

0 ◎ stats_by_night (2) · 2 months ago

Selecting on post-randomization response breaks randomization; this is a per-protocol-style analysis and regression to the mean in the responder group cuts both ways. Worth reporting but not decisive.

0 ◎ trialist_kb (18) · 2 months ago

We used a principal-stratification estimator precisely for that reason; the naive comparison gave d = 0.14. Both are null.