Cognitive bias modification (CBM) is a “hot” topic, both among researchers, as well as with clinicians and end-users. It was also the subject of several blog posts (here, here and here) and absorbing subsequent discussions at The Mental Elf.
The term is an umbrella name for experimental procedures, designed to promote the modification of specific cognitive biases considered critical in the onset and maintenance of psychopathology both in adults and in children. These procedures are usually tasks delivered via a computer, intended to teach participants, generally without their awareness, to either orient their attention away from threatening stimuli, or to interpret ambiguous situations in a positive or benign manner. In this way, negative biases are presumably modified or at least “de-activated”, which in turn should lead to improvements in negative mood and in symptoms.
In a recent trial reported in Clinical Psychological Science, Blackwell and colleagues describe a specific CBM paradigm previously developed by them: imagery CBM. This procedure involves repeated practice in using mental imagery for generating positive interpretations of ambiguous stimuli, with the goal of creating an adaptive “bias” that participants would then presumably apply in their everyday life, by automatically imagining positive resolutions to novel ambiguous stimuli. Blackwell and colleagues aimed to test imagery CBM as a potential low-intensity treatment tool for depression. The study was prospectively registered (clinicaltrials.gov identifier NCT01443234).
Participants were 150 adults (age range 18-65), who met criteria for a current major depressive episode, assessed with a semi-structured clinical interview. Participants were excluded if they met criteria for a psychotic or substance-abuse disorder, had a history of mania or hypomania, had started or changed antidepressant medication during the past month, were receiving psychological treatment or were involved in other trials of depression. Participants were randomly allocated to an intervention and control group.
The intervention consisted of 12 sessions, over a 4 week period, administered via the internet. Half of the sessions were in auditory form, and the other half in visual form. In each session, descriptions of ambiguous everyday situations were presented.
- In the intervention group, participants were asked to imagine these situations as vividly as possible and while the descriptions were initially ambiguous, their resolution was always positive.
- In the control group, in order to try to remove the mental imagery component of the training, participants were asked to focus on the meaning of the words in the auditory sessions and to generate sentences combining pictures with words in the visual sessions.
Across all sessions, half of the situations were resolved positively, and half negatively.
The primary outcome measure was change in depressive symptoms, as measured by the Beck Depression Inventory-II (BDI-II) at post-treatment. Apart from this, the investigators also assessed process cognitive measures: negative interpretative bias (using the Scrambled Sentences Test), vividness of positive future imagery (Prospective Imagery Test), as well as participants’ engagement and satisfaction with the intervention. Expectancies about the interventions were measured at baseline.
Post-treatment assessments were conducted by a researcher blind to group allocation. Participants also completed follow-up evaluations online at 1, 3 and 6 months. Participants were compensated for their completion of post-treatment and follow-up assessments.
The main efficacy analysis was conducted following the intent to treat principle, meaning that all randomized participants were included in the analysis. However, the authors also reported secondary analysis (i.e., per protocol) for those participants judged to have received an adequate “dose” of the intervention, which was defined as the completion of at least 6 sessions.
- Adherence rates to the program were similar for the two groups: 88% of the participants in the imagery CBM group and 93% of the those in the control group completed more than 6 of the 12 sessions.
- There were no significant differences between the two groups for change in depressive symptoms from pre- to post-treatment.
- Depressive symptoms decreased from pre- to post-treatment and follow-up in both groups.
- A similar pattern was reported for the cognitive process measures considered: negative interpretive bias decreased and positive future imagery increased in both groups, over time.
- Per protocol analysis using only subjects judged to have received an adequate intervention “dose” yielded the exact same pattern of results for depressive symptoms and cognitive process measures.
- Exploratory, post-hoc analysis looking only at the per protocol sample and focusing only on the anhedonia items of the BDI-II did reveal a greater reduction in the imagery condition compared to the control condition from baseline to post-treatment, even if this effect was not maintained during follow-up. Other exploratory analyses also found a superior efficacy of imagery CBM for participants with fewer than 5 depressive episodes.
The authors synthesize the results of their study:
We did not find the expected superiority of the imagery CBM over the control condition in reducing symptoms of depression (our primary outcome measure), reducing negative interpretive bias, or increasing vividness of positive future imagery. Participants in both conditions showed equal improvements on all these measures.
However, they also state:
These results from the trial highlight potentially fruitful pathways for further development of imagery CBM as a low-intensity treatment tool in the context of depression.
I appreciate that the authors highlight the strong points of their trial, which was conducted following the Cochrane Collaboration’s recommendations for minimizing risk of bias in RCTs, like the use of independent randomization, concealed allocation, blinding of outcome assessors, prospective registration and proper treatment of missing data. Power was computed a priori and was also adequate.
As the authors also note, this is in stark contrast with other trials of CBM interventions. A recent meta-analysis of CBM interventions (Cristea et al., 2015) also found that for depression the quality of included trials (translated in the number of criteria for which an included trial presented low risk of bias) was inversely correlated with the pooled effect sizes: trials with lower risk of bias found smaller effects of CBM for outcomes related to depression.
I remain dubious as to why, given all these observations, the authors chose to still put a positive spin on their results, and focus on exploratory, post hoc, sub-group or correlational analysis.
This adequately powered, low risk of bias, randomized controlled trial of a CBM intervention for properly diagnosed clinically depressed participants found no effects. From an empirical, science-based point of view, what is more likely? That some exploratory moderators, discovered post hoc, that had not been directly tested in previous prospective research hold the key to explaining the lack of significant effects? Or rather that these effects just aren’t there?
Maybe researchers in the CBM field should start making more use of Occam’s razor.
Blackwell SE, Browning M, Mathews A, Pictet A, Welch J, Davies J, Watson P, Geddes JR, Holmes EA, (2015) Positive Imagery-Based Cognitive Bias Modification as a Web-Based Treatment Tool for Depressed Adults A Randomized Controlled Trial. Clinical Psychological Science 3, 91–111.
Cristea IA, Kok RN, Cuijpers P. (2015) Efficacy of cognitive bias modification interventions in anxiety and depression: meta-analysis. Br J Psychiatry 206, 7–16. [Abstract]