Using Attentional Bias Modification as a Cognitive Vaccine Against Depression

Background Negative attentional biases are thought to increase the risk of recurrence in depression, suggesting that reduction of such biases may be a plausible strategy in the secondary prevention of the illness. However, no previous study has tested whether reducing negative attentional bias causally affects risk factors for depressive recurrence. The current experimental medicine study reports the effects of a computerized attentional bias modification (ABM) procedure on intermediate measures of the risk of depressive recurrence (residual depressive symptoms and the cortisol awakening response) in patients with recurrent depression. Methods Sixty-one patients with at least two previous episodes of depression who were currently in remission were randomized to receive either an active (positive) or placebo computer-based ABM regime. The ABM regime presented either pictures of faces or words. Residual depressive symptoms, measured using the Beck Depression Inventory and the cortisol awakening response were measured immediately before and after completion of the bias modification and then again after 4 weeks' follow-up. Results Positive, face-based ABM reduced both measures of recurrence risk (Beck Depression Inventory and cortisol awakening response). This effect occurred during the month following completion of bias modification. Word-based modification did not influence the outcome measures. Conclusions Positive face-based ABM was able to reduce intermediate measures of recurrence risk in previously depressed patients. These results suggest that ABM may provide a “cognitive vaccine” against depression and offer a useful strategy in the secondary prevention of the illness.

Stimuli Selection. Face stimuli were taken from a range of previously published sources (2)(3)(4)(5). Happy faces were used as positive stimuli, neutral faces as neutral stimuli and angry and fearful faces as negative stimuli. Positive, neutral and negative (including both physically and socially threatening) word stimuli were selected from a range of previously published sources . A total of 288 word pairs were created (96 pairs each of positiveneutral, negative-neutral and positive-negative words). The word pairs were split into 4 separate sets of stimuli as described in the main text, with each set containing 72 pairs (24 each of the three possible valence combinations). The specific words used in the pairs were selected to be identical in length. In addition the pairs of words used in each set were balanced in terms of Kucera-Francis Frequency [all t(71) < 1.6, p > 0.12].
In order to ensure that tests of attentional bias used novel stimuli, the lists of word and face stimuli were split into four separate sets with each of the three assessment sessions deploying a novel set of stimuli (the sets used were counterbalanced across patients). The stimuli used in ABM were taken from the fourth set combined with the set of stimuli used in the pre-ABM assessment.
Compliance with Bias Modification. Compliance with the bias modification interventions was encouraged by planning with patients when during the day they would complete the task (35). In addition patients were contacted by researchers on two occasions during the treatment fortnight to check that they were complying with the study protocol. The number of times a participant completed ABM was recorded automatically. For the purpose Browning et al. 2 of analysis, compliance with the ABM regime was defined as completing at least 25 out of a total of 28 sessions.

Cortisol Assay
Salivary cortisol was measured, blind to subject status, by an in-house doubleantibody radioimmunoassay with utilization of 125I-cortisol as the ligand. The intra-and inter-assay coefficients of variation were 4.1% and 7.8% respectively and the minimum detectable concentration was 0.2 nmol/L when a 0.1 ml volume was assayed.

Duration?
The analysis reported in the main manuscript described a time x ABM type effect on attentional bias as measured by the word based visual probe. This indicates that the positive and placebo ABM tasks had a differential effect on attentional bias. However the visual probe task included trials with three different pairs of stimuli valence (positive-neutral, positivenegative, negative-neutral) and two different durations (500 ms and 1000 ms) which makes it possible to test for more specific effects of the ABM task. For example, it allows assessment of whether the ABM tasks specifically altered attentional bias to brief (500 ms) as opposed to longer (1000 ms) stimuli presentations. This possibility was tested by calculating individual bias scores for each stimuli pair type and each duration. These were then entered into a repeated measures analysis of variance which included the between subject factors of ABM type (positive, placebo) and ABM stimulus (faces, words) and the within subject factors of time (before ABM, after ABM, follow-up), assessment stimuli valence (positive-neutral, positive-negative, negative-neutral) and assessment stimuli duration (500 ms, 1000 ms).
Differential effects of the ABM task on attentional bias would be detected in this analysis as a time x ABM type x assessment stimulus parameter effect. This analysis confirmed the reported time x ABM type effect [F(2,112) = 3.1, p = 0.05], and found no evidence that this effect was modified by any of the assessment stimulus parameters [all F < 1, p > 0.6]. This indicates that the ABM procedure produces a generalized alteration of affective bias rather than specifically altering bias to a subset of stimulus valences/durations. This is consistent Browning et al. 3 with the effects found in a previous study of healthy volunteers which used a similar ABM paradigm (1).

Did the Effects of ABM Depend on Whether Participants Became Aware of the Training Contingency?
The ABM task was originally designed as a method for implicitly modifying attentional bias (24). In other words, bias modification should occur even though participants are unable to report the training contingency of the task (i.e. that during positive ABM the probes replace the positive faces). However, some participants will inevitably notice the contingency while completing the task raising the possibility that the impact of bias modification may be accounted for by demand effects specifically in this "aware" subgroup of participants. In the current study, after completion of the final follow-up session, participants were asked whether they had noticed the training contingency. This question was relevant only to those in the positive ABM groups, as there was no contingency to detect in the placebo groups (and no illusory correlations were reported). Of the 16 participants in the positive face-ABM group only three noticed the training contingency, and of the 16 participants in the positive word-ABM group only 4 noticed the contingency (difference between groups using Fisher's Exact Test, not significant). The number of participants who detected the contingency was therefore too small to allow a direct statistical comparison between "aware" and "unaware" groups. We therefore assessed whether aware participants These results provide robust evidence, at least for the mood outcomes, that the impact of ABM was not dependent on participant awareness and therefore cannot be accounted for demand effects.

Did the Number of Sessions of Bias Modification Completed Influence the Results?
Rerunning all the analyses reported in the main paper with an additional covariate coding for the number sessions of ABM completed did not alter any of the reported results. In addition the number of sessions completed did not significantly predict any of the critical outcomes (i.e. change in Beck Depression Inventory, trait-Spielberger State-Trait Anxiety Inventory or cortisol awakening response during the follow-up period or change in attentional bias during the bias modification period) indicating that the effects of ABM did not depend on high levels of compliance.