Effects of Attentional Bias Modification on residual symptoms in depression: a randomized controlled trial

Patients who had previously been identified as suffering from depression were randomised to receive two weeks of either ABM- or placebo training. Participants completed two sessions of ABM daily using laptops provided by the research team. A detailed calendar was used to specify the scheduling of the training sessions for each participant. Symptoms of depression and AB were assessed at baseline and immediately after the intervention.

The main recruitment base was an outpatient clinic in the Department of Psychiatry, Diakonhjemmet Hospital in Oslo. Participants were also recruited from other clinical sites, by local advertisements, and via social media. Candidates were pre-screened by phone for exclusion criteria before in person formal clinical evaluation and enrollment. Individuals diagnosed with current- or former neurological disorders, psychosis, bipolar spectrum disorders, substance use disorders, attention deficit disorder, and head trauma were excluded.

A total of 377 participants between 18 and 65 years old were recruited for clinical evaluation. Participants had experienced more than one previous episode of depression as defined using the Mini International Neuropsychiatric Interview, version 6.0.0. (M.I.N.I). A total of 56 subjects were excluded following the clinical evaluation. The period of recruitment and follow-up was January 2015 to October 2016 and was first posted in ClinicalTrials.gov in January 2016, and was therefore retrospectively registered as registration took place after the first participant was enrolled.

A total of 321 participants with a history of Major Depressive Episodes (MDE) were randomised to receive ABM or placebo (Table 1). After formal diagnostic evaluation, thirty-seven participants also fulfilled the M.I.N.I criteria for a current MDE. The study used an Intention to treat (ITT) approach, meaning that we included everyone who was recruited, including those people who were recruited incorrectly (which is a protocol violation).

The ABM task was a computerized visual dot-probe procedure adopted from [14]. Paired images of faces (the stimuli), were presented followed by one or two dots (a probe), which appeared in the same screen location where one of the two faces was previously exposed. Participants were required to press one of two buttons as quickly as possible to indicate the number of dots in the probe. The types of stimuli used during the ABM task were pictures of emotional faces of three valences; positive (happy), neutral, or negative (angry and fearful). A single session of the task involved 96 trials with equal numbers of the three stimulus pair types. In addition, there were equal numbers of trials in which the stimuli were randomly presented for 500- or 1000 ms before the probe was displayed. In each trial of the task, stimuli from two valences were displayed, in one of the following pairing types: positive-neutral, positive-negative, and negative-neutral. In the ABM condition, probes were in the same locus as the more positive/less negative stimuli in 87% of the trials, as opposed to 13% with probes in the same locus as more negative/less positive stimuli. Thus, when completing the ABM, participants should learn to deploy their attention toward the relatively more positive stimuli, without being explicitly aware of allocation to intervention, and in this way develop a more positive AB. The neutral ABM control (placebo) condition was identical in every respect, other than the probe location, that had the same locus as the more positive/less negative 50% -, and same locus as more negative/less positive stimuli in 50% of the trials. Participants completed two sessions of ABM at home daily during the course of fourteen days (28 sessions in total) on identical notebook computers that were set up and used exclusively for ABM-training.

The AB assessment task was identical to a single session of the placebo-training task but used novel face stimuli. Attentional biases (AB) was calculated as the difference in reaction time in milliseconds (RT) between trials in which the probe replaced the relatively more negative face vs. the more positive face ([(SUM (more positive face in upper screen position - locus of probe in lower screen position, more positive face in lower screen position, locus of probe upper screen position) – SUM (more positive face in upper screen position - locus of probe in upper screen position, more positive face in lower screen position- locus of probe in lower screen position)]/2.). Thus a more positive score reflects a greater bias towards the more positive stimuli. The ABM assessment task was conducted in the same lab environment for all participants.

An independent lab technician (not involved in the day to day collection of data) prepared training computers to deliver either ABM or placebo treatment according to a randomization list in a 1:1 ratio ensuring that allocation was concealed from all researchers involved in screening procedures and all participants. Allocation to intervention is stored in the ABM raw data file that was opened and merged with screening data for statistical analyses first after the data collection period was finished in October 2016. Outcome assessors and participants were therefore blind to allocation during the whole study.

The primary outcome measures were The Hamilton Rating Scale for Depression (HRSD) [15] and Beck Depression Inventory (BDI-II) [16] administered to assess both clinician - and self-evaluations of symptoms. Secondary outcome was AB measured by the AB assessment task at baseline and after two weeks follow-up. Comorbid anxiety symptomatology was screened by the use of The Beck Anxiety Inventory (BAI-II) [17].

All data were analysed using PASW 25.0 (IBM). The primary outcomes for the ITT sample was analysed by using repeated measures ANOVA with intervention (ABM versus placebo) as a fixed factor. Symptom at baseline and at two weeks follow-up (HRSD or BDI-II) was entered as within subject variable (time). To test the interaction between changes in symptoms and changes in AB, statistically significant symptom changes between ABM and placebo were followed up by repeated measures ANOVA by adding AB (two weeks follow-up minus baseline across valences and stimulus durations) as a second time factor in the model. Thereafter, the individual distribution of symptom change and AB was explored within groups with linear regression models.

A requirement of the national ethical committee was that study participants could elect to have all data stored on them during the study erased (i.e. including randomisation information) which precludes imputation of missing data. Of the 20 participants who did not complete the study following randomisation, 10 elected to have their data erased and thus data was imputed only for the remaining 10 remaining allocated to intervention. Thus, Thus the ITT analyses included 311 participants while participant with full data sets included 301 participants (see Additional file 1: Figure S1). A detailed description of procedures for data reduction, imputation and a post hoc analysis of the potential impact of noncompliance, stimulus valence- and duration, and sensitivity analysis are provided as Additional file 2.

Two weeks of ABM training significantly reduced blinded clinician-rated symptoms as compared to a control condition in a group of patients previously treated for depression and with various degrees of residual symptoms. Relatively more positive change in AB was associated with symptom improvement. ABM versus placebo did not differentiate changes in AB without taking symptom changes into account, meaning that a beneficial effect of ABM despite no intervention dependent change in AB was found. This means either than ABM is working by some other mechanism or that the measurement of AB is not very robust. Corroborating cognitive models of emotional disorders, the degree of symptom improvement increased with degree of relatively more positive bias across groups. AB change may therefore be a useful clinical measure since it is sensitive to individual differences in clinician rated symptoms, and may predict treatment outcome.

A larger intervention independent effect was also found for self-reported depression and self-reported anxiety. Post hoc exclusion of participants with high HRSD scores had no impact on the ABM effect, and further analyses showed that the results are robust against imputations and compliance rates.

In addition to being among the strongest predictors for recurrence in depressive disorder [4] residual symptoms also cause significant functional impairments, manifested in a variety of domains, including work and leisure activities [18]. The latter is important because one aim of depression treatment is to restore the patients’ previous level of functioning. However, residual symptoms in depression have traditionally not been the target of treatment trials. Given the recurrent nature of the illness, we suggest that this should change and that our study provides an example of how this may be done.

It is not clear why ABM versus placebo did not differ as measured by the BDI-II self-report scale, but instead showed a relatively large improvement in both the ABM- and placebo. In another recent study, fifty-two participants in a major depressive episode (MDE) was recruited for ABM training and a similar significant change in BDI-II in both the ABM- and placebo was found [19]. The lack of association between ABM and self-reported symptoms was not expected from the pre-registered hypothesis and interpreting the source of this lack of evidence is speculative. However, self-report may be more influenced by placebo or expectation effects. A general effect of the intervention may also be linked to cognitive training effects present in both training conditions. In contrast to general improvement in self-reported symptoms, increased depression was found within the placebo group for clinician rated depression. Further explorations of this pattern showed that a HRSD change was only statistically significant within the ABM group, indicating that the negative changes within the placebo group did not drive the observed ABM effect. The intervention may target the exact thing that worsens in the placebo group, but may also improve something else and thereby influence different domains that cancel each other out. Hereunder, clinical assessment might drive the participants’ attention towards depression symptoms and both positive- and negative assessment effects could therefore influence the change patterns in ABM and placebo. An assessment only group will help in distinguishing assessment effects- from ABM in future studies.

Symptom assessment based on both syndrome diagnoses, self-rated and blinded clinician-rated symptoms represent strength in the current study and the results clearly underpin the importance of comprehensive clinical evaluations in future research. The current study used a training task that earlier has shown to reduce self-rated symptoms first after 4 weeks [14]. The sample size (statistical power) and sample characteristics, like degree of residual symptoms, may explain why the current study found intervention dependent symptom changes immediate after training.

Participants completed the training in their homes, which is an advantage as it makes it more feasible as a treatment. Furthermore, compliance was high (see Additional file 2) which is important and contrary to the conclusion in a recent review of meta-analyses [11]. These authors conclude that ABM paradigms are most effective when delivered in the laboratory rather than at home. Specifying the scheduling of the training sessions individually might have increased motivation to do and focus on the task.

The results suggest that ABM does indeed exert an effect on blinded clinician-rated symptoms. The small effect sizes could be considered clinically non significant in treatment trials. However, it is not clear how to interpret clinical relevance of small HRSD score changes in this group. The HRSD and BDI-II is only moderately correlated and may partly reflect different depression constructs [20]. It is unclear whether the effect of ABM would increase and generalize across self-rated and clinician-rated symptoms if treatment continued for longer. Although residual symptoms are predictive of risk of relapse, it will be important to test in longer-term prevention studies whether the observed beneficial effect of ABM could translate into clinical relevance like reduction of relapse.

There is a pressing need to improve treatment and thus clinical trials should focus not only on efficacy, but also on identification of the underlying mechanisms through which treatments operate [21]. We observed a number of people with large changes in AB that go along with symptoms improvement, but also with no change in HRSD or even worsening of HRSD scores. Identification of this individual variability may be useful in evaluations of treatment efficacy and in personalized treatment. The current study reports the pre-registered primary outcome, but stratification based on degree of AB, changes in dispersion and/or distribution of AB, or paradigms sensitive to the temporal expression of AB may help advance basic knowledge on the conditional benefit of ABM [22].

The study has several limitations that should be mentioned. Sparse research on this population impede calculations of sample sizes from prior studies but would further increase the stringency of the study design in accordance with CONSORT guidelines [23]. Inclusion of 37 patients that also fulfilled the formal criteria for current depression represents a deviation from the pre-registered protocol. However, sensitivity analyses that excluded participants at the higher end of symptom scores did not explain the results and increases the generalizability of the reported findings. Residual depression was widely defined as the study did not assess symptom change during former treatment and classify all symptoms as residual independent of treatment response or time since the last episode. The stimuli used for ABM training- and AB assessment was adopted from earlier studies and this combination of stimuli was not cross-balanced or validated based on ratings of stimulus valence- and arousal.