Primary feasibility outcomes
Recruitment was completed within a pre-defined timeframe (February 2019-January 2020). For baseline characteristics, see Table 1, and for patient flow through the study, see Fig 1. Of 156 people who expressed interest, 54 were screened and 45 (n = 34 female, n = 11 male) were randomly allocated to one of the trial arms (WL n = 14, MT n = 16, ABMT n = 15). Study retention rates were 86.7% (39/45) and 84.5% (38/45) at post-intervention and follow-up respectively. Session completion rates (defined as attending 8/8 sessions) were 87% and 94% respectively for the ABMT and MT groups. In the MT condition, one individual stopped the intervention after one training session. In the ABMT group, two individuals dropped-out after 3 and 2 sessions respectively.
Daily home practice of training was recommended (6 days/ week x 8 weeks = 48 home training sessions). MT participants completed the home practice 51/48 times on average versus 17/48 times on average in the ABMT group (i.e., 106% vs 35% of recommended practice respectively).
Secondary outcomes: Baseline to post-intervention (Week 8)
Means, standard deviations and effect sizes (including confidence intervals) of the clinical outcomes from baseline to post-intervention can be seen in Tables 2 and 3 for between-groups and within-groups respectively.
Weight and body composition
Between-group effect sizes for change in BMI were small from baseline to post-intervention. Within-group comparisons showed a small effect size on BMI reduction from baseline to post-intervention in all groups.
For body fat mass percentage, between-group effect sizes ranged from medium to small. There was a small effect size increase from baseline to post-intervention within the WL and ABMT groups. Within the MT group, there was a small effect size reduction in body fat.
Eating behaviour
There was a medium size difference on the EDE-Q global scores between WL and MT (favouring WL), and all other between-group comparisons were of a small effect size. Similarly, for EDE-Q binge eating episodes, all between-group effect sizes were small. Within-group analyses on the global score of the EDE-Q showed a large size decrease (confidence intervals do not cross zero) in the WL group (d = .90, 95% CI .03 to 1.76). There was a reduction on EDE-Q global scores with small effect size within both the MT and the ABMT groups. The number of binge eating episodes from baseline to post-intervention had a small to medium effect size reduction within all groups.
The total change score of the PFS (assessing psychological impact of food availability) from baseline to post-intervention showed a large effect size of d = .88 (95% CI .05 to 1.69) between WL and MT (confidence intervals not including zero), and a medium effect size between MT and ABMT, favouring MT in both cases.
Within-groups, there was a small effect reduction in PFS scores (lower scores indicate less hedonic hunger) from baseline to post-intervention in the WL group. This reduction was of a medium effect size for the ABMT group and of a large effect of d = .82 (95% CI .07 to 1.54) in the MT group (confidence intervals do not cross zero).
On a behavioural level, the change scores of food consumed (grams) during the Bogus taste test from baseline to post-intervention revealed a large effect size (confidence intervals do not include zero) between the WL and the ABMT groups (d = .87, 95% CI .02 to 1.71), favouring ABMT. However, the reduction of food consumption within intervention groups (MT and ABMT) was of a small size. Within the WL group, there was a small size increase of food intake.
Mood
Between-group effect sizes for mood symptoms ranged from small to large. The greatest difference was found in the state-anxiety scores measured by the STAI-S, showing a large effect size between WL and MT, favouring WL (d = -.77, 95% CI -1.60 to .07). Within-groups effects sizes for mood outcomes ranged from small to medium in all groups.
Mindfulness
Effect sizes for dispositional mindfulness measured by the MAAS ranged from small to medium between-groups. Within-groups effect sizes showed a small to medium increase in MAAS scores across groups.
Global MEQ scores from baseline to post-intervention showed a large effect size in the difference between WL and MT (d = -.84, 95% CI -2.31 to .69), favouring MT. Within both WL and ABMT, MEQ scores had a small increase from baseline to post-intervention and a large effect increase within the MT group.
Secondary outcomes: Baseline to follow-up (week 12)
Data on the clinical outcomes from baseline to follow-up are shown in Tables 4 and 5 for between-groups and within-groups respectively.
BMI
Between-group effect sizes for BMI change were small (baseline to follow-up). There was a medium size BMI reduction within the MT group.
Eating behaviour
For change in EDE-Q global scores (baseline to follow-up), the effect size between MT and ABMT was large (d = -.78, 95% CI -1.62 to .07), favouring ABMT. There was a reduction on EDE-Q global scores with small effect size within both the MT and ABMT groups. The number of binge eating episodes (baseline to follow-up) had a small to medium effect size reduction within both groups.
The total change score of the PFS (baseline to follow-up) was of medium effect between MT and ABMT, favouring MT. Decreases in hedonic hunger remained as a large effect size (d = .94, 95% CI .18 to 1.7) within the MT group.
Mindfulness
Global MEQ scores (baseline to follow-up) showed a medium effect size in the difference between MT and ABMT, favouring MT, with higher levels of mindful eating scores at follow-up. There was a large increase (confidence intervals do not cross zero) in global mindful eating scores within the MT group (d = -1.36, 95% CI -2.4 to -.29), and a small effect increase within the ABMT group.
Credibility and acceptability of training
Participants rated the credibility and usefulness of the training using VAS scales (0-100). The mean credibility score was 75.5 (SD = 13.3, n = 15) for MT and 71.3 (SD = 12.4, n = 12) for ABMT. The mean usefulness score was 70.1 (SD = 26.2, n = 15) for MT participants and 68.5 (SD = 18.3, n = 12) for ABMT. In terms of perceived benefits, 78.5% (n = 11/14) of MT participants and 81.8% (n = 9/11) of ABMT participants said they benefited from the training, with the remainder saying they did not. When asked whether they would recommend the training, 86.7% (13/15) of MT participants and 91.7% (n =11/12) of ABMT participants and said “yes”. When asked whether they would take up the training as a treatment if there was evidence of its benefits on eating habits and weight loss, 86.7% (13/15) of MT participants said “yes” compared to 100% of ABMT participants (n = 12/12). From people in the WL group, 54.5% (n = 6/11) opted for MT and 45.4% (n = 5/11) opted for ABMT after their 8-week waitlist period.
Attention bias change: an exploration of potential mechanisms underpinning the interventions
We explored the effects of the interventions on AB change from baseline to post-intervention within the whole sample (n = 39). As an additional sensitivity analysis, and based on the rationale that attention trainings would modify initial AB to High-Cal food cues, we also assessed AB change in those with an initial attentional approach (i.e., AB scores greater than zero) to High-Cal food cues at baseline (i.e., “approach” subsample, n=28). Results from the “approach” subsample mirrored the findings from the whole sample. Hence, only the whole sample data are described below. Means and standard deviations for all groups at baseline and post-intervention are shown in Supplementary Table 1).
Initial gaze fixation bias for high caloric food
We found a significant interaction of time x group [F(2,36) = 7.68, p = 0.002, partial η2 = .299] (Supplementary Figure 1.1) and a significant main effect of time [F(1,36) = 5.613, p = 0.023, η2 = .135], associated effect sizes were large. No main effect of group was found [F(2,36) = .207, p = 0.814, η2 = .011], and the associated effect size was small.
As a post-hoc analysis, we used paired-samples T-tests to look at the change of initial fixation bias to High-Cal food at baseline vs post-intervention within each group. Results showed a significant mean difference of 459.08ms (95% CI, 123.59 to 794.57) from baseline to post-intervention within the ABMT group [t(12) = 2.981, p = 0.011, d = 0.82] with a large effect size, indicating a reduction in AB to High-Cal food cues. However, no significant differences were found within the MT group, with a mean difference from baseline to post-intervention of -1.47ms (95% CI, -106.92 to 103.98), [t(14) = -.030, p = 0.977, d = 0.00]. As expected, the mean difference from baseline to post-intervention within the WL condition [-44.03ms (95% CI, -189.94 to 101.86)] was non-significant [t(10) = -.673, p = 0.516, d = -0.20], associated effect sizes were small.
Gaze duration bias to high caloric food
We found a significant time x group interaction [F(2,36) = 8.619, p > 0.001, η2 = .324] (Supplementary Figure 1.2), and a main effect of time [F(1,36) = 6.936, p = 0.012, η2 = .162], both associated effect sizes were large. No main effect of group was found [F(2,36) = 0.697, p = 0.505, η2=.037] with a small effect size.
A post-hoc analysis into the mean differences from baseline to post-intervention within groups in the whole sample revealed a significant mean difference in the ABMT group of 640.29ms (95% CI, 227.34 to 1053.24), [t(12)= 3.378, p = 0.005, d = 0.93] with a large associated effect size, revealing a significant decrease in AB to High-Cal food cues. The mean difference within the MT group [-27.39ms (95% CI, -211.98 to 157.19)] was non-significant [t(14) = -.318, p = 0.755, d = -.08]. Similarly, the mean difference from baseline to post-intervention within the WL group was non-significant [-14.00ms (95% CI, -201.09 to 173.07)], [t(10) = -.167, p = 0.871, d = 0.05], both associated effect sizes were small.
AB change and binge eating episodes
Based on the reduction in AB to High-Cal food cues within the ABMT group, an exploration of the correlational relationship between change in initial fixation bias to High-Cal food cues and change in binge eating episodes (EDE-Q, item 15) was made. Analyses revealed a significant correlation between change in AB to High-Cal food cues and change in binge eating episodes in the ABMT group [r(11) = -0.68, p = 0.01], i.e., the greater the reduction on attentional approach to High-Cal food cues, the greater the reduction of binge eating episodes at post-intervention. No correlations between AB change and binge eating episodes were found in the MT or WL groups.