Negative and positive affective states are more labile during adolescence (10–24 years; Sawyer, Azzopardi, Wickremarathne, & Patton, Reference Sawyer, Azzopardi, Wickremarathne and Patton2018) compared toin adulthood (Bailen, Green, & Thompson, Reference Bailen, Green and Thompson2019; Green et al., Reference Green, van de Groep, Sweijen, Becht, Buijzen, de Leeuw and Crone2021; Griffith, Clark, Haraden, Young, & Hankin, Reference Griffith, Clark, Haraden, Young and Hankin2021). Dysregulated affect is a core characteristic of common mental health disorders, including depressive and anxiety disorders. Regulation of affective states depends on the deployment of situationally appropriate emotion regulation strategies (Silvers & Guassi Moreira, Reference Silvers and Guassi Moreira2019). The ability to select adaptive regulatory strategies depending on contextual demands (e.g. engaging in action-focused strategies is not adaptive when the situation is not changeable) relies on affective control. Affective control refers to the application of cognitive control in affective contexts. That is, the capacity to inhibit affective information that is in conflict with current goals, while attending and responding to goal relevant environmental and internal inputs (Schweizer, Parker, Leung, Griffin, & Blakemore, Reference Schweizer, Parker, Leung, Griffin and Blakemore2020b). Affective control is still developing during adolescence (Aïte et al., Reference Aïte, Cassotti, Linzarini, Osmont, Houdé and Borst2018) and might constitute a promising target for prevention and early intervention (Schweizer, Gotlib, & Blakemore, Reference Schweizer, Gotlib and Blakemore2020a). Here, we explore the preventative potential of affective control training (AffeCT) for reducing mental health symptoms in adolescents.
Computerised AffeCT has been shown to improve emotion regulation capacity and mood in healthy and clinical adult samples (Krause-Utz et al., Reference Krause-Utz, Walther, Schweizer, Lis, Hampshire, Schmahl and Bohus2020; Lotfi et al., Reference Lotfi, Ward, Ayazi, Bennett, Larson and Lee2021; Pan, Hoid, Gu, & Li, Reference Pan, Hoid, Gu and Li2020a; Pan, Hoid, Wang, Jia, & Li, Reference Pan, Hoid, Wang, Jia and Li2020bb; Veloso & Ty, Reference Veloso and Ty2021). Neuroimaging evidence has shown that these training-related affective benefits are associated with increased recruitment of the cognitive control network, particularly the ventrolateral node (Schweizer, Grahn, Hampshire, Mobbs, & Dalgleish, Reference Schweizer, Grahn, Hampshire, Mobbs and Dalgleish2013). This region, the inferior frontal gyrus of the ventrolateral prefrontal cortex, is recruited more frequently during affective control when compared with neutral (cool) cognitive control (Schweizer et al., Reference Schweizer, Satpute, Atzil, Field, Hitchcock, Black and Dalgleish2019b). The cognitive control network develops throughout adolescence, with the inferior frontal gyrus being one of the structures to show the latest structural maturation (Dong, Margulies, Zuo, & Holmes, Reference Dong, Margulies, Zuo and Holmes2021). Training affective control during adolescence might be especially advantageous as brain development is experience-dependent (Frankenhuis & Walasek, Reference Frankenhuis and Walasek2020). Training could offer additional opportunities to apply affective control during a stage when this capacity and its underlying neural substrates are still developing.
Before discussing the AffeCT used in the current study, it should be noted that ‘cool’ cognitive control training, where cognitive control is trained on valence-neutral tasks (including, digits, letters and other valence neutral stimuli), has also been shown to benefit mental health in both adolescents and adults. For example, working memory training (with no affective component) was shown to reduce the onset of depressive symptoms in a school-based sample of adolescents (Beloe & Derakshan, Reference Beloe and Derakshan2020). Research with adults has shown that cognitive control training improves both negative affect and affect regulation (Calkins, McMorran, Siegle, & Otto, Reference Calkins, McMorran, Siegle and Otto2015; Hoorelbeke, Koster, Demeyer, Loeys, & Vanderhasselt, Reference Hoorelbeke, Koster, Demeyer, Loeys and Vanderhasselt2016; Hoorelbeke, Koster, Vanderhasselt, Callewaert, & Demeyer, Reference Hoorelbeke, Koster, Vanderhasselt, Callewaert and Demeyer2015; Onraedt & Koster, Reference Onraedt and Koster2014; Owens, Koster, & Derakshan, Reference Owens, Koster and Derakshan2013; Siegle, Thompson, Carter, Steinhauer, & Thase, Reference Siegle, Thompson, Carter, Steinhauer and Thase2007). Considering the efficacy of cool cognitive training, is there a need to train affective control? Preliminary evidence suggests so: Affective control has been shown to be uniquely associated with clinical endpoints that are central to the onset, relapse and maintenance of depression in young people, such as rumination (Hilt, Leitzke, & Pollak, Reference Hilt, Leitzke and Pollak2014, Reference Hilt, Leitzke and Pollak2017; Hilt & Pollak, Reference Hilt and Pollak2013). Training affective control, therefore, may confer benefits to mental health over and above those observed for cool cognitive control interventions.
In the present study we tested AffeCT (Schweizer et al., Reference Schweizer, Auer, Hitchcock, Lee-Carbon, Rodrigues and Dalgleish2022), which trains affective control on an affective working memory task. The training requires participants to continuously update affective information (words and faces) in working memory. AffeCT includes three tasks, single modality n-back tasks (one auditory, the other visuospatial) as well as a dual n-back task in which the two modalities have to be attended to simultaneously. The n-back paradigm requires individuals to indicate whether the current stimuli they are seeing and/or hearing are the same as those presented a specified number of trials ago (i.e. n-back). Together, these AffeCT tasks train engagement with task-relevant affective information (auditory modality) and disengagement from task-irrelevant affective properties (visuospatial modality), or both.
To evaluate the efficacy of AffeCT in improving our clinical outcomes of interest (mental health, affect and emotion regulation), we adopted the ‘Science of Behaviour Change framework’ experimental medicine approach (Nielsen et al., Reference Nielsen, Riddle, King, Aklin, Chen and Weber2018). The premise of the framework is to evaluate interventions by identifying a target mechanism, in this case affective control, and to investigate whether change in the target mechanism drives change in the clinical outcome of interest. To this end, a reliable assay of the target mechanism is required. Affective control here is operationalised as including three different facets: affective inhibition, affective shifting and affective working memory (Schweizer et al., Reference Schweizer, Gotlib and Blakemore2020a). Any training that successfully improves affective control might lead to improvements in any, or all, of the facets of affective control. The present study therefore included a multifaceted assessment of affective control, including the affective backward digit span task (Schweizer et al., Reference Schweizer, Leung, Kievit, Speekenbrink, Trender, Hampshire and Blakemore2019a) as a measure of affective working memory updating, the affective Stroop task as a measure of affective inhibition (Preston & Stansfield, Reference Preston and Stansfield2008), and the affective card sorting task (Schweizer et al., Reference Schweizer, Parker, Leung, Griffin and Blakemore2020b) as a measure of affective shifting. In addition to examining the effect of training on each facet separately, we examined the structure of affective control in this sample to extract a meaningful assay of affective control. This index of affective control is essential to test whether any improvements in the clinical outcomes of interest vary as a function of changes in the index of affective control. As per protocol (Schweizer et al., Reference Schweizer, Leung, Kievit, Speekenbrink, Trender, Hampshire and Blakemore2019a), we predicted that performance on these affective control tasks would dissociate into separable affective control and cognitive control factors. This two-factor model was compared to a single factor model, which if superior would indicate that there is no difference between affective control applied in affective or neutral contexts.
Pre- to post-training changes on these indices of affective control were compared between the AffeCT group and a group undergoing placebo training (Placebo). The Placebo group received the same narrative regarding the potential benefits of training on mental well-being and affect regulation. Placebo training was included to control for any effects of engaging repeatedly in a computerised cognitive training activity purported to benefit well-being and emotion regulation.
This design allowed us to test the following pre-registered hypotheses (Schweizer et al., Reference Schweizer, Leung, Kievit, Speekenbrink, Trender, Hampshire and Blakemore2019a): Affective control training hypothesis (H1): affective control can be improved in adolescents. To examine the first hypothesis, we compared pre- to post-training affective n-back performance across the two training groups. Affective control facets hypothesis (H2): Compared to Placebo, AffeCT would lead to greater improvements in all facets of affective control. To investigate this hypothesis, changes in performance on the affective transfer tasks were compared between the training groups. Age-related change hypothesis (H3): Training-related changes in affective control would vary as a function of age. Mental health hypothesis (H4): Improved affective control from pre- to post-training would be associated with fewer self-reported mental health problems, emotion regulation difficulties and self-control ability.
Methods
Participants
Two-hundred and forty-two participants aged 11–19 years old were recruited from eleven schools from Greater London, as well as through advertisements at University College London and the University of New South Wales. 43 participants were excluded due to technical issues or not meeting inclusion criteria, for details see the participant inclusion flowchart online Supplementary Fig. S1. The final sample included 199 participants (159 female, mean age = 14.32 years, s.d. = 2.31 years, Table 1 for participant characteristics) who were randomised to one of the two training groups: Affective Control Training (AffeCT; n = 101), or the Placebo group (Placebo; n = 98). Training allocation was based on a computer-generated condition assignment (using Sealed Envelope simple randomisation service) stratified by age (young adolescents 11–14 years and mid-late adolescents 15–19 years). Following randomisation, the groups were matched on all baseline characteristics including age, gender, socioeconomic status, ethnicity, fluid intelligence, mental health, emotion regulation and self-control.
Table 1. Baseline characteristics across groups
Note: The table reports baseline characteristics and demonstrates that randomisation was successful with Bayesian t tests (continuous variables) and χ2 tests (categorical variables) showing no significant group differences on any baseline characteristics. Fluid intelligence = IQ score derived from the Raven's Advanced Progressive Matrices (Raven et al., Reference Raven, Court and Raven1988); Mental health difficulties = Difficulties score on the Strengths and Difficulties Questionnaire (Goodman, Reference Goodman1997); Emotion regulation difficulties = Total score on the Difficulties in Emotion Regulation Scale (Gratz & Roemer, Reference Gratz and Roemer2004); Self-control = Total score on the Brief Self-control Scale (Tangney et al., Reference Tangney, Baumeister and Boone2004); Parental education = Highest parental education was measured as a proxy of socioeconomic status (SES); Asian = Included individuals selecting any of these answer options: Asian-other, Bangladeshi, Chinese, Indian, Pakistani; Black = Black is a term used in Britain to refer to citizens of African or African-Caribbean decent, here it included individuals who selected any of the following to describe their ethnicity: Black-African, Black-British, Black-Other; White = here refers to individuals who identified as White-British or White-other; Mixed/Other = here includes individuals who identified as being of mixed or other ethnicity than the available options by selecting Mixed/Other; Group size = average number of participants in the pre- and post-training assessment sessions.
Testing procedure
Informed consent was obtained from parents if the participant was under 18 years and from participants aged 18 or over; participants under 18 also provided informed assent. The pre- and post-training sessions each lasted 1.5 h and took place at the school or in the research lab in groups between 2 and 42 participants (with 1–4 researchers in each session). During the pre- and post-training sessions, participants completed a range of cognitive tasks and questionnaires. The mental health, emotion regulation and self-control questionnaires were administered to participants again online at two follow-up assessments, oneand twelve months after the post-training assessment. Researchers involved in the post-training sessions were blinded to training-group allocation. Participants were compensated £10 for each pre- and post-training session, £2/training day (£5 if they completed more than one training session in a single day), and £5 for each follow-up questionnaire.
Training procedure
Participants were asked to complete 14 days of online training on their own device in a quiet space within a 4-week period. On the first three day both groups completed a different version (A–C) of the training tasks. From day four onwards, participants freely choose any version of the training. Both groups were told to spend as much time as possible training on version C due to its benefits to attention, memory, and emotion regulation. However, to maximise training engagement participants had the option to continue using the other versions. A full training session took between 20 and 30 min, depending on the performance levels achieved. Training could be ended after 10 min. Training sessions under 10 min were not considered a full training session and not included in the analyses. Participants received a daily training reminder at 8am. Those who did not complete the training by 5pm received an additional reminder.
Tasks and measures
Affective control training task
AffeCT consisted of three versions of the n-back task (Fig. 1): a visuospatial n-back (A), an auditory n-back (B) and a dual (including the auditory and visuospatial modalities) n-back (C). Across all three versions of the task, stimuli were presented on a 4 × 4 grid and/or over headphones. Participants had to respond via a ‘Match’ or ‘No match’ button press to indicate if the stimuli matched the corresponding stimuli presented n trials back (for details see Supplementary Materials).
Figure 1. AffeCT Tasks Including a Visuospatial (a), Auditory (b) and Dual (c) n-back Task.
Note: The figure depicts sample trials for each of the three training tasks: (a) visuospatial n-back, (b) auditory n-back, and (c) dual n-back task. Trials depicted with a light blue background require a ‘No Match’ response. Yellow backgrounds indicate ‘Match’ (i.e. target) trials. The example block in Fig. 1 is depicted for n = 1.
Placebo training task
The Placebo task required participants to indicate via button press (‘Match’, ‘No Match’) whether two panels displayed exactly the same stimuli in the same positions on a grid. There were three versions of the task that differed in the type of stimuli they included, namely: shapes, faces and word. The faces and words versions include the same stimuli as AffeCT. The initial trial included five items per panel, the number of items per panel subsequently increased with participants' performance.
Affective control tasks
To assess the different facets of affective control three measures of affective inhibition, updating and shifting were included. These measures were administered before and immediately after training.
Inhibition. The affective Stroop task was used to assess inhibition of affective interference (Preston & Stansfield, Reference Preston and Stansfield2008). Pictures of faces were presented to the participants with words superimposed on the image. Then, participants were asked to indicate whether the adjectives were happy or sad. Feedback was provided after each trial with a red or green boarder around the image for 200 ms to indicate an incorrect or correct response. The task was self-paced and there were 96 trials in total. Trials were considered inaccurate if no response was detected after 4 s. The performance of the task was operationalised as task accuracy (i.e. percentage trials correct) and reaction time was recorded.
Updating. The affective backward digit-span task was used to assess updating. Participants were presented with a series of digits (1500 ms) displayed over negative or neutral background images. Participants were then asked to recall the digits in reverse order. The task started at two digits, with each span level presented twice. To progress to the next span level, participants had to get at least one out of the two trials correct. The task was terminated if both trials were incorrect. The performance of the task was operationalised as the highest span level achieved in the negative and neutral condition.
Set-shifting. The affective set-shifting task, which was adapted from the Madrid Card Sorting Task (Schweizer et al., Reference Schweizer, Parker, Leung, Griffin and Blakemore2020b), was used to assess individuals' capacity to switch between task demands. Participants were dealt with a card and were asked to assign it to one of the four decks according to the three possible sorting rules: (1) card colour, (2) number of items on the card, and (3) shapes (for neutral condition) or emotional expressions (for affective condition). There were 96 trials in total and the sorting rule changed randomly after 6 to 9 trials. Participants had to respond within 30 s or the trial would be recorded as an error. Performance of the task was operationalised as random errors, which refers to errors that occur on any trial in the series from the third trial onwards (as the first two trials were needed to establish the correct sorting rule).
Mental health, emotion and self-regulation
The 25-item Strengths and Difficulties Questionnaire (SDQ; Goodman, Reference Goodman1997) was used to assess mental health difficulties. The scale has been developed for use in children and adolescents (Goodman, Reference Goodman1997). Participants indicated the extent to which the statements were true of them from ‘0’ Not true to ‘2’ Certainly true. Internal consistency in the current sample was acceptable, ωT = 0.81.
The Difficulties in Emotion Regulation Scale (DERS; Gratz & Roemer, Reference Gratz and Roemer2004) assessed emotion regulation. The 36-item scale measures a range of affective processes and has been shown to be valid for use in adolescents (Charak et al., Reference Charak, Byllesby, Fowler, Sharp, Elhai and Frueh2019). Participants indicated how often from ‘0’ Almost never to ‘4’ Almost always they experienced difficulties with different facets of emotion regulation. Internal consistency in the current sample was good, ωT = 0.93.
Self-regulation was assessed using the Brief Version of the Self-Control Scale (Tangney, Baumeister, & Boone, Reference Tangney, Baumeister and Boone2004), which has been successfully used with adolescents (Duckworth, Kim, & Tsukayama, Reference Duckworth, Kim and Tsukayama2013). The scale includes 13 statements and participants had to indicate, on a 5-point scale from ‘0’ Not at all to ‘4’ Very much, how well each statement described them. Internal consistency in the current sample was acceptable, ωT = 0.89.
Fluid intelligence
At pre-training we additionally administered the 12-item version of the Raven's Advanced Progressive Matrices to compare the groups on pre-training fluid intelligence (Raven, Court, & Raven, Reference Raven, Court and Raven1988). Participants were instructed to complete the task as quickly as possible. The measure has good psychometric properties (Raven, Reference Raven2000).
Statistical analysis
To investigate gain on the AffeCT task, our protocol specified that this would be analysed investigating dprime (d’) scores as performance index. However, the inclusion of a no-match button in the task design meant that hit rates were at ceiling, rendering the d’prime scores non-informative. We therefore opted to examine the maximum level of n-back achieved instead, as is conventional for n-back training studies (Soveri, Antfolk, Karlsson, Salo, & Laine, Reference Soveri, Antfolk, Karlsson, Salo and Laine2017).
The protocol specified a multivariate mixed effects model for the analyse of the three facets. However, as the primary outcomes varied across tasks (accuracy for the affective digit span task and reaction time for the affective Stroop and Card Sorting tasks) the facets were analysed in individual mixed effects models. The results were Bonferroni-corrected (0.05/3) for three separate comparisons and therefore the statistical threshold was α = 0.017. All analyses were conducted in R, version 4.1.0 (R Core Team, 2013).
Results
Baseline affective control
In line with our pre-registration, we examined the structure of affective control at baseline. Specifically, we predicted affective control and cognitive control to be correlated but separate factors. The predicted two-factor model including performance on the transfer measures of shifting, inhibition and updating did not converge, even after scaling the reaction time data. Removing the latent congruency factor from the latent affective control factor allowed the model to converge but showed a very poor fit [Χ 2(96) = 800.81; CFI = 0.37; TLI = 0.21; RMSEA = 0.21; SRMR = 0.20; AIC = 567.38]. We therefore examined the structure for accuracy and reaction time separately. The two-factor structure provided a good fit for the reaction time data. However, the fit was not significantly better when compared to a model including a single cognitive control factor, ΔΧ 2(3) = 4.27; p = 0.234. For accuracy, the two-factor showed borderline acceptable fit (online Supplementary Table S1) and outperformed the poorly fitting one-factor model ΔΧ(3) = 41.32; p < 0.001. The latent affective control factors (from the two-factor models) for accuracy and reaction time were retained to examine hypothesis 4, as per protocol.
Training characteristics and results
There were no significant differences (see supplementary materials) in training adherence with the exception of average number of sessions completed, which was significantly higher in the Placebo (M = 12.08, s.d. = 13.39) compared to the AffeCT group (M = 7.19, s.d. = 9.12), t(187) = 2.90, p = 0.004. Number of training sessions completed was therefore included as a covariate.
In line with our affective control training hypothesis, which predicted that the AffeCT group would perform better on the affective n-back task following training, there was a training group × time interaction, b = 0.48, s.e. = 0.13, t = 3.70, p < 0.001. The interaction remained significant when controlling for number of sessions trained, b = 0.47, s.e. = 0.13, t = 3.55, p < 0.001 (for full model statistics see online Supplementary Table S2).
Effects of training on the three components of affective control
We found no support for our second affective control facets hypothesis: AffeCT did not lead to greater gains in affective inhibition, shifting or updating. As indicated by non-significant interactions between the effects of training group (AffeCT v. Placebo) and time (Pre v. Post-training) reported in Table 2.
Table 2. Mixed effects models investigating the effects of training group on affective control facets from pre- to post-training
Note: Time = Pre-training v. Post-training; Group = AffeCT v. Placebo. The models are reported for accuracy and reaction time performance on affective inhibition (affective Stroop task), shifting (affective card sorting task) and updating task (affective backward digit span task). For means and standard deviations across the different conditions at pre- and post-training on the affective inhibition, shifting and updating tasks see online Supplementary Table S3.
Age-related differences in affective control training
In line with our third age-related change hypothesis, there were age-related differences in training gains on the affective n-back task. That is, age group moderated the significant group by time interaction reported in H1, b = 0.24, s.e. = 0.07, t = 3.56, p < 0.001 (for the full model estimates see online Supplementary Table S4). This effect was reduced, but remained significant when correcting for number of training sessions completed, β = 0.06, s.e. = 0.02, t = 3.50, p < 0.001. Analyses of the estimated marginal means trends revealed a significant age by time interaction effect in the AffeCT group, b = −0.17, s.e. = 0.08, t = −2.30, p = 0.022, but not in the Placebo group, b = 0.08, s.e. = 0.07, t = 1.23, p = 0.219. In the AffeCT group increasing age showed a small, non-significant association with improvements on the affective n-back task, r = 0.22, p = 0.069.
The effect of training on mental health, emotion regulation and self-control
Applying a multi-group latent growth curve model showed that AffeCT, but not Placebo, training led to significant covariance of change in affective control index identified at baseline in the reaction time and accuracy models and mental health difficulties (online Supplementary Table S5). The formal comparison of the free model to a model with constrained variance of the latent variables was significant for both affective control indices: reaction time: ΧDiff 2 = 11.07, p = 0.004; accuracy: ΧDiff 2 = 10.04, p = 0.007.
Extracting the indices of change from the latent growth curve model revealed that post-training mental health difficulties in the AffeCT group were negatively associated with change in affective control, (reaction time: r = −0.48, p < 0.001; accuracy: r = −0.57, p < 0.001). That is, greater change in affective control was associated with fewer mental health problems. In contrast, the Placebo group showed small, non-significant associations between change in affective control and post-training mental health difficulties, (reaction time: r = 0.21, p = 0.065, accuracy: r = −0.28, p = 0.014). However, the effects of training on mental health difficulties were not maintained at the 1-month follow-up (reaction time: ΧDiff 2 = 4.54, p = 0.10, accuracy: ΧDiff 2 = 2.43, p = 0.30), nor at 1-year follow-up (reaction time: ΧDiff 2 = 0.87, p = 0.65, accuracy: ΧDiff 2 = 0.89, p = 0.64).
There was no effect of AffeCT compared to Placebo on emotion regulation difficulties or self-control capacity. This was true for latent growth curve models including the latent affective control index for reaction time (emotion regulation: ΧDiff 2 = 3.61, p = 0.163; self-control: ΧDiff 2 = 3.67, p = 0.163) and accuracy (emotion regulation: ΧDiff 2 = 0.00, p = 0.999; self-control: ΧDiff 2 = 4.98, p = 0.081).
Discussion
Identifying novel and scalable avenues for prevention of mental problems is essential (Hagan et al., Reference Hagan, Graham, Wilkinson, Midgley, Suckling, Sahakian and Goodyer2015), as mental ill health has become the leading burden of disease in young people worldwide (Gore et al., Reference Gore, Bloem, Patton, Ferguson, Joseph, Coffey and Mathers2011). The present study tested the potential of an app-based AffeCT to benefit adolescents' mental health. The study found that adolescents' performance on an AffeCT task improved from pre- to post-training, with older adolescents benefiting more from training compared to younger adolescents. These training gains did not lead to improvements on non-trained measures of affective control, across affective updating, inhibition and shifting. However, variance of change in affective control was related to reduced mental health difficulties at pre-training in the AffeCT but not Placebo group. This training-related benefit was not maintained one month or one year later. We discuss the implications of these findings for the potential of app-based AffeCT in youth mental health prevention.
Improving affective control with minimal effort
The observed improvement on the trained AffeCT task, was noteworthy as the AffeCT group on average completed half as many training sessions as the Placebo group. Moreover, both groups predominantly completed the ‘simpler’ training tasks. For the AffeCT group these were the two single modality training versions. That is, adolescents limited the cognitive effort involved in training by selecting the less demanding option. Ganesan and Steinbeis (Reference Ganesan and Steinbeis2022) argued that effort exerted in cognitive (training) tasks is guided by an individual's cost-value computation. This suggests that the perceived value of training, especially on the dual n-back version of AffeCT, was insufficient to motivate most participants. Engagement in cognitive training tasks, such as AffeCT, should therefore be motivated by providing appropriate incentives to exert cognitive effort. The incentives are not limited to monetary incentives or other rewards, instead the relative value of ‘training’ can be increased by changing adolescents' mindset about the benefits of a specific training regime (Yeager et al., Reference Yeager, Bryan, Gross, Murray, Krettek Cobb, H. F. Santos and Jamieson2022).
Age-related training improvement
Pre- to post-training gains in affective control (i.e. change in maximum level of n back achieved from pre- to post-training) increased as a function of age. This is in line with non-affective cognitive training findings, which have shown greater benefits of cognitive training in older (16–18 years) compared to younger (11–13 years) adolescents (Knoll et al., Reference Knoll, Fuhrmann, Sakhardande, Stamp, Speekenbrink and Blakemore2016). A comprehensive review of the literature on brain plasticity in adolescence, however, suggests that brain plasticity-related changes in higher cognitive functions during adolescence are dependent on the brain regions involved in the specific cognitive domain that is being trained and a range of factors, including the type of training and the trained individuals' gender (Laube, van den Bos, & Fandakova, Reference Laube, van den Bos and Fandakova2020). Affective control recruits the cognitive control network, in particular the inferior frontal gyrus (Schweizer et al., Reference Schweizer, Satpute, Atzil, Field, Hitchcock, Black and Dalgleish2019b), which shows protracted development throughout adolescence. The age-related differences in training gains observed in the present study support training-induced functional changes and possibly structural plasticity of the neural substrates of affective control throughout adolescence. However, examinations of training-induced functional or structural brain changes in typically developing adolescents are scarce (Lee, Kwak, & Chey, Reference Lee, Kwak and Chey2019). Examining training-induced neural changes will be important to determine potentially sensitive periods for training (Frankenhuis & Walasek, Reference Frankenhuis and Walasek2020).
Performance on different facets of affective control remains unchanged by AffeCT
AffeCT did not transfer to significantly greater improvements in affective inhibition (Preston & Stansfield, Reference Preston and Stansfield2008), affective working memory (Schweizer et al., Reference Schweizer, Leung, Kievit, Speekenbrink, Trender, Hampshire and Blakemore2019a) and affective shifting (Schweizer et al., Reference Schweizer, Parker, Leung, Griffin and Blakemore2020b). The lack of training-related transfer to untrained measures of affective control is in line with meta-analytic reviews of the cognitive control training literature, which show that training leads to improvements on the trained task but typically does not extend to untrained measures of neutral cognitive control (e.g. Soveri et al., Reference Soveri, Antfolk, Karlsson, Salo and Laine2017). Transfer effects (i.e. training-induced changes) to affective control have received comparatively less attention. In older adolescents (16–24 years), Roberts, Mostazir, Moberly, Watkins, and Adlam (Reference Roberts, Mostazir, Moberly, Watkins and Adlam2021) showed that neutral, but not AffeCT, led to improvements in affective updating. In adults, cognitive (Cohen et al., Reference Cohen, Margulies, Ashkenazi, Schaefer, Taubert, Henik and Okon-Singer2016; Sari, Koster, Pourtois, & Derakshan, Reference Sari, Koster, Pourtois and Derakshan2016) and affective (Schweizer et al., Reference Schweizer, Grahn, Hampshire, Mobbs and Dalgleish2013; Schweizer, Hampshire, & Dalgleish, Reference Schweizer, Hampshire and Dalgleish2011) control training paradigms have been shown to improve affective inhibition. While the current study showed no transfer on untrained affective control measures, the existent literature suggests that with appropriate training regimes facets of affective control may be malleable and benefit from computerised control training.
Affective control gains and mental health
Despite the lack of training-related transfer to the individual facets of affective control, pre- to post-training change on a composite index of affective control and pre- to post-training change in mental health problems covaried in the AffeCT group but not the Placebo. That is, greater improvement in affective control was associated with mental health benefits in adolescents who had trained with AffeCT but not Placebo. This is in line with a review of the literature showing that computerised cognitive control training, in particular training that induces improvements in affective control, is associated with benefits in symptoms of depression and anxiety in children and adolescents (Edwards et al., Reference Edwards, Zec, Campbell, Hoorelbeke, Koster, Derakshan and Wynne2022) and adults (Koster, Hoorelbeke, Onraedt, Owens, & Derakshan, Reference Koster, Hoorelbeke, Onraedt, Owens and Derakshan2017).
Of note is the per protocol use of multi-group latent growth curve modelling to analyse the impact of training. Traditional methods to analyse training effects have typically included repeated measures analyses of variance and mixed effects models that examine average performance or symptom levels (Soveri et al., Reference Soveri, Antfolk, Karlsson, Salo and Laine2017). Multi-group latent growth curve modelling instead compared the groups on the extent to which individual variance in the change in affective control was associated with variance in change in mental health benefits. This training effect would not have been captured by traditional inference methods and despite not extending beyond the immediate post-training assessment, it warrants future investigation into the potential of AffeCT as a preventative intervention in adolescent mental health. A particularly promising avenue of investigation is whether the training benefits of AffeCT can be augmented in magnitude (i.e. more reduction in symptoms) and temporally extended when increasing engagement and participation in the more demanding version of the training task (i.e. dual v. single n-back). Beyond increasing engagement by an improved rationale for training as proposed above, participation can be boosted through gamification. Gamification refers to the process of enhancing training with affordances to create engaging experiences (Hamari, Reference Hamari2013). Meta-analytic evidence demonstrates that it is an effective tool to reliably encourage attentional engagement and motivation to increase rates of training (Lumsden, Edwards, Lawrence, Coyle, & Munafò, Reference Lumsden, Edwards, Lawrence, Coyle and Munafò2016).
The present findings need to be considered within the context of the study's limitations. While the study was adequately powered, limited per protocol engagement with the task restricts the inferences that can be drawn regarding the effectiveness of AffeCT. Moreover, for over half the sample the 1-year follow-up was conducted during the COVID-19 pandemic, so the potential to detect training-effects over and above the adverse impact that the pandemic has had on adolescent mental health (Racine et al., Reference Racine, McArthur, Cooke, Eirich, Zhu and Madigan2021) is limited. Moreover, the assessment of mental health in the current study arguably does not capture potential effects of AffeCT on affective fluctuations in adolescents' everyday lives. Future research should therefore consider assessing the impact of AffeCT on dynamically changing mood states by sampling affect using ecological momentary assessments.
In sum, the present study showed that AffeCT in adolescents, especially older adolescents, led to improvements in performance on the training task, but did not transfer to untrained measures of individual facets of affective control. Encouragingly, the covariance of change on the non-trained composite index of affective control and mental health difficulties form pre- to post-training showed a significant benefit of AffeCT over Placebo. That is, the present study provides preliminary evidence that AffeCT may confer short-term preventative benefits for adolescent mental health. App-based training is easy to disseminate and can therefore be delivered at scale anywhere in the world. Even small and short-term benefits are therefore potentially meaningful if they can be delivered at the population level. If engagement with AffeCT can be further boosted through gamification and incentives, these benefits may be extended beyond the period immediately following training.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723002167.
Authors’ contributions
S. S. and S.-J. B. designed the study. W. T. and A. H. designed and maintained the software. J. T. L. and S. S. collected the data. S. S., J. T. L., R. K. and S.-J. B. analysed the data. S. S. wrote the manuscript, and all authors commented on the manuscript.
Financial support
The work was funded by a Sir Henry Wellcome Fellowship [Wellcome Trust: 209127/A/17/Z] to S. S. S.-J. B. is funded by Wellcome (grant number WT107496/Z/15/Z), the MRC, the Jacobs Foundation, the Wellspring Foundation and the University of Cambridge. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission.
Competing interests
A. H. is a co-founder of the company that developed the app software. He was not involved in the study design or data analyses.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the Helsinki Declaration of 1975, as revised in 2008. The study was approved by the University College London [Ref: 12753/002] and UNSW Research Ethics Committees [HC3231].
Open access
