Participants were recruited through an outpatient clinic for affective disorders at the Psychiatric unit of Haukeland University Hospital and through advertisements and previous research projects. A total of 29 participants were included in the study, and of these, 20 completed the intervention and retest. Inclusion criteria were age (20–60), previous treatment of MDD, currently remission, or recovery as indicated by ≤ 12 on The Montgomery Åsberg Depression Rating Scale (MADRS; Montgomery and Åsberg, 1979). Exclusion criteria were brain damage/pathology, severe somatic disorders, alcohol, or substance abuse. The Mini-International Psychiatric Structural Interview (Sheehan et al., 1998) was used to confirm diagnosis and history of MDD, and screen for exclusion criteria. Other clinical data were assessed by a structured form developed for the study. A psychologist or a psychiatric nurse assessed inclusion criteria at the outpatient clinic for affective disorders or the neuropsychological clinic at the University of Bergen. Neuropsychological and experimental assessment was performed at the neuropsychological clinic, both preceding (T1) and following the intervention (T2). The CWMT intervention was delivered at home online on a computer or tablet. Participants started training within 2 weeks following assessment. After the intervention participants were invited to a posttest assessment of e-WM, EF, and PS, depressive symptoms, and rumination.

Cogmed™ (Pearson, 2022) was chosen as the CWMT intervention. It is a commercial CWMT program that has several advantages: It is performed online, so participants can use it at home, it has a considerable duration, with incremental difficulty depending on past performance (successfully completed tasks), both aspects that are considered important for successful CWMT (Koster et al., 2017; Launder et al., 2021). Cogmed targets “cold cognition,” i.e., working memory span. The program consisted of several different number, letter, and spatial, forward and backward sequencing tasks. For example, Letters are read out loud, while one of several rotating bubbles lights up on screen, participants remember the sequence of the lighted bubbles, clicking on the corresponding bubbles sequentially following presentation. In another task, participants attend to a sequence of numbers read out and then click on the numbers in reverse sequence on screen. In a different task, participants remember the sequence of panels on a cube that light up one by one, then clicking on the panels in the correct sequence after presentation. The intervention consisted of 25 sessions, lasting ~5 weeks, with five weekly 30–40-min sessions. In addition, there was weekly telephone contact between participants and a trained Cogmed coach, discussing progress, motivation, and questions related to the training.

All participants completed a neuropsychological test battery pre- and post-intervention measuring general intellectual ability, EF, and PS. General intellectual ability was measured by the Norwegian version of the Wechsler Abbreviated Scale of Intelligence (Wechsler, 1999), to better describe the sample and potentially explain dropout (refer to Table 1). The Delis–Kaplan Executive Function System (Delis et al., 2001) subtests TMT, and the Color Word Interference Test (CWIT), an updated version of the Stroop task, was used to measure PS and EF. The TMT is a paper and pencil task consisting of five conditions. The CWIT consists of four conditions with verbal responding to visual stimuli.

Emotional working memory was measured by a computerized paradigm pre–post-intervention (refer to Figure 2). Participants were presented with a series of three happy or sad faces (Lundqvist et al., 1998) and then instructed to hold the sequence in mind in either a forward- (low WM loading) or manipulate in a backward (high WM loading) sequence. Finally, participants were shown one of the three faces and asked to specify where in the (rotated) picture sequence, the face was presented by pressing key “1,” “2,” or “3.” The experiment was run in E-prime (version 2), and percent accuracy was used as an outcome measure. For further description of the paradigm, refer to Ronold et al. (2020a).

Several self-report measures were administered pre- and post-intervention. The Ruminative Responses Scale (RRS) and Rumination Reflection Questionnaire (RRQ) measured depressive and neurotic rumination, respectively. Depression was measured post-test by MADRS self-report.

The Norwegian version of the form RRS (Treynor et al., 2003) was administered pre- and post-intervention. It consisted of 22 questions in a four-point scale from 1 “almost never” to 4 “almost always” (22–88 range) and asked participants lifetime proclivity for rumination when “feeling sad or depressed,” e.g., “think about how sad you feel.”

The rumination subscale from the Norwegian version of the RRQ was used to measure lifetime proclivity for self- ruminative though, independent of depressive mood, e.g., “My attention is often focused on aspects of myself I wish I'd stop thinking about.” Here, 12 items are rated on a five-point scale ranging from 1 “strongly disagree” to 5 “strongly agree” (12–60 range). The scale has been shown to correlate with the personality trait neuroticism and could thus measure neurotic rumination relatively independent of mood (Trapnell and Campbell, 1999).

Clinician administered MADRS was used to assess depression pre-intervention. Following the intervention, the Montgomery Åsberg Depression Rating Scale Self-report (Svanborg and Åsberg, 2001) was used to assess depression. This measure is comparable to the MADRS with the exception of one item (item 1) that assess apparent sadness. Depressive symptoms are scored on a seven-point scale from zero to six (range 0–54 relative to MADRS 0–60) and have proven to be useful and sensitive to change in depressive symptoms in clinical trials (Cunningham et al., 2011). Items 6 and 5 measure self-subjectively experienced difficulties with attention on MADRS and MADRS Scale Self-report, respectively.

All participants provided informed signed consent prior to participating in the study. The study was conducted in accordance with the Helsinki Declaration of Ethical Research regarding Ethical Principles for Medical Research Involving Human Subjects (World Medical Association, 2013) and was approved by the committee for ethical research in Western Norway (2014/1079), in addition to the Norwegian Data Inspectorate. Participants were awarded with a gift card valued at 400 Norwegian Kroner (~50 dollars), for their participation. In addition, participants were given access to the CWMT program free of charge.

E-Data aid (version 2) was used to calculate accuracy in the e-WM paradigm and exported to a statistical program used for all analyses (version 25). For the neuropsychological tests, raw scores consisting of seconds to complete tasks were used. Data were screened for outliers using boxplots and histograms. Normality was assessed by Kolmogorov–Smirnov test, and non-parametric tests were used when assumptions were violated. To investigate the pre-intervention hypotheses, associations between cognition and symptoms post-intervention, in addition to the explorative analyses, the bivariate correlation coefficient, Pearson's r was used. To investigate post-intervention hypothesis and attrition, paired sample t-tests were used, with Wilcox signed rank test as a non-parametric alternative. Spearman's rank-order correlation ρ was calculated for non-parametric data. To avoid risk for type 2 errors, Bonferroni correction was not strictly implemented (Nakagawa, 2004), but it has been noted which improvements remained significant after this correction (refer to Table 2). Significance level was set at p = 0.05 (Bonferroni correction for Table 2 = 0.05/16 = 0.003), and one-tailed significance levels were used to assess pre-registered hypotheses. Effect sizes (Cohen's d), where reported, for non-parametric tests r, were calculated from the following formula r = Z/√ total number of observations. Effect sizes were reported as small (d = 0.2, r/ρ =0.1), moderate (d = 0.5, r/ρ = 0.3), and large (d = 0.8, r/ ρ = 0.5) according to Cohen (1988).

Z-score transformation of MADRS pre- and post-intervention was calculated to compare the change in depression from pre- to post-test. A depression history composite was computed from the following Z-transformed depression and rumination measures (T1 MADRS + T2 MADRS + MDD length in months + number of MDD episodes + T1 RRS + T2 RRS + T1 RRQ + T2 RRQ)/8. Change scores for neuropsychological tests, RRS, RRQ, and MADRS were calculated by subtracting T2 scores from T1 scores. Change scores in e-WM were calculated by subtracting scores at T1 from scores at T2. Thus, positive values indicated post-intervention improvements. Separating EF from PS: Contrast scores were calculated to separate EF from PS: TMT Switching (TMT Switching contrast = Number-Letter Switching – (Visual Scanning + Letter Sequencing + Number Sequencing + Motor Speed)/4), Inhibition (Inhibition Contrast = Inhibition – (Color Naming + Word Reading)/2) CWIT Switching condition (CWIT Switching Contrast = Inhibition/Switching – (Color Naming + Word Reading + Inhibition)/3). CWMT improvement score: Improvement in CWMT was calculated through the formula: highest performance score (i.e., session with most completed tasks) – starting performance score/starting performance score.

There were no significant differences in demographic characteristics in Table 1, depressive symptoms, EF, PS, e-WM, with small effect sizes, between participants who completed the intervention and participants that dropped out, suggesting that attrition effects were limited.

There was partial support for an association between rumination and e-WM. Neurotic rumination showed a moderate negative association with negative high WM loading material in e-WM, supporting associations between rumination and difficulties in manipulation of negative material in WM (Joormann et al., 2011; Ronold et al., 2020a). Inhibition was independent of depressive symptoms as expected (Snyder, 2013; Ronold et al., 2020b). The lack of association between depressive rumination measured by RRS and e-WM could be due to the low level of depressive symptoms in the group at T1 (remission as inclusion criteria) and could help explain the lack of depressive state effects on cognitive functions such as PS and switching. The direction of associations between depressive symptoms and cognition went opposite of predictions, and from what is usually reported in the literature (McDermott and Ebmeier, 2009: Snyder, 2013; Semkovska et al., 2019). It is thus likely that participants must show higher levels of depressive symptoms for state effects to appear (Dotson et al., 2020), which is noteworthy given the current remitted sample with low levels of depressive symptoms. Supporting this were the findings at T2 when symptoms and variance were higher. Here, some of the expected associations between depressive symptoms and switching and rumination were more evident. This could indicate that some associations between symptoms and cognition remain following CWMT and thus are impacted to a limited degree by improvements in cognition, suggestive of trait or scar effects. Thus, e-WM showed expected associations with rumination, whereas EF and PS did not show expected associations with depressive symptoms. Limited support for state effects on cognition in remitted MDD appeared, and inhibition emerged as a possible trait, independent of variations in current depression symptoms, but state effects were probably difficult to detect due to low variance and score of depression in the current sample initially.

Hypotheses about improvements in e-WM, PS, and EF following CWMT were supported. Participants improved across all e-WM conditions, most PS conditions, and most EF conditions, following CWMT. Effects were largest for CWIT Inhibition and TMT Number Letter Switching, and these were the only conditions statistically significant after Bonferroni correction, supporting larger effects for CWMT in EF tasks, and potential for the remediation of trait deficits associated with psychiatric disorders such as MDD. This is in line with stronger associations between EF and WM, relative to PS (McCabe et al., 2010), supporting transfer between CWMT and EF improvements. This could be important since EF deficits are of particular clinical significance (Snyder et al., 2015; Groves et al., 2018). However, the improvements compared to normative data indicated that that only inhibition improved descriptively, i.e., from “low average” to “average” (Delis et al., 2001). Other studies, in contrast, find superior improvements in PS following training (Webb et al., 2018). Of note, improvements were larger than retest effects reported in the literature (Scharfen et al., 2018), which support causal effects of CWMT on both PS and EF. An association between the EF switching, measured by TMT, and depressive symptoms, emerged following CWMT, suggesting that some associations between symptoms and EF does not disappear following CWMT.

Post-intervention, e-WM showed improvements with similar effects across conditions. The correlation between negative high loading WM condition and rumination disappeared following CWMT, as expected. This suggests that improvements in manipulation of negative material in WM, suspected of contributing to rumination (Ronold et al., 2020a), could improve ruminative tendencies. Improvements in hot cognition following CWMT were found, which could improve the clinical course of remission from MDD (Ahern et al., 2019). Processing of emotional faces could be of particular interest due to its implications for social interaction, and improvements could have positive consequences for interpersonal function (Bourke et al., 2010). This study did not find improvements in symptoms, however, precluding the clinical implications of the reduced associations between e-WM and rumination. Thus, the clinical importance of cognitive improvements remains unclear. In addition, improvements in objective cognitive functions did not transfer to subjective ratings of attention, and there were suggestions that CWMT improvements exacerbated self-rated attention. Objective and subjective cognition are largely distinct (Snyder et al., 2020) however, and findings regarding CWMT and symptomatic improvements are mixed (Barkus, 2020; Legemaat et al., 2021). This could suggest that broader interventions (Myklebost et al., 2021), selection, and personalized interventions for individuals and subgroups with varying degrees of residual symptoms are necessary for optimal effects to appear.

Hypothesis about improvements in symptoms was not supported. Thus, transfer effects, beyond cognition, were not found (Grinberg et al., 2021). This is contrary to the findings of Lengvenyte et al. (2020), where the authors found improvements in depressive symptoms including subjective attention, using similar methods and interventions as this study. However, the authors used subjective cognitive complaints as an inclusion criterion. In addition, it is not evident that MADRS is an optimal measure of subjective cognition (Russo et al., 2015). A recent meta-analysis found that improvements in symptoms following cognitive remediation were related to higher baseline symptomatology (Legemaat et al., 2021), which could help explain our findings. This study had low initial depressive symptoms that could have resulted floor effects in improvement, also influencing sensitivity to state effects. However, the main aim of the study was to investigate CWMT effect on residual cognitive symptoms and not mood symptoms per se, resulting in the low mood symptom load in the fully remitted sample at inclusion. Sample size also probably affected results. Meta-studies generally find small associations between EF and rumination (Vălenaş and Szentágotai-Tătar, 2017; Yang et al., 2017; Zetsche et al., 2018), and thus, associations might be too small to detect with the current sample. Rumination did not improve, which could suggest that CWMT does not target this function in remission from MDD. Supporting this, a recent randomized controlled trial did not find effects of cognitive control training on rumination (Ferrari et al., 2021). Possibly, effects could only be evident for a subsample of remitted individuals (Hammar et al., 2020; Woolf et al., 2021), with differing degrees of residual symptomatology and cognitive deficits (Paelecke-Habermann et al., 2005; Pu et al., 2018; Ronold et al., 2021). Thus, inclusion through some sort of objective or subjective deficit or criteria could be essential for optimal effects of CWMT to manifest. The CWMT intervention targeted cold cognition. Targeting hot cognition instead might show larger effects on symptoms as suggested by a recent meta-analysis (Sociali et al., 2022). Future research should compare the effects on interventions targeting hot and cold cognition on several clinically relevant residual symptoms to examine utility for relapse prevention.

Explorative analyses found associations between improvements in CWMT and improvements in EF and PS from T1 to T2. Some of the changes in PS and EF showed moderate associations to improvements in CWMT. EF also showed moderate associations to improvements in CWMT, even when controlling for PS. This supports that improvements are related to CWMT and not merely retest effects, and that CWMT improves both EF and PS. In addition, the depression history composite score showed negative associations to the CWIT Inhibition Switching change score, thus supporting an EF deficit associated with depression that could also impair EF improvements from CWMT. This could partly explain the relatively small improvement in this condition and suggests that previous history of illness could impair improvements in parts of the EF domain. These findings could have implications for the timing of interventions and could suggest that some patients may show better training effects for some domains, by utilizing CWMT early in their illness (Ahern and Semkovska, 2017). Perhaps, other interventions such as attention bias modification training (Koster and Hoorelbeke, 2015) or interventions targeting hot cognition are more useful than CWMT in later MDD. The CWMT improvement score also showed somewhat paradoxical associations to self-reported attention deficits, with gains being associated with poorer self-assessed cognition. Potential negative effects from CWMT underscore the importance of therapist-assisted interventions, psychoeducation, and/or broader interventions targeting functions beyond WM alone, in interventions targeting residual symptoms in remission from MDD. However, the small, strictly remitted sample, cautions firm conclusion regarding this.

This is the first study to investigate how CWMT effects e-WM, PS, EF, and symptoms and different forms of rumination combined. Pre-registration of hypotheses was done before analysis. In addition, the sample was selected and the intervention comprehensive. Demand characteristics could have resulted in underreporting of symptoms and could thus have influenced expected associations between MADRS and PS. In addition, dropout was considerable, but not higher than comparable studies. To reduce the chance of type 2 errors, correction of multiple comparisons was not strictly implemented, increasing risk for type 1 errors. The small sample size makes correlation estimates unstable (Schönbrodt and Perugini, 2013), and novel findings in this study should be considered preliminary before being replicated in larger samples. However, it should be noted that this study has higher sample size than 75% of the studies in a recent meta-analysis of cognitive remediation in MDD (Legemaat et al., 2021).

This study shows some overlap with a pilot study on the acceptability and feasibility of CWMT (Hammar et al., 2020). Hammar et al. (2020) investigated some cognitive outcomes of CWMT, namely, WM, inhibition, and rumination, with the two latter similar to this study. However, analyses in the pilot study were done before all the data were collected on these measures. In addition, this study investigated several other relevant aspects of cognitive function: e-WM, EF, PS, and depressive symptoms, in a complete dataset. This study focused on the clinically relevant distinctions of hot cognition, PS, and EF, in addition to the differing perspective of investigating the mechanisms of interrelationships between cognitive function and symptoms pre- and post-CWMT. To the best of the authors knowledge, this study is the first CWMT study to do to so, supporting some previous findings (Joormann et al., 2011; Ronold et al., 2020a), while not others (Ronold et al., 2020b), regarding the association between cognitive functioning and depressive symptoms and rumination.

This study design, lacking control group and randomization, makes it hard to control for placebo effects/demand characteristics and retest effects in the sample. However, given the size of improvements, and associations between improvements in CWMT and cognition, it is suggested that both PS and EF are improved by this intervention. Large longitudinal double-blinded randomized controlled studies should replicate and improve on the findings, investigate to what degree effects are clinically significant beyond placebo effects, transfer to daily functioning, should investigate how long effects last, and what clinical implications they have for relapse, recurrence, quality of life and daily functioning in the remitted phase of MDD. In addition, the change scores utilized in the explorative hypotheses do not correct for the correlated nature of the pre- and post-measures and must be interpreted with caution (Hayes and Rockwood, 2017). Larger samples should thus investigate mediating effects of MDD history on the effects of CWMT, and the associations between EF PS and CWMT.