Depression is among the major causes of global disease burden (Vos et al., 2012). In 2010, it was the second leading medical cause of disease burden, with particularly high numbers of disability in working age (Ferrari et al., 2013), making it a particularly costly disease for society. Although affective symptoms, namely low mood accompanied by low self-esteem and a loss of interest in activities that others find enjoyable, are at the core of current diagnostic criteria for depression, cognitive dysfunction also plays a crucial role. Importantly, while treatments such as psychotherapy or medication have proven effective in improving mood, cognitive deficits oftentimes persist (Baune et al., 2010). Objective measures of executive function, attention, and short- and long-term memory reveal cognitive impairments in individuals with major depression (Hammar and Ardal, 2009; Gonda et al., 2015). Additionally, there is a discordance between the objective measures of cognitive dysfunction and the subjectively reported cognitive complaints, the latter being even more pronounced (Moritz et al., 2004), which may be due to a bias toward negative cognitive schemata that is commonly observed in major depression (Svendsen et al., 2012; Lam et al., 2014).

Recent studies investigating the effects of training interventions using video games have shown improvements in cognitive function, although these studies have almost exclusively been performed on healthy young adults. Interestingly, these prior studies consistently show that action video game interventions in particular are associated with specific improvements in attention and processing speed (Green and Bavelier, 2003; Dye et al., 2009) as well as in executive functions such as flexible updating of task-relevant information (Colzato et al., 2013). A recent meta-analysis on the cognitive domains that benefit from commercial video game playing has shown that particularly executive functions (or “information processing” as they call it) show improvements (Powers et al., 2013). Many of these previous studies on video games have investigated expertise effects, that is, they compared habitual gamers with non-gamers cross-sectionally. However by now a considerable number of studies also demonstrate beneficial effects of video game training on cognition in longitudinal designs, supporting a causal relationship.

Previous studies investigating the effects of video games in the context of depression have mostly used specific custom-made games for training that explicitly included elements of cognitive behavioral therapy (CBT) and psychoeducation such as SPARX (Merry et al., 2012; Poppelaars et al., 2016). Other studies have used so-called exergames, which are video games that promote physical exercise. Since regular fitness training is known to improve depression (Kvam et al., 2016), it may not be overly surprising that these exergames likewise lead to an alleviation of symptoms (Li et al., 2016). However, we suspect that compliance with physical exercise regimes is generally lower and more patients will enjoy a video game rather than an exercise intervention. A recent survey conducted in the United States reported that in all age groups, more than 185 million people play and enjoy videogames (≈ 54%; Entertainment-Software-Association, 2017). First evidence for potentially positive effects of commercial video game playing on depression comes from a study that gave patients with major depression the choice between three different commercial puzzle video games (Russoniello et al., 2013). After an instructed 6 h of game play spread over 1 month, the video game group showed significant decreases in depression severity compared with a control group that surfed Internet websites on depression for the same amount of time. However, puzzle-like video games have not been associated as systematically with cognitive benefits encompassing several executive functions as action video games in healthy samples (Green and Bavelier, 2012; Green et al., 2012; Strobach et al., 2012). We therefore set out to test the effects of action video gaming on depressive symptoms and cognitive function in a depressed population.

Based on this prior evidence of improvement in the domain of executive functioning, in response to training with action video games in healthy populations, the present study set out to investigate the effects of action video game training on cognition and depression severity in depressed patients. Furthermore we also assessed symptoms associated with depression such as rumination, since rumination is an important predictor of depressive symptoms (Nolen-Hoeksema, 2000). We hypothesized that depressed patients who complete an action video game intervention would show a larger reduction in depressive symptoms and rumination, and an increase in both objective and subjective measures of cognition, compared with depressed patients in a passive (wait-list) control group.

Patients' demographic characteristics and clinical parameters at pretest are reported in Table 1. The only significant difference at pretest was the distribution of sex (χ² = 0.69, p = 0.012) and on the PHQ-9 score [t₍₄₈₎ = 2.41, p = 0.020], which was higher in the control group. Moreover, we found a trend for BDI score differences at pretest with the same direction [t₍₄₈₎ = 2.01, p = 0.050] and a tendency for more habitual video game playing in the video gaming group [t₍₄₈₎ = 1.99, p = 0.052]. Participants in the video game group on average reported to have played two to three times a week (mean = 3.77, SD = 2.1). However, due to the fact that participants were randomly assigned to the groups we assume that the differences in sex and PHQ-9 arose by chance and are not systematically related to the dependent variable which makes it feasible to use ANCOVA as a statistical procedure.

Our main focus was on potential changes in depressive symptoms but also on associated symptoms such as rumination and subjective cognition due to our video game intervention. In separate ANCOVAs we compared the video game group against the passive control group at posttest and found a significant effect of group reflecting lower rumination score in the video gaming group [F_{(1, 44)} = 72.65, p = 0.017, η_p² = 0.123, with the covariate of pretest rumination score being significant, p < 0.001, see Table 2], as well as significantly higher subjective cognition in the video gaming group at posttest [F_{(1, 44)} = 162.31, p = 0.043, η_p² = 0.090, with the covariate of pretest subjective cognition score, p < 0.001] and habitual video gaming (p = 0.035) being significant. For the same analysis on BDI scores [F_{(1, 44)} = 1.67, p = 0.167, η_p² = 0.043, with the covariate of pretest BDI score being significant, p < 0.001] and PHQ-9 scores [F_{(1, 44)} = 0.503, p = 0.482, η_p² = 0.011, with the covariate of pretest PHQ-9 score, p < 0.001] we did not find a significant group effects at posttest. When correcting for multiple testing by means of Bonferroni correction (three more or less independent tests with an overall p-value criterion of p < 0.05 results in a corrected p-value of 0.0125) none of the analyses presented would survive this correction. The changes from pre- to posttest in subjective cognition, BDI, PHQ9 and rumination were not associated with self-reported playtime at posttest (p > 0.354).

When exploring the subsample with data on cognitive performance (n = 10 control group, n = 9 video game training group, Table 3) we found a significant posttest group difference in performance on TMT performance when controlling for habitual video game playing, pretest TMT performance and accounting for sex, with the video game group showing faster performance compared with the control group [F_{(1, 19)} = 5.06, p = 0.030, η_p² = 0.103, the covariate of pretest performance, p = 0.034, was significant and sex was close to significance, p = 0.052]. We found no significant effects in the performance data of the other cognitive tests (p > 0.261). The TMT result does not survive when applying Bonferroni correction to adjust for multiple testing. TMT performance data was uncorrelated to the subjective cognition score at pretest [r₍₁₉₎ = −0.026, p = 0.916] as well as at posttest [r₍₁₉₎ = −0.273, p = 0.258].

To explore the finding of an increase in subjective cognition in the experimental training group in more detail, we decomposed the SSTICS total score according to the cognitive domains that the items of the SSTICS referred to. While controlling for habitual video game playing and the pretest score and accounting for sex, as described above, we found that the subscore of episodic memory was significantly higher [F_{(1, 44)} = 4.691, p = 0.036, η_p² = 0.096], and the semantic memory showing a tendency for higher values in the experimental group at posttest [F_{(1, 44)} = 3.368, p = 0.073, η_p² = 0.071].

Within the scope of the present training study we found results suggesting that a 6-week training with a fast paced action video game (Boson X) in depressed patients contributes to lower levels of rumination and higher levels of subjective cognitive ability at posttest in comparison to a control group (although these results did not survive multiple comparison correction). On a small subsample, we detected better processing speed and cognitive flexibility (TMT) in the training group at posttest compared with a passive control group when controlling for pretest performance, pointing into the direction of our hypothesis that fast paced action video games may train executive functioning.

Although we did not find effects on the measure of depression severity (BDI, PHQ-9), the observed reduction in rumination is highly relevant with respect to depression. The measure we used to assess rumination is based on the response style theory (Nolen-Hoeksema, 1987) according to which rumination is characterized by self-reflection and a passive focus on negative emotions. Rumination can exacerbate negative moods, and can both trigger and prolong depressive episodes (Nolen-Hoeksema et al., 2008).

Taken together our results are an interesting add on to findings of Russoniello and colleagues, who trained depressed patients with casual puzzle-like video games (Russoniello et al., 2013). But differently to Russoniello's finding of a reduction in PHQ-9 scale, our data suggest a reduction of rumination. The difference in outcome between this and our study might be due to the game genre used in the intervention. The previous literature on action video games suggests improvements on executive functions (Powers et al., 2013). Executive functions comprise the ability to inhibit ongoing actions as well as thoughts. Rumination in turn has been associated with deficits in brain areas associated with inhibition (Kühn et al., 2012, 2014b; Yang et al., 2017). Action video games may improve the ability to supress rumination via improvements in executive function, but less effectively target the affective symptoms of depression. In contrast, puzzle-like video games might be more rewarding and less frustrating, and therefore better suited to entertain patients and therewith cause changes in mood. Unfortunately, our cognitive battery did not include any inhibition tasks to explore this hypothesis further. However, this is a post-hoc speculation that may inspire future research. Interestingly, a previous study that applied a working memory training regime targeted at rumination to depressed patients failed to show a reduction in rumination (Wanmaker et al., 2015). This finding underlines the recently emerging notion that cognitive training regimes may train certain processes but often show little transfer effects to other non-trained domains (Melby-Lervag et al., 2016). One may interpret the finding as in line with the fact that we do observe a reduction of rumination, but no effect on our spatial working memory task (Corsi block tapping).

In addition to the observed reduction in rumination we found an effect on the subjective assessment of cognitive functioning, albeit neither of the effects survive Bonferroni correction. Although the latter subjective cognition scale we used was originally used in schizophrenia patients, the questions are not disease specific and the same cognitive domains (working memory, episodic memory, semantic memory, attention, executive functioning) are also highly relevant in depression. Although the size of the subsample from which we obtained cognitive data was admittedly fairly small, we did not observe an association between objective performance and subjective assessment of cognition. This is in line with previous research on depressed patients failing to show correlations between objective and subjective cognitive performance data (Srisurapanont et al., 2017). The authors demonstrated that age and depression severity may be predictive of the discrepancy between subjective and objective measures. In a study on a large healthy population cohort of whom 87% reported at least one symptom of depression, likewise no association between subjective memory complaints and objective memory performance was shown, although participants showed a decline in memory for pictures presented in a negative context (Schweizer et al., 2017).

Interestingly, in an additional analysis in which we set out to explore which cognitive domain contributed most to the overall effect of an increase in subjective cognition, we computed separate analyses for each domain and found that in particular episodic memory performance subjectively increased in response to the video game intervention. Unfortunately, we did not assess any neuropsychological tasks that tap into the episodic memory domain in the present study.

In the subsample from which we obtained cognitive performance data, we found a reduction in reaction time in the TMT, which is a measure of executive functioning. However, we did not find any effects on the spatial working memory task (Corsi block tapping task), the speed of processing task (DSST) or the two spatial processing tasks (Manikin task, Spatial relations task). The absence of an effect on processing speed is in line with a recent study on healthy adults who have been trained with an action video game that likewise found no improvement in a speed task (van Ravenzwaaij et al., 2014). However, a previously published study did show effects of action video gaming on speed of processing (Dye et al., 2009) and since Trail Making, for which we found a significant improvement in the present study, also captures speed of processing to a certain extent one may have expected to observe improvement in the Digit symbol substitution task as well.

We selected three spatial tasks (Corsi block tapping, Manikin task, Spatial relations task) based on the reasoning that the spatial nature of the tasks may be a proxy of hippocampal functioning and previous studies have demonstrated positive effects of action video gaming on spatial processing. In one study, pre-existing gender differences in spatial processing were eliminated after 10 h of action video game training and in another study, training with action video games was demonstrated to improve performance on a virtual surgery task in medical students (Feng et al., 2007; Schlickum et al., 2009). Since depression has repeatedly been associated with volume decrements in the hippocampal formation, and antidepressant use with neurogenesis in the hippocampus (Jacobs et al., 2000; Willner et al., 2013) we reasoned that spatial tests could be an indirect measure of brain structural changes known to be elicited by action video games (Kühn et al., 2014a).

A profound limitation of the study is the overall sample size, as well as the small sample size in the subgroup that completed the neuropsychological test battery before and after the intervention. This may have considerably reduced the power of the present design to detect effects.

Another critical point is that the observed effects, in particular on self-reported subjective cognition, may have been caused by the non-blind nature of the intervention and may reflect the participants' expectation that the game improved their cognition. However, at least in Germany, where the study was performed, video games are not very highly valued, which may also have lead participants to believe that video games should impair their cognition, although this was not assessed. Future work may usefully explore how factors such as expectation and motivation may influence the efficacy of video game interventions. Another limitation is that we were not able to evaluate potential associations between performance in the video game to the outcome, since we did not obtain the game logging information from all participants. Future studies should carefully log performance data as it may reveal insights into the particular elements of the game that drive the observed effects. Furthermore, the fact that the testing and training was accomplished by the participants at their home computer is a disadvantage because it is not well-controlled how much participants trained and whether participants engaged in other activities while filling in questionnaires or during the training. Future studies should likewise control the training time to investigate dose-response relationships.

Due to the fact that the control group in the present study did not perform any task, we cannot finally exclude that the observed effects are due to distraction in the experimental group that was absent in the control group. However, this is unlikely to be the only mechanism at work here, as we observed the predicted effects of action video games on executive function, similar to previous work in healthy individuals (Green and Bavelier, 2003; Dye et al., 2009; Colzato et al., 2013). Nevertheless, rather than a passive control group, future studies should use an active control group that gets an instruction to pursue a certain activity e.g., reading books on a tablet to control for differences in motivation and to exclude the possibility that the observed effects are purely due to distraction. As our questionnaires retrospectively ask about a time frame of 1–2 weeks, responses may have differed depending on whether participants were still engaged in the game or not. To prevent this, future studies should consider administering the posttest questionnaires 2 weeks after the end of the training interval.

Taken together, the present results suggest beneficial effects of training with an action video game on executive function, rumination and subjective cognition in a sample of depressed patients, albeit the results do not survive multiple comparison correction. These results pave the way for future studies that may set out to unravel the potentially mediating role of cognitive changes in the beneficial effects of action video game training on rumination in depressed patients. We did not observe significant improvements in affective symptoms following an action video game intervention. Other games, such as puzzle-like video games, might be better suited to improve mood in depression (Russoniello et al., 2013). Future work on video game interventions for depression may seek to combine both action and puzzle games, to target affective and cognitive symptoms simultaneously. A clear advantage of video game interventions is that they constitute a low-cost, feasible intervention that can be easily offered to patients alongside regular therapy, and even as an online offer to extend treatment to their home computer. Administration does not require a clinician, and can even be self-administered. Video games are enjoyable to play, and, unlike pharmacological interventions for depression, have not been found to produce adverse side effects so compliance is likely to be high. The games are also adaptive; as an individual continues to play the difficulty level (e.g., the speed at which the avatar moves) automatically increases until the game ends (the avatar falls out of the tunnel), at which point the game restarts. As such, through adaptation of the level of difficulty to the skill of the player, frustration that the game is too difficult and boredom that the game is too easy should be kept at a minimum and the player should remain engaged in the game. As such, video games provide a promising, novel, and highly practical therapeutic intervention for individuals with depression.

Contribution to the design of the study: SK, SM, and JG; to the acquisition of the data: SK, SM, FB, and TL; to data analysis: SK, SM, and FB; to drafting the manuscript: SK, SM, FB, TL, and JG. All authors take full responsibility for the content of the paper.