Introduction

Depression is the most common mental health disorder worldwide [] and is now the most common factor leading to lower workplace productivity [] and absence from work []. Targeting people with subclinical depression (ie, those “at risk”) is considered a priority for preventing future episodes [], and as subclinical depression and anxiety may impact workplace performance, companies are increasingly turning to smartphone apps targeting mental health and well-being to support their employees []. These typically draw on principles from cognitive behavioral therapy (CBT) or other psychological therapies, and purport to target the core characteristics of depression, such as low mood, negative biases, and anhedonia. Though not directly targeted by these apps, absenteeism from the workplace, low productivity, and engagement in unhealthy working patterns are linked to deficits in executive function in depressed populations, suggesting that cognitive function must also be targeted to improve work-related outcomes [].

Cognitive models suggest that impaired executive functions, such as inhibitory control, attention, and working memory, underlie both the onset and maintenance of depression [] and anxiety []. These models suggest that impairments in executive function result in a reduced ability to inhibit negative thoughts, shift attention away from them, or update negative information currently held in working memory in response to new, potentially positive, information, leading to the onset or maintenance of symptoms []. In the working population, executive function is positively related to employment status and the ability to effectively balance tasks and responsibilities []. Deficits in executive function have been identified in populations with depression [] and anxiety [], with the severity of symptoms being correlated with the magnitude of impairment []. Given that executive function has been found to have a reciprocal relationship with depressive symptoms [], with impairments in executive function predicting future depressive symptoms [,], continuing into remission, and worsening with repeated episodes [], executive function is a key target for intervention.

Computerized interventions targeting different facets of executive function, such as cognitive control, aim to target putative neurocognitive mechanisms underlying mental health disorders. By enhancing executive function, it is hypothesized that the affected individual is better able to inhibit or shift attention away from negative information, reducing the cognitive symptoms of depression and anxiety and improving mood. The appeal of executive function training lies in the promise of a low-cost, easily accessible intervention that translates findings from basic research to clinical utility. Accordingly, systematic reviews have found broadly positive effects of computerized executive function training on depressive symptoms [,]. For example, a commercially available app, NeuroNation (Synaptikon GmbH), was found to improve working memory capacity after 21 sessions of training []. However, the malleability of executive function has been debated [], and although research has found that some aspects of executive function can be improved by training (eg, working memory) [], this has not been consistently shown (eg, working memory did not improve after 2 weeks of training []). There are also concerns regarding whether the effects of these interventions generalize beyond the trained tasks []. In addition, owing to a lack of follow-up data, the duration of effects is unknown []. Overall, the literature has yielded mixed findings regarding the effects of executive function training, and explicit tests of the mechanisms of action would help in resolving this debate [].

Self-paced digital CBT offers another low-cost, accessible intervention, and it has been proposed as a solution to situations where people are currently on waiting lists for treatment or where clinicians are engaged in “watchful waiting” []. Executive function has been found to predict the response to CBT [,], enhance compliance with CBT homework practices [], and improve after successful treatment [], which may be due to CBT targeting and restructuring negative thoughts, thereby reducing the burden on executive function, which can support further change. Therefore, improving executive function may be a common pathway by which psychological interventions reduce the symptoms of depression and anxiety. Systematic reviews have found reductions in depressive symptoms with CBT delivered via automated text messages [] or via CBT-based apps [], though there is a lack of research investigating their effects on executive function. As with studies on executive function training, there are also concerns regarding the effectiveness of these interventions, given notably high attrition rates [] and a lack of research in real-world settings, with studies often conducted in undergraduate populations or often excluding participants who do not engage with the intervention. It is not clear whether these results would generalize to working adults who face, for example, multiple time pressures.

Positive outcomes, such as well-being, are equally important outcomes of interest, both as predictors of future depressive episodes [] and as intervention targets []. Work-related stress and well-being also predict future episodes of depression [], though they are rarely included as outcomes in intervention research. Understanding how interventions affect not only negative outcomes but also positive outcomes could provide a more complete picture of an intervention’s effectiveness and clarify for whom it is most effective. Additionally, though anxiety is highly prevalent in the working population and comorbid with depression, it is relatively underresearched compared with depression []. As anxiety is also associated with long-term absenteeism and reduced productivity [], exploring the effects of app-based interventions on anxiety symptoms and the potential mechanisms of action is of interest. Specifically, drawing on the Research Domain Criteria framework, executive function may be a transdiagnostic mechanism of action [] through which diverse interventions improve mental health and daily functioning. For example, improving executive function may result in more efficient filtering and regulation of negative information in working memory, promoting more neutral or positive interpretations and improving mood.

We therefore aimed to test the effectiveness of an executive function training app and a CBT-based app on mental health, well-being, and executive functioning in a working adult sample at risk of depression and anxiety disorders. We also aimed to test whether changes in these outcomes are mediated by changes in executive function, which would support models proposing that executive function is a common pathway through which treatments may work. This study, therefore, is a 3-arm randomized controlled trial investigating the effectiveness of an executive function training app (NeuroNation) and a CBT-based app (Moodfit, Roble Ridge LLC), compared with a waitlist control. NeuroNation delivers a variety of gamified executive function tasks each day, with gamification proposed to improve intrinsic motivation and training protocol adherence []. Moodfit is a self-paced mental health app using a variety of techniques from CBT, such as cognitive restructuring, as well as mindfulness and positive psychology. By assessing both negative symptoms, such as mood and anxiety, and positive outcomes, such as workplace well-being, at baseline, postintervention (after using the apps for 4 weeks), and follow-up (12 weeks), we aimed to provide a more comprehensive assessment of their effectiveness in real-world settings. We therefore tested the following hypotheses:

Hypothesis 1: There will be greater reductions in depressive (hypothesis 1a) and anxiety (hypothesis 1b) symptoms and greater increases in workplace well-being (hypothesis 1c) in the executive function training and self-guided CBT groups at postintervention and follow-up (12 weeks), compared with the findings in the waitlist control group.Hypothesis 2: Intervention-related changes in depressive (hypothesis 2a) and anxiety (hypothesis 2b) symptoms and workplace well-being (hypothesis 2c) will be mediated by improvements in executive function.
MethodsTransparency and Openness

Materials, code, and preregistration data for this study are available on the Open Science Framework (OSF) []. This study was preregistered on the OSF prior to commencing data collection, and data are available via the University of Bath’s Research Data Archive []. The preregistered sample size rationale is provided in .

Design

This study was a 3-armed randomized controlled trial with intervention (executive function training app, self-guided CBT app, and waitlist control) as the group variable and time (baseline, postintervention, and follow-up) as the repeated measure. Participants were allocated to groups via block randomization [], with block sizes of 6, 9, and 12, and a list created using Sealed Envelope (Sealed Envelope Ltd). The primary researcher (AM) had access to the allocation sequence and was not blinded to intervention assignment.

Ethical Considerations

This study was approved by the University of Bath Psychology Research Ethics Committee (number: 22-049). Participants were presented with study information and researcher contact details in the event of any questions, prior to signing a consent form online. Participants were reimbursed £10 (approximately US $13) per hour for completing testing sessions in the study or were offered course credit if recruited through the university’s research participation scheme. Participants were provided with links to local and national mental health support charities after each testing session and as part of the debriefing process. No identifying information of participants was recorded in the final dataset, and participants provided consent for sharing of their anonymized data.

Participants

Participants were eligible for inclusion if they were aged between 18 and 67 years (the state retirement age in the United Kingdom) and were either employed part-time or full-time or self-employed. Participants were included if they scored above 5 on the Patient Health Questionnaire-9 (PHQ-9) and above 4 on the Generalized Anxiety Disorder-7 (GAD-7), indicating clinically relevant mild symptoms of anxiety and depression []. Participants were ineligible if they were receiving talking therapy or were due to start talking therapy in the next 12 weeks. Computer literacy was assumed but was not an explicit inclusion criterion. A total of 240 participants were recruited via advertisements placed on social media and mental health charity research boards (eg, MQ Mental Health), research participation schemes (eg, SONA), the University of Bath’s community research participation panel, and Prolific Academic, as well as through word of mouth. Emails or Prolific messages were exchanged with applicants to protect against multiple or fraudulent submissions. After exclusion, our final sample included 228 participants (mean age 33.79, SD 11.50 years; female participants: 147/228, 64.5%).

Study GroupsWaitlist Control

Participants assigned to the waitlist control group completed questionnaires and the executive function task at the same time points as those receiving the active interventions and were not instructed to download an app. Previous research has found that waitlist control conditions may exaggerate treatment effects in randomized controlled trials []; however, given the differing “active” components of our 2 intervention conditions, a waitlist control was considered as a pragmatic option for comparison.

Executive Function Training

Participants allocated to the executive function training intervention were provided with a subscription to NeuroNation, an executive function and brain training app, on their Android or iOS phone. The app presents participants with a selection of games to complete each day, aimed at improving attention, speed, and memory. The app provides psychoeducation and details how each game could improve functioning in everyday situations, such as conversations with loved ones. Each game takes around 120 seconds, with 10 games in each training session. Participants were instructed to complete a minimum of 21 sessions during the 28-day intervention period. In each session, participants completed gamified executive function tasks such as modified digit span or trail-making tasks. NeuroNation has been used in previous research investigating the effects of executive function training on working memory and broader cognitive function [,,], with effect sizes ranging from f=0.23 [] to f=0.30 []. Screenshots from the app are provided in .

Self-Guided CBT

Participants allocated to the CBT intervention were provided with a subscription to Moodfit, a self-guided CBT and well-being app, on their Android or iOS phone. Moodfit was selected based on its layout, provision of core CBT techniques such as guided cognitive restructuring, and popularity. Participants were instructed to complete the mood journal each day and spend at least 10 minutes recording cognitive restructuring entries twice a day, for a minimum of 21 days over the 28-day intervention period. Screenshots from the Moodfit app are provided in .

Primary Outcomes

Depressive symptoms were measured with the PHQ-9 [], which is a 9-item rating scale, with possible response options ranging from 0 (“not at all”) to 3 (“nearly every day”). Questions are related to general depressive symptoms experienced over the previous 2 weeks (eg, “Little interest or pleasure in doing things?”). Higher scores represent greater symptom severity. This measure was chosen owing to its reliability and validity in the general population [], and it had a Cronbach α value of 0.82 in our sample at baseline.

Anxiety symptoms were measured with the GAD-7 [], which is a 7-item rating scale, with possible response options ranging from 0 (“not at all”) to 3 (“nearly every day”). Questions are related to the frequency of anxiety symptoms over the past 2 weeks (eg, “Not being able to stop or control worrying?”). Higher scores represent greater symptom severity. This measure was chosen owing to its reliability and validity in the general population [], and it had a Cronbach α value of 0.86 in our sample at baseline.

Workplace well-being was measured with the Utrecht Work Engagement Scale [], which is a 9-item rating scale, with responses ranging from 1 (“never”) to 7 (“every day”). Questions are related to engagement and motivation at work (eg, “I am immersed in my work”). This scale has been found to have good reliability and validity in previous research [], and it had a Cronbach α value of 0.93 in our sample at baseline.

Working memory capacity was assessed with the Operation Span (OSPAN) task []. Participants solved a set of simple arithmetic problems while simultaneously remembering a string of letters, with set sizes ranging from 3 to 7 letters. Set sizes were repeated thrice, for a total of 15, with the set order randomized by Inquisit 6 (Millisecond). Absolute scoring (the sum of all letters recalled in the correct position) was used, with higher OSPAN scores indicating greater working memory capacity.

Secondary Outcomes

Daily life stress was measured with the Survey of Recent Life Experiences [], which is a 41-item scale asking about the frequency of daily life stressors (eg, “being ignored” and “financial burdens”) over the past 4 weeks. Responses range from 1 (“not at all part of my life”) to 4 (“very much part of my life”), with higher scores indicating more stress over the past month. This scale has been shown to have good construct validity [], and it had a Cronbach α value of 0.90 in our sample at baseline.

Social workplace well-being was measured with the Eudaimonic Workplace Wellbeing Scale [], which is an 8-item, 5-point rating scale designed to measure the social dimension of well-being in a workplace context and general workplace well-being (eg, “I feel close to the people in my work environment”). Responses range from 1 (“strongly disagree”) to 5 (“strongly agree”), with higher scores indicating higher or more positive eudaimonic well-being in the context of the workplace. This scale has been shown to have good construct validity [], and it had a Cronbach α value of 0.85 in our sample at baseline.

General mental well-being was measured with the Short Warwick and Edinburgh Mental Well-being Health Scale []. This is a 7-item, 5-point rating scale designed to measure mental well-being, with questions related to the frequency of positive emotions or behaviors over the past 2 weeks (eg, “I’ve been feeling optimistic about the future” and “I’ve been dealing with problems well”). This scale has been shown to have good construct validity [], and it had a Cronbach α value of 0.81 in our sample at baseline.

Workplace absence, stressful events, and behaviors were measured via a series of binary response questions at each time point. Participants were asked whether they had experienced a stressful event at work in the last 7 days, taken time off work due to stress, worked outside of contracted hours or annual leave, or come in to work when they were ill in the past month.

Intervention adherence was measured via self-report as part of the postintervention testing session. Participants were asked if they had used the app on at least 21 days over the 4-week training period and asked about the number of sessions they thought they had completed.

Procedure

Participants completed baseline, postintervention (4 weeks after baseline), and follow-up (12 weeks after baseline) testing sessions online. Participants were presented with an information sheet on the recruitment site (available with data) prior to signing a consent form online. After gaining informed consent, questionnaires were completed online using Qualtrics (Qualtrics, LLC) and QuestionPro (QuestionPro Inc), followed by the OSPAN task on Inquisit 6. Upon completion of their first testing session, participants were emailed a link revealing their training group and provided with links to download and register their app, if they were assigned to one of the active intervention groups.

Statistical Analysis

All analyses were carried out using RStudio (Posit; see for packages and references). Missing data were assessed for randomness using the Little test [] and found to be missing completely at random at both postintervention (168/226 complete participants; 26% missing; χ28=6.94; P=.54) and follow-up (95/228 complete participants; 58% missing; χ28=8.00; P=.33). Missingness was similar across groups. Simulation-based power analyses found that the smallest detectable postintervention interaction effect size with 80% power would be β=3.29, and this would increase to β=3.95 after attrition at follow-up.

In line with our preregistered analysis plan, mixed linear models were conducted to test hypotheses 1a-c. In all models, participant was set as a random effect and compared to an intercept-only model to determine if model fit was improved by its inclusion. Group (dummy coded with waitlist control set as the reference), time (dummy coded with baseline set as the reference), and group × time interactions were included as predictor variables. We also conducted regression models for postintervention and follow-up scores of depressive and anxiety symptoms and workplace well-being while adjusting for baseline scores. These results are reported in . In all analyses, we controlled for multiple comparisons with Benjamini-Hochberg correction for the false discovery rate (FDR []) for each hypothesis separately (eg, hypothesis 1a, hypothesis 1b, and hypothesis 1c), in addition to related exploratory variables.

To test hypotheses 2a-c, mediation models determined whether intervention-related changes in depressive symptoms were mediated by changes in working memory capacity. Following information provided by Hayes and Rockwood [], we used the causal step approach for each outcome variable, and the significance of the mediation was determined with the bootstrap method (5000 bootstrap samples).

Exploratory Sensitivity Analysis

To explore any clinical relevance of our results, we conducted binomial logistic regression in a subsample (n=122) of participants who scored above 10 on the PHQ-9 and above 8 on the GAD-7 at baseline to explore the effect of intervention group on recovery and the minimal clinically important difference in depressive symptoms. Recovery, according to the National Health Service Improving Access to Psychological Therapies definition, is moving from caseness (PHQ-9 score above 10 or GAD-7 score above 8) to below caseness after treatment []. Given our subclinical population, the minimal clinically important difference was set as a 20% reduction in the PHQ-9 score [,] for this subsample.

Exploratory Analysis

We explored the effect of the intervention group on general well-being and daily stress with linear models adjusted for baseline well-being and daily stress scores, respectively. We additionally explored the effect of training group on incidents of workplace stress, absenteeism, presenteeism, and working while on leave. Finally, we explored the relationships of baseline depression and anxiety with OSPAN task performance. These results are presented in .

ResultsSample Characteristics

Our sample included 228 participants (mean age 33.79, SD 11.50 years; range 18‐67 years; female participants: 147/228, 64.5%). The CONSORT (Consolidated Standards of Reporting Trials) checklist is presented in , and the CONSORT flow diagram is presented in . No baseline differences in age (F2,225=0.01; P=.97) or gender (χ22=0.28; P=.87) were found between the groups, with most participants being female, White, single, and an employee of a company (rather than self-employed; see ). Most participants (155/228, 68.0%) had experienced at least one stressful event in the month leading up to their participation in the study, and 161 (71%) reported taking at least 1 day off work due to stress or mental health concerns. Posttraining self-reported intervention adherence was reasonable, with 89% (48/54) adherence to the NeuroNation (executive function training) app and 96% (52/54) adherence to the Moodfit (CBT-based) app.

‎Figure 1. CONSORT diagram showing participant flow through the study. Table 1. Sample characteristics at baseline.VariableWaitlist control group (n=75)Executive function training (NeuroNation) group (n=74)Self-guided CBT (Moodfit) group (n=79)Age (years), mean (SD)33.71 (12.44)33.66 (11.65)33.98 (10.65)Female, n (%)46 (61)49 (65)52 (66)Ethnicity, n (%)Asian or British Asian8 (11)6 (8)9 (11)East Asian0 (0)3 (4)2 (3)Black/Black British5 (7)0 (0)1 (1)Mixed3 (4)1 (1)2 (3)White45 (60)53 (72)56 (71)Other3 (4)1 (1)0 (0)Did not answer11 (14)10 (14)9 (11)Marital status, n (%)Divorced3 (4)2 (3)1 (1)Married/living with partner24 (32)27 (36)28 (35)Single30 (40)30 (41)34 (43)Separated/widowed3 (4)2 (3)2 (3)Did not answer15 (20)13 (17)14 (18)Employment status, n (%)Employee48 (64)49 (66)51 (65)Management5 (7)7 (9)6 (8)Self-employed5 (7)4 (5)8 (10)Did not specify17 (22)14 (20)14 (17)Events experienced in the last month, n (%)Stressful event51 (68)48 (66)56 (71)Time off work due to stress50 (67)52 (71)59 (75)Training adherence (self-reported at posttraining; n=54)Completed >21 sessions, n (%)—48 (89)52 (96)Sessions completed, mean (SD)—22.59 (11.39)24.57 (10.95)

aCBT: cognitive behavioral therapy.

bNot applicable.

Effects on Mental Health and Workplace Well-Being OutcomesOverview

Descriptive statistics for all primary outcomes are presented in . The groups did not differ in terms of outcome measures at baseline.

Table 2. Results for the primary outcome variables.Variable and time pointWaitlist control group (n=75), mean (SD)Executive function training (NeuroNation) group (n=74), mean (SD)Self-guided CBT (Moodfit) group (n=79), mean (SD)P valueDepressive symptoms (PHQ-9)Baseline10.08 (5.78)10.03 (5.24)9.96 (4.97).99Postintervention8.63 (5.64)7.72 (5.37)6.85 (5.42)—Follow-up8.86 (5.38)7.52 (6.07)8.29 (4.75)—Anxiety symptoms (GAD-7)Baseline9.11 (5.30)9.41 (4.67)9.15 (4.33).92Postintervention7.97 (4.85)7.00 (4.35)6.74 (4.17)—Follow-up8.63 (5.57)6.58 (4.91)7.15 (4.48)—Workplace well-being (UWES-9)Baseline41.09 (14.38)42.03 (11.33)41.41 (11.38).90Postintervention38.97 (13.69)43.80 (12.24)43.81 (11.53)—Follow-up40.31 (13.32)43.81 (13.87)43.47 (11.85)—Working memory capacity (OSPAN)Baseline40.30 (18.59)36.11 (18.69)40.87 (19.34).25Postintervention45.72 (17.99)44.32 (20.00)42.08 (17.91)—Follow-up44.61 (19.56)42.36 (21.64)49.38 (16.57)—

aCBT: cognitive behavioral therapy.

bBaseline differences were tested using 1-way ANOVA.

cPHQ-9: Patient Health Questionnaire-9.

dNot applicable.

eGAD-7: Generalized Anxiety Disorder-7.

fUWES-9: Utrecht Work Engagement Scale-9.

gOSPAN: Operation Span.

Effects of Executive Function Training or Self-Guided CBT on Depressive Symptoms

Addressing hypothesis 1a, a mixed linear model testing the effects of executive function training or self-guided CBT on depressive symptoms () found a main effect of time (χ²2=35.69; P<.001), suggesting that depressive symptoms decreased in all groups across the study, though there was no group × time interaction (χ²4=7.63; P=.11). We found an effect of assignment to the NeuroNation group on depressive symptoms at the 12-week follow-up (b=−2.77, 95% CI −4.96 to −0.56; P=.02), suggesting an average reduction of 2.77 points in the PHQ-9 score compared with the finding in the waitlist control group, partially supporting hypothesis 1a (see ), though these effects did not survive Benjamini-Hochberg correction (adjusted P=.06). There was no effect of assignment to the self-guided CBT app at either time point.

Table 3. Parameter estimates for mixed linear models involving depressive symptoms (PHQ-9), anxiety symptoms (GAD-7), and workplace well-being (UWES-9).VariableβSE95% CIP valueAdjusted P valueDepressive symptoms (PHQ-9)Group1.67——.44—Waitlist control (reference)—————NeuroNation−0.150.87−1.85 to 1.56.87—Moodfit0.090.87−1.61 to 1.79.92—Time35.69——<.001—Baseline (reference)—————Postintervention−1.320.64−2.57 to −0.08.04—Follow-up−0.300.79−1.83 to 1.23.71—Group × time7.63——.11—NeuroNation × postintervention−1.090.93−2.89 to 0.72.24.24Moodfit × postintervention−1.510.92−3.30 to 0.29.10.26NeuroNation × follow-up−2.771.13−4.96 to −0.56.02.06Moodfit × follow-up−1.561.12−3.74 to 0.62.16.22Anxiety symptoms (GAD-7)Group2.01——.37—Waitlist control (reference)—————NeuroNation0.220.77−1.28 to 1.72.78—Moodfit−0.210.76−1.70 to 1.28.78—Time41.16——<.001—Baseline (reference)—————Postintervention−1.150.55−2.22 to −0.08.04—Follow-up−0.070.67−1.39 to 1.24.91—Group × time9.58——.048—NeuroNation × postintervention−1.340.79−2.88 to 0.21.09.12Moodfit × postintervention−1.240.79−2.78 to 0.29.12.12NeuroNation × follow-up−2.790.97−4.68 to −0.91.004.02Moodfit × follow-up−1.860.96−3.73 to 0.01.054.11Workplace well-being (UWES-9)Group1.34——.51—Waitlist control (reference)—————NeuroNation0.672.10−3.43 to 4.76.75—Moodfit−0.112.09−4.19 to 3.97.96—Time1.04——.59—Baseline (reference)—————Postintervention−1.431.10−3.58 to 0.72.20—Follow-up−2.231.36−4.88 to 0.43.10—Group × time8.64——.07—NeuroNation × postintervention2.701.61−0.43 to 5.83.09.09Moodfit × postintervention3.721.590.62 to 6.82.02.046NeuroNation × follow-up3.671.98−0.17 to 7.51.06.09Moodfit × follow-up4.461.940.67 to 8.24.02.046

aPHQ-9: Patient Health Questionnaire-9.

bGAD-7: Generalized Anxiety Disorder-7.

cUWES-9: Utrecht Work Engagement Scale-9.

dNot applicable.

eSignificant (P<.05).

fResult survived Benjamini-Hochberg correction for the false discovery rate.

‎Figure 2. Change in depressive symptoms in each intervention group over time. PHQ-9: Patient Health Questionnaire-9.

In post hoc sensitivity analyses of recovery and minimal clinically important differences (), we found that those assigned to the executive function training and self-guided CBT groups were 4.77 times and 3.82 times more likely than the waitlist control group to no longer meet caseness criteria after the intervention, respectively. When assessing minimal clinically important differences, the executive function training and self-guided CBT groups were 3.54 and 3.52 times more likely than the waitlist control group to experience a clinically meaningful improvement in depressive symptoms, respectively (). These results survived FDR correction (all P<.05; detailed results are presented in ). However, there was no effect of group assignment at the 12-week follow-up testing session.

Table 4. Results for binomial logistic regression models investigating the effect of training on the likelihood of recovery and experiencing a minimal clinically meaningful reduction in symptoms.ModelTime pointExecutive function training (NeuroNation) group (reference: waitlist control group)Self-guided CBT (Moodfit) group (reference: waitlist control group)Exp(b), value (95% CI)P valueExp(b), value (95% CI)P valuePHQ-9, recovery4 weeks4.77 (1.59-14.30).005,3.82 (1.25-11.69).02,PHQ-9, recovery12 weeks5.33 (1.28-22.19).02,2.00 (0.49-8.24).34PHQ-9, MCID4 weeks3.54 (1.19-10.50).02,3.66 (1.13-10.88).03,PHQ-9, MCID12 weeks1.78 (0.44-7.18).420.83 (0.21-3.35).80GAD-7, recovery4 weeks3.75 (1.38-10.17).009,1.96 (1.58-12.14).005,GAD-7, recovery12 weeks3.18 (0.72-12.94).113.50 (0.85-14.41).08GAD-7, MCID4 weeks1.84 (0.69-4.87).223.13 (1.07-9.09).04GAD-7, MCID12 weeks1.33 (0.38-4.73).662.33 (0.62-8.82).21

aCBT: cognitive behavioral therapy.

bPHQ-9: Patient Health Questionnaire-9.

cSignificant (P<.05).

dResult survived Benjamini-Hochberg correction for the false discovery rate.

eMCID: minimal clinically important difference.

fGAD-7: Generalized Anxiety Disorder-7.

With regard to hypothesis 2a, the causal step mediation model found no mediating effect of change in executive function on intervention-related changes in depressive symptoms from pre- to postintervention (b=0.02, 95% CI −0.16 to 0.26; P=.84; ). Taken together, these results suggest that assignment to the executive function training group reduced depressive symptoms when compared with assignment to the waitlist control group, partially supporting hypothesis 1a, though there was no effect of assignment to the self-guided CBT group in our preregistered analyses. Exploratory analyses suggested that among those with moderate symptoms of depression, assignment to an intervention group resulted in improved outcomes. However, these improvements were not mediated by changes in executive function, and thus, the findings fail to support hypothesis 2a.

Table 5. Parameter estimates for mediation models.Variableβ95% CIP valueΔOSPANNeuroNation1.29−6.40 to 8.99.74Moodfit−7.16−14.85 to 0.54.07ΔPHQ-9NeuroNation−1.30−3.40 to 0.79.22Moodfit−1.32−3.44 to 0.80.22ΔOSPAN0.01−0.03 to 0.06.51Indirect effect0.02−0.17 to 0.26.84ΔGAD-7NeuroNation−1.49−3.24 to 0.25.09Moodfit−0.86−2.61 to 0.90.34ΔOSPAN0.03−0.00 to 0.07.07Indirect effect0.04−0.20 to 0.31.71ΔUWES-9NeuroNation2.23−1.16 to 5.61.20Moodfit3.520.10 to 6.94.04ΔOSPAN−0.05−0.12 to 0.02.17Indirect effect−0.06−0.52 to 0.25.70

aChange scores are calculated from pre- to postintervention.

bOSPAN: Operation Span.

cPHQ-9: Patient Health Questionnaire-9.

dGAD-7: Generalized Anxiety Disorder-7.

eUWES-9: Utrecht Work Engagement Scale-9.

fSignificant (P<.05).

Effects of Executive Function Training or Self-Guided CBT on Anxiety Symptoms

Addressing hypothesis 1b, the mixed linear model testing the effects of executive function training or self-guided CBT on anxiety symptoms () found a main effect of time (χ²2=41.16; P<.001), suggesting that anxiety symptoms d