Executive function (EF) is defined as a collection of advanced cognitive processes to manage purpose-guided actions and thoughts, encompassing abilities such as adaptability, memory, and planning (Friedman & Miyake, Reference Friedman and Miyake2017). EF has been theorized to include inhibitory control (or inhibition; managing undesired or unhelpful behaviors), shifting (cognitively transitioning between various tasks), and working memory (WM; processing real-time inputs while attending to the task at hand; Miyake & Friedman, Reference Miyake and Friedman2012) in related yet dissociable ways (Rodríguez-Nieto et al. Reference Rodríguez-Nieto, Seer, Sidlauskaite, Vleugels, Van Roy, Hardwick and Swinnen2022). Verbal fluency (VF) has been further recognized as a distinguishable EF component, connecting to more specialized language-based WM domains (Gustavson et al., Reference Gustavson, Panizzon, Franz, Reynolds, Corley, Hewitt and Friedman2019). Relatedly, attentional control (AC; subjective ability to maintain stable focus on the task at hand) has been posited to underpin EF, as it is closely related to meta-cognitive processes (Drigas & Karyotaki, Reference Drigas and Karyotaki2017). Since problems with EF and AC are linked to various biopsychosocial issues (Lawson, Hook, & Farah, Reference Lawson, Hook and Farah2018; Woon, Farrer, Braman, Mabey, & Hedges, Reference Woon, Farrer, Braman, Mabey and Hedges2017; Yang, Shields, Guo, & Liu, Reference Yang, Shields, Guo and Liu2018), developing efficacious interventions to enhance EF outcomes is crucial.
Mindfulness-based interventions (MBIs) have been theorized as potentially efficacious for EF by possibly enhancing resting-state functional connectivity between brain regions linked to EF (Taren et al., Reference Taren, Gianaros, Greco, Lindsay, Fairgrieve, Brown and Creswell2017). Further, intentionally focusing on the current moment without judgment and fully accepting emotions could aid in reorienting attention to the task and intended aims (Lutz et al., Reference Lutz, Slagter, Rawlings, Francis, Greischar and Davidson2009), possibly enhancing AC and EF. Concordant with these propositions, a qualitative review observed that even brief MBIs could strengthen inhibition and WM EF domains (Zhou, Liu, & Deng, Reference Zhou, Liu and Deng2020). Extending that report, a meta-analysis of 111 randomized controlled trials (RCTs) found that MBIs v. active controls produced small-to-medium efficacy on executive attention, WM, inhibition, and set-shifting but did not differentially affect VF (Zainal & Newman, Reference Zainal and Newman2024b). Together, enhanced AC and EF might function as change mechanisms of the impact of MBIs on clinical outcomes, such as reductions in depression and anxiety symptoms (Blanck et al., Reference Blanck, Perleth, Heidenreich, Kröger, Ditzen, Bents and Mander2018; Spijkerman, Pots, & Bohlmeijer, Reference Spijkerman, Pots and Bohlmeijer2016).
Mediation research offers invaluable perspectives into proxy change mechanisms, frequently recognized as the central methodology in the exploration of how or why MBIs or other therapies work (Kazdin, Reference Kazdin2007). It determines whether a mediator variable statistically elucidates the link between treatment and its outcome. The implication is that enhancing our grasp of how MBI change mechanisms occur enables clinicians to focus more precisely on essential EF elements while eliminating ineffective targets. Such efforts might lead to more effective treatments that offer quicker and stronger clinical enhancements (Maddock & Blair, Reference Maddock and Blair2021).
Only a few RCTs examined EF as a mediator of the effect of MBIs. In Noone and Hogan (Reference Noone and Hogan2018), performance-based EF did not mediate the effect of MBI on critical thinking and associated skills in undergraduates. Relatedly, in a cross-sectional study, higher trait mindfulness predicted more positive affect and less negative affect via better self-reported EF in everyday contexts in undergraduates (Short, Mazmanian, Oinonen, & Mushquash, Reference Short, Mazmanian, Oinonen and Mushquash2016). However, both studies recruited unselected undergraduates. Unselected samples are problematic since task-based EF and AC have been shown to correlate frequently with high levels of psychiatric symptoms (Abramovitch, Short, & Schweiger, Reference Abramovitch, Short and Schweiger2021).
Improvement in AC and EF could plausibly be a proxy change mechanism of an MBI, and such changes would precede the alleviation of worry and other repetitive negative thinking (RNT) tendencies for people with generalized anxiety disorder (GAD). Attentional control theory (ACT; Eysenck & Derakshan, Reference Eysenck and Derakshan2011) and the cognitive model (Hirsch & Mathews, Reference Hirsch and Mathews2012) posit that EF and associated issues, such as biased attention toward threat and unhelpful interpretations, generate excessive and uncontrollable worry (core GAD symptoms). Buttressing ACT, a meta-analysis showed negative correlations between rumination and both inhibition (Pearson's r = −0.23) and set-shifting (r = −0.19; Yang, Cao, Shields, Teng, and Liu, Reference Yang, Cao, Shields, Teng and Liu2017). Supporting the cognitive model, compromised set-shifting, inhibition, WM, and inductive reasoning abilities predicted GAD diagnosis and increased symptom severity after nine years among community adults (Zainal & Newman, Reference Zainal and Newman2018).
Additionally, scar theories (McEwen & Gianaros, Reference McEwen and Gianaros2011; Ottaviani et al., Reference Ottaviani, Thayer, Verkuil, Lonigro, Medea, Couyoumdjian and Brosschot2016) posit that worry and RNT trigger persistent activation and disturbance of interconnected neuroendocrine and immune regulatory systems that may build up allostatic load over time. Allostatic load is defined as the gradual deterioration of the hypothalamic-pituitary-adrenal axis (HPA) and related systems across time (McEwen & Seeman, Reference McEwen and Seeman1999), potentially impacting EF-implicated brain areas (Juster, McEwen, & Lupien, Reference Juster, McEwen and Lupien2010). Concurring with scar theories, heightened excessive worry and other GAD symptoms predicted future EF declines (Zainal & Newman, Reference Zainal and Newman2021), and increased inflammation consistently mediated this prospective association in separate samples (Zainal & Newman, Reference Zainal and Newman2022a, Reference Zainal and Newman2022b). Collectively, since AC and EF issues are bidirectionally related to pathological worry and other RNT habits, MBIs might be efficacious for GAD by improving AC and EF as proxy change mechanisms.
On that note, typical MBIs included 8–16 weeks of mindfulness-based stress reduction (MBSR; Kabat-Zinn, Reference Kabat-Zinn1990) and mindfulness-based cognitive therapy (MBCT; Williams, Russell, & Russell, Reference Williams, Russell and Russell2008), coupled with daylong meditation retreats (Creswell, Reference Creswell2017). Nevertheless, research uniformly showed that most people with GAD would not seek out and attend traditional face-to-face psychotherapy (including MBIs; e.g. Olfson, Blanco, Wall, Liu, & Grant, Reference Olfson, Blanco, Wall, Liu and Grant2019), and some of them would instead prefer to solve their mental health struggles independently (Goetter et al., Reference Goetter, Frumkin, Palitz, Swee, Baker, Bui and Simon2020; Rackoff, Fitzsimmons-Craft, Taylor, Wilfley, & Newman, Reference Rackoff, Fitzsimmons-Craft, Taylor, Wilfley and Newman2023). Scalable, evidence-based mental health apps (or ecological momentary interventions; EMIs) might somewhat solve this issue (Marciniak et al., Reference Marciniak, Shanahan, Rohde, Schulz, Wackerhagen, Kobylinska and Kleim2020). EMIs use experience sampling methods to offer individualized assistance in real-time by recognizing a person's inner and outer contexts and emotional struggles (Henry et al., Reference Henry, Hansen, Chimoff, Pokstis, Kiderman, Naim and Brotman2024). Mindfulness EMIs (MEMIs), particularly those with mood-tracking attributes, exhibited modest yet notable efficacy against self-monitoring placebos (SM) on GAD symptoms (cf. recent meta-analysis by Linardon et al., Reference Linardon, Torous, Firth, Cuijpers, Messer and Fuller-Tyszkiewicz2024). These findings suggested that a MEMI against SM might reduce GAD symptoms and RNT tendencies by enhancing AC and EF.
The current study was a secondary analysis of an RCT for GAD. In prior reports, a 14-day MEMI, compared to SM, reduced RNT and GAD severity (Zainal & Newman, Reference Zainal and Newman2024c, Reference Zainal and Newman2023b) and enhanced various empathy domains (Zainal & Newman, Reference Zainal and Newman2024a) to a greater degree from pre-treatment to one-month follow-up (1MFU). Two new hypotheses based on theory and evidence reviewed were tested. Hypothesis 1 predicted that a MEMI would significantly outperform SM in enhancing pre-post AC, inhibition, set-shifting, VF, and WM. Hypothesis 2 predicted that the effect of a MEMI against SM on reducing pre-1MFU GAD severity and trait RNT would be significantly mediated by improved pre-post AC, inhibition, set-shifting, VF, and WM.
Method
Study design
The Pennsylvania State University Institutional Review Board granted ethical permission to conduct our study. Our pre-registered RCT (NCT04846777 and https://osf.io/7g4su) used a mixed design involving two treatments (MEMI and SM) and three time points (pre-randomization, post-intervention, and 1MFU). Online Supplementary Appendix A offers an extensive summary of the methodology, detailing compensation details, power analysis, and pre-randomization measures. Figure 1 presents the CONSORT (Consolidated Standards of Reporting Trials) diagram illustrating participant flow from enrollment through study completion (Montgomery et al., Reference Montgomery, Grant, Mayo-Wilson, Macdonald, Michie, Hopewell and Group2018).
Figure 1. CONSORT flowchart of participant recruitment and progress.
Note: CONSORT, Consolidated Standards of Reporting Trials; GAD, generalized anxiety disorder; GADQ-IV, GAD questionnaire- fourth edition; MEMI, mindfulness ecological momentary intervention; SMP, self-monitoring app or placebo.
Eligibility criteria
Individuals were recruited from the subject pool and local community via initial screening of GAD criteria based on the Generalized Anxiety Disorder Questionnaire-IV (GADQ-IV; Newman et al., Reference Newman, Zuellig, Kachin, Constantino, Przeworski, Erickson and Cashman-McGrath2002), a minimum age of 18 years, and ownership of an iPhone or Android smartphone. They also had to be actively seeking treatment but not presently receiving any intervention (e.g. psychotropic medications) for mental health. Relatedly, they had to be meditation-naïve and report no prior experience with structured mindfulness practices. Those who consented were invited to participate in a 30-minute clinical interview using the Anxiety and Related Disorders Interview Schedule for DSM-5 (ADIS-5; Brown & Barlow, Reference Brown and Barlow2014) to confirm Diagnostic and Statistical Manual-Fifth Edition Text Revision (DSM-5-TR; American Psychiatric Association, 2022) GAD diagnosis. Exclusion criteria were presence of substance abuse disorders, suicidal thoughts, manic episodes, or psychotic disorders.
Participants
We enrolled a total of 110 participants diagnosed with GAD and meeting study eligibility criteria; 42 were randomly assigned to SM and 68 to MEMI. Table 1 presents the attributes of recruited participants. There was no between-group difference in baseline diagnoses (alcohol use disorder, anorexia nervosa, binge-eating disorder, major depressive episode [current or recurrent], obsessive-compulsive disorder [OCD], panic disorder, post-traumatic stress disorder [PTSD], social anxiety disorder [SAD], substance use disorder).
Table 1. Sociodemographic data of study participants in the mindfulness ecological momentary intervention (MEMI) and self-monitoring app (SM) (N = 110)
Note. GAD-Q-IV, generalized anxiety disorder questionnaire-fourth edition; OCD, obsessive-compulsive disorder; PTSD, post-traumatic stress disorder.
Pre–post mediator measures
AC
The Attentional Control Scale (ACS; Derryberry & Reed, Reference Derryberry and Reed2002) comprised 20 self-assessed items, combining a 9-item attentional focus measure with an 11-item attentional shifting subtest. It demonstrated strong convergent validity (agreement with related measures), good predictive validity (Judah, Grant, Mills, & Lechner, Reference Judah, Grant, Mills and Lechner2014), satisfactory discriminant validity (differentiation from unrelated constructs; Williams, Rau, Suchy, Thorgusen, & Smith, Reference Williams, Rau, Suchy, Thorgusen and Smith2017), and high retest reliability (Abasi, Mohammadkhani, Pourshahbaz, & Dolatshahi, Reference Abasi, Mohammadkhani, Pourshahbaz and Dolatshahi2017). The alpha (internal consistency) values in the present study were .87, .90, and .90 at pre-randomization, post-intervention, and 1MFU, respectively. Higher scores denoted better AC skills.
Inhibition
The Delis-Kaplan Executive Function System (D-KEFS; Delis, Kaplan, & Kramer, Reference Delis, Kaplan and Kramer2001) Color-Word Inhibition Test (CWIT) Condition 3, a modification of the classic Stroop test (Stroop, Reference Stroop1935), measured performance-based inhibition. Participants viewed a matrix of 50 color words with three ink colors (blue, red, and green). They were directed to verbalize the ink color as quickly and accurately as possible and to abstain from naming the word. The CWIT offers an advantage in measuring inhibition by eliminating the confounding effect of triggering symptoms with emotional words (Hsu & Davison, Reference Hsu and Davison2017). CWIT scores were correlated with GAD symptoms (Beaudreau et al., Reference Beaudreau, Hantke, Mashal, Gould, Henderson and O'Hara2017; Dorenkamp, Irrgang, & Vik, Reference Dorenkamp, Irrgang and Vik2023). The D-KEFS CWIT had good reliability and construct validity in assessing EF (Delis, Kramer, Kaplan, & Holdnack, Reference Delis, Kramer, Kaplan and Holdnack2004; Homack, Lee, & Riccio, Reference Homack, Lee and Riccio2005). Higher scores were indicative of poorer performance.
Set-shifting
The paper-based Trail Making Test-B (TMT-B) assessed set-shifting, attention, and processing speed (Army Individual Test Battery, 1944). Participants were required to alternately connect numbers and letters sequentially in ascending order as quickly as possible, with the assessor recording their completion time or speed (Spreen & Strauss, Reference Spreen and Strauss1998). The TMT-B had good retest reliability (Bracken, Mazur-Mosiewicz, & Glazek, Reference Bracken, Mazur-Mosiewicz and Glazek2019) and strong convergent validity (Osuka, Kojima, Sakurai, Watanabe, & Kim, Reference Osuka, Kojima, Sakurai, Watanabe and Kim2020; Plotnik et al., Reference Plotnik, Doniger, Bahat, Gottleib, Gal, Arad and Heled2017). Higher TMT-B scores indicated poorer set-shifting skills.
VF
VF was assessed via the D-KEFS VF subtest, comprising three progressively challenging subtests (Delis et al., Reference Delis, Kaplan and Kramer2001). This encompassed Letter Fluency (swiftly naming words beginning with a designated letter), Category Fluency (swiftly naming words within a specified domain), and Category Switching (promptly transitioning between semantic domains). The D-KEFS VF subtest evidenced good psychometric reliability and construct validity (Suchy & Brothers, Reference Suchy and Brothers2022). Participants were allotted 1 min in each test to generate as many words as possible. Higher sum scores on these subtests indicated better VF ability.
WM
Letter-number sequencing (LNS) in the Wechsler-Adult Intelligence Scale Fourth Edition (WAIS-IV; Wechsler, Reference Wechsler2008) was used because it is considered the best WM assessment (Crowe, Reference Crowe2000; Salthouse, Reference Salthouse1996; Shelton, Elliott, Hill, Calamia, & Gouvier, Reference Shelton, Elliott, Hill, Calamia and Gouvier2009). The evaluator verbally presented a sequence of alphanumeric strings at a pace of about one string per second. Participants were instructed to listen and reiterate successively longer alphanumeric strings by verbalizing the numbers before the letters in ascending order (Reynolds, Reference Reynolds1997). LNS captured WM, auditory processing speed, attention span, and cognitive manipulation. It has strong internal consistency, retest reliability, and construct validity (Vora, Varghese, Weisenbach, & Bhatt, Reference Vora, Varghese, Weisenbach and Bhatt2016). The sum of correct responses from all three trials was calculated, and higher scores reflected better WM ability.
Pre-1MFU outcome measures
GAD severity
GAD symptom severity was evaluated using the 16-item GAD-Q-Dimensional assessment, which closely resembled and was adapted from the GAD-Q-IV (Newman et al., Reference Newman, Zuellig, Kachin, Constantino, Przeworski, Erickson and Cashman-McGrath2002) but consistently used 9-point Likert scale response options (e.g. 0 = Never to 8 = Almost Every Day). The first eight items assessed individuals' levels of trait anxiety, with respondents evaluating the extent, frequency, intensity, and manageability of their worry across their lifetime. The subsequent eight questions posed similar queries regarding the preceding half-year (α values: .96, .97, .97).
Trait RNT
The 45-item Perseverative Cognitions Questionnaire (PCQ) measured enduring cognitive attributes associated with brooding, obsessive thinking, and worry (Szkodny & Newman, Reference Szkodny and Newman2019). Participants rated items on a 6-point Likert scale (0 = Strongly Disagree to 5 = Strongly Agree). The PCQ has six factors: anticipating adverse outcomes, exploring reasons and meanings, perceiving lack of control, preparing for the future, ruminating on past events, and thoughts incongruent with one's ideal self. The overall score was determined by summing items of each subscale. It had robust retest reliability over two weeks, strong convergent and discriminant validity (Szkodny & Newman, Reference Szkodny and Newman2019), and good cross-cultural equivalence (Zainal, Newman, & Hong, Reference Zainal, Newman and Hong2021; α values = .96, .97, .97).
Group arms
MEMI
MEMI participants accessed a video featuring the principal researcher, a doctoral-level clinical psychologist. The video taught them to use evidence-supported components in accordance with the principles derived from MBSR (Kabat-Zinn, Reference Kabat-Zinn1990). They were introduced to the concept of mindfulness and given precise guidance to immerse themselves in their immediate environment and actively participate in meaningful activities. This portion was designed to give them the capability of open monitoring and acceptance, improving their capacity to concentrate on intricate details. Following that, the video therapist guided unhurried diaphragmatic breathing techniques with a practical demonstration of the optimal procedure. This segment encompassed instructions on cultivating tranquility via optimized breathing routines and nurturing mindfulness qualities such as non-reactive observation and non-judgment, borrowing inspiration from the principles of MBCT (Segal, Williams, & Teasdale, Reference Segal, Williams and Teasdale2002). Subsequently, the video therapist stressed integrating mindfulness into daily activities. The instructional material explicitly asked them to consistently review and actively participate in mindfulness via MEMI exercises (online Supplementary Appendix B).
Self-monitoring app (SM)
The SM video commenced with an introduction to self-awareness as the heightened recognition of one's feelings and thinking states. Subsequently, SM suggested that practicing self-observation and recording of thoughts emotional discomfort could potentially aid in nurturing more beneficial thought-feeling repertoires. It conveyed that self-observation by itself could reduce anxiety symptoms. The core foundation was derived and adapted from the rationale employed in a recent app-based intervention (LaFreniere & Newman, Reference LaFreniere and Newman2016, Reference LaFreniere and Newman2020). Therefore, it purposefully avoided any reference to the concept of mindfulness. SM avoided any directives to enhance awareness and perception of current experiences. Instead, it concentrated on monitoring distressing thoughts and feelings. Participants were not guided to focus exclusively on their ongoing tasks or be in the moment. Although SM participants were encouraged to observe distress-associated thoughts and feelings, we omitted any instructions regarding accepting cognitive-affective states as they arose. SM also did not offer guidance on techniques for regulating and optimizing breath or prolonging self-observation practices after the 14-day intervention phase ended (online Supplementary Appendix C). This placebo control approach sought to increase credibility of the SM and to prevent any potential amplification of between-group effect sizes observed with a no-treatment/waitlist control (Lutz, Offidani, Taraboanta, Lakhan, & Campellone, Reference Lutz, Offidani, Taraboanta, Lakhan and Campellone2022).
Procedures
Those meeting eligibility criteria completed a series of initial self-assessments and participated in behavioral tests capturing EF. After completing the 14-day intervention (with five prompts/day), all participants completed the same measures at post-intervention and 1MFU (six weeks from the start). Measures were counter-balanced to prevent order effects. To maintain assessor-blinding to randomized arms, assessors either physically left the room (before COVID-19) or instructed participants to mute their Zoom before accessing the designated video link (during and post-COVID-19). Participants installed the PACO mobile application, which came preloaded with MEMI or SM (https://github.com/google/paco). The assessor answered questions concerning protocols, such as upcoming appointments or technical problems regarding installing the PACO app on their mobile devices. Nonetheless, the assessor did not attend when participants were informed about their assigned treatment and its constituents. Participants were provided a MEMI or SM intervention rationale document automatically delivered via Qualtrics to uphold assessor-blinding. They received compensation in credit hours, monetary remuneration, or their mixture (online Supplementary Appendix A). On the seventh intervention day, the team conducted an assessment to verify whether participants had adhered to the instruction of completing a minimum of 56 out of 70 prompts.
Data analyses
Random forest imputation utilizing the missRanger R package was employed to address missing data, which accounted for 11% of the dataset (Mayer, Reference Mayer2023). To evaluate MEMI against SM concerning their effects on specific EF mediators, we applied an intent-to-treat approach similar to the primary efficacy analysis (Zainal & Newman, Reference Zainal and Newman2023b). This entailed a two-tiered multilevel model, which examined changes from pre-treatment to 1MFU in GAD severity or trait RNT, with group as the between-person factor. We utilized a counterfactual causal approach in multilevel mediation (online Supplementary Appendix A and VanderWeele, Reference VanderWeele2016).
The analysis examined three multiplicative routes: effect of treatment group assignment on the pre–post mediator (a path), pre-post mediator's effect on pre-1MFU outcome (b path), and treatment group effect on pre-1MFU outcome (c path or direct effect). Simultaneously adjusting for treatment group effect, the mediation effect is called the indirect effect (VanderWeele, Reference VanderWeele2016). Temporal precedence was set such that treatment assignment preceded the pre–post EF mediator, and the pre–post EF mediator preceded the pre-1MFU endpoint (Winer et al., Reference Winer, Cervone, Bryant, McKinney, Liu and Nadorff2016). Given the theoretical significance of each plausible mediator and their interconnectedness, we refrained from adjusting for additional mediators (Vansteelandt & Daniel, Reference Vansteelandt and Daniel2017). We displayed the non-standardized regression coefficients (β) with 95% confidence intervals (CIs) and used bootstrapping with 1000 resampling repetitions (Cheung & Lau, Reference Cheung and Lau2008). We performed sensitivity assessments using non-linear generalized additive multilevel models to examine the extent to which the observed results remained consistent (Imai, Keele, & Tingley, Reference Imai, Keele and Tingley2010). Effect size was calculated as proportion of the indirect effect in relation to the total effect (Wen & Fan, Reference Wen and Fan2015). For mediation analyses, we employed three R packages: intmed (Chan, Reference Chan2020), mgcv (Wood, Reference Wood2014), and mediation (Tingley, Yamamoto, Hirose, Keele, & Imai, Reference Tingley, Yamamoto, Hirose, Keele and Imai2014) as well as adapted published tutorials for our analyses (http://tinyurl.com/missRanger; http://tinyurl.com/codesintmed; http://tinyurl.com/codesmediation).
Results
Path a: random assignment predicting each pre–post EF mediator
MEMI v. SM had significantly stronger effects on enhancing pre-post inhibition (β = −2.075, p = .002), WM (β = 0.512, p = .045), and set-shifting (β = −2.916, p = .010), but not VF (β = −0.991, p = .547) and AC (β = 0.866, p = .337; Figure 2). MEMI yielded significant within-group pre-post improvements in inhibition (β = −3.958, p < .001), VF (β = 4.331, p = .004), WM (β = 0.551, p = .017), and set-shifting (β = −8.566, p < .001), but not AC (β = 1.201, p = .151). By comparison, SM produced significant within-group pre-post enhancements in inhibition (β = −2.276, p = .002) and set-shifting (β = −8.566, p < .001) but not VF (β = 4.179, p = .065), WM (β = 0.571, p = .103), or AC (β = 1.201, p = .151; Table 2).
Figure 2. Significant comparative efficacy of mindfulness EMI against SM on pre-post EF domains.
Note: EF, executive functioning; EMI, ecological momentary intervention; SM, self-monitoring placebo.
Table 2. Simple slope analyses of counterfactual mediation analysis of EF domains mediating the effect of MEMI against SM on pre-1MFU GAD severity
Note. * p < 0.05; ** p < 0.01; *** p < 0.001.
EF, executive functioning; MEMI, ecological momentary intervention; SM, self-monitoring control; 1MFU, one-month follow-up; β, unstandardized regression estimate; GAD, generalized anxiety disorder; LCI, lower limit of the 95% confidence interval (CI); UCI, upper limit of the 95% CI. All b path (mediator-outcome associations) models adjusted for the specific pre-post mediator.
Path b: pre–post EF predicting pre-1MFU change in GAD severity
Treatment group assignment did not significantly moderate the association of pre-post change in inhibition (β = 0.565, p = .057), VF (β = 0.011, p = .924), WM (β = −0.126, p = .867), set-shifting (β < 0.001, p = .999), and AC (β = −0.063, p = .760) predicting pre-1MFU change in GAD severity. Within the MEMI arm, pre-post improvements in inhibition (β = −5.072, p = .005), VF (β = −6.996, p < .001), WM (β = −6.798, p < .001), set-shifting (β = −6.193, p = .001), and AC (β = −6.148, p < .001) significantly predicted pre-1MFU declines in GAD severity (Table 2). However, within the SM arm, pre-post change in inhibition (β = −1.076, p = .627), VF (β = −1.083, p = .617), WM (β = −1.064, p = .622), set-shifting (β = 0.161, p = .944), and AC (β = −0.396, p = .849) did not significantly predict pre-1MFU change in GAD severity.
Path b: pre–post EF predicting pre-1MFU change in trait RNT
Treatment group assignment significantly moderated the association of pre-post inhibition (β = 0.036, p = .003) predicting pre-1MFU change in trait RNT. However, treatment group did not moderate the association of pre-post VF (β < 0.001, p = .832), WM (β = −0.042, p = .172), set-shifting (β = 0.007, p = .316), and AC (β = 0.003, p = .708) predicting pre-1MFU change in trait RNT. Within the MEMI arm, pre-post improvements in inhibition (β = −0.236, p = .002), VF (β = −0.309, p < .001), WM (β = −0.300, p < .001), set-shifting (β = −0.278, p < .001), and AC (β = −0.283, p < .001) significantly predicted pre-1MFU declines in trait RNT (Table 3). However, within the SM arm, pre-post change in inhibition (β = −0.100, p = .261), VF (β = −0.055, p = .521), WM (β = −0.076, p = .378), set-shifting (β = −0.059, p = .524), and AC (β = −0.028, p = .715) did not significantly predict pre-1MFU change in trait RNT.
Table 3. Simple slope analyses of counterfactual mediation analysis of EF domains mediating the effect of MEMI against SM on pre-1MFU trait repetitive thinking
Note. * p < 0.05; ** p < 0.01; *** p < 0.001.
EF, executive functioning; MEMI, ecological momentary intervention; SM, self-monitoring control; 1MFU, one-month follow-up; β, unstandardized regression estimate; LCI, lower limit of the 95% confidence interval (CI); UCI, upper limit of the 95% CI.
Indirect effect: treatment effect on pre-1MFU change in GAD severity via EF
The effect of MEMI compared to SM on reduced pre-1MFU GAD severity was significantly mediated by enhanced pre-post inhibition performance (β = −0.605, p = .030; proportion mediated = 7.1%). However, pre-post AC (β = −0.409, p = .340), set-shifting (β = −0.302, p = .236), VF (β = −0.046, p = .746), and WM (β = 0.045, p = .808) were non-significant mediators of the effect of MEMI against SM on decreased pre-1MFU GAD severity. Effect sizes ranged from 0.2% to 4.7% for non-significant mediation paths.
Indirect effect: treatment effect on pre-1MFU change in trait RNT via EF
The effect of MEMI compared to SM on decreased pre-1MFU trait RNT was significantly mediated by enhanced pre-post inhibition performance (β = −0.024, p = .040; proportion mediated = 7.4%). However, pre-post AC (β = 0.005, p = .900), set-shifting (β = −0.012, p = .222), VF (β = 0.003, p = .570), and WM (β = −0.008, p = .308) were non-significant mediators of the effect of MEMI against SM on reduced pre-1MFU trait RNT. Effect sizes ranged from 0% to 3.6% for non-significant mediation paths.
Discussion
Partially supporting Hypothesis 1, MEMI generated greater pre-post enhancements in inhibition, WM, and set-shifting but not VF and AC. Although there was no significant difference between MEMI and SM on increased VF, simple slope analysis revealed pre-post VF improvements in MEMI but not SM. Hypothesis 2 also received partial support, as pre-post improvement in inhibition uniquely mediated the effect of MEMI compared to SM on pre-follow-up reductions in GAD symptom severity and trait RNT. The combination of therapy elements in MEMI, such as acceptance, diaphragmatic breathing retraining, engagement with present activity, non-reactivity, and open monitoring, likely contributed to any observed differential efficacy over SM rather than any single component alone.
Other potential theoretical propositions are considered to inform optimization efforts of MBIs for GAD. Pre-post enhancements in inhibition accounted for the effect of MEMI against SM on pre-follow-up GAD severity and RNT. Further, pre-post enhancements in inhibition mediated pre-follow-up decline in RNT more strongly in MEMI than SM. Since people with GAD experience their worries as uncontrollable (Hallion & Ruscio, Reference Hallion and Ruscio2013) and have negative beliefs about worry (LaFreniere & Newman, Reference LaFreniere and Newman2019), it may be necessary to improve inhibition skills to put a brake on worry and other perseverative cognitions. Relatedly, people with GAD worry in autopilot ways that heighten and prolong distress to avoid negative emotional contrasts, i.e. sharp rises from positive or neutral to negative affect states (cf. contrast avoidance theory; Newman & Llera, Reference Newman and Llera2011; Newman, Llera, Erickson, Przeworski, & Castonguay, Reference Newman, Llera, Erickson, Przeworski and Castonguay2013). Thus, MEMI might have fostered tolerance of intense surges in distress, relinquishing the usual disinhibited, reflexive urges to avoid or resist negative emotional contrasts and, instead, allowing these experiences to fully register in one's awareness by enhancing inhibition. Future studies should investigate this idea by examining the link between inhibition and contrast avoidance in GAD.
Additional issues in GAD could explain the salience of inhibition as a potential change mechanism (cf. attentional control theory; Eysenck, Derakshan, Santos, & Calvo, Reference Eysenck, Derakshan, Santos and Calvo2007). People with GAD (v. controls) displayed poorer decision-making on inhibition-based reinforcement learning tasks (White et al., Reference White, Geraci, Lewis, Leshin, Teng, Averbeck and Blair2017). Relatedly, higher clinician-assessed GAD severity was correlated with slower and less precise performance in the Stroop task (Hallion, Tolin, Assaf, Goethe, & Diefenbach, Reference Hallion, Tolin, Assaf, Goethe and Diefenbach2017). Inhibitory dyscontrol has been observed to be a correlate (Majeed et al., Reference Majeed, Chua, Kothari, Kaur, Quek, Ng and Hartanto2023), predictor (Zainal & Newman, Reference Zainal and Newman2018), and longitudinal outcome of increase in pathological worry (Zainal & Newman, Reference Zainal and Newman2023a, Reference Zainal and Newman2021). These problems might translate to ample opportunities for MEMI to remedy inhibition deficits in GAD, thereby alleviating future worry and other RNT. Further, pathological worry preceded and increased future inhibition deficits within individuals across time (cf. scar theories; Zainal & Newman, Reference Zainal and Newman2021). Thus, MEMI likely improved inhibition by teaching and reinforcing the skill of resisting the habit of worrying, ruminating, or obsessing (Gallant, Reference Gallant2016). Enhancing inhibition by focusing on the here-and-now instead of the past/future with perseverative cognitions may have led to pre-follow-up decreases in worry and other RNT propensities.
Why did MEMI yield stronger pre–post improvements in set-shifting and WM relative to SM? Perhaps MEMI liberated cognitive resources that were otherwise consumed by suppressing worry-related, task-irrelevant thoughts, leading to an overall enhancement in cognitive efficiency to deploy better set-shifting and updating WM skills (Course-Choi, Saville, & Derakshan, Reference Course-Choi, Saville and Derakshan2017; Jankowski & Holas, Reference Jankowski and Holas2020). The potential for MEMI to induce better meta-cognitive skills, such as non-identification with feelings and thoughts (McEvoy, Graville, Hayes, Kane, & Foster, Reference McEvoy, Graville, Hayes, Kane and Foster2017), could also explain these results. Further, these outcomes might make sense given how patients with (v. without) GAD continually exhibited worse WM task performance under threat conditions (Vytal, Arkin, Overstreet, Lieberman, & Grillon, Reference Vytal, Arkin, Overstreet, Lieberman and Grillon2016). When exposed to emotion-inducing distractions, they displayed reduced activity and white matter volume in WM-linked dorsolateral prefrontal cortex areas (Moon & Jeong, Reference Moon and Jeong2017) and struggled with cognitively retaining materials germane to present objectives (Moon, Sundaram, Choi, & Jeong, Reference Moon, Sundaram, Choi and Jeong2016; Yoon, LeMoult, Hamedani, & McCabe, Reference Yoon, LeMoult, Hamedani and McCabe2018). Together, MEMI might reverse worry-triggered set-shifting and WM deficits by freeing cognitive processing assets (cf. resource allocation theory; Levens, Muhtadie, & Gotlib, Reference Levens, Muhtadie and Gotlib2009) and instructing focus on the here-and-now and task-switching flexibility.
Nevertheless, set-shifting and WM did not mediate treatment effects. MBIs were also found to be more efficacious for accuracy (v. latency), set-shifting, and WM scores (cf. meta-analysis; Zainal & Newman, Reference Zainal and Newman2024b). It is possible that instead of the TMT-B, using other set-shifting measures based on accuracy rather than latency might have increased the chances of detecting a mediation effect. Likewise, WM measures apart from the WAIS-IV LNS might have been more sensitive for mediation purposes, such as the automated operation span task (Dubert, Schumacher, Locker, Gutierrez, & Barnes, Reference Dubert, Schumacher, Locker, Gutierrez and Barnes2016; Unsworth, Heitz, Schrock, & Engle, Reference Unsworth, Heitz, Schrock and Engle2005). Alternatively, the lack of mediation effects with set-shifting and WM might be due to the present RCT being underpowered to detect small effect sizes for these domains (cf. method paper by Qin, Reference Qin2024).
Although no between-group effects emerged, within-group analyses revealed notable pre-post improvements in VF in MEMI but not SM. These findings might be explained by evidence that brief MBIs could enhance verbal learning and memory via refinements in the ability to register information (Lueke & Lueke, Reference Lueke and Lueke2019). Relatedly, since VF is associated with aptitude to efficiently recall words linked to emotions (Hegefeld, Satpute, Ochsner, Davidow, & Nook, Reference Hegefeld, Satpute, Ochsner, Davidow and Nook2023), MEMI might have strengthened VF of emotion and non-emotion words more than SM (Edwards, Shivaji, & Wupperman, Reference Edwards, Shivaji and Wupperman2018). This interpretation could be understood in the context of struggles to recognize and describe emotions in GAD (Paniccia et al., Reference Paniccia, Gaudio, Puddu, Di trani, Dakanalis, Gentile and Di ciommo2020). To confirm these interpretations, experimental work is required to test these conjectures.
Unexpectedly, neither between- nor within-group effects on self-reported AC occurred. Although a prior cross-sectional study showed that higher AC mediated the inverse anxiety-mindfulness correlation (MacDonald & Olsen, Reference MacDonald and Olsen2020), such findings did not extend to our longitudinal RCT of MEMI for GAD. These outcomes might be accounted for by weak correlations between self-reported and performance-based AC measures (Snyder, Friedman, & Hankin, Reference Snyder, Friedman and Hankin2021). In addition, based on recent theoretical formulations (Prakash, Reference Prakash2021) and evidence of the efficacy of 8-week MBSR on AC (Chin et al., Reference Chin, Lindsay, Greco, Brown, Smyth, Wright and Creswell2021; Lee et al., Reference Lee, Wong, Chan, Zhang, Sun, Chan and Wong2021), lengthier and more rigorous forms of MEMI might be needed to improve AC for GAD. Alternatively, other measures, such as task-unrelated mind-wandering probes (Mrazek, Franklin, Phillips, Baird, & Schooler, Reference Mrazek, Franklin, Phillips, Baird and Schooler2013), might better capture the effect of brief MEMI on AC. Relatedly, based on a meta-analysis data of robust inverse relations between AC/EF and RNT (Mennies, Stewart, & Olino, Reference Mennies, Stewart and Olino2021), another AC measure, such as the self- or parent-reported Behavior Rating Inventory of EF (Gioia, Isquith, Guy, & Kenworthy, Reference Gioia, Isquith, Guy and Kenworthy2000; Guy, Isquith, & Gioia, Reference Guy, Isquith and Gioia2004), might have mediated treatment effects.
Interpreting results in light of the broader literature on the relations between RNT and AC/EF constructs is also essential. We tested how specific EF domains mediated the efficacy of MEMI on reductions in GAD severity and RNT. Our findings regarding treatment predictors or mediators might differ if other domain-specific RNT outcomes were examined, such as anger and depressive rumination (du Pont, Rhee, Corley, Hewitt, & Friedman, Reference du Pont, Rhee, Corley, Hewitt and Friedman2019) or job-related rumination (Cropley & Collis, Reference Cropley and Collis2020), which exhibited modest yet meaningful negative correlations with a global EF. Heterogeneity also exists in the literature, such that global EF was often (Abramovitch et al., Reference Abramovitch, Short and Schweiger2021), but not always (du Pont et al., Reference du Pont, Rhee, Corley, Hewitt and Friedman2019), linked to the internalizing symptom constructs that subsume worry.
Several limitations deserve consideration. First, future studies should examine additional factors that might explain outcomes and maximize the potential to identify differential mediator effects in the context of GAD, such as self-reported WM (Adamis & Olatunji, Reference Adamis and Olatunji2024) and performance-based composites of various domain-specific EF tasks (e.g. anti-saccade and go-no-go tasks; Gustavson et al., Reference Gustavson, Elman, Panizzon, Franz, Zuber, Sanderson-Cimino and Kremen2020). The proportion mediated estimate for inhibition was 7%, which might be considered a meaningful yet small effect size (Preacher & Kelley, Reference Preacher and Kelley2011). This magnitude prompts the question of alternative EF-related pathways through which MEMI v. SM affects the outcome (VanderWeele, Reference VanderWeele2013). Despite its sensitivity in correlating with worry symptoms in other samples with GAD (Beaudreau et al., Reference Beaudreau, Hantke, Mashal, Gould, Henderson and O'Hara2017; Dorenkamp et al., Reference Dorenkamp, Irrgang and Vik2023), the neutral CWIT measure of inhibition might not sufficiently capture inhibition skills needed to curtail experiential avoidance of negative emotions or thoughts inherent in our GAD sample. Future studies should thus assess the mediation potential of AC/EF using ambulatory assessments (Hernandez et al., Reference Hernandez, Hoogendoorn, Gonzalez, Jin, Pyatak, Spruijt-Metz and Schneider2023) or AC/EF tasks that capture emotional states (Kalanthroff, Reference Kalanthroff2024). The lack of tasks capturing emotional states might explain some of our null or small effect size findings. For instance, if the stimuli in the EF/AC tasks had been affect-based (e.g. emotional Stroop task; Smolker et al., Reference Smolker, Wang, Luciana, Bjork, Gonzalez, Barch and Banich2022) or if clients with GAD had performed the tasks under induced anxiety or other emotional states (Azab, Reference Azab2022) findings might have notably varied. Results might also have differed had other EF indices, such as cognitive flexibility (Baussay et al., Reference Baussay, Di Lodovico, Poupon, Doublet, Ramoz, Duriez and Gorwood2024) and self-regulation (Short et al., Reference Short, Mazmanian, Oinonen and Mushquash2016), been in the equation for all mediational analyses. Second, the 14-day intervention duration might have been inadequate to identify or generate mediation effects of all examined EF mediators, given null treatment effects on AC and VF. Eight-week MBI RCTs suggest that maximizing the detection of between-group differences in AC and VF could require more time and practice to improve present-mindedness and express emotions and thoughts more clearly (Chin et al., Reference Chin, Lindsay, Greco, Brown, Smyth, Wright and Creswell2021). Third, future research should explore whether ongoing mindfulness practices could have yielded any distinct mediation effects during follow-up without repeated guidance through MEMI. Fourth, the inferences drawn from our study may not apply to a broader population beyond White females, emphasizing the need for future digital mental health EMI RCTs to include more culturally diverse participants.
Despite these limitations, several strengths were noteworthy. Our study adhered to rigorous CONSORT guidelines (Calvert, Brundage, Jacobsen, Schünemann, & Efficace, Reference Calvert, Brundage, Jacobsen, Schünemann and Efficace2013; Montgomery et al., Reference Montgomery, Grant, Mayo-Wilson, Macdonald, Michie, Hopewell and Group2018), thus leveraging the methodological strengths of RCTs to eliminate bias and confounding sources. A placebo control and assessor-blinding to random assignment were also included, further reducing the potential for confounding and selection biases. Also, because we controlled for focus on and monitoring of thoughts and emotions, which is a powerful treatment in and of itself, we can more confidently attribute differential treatment effects to unique components of mindfulness. The engagement rates were also high, with a dropout rate (11%) far lower than the 24–50% dropout rates observed in app RCTs (Linardon, Reference Linardon2023; Linardon & Fuller-Tyszkiewicz, Reference Linardon and Fuller-Tyszkiewicz2020). Further, the current study also enrolled a clinician-diagnosed sample with GAD, ensured adequate power, and incorporated a follow-up assessment.
In conclusion, MEMI was more efficacious than SM in enhancing pre-post inhibition, WM, and set-shifting, though it did not show superiority in AC and VF. Despite the lack of between-group differential efficacy, within-group analyses showed MEMI improved VF but not SM. Only inhibition mediated the effect of treatment on reductions in GAD severity and RNT. If replicated, the present study has possible practical applications in clinical contexts. Brief MEMIs for GAD might be optimized by prioritizing the targeting of inhibition rather than other EF domains. Several approaches could be tried to attain this goal. First, adding dialectical behavioral therapy components to MEMI by inviting clients to practice acceptance of life stressors and commit to inhibiting the urge to worry or ruminate might optimize brief MEMIs for GAD (Afshari et al., Reference Afshari, Jafarian Dehkordi, Asgharnejad Farid, Aramfar, Balagabri, Mohebi and Amiri2022; Vijayapriya & Tamarana, Reference Vijayapriya and Tamarana2023). Second and related, GAD should change the tendency to worry in order to create and maintain negative moods to avoid sharp rises in negative emotions (cf. contrast avoidance theory; Newman & Llera, Reference Newman and Llera2011; Newman et al., Reference Newman, Llera, Erickson, Przeworski and Castonguay2013). Plausibly, instructing clients with GAD to let go of worrying and allow experiences of emotional fluctuations, including negative emotional contrasts, via a higher-intensity version of MEMI might have a positive effect of improving inhibition. Third, as sustained worry induces wear-and-tear of physiological systems in ways that adversely affect EF over time (Zainal & Newman, Reference Zainal and Newman2022a, Reference Zainal and Newman2022b), MEMI should be merged with EF-enhancing physical exercise (cf. a meta-analysis; Moreau & Chou, Reference Moreau and Chou2019) among people with GAD. Fourth, future research should address the perennial inquiry of which subgroup with GAD would benefit most from inhibition-boosting exercises in conjunction with MEMI.
The current study received funding from the National Institute of Mental Health (NIMH) (R01 MH115128), the Pennsylvania State University RGSO Dissertation award, Penn State Susan Welch/Nagle Family Graduate Fellowship, the National University of Singapore (NUS) Development Grant, and the Association for Behavioral and Cognitive Therapies (ABCT) Leonard Krasner Student Dissertation Award.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291724001958.
Funding statement
The current study received funding from the National Institute of Mental Health (NIMH) (R01 MH115128), the Pennsylvania State University RGSO Dissertation award, Penn State Susan Welch/Nagle Family Graduate Fellowship, the National University of Singapore (NUS) Development Grant, and the Association for Behavioral and Cognitive Therapies (ABCT) Leonard Krasner Student Dissertation Award.
Competing interests
None.
Open access
