Work Engagement and Well-being Study (SWELL): a randomised controlled feasibility trial evaluating the effects of mindfulness versus light physical exercise at work

Employers increasingly adopt mindfulness-based programmes (MBPs) to support staff health and well-being, in line with public health guidance. MBPs aim to improve attention, self-regulation, and compassion by training present-moment awareness and decentering. Evidence indicates MBPs benefit mental health compared to passive controls, but effects on work performance (WP) are unclear, with heterogeneous measures used (eg, resilience, engagement, absenteeism/presenteeism). Potential mechanisms for WP improvement include: (1) mental health enhancement, which may improve engagement and reduce absenteeism/presenteeism; and (2) cognitive control improvements, facilitating self-regulation and prioritisation of work goals, particularly in affective contexts. However, prior cognitive control findings largely derive from trials with passive controls, and transfer to WP is uncertain. Given these gaps, a CONSORT-adherent, pre-registered feasibility RCT with an active control was conducted to clarify methodological uncertainties and estimate preliminary effect sizes of an online, self-guided MBP on WP, with cognitive control as a hypothesised mediator.

Prior systematic reviews show MBPs in nonclinical settings improve mental health relative to passive controls (eg, reduced anxiety, depression, stress). Workplace trials rarely assess WP and use diverse operationalisations (resilience, engagement, absenteeism/presenteeism), making synthesis difficult. Beliefs in one’s task efficacy predict task performance, potentially via cognitive control, suggesting MBPs might affect task performance. Meta-analyses suggest mindfulness training may enhance cognitive control versus passive controls (Hedges g ≈ −0.03 to 0.42), though generalisation to emotionally valenced stimuli and to WP remains unclear. Evidence comparing MBPs with active controls is limited; prior reviews call for CONSORT-standard, pre-registered trials using active comparators to control for non-specific and mental health effects. Existing active-control trials report trivial differences in WP outcomes (eg, health-related absences, engagement) when comparing MBPs to other active interventions.

Design: Pragmatic randomised controlled feasibility trial adhering to CONSORT guidance for pilot/feasibility trials; prospectively registered at ClinicalTrials.gov (NCT04631302), with full protocol published. Participants and recruitment: Eight employers with predominantly desk-based staff disseminated study information internally. Eligibility: Current employee of a participating UK-based employer; advised not to participate if on long-term leave, experiencing severe mental illness, recent bereavement, or prior substantial meditation experience (>10 hours in past 10 years). Written informed consent obtained. No incentives for intervention completion; vouchers provided for assessments (£10 postintervention, £15 at 12-week follow-up). Randomisation and masking: Participant-level 1:1 randomisation to MBP or light physical exercise (LE), stratified by employer, via automated REDCap allocation using random blocks (randomizeR in R), concealed from researchers; allocation emailed to participants. Participants and study manager were not blinded to assignment; materials introduced both arms equivalently; primary analysis conducted by a statistician blinded to allocation. Sample size: Target n=240 to assess feasibility parameters and provide preliminary effect size estimates; anticipated attrition typical of online trials. Interventions: MBP arm completed Be Mindful (Wellmind Media), a 4-week, pre-recorded, fully automated online mindfulness course accessed via bemindfulonline.com, matching prior evidence of mental health benefits. LE arm completed a 4-week, pre-recorded, fully automated online light physical exercise programme aimed at mobility and strain prevention, matched for time and engagement, and previously shown to benefit mental health; chosen to control for non-specific and mental health effects without evidence of improving cognitive control in healthy adults. Outcomes and timing: Assessments at baseline, postintervention (primary endpoint), and 12-week follow-up. Daily monitoring invited at 15:00 each workday using the WRFQ-5 to index day-level WP fluctuations. Primary outcome: Work Role Functioning Questionnaire (WRFQ) 25-item version 2 total score at postintervention; prior validations of non-English versions show good reliability; internal consistency in this study α=0.93. Secondary outcomes: Work-related outcomes including health impacts on ability to work and Work and Social Adjustment Scale (WSAS) among those reporting health problems; daily WRFQ-5. Cognitive control: Affective stop-signal task (SSRT; exclusion criteria for implausible RTs and short delays per consensus guidelines) and affective probabilistic reversal learning (proportion correct). Mental health and process measures: Perceived Stress Scale (PSS), PHQ-9 (depression), GAD-7 (anxiety), Experiences Questionnaire (decentring), Mindful Attention Awareness Scale (MAAS). Planned Short Warwick-Edinburgh Mental Well-Being Scale data were not obtained due to technical error. Data collection platforms: REDCap and jsPsych hosted on JATOS. Analysis: Intention-to-treat analyses. Multiple linear regression for primary and most secondary outcomes (postintervention and follow-up) adjusted for baseline and employer using miceadds; independent blinded statistician analysed postintervention questionnaire data (except MAAS and decentring). Linear mixed-effects models used for cognitive tasks due to non-normality and sphericity violations. Mediation analysis using unimputed complete cases tested whether postintervention SSRT (negative valence) mediated effects of group allocation on WRFQ at follow-up.

Feasibility and acceptability: Eight employers and 241 employees participated; median employer size 2,130 (range 180–7,500). Participation among staff was modest (median 0.91%, range 0.27–2.85%), with higher uptake in local authorities (mean 2% vs 0.58% in other industries). 87% started their assigned course (MBP 87.7%; LE 86.55%). Retention: 64% completed postintervention outcomes (MBP 60.66%; LE 68.07%); 30% completed 12-week follow-up (MBP 32.79%; LE 27.87%). Median engagement length was 3 of 4 weeks (IQR=2). Six participants abandoned the programme but still provided outcomes; none withdrew from the study. Preferences: At postintervention, LE participants expressed greater desire to have been assigned MBP (W=3523, p=0.02); MBP participants showed no strong preference. Contamination: Minimal; MBP participants reported slightly more cross-arm discussions (mean 5.07/100 vs 2.81/100; p=0.93). Mindfulness practice ≥ up to 3 hours/week was more common in MBP (54.91%) than LE (16.8%) at postintervention (χ²(2,246)=12.8, p=0.002). Primary outcome (work performance): Adjusted intention-to-treat showed negligible between-arm difference at postintervention (d=0.06; t(237)=0.49, p=0.63; 95% CI −0.19 to 0.32) and at 12-week follow-up (d=0.02; p=0.91; 95% CI −0.30 to 0.26). Within-arm changes in WRFQ were trivial: baseline to postintervention (d=0.10, p=0.28) and baseline to follow-up (d=0.14, p=0.12). Daily monitoring: Participants completed fewer than half of daily surveys on average (MBP mean 12.2 of 28 days; LE mean 9.95). Work functioning improved across days in both arms (day effect beta=0.20, SE=0.05, t=4.03, p=0.0004), with negligible between-arm difference (beta=0.15, SE=0.94, t=0.16, p=0.873). Mental health and process outcomes: Between-arm differences at postintervention and follow-up were trivial and non-significant for stress (PSS), anxiety (GAD-7), depression (PHQ-9), and mindfulness (MAAS) (ds<0.10). Decentring showed small effects in favour of MBP (postintervention d=0.24, p=0.07; follow-up d=0.22, p=0.09). Within-arm improvements across both arms were moderate and significant: for example, PSS baseline to postintervention d=−0.58 (p<0.001) across arms; sustained improvements at 12 weeks (see table of within-arm ds for PSS, GAD-7, PHQ-9, decentring, MAAS). Cognitive control: Trivial between-arm effects in affective stop-signal task (interaction beta=−0.82, SE=2.04, t=−0.40, p=0.69, d=−0.05 at postintervention; beta=−1.16, SE=2.17, t=−0.53, p=0.59, d=−0.01 at follow-up) and affective reversal learning (accuracy beta=0.001, SE=0.003, t=0.34, p=0.74, d=0.04 postintervention; beta=0.0002, t=0.09, p=0.93, d=0.07 at follow-up). Mediation: No evidence that postintervention SSRT (negative valence) mediated the effect of allocation on WRFQ at follow-up (indirect effect −0.71, p=0.54; direct effect 0.79, p=0.84; total effect 0.08, p=0.98).

This feasibility trial addressed whether an online self-guided MBP improves work performance compared to an active light physical exercise programme and explored cognitive control as a mechanism. The interventions were acceptable and a trial in workplace settings is feasible; however, the MBP did not confer superior benefits over light exercise for work performance at postintervention or 12-week follow-up, and within-arm changes were minimal. Both interventions improved mental health outcomes with similar magnitudes, indicating that MBPs may not outperform other active workplace well-being strategies when delivered online and self-guided. Trivial differences in affective cognitive control do not support a mechanistic pathway via cognitive control under these conditions; observed small advantages in decentring for MBP suggest this process could be a more promising mechanism to investigate. The findings imply that a full-scale efficacy trial comparing MBP to light exercise is not warranted, and that organisations should consider comparative effectiveness, user preferences, and engagement strategies when selecting workplace well-being interventions. Measurement challenges were evident: the WRFQ facilitated cross-role comparisons but showed potential ceiling effects, and daily monitoring had low adherence, complicating detection of effects. Enhancing intervention engagement (eg, co-design with target communities) and refining outcome measures for work performance may be critical to capturing potential benefits in future research.

An online, self-guided MBP and a matched light physical exercise programme were feasible and acceptable in workplace settings, but the MBP provided negligible additional benefits for work performance compared to light exercise at postintervention and 12-week follow-up. Both interventions yielded significant improvements in mental health outcomes, with small non-significant advantages for decentring in the MBP arm. Given the trivial between-arm differences on work performance and cognitive control, progression to a later-phase superiority trial of MBP versus light exercise is not supported. Future research should prioritise improving participant engagement through participatory design, investigate decentring as a potential mechanism, and develop or select robust, sensitive, and cross-role-comparable measures of work performance to better evaluate workplace interventions.

Attrition and missing data were substantial (64% postintervention, 30% follow-up completion), potentially limiting external validity and attenuating randomisation’s protection against confounding. The primary WP measure (WRFQ v2) is not validated in English and may exhibit ceiling effects in generally well populations. Daily monitoring adherence was low, reducing sensitivity to detect day-level effects. Cognitive task analyses required robust mixed-effects approaches due to assumption violations, and mediation analyses were based on a limited complete-case sample (n=43), reducing power. The interventions were self-guided online with typical engagement (median 3 weeks), which may not reflect optimal efficacy conditions. Participants and study manager were not blinded to allocation (although primary analysis was blinded), and contamination, while low, cannot be fully excluded. As a feasibility study, it was not powered to detect small between-arm differences.