Mindfulness-Based Programs in the Workplace: a Meta-Analysis of Randomized Controlled Trials

**Interventions:** Any mindfulness-based program (MBP) offered at or initiated by an employer, requiring at least 2 hours of training with mindfulness constituting at least 50% of the content. This included full MBSR (8-week, ~26 hours), shortened MBSR versions, mindfulness combined with yoga, ACT/MBCT hybrids, meditation apps, and online programs.

**Comparators:** Mostly passive wait-list or no-treatment controls (79% of studies); only 21% used active controls such as psychoeducation or stress management workshops.

**Outcome domains tested:**

- Mindfulness skill (self-report scales)

- Perceived stress

- Subsyndromal symptoms (anxiety, depression, fatigue, sleep, rumination)

- Burnout (emotional exhaustion, depersonalisation)

- Somatisation / physical complaints

- Well-being and life satisfaction

- Compassion (self and other)

- Job satisfaction

- Work engagement

- Productivity (self-report and some external metrics)

- Resilience

---

**Total participants:** 5,161 across 56 articles (53 distinct studies, 57 independent intervention samples)

**Intervention group:** n = 2,689; **Control group:** n = 2,472

**Age range across studies:** 19.5–50.5 years old (mean = 40.1 years, SD = 7.0)

**Gender:** Samples averaged 73% women

**Education:** Average roughly equivalent to a Bachelor's degree (~15.4 years of education)

**Professions:** Healthcare workers (36%), teachers (17%), industry (8%), finance (6%), call centre (4%), public administration (4%), mixed/all-employee programmes (21%)

**Geography:** ~42% USA, 8% UK, 8% Spain, plus Canada, Australia, Netherlands, China, Germany, India, and others

**Setting:** Mostly employed adults working >30 hours/week; mostly healthy (not clinical populations, though four studies recruited employees with poor mental health specifically)

---

Outcomes were grouped into clusters from whatever scales the original studies used. Common instruments across primary studies included:

**Mindfulness:** Five Facet Mindfulness Questionnaire (FFMQ), Mindful Attention Awareness Scale (MAAS)

**Perceived stress:** Perceived Stress Scale (PSS)

**Subsyndromal symptoms:** General Health Questionnaire (GHQ), Depression Anxiety Stress Scales (DASS), Patient Health Questionnaire (PHQ), Pittsburgh Sleep Quality Index (PSQI)

**Burnout:** Maslach Burnout Inventory (MBI) — emotional exhaustion, depersonalisation, personal accomplishment subscales

**Well-being:** Satisfaction with Life Scale (SWLS), WHO-5 Well-Being Index, Positive and Negative Affect Schedule (PANAS)

**Compassion:** Self-Compassion Scale (SCS)

**Job satisfaction:** Various single-item and multi-item job satisfaction scales

**Work engagement:** Utrecht Work Engagement Scale (UWES)

**Productivity:** Heterogeneous — ranging from self-report (Work Limitations Questionnaire) to observer-rated performance, absenteeism records, and call-centre metrics

When a study used multiple instruments for the same construct, the authors computed a weighted pooled mean and standard deviation to avoid double-counting.

---

**Study design:** This is a pre-registered (PROSPERO: CRD42015019282) systematic review and meta-analysis conducted according to PRISMA guidelines and Cochrane recommendations. It synthesises data from 56 randomised controlled trials (RCTs) published up to November 2018.

**Literature search:** Four electronic databases (PsychInfo, PubMed, Web of Science, Academic Search Premier) plus manual reference screening and author contact for unpublished data.

**Effect size calculation:** Hedges' g (a bias-corrected standardised mean difference) was computed for each outcome, comparing the mindfulness group to the control group at post-intervention and at follow-up. Positive g = better outcome in the mindfulness group; negative g = reduction in a bad thing (stress, burnout). Standard conventions: g = 0.2 small, g = 0.5 medium, g = 0.8 large.

**Statistical model:** Random-effects models (restricted maximum likelihood estimator), which assume true effect sizes vary across studies and do not force a single "true" effect — appropriate given the large variability in programmes and populations.

**Heterogeneity:** Assessed using I² (the proportion of variance due to between-study differences rather than chance). I² was medium-to-high for most outcomes (e.g., I² = 89.8% for stress, 91.5% for subsyndromal symptoms), meaning results varied substantially across studies.

**Sensitivity analyses:** Cook's distance outlier detection was applied to every outcome. Analyses were repeated after removing influential studies to check robustness.

**Moderator analyses:** Conducted for outcomes with at least 10 primary effect sizes and at least moderate heterogeneity (I² > 50%). Tested programme type, hours of attendance, weeks of duration, delivery mode, location, gender, age, profession, education, work experience, control group type, and ITT handling.

**Follow-up data:** Only 18 studies (34%) provided follow-up data within 12 weeks post-programme; only 7 provided longer follow-ups (16 weeks to 3 years), too few for meta-analytical pooling beyond 12 weeks.

**What this design CAN prove:** That, on average across diverse programmes and workplaces, employees randomised to a mindfulness programme score better than controls on well-being, stress, and job satisfaction outcomes — immediately after the programme and for up to 12 weeks after.

**What this design CANNOT prove:**

Long-term durability beyond 3 months (insufficient data)

Causality at the individual level — this is a group average

That mindfulness is uniquely effective compared to other active interventions, because 79% of control groups were passive wait-lists (a weak comparison that inflates effect sizes)

Generalisability to male-dominated, blue-collar, or lower-education workplaces, given the skew towards female, educated, healthcare/education populations

That productivity and work engagement genuinely improve — these findings are fragile (see Key Findings)

**Major weaknesses:**

Only 40% of studies used intention-to-treat analysis, meaning dropouts are handled inconsistently and effects may be overstated

Possible publication bias detected for stress and mindfulness outcomes specifically

Most studies did not adequately report randomisation sequence generation or allocation concealment

Heterogeneity is very high for most outcomes (I² often > 80%), meaning the pooled number describes a wide range of real-world results, not a single reliable effect

---

**Post-intervention between-group effects (mindfulness vs. control):**

**Mindfulness skill:** g = 0.44, 95% CI [0.32, 0.56], p < .001, k = 32 studies; I² = 52%

**Well-being / life satisfaction:** g = 0.68, 95% CI [0.24, 1.12], p = .002, k = 22; I² = 95% (very high heterogeneity — treat this number with caution)

**Compassion:** g = 0.61, 95% CI [0.37, 0.85], p < .001, k = 8; I² = 31%

**Perceived stress:** g = −0.66, 95% CI [−0.88, −0.44], p < .001, k = 43; I² = 90%

**Subsyndromal symptoms (anxiety, depression, fatigue, sleep):** g = −0.56, 95% CI [−0.79, −0.33], p < .001, k = 40; I² =