記住我
Even before the COVID-19 pandemic, findings from meta-analyses demonstrated a high prevalence of stress in health care workers worldwide [-]. Stress is a vulnerability factor for work-related burnout [], anxiety, and depression [], all of which are disproportionately prevalent among health care workers [-], and stress also increases the risk of several long-term physical health conditions [-]. In the National Health Service (NHS) in England, which employs >1.3 million health care staff [], 46.8% of staff reported feeling unwell because of work-related stress [], a figure that has steadily risen since 2016. Almost one-quarter of the days lost to staff sickness in the NHS are because of stress, anxiety, depression, or other mental health problems [], and similar concerns have been noted in health care systems worldwide []. Moreover, stress among health care workers can compromise patient outcomes and safety []. The COVID-19 pandemic has further exacerbated stress and distress for health care workers [,]; therefore, there is an urgent need to find effective, accessible, and affordable ways of reducing health care workers’ stress.
Mindfulness involves intentionally bringing curiosity and nonjudgmental awareness to present-moment experiences such as thoughts, feelings, and physical sensations as they arise [,]. Mindfulness-based interventions (MBIs) typically involve teaching mindfulness in in-person group settings through 8-week courses such as mindfulness-based cognitive therapy (MBCT) [] and mindfulness-based stress reduction (MBSR) [], with mindfulness practice and teacher-led discussion of practice being core intervention ingredients. There is substantial evidence from meta-analyses of randomized controlled trials (RCTs) that MBCT reduces the risk of relapse in people with a history of recurrent depression [] and that MBIs improve symptoms of a range of mental health problems []. The degree of engagement in mindfulness practice during MBIs is associated with treatment outcomes [], and MBI mechanisms of action include mindfulness, rumination, worry, and self-compassion [].
The benefits of MBIs extend beyond clinical populations, with RCTs demonstrating beneficial effects on stress in nonclinical populations [], including working adults [] and, specifically, health care workers [-]. However, there are several barriers to health care workers attending in-person MBIs, including the lack of availability []; high workplace demands [,] that make it difficult for health care workers to find the time to attend; and stigma-related concerns regarding negative social judgments and disclosure and confidentiality, which are more common among health care workers than among those working in other settings [].
Fortunately, mindfulness-based self-help (MBSH) has the potential to increase opportunities for engagement with MBIs through a plethora of MBSH books, web-based courses, and available smartphone apps. In addition, meta-analyses of RCTs of MBSH have indicated promising effects on stress and mental health outcomes across a range of populations [,]. Digital MBSH using smartphone apps has the potential to be particularly accessible as it does not rely on the user having a computer or book on hand to engage with the intervention when needed. Headspace [] is a smartphone app with >70 million users to date worldwide []. There is emerging empirical literature exploring the effectiveness of MBSH apps, including Headspace []. Preliminary findings show potential benefits in nonclinical samples, including health care workers; however, the study sample sizes were too small to draw definitive conclusions regarding this working population. Given the early stage of research in this area and studies with small sample sizes, the potential of unguided digital MBSH as a health care–wide solution to reduce health care worker stress is yet to be explored in an adequately powered trial. Although MBSH can effectively reduce stress in a range of nonclinical populations, it is possible that the particularly high demands of working in health care [,] will mean that when offered at scale, health care staff may struggle to engage with the intervention, leading to disappointing outcomes. The learnings available from a definitive trial of unguided digital MBSH are particularly important in the current context of rising health care worker stress during the COVID-19 pandemic.
ObjectivesThis study sought to overcome some of the methodological limitations of previous related studies and extend our understanding of the potential effects of unguided MBSH among health care workers. The aim of this large multisite RCT was to explore the effectiveness of unguided digital MBSH in comparison with an active control condition (it should be noted that comparisons with active controls are lacking in RCTs of MBIs []) for health care workers in targeting stress (primary outcome), mental health outcomes (depression, anxiety, and well-being), work-related outcomes (work-related burnout, sickness absence, and compassion for others), and proposed mechanisms of action (intervention engagement, rumination, worry, mindfulness, and self-compassion). To explore its potential as a health care–wide intervention to reduce health care worker stress, the trial recruited across the full range of NHS organization types (general practitioner or primary care, hospital trusts, community trusts, mental health and/or learning disability trusts, and ambulance trusts), across geographically and sociodemographically diverse regions of England, and across a range of NHS job roles (medical, nursing, allied health professions, and psychological and wider health care support roles). The primary hypothesis was that participants allocated to unguided digital MBSH will show greater reductions in stress from the baseline to postintervention time points (4.5 months following randomization) in comparison with participants in the active control trial arm. The secondary hypotheses were that unguided digital MBSH will be more effective than active control in improving mental health outcomes, work-related outcomes, and potential mechanisms of action from baseline to after the initial intervention period (1.5 months after randomization) and from the baseline to postintervention time points. Analyses examining whether intervention engagement and improvements in mindfulness, self-compassion, worry, and rumination mediated the effects of the intervention on improvements in stress were planned to ascertain intervention-specific mechanisms of action.
This study was a 2-arm superiority definitive RCT, with a 1:1 allocation and no stratification, comparing unguided digital MBSH (Headspace []) with an active control group (the NHS digital platform for work-related stress, Moodzone []). Assessments were performed at 3 time points: baseline (time point 1 [T1]), after the initial intervention period (time point 2 [T2]; 1.5 months after randomization), and at the postintervention time point (time point 3 [T3]; 4.5 months after randomization).
Ethics approval (reference ER/HT207/8) was provided by the University of Sussex, and study approval was granted by the Health Research Authority (reference 16/HRA/5525). The study was prospectively registered on the International Standard Randomised Controlled Trial Number register (reference number: 15424185) [].
Participants and RecruitmentParticipants had to (1) be employed within an NHS Trust or general practitioner practice in England, (2) be working in roles that involved direct contact with patients for a minimum of 1 day per week, (3) be currently in work (ie, not on long-term sickness absence), (4) be willing to refrain from engaging in other psychological interventions during the course of the study, (5) have regular personal access to an Apple, Android smartphone, or tablet or a computer with internet access, (6) be aged ≥18 years, and (7) have sufficient English language skills to read and understand the intervention materials. There were no additional exclusion criteria. Recruitment took place between February 21, 2017, and September 18, 2018.
Sample size calculations were conducted using G*Power [], which indicated that 527 complete cases per study arm (1054 total) would be needed to detect a small between-group difference of Cohen d=0.20 (P=.05; 90% power; 2-tailed) on the primary outcome (stress at T3), with this estimate based on a meta-analysis of MBSH on stress outcomes []. A conservative estimate of a 50% study dropout rate was assumed [], giving a total required sample size of 2108 (n=1054 per arm).
A total of 2182 participants were enrolled in the study (completed baseline measures and were randomized); 1095 (50.18%) were randomized into the Headspace arm, and 1087 (49.82%) were randomized into the Moodzone arm. The participant flow is shown in the CONSORT (Consolidated Standards of Reporting Trials) diagram (), and further participant details are reported in the Results section.
Figure 1. CONSORT (Consolidated Standards of Reporting Trials) diagram showing participant flow. ITT: intention-to-treat. View this figureInterventionsHeadspaceThe Headspace MBSH digital program [] offers a range of brief mindfulness-based practices alongside psychoeducational materials. The Headspace MBSH digital program can be accessed via a website [] or an app available on the Apple app store or Android Play store. Headspace offers a range of mindfulness-based practices and psychoeducational animations, including an introductory series that comprises daily sessions designed to teach foundational mindfulness principles and practices, as well as packs designed for more specific emotional difficulties (eg, stress and anxiety) and brief SOS mindfulness practices designed to be used in times of acute stress. Headspace also offers guidance on informal mindfulness practices that can be undertaken while performing everyday activities, such as running and cycling, and there is written information, including research evidence, related to mindfulness and a frequently asked questions section. At the time of the study, mindfulness practices were verbally guided by Andy Puddicombe, a founder of Headspace with many years of experience in mindfulness practice. For the introductory sessions, users were verbally guided to bring nonjudgmental awareness to the body, breath, thoughts, and feelings, with later sessions also inviting users to bring awareness to difficulties arising during practice (eg, boredom and restlessness) and behavioral choices. At the time of recruitment, users were invited to start the Headspace program by completing the Take Ten introductory pack, which involved undertaking guided 10-minute mindfulness practices daily for 10 consecutive days. Upon completion of the Take Ten, participants were provided with unlimited access to the full range of Headspace content. Although participants were free to choose the content they engaged with, they were invited to perform at least one 10-minute mindfulness practice daily for the duration of the study. Although practices range in length from 3 to 20 minutes, users can select the duration of most sessions. As Headspace is a live product, the program structure was nonstatic, and participants were able to access newly released and changing content as it became available.
MoodzoneThe NHS Moodzone psychoeducational digital platform [] was used as an active control. At the time of recruitment, the website offered a range of evidence-based psychosocial recommendations, advice, and guidance on how to manage work-related stress and mental health difficulties effectively. The initial web page was divided into the following sections: “What causes work stress?” “How to manage work stress,” “Learn to speak out,” “Spot the signs of work stress,” and “Who else can help with work stress?”; each provided information and recommendations or guidance relevant to the respective questions. Moodzone also included information, videos, audio tracks, podcasts, and links to other related resources. Participants were invited to engage with the Moodzone website for 10 minutes per day for the duration of the study. It should be noted that although very similar content is still available [], the Moodzone website used in this study is no longer active. As with Headspace, a live nonstatic version of Moodzone was used in the study, meaning that participants could access new and changing content as it became available. Before this study, adequately powered Moodzone trials were not undertaken. However, related evidence from a meta-analysis of RCTs identified a significantly small effect (Cohen d=0.20; P=.04) of passive psychoeducational interventions compared with control conditions in reducing depression and psychological distress at the postintervention time point [].
MeasuresFull details of the measures are shown in [,-].
Participants completed the measures described in at T1, T2, and T3 unless stated otherwise.
Participant measures.Participant measures
Short version of the 21-item Depression, Anxiety, and Stress Scale []; the Stress subscale was the primary outcome, with time point 3 (T3) being the primary end pointShort Warwick Edinburgh Mental Well-being Scale []Maslach Burnout Inventory []15-item version (minus “observe”) of the Five Facets of Mindfulness Questionnaire []Self-Compassion Scale–Short Form []Compassionate Love Scale []Penn State Worry Questionnaire []Brooding subscale of the Ruminative Response Scale []Sickness absence measured at time point 1 [T1] and T3 was assessed using 1 item that asked participants to report how many days they had been absent from work because of sickness during the past 3 monthsDemographic information assessed at T1 included participants’ age; gender; marital status; number of children aged <18 years; number of children aged ≥18 years; National Health Service job role; trust and team; number of hours worked per week in the National Health Service job role; highest level of education; individual and household annual incomes; ethnicity; and perceived relative socioeconomic status, with response options from 1 (lowest) to 10 (highest) perceived socioeconomic status []Intervention expectancy at T1 (Credibility and Expectancy Questionnaire [])Self-reported intervention engagement at time point 2 [T2] and T3:Formal engagement: self-reported average number of days per week spent following guided mindfulness meditation on Headspace or following a recommended stress management or well-being strategy on the Moodzone web pageInformal engagement: self-reported average number of days per week participants brought mindfulness to daily activities or recommended stress management and well-being strategies from Moodzone into their daily lives; at T2, these questions were asked in relation to the past month, and at T3, they were asked in relation to the past 3 monthsIntervention evaluations at T2 and T3: participants asked how likely they were to recommend the intervention to friends and family, how much they really felt that their allocated intervention had helped their well-being, and how likely they were to continue practicing mindfulness (Headspace participants) or stress management and well-being strategies (Moodzone participants) over the following 6 monthsHypothesis guess at T3: participants asked to state what they thought the purpose of the study wasIntervention deviations at T3: participants asked to indicate whether they had engaged in the alternative study intervention during the course of the studyPrior mindfulness experience at T3: participants asked to indicate their experiences of mindfulness before the study, including mindfulness-based cognitive therapy, stress reduction, mindfulness-based self-help, and Headspace, and how often they practiced mindfulnessSerious adverse events were recorded in accordance with the National Institute for Health Research Good Clinical Practice guidelines []Participants were also asked to indicate the extent to which they agreed or disagreed that they had experienced “lasting bad effects” from using their allocated intervention (based on Crawford et al []); if the participants agreed or strongly agreed, they were asked to provide further detailsTextbox 1. Participant measures.ProcedureNHS staff were recruited via posters and leaflets in NHS settings, invitation emails sent through NHS organizations, and study advertisements on staff web pages or newsletters. Potential participants were directed to the study website hosted by Qualtrics XM [], where they could read the participant information and confirm their eligibility and informed consent (). After consenting, the participants were emailed a weblink along with a unique ID code and asked to self-complete the T1 measures on Qualtrics. Participants completed T1 measures, which were allocated automatically to Headspace or Moodzone using a 1:1 block randomization with a block size of 4 by Qualtrics. To ensure allocation concealment, the members of the research team responsible for collecting data and communicating with participants were blinded to the block size. Participants were informed of their random allocation and subsequently asked to indicate their views on the credibility and expectations of their assigned intervention.
Following the completion of the T1 assessment, participants were emailed information on how to access their allocated intervention. Intervention participants were given 12 months of free access to Headspace, and Moodzone was available free of charge. Allowing 5 days for participants to receive this information or download their intervention, participants were invited to engage with their allocated intervention for 10 minutes per day, every day during the initial 30-day study period. At 35 days after randomization, participants were emailed a link to complete the T2 assessments on Qualtrics and invited to continue engaging with their allocated intervention for 10 minutes per day during the remaining 90-day study period. On average, T2 was completed at 1.5 months (SD 0.57) after randomization. At 125 days after randomization, participants were emailed a link to complete the T3 assessment on Qualtrics, with T3 completed at an average of 4.5 months (SD 0.53) from randomization. At this point, the participants who completed the study were given access to the alternative intervention.
Participants who did not complete assessments within 1 week of them being sent were reminded to do so via email. One reminder email was sent for completion of the T1 assessments, and a maximum of 4 reminder emails at weekly intervals were sent for T2 and T3 assessments. The research team was available to answer technical questions or queries via email. No further support was provided.
To improve trial quality and blind participants to the study condition and direction of study hypotheses, advertisements about the study simply referred to both conditions as “online interventions to reduce NHS staff stress,” and details of the alternative or nonallocated intervention were not communicated to participants until T3 assessments (after outcome and engagement measures had been taken). As all assessments were completed on the web without researchers present, the potential for researcher bias to influence assessment outcomes was minimized. All but the mediation analysis was conducted blind to the study arm.
Participants were given the option to enter a prize draw to win 1 of 5 gift vouchers for £50 (US $60).
Data Analysis PlanDescriptive statistics are reported by trial arm and time as means and SDs (for continuous data), medians and IQRs (for ordinal data), and counts and percentages (for categorical data). Data analysis was conducted using SPSS (version 25; IBM Corp) [] and R (version 4.0.2; R Foundation for Statistical Computing) [] and the following packages: emmeans [], lme4 [], mice [], papaja [], and tidyverse [].
Handling Missing DataA minimal number of items were missing at the item level, and missing values for missing items were imputed (using a single imputation) using predictive mean matching in mice []. At the scale level, multiple imputation was used to handle missing values. Further details are provided in [-].
Model SelectionAs participants were nested within job roles (level 3), there are good reasons for model variations in intervention effects between job roles []. There is participant-level randomization to intervention arms in such a model, and job roles act as a crossed effect. We can think of time (i) as being nested within participants (j), which is nested within job roles (k); however, the effect of the treatment arm occurs at level 2 (the participant level), not level 3 (the job role level), of the hierarchy. This situation is described by the model given in .
This saturated model includes random effects for time, trial arm, and their interaction at level 3. However, this model resulted in convergence problems that yielded erratic estimates of random effects involving the trial arm in the raw sample and nearly all imputed samples. On the basis of this preanalysis, a simpler model seemed more appropriate, in which only time was treated as a random effect and only at level 2. However, to model level 3 variability in outcomes, a random intercept (at level 3) was included. This simpler model converged in all the imputed samples. The resulting model is described in (notice that at level 3, a total of 2 random effects have been knocked out).
To sum up, the hypotheses were tested using a growth model fit as a general linear mixed model, with observations (level 1) nested within participants (level 2) nested within job roles (level 3). Time (time from baseline at which responses were recorded) and trial arm were predictors. The effect of the intervention was quantified and tested with the interaction between time and trial arm, which shows the degree to which the change in the outcome over time is different between the 2 trial arms. Between-group effects were reported separately at T2 and T3 in the event of significant (P<.05) trial arm × time interactions. The primary analysis was conducted on the intention-to-treat (ITT) sample with multiple imputed data sets. Secondary analysis was conducted on the per-protocol sample (formal engagement T1-T2 on at least 3 days per week []) with the multiple imputed data sets.
Details of the plan for reliable change analysis, mediation analysis, and randomization check can be found in [,,,,,,,-].
The saturated model showing the data structure.Level 1
Depression, Anxiety, and Stress Scale–Stressijk= π0jk+ π1jkTimeijk+
ijkLevel 2
π0jk= γ00k+ γ01kTrial armjk+ ζ0jkπ1jk= γ10k+γ11kTrial armjk+ ζ1jkLevel 3
γ00k= δ000+υ0kγ10k= δ100+υ1kγ01k= δ010+υ2kγ11k= δ110+υ3kTextbox 2. The saturated model showing the data structure.The fitted model.Level 1
Depression, Anxiety, and Stress Scale–Stressijk= π0jk+ π1jTimeijk+ijkLevel 2
π0jk= γ00k+ γ01Trial armjk+ ζ0jkπ1j=γ11Trial armjk+ ζ1jkLevel 3
Textbox 3. The fitted model.presents the demographic characteristics of the participants by study arm, and presents descriptive statistics on all outcome measures at all time points by study arm. shows that participants represented a broad range of NHS Trust types and health care professions. As would be expected of a health care workforce, most participants were educated to at least an undergraduate degree level and were earning, on average, the median UK annual salary; most participants were working full-time. Participants covered the full working age spectrum, although they were disproportionately White and female. For the randomization check, all Bayes factors were very close to 0, suggesting very strong evidence for the null hypothesis: randomization was successful in balancing demographic and baseline measurements across the 2 trial arms (). There were also no differences in dropout rates between the trial arms. A formal analysis using a multilevel generalized linear model, with a random intercept, predicting dropout (1=in the study and 0=dropped out) from the trial arm, study wave (as a categorical variable), and their interaction showed no significant effects.
Table 1. Demographic characteristics of participants (N=2182).CharacteristicsMoodzoneHeadspaceHighest educational achieved, n (%)aGCSE: General Certificate of Secondary Education.
bNVQ 2: National Vocational Qualification level 2.
cNHS: National Health Service.
dGP: general practitioner.
Table 2. Descriptive statistics on all outcome measures at all time points (raw complete case data; N=2182).Measure and armTime point 1 (baseline)Time point 2 (1.5 months)Time point 3 (4.5 months)aDASS-21: 21-item Depression, Anxiety, and Stress Scale.
bSWEMWBS: Short Warwick Edinburgh Mental Well-being Scale.
cMaslach Burnout Inventory.
dFFMQ15: 15-item Five Facets of Mindfulness Questionnaire.
eSCS-SF: Self-Compassion Scale–Short Form.
fPSWQ: Penn State Worry Questionnaire.
gRRS: Ruminative Response Scale.
hCLS: Compassionate Love Scale.
iNot available.
jN/A: not applicable.
kCEQ: Credibility and Expectancy Questionnaire.
lMeans created from subscale totals of z scores [].
Primary Outcome (Stress)ITT Analysisshows that the main effects of trial arm (Headspace or Moodzone) and time (months) were significant, as was the crucial trial arm × month interaction, which indicates that the trajectories of the 21-item Depression, Anxiety, and Stress Scale (DASS-21) Stress scores over time differed significantly between the 2 trial arms for the ITT sample (). The parameter value (b=−0.31) tells us that the rate of change (gradient) over time was −0.31 points greater on the DASS-21 Stress subscale per month in the Headspace arm than in the Moodzone arm. Specifically, for every month that passed, DASS-21 stress scores changed by −0.23 units on the scale in the Moodzone group compared with a corresponding change of −0.54 units in the Headspace group (ie, a difference between arms of −0.31 units per month).
To break down this effect, comparisons were made between the estimated marginal means of the outcome from the model at 1.5 (T2) and 4.5 (T3) months in the 2 arms. In the Moodzone arm, stress was significantly higher at baseline than at both 1.5 months (b=0.34, SE 0.09; P<.001) and 4.5 months (b=1.03, SE 0.26; P<.001). Stress was also significantly higher at 1.5 months than at 4.5 months (b=0.69, SE 0.18; P<.001). Similarly, in the Headspace arm, stress was significantly higher at baseline than at both 1.5 months (b=0.81; SE 0.08; P<.001) and 4.5 months (b=2.42, SE 0.25; P<.001), and significantly higher at 1.5 months than at 4.5 months (b=1.61, SE 0.17; P<.001). The b values represent the difference in the estimated marginal means; they show that, for example, at 4.5 months, the decrease in DASS-21 Stress compared with baseline was 1.03 points in the Moodzone arm and 2.42 points in the Headspace arm. In other words, at 4.5 months after randomization, Moodzone reduced DASS-21 Stress scores by approximately 1 point along the 42-point scale, and the equivalent change for Headspace was a reduction of approximately 2.5 points along the scale. In addition, the difference in estimated marginal means between the 2 arms was b=0.62 (SE 0.31; P=.045) at baseline, b=1.08 (SE 0.30; P<.001) at 1.5 months, and b=2.00 (SE 0.42; P<.001) at 4.5 months (the preregistered primary end point).
Table 3. Model for the 21-item Depression, Anxiety, and Stress Scale Stress (intention-to-treat sample with multiple imputation).EffectUnstandardized b (SE; 95% CI)t test (df)P valueIntercept15.33 (0.40; 14.55 to 16.11)38.41 (6054.49)<.001Trial arm−0.62 (0.31; –1.23 to –0.01)−2.01 (5129.90).045Months−0.23 (0.06; –0.35 to –0.11)−3.92 (165.07)<.001Trial arm × month−0.31 (0.08; –0.47 to –0.14)−3.64 (151.13)<.001
Figure 2. DASS-21 Stress scores over time (intention-to-treat complete case sample). Each triangle represents a Headspace participant, and each circle represents a Moodzone participant. DASS-21: 21-item Depression, Anxiety, and Stress Scale. View this figurePer-Protocol AnalysisThe per-protocol sample included only participants who formally engaged with their allocated intervention at least 3 days per week during the initial intervention period (T1-T2). shows a significant trial arm × month interaction, indicating that the trajectories of stress over time differed significantly between the 2 trial arms. The rate of change over time was −0.28 DASS-21 Stress units greater per month in the Headspace arm than in the Moodzone arm. Specifically, in the Moodzone arm, the rate of change over time was −0.42, which means that for every month that passed, DASS-21 Stress scores decreased by 0.42 points; however, in the Headspace, arm the rate of change over time was −0.70 (a difference of −0.28 between arms), which means that for every month that passed, DASS-21 Stress decreased by 0.70 points.
In the per-protocol sample in the Moodzone arm, stress was significantly higher at baseline than at both 1.5 months (T2; b=0.63, SE 0.16; P<.001) and 4.5 months (T3; b=1.88, SE 0.47; P<
留言 (0)