If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Oregon Institute of Occupational Health Sciences, Oregon Health & Science University, 3181 SW Sam Jackson Park Rd, L606, Portland, OR 97239Department of Public Health & Preventive Medicine, Oregon Health & Science University, 3181 SW Sam Jackson Park Rd, Mail Code CB 669, Portland, OR 97239Department of Psychology, Portland State University, 1721 SW Broadway, Rm 317, Portland, OR 97201
Population Dynamics Branch, Eunice Kennedy Shriver National Institute of Child Health and Human Development, 6100 Executive Blvd., Bethesda, MD 20892-7510
Department of Biobehavioral Health and Penn State Institute of the Neurosciences, Pennsylvania State University, 221 Biobehavioral Health Bldg, University Park, PA 16802
Harvard Center for Population and Development Studies, Harvard University, 9 Bow Street, Cambridge MA 02138Department of Social and Behavioral Sciences, Harvard T. H. Chan School of Public Health, 677 Huntington Ave, Kresge Building, Boston, MA 02115
Department of Biobehavioral Health and Penn State Institute of the Neurosciences, Pennsylvania State University, 221 Biobehavioral Health Bldg, University Park, PA 16802Harvard Center for Population and Development Studies, Harvard University, 9 Bow Street, Cambridge MA 02138Department of Social and Behavioral Sciences, Harvard T. H. Chan School of Public Health, 677 Huntington Ave, Kresge Building, Boston, MA 02115Department of Medicine, Brigham and Women's Hospital, 221 Longwood Ave, Boston, MA 02115Division of Sleep Medicine, Harvard Medical School, 221 Longwood Ave, Boston, MA 02115
The Work, Family, and Health Network Study tested the hypothesis that a workplace intervention designed to increase family-supportive supervision and employee control over work time improves actigraphic measures of sleep quantity and quality.
Design
Cluster-randomized trial.
Setting
A global information technology firm.
Participants
US employees at an information technology firm.
Interventions
Randomly selected clusters of managers and employees participated in a 3-month, social, and organizational change process intended to reduce work-family conflict. The intervention included interactive sessions with facilitated discussions, role playing, and games. Managers completed training in family-supportive supervision.
Measurements and results
Primary outcomes of total sleep time (sleep duration) and wake after sleep onset (sleep quality) were collected from week-long actigraphy recordings at baseline and 12 months. Secondary outcomes included self-reported sleep insufficiency and insomnia symptoms. Twelve-month interviews were completed by 701 (93% retention), of whom 595 (85%) completed actigraphy. Restricting analyses to participants with ≥3 valid days of actigraphy yielded a sample of 473-474 for intervention effectiveness analyses. Actigraphy-measured sleep duration was 8 min/d greater among intervention employees relative to controls (P < .05). Sleep insufficiency was reduced among intervention employees (P = .002). Wake after sleep onset and insomnia symptoms were not different between groups. Path models indicated that increased control over work hours and subsequent reductions in work-family conflict mediated the improvement in sleep sufficiency.
Conclusions
The workplace intervention did not overtly address sleep, yet intervention employees slept 8 min/d more and reported greater sleep sufficiency. Interventions should address environmental and psychosocial causes of sleep deficiency, including workplace factors.
The recent Institute of Medicine report “Sleep disorders and sleep deprivation: an unmet public health problem” highlighted improving sleep health in the United States as a critical public health need.
For these reasons and others, increasing the proportion of US adults and adolescents obtaining adequate sleep has become a new federal priority in Healthy People 2020
Strategic opportunities in sleep and circadian research: report of the Joint Task Force of the Sleep Research Society and American Academy of Sleep Medicine.
Clinical sleep medicine focuses primarily on treatment of individual patients' sleep disorders; however, individual-level treatments are limited in reach for meeting this challenge.
The role of workplace factors in sleep loss has been studied for decades.
Most studies focus on the individual level and use stress-based models. Extreme work hours have negative impacts on sleep duration that can be alleviated to some degree by changes in work schedules.
Demands of the global recession coupled with mobile technologies have extended work hours and blurred the boundaries between work and nonwork, especially within information technology (IT) industries.
“Impoverished” environments, such as workplaces where employees lack social support and/or have limited control over their time, are associated with a range of negative health outcomes.
Employees in low support/control work environments are particularly vulnerable to work-family conflict, which occurs when the demands of work are incompatible with the nonwork demands of family and personal life.
Work-family demands have increased for a growing number of dual-earner families, single mothers, and “sandwich” families who must simultaneously provide care for young and old.
In an extended care (nursing home) workplace setting, employees whose managers were less supportive of work-family integration averaged about 30 minutes less daily sleep (measured with actigraphy) than employees with supportive managers.
Limited work-family intervention research has shown that improving supervisor support and employee control over work time benefits worker health and/or sleep. A randomized field experiment with grocery workers evaluated the effects of training supervisors on family supportive behaviors. For workers experiencing high work-family conflict, the intervention improved self-reported health and job satisfaction and reduced turnover intentions.
A longitudinal study of white-collar employees at the headquarters of a Fortune 500 retail organization found that an intervention promoting greater employee control over time increased self-reported sleep on nights before work by almost an hour
However, structural workplace interventions to reduce work-family conflict remain scarce, and no prior work-family intervention has measured impacts on objective sleep outcomes within a randomized controlled trial.
The present study evaluated the effects of a theoretically informed workplace intervention on objectively measured employee sleep with a randomized controlled design. The study was implemented at an IT company, and measurements were collected at baseline, 6 months, and 12 months. Our primary hypotheses were that the intervention would improve actigraphically measured total sleep time and wake after sleep onset (WASO; a marker of insomnia symptoms), as well as self-reported measures of sleep insufficiency and insomnia symptoms at the 12-month time point relative to the usual practice (control) condition. Our secondary hypothesis was that intervention effects on sleep at the 12-month time point would be at least partially mediated by increased control over work hours and subsequently reduced work-family conflict at the 6-month time point.
Methods
Study methods were approved by appropriate institutional review boards. Primary study outcomes were actigraphy-based measures of total sleep time per day and WASO (in minutes) and self-reported measures of sleep insufficiency and insomnia symptoms measured at baseline and 12 months. Of interest was the change in these outcomes over the study year in the intervention and usual practice (control) arms of the study.
Design and data collection
The study used a cluster-randomized design with 3 measurement time points reported here (baseline, 6 months, and 12 months). Recruitment spanned from September 2009 to September 2010, and 12-month follow-up was completed on September 2011. Fifty-six “study groups” or clusters of managers and employees were identified with company representatives as eligible for randomization. Adaptive randomization occurred after baseline data collection for each study group, as previously described.
Bray JW, Kelly E, Hammer LB, Almeida DM, Dearing JW, Berkowitz King R, et al. An integrative, multilevel, and transdisciplinary research approach to challenges of work, family, and health. Research Triangle Park, NC 2013. RTI Press publication No. MR-0024-1303.
All employees within these groups were eligible to participate. Some study groups involved large teams under 1 leader, whereas other study groups involved multiple teams who worked closely together or reported to the same senior leader. We refer to randomization units as study groups to denote that they are aggregations of existing functional work groups that operated in the organization. Baseline demographic descriptive statistics for the sample are provided in Table 1.
Table 1Percentage or mean ± SD for demographic characteristics by condition (n = 474).
Usual practice (n = 240)
Intervention (n = 234)
Female
37.9%
42.7%
Age
46.6 ± 8.4
46.8 ± 8.8
Race/ethnicity
White, non-Hispanic
72.1%
70.5%
Black or African American, non-Hispanic
1.3%
1.7%
Asian Indian
13.8%
12.4%
Other Asian
4.2%
5.1%
Other Pacific Islander
0.8%
1.3%
Hispanic
6.7%
8.1%
More than 1 race
1.3%
0.9%
Married or living with partner
79.2%
81.2%
No. of children
1.0 ± 1.2
1.0 ± 1.0
Elder care
25.8%
24.4%
Education
High school graduate
2.5%
3.0%
Some college or technical school
17.9%
22.7%
College graduate
79.6%
74.4%
Hours worked per week
45.5 ± 6.0
45.6 ± 5.5
Shift
Variable schedule
21.3%
20.9%
Regular daytime
77.9%
78.2%
Rotating
0.4%
0.9%
Split shift
0.4%
0.0%
Note: Descriptives shown for all individuals included in analyses (n = 474).
Recruitment materials emphasized the value of the research for employees and the organization as well as for scientific knowledge. Trained study site managers introduced the study to employees at work sites and coordinated project implementation. To minimize bias, separate and blinded study field interviewers obtained written informed consent and collected data from employees within intervention and usual practice study groups at baseline, 6-month, and 12-month time points.
Bray JW, Kelly E, Hammer LB, Almeida DM, Dearing JW, Berkowitz King R, et al. An integrative, multilevel, and transdisciplinary research approach to challenges of work, family, and health. Research Triangle Park, NC 2013. RTI Press publication No. MR-0024-1303.
Baseline data were collected approximately 1 month before the onset of the 3-month intervention. Self-reported measures were collected as part of a 60-minute interview at each time point. Employees were asked to wear a wrist actigraph for 1 week, as described below. Employees received up to $60 for completing all worksite data collection components at each time point. Figure 1 depicts study recruitment and retention.
Fig. 1Flow diagram of study recruitment and retention. A small work group was randomized to intervention (n = 16) but never invited to STAR activities, by error. These individuals were excluded from analyses. rr, retention rate.
While randomization occurred at the study group level (m = 56), multilevel analyses used coding for the smaller unit of work group level to best account for functional day-to-day work at the organization (m = 107). Given the nonindependence of measures due to the nesting of observations within individuals across time and the nesting of individuals within work groups in the organization, we used a recommended general linear mixed modeling approach for cluster-randomized designs
in SAS (version 2.2, 2010; SAS Institute, Cary, NC) with restricted maximum likelihood estimation. The key model parameter was the interaction between assessment wave and the intervention indicator that represents the differential change in the sleep outcomes across time and intervention conditions. These models take into account baseline values of the outcome variable. Furthermore, differences between intervention and control groups on demographic variables are irrelevant because multiple waves are included in the models (ie, each person controls for themselves, despite what group they are in). To test our secondary hypothesis of mechanisms underlying intervention effects, we used multilevel structural equation modeling approaches accounting for longitudinal and nesting features of the data
in Mplus (version 4.2, 2006; Muthen & Muthen, Los Angeles, CA). Specifically, we modeled whether the intervention directly increased control over work hours at the 6-month time point, followed by whether these changes, in turn, reduced work-family conflict at the 6-month time point and then whether reductions in work-family conflict were directly related to sleep improvements at 12 months. The product of these direct effects defines the indirect effect (ie, mediation) in our model.
revealed no substantive differences in mediation model results (data not reported).
Intervention
The intervention was a 3-month structural and social change process designed to increase (1) employee control over work time and (2) family supportive supervisory behaviors.
The change process was an integration of 2 interventions that, in prior evaluations, had independently addressed family supportive supervisor behaviors and employee control, respectively.
For implementation, the new intervention was named STAR (Support. Transform. Achieve. Results.). See Figure 2 for an intervention overview.
Fig. 2Overview of the intervention process. The 3-month, workplace-based cultural change process included multiple steps and training sessions for both employees and managers. Adapted from Kossek et al.
During the intervention period, a facilitator led employees and managers through 8 hours of participatory sessions to transition them from a time-based to a result-based work culture. This change was expected to impact employee sleep by enabling greater control over wake time. Participatory sessions included structured presentations about the reasons for workplace changes as well as facilitated discussions, role playing, and games. During this process, leaders and employees were asked to make structural changes and exercise greater freedom to work at whatever time and whatever place they wanted, as long as they produced their expected work results. After being challenged to enact changes, work groups participated in daily Web polls where they self-monitored and viewed feedback about their collective actions.
Managers/supervisors participated in all change activities plus 4 hours of training in family supportive supervisor behaviors and meetings to discuss the change process. Increased supervisor support was expected to impact employee sleep primarily by reducing employee stress, thereby enhancing employee psychological and physical readiness for sleep onset in the evening and quality sleep through the night. The training in supportive supervision, which was named “weSupport for Supervisors” for implementation, began early in the overall intervention process by having managers meet with a facilitator individually to complete a 1-hour self-paced computer-based training, set goals, and start a self-monitoring activity. The individual training was computer based (cTRAIN; NWeta, Lake Oswego, OR) and designed according to behavioral principles (eg, self-paced, mastery required, or frequent quizzes). Goal setting and self-monitoring of supportive behaviors were implemented using an Enterprise application for iPhone/iPod Touch (HabiTrack; Oregon Health & Science University, Portland, OR) that was designed based on best practices in clinical and workplace self-monitoring methods.
Each supervisor completed 2 rounds of self-monitoring (each round = 2 weeks); the first occurring immediately after the computer-based training and the second occurring toward the end of the change process.
During intervention implementation, the participating organization announced a pending merger. During mergers, employees are likely to experience heightened job insecurity, which could impact work-family conflict and sleep. Therefore, to control for the possible effect of the merger announcement, multilevel models included an indicator of the timing of employee baseline interviews (premerger or postmerger announcement; Table 3).
Throughout the study, qualitative process evaluation data were collected via observation of training sessions and semistructured interviews with employees and managers. The relevant interviews were conducted with participants in the intervention, including both employees and managers, covering their responses to the intervention activities and descriptions of how the intervention was being implemented in their work group or by them individually. In each study group, trained observers identified employees and managers who seemed engaged during the training sessions and invited them to participate in the interviews. The researchers recruited intervention participants who ranged from being enthusiastic, skeptical (but interested), or resistant to the intervention. More than 90% of invited participants completed these qualitative interviews (n = 128). The interview guide did not ask explicitly about sleep but did ask about health broadly. Selected excerpts from these discussions relevant to sleep and the intervention are included below.
Measures
Actigraphy outcome measures
Wrist actigraphy (Actiwatch Spectrum; Philips/Respironics, Murrysville, PA) data were analyzed using the recently validated manufacturer's standard algorithm at medium sensitivity
for 24-hour total sleep time (including naps) and WASO during the main sleep period for all study days with a valid recording, as described below.
Data from each subjects' Actiwatch was uploaded to databases (Respironics Actiware sleep scoring program version 5.61) and analyzed by at least 2 members of the scoring team. All analyses were completed blinded to condition. Scorers first determined the validity of each recording and the validity of each day of the recording and then manually inserted sleep periods (main sleep intervals and naps) based on study-specific standard operating procedures applied similarly to all recordings. Scorers started by finding points of decreased activity levels and also used sudden, decreased light levels (lux) to help suggest, but not confirm, bedtimes. By visually reviewing the entire sleep recording before manually setting intervals, scorers took into consideration the activity intensity of the subject for determining naps and sleep periods to be able to accurately define sleep periods rather than low movement activities such as watching TV or reading in bed. Sleep periods began at the last epoch of high activity (>10 activity counts) preceding at least five 30-second epochs of <10 activity counts, indicating little to no movement. Wake times were determined the same way by finding the first epoch of sustained high activity (>10 activity counts) after at least five 30-second epochs of <10 activity counts. A recording was deemed invalid if there was constant false activity (a device malfunction) or if the data were irretrievable. Reasons for invalid days within a recording include watch error, such as false activity, and subject noncompliance (>4 hours of Actiwatch off-wrist time throughout the day or an off-wrist period >60 minutes within 10 minutes of the determined beginning or end of the main time in bed period for that day). Sleep diaries were not used due to subject burden, recall bias, and low response rates in other studies.
Concordance between at least 2 scorers was measured for the validity of the recording and the number of valid days. All scorers used the same cut time to define 24-hour days. If there were no discrepancies between scorers in any of those variables, the analyses were then checked that all scorers determined that the recording had the same number of sleep periods and that each sleep period as a main sleep or nap was labeled in an identical manner. Any discrepancies were resolved among the scoring team with the final determination by the last author. Finally, each of the sleep periods was checked on an interval-by-interval basis. Any corresponding intervals that exceeded a 15-minute difference in duration or exceeded 15 minutes of either total sleep time or WASO were rescored with all final adjudications by the last author. The Actiware sleep scoring program separates the recording into 30-second epochs and determines sleep or wake using weighted activity counts using the medium sensitivity standard algorithm, as recently validated vs the criterion standard of polysomnography.
If the total activity count exceeded the wake threshold level determined by the researchers (medium wake threshold level selection uses a wake threshold value of 40 total activity counts), then the epoch was determined to be wake, and if the total activity count was below the set wake threshold level, the epoch was determined to be sleep. Total sleep time was the total number of epochs determined to be sleep multiplied by the set epoch length, whereas WASO was the total number of epochs determined to be wake multiplied by the set epoch length, with both variables being presented as a value in minutes for both main sleep intervals and nap sleep intervals. Actigraphy measures were screened for outliers using box and whisker plots, but no outliers were present 3 times the interquartile range below or above the 25th and 75th percentile, respectively. More conservative histograms revealed 2 cases <200 minutes for total sleep time and 1 case >125 minutes for WASO. Analyses were then conducted with and without outliers found in these more conservative histograms, but results did not change. Thus, the current analyses are based on the larger sample.
Self-report outcomes
Participants answered questions regarding their sleep for the previous 4 weeks. Sleep insufficiency was assessed with 1 question: “How often during the past 4 weeks did you get enough sleep to feel rested upon waking up?” with response categories of never, rarely, sometimes, often, and very often,
: “During the past four weeks, how often could you not get to sleep within 30 minutes?” and “During the past four weeks, did you wake up in the middle of the night or early morning?” with response categories of never, less than once a week, once or twice a week, and 3 or more times a week (α = .32).
Mediation measures
All mediators were assessed with means, computed if participants provided at least 75% of valid responses to the items within a scale. Work-family conflict, a well-validated scale,
was assessed using the mean of 5 items; a sample item reads “The demands of your work interfere with your family or personal time.” Each item used the following response categories: “strongly agree, agree, neither agree nor disagree, disagree, and strongly disagree” (α = .91). Control over work hours
was measured with the mean of 8 items; a sample item is “How much choice do you have over when you take vacations or days off?” Response categories included “very little, little, a moderate amount, much, and very much” (α = .82).
Results
Baseline interviews were completed by 823 individuals (70% response rate), of whom 710 (86%) also completed wrist actigraphy. Among these individuals, a minimum of 3 valid days of actigraphy, a quality metric for reliable sleep estimates, was available in a final baseline sample of 618 individuals (intervention, n = 313; control, n = 305). The 12-month interviews were completed by 701 individuals (93% retention rate), of whom 595 (85%) completed wrist actigraphy. There is no evidence of differential retention/attrition; the wave-1-only sample was not significantly different (at baseline) from those who completed both waves on experimental condition, schedule control, supervisor support, work-family conflict, time adequacy, work hours, or other demographics. A minimum of 3 valid days of actigraphy was available at both baseline and 12-month time points from a sample of 473-474 (intervention, n = 233-234; control, n = 240), which was the functional sample size for the current intervention analyses. On average, individuals had 7.3 ± 2.1 (mean ± SD) valid days of actigraphy, with only 14 individuals having 3 valid days and 23 with 4 valid days. A smaller sample (n = 456-457; intervention, n = 227-228; control, n = 229), due to nonparticipation at 6-month wave, was used in mediation path models. We provide sample size ranges because the sample size varied depending on which sleep outcome was being used in each model. As noted in the statistical analysis section, study groups (m = 56; intervention, 27; control, 29) were the unit of randomization (see Fig. 2 for study flow diagram), but smaller clusters of work groups of employees working together (m = 107; intervention, m = 54; control, m = 53) were used as the appropriate nesting variable in multilevel analyses.
The baseline burden of sleep deficiency using actigraphy and self-reported data is depicted in Figure 3: 65% of the sample exhibited at least 1 of the 3 components of sleep deficiency, including 18% with short actigraphic mean sleep duration, 41% with mean actigraphic WASO >45/min/d, and 22% self-reported sleep insufficiency. In qualitative interviews, common problems that emerged as self-reported reasons for losing sleep included smartphones, staying in bed and thinking about work, not being able to shake a cold because of sleep deprivation, and merger-related sleep deprivation. One participant volunteered the fact that he has trouble sleeping every night because he is constantly on his smartphone, right up until bedtime. He answers his last e-mail for the night, turns it off, and spends hours trying to unwind as he lies there. Then when he gets up in the morning, he's already got e-mails waiting that came in through the night. He said that it started a few years ago, that things have gotten much more hectic and the pace of work has sped up over the past few years, and that his lack of sleep is a result of that.
Fig. 3Venn diagram of the burden of sleep deficiency in the cohort at baseline. Sleep deficiency at baseline (insufficient sleep duration and/or inadequate sleep quality) was defined as having at least 1 of any components: mean Wake After Sleep Onset (WASO) >45 minutes per main sleep period (measured with wrist actigraphy and indicative of insomnia), self-reported sleep insufficiency (never or rarely feeling rested upon waking), and/or mean total sleep time <6.5 h per 24 hours (measured from actigraphy). Due to rounding, totals do not add up to 100%.
Unadjusted means and standard deviations of study variables by condition over time are shown in Table 2. Relative to the control condition, the intervention resulted in significant improvements in employees' actigraphy-measured total sleep time (γ = 8.17, t = 2.07, P = .041) and reported sleep insufficiency (γ = −0.25, t = −3.11, P = .002) at follow-up. Significant intervention-by-time interactions for respective models are shown in Table 3. Intervention participants demonstrated an average of 8 minutes more total sleep time per night (approximately 1 hour per week increase in sleep) and an average of 0.25 point lower sleep insufficiency ratings (5-point scale) compared to control participants (Fig. 4). Although parameter estimates were in the anticipated direction, the intervention did not have a significant impact on changes in actigraphy-measured WASO (γ = 1.31, t = 1.15, P = .25) or reported insomnia symptoms (γ = −0.01, t = −0.20, P = .85) at follow-up.
Table 2Mean ± SD for work-family conflict and sleep outcomes by condition over time (n = 457-474).
Usual practice
Intervention
Unit
Baseline
6 mo
12 mo
Change
Baseline
6 mo
12 mo
Change
Control over work hours
Rating 1-5
3.7 ± 0.7 (n = 240)
3.7 ± 0.7 (n = 229)
0.0 ± 0.5
3.6 ± 0.7 (n = 234)
3.8 ± 0.7 (n = 228)
0.3 ± 0.6
Work-family conflict
Rating 1-5
3.1 ± 1.0 (n = 240)
3.0 ± 0.9 (n = 229)
−0.1 ± 0.8
3.2 ± 0.9 (n = 234)
2.9 ± 0.9 (n = 228)
−0.3 ± 0.9
Total sleep time
Minutes
440.9 ± 49.8 (n = 240)
436.8 ± 51.5 (n = 240)
−4.2 ± 42.2
433.4 ± 58.7 (n = 234)
437.4 ± 58.6 (n = 234)
4.0 ± 43.8
Sleep insufficiency
Rating 1-5
2.8 ± 0.9 (n = 240)
2.8 ± 0.9 (n = 240)
0.0 ± 0.9
2.9 ± 0.9 (n = 234)
2.6 ± 0.9 (n = 234)
−0.2 ± 0.9
WASO
Minutes
42.7 ± 16.3 (n = 240)
40.4 ± 13.7 (n = 240)
−2.3 ± 12.8
43.6 ± 17.0 (n = 234)
42.5 ± 15.7 (n = 234)
−1.0 ± 11.8
Insomnia symptoms
Rating 1-4
2.8 ± 0.8 (n = 240)
2.7 ± 0.8 (n = 240)
−0.0 ± 0.7
2.7 ± 0.8 (n = 233)
2.6 ± 0.7 (n = 234)
−0.1 ± 0.7
Note: Sample size range reflects the number of individuals included in the analyses at baseline, 6 months, and 12 months.
Note: Intervention (0, usual practice; 1, intervention). Wave (0, baseline; 1, 12 months). Merger (0, collection before merger announcement; 1, collection after merger announcement). Analyses also controlled for (1) the number of employees for randomization and (2) groups where the most individuals were involved in software development (groups dominated by other IT jobs are the reference group), but neither of the coefficients were significant and are thus not included in this table. Sample size range reflects the number of individuals included in the analyses at baseline at 12 months, with 1 less person providing reports of insomnia symptoms. Significant P values are in boldface. Abbreviations: γ, unstandardized regression coefficient; CI, confidence interval.
Fig. 4Effects of the intervention on sleep. Adjusted mean changes in (A) actigraphic measures of sleep duration in minutes per day ± SE of the change and (B) mean changes in self-reported sleep insufficiency ± SE of the change. Sleep insufficiency was a self-report of never or rarely feeling rested upon awakening in the past month.
] at 6 months, serving as mediating variables for the effects of the intervention on changes in sleep outcomes at 12 months. We observed a significant indirect effect of the intervention on changes in self-reported sleep insufficiency through increases in control over work hours and subsequent reductions in work-family conflict (full mediation; indirect effect, −0.03; Z = −2.75; P = .01; see Fig. 5). Path modeling confirmed the significant direct effects of the intervention on actigraphy-based total sleep time.
Fig. 5Intervention path model: reduced sleep insufficiency at 12 months fully mediated by increased control over work hours and reduced work-family conflict at 6 months (n = 456). Significant direct effects of the intervention on actigraphic total sleep time (9 min/d in the path analysis) were not mediated by these intervention targets. Dashed bold arrow depicts significant indirect effects (mediation), whereas solid bold arrow depicts significant direct effects. Black arrows = ns. ⁎P < .05.
In qualitative interviews, employees also offered examples of how the intervention had helped them improve their sleep. In interviews conducted after the intervention, sleep was mentioned in 27 of 128 interviews (21%, combining 9- and 15-month rounds of interviews). A theme observed in these qualitative data was how control over work time allowed employees to adjust their sleep for a better fit with acute and chronic demands. For example, 1 employee said, “I don't have to get up at 5 o'clock to leave the house at seven, I can get up at a quarter to seven and be online, or sleep until ten till 8 or whatever…” Another employee reported that before the intervention she had to get up at 4:30 to get an early start at work and avoid evening rush hour traffic. After the intervention, she reported that “if I'm working from home I don't get up until 6:00 or 6:30 and I start working at 7:00… I think that's been extremely beneficial to me in that I do get more sleep than I've had in years.”
Discussion
The current study was a randomized experimental evaluation of the effects of a workplace intervention on directly measured sleep outcomes 12 months after baseline. Relative to control, intervention employees increased total sleep time by an hour per week, and their perceptions of sleep insufficiency improved. Sleep insufficiency changes were mediated by increases in control over work hours and reductions in work-family conflict, which were intervention change targets. Various interventions with other populations have increased actigraphy-measured total sleep time, including delayed school start times in children,
In contrast, the current study evaluated the effects of an intervention on actigraphy-measured total sleep time among working adults in a stressful occupation. Our findings suggest that there is potential to improve sleep duration in the general US population, a federal priority articulated in Healthy People 2020,
by addressing environmental and contextual causes of insufficient sleep, particularly within workplace psychosocial environments.
Our Work, Family, and Health Network (WFHN) intervention was informed by the theoretical expectation that increases in (a) employee control over work time and (b) supervisor support for integrating work and nonwork demands would reduce work-family conflict.
Bray JW, Kelly E, Hammer LB, Almeida DM, Dearing JW, Berkowitz King R, et al. An integrative, multilevel, and transdisciplinary research approach to challenges of work, family, and health. Research Triangle Park, NC 2013. RTI Press publication No. MR-0024-1303.
to test the hypothesis that the WFHN intervention would improve employee sleep quantity and quality and that intervention effects on sleep would be partially mediated by reductions in work-family conflict. We observed full mediation of intervention effects on perceived sleep sufficiency via reductions in control over work hours and work-family conflict. However, the significant direct effect of the intervention on total sleep time was not mediated by these particular work-related resources and stressors, suggesting that factors other than control over work hours and work-family conflict mediated this direct intervention effect. We also observed no direct intervention effects on WASO or insomnia symptoms. A potential explanation for these null results is that the intervention did not create sufficiently large reductions in work-family conflict to impact WASO or insomnia symptoms. It is also possible that work-family conflict is simply unrelated to these metrics of sleep quality; however, a number of prior studies show positive relationships between work-family conflict and sleep problems.
The intervention's main effect of 8 minutes more sleep per day (or ~ 1 hour more sleep per week) relative to the control condition is socially important and may have clinically significant health benefits over time. The clinical health benefits of an extra hour of sleep per week are not clearly known, but an extra hour is important at the population level, especially because work-related factors remain important predictors of sleep even after controlling for home-based factors.
CDC analyses of 2010 National Health Interview Survey data revealed that 30.0% of employed US civilian adults report short sleep duration (≤6 hours/day), with significant variation by sector.
The full mediation of the intervention effect on sleep sufficiency via reductions in control over work hours and work-family conflict suggests that we should also expect other generalized benefits of reduced work stress for employees and employers, such as improved health and reduced absenteeism/turnover. If a collectively administered workplace intervention, such as the group-level WFHN intervention, was enhanced with an individual-level sleep intervention component at work or elsewhere, even larger effects might be anticipated. Indeed, CDC/National Institute for Occupational Safety and Health (NIOSH) has also identified the workplace as a locus of sleep interventions, concluding that “further explorations of the relationship between work and sleep are needed.”
We believe that these findings would generalize to at least other professional, knowledge work environments where employees could be given greater control over work time. As described in detail elsewhere, stressors in the environment included a project-oriented and deadline-driven work culture including coordination with coworkers overseas and job insecurity common in the US workforce with occasional “right-sizing” of the workforce and trends for offshoring of some functions.
In terms of work hours, we compared our sample (nonsupervisory employees who participated in the baseline survey) with IT employees as well as other professional employees using the Current Population Survey.
Employees put in slightly longer hours on the job (45 hours per week) than the average national average IT workers (43 hours per week) and those in other professional jobs (43 hours per week). Approximately 29% work at least 50 hours per week, whereas the corresponding figure is 16% for IT employees and 19% for other professional employees at the national level. Future studies are needed to replicate and expand upon these findings in other contexts.
Strengths/limitations
Primary strengths of the study include the randomized controlled design, actigraphy-measured sleep outcomes, 1-year follow-up, qualitative and quantitative data, and a focus on working adults as the target population. Additional strengths include mediation analyses involving a target of the intervention collected at a separate, earlier time point from outcomes. Limitations include using only 2 weeks of actigraphy per participant (minimum 3 valid days for each week-long recording) and the IT worker sample not representative of all workforces. Although actigraphy is superior to self-report data, the methodology has limitations and measurement error, including a particular weakness for detecting wakefulness during sleep.
Innovative aspects of study included the randomized and controlled design, intervening at the workplace to improve individual health outcomes and intervening at multiple levels (employee and manager) and over time to produce true cultural change. In our zeal to maintain a “pure” work-family focused workplace intervention, we did not also specifically intervene on individual health behaviors that were primary a priori outcomes. Future studies should, of course, take this combined approach and intervene at both organizational and individual levels, and we recognize this as a limitation of the study.
Conclusion
The current study, a randomized experimental evaluation of a work-family intervention, did not overtly address participant sleep as a part of the intervention process. Yet, after organizational and policy changes at the manager and team levels, employees' sleep time increased by an hour per week, as did perceptions of sleep sufficiency. Improvements in sleep sufficiency were fully mediated by reductions in control over work hours and work-family conflict, both targets of the intervention. Increasing family supportive supervision and employee control over work may reduce stress and improve sleep. Workplace interventions may be a particularly effective approach with sufficiently broad reach for achieving the Healthy People 2020 goal to increase the number of adults who are getting sufficient sleep (US baseline, 69.6%; 2020 goal, 70.9%
). Work, although a source of financial security and fulfillment, has modifiable facets that may hinder a healthful lifestyle. In the current study, we modified work and improved measured sleep duration and perceived sleep sufficiency, strongly suggesting workplaces as foci for future sleep interventions.
Support
This research was conducted as part of the WFHN (www.WorkFamilyHealthNetwork.org), which is funded by a cooperative agreement through the National Institutes of Health and the Centers for Disease Control and Prevention: Eunice Kennedy Shriver National Institute of Child Health and Human Development (U01HD051217, U01HD051218, U01HD051256, and U01HD051276); National Institute on Aging (U01AG027669); Office of Behavioral and Social Sciences Research and National Institute for Occupational Safety and Health (U01OH008788 and U01HD059773). Grants from the National Heart, Lung and Blood Institute (R01HL107240), William T. Grant Foundation, Alfred P Sloan Foundation, and the Administration for Children and Families have provided additional funding. The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of these institutes and offices.
Disclosures
Buxton (last 3 years): OMB reports research grant support from the National Institutes of Health; outside of the submitted work, investigator-initiated research grant support from Sepracor (now Sunovion) and Cephalon (now Teva); personal fees from Takeda Pharmaceuticals North America, Dinsmore LLC (expert witness testimony), Matsutani America (scientific advisory board), Wake Forest University Medical Center (consulting); travel support and/or honoraria for speaking from American Academy of Craniofacial Pain; National Institute of Heart, Lung and Blood Institute; National Institute of Diabetes and Digestive and Kidney Diseases; National Postdoctoral Association; Oklahoma State University; Oregon Health Sciences University; SUNY Downstate Medical Center; American Diabetes Association; New York University; Academy of Nutrition and Dietetics; and Harvard T.H. Chan School of Public Health. All other authos have no conflicts of interest or disclosures to report.
Acknowledgments
Special acknowledgement goes to Extramural Staff Science Collaborator, Rosalind Berkowitz King, PhD, and Lynne Casper, PhD, for design of the original Workplace, Family, Health and Well-Being Network Initiative (responsible for generating and encompassing the WFHN). We wish to express our gratitude to the worksites, employers, and employees who participated in this research and made this publication possible. The National Institutes of Health, CDC, and other funders did not play a directive role in study design, conduct, reporting, or decision to submit a manuscript. We thank the members of the Biomarker and Actigraphy Data Coordinating Center, James H. Porter, Deirdre T. McLaren, Richard M. Lilienthal, and Chloe M. Beverly for assistance with data collection, scoring, and preparation of this manuscript and Rachel Magennis and Yagmur Karayaka for assistance with qualitative data.
Strategic opportunities in sleep and circadian research: report of the Joint Task Force of the Sleep Research Society and American Academy of Sleep Medicine.
Bray JW, Kelly E, Hammer LB, Almeida DM, Dearing JW, Berkowitz King R, et al. An integrative, multilevel, and transdisciplinary research approach to challenges of work, family, and health. Research Triangle Park, NC 2013. RTI Press publication No. MR-0024-1303.
Is work-family conflict a multi-level stressor linking job conditions to mental health? Evidence from the work, family and health network.
in: Ammons S.K. Kelly E.L. Research in the Sociology of Work: Work and Family in the New Economy. Emerald Publishing Group,
Bradford, U.K.2015 (Forthcoming)
P < .05.
00004-7&id=gr1.jpg){kind=link}
00004-7&id=gr2.jpg){kind=link}
00004-7&id=gr3.jpg){kind=link}
00004-7&id=gr4.jpg){kind=link}
00004-7&id=gr5.jpg){kind=link}