Association of Behavioral Phenotypes With Changes in Sleep During a Workplace Wellness Program

Abstract

Purpose:

Examine changes in sleep duration by 3 behavioral phenotypes during a workplace wellness program with overweight and obese adults.

Design:

Secondary analysis of a randomized clinical trial

Setting:

Remotely monitored intervention conducted across the United States

Subjects:

553 participants with a body mass index ≥25

Intervention:

Participants were randomized to 1 of 4 study arms: control, gamification with support, gamification with collaboration, and gamification with competition to increase their physical activity. All participants were issued a wrist-worn wearable device to record their daily physical activity and sleep duration.

Measures:

The primary outcome was change in daily sleep duration from baseline during the 24 week intervention and follow-up period by study arm within behavioral phenotype class.

Analysis:

Linear mixed effects regression.

Results:

Participants who had a phenotype of less physically active and less social at baseline, in the gamification with collaboration arm, significantly increased their sleep duration during the intervention period (30.2 minutes [95% CI 6.9, 53.5], P = 0.01), compared to the control arm. There were no changes in sleep duration among participants who were more extroverted and motivated or participants who were less motivated and at-risk.

Conclusions:

Changes in sleep during a physical activity intervention varied by behavioral phenotype. Behavioral phenotypes may help to precisely identify who is likely to improve sleep duration during a physical activity intervention.

Keywords

sleep physical activity wearable devices gamification phenotype obesity workplace wellness

Introduction

An estimated 35-40% of adults in the United States report difficulties falling or staying asleep in addition to increased daytime sleepiness.¹ Insufficient sleep is associated with elevated rates of several chronic health conditions, including obesity and cardiovascular disease^2-5 and results in billions of direct and indirect medical and societal costs annually.^6-8 Additionally, insufficient sleep negatively impacts employee performance resulting in lost worker productivity, increased employer healthcare costs, and slowed cognitive processing throughout the work day.^7,9,10 The increasing prevalence of sleep disorders and inadequate sleep and its contribution to adverse workplace outcomes has led to a growing emphasis on workplace wellness programs designed, either directly or indirectly, to improve sleep.^9,11 Physical activity is generally considered to have positive effects on sleep and is a common target of workplace wellness programs.^12-15

Among previous workplace wellness programs that examined the impact of a physical activity intervention on sleep, few reported slightly improved sleep quality^16,17 while others reported no improvements in sleep quality¹⁰ or sleep duration.¹⁸ In the general population, previous studies reported an increase in sleep quality,^19-25 sleep efficiency,²⁵ and sleep duration^22,25 while others reported sleep improvements that were limited to participant subgroups^26,27 or reported no change in sleep²⁸ following a physical activity intervention. These mixed results emphasize the complex relationship between physical activity and sleep and suggests there may be subgroups of individuals who stand to benefit the most from these interventions.

Novel segmentation methods that combine demographic, behavioral, and psychological characteristics to cluster individuals can help explain who may be more likely to respond to interventions and lead to more individualized, targeted programs for improving complex behaviors such as physical activity and sleep.^29-32 Segmenting individuals into behavioral phenotypes using a combination of sociodemographic, behavioral, and psychological variables has been previously used to identify phenotypes that help explain healthcare utilization, health outcomes, and heterogeneity of intervention responsiveness.^29,30,32-34 For example, previous work segmented high-cost and high-need patients into distinct clinical phenotypes (e.g. end-stage renal disease, multiple comorbidities) that better explained health utilization and survival outcomes over time. This work directly informs the design of future medical care interventions that are uniquely tailored to each clinical phenotype.²⁹ Additionally, recent work segmented physicians into 2 unique phenotypes, high clinical workload and low clinical workload, that helped explain responsiveness to an intervention to improve flu vaccination rates.³⁴ These methods are emerging in the health behavior field and provide a novel, scientifically rigorous approach to identifying sub-groups of individuals that may be more likely to respond to interventions designed to improve health behaviors such as physical activity and sleep.

The Social Incentives to Encourage Physical Activity and Understand Predictors (STEP UP) clinical trial tested the effectiveness of gamification with different social incentives (competition, collaboration, and support) to improve physical activity among overweight or obese adults at a large consulting firm.³⁵ Gamification, the use of game design elements in non-game contexts such as points and levels, is an increasingly popular intervention to motivate behavior change.^35-38 A recent secondary analysis from the STEP UP trial clustered participants according to baseline behavioral, psychological, and demographic characteristics using latent class analysis to identify participant phenotypes and examine responsiveness to a physical activity intervention.³¹ While the gamification with competition arm had the largest, significant increase in daily steps in the primary trial,³⁵ intervention responsiveness varied between the 3 behavioral phenotype classes. For participants who were extroverted and motivated (Class 1), the gamification with competition arm was the only study arm to significantly increase their daily steps during the intervention period but this improvement was not sustained during the follow-up period. For participants who were less active and less social (Class 2), all gamification arms significantly increased their daily steps during the intervention and these changes were sustained during the follow-up period. There were no changes in physical activity for participants who were less motivated and at-risk (Class 3).³¹ This previous study focused exclusively on physical activity outcomes and an important next step is to extend this work to other health domains, such as sleep, to better understand how intervention responsiveness varies across other complex health behaviors. Therefore, the purpose of this analysis was to examine changes in daily sleep duration during a workplace wellness physical activity intervention by these established behavioral phenotype classes.

Methods

This was a secondary analysis from the STEP UP clinical trial.^35,39 Participants were recruited from across the United States in collaboration with Deloitte Consulting from February 2018 to March 2019. Full details of the clinical trial design have been previously reported.^35,39 Individuals were enrolled in the trial if they were at least 18 years of age, had a body mass index (BMI) of ≥25, owned a smartphone or tablet that was compatible with the Withings/Nokia Steel activity tracker application, and were able to provide informed consent. Individuals were excluded from the trial if they had a medical condition that made participation unsafe or compromised their ability to complete the full 36 week study, were currently enrolled in another physical activity program, or unable to provide informed consent. This study was approved by the Institutional Review Board at the University of Pennsylvania and all participants provided informed consent.

Briefly, this was a 4-arm randomized controlled trial testing the effectiveness of gamification with social incentives (support, collaboration, and competition) for increasing physical activity among a national sample of overweight or obese adults. Participants in the gamification with support arm identified a support partner who received weekly updates about the participant’s performance and provided encouragement to meet their activity goal throughout the duration of the study. Those in the gamification with collaboration arm were placed in groups of 3, and each day a different participant from the group was randomly selected to represent the team. Lastly, participants in the gamification with competition arm were also placed in groups of 3, but used a weekly leaderboard to create a sense of competition within the group for achieving daily activity goals. The control arm received daily feedback on their step goal but no additional intervention.

This 36 week study included a 2 week run-in period, 24 week intervention period, and 12 week follow-up period. The study was conducted using Way to Health, a research technology platform at the University of Pennsylvania and previously used for remote monitoring and physical activity studies.^40-42 Participants completed enrollment and received all study related communications via the Way to Health platform. Eligible participants were mailed a wrist-worn wearable activity tracker (Withings Activite Steel) and asked to wear the device during the day and at night for the entire 36 week study duration. Participants were provided with instructions on how to authorize their device and sync their daily step counts and total sleep minutes to the Way to Health platform. The reliability of commercially available wearable devices for quantifying total sleep time has been established.⁴³

Sleep Duration

Total daily sleep was defined as sleep that occurred between 12 pm to 12 pm the following day (24 hour period). All sleep minutes were summed during this 24-hour period to yield a daily sleep value. All participants completed the 2 week run-in period, which provided an opportunity to become comfortable with wearing the device. We used data from the second week of the run-in period to calculate a baseline sleep value for all participants. The first week of data was ignored due to the potential upward bias or atypical sleep patterns that may result from wearing the device. Consistent with the primary trial analysis,³⁵ each participants’ baseline sleep value was calculated using data from the full second week of the baseline period (7 days). If a participant did not have 7 days of sleep data, a minimum of 4 days of valid sleep data were required to calculate a baseline sleep value.

Behavioral Phenotyping

A previous study³¹ clustered participants, using latent class analysis (LCA), from the STEP UP clinical trial using validated baseline sociodemographic (e.g. age, gender), behavioral (baseline physical activity and sleep quality), and psychological variables (Big 5 Personality Inventory,⁴⁴ Exercise Self-Efficacy Scale,⁴⁵ Medical Outcomes Social Support Survey,⁴⁶ Domain Specific Risk-Taking Scale,⁴⁷ and the Grit Scale⁴⁸) that are relevant to behavior change or associated with physical activity.^49,50 Latent class analysis requires categorical variables, therefore all continuous variables were converted to discrete categorical variables for the analysis. All variables were used to create a series of latent class models that were then evaluated for their fit using an iterative approach.³¹ To better understand which variables were most influential in differentiating classes, probability weights were generated for each variable used in the LCA to calculate the proportion of participants in each variable category within each class, compared to the overall sample.³¹ Variables with a probability weight of 1.33 or higher were considered drivers of class distinction and were labeled as defining characteristics.³¹

A total of 3 participant classes emerged and were named according to their defining characteristics: extroverted and motivated (Class 1), less physically active and less social (Class 2), and less motivated and at-risk (Class 3). The defining characteristic of less social for Class 2 is a comprised of both lower social support and social risk-taking. Participants who were in class 1 were also, on average, older, had higher grit, exercise self-efficacy, and extroversion. Those who were in class 2 had medium grit, lower exercise self-efficacy, and lower extroversion. Participants in class 3 were, on average, younger with lower grit, lower exercise self-efficacy, higher risk-taking and neuroticism.

Outcome Measure

The primary outcome for this analysis was change in daily sleep minutes from baseline through the main intervention period (weeks 5-24) by behavioral phenotype class. Weeks 1 to 4 of the intervention were considered a ramp-up period when participants in all gamification arms had their daily step goal gradually increased by 25% each week.³⁹ Consistent with the primary trial analysis,³⁵ the ramp-up period was excluded from the main analysis for this study.

Statistical Analysis

All participants with baseline sleep data and at least 1 sleep event during the intervention period were included in this analysis. Total sleep minutes were obtained for each participant on each study day as a continuous variable. Data were considered missing if the participant did not upload sleep data to the Way to Health platform, did not wear their device at night, or registered an errant number. We used multiple imputation for daily sleep minutes that were missing (46% missing rate). Five imputations were conducted in R⁵¹ using the mice package.⁵² The following covariates were used to impute the missing data: age, race/ethnicity, education level, marital status, study arm, participant week in the study, calendar month, baseline sleep, prior use of smartphone or wearable devices to track daily steps or sleep, experience with wireless technology to track activity levels, commute status, BMI, daily caffeine consumption, and baseline self-reported sleep quality (Pittsburgh Sleep Quality Index). All model results were pooled using Rubin’s standard rules⁵³ and sensitivity analyses were conducted using the collected data without multiple imputation.

We examined changes in total sleep duration by intervention arm, compared to the control arm, within each behavioral phenotype class for both the intervention and follow-up periods. Linear mixed effects regression models were fit with a participant random effect and calendar month, study arm, and baseline sleep as fixed effects. Lastly, we removed the behavioral phenotype classes and tested for change in sleep by study arm, compared to the control arm, using a linear mixed effects regression analysis with a participant random effect and calendar month, study arm, and baseline sleep as fixed effects.

All analyses were conducted in R⁵¹ and we used a conservative Bonferonni adjusted P value due to the multiple comparisons. The significance level was established at P < 0.017 for all analyses.

Results

Of the 602 participants in the primary trial, 553 had available data for this analysis. Participants were excluded from this analysis if they did not have baseline sleep data (n = 47) or did not have at least 1 recorded sleep observation during the intervention period (n = 2). Overall, participants had a mean (SD) age of 39 (10.3) years and mean (SD) BMI of 29.6 (4.9). Participants’ self-reported good overall sleep quality at baseline using the Pittsburgh Sleep Quality Index (4.8 (2.8) points, (Table 1).

Table 1.

Participant Demographics.

	Class 1 (n = 301)	Class 2 (n = 110)	Class 3 (n = 142)	P value	All classes (n = 553)
Sociodemographics
Age, years	40.4 (11.0)	38.0 (8.0)	36.5 (10.7)	<0.001	39.0 (10.3)
Male, No. (%)	208 (69.1)	89 (81.0)	93 (65.5)	0.02	390 (70.5)
Race/ethnicity, No. (%)
White non-Hispanic	171 (56.8)	39 (35.5)	80 (56.3)	<0.001	290 (52.4)
Black non-Hispanic	30 (10.0)	4 (3.6)	2 (1.4)		36 (6.5)
Asian non-Hispanic	82 (27.2)	57 (51.8)	48 (33.8)		187 (33.8)
Hispanic	12 (4.0)	4 (3.6)	9 (6.3)		25 (4.5)
Other	6 (2.0)	6 (5.5)	3 (2.1)		15 (2.7)
Self-Reported Measures
Depression (PHQ-9), No. (%)
Minimal or none	238 (79.1)	67 (61.0)	56 (39.4)	<0.001	361 (65.2)
Mild	51 (16.9)	38 (34.5)	59 (41.5)		148 (26.8)
Moderate	12 (4.0)	5 (4.5)	20 (14.1)		37 (6.7)
Moderately severe or severe	0 (0.0)	0 (0.0)	7 (5.0)		7 (1.3)
Prior wearable device use, No. (%)	206 (68.4)	54 (49.1)	87 (61.3)	0.001	347 (62.7)
Work and reside in same place, No. (%)	120 (39.9)	60 (54.5)	68 (47.9)	0.02	248 (44.8)
Daily caffeine intake, No. (%)
None	62 (20.6)	27 (24.5)	35 (24.6)	0.50	124 (22.4)
1-2 cups	157 (52.2)	55 (50.0)	67 (47.2)		279 (50.5)
3-4 cups	70 (23.3)	26 (23.7)	34 (23.9)		130 (23.5)
>5 cups	12 (4.0)	2 (1.8)	6 (4.2)		20 (3.6)
Baseline Measures
Baseline steps	6189 (2557)	5354 (2130)	6498 (3286)	0.002	6248 (2629)
Body mass index	30.0 (4.7)	28.5 (3.4)	29.8 (6.1)	0.01	29.6 (4.9)
Baseline Sleep Measures
Minutes of sleep	441 (63)	446 (67)	442 (72)	0.83	445 (64)
Sleep quality (Global PSQI score)	4.2 (2.7)	4.6 (2.4)	6.0 (2.9)	<0.001	4.8 (2.8)
Baseline Assessments
Extroversion^a	3.6 (0.8)	2.9 (0.7)	3.1 (0.8)	<0.001	3.3 (0.8)
Agreeableness^a	4.2 (0.6)	3.9 (0.4)	3.6 (0.6)	<0.001	4.0 (0.6)
Conscientousness^a	4.2 (0.5)	3.8 (0.4)	3.3 (0.6)	<0.001	3.9 (0.6)
Neuroticism^a	2.2 (0.6)	2.7 (0.6)	3.2 (0.7)	<0.001	2.5 (0.8)
Openness^a	3.8 (0.5)	3.3 (0.4)	3.7 (0.6)	<0.001	3.7 (0.6)
ESE Sticking to it^b	4.1 (0.7)	3.5 (0.6)	3.5 (0.7)	<0.001	3.8 (0.7)
MOS Social support^c	4.3 (0.8)	3.8 (0.8)	3.9 (0.9)	<0.001	4.1 (0.8)
DOSPERT health/safety^d	2.5 (1.0)	2.3 (0.8)	3.3 (1.1)	<0.001	2.6 (1.1)
DOSPERT Social^d	5.1 (1.0)	4.1 (0.8)	4.9 (0.9)	<0.001	4.8 (1.0)
Grit^e	4.0 (0.4)	3.6 (0.3)	3.0 (0.5)	<0.001	3.7 (0.6)

Values reported as mean (SD) unless otherwise indicated.

Abbreviations: PHQ-9 = Patient Health Questionnaire, 9-item; PSQI = Pittsburgh Sleep Quality Index; ESE = Exercise self-efficacy; MOS = Medical Outcomes Survey; DOSPERT = Domain Specific Risk-Taking scale.

^a Big 5 Personality Inventory, scores range from 0-5 where 5 = more extroversion, agreeableness, conscientiousness, neuroticism, and openness.

^b Exercise self-efficacy scale, scores range from 0-5 where 5 = more likely to stick to an exercise routine.

^c MOS Social Support survey, overall score ranges from 0-5 where 5 = more social support.

^d DOSPERT survey, scores range from 1-7 where 7 = more likely to engage in a risky behavior.

^e Grit Scale, scores range from 0-5 where 5 = more gritty.

The majority of participants were in phenotype Class 1 (n = 301), followed by Class 3 (n = 142) while Class 2 had the fewest number of participants (n = 110). Differences in baseline sleep were not significantly different between Class 1 (441 [63] minutes), Class 2 (446 [67] minutes), and Class 3 (442 [72] minutes). Baseline sleep quality was significantly different between Class 1 (4.2 [2.7] points), Class 2 (4.6 [2.4] points), and Class 3 (6.0 [2.9] points). As expected, there were significant differences between classes on the baseline sociodemographic, behavioral, and psychological assessments (Table 1). The characteristics within each behavioral phenotype class are listed in Table 2 and have also been previously reported.³¹

Table 2.

Characteristics of Each Behavioral Phenotype Class.

Behavioral phenotype class	Defining characteristics	Additional characteristics
Class 1 (n = 301)	More extroverted & motivated	Higher grit^a
		Higher exercise self-efficacy^b
		Higher extroversion^c
		Higher conscientiousness^c
		Higher openness^c
Class 2 (n = 110)	Less physically active & less social	Medium grit^a
		Lower exercise self-efficacy^b
		Lower extroversion^c
		Medium conscientiousness^c
		Lower openness^c
		Lower social risk-taking^d
		Lower social support^e
		Less prior wearable use
		Lower baseline steps
Class 3 (n = 142)	Less motivated & at-risk	Lower grit^a
		Lower exercise self-efficacy^b
		Lower extroversion^c
		Lower conscientiousness^c
		Higher neuroticism^c
		Higher health/safety risk-taking^d
		Lower social support^e
		Lower agreeableness^c
		Lower sleep quality^f
		Younger age

Key drivers of each behavioral phenotype class that have been previously reported.³¹ The corresponding assessments are as follows: ^aGrit scale⁴⁸, ^bExercise self-efficacy scale⁴⁵, ^cBig 5 Personality Inventory⁴⁴, ^dDomain Specific Risk-Taking scale (DOSPERT)⁴⁷, ^eMedical Outcomes Survey Social Support (MOS SS)⁴⁶, ^fPittsburgh Sleep Quality Index (PSQI).⁵⁴

Results from the adjusted analysis indicate that there was not a significant increase in daily sleep duration for any study arm within Class 1 (participants who were more extroverted and motivated) during the intervention or follow-up period (Table 3). Within Class 2 (participants who were less active and less social), there was a significant increase in sleep duration among participants in the gamification with collaboration arm, compared to the control arm, during the intervention period (30.2 minutes [6.9, 53.5]; P = 0.01) that continued through the follow-up period, although not statistically significant due to the adjusted P value (28.8 minutes [95% CI 4.1, 53.5], P = 0.02, (Table 4). Similar to Class 1, there were no observed changes in sleep duration for any study arm within Class 3 (participants who were less motivated and at-risk, (Table 5).

Table 3.

Adjusted Differences in Sleep Duration Within Phenotype Class 1, by Study Arm.

	Control (n = 65)	Gamification with support (n = 72)	Gamification with collaboration (n = 82)	Gamification with competition (n = 82)
Baseline
Sleep per day, mean (SD)	443 (53)	435 (72)	448 (65)	441 (59)
Main intervention period^a
Sleep per day, mean (SD)	430 (112)	415 (121)	433 (121)	422 (114)
Main adjusted model^b
Difference relative to control and adjusted for baseline (95% CI)		-12.9 (-28.8, 3.0)	-0.64 (-18.0, 16.8)	-8.5 (-24.0, 6.9)
P value		0.12	0.94	0.28
Follow-up period^c
Sleep per day, mean (SD)	430 (115)	414 (124)	434 (125)	419 (117)
Main adjusted model
Difference relative to control and adjusted for baseline (95% CI)		-12.5 (-29.9, 4.8)	0.14 (-18.5, 18.8)	-10.6 (-28.6, 7.4)
P value		0.16	0.99	0.25

Class 1 includes participants who were more extroverted and motivated.

^a Main intervention period included weeks 5 to 24 and excluded the ramp-up phase.

^b Main adjusted model adjusted for baseline sleep, calendar month, and study arm.

^c The follow-up period included weeks 25 to 36.

Table 4.

Adjusted Differences in Sleep Duration Within Class 2, by Study Arm.

	Control (n = 31)	Gamification with support (n = 28)	Gamification with collaboration (n = 24)	Gamification with competition (n = 27)
Baseline
Sleep per day, mean (SD)	453 (58)	439 (80)	437 (63)	454 (62)
Main intervention period^a
Sleep per day, mean (SD)	413 (122)	423 (120)	431 (106)	420 (121)
Main adjusted model^b
Difference relative to control and adjusted for baseline (95% CI)		17.2 (-12.4, 46.8)	30.2 (6.9, 53.5)	5.7 (-20.0, 31.3)
P value		0.25	0.01	0.32
Follow-up period^c
Sleep per day, mean (SD)	414 (129)	418 (123)	433 (105)	419 (122)
Main adjusted model
Difference relative to control and adjusted for baseline (95% CI)		12.0 (-19.8, 43.8)	28.8 (4.1, 53.5)	4.7 (-21.9, 31.4)
P value		0.46	0.02	0.73

Class 2 includes participants who were less active and less social at baseline.

^a Main intervention period included weeks 5 to 24 and excluded the ramp-up phase.

^b Main adjusted model adjusted for baseline sleep, calendar month, and study arm.

^c The follow-up period included weeks 25 to 36.

Table 5.

Adjusted Differences in Sleep Duration Within Class 3, by Study Arm.

	Control (n = 43)	Gamification withsupport (n = 36)	Gamification with collaboration (n = 34)	Gamification with competition (n = 29)
Baseline
Sleep per day, mean (SD)	448 (67)	431 (84)	446 (74)	443 (57)
Main intervention period^a
Sleep per day, mean (SD)	430 (125)	418 (124)	435 (117)	431 (115)
Main adjusted model^b
Difference relative to control and adjusted for baseline (95% CI)		-5.4 (-27.5, 16.8)	4.2 (-16.4, 24.8)	4.0 (-17.6, 25.5)
P value		0.63	0.69	0.72
Follow-up period^c
Sleep per day, mean (SD)	431 (127)	419 (127)	438 (123)	435 (111)
Main adjusted model
Difference relative to control and adjusted for baseline (95% CI)		-6.5 (-31.8, 18.8)	6.0 (-18.6, 30.4)	4.8 (-20.1, 29.7)
P value		0.61	0.63	0.71

Class 3 includes participants who were less active and less social at baseline.

^a Main intervention period included weeks 5 to 24 and excluded the ramp-up phase.

^b Main adjusted model adjusted for baseline sleep, calendar month, and study arm.

^c The follow-up period included weeks 25 to 36.

Results of the impact of gamification alone, without behavioral phenotype class, indicate there were no changes in sleep duration in any gamification arm compared to the control arm during the intervention period (gamification with support, -4.2 minutes [95% CI-16.2, 7.7]; P = 0.48, gamification with collaboration, 6.8 minutes [95% CI-4.9, 18.5]; P = 0.25, gamification with competition,-2.0 minutes [95% CI-13.3, 9.2]; P = 0.72) or the follow-up period (gamification with support, -5.5 minutes [95% CI-18.7, 7.7]; P = 0.42, gamification with collaboration, 6.9 minutes [95% CI-6.2, 20.1]; P = 0.29, gamification with competition (-3.9 [95% CI-16.8, 9.0]; P = 0.58).

Discussion

In this sample of 553 overweight or obese adults from across the United States, we found that participants in the gamification with collaboration arm, who were less physically active and less social at baseline (phenotype Class 2), experienced a significant increase in sleep duration during a physical activity intervention that was mostly sustained, although no longer significant, during the follow-up period. There were no changes in sleep duration within phenotype Class 1 (extroverted and motivated) or phenotype Class 3 (less motivated and at-risk). To our knowledge, this is the first study to examine changes in sleep duration during a workplace physical activity intervention by behavioral phenotype class. Additionally, after removing behavioral phenotype class and testing for changes in sleep by gamification arm, we found no change in sleep duration during the intervention or follow-up periods, compared to the control group. Despite all gamification arms experiencing a significant increase in daily steps,³⁵ these improvements did not carry over to sleep. Together, these findings suggest that the relationship between physical activity and sleep is complex, but individual behavioral, demographic, and psychosocial characteristics may help explain who is most likely to experience improvements in sleep duration during a physical activity intervention.

Participants in the gamification with collaboration arm, who were less active and less social at baseline (Class 2), significantly increased their sleep duration during the main intervention period. These increases were mostly sustained during the follow-up period. These participants, compared to those in Class 1 and 3, were less active at baseline and experienced, on average, an 1119 step increase during the intervention period, compared to the control group.³¹ This step increase was not significantly different from the other gamification arms within Class 2 but future work may want to further explore characteristics of the participants who experienced a significant increase in sleep to better understand these changes. It is possible that individuals who are less active at baseline may be more likely to experience increased sleep duration during a physical activity intervention. The growing body of research examining variations in intervention responsiveness^30-32 and healthcare utilization²⁹ by behavioral phenotype contributes to the science of personalized medicine and directly informs how future interventions and medical services can be tailored to the unique needs of different phenotypes. Indeed, individual phenotypes afford the opportunity to prospectively randomize participants to an intervention they may be most likely to benefit from, reducing heterogeneity of responses and leading to a more targeted approach. Future studies may consider how baseline sociodemographic, behavioral, and psychological variables could be leveraged to prospectively randomize participants to interventions that may be most appropriate for their phenotype.

Disordered or limitations with sleep are common among individuals who are overweight or obese.⁵⁵ To date, few studies²³ have examined the impact of a physical activity intervention on sleep duration for adults who are overweight or obese. There are several possible explanations for why the majority of participants did not experience a significant increase in sleep duration. First, there may be a dose-response relationship between physical activity and sleep,¹² and perhaps our sample did not achieve the necessary threshold of improved physical activity to induce improved sleep duration. Additionally, there may be other factors that influenced sleep duration. This cohort was relatively young, with nearly half of all participants regularly commuting to a different city for work. The nature of work in a large consulting firm could impact stress or anxiety levels which may influence sleep. The relationship between physical activity and sleep may be mediated by these factors and others, and future work may want to further explore these potential relationships.

This study quantified both physical activity and sleep duration using a commercially available wearable device, decreasing the threat of recall and social desirability bias that are well-documented with self-reported outcomes.^56,57 Commercially available wearable devices are an emerging technology for sleep tracking and currently lack the capacity to reliably quantify all sleep domains. It is possible that participants in this study improved in other sleep domains, such as sleep quality, sleep latency, or sleep efficiency but the wearable device was unable to quantify this change. Undoubtedly, the reliability of wearable devices for quantifying other sleep domains will continue to improve, affording new opportunities to examine the relationship between physical activity and other sleep domains using wearable devices.

There are several limitations that influence the interpretation of these findings. Overall, this was a younger, college-educated cohort with a higher socioeconomic status. This limits generalizability to other clinical populations. This cohort was comprised of overweight and obese adults, who make up 70% of the United States’ population,⁵⁸ but limits the generalizability of these findings to individuals of other weight categories. Additionally, we did not have the ability to examine other forms of exercise on sleep duration in this cohort due to device limitations. It is possible that some participants increased their moderate to vigorous physical activity, which may impact sleep duration. Future work may want to examine the relationship between different forms of exercise and sleep duration. Lastly, previous work has hypothesized that physical activity interventions may need to exceed 16 weeks to improve sleep.¹² While the 24 week intervention period for this study exceeded this threshold, it is possible that 24 weeks was insufficient for improving sleep duration in this cohort. A longer intervention period may be necessary to induce longer-term changes in sleep.

In conclusion, we found an association between a small subset of individuals, who were overweight or obese, less active and social at baseline, and changes in sleep duration. The relationship between physical activity and sleep is complex but baseline sociodemographic, behavioral, and psychological factors may aide in identifying subgroups of participants who are most likely to experience improved sleep from physical activity interventions. Behavioral phenotypes may help to precisely identify who is most likely to benefit from various interventions to improve overall health and offer a promising approach to individualizing care for clinical populations.

So What?

What is already known on this topic?

Workplace wellness programs are designed to improve employee health behaviors such as physical activity and nutrition. In recent years, workplace wellness programs have examined the impact of physical activity interventions on sleep. Changes in physical activity and sleep by behavioral phenotype are unknown.

What does this article add?

This study examined changes in sleep duration by behavioral phenotype during a physical activity intervention. Novel segmentation methods that use baseline sociodemographic, behavioral, and psychological variables can help identify groups of participants who may be most likely to benefit from interventions to improve complex health behaviors.

What are the implications for health promotion practice or research?

Using behavioral phenotypes to examine the impact of workplace wellness programs is a promising approach to better understand intervention responsiveness.

Footnotes

Authors’ Note

Patel reports personal fees and other from Catalyst Health, personal fees and other from Healthmine Services Inc., other from LifeVest Health, personal fees from Holistic Industries, outside the submitted work. Dr. Godby and Mr. Szwartz are employed by Deloitte Consulting, LLP. No other disclosures were reported.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by Deloitte Consulting LLP and the University of Pennsylvania Health System through the Penn Medicine Nudge Unit. Dr. Waddell was supported by the Department of Veterans Affairs Advanced Fellowship Program in HSR&D. Dr. Patel was supported by a career development award from the Department of Veterans Affairs HSR&D.

ORCID iD

Kimberly J. Waddell, PhD, MS

References

Hossain

Shapiro

. The prevalence, cost implications, and management of sleep disorders: an overview. Sleep Breath. 2002;6(2):85–102.

National Center for Chronic Disease Prevention and Health Promotion. Data and Statistics: Short Sleep Duration Among US Adults. Centers for Disease Control. Published 2017. Accessed September 11, 2020. https://www.cdc.gov/sleep/data_statistics.html#:∼:text=Short%20Sleep%20Duration%20Among%20US,sleep%20per%2024%2Dhour%20period

Beccuti

Pannain

. Sleep and obesity. Curr Opin Clin Nutr Metab Care. 2011;14(4):402–412. doi:10.1097/MCO.0b013e3283479109

Loprinzi

Cardinal

. Association between objectively-measured physical activity and sleep, NHANES 2005–2006. Mental Health Physical Activity. 2011;4(2):65–69.

Zhou

Zhang

. Dose-response association between sleep duration and obesity risk: a systematic review and meta-analysis of prospective cohort studies. Sleep Breath. 2019;23(4):1035–1045.

National Academies of Sciences, Engineering and Medicine. Commercial Motor Vehicle Driver Fatigue, Long-Term Health, and Highway Safety: Research Needs. National Academies Press; 2016.

Kessler

Berglund

Coulouvrat

, et al. Insomnia and the performance of US workers: results from the America Insomnia Survey. Sleep. 2011;34(9):1161–1171.

Altevogt

Colten

. Sleep Disorders and Sleep Deprivation: An Unmet Public Health Problem. National Academies Press; 2006.

Robbins

Jackson

Underwood

Vieira

Jean-Louis

Buxton

. Employee sleep and workplace health promotion: a systematic review. Am J Health Promot. 2019;33(7):1009–1019.

10.

Song

Baicker

. Effect of a workplace wellness program on employee health and economic outcomes: a randomized clinical trial. JAMA. 2019;321(15):1491–1501.

11.

Redeker

Caruso

Hashmi

Mullington

Grandner

Morgenthaler

. Workplace interventions to promote sleep health and an alert, healthy workforce. J Clin Sleep Med. 2019;15(4):649–657.

12.

Buman

King

. Exercise as a treatment to enhance sleep. Am J lifestyle Med. 2010;4(6):500–514.

13.

Kredlow

Capozzoli

Hearon

Calkins

Otto

. The effects of physical activity on sleep: a meta-analytic review. J Behav Med. 2015;38(3):427–449. doi:10.1007/s10865-015-9617-6

14.

Youngstedt

. Effects of exercise on sleep. Clin Sports Med. 2005;24(2):355–365.

15.

Mattke

Schnyer

Van Busum

. A Review of the U.S. Workplace Wellness Market. RAND Corporation. Published 2012. Acessed November 17, 2020. https://www.rand.org/pubs/occasional_papers/OP373.html

16.

Mackey

Foley

, et al. Reducing office workers’ sitting time at work using sit-stand protocols: results from a pilot randomized controlled trial. J Occup Environ Med. 2017;59(6):543–549.

17.

Bouchard

Strachan

Johnson

, et al. Using shared treadmill workstations to promote less time spent in daily low intensity physical activities: a pilot study. J Phys Act Health. 2016;13(1):111–118.

18.

Hallman

Holtermann

Søgaard

Krustrup

Kristiansen

Korshøj

Effect of an aerobic exercise intervention on cardiac autonomic regulation: a worksite RCT among cleaners. Physiol Behav. 2017;169:90–97.

19.

Bisson

ANS

Robinson

Lachman

. Walk to a better night of sleep: testing the relationship between physical activity and sleep. Sleep Health. 2019;5(5):487–494.

20.

King

Oman

Brassington

Bliwise

Haskell

. Moderate-intensity exercise and self-rated quality of sleep in older adults: a randomized controlled trial. JAMA. 1997;277(1):32–37.

21.

King

Baumann

O’Sullivan

Wilcox

Castro

. Effects of moderate-intensity exercise on physiological, behavioral, and emotional responses to family caregiving: a randomized controlled trial. J Gerontol Series A Biol Sci Med Sci. 2002;57(1):M26–M36.

22.

Baron

Reid

Zee

. Exercise to improve sleep in insomnia: exploration of the bidirectional effects. J Clin Sleep Med. 2013;9(8):819–824.

23.

Tan

Alén

Wiklund

Partinen

Cheng

. Effects of aerobic exercise on home-based sleep among overweight and obese men with chronic insomnia symptoms: a randomized controlled trial. Sleep Med. 2016;25:113–121.

24.

Durcan

Wilson

Cunnane

. The effect of exercise on sleep and fatigue in rheumatoid arthritis: a randomized controlled study. J Rheumatol. 2014;41(10):1966–1973.

25.

Fisher

Harmer

Irbe

Tearse

Weimer

. Tai Chi and self-rated quality of sleep and daytime sleepiness in older adults: a randomized controlled trial. J Am Geriat Soc. 2004;52(6):892–900.

26.

Irwin

Olmstead

Motivala

. Improving sleep quality in older adults with moderate sleep complaints: a randomized controlled trial of Tai Chi Chih. Sleep. 2008;31(7):1001–1008.

27.

Tworoger

Yasui

Vitiello

, et al. Effects of a yearlong moderate-intensity exercise and a stretching intervention on sleep quality in postmenopausal women. Sleep. 2003;26(7):830–836.

28.

Guilleminault

Clerk

Black

Labanowski

Pelayo

Claman

. Nondrug treatment trials in psychophysiologic insomnia. Arch Int Med. 1995;155(8):838–844.

29.

Davis

Shen

Shah

, et al. Segmentation of high-cost adults in an integrated healthcare system based on empirical clustering of acute and chronic conditions. J Gen Int Med. 2018;33(12):2171–2179.

30.

Calfee

Delucchi

Parsons

, et al. Subphenotypes in acute respiratory distress syndrome: latent class analysis of data from two randomised controlled trials. Lancet Respir Med. 2014;2(8):611–620.

31.

Chen

Changolkar

Navathe

, et al. Association between behavioral phenotypes and response to a physical activity intervention using gamification and social incentives: secondary analysis of the STEP UP randomized clinical trial. PloS One. 2020;15(10):e0239288. doi:https://doi.org/10.1371/journal.pone.0239288

32.

Lienert

Patel

. Patient phenotypes help explain variation in response to a social gamification weight loss intervention. Am J Health Promot. 2020;34(3):277–284.

33.

Keeney

Belanger

Jones

Joyce

Meyers

Mor

High-need phenotypes in Medicare beneficiaries: drivers of variation in utilization and outcomes. J Am Geriat Soc. 2020;68(1):70–77.

34.

Changolkar

Rewley

Balachandran

, et al. Phenotyping physician practice patterns and associations with response to a nudge in the electronic health record for influenza vaccination: a quasi-experimental study. Plos One. 2020;15(5):e0232895.

35.

Patel

Small

Harrison

, et al. Effectiveness of behaviorally designed gamification interventions with social incentives for increasing physical activity among overweight and obese adults across the United States: the STEP UP randomized clinical trial. JAMA Int Med. 2019;179(12):1624–1632. doi:10.1001/jamainternmed.2019.3505

36.

Edwards

Lumsden

Rivas

, et al. Gamification for health promotion: systematic review of behaviour change techniques in smartphone apps. BMJ Open. 2016;6(10):e012447.

37.

Kawachi

. It’s all in the game—the uses of gamification to motivate behavior change. JAMA Int Med. 2017;177(11):1593–1594.

38.

Kurtzman

Day

Small

, et al. Social incentives and gamification to promote weight loss: the LOSE IT randomized, controlled trial. J Gen Int Med. 2018;33(10):1669–1675.

39.

Harrison

Jones

Small

, et al. Social incentives to encourage physical activity and understand predictors (STEP UP): design and rationale of a randomized trial among overweight and obese adults across the United States. Contemp Clin Trials. 2019;80:55–60.

40.

Patel

Benjamin

Volpp

, et al. Effect of a game-based intervention designed to enhance social incentives to increase physical activity among families: the BE FIT randomized clinical trial. JAMA Int Med. 2017;177(11):1586–1593.

41.

Patel

Asch

Rosin

, et al. Framing social comparison feedback with financial incentives for physical activity promotion: a randomized trial. J Phys Act Health. 2020;17(6):641–649.

42.

Chokshi

Adusumalli

Small

, et al. Loss-framed financial incentives and personalized goal-setting to increase physical activity among ischemic heart disease patients using wearable devices: the active reward randomized trial. J Am Heart Assoc. 2018;7(12):e009173. doi:10.1161/JAHA.118.009173

43.

Gruwez

Libert

Ameye

Bruyneel

. Reliability of commercially available sleep and activity trackers with manual switch-to-sleep mode activation in free-living healthy individuals. Int J Med Informat. 2017;102:87–92.

44.

John

Srivastava

The big five trait taxonomy: history, measurement, and theoretical perspectives. Hand Personal Theory Res. 1999;2(1999):102–138.

45.

Sallis

Pinski

Grossman

Patterson

Nader

. The development of self-efficacy scales for healthrelated diet and exercise behaviors. Health Educat Res. 1988;3(3):283–292.

46.

Sherbourne

Stewart

. The MOS social support survey. Soc Sci Med. 1991;32(6):705–714.

47.

Blais

A-R

Weber

. A Domain-Specific Risk-Taking (DOSPERT) scale for adult populations. Judgm Decis Mak. 2006;1(1):33–47.

48.

Duckworth

Peterson

Matthews

Kelly

. Grit: perseverance and passion for long-term goals. J Pers soc psychol. 2007;92(6):1087.

49.

Rhodes

Smith

. Personality correlates of physical activity: a review and meta-analysis. Brit J Sports Med. 2006;40(12):958–965.

50.

Rhodes

Janssen

Bredin

Warburton

Bauman

Physical activity: health impact, prevalence, correlates and interventions. Psychol & Health. 2017;32(8):942–975.

51.

R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing. Published 2019. Accessed November 16, 2020. https://www.R-project.org/

52.

van Buuren

Groothuis-Oudshoorn

Mice: multivariate imputation by chained equations in r. J Statistl Soft. 2011;45(3):1–67.

53.

Rubin

. Multiple Imputation for Nonresponse in Surveys. NY Wiley. 1987.

54.

Buysse

Reynolds

III Monk

Berman

Kupfer

. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psych Res. 1989;28(2):193–213.

55.

Palm

Janson

Lindberg

. The impact of obesity and weight gain on development of sleep problems in a population-based sample. Sleep Med. 2015;16(5):593–597.

56.

Bradburn

Rips

Shevell

. Answering autobiographical questions: the impact of memory and inference on surveys. Science. 1987;236(4798):157–161.

57.

Adams

Matthews

Ebbeling

, et al. The effect of social desirability and social approval on self-reports of physical activity. Am J Epidemiol. 2005;161(4):389–398.

58.

Fryar

Carroll

Ogden

. Prevalence of Overweight, Obesity, and Extreme Obesity among Adults Aged 20 and Over: United States, 1960-1962 through 2011-2014. National Center for Health Statistics Data, Health E-Stats. Published 2016. Accessed January 13, 2020. https://www.cdc.gov/nchs/data/hestat/obesity_adult_13_14/obesity_adult_13_14.htm