Keywords
pregnancy - sleep - activity tracking device
Disordered and deficient sleep is common during pregnancy and linked with several
adverse outcomes,[1]
[2] yet sleep disorders are often dismissed as an expected physiological change in pregnancy.
Furthermore, measures of sleep during pregnancy are often limited to self-reported
surveys which may overestimate sleep duration compared with objective measures.[1]
[3] However, laboratory-based sleep studies such as polysomnography (PSG) may not reflect
free-living conditions. Commercially available activity tracking devices (ATD) have
the capability not only to measure activity including steps taken, heart rate, and
energy expenditure but also to measure sleep duration. However, prior studies of ATD
use in pregnancy have not evaluated sleep duration.[4]
[5]
[6]
The primary objective of this study was to describe sleep duration, as measured by
the main sleep event in a 24-hour period, across gestation in women who wore a commercially
available ATD during pregnancy. A secondary objective was to study the association
between sleep duration and adverse maternal and neonatal outcomes such as excessive
gestational weight gain, pregnancy associated hypertension, and birthweight. We hypothesized
that women with < 7 hours of sleep per 24 hours would have more frequent adverse outcomes
compared with those with ≥ 7 hours per 24 hours.
Materials and Methods
For this study, women who were enrolled in prenatal care at one of two sites (one
federally qualified health center and one academic medical center with an outpatient
clinic, both located in Chicago, IL) were approached during their prenatal visit and
asked to participate in a study about “activity monitoring devices and pregnancy.”
Other inclusion criteria were English or Spanish speaking, ≥18 years old, and personal
ownership of a smartphone. Exclusion criteria were restrictions or inability to exercise,
defined as at least 30 minutes of walking per day. The study period extended from
April 2016 to December 2017.
After written informed consent was obtained, participants chose a wrist Fitbit Flex
(San Francisco, CA) (i.e., the ATD) in their color preference. In-person instructions
were given on how to install the ATD app on their smartphone, charge and sync the
ATD, wear the ATD, and interpret the ATD data from the dashboard. Members of the research
team registered the participants' ATD online and created user accounts authorizing
access to the ATD data for the research personnel. The user accounts were available
in both English and Spanish according to the participants' preferences. No specific
instructions were given regarding sleep monitoring. Steps, sedentary hours, and total
sleep duration in hours were wirelessly transmitted via cellular and bluetooth technology
and plotted on a graph in the ATD app that participants could view on their device
or personal dashboard at all times (e.g., participants not blinded to data). Members
of the research team contacted the participant via text, phone calls, or email if
more than 72 hours lapsed since the ATD was synced. In-person visits with the research
team also occurred when these contact methods were not successful. Lost, stolen, or
broken ATDs were not replaced. The research team provided technical support for the
ATD throughout the pregnancy with text messages, phone calls, emails, or in-person
troubleshooting sessions.
The total sleep duration in hours was measured from the main sleep episode in a 24-hour
period. Other sleep episodes that were shorter in duration, which may have represented
naps, were not included in the total sleep duration measure. Also, we did not evaluate
when the main sleep episode occurred (e.g., daytime vs. nighttime). Similar to Xu
et al, we required that a minimum value to count a record as having sufficient sleep
data was more than 2 hours, but there was no upper limit on the maximum amount of
hours of sleep per day.[7] For primary analyses, we averaged all sleep times meeting these criteria by gestational
week for each participant.
We calculated descriptive statistics for all variables of interest. Categorical variables
were summarized with frequencies and percentages, and continuous variables were summarized
with means and standard deviations or medians and interquartile ranges, as appropriate.
Primary analyses utilized mixed effect models to examine the trajectory of mean weekly
hours of sleep by gestational age. Specifically, models included a fixed effect for
gestational age and a random subject effect to account for repeated measures. We evaluated
whether site should be included as a fixed effect, with the plan to retain it in the
multivariable models if it was significantly associated with sleep duration.
Insufficient sleep was defined as < 7 hours in a 24-hour period per the American Academy
of Sleep Medicine and Sleep Research Society.[8] Secondary analyses evaluated differences in pregnancy outcomes between insufficient
and sufficient sleep groups, based on mean hours of sleep within the first 7 days
of ATD use. Women were included in the secondary analyses if they had at least two
main sleep observations within the first 7 days. The means of the sleep times for
each possible day were calculated to determine if the sleep time was < 7 or ≥ 7 hours.
Sedentary minutes were defined as < 1.5 metabolic equivalents (METs), such as seated
activities, according to the manufacturer's descriptions. In the secondary analyses,
we assessed whether activity (daily steps and sedentary time) and pregnancy outcomes
(total gestational weight gain, gestational weight gain categories, gestational diabetes,
pregnancy associated hypertension, cesarean delivery, and birthweight) differed between
women with < 7 hours and women with ≥ 7 hours of sleep during the first week of ATD
use with either t-tests or Wilcoxon's rank sum tests for continuous predictors and Chi-square or Fisher's
exact tests as appropriate.
All analyses assumed a two-sided type one error rate of 0.05. Analyses were performed
with SAS, version 9.4. The study was approved was the Northwestern University and
the Erie Family Health Center Institutional Review Board.
Results
Of the 174 women approached at both clinical sites, 75 declined to participate and
7 were lost to follow-up, leaving 94 women with at least 1 day of ATD data available.
These 94 women had a median of 24.5 sleep observations meeting inclusion criteria.
The majority of the participants self-reported as belonging to a minority group (33%
non-Hispanic black and 51% Hispanic), had government insurance (83%), were nulliparas
(60%), and were enrolled in the second trimester (79%) ([Table 1]). Most women also were overweight or obese prior to pregnancy (56%), felt “very
comfortable using a computer or the internet” (84%), and rarely exercised every day
prior to pregnancy (9.6%).
Table 1
Maternal demographics and characteristics
|
Variable
|
Response
|
|
Age (y) (mean ± SD)
|
26.2 ± 5.2
|
|
Race-ethnicity, n (%)
|
|
Asian American
|
2 (2.1)
|
|
Black/African American
|
31 (33.0)
|
|
Hispanic/Latino
|
48 (51.1)
|
|
White or European American
|
4 (4.3)
|
|
Other/unknown
|
9 (9.6)
|
|
Education, n (%)
|
|
Grades 9–11
|
3 (3.2)
|
|
High school graduate/GED
|
26 (27.7)
|
|
Some college/technical school
|
40 (42.6)
|
|
Four year college degree or more
|
14 (14.9)
|
|
Unknown
|
11 (11.7)
|
|
Health insurance, n (%)
|
|
Medicaid or Medicare
|
73 (77.7)
|
|
Private insurance
|
5 (5.3)
|
|
Other/unknown
|
16 (17.0)
|
|
Employed outside of the home for a salary, n (%)
|
|
Yes
|
45 (47.9)
|
|
No
|
38 (40.4)
|
|
Unknown
|
11 (11.7)
|
|
Marital status, n (%)
|
|
Married
|
22 (23.4)
|
|
Single
|
30 (31.9)
|
|
Living with partner, but not married
|
31 (33.0)
|
|
Unknown
|
11 (11.7)
|
|
Nullipara, n (%)
|
57 (60.6)
|
|
Gestational age at enrollment (wk) (mean ± SD)
|
17.5 ± 5.4
|
|
Trimester at enrollment, n (%)
|
|
First
|
14 (14.9)
|
|
Second
|
74 (78.7)
|
|
Third
|
6 (6.4)
|
|
Pre-pregnancy BMI kg/m2 (mean ± SD)
|
28.3 ± 7.4
|
|
Pre-pregnancy BMI, n (%)
|
|
Underweight: BMI < 18.5 kg/m2
|
12 (12.8)
|
|
Normal: 18.5 ≤ BMI < 25 kg/m2
|
29 (30.9)
|
|
Overweight: 25 ≤ BMI < 30 kg/m2
|
24 (25.5)
|
|
Obese: BMI ≥ 30 kg/m2
|
29 (30.9)
|
|
History of regular cigarette use, n (%)
|
|
Yes
|
12 (12.8)
|
|
No
|
73 (77.7)
|
|
Unknown
|
9 (9.6)
|
|
Self-reported daily internet use, n (%)
|
76 (80.9)
|
|
Self-reported “very comfortable” using a computer and/or the internet, n (%)
|
79 (84.0)
|
|
Type of smartphone owned, n (%)
|
|
iPhone
|
55 (58.5)
|
|
Android
|
29 (30.9)
|
|
Other/unknown
|
10 (10.6)
|
|
“Before pregnancy, how much did you exercise?,” n (%)
|
|
Not at all
|
16 (17.0)
|
|
Occasionally
|
27 (28.7)
|
|
Once a month
|
5 (5.3)
|
|
Once a week
|
9 (9.6)
|
|
More than 1 time a week
|
20 (21.3)
|
|
Everyday
|
9 (9.6)
|
|
Unknown
|
8 (8.5)
|
Abbreviations: BMI, body mass index; GED, general equivalency development; SD, standard
deviation.
The mean sleep duration was 7.2 ± 2.4 hours with a median of 7.3 and an interquartile
range (IQR) of 5.67 to 8.55 per 24 hours. [Fig. 1] shows the observed longitudinal trends in mean hours of sleep, averaged across women,
by gestational age. There was no significant difference in sleep duration by site
(p-value = 0.26), and thus, remaining analyses did not include site as a fixed effect.
Mixed model analyses showed that gestational age had a significant inverse association
with mean hours of sleep (β = − 0.02, 95% confidence interval [CI]: −0.04 to −0.01, and p-value < 0.001, [Fig. 2]). After adjustment for possible confounders such as age, race–ethnicity, employment
status, marital status and pre-pregnancy obesity, the association between gestational
age, and sleep duration remained significant (β = − 0.02, 95% CI: −0.04 to −0.01, and p-value < 0.001).
Fig. 1 Scatterplot of mean hours of sleep (across women) by gestational age, with locally
weighted scatterplot smoothing curve overlaid.
Fig. 2 Predicted hours of sleep versus gestational age in weeks with 95% confidence intervals
(unadjusted model).
A total of 11 (12%) of the original 94 participants were excluded from further analyses
of outcomes as they did not have at least two sleep records within the first 7 days
of Fitbit use. A total of 30 (36.1%) women had < 7 hours and 53 (63.9%) had ≥ 7 hours
of sleep during the first week of ATD use. Eligible women who had <7 hours of sleep
had greater mean number of steps compared with those who had ≥ 7 hours of sleep (median:
7,122; IQR: 5,167–8,338 vs. median: 5,005; IQR: 4,115–7,059; p-value < 0.01). Conversely, there were no significant differences in other activity
or pregnancy outcomes (sedentary time, median gestational weight gain, pregnancy associated
hypertension, gestational diabetes, median gestational age at delivery, cesarean delivery,
or mean birthweight; [Table 2]).
Table 2
Perinatal outcomes according to total sleep duration in the first 7 days of use
|
Variable
|
<7 hours of sleep
n = 30
|
≥ 7 hours of sleep
n = 53
|
p-Value
|
|
Daily steps (median, IQR)
|
7,122 (5,167–8,338)
|
5,005 (4,115–7,059)
|
<0.01
|
|
Sedentary time/day (h) (mean ± SD)
|
887 ± 157
|
859 ± 125
|
0.37
|
|
Total gestational weight gain (kg) (median, IQR)
|
n = 28
11.4 (9.2–14.6)
|
n = 46
14.1 (8.8–17.4)
|
0.40
|
|
Gestational weight gain categories, n (%)
|
|
Inadequate
|
9 (32.1)
|
8 (17.4)
|
0.34
|
|
Adequate
|
8 (28.6)
|
15 (32.6)
|
|
|
Excessive
|
11 (39.3)
|
23 (50.0)
|
|
|
Gestational diabetes, n (%)
|
2 (6.9)
|
1 (2.1)
|
0.55
|
|
Gestational hypertension or preeclampsia, n (%)
|
3 (10.3)
|
7 (14.9)
|
0.73
|
|
Gestational age at delivery (wk) (median, IQR)
|
n = 29
39.1 (37.7–39.4)
|
n = 48
39.3 (38.4–40.2)
|
0.09
|
|
Preterm delivery, n (%)
|
6 (20.0)
|
9 (17.0)
|
0.73
|
|
Cesarean delivery, n (%)
|
5 (17.2)
|
14 (29.2)
|
0.29
|
|
Birthweight (g) (mean ± SD)
|
n = 28
3,142 ± 580
|
n = 46
3,234 ± 427
|
0.44
|
Abbreviations: IQR, interquartile range; SD, standard deviation.
Data presented as median (IQR), mean ± SD or n %.
Discussion
In this study of ATD use in 94 women who were predominantly nulliparas and overweight
or obese prior to pregnancy, we found that the mean duration of the main sleep episode
in a 24-hour period was 7.2 ± 2.4 hours and decreased over gestational age. Our findings
are similar to studies that have used actigraphy during pregnancy. In an actigraphy-based
study of 80 low-income pregnant women, mean sleep duration recorded by actigraphy
was 6.8 hours.[3] In another study, the mean second-trimester sleep duration was 6.6 hours.[9] In the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be (nuMoM2b)
sleep activity substudy, 782 women wore a wrist activity monitor and completed a sleep
log for 7 consecutive days. Their median actigraphy-recorded sleep duration was 7.4 hours,
with 27.9% having a sleep duration <7 hours. Age, race-ethnicity, BMI, insurance,
and recent smoking history were significantly associated with sleep duration in multivariable
models.[10]
Self-reported sleep data were not available for the current study, but our findings
are also similar to data derived from sleep questionnaires or sleep logs. For example,
Facco et al evaluated 189 nulliparous women with several sleep questionnaires and
compared differences in baseline and third trimester sleep characteristics.[11] Mean sleep duration was significantly shorter in the third trimester compared with
baseline (7.0 ± 1.3 vs. 7.4 ± 1.2 hours, p-value < 0.001). Overall poor sleep quality, as defined by a Pittsburgh Sleep Quality
Index score greater than 5, also was more common as pregnancy progressed (39.0 vs.
53.5%, p-value = 0.001).
In nonpregnant populations, poor sleep quality and quantity is associated with significant
morbidities including obesity, diabetes, pregnancy associated hypertension, as well
as mortality.[12]
[13]
[14]
[15] We found no differences in maternal and neonatal outcomes between women with < 7
and ≥ 7 mean hours of sleep in the first week of ATD use, but we also realize the
low frequency of several outcomes such as hypertensive disorders of pregnancy and
gestational diabetes which limits the power to detect differences between the groups.
The gold standard for documenting sleep is PSG. The correlation between actigraphy
and PSG in measuring sleep duration is high (0.7–0.97), and the American Academy for
Sleep Medicine considers actigraphy to be a valid method to measure sleep patterns
in healthy adults.[8] The mechanism for how the Fitbit ATD measures activity and sleep is proprietary
and not available for a commercial user; however, most ATD use a three-axis microelectromechanical
systems (MEMS) accelerometer that measures acceleration caused by movement of the
accelerometer unit along three axes.[16] The Fitbit Flex device has been validated for measuring sleep against PSG in a healthy
adult population, with a correlation of r = 0.96 for total sleep time.[17] However, other studies have reported differences in sleep measures between PSG,
actigraphy, and Fitbits. In one study evaluating the accuracy of Fitbit in determining
various sleep parameters, Montgomery-Downs et al compared Fitbit against PSG and actigraphy
in 24 healthy adults (mean age = 26.1; range = 19–41 years) with no history or symptoms
of sleep disorders. Fitbit and actigraphy differed significantly from PSG and from
each other (p-value < 0.001; d = 1.6). Fitbit overestimated total sleep time by 67.1 ± 51.3 minutes
compared with PSG. Fitbit also overestimated total sleep time compared with actigraphy
by 24.1 minutes.[18] In another review of consumer tracking devices, Fitbit was good at detecting sleep
but poor at detecting wake and tended to overestimate total sleep time.[16] In summary, the current studies of sleep measures and validity of different instruments
have small numbers of participants and often use a single night of recording in a
laboratory setting. In general, data to support the use of ATD in a clinical setting
are limited. If our sleep data measurements are overestimated, then total sleep time
would be slightly less than most studies that used either PSG or self-report during
pregnancy. Further evaluation of commercially available ATD and sleep is important
given their widespread use and ease of use (e.g., data accessible to participants,
sleep recorded in actual sleep environment) in pregnant women.
We acknowledge several limitations to this study including the small number of participants
and the number of valid sleep observations per person, especially in the context of
the comparison of maternal and neonatal outcomes. We did not collect self-reported
sleep or activity measures or perform PSG in this study. Participants also did not
log sleep start and stop times on the ATD, as such, we opted not to evaluate other
components of sleep such as sleep patterns (e.g., number of sleep periods per day,
daytime, vs. nighttime sleep periods, etc.). Furthermore, we were not aware of participant's
work schedules (e.g., day vs. night shifts) or prior diagnoses of sleep disorders.
However, this study does add to the sleep literature among pregnant women with respect
to commercially available ATD in a diverse population of women who collect the data
in real-life settings and not from a laboratory.
In conclusion, sleep duration was similar to studies from self-reported or actigraphy
sleep data in our study of 94 women who wore a Fitbit Flex device during pregnancy.
Given the increasing ease of access to these devices, further research is required
to evaluate the validity of commercially available ATD in pregnant women and how sleep
duration is related to maternal and neonatal outcomes.
