Sleep characteristics modify the associations of physical activity during pregnancy and gestational weight gain

Excessive gestational weight gain (eGWG) is associated with adverse long-term maternal outcomes. Most lifestyle interventions that incorporate physical activity have been ineffective at reducing eGWG. The purpose of this study was to determine if sleep modified the relationships between physical activity change from the 2nd to 3rd trimester and the odds of excessive gestational weight gain (eGWG). This was a secondary data analysis of a prospective cohort study of pregnant birthing people with overweight or obesity (n = 105). We estimated physical activity energy expenditure (PAEE) in the 2nd and 3rd trimesters of pregnancy and sleep characteristics (i.e., sleep quality, daytime dysfunction, sleep efficiency, sleep duration) in the 2nd trimester of pregnancy with validated measures. We used regression models with sleep and PAEE change (increase/stable vs. decrease) interaction terms to examine the impact of sleep on PAEE change and eGWG. Mean GWG was 37.02 ± 16.76 lbs. and 80% of participants experienced eGWG. Eighteen percent of participants increased their PAEE from the 2nd to the 3rd trimester. Increasing (vs. decreasing) PAEE was associated with lower log-odds of eGWG only among participants that slept at least 8 h/night (p = 0.06), had at least 85% sleep efficiency (p = 0.03), or reported less daytime dysfunction (p = 0.08). Sleep quality did not moderate the association between PAEE change and eGWG. Weight management interventions in pregnancy should consider screening for and addressing poor sleep in the second trimester.


Introduction
Birthing people with pre-pregnancy overweight or obesity who have excessive gestational weight gain (GWG) (i.e., GWG that exceeds amounts established by the Institute of Medicine) are at greater risk for adverse maternal outcomes, including cesarean section, gestational diabetes, and postpartum weight retention [1][2][3], than other birthing people. These adverse outcomes further increase longterm cardiovascular disease risk among individuals with overweight or obesity before pregnancy. Interventions designed to attenuate excessive GWG in pregnancy have focused on improving diet and physical activity behaviors but have largely been unsuccessful.
Specifically, interventions that have incorporated physical activity to attenuate weight gain have been ineffective at reducing excessive GWG (eGWG) among birthing people with overweight or obesity [4][5][6][7][8][9][10][11]. For example, a meta-analysis of six randomized controlled trials (RCTs) of exercise interventions among pregnant people with overweight and obesity found an overall null effect on GWG [10]. However, the authors identified several intervention design factors that modified the intervention effects. Specifically, supervised (vs. unsupervised) and structured (vs. unstructured) interventions had slightly stronger effects [10]. In addition to intervention design factors, behavioral factors can modify intervention effectiveness. Indeed, the growing literature on multiple health behavior change provides evidence that considering the interaction of health behaviors can lead to more effective interventions [12]. Thus, identifying factors that modify the relationship between exercise or physical activity during pregnancy and GWG is important.
Sleep is a behavior that could potentially modify associations of physical activity during pregnancy and GWG. Poor sleep characteristics such as insufficient sleep duration (i.e., < 7 h/night), low sleep efficiency (i.e., < 85% of the time in bed asleep), and poor self-reported sleep quality are common in pregnancy [13][14][15]. In non-pregnant populations, these sleep characteristics are independently associated with obesity and weight gain [16]. Although the data on the effects of sleep during pregnancy and GWG is limited, there is evidence that short sleep duration and perceived poor sleep quality in mid-pregnancy is associated with inadequate GWG. Poor sleep has also been associated with reduced physical fitness [17], lower exercise frequency [18], and lower next-day physical activity [19][20][21], which can reduce intervention effectiveness. Also, there is evidence from clinical trials that poor sleep moderates the effectiveness of lifestyle interventions. There are no data from observational or experimental studies among pregnant people to the authors' knowledge [22][23][24][25].
Given the links between sleep characteristics, obesity, and weight-related behaviors, data related to the impact of sleep on physical activity and weight gain among pregnant populations are needed. Such data would inform the design of definitive clinical trials. We conducted a secondary data analysis of a prospective cohort study that examined factors related to weight gain during pregnancy among participants with pre-pregnancy overweight or obesity. Throughout pregnancy, participants self-reported their physical activity and sleep behaviors. Using this observational data, we explored patterns of a potential interaction between sleep, physical activity change, and GWG. Accordingly, the primary aim of this study was to examine the impact of sleep characteristics in the second trimester of pregnancy on the association between physical activity change and GWG in a cohort of participants with prepregnancy overweight or obesity. We hypothesized that increasing physical activity would be associated with lower odds of eGWG among participants with better sleep characteristics. Our overall goal is to examine patterns of interactions of multiple sleep characteristics to generate hypotheses for future studies.

Study design and population
We recruited participants (N = 257) between 12 and 20 weeks' gestation from September 2012 to January 2017 at obstetric clinics affiliated with a large urban hospital. Participants were eligible if they (1) were ≥ 14 years of age, (2) had a pre-pregnancy body mass index (BMI) ≥ 25 kg/m 2 , and (3) had a singleton pregnancy. Exclusion criteria were (1) type I diabetes, (2) taking medications or diagnosed with conditions known to influence weight, (3) participating in a weight management program, or (4) reported psychiatric symptoms requiring immediate treatment. For this secondary analysis, we only included participants with data on sleep, physical activity, and GWG. Sleep assessment (further described below) was introduced later in data collection (February 2015, after n = 130 participants had completed the study). Sleep data were available on 100% of participants enrolled after we began sleep assessments. We excluded participants aged < 18 years (n = 2) because sleep guidelines may differ for adults and young adults [26]. Additionally, we excluded participants with inadequate GWG (i.e., weight gain below IOM recommendations; n = 20) because our primary outcome was eGWG. Participants aged 18 + years provided written informed consent before initiating any procedures. The University's Institutional Review Board approved all study procedures.

Gestational weight gain
Participants self-reported their pre-pregnancy weight during the initial phone screen. We abstracted the final measured weight before delivery from medical records. Total GWG was defined as the difference in final measured weight before delivery and self-reported pre-pregnancy weight. We defined eGWG according to the Institute of Medicine (IOM) guidelines [27]. Specifically, these guidelines define adequate GWG as weight gain between 15 and 25 lbs for people with pre-pregnancy overweight and between 11 and 20 lbs for people with pre-pregnancy obesity. IOM defines eGWG as weight gain greater than 25 lbs and 20 lbs for people with overweight and obesity, respectively. We measured the participant's height via a calibrated stadiometer. Pre-pregnancy BMI was calculated as weight in kilograms divided by height in meters squared.

Physical activity
We estimated physical activity energy expenditure with the interviewer-administered Paffenbarger questionnaire. The Paffenbarger is a widely used instrument that moderately correlates with objective physical activity measures among healthy adults with overweight and obesity [28]. The Paffenbarger assesses activities of daily living and leisure-time physical activity in the past 7 days. To assess activities of daily living, participants reported the number of flights of stairs climbed, and the number of city blocks walked outside each day in the past week. To assess leisure-time physical activity, participants reported their frequency, intensity, and duration of engaging in sports, recreational, or physical activities other than walking outside. As needed, the interviewers probed responses to clarify intensity levels.
We estimated weekly energy expenditure from stairs climbed (stairs climbed per day × (4 kcal × 7 days)) and blocks walked (blocks walked per day × (8 kcal × 7 days)) using a standard formula. To estimate energy expenditure from leisure-time physical activity, two trained research assistants independently assigned each activity a metabolic equivalent of task (MET) level based on the Compendium of Physical Activities [29]. The Compendium is a widely used coding system for classifying the intensity of various forms of physical activity. Infrequently, reported activities fell into a general category that required interpretation. Given that data were double-entered, if two different MET levels were assigned for an activity, we chose the lower MET value because it is a more conservative estimate of PAEE. Next, we estimated weekly energy expenditure using the following formula: MET × body weight (kg) × weekly duration (hrs). We estimated total PAEE by taking the sum of energy expenditure from blocks walked, stairs climbed, and sports/ recreational activities. We estimated PAEE change as the difference in PAEE between the second and third trimesters.

Sleep characteristics
The Pittsburgh Sleep Quality Index (PSQI) was used to assess sleep characteristics. The PSQI is one of the most commonly used instruments to measure self-reported sleep and has high sensitivity (89.6%) and specificity (86.5%) at distinguishing between good and poor sleep quality [30]. The PSQI has been used in prior studies in pregnancy to assess sleep associations with physical activity [31,32] and maternal health outcomes [33][34][35].
We selected sleep moderators that assessed sleep health constructs (satisfaction, alertness/daytime function, timing, efficiency, and duration) proposed by Buysse [36]. These sleep health dimensions are independently associated with health outcomes in adult populations. We used the overall PSQI global score (i.e., sleep quality) as our proxy for sleep satisfaction. The global PSQI score is comprised of seven components (i.e., subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleeping medications, and daytime dysfunction over the last month), each scored on a scale of 0 to 3. Overall sleep quality is calculated by taking the sum of all components (possible range of 0-21), with higher scores indicating worse sleep quality. To assess daytime dysfunction, participants were asked about the frequency of (1) trouble staying awake while driving, eating meals, or (2) engaging in social activity, and problems maintaining enthusiasm for getting things done. Each question was scored on a scale of 0 to 3, with higher scores indicating worse daytime dysfunction. Sleep efficiency, defined as the percentage of time in bed spent asleep, was estimated by dividing the self-reported actual sleep time by the time in bed. A lower percentage indicates worse sleep efficiency. Sleep duration was assessed by asking participants, "How many hours of actual sleep did you get at night?" during the past month.

Covariates
At baseline, participants self-reported demographic, behavioral, and pregnancy-related information, including their age, relationship status, race, annual household income, education level, gestational age, and parity. We abstracted gestational age at delivery from participants' medical charts.

Statistical analysis
Descriptive statistics (i.e., means, standard deviations, and proportions) were calculated for participant demographics and baseline characteristics. Chi-square test and Student's T-test were used to compare demographic and baseline characteristics between PAEE change groups, increased/stable or decreased from 2nd to 3rd trimester. For simplicity, we will refer to the increased/stable group as "increased".
For the primary analysis, we used multivariable logistic regression to examine associations between PAEE change and eGWG. For each analysis, the reference group was participants that decreased their PAEE. We ran adjusted models that included covariates related to GWG identified in prior research [37][38][39] or based on the research team's knowledge. Specifically, we adjusted for age, education, income, prepregnancy BMI, and gestational length. We also adjusted for baseline PAEE. We treated each sleep variable as a continuous moderator. We fit separate sleep (i.e., overall quality, daytime dysfunction, efficiency, and duration) and PAEE change interaction terms in the model to identify effect modification. We will use a likelihood ratio test to compare model fit with and without the interaction term. We used the Johnson-Neyman approach to probe statistically significant (p < 0.10) interactions as an exploratory analysis. Probing interactions for continuous moderators allow us to identify where the conditional association between PAEE and GWG is statistically significant in the distribution of sleep. Identifying regions of statistical significance could be informative for identifying potential thresholds for sleep screening in future physical activity weight management interventions.
All statistical analyses were computed in the R statistical computing environment (Vienna, Austria. http:// www.Rproje ct. org, Version 1.3.1073). For the primary analysis, we used PROCESS, a freely available macro that aids in computing mediation and moderation analysis [40].

Results
The final sample included 105 participants with a mean GWG of 37.02 ± 16.76 lbs, and approximately 80% (n = 84) had eGWG. Participants included in the final sample were slightly older, had a lower pre-pregnancy BMI, and had a higher household income than participants not included in the final sample. There were no statistically significant differences in baseline PAEE, PAEE change, or other demographic characteristics between participants included in the final sample compared with those not included.
The sample median PAEE was 820.75 [IQR = 438.67, 1733.44] kcal/week in the second trimester. The median change in PAEE from the second to the third trimester was -454.42 kcal/week (IQR = -1168.29, -57.04 kcal/ week). Approximately 18% (n = 19) of participants increased their PAEE from the second to the third trimester. There were no statistically significant differences in demographic or sleep characteristics between participants who increased their PAEE and participants who decreased their PAEE (Table 1). However, participants who decreased their PAEE (vs. increased) were more physically active in the second trimester. Among participants who decreased their PAEE, the median change was -664.71 kcal/week (95% CI -1347.99, -192.05). Among participants who increased their PAEE, the median change was 224.00 kcal/week (95% CI 121.33, 408.32).
As hypothesized, sleep characteristics moderated the associations between PAEE change and eGWG (Table 2; Fig. 1). PAEE change was not independently associated with eGWG. Higher daytime dysfunction was the only sleep characteristic independently associated with a higher log-odds of eGWG ( Table 2). The log-odds for eGWG were lower among participants who increased (vs. decreased) their PAEE when their daytime dysfunction score was above 1.4 on a 0 to 3 scale (higher values indicated more daytime dysfunction). Approximately 10% of participants had daytime dysfunction values above this threshold. The log-odds of eGWG were lower among participants who increased (vs. decreased) their PAEE when their sleep efficiency and sleep duration were above 85.7% and 8 h/night, respectively. Approximately 37% and 5% of participants had sleep efficiency and sleep duration levels above these thresholds, respectively. It appeared that participants who increased their PAEE had lower odds of eGWG than participants who decreased their PAEE when sleep quality was high (i.e., lower PSQI score), but the relationship was not statistically significant.

Discussion
In this secondary data analysis of a prospective cohort during pregnancy, we examined the impact of sleep characteristics in the second trimester of pregnancy on the association between physical activity change and GWG. We found that increasing PAEE was associated with lower log-odds of eGWG, only among participants with at least 85% sleep efficiency or who slept at least 8 h/night. These sleep efficiency and duration thresholds are consistent with the American Academy of Sleep Medicine, Sleep Research Society, and National Sleep Foundation recommendations [41][42][43]. Also, daytime dysfunction moderated associations between PAEE change and eGWG. Specifically, participants who decreased their PAEE and reported more daytime dysfunction had the highest log-odds of eGWG. Last, participants who increased their PAEE had lower log-odds eGWG when sleep quality was higher, but the relationship was not statistically significant.
These novel findings are consistent with some studies among non-pregnant populations related to the impact of sleep on weight-related lifestyle interventions [22-25, 44, 45]. For example, in a 2-phase RCT of a lifestyle intervention among 472 adults with obesity, participants who slept 6-8 h/night (vs. < 6 h/night) had a greater odds of successful weight loss (4.5 kg or greater) at 6 months follow-up [22]. Likewise, in a 6-month lifestyle intervention among adults with metabolic syndrome, short sleep duration (< 6 h/night) and high sleep variability (highest tertile of variability in sleep duration) were associated with smaller decreases in BMI and waist circumference [23]. Other studies have also found that measures of sleep continuity are associated with weight loss. In a 7-month RCT of a lifestyle intervention among 90 participants with overweight and obesity, actigraphy-assessed sleep fragmentation (i.e., 5 or more nocturnal awakenings) was associated with greater BMI reductions [24]. Although the current study's findings are based on observational data, they show consistent results in a population who commonly experience poor sleep.
To the authors' knowledge, these data are the first to evaluate the impact of sleep on the associations of physical activity change during pregnancy and GWG among birthing people with overweight/obesity prior to pregnancy. This study's findings are important, as most behavioral weight management interventions during pregnancy do not screen for or simultaneously address poor sleep among pregnant participants, which may have attenuated their effectiveness. These results must be interpreted with caution because they were derived from an observational study design. Our findings  . 1 The associations of PAEE change and the log-odds eGWG across the distribution of sleep characteristics may be biased due to unmeasured confounders associated with the change in PAEE and GWG. However, we did not observe any demographic or behavioral differences between participants that increased vs. decreased their PAEE, except for second-trimester PAEE. We used covariate adjustment in our regression models to account for differences in second trimester PAEE and other covariates selected based on the experience of the research team and factors identified in prior studies. We acknowledge the possibility of unknown and unmeasured confounders, which would be accounted for in a randomized controlled trial study design. However, the findings support the hypothesis of our research team and others [46] that sleep and sleep health should be considered an essential part of weight management programs. Our findings will need to be confirmed in the context of well-designed RCTs of physical activity interventions conducted during pregnancy or observational studies where causal effects are identifiable. This study had several additional strengths worth noting. First, the sample was focused on birthing people with pre-pregnancy overweight and obesity, a group with a high risk for excessive GWG. Also, the data involve repeated measures of PAEE and sleep during pregnancy using validated questionnaires. However, there are some limitations to consider when interpreting these findings. First, we relied on self-reported measures of physical activity and sleep. The Paffenbarger physical activity questionnaire relies on participant recall of walking behaviors and indirect energy expenditure estimation. Indirect assessment of PAEE may over or under-estimate true PAEE and may fail to capture change. However, the Paffenbarger is a commonly used measure with reasonable validity and reliability and is associated with outcomes known to be impacted by physical activity [47][48][49]. We also relied on self-reported measures to calculate sleep duration and efficiency. Self-reported sleep duration and efficiency measures may be less accurate than objective assessments of sleep characteristics, particularly among individuals with shorter sleep duration [50]. Misclassification of sleep duration would likely be similar between PAEE groups, biasing the results towards the null hypothesis. However, the PSQI is the most commonly used measure of sleep quality and has been used in several studies during pregnancy [51]. The recommendations for optimal sleep duration are also based largely on self-reported sleep. Thus, the findings in this study are comparable to other studies.
Last, the majority of the participants in this study were physically inactive. We previously reported that only 21% of the participants in this cohort achieved recommended physical activity levels, lower than estimates from nationally representative samples of pregnant people [52]. However, despite the small sample of participants reporting increases in physical activity, we still observed statistically significant associations consistent with our a priori hypotheses.
The results of this study have important implications for practice. Specifically, we provide evidence that sleep screenings should be considered when designing and implementing weight management interventions during pregnancy. Sleep screenings have the potential for identifying populations who would most benefit from existing weight management strategies or those who need interventions that also address sleep disturbances.

Conclusion
In conclusion, poor sleep health is common during pregnancy and modifies associations between physical activity change and GWG. Future research should investigate if identifying and managing poor sleep before or simultaneously with physical activity interventions reduces eGWG. Given the morbidity associated with eGWG, such interventions may significantly impact maternal and infant health.
Author contributions MSH contributed to the conception of the work, data analysis and interpretation, drafting and revising the work; RPC contributed to the data interpretation and revising the work. SD contributed to the data interpretation and revising the work; DJB contributed to the data interpretation and revising the work; EMV contributed to the data interpretation and revising the work; YC contributed to the data interpretation and revising the work; MDL contributed to the data acquisition, conception of the work, data interpretation, and revising the work. All authors approved the final version to be published.

Declarations
Conflict of interest Over the past five years, Buysse has served as a paid consultant to Bayer, BeHealth Solutions, Cereve/Ebb Therapeutics, Emmi Solutions, National Cancer Institute, Pear Therapeutics, Philips Respironics, Sleep Number, Idorsia, and Weight Watchers International. He has served as a paid consultant for professional educational programs developed by the American Academy of Physician Assistants and CME Institute and received payment for a professional education program sponsored by Eisai (content developed exclusively by Buysse). Buysse is an author of the Pittsburgh Sleep Quality Index, Pittsburgh Sleep Quality Index Addendum for PTSD (PSQI-A), Brief Pittsburgh Sleep Quality Index (B-PSQI), Daytime Insomnia Symptoms Scale, Pittsburgh Sleep Diary, Insomnia Symptom Questionnaire, and RU_SATED (copyright held by the University of Pittsburgh). These instruments have been licensed to commercial entities for fees. He is also co-author of the Consensus Sleep Diary (copyright held by Ryerson University), licensed to commercial entities for a fee. He has received grant support from NIH, PCORI, AHRQ, and the VA. Over the past three years, Buysse has served as a paid consultant to National Cancer Institute, Pear Therapeutics, Sleep Number, Idorsia, and Weight Watchers International. None of the authors have conflicts of interest to disclosure except for Dr. Daniel Buysse, who reported his conflicts in the attached ICMJFE disclosure form.
Ethical approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
Informed consent Informed consent was obtained from all individual participants included in the study.