History of Weight Cycling Is Prospectively Associated With Shorter and Poorer-Quality Sleep and Higher Sleep Apnea Risk in Diverse US Women

Abstract

Background:

Poor sleep and history of weight cycling (HWC) are associated with worse cardiovascular health, yet limited research has evaluated the association between HWC and poor sleep patterns.

Methods:

The American Heart Association Go Red for Women Strategically Focused Research Network cohort at Columbia University (n = 506; mean age, 37 ± 15.7 years; 61% racial/ethnic minority) was used to evaluate the cross-sectional associations of HWC and sleep at baseline and the prospective associations of HWC from baseline with sleep at the 1-year visit. History of weight cycling, defined as losing and gaining 10 lb or more at least once (excluding pregnancy), was self-reported. Sleep duration, sleep quality, insomnia severity, and obstructive sleep apnea risk were assessed using the validated Pittsburgh Sleep Quality Index, Insomnia Severity Index, and Berlin questionnaires. Linear and logistic regression models, adjusted for age, race/ethnicity, education, health insurance status, pregnancy history, and menopausal status, were used to evaluate the relation of HWC with sleep.

Results:

Most women reported 1 or more episodes of weight cycling (72%). In linear models of cross-sectional and prospective data, each additional weight cycling episode was related to shorter sleep duration, poorer sleep quality, longer sleep onset latency, greater insomnia severity, more sleep disturbances, lower sleep efficiency, and higher sleep medication use frequency. In the logistic models, HWC (≥1 vs 0 episodes) was associated with greater odds for short sleep, poor sleep quality, long sleep onset latency (≥26 minutes), high obstructive sleep apnea risk, and sleep efficiency lower than 85%.

Conclusion:

History of weight cycling predicted poor sleep among women, suggesting that weight maintenance may represent an important strategy to promote sleep health. The potential bidirectional relationship between HWC and sleep requires further investigation.

Sleep is an important component of health, and inadequate sleep, both in duration and quality, has been associated with higher risk for obesity, hypertension, and type 2 diabetes,^1,2 predisposing to cardiovascular disease.^3–5 Women may be at greater risk for sleep problems than men potentially because of psychosocial factors such as higher perceived stress at work and home, life events unique to women such as menopause and pregnancy, and hormone changes across the life course.^5–8 Indeed, previous research has shown that women have a higher prevalence of insomnia across their lifetime⁹ and are more likely to have shorter sleep duration than men.¹⁰

Poor sleep is strongly linked to increased risk for obesity.^1,2 Among US women, the prevalence of obesity is 41.1%, and women are at greater risk of developing obesity than men during adulthood.^11,12 Thus, the observed sex disparities in sleep health may contribute to differences in the prevalence of obesity. Lifestyle and weight loss interventions are recommended for patients with obesity to improve cardiometabolic risk factors, but previous studies have documented that most individuals are unable to maintain weight loss in the long-term.^13–15 Furthermore, there are physiological mechanisms that may prevent weight loss and favor weight regain, such as compensatory changes to appetite regulation.¹⁶ The resulting recurring pattern of weight loss and weight regain is known as weight cycling (or less commonly, “yo-yo dieting”) and has been associated with long-term adverse health outcomes, including elevated cardiovascular risk, in research by our group and others.^14,17

In observational studies, better sleep quality and longer sleep duration are associated with increased likelihood of weight loss success.^18,19 In addition, body weight status may influence sleep, as the relation between sleep and obesity risk is likely bidirectional. In a randomized controlled trial of young adult Black women with obesity, weight maintenance interventions improved sleep quality over time.²⁰ It has also been documented that adults with overweight and obesity have poorer sleep.^21,22

Because women are more likely to partake in dieting behaviors owing to a higher perceived social pressure to lose weight, they may be more likely to weight cycle.^23–25 However, although body weight status has been examined in relation to sleep patterns, little attention has been given to the potential association between weight cycling and sleep. To our knowledge, this is the one of the first studies to evaluate the relation between history of weight cycling (HWC) and sleep. The purpose of the current study was to examine the prospective associations between HWC and sleep characteristics (sleep duration, sleep quality, insomnia, sleep onset latency, snoring, and risk of obstructive sleep apnea [OSA]) among a diverse population of women.

METHODS

Design and Study Population

This was a cross-sectional analysis of baseline data and a prospective analysis of HWC, assessed at baseline, with sleep characteristics at 1 year in a diverse community-based cohort of women aged 20 to 76 years. This prospective cohort of 506 English- and Spanish-speaking women were recruited as part of the American Heart Association Go Red for Women Strategically Focused Research Network population study at Columbia University Irving Medical Center. The Go Red for Women study was established to investigate sleep patterns in relation to cardiometabolic risk factors among women throughout the adult lifecycle, as previously described.⁶ Women from all adult life stages, premenopausal, pregnant, menopausal, and postmenopausal, were included. The study was approved by the Columbia University Irving Medical Center Institutional Review Board and written informed consent was required from all participants.

Participants in this sample were recruited from New York Presbyterian Hospital/Columbia University Irving Medical Center and nearby hospitals, living in the neighboring communities, or referred to the study by affiliated physicians. Participants were excluded because of any of the following reasons: (1) currently pregnant or less than 6 months postpartum; (2) unable to read or understand English or Spanish; (3) unable to complete baseline, 6-month, or 1-year evaluations; and (4) have dementia or significant cognitive impairment. At 1 year, 90% of the participants (n = 456) completed 1-year follow-up visits.

History of Weight Cycling Assessments

The primary exposure of interest was HWC, which was assessed using the self-reported response to the following question, “How many times in your life have you lost and gained at least 10 lb in 1 year (do not include pregnancies)?”^17,26 The 10-lb minimum threshold for defining significant weight gain is consistent with definitions used in previous studies.^17,26,27

Sleep Assessments

The Pittsburgh Sleep Quality Index (PSQI) is a validated instrument that was used to assess quality and duration of habitual sleep.²⁸ Based on their responses to the PSQI domains, participants received scores for 7 components that include subjective sleep quality, sleep onset latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleep medication, and daytime dysfunction. The sum of all scores, that is, the global PSQI score, was then used to categorize individuals as either good sleepers (≤5 global sum) or poor sleepers (>5 global sum). Short sleep was defined as less than 7 hours per night, based on the American Heart Association and American Academy of Sleep Medicine’s recommendation of 7 or more hours of sleep to promote optimal health.^4,29 The Insomnia Severity Index Questionnaire was used to evaluate the presence and severity of insomnia.³⁰ The total Insomnia Severity Index scores, which ranged from 0 to 28, were used to categorize individuals into 1 of the following: (1) no clinically significant insomnia (score <8); (2) subthreshold insomnia (score 8–14); (3) clinical insomnia with moderate severity (score 15–21); and (4) severe clinical insomnia (score 22–28). In this study, insomnia was also assessed as a dichotomous variable: no clinically significant insomnia (<8) or subthreshold, moderate, severe insomnia (≥8). The Berlin Sleep Questionnaire was used to identify snoring status and prevalence of high risk phenotype for OSA.³¹ Snoring status was identified by response to the question, “Do you snore?” High risk phenotype for OSA was identified by using responses to questions related to snoring, daytime somnolence, hypertension, and body mass index. Participants with a positive score in 2 or more categories of the Berlin Sleep Questionnaire were considered being at high risk for OSA.

Lifestyle Behaviors and Clinical Risk Factors Assessments

A standardized health questionnaire was used to assess participants’ sociodemographic characteristics (eg, education and health insurance status) and medical history, including pregnancy history and menopausal status. The International Physical Activity Questionnaire was used to assess time spent in moderate and vigorous physical activity.^32,33 Height and weight were measured using standardized height rod and research grade scale, respectively. Body mass index was calculated using the standardized equation (weight [kg]/height [m²]).

Statistical Analysis

All data were gathered on standardized forms, entered into a secure RedCap database, and exported into SAS version 9.4 (SAS Institute, Inc, Cary, North Carolina) for statistical analyses. Descriptive statistics for participant demographics, lifestyle, and medical characteristics were generated using mean ± standard deviation (SD) for continuous variables and frequencies for categorical variables. Multivariable-adjusted linear and logistic regression models were used to evaluate associations between HWC and sleep characteristics at baseline. History of weight cycling, the primary exposure of interest, was evaluated as a continuous variable in linear models (number of weight cycling episodes) and as a categorical variable in logistic models (number of episodes ≥1 vs 0). The outcomes were sleep characteristics, analyzed as both continuous and categorical variables in linear and logistic regression models, respectively. The following sleep characteristic outcomes were used in linear models: sleep duration (hours), sleep quality (PSQI score), sleep onset latency (minutes), insomnia severity (Insomnia Severity Index score), snoring (yes vs no), and risk of sleep apnea (OSA risk score). The following dichotomous outcomes were used in the logistic models: poor versus good sleep quality (PSQI >5 vs ≤5), short versus sufficient sleep (<7 vs ≥7 hours), snoring (yes vs no), and OSA phenotype (high risk vs low risk). The other individual domains of the PSQI were calculated and categorized for logistic models according to the scoring guidelines of the PSQI. For linear models, the individual domains of the PSQI were analyzed on a continuous scale. These included sleep efficiency (time spent sleeping divided by the time spent in bed, ≤85% vs >85%), sleep onset latency (how long it took to fall asleep, ≥26 vs <26 minutes), subjective sleep quality (fairly bad/very bad vs very good/fairly good), use of sleep medication (none vs ≥1 times per week), sleep disturbances due to various factors (assigned score of 2/3 vs 0/1), and daytime dysfunction (assigned score of 2/3 vs 0/1).

Linear and logistic regression analyses were also used to evaluate the prospective associations between HWC, as reported at baseline, and sleep characteristics assessed at 1 year. In model 1, linear and logistic regression models were unadjusted. In model 2, linear and logistic regression models were adjusted for age (years), race/ethnicity (white/non-Hispanic vs racial minority/Hispanic), education (≤college vs >college), and health insurance status (have insurance vs do not have insurance). In model 3, linear and logistic regression models were adjusted for age, race/ethnicity, education, health insurance status, pregnancy history (history of ≥1 pregnancies vs none), and menopausal status (postmenopausal vs premenopausal). Pregnancy history and menopausal status are factors known to affect women’s weight history and were adjusted for in addition to sociodemographic factors (model 3).^34,35

RESULTS

Baseline Characteristics

The baseline characteristics of the study population are presented in Table 1. The mean (SD) age was 37.0 (15.7) years and more than half the sample identified as a racial and/or ethnic minority (61%). Less than a quarter of participants identified as a current or former smoker (23%). The mean BMI and waist circumference were 25.9 (5.7) kg/m² and 35.4 (5.5) inches, respectively. On average, participants spent 281.8 (557.2) min/wk engaged in moderate or vigorous physical activity.

TABLE 1.

Descriptive Characteristics of Study Population at Baseline (N = 506)

Demographic Characteristics	Overall n (%) or Mean ± SD
Age, y	37.0 ± 15.7
Age (≥55 vs <55 y)	96 (19)
Race
White	288 (57)
Black	101 (20)
Asian	94 (19)
Other	22 (4)
Race/ethnicity combined
Minority/Hispanic	311 (61)
Married/living with partner	146 (29)
Socioeconomic status
Have health insurance	317 (63)
Education
≤College	337 (67)
Clinical characteristics
History of pregnancy ≥1	152 (30)
Postmenopausal	146 (29)
Waist circumference, in	35.4 ± 5.5
Waist circumference >35 in	293 (58)
BMI, kg/m2	25.9 ± 5.7
BMI overweight/obese	243 (49)
Sleep habits
Sleep duration (h/night)	6.8 ± 1.2
Sleep duration <7 h	218 (43)
Sleep PSQI score	5.6 ± 3.7
Poor sleep quality (PSQI >5)	200 (40)
How long to fall asleep, min	25.1 ± 30.7
≥26 min to fall asleep	168 (33)
ISI score	7.1 ± 6.0
Insomnia: somewhat, moderate, severe (ISI ≥8)	190 (38)
High risk for OSA	86 (17)
Reported snoring	152 (30)
Habitual sleep efficiency score	0.49 ± 0.87
Sleep disturbances score	1.15 ± 0.55
Use of sleep medications (times per week)	0.38 ± 0.85
Daytime dysfunction score	0.81 ± 0.73
Weight cycling
Self-reported weight problem (yes vs no/unsure)	158 (32)
Weight cycling episodes	2.1 ± 2.6
Episodes ≥1	358 (72)
Lifestyle habits
Moderate/vigorous PA (min/week)	281.8 ± 557.2
Current or former smoker	115 (23)

Abbreviations: BMI, body mass index; PSQI, Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; OSA, obstructive sleep apnea; PA, physical activity.

Most participants reported at least 1 episode of weight cycling (72%). The mean (SD) nightly sleep duration was 6.8 (1.2) hours per night. Nearly half reported sleeping less than 7 hours per night and 40% reported poor sleep quality. More than one-third of participants reported some level of insomnia (38%) and 30% of participants reported snoring. The prevalence of a high-risk phenotype for OSAwas 17%.

Association of History of Weight Cycling With Sleep Characteristics at Baseline

In the multivariable-adjusted logistic regression models of cross-sectional associations between HWC and sleep characteristics at baseline (Table 2), women who reported HWC had 61% higher odds of having a sleep duration of less than 7 hours per night (odds ratio [OR] [95% confidence interval (CI)], 1.61 [1.04–2.47]). History of weight cycling was significantly associated with greater than 2-fold greater odds of having high risk of OSA (OR [95% CI], 2.31 [1.20–4.47]). However, null results were observed for HWC in relation to sleep onset latency, Insomnia Severity Index score, and snoring. There was a borderline significant association between HWC and overall PSQI score (OR [95% CI], 1.52 [0.99–2.34]). When HWC was examined in relation to other individual PSQI domains, women who reported HWC had an 88% higher odds of fairly bad/very bad sleep quality (OR [95% CI], 1.88 [1.09–3.25]), 74% higher odds of poor habitual sleep efficiency (OR [95% CI], 1.74 [1.08–2.80]), and 93% higher odds of more sleep disturbances (OR [95% CI], 1.93 [1.08–3.45]). Results remained largely unchanged after further adjustment for pregnancy history and menopausal status (model 3). Null results were observed for HWC in relation to sleep latency, use of sleep medications, and daytime dysfunction.

TABLE 2.

Logistic Regression Models for Cross-sectional Associations of History of Weight Cycling (≥1 vs <1 Episode) With Sleep Characteristics at Baseline (N = 506)

	Model 1		Model 2		Model 3
Sleep Characteristics	OR (95% CI)	P a	OR (95% CI)	P a	OR (95% CI)	P a
Sleep duration (h) (<7 vs ≥7)	1.55 (1.03–2.33)	.0369a	1.61 (1.04–2.47)	.0314a	1.66 (1.07–2.56)	.0240a
Sleep quality (PSQI score) (>5 vs ≤5)	1.52 (1.00–2.32)	.0482a	1.52 (0.99–2.34)	.0560	1.54 (1.00–2.36)	.0511
How long to fall asleep (sleep onset latency, min) (≥26 vs <26)	1.37 (0.89–2.11)	.1577	1.37 (0.88–2.13)	.1630	1.38 (0.89–2.16)	.1514
Insomnia severity (ISI score) (≥8 vs <8)	1.46 (0.96–2.22)	.0754	1.48 (0.96–2.27)	.0762	1.51 (0.98–2.32)	.0628
Snoring (yes vs no)	1.16 (0.75–1.79)	.5076	1.15 (0.72–1.83)	.5559	1.18 (0.74–1.90)	.4908
Risk of sleep apnea (OSA risk score) (high vs low)	2.04 (1.11–3.77)	.0223a	2.31 (1.20–4.47)	.0126a	2.32 (1.20–4.51)	.0129a
Subjective sleep quality (fairly bad/very bad vs very good/fairly good)	1.87 (1.10–3.18)	.0214a	1.88 (1.09–3.25)	.0233a	1.95 (1.12–3.38)	.0180a
Sleep disturbances (2/3 vs 0/1)	1.88 (1.08–3.27)	.0264a	1.93 (1.08–3.45)	.0273a	1.94 (1.08–3.49)	.0262a
Daytime dysfunction (2/3 vs 0/1)	0.88 (0.50–1.55)	.6556	0.89 (0.50–1.56)	.6744	0.88 (0.50–1.55)	.6510
Habitual sleep efficiency (≤85% vs >85%)	1.72 (1.08–2.72)	.0220a	1.74 (1.08–2.80)	.0230a	1.75 (1.08–2.82)	.0222a
Use of sleep medication (none vs ≥1 times/week)	1.16 (0.70–1.91)	.5706	1.11 (0.66–1.85)	.6930	1.14 (0.68–1.91)	.6218

Model 1: unadjusted. Model 2: adjusted for age, race/ethnicity, education, and health insurance. Model 3: adjusted for age, race/ethnicity, education, health insurance, pregnancy history, and menopausal status.

Abbreviations: OR, odds ratio; CI, confidence interval; PSQI, Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; OSA, obstructive sleep apnea.

^aP < .05.

In multivariable-adjusted linear regression models of the cross-sectional associations of weight cycling episodes and sleep characteristics at baseline (Table 3), a greater number of weight cycling episodes was associated with shorter sleep duration (ß [SE] = −0.08 [0.02]; P < .0001), higher PSQI score indicative of worse sleep quality (ß [SE] = 0.29 [0.06]; P < .0001), longer sleep onset latency (ß [SE] = 1.05 [0.48]; P = .0282), higher Insomnia Severity Index score indicative of greater insomnia severity (ß [SE] = 0.49 [0.10]; P < .0001), and a higher OSA risk score (ß [SE] = 0.08 [0.01]; P < .0001). Snoring was not associated with increased number of weight cycling episodes. In the multivariable-adjusted linear regression models examining the associations between HWC and additional individual domains of the PSQI (Table 3), greater episodes of weight cycling were associated with worse habitual sleep efficiency (ß [SE] = −0.61 [0.22]; P = .0046). Greater episodes of weight cycling were also associated with greater sleep disturbances and worse subjective sleep quality (ß [SE] = 0.22 [0.07]; P = .0022 and ß [SE] = 0.06 [0.01]; P < .0001, respectively).

TABLE 3.

Linear Regression Models for Cross-sectional Associations Between Weight Cycling Episodes and Sleep Characteristics at Baseline (N = 506)

	Model 1		Model 2		Model 3
Sleep Characteristics	ß (SE)	P a	ß (SE)	P a	ß (SE)	P a
Sleep duration (h)	−0.01 (0.02)	<.0001a	−0.08 (0.02)	<.0001a	−0.08 (0.02)	<.0001a
Sleep quality (PSQI score)	0.35 (0.06)	<.0001a	0.29 (0.06)	<.0001a	0.28 (0.06)	<.0001a
Sleep onset latency (min)	1.33 (0.48)	.0060a	1.05 (0.48)	.0282a	1.03 (0.45)	.0232a
ISI score	0.57 (0.10)	<.0001a	0.49 (0.10)	<.0001a	0.49 (0.10)	<.0001a
Snore (yes vs no)	0.02 (0.01)	.0107a	0.01 (0.01)	.1275	0.01 (0.01)	.1638
OSA risk score (high vs low)	0.10 (0.01)	<.0001a	0.08 (0.01)	<.0001a	0.08 (0.01)	<.0001a
Subjective sleep quality (score)	0.06 (0.01)	<.0001a	0.06 (0.01)	<.0001a	0.05 (0.01)	<.0001a
Sleep disturbances (score)	0.30 (0.07)	<.0001a	0.22 (0.07)	.0022a	0.20 (0.07)	.0060a
Daytime dysfunction (score)	0.06 (0.02)	.0031a	0.06 (0.02)	.0045a	0.06 (0.02)	.0088a
Habitual sleep efficiency (%)	−0.75 (0.23)	.0006a	−0.61 (0.22)	.0046a	−0.63 (0.22)	.0040a
Use of sleep medication (times per week)	0.05 (0.01)	.0016a	0.04 (0.01)	.0092a	0.03 (0.01)	.0182a

Model 1: unadjusted. Model 2: adjusted for age, race/ethnicity, education, health insurance. Model 3: adjusted for age, race/ethnicity, education, health insurance, pregnancy history, menopausal status.

Abbreviations: PSQI, Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; OSA, obstructive sleep apnea.

^aP < .05.

Association of Baseline History of Weight Cycling With Sleep Characteristics at 1 Year

In the logistic regression models of the prospective association between HWC at baseline and sleep characteristics at 1 year (Table 4), 1 or more episodes versus fewer than 1 episode of HWC was associated with short sleep duration (<7 hours) (OR [95% CI], 1.88 [1.13–3.12]), poor sleep quality (PSQI >5) (OR [95% CI], 1.75 [1.07–2.86]), high risk of sleep apnea (OR [95% CI], 5.35 [2.32–12.33]), longer sleep onset latency (≥26 minutes) (OR [95% CI], 1.85 [1.04–3.31]), and worse habitual sleep efficiency (≤85%) (OR [95% CI], 1.74 [1.06–2.87]) at 1 year. Null results were observed for associations between baseline HWC and subjective sleep quality, daytime dysfunction, and use of sleep medication at 1 year. Results remained the same after further adjustment for pregnancy history and menopausal status (model 3).

TABLE 4.

Logistic Regression Models for Prospective Associations Between Baseline Weight Cycling Episodes and Sleep Characteristics at 1 Year (N = 456)

	Model 1		Model 2		Model 3
Sleep Characteristics	OR (95% CI)	P a	OR (95% CI)	P a	OR (95% CI)	P a
Sleep duration (h) (<7 vs ≥7)	1.75 (1.08–2.84)	.0236a	1.88 (1.13–3.12)	.0146a	1.77 (1.07–2.95)	.0273a
Sleep quality (PSQI score) (>5 vs ≤5)	1.71 (1.06–2.76)	.0280a	1.75 (1.07–2.86)	.0251a	1.69 (1.03–2.76)	.0382a
How long to fall asleep (sleep onset latency, min) (≥26 vs <26)	1.88 (1.06–3.33)	.0315a	1.85 (1.04–3.31)	.0372a	1.83 (1.02–3.23)	.0434a
Insomnia severity (ISI score) (≥8 vs <8)	1.62 (0.99–2.64)	.0552	1.62 (0.99–2.66)	.0569	1.58 (0.96–2.61)	.0722
Snoring (yes vs no)	1.62 (0.95–2.76)	.0763	1.59 (0.90–2.83)	.1138	1.59 (0.89–2.87)	.1200
Risk of sleep apnea (OSA risk score) (high vs low)	4.65 (2.08–10.41)	.0002a	5.35 (2.32–12.33)	<.0001a	5.13 (2.22–11.85)	.0001a
Subjective sleep quality (fairly bad/very bad vs very good/fairly good)	1.62 (0.88–2.97)	.1195	1.60 (0.87–2.96)	.1310	1.54 (0.83–2.85)	.1709
Sleep disturbances (2/3 vs 0/1)	0.73 (0.43–1.25)	.2485	0.65 (0.37–1.14)	.1343	0.62 (0.35–1.11)	.1099
Daytime dysfunction (2/3 vs 0/1)	1.47 (0.71–3.03)	.3002	1.51 (0.72–3.14)	.2766	1.46 (0.70–3.06)	.3162
Habitual sleep efficiency (≤85% vs >85%)	1.70 (1.04–2.77)	.0343a	1.74 (1.06–2.87)	.0287a	1.67 (1.01–2.76)	.0462a
Use of sleep medication (none vs ≥1 times/week)	1.43 (0.80–2.55)	.2328	1.36 (0.75–2.45)	.3111	1.33 (0.74–2.42)	.3439

Model 1: unadjusted. Model 2: adjusted for age, race/ethnicity, education, health insurance status. Model 3: adjusted for age, race/ethnicity, education, health insurance status, pregnancy history, and menopausal status.

Abbreviations: OR, odds ratio; CI, confidence interval; PSQI, Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; OSA, obstructive sleep apnea.

^aP < .05.

In the linear regression models that examined prospective associations of baseline HWC with sleep characteristics at 1 year (Table 5), greater episodes of weight cycling were associated with shorter sleep duration (ß [SE] = −0.05 [0.02]; P = .0091), poorer sleep quality (ß [SE] = 0.23 [0.05]; P < .0001), higher Insomnia Severity Index score (ß [SE] = 0.39 [0.09]; P < .0001), poorer subjective sleep quality (ß [SE] = 0.04 [0.01]; P = .0005), and greater sleep disturbances (ß [SE] = 0.23 [0.07]; P = .00). Greater episodes of weight cycling were also associated with worse daytime dysfunction and greater use of sleep medications at 1 year (ß [SE] = 0.10 (0.02); P < .0001 and ß [SE] = 0.03 [0.01]; P = .03, respectively). Results remained similar after further adjustment for pregnancy history and menopausal status, except for the association between HWC and use of sleep medication, which was no longer significant (model 3).

TABLE 5.

Linear Regression Models for Prospective Associations Between Baseline Weight Cycling Episodes and Sleep Characteristics at 1 Year (N = 456)

	Model 1		Model 2		Model 3
Sleep Characteristics	ß (SE)	P a	ß (SE)	P a	ß (SE)	P a
Sleep duration (h)	−0.06 (0.02)	.0008a	−0.05 (0.02)	.0091a	−0.05 (0.02)	.0159a
Sleep quality (PSQI score)	0.26 (0.05)	<.0001a	0.23 (0.05)	<.0001a	0.22 (0.06)	.0001a
How long to fall asleep (sleep onset latency, min)	0.51 (0.36)	.1551	0.34 (0.37)	.3568	0.30 (0.37)	.4198
Insomnia severity (ISI score)	0.44 (0.09)	<.0001a	0.39 (0.09)	<.0001a	0.37 (0.09)	<.0001a
Snoring (yes vs no)	0.03 (0.01)	.0004a	0.02 (0.01)	.0082a	0.01 (0.01)	.0656
Risk of sleep apnea (OSA risk score, high vs low)	0.08 (0.01)	<.0001a	0.07 (0.01)	<.0001a	0.07 (0.01)	<.0001a
Subjective sleep quality (score)	0.05 (0.01)	.0001a	0.04 (0.01)	.0005a	0.04 (0.01)	.0009a
Sleep disturbances (score)	0.32 (0.07)	<.0001a	0.23 (0.07)	.0009a	0.22 (0.07)	.0013a
Daytime dysfunction (score)	0.09 (0.02)	<.0001a	0.10 (0.02)	<.0001a	0.09 (0.02)	<.0001a
Habitual sleep efficiency (%)	−0.25 (0.22)	.2638	−0.14 (0.22)	.5379	−0.11 (0.23)	.6249
Use of sleep medication (times per week)	0.04 (0.01)	.0091a	0.03 (0.01)	.0303a	0.03 (0.01)	.0591

Model 1: unadjusted. Model 2: adjusted for age, race/ethnicity, education, and health insurance status. Model 3: adjusted for age, race/ethnicity, education, health insurance status, pregnancy history, and menopausal status.

Abbreviations: PSQI, Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; OSA, obstructive sleep apnea.

^aP < .05.

DISCUSSION

To our knowledge, this represents one of the earliest reports on HWC in relation to sleep characteristics in US women of different life stages. The key findings of this study are that HWC was prospectively associated with several measures of poor sleep health, including short sleep duration, worse sleep quality, greater insomnia severity, and more sleep disturbances and daytime dysfunction. In addition, HWC predicted a high risk phenotype for OSA. Associations persisted after statistical adjustment for factors known to affect women’s weight history, including pregnancy history and menopausal status.

The relationship between overweight/obesity and short sleep duration has been well documented.^1,2,36 In a meta-analysis of observational cohort studies, adults with short sleep duration had a 55% higher odds of obesity.² Furthermore, short sleep duration and poor sleep quality have been associated with dietary factors that are linked to weight gain and weight cycling, such as increased intake of total fat, unsaturated fat, and overall quality of food and caloric intake.^36,37 Prevention of overweight/obesity has also been shown to influence sleep quality. In a 12-month randomized controlled trial of a weight gain prevention intervention among 184 Black women with obesity, the intervention group reported greater improvements in sleep disturbance and sleep problems at 12 months.²⁰ In addition to the well-established link between overweight/obesity and sleep, we have previously shown that HWC was associated with higher odds of poor cardiovascular health among women in our cohort. Although not formally tested, it is possible that sleep may be a potential mechanism underlying this previously documented relation.¹⁷

Although we are not aware of any studies to date that have examined the relation between HWC and sleep, specifically, one potential mechanistic pathway underlying the relation of HWC with sleep could be physiological adaptations that favor weight regain.¹⁶ Previous studies have shown that the hormones that regulate metabolism and appetite, such as ghrelin and leptin, are influenced by sleep^38,39 and may be altered in participants who have HWC.^22,40 Weight cycling has also been shown to lead to preferential deposition of visceral fat⁴¹ or perivascular adipose tissue.⁴² Visceral fat accumulation has been shown to cause reduction of arterial oxygen tension during sleep and is noted to be a better indicator of OSA prevalence and severity than BMI alone.⁴³ The weight loss is countered by metabolic adaptations to the weight loss and may lead not only to a higher BMI in the long-term,^17,44,45 which is a known risk factor for OSA, but also to poor sleep patterns including shorter duration, poorer sleep quality, marked sleep disturbances, and more fragmented sleep.^1,46

Other factors that are associated with weight change,⁴⁷ including lower physical activity,⁴⁸ and increased stress,^49,50 are also known to influence sleep. Physical activity protects against weight gain, especially when combined with decreased food intake over time.⁵¹ Previous research in our cohort showed that not meeting recommended guidelines for physical activity was associated with sleep duration of less than 6 hours per night.⁴⁸ Increased stress and stressful events or stressors have been shown to be associated with poor sleep.⁴⁹ Stress also causes changes in food intake patterns and the effects of weight loss.⁵⁰ These behavioral factors, although not tested in this study, may help to explain in part the relation we found between HWC and future sleep disturbances among women.

Strengths of this study include the racially and ethnically diverse nature of this community-based cohort of women and the use of well-validated surveys and questionnaires to ascertain sleep characteristics. The prospective design also enabled the establishment of temporality. Limitations of the study include the observational design, which limits our ability to determine causal relations between weight cycling and sleep characteristics. Potential recall bias of collecting data via self-reported surveys and questionnaires is another limitation. Although the Berlin Questionnaire assesses the risk of OSA, it does not provide information on diagnosis and severity of OSA, which is typically ascertained from polysomnography. Although the sample is diverse, this was a community-based sample that may not be representative of the broader US population and the moderate sample size limited the power for any subgroup analyses by life stage, which may modify associations of HWC with sleep.

In conclusion, this study provides evidence that HWC predicts poorer sleep health among US women. A greater number of weight cycling episodes was related to shorter sleep duration, poorer sleep quality, longer sleep onset latency, greater insomnia severity, greater OSA risk, worse subjective sleep quality score, more sleep disturbances and daytime dysfunction, lower sleep efficiency, and greater use of sleep medication. These results remained in prospective analyses. Any HWC was related to greater odds of short sleep duration, worse subjective sleep quality, greater sleep disturbances, and poorer sleep efficiency. These results remained in prospective analyses along with greater odds of poorer sleep quality, longer sleep onset latency, and greater OSA risk. These data represent one of the earliest reports to demonstrate such associations in women and suggest that screening for HWC may be helpful for identifying risk for poor sleep and that maintaining a stable weight over time may be a strategy for the promotion of sleep health, but our findings warrant confirmation in other populations. The relation between HWC and sleep is likely bidirectional, and thus, the intricate interplay between sleep and weight loss and maintenance requires further investigation. Additional research in larger samples and with longer follow-up is also necessary to identify critical periods of exposure across the life course during which HWC may impact sleep. In addition, racially/ethnically diverse cohorts of both men and women are needed to better understand sex and racial/ethnic differences in these relations to inform more targeted weight maintenance interventions for sleep health and cardiovascular health promotion.

What’s New and Important.

• History of weight cycling was prospectively associated with several measures of poor sleep, including short sleep duration, worse sleep quality, greater insomnia, greater sleep disturbances, and greater daytime dysfunction among diverse US women across various life stages.

• Findings suggest that screening for HWC may be helpful for identifying risk for poor sleep and that maintaining a stable weight over time may promote better quality sleep.

• Future research can potentially inform more targeted weight maintenance interventions for sleep health and cardiovascular health promotion.