Preconception sleep, pregnancy loss, and adverse pregnancy outcomes among women with a history of pregnancy loss

Abstract

STUDY QUESTION

Are preconception sleep characteristics associated with pregnancy loss and adverse pregnancy outcomes?

SUMMARY ANSWER

Preconception sleep characteristics were not associated with pregnancy loss, but earlier sleep midpoints were associated with lower risk of adverse pregnancy outcomes, while social jetlag >1 h was associated with greater risk of a composite of adverse pregnancy outcomes.

WHAT IS KNOWN ALREADY

Short sleep duration in mid-pregnancy has been associated with risk of second-trimester pregnancy loss, preterm birth (PTB), and hypertensive disorders of pregnancy (HDP). The relationships between preconception sleep and pregnancy loss, and adverse pregnancy outcomes have not been well characterized, despite plausible links.

STUDY DESIGN, SIZE, DURATION

This was a secondary analysis of a randomized controlled trial conducted between 2006 and 2012 that prospectively followed 1228 women who were attempting to become pregnant after a history of pregnancy loss. Women were followed for ≤6 cycles while attempting pregnancy, and throughout pregnancy if they conceived. Over the follow-up, 140 women withdrew from the study.

PARTICIPANTS/MATERIALS, SETTING, METHODS

This study evaluated baseline, self-reported preconception sleep duration, sleep latency, sleep midpoint, and social jetlag with risk of pregnancy loss and adverse pregnancy outcomes (e.g. PTB, HDP, and gestational diabetes (GDM)) among 1228 women with a history of pregnancy loss in the EAGeR trial. Pregnancy was documented by hCG tests; 797 women became pregnant over the follow-up. Pregnancy losses were defined as any loss after a positive hCG test; there were 188 pregnancy losses. PTB, HDP, and GDM cases were ascertained via medical record abstraction. PTB (n = 53), HDP (n = 62), and GDM (n = 22) were examined as a composite outcome (n = 118) and PTB and HDP were examined individually in exploratory analyses. GDM was not examined individually due to insufficient numbers. Log-Poisson models were used to estimate relative risks (RR) and 95% CIs for associations between preconception sleep characteristics, and pregnancy loss or adverse pregnancy outcomes with adjustment for age, BMI, lifestyle, and sociodemographic factors. Stabilized inverse probability weights were applied to address potential selection bias from loss to follow-up and from restricting to pregnancy.

MAIN RESULTS AND THE ROLE OF CHANCE

Preconception sleep characteristics were not associated with risk of pregnancy loss. Preconception sleep duration and sleep latency were not associated with risk of the composite adverse pregnancy outcome. Early preconception sleep midpoints were associated with a lower risk of the composite adverse pregnancy outcome (first vs second tertile RR; 0.63, 95% CI: 0.40, 0.98) and preconception social jetlag was associated with a higher risk of the composite adverse pregnancy outcome (>1 vs ≤1 h RR; 1.65, 95% CI: 1.11, 2.44).

LIMITATIONS, REASONS FOR CAUTION

Preconception sleep was restricted to baseline self-report, which may be non-differentially misclassified and may underestimate these associations. The EAGeR study did not measure sleep during pregnancy. There were few adverse pregnancy outcomes and thus limited power to evaluate individual outcomes; the findings could be due to chance.

WIDER IMPLICATIONS OF THE FINDINGS

These findings suggest that preconception sleep is not associated with pregnancy loss, but preconception sleep timing may be relevant for risk of adverse pregnancy outcomes. Additional studies on preconception sleep and adverse pregnancy outcomes are needed given the potential impact of poor sleep on pregnancy outcomes.

STUDY FUNDING/COMPETING INTEREST(S)

Joshua R. Freeman and this work were supported by the Intramural Research Program Cancer Research Training Award, National Cancer Institute, National Institutes of Health (ZIA CP010197), and the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland (Contract numbers: HHSN267200603423, HHSN267200603424, HHSN267200603426, HHSN275201300023I). Dr Silver received NIH funding through the listed contracts as site-PI for the original EAGeR trial at the University of Utah. Dr O’Brien reports receiving funding from the Star Legacy Foundation (paid to institution); an advisory board role at the Star Legacy Foundation; and receiving travel support from the Star Legacy Foundation. Dr Dunietz reports a role as Associate Editor at Human Reproduction and a role on the Journal Editorial Board of SLEEP. Dr Purdue-Smithe is an employee of Merck & Co. and has received stock compensation as an employee of Merck & Co. in the past 36 months. The work in this manuscript was completed before Dr Purdue-Smithe’s employment at Merck & Co. and is unrelated to Dr Purdue-Smithe’s work at the company. Dr Silver reports royalties or licenses from BJOG and UpToDate, Inc. in the past 36 months, receiving payment or honoraria for Grand Rounds in the past 36 months, and participating on a Data Safety Monitoring Board or Advisory Board for a National Institutes of Health-funded Apple Trial in the past 36 months. The other authors report there are no competing interests to declare.

TRIAL REGISTRATION NUMBER

Clinicaltrials.gov NCT00467363

Introduction

It is estimated that ∼27% of human conceptions end in pregnancy loss, and among pregnancies, approximately 8–16% are complicated by preterm birth (PTB), hypertensive disorders of pregnancy (HDP), or gestational diabetes (GDM) (Wilcox et al., 1988; Wilcox et al., 1999; Ford et al., 2022; Martin and Gregory, 2023; CDC, 2024a). Collectively, these adverse pregnancy outcomes may negatively influence maternal and neonatal short and long-term health (Stewart et al., 1999; Hauspurg et al., 2018; Kong et al., 2020; Gow et al., 2024). Given the potential consequences of these outcomes, there is a need to identify modifiable risk factors.

An example of a modifiable risk factor is a daily behavior such as sleep. Among US women in 2020, 35.4% reported short sleep duration (i.e. <7 h/night), and it is estimated that 44.1% of women in the USA have at least 1 h of social jetlag, a marker of circadian misalignment (Di et al., 2022; CDC, 2024b). Importantly, sleep and circadian rhythms prior to conception are related to reproductive hormone fluctuations, which may impact reproductive functions (Hall et al., 2005; Lateef and Akintubosun, 2020; Beroukhim et al., 2022). Sleep during the preconception and periconception windows may also be relevant for placentation as developmental changes begin as early as 3–5 weeks’ gestation (Louis et al., 2008; James et al., 2012; Stephenson et al., 2018). This critical window of exposure may reflect a period in which preconception sleep could alter endometrial decidualization and placental tissues (Louis et al., 2008; James et al., 2012; Stephenson et al., 2018; Cui et al., 2022). Endometrial decidualization is critical for establishing a robust microvasculature at the interfacing tissues of the blastocyst and uterus (Fournier et al., 2021; Greenbaum et al., 2023; Zambuto et al., 2024). Inadequate decidualization has been linked to infertility, pregnancy loss, preeclampsia, and PTB; thus, sleep variability during preconception may be related to a continuum of adverse pregnancy outcomes through pathways involving defective endometrial decidualization and shallow placentation (Larsen et al., 2013; Schatz et al., 2016; Sang et al., 2020; Garrido-Gómez et al., 2022).

To date, there is evidence that generally supports a role of mid-pregnancy short sleep duration, mid-pregnancy poor sleep quality, mid-pregnancy later sleep timing, circadian disruption from shift work, or work during typical sleep hours, with greater risk of pregnancy loss and adverse pregnancy outcomes (Infante-Rivard et al., 1993; Zhu et al., 2004; Samaraweera and Abeysena, 2010; Williams et al., 2010; Cai et al., 2017; Facco et al., 2017, 2018; Xu et al., 2018; Begtrup et al., 2019; Wang and Jin, 2020; Zhang et al., 2020; Beroukhim et al., 2022). However, the studies evaluating mid-pregnancy sleep and pregnancy loss recruited women near the end of the first trimester after most pregnancy losses have occurred (Wilcox et al., 1988, 1999; Zinaman et al., 1996; Samaraweera and Abeysena, 2010), possibly missing a critical early window of susceptibility. Preconception studies, which can prospectively evaluate early pregnancy loss, including implantation failures, may be able to overcome these limitations (Weinberg et al., 1992). In addition, it is unclear whether sleep assessed in mid-pregnancy is a cause or consequence of adverse pregnancy outcomes. It is possible that preconception sleep may reflect the influence of long-term sleep patterns on pregnancy establishment and may be different in its association with pregnancy loss and adverse pregnancy outcomes than sleep measured mid-pregnancy, which may reflect short-term impacts of pregnancy on sleep (Mindell et al., 2015; Won, 2015). However, the role of preconception sleep in adverse pregnancy outcomes has rarely been evaluated despite the proximity of the preconception period to a critical window of susceptibility for pregnancy (Louis et al., 2008; Stephenson et al., 2018; Nakahara et al., 2020; Bond et al., 2024).

Given these knowledge gaps on the potential impact of preconception sleep on pregnancy loss and adverse pregnancy outcomes, the objective of this work was to examine how preconception sleep is associated with pregnancy loss and adverse pregnancy outcomes including PTB, HDP, and GDM. We hypothesized that short (<7 h) and long (≥9 h) sleep duration, longer sleep latency as a measure of poor sleep quality, earlier and later sleep midpoints, and higher social jetlag could be associated with pregnancy loss and adverse pregnancy outcomes.

Materials and methods

Study population

Women in this analysis are from the Effects of Aspirin in Gestation and Reproduction (EAGeR) trial. The EAGeR (n = 1228) trial was a block-randomized, double-blind, placebo-controlled trial designed to examine the effect of low-dose aspirin on probability of live birth (Schisterman et al., 2013, 2014). The study included US women recruited from Salt Lake City, Utah; Buffalo, New York; Scranton, Pennsylvania; and Denver, Colorado from 2006 to 2012. To be eligible for recruitment, women must have had 1–2 prior pregnancy losses, be aged 18–40 years at baseline, have an intact uterus, tubes, and ovaries, and have regular menstrual cycles ranging from 21 to 42 days (Schisterman et al., 2013, 2014). Women must have also been attempting to conceive and must not have been pregnant at enrollment. Women were excluded if they had allergies to non-steroidal anti-inflammatory drugs or aspirin, major medical disorders including uncontrolled depression and anxiety, a history of infertility, or were undergoing or planning to utilize fertility therapies (Schisterman et al., 2013, 2014). Women were followed for up to six cycles while attempting to conceive, and throughout pregnancy if they conceived. Fertility monitors (ClearBlue Easy; Inverness Medical) were used to coordinate intercourse timing and to schedule follow-up clinic visits (Schisterman et al., 2013, 2014; Mumford et al., 2016). In Fig. 1, we provide a visual of participant follow-up over the study documenting pregnancy, pregnancy loss, adverse pregnancy outcomes, and withdrawal.

Figure 1.

Flow chart of participant status over follow-up.  ^aOf 133 pregnancy losses detected after the 6–7 weeks’ ultrasound, 127 were confirmed to have implanted in the uterus and were considered for analyses. Numbers do not account for all pregnancy endpoints.

Ethical approval

The study protocol was approved by the institutional review board at each study site and data coordinating center and all participants provided informed consent prior to study enrollment.

Assessment of preconception sleep characteristics

Preconception sleep characteristics were measured at baseline via self-report on a lifestyle questionnaire of habitual sleep patterns specific to weekdays and weekends. Sleep characteristics included habitual bedtime, wake time, and sleep latency. Derived sleep characteristics were calculated using these data and included sleep duration, sleep midpoint, and social jetlag. Details on sleep characteristics are provided in Supplementary Table S1. In brief, sleep duration, a measurement for the quantity of sleep, was calculated from the difference in usual bedtimes and wake times, and further subtracted away sleep latency, or how long it took participants to fall asleep (Watson et al., 2015). Sleep duration was weighted for weekdays and weekends. We additionally evaluated sleep latency individually, as a measure of sleep quality, which was also weighted for weekdays and weekends (Buysse et al., 1989; Grandner et al., 2006). Sleep midpoint, defined as the midway clock time between sleep onset and waking up, is often used to study chronotype, or the tendency to be a morning person or evening person (Roenneberg et al., 2003, 2004, 2019). Sleep midpoint was calculated based on the Munich Chronotype Questionnaire assessment of chronotype and assumed weekend days represented ‘free’ days; this calculation takes the average weekend sleep midpoint and subtracts one-half the difference between weekend and average sleep duration (Roenneberg et al., 2003, 2004, 2019). This method accounts for sleep debt accrued on weekdays, but if women slept longer on weekdays than weekends then sleep midpoints were not corrected (Roenneberg et al., 2019). We also calculated social jetlag, a measure of sleep variability between weekdays and weekends. Social jetlag reflects chronic circadian misalignment or the difference between circadian rhythms and social schedules (e.g. work, school, and social life demands), which may impact sleep–wake tendencies (Wittmann et al., 2006; Roenneberg et al., 2019). Social jetlag is calculated by taking the absolute difference between weekend and weekday sleep midpoints.

Assessment of pregnancy loss and adverse pregnancy outcomes

Pregnancies were determined using urine hCG pregnancy tests (Quidel Quickvue, Quidel Corporation), conducted at home or in clinic near the time of expected menses and were supplemented with information from first-morning urine specimens collected daily over the first two study cycles and spot urine pregnancy tests conducted during clinic visits that occurred monthly during the study (Schisterman et al., 2013, 2014; Mumford et al., 2016). Ongoing pregnancies at 6–7 weeks were confirmed with clinical ultrasound. Pregnancy losses were identified using this information and included hCG-detected pregnancy losses and clinical pregnancy losses (Mumford et al., 2016). HCG-detected pregnancy loss was defined as: (i) positive urine hCG pregnancy tests followed by the absence of pregnancy at the clinical ultrasound at 6–7 weeks’ pregnancy, or (ii) positive free β-hCG tests from batched augmented assays of urine samples followed by the absence of pregnancy from pregnancy tests conducted at home or in clinic (Mumford et al., 2016). Clinical pregnancy losses were losses that were observed after the clinical ultrasound confirmation of pregnancy.

Date of delivery, HDP, including preeclampsia, and GDM were ascertained from phone interviews conducted postpartum and from medical record abstraction (Schisterman et al., 2013; Silver et al., 2015). Medical record abstraction was completed among women who became pregnant over follow-up to ascertain clinical outcomes. Gestational age was based on the 6- to 7-week clinical visit ultrasounds or study-provided, home-based fertility monitor peak day or last menstrual period, or clinical chart abstraction. PTB was defined as any birth that occurred between 20 weeks–0 days and 36 weeks–6 days of gestation (Silver et al., 2015).

Assessment of covariates

In EAGeR, participant demographics, lifestyle factors, and reproductive histories were obtained from questionnaires completed at baseline (Schisterman et al., 2013). BMI was calculated from weight (kg) and height (m) measurements, which were measured using standardized protocols. Smoking status was assessed via urine cotinine biomarkers (Mumford et al., 2021). Biomarkers of antidepressant, marijuana, and opioid use were obtained from urine samples collected at baseline. Positive screens were determined by the Drug of Abuse IV Ultra chemiluminescence-based immunoassay on the Evidence Investigator benchtop analyzer (Randox Toxicology Ltd., County Antrim, UK) using manufacture-based cutoffs (Mumford et al., 2021). Current employment status was assessed at baseline (not employed, full time, part time, or other). Participants were also asked whether their current or most recent job involved night work and rotating shift work.

Statistical analysis

Participant characteristics were summarized by categories of preconception sleep duration. Log-Poisson models were used to estimate relative risks (RR) and 95% CIs for associations between preconception sleep characteristics, pregnancy loss, and adverse pregnancy outcomes. We examined sleep duration (<7, 7 to <9 (ref), ≥9 h), sleep latency (<30 (ref) vs ≥30 min), sleep midpoint (tertiles; Tertile 1: Median (M): 2:45 a.m., interquartile range (IQR) 1: 2:20 a.m.–3:00 a.m.; Tertile 2 (ref): M: 3:36 a.m., IQR: 3:25 a.m.–3:48 a.m.; Tertile 3: M: 4:40 a.m., IQR: 4:21 a.m.–5:17 a.m.), and social jetlag (≤1 (ref) vs >1 h) categorically. For pregnancy loss analyses, we additionally stratified by pregnancy loss type (i.e. hCG-detected losses and clinical pregnancy losses). Given there were few PTB, HDP, and GDM cases, and that these outcomes may be related to preconception sleep characteristics through similar mechanisms, we evaluated a composite outcome of PTB, HDP, and GDM (Hall et al., 2005; Ratajczak et al., 2010; Lateef and Akintubosun, 2020; Beroukhim et al., 2022; Cui et al., 2022). We also conducted exploratory analyses to evaluate PTB and HDP separately. Due to a limited number of GDM cases, we did not evaluate GDM as an individual outcome in exploratory analyses. Furthermore, all analyses of adverse pregnancy outcomes utilized a binary form of sleep duration (<7 vs ≥7 h (ref)) given the limited number of cases with sleep duration ≥9 h. Pregnancy loss and adverse pregnancy outcomes models were adjusted for a priori selected confounders including age, BMI, parity, opioid use, marijuana use, antidepressant use, education, smoking (i.e. cotinine), employment, alcohol consumption, caffeine consumption, season at baseline, exercise, and time since last pregnancy loss. We used multiple imputation to address missing exposure and covariate information and accounted for covariance between imputed datasets (Austin et al., 2021).

Models of pregnancy loss and adverse pregnancy outcomes are based on multiple conditional probabilities including: (i) remaining in the cohort, (ii) becoming pregnant, and (iii) remaining pregnant over follow-up to clinical confirmation or ≥20 weeks’ gestation (Mumford et al., 2016; Purdue-Smithe et al., 2021). Sleep during the preconception period may influence each step of this process and may give rise to selection bias. To address this, we used stabilized inverse probability weights to account for this potential selection bias as relevant for a given analysis (Hernan et al., 2004; Cole and Hernan, 2008; Mumford et al., 2016; Purdue-Smithe et al., 2021). The models of stabilized inverse probability weights for early withdrawal, pregnancy loss, clinical pregnancy loss, and adverse pregnancy outcomes included sleep characteristics, age, BMI, history of previous pregnancy losses, number of live births, treatment group, marital status, and parity. Weights were multiplied to account for potential selection biases from conditioning on multiple probabilistic events in each analysis and were truncated at the 1st and 99th percentiles in models of clinical pregnancy loss and adverse pregnancy outcomes (Cole and Hernan, 2008). Specifically, analyses of all pregnancy losses and hCG pregnancy losses included weights to account for withdrawal and pregnancy, and analyses of clinical pregnancy losses included weights to address withdrawal, pregnancy, and clinical confirmation of pregnancy. Analyses of adverse pregnancy outcomes included weights to address withdrawal, pregnancy, and pregnancies lasting to at least 20 weeks’ gestation. In Supplementary Table S2, we provide summary statistics of the inverse probability weights as relevant for each analysis.

We also conducted several sensitivity analyses. As shift work may confound associations of interest, we excluded women with a current or recent job that involved shift work. Preconception sleep, pregnancy loss, and adverse pregnancy outcomes associations may also differ within specific age groups or by parity. Thus, we evaluated statistical interactions with parity and age, and also present stratified analyses (nulliparous and parous; age <30 and ≥30 years). Although participant stress levels were assessed through daily diaries during the study, these measurements occurred after the baseline measurement of sleep. Daily stress was not utilized for confounder control in these analyses as daily stress may be considered part of the causal pathway. However, usual, or historical, stress levels could be considered a potential confounder. Since baseline assessments of usual stress levels were unavailable in EAGeR, we conducted an e-value analysis to estimate the strength of association that a potential confounder would require to fully explain the observed point estimates and to shift the CI to the null (VanderWeele and Ding, 2017; Mathur et al., 2018). Statistical analyses were conducted using SAS v 9.4 software (SAS Institute Inc., Cary, NC, USA).

Results

Participant characteristics

Among 1228 women who enrolled in EAGeR, 797 women became pregnant, of whom 732 had a clinically confirmed pregnancy; 598 women were pregnant at 20 weeks’ gestation and had data to determine gestational age. In total, 140 women withdrew over follow-up (Fig. 1). In Table 1, sleep characteristics and covariates are presented by sleep duration categories (i.e. <7, 7 to <9, ≥9 and ≥7 h) among 1220 women with complete sleep data. Women who slept ≥9 h were younger, had lower BMI, and lower educational attainment compared to women who slept 7 to <9 h. Women who slept ≥9 h were more likely to use opioids, marijuana, and antidepressants compared to women with 7 to <9 h sleep duration. Women who slept <7 h or ≥9 h were more likely to withdraw from the study, and to smoke compared to women who slept 7 to <9 h.

Table 1.

Characteristics of participants (N = 1220) by preconception sleep duration in EAGeR (2006–2012)^a.

	Sleep Duration
	Sleep <7 h	Sleep 7 to <9 h	Sleep ≥9 h	Sleep ≥7 h
Characteristics	n = 180	n = 906	n = 134	n = 1040
	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)
Age (years)	28.9 (5.3)	29.1 (4.6)	26.4 (4.5)	28.7 (4.7)
BMI (kg/m2)	27.7 (7.3)	26.2 (6.3)	25.0 (6.2)	26.1 (6.3)
Sleep latency (min)b	27.1 (12.7, 60)	15 (10, 30)	15 (9.3, 30)	15 (9.7, 30)
Sleep midpoint (time)b	3:41 a.m.	3:33 a.m.	4:10 a.m.	3:35 a.m.
	(3:07 a.m., 4:39 a.m.)	(2:55 a.m., 4:13 a.m.)	(3:23 a.m., 5:17 a.m.)	(3:00 a.m., 4:18 a.m.)
Social jetlag (min)b	60 (0, 105)	45 (0, 90)	45 (0, 83)	45 (0, 90)
	n (%)	n (%)	n (%)	n (%)
Number of previous live births
0	80 (44.4)	418 (46.1)	68 (50.8)	486 (46.7)
1	62 (34.4)	330 (36.4)	49 (36.6)	379 (36.4)
2	38 (21.1)	158 (17.4)	17 (12.7)	175 (16.8)
Number of previous losses
1	122 (67.8)	612 (67.6)	86 (64.2)	698 (67.1)
2	58 (32.2)	294 (32.5)	48 (35.8)	342 (32.9)
Time since last pregnancy loss (months)
<9	124 (70.9)	643 (72.0)	99 (74.4)	742 (72.3)
≥9	51 (29.1)	250 (28.0)	34 (25.6)	284 (27.7)
Withdrew	27 (15.0)	81 (8.9)	27 (20.2)	108 (10.4)
Treatment group
Placebo	87 (48.3)	460 (50.8)	64 (47.8)	524 (50.4)
Aspirin	93 (51.7)	446 (49.2)	70 (52.2)	516 (49.6)
Married	157 (87.2)	847 (93.5)	112 (83.6)	959 (92.2)
Parity
Nulliparous	75 (41.7)	389 (42.9)	57 (42.5)	446 (42.9)
1	63 (35.0)	319 (35.2)	49 (36.6)	368 (35.4)
2+	42 (23.3)	198 (21.9)	28 (20.9)	226 (21.7)
More than high school education	146 (81.1)	807 (89.1)	99 (74.4)	906 (87.2)
Marijuana usec	—d	38 (4.2)	13 (9.7)	51 (4.9)
Opioid usec	18 (10.0)	54 (6.0)	13 (10.1)	67 (6.6)
Antidepressant usec	20 (11.1)	154 (17.0)	32 (23.9)	186 (17.9)
Smoker (cotinine)	36 (20.0)	84 (9.40)	21 (16.3)	105 (10.3)
No alcohol consumption in past year	116 (65.2)	599 (66.5)	91 (68.4)	690 (66.7)
Caffeine consumption
Nondrinker	34 (18.9)	238 (26.3)	33 (24.6)	271 (26.1)
1–3 cups/day	107 (59.4)	533 (58.8)	73 (54.5)	606 (58.3)
≥3 cups/day	39 (21.7)	135 (14.9)	28 (20.9)	163 (15.7)
Exercise
Low	36 (20.0)	237 (26.2)	47 (35.1)	284 (27.3)
Moderate	68 (37.8)	384 (42.4)	44 (32.8)	428 (41.2)
High	76 (42.2)	285 (31.5)	43 (32.1)	328 (31.5)
Employment
Not employed (e.g. students, homemakers/parents, temporarily unemployed, unable to work due to disability, other)	31 (18.1)	207 (23.4)	49 (39.2)	256 (25.4)
Currently employed part time or full time	140 (81.9)	676 (76.6)	76 (60.8)	752 (74.6)
Rotating shift worke	34 (20.1)	138 (15.7)	34 (27.4)	172 (17.1)
Night shift worke	47 (28.0)	182 (20.7)	44 (35.2)	226 (22.5)
Season at baseline
Winter	36 (20.0)	201 (22.2)	40 (29.9)	241 (23.2)
Spring	46 (25.6)	269 (29.7)	33 (24.6)	302 (29.0)
Summer	48 (26.7)	195 (21.5)	34 (25.4)	229 (22.0)
Fall	50 (27.8)	241 (26.6)	27 (20.2)	268 (25.8)

^aData in the table presented as complete case data prior to missing data imputation. Variables with no missingness: Age, Number of Previous Live Births, Number of Previous Losses, Withdrew, Treatment Group, Married, Parity, Marijuana Use, Antidepressant Use, Season at baseline, Any Pregnancy Loss, hCG Pregnancy Loss, Clinical Pregnancy Loss, Composite of Adverse Pregnancy Outcomes, Preterm Birth, Hypertensive Disorders of Pregnancy, Gestational Diabetes. Variables with missingness >0 to <1%: Sleep Duration, Sleep Latency, Sleep Midpoint, Social Jetlag, More Than High School Education, Exercise, Caffeine Consumption, Rotating Shift Work^e, Night Shift Work^e. Variables with missingness between 1% and <5%: BMI, Time Since Last Pregnancy Loss, Opioid Use, Cotinine, Alcohol Consumption in the Past Year, and Employment.

^bMedian (Quartile 1, Quartile 3).

^cMarijuana, opioid, and antidepressant metabolites were measured from urine samples collected at baseline using the Drug of Abuse IV Ultra chemiluminescent immunoassay measured on the Evidence Investigator (Randox Toxicology, County Antrim, UK). Marijuana was also assessed via self-report on the baseline questionnaire. Endorsement of marijuana use, or positive screening tests were used to classify marijuana use.

^dCell size was omitted due to small numbers.

^eRestricted to participants with complete employment data.

Pregnancy loss

Overall, 24% (188/797) of pregnant women in EAGeR experienced a pregnancy loss, of which 29% (55/188) were hCG pregnancy losses and 71% (133/188) were clinically recognized or ectopic pregnancy losses. After adjustment for confounders, preconception sleep duration, sleep latency, sleep midpoint, and social jetlag were not associated with risk of pregnancy loss (Table 2). When stratified by type of loss (i.e. hCG-detected or clinically recognized pregnancy losses), the associations remained similarly null (Supplementary Table S3).

Table 2.

Association between preconception sleep characteristics and any pregnancy loss (n = 188) in EAGeR (2006–2012)^a,b.

		Unadjusted			Multivariable adjustedc
Any pregnancy loss	n (%)	RR	95% CI		RR	95% CI
Sleep duration
<7 h (n = 111)	29 (26.1%)	1.04	0.71	1.53	0.94	0.63	1.41
7 to <9 h (n = 592)	135 (22.8%)	Ref	—	—	Ref	—	—
≥9 h (n = 80)	23 (28.8%)	1.22	0.82	1.81	1.21	0.80	1.85
Sleep latency
≤30 min (n = 676)	158 (23.4%)	Ref	—	—	Ref	—	—
>30 min (n = 106)	28 (26.4%)	1.15	0.80	1.65	1.11	0.76	1.62
Sleep midpoint
Tertile 1 (n = 272)	60 (22.1%)	0.97	0.68	1.38	0.96	0.67	1.37
Tertile 2 (n = 270)	64 (23.7%)	Ref	—	—	Ref	—	—
Tertile 3 (n = 241)	63 (26.1%)	1.05	0.75	1.47	1.06	0.74	1.50
Social jetlag
≤1 h (n = 520)	123 (23.7%)	Ref	—	—	Ref	—	—
>1 h (n = 263)	64 (24.3%)	1.05	0.79	1.40	0.93	0.68	1.27

^aNumbers and percentages reflect complete case data; imputed data for missing sleep and covariates were used in this analysis.

^bWeighted for loss to follow-up and pregnancy; denominator is 785 women at risk of any pregnancy loss who did not withdraw from the study.

^cAdjusted for: age (continuous; years), BMI (continuous; kg/m²), parity (categorical; Nulliparous, 1, ≥2), opioid use (binary; Yes, No), marijuana use (binary; Yes, No), antidepressant use (binary; Yes, No), education (binary; <High School, ≥High School), smoking (cotinine; binary; Yes, No), employment (binary; Currently Employed, Not Employed), alcohol consumption (binary; Yes, No), caffeine consumption (categorical; Nondrinker, 1–3 cups/day, >3 cups/day), season at baseline (categorical; Fall, Winter, Spring, Summer), exercise (categorical; Low, Moderate, High), and time since last pregnancy loss (categorical; ≤4, 5–8, 9–12, and >12 months).

Adverse pregnancy outcomes

Overall, 20% (118/598) of women with pregnancies lasting ≥20 weeks experienced an adverse pregnancy outcome. Of the 598 women, 9% (53/598) had a PTB, 10% (62/598) experienced a HDP, and 4% (22/598) developed GDM.

Sleep duration <7 h was not associated with risk of the composite adverse pregnancy outcome (vs ≥7 h; RR: 1.19, 95% CI: 0.74, 1.89; Table 3). When evaluating HDP and PTB separately, sleep duration <7 h was not associated with HDP, but sleep duration <7 h was associated with greater risk of PTB (vs ≥7 h; RR: 2.74, 95% CI: 1.43, 5.23; Supplementary Table S4). Sleep latency >30 min was not associated with the composite adverse pregnancy outcome (vs ≤30 min; RR: 0.80, 95% CI: 0.50, 1.29; Table 3) or any individual adverse pregnancy outcome (Supplementary Table S4).

Table 3.

Association between preconception sleep characteristics and the composite of adverse pregnancy outcomes (n = 118) in EAGeR (2006–2012)^a,b.

		Unadjusted			Multivariable Adjustedc
Composite outcome	n (%)	RR	95% CI		RR	95% CI
Sleep duration
<7 h (n = 81)	19 (23.5%)	1.39	0.94	2.04	1.19	0.74	1.89
≥7 h (n = 516)	98 (19.0%)	Ref	—	—	Ref	—	—
Sleep latency
≤30 min (n = 519)	103 (19.9%)	Ref	—	—	Ref	—	—
>30 min (n = 78)	14 (18.0%)	0.89	0.56	1.39	0.80	0.50	1.29
Sleep midpoint
Tertile 1 (n = 213)	34 (16.0%)	0.67	0.44	1.03	0.63	0.40	0.98
Tertile 2 (n = 205)	44 (21.5%)	Ref	—	—	Ref	—	—
Tertile 3 (n = 179)	39 (21.8%)	0.86	0.59	1.24	0.95	0.63	1.44
Social jetlag
≤1 h (n = 397)	64 (16.1%)	Ref	—	—	Ref	—	—
>1 h (n = 200)	53 (26.5%)	1.96	1.40	2.73	1.65	1.11	2.44

^aNumbers and percentages reflect complete case data; imputed data for missing sleep and covariates were used in this analysis.

^bWeighted for loss to follow-up, pregnancy, and women who were still pregnant at ≥20 weeks’ gestation; denominator is 598 women at risk of an adverse pregnancy outcome who did not withdraw from the study.

^cAdjusted for: age (continuous; years), BMI (continuous; kg/m²), parity (categorical; Nulliparous, 1, ≥2), opioid use (binary; Yes, No), marijuana use (binary; Yes, No), antidepressant use (binary; Yes, No), education (binary; <High School, ≥High School), smoking (cotinine; binary; Yes, No), employment (binary; Currently Employed, Not Employed), alcohol consumption (binary; Yes, No), caffeine consumption (categorical; Nondrinker, 1–3 cups/day, >3 cups/day), season at baseline (categorical; Fall, Winter, Spring, Summer), exercise (Categorical; Low, Moderate, High), and time since last pregnancy loss (categorical; ≤4, 5–8, 9–12, and >12 months).

Earlier sleep midpoints were associated with lower risk of the composite adverse outcome (first vs second tertile; RR: 0.63, 95% CI: 0.40, 0.98), but later sleep midpoints were not associated with risk of the composite outcome (third tertile vs second tertile; RR: 0.95, 95% CI: 0.63, 1.44; Table 3). When evaluating individual outcomes, the magnitudes of association with earlier sleep midpoints were similar for PTB and HDP, but CIs were wide (PTB RR: 0.54, 95% CI: 0.26, 1.13; HDP RR: 0.60, 95% CI: 0.32, 1.13; Supplementary Table S4). Social jetlag was associated with greater risk of the composite adverse outcome (>1 vs ≤1 h; RR: 1.65, 95% CI: 1.11, 2.44; Table 3). In analyses of individual adverse outcomes (Supplementary Table S4), associations with social jetlag were similar for HDP (>1 vs ≤1 h; RR: 1.82, 95% CI: 1.08, 3.07). Social jetlag was not associated with PTB (>1 vs ≤1 h; RR: 0.96, 95% CI: 0.52, 1.79; Supplementary Table S4).

Sensitivity analyses

Excluding women who reported that their current or most recent job involved night or rotating shift work did not change the primary findings (Supplementary Table S5). There was a significant interaction of parity only with social jetlag and the composite adverse pregnancy outcome (P = 0.02). In stratified analyses, we found that the association between preconception social jetlag and the composite adverse pregnancy outcome was stronger in magnitude of association among nulliparous women (>1 vs ≤1 h; RR: 2.48, 95% CI: 1.27, 4.84), but was attenuated among parous women (>1 vs ≤1 h; RR: 0.82, 95% CI: 0.45, 1.50). There was no consistent evidence of age-related effect modification based on tests for interaction with age or stratification by age (<30 and ≥30 years).

E-value analyses revealed that an unmeasured confounder, like usual stress, would need to have an RR of 1.16 and 2.55 to shift the preconception sleep midpoint and composite adverse pregnancy outcome association upper confidence bound to the null, and to fully explain the point estimate, respectively (Supplementary Table S6). For the preconception social jetlag and composite of adverse pregnancy outcomes association, an unmeasured confounder would need an RR of 1.46 and 2.69 to shift the lower confidence bound to the null, and to fully explain the point estimate, respectively (Supplementary Table S6).

Discussion

This study examined preconception sleep, pregnancy loss, and adverse pregnancy outcomes in a well-characterized cohort of women with a history of pregnancy loss. Preconception sleep duration, sleep latency, sleep midpoint, and social jetlag were not associated with pregnancy loss. However, preconception sleep timing measures, including sleep midpoint and social jetlag, were associated with the composite adverse pregnancy outcome. In exploratory analyses, associations of preconception sleep midpoint and the individual outcomes of HDP and PTB were largely similar in magnitude and direction, though estimates were imprecise due to small number of cases. We also found that associations with preconception social jetlag and adverse pregnancy outcomes were attenuated among parous women, but were stronger among nulliparous women. This finding may suggest that parous women may have more consistent sleep timing and thus less social jetlag. Alternatively, this finding may be due to characteristics of women in the EAGeR study instead of an interaction; additional studies are needed to replicate this finding. Overall, these results suggest that preconception sleep timing may be a risk factor for adverse pregnancy outcomes but may be less relevant for pregnancy loss.

Prior work on sleep and risk of pregnancy loss has been scant. A single case–control study reported that sleep duration ≤8 h/day in mid-pregnancy was associated with greater risk of pregnancy loss in both the first and second trimesters (Samaraweera and Abeysena, 2010). However, these results may be biased as most pregnancy losses occur in early gestation, and retrospective studies evaluating mid-pregnancy sleep and pregnancy loss are unable to capture these early pregnancy losses (Wilcox et al., 1988, 1999; Weinberg et al., 1992; Zinaman et al., 1996). Recent findings from the Pregnancy Study Online (PRESTO), a prospective cohort which measured preconception sleep duration, suggest that preconception sleep duration was not associated with risk of pregnancy loss (Bond et al., 2024). The findings in EAGeR for preconception sleep duration and pregnancy loss are consistent with this study. However, the study from PRESTO also reported a modest, suggestive association between male partner self-reported sleep duration of <6 h (vs 7 to <9 h) and higher risk of pregnancy loss (Bond et al., 2024). Unfortunately, male partner characteristics were not assessed in EAGeR, and additional studies should consider assessing sleep in both partners and evaluating risk of pregnancy loss.

We evaluated associations between preconception sleep duration, sleep latency, and adverse pregnancy outcomes and found no overall association. In exploratory sub-analyses, sleep duration <7 h was associated with PTB risk only. A prior study of retrospectively recalled preconception sleep duration and mid-pregnancy sleep duration found no association with PTB risk (Nakahara et al., 2020). Preconception sleep latency was not associated with adverse pregnancy outcomes. However, studies evaluating mid-pregnancy poor sleep quality using detailed sleep quality scales have observed associations with adverse pregnancy outcomes (Cai et al., 2017; Wang and Jin, 2020; Ma et al., 2024). Additional studies are needed to supplement these findings and should consider evaluating multiple dimensions of sleep quality using validated instruments during the preconception period.

We found that measures of preconception sleep midpoint and social jetlag may be relevant to adverse pregnancy outcomes. Specifically, early sleep midpoints were associated with lower risk of the composite adverse pregnancy outcome and findings were similar in direction and magnitude of association among both PTB and HDP. These findings are consistent with results published in the NuMoM2b cohort, which found that later sleep midpoints in mid-pregnancy were associated with greater risk of PTB and GDM (Facco et al., 2017, 2018, 2019). To date, social jetlag has rarely been evaluated in the context of adverse pregnancy outcomes; though, social jetlag has been associated with obesity, which is a risk factor for many adverse pregnancy outcomes (Poston et al., 2011; Hashemipour et al., 2022; Arab et al., 2024). Overall, the results from these exploratory analyses underscore the importance of sleep midpoint and social jetlag for adverse pregnancy outcomes. However, findings were primarily based on a composite outcome, and only sleep midpoint associations generally replicated among all individual outcomes comprising the composite outcome. These findings should be confirmed in additional prospective studies with a larger number of adverse pregnancy outcomes.

Mechanistic evidence linking preconception sleep to pregnancy loss and adverse pregnancy outcomes is sparse and based on research in animal models and humans. Preconception sleep may be related to adverse pregnancy outcomes through mechanisms in early pregnancy that involve reproductive hormones, which have circadian rhythmicity, and circadian clock mechanisms along the hypothalamus-pituitary-ovary axis and the uterus (Hall et al., 2005; Lateef and Akintubosun, 2020; Beroukhim et al., 2022). Evidence from animal models and human tissue samples has shown that clock genes are found in the uterus, placenta, and fetal tissues (Ratajczak et al., 2010; Cui et al., 2022). This may suggest that sleep and circadian disruption could influence reproductive hormone and inflammatory cytokine signaling with clock mechanisms that participate in key processes during pregnancy, gestation, and birth (Ratajczak et al., 2010; Cui et al., 2022). Sleep is linked with innate immune function and findings from recent work in a cross-sectional study have highlighted that later sleep timing (i.e. both later sleep midpoint and greater social jetlag) is associated with greater inflammatory cytokine levels among adults after adjustment for the apnea–hypopnea index (Irwin and Opp, 2017; Girtman et al., 2022). Imbalance in inflammatory processes is hypothesized to contribute to PTB and preeclampsia (Peltier, 2003; Harmon et al., 2016). However, more direct evidence is needed to mechanistically link sleep timing, social jetlag, inflammatory profiles, and adverse pregnancy outcomes. Preconception sleep may serve as an indicator of the influence of sleep on early pregnancy health before pregnancy induces changes in sleep patterns, due to the release of hormones that promote daytime sleepiness and sleep fragmentation, and due to pregnancy-related physiological changes that impact sleep (Mindell et al., 2015; Won, 2015). However, preconception sleep may also proxy mid-pregnancy sleep. Further research is needed to understand sleep pattern trajectories and their relationship with pregnancy health from preconception through birth.

This study has many strengths. The study design of EAGeR allowed for examination of preconception sleep and for detection of early pregnancy losses through high sensitivity β-hCG testing. There was minimal participant dropout and we used stabilized inverse probability weights to address potential selection biases from conditioning on sequential events that occur in human reproduction (Hernan et al., 2004; Cole and Hernan, 2008; Mumford et al., 2016; Purdue-Smithe et al., 2021). Furthermore, we calculated sleep debt corrected sleep midpoints as a proxy of chronotype, which has moderate–strong correlation with the gold standard measure of chronotype, dim light melatonin onset (Kitamura et al., 2014; Kantermann et al., 2015).

This study also has limitations. Self-reported sleep was based on a single assessment during preconception and no measures were available during pregnancy; however, we expect this to be non-differentially misclassified with respect to outcomes given sleep was measured prior to pregnancy and is not subject to potential changes in sleep due to pregnancy (Won, 2015). Self-reported sleep characteristics may be reported with error and leveraging more objective, actigraphy-based assessments may help overcome these limitations (Cespedes et al., 2016; St-Onge et al., 2019; Jackson et al., 2020; Zak et al., 2022). The measures of sleep midpoint and social jetlag assumed that weekend sleep midpoints represent the midpoints on ‘free’ days (Roenneberg et al., 2004, 2019; Wittmann et al., 2006). Participant stress was assessed via a daily diary that started after the baseline assessment of sleep in EAGeR. As such, these acute measures of daily perceived stress may be on the causal pathway from usual sleep to pregnancy loss and adverse pregnancy outcomes. Thus, we did not adjust for daily stress in these analyses. There may also be some residual confounding by rotating and night shift work given the questionnaire only asked about current or most recent jobs. However, findings from e-value analyses suggest that confounding from stress or shift work would need to be strong to attenuate or fully explain the key findings. Sleep and individual adverse pregnancy outcomes associations were based on few cases and could be due to chance. To address this limitation and improve statistical precision, we combined each adverse pregnancy outcome into a composite outcome, hypothesizing that sleep is related to the individual outcomes through similar pathways. The study may have been underpowered to detect interactions between age and parity with some of the primary findings. The EAGeR population was comprised of mostly healthy, white women attempting pregnancy who had a history of pregnancy loss. Women with a history of pregnancy loss may reflect a higher risk group for pregnancy loss and adverse pregnancy outcomes compared with women with no history of pregnancy loss (Ahrens et al., 2016; Magnus et al., 2019). The findings are only generalizable to women with a history of pregnancy loss; however, pregnancy loss is common and affects a large proportion of reproductive-age women.

Preconception sleep was not associated with risk of pregnancy loss, but preconception sleep duration and timing may be relevant for adverse pregnancy outcomes. Importantly, this work extends upon prior studies by evaluating preconception sleep characteristics, which allows for examination of sleep before pregnancy-induced changes to sleep patterns, and for assessment of early pregnancy losses from a preconception timepoint (Weinberg et al., 1992; Mindell et al., 2015; Won, 2015). These findings highlight the importance of understanding sleep during the preconception period for adverse pregnancy outcomes. Further research is needed to leverage actigraphy-based sleep assessments collected longitudinally from preconception through pregnancy and postnatal periods among a diverse cohort of reproductive-aged women to confirm and supplement these findings.

Data availability

The data underlying this article have been provided by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), National Institutes of Health under a data use agreement. We do not have permission to share it with other individuals, within or outside the primary author’s Institution, or with commercial enterprises. Data from the EAGeR study are available to approved researchers through a data use agreement. Information about the study and data are available here: https://www.nichd.nih.gov/about/org/dir/dph/officebranch/eb/effects-aspirin. Data used in this study were created using pre-existing questionnaires, and derived variables from this data were previously returned to NICHD contacts for this project. This work was supported by the Intramural Research Program of the Eunice Kennedy Shriver National Institute of Child Health and Human Development (Contract numbers: HHSN267200603423, HHSN267200603424, HHSN267200603426, HHSN275201300023I).

Authors’ roles

J.R.F., B.W.W., E.R.B.-J., and S.L.M. planned the study. The analytic strategy was designed by J.R.F., B.W.W., E.R.B.-J., and S.L.M. The statistical analysis was performed by J.R.F. J.R.F. wrote the first draft. All authors interpreted the data, supported discussion, and critically revised the manuscript. All authors approved and accepted responsibility for the final manuscript.

Funding

J.R.F. and this work were supported by the Intramural Research Program Cancer Research Training Award, National Cancer Institute, National Institutes of Health (ZIA CP010197), and the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland (Contract numbers: HHSN267200603423, HHSN267200603424, HHSN267200603426, HHSN275201300023I). R.M.S. received NIH funding through the listed contracts as site-PI for the original EAGeR trial at the University of Utah.

Conflict of interest

L.M.O. reports receiving funding from the Star Legacy Foundation (paid to institution); an advisory board role at the Star Legacy Foundation; and receiving travel support from the Star Legacy Foundation. G.L.D. reports a role as Associate Editor at Human Reproduction and a role on the Journal Editorial Board of SLEEP. A.C.P.-S. is an employee of Merck & Co. and has received stock compensation as an employee of Merck & Co. in the past 36 months. The work in this manuscript was completed before A.C.P.-S.’s employment at Merck & Co. and is unrelated to A.C.P.-S.’s work at the company. R.M.S. reports royalties or licenses from BJOG and UpToDate, Inc. in the past 36 months, receiving payment or honoraria for Grand Rounds in the past 36 months, and participating on a Data Safety Monitoring Board or Advisory Board for a National Institutes of Health-funded Apple Trial in the past 36 months. The other authors report there are no competing interests to declare.