Prospectively assessed perceived stress associated with early pregnancy losses among women with history of pregnancy loss

Abstract

STUDY QUESTION

What is the association between perceived stress during peri-conception and early pregnancy and pregnancy loss among women who have experienced a prior pregnancy loss?

SUMMARY ANSWER

Daily perceived stress above the median is associated with over a 2-fold risk of early pregnancy loss among women who have experienced a prior loss.

WHAT IS KNOWN ALREADY?

Women who have experienced a pregnancy loss may be more vulnerable to stress while trying to become pregnant again. While prior research has indicated a link between psychological stress and clinically confirmed miscarriages, research is lacking among a pre-conceptional cohort followed prospectively for the effects of perceived stress during early critical windows of pregnancy establishment on risk of both hCG-detected pregnancy losses and confirmed losses, while considering important time-varying confounders.

STUDY DESIGN, SIZE, DURATION

Secondary data analysis of the EAGeR trial (2007–2011) among women with an hCG-detected pregnancy (n = 797 women).

PARTICIPANTS/MATERIALS, SETTING, METHODS

Women from four US clinical centers enrolled pre-conceptionally and were followed ≤6 cycles while attempting pregnancy and, as applicable, throughout pregnancy. Perceived stress was captured via daily diaries and end-of-month questionnaires. Main outcome measures include hCG-detected and clinically recognized pregnancy losses.

MAIN RESULTS AND THE ROLE OF CHANCE

Among women who had an hCG-confirmed pregnancy, 188 pregnancies (23.6%) ended in loss. Women with high (>50th percentile) versus low (≤50th percentile) peri-implantation or early pregnancy weekly perceived stress had an elevated risk of experiencing any pregnancy loss (hazard ratio (HR): 1.69, 95% CI: 1.13, 2.54) or clinical loss (HR: 1.58, 95% CI: 0.96, 2.60), with higher risks observed for women experiencing an hCG-detected loss (HR: 2.16, 95% CI: 1.04, 4.46). Models accounted for women’s age, BMI, employment, marital status, income, education, race, parity, prior losses, exercise and time-varying nausea/vomiting, caffeine, alcohol and smoking.

LIMITATIONS, REASONS FOR CAUTION

We were limited in our ability to clearly identify the mechanisms of stress on pregnancy loss due to our sole reliance on self-reported perceived stress, and the lack of biomarkers of different pathways of stress.

WIDER IMPLICATIONS OF THE FINDINGS

This study provides new insight on early pregnancy perceived stress and risk of pregnancy loss, most notably hCG-detected losses, among women with a history of a prior loss. Our study is an improvement over past studies in its ability to account for time-varying early pregnancy symptoms, such as nausea/vomiting, and lifestyle factors, such as caffeine, alcohol and smoking, which are also risk factors for psychological stress and pregnancy loss.

STUDY FUNDING/COMPETING INTEREST(S)

This work was supported by the Intramural Research Program of the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland (Contract numbers: HHSN267200603423, HHSN267200603424, HHSN267200603426, HHSN275201300023I). Additionally, K.C.S. was supported by the National Institute on Aging of the National Institutes of Health under Award Number K01AG058781. The authors have no conflicts of interest to disclose.

TRIAL REGISTRATION NUMBER

#NCT00467363.

Introduction

Women who have suffered a pregnancy loss are particularly vulnerable to experiences of stress, anxiety and depressive symptoms as they plan or begin trying again to conceive (Kolte et al., 2015). While pregnancy losses are common, estimated to occur in 20% of all pregnancies and affecting over 40% of women at some point in their reproductive lives (Wilcox et al., 1988; García-Enguídanos et al., 2002; Savitz et al., 2002; Cohain et al., 2017), miscarriages can nevertheless cause significant distress for women, their partners and their families (Kolte et al., 2015). Since pregnancy loss remains largely unexplained, research that can contribute to our understanding of the etiology of pregnancy loss and modifiable factors amenable to interventions to mitigate loss are needed (Qu et al., 2017).

A recent systematic review and meta-analysis of eight prior studies indicated that psychological stress may lead to a 1.42-fold increase in miscarriage probability (Qu et al., 2017). However, limitations of prior studies include being predominately of retrospective nature (O’Hare and Creed, 1995; Bashour and Abdul Salam, 2001; Nelson et al., 2003; Maconochie et al., 2007) with a focus on work-related psychological stress (Brandt and Nielsen, 1992; Fenster et al., 1995; Maconochie et al., 2007). No prior study that we are aware of has recruited women pre-conceptionally and prospectively assessed the effects of perceived stress during early critical windows of pregnancy establishment on risk of both hCG-detected pregnancy losses and confirmed losses, taking into account important time-varying confounders, such as nausea, caffeine, alcohol and smoking (Brandt and Nielsen, 1992; Fenster et al., 1995; O’Hare and Creed, 1995; Hamilton Boyles et al., 2000; Bashour and Abdul Salam, 2001; Nelson et al., 2003; Maconochie et al., 2007; Meaney et al., 2014; Qu et al., 2017).

Several mechanisms have been proposed to explain the potential adverse effects of psychological stress on early pregnancy including activation of the hypothalamic–pituitary–adrenal (HPA) axis with consequent secretion of adrenal cortisol, known to act on decidual and placental metabolism (Parker and Douglas, 2010). Additionally, HPA activation leading to suppression of the hypothalamic–pituitary–gonadal axis may interfere with progesterone synthesis, crucial for the maintenance of early pregnancy (Parker and Douglas, 2010). Despite these postulated mechanisms, evidence of a prospective link between psychological stress and early pregnancy loss is lacking.

To address these large research gaps, we aimed to (i) describe daily perceived stress across the critical and acute windows of pregnancy establishment beginning at conception (i.e. time of ovulation) through delivery among a prospectively followed preconception cohort of women with a prior loss; and (ii) assess the association between daily stress across pregnancy and pregnancy loss, taking into account important time-varying confounders such as pregnancy symptoms and lifestyle factors that are in themselves risk factors for psychological stress and pregnancy loss.

Materials and methods

Study population

The Effects of Aspirin in Gestation and Reproduction (EAGeR) trial (2007–2011), was a multicenter, block-randomized, double-blinded, placebo-controlled trial to evaluate the effect of maternal preconception through pregnancy daily low dose aspirin (81 mg/day) on reproductive outcomes (Schisterman et al., 2013, 2014). Women were eligible if they met the following criteria: aged 18–40 years, experienced 1–2 prior pregnancy losses <20 weeks gestation, had regular menstrual cycles (21–42 days), no known history of infertility, intended to conceive and had no known major medical disorders. Women were enrolled at the following study sites: University of Buffalo (Buffalo, NY), University of Colorado (Denver, CO), University of Utah (Salt Lake City, UT) and Moses Taylor Hospital (Scranton, PA). Eligible women were randomized and assigned to low-dose aspirin (LDA) (81 mg per day) plus folic acid or placebo plus folic acid prior to conception. Twelve hundred twenty-eight women consented and enrolled in the trial; however, 14 women withdrew on the day of randomization leaving a total of 1214 women. Details of the study protocol have been published previously (Schisterman et al., 2013).

Ethics approval

Written informed consent was acquired from all participants and Institutional Review Board approval was provided at each study site and the data coordinating center. The design, methods and participant characteristics of the EAGeR Study have been previously described (Schisterman et al., 2013); the trial was registered with clinicaltrials.gov (number NCT00467363).

Study follow-up

After randomization, participants were followed for up to six menstrual cycles while attempting pregnancy. Participants who became pregnant were followed throughout pregnancy with periods of active and passive follow-up pre-conceptionally and during pregnancy (Hinkle et al., 2016). Women were in active follow-up during the first two menstrual cycles after enrollment. During active follow-up, women completed a preconception daily diary including perceived stress, provided daily urine samples, used a fertility monitor (Clear Blue Easy; Swiss Precision Diagnostics, Geneva, Switzerland) and completed clinic visits in the middle and at the end of each cycle. Women received their preconception and pregnancy daily diaries at baseline randomization visit, with trained staff instructing participants on how to complete the 13-item questionnaire (Supplementary Data File S1), including instructing them to complete the diary for the prior day. Women entered passive follow-up if they did not become pregnant during the first two cycles. During preconception passive follow-up, women used the fertility monitor and completed clinic visits at the end of each cycle, at which they reported their perceived stress experienced during the prior cycle.

After an hCG positive pregnancy test result, women entered active pregnancy follow-up for 4 weeks, during which a pregnancy-specific daily diary assessing perceived stress, nausea/vomiting and various lifestyle factors including caffeine consumption, smoking and alcohol intake were completed (Hinkle et al., 2016). Study ultrasonography was performed during gestational week 6 or 7. We determined gestational age based on ultrasonographic findings. For participants who had a pregnancy loss before ultrasonography could be performed, pregnancy dating was based on the first day of the last menstrual period as recorded with the fertility monitor or the preconception diary. Pregnancy study clinic visits were performed at gestational weeks 8, 12, 20 and 28 and telephone contact at gestational weeks 16, 24, 32 and 36 (Hinkle et al., 2016). This secondary data analysis of the EAGeR trial is limited to the 797 women who had an hCG-detected pregnancy.

Perceived stress assessment

During active follow-up, in their daily preconception and pregnancy diaries, participants reported their daily stress levels via a Likert scale: 0 = no stress, 1 = little stress, 2 = moderate stress and 3 = a lot of stress. The preconception diary was used to quantify perceived stress from gestational week 2 (i.e. conception) until a positive pregnancy test, at which time the pregnancy diary was initiated and used to quantify perceived stress up to Week 8. All diaries were paper forms that were completed at home, returned during the clinic visits, and reviewed by study coordinators at each visit.

From gestational weeks 8 to 36, participants completed monthly questionnaires describing their average daily perceived stress during the prior 4 weeks on a scale from 0 (no stress) to 10 (maximum stress). Only participants who remained pregnant at the time of the questionnaire assessment were asked to complete it for the preceding 4 weeks, ensuring that all data were collected before the pregnancy outcome. For all analyses, daily diary and questionnaire data were averaged into gestational week-level measures and ranked into quartiles. Weekly averages were used in lieu of the daily values to not exclude women in the analyses who missed a daily entry.

Pregnancy outcome

Pregnancies were identified by hCG positive urine pregnancy test and a clinical pregnancy was confirmed by study ultrasonography at 6 to 7 weeks with demonstrated pregnancy signs (e.g. a visible gestational sac, clinical documentation of fetal cardiac activity, or later-stage confirmation of pregnancy) (Schisterman et al., 2015). We categorized pregnancy loss as hCG-detected loss or clinically recognized loss (Mumford et al., 2016). hCG-detected loss was defined as (i) an hCG positive urine pregnancy test result at home or the clinic visit followed by the absence of pregnancy signs at the study ultrasonography, with or without missed menses, or (ii) an hCG positive pregnancy test result via retrospective laboratory analysis with highly sensitive urine testing from daily first morning urine collected at home on the last 10 days of women’s first and second cycle of study participation and on spot urine samples collected at the end of each cycle, followed by the absence of a positive pregnancy test result at home or in the clinic (Schisterman et al., 2015; Mumford et al., 2016). Clinically recognized pregnancy losses were defined as a pregnancy loss after ultrasonographic confirmation. We included ectopic pregnancies; pre-embryonic, embryonic and fetal losses; and stillbirths in our definition of clinically recognized pregnancy loss. No molar pregnancies occurred in this cohort.

Covariate assessment

Baseline data collection and randomization occurred at a study visit on menstrual cycle days 2–4. Women completed questionnaires capturing sociodemographic and reproductive history including age, race, income, student status, attained educational level, employment status, marital status, number of prior pregnancy losses, time since last pregnancy loss, number of prior live births and gestational age of previous fetal loss. Physical activity was reported by participants at baseline using the using the self-administered, last 7-day, short-form version of the International Physical Activity Questionnaire (International Physical Activity Questionnaire, 2005; Russo et al., 2018). Anthropometric measurements including height and weight (used to calculate BMI) were measured at the baseline visit (Schisterman et al., 2013). Time-varying covariates collected in diaries and pregnancy questionnaires included nausea/vomiting, smoking and caffeine and alcohol intake.

Statistical analysis

Variations in perceived stress over pregnancy, Weeks 2 to 8 (during active follow-up) and Weeks 9 to 36 (during passive follow-up), were assessed with the use of linear mixed models. Pairwise comparisons between gestation weeks were made with the use of Bonferroni’s method to account for multiple comparisons. Participant characteristics by early pregnancy (Weeks 2 to 8) average stress quartile were also generated.

The association between weekly perceived stress (quartiles and <median versus ≥median) and the time to pregnancy loss was assessed using discrete Cox proportional hazards regression models. Quartiles were used to standardize the two different stress measures used during active and passive follow-up. Given that studies evaluating psychological stress and pregnancy loss have yet to consider potential thresholds (Bashour and Abdul Salam, 2001; Nelson et al., 2003; Meaney et al., 2014), we examined the possibly nonlinear relationship between continuous stress exposure and pregnancy loss using restricted cubic spline models (with four knots specified). We examined hCG-detected and clinical pregnancy losses in aggregate, then separately. Important time-fixed and time-varying confounding factors, thought to be common causes of our exposure (perceived stress) and outcome (pregnancy loss), were determined based on prior literature (Qu et al., 2017). Our fully adjusted models accounted for the following potential confounding factors: woman’s age (continuous), BMI (continuous), employment (yes/no), marital status (married versus other), income (continuous), education (greater versus less than high school graduate), race (White versus non-White), parity (0, 1, 2), prior losses (1 or 2), physical activity (low, moderate high) (International Physical Activity Questionnaire, 2005) and daily time-varying nausea/vomiting (none, nausea only, vomiting once/day, vomiting more than once/day), caffeinated beverages (yes/no), alcohol (yes/no) and smoking (yes/no). Additional adjustment for blinded treatment status (daily low dose aspirin versus placebo) or income did not alter our findings.

Daily diary adherence for stress exposure was high with most women (64%) completing all 7 days in a given week and 82% completing at least 4 days in a given week. Ninety percent (n = 720/797 women) of women completed daily diary perceived stress assessment during gestational weeks 2 to 7, with at least 87% completing all 13 factors assessed. Missing data for baseline factors were low, with no missing values for women’s age, race, treatment, income, marital status, education, prior number of live births or physical activity. There were 14 observations missing for BMI, 7 missing employment status, 8 missing student status, 8 missing prior loss number and 1 missing smoking or alcohol consumer status. To address potential missing data bias, we used multiple imputation with 50 replicates and included women’s baseline characteristics and relevant longitudinal data for the time-varying covariates for the imputation models. Since 330 women became pregnant during passive preconception follow-up and, thus, did not provide prospectively assessed daily perceived stress data for gestational weeks 2 to 4 until the pregnancy diary was initiated, we used left censoring in our Cox proportional hazards regression models to account for missing by design due to passive follow-up (Cain et al., 2011). Women who withdrew from the study and had an unknown pregnancy outcome were censored at the gestational age of their last study contact.

Secondary analyses were conducted to evaluate potential residual confounding and selection bias. To address unmeasured confounding (e.g. dietary or environmental factors), we used the e-value method (with outcome prevalence set to >15%) (VanderWeele and Ding, 2017; Mathur et al., 2018). The e-value for our hazard ratio (HR) estimate is the minimum strength of association needed for an unmeasured confounder to have with our stress exposure and pregnancy loss outcome to completely explain away any associations found. The e-value for the upper confidence interval is the minimum strength of association an unmeasured confounder would need to have with our stress exposure and pregnancy loss outcome to completely explain away the upper confidence interval (i.e. to include the null). We also used the e-value method to assess potential selection bias by excluding women who did not achieve pregnancy (Supplementary Table SI). The calculated e-value for selection bias determines the strength of association a selection factor would have to have with our stress exposure and pregnancy loss outcome to produce a significant finding when the true association is null (Smith and VanderWeele, 2019).

All analyses were conducted using SAS software (version 9.4; SAS Institute Inc).

Results

Of the 797 women with an hCG pregnancy, 188 pregnancies (24%) ended in a loss. Of the losses, 55 (29%) were hCG-detected losses and 133 (71%) were clinically recognized losses. There was no appreciable change in average perceived stress over the early or later pregnancy period, with the exception of a slight rise in median stress levels after Week 17 (Figs 1 and 2). Median (interquartile range) early pregnancy (Weeks 2 to 8, 0 to 3 scale) perceived stress was 0.67 (0.27–1.13) and later pregnancy perceived stress (Weeks 9 to 36, 0 to 10 scale) was 4.83 (3.43–6.00). The study population had a mean ± SD age of 28.7 ± 4.6 years and were predominately of White race/ethnicity (97%) and parous (58%) (Table I). Women with higher early pregnancy stress (average weeks 2–8) tended to be White, married, more highly educated, have higher nausea/vomiting symptoms and were less likely to consume caffeine.

Figure 1.

Variations in perceived stress during early pregnancy (Weeks 2–8). (a) Boxplot of perceived stress via daily preconception and pregnancy diaries recorded via a Likert scale: 0 = no stress, 1 = little stress, 2 = moderate stress and 3 = a lot of stress. The length of the box represents the interquartile range (the distance between the 25th and 75th percentiles). The symbol in the box interior represents the group mean. The horizontal line in the box interior represents the group median. The vertical lines (whiskers) issuing from the box extend to the group minimum and maximum values. (b) Mean ± SD variations in pregnancy perceived stress. Comparisons were made with the use of linear mixed models to account for repeated measures within women.

Figure 2.

Variations in perceived stress during later pregnancy (Weeks 9–36). (a) Boxplot of perceived stress via monthly questionnaires during pregnancy passive follow-up recorded via a Likert scale: 0 (no stress) to 10 (maximum stress). The length of the box represents the interquartile range (the distance between the 25th and 75th percentiles). The symbol in the box interior represents the group mean. The horizontal line in the box interior represents the group median. The vertical lines (whiskers) issuing from the box extend to the group minimum and maximum values. (b) Mean ± SD variations in pregnancy perceived stress. Comparisons were made with the use of linear mixed models to account for repeated measures within women.

Table I.

Participant characteristics by early pregnancy (Weeks 2–8) perceived stress quartile during active follow-up (n = 720).^a

Characteristic		Perceived early pregnancy stress quartile
	Total (n = 720)	1 0.00–0.26 (n = 179)	2 0.26–0.66 (n = 180)	3 0.67–1.10 (n = 180)	4 1.11–3.0 (n = 181)
Demographics
Age (years), mean ± SD	28.7 ± 4.6	28.8 ± 4.9	28.4 ± 4.7	28.6 ± 4.2	28.7 ± 4.5
Race, n (%)
White	695 (97)	171 (96)	169 (94)	176 (98)	179 (99)
Non-White	25 (3)	8 (5)	11 (6)	4 (2)	2 (1)
BMI (kg/m2), mean ± SD	25.4 ± 5.9	25.5 ± 6.1	25.6 ± 6.4	25.3 ± 5.7	25.0 ± 5.5
Education, n (%)
>High school graduate	645 (90)	156 (87)	155 (86)	167 (93)	167 (93)
≤High school graduate	75 (10)	23 (13)	25 (14)	13 (7)	14 (8)
Marital status, n (%)
Living with partner	25 (3)	7 (4)	9 (5.0)	4 (2)	5 (3)
Married	683 (95)	165 (92)	168 (93)	175 (97)	175 (97)
Other	12 (2)	7 (3)	3 (2)	1 (1)	1 (1)
Income (annual), n (%)
≥$100 000	297 (41)	78 (44)	71 (39)	78 (43)	70 (39)
$75 000–$99 999	105 (15)	22 (12)	27 (15)	26 (14)	30 (17)
$40 000–$74 999	103 (14)	27 (15)	29 (16)	24 (13)	23 (13)
$20 000–$39 999	171 (24)	44 (25)	42 (23)	41 (23)	44 (24)
≤$19 999	44 (6)	8 (5)	11 (6)	11 (6)	14 (8)
Employed, n (%)	517 (73)	123 (70)	133 (75)	133 (75)	128 (71)
Currently student, n (%)	103 (14)	25 (14)	27 (15)	26 (14)	25 (14)
Treatment arm, n (%)
Treatment	377 (52)	85 (48)	95 (53)	105 (58)	92 (51)
Placebo	343 (48)	94 (52)	85 (47)	75 (42)	89 (49)
Reproductive history
No. of previous live births, n (%)
0	302 (42)	72 (40)	76 (42)	72 (40)	82 (45)
1	277 (38)	71 (40)	71 (39)	73 (41)	62 (34)
2	141 (20)	36 (20)	33 (18)	35 (19)	37 (20)
No. of prior losses, n (%)
1	469 (66)	114 (64)	114 (64)	121 (67)	120 (68)
2	243 (34)	64 (36)	63 (36)	59 (33)	57 (32)
GA of last loss	9.3 ± 4.9	9.4 ± 5.8	9.6 ± 5.7	9.0 ± 4.3	9.2 ± 3.1
Lifestyle/pregnancy symptoms of current pregnancy
Nauseab (median, Q1, Q3)	0.62 (0.32, 0.90)	0.59 (0.29, 0.88)	0.61 (0.31, 0.86)	0.60 (0.32, 0.86)	0.74 (0.40, 1.00)
Alcohol Consumerc (n (%))	28 (4)	4 (2)	7 (4)	9 (5)	8 (4)
Smokerd (n (%))	39 (5)	11 (6)	11 (6)	9 (5)	8 (4)
Caffeine consumere (n (%))	392 (54)	97 (54)	110 (61)	102 (57)	83 (46)
Perceived stressf (median, Q1, Q3)	0.67 (0.26, 1.11)	0.10 (0.03, 0.17)	0.46 (0.36, 0.56)	0.93 (0.79, 1.00)	1.41 (1.23, 1.72)
Physical activityg (n (%))
Low	233 (32)	61 (34)	53 (29)	56 (31)	63 (35)
Moderate	311 (43)	78 (45)	75 (46)	81 (44)	83 (38)
High	176 (24)	40 (21)	45 (24)	45 (25)	47 (28)

^aIncluded women who recorded perceived stress in active pregnancy follow-up who had an hCG positive pregnancy test (n = 720/797 (91%) women). Among the included women, there were no missing values for women’s age, race, treatment, income, marital status, education, prior number of live births or exercise. There were 14 observations missing for BMI, 7 missing employment status, 8 missing student status, 8 missing prior loss number and 1 missing smoking or alcohol consumer status.

^bWomen reported nausea and vomiting symptoms as 0 = none, 1 = nausea only, 2 = vomiting once/day, 3 = vomiting more than once/day.

^cWomen reported alcohol drinks consumed per day, which was dichotomized to any alcohol or none (yes/no).

^dWomen reported tobacco smoking per day (yes/no).

^eWomen reported caffeinated drinks consumed per day, which was dichotomized to any caffeine or none (yes/no).

^fWomen recorded their daily stress levels via a Likert scale with 0 = no stress, 1 = little stress, 2 = moderate stress and 3 = a lot of stress.

^gAt baseline, women completed the short form of the International Physical Activity Questionnaire, from which low, moderate and high physical activity was categorized via standard protocol.

Women with high (>median) versus low (≤median) perceived stress during pregnancy had a statistically significantly higher HR of total losses (HR: 1.69, 95% CI: 1.13, 2.54) and hCG-detected losses (HR: 2.16, 95% CI: 1.04, 4.46), but a null relationship with risk of clinical losses (HR: 1.58, 95% CI: 0.96, 2.60) after adjusting for woman’s age, BMI, employment, marital status, income, education, race, parity, prior losses, exercise and time-varying nausea/vomiting, caffeine, alcohol and smoking (Fig. 3). Results remained consistent after only adjusting for baseline confounding factors. Assessing weekly perceived stress and pregnancy loss by quartiles did not reveal any clear dose–response effect (Table II). Restricted cubic spline analyses indicated an initial sharp then gradual rise in the adjusted HR for total losses by continuous perceived stress (Supplementary Fig. S1).

Figure 3.

Hazard ratio of pregnancy loss, high (>median) versus low (≤median) perceived stress.

Table II.

Hazard ratio of pregnancy loss by type of pregnancy loss by perceived stress quartiles.

Perceived stress	Unadjusted HR (95% CI)	Adjusteda HR (95% CI)
Any pregnancy loss
Q1	1.0	1.0
Q2	1.10 (0.59, 2.04)	1.17 (0.62, 2.19)
Q3	1.69 (0.92, 3.06)	1.89 (1.02, 3.49)
Q4	1.51 (0.81, 2.83)	1.73 (0.91, 3.30)
hCG pregnancy loss
Q1	1.0	1.0
Q2	1.40 (0.37, 4.30)	1.39 (0.36, 5.33)
Q3	2.82 (0.88, 8.98)	3.02 (0.93, 9.79)
Q4	2.36 (0.75, 7.41)	2.40 (0.75, 7.72)
Clinically confirmed pregnancy loss
Q1	1.0	1.0
Q2	1.03 (0.51, 2.08)	1.14 (0.56, 2.31)
Q3	1.42 (0.69, 2.90)	1.63 (0.78, 3.41)
Q4	1.32 (0.62, 2.84)	1.59 (0.73, 3.49)

^aAdjusted analyses accounted for woman’s age (continuous), BMI (continuous), employment (yes/no), marital status (married or living with partner/other), income (5 categories), education (high school or below versus higher than high school), race (White versus non-White), parity (0, 1, 2), prior losses (1 or 2), exercise (low, moderate, high) and time-varying nausea/vomiting, caffeine, alcohol and smoking. Weekly perceived stress assessed from gestational weeks 2 to 36 and categorized into quartiles.

Regarding unmeasured confounding, the observed HR of 1.69 for total losses and 2.16 for hCG losses could be explained away by an unmeasured confounder that was associated with both the exposure and the outcome by an HR of 2.23 (CI: 1.40) and 2.79 (CI: 1.20) respectively, above and beyond the measured confounders, but weaker confounding could not do so. Regarding selection bias, the observed HR of 1.69 for total losses and 2.16 for hCG losses could be explained away by a selection factor that was associated with both the exposure and the outcome by an HR of 2.77 (CI: 1.51) and 3.74 (CI: 1.24), but weaker selection bias could not do so.

Discussion

In this secondary data analysis among pregnant women with a history of pregnancy loss who were enrolled in the EAGeR trial, we found that higher perceived stress in early pregnancy was associated with an elevated risk of pregnancy loss, with higher risks observed for early hCG-detected pregnancy losses. Notably, these findings are based on daily reports of perceived stress during these early critical windows of development and consider important lifestyle factors that also vary in relation to stress levels during this period. Our finding of an estimated 2.2-fold elevated risk of an hCG-detected pregnancy loss is novel and suggests that the associations between stress and pregnancy loss may be higher in early pregnancy; however, given our overlapping confidence intervals, more research assessing peri-conceptional and early pregnancy stress and loss are needed before definitive conclusions can be made. With the recent rise in stress levels among reproductive age women (Berthelot et al., 2020; Corbett et al., 2020; Mappa et al., 2020) routine screening during early pregnancy in women with previous pregnancy loss, coupled with novel and effective interventions including meditation and mindfulness (Jensen et al., 2021) may lead to improved pregnancy outcomes (Cardwell, 2013).

Our results are in line with a recent systematic review and meta-analysis showing a 1.42 pooled odds ratio (95% CI: 1.19, 1.70) of any pregnancy loss among women with a history of exposure to psychological or work-related stress (Qu et al., 2017). While this review supports the positive association between stress and miscarriage risk (Nakamura et al., 2008; Qu et al., 2017), the authors note that prior results have been mixed due to differences in study design, stress exposure assessment, proper control for potential confounding factors and how pregnancy loss was assessed. The majority of studies to date enrolled women at the time of their first prenatal appointment or after, consequently missing early pregnancy losses (Nakamura et al., 2008; Qu et al., 2017). Given that different mechanisms may be at work with regards to how stress contributes to an hCG detected versus clinically confirmed loss, risk estimates are not comparable across studies with unspecified or widely varying failure time windows (Nakamura et al., 2008). Case-control studies (O’Hare and Creed, 1995; Hamilton Boyles et al., 2000; Bashour and Abdul Salam, 2001; Nelson et al., 2003; Maconochie et al., 2007) whereby exposure to stress is assessed retrospectively, have shown more variability and reduced estimates in relation to stress and pregnancy loss as compared to cohort studies (Brandt and Nielsen, 1992; Fenster et al., 1995; Hjollund et al., 2000; Nepomnaschy et al., 2004; Meaney et al., 2014; Qu et al., 2017; Lynch et al., 2018). The majority of studies have measured life events (O’Hare and Creed, 1995; Hamilton Boyles et al., 2000) or work stress (Brandt and Nielsen, 1992; Fenster et al., 1995; Maconochie et al., 2007) as compared to psychological stress (Bashour and Abdul Salam, 2001; Nelson et al., 2003; Meaney et al., 2014). Perceived or psychological stress, broadly thought of as when an individual perceives that environmental demands, tax or exceed his or her adaptive capacity (Cohen, 2007), depends on a host of factors including internal resources, and social and material support available (Qu et al., 2017), and was previously found to be more strongly associated with miscarriage as compared to the objective measures of life events or work stress (Qu et al., 2017). Additionally, prior studies have focused on a single assessment of perceived stress, either at baseline (Nelson et al., 2003) or in the latter part of the first trimester (Brandt and Nielsen, 1992; Meaney et al., 2014). Given that reproductive events, such as pregnancy loss, are susceptible to critical windows of exposure, prospective repeated measurements, such as our use of a daily perceived stress assessment, are warranted.

Only three prior studies, that we are aware of, have enrolled women pre-conceptionally and assessed associations between various measures of stress and early hCG pregnancy losses, two of which found risk estimates above 2-fold for the relationship between reported stress and early pregnancy loss (Hjollund et al., 2000; Nepomnaschy et al., 2006; Lynch et al., 2018). While findings from these three studies and ours are difficult to compare due to differing methods of stress exposure assessment, appropriately accounting for time-varying factors known to affect psychological stress and pregnancy loss, and heterogeneity of study populations, it is important to note that our results are consistent with the two studies that measured stress during the critical window of early pregnancy establishment (i.e. from implantation to clinically confirmed pregnancy) (Hjollund et al., 2000; Nepomnaschy et al., 2006) as compared to measuring preconception stress (Lynch et al., 2018). In a study of 22 observed pregnancies among a community of Mayan women, pregnancies exposed to increased versus normal cortisol levels during the first 3 weeks after conception were 2.7 times more likely (95% CI: 1.2, 6.2) to result in an early hCG pregnancy loss (i.e. up to 3 weeks after ovulation) (Nepomnaschy et al., 2006). In a Danish preconception cohort including 181 pregnancies, 32 of which were subclinical pregnancies detected by hCG only, women who reported a high versus low physical strain score across the menstrual cycle (Days 6 to 9 after ovulation) had a 2.5 risk ratio (95% CI: 1.3, 4.6) for spontaneous abortion after adjusting for study site, women’s age, body mass index, smoking, caffeine and alcohol consumption, female reproductive disease and partner’s sperm count (Hjollund et al., 2000). Conversely, in a recent study among 337 healthy women from the USA who became pregnant, for which 97 ended in loss (43% prior to 6 weeks gestation), there was no clear pattern of association between pre-conceptional stress biomarkers (salivary cortisol and alpha-amylase) and pregnancy loss (Lynch et al., 2018). In prior work, we found poor agreement between a baseline assessment of stress and prospectively measured stress over the menstrual cycle, underscoring the need for temporally relevant stress assessments (Schliep et al., 2015).

Our study had many strengths. Our relatively large sample size of 797 pregnancies (with 188 ending in a pregnancy loss) along with our systematic, prospective stress data collection from the time of conception through pregnancy, allowed us to uniquely assess the association between psychological stress and pregnancy loss during critical windows of exposure. Additionally, our assessment of daily stress up to the first 8 weeks of pregnancy, which were averaged into a weekly measure, and monthly thereafter allowed us to capture potential varying stress levels throughout critical windows of early pregnancy, as compared to prior studies assessing stress levels at a single time point, either retrospectively or at baseline prior to pregnancy (Nakamura et al., 2008; Qu et al., 2017). Furthermore, we were able to account for not only important time-fixed sociodemographic confounders but time-varying lifestyle and physiologic factors, such as nausea/vomiting, caffeine, alcohol and smoking that may impact both stress levels as well as risk of pregnancy loss. Finally, all pregnancy losses were thoroughly identified through daily urine hCG tests for peri-implantation losses and by the woman or her primary care provider after clinical recognition of pregnancy for clinical losses (Mumford et al., 2016). Our study, however, is not without limitations. Women enrolling in the EAGeR trial were predominately White and of high socioeconomic status. The effects of perceived stress on pregnancy loss among non-White women of lower socioeconomic status who may experience more extreme chronic stress is still yet to be fully explored and warrants investigation in future studies. Additionally, we were limited in our ability to clearly identify the mechanisms of stress on pregnancy loss due to our sole reliance on self-reported perceived stress, and the lack of biological stress biomarkers. Finally, while we were able to address many confounding factors that prior studies left out, we cannot completely rule out potential residual confounding that could have biased our findings nor can we rule out selection bias due to restricting the analysis to women who achieved pregnancy and were, therefore, at risk of a pregnancy loss. However, our e-values for unmeasured confounding and selection bias for total losses (2.23 and 2.79, respectively) and for hCG-detected losses (2.77 and 3.74, respectively) are relatively strong. We are not aware of unmeasured confounding factors or selection factors that could completely explain away our results.

In conclusion, our findings indicate that daily perceived stress in early pregnancy may be associated with pregnancy loss, most notably hCG-detected losses, among women with a history of a prior loss. While simply telling women at risk for pregnancy loss not to be stressed is unadvisable, effective interventions for reducing preconception and prenatal stress have been recently reported (Jensen et al., 2021). Most notably, a recent randomized control trial showed reduced perceived stress among non-pregnant women experiencing recurrent pregnancy loss who were assigned to a 7-week meditation and mindfulness program compared to standard supportive care (Jensen et al., 2021). Rising stress levels among women of reproductive age (Berthelot et al., 2020; Corbett et al., 2020; Mappa et al., 2020) along with the serious morbidities already shown to be tied to prenatal stress highlight the potential benefits of identifying, preventing and treating the psychological stress contributing to adverse pregnancy outcomes, including early pregnancy loss (Qu et al., 2017).

Data Availability

The data underlying this article cannot be shared publicly. The data will be shared on reasonable request to the corresponding author.

Authors’ roles

K.C.S.: design, analysis and interpretation of data; drafting of article; final approval of the version to be published. S.L.M., R.M.S. and E.F.S.: conception and design of study; acquisition and interpretation of data; revising the article critically for important intellectual content, final approval of the version to be published. S.N.H., K.K., L.A.S., J.B.S., A.P.-S. and T.C.P.: interpretation of data; revising the article critically for important intellectual content; final approval of the version to be published.

Funding

This work was supported by the Intramural Research Program of the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, Maryland (Contract numbers: HHSN267200603423, HHSN267200603424, HHSN267200603426, HHSN275201300023I). Additionally, K.C.S. was supported by the National Institute on Aging of the National Institutes of Health under Award Number K01AG058781.

Conflict of interest

The authors have no conflicts of interest to disclose.