Allostatic load, a measure of chronic physiological stress, is associated with pregnancy outcomes, but not fertility, among women with unexplained infertility

Abstract

STUDY QUESTION

Among infertile women undergoing ovarian stimulation, is allostatic load (AL), a measure of chronic physiological stress, associated with subsequent fertility and pregnancy outcomes?

SUMMARY ANSWER

AL at baseline was not associated with conception, spontaneous abortion or live birth, however, it was significantly associated with increased odds of pre-eclampsia and preterm birth among women who had a live birth in the study.

WHAT IS KNOWN ALREADY

Several studies have linked AL during pregnancy to adverse outcomes including preterm birth and pre-eclampsia, hypothesizing that it may contribute to well-documented disparities in pregnancy and birth outcomes. However, AL biomarkers change over the course of pregnancy, raising questions as to whether gestational AL assessment is a valid measure of cumulative physiologic stress starting long before pregnancy. To better understand how AL may impact reproductive outcomes, AL measurement in the non-pregnant state (i.e. prior to conception) is needed.

STUDY DESIGN, SIZE, DURATION

A secondary data analysis based on data from 836 women who participated in Assessment of Multiple Intrauterine Gestations from Ovarian Stimulation (AMIGOS), a multi-center, randomized clinical trial of ovarian stimulation conducted from 2011 to 2014.

PARTICIPANTS/MATERIALS, SETTING, METHODS

Ovulatory women with unexplained infertility (ages 18–40) were enrolled and at baseline, biological and anthropometric measures were collected. AL scores were calculated as a composite of the following baseline variables determined a priori: BMI, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, dehydroepiandrosterone sulfate, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, triglycerides, C-reactive protein and HOMA score. Participants received ovarian stimulation for up to four cycles and if they conceived, were followed throughout pregnancy. We fit multi-variable logistic regression models examining AL (one-tailed and two-tailed) in relation to the following reproductive outcomes: conception, spontaneous abortion, live birth, pre-eclampsia, preterm birth and low birthweight.

MAIN RESULTS AND THE ROLE OF CHANCE

Adjusting for covariates, a unit increase in two-tailed AL score was associated with 62% increased odds of pre-eclampsia (OR: 1.62, 95% CI: 1.14, 2.38) 44% increased odds of preterm birth (OR: 1.44, 95% CI: 1.02, 2.08), and 39% increased odds of low birthweight (OR: 1.39, 95% CI: 0.99, 1.97). The relationship between AL and preterm birth was mediated by pre-eclampsia (P = 0.0003). In one-tailed AL analyses, associations were similar, but slightly attenuated. AL was not associated with fertility outcomes (conception, spontaneous abortion, live birth).

LIMITATIONS, REASONS FOR CAUTION

Results may not be generalizable to fertile women who conceive naturally or women with other types of infertility. Comparisons to previous, related work are difficult because variables included in AL composite measures vary across studies. AL may be indicative of overall poor health, rather than being specific to chronic physiological stress.

WIDER IMPLICATIONS OF THE FINDINGS

Our results suggest that chronic physiological stress may not impact success of ovarian stimulation, however, they confirm and extend previous work suggesting that AL is associated with adverse pregnancy outcomes. Physiological dysregulation due to chronic stress has been proposed as a possible mechanism underlying disparities in birth outcomes, which are currently poorly understood. Assessing biomarkers of physiological dysregulation pre-conception or in early pregnancy, may help to identify women at risk of adverse pregnancy outcomes, particularly pre-eclampsia.

STUDY FUNDING/COMPETING INTEREST(S)

Support for AMIGOS was provided by: U10 HD39005, U10 HD38992, U10 HD27049, U10 HD38998, U10 HD055942, HD055944, U10 HD055936 and U10HD055925. Support for the current analysis was provided by T32ES007271, R25HD075737, P30ES001247 and P30ES005022. This research was made possible by funding by American Recovery and Reinvestment Act. The content is solely the responsibility of the authors and does not necessarily represent the official views of NICHD, NIEHS or NIH. E.B., W.V., O.M., R.A., M.R., V.B., G.W.B., C.C., E.E., S.K., R.U., P.C, H.Z., N.S. and S.T. have nothing to disclose. R.L. reported serving as a consultant to Abbvie, Bayer, Kindex, Odega, Millendo and Fractyl and serving as a site investigator and receiving grants from Ferring. K.H. reported receiving grants from Roche Diagnostics and Ferring. R.R. reported a grant from AbbVie. M.D. reported being on the Board of Directors of and a stockholder in Advanced Reproductive Care.

TRIAL REGISTRATION NUMBER

Clinical Trials.gov number: NCT01044862.

Introduction

It is widely accepted that stress negatively affects health and many studies have explored associations between psychosocial stressors, fertility and pregnancy outcomes (Hobel et al., 2008; Boivin et al., 2011; Lynch et al., 2014; Nillni et al., 2016). Stressors trigger an integrated physiological response (including neuroendocrine, cardiometabolic and immune activation) that is adaptive in the short term because it allows the body to respond to a changing environment (McEwen and Seeman, 2009). However, under conditions of chronic, long-term stress, these systems can become dysregulated. The cumulative physiological ‘wear and tear’ that occurs as a result of chronic stress is known as allostatic load (AL) (McEwen and Stellar, 1993), and is operationalized as a composite measure based on subclinical neuroendocrine, cardiovascular, metabolic and immune measures (Edes and Crews, 2017).

Because conception and maintenance of pregnancy requires a complex suite of neuroendocrine, immune and cardiometabolic changes, when these systems are dysregulated (as indicated by high AL), fecundity may be impaired and the course of pregnancy altered. To date, no published study has examined AL in relation to measures of fertility. Epidemiological evidence examining AL in relation to pregnancy outcomes has been mixed. Several studies measuring AL during or after pregnancy have shown increased odds of adverse pregnancy and birth outcomes including pre-eclampsia, preterm birth and low birthweight or small for gestational age infants (Wallace and Harville, 2013; Hux et al., 2014; Hux and Roberts, 2015; Accortt et al., 2017). However, AL assessment during pregnancy is non-optimal because altered AL biomarker profiles may not represent the physiological manifestation of chronic stress over years, but rather, typical pregnancy-specific physiological changes (Morrison et al., 2013). To our knowledge, only one study has examined pre-pregnancy AL in relation to birth outcomes, finding no associations, however, AL was measured in adolescence, years prior to pregnancy, raising additional questions about timing of exposure assessment (Wallace et al., 2013a, b).

Here we build upon this small body of literature to examine the relationships between AL, fertility and birth outcomes using secondary data from a randomized controlled trial of infertile women trying to conceive. This study design improves upon the temporal issues encountered by previous studies by assessing AL based on biomarkers measured at baseline, in the non-pregnant state. This approach also enables us to examine the relationship between AL and fertility outcomes. Our a priori hypothesis is that baseline AL will be associated with decreased odds of conception, increased odds of pregnancy loss, and increased odds of adverse pregnancy and birth outcomes including pre-eclampsia, preterm birth and low birthweight.

Materials and Methods

Study population

The Assessment of Multiple Intrauterine Gestations from Ovarian Stimulation (AMIGOS) was a multi-center randomized controlled trial to compare rates of multiple gestations among women with unexplained infertility across three ovarian stimulation treatment groups (clomiphene, letrozole and gonadotropins) (Diamond et al., 2015a,b). From 2011 to 2013, healthy infertile couples trying to conceive were recruited. Inclusion criteria included: unexplained infertility in which the female partner was ovulating regularly (at least nine menses per year), female partner age 18–40, at least 1 year of infertility history, normal uterine cavity, at least one patent Fallopian tube, and male partner with at least 5 million total motile sperm in the ejaculate (Diamond et al., 2011). Women with serious medical conditions that were poorly controlled or would threaten a pregnancy were excluded from participation.

Ethical approval

An Advisory Board and a Data Safety Monitoring Committee appointed by the National Institutes of Health and the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) approved the study protocol prior to implementation. The Institutional Review Board (IRB) of each recruitment, treatment and data collection center also approved all study activities and all subjects signed informed consent.

Baseline assessments

Participating women underwent a single standardized physical examination at baseline including height, weight, waist circumference, hip circumference, and blood pressure (diastolic and systolic). BMI was calculated as weight/(height)². Fasting serum samples were obtained, frozen at −80°C, and shipped on dry ice for analysis at the University of Virginia Ligand Assay and Analysis Core Laboratories. Dehydroepiandrosterone sulfate (DHEAS), fasting insulin, and C-reactive protein (CRP) were assayed using the Immulite system (Siemens Diagnostics, Los Angeles, CA). Fasting glucose was measured using a glucose oxidase assay (Analox Instruments, Inc., Lunenburg, MA). The sensitivity limits for the DHEAS, insulin, CRP and glucose assays were 7 μg/dl, 2.0 uIU/ml, 0.02 mg/dl and 1.0 mg/dl, respectively. Homeostatic model assessment (HOMA) score, a measure of insulin resistance, was calculated as insulin*glucose/405 (Matthews et al., 1985).

Women self-reported race, income, education, alcohol use and smoking. In the current analyses, race was dichotomized as white/non-white and household income was dichotomized as <$50 000/year vs ≥$50 000/year. Alcohol use was categorized as any vs none and cigarette smoking was categorized as never, former and current.

In the parent AMIGOS study, participating women were then randomly assigned to one of three treatment groups: gonadotropin, clomiphene or letrozole. Treatment began on Day 3, 4 or 5 of the menstrual cycle and ovarian stimulation continued for up to four treatment cycles or until pregnancy occurred.

Assessment of AL

Drawing on commonly used measures, AL scores were calculated using 10 variables measured at baseline: BMI, waist-to-hip ratio (WHR), systolic blood pressure (SBP), diastolic blood pressure (DBP), DHEAS, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides, CRP and HOMA score (Edes and Crews, 2017). Two sets of AL scores were calculated: one- and two-tailed. In one-tailed AL indices, which are most common in the literature, for each biomarker, the highest quartile of values in the population distribution are considered ‘high risk’ (McEwen and Seeman, 2009). For each biomarker, values at or above the 75th percentile are scored as ‘1’, while values below the 75th percentile are scored as ‘0’. The exceptions to this are DHEAS and HDL cholesterol, where the scoring was reversed such that the lowest 25th percentile is considered higher risk and scored as ‘1’. The 10 individual biomarker scores are then totaled to create a one-tailed AL score for each subject ranging from 0 to 10. Two-tailed AL indices are formulated to consider the possibility that any extreme values (high or low) may signal dysregulation of homeostatic mechanisms (Seplaki et al., 2005). In two-tailed AL analyses, for each biomarker, values below the 10th percentile or above the 90th percentile are considered high risk and scores as ‘1’ while values between the 10th and 90th percentile are scored as ‘0’. The 10 individual biomarker scores are then totaled to create a two-tailed AL score ranging from 0 to 10 for each subject. For both one- and two-tailed indices, higher AL scores correspond to greater physiological dysfunction.

Outcome assessment

We assessed six binary outcome measures through clinical chart abstraction, patient report and medical record review. Clinical pregnancy was defined as an ultrasound-confirmed fetal heartbeat confirmed on ultrasound 4–6 weeks after fertility treatment (singleton or multiple). Spontaneous abortion was a pregnancy loss following confirmation of clinical pregnancy. Live birth was defined as having an infant born alive (singleton or multiple). Among women with a singleton live birth, we assessed three additional outcomes: pre-eclampsia (as diagnosed by physicians according to criteria recommended by the American College of Obstetricians and Gynecologists) (ACOG, 2013), preterm birth (gestational age at birth <37 weeks), and low birthweight (<2500 g). Pregnancies resulting in multiple births were not included in those analyses because carrying multiples is an independent risk factor for all three outcomes (2012). Pregnancies in which there was a live birth as well as a pregnancy loss were only included in models with clinical pregnancy as the outcome.

Statistical analysis

We calculated summary statistics for all variables of interest and used scatterplots to examine the relationships between continuous variables. Crude logistic regression models examined relationships between one-tailed AL scores and outcome measures. In multi-variable logistic regression models, we adjusted for a set of covariates chosen a priori based on their relationships to measures of stress and fertility: age, study arm, race, alcohol use, smoking status and income level. For models examining pre-eclampsia, preterm birth and low birthweight, the same covariates were included, however, analyses were limited to live, singleton births. We then refit all models using two-tailed AL scores.

We evaluated assumptions of linearity between the continuous predictors and all outcomes using generalized additive models (Hastie and Tibshirani, 1990). All relationships between predictors and outcomes were deemed to be approximately linear (P-values for testing the null hypothesis of linearity were all >0.05), thus generalized additive models were not considered further (not shown). Additionally we fit exploratory models including a fetal sex–AL interaction term to examine the possibility that the relationship between AL and pregnancy/birth outcomes differs by fetal sex with males being more vulnerable to adverse outcomes, but observed no interaction nor main effect of fetal sex and these models were also not considered further (not shown) (Bale, 2011; Bronson and Bale, 2014).

We conducted a set of secondary analyses to generate an alternative AL composite measure comprised of only those variables that most strongly predicted outcomes. To do so, we first examined correlations between the 10 individual variables in our AL composite and the outcome measures. For each outcome measure, the variable with the strongest association was considered the most important predictor and coded as 0/1. We then selected the variable with the second strongest correlation with the outcome measure, coded it as 0/1, and created a composite based on those two variables with a range of values from 0 to 2. We repeated this incremental approach until all 10 variables had been entered into the composite, matching our primary models. At each increment, we evaluated the P-value associated with that composite score, which allowed us to identify the composite measure that was most strongly associated with our outcomes. Both one-tailed and two-tailed incremental composite measures were generated, adjusting for covariates as in our primary models.

Finally, because pre-eclampsia is a risk factor for preterm birth (Sibai et al., 2005; Auger et al., 2011), we conducted mediation analyses to examine the possibility that pre-eclampsia mediates the relationship between AL and preterm birth. To do so, we fit four regression models: (i) AL predicting preterm birth; (ii) AL predicting pre-eclampsia; (iii) pre-eclampsia predicting preterm birth; and (iv) AL and pre-eclampsia predicting preterm birth. If any of the relationships in steps 1–3 (establishing zero-order relationships among the variables) is non-significant, then mediation is unlikely or impossible. If models 1–3 demonstrate significant relationships, the analysis proceeds to Step 4. Full mediation is supported if AL is no longer significant after adjusting for pre-eclampsia (Baron and Kenny, 1986). Statistical significance was assessed based on α = 0.05. Analyses were conducted using R version 3.3.2.

Results

Overall, 900 women participated in AMIGOS. Of these, 836 women had complete data on variables used to calculate AL scores. The remaining 64 women were missing data on at least one AL variable [waist to hip ratio (n = 4), DHEAS (n = 26), CRP (n = 28), cholesterol and triglycerides (n = 56)], and thus did not contribute to subsequent analyses. Women for whom AL could not be calculated were more likely to be non-white than those with calculated AL scores. They were also older, less likely to smoke or drink, and had lower education, income and BMI.

At baseline, participants were 32.2 ± 4.3 years old and had a BMI of 26.8 ± 6.5 kg/m² (Table I). Most participants (81.0%) were White, had at least a college education (92.0%), and had an annual household income of at least $50 000 (66.0%). The majority of women reported alcohol use (88.5%), and 92.2% were never or former smokers. Subjects were evenly split across the study treatment arms (33.6% gonadotropin; 33.8% clomiphene; 32.5% letrozole). The mean one-tailed and two-tailed AL scores at baseline were 2.6 ± 2.0 and 2.1 ± 1.5, respectively.

Table I.

Characteristics of AMIGOS participants.

Continuous variables	All subjects (N = 836)			Live singleton births only (N = 171)
	Mean (SD)	Median	Min–max	Mean (SD)	Median	Min–max
Age (years)	32.2 (4.3)	32.0	20.0–41.0	31.7 (4.3)	32.0	21.0–40.0
BMI (kg/m2)	26.8 (6.5)	25.2	16.5–52.0	27.3 (6.4)	26.2	17.4–49.1
Waist-to-hip ratio	0.8 (0.1)	0.8	0.6–1.5	0.9 (0.1)	0.8	0.7–1.2
Systolic blood pressure (mmHg)	116.0 (12.6)	115.0	74.0–159.0	117.1 (13.2)	117.0	85.0–156.0
Diastolic blood pressure (mmHg)	74.6 (9.8)	74.0	40.0–102.0	74.0 (9.2)	73.0	54.0–100.0
DHEAS (μg/dl)	130.7 (65.8)	120.0	7.0–437.0	128.8 (62.7)	119.0	21.0–437.0
C-reactive protein (mg/l)	3.2 (4.7)	1.6	0.2–54.3	3.4 (4.9)	1.5	0.2–42.2
HDL cholesterol (mg/dl)	45.9 (11.1)	46.0	18.0–120.0	46.7 (13.5)	46.0	20.0–120.0
LDL cholesterol (mg/dl)	109.4 (33.1)	107.0	17.0–400.0	113.6 (42.1)	109.0	17.0–400.0
Triglycerides (mg/dl)	94.9 (52.5)	81.0	15.0–605.0	98.3 (58.2)	84.0	15.0–483.0
HOMA scorea	2.1 (3.6)	1.1	0.2–45.9	2.3 (3.6)	1.2	0.2–23.4
Allostatic load score (one-tailed)b,c	2.6 (2.0)	2	0.0–10.0	2.7 (2.0)	2.0	1.0–8.0
Allostatic load score (two-tailed)b,d	2.1 (1.5)	2	0.0–7.0	2.1 (1.6)	2.0	0.0–7.0
Gestational age at birth (weeks)e	37.9 (2.7)	39	23.0–42.0	38.6 (2.2)	39.0	23.0–42.0
Birthweight (g)e	3035.0 (741.5)	3146.8	226.8–4507.6	3229.2 (226.8)	3288.5	226.8–4507.6

Categorical Variables	All subjects (N = 836)	Live singleton births (N = 171)
	N (%)	N (%)
Race
White	677 (81.0)	144 (84.2)
Non-white	159 (19.0)	27 (15.8)
Education
High school or less	66 (7.9)	14 (8.2)
College/some college	551 (65.8)	120 (70.2)
Graduate degree	219 (26.2)	37 (21.6)
Annual household income
<$50 000	142 (17.0)	21 (12.3)
≥$50 000	552 (66.0)	123 (71.9)
Unknown	142 (17.0)	27 (15.8)
Alcohol use (any)	740 (88.5)	158 (92.4)
Smoking
Never	548 (65.6)	108 (63.2)
Former	222 (26.5)	43 (25.1)
Current	66 (7.9)	20 (11.7)
Treatment group
Clomiphene	283 (33.8)	63 (36.8)
Letrozole	272 (32.5)	47 (27.5)
Gonadotropin	281 (33.6)	61 (35.7)
Conceptions
Achieved pregnancy	312 (37.3)	171 (100.0)
Achieved clinical pregnancy	244 (29.2)	171 (100.0)

	Clinical pregnancies (N = 244)	Live singleton births (N = 171)
	N (%)	N (%)
Pregnancy outcomes
Live birth	211 (86.5)	171 (70.1)
Induced abortion	1 (0.4)	N/A
Spontaneous abortion	19 (7.8)	N/A
Ectopic	2 (0.8)	N/A
Unknown	11 (4.5)	N/A
Pregnancy complications	(out of n = 244)	(out of n = 171)
Pre-eclampsia	24 (9.8)	17 (9.9)
Low birthweight	44 (18.0)	19 (11.1)
Preterm birth	39 (16.0)	16 (9.4)

^aHOMA=insulin*glucose/405.

^bThe following variables are including in allostatic load scores: BMI, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, DHEAS, CRP, HPL cholesterol, LDL cholesterol, triglycerides, HOMA score.

^cOne-tailed allostatic load: for each biomarker, values at or above the 75th percentile are scored as ‘1’, otherwise they are scored as ‘0’ (with the exception of DHEAS and HDL-cholesterol), for which the lower 25th percentile signifies highest risk and the scores are reversed. Scores are then summed to create an AL index ranging from 0 to 11.

^dTwo-tailed allostatic load: for each biomarker, values at or above the 90th percentile or at or below the 10th percentile are scored as ‘1’. Values between the 10th and 90th percentile are scored as ‘0’. Scores are then summed to create an AL index ranging from 0 to 11.

^eGestational age at birth (n = 211); birthweight (n = 210).

Of the women in the current analysis, 312 women (37.3%) conceived a study pregnancy and 78.2% of detected conceptions resulted in a clinically confirmed pregnancy (n = 244). The clinical conceptions resulted in 33 losses, 171 singleton births and 40 multiple births. Of the singleton pregnancies, 9.8% of mothers had pre-eclampsia, 9.4% of infants were born preterm and 11.1% were born low birthweight.

In crude and adjusted analyses, one-tailed or two-tailed AL scores were not associated with clinical pregnancy, spontaneous abortion or live birth (Table II). In unadjusted models limited to singleton births, a one unit increase in one-tailed AL score was associated with 30% increased odds of pre-eclampsia (OR: 1.30, 95% CI: 1.03, 1.65) and the association was stronger after adjusting for covariates (OR: 1.41, 95% CI: 1.08, 1.86) (Table III). Relationships were even stronger when two-tailed AL score was considered. A one unit increase in two-tailed AL was associated with 46 and 62% increased odds of pre-eclampsia in unadjusted (OR: 1.46, 95% CI: 1.07, 2.02) and adjusted models (OR: 1.62, 95% CI: 1.14, 2.38) respectively (Table III). One-tailed AL was associated with increased odds of preterm birth in crude (OR: 1.23; 95% CI: 0.97, 1.57) and adjusted models (OR: 1.27, 95% CI: 0.98, 1.66). Relationships were significantly associated for two-tailed AL scores. In unadjusted models, the odds of preterm birth rose by 49% with a one-unit increase in two-tailed AL score (OR: 1.49, 95% CI: 1.08, 2.08) and in adjusted models, the relationships were similar (OR: 1.44; 95% CI: 1.02, 2.08) (Table III). Similar associations were observed for low birthweight.

Table II.

Odds ratio estimates for fertility outcomes in relation to allostatic load at baseline (prior to conception).

Outcome	n	One-tailed AL scores		Two-tailed AL scores
		Crude analyses	Adjusted analysesa	Crude analyses	Adjusted analysesa
		Point estimate (95% Wald CI)	Point estimate (95% Wald CI)	Point estimate (95% Wald CI)	Point estimate (95% Wald CI)
Clinical pregnancy	836	1.00 (0.93–1.08)	1.03 (0.95, 1.11)	1.00 (0.90–1.10)	1.02 (0.92, 1.13)
Spontaneous abortionb	243	1.09 (0.92–1.29)	1.07 (0.90, 1.28)	1.01 (0.80–1.26)	1.00 (0.78, 1.28)
Live birth	835	1.00 (0.92–1.08)	1.03 (0.95, 1.11)	1.00 (0.90–1.11)	1.02 (0.91, 1.13)

^aAdjusted for age, study arm, race (white/non-white), alcohol use (any/none), smoking (current/former/never) and income (<$50 000, ≥$50 000, did not respond).

^bOnly includes women who conceived.

Table III.

Odds ratio estimates for pregnancy and birth complications in singleton births in relation to allostatic load at baseline (prior to conception).

Complication	nb	One-tailed AL scores		Two-tailed AL scores
		Crude analyses	Adjusted analysesa	Crude analyses	Adjusted analysesa
		Point estimate (95% Wald CI)	Point estimate (95% Wald CI)	Point estimate (95% Wald CI)	Point estimate (95% Wald CI)
Pre-eclampsia	170	1.30 (1.03, 1.65)	1.41 (1.08, 1.86)	1.46 (1.07, 2.02)	1.62 (1.14, 2.38)
Preterm birthc	170	1.23 (0.97, 1.57)	1.27 (0.98, 1.66)	1.49 (1.08, 2.08)	1.44 (1.02, 2.08)
Low birthweightd	169	1.05 (0.83, 1.32)	1.09 (0.85, 1.39)	1.35 (1.01, 1.83)	1.39 (0.99, 1.97)

^aAdjusted for age, study arm, race (white/non-white), alcohol use (any/none), smoking (current/former/never) and income (<$50 000, ≥$50 000, did not respond).

^bIncludes only women who had a live birth.

^c<37 weeks gestation.

^d<2500 g.

In secondary analyses, we explored creating an optimal AL score by incrementally adding variables based on the strength of their correlation with the outcome measures. This allowed us to identify subsets of variables that, used together as a composite, most strongly predicted the outcome measures (Fig. 1). The subset of variables that most strongly predicted the outcome differed by outcome and type of analysis (one-tailed vs two-tailed). For preterm birth, the best fitting one-tailed AL composite included CRP, BMI, SBP, HDL and LDL whereas the best fitting two-tailed composite included BMI, SBP, DBP, CRP and WHR ratio. For pre-eclampsia, the best fitting one-tailed AL composite included BMI, DBP and SBP, whereas the best fitting two-tailed composite included SBP, BMI, DBP, WHR and HDL.

Figure 1.

Secondary analyses examining P-values for the AL composite score in relation to the number of variables in that score (added incrementally based on strength of their individual associations with the outcome measures). In each panel, the horizontal line corresponds to a P-value of 0.05. (A) One-tailed AL scores predicting preterm birth. (B) Two-tailed AL scores predicting preterm birth. (C) One-tailed AL scores predicting pre-eclampsia. (D) Two-tailed AL scores predicting pre-eclampsia.

In additional secondary analyses, we examined whether the relationship between two-tailed AL and preterm birth was mediated by pre-eclampsia (Supplementary Table S1). We did not conduct full mediation analyses for one-tailed AL given that the non-significant results of Step 1 (simple regression analysis with one-tailed AL predicting preterm birth; see Statistical analysis for mediation analysis steps). However, in two-tailed AL analyses, the relationships in Steps 1–3 of the mediation analysis were significant and we therefore proceeded to Step 4. In unadjusted analyses including two-tailed AL and pre-eclampsia in the same model, pre-eclampsia, but not AL, remained a significant predictor of preterm birth suggesting full mediation (β = 2.20; P = 0.0004). Similar results were observed in adjusted models (β = 2.81; P = 0.0003).

Discussion

This study examined AL in non-pregnant women trying to conceive in relation to subsequent fertility and pregnancy outcomes. We demonstrated that baseline AL was associated with adverse pregnancy outcomes; specifically, a one unit increase in two-tailed AL was associated with 62% increased odds of pre-eclampsia, 44% increased odds of preterm birth and 39% increased odds of low birthweight. Given the tremendous maternal and infant morbidity and mortality associated with these conditions, better understanding of their origins is essential (Frey and Klebanoff, 2016; Shih et al., 2016). Early detection of women at risk of poor pregnancy outcomes could have substantial public health benefits in light of a growing literature suggesting transgenerational transmission of adverse reproductive and cardiovascular outcomes (Castrillio et al., 2014; Dabelea and Crume, 2011).

Previous studies of AL in relation to pregnancy outcomes have reported mixed results. Only two studies (both linking Bogalusa Heart Study data to subsequent state birth records) have examined pre-pregnancy AL in relation to birth outcomes and neither observed associations with preterm birth or low birthweight. However, AL assessment occurred many years prior to the index pregnancy (typically in early adolescence), which may have been too early to detect meaningful physiologic dysregulation that could impact the course of pregnancy years later (Wallace et al., 2013a, b). By contrast, in the current study, AL biomarkers were measured at mean age 32, a maximum of 4 months prior to pregnancy, potentially allowing more time for chronic ‘wear and tear’ to dysregulate the stress response.

Two small studies have measured AL in mid-late pregnancy in relation to birthweight and gestational age at birth (Wallace and Harville, 2013; McKee et al., 2017). In the first (n = 42), AL was inversely associated with gestational age at birth but not birth size (Wallace and Harville, 2013), while in the second (n = 111), AL was not associated with birthweight or gestational age (McKee et al., 2017). Given the sizes of the cohorts, neither study could examine preterm birth, low birthweight or pre-eclampsia and in both cases, the neuroendocrine component of AL was insufficiently captured.

Perhaps the biggest challenge in this field is that due to normal pregnancy-related changes in the neuroendocrine, cardiovascular, metabolic and immune systems, conventional AL measures may not be informative (Morrison et al., 2013). Two studies circumvented this issue by measuring post-partum AL in relation to history of adverse birth outcomes. The first, using National Health and Nutrition Examination Survey (NHANES) data, found that women reporting a history of adverse birth outcomes had significantly higher AL than women with a history of only uncomplicated births (Hux et al., 2014). The second study, based on a diverse birth cohort, measured AL at 6–12 months post-partum, similarly finding associations with adverse birth outcomes (Accortt et al., 2017). However, the fact that AL was assessed after birth in both studies makes it impossible to establish temporality or rule out the possibility that having an adverse birth outcome is itself a stressor that contributes to a subsequent elevated AL.

The strongest association observed in the current study was between AL and pre-eclampsia, a disorder characterized by hypertension, proteinuria, and inflammation (Phipps et al., 2016). Mediation analysis, moreover, suggested that relationships observed between AL and preterm birth were mediated by pre-eclampsia, further underscoring the importance of the disorder. To our knowledge, only one study has examined AL and pre-eclampsia. In a small case–control study, AL in early pregnancy (<15 weeks gestation) was significantly higher among preeclamptic women than in matched controls (Hux and Roberts, 2015). Consistent with these results, among women with a history of pre-eclampsia, pre-pregnancy metabolic syndrome (which includes many metabolic AL biomarkers) predicts recurrence of pre-eclampsia in the next pregnancy (Stekkinger et al., 2013).

While incorporating numerous relevant pathways and biomarkers may be seen as a strength of AL composite measures in the research setting, it may limit the utility of AL in the clinical setting. In light of this limitation, our secondary analyses identified a subset of variables that when integrated into an AL composite, most strongly predicted the outcomes of interest. Results differed by the outcome of interest and whether the composite was created based on one-tailed or two-tailed scores. In general, BMI, CRP, SBP, DBP, HDL and WHR emerged as most important, although their relative importance to one another varied across models. For example, while BMI was the strongest predictor in two-tailed models examining preterm birth and one-tailed models examining pre-eclampsia, CRP emerged as a stronger predictor of preterm birth in one-tailed models and SBP was most strongly associated with pre-eclampsia in two-tailed models examining pre-eclampsia. Triglycerides, HOMA and DHEAS were less important across all models. This may indicate that those pathways are less relevant to the birth outcomes of interest, but it could also reflect biomarker measurement issues. Importantly, with the exception of CRP, the variables that emerged as most important are routinely assessed in clinical settings, suggesting that in the future, predictive algorithms based on these kinds of measures could be developed for clinical use.

We observed no associations between baseline AL and measures of fertility, which is interesting in light of the literature on psychosocial stress and fertility in women (reviewed in Santoro et al., 2016). Our results suggest that AL does not impact the odds of conception, pregnancy loss or live birth in women undergoing fertility treatments. These results warrant replication in a cohort attempting to conceive naturally. It is possible, for example, that AL dysregulates ovarian dysfunction, but that those effects were overridden by the fertility treatments administered in AMIGOS and therefore undetectable in this population.

Our study has several strengths, most notably the study design, which involved biomarker assessment in the non-pregnant state. Pre-pregnancy recruitment allowed us to assess fertility outcomes and circumvented the limitations of AL assessment during pregnancy. Unlike some previous work, our sample size was large enough that we were able to examine adverse pregnancy outcomes. We also examined both one- and two-tailed AL measures, and our stronger findings for the latter suggest that physiological dysregulation at both ends of the spectrum may adversely impact pregnancy. Finally, of the studies on AL and pregnancy, ours is among the few to include biomarkers from all of the key domains included in the stress response (Supplementary Table S2).

Our study has several limitations of note. First, our analyses were conducted within the context of a randomized clinical trial for women with unexplained infertility. Although treatment arm is unlikely to have affected our results (and was not associated with birth outcomes), the generalizability of our results may be limited. We cannot rule out the possibility that AL differs in fertile and infertile women. If infertile women have higher baseline AL than fertile women, for example, the additional physiological burden of a pregnancy may further compound the dysregulation and lead to adverse outcomes. In addition, results may not be generalizable to women with other types of infertility. Much of the interest in AL in relation to pregnancy outcomes has focused on understanding perinatal disparities, but our population was predominantly white and of high socioeconomic status. Nevertheless, our results suggest that AL may be a meaningful measure for understanding disparities as well as unexplained variation in perinatal health within demographic groups. Finally, it should be noted that although previous work has demonstrated associations between chronic exposure to psychosocial stress and AL, AL may also reflect overall poor health. Future work in this area should consider comparing questionnaire-based measures of chronic adversity and stress to AL scores.

Our study raises several questions meriting further research. Beyond replicating this work in fertile women, it will be important to determine whether AL in early (or even mid-) pregnancy is correlated with pre-pregnancy AL (therefore reflecting cumulative physiological dysregulation) which would make future work in this field more feasible. One of the intrinsic problems with AL studies is the tremendous difference in how scores are calculated (Duong et al., 2016). This lack of convention precludes the comparison of AL scores across studies and is exemplified by the variation in measures used just within the perinatal literature (Supplementary Table S2). We adhered to widely followed guidelines for AL score calculation established by the MacArthur Research Network on SES and Health (McEwen and Seeman, 2009). Guidelines include selection of primary and secondary measures of physiological dysregulation representing four main domains (endocrine, metabolic, cardiovascular and immune), summing the number of parameters, and using distributions to classify high and low function. However, even in that seminal work, alternative criteria for AL assessment are discussed and the need to define a gold standard approach to guide future research is evident.

Ultimately, these results support the growing evidence suggesting that chronic stress may impact perinatal outcomes. A substantial literature has suggested that psychosocial stress and related constructs (such as depression and anxiety) may affect the course of pregnancy (reviewed in Hobel et al., 2008), while a parallel literature has explored the hypothesis that chronic exposure to stressors like racism and violence result in physiological wear and tear (or ‘weathering’) that ultimately leads to racial disparities in pregnancy outcomes (Geronimus, 1992; Holzman et al., 2009). Our study is relevant to both of these questions in that it provides evidence that in reproductive age women, subclinical dysregulation of stress-related systems may indeed contribute to pre-eclampsia, preterm birth and low birthweight, arguably today’s most pressing reproductive health concerns. Moving forward, AL could theoretically be used to identify women in early pregnancy (or even pre-conception) who may be at increased risk of adverse pregnancy outcomes and require closer monitoring throughout pregnancy.

Authors’ roles

E.B. designed the current analysis and was responsible for drafting and revising the article. W.V. assisted in the design and contributed to drafting and revising the article. OM and S.T. conducted data analysis and contributed to data interpretation and article preparation. M.D., N.S., R.A., R.R., P.C., R.L., C.C, K.H., V.B., R.U., G.W.B., M.R., E.E., S.K. and H.Z. contributed to the design, implementation and data collection in the AMIGOS study and participated in drafting the current article. All authors reviewed and approved the final article.

Funding

Support for the AMIGOS study was provided by the National Institutes of Health (NIH) grants U10HD39005, U10HD38992, U10HD27049, U1038998, U10HD055942, U10HD055944, U10HD055936, U10HD055925, and the American Recovery and Reinvestment Act. Additional support for the current analyses was provided by National Institutes of Health grants: T32ES007271, R25HD075737, P30ES001247, and P30ES005022. The content is solely the responsibility of the authors and does not necessarily represent the official view of NICHD, NIEHS, or NIH.

Conflict of interest

E.B., W.V., O.M., R.A., M.R., V.B., G.W.B., C.C., E.E., S.K., R.U., P.C, H.Z., N.S. and S.T. have nothing to disclose. R.L. reported serving as a consultant to Abbvie, Bayer, Kindex, Odega, Millendo and Fractyl and serving as a site investigator and receiving grants from Ferring. K.H. reported receiving grants from Roche Diagnostics and Ferring. R.R. reported a grant from AbbVie. M.D. reported being on the Board of Directors of and a stockholder in Advanced Reproductive Care.