Allostatic Load in Women with a History of Low Birth Weight Infants: The National Health and Nutrition Examination Survey

Abstract

Background: The purpose of our study was to determine whether women of reproductive age with history of low birth weight (LBW) deliveries have higher allostatic load (AL), a measure of the cumulative toll of chronic stress, than those with normal-weight deliveries.

Methods: We used data from women ages 17–35 who responded to the National Health and Nutrition Examination Survey (NHANES) reproductive-health questionnaire, 1999–2006. Women reported history of LBW infants and those who were preterm. We classified preterm-LBW and term-LBW as surrogates for preterm birth (PTB) and small for gestational age (SGA), respectively. Normal weight included those without LBW infant history. We utilized nine biomarkers measured in NHANES to determine AL and used linear regression to compare unadjusted and adjusted means.

Results: We identified 877 women divided among SGA (2%), PTB (10%), and normal groups (88%). The SGA group had higher unadjusted and adjusted AL scores than did the normal group (2.82±0.35 vs. 1.92±0.07, p=0.011); women in the PTB group had higher AL scores than did the referent in adjusted analyses (2.58±0.21 vs. 1.92±0.07, p=0.001).

Conclusions: Women with history of SGA or PTB had higher AL than did those with normal birth weight outcomes. This suggests a link between adverse pregnancy outcomes, chronic stress, and subclinical disease.

Introduction

Preterm birth and fetal growth restriction remain significant US public health concerns and contribute significantly to neonatal morbidity and mortality. Infants who were preterm births (PTB) or small for gestational age (SGA) are also at increased risk for cardiovascular disease (CVD), diabetes, and obesity as adults. Additionally, mothers of these infants have increased risk for CVD in later life.^1,2 Despite decades of study, the etiology of PTB and SGA is not completely understood. PTB occurred in 11.7% of births in the United States in 2011; low birth weight (LBW), a broad classification describing infants <2,500 g and including both PTB and SGA or growth-restricted infants, occurred in 8.1% of births.³ African Americans and individuals of lower socioeconomic status bear a disproportionate amount of this burden,^3,4 and some propose that chronic stress may explain persistent racial and socioeconomic disparities in these pregnancy outcomes.^5–8

Allostatic load (AL) reflects the cumulative burden of chronic physiologic and psychologic stress and is measured by combining multiple subclinical biomarkers of systemic function into a single continuous index score of biologic risk.^9–11 AL increases with age and is higher among African Americans and individuals of lower socioeconomic status, populations believed to encounter daily chronic stress.^12,13 Higher AL is associated with CVD and may be a better predictor of CVD than individual risk factors.¹⁴ Given that mothers of PTB and SGA infants have increased risk for CVD, these women may also have higher AL as a subclinical multisystemic measure of risk. Furthermore, AL may provide a biologically plausible mechanism for the impact of chronic stress on pregnancy outcomes and later-life CVD.

Few studies have examined AL in pregnancy and its relation with pregnancy outcomes.^15,16 We aimed to determine whether past history of having had a pregnancy with a LBW outcome (SGA or PTB) is associated with higher AL scores in a nationally representative data set. We hypothesized that women reporting a history of LBW outcomes would have higher AL in comparison to women with normal birth weight outcomes, even accounting for covariates.

Materials and Methods

National Health and Nutrition Examination Survey

NHANES, a cross-sectional study that has been conducted in 2-year cycles since 1999, incorporates a complex survey design, with weighting of participants to make wider conclusions about the US population. We used data collected from 1999 through 2006. Continuous NHANES 1999–2006 was approved by the National Center for Health Statistics Research Ethics Review Board under protocols #98-12 and #2005-06. Questionnaire and examination components have been previously described.¹⁷ From 1999 through 2006, the reproductive-health questionnaire asked women about their history of LBW deliveries. We generated an 8-year weighting variable using 2- and 4-year mobile examination-assigned weights as described in the NHANES tutorial.¹⁸

Subpopulation

Because we wanted to study AL in young, relatively healthy nonpregnant women, we limited the study to women ages 17–35 with negative urine pregnancy tests. AL is known to increase with age; however, the rate of increase is not stable over one's life span.¹² The rate of increase is relatively constant in the younger adults; thus, we limited our study to younger women in the peak of their reproductive years so that we could also account for age in a linear regression model. We used two subpopulations: an index group (n=3,546) for determining high-risk cutoffs in this younger population and a study group (n=877), a smaller subset of women within the index group who also had a history of pregnancy, values for all components of AL, and responses to the reproductive-health questionnaire that would allow them to be assigned to a birth weight outcome group.

LBW outcomes

We used two questions—“Did any of your children weigh less than 5½ pounds (2,500 g) at birth?” and “How many of these babies were born preterm?”—from the Reproductive Health Questionnaire to categorize subjects into three birth weight groups. Preterm was designated as a birth at 36 weeks or less. We considered infants of women reporting LBW and no history of PTB as a surrogate for SGA. Women reporting LBW and PTB were considered as the PTB group, and those with no history of LBW were categorized as normal birth weight outcomes. Women who could not report a history (“I don't know”) were excluded.

AL

We calculated an AL score using nine individual measures of system function used in several previous studies of AL.^11,13,16,19 These selected components were systolic blood pressure, diastolic blood pressure, body mass index (BMI), glycosylated hemoglobin (HbA1c), high-density lipoprotein (HDL), total cholesterol, C-reactive protein (CRP), serum albumin, and creatinine clearance. We calculated creatinine clearance from serum creatinine as determined in NHANES.²⁰ All component values came from the NHANES data set as determined by NHANES standard examination (systolic and diastolic blood pressure and BMI) and laboratory procedures (HbA1c, HDL, total cholesterol, CRP, serum albumin, and serum creatinine).²¹ There is no “gold standard” for the measurement of AL; we used the group AL index formulation, a standard count-based algorithm for determining AL score.^11,22 We first defined high-risk values as the highest (or lowest) quartile among the index population. We assigned a point for each component for which each subject was in the high-risk quartile. We generated a composite AL score by summing the high-risk category scores. AL score ranged from 0 to 9. We also generated a binary AL variable, with elevated AL defined as a score at or above the 75th percentile.

Other covariates

We evaluated several demographic and medical-history characteristics of subjects for description and consideration as covariates in our analysis. Age, race/ethnicity, annual household income, education, and marital status were reported in the demographic data set. We also considered BMI as a continuous variable as a potential covariate and included it in our analysis. We combined individuals reporting either Mexican American or Hispanic into a single racial/ethnic category “Latino/Hispanic.” Other self-report races/ethnicities were white, African American, and other (including Asian and multiracial). Subjects self-reported smoking history (>100 lifetime cigarettes), history of hypertension, and breastfeeding history (yes/no) in the smoking, cardiovascular health, and reproductive-health questionnaires, respectively.

Statistical analyses

All statistical analyses were weighted by established strategies of NHANES.¹⁸ We analyzed AL score, using the accepted statistical methods described in the existing literature.^13,14,23 For all calculated means, we have provided the linearized standard error (SE). We compared demographic and descriptive data between the three birth weight outcome groups, using chi-square tests for proportions and linear regression for mean values to account for the weighting methodology. We also evaluated the relationship of each of these demographic components to AL, using linear regression, and included significant predictors in the model. We then used linear regression to compare AL among the three outcome groups. We calculated means for each group, using linear combinations. We selected covariates as those that were either different (p<0.10) between groups or significant (p<0.10) predictors of AL. Age, African American race, and BMI met criteria for inclusion in the model. However, because of the small number of women in the SGA group (n=24), we were concerned that adjusting for three covariates would overfit the model. As our analysis of characteristics showed no difference in BMI between the SGA and normal birth weight groups, the most parsimonious model for these comparisons included only African American race and age as covariates. We therefore ran both the two- and three-covariate linear regression models for comparing the PTB and SGA groups to the normal birth weight group. We then used logistic regression to determine adjusted and unadjusted odds ratios (ORs) for having elevated AL between the SGA and preterm groups, using the normal weight outcome group as a referent. Significance was defined as a p-value <0.05. All statistical analyses were performed using Stata Statistical Software Version 12 (StataCorp. LP, College Station, TX).

Results

We identified 3,546 nonpregnant women between the ages of 17 and 35 for the index group; 877 women met the inclusion/exclusion criteria for the study group. The leading reason for exclusion from the smaller subset of women included in our analyses was no prior pregnancy history (n=1,622). Other women did not have data available for all individual components (n=760), did not complete the reproductive-health questionnaire (n=286), or could not report an outcome (n=1). Women included in the study group tended to be older (29.1±0.2 years vs. 25.1±0.1 years, p<0.001) than those excluded from the study, but racial distributions were similar. Weighted proportions of women by birth outcome for the SGA, PTB, and normal birth weight groups were 2%, 10%, and 88%, respectively; characteristics by outcome group are included in Table 1. There was a statistically significant difference in BMI among women in these groups. Post hoc analyses indicated that women in the PTB group had a lower BMI than those in the SGA (p=0.008) and normal birth weight (p=0.003) outcome groups. There were no other significant differences.

Table 1.

Demographic and Descriptive Characteristics of Subjects by Birth Outcome

	Normal birth weight infant (n=776)	SGA infant (n=24)	Preterm infant (n=77)	p-value
Weighted percentage, %	87.5	2.1	10.4	—
Mean age±SE, years	29.0±0.2	29.0±0.9	29.4±0.7	0.85
Race				0.26
White	62.8	36.3	66.4
Latino/Hispanic	20.1	34.1	14.4
Black	14.4	29.5	15.7
Other	2.8	0.0	3.4
Annual household income <$20,000	21.6	36.5	18.6	0.36
Education				0.31
<High school diploma	21.2	26.2	32.7
High school diploma	28.2	25.7	27.4
Some college and beyond	50.5	48.1	39.9
Married/living with partner	69.1	75.2	64.4	0.65
Positive smoking history	42.3	43.8	54.8	0.52
Hypertension diagnosis	9.7	14.8	20.1	0.10
History of Breastfeeding	62.7	41.9	57.6	0.25
BMI, kg/m2	27.9±0.3	29.7±1.5	25.6±1.0	0.04a

^aSignificant difference between the preterm group and both SGA and normal-weight group. There is no statistically significant difference in the mean BMI in the SGA and normal- weight groups.

BMI, body mass index; SE, standard error; SGA, small for gestational age.

Using the index group, we determined high-risk cut-point values (Table 2). In the study population (n=877), AL scores ranged from 0 to 8, with a median score of 2 and an interquartile range of 1 to 4. The mean AL score and standardized linear error were 2.49 and 0.07, respectively. As validation of our score, we then confirmed the well-established relationship among AL, age, and African American race.^12,13 African American women had higher AL (p<0.001), and AL score increased with age (p<0.001). In addition, there was a positive association between AL and BMI (Table 3). Therefore, we included age, African American race, and BMI (PTB only) in adjusted analyses.

Table 2.

Values for High-Risk Quartiles for Allostatic Load Components

Component	High-risk valuea	Subjects meeting threshold, %b
Systolic blood pressure, mm Hg	≥115	29.3
Diastolic blood pressure, mm Hg	≥73	30.5
BMI, m/kg2	≥30.6	28.9
Creatinine clearance, ml/min	<107.1	22.4
Serum albumin, g/dL	<4.1	24.0
HbA1c, %	≥5.3	27.2
HDL, mg/dL	<45	26.0
Total cholesterol, mg/dL	≥201	30.9
CRP, mg/dL	≥0.44	29.5

^aValues determined as the weighted 75th percentile or 25th percentile (HDL and creatinine clearance only) value of women ages 17–35 in NHANES, 1999–2006 (n=3,546).

^bWeighted percentages for study population (n=877) shown.

CRP, C-reactive protein; HbAlc, glycosylated hemoglobin; HDL, high-density lipoprotein.

Table 3.

Univariable Analyses of Allostatic Load and Demographic Characteristics of Study Population (n=877)

	Coefficient	Linearized SE	p-value
Age, years	0.061	0.014	<0.001
African American race	0.92	0.16	<0.001
BMI, kg/m2	0.15	0.0079	<0.001
Married/marriage-like	−0.21	0.15	0.17
Household income <$20,000	0.21	0.17	0.22
History of breastfeeding	−0.13	0.15	0.37
Educationa	−0.045	0.17	0.61
Positive smoking history	0.064	0.15	0.66

^aEducation compared categorically to <high school education as the referent. Overall p-value and coefficient for comparison of college education to <high school shown.

Women in the SGA group had higher AL scores than did women in the normal birth weight group (3.66±0.42 vs. 2.42±0.07, p=0.003). This difference persisted after adjusting for both African American race and age (3.39±0.36 vs. 2.29±0.08, p=0.003) and African American race, age, and BMI (2.82±0.35 vs. 1.92±0.07, p=0.011). Results for both adjusted analyses are shown in Table 4. Our selected models for SGA and PTB were age and race/ethnicity and age, race/ethnicity, and BMI, respectively. Crude AL scores in the PTB group were no different when compared to the normal group (p=0.22); however, after adjustment for age, race/ethnicity, and BMI, AL scores for the PTB group were higher (2.58±0.21 vs. 1.92±0.07, p=0.001) (Table 4). This finding appeared to be related to higher scores among normal-weight women (BMI <25 kg/m²) with preterm vs. normal-weight births. In exploratory analyses of AL scores stratified by normal weight (BMI <25 kg/m²) and overweight/obese (BMI ≥25 kg/m²) women in the PTB group had higher AL than those in the normal birth weight group (2.30±0.22 vs. 1.50±0.08, p<0.001). Among overweight/obese women, AL scores in the PTB group were not significantly higher (3.43±0.47 vs. 3.03±0.10, p=0.40).

Table 4.

Mean Allostatic Load Scores and Odds Ratios for Normal Birth Weight, Small for Gestational Age, and Preterm Birth Groups

	Mean AL± linearized SE						OR (95% CI)
	Unadjusted	p	Age- and race-adjusteda	p	Age-, race-, and BMI-adjustedb	p	Unadjusted	p	Adjustedc	p
Normal birth weight (n=776)	2.42±0.07	—	2.29±0.08	—	1.92±0.07	—	1.00 (referent)	—	1.00 (referent)	—
SGA (n=24)	3.66±0.42	0.003	3.39±0.36	0.003	2.82±0.35	0.011	3.67 (1.39–9.68)	0.009	3.46 (1.36–8.79)	0.009
Preterm birth (n=77)	2.73±0.23	0.19	2.56±0.22	0.22	2.58±0.21	0.001	1.23	0.5	2.22 (1.04–4.79)	0.04

^aAge- and race-adjusted: age (continuous) centered at 29 years and African American race; adjusted values shown are for the referent group (nonwhite, 29 years of age).

^bAge-, race-, and BMI-adjusted: age (continuous) centered at 29 years, African American race, and BMI (continuous) centered at 25; adjusted values shown are for the referent group (nonwhite, 29 years of age).

^cAdjusted values presented as age- and race-adjusted for SGA; age-, race-, and BMI-adjusted for preterm birth.

AL, allostatic load; CI, confidence interval; OR, odds ratio.

We then defined elevated AL, using a 75th percentile score (greater than or equal to 4). The odds of falling into this elevated AL group for women reporting a history of SGA were higher compared to those with normal birth weight outcomes in the unadjusted (OR 3.67, 95% confidence interval [CI] 1.39–9.68) and adjusted models (OR 3.46, 95% CI 1.36–8.79). Again, excess risk was associated with past history of a preterm infant delivery in the adjusted model (OR 2.22, 95% CI 1.04–4.79) (Table 4).

Discussion

We demonstrate in a representative US population that nonpregnant women of reproductive age with history of LBW outcomes have higher AL scores than do their peers who delivered normal-weight infants. We observed this difference in both unadjusted and adjusted analyses. This novel study examines AL in a population of young, reproductive-age women and relates AL to a previous reproductive outcome in women postpregnancy. Notably, the component values used for defining high risk in this population fall below those for defining clinically significant disease; nonetheless, those women with subtly higher values relative to their peers were more likely to have had an SGA or preterm infant. As AL itself is a known risk factor for CVD, these findings align with the association between PTB and SGA and increased maternal risk for later-life CVD and metabolic syndrome.^1,24 Because CVD typically presents in later life, it is difficult to capture increased CVD risk in a young population. Based upon our demonstration of an association of adverse pregnancy outcomes with higher AL, we suspect that higher AL scores in our population may identify at-risk women with subclinical CVD.

This work supports a growing body of literature that psychosocial stress is associated with adverse pregnancy outcomes. Prior studies using subjective assessments of stress have found associations between stress and anxiety during pregnancy and perinatal outcomes.^25–28 Our findings using AL score as an objective measure of chronic stress are consistent with those findings. AL conceptually captures the “cost” of adaptation to stressors. The premature wear-and-tear experienced by individuals exposed to chronic stress can manifest itself in both long-lasting changes in measurable biomarkers and organ damage.²⁹ This model of physiologic and psychologic stress may provide insights into the increased risk of PTB and SGA for African American women, as it could represent a biologically plausible mechanism for increased risk present prior to pregnancy for adverse pregnancy outcomes. Persistence of this association despite adjustments for age, race/ethnicity, and BMI may be indicative of another nonquantified, nondemographic contributor to AL score, which plausibly could be the physiologic impact of chronic stress.

This study is one of the few to investigate the relationship between AL and pregnancy outcomes. A small study by Wallace and Harville examined the relationship of birth outcomes to AL measured at 26 to 28 weeks gestational age.¹⁵ They found a borderline negative association between AL and gestational age and a nonsignificant positive trend between AL and birth weight,¹⁵ which is not consistent with the findings of our study. Furthermore, other trends in their data were opposite to those expected. They found that African Americans tended to have lower, not higher, AL scores. Morrison et al. also measured AL in pregnant women in NHANES and could not demonstrate an association between higher AL and African American race.¹⁶ Neither study demonstrated an association with AL in pregnancy and characteristics well associated with higher AL. Their unexpected findings may relate to the timing of samples examined during pregnancy. The Morrison study measured variables throughout pregnancy; the Wallace and Harville study, in late pregnancy. These biomarkers change with advancing pregnancy, and the indices of AL in these studies may reflect adaptation to pregnancy instead of cumulative lifetime stress. We studied nonpregnant women and related postpregnancy AL to history of LBW outcomes. In this population, we replicated well-known associations of age and race/ethnicity with AL and described an association between adverse pregnancy outcomes and AL. By designing a study of nonpregnant women to evaluate the relationship of AL to adverse pregnancy outcomes, we eliminated the impact of pregnancy physiology on AL. Although our findings demonstrate this relationship in postpregnancy, we posit that these subtle differences may have been present prior to pregnancy and contributed to these adverse outcomes.

A peculiarity in these findings is the association between PTB and higher AL observed only in BMI-adjusted analyses. We noted that the BMI in this group was lower in comparison to that in the normal group. In analysis not adjusted for BMI, there appeared to be no difference in AL between the groups; however, this difference is observed when we controlled for BMI, which suggests that BMI impacts multiple systems and AL scores. Previous literature has demonstrated an association between BMI and inflammation independent of other metabolic factors.³⁰ We believe that we initially observed no significant difference, as the higher BMI in the normal birth outcome group masked the effect within both the BMI and CRP components; once controlled, we then observed a difference in AL between the normal and preterm groups.

A limitation of the study is the self-report definitions we used for SGA and PTB. We defined SGA and PTB outcomes based upon available answers to two questions related to birth outcome in NHANES. Unfortunately, there was no additional information providing either the exact infant birth weight or gestational age. We assumed recall of birth weight history and PTB to be reliable in this population, given previous work examining high reliability between recall and registered birth weight and gestational age.³¹ In defining the LBW outcome categories, we suspect that the number of SGA outcomes is underestimated; for term infants, 2,500 g is far below the 10th and even 3rd percentiles, depending on gestation week and infant sex.³² Given that SGA can be defined at the 3rd, 5th, or 10th birth weight percentile, the SGA group we identified in NHANES would include only the smallest term infants and would exclude some larger infants who would have typically been classified as SGA. Likewise, we may have also underestimated the number of preterm births in our population, because subjects were first asked to identify LBW deliveries and then subsequently PTB. At 36 weeks gestation, the median birth weight is ∼2,700 g; this median does not cross the 2,500 g threshold for LBW until less than 35 weeks gestation.³² Therefore, we suspect that our analysis of PTB may be more heavily weighted toward infants born at less than 34 weeks gestation. Thus, although our definitions were limited, our SGA and PTB categories likely have more severely growth-restricted and earlier preterm infants, respectively, than the standard definitions for either category.

An additional limitation in the self-report LBW survey in NHANES is our inability to differentiate those who had normal birth weight deliveries from those who had macrosomic infants. Although the 1999–2006 NHANES surveys include questioning regarding LBW, there is no questioning regarding delivery of large infants. Therefore, our normal group includes these deliveries as well. As we were able to adjust for BMI in our study population, we likely reduced the impact of deliveries of macrosomic infants in our normal birth weight group.

The most important question that we cannot answer is whether these differences in AL (1) existed prior to pregnancy or (2) developed during or after pregnancy as a result of the adverse outcome. We hypothesize that the former is more likely, which is also consistent given the shared risk factors for these adverse outcomes, AL, and later-life CVD.³³ This question needs to be resolved with prospective studies. Our study also informs the design of such studies. Although assessing AL before pregnancy would be the definitive test, doing so is extraordinarily difficult without introducing selection bias for patients with planned pregnancies. It is likely that future studies will be done during pregnancy. The fact that we did replicate usual relationships and identify an association to pregnancy outcomes in these nonpregnant women after pregnancy, whereas studies in late pregnancy did not, suggests that the timing of the measurement of AL is particularly important in pregnancy and that AL may be better interpreted during a point in pregnancy when the maternal physiology is most similar to a nonpregnant state. Given the dynamic changes in the cardiovascular, metabolic, neuroendocrine, and inflammatory systems that occur as a part of normal pregnancy and current understanding of maternal physiology, we believe that early pregnancy is the time point in pregnancy that most closely resembles the nonpregnant state.³⁴ We suggest that future studies of AL and pregnancy outcomes consider early pregnancy as a time point in pregnancy that may provide a more accurate estimation of prepregnancy AL and that may replicate well-accepted associations between AL and socioeconomic and demographic variables.

Conclusions

To our knowledge, ours is the first study to demonstrate an association between the adverse pregnancy outcomes SGA and PTB and higher AL. We demonstrate this association in nonpregnant women postpregnancy in younger women without overt clinical disease. Practically, this study informs future study of AL and adverse pregnancy outcomes, particularly given that previous studies of AL during pregnancy have not been consistent with well-accepted associations. As AL is a measure of the cumulative impact of chronic stress, this work suggests that chronic stress, in addition to contributing to disease burden in older adults, contributes to subtle dysregulation in younger women with a history of PTB or SGA. Higher AL among these women may explain not only the increased CVD risk among these women but also the persistent racial and socioeconomic disparities in health and pregnancy outcomes in the United States.

Acknowledgments

This work was supported by a grant from the Doris Duke Charitable Foundation to the University of Pittsburgh to fund Vanessa J. Hux as a Doris Duke Clinical Research Fellow; an NIMH Research Education Grant (5R25 MH054318) to Gretchen Haas, PhD; and the Preeclampsia Program Project (NICHD P01 HD030377). We also gratefully acknowledge Jia Xu for her guidance in NHANES statistical methods.

Author Disclosure Statement

No competing financial interests exist.