Childhood adversity and spontaneous abortion in a North American preconception cohort study

Pregnancy Study Online (PRESTO) is an ongoing internet-based prospective cohort study of pregnancy planners in North America (2013-present). 24 Eligible participants are aged 21–45, assigned female at birth, residing in the United States or Canada, and trying to conceive without fertility treatment. Recruitment through paid social media advertisements has been described previously. 24 After screening eligible, participants complete a baseline questionnaire (BQ; available in English and Spanish) on demographics, lifestyle factors, reproductive and medical history, and follow-up questionnaires every 8 weeks for up to 12 months or until a reported pregnancy. If they conceive, they are prompted to complete the early pregnancy (~8–12 weeks’ gestation), late pregnancy (~32 weeks’ gestation), and 6-month postpartum questionnaires. Thirty days after baseline, participants are invited to complete the supplemental Life Course Experiences Questionnaire (LCEQ; only available in English). Childhood adversity was assessed on the LCEQ via the ACE scale and Brief Trauma Questionnaire (BTQ), 7 , 25 , 26 which we analyzed separately. The ACE scale was adapted from the Behavioral Risk Factor Surveillance System (BRFSS) ACE module and elicited data on 8 experiences <age 18 (hereafter “substantive domains”): having a household member with mental illness, a substance abuse problem, or who had been incarcerated, parental separation/divorce, intimate partner violence (IPV), and abuse (physical, emotional, sexual) ( Table S1 ). 25 We calculated participants’ cumulative ACE scores by summing affirmative responses to each substantive domain question (range: 0–8). The ACE substantive domains and resulting ACE scores have good to excellent test–retest reliability with weighted-kappa coefficients ranging from 0.41 to 0.76. 27 , 28 We further assessed physical and sexual abuse before age 18 years using the BTQ where participants were asked to report the frequency (once, a few times, or more than a few times) and timing (as a child [≤11 years], teen [12–17 years] or adult [≥18 years]) of their experiences ( Table S2 ). The BTQ has shown to have good to excellent interrater reliability (kappa coefficient range = 0.74–1.00). 26 Pregnancy outcomes were reported on follow-up, early pregnancy, late pregnancy, and postpartum questionnaires. We asked participants if they experienced any of the following outcomes since their last questionnaire: miscarriage (including chemical pregnancy), induced abortion, ectopic/tubal pregnancy, or a blighted ovum. Spontaneous abortion was defined as pregnancy loss <20 weeks’ gestation and comprised of questionnaire responses of “miscarriage,” “chemical pregnancy,” or “blighted ovum.” If the participant reported a pregnancy loss, they were asked the date of their pregnancy loss and gestational week at loss. If they did not report their gestational week at loss, we estimated gestational age using the following equation: ( p r e g n a n c y e n d d a t e − ( p r e g n a n c y d u e d a t e − 280 d a y s ) ) / 7 . If they did not report a due date, we estimated gestational age with the following equation: ( p r e g n a n c y e n d d a t e − l a s t m e n s t r u a l p e r i o d d a t e ) / 7 . We attempted to gather data on loss-to-follow-up participants by phone or email, through birth registry linkage in states with the largest numbers of participants (CA, FL, MA, MI, NY, OH, PA, TX), and by searching for baby registries and birth announcements online (<2%). 29 We measured childhood SSI using an adapted version of the Berkman–Syme Social Network Index (SNI) on the LCEQ. 30 The SNI asks participants about eight types of SSI, such as having friends, relatives, or community groups to confide in. 30 We asked participants to report on their experiences during childhood (<18 years) and adulthood (≥18 years; “in the year before study enrollment”). In this paper, we analyzed SSI during childhood (<18 years) due to insufficient sample size for low SSI in adulthood. We calculated childhood SNI scores by summing affirmative responses for SSI (range 0–8). Childhood SNI scores were dichotomized as higher (≥4) and lower (<4) SSI in line with definitions of “socially integrated” and “socially isolated” individuals in previous research. 30 Participants completed preconception questionnaires to report their alcohol consumption, smoking status, perceived stress (via the Perceived Stress Scale-10 [PSS-10]) 31 ; depressive symptoms (via the Major Depression Inventory [MDI]) 32 ; diagnosis of depression, anxiety, or PTSD; probable diagnosis of PTSD (via the Primary Care PTSD-5); adult trauma (via the BTQ); diagnosis of polycystic ovary syndrome (PCOS), endometriosis, uterine leiomyomata, and body mass index (BMI; kg/m 2 ). The PSS-10 is a widely used scale with proven internal consistency reliability through a Cronbach’s alpha >0.7 in 12 studies. 33 The MDI and PC-PTSD-5 have been validated showing acceptable sensitivity (0.86 and 0.95, respectively) and specificity (0.86 and 0.85, respectively). 32 , 34 Our selection of potential confounders was guided by the published literature and a directed acyclic graph ( Figure S1 ). 35 Participants reported their age (years), race, and ethnicity (categorized as Hispanic, non-Hispanic White, non-Hispanic Black, non-Hispanic other [defined as American Indian or Alaskan Native; Asian; Pacific Islander; Native Hawaiian; multiracial, Middle Eastern or North African; or some other race]) and highest level of parental education (≤high school, some college, Bachelor’s degree, graduate school) on the BQ. We used the age closest to the time of conception for this analysis. Participants reported childhood experiences of financial hardship (based on reports of limited income for food/housing, loan for medical expenses, used public assistance or welfare) through the LCEQ. We used multiple imputation with fully conditional specification methods to account for missing data and created 20 imputed datasets. We averaged the estimates across imputed datasets for the final analysis. 36 Missingness for the ACE module and BTQ was <1.3% and <0.05%, respectively. We imputed gestational age at loss for 43 participants (<1%). Childhood SNI questions had <1% missingness. All other covariates had 60 days after screening or completed the BQ >60 days after starting ( n = 51), did not conceive during follow-up ( n = 4106), or were lost to follow-up before conception ( n = 5128). Of the remaining 10 333 participants, 6100 completed the LCEQ and were included in this analysis. The 5128 participants lost to follow-up prior to conception were similar to the analytic sample with respect to age (30.1 vs 30.6), alcohol consumption (2.3 vs 2.9 drinks/week), diagnosed depression history (27.8% vs 26.1%), and diagnosed anxiety history (28.0% vs 28.1%), but differed with respect to race/ethnicity (non-Hispanic White: 72.0% vs 86.9%), education (<Bachelor’s degree: personal: 15.7% vs 2.1%; parental: 28.3% vs 11.3%), household income (<$50 000: 37.3% vs 10.3%), BMI (31.1 vs 26.6 kg/m 2 ), current smoking (20.6% vs 2.2%), perceived stress (PSS score ≥25: 15.1% vs 7.6%), depressive symptoms (MDI score ≥30: 12.0% vs 3.1%), PCOS diagnosis (17.0% vs 6.3%), and history of SAB (32.8% vs 24.8%; Table S3 ). The magnitude of these differences was much smaller comparing those who were and were not lost to follow-up after conception ( Table S3 ). The Boston University Medical Campus Institutional Review Board approved the study protocol. All participants provided online informed consent. We fit Cox proportional hazards regression models to calculate HRs and 95% CIs to estimate the association between childhood adversity and SAB. Participants contributed gestational weeks at risk starting from their first positive pregnancy test and were followed until SAB, 20 gestational weeks, or a censoring event (eg, induced abortion, ectopic pregnancy, study withdrawal, loss to follow-up), whichever came first. We used an Andersen–Gill data structure with one observation per at-risk gestational week. We modeled ACEs as a polytomous categorical variable (0: reference group, 1–3, ≥4), as a continuous variable (ie, per 1-unit increase in ACE score), and as individual ACE domains with 0 ACEs as the referent. ACE score categories were informed by previous literature. 7 , 12 , 37 , 38 Using the BTQ data, we calculated HRs for having experienced no abuse (reference group), physical abuse only, sexual abuse only, or both during childhood. We stratified these models by higher and lower childhood SSI (SNI score ≥4 vs <4) to observe potential modifying effects. Additionally, we leveraged the BTQ data to evaluate the association of frequency (once, a few times, more than a few times), timing (first exposure as a child, first exposure as a teen), and chronicity (only as a child, only as a teen, both as a child and teen) of childhood physical and sexual abuse on SAB risk, with no childhood physical or sexual abuse as the referent. We adjusted analyses for race/ethnicity, highest level of parental education, economic resources as a child, and age at conception. Race/ethnicity is a proxy for exposure to systemic racism (structural, institutional, interpersonal)—a known predictor of health outcomes and access to healthcare and related resources (eg, affordable housing, job opportunities, health care). 39 Moreover, people of color have reported higher prevalence of childhood adversity than those who identify as non-Hispanic White. 1 Therefore, race/ethnicity was modeled as a potential confounder. Given that childhood socioeconomic position has been associated with increased risk of both childhood adversity and inflammation in adulthood, 40 , 41 we adjusted for the highest level of parental education and childhood economic resources as measures of childhood socioeconomic position. The analyses for frequency, timing, and chronicity of abuse were also mutually adjusted for the other type of abuse (eg, physical abuse analyses were adjusted for sexual abuse). We implemented inverse probability weighting to address potential selection bias by differential completion of the LCEQ. We predicted the probabilities of LCEQ completion using logistic regression models that included covariates that potentially influence participation. When the LCEQ was launched in July 2019, former participants were invited to complete it retrospectively while newly enrolled participants completed it prospectively. Therefore, we calculated nonstabilized weights by retrospective vs prospective completion of the LCEQ. Participants with a lower probability of completing the LCEQ based on the factors in the model (eg, race/ethnicity, educational attainment, anxiety diagnosis) were given larger weights. Childhood adversity has been associated with health outcomes later in life. We conducted exploratory mediation analyses to better understand the relationship between having experienced both childhood physical and sexual abuse (vs no abuse) and SAB. We estimated the natural indirect and direct effects, overall and restricting to participants with lower childhood SSI since we observed the largest effect sizes among this group in the regression analyses. We estimated the percentages mediated, calculated as Natural Indirect Effect/Total Effect × 100%. 42 We explored potential mediators including alcohol consumption closest to conception (≥7 vs <7 drinks/week), high perceived stress closest to conception (PSS score ≥25 vs <25), severe depressive symptoms (MDI score ≥30 vs <30), lower BMI (<18.5 kg/m 2 ), higher BMI (≥25 kg/m 2 ), and current smoking at conception, depression diagnosis, anxiety diagnosis, probable PTSD diagnosis, adult trauma, PCOS, endometriosis, uterine leiomyomata—all coded as binary variable (yes vs no). In the mediation analyses, we adjusted for race/ethnicity and highest level of parental education. We also modeled the ACE scores with restricted cubic splines, which did not yield evidence for nonlinearity, so we retained its linear specification ( Figure S2 ). We conducted all analyses using SAS statistical software (version 9.4, SAS Institute).

Participants with ACE scores ≥4 were more likely to identify as Hispanic, non-Hispanic Black, or non-Hispanic other race; currently smoke; have education <16 years, household income <$50 000, highest level of parental education <16 years, higher mean BMI, higher perceived stress, severe depressive symptoms, diagnosis of depression, anxiety, and PCOS, and a history of SAB, compared to participants with ACE scores of 0 and 1–3 ( Table 1 ). The prevalence of higher childhood SSI was lowest among participants with ACE score ≥4. Compared with participants reporting no abuse during childhood, a higher proportion of participants who experienced physical abuse, sexual abuse, or both during childhood were Hispanic, non-Hispanic Black, or non-Hispanic of some other race. Participants reporting physical and/or sexual abuse during childhood were more likely to have household incomes <$50 000; lower parental education; severe depressive symptoms; a history of diagnosed depression, anxiety, probable PTSD, or PCOS; and a history of SAB ( Table 1 ). The prevalence of higher childhood SSI was lower among participants who experienced physical abuse only, sexual abuse only, or both, compared with no abuse. We observed some associations between ACEs and SAB in unadjusted analyses, but all were attenuated after covariate adjustment ( Table 2 ). Compared with those who reported no childhood abuse on the BTQ, those who experienced physical abuse only, sexual abuse only, and both had a slightly increased risk of SAB (HR = 1.11 [95% CI, 0.92–1.35]; HR = 1.12 [95% CI, 0.96–1.30]; HR = 1.09 [95% CI, 0.90–1.32], respectively). In analyses stratified by SSI, associations between childhood adversity and SAB were consistently stronger among participants with lower childhood SSI than among those with higher childhood SSI ( Table 2 ). Among participants with higher childhood SSI, we observed slight increases in SAB incidence among those who reported physical abuse only (HR = 1.11 [95% CI, 0.90–1.37]) and sexual abuse only (HR = 1.10 [95% CI, 0.93–1.29]) compared with no abuse in childhood on the BTQ. Among participants with lower childhood SSI, we observed slight-to-moderate associations between all ACE domains and SAB when compared with those with ACE score 0. Relative to participants with no childhood abuse on the BTQ, we observed associations between experiencing physical abuse only (HR = 1.21 [95% CI, 0.73–2.01]), sexual abuse only (HR = 1.34 [95% CI, 0.85–2.11]), and both (HR = 1.76 [95% CI, 1.15–2.68]) with SAB incidence. For frequency of physical and sexual abuse measured via the BTQ, we observed a slightly increased risk of SAB for participants who experienced physical abuse once (HR = 1.17 [95% CI, 0.86–1.59]) or a few times (HR = 1.14 [95% CI, 0.90–1.45]), and sexual abuse a few times (HR = 1.17 [95% CI, 0.96–1.42]) compared with no abuse ( Table 3 ). For timing of abuse, we observed a slight increase in SAB risk for first exposure to sexual abuse as a child compared with no abuse (HR = 1.27 [95% CI, 1.03–1.56]). For chronicity, when compared with no abuse, we observed slightly increased SAB risk for physical abuse both as a child and teen (HR = 1.21 [95% CI, 0.95–1.56]) and for sexual abuse only as a child (HR = 1.15 [95% CI, 0.89–1.48]), but the strongest association was observed for those who experienced sexual abuse both as a child and teen (HR = 1.48 [95% CI, 1.10–1.98]). In the overall mediation analysis, probable PTSD diagnosis mediated the largest percentage of the association between BTQ-measured childhood physical and sexual abuse and SAB risk compared with no abuse (51.5%), followed by anxiety diagnosis (27.7%), depression diagnosis (26.5%), adult trauma (24.7%), and MDI score ≥30 (21.1%) ( Table 4 ). When restricted to participants with lower childhood SSI, probable PTSD diagnosis still mediated the largest percentage of the association (42.8%), followed by adult trauma (14.3%) and anxiety diagnosis (13.2%). The percentage mediated was <10% for all other mediators explored.

In this North American preconception cohort study, there was little overall association between ACE scores and SAB risk. Some ACE point estimates suggested an increased risk of SAB but were imprecise and should be interpreted cautiously. However, BTQ-measured experiences of physical and sexual abuse before age 18, particularly sexual abuse, were associated with higher SAB risk. We observed stronger associations between BTQ-measured physical and sexual abuse and SAB risk among participants reporting lower SSI in childhood. Mediation analyses among participants with lower childhood SSI indicated that probable diagnosis of PTSD, adult trauma, and a history of diagnosed anxiety mediated the largest percentages of the association between childhood physical and sexual abuse and SAB risk. Our results support other findings from the literature. Several studies found associations between cumulative ACE scores and SAB risk, 12 , 13 , 43 , 44 similar to the current study’s results among participants with lower SSI. Previous studies yielded similar results for childhood adversity domains such as associations between SAB risk and household mental health, household substance abuse, and physical and sexual abuse in childhood. 12 – 14 Other published findings contradict our results. For instance, while Li et al . 43 found no association among those experiencing abuse in childhood only, those who experienced abuse in both childhood and adulthood had slightly increased odds of SAB. However, their single measure of abuse included having witnessed violence, 43 which could have weaker effects on SAB than experiencing violence. We observed slightly stronger associations of physical and sexual abuse with SAB risk when measured by the BTQ (compared with the ACE). Differences between ACE and BTQ likely reflect the scope of measurement. The BTQ assesses abuse by anyone, including peers or dating partners, whereas the ACE items are restricted to physical abuse by adults in the home or sexual abuse by someone 5 years older ( Tables S1 and S2 ). As a result, the BTQ may better capture the full spectrum of childhood adversity relevant to reproductive outcomes and ACE-based associations may be biased toward the null. Among those who experienced lower childhood SSI, mental health factors (ie, probable PTSD diagnosis, anxiety diagnosis) and adult trauma suggest a larger mediating role between childhood physical and sexual abuse and SAB risk than other lifestyle and reproductive factors explored (eg, alcohol consumption, smoking, menstrual disorders, BMI). Individuals who experience childhood adversity are at an increased risk of experiencing mental illness in adulthood (eg, PTSD, anxiety). 45 , 46 Additionally, experiencing childhood adversity may heighten individuals’ perceptions of traumatic events in adulthood compared with unaffected individuals. 47 These findings support previous literature indicating that PTSD is associated with higher odds of SAB. 48 This underscores the importance of addressing long-term mental health impacts of childhood adversity. This study had several limitations. First, participants lost to follow-up before conception were generally more socioeconomically disadvantaged and were more likely to have risk factors for SAB than participants included in the analysis. If participants lost to follow-up before conception differed with respect to ACEs and risk of SAB, selection bias may have influenced our findings. However, LCEQ respondents who were lost to follow-up after conception were much more similar to the analytic sample. Second, there are some drawbacks to measuring ACEs as a cumulative score. For instance, calculating a summed ACE score assumes each adverse event carries equal weight. 35 , 49 Given that not all experiences of childhood adversity are the same and may have different effects on SAB, we evaluated associations for each individual ACE domain. Third, inaccurate reporting of childhood adversity is likely due to stigma and the retrospective collection of childhood adversity data. Fourth, we were unable to determine the timing of childhood economic resources in relation to exposure to childhood adversity. However, adjusting for childhood economic resources was important due to its role as an early life stressor and its association with childhood adversity. 50 To address this limitation, we ran models with and without adjustment for childhood economic resources and observed similar patterns. Although, we observed some attenuation of selected associations for ACE score and ACE-ascertained physical and sexual abuse in models unadjusted for childhood economic resources (data not shown). Finally, as the cohort was recruited through social media and is predominantly White and highly educated and economically advantaged compared to the general US and Canadian populations, 51 – 54 the findings may not generalize to more diverse populations. This is one of the first prospective cohort studies to examine the association between childhood adversity and SAB. Prospective data collection allowed for identification of early SABs often missed in studies beginning follow-up later in pregnancy. A key strength was our careful control of potential confounders. We adjusted only for SAB risk factors that preceded childhood adversity, while prior studies adjusted for potential mediators like BMI and smoking. We also evaluated potential modification by SSI across the life course and explored mediators of the childhood adversity–SAB association. In summary, cumulative ACE scores were not consistently associated with SAB risk. In contrast, BTQ-measured physical and sexual abuse were associated with increased SAB risk, of which childhood sexual abuse was the strongest association especially among those with lower SSI. These findings suggest that childhood SSI buffers the deleterious impact of early life abuse on SAB risk. Mental health and adult trauma emerged as important mediators. Clarifying the buffering effects of SSI and key mediating pathways can guide targeted interventions to reduce adverse reproductive outcomes in adulthood following exposure to childhood adversity.

Childhood adversity—defined as potentially traumatic events occurring before age 18—is highly prevalent in the United States, with nearly two-thirds of adults reporting at least 1 adverse childhood experience (ACE), and 1 in 6 reporting ≥4 ACEs. 1 Childhood adversity is linked to poor adulthood (≥18 years) health outcomes including obesity, 2 substance use (eg, heavy alcohol consumption, smoking), sexual risk-taking, 3 and various reproductive health outcomes including age at menarche, 4 , 5 sexually transmitted infections, 6 and fertility. 7 Childhood adversity may cause chronic stress that disrupts immune and inflammatory responses. Given that appropriate immune responses are vital to establishing and maintaining a pregnancy, these effects may heighten spontaneous abortion (SAB) risk. 8 Studies have demonstrated that adults who experienced childhood adversity had resistance to glucocorticoid signaling. 8 – 11 Typically, for acute stressors, glucocorticoid signaling terminates the body’s inflammatory response once the stressor is alleviated, but these findings suggest that maltreated children may exhibit increased inflammation levels in adulthood after experiencing chronic stress from adverse childhood events. 8 – 11 Chronic stress can also disrupt the autonomic nervous system, and hypothalamic–pituitary–gonadal and hypothalamic–pituitary–adrenal axes, which coregulate immune function. 8 Epidemiologic research generally supports the hypothesis that childhood adversity increases SAB risk though research is limited. In a cross-sectional study, odds of SAB at first pregnancy increased by 4% for each additional adverse event experienced among 1511 women aged 18–45 from Louisiana. 12 In a retrospective cohort study of 2795 English women, those who experienced ≥3 ACEs were more than 3 times as likely to have experienced recurrent SAB. 13 In an Australian prospective birth cohort study, researchers observed increased miscarriage risk among those who experienced childhood maltreatment compared to those who experienced none. 14 Several studies have found an increased risk of pregnancy loss among people that experienced physical and/or sexual abuse 15 – 17 or other adverse experiences in childhood, 18 – 21 although the studies did not focus exclusively on SAB. Key limitations of existing studies include retrospective ascertainment of self-reported SAB, which may not capture early pregnancy losses (<8 gestational weeks) and introduce recall bias, and limited evaluation of potential modifiers and mediators of the association. We investigated the association between childhood adversity and SAB incidence in a North American preconception cohort study of pregnancy planners and the extent to which childhood social support and integration (hereafter “SSI”) modifies the association. We also evaluate potential mediators of the association, including post-traumatic stress disorder (PTSD), depression, and anxiety. Studies have shown that SSI is a potential buffer against the adverse effects of stress by regulating glucocorticoid responses. 22 Childhood SSI buffers stress responses for those whose temperaments would make the stressor especially damaging, indicating that stress responses are sensitive to SSI during early development. 23 We hypothesized that childhood adversity would increase SAB incidence, and childhood SSI would attenuate this association.

Supplementary material is available at the American Journal of Epidemiology online.