Discrimination and risk of spontaneous abortion in a preconception cohort study

Data were derived from Pregnancy Study Online (PRESTO), a web-based preconception cohort study of pregnancy planners who reside in the U.S. and Canada. 21 Beginning in June 2013, PRESTO recruited participants through social networking sites ( e.g ., Facebook). Eligible participants were assigned female at birth, aged 21–45 years, and were attempting to conceive spontaneously without fertility treatment at enrollment. Participants completed a baseline questionnaire on socio-demographics, lifestyle, reproductive, and medical histories. Participants also completed follow-up questionnaires every 8 weeks until pregnancy or up to 12 months (whichever came first). Among participants that conceived, additional questionnaires were administered in early pregnancy (~8 weeks’ gestation), late pregnancy (~32 weeks’ gestation), and at 6 months postpartum. In July 2019, we launched an optional supplemental Life Course Experiences Questionnaire (LCEQ). 19 All former participants were recontacted and invited to complete the LCEQ retrospectively, while newly-enrolled participants were invited to complete the LCEQ 30 days after enrollment. This study protocol was approved by the Boston University Medical Campus Institutional Review Board and participants provided informed consent. On the LCEQ, we ascertained everyday and lifetime discrimination using adapted versions of Williams’ Everyday Discrimination and Major Experiences of Discrimination scales. 19 , 22 – 24 For everyday discrimination, we queried participants about five experiences in their daily life before enrolling in PRESTO ( i.e ., “treated with less courtesy or respect than other people,” “received poorer service than other people at restaurants or stores,” “people acted as if they thought you were not smart,” “people acted as if they were afraid of you,” “threatened or harassed”). Response options included “never,” “less than once a year,” “a few times a year,” “a few times a month,” “at least once a week,” and “almost every day.” For lifetime discrimination, we asked participants about their experiences in the following settings: “on the job (hiring, promotion, firing),” “in housing (renting, buying, mortgage),” “by the police (stopped, searched, threatened),” “in the courts,” “at school,” or while “getting medical care.” Response options were “yes” or “no.” Participants who reported discrimination were asked to report the main reason(s) for their experiences (choose all that apply): “race or ethnicity,” “sex or gender,” “education or income level,” and “other.” 24 , 25 Guided by Krieger et al., participants were also asked how they responded when they felt they were treated unfairly (choose one from each set of response options): “keep it to yourself” (hereafter “quiet”) or “talk to other people about it” (hereafter “talk”) and “accept it as a fact of life” (hereafter “accept”) or “try to do something about it” (hereafter “act”). 25 , 26 We collected information on pregnancy outcomes ( e.g ., viable pregnancy, miscarriage/chemical pregnancy, induced abortion, ectopic/tubal pregnancy, blighted ovum) on follow-up and pregnancy questionnaires. We defined SAB as miscarriage (including chemical pregnancy) and blighted ovum <20 weeks’ gestation. Among lost-to-follow-up participants, we sought additional outcome information by emailing and phoning participants, searching social media, and linking with birth registries in selected states (CA, FL, MA, MI, NY, OH, PA, TX). 27 We also collected information on gestational weeks at loss and the pregnancy end date. Among participants who provided their pregnancy due date but not their gestational weeks at SAB, we estimated gestational weeks as: (pregnancy end date – (pregnancy due date – 280 days))/7. 28 When participants’ pregnancy due date and gestational weeks at SAB were missing, we estimated gestational weeks as: (pregnancy end date – last menstrual period date)/7. We censored participants reporting an induced abortion or ectopic/tubal pregnancy at their date of pregnancy loss. At baseline, participants reported their residential address, race, ethnicity, mother’s age at their birth, highest level of education completed by themselves and their parent(s)/caregiver, and annual household income. Participants also reported data on lifestyle factors ( e.g ., cigarette smoking, cannabis use, intake of caffeine or sugar-sweetened beverages, multivitamin use, physical activity, sleep duration “in the past month,” frequency of sleep trouble “in the past 2 weeks”), reproductive factors ( e.g ., age at menarche; menstrual cycle regularity and length; parity; last method of contraception; doing something to improve chances of conception [ e.g ., electronic fertility monitoring]; history of infertility, SAB, gynecologic conditions), and healthcare history and utilization ( e.g ., health insurance, perceived stress “in the past month” [10-item Perceived Stress Scale ranging 0–40, PSS], 29 depressive symptoms “in the past 2 weeks” [Major Depression Inventory ranging 0–50, MDI], 30 post-traumatic stress disorder [PTSD], depression, anxiety). We calculated diet quality via the Healthy Eating Index ranging 0–100 from a food frequency questionnaire completed ten days after baseline, 31 and body mass index using weight in kilograms divided by height in meters squared. On the LCEQ, participants reported adverse childhood experiences via items adapted from the Behavioral Risk Factor Surveillance System; 32 , 33 childhood financial hardship (defined as not having enough money for living expenses, having to borrow money for medical expenses, or receiving public assistance before age 18) via the sociocultural module of the U.S. Department of Health and Human Services’ Breast Cancer Core Questionnaire; 34 , 35 and length of time the participants’ standard of living could be maintained if they suddenly lost all sources of household income. Between June 2013 and March 2025, 21,804 eligible participants completed the baseline questionnaire. We excluded participants who completed the baseline questionnaire >60 days after screening (n=58), who did not conceive during the study period (n=4,214), and who were lost to follow-up (n=5,762). We also excluded participants who did not complete the LCEQ (n=4,411). LCEQ non-respondents were slightly more likely to be racial and ethnic minorities and have a SAB history ( eTable 1 ). Our analytic sample included 7,359 participants. We defined time zero as the week of first positive pregnancy test (when available) or 4 weeks’ gestation (cohort median). Participants contributed gestational weeks from time zero until the date of SAB or a censoring event ( i.e ., induced abortion, ectopic/tubal pregnancy, loss to follow-up, or 20 weeks’ gestation), whichever came first. We used Cox models to estimate hazard ratios (HRs) and 95% confidence intervals (CIs), with gestational weeks as the time scale. To account for differential response to the LCEQ, we calculated and applied stabilized inverse probability of selection weights among SAB eligible participants in the PRESTO cohort ( eTable 2 ). 36 To account for left truncation from differential timing of pregnancy recognition, we used the Andersen-Gill data structure, 37 with each observation representing one gestational week at risk. 38 , 39 We used a directed acyclic graph to guide the selection of potential confounders ( eFigure 1 ). 40 – 43 The first multivariable model (“minimally adjusted”) included age (<25, 25–29, 30–34, ≥35 years) and race/ethnicity (non-Hispanic White; non-Hispanic Black; non-Hispanic Asian; non-Hispanic American Indian, Alaskan Native, or Indigenous [AI/AN]; non-Hispanic Multiracial; non-Hispanic other race (defined as Middle Eastern or North African; Native Hawaiian or Pacific Islander; some other race; unknown race); Hispanic White; Hispanic Black; Hispanic Multiracial; and Hispanic Other Race (defined as Asian; AI/AN; Middle Eastern or North African; Native Hawaiian or Pacific Islander; some other race; unknown race). The second model (“fully adjusted”) further adjusted for highest level of parental education (≤12, 13–15, 16, ≥17 years), adverse childhood experiences (score: 0, 1–3, ≥4), and childhood financial hardship (yes vs. no). Assuming discrimination is a common cause of both prior and index reproductive outcomes, we did not adjust for reproductive history to avoid collider or overadjustment bias. 40 , 44 In categorical analyses, we modeled everyday discrimination ( i.e ., disrespect, poor service, not smart, afraid of you, harassed) by type and frequency of experiences using a floating reference group ( i.e ., not experiencing that specific type of discrimination). We also characterized everyday discrimination using a summary variable, assigning a score to each value of the Likert scale: 0=never, 1=less than once a year, 2=a few times a year, 3=a few times a month, 4=at least once a week, and 5=almost every day, summing across each type of discrimination (range: 0–25). We then divided this summary score variable into five categories based on the cohort distribution: none (0; referent), low (1–2), medium (3–4), high (5–6) and very high (≥7). 7 , 19 We modeled lifetime discrimination as any experience (yes vs. no), type of experience (on the job, in housing, by police, in the courts, at school, and getting medical care vs. not that type), and number of experiences (1, 2, ≥3 vs. 0). For summary variables of discrimination, we also generated restricted cubic splines to visualize the shape of these associations with SAB. We conducted mediation analyses with exposure-mediator interaction. 45 , 46 We investigated the effects of short sleep duration (<7 hours/night) and indicators of mental health ( i.e ., perceived stress, depressive symptoms) that operate as mediating pathways ( i.e ., natural indirect effects) of the discrimination-SAB association. What remains is the natural direct effect of everyday discrimination (score ≥7 vs. 0) and lifetime discrimination (≥3 events vs. 0) on SAB, representing the pathway that operates independently of the potential mediator. We computed the percentage mediated as: (HR NDE × [HR NIE – 1])/(HR NDE × HR NIE – 1). Finally, we cross-classified summary variables of everyday and lifetime discrimination with participants’ attribution ( e.g ., “race or ethnicity,” “sex or gender,” or “education or income level” vs. “other attribution”). We repeated the subgroup analyses of “race or ethnicity” attribution excluding non-Hispanic White participants reporting “race or ethnicity” as their main reason for discriminatory experiences to improve interpretability of the effects of race-based discrimination among racial and ethnic minorities. We also created cross-classified summary variables of everyday and lifetime discrimination with participants’ response type as “quiet/accept” or “talk/act.” We stratified analyses by race/ethnicity (non-Hispanic White; non-Hispanic other race [defined as Black; Asian; AI/AN; Middle Eastern or North African; Native Hawaiian or Pacific Islander; multiracial; some other race; unknown race; combined due to small sample sizes], Hispanic) to assess ethnic differences, parity (yes vs. no, as parous participants may more consistently recognize early pregnancy), LCEQ timing (retrospective vs. prospective completion, assuming more valid associations among participants who completed the LCEQ prospectively), and before vs. after June 2020 for LCEQ completion (coinciding with COVID-19 shutdowns and civil unrest following the death of George Floyd). We repeated analyses restricted to U.S. residents, participants who conceived without fertility treatment (as SAB etiology may differ for participants who did vs. did not conceive with fertility treatment), and participants with SAB at <8 gestational weeks. Missing data was minimal, with <1% for discrimination and gestational weeks at SAB. We imputed missing values using fully conditional specification methods. 47 , 48 We created 20 imputed datasets, then pooled effect estimates and standard errors using Rubin’s rule. 49

Over follow-up, 22.2% of participants reported SAB and 69.7% of those SABs occurred <8 weeks’ gestation. Mean age was 30.7 years, 54.1% of participants had ≥17 years education, and 61.6% had ≥$100,000/year household income. Most participants (85.9%) identified as non-Hispanic White, 4.1% as Hispanic White, 1.7% as Black, and 8.3% as some other race. Nearly 15% reported very high everyday discrimination (scores ≥7), whereas 9% reported ≥3 events of lifetime discrimination ( Table 1 ). Participants with higher levels of everyday and lifetime discrimination were more likely to be racial and ethnic minorities and have lower education and income. Similarly, participants with higher discrimination had higher prevalence of high perceived stress and severe depressive symptoms. Lost-to-follow-up participants were similar to those included across most characteristics ( eTable 3 ). Participants who completed the LCEQ prospectively were less likely to be parous and use hormonal contraception, but more likely to have private health insurance and be diagnosed with PTSD or anxiety ( eTable 3 ). The most prevalent types of everyday discrimination were people acting as if you were not smart (62.6%) and being treated with disrespect (61.6%; eTable 4 ). For lifetime events, job and school-related discrimination was most prevalent at 31.9% and 20.3%, respectively. Among participants responding affirmatively to discrimination, “sex or gender” was the most common attribution (80.9%) while “talk to other people about it” (59.2%) and “accept it as a fact of life” (71.3%) were the most common responses. Everyday discrimination was associated with greater incidence of SAB ( Figure 1 ; eTable 5 ). Across different types of everyday discrimination, HRs generally ranged from 1.20 to 1.46 when comparing experiences of discrimination ≥a few times a month vs. never. For example, experiences of people acting as if they were afraid of you were associated with 46% higher incidence of SAB (95% CI: 0.98–2.17). Conversely, experiences of disrespect were not associated with SAB. Higher scores of everyday discrimination showed weak positive associations with SAB incidence. Relative to no everyday discrimination, HRs for low, medium, high, and very high scores were 1.12 (95% CI: 0.95–1.31), 0.96 (95% CI: 0.81–1.12), 1.07 (95% CI: 0.91–1.27), and 1.15 (95% CI: 0.96–1.37), respectively. When we stratified by race/ethnicity, associations between everyday discrimination and SAB incidence persisted but were imprecise ( eTable 6 ). Multiple lifetime discrimination events were associated with higher incidence of SAB ( Figure 1 ; eTable 5 ). Relative to 0 events, HRs for 1, 2, and ≥3 events were 0.98 (95% CI: 0.86–1.12), 1.21 (95% CI: 1.04–1.42), and 1.19 (0.99–1.43), respectively. Police-related discrimination showed the strongest association with SAB (HR=1.24, 95% CI: 1.01–1.53), followed by housing-related discrimination (HR=1.19, 95% CI: 0.96–1.48) and discrimination when receiving medical care (HR=1.19, 95% CI: 1.03–1.37). Across racial and ethnic groups, associations of police-related discrimination and SAB persisted, but were strongest among Hispanic participants (HR=1.66, 95% CI: 0.92–3.00) and non-Hispanic other race participants (HR=1.36, 95% CI: 0.74–2.51; eTable 6 ). Associations between medical care-related discrimination and SAB were more pronounced among non-Hispanic other race participants (HR=1.37, 95% CI: 0.78–2.40) and non-Hispanic White participants (HR=1.21, 95% CI: 1.04–1.40). “Severe depressive symptoms” were the strongest and only consistent mediator of the association between discrimination and SAB: 10.8% for everyday discrimination (score ≥7 vs. no) and 7.8% for lifetime discrimination (≥3 events vs. 0; eTable 7 ). Mediation by short sleep duration and perceived stress varied by the type and timing of discrimination exposure. Greater everyday and lifetime discrimination was not associated with a greater incidence of SAB for those who reported attributions of “race or ethnicity” ( Figure 2 ; eTable 8 ) or “sex or gender” ( eTable 9 ). After excluding non-Hispanic White participants who reported “race or ethnicity” as an attribution, results were similar ( eTable 8 ). However, everyday and lifetime discrimination was more strongly associated with SAB incidence for those who reported attributions related to socioeconomic position (very high everyday discrimination [score ≥7] plus “education or income level” attribution): HR=1.30, 95% CI: 0.88–1.91 vs. no everyday discrimination and 2 lifetime events plus “education or income level” attribution: HR=1.24, 95% CI: 0.91–1.70 vs. no lifetime discrimination; eTable 10 ). Associations varied by discrimination type and response ( Figure 2 ; eTable 11 ). Participants with very high scores for everyday discrimination and “talk/act” response types vs. no everyday discrimination had the highest incidence of SAB, although imprecise (HR=1.28, 95% CI: 0.86–1.91). Conversely, among participants with higher levels of lifetime discrimination, those with “quiet/accept” response types to discrimination experienced the highest incidence of SAB: 2 events plus “quiet/accept” (HR=1.37, 95% CI: 1.07–1.75) and ≥3 events plus “quiet/accept” (HR=1.19, 95% CI: 0.88–1.59) vs. no lifetime discrimination. Associations of lifetime events were stronger among parous participants ( e.g ., housing, police, and school-related discrimination; eTable 12 ). Associations of everyday and lifetime discrimination with SAB incidence were generally stronger among participants who completed the LCEQ prospectively ( eTable 13 ) and before June 2020 ( eTable 14 ). When we restricted to 1) U.S. residents, 2) participants who conceived during follow-up without fertility treatment, and 3) SABs that occurred during the first 8 weeks of gestation, results were similar to the main analyses ( eTables 15 – 17 ).

In this preconception cohort study, we observed small positive associations between everyday discrimination and SAB incidence, and moderate positive associations between selected types of lifetime discrimination ( i.e ., police, housing, medical care) and SAB incidence, especially among certain racial and ethnic groups ( e.g ., police-related discrimination among Hispanic and non-Hispanic other race participants). Other notable findings included moderate associations with discrimination in housing among non-Hispanic White participants and in medical care among participants who identified as non-Hispanic other race. Only severe depressive symptoms consistently mediated associations between discrimination (everyday and lifetime) and SAB. Associations between discrimination and SAB incidence were generally strongest among participants who attributed their discriminatory experiences to socioeconomic position. We observed inconsistent patterns of SAB risk when discrimination was cross-classified with response types. This is the first study to estimate associations of discrimination and SAB incidence, adding to the growing literature on how systems of oppression ( e.g ., racism, sexism, classism) can have lasting impacts. Our results agree with previous studies identifying associations of discrimination with gynecologic conditions, 6 fertility, 7 and perinatal outcomes. 8 – 18 Although the landmark passing of the Civil Rights Act of 1964 may suggest discrimination is a relic of the past, persistent racial and socioeconomic disparities in health reveal a contrasting reality. In the present study, racial and ethnic minorities continue to bear the brunt of higher frequency of discriminatory experiences and associations varied by discrimination. The stronger associations observed for police, housing, and medical care-related discrimination suggest that the health effects of interpersonal discrimination can be embodied, 41 distinct from measures of structural or institutional discrimination ( e.g ., redlining, segregation), which have been associated with fertility and stillbirth. 50 – 53 These social determinants of health can lead to subsequent individual and neighborhood-level exposures and stressors that influence structural to biological pathways to SAB ( eFigure 1 ). 54 , 55 Our results also demonstrate the complexities of type and timing of discrimination, congruent with the broader literature on type and timing of stressors across the life course. 55 , 56 While this study indicates preconception discrimination is associated with SAB, associations were strongest among selected types of everyday discrimination (others acting afraid of you, poor service, harassment, or being perceived as not smart: ≥a few times a month vs. never), selected types of lifetime discrimination (police, housing, or medical care), and when participants attributed discrimination to “education or income level” instead of other attributions such as race or gender. The observed associations using summary variables of everyday and lifetime discrimination were generally weaker. The triangulation of these results suggests that cumulative stress may be characterized by time- and intensity-dependent critical windows, 54 not always reflected as monotonic associations, advancing our understanding of weathering pre-pregnancy stress on risk of SAB. Furthermore, systems of oppression have complex intertwined histories ( e.g ., police brutality, medical apartheid) and disentangling upstream factors of each discrimination exposure is challenging. The prevalence of attributions and responses to discrimination also vary across racial and ethnic groups. One potential explanation is how discrimination is differently received, internalized, and processed. 19 , 24 For example, through racial socialization, Black Americans are given “the talk” from a young age on how to keep themselves safe during encounters with police or healthcare professionals, 57 and are taught in their later years the unspoken rules of being “seen and not heard” in predominantly White spaces as a tactic of survival. 58 Moreover, previous work in PRESTO showed Black and Hispanic participants most often accepted the discriminatory experiences as a fact of life vs. took action. 19 In this study, we found that SAB incidence was higher for those experiencing multiple lifetime discrimination events and “quiet/accept” response types. It was also, therefore, unsurprising to observe stronger associations of police and medical care-related discrimination among non-Hispanic other race participants. Conversely, several scholars highlight the paradox of relative advantage when associations of discrimination are stronger among non-Hispanic White participants. 24 , 59 Briefly, they suggest 1) racial and ethnic minorities may be more likely to notice and adapt to discriminatory scenarios following exposure in early life (conditioning their later life exposure), 2) differential access to coping resources, such as religious service, 60 , 61 mitigate discrimination, and 3) greater perception, appraisal, and response to discrimination may facilitate recovery. Mechanisms by which discrimination affects health are well-theorized but largely understudied in empirical research. The effects of discrimination operate across socioecological levels and may occur intergenerationally. 41 , 62 In mediation analyses, we prioritized exploration of sleep deprivation and mental health measures as imperfect proxies of allostatic load. Allostatic load is a framework for understanding “wear and tear” on the body and other cumulative effects of life course stressors, 54 , 63 undermining the complex relations of discrimination and SAB. Herein, we observed severe depressive symptoms were important mediators of discrimination-SAB associations. Our findings parallel earlier work that identified depressive symptoms as a mediating pathway between everyday discrimination and infant birthweight in a randomized trial of pregnant U.S. residents. 16 In contrast, in the Coronary Artery Risk Development in Young Adults (CARDIA) study, depressive symptoms did not mediate associations between lifetime discrimination and low birthweight. 64 Study limitations include the lack of data on early life determinants ( e.g ., parental discrimination) and perpetrators of discrimination. This information would facilitate further investigation of participants’ repeated cycles of stress via proximity and imbalanced power dynamics ( e.g ., intimate partner, work supervisor). 65 We also did not ascertain frequency or severity of lifetime events ( e.g ., job, housing, police). Our inverse probability of selection weights could not be calculated as a function of discrimination exposure because these data were only ascertained on the LCEQ. To compensate for underrepresentation of selected participants in our study, we can assume our approach is valid if 1) LCEQ response is independent of discrimination exposure, conditional on the factors included in our probability model ( eTable 2 ), and 2) these factors sufficiently block all paths between LCEQ response and SAB. 36 We acknowledge that some associations may have been observed by chance and there is potential for greater recall bias among participants who completed the LCEQ retrospectively. Broad groupings of racial and ethnic groups in stratified analyses may have obscured our findings. We could not explore mediation by mental health diagnoses ( e.g ., PTSD, depression, anxiety) because they may have occurred before experiences of discrimination. We assumed that perceived stress and severe depressive symptoms were downstream effects because participants were asked to recall their mental health during the month or 2 weeks before enrollment, whereas our discrimination exposure captures experiences occurring at any point during adulthood or across the life course. 19 PRESTO is a web-based convenience sample of pregnancy planners and is not representative of the general population or all women of reproductive age. Although most PRESTO participants identify as non-Hispanic White and have higher-than-average socioeconomic status ( e.g ., education, income), measures of associations are not necessarily biased by self-selection. 21 , 66 – 68 We acknowledge that conditioning on pregnancy may induce collider stratification bias if 1) discrimination is associated with conception, and 2) unmeasured confounders influence both conception and SAB. 7 , 69 There is debate in the literature on best practices for accounting for this bias, if at all, and the implications of its causal interpretation. 69 – 71 Study strengths include early ascertainment of SAB in a geographically diverse cohort, validated discrimination measures, 19 , 22 – 24 analyses of attributions for and responses to discrimination, and everyday vs. lifetime discrimination-specific mediation. In conclusion, discrimination was associated with SAB among pregnancy planners, and the magnitude of association differed by race/ethnicity and attribution. Associations showed mediation by severe depressive symptoms. Future studies investigating the reproductive health consequences of discrimination are warranted, including exploration of additional mediators and resilience factors. Findings emphasize the need for interventions to reduce discrimination and its downstream effects.

Spontaneous abortion (SAB), defined as pregnancy loss <20 weeks’ gestation, 1 affects up to 20% of clinically-recognized pregnancies and 30% of post-implantation pregnancies. 1 – 3 SAB can have physical, psychological, and financial sequelae, 1 and is predictive of women’s long-term health. 1 , 4 Only age and history of SAB have been consistently identified as risk factors for SAB, warranting identification of other risk factors. 1 , 5 Multiple studies have evaluated discrimination in relation to reproductive and perinatal outcomes ( e.g ., fibroids, 6 fecundability, 7 preterm birth, 8 – 14 low birthweight, 8 – 11 , 15 , 16 small-for-gestational-age birth, 11 , 17 stillbirth 18 ). However, no studies have evaluated the effects of discrimination, from racism and/or other systems of oppression ( e.g ., sexism, classism), on SAB. The extent to which attributions ( e.g ., race, gender, education, income) and response to discrimination ( e.g ., take action, talk to other people) could modify these effects is also unclear. Discrimination could increase SAB incidence through several mechanisms, including physiologic stress, vascular resistance, inflammation, short sleep duration, mental health, and reduced access to societal resources ( e.g ., healthcare, education, income; eFigure 1 ). 19 Discrimination may further affect SAB incidence through adverse socioeconomic conditions that promote stress-related coping and impede the uptake of healthy behaviors ( e.g ., smoking, alcohol consumption, poor diet quality). 3 , 20 To expand the evidence on discrimination and health, we estimated the effects of everyday and lifetime discrimination on SAB incidence in a preconception cohort study. We also explored 1) the extent to which attributions for and responses to discrimination modified the associations, and 2) mediation by sleep and mental health measures.