Racial/Ethnic Differences in Associations Between Traumatic Childhood Experiences and Both Metabolic Syndrome Prevalence and Type 2 Diabetes Risk Among a Cohort of U.S. Women

Early life exposures such as traumatic childhood experiences (TCEs) are understudied, preventable contributors to type 2 diabetes (1). TCEs are substantially stressful events, often resulting in enduring emotional, psychological, social, and/or physiological impacts (2). They may likely contribute to type 2 diabetes in adulthood through, for example, triggering biological stress-response pathways (e.g., the autonomic nervous system, the hypothalamus-pituitary-adrenal axis) leading to inflammation and altered insulin sensitivity (3). Furthermore, psychological stress during childhood or adolescence, in part through affecting the development of brain regions involved in emotional regulation, may prompt the adoption of unhealthy behaviors (e.g., overeating) that have been identified as risk factors for type 2 diabetes (3–5).

Like type 2 diabetes, TCEs are prevalent in the U.S., and there are sex/gender as well as racial disparities in TCE exposure. A recent nationally representative study reported that approximately 61% of U.S. children endured at least one TCE in the year prior to the study, and adolescent girls were more likely than boys to report certain TCEs (e.g., sexual abuse) (6). Furthermore, minoritized racial/ethnic groups in the U.S. had a higher prevalence of TCEs compared with non-Hispanic White (NHW) adults in a nationally representative sample, similar to the observed disparities in both well-identified type 2 diabetes risk factors (e.g., metabolic syndrome [MetS]) and type 2 diabetes (7–9). Yet, disparities research has been sparse.

Prior studies of the relationship between childhood adversities and type 2 diabetes have yielded mixed results, which are likely related to variation in measured traumas and in study populations (2,10–13). Much of the prior U.S. literature has focused on less generalizable populations, such as clinical samples (e.g., patients with mood disorders) or veterans; some of the studies had small, nondiverse samples, were cross-sectional, and few investigated associations by sex/gender or evaluated potential differences by race/ethnicity (2,10,12,14–20). Last, to our knowledge, U.S. studies have not yet investigated a robust range of TCEs that may uniquely contribute to type 2 diabetes risk (e.g., natural disaster, major illnesses) (2).

To address research gaps from prior studies, we investigated the association between TCEs and incident type 2 diabetes in a prospective cohort of NHW, Black/African American (BAA), and Hispanic/Latina (hereafter, Latina) middle- to older-aged U.S. women. We also assessed potential racial/ethnic variation in these associations. We hypothesized that 1) TCEs would be more prevalent and type 2 diabetes incidence would be higher among minoritized racial/ethnic groups compared with NHW women; 2) TCEs would be associated with a higher incidence of type 2 diabetes; and 3) these associations would be stronger among minoritized racial/ethnic groups compared with NHW women. MetS also disproportionately affects minoritized racial/ethnic groups, is one of the strongest risk factors for type 2 diabetes, and likely shares the same biological pathways as those between TCEs and type 2 diabetes (21,22). Therefore, we investigated prevalent MetS as both a potential effect modifier and mediator of the TCEs and type 2 diabetes relationship. We hypothesized that the TCEs and type 2 diabetes relationship would be stronger among women with versus without prevalent MetS and that MetS would also mediate associations.

The Sister Study is a prospective cohort study of 50,884 women who reside in the United States, including Puerto Rico. Details of the study protocol are described elsewhere (23). Briefly, the Sister Study was designed to identify environmental and genetic risk factors for breast cancer among women. To be eligible, women aged 35–74 years had to have a sister diagnosed with breast cancer but be free from breast cancer themselves at baseline data collection (2003–2009). Data collection included computer-assisted telephone interviews, a self-administered questionnaire, and an in-home visit at baseline. Brief follow-ups occurred annually, and detailed follow-ups occurred every 2–3 years (Supplementary Figure 1). The study is overseen by the Institutional Review Board of the National Institutes of Health. All participants provided informed consent, and data used for this analysis were deidentified (data release 9.1).

As detailed in Supplementary Figure 2, ineligible participants excluded in a stepwise manner met the following criteria: withdrew from the study, had no data for the Brief Betrayal Trauma Survey (BBTS; administered during follow-up, 2008–2012), had type 1 diabetes, had type 2 diabetes prior to baseline, were missing or had unknown timing for type 2 diabetes, self-identified race/ethnicity other than NHW, BAA, or Latina (due to small sample size), or had missing data for race/ethnicity. The final analytic sample comprised 42,172 participants. Compared with eligible participants, ineligible Sister Study participants were more likely to belong to minoritized racial/ethnic groups, to have a lower socioeconomic status (SES), to have worse health behaviors and clinical characteristics, and, likely related to type 2 diabetes exclusion criteria, were more likely to have a metabolic profile consistent with MetS at baseline (Supplementary Table 1).

Participants completed a self-administered modified version of the BBTS (24) that assessed TCEs at ages ≤12 and 13 to <18 years. We combined participants’ reports to assess whether they ever experienced traumatic events at age <18 years (yes or no), as detailed in Supplementary Table 2. The eight major traumatic-event categories included: 1) any trauma (a summary measure capturing any of the following seven trauma categories or any other trauma type that was not listed); 2) psychological/emotional trauma (e.g., mistreatment); 3) household dysfunction (e.g., family substance use); 4) sexual trauma; 5) physical trauma (e.g., hit or attacked); 6) natural disaster; 7) major accident; and 8) major illness. Categories were 1) no childhood trauma (reference); 2) no for the trauma category but yes for another trauma category (beyond the study’s focus and not presented); and 3) yes for the specific trauma category.

Participants reported diabetes status over follow-up by providing a yes or no response to the following question: “Has a doctor or other health care provider ever told you that you had diabetes?” If applicable, participants also provided the diagnosis date. Given the high positive and negative predictive values of self-reported type 2 diabetes diagnoses found in the literature, participants were assigned type 2 diabetes status (yes, no) on the basis of self-report (25).

We identified potential confounders through prior literature and construction of directed acyclic graphs (DAGs), tools used to identify adjustment sets necessary to estimate causal relationships between exposures and outcomes (Supplementary Figure 3). The following potential confounders were self-reported at baseline and are listed in Table 1: age (in years), race/ethnicity, childhood region of residence, and indicators of low childhood SES (namely, food insecurity, low educational attainment in household, and low income). Although early-life low SES may be considered a TCE, our focus was on TCEs derived from the BBTS.

Participants who self-identified as White alone and of non-Hispanic ethnicity were considered NHW, those who identified as non-Hispanic Black or African American were considered BAA, and participants who identified as any race and of Hispanic ethnicity were considered Latina.

We ascertained a metabolic profile consistent with MetS at baseline and considered it a potential modifier because participants with versus without prevalent MetS are at a higher risk of developing type 2 diabetes (26). Described in detail elsewhere and consistent with the International Diabetes Federation harmonized definition (27, 28), the presence of at least three metabolic abnormalities, which included abdominal obesity, prehypertension or hypertension, dyslipidemia, and glucose intolerance, defined MetS (yes, no). Last, low childhood social support was ascertained for sensitivity analyses (described in Table 1).

We also considered prevalent MetS as a potential mediator because TCEs are likely linked to both MetS and type 2 diabetes through largely the same biological pathways. Adulthood sociodemographic characteristics and health behaviors were potential mediators, based on prior literature and our DAG construction. As described in Table 1, sociodemographic characteristics in adulthood included region where participant lived longest in adulthood, educational attainment of participant, and, if applicable, participant’s partner. Health behaviors included smoking status, alcohol consumption in the past 12 months, physical activity, diet, and sleep health (27). Other adulthood characteristics that were considered but were not identified as necessary adjustment factors after DAG construction are listed in Table 1.

Thirty-two percent of participants were missing data, mostly due to missing self-reported diagnosis of depression or bipolar depression (11%), at least one TCE (11%), or a component of MetS (13%). There was 0.1–5% missingness for other variables (Supplementary Table 1), and we used a doubly robust analytic approach to address missing data (29). Using standard methods, we examined missingness and missing data patterns prior to imputing 32 data sets using multiple imputation by chained equations. To account for whether a participant was missing data and had imputed values, we estimated and applied inverse probability weights for missingness to all models (29).

As a data reduction method, and detailed in Supplementary Table 3, we used Mplus, version 8.4 (Muthén & Muthén) to perform a latent-class analysis to identify and group participants with similar patterns of co-occurring TCEs. Briefly, we used recommended approaches to sequentially estimate two-class to six-class models, identify the best models using model fit criteria, and assign participants to latent classes on the basis of their highest posterior probability of latent class membership.

We used Cox proportional hazards regression models with age as the time scale to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for incident type 2 diabetes from baseline to whichever came first: age at type 2 diabetes diagnosis, censoring (e.g., loss to follow-up), or the end of follow-up (30 September 2019). We then performed Wald tests on trauma category by race/ethnicity cross-product interaction terms and stratified models by race/ethnicity. Using modified Poisson regression with robust variance, we estimated prevalence ratios (PRs) and 95% CIs for MetS, overall and by race/ethnicity. Models for associations between TCEs and type 2 diabetes were then stratified by MetS after Wald tests for two-way interaction. We then applied a weighted inverse odds ratio approach to test for mediation of the any trauma and type 2 diabetes association by MetS (30). We considered only the any-trauma measure, because TCEs likely share mechanistic pathways through physiological and psychosocial stress that affect MetS development to contribute to type 2 diabetes risk. Details of the approach are provided in the Supplemental Text. All models were adjusted for childhood region of residence and childhood SES. Models in the total population were additionally adjusted for race/ethnicity. Using Wald tests of TCEs by time cross-product interaction terms, we found no violations of the proportionality assumption.

We conducted four sensitivity analyses (detailed in the Supplemental Text): multiple comparison correction; stratification by a potential buffer, childhood social support; adjustment for potential adulthood mediators; and estimation of HRs with follow-up time beginning at the time of TCE reporting. Analyses were conducted using SAS, version 9.4 (SAS Institute, Cary, NC), and we used the computational resources of the National Institutes of Health High-Performance Computing Biowulf cluster (https://hpc.nih.gov). A two-sided P value of 0.05 determined statistical significance.

The data sets analyzed during the present study are not publicly available, because of privacy concerns. However, requests for data may be made following procedures described on the Sister Study website (www.sisterstudy.niehs.nih.gov).

Among 42,173 participants who contributed 470,226 person-years of follow-up (mean ± SD, 11.1 ± 2.7 years), there were 3,158 incident cases of type 2 diabetes (n = 2,479 among NHW, 461 among BAA, and 218 among Latina participants; Table 1). At baseline data collection, mean age ± SE was 55 ± 0.04 years; 88% of participants self-identified as NHW, 7% as BAA, and 4% as Latina. During childhood, the largest proportion of NHW women resided in the Midwest (38%), 55% of BAA women resided in the South, and 57% of Latina women resided in Puerto Rico or outside the United States. MetS prevalence was approximately 10% (Table 1, Supplementary Table 4). Compared with NHW participants, BAA and Latina participants had higher prevalence of low SES indicators in both childhood and adulthood, reported poorer health behaviors, and had higher prevalence of MetS. Prevalence of any TCEs was 53% among BAA, 51% among Latina, and 50% among NHW participants; was higher among type 2 diabetes cases versus the full cohort (55% vs. 50%); and was higher among participants with versus those without prevalent MetS (54% vs. 50%; Tables 2–4, Supplementary Table 5). In the latent class analysis, a two-class model was the best fit (Supplementary Table 3 and Supplementary Figure 4). TCE categories included high trauma (i.e., psychological/emotional mistreatment and moderate probabilities of family issues as well as sexual and physical trauma; 13% of participants) and low trauma (i.e., low probability of all TCEs; 87% of participants).

Overall, experiencing any TCE was associated with a higher incidence of type 2 diabetes, and associations were strongest among Latina participants (HR 1.64 [95% CI 1.21–2.22] vs. HR_NHW 1.09 [95% CI 0.99–1.20] and HR_BAA 1.09 [95% CI 0.90–1.33]; P for interaction <0.05; Table 2). Psychological/emotional trauma was associated with higher risk of type 2 diabetes across all races/ethnicities (HR 1.16; 95% CI 1.06–1.27) with some suggestion of stronger associations among Latina participants (P for interaction <0.10). Although CIs overlapped with estimates among NHW and BAA participants, Latina participants had the highest HR associated with household dysfunction and sexual trauma, which were associated with 19% higher risk overall (HR_{sexual  trauma} 1.19; 95% CI 1.06–1.33]). Physical trauma was associated with higher type 2 diabetes incidence and varied by race/ethnicity (HR_NHW 1.46 [95% CI 1.22–1.74]; HR_BAA 0.82 [95% CI 0.54–1.24]); and HR_Latina 1.61 [95% CI 0.93–2.79]; P for interaction <0.05). Natural disaster during childhood was associated with a 24% higher incidence of type 2 diabetes overall (HR 1.24; 95% CI 1.02–1.51), and despite overlapping CIs with NHW and BAA participants, HR estimates were highest among Latina participants. Higher incidence associated with major illness was suggestively strongest among BAA participants (HR 1.91; 95% CI 1.10–3.31; P for interaction <0.10). Overall, results for high trauma (HR 1.13; 95% CI 1.01–1.27) were similar to the higher type 2 diabetes risk associated with any trauma.

Compared with participants without any reported childhood trauma, participants with at least one childhood trauma had an 11% higher prevalence of MetS (PR 1.11; 95% CI 1.04–1.19) at baseline (Table 3). All TCEs, including high versus low TCEs, were either marginally associated or associated with a higher prevalence of MetS during adulthood and did not statistically significantly vary by race/ethnicity. MetS did not modify associations between TCEs with type 2 diabetes (Table 4).

MetS acted as a mediator (indirect effect HR 1.01 [95% CI 1.00–1.01], P = 0.02; total effect HR 1.13 [95% CI 1.04–1.22], P = 0.01; direct effect HR 1.11 [95% CI 1.02–1.20], P = 0.02) (Supplementary Table 6) and explained 14% of the association between reporting any TCE and type 2 diabetes (31).

TCEs that remained statistically significantly associated with both higher type 2 diabetes risk and MetS prevalence after multiple comparison correction included any, psychological/emotional, sexual, and physical trauma, as well as major illness (Supplementary Text and Tables 7 and 8). Childhood social support did not modify associations (Supplementary Table 9). Attenuated associations with wider CIs were consistent with the main analysis after adjustment for adulthood characteristics and after beginning follow-up at TCE reporting (Supplementary Tables 10–12). Estimates did not significantly differ by race/ethnicity after beginning follow-up at TCE reporting. However, the any trauma-type 2 diabetes associations remained suggestively strongest among Latina women in all sensitivity analyses.

Among a racially and ethnically diverse cohort of middle-aged to older-aged women, we found that early-life trauma was associated with a higher risk of incident type 2 diabetes. BAA and Latina women had a higher burden of TCEs, MetS, and type 2 diabetes compared with NHW counterparts. Moreover, associations of reporting any TCEs with type 2 diabetes were strongest among Latina women. TCEs were also associated with higher prevalence of MetS, which did not modify, but partially mediated, associations after adjustment. This study expanded upon prior studies by, to our knowledge, being the first to longitudinally investigate understudied but increasingly relevant TCEs beyond childhood abuse and by demonstrating novel associations (i.e., natural disaster, major illness) with both MetS and incident type 2 diabetes among U.S. women that varied by race/ethnicity. Importantly, our study suggests that Latina women may be an understudied population at particularly higher risk of TCE-associated type 2 diabetes incidence.

Our observations of associations between experiencing childhood trauma, including psychological emotional, sexual, and physical trauma, and higher type 2 diabetes incidence were consistent with most of the prior literature (2,11,12). Exceptions include one cross-sectional study of young, postpartum women in Rhode Island and a cross-sectional study of adults with low income in Oregon that each found no associations (10,13). Associations between childhood trauma and a higher prevalence of MetS observed in our study are consistent with findings of most of the few U.S. studies conducted; however, studies of special populations and a small (n = 342) longitudinal study of Black and White U.S. women found no association between either emotional or sexual abuse with MetS (14–16,18,20,32–36). Our results may differ because of the large sample size, prospective study design, inclusion of understudied TCEs beyond abuse (e.g., natural disaster), and subpopulation of Latina women who were not included in the prior, most comparable U.S. study (32).

Our observation of MetS as a mediator is also consistent with prior literature, particularly a study that suggested a pathway from stressful childhood experiences to obesity and, thereby, increased risk of type 2 diabetes (37). Furthermore, TCEs may, for instance, alter insulin sensitivity through activating stress pathways that contribute to inflammation (3) and may affect development of the amygdala, which is involved in emotional regulation, thus potentially affecting coping and fostering unhealthy behaviors that have been linked to both MetS and type 2 diabetes, such as diet, physical inactivity, and poor sleep (3–5,22,27). Biological impacts of poor health behaviors include components of MetS, such as elevated circulating levels of triglycerides, free fatty acids, glucose, and insulin (2). These changes, in part or in combination, and with other pathways including epigenetic alterations (3), may result in type 2 diabetes development.

Consistent with our hypothesis, relationships between any TCE and higher incidence of type 2 diabetes were stronger among Latina participants than among NHW women. Latina participants largely resided in Puerto Rico and were likely of Puerto Rican descent. Puerto Rican adults are U.S. citizens, and lower risk of poor health outcomes known as the “Hispanic paradox” found in Latino populations with more representation of immigrants may not occur among this group. Furthermore, although we adjusted for region of residence, unmeasured differences in access to health-promoting resources in the U.S. territory, compared with the continental United States, may partially explain differences. Relatedly, modifiable factors (e.g., lower SES, poor health behaviors and metabolic profiles among Latina compared with NHW women) may have contributed to disparities. More study to guide potential tailored interventions is warranted.

After adjustment for adulthood characteristics that likely act as mediators, associations with type 2 diabetes, although attenuated, remained suggestively strongest among Latina participants. The attenuated results may be related to both imprecision due to saturating models with a small number of cases and the adjustment for factors that are likely along the causal pathway. Imprecision also increased and sample size decreased after beginning follow-up at the time of TCE reporting, which resulted in reduced power to identify racial/ethnic differences; however, point estimates remained higher among Latina participants than NHW women, supporting the robustness of our findings.

Counter to our hypothesis, associations between TCEs with both MetS and type 2 diabetes were rarely stronger among BAA women compared with NHW women. Nonetheless, BAA women had the highest prevalence of both TCEs and MetS, as well as type 2 diabetes incidence. Although numerous explanations are conceivable, higher cumulative adversity among historically minoritized groups may promote more resiliency and psychobiological adaptive responses, thus contributing to weaker relative associations between individual adversities and poor health manifestations (38). This possibility may have influenced our observation that physical trauma was associated with higher risk of type 2 diabetes among NHW and Latina women, whereas there was no association among BAA women. More investigations are warranted to explore this possibility.

Pertaining to study limitations, unknown impacts of unobservable, unmeasurable variables are possible due to the observational nature of this study. Despite our large sample size, the few cases, upon stratification when investigating modifiers, limited our power to detect some associations. Some relevant data were not available (e.g., additional TCEs; duration, frequency, and intensity of trauma; posttraumatic stress), participants missing TCE data were excluded, and most data were self-reported. Recall bias and social desirability bias, although reduced by use of self-administered questionnaires, may affect results. However, studies using test–retest reliability have shown consistent reporting of certain TCEs (e.g., household substance abuse) over time (39). Furthermore, TCE status reporting is likely nondifferentially misclassified by MetS and type 2 diabetes status, which would result in underestimation of associations. Self-reported diagnosis of diabetes is a reliable measure (25) but may have led to an underestimation of diabetes cases that varies by race/ethnicity, due to differential access to health care. However, women of all races and ethnicities in this study had higher SES, on average, than the general U.S. population, reducing the likelihood of racial/ethnic differences in undiagnosed diabetes. Nonetheless, study population and design characteristics may have limited generalizability of our results. First, follow-up began when women were middle-aged and older; therefore, our sample excludes women who developed type 2 diabetes earlier in life, particularly those who experienced TCEs, if our observed associations are accurate. Second, although Sister Study participants were without breast cancer and diabetes when the study began, they had a sister with a prior diagnosis of breast cancer. Both breast cancer and type 2 diabetes have shared risk factors; therefore, unknown bias may be present. Last, our study of mostly NHW, higher-SES women in which the Latina population largely resided in Puerto Rico limited generalizability. However, we hypothesize that biological mechanisms would be consistent across all populations of women. To overcome our study limitations, future prospective studies over the life course are warranted, with robust repeated measures among large, general population samples that are inclusive of sufficient representation of all minoritized racial/ethnic groups.

Despite the limitations, study strengths include our use of a large, racially and ethnically diverse cohort of women from across the United States, as well as the investigation of understudied but increasingly relevant TCEs (e.g., natural disaster). We also used a longitudinal study design with over 10 years of follow-up and addressed an important literature gap by stratifying by race/ethnicity. We additionally used widely accepted methods to address missing data and avoid potential bias related to complete case analyses. Last, our investigation of mediation elucidated a likely pathway by which childhood trauma may contribute to type 2 diabetes risk.

Our findings offer several implications. A novel finding, natural disasters during childhood were associated with higher type 2 diabetes incidence among U.S. women. As climate change continues, children will be at increasing risk of experiencing disasters. Policy makers, public health professionals, and clinicians should consider TCEs while designing, implementing, and evaluating disaster preparedness and response, and consider environmental vulnerability, because data suggested possibly stronger associations among Latina women who largely resided in Puerto Rico during childhood. Clinicians may also need to provide trauma-informed care to aid in type 2 diabetes prevention efforts. Racially or ethnically minoritized women had the highest burdens of TCEs, MetS, and type 2 diabetes, and thus may particularly benefit. Replication of our study and further study of biological mechanisms (e.g., epigenetic pathways) are warranted to further elucidate mediation pathways and develop additional prevention strategies. For instance, major illness during childhood was associated with higher type 2 diabetes risk in adulthood. Emerging research links certain cancer treatments to metabolism disruption, inflammation, and subsequent diabetes risk; future research of specific childhood illnesses along with potential mechanisms is warranted. Last, continuing the promotion of healthy coping behaviors along with environments that foster such behaviors, even in the presence of adversity, is warranted.

TCEs were associated with higher risk of developing type 2 diabetes among a racially and ethnically diverse cohort of middle-aged to older-aged women in the United States. Furthermore, racially/ethnically minoritized women had the highest burdens of TCEs, MetS, and type 2 diabetes, and TCEs-associated type 2 diabetes risk was often higher among Latina participants than among NHW women. Although additional studies are warranted, tailored primary prevention and intervention efforts of both protecting youth from trauma and inhibiting likely mediators (i.e., metabolic abnormalities) may help ease the burden of type 2 diabetes among women.

Acknowledgments. In this work, the computational resources of the National Institutes of Health High Performance Computing Biowulf cluster (https://hpc.nih.gov) were used. The authors thank Nat MacNell of DLH, Inc., for technical assistance during the development of this article. The authors also acknowledge Deborah Bookwalter and Cathy Sampson of Westat for assistance with data curation and preparation, Stacey Mantooth and Erin Knight of the National Institute of Environmental Health Sciences Library Staff for assistance with the literature search, and DLH, Inc., staff for quality control of the data analysis. Last, the authors thank the Sister Study participants for their participation.

Funding. This work was funded by the Intramural Program at the National Institutes of Health, National Institute of Environmental Health Sciences (grants Z1AES103325 to C.L.J. and Z01ES044005 to D.P.S.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. S.A.G. contributed to study conceptualization, designed and conducted the analysis, drafted the manuscript, and revised the manuscript. N.M.R. drafted parts the manuscript and reviewed and edited the manuscript. C.G.P. reviewed and edited the manuscript. J.M.P.W. conducted data analysis and reviewed and edited the manuscript. D.P.S. provided resources and funding and reviewed and edited the manuscript. C.L.J. conceptualized the study, provided resources, funding, and supervision, and reviewed and edited the study design and manuscript. S.A.G. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.