Cumulative childhood adversity and adult cardiometabolic disease: A meta-analysis.

Method

A literature search was conducted in the PubMed database for articles appearing through September 2017 using the following search terms: (“adverse childhood experiences” or “cumulative adversity” or “childhood adversity” or “childhood maltreatment” or “childhood trauma” or “risky families” or “early life adversity” or “stressful life events” or “psychosocial stress”) AND (“cardiovascular” or “hypertension” or “diabetes” or “CHD” or “MI” or “metabolic syndrome” or “stroke”). Reference lists of relevant articles were used to identify additional studies. The search returned 3,802 possible abstracts (see Figure 1 for the number of articles identified, screened, and excluded). Abstracts were reviewed to assess for the presence of a measure of cumulative adversity (i.e., an index that included at least two adverse childhood experiences) and at least one measure of cardiometabolic disease obtained at age 18 or beyond in adult populations without severe psychiatric illness (i.e., schizophrenia and psychosis).

Cardiometabolic disease included the following: 1) CVD clinical outcomes: clinically-diagnosed hypertension (but not continuous measures of blood pressure, as clinical cutoffs could not be applied without obtaining and grouping individual-level data from the parent study), coronary heart disease (CHD), ischemic heart disease (IHD), myocardial infarction (MI), stroke or cerebrovascular disease, or combinations of the aforementioned endpoints; and 2) metabolic outcomes: type 2 diabetes and metabolic syndrome. Measurement of outcomes typically involved diagnosis by a nurse, physician, or other healthcare provider (as indexed in national registers, hospital registers, or medical records); assessment by trained research staff; self-report of a previous diagnosis made by a physician or other healthcare provider; or a composite of items assessed by physician, research staff, and/or self-report.

The present meta-analysis does not include pre-clinical CVD risk factors per se, such as weight-related outcomes (e.g., body mass index, waist circumference, obesity), sub-clinical indicators of CVD (e.g., carotid intima-media thickness), or inflammation, as these outcomes have previously been reviewed in association with early life adversity in adults (Appleton et al., 2017; Basu, McLaughlin, Misra, & Koenen, 2017; Baumeister et al., 2015; Danese & Tan, 2014; Norman et al., 2012; Su, Jimenez, Roberts, & Loucks, 2015), and are predictors of later CVD, rather than CVD, which is the focus of this study.

Our search resulted in 93 articles that were assessed in detail for eligibility. Of these full-text articles, 29 were excluded because they did not have a cumulative measure of adversity; 6 involved ineligible populations (e.g., adults with severe psychiatric illness; children); 16 studies did not have an appropriate measure of cardiometabolic disease; and 5 studies did not report relevant effects. Thus, 37 studies met the stated inclusion criteria. Surprisingly, there were no articles evaluating CVD mortality.

The first author (KJ) performed the data searches and coding in consultation with KM. The following information was extracted from each study: 1) sample size; 2) sample demographics (age, sex [i.e., 80% or more of one sex], and geographic region [U.S. vs. non-U.S. samples]); 3) study design (prospective vs. retrospective report of adversity); 4) type of cardiometabolic disease (i.e., CVD clinical outcomes vs. metabolic outcomes); 5) type of cumulative adversity measure and whether it included childhood SES, household dysfunction (defined as domestic violence, substance use, mental illness, and incarceration), and abuse (defined as physical, sexual, or emotional/psychological/verbal abuse or neglect); 6) measurement of cardiometabolic disease (independently-assessed disease [i.e., outcomes assessed in a laboratory and/or medical setting or diagnosis made by a healthcare provider and indexed in national registers, hospital registers, or medical records]) vs. self-report of a previous diagnosis made by a physician or other healthcare provider); 7) analytic strategy (i.e., comparing individuals who reported the greatest number or severity of adversities to a reference group of those who reported the fewest or least severe adversities or none at all vs. using a continuous total/average score on an adversity-related scale); and 8) total number and type of covariates included in analytic models, including adult SES, adult psychosocial risk factors for CVD (e.g., depression or anxiety symptoms, marital status, social support), and adult health behaviors (e.g., BMI, smoking, alcohol use, diet, physical activity, sleep); these variables were counted as covariates if they were tested as mediators or part of the pathway between adversity and cardiometabolic disease.

In order to calculate reliability statistics, a random sub-sample of eight studies (n=3 HR studies; n=5 OR studies), or 13% of all HR and OR studies, were double-coded for reliability by KJ and KM. Overall reliability across extracted study data was 95%, with individual category reliabilities ranging from 75% to 100%.

The reported effect size was extracted for each health outcome examined. Effect sizes were reported as regression weights, odds ratios (OR), relative risks (RR), or hazard ratios (HR), and were drawn from fully adjusted models. Effects involving regression weights or RR were pooled with OR effects. An OR reflects the ratio of the odds of an outcome in exposed persons (i.e., those exposed to cumulative childhood adversity) relative to the odds in non-exposed persons, thus, it can be interpreted as the risk of developing a disease given a certain exposure. In contrast, a HR reflects time to event for a particular outcome, or a survival analysis. Thus, one might hypothesize that persons exposed to greater cumulative adversity would demonstrate greater risk of cardiometabolic disease (as indicated by OR analyses) and earlier/faster disease onset (as indicated by HR analyses). Given differences in the interpretation of OR versus HR and the inability to harmonize these effect sizes statistically, these effects were pooled separately, as has been the case in other meta-analyses (Bhattacharjee, Bhattacharya, Kelley, & Sambamoorthi, 2013; Kronish et al., 2011). The direction of effects was consistently coded such that values indicate a relationship between increased exposure to cumulative adversity and greater risk of cardiometabolic disease or faster time to disease event.

It was common for studies to analyze multiple cumulative adversity-health relationships (i.e., multiple health outcomes were tested) or the same adversity-health relationship within multiple subgroups (men vs. women) so each of these effects was coded. Non-overlapping groups were treated as independent effects. If multiple comparisons were tested within the same group for the same outcome (e.g., 0 adversities vs. 1–2 adversities vs. 3–4 adversities), the most extreme comparison was retained. The majority of studies used an extreme groups approach with a reference group of 0 adversities (HR studies: K = 7; OR studies: K = 17), although three studies used a reference group other than 0 adversities (Halonen et al., 2015; Rich-Edwards et al., 2010; Wilson et al., 2012). All remaining studies (HR studies: K = 1; OR studies: K = 10) did not use an extreme groups approach, and instead analyzed continuous associations between a total/average score on a trauma/adversity-related scale with no reference group and cardiometabolic outcomes.

Overall, 37 studies were included: 9 HR studies (15 effects; total N = 203,017) and 29 OR studies (62 effects; total N = 555,532). One study (McCrory, Dooley, Layte, & Kenny, 2015) was included in both HR and OR pooled effects because it repeated both types of analyses in the health outcomes tested. A description of samples, measures, and analytic strategies used in HR and OR studies are listed in Supplemental Tables 1 and 2, respectively, by author and year of publication.

Treating multiple effects from the same study sample as independent can artificially reduce the SE, making it more likely that results will be accompanied by a lower p value. However, aggregating across effects reduces power, could artificially deflate estimates, and makes it more difficult to adequately test whether the strength of the association may vary depending on the type of outcome assessed (Borenstein, Hedges, Higgins, & Rothstein, 2009). Thus, meta-analytic results are presented in two ways: (a) using a more conservative approach, which aggregated effects within studies across dependent measures (HR studies: K = 9; OR studies: K = 29), and (b) using a less conservative approach which treated each cumulative adversity-cardiometabolic disease effect as independent (HR studies: k = 15; OR studies: k = 62). By necessity, moderators were tested using the less conservative approach. To be thorough, we also performed analyses assuming a correlation of .4 among outcomes within the same study, for HR and OR studies, respectively. This correlation was chosen because it represents an analytic strategy in between the extreme strategies presented above (treating outcomes as independent is akin to correlating outcomes within studies at 0, and treating them as redundant is akin to correlating outcomes within studies at 1).

Analyses were conducted with Comprehensive Meta-Analysis analytic software version 3 (BiostatTM, USA) to employ random effects modeling of associations between adversity and cardiometabolic disease. Random effects models assume that samples are drawn from populations with different effect sizes, thus allowing for both random variance and variance due to true differences between populations. Accordingly, the sample size in a study contributes less weight in random effects models compared to fixed effects models (Borenstein, Hedges, & Rothstein, 2007). Furthermore, random effects models yield more accurate confidence intervals in meta-analysis (Schmidt, Oh, & Hayes, 2009). The additional analyses assuming a correlation of .4 were also conducted in the CMA program; however, within study effects were first aggregated in excel using a template provided by the CMA software’s creator, who is an expert in meta-analysis (M. Borenstein).

The heterogeneity statistic (QT) provides an estimate of variability of effect sizes. If there was significant heterogeneity of effects (i.e., significant QT coefficient), then moderators were examined as a potential explanation for variability of effect size (see “Data Extraction” section for specific moderators). Because of different numbers of total effects in HR (k=15) and OR (k=62) analyses, arbitrary cutoffs of “3” and “12,” respectively, were chosen for the minimum number of effects needed in each category to test moderators. This corresponded to cutoffs of at least 20% and 19% of available effects for HR and OR analyses, respectively. These cutoffs were not applied to moderation by sex analyses, which involved fewer total HR (k = 5) and OR (k = 25) effects.

Finally, because null and negative findings are less likely to be submitted for publication by authors and/or less likely to be accepted for publication, there may be biased effects in the available literature. Proposed guidelines (Rothstein, Sutton, & Borenstein, 2006) were used to examine the presence and impact of publication bias in both HR and OR pooled effects, including examination of (1) forest plots for individual effects that appear to be outliers; (2) funnel plots of analyzed HR and OR effects (Light & Pillemer, 1984); (3) Kendall’s tau, which provides a statistic of the relationship between standard errors and standardized effect sizes (Begg & Mazumdar, 1994); and (4) Egger’s regression, where the intercept reflects the slope of the association (Egger, Smith, Schneider, & Minder, 1997).

Results

Table 1 provides a summary of the frequencies of demographic, moderator, and outcome variables for both HR and OR pooled effects. The analysis for HR effects were based on 203,017 participants across all effects (i.e., counted for each outcome and males and females separately for sex-stratified analyses) and 179,612 participants across studies, whereas the analysis for OR effects were based on 555,532 across effects and 247,393 participants across studies. Overall, HR and OR effects were based on predominantly female samples and retrospectively-reported adversity data. The majority of HR studies were conducted in European samples, while the majority of OR studies were conducted in U.S. samples, with non-U.S. samples spread amongst several European countries, Saudi Arabia, New Zealand, Canada, and the Philippines. The Philippines (Ramiro, Madrid, & Brown, 2015) is the only country that is grouped with low and middle-income countries (LMIC) for the current 2018 fiscal year by the World Bank Atlas method (World Bank Group, 2017). The majority of both HR and OR effects reflected CVD clinical outcomes, compared to metabolic outcomes. Roughly half of HR effects reflected independently-assessed disease, compared to 36% of OR effects. The majority of both HR and OR effects utilized an analytic strategy in which they compared extreme categories of adversities, as opposed to studies that used a total count or average. Five OR studies utilized prospectively-collected data on cumulative adversity, primarily reported by parents and/or teachers, and longitudinal associations with cardiometabolic disease; follow-up length for these studies ranged from 27 to 38 years (M = 31.8 years, SD = 4.02).

Results for HR studies are presented in Table 2. A forest plot examining all effects individually is presented in Figure 2 and a forest plot of the individual effects aggregated within study is pictured in Supplemental Figure 1. Results treating all effects as independent (k = 15) and results aggregating effects within studies (K = 9) both revealed that exposure to cumulative adversity in childhood was associated with decreased time to event for all outcomes combined. The cumulative effect size was similar in magnitude across these two analytic strategies (Table 2, first two rows). Results assuming a correlation of .4 among outcomes tested in the same sample (HR=1.35, 95% CI [1.14, 1.60]) were very similar to the estimated effect sizes from analyses treating all outcomes as independent or redundant (Table 2, first two rows), likely due to the number of large samples in our analyses.

Examining the forest plot raised the possibility that the Bellis et al. (2015) study may be an outlier, given relatively large effects across multiple outcomes. Thus, analyses were re-run excluding Bellis et al. (2015). Results indicated a reduced, but still significant, effect size whether treating all effects as independent or aggregated within study (see Figure 2 and Supplemental Figure 1 captions).

Results for OR studies are presented in Table 3. A forest plot examining all effects individually is presented in Figure 3 and a forest plot of the individual effects aggregated within study is pictured in Supplemental Figure 2. Results treating all effects as independent (k = 62) and results aggregating effects within studies (k = 29) both revealed that exposure to cumulative adversity in childhood was associated with increased risk for all outcomes combined, with almost identical cumulative effect sizes (Table 3, first two rows). Results assuming a correlation of .4 among outcomes tested in the same sample (OR=1.34, 95% CI [1.24, 1.45]) were very similar to the estimated effect sizes from analyses treating all outcomes as independent or redundant (Table 3, first two rows), likely due to the number of large samples in our analyses.

Results of moderator analyses for HR studies are presented in Table 2 and revealed significant between-group differences for 5 of the 10 moderators tested: larger effects were observed for the ACE questionnaire compared to other adversity questionnaires or ad-hoc composites; for studies that did not include an indicator(s) of child SES in the composite index of cumulative adversity compared to those that did; for studies that did not adjust for adult psychosocial factors or health behaviors, respectively, in analytic models, compared to models that did; and for studies that had 5 or fewer covariates included in analytic models, compared to 6 or more covariates.

Results of moderator analyses for OR studies are presented in Table 3 and revealed significant between group differences for 7 of the 12 moderators tested: effects were larger for non-U.S. vs. U.S. samples; for CVD clinical outcomes compared to metabolic outcomes; for self-reported compared to independently-assessed disease; for studies that used an analytic strategy comparing individuals who reported the most adversities to a reference group of those who reported the fewest adversities (usually categorized as 0 adversities) vs. using a continuous total/average score on a trauma/adversity-related scale with no reference group; for studies that did not adjust for adult health behaviors in analytic models, compared to models that did; for studies that did adjust for adult SES in analytic models, compared to those that did not; and for studies that had 5 or fewer covariates, compared to 6 or more covariates.

Results suggested no evidence of moderation by sex in HR or OR effects, but there were relatively few studies that had at least 80% males or females. Given limited variability in effects (see Table 1), for HR effects it was not possible to test moderation by adult SES, study design, analytic approach, and U.S. vs. non-U.S. samples, while for OR effects it was not possible test study design and whether the cumulative index included abuse/neglect.

Evaluation of the funnel plot for HR effects (Supplemental Figure 3) indicated evidence of bias, such that smaller studies (i.e., those closer to the bottom of the plot) showed larger effects (because effects would need to be larger in order to reach statistical significance) and effects were less symmetrically distributed about the mean. Additionally, the funnel plot for OR effects (Supplemental Figure 4) suggests there may be some bias. Notably, results for Kendall’s Tau did not suggest significant publication bias in HR effects (Tau = .24, p = .23) or in OR effects (Tau = .05, p = .53), however, results for Egger’s regression did suggest bias in both HR (Intercept = 2.14, SE = .67, p<.05) and OR effects (Intercept = 1.62, SE = .36, p<.05).

Given evidence that results are influenced by publication bias, Duval and Tweedie’s (2000) trim-and-fill procedure was used to provide a bias-corrected estimate of the cumulative effect size; see Supplemental Figures 3 and 4. Overall, for both HR and OR studies, results suggest there may be bias in publication, depending on the approach used. However, statistically correcting for this bias did not nullify associations found here, but reduced the estimated effect size in OR analyses to 1.31, 95% CI [1.22, 1.40].

Discussion

The present review provides a quantitative analysis of the relationship of cumulative childhood adversity with cardiometabolic disease in adulthood. Based on 179,612 participants in 9 HR studies and on 247,393 participants in 29 OR studies that met eligibility criteria, results show overall effect sizes of 1.38 for HR studies and 1.36 for OR studies (Tables 2 and 3). The estimate for OR studies was reduced to 1.31 after correcting for publication bias. These results are based on retrospectively-reported adversity for the majority of HR (100%) and OR (90%) studies. Overall, results are surprisingly robust considering that most of the studies adjusted for multiple covariates (some of which are likely mediators of the association of interest) as well as the diversity of samples, study designs, and outcomes. Furthermore, the effect sizes are comparable to the aggregated effects of other psychosocial risk factors on cardiovascular morbidity and mortality. For example, in healthy populations (i.e., individuals without pre-existing CVD), meta-analytic results demonstrate a significant effect of anger and hostility (HR = 1.19, 95% CI [1.05, 1.35]; Chida & Steptoe, 2009), depression (RR = 1.81 [1.53, 2.15]; Nicholson, Kuper, & Hemingway, 2006), and anxiety (HR = 1.26 [1.15, 1.38]; Roest, Martens, de Jonge, & Denollet, 2010) on CVD events, as well as an effect of depression on risk of stroke (RR = 1.34 [1.17, 1.54]; J. Y. Dong, Zhang, Tong, & Qin, 2012) and anxiety on cardiac death (HR = 1.48 [1.14, 1.92]; Roest et al., 2010). Additionally, similar to effect sizes found here, recent meta-analytic results suggest a moderate combined estimated effect of loneliness, social isolation, and living alone on risk of incident CVD (RR = 1.29 [1.04, 1.59]) and stroke (RR = 1.32 [1.04, 1.68]; Valtorta, Kanaan, Gilbody, Ronzi, & Hanratty, 2016), as well as mortality (OR = 1.30 [1.16, 1.46]; Holt-Lunstad, Smith, Baker, Harris, & Stephenson, 2015).

The present meta-analysis also examined moderators. Of particular interest was whether childhood adversity was more closely associated with CVD clinical outcomes or metabolic outcomes. Findings suggest that cumulative adversity may be more closely associated with CVD clinical outcomes than metabolic outcomes (based on OR effects), although associations were significant for both types of outcome. Notably, there was no evidence of moderation by type of outcome for HR effects.

This weaker effect for metabolic results is consistent with the small cumulative association between abuse and neglect and diabetes reported in Huang et al. (2015). It is also consistent with the results of five studies (k = 6 effects) in the present review that involved “exemplary” measurement of both predictor and outcome variables (i.e., prospectively-reported childhood adversity and objective metabolic outcomes). Null relationships with metabolic disorders were found in four of these five studies (Danese et al., 2009; Delpierre et al., 2016; Gustafsson & Hammarstrom, 2012; Melchior, Moffitt, Milne, Poulton, & Caspi, 2007). Further, although not included in this meta-analysis, several recent studies using exemplary measurement have reported positive associations between prospective measures of childhood adversity and subclinicial CVD in adulthood (Hakulinen et al., 2016; Juonala et al., 2016), providing strong support for increased risk for later clinical events among those who experienced childhood adversity.

Regarding vulnerable or resilient subgroups, the meta-analysis found no evidence that cumulative childhood adversity has a more negative effect on men or women, although this is based on a limited number of HR and OR effects that provided sex-specific estimates. In contrast, previous meta-analytic evidence suggests stronger effects of abuse and neglect in females relative to males for obesity (Danese & Tan, 2014) and CVD problems (Wegman & Stetler, 2009). Ultimately, it is important that future studies examine sex-specific effects and not simply statistically adjust for sex, as males and females may be vulnerable to different types of adversity, such as child maltreatment versus household dysfunction. Importantly, it was not possible to test for moderation by race in HR and OR analyses because most studies (both prospective and retrospective) are conducted in predominantly non-Hispanic White samples. Consequently, little is known about the impact of cumulative childhood adversity on cardiometabolic disease in ethnically and racially diverse samples.

Regarding method issues, although overall effects were similar for HR and OR studies, effects for some moderators varied by HR versus OR analyses. The total number of covariates included in analytic models moderated associations with cardiometabolic disease for both types of analyses, suggesting indirectly that some of the covariates (particularly those related to health behaviors) may be part of the pathways connecting cumulative adversity with health outcomes. However, smaller effects were seen in HR studies that adjusted for psychosocial covariates. As much as possible, future studies should adjust for known correlates of both childhood adversity and long-term disease risk, in order to better understand the direct association of childhood adversity on cardiometabolic disease and identify potential mediators that may serve as intervention targets (see also Su et al., 2015, for a review of these issues).

Interestingly, the type of analytic strategy emerged as a significant moderator, such that OR effects were larger when analytic models involved comparison of extreme groups of adversities relative to a total score or average of a trauma/adversity scale, although it was not possible to test this moderator in HR effects (93% used an extreme groups approach). Given that all studies rely on the distribution of total number of adversities reported within their own sample in order to categorize individuals, there are stark differences among studies in how many adversities are included in the “extreme” category; for example, using 2+ adversities (e.g., Danese et al., 2009) vs. 4+ (Felitti et al., 1998) vs. 7–9 (e.g., Gilbert et al., 2015). As the field develops, it would be helpful to develop standard ways to measure and describe exposures to childhood adversity to allow comparisons across studies and populations, including using both continuous measures and discrete categories of exposure.

Considering the broader literature, there are significant methodological limitations that should be addressed and also limit the strength of the findings of this meta-analysis (see also Appleton et al., 2017 and Basu et al., 2017 for excellent discussions of these issues). First, there is wide variability regarding the number of adversity items included in cumulative indices, with some studies assessing as many as 27 items. Thus, there are many unique combinations of adversities that could be endorsed, and it is unclear whether it is actually the combination of certain types of adversities, a particular number of adversities, or the duration (c.f. Alastalo et al., 2009; Pesonen et al., 2007), chronicity (c.f. Slopen, Koenen, & Kubzansky, 2014), severity (c.f. Schilling, Aseltine, & Gore, 2008), or timing of exposure (c.f. Friedman, Montez, Sheehan, Guenewald, & Seeman, 2015; McLaughlin et al., 2015; Pesonen et al., 2010; Slopen et al., 2014; Slopen, Kubzansky, McLaughlin, & Koenen, 2013) that is most health-damaging. Future research should take a more exhaustive approach to measurement and analysis to overcome these limitations, as these details may provide a window into how and when to best intervene.

Second, not all studies include a measure of childhood SES in cumulative indices of adversity, despite established literatures from the areas of developmental and health psychology that indicate associations between childhood SES and poor long-term physical health and psychosocial outcomes (Cohen et al., 2010). Moreover, some researchers include childhood poverty as an adverse exposure. Thus, it is important for researchers to come to consensus about whether poverty or low childhood SES should be consistently included in cumulative measures of adversity. Given that the inclusion of childhood SES in measures of adversity may influence estimated cumulative effects (e.g., Table 2), it seems wise to investigate SES and adversity as independent variables in the same analysis, in order to inform our understanding of their overlap and their independent and aggregate predictive utility (see Appleton et al., 2017).

Third, the majority of studies analyzed here (over 90%) used retrospective self-reports of childhood adversity. Some evidence suggests that this may lead to inflated results, both in terms of the report of adversity and relationships with self-reported health outcomes, and that retrospectively-reported adversity may be less related to objectively-measured outcomes than prospectively-reported adversity (see Reuben et al., 2016; Widom, Raphael, & DuMont, 2004). Prospective adversity data have their own set of limitations, such as the possibility of under-reporting by parents or children due to fear of legal or social consequences. However, at face value, one may hypothesize that prospective data are “closer to the truth” and less subject to poor memory, biased post-hoc interpretation of past events, or the influence of concurrent negative mood or psychosocial adversity. Thus, it is important in future research, when feasible, to measure exposure to adversity both prospectively and retrospectively to elucidate the role of childhood adversity in adult cardiometabolic disease.

Ultimately, although exposure to cumulative childhood adversity appears to have long-term negative consequences for adult cardiometabolic disease, there are several promising avenues with regard to the mitigation of cardiometabolic risk in youth and adults. First, several studies in racially diverse samples indicate that interventions designed to improve parenting may have indirect but salubrious relationships on health outcomes, including decreased BMI (Smith, Montano, Dishion, Shaw, & Wilson, 2015), blood pressure (Brotman et al., 2012), and inflammation (Miller, Brody, Yu, & Chen, 2014) in youth samples, as well as decreased CVD risk factors and clinical events in adults, decades after the intervention (F. Campbell et al., 2016). Second, extant data suggests that adults who report a history of early adversity but also report greater resilience may demonstrate relatively better mental and physical health (e.g., Surtees & Wainwright, 2007; Wingo et al., 2010). Indeed, a recent review describes the scientific and clinical utility of investigating domains of development that may be shaped by exposure to stressful life experiences (e.g., learning, memory, attention) to potentially enhance long-term social, emotional, and functional outcomes (Ellis, Bianchi, Griskevicius, Frankenhuis, 2017).

In conclusion, the present analysis adds to the existing literature in a number of ways. First, it is the most thorough and recent review of the childhood adversity literature that also includes relevant papers from the child stress literature, and it provides the first quantitative analysis focused on clinical cardiometabolic disease. Second, it goes beyond the definition of adversity in terms of individual types of maltreatment to include other forms of adversity, which resulted in a large number of studies and effects to include. Finally, it examined a broad range of potential moderators of the effects. Based on the reviewed literature, this meta-analysis revealed a moderate cumulative effect of childhood adversity on risk for cardiometabolic diseases, with the evidence stronger for CVD clinical outcomes than for metabolic outcomes.