Associations between childhood maltreatment and mid-life health risks and risky health behaviors in a nationally representative sample

Abstract

Millions of U.S. children experience maltreatment each year, which is associated with subsequent health risks, including psychopathology. We use data from up to 9692 participants in the National Longitudinal Study of Adolescent to Adult Health (Add Health) to update and extend knowledge about such health risks into the start of midlife. These data include detailed retrospective measures of supervisory neglect, physical neglect, physical abuse, and sexual abuse, and also include data on health and health-relevant behaviors that were self-reported (general health, depression measured as CES-D negative affect subscale, smoking, alcohol use, and illegal drug use) and directly measured (BMI, hypertension, renal function, and cardiovascular disease risk) at ages 32-42. We use these data to (1) provide the strongest evidence to date about the persistence of health risks among formerly maltreated individuals approaching midlife and (2) examine the prevalence of such risks among two particularly vulnerable groups: those who experienced multiple types of maltreatment, and those who experienced frequent maltreatment. Our analyses reveal depression and smoking as salient midlife health risks among individuals who experienced physical neglect and sexual abuse. We also find evidence that multi-victimization and recurring maltreatment increases such risks but find no evidence of linear dose-response relationships between maltreatment frequency and health. Findings are robust to the inclusion of covariates that control for other individual-, family-, and neighborhood-level risk factors.

1. Introduction

Around 40% of American children will experience abuse or neglect before age 18, and around 12% will have their maltreatment confirmed by Child Protective Services (CPS) (Finkelhor et al., 2015; Yi et al., 2020). Such maltreatment is associated with adverse health and behavioral outcomes in the short-term, including delayed developmental trajectories, increased substance use, risky sexual behavior, higher rates of depression and personality disorders, and poor physical health (Arnow, 2004; Hussey et al., 2006; Jaffee, 2017; Jaffee et al., 2004; Jonson-Reid et al., 2012; Miller and Lacey, 2022; Nanni et al., 2012; Norman et al., 2012; Segal et al., 2021; Shonkoff and Garner, 2012). Moreover, individuals with histories of childhood maltreatment may suffer some deleterious consequences deep into adulthood (Corso et al., 2008; Costello et al., 2011; Danese et al., 2007; Gilbert et al., 2009; Jackisch et al., 2021). Recent research in this domain has focused on the potential persistence of mental and physical health consequences (Jaffee, 2017; Lee et al., 2023; Miller and Lacey, 2022; O'Shields et al., 2024; Shen et al., 2025) and on potentially elevated vulnerabilities among individuals who experienced multiple types or episodes of maltreatment during childhood (Bayly et al., 2022; Goncalves Soares et al., 2021; Hughes et al., 2017; Miller and Lacey, 2022; Steine et al., 2017).

In this study, we leverage uniquely suitable data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) to update and expand knowledge of such long-term health vulnerabilities. We make two advancements. First, we examine the health consequences of childhood maltreatment among adults approaching midlife (ages 32-42). While many relevant, widely-cited studies focus on outcomes during adolescence or the transition to early adulthood (Hussey et al., 2006; Jaffee, 2017; Jonson-Reid et al., 2012; Segal et al., 2021), the longer-term health consequences of childhood adversity are a pressing research domain—and one that can now be more systematically addressed, as more data on midlife health are available.

Our study is not the first to consider this part of the life course. However, prior knowledge on midlife health is based on studies that use select local samples (Corso et al., 2008; Felitti et al., 1998; Shea et al., 2022), comparatively small and more outdated national samples such as the National Survey of Midlife Development in the United States (MIDUS) (Morton et al., 2014; O'Shields et al., 2024), or data from European countries with large public healthcare systems and near-universal healthcare coverage (Cheong et al., 2017; Jackisch et al., 2021). Research in this area that leverages Add Health's large, broadly representative longitudinal sample and detailed measures of health and maltreatment in the United States stops in early adulthood, when respondents were aged 24 to 32 (Shen et al., 2025). We add to this research by using models that are sufficiently well-powered to detect associations between specific types of childhood maltreatment and relatively rare midlife health risks, using a nationally representative sample.

Second, we examine health risks and risky health behaviors among individuals who experienced multiple types or episodes of childhood maltreatment. Researchers have previously analyzed whether multi-victimization matters for health outcomes (Goncalves Soares et al., 2021; Hughes et al., 2017; Miller and Lacey, 2022; Turner et al., 2013), using surveys that distinguish physical neglect, physical abuse, and sexual abuse (such as the National Survey of Children's Exposure to Violence, the Rochester Youth Development Study, the Lehigh Longitudinal Study, the Minnesota Longitudinal Study of Parents and Children, or the California Health Interview Survey). However, these surveys do not allow researchers to measure how frequently or persistently abuse and neglect occurred. In light of research that finds evidence of dose-response relationships yet is often limited by small and local samples (English et al., 2005; Jonson-Reid et al., 2012; Manly et al., 1994; Steine et al., 2017), it is crucial to understand the impact of multi-victimization as well as cumulative maltreatment exposure on health outcomes. Add Health data allow us analyze these potential sources of vulnerability within a single research design, which makes it possible to quantify health risks among individuals who experienced multi-victimization while also testing for dose-response relationships between childhood maltreatment and health risks.

Our research design requires us to give serious consideration to the validity of self-reported data about childhood adversity and midlife health risks. Retrospective self-reports of traumatic childhood events—such as those included in Add Health datasets—are subject to potential recall bias and may differ from prospective measures (Baldwin et al., 2019; Shen et al., 2025). However, such data are still highly useful for population studies, are widely used to study the consequences of childhood adversity, and are often preferable over alternatives (Gilbert et al., 2009; Hardt and Rutter, 2004; Hussey et al., 2006; Li et al., 2019; Shen et al., 2025; Suglia et al., 2014; Yau et al., 2022). Official maltreatment data from CPS agencies only count instances of maltreatment that were reported to CPS, leading to severe undercounts (Saunders and Adams, 2014). Alternatively, collecting prospective maltreatment reports from parents or very young children would come with significant practical and ethical challenges and is unlikely to yield accurate results due to fears of social stigma or criminal sanctions, parental pressures on children, or young children's inability to identify maltreatment in real time. Add Health maltreatment data have also been rigorously validated by confirming that respondents' age at enumeration did not affect retrospectively self-reported maltreatment frequencies (which provides suggestive evidence against age-dependent recall bias) and by comparing the observed prevalence of maltreatment in Add Health to the prevalence observed in other surveys (Currie and Tekin, 2012; Eiermann et al., 2025; Tekin and Markowitz, 2008).

Unlike other formative early-life course events that are widely studied, such as educational performance or contact with the legal system, measuring maltreatment in an objective manner is functionally impossible, since discretionary judgment is often involved in deciding whether any particular instance meets statutory criteria and/or rises to the level of actionable abuse and neglect (LaSelle, 2025). However, clinical measurements of cardiovascular and other risk factors can provide an objective snapshot of a person's health, and information about prescription medications can offer clues about underlying health conditions (Marasine et al., 2021). Since patients may be unaware of their true health or may not accurately self-report the symptoms they experience (Celis-Morales et al., 2012; Goldman et al., 2003; Thompson et al., 1999; Xie and Wang, 2020), such measurements can help researchers to address concerns about the validity of self-reported health information.

Add Health data—which are widely used to study health trajectories and health risks (Hussey et al., 2006; Morris et al., 2006; Shen et al., 2025; Shin and Miller, 2012; Udry et al., 2003)—include in-person measurements of BMI, health measures obtained from laboratory analyses of biospecimen, and detailed records of medical prescriptions. This allows us to combine validated self-reported measures (e.g., the Center for Epidemiological Studies-Depression CES-D scale) with information about directly measured health outcomes and data on prescriptions. Importantly, Add Health also includes information about family and community contexts, which have been found to mediate the health consequences of ACEs (Negriff, 2020; Pereira and Zhu, 2025). Taken together, the large sample size and the wide range of health, maltreatment, and contextual data make it possible to implement a research design that yields robust and nationally representative estimates of midlife health risks among formerly maltreated individuals while accounting for childhood adversities, multi-victimization, and the type and dosage of maltreatment.

2. The known health consequences of childhood maltreatment

Adverse childhood experiences (ACEs) can take many forms, including parental incarceration, frequent parental drug and alcohol use, divorce, abuse, and neglect (Boullier and Blair, 2018; Cheong et al., 2017; Lee et al., 2023). But among the full sweep of ACEs that U.S. children encounter each year, maltreatment stands out as a likely key driver of future health problems (Danese et al., 2007; Negriff, 2020). Such maltreatment is regrettably common, with more than 10% of children experiencing some form of maltreatment in any given year (Finkelhor et al., 2014; Gilbert et al., 2009). During childhood and adolescence, maltreated children are more likely to experience psy-chopathology, anxiety disorder, posttraumatic stress disorders, psy-chotic symptoms, personality disorders, and substance use (Cohen et al., 2001; Jaffee, 2017; Jaffee et al., 2004; Jonson-Reid et al., 2012; Milot et al., 2010). Maltreated girls are also more likely than never-maltreated girls to be diagnosed with a sexually transmitted disease or experience teenage pregnancy (Cohen et al., 2001; Font et al., 2022). Such vulnerabilities can be effected through injuries or developmental delays caused by maltreatment, physiological pathways that include allostatic load and hormonal imbalances, and behavioral pathways that include a greater likelihood to engage in risky behaviors (Breton et al., 2022; Corso et al., 2008; Felitti et al., 1998; Hughes et al., 2017; Milot et al., 2010; Morton et al., 2014; Umberson and Montez, 2010). Concurrent ACEs—such as domestic violence, parental substance abuse, parental incarceration, or housing instability—can exacerbate those health vulnerabilities by introducing additional risk factors (Jaffee and Maikovich-Fong, 2011; Jonson-Reid et al., 2012; Lee et al., 2023), while supportive social environments and remedial CPS interventions may mitigate associations between childhood adversity and subsequent health problems (Carroll et al., 2013; Font and Gershoff, 2020; O'Shields et al., 2024; Pereira and Zhu, 2025).

Prior studies have focused on four types of maltreatment: supervisory neglect, physical neglect, physical abuse, and sexual abuse (Hussey et al., 2006; Negriff, 2020). Sexual abuse has been linked to particularly severe health consequences during childhood and adolescence, which can also persist in adulthood (Arnow, 2004; Cohen et al., 2001; Maniglio, 2009; Steine et al., 2017). Victims of sexual abuse are at greater risk of experiencing subsequent mental health problems (Jaffee, 2017; Negriff, 2020), while individuals with histories of physical abuse and physical neglect have been found to experience worse cardiometabolic health than never-maltreated individuals (Li et al., 2019; Miller and Lacey, 2022).

Several health disadvantages have been shown to persist into adulthood for maltreated individuals generally (Hussey et al., 2006) and for the subset of maltreated individuals who experienced contact with CPS or other forms of state intervention in family life (Jackisch et al., 2021). Affected adults are at higher risk of depression (Arnow, 2004; Jaffee, 2017; O'Shields et al., 2024) and are more likely to suffer from poor physical health, including inflammation (Danese et al., 2007; O'Shields et al., 2024), cardiometabolic risk factors like elevated BMI and cholesterol (Li et al., 2019; Shin and Miller, 2012), higher triglyc-eride concentrations (Li et al., 2019; Miller and Lacey, 2022), and strokes or heart attacks (Lee et al., 2023; Morton et al., 2014). Some differences between ever-maltreated and never-maltreated individuals—for example, differences in the prevalence of depression and heart attacks—remain observable at midlife (Morton et al., 2014; O'Shields et al., 2024). However, recent research also suggests that associations between childhood maltreatment and depressive symptoms may weaken over time (Shen et al., 2025; Steine et al., 2017; Umberson and Montez, 2010).

Individuals who suffered from more than one type of mal-treatment—also known as multi-victimization—are potentially at greater risk of experiencing adverse health outcomes than individuals who suffered from a single type of maltreatment (Felitti et al., 1998; Goncalves Soares et al., 2021; Hughes et al., 2017; Miller and Lacey, 2022; Turner et al., 2013). Elevated vulnerability has also been observed for those who experienced maltreatment during multiple developmental periods (Jaffee and Maikovich-Fong, 2011) and who experienced multiple CPS investigations or multiple ACEs during childhood and adolescence (Felitti et al., 1998; Font et al., 2022).

Given such evidence of elevated vulnerability, several studies have probed for dose-response relationships between maltreatment and health outcomes, based on the premise that each additional episode of abuse and neglect may increase subsequent health risks. Using data from the Longitudinal Investigation of Sexual Abuse (LISA) in Norway, one study identified dose-response relationships between sexual abuse and mental health outcomes (Steine et al., 2017). Another study, using local data from a low-income sample in the St. Louis area in the US, also identified dose-response relationships between maltreatment and mental health risks but additionally found that those relationships were attenuated in adulthood once other ACEs were added into the analysis (Jonson-Reid et al., 2012).

One challenge when evaluating the evidentiary basis for such claims is that definitions and measurements of dosage vary. Many studies sidestep this issue by simply dichotomizing maltreatment experiences into “ever experienced” vs. “never experienced” (Hussey et al., 2006; Morton et al., 2014) or by using cutoff scores (Brumley et al., 2019). Others focus on chronicity, defined as the occurrence of maltreatment during successive developmental periods or years (Bayly et al., 2022; English et al., 2005; Jaffee and Maikovich-Fong, 2011; Manly et al., 1994). Finally, some studies take ordinal measures of childhood exposure to ACEs or maltreatment (e.g., experiencing maltreatment “often” or “very often”, see Wang et al., 2020) and translate those categories into integer variables by using the numeric value associated with each ordinal category (Carroll et al., 2013; O'Shields et al., 2024; Shen et al., 2025; Steine et al., 2017; Wang et al., 2020). This is based on the plausible assumption that higher ordinal categories are a useful proxy for higher dosage and that corresponding integer variables can therefore be used to compute aggregate “trauma scores” (Steine et al., 2017). However, this approach has the disadvantage of giving equal weight to each ACE or maltreatment type (e.g., a single instance of supervisory neglect increases trauma scores by as much as a single instance of sexual abuse) and also assumes that differences between ordinal categories are substantively equal (i.e., moving from 1 instance to 2 instances of maltreatment can have the same impact on trauma scores as moving from 6 instances to 16 instances) (Shen et al., 2025), both of which are strong assumptions.

3. Methods

3.1. Data source and participants

We intervene in this literature by analyzing data from the Add Health study, which has followed a nationally representative sample across multiple waves of data collection (henceforth W1-W5) from ages 12-19 to ages 32-42. Add Health includes detailed questions about exposure to maltreatment during childhood and adolescence, as well as a large suite of variables that capture health and health-relevant behaviors based on self-reports, on-site measurements, and laboratory analyses of biospecimen. The original W1 sample (n = 20,745) was collected in 1994-95 from students in grades 7-12 in 132 schools across 80 US communities, using a school-based stratified sampling strategy (Harris et al., 2019). Data for W3 (n = 15,197) were collected in 2001-21 and data for W4 (n = 15,701) were collected in 2008. W5 data (n = 12,300)—the source of our midlife health outcome measures—were collected in 2016-18 when those respondents were aged 32-42. Separate survey weights are available for each wave sample. Because this research involved existing deidentified survey data, it was considered exempt by our IRB boards.

Following prior Add Health-based research on the health consequences of childhood adversity (Hussey et al., 2006; Yau et al., 2022) we use demographic and socioeconomic measures recorded in closest proximity to childhood during W1, data on family and neighborhood contexts collected from parents of W1 respondents (usually the mother or female head of household) at the same time as the W1 survey in 1995, and data on childhood maltreatment collected retrospectively during W3 and W4. We use midlife health data from W5. Maltreatment measures at W3 or W4 and health measures and survey weights at W5 are available for n = 9692 respondents (Table 1). The demographics of this analytic sample closely resemble the full W5 sample. Maltreatment data and complete biomarkers based on laboratory measurements are only available for 3703 respondents who had additionally consented to the collection and analysis of biospecimen. The composition of this biomarker subsample (which is still considerably larger than survey samples used in recent related research, see O'Shields et al., 2024) is shown in Table 1 and also resembles the complete W5 sample.

Table 1.

Sample statistics.

	By Add Health sample			By maltreatment type					By maltreatment severity
	Full W5 sample	Analytic sample	Biomarker subsample	Any abuse/ neglect	Supervisory neglect	Physical Neglect	Physical Abuse	Sexual Abuse	Multi- victimization	6+ instances
n	12300	9692	3703	4818	3649	963	1819	482	1637	1951
Male (%)	50.4	49.1	48.9	50.8	51.1	60.4	51.0	29.2	52.5	51.1
Female (%)	49.6	50.9	51.1	49.2	48.9	39.6	49.0	70.8	47.5	48.9
White (%)	65.5	66.4	66.7	64.6	64.4	58.5	62.1	60.3	59.2	60.5
Black (%)	15.5	14.8	15.8	14.6	14.4	18.7	14.4	18.7	16.3	14.7
Hispanic (%)	12.3	11.7	10.9	12.7	12.8	13.9	13.7	12.2	14.1	14.4
Asian (%)	4.2	4.4	3.2	4.9	4.9	5.8	5.6	4.4	5.7	6.8
AIAN (%)	2.6	2.8	3.3	3.3	3.5	3.1	4.3	4.4	4.7	3.5
Living below poverty line (%)	14.9	14.2	13.8	15.7	15.5	23.4	16.7	24.4	20.7	17.4
Parent without high school degree (%)	14.6	14.3	11.8	14.7	14.3	21.1	15.9	22.1	18.6	15.3
Rural (%)	28.5	29.2	28.3	27.2	26.8	27.2	26.6	27.7	26.5	24.5
Suburban (%)	38.4	38.4	38.6	38.2	39.0	34.4	37.3	34.7	36.1	39.9
Urban (%)	30.1	29.7	30.2	31.7	31.2	34.1	32.9	34.0	33.4	32.4
Northeast (%)	16.4	16.2	15.2	18.0	18.7	18.7	17.8	21.2	19.3	20.9
Midwest (%)	31.3	32.3	35.8	32.0	32.0	30.8	33.3	27.0	32.3	33.3
South (%)	38.9	39.0	39.0	37.2	36.9	38.1	33.9	39.3	34.7	32.6
West (%)	13.4	12.4	10.1	12.8	12.4	12.5	15.0	12.5	13.8	13.2

Notes: Percentages are weighted, using Add Health cross-sectional weights. Racial categories (White, Black, Asian, AIAN) are non-Hispanic. Maltreatment categories are not mutually exclusive, i.e. respondents can self-report more than one maltreatment type.

3.2. Measures

Maltreatment.

Respondents retrospectively self-reported supervisory neglect, physical neglect, physical abuse, and sexual abuse prior to 6th grade during W3 and also self-reported physical and sexual abuse prior to their 18th birthday during W4 (Table 2). These measures mirror maltreatment classifications used in other surveys and in prior research (Currie and Tekin, 2012; Hussey et al., 2006; Shen et al., 2025; Suglia et al., 2014; Yau et al., 2022). In our main models, we focus on W3 measures of supervisory neglect and physical neglect and W4 measures of physical and sexual abuse. This is because the former two maltreatment types are applicable mostly to young children, making the 6th grade cutoff imposed by W3 questionnaires appropriate. In the Supplemental Appendix (SA) we present results from models that only use maltreatment data from W3, including measures of physical and sexual abuse.

Table 2.

Maltreatment measures.

Wave	Age range	Variable	Definition
3	To the start of sixth grade	Supervisory neglect	Left home alone when parents or other adult care-givers should have been with the child
		Physical neglect	Not having basic needs met by parents or other adult care-givers, such as personal hygiene, food, and clothing
		Physical abuse	Slapped, hit, or kicked by parents or other adult care-givers
		Sexual abuse	Touched in a sexual way by parents or other adult care-givers, forced to touch them in a sexual way, or forced to have sexual relations
4	Before the 18th birthday	Physical abuse	Slapped, hit, or kicked by parents or other adult care-givers
		Sexual abuse	Touched in a sexual way by parents or other adult care-givers, forced to touch them in a sexual way, or forced to have sexual relations

We count respondents as mono-victimized if they self-reported exactly one type of maltreatment and count them as multi-victimized if they self-reported two or more types of maltreatment. Maltreatment frequencies were reported—separately for each type—in six bins: “Never”, “1 time”, “2 times”, 3-5 times”, “6-10 times”, and “More than 11 times.” Our main models (a) collapse all frequency bins into a binary yes/no measure of maltreatment and (b) alternatively translate binned frequencies into numeric counts by assigning to each respondent the mean count of their bin (e.g., respondents reporting 6-10 instances are counted as having experienced 8 instances). We use this frequency count to test for linear dose-response relationships. In a supplemental model (shown in the SA), we additionally collapse those counts into broader categories of frequent (6+ instances) vs. less frequent (less than 6 instances) maltreatment.

Self-reported health risks

All self-reported measures of health risks come from W5, when respondents were aged 32-42 (Table 3). Following prior research, we measure mental health with the Center for Epide-miologic Studies-Depression (CES-D) negative affect subscale, using five questions about the number of times that respondents felt “depressed,” “sad,” “like they could not shake off the blues,” “happy (reverse coded),” and “like life was not worth living” during the previous week (never or rarely, sometimes, a lot of the time, most of the time/all the time) (Perreira et al., 2005). We use this subscale because the full CES-D scale is not available in Add Health. We use ≥4 as the CES-D cutoff score for depression based on prior validations of 5-item CES-D subscales (Zauszniewski and Graham, 2009), but results are substantively the same when we use adjacent cutoffs and replace the CES-D subscale with a single W5 question about having felt “depressed” during the past week. We recode all self-reported health risks as dichotomous outcome variables, shown in Table 3.

Table 3.

Dichotomous health outcome measures.

Wave	Method	Variable	Definition
5	Self-reported	Poor general health	1 = fair or poor general health
		Depressiona	1 = CES-D negative affect subscale score ≥4
		Smoking	1 = any smoking during preceding 30 days
		Daily alcohol use	1 = drinking every day or almost every day during preceding 30 days
		Illegal drug use	1 = any use of LSD, PCP, ecstasy, mushrooms, speed, ice, heroin, pills, cocaine, crystal meth, or inhalants during preceding 30 days
	Measured at interview	Obesity	1 = BMI ≥30 (BMI computed by authors from measured height and weight)
	Laboratory analyses	Hypertension+	1 = stage 1 or 2 hypertension, using guidelines from the Seventh Report of the Joint National Committee on Prevention Detection, Evaluation, and Treatment of High Blood Pressure
		Reduced kidney function	1 = decreased renal function, using the creatinine-based 2021 CKD-EPI eGFR equation
		Cardiovascular risk+	1 = cardiovascular disease risk rated “high”, using classifications of hsCRP concentrations in accordance with the American Heart Association/Centers for Disease Control recommendations

⁺In line with prior work, measures of hypertension and cardiovascular risk are adjusted based on self-reported prescription medications. SA models use unadjusted measures.

^aSA models replace CES-D scores with an alternative measure of mental health (“During the past week, how often have you felt depressed?” 1 = a lot/most of the time/all of the time).

Measured and biomarker-based health risks.

Respondents’ weight and height were directly measured during W5 at-home exams. We use these measurements to compute the Body Mass Index (BMI) and categorize all respondents with BMI≥30 as obese. For respondents in the biomarker subsample, we additionally analyze three variables that were computed from laboratory analyses of biospecimen, also shown in Table 3: Hypertension, renal function, and cardiovascular disease risk. We focus on these variables because they have been linked to ACEs in prior research (Goncalves Soares et al., 2021; Hughes et al., 2017; Li et al., 2019; Miller and Lacey, 2022; Suglia et al., 2014). Our models also include two health-specific covariates that measure whether a respondent was born with a low birth weight (clinically defined as less than 5.5 pounds) and lacked health insurance coverage during childhood.

Prescription medications.

Prescription medication use can push abnormal biomarkers into the normal range. Following prior studies (Kraav et al., 2019; Li et al., 2019; Suglia et al., 2014; Xie and Wang, 2020), we therefore reclassify respondents as having stage 1-2 hypertension and high cardiovascular disease risk if they reported relevant prescription medications (antihypertensive and and/or anti-hyperlipidemic medications) during the W5 Medications Exam, which was administered to respondents in the biomarker subsample. In sensitivity checks (shown in the SA), we also use data on prescribed antidepressants to derive an adjusted CES-D measure, counting respondents as experiencing depression if they reported the use of antidepressants even if their symptomology fell below the CES-D cutoff score.

Other ACEs:

Maltreatment can co-occur with, or be preceded by, other ACEs, making it difficult to ascribe health consequences specifically to childhood abuse and/or neglect (Hughes et al., 2017; Negriff, 2020). While our data do not have as precise temporal granularity as would be optimal, we address this by adjusting for the co-presence of a wide range of ACEs that include experiencing, before age 18: maternal incarceration (2.1% in the weighted sample), paternal incarceration (9.5%), parental divorce (13.8%), maternal death (1.3%), paternal death (3.6%), and easy drug availability in the home (3.0%). We use the latter—which was reported by respondents during W1—as a proxy for parental drug use, since parents were not asked to self-report illegal drug use. We also adjust for the frequent presence of drug dealers in the childhood neighborhoods of respondents (8.2% in the weighted analytic sample) as a proxy for environmental dysfunction, a potential stressor during childhood. This information was collected from the parents of W1 respondents.

Controlling for a wide range of ACEs that have been plausibly linked to health disadvantages minimizes the risk of omitted variable bias in our analyses but risks introducing overcontrol bias, especially since Add Health data do not include detailed information on the sequencing of maltreatment and other ACEs (e.g., maltreatment may have preceded a divorce or may have occurred after the divorce). We present results from models that do not adjust for other ACEs, and thus minimize overcontrol bias, in the SA. We also implement additional subgroup analyses (shown in the SA) to test whether the health consequences of maltreatment vary by gender (male vs. female) and by the onset of maltreatment (before/after age 7, which is approximately the age at which children enter elementary school).

Other background covariates.

We control for sociodemographic covariates measured in closest proximity to childhood during W1. Specifically, we adjust for sex (male, female), Hispanic ethnicity, race (White, Black, Asian/Pacific Islander, American Indian/Alaska Native), parental reliance on welfare assistance as a proxy for socioeconomic status (SES), highest level of parental educational attainment, usually collected from the mother (less than high school, high school or more), and Census region. We impute missing demographic data for a small number of respondents (e.g., ethnicity: 0.3% missing; race: 0.1% missing) and describe the multiple imputation by chained equations (MICE) procedure in the SA.

3.3. Analytic strategy

Our analyses proceed in three stages. First, we tabulate the counts and weighted percentages of respondents who had experienced different types and frequencies of maltreatment (Fig. 1). Second, we tabulate the observed prevalence of self-assessed and measured health risks for populations of respondents who had experienced different maltreatment patterns—by type and frequency—and identify differences in the prevalence of health risks across groups using Wald tests (Table 4). Third, we fit binomial regression models that estimate the association between childhood maltreatment and health risks during early midlife (Table 5). We use W5 cross-sectional survey weights in line with Add Health research design recommendations. Results are reported as prevalence ratios (PR, which measure relative risk) with 95% confidence intervals after applying a Holm-Bonferroni correction of p-values, which controls the family-wise error rate. We show prevalence ratios rather than odds ratios because odds ratios can overestimate the strength of an association when the outcome of interest is relatively common in the population (>10-15%), which in the case for several health outcomes (Nordahl-Hansen et al., 2018; Suglia et al., 2014). Standard errors are clustered at the school-level to account for the Add Health stratified sampling strategy.

Fig. 1.

Maltreatment frequencies, overall and by type.

Table 4.

Prevalence of health risks in early midlife.

	By Add Health sample			By maltreatment type					By maltreatment severity
	Full W5 sample	Analytic sample	Biomarker subsample	Any abuse/ neglect	Supervisory neglect	Physical Neglect	Physical Abuse	Sexual Abuse	Multi- victimization	6+ instances
Self-reported health risks
Poor general health (%)S	14.1 (0.6)	14.0 (0.7)	14.0 (0.8)	16.0 (0.9)	15.8 (1.0)	20.9 (1.9)	18.7 (1.3)	19.9 (2.5)	20.9 (1.5)	17.4 (1.2)
Depression (CES-D subscale) (%)S	36.0 (0.7)	35.8 (0.8)	36.0 (1.0)	39.0 (1.0)	39.5 (1.1)	45.8 (2.4)	42.8 (1.7)	49.9 (2.8)	46.8 (1.7)	42.0 (1.6)
Smoking (%)S	27.8 (1.0)	27.4 (1.0)	25.7 (1.2)	32.2 (1.1)	31.3 (1.3)	42.8 (2.5)	35.1 (1.8)	37.0 (3.2)	38.4 (2.0)	33.9 (1.6)
Daily alcohol use (%)S	5.8 (0.4)	5.5 (0.4)	5.7 (0.5)	5.0 (0.5)	5.1 (0.6)	5.1 (1.0)	5.1 (0.7)	2.7 (1.0)	4.9 (0.8)	4.9 (0.7)
Illegal drug use (%)S	3.9 (0.3)	3.9 (0.3)	4.1 (0.4)	4.8 (0.5)	4.8 (0.5)	5.1 (1.0)	5.0 (0.7)	7.0 (1.8)	5.5 (0.8)	4.8 (0.7)
Measured health risks
Obesity (%)M	40.6 (0.9)	40.7 (1.0)	42.4 (1.2)	44.6 (1.5)	43.6 (1.7)	51.5 (3.7)	49.1 (2.5)	46.9 (4.3)	48.8 (2.7)	49.5 (2.2)
Hypertension (%)M			28.9 (1.0)	28.0 (1.4)	28.7 (1.5)	33.5 (3.3)	30.8 (2.3)	25.6 (3.8)	31.3 (2.2)	27.0 (2.1)
Reduced kidney function (%)M			18.1 (1.1)	20.0 (1.4)	19.6 (1.4)	23.7 (3.8)	21.0 (2.2)	18.3 (4.0)	21.6 (2.3)	19.6 (1.9)
Cardiovascular risk (%)M			45.0 (1.2)	46.8 (1.5)	46.4 (1.7)	56.3 (3.6)	51.6 (2.8)	54.8 (4.5)	54.4 (3.0)	48.8 (2.2)

Notes:

^{S =}self-assessed by respondents

^{M =}measured during interviews or through laboratory analyses. Percentages are weighted, using Add Health cross-sectional weights. SEs are clustered at the school level and account for Add Health’s regionally stratified school samples. Maltreatment categories are not mutually exclusive, i.e. respondents can self-report more than one maltreatment type.

Table 5.

Associations between maltreatment (any/all types) and health risks in early midlife.

Model	Maltreatment measure	Poor general healthS	Depression (CES- D subscale)S	SmokingS	Daily alcohol useS	Illegal drug useS	ObesityM	Hyper- tensionM	Reduced kidney functionM	Cardio- vascular riskM
(1)	Supervisory neglect (binary)	1.12 (0.91, 1.38)	1.12 (1.01, 1.24)	1.17 (1.03, 1.32)	0.88 (0.63, 1.23)	1.61 (1.08, 2.40)	1.07 (0.98, 1.17)	1.01 (0.86, 1.19)	1.14 (0.90, 1.45)	1.09 (0.98, 1.23)
	Physical neglect (binary)	1.13 (0.85, 1.49)	1.15 (0.98, 1.34)	1.38* (1.15, 1.65)	1.03 (0.65, 1.65)	0.77 (0.42, 1.41)	1.00 (0.87, 1.15)	1.20 (0.94, 1.54)	1.02 (0.69, 1.53)	1.13 (0.95, 1.33)
	Physical abuse (binary)	1.21 (0.97, 1.50)	1.17 (1.05, 1.31)	1.22 (1.06, 1.40)	1.08 (0.73, 1.60)	1.05 (0.70, 1.57)	1.11 (0.99, 1.26)	1.06 (0.85, 1.32)	1.16 (0.89, 1.51)	1.18 (1.01, 1.38)
	Sexual abuse (binary)	1.49 (1.09, 2.03)	1.32* (1.13, 1.53)	1.13 (0.87, 1.46)	0.50 (0.20, 1.21)	1.52 (0.68, 3.41)	1.05 (0.91, 1.21)	1.08 (0.72, 1.61)	0.80 (0.46, 1.39)	1.06 (0.88, 1.27)
(2)	Supervisory neglect (count)	1.00 (0.98, 1.02)	1.00 (0.99, 1.02)	1.01 (0.99, 1.03)	1.00 (0.96, 1.04)	1.02 (0.98, 1.06)	1.01 (1.00, 1.02)	0.99 (0.97, 1.02)	0.98 (0.94, 1.01)	1.01 (1.00, 1.03)
	Physical neglect (count)	1.03 (1.00, 1.06)	1.02 (1.00, 1.04)	1.03 (1.01, 1.05)	0.97 (0.90, 1.04)	1.00 (0.91, 1.09)	1.00 (0.98, 1.02)	1.02 (0.98, 1.05)	1.02 (0.96, 1.08)	0.99 (0.96, 1.02)
	Physical abuse (count)	1.01 (0.98, 1.04)	1.02 (1.00, 1.04)	1.03 (1.01, 1.05)	0.99 (0.93, 1.05)	0.97 (0.91, 1.03)	1.01 (1.00, 1.03)	1.01 (0.97, 1.04)	1.04 (1.01, 1.07)	1.02 (1.00, 1.05)
	Sexual abuse (count)	1.05 (1.01, 1.10)	1.03 (1.01, 1.05)	1.00 (0.96, 1.04)	0.79 (0.63, 0.99)	1.08 (0.97, 1.20)	1.01 (0.99, 1.04)	1.01 (0.95, 1.06)	1.00 (0.93, 1.06)	1.00 (0.98, 1.03)
(3)	Mono-victimization	1.17 (0.94, 1.47)	1.08 (0.98, 1.20)	1.29** (1.13, 1.47)	0.74 (0.55, 0.99)	1.37 (0.93, 2.00)	1.06 (0.97, 1.17)	0.93 (0.77, 1.13)	1.17 (0.95, 1.44)	1.04 (0.90, 1.21)
	Multi-victimization	1.50** (1.22, 1.84)	1.41** (1.27, 1.57)	1.66** (1.44, 1.90)	0.87 (0.56, 1.35)	1.45 (0.91, 2.33)	1.17* (1.06, 1.30)	1.09 (0.88, 1.34)	1.22 (0.90, 1.56)	1.33* (1.11, 1.58)

Notes: Shown as prevalence ratios (PR) with 95% confidence intervals (CI).

^{S =}self-assessed by respondents

^{M =}measured during interviews or through biological samples.

^{* =}significance at p = 0.05

^{** =}significance at p = 0.01. P-values are adjusted, using a Holm-Bonferroni correction. All models additionally include covariates for other ACEs, ethnicity, race, sex, parental educational attainment, parental welfare dependence, low birth weight, childhood health insurance, and census region. Covariate coefficients not shown.

The large Add Health sample size and prevalence of maltreatment and health risks mean that the minimum detectable effect size at power = 0.8 and alpha = 0.05 ranges from PR = 1.1 for more common risks like depression (36% prevalence in the sample) to PR = 1.3 for relatively rare health risks like daily alcohol use (6% prevalence). Minimum detectable effect sizes are larger for sexual abuse (which is itself relatively rare), ranging from PR = 1.2 for more common risks (like depression) to PR = 1.6 for relatively rare health risks (like daily alcohol use). This implies that we can identify small but clinically meaningful associations even for relatively rare health risks but may not be able to identify those associations specifically for survivors of sexual abuse. In analyses that use maltreatment frequency counts, our sample allows us to detect a minimum PR per one standard deviation (SD) increase in maltreatment frequency of approximately 1.05 for more common health outcomes and approximately 1.1 for less common health outcomes.

4. Results

4.1. Maltreatment patterns

4818 respondents in the analytic sample (49.7%) reported any instance of maltreatment, 3649 reported supervisory neglect (37.7%), 963 reported physical neglect (9.9%), 1819 reported physical abuse (18.7%), and 482 reported sexual abuse (5.0%). Among the 4818 respondents who ever experienced maltreatment, 3176 experienced mono-victimization and 1642 experienced multi-victimization. Sample statistics from Table 1 show that respondents who experienced multi-victimization, physical neglect, and sexual abuse were most likely to come from low-SES families. Respondents who experienced physical neglect were more likely to be male; respondents who experienced sexual abuse were more likely to be female; and respondents who experienced physical neglect and sexual abuse were more likely to identify as Non-Hispanic Black.

Looking at the cumulative dosage of maltreatment, 1960 respondents experienced six or more instances of maltreatment. This high frequency count was commonly due to multi-victimization rather than high-frequency mono-victimization. Fig. 1 illustrates this by plotting the weighted percentages of respondents who reported different frequencies. It shows that the modal maltreated respondent experienced frequent maltreatment but that, within each maltreatment type, a majority experienced 5 instances or fewer.

4.2. Prevalence of health risks

Table 4 shows the weighted prevalence of health risks during early midlife (ages 32-42). We find that the prevalence of all studied health risks except daily alcohol use and hypertension was greater among ever-maltreated individuals than among never-maltreated individuals. Comparing multi-victimized against mono-victimized respondents, we similarly find a greater prevalence of all health risks except daily alcohol use among the former. We also find that respondents who experienced six or more instances of maltreatment experienced a greater prevalence of multiple health risks, including poor general health, depression, and smoking, than respondents who reported less than six instances of maltreatment.

Across populations with different maltreatment experiences (by type, multi-victimization, and frequency), we find that fair/poor general health in early midlife was most common among those who had experienced physical neglect (20.9% reporting fair/poor general health) and multi-victimization (20.9%); depression was most common among those who had experienced sexual abuse (49.9%) and multi-victimization (46.8%); and smoking was most common among those who had experienced physical neglect (42.8%) and multi-victimization (38.4%). For measured health risks, we find that obesity was most common among those who had experienced physical neglect (51.5%) and recurring maltreatment (49.5%) while cardiovascular disease risk was most common among those who had experienced physical neglect (56.3%) and sexual abuse (54.8%). Collectively, these prevalence estimates hint at concentrated health disadvantage among individuals who experienced physical neglect and sexual abuse and who experienced more than one type of maltreatment.

4.3. Associations between childhood maltreatment and health risks

Next, we estimate associations between childhood maltreatment and midlife health risks with a series of binomial regression models (Table 5). We present three sets of models below and include additional sensitivity checks in the SA. The first set of models treats each maltreatment type as a binary (yes/no). The second set of models re-places this binary measure with a maltreatment count, allowing us to test for linear dose-response relationships. The third set of models compares multi-victimized respondents against mono-victimized and never-maltreated respondents. All models adjust for potentially co-occurring ACEs and socio-demographic background.

We find (Model 1) that sexual abuse was associated with an increased risk of midlife depression (PR = 1.32, CI=(1.13, 1.53)) and that physical neglect was associated with an increased risk of smoking (PR = 1.38, CI=(1.15, 1.65)). We do not find evidence that a higher dosage of maltreatment—measured as the frequency count of maltreatment episodes of each type—was linearly associated with increased health risks at midlife (Model 2). Models that omit the Holm-Bonferroni correction do show linear associations (at p = 0.05) between the dosage of physical neglect, physical abuse, sexual abuse, and midlife depression and between physical neglect, physical abuse, and smoking, However, those findings disappear once the family-wise error rate is taken into account (as is statistically appropriate).

In a supplemental model (shown in the SA), we also compute and then compare associations between maltreatment and health risks for respondents who experienced frequent (6+ instances) vs. less frequent (less than 6 instances) maltreatment. A majority (56%) of respondents in the “frequent” category are also multi-victimized, reflecting the fact that high cumulative maltreatment counts are often the result of multi-victimization rather than high-frequency mono-victimization. Results indicate that frequently maltreated individuals were more likely to experience midlife depression and obesity (but were not more likely to smoke) than infrequently maltreated individuals. We return to this point in the Discussion below.

We do find strong evidence that multi-victimization was associated with multiple health risks. Multi-victimized individuals (but not mono-victimized individuals) were more likely than never-maltreated respondents to experience poor general health (PR = 1.50, CI=(1.22, 1.84)), depression (PR = 1.41, CI=(1.27, 1.57)), obesity (PR = 1.17, CI=(1.06, 1.30)), and cardiovascular disease risk (PR = 1.33, CI=(1.11, 1.58)). Mono- and multi-victimized respondents were each more likely than never-maltreated respondents to smoke at midlife, although the prevalence ratio is significantly higher for multi-victimized respondents (PR = 1.66, CI=(1.44, 1.90)) than for mono-victimized respondents (PR = 1.29, CI=(1.13, 1.47)).

Core findings—i.e., associations between physical neglect and smoking; between multi-victimization, depression, and smoking; and between multi-victimization and depression—are replicated in supplemental models that use alternative maltreatment measures (using W3 data only) and drop ACE covariates to minimize the risk of overcontrol bias. The significant association between sexual abuse and depression is replicated in models that only use W3 data to measure maltreatment but not in models that omit ACE controls. Null findings about linear dose-response relationships are also replicated in supplemental models that use unadjusted measures of hypertension and cardiovascular disease risk (i.e., measures that do not take prescription medications into account).

In supplemental models, we do not find differences in associations between children who first experienced physical or sexual abuse early in life (prior to age 7), vs. children who first experienced a later onset of abuse. Subgroup analyses suggest that some health risks (e.g., midlife depression) may be elevated specifically among women. However, we note that those analyses are statistically underpowered—even with the large Add Health sample—and cannot reliably detect small or moderate gender-specific associations between relatively uncommon ACEs (e.g., sexual abuse) and midlife health outcomes. Hence, we consider results from those models merely as suggestive and encourage research that adds to existing studies (Cohen et al., 2001; Font et al., 2022) on the gender-specific consequences of ACEs.

5. Discussion

Prior research has uncovered a robust body of evidence linking childhood maltreatment to multiple adolescent and adult health risks (Felitti et al., 1998; Goncalves Soares et al., 2021; Hughes et al., 2017; Hussey et al., 2006; Jaffee, 2017; Jaffee and Maikovich-Fong, 2011; Jonson-Reid et al., 2012; Steine et al., 2017). We contribute to this line of research by (1) extending analyses into midlife, using the large and nationally representative sample of the Add Health survey and its wide range of self-reported and directly measured health outcomes, and (2) considering the importance of multi-victimization and maltreatment dosage for midlife health risks.

Building on research that has uncovered an association between childhood maltreatment and depression (Jaffee, 2017; Nanni et al., 2012; Negriff, 2020; Norman et al., 2012), our study adds to the evidentiary basis for the claim that such effects can persist deep into adulthood (Arnow, 2004; Jaffee, 2017; O'Shields et al., 2024). We specifically find that the midlife risk of experiencing depression is concentrated among those who experienced sexual abuse and/or multi-victimization. We additionally show that maladaptive coping strategies, such as smoking, also remain prevalent among some former maltreatment victims at midlife—specifically those who experienced physical neglect and/or multi-victimization. Prior research has argued that smoking behavior among formerly maltreated individuals may be a salient behavioral pathway linking childhood misfortune to an elevated risk of midlife heart attacks (Morton et al., 2014). Our analyses do not address this argument directly but hint at its plausibility.

Several prior studies have proposed that the dosage of maltreatment or other ACEs may matter for subsequent health outcomes, sometimes by focusing on recurring episodes across multiple calendar years or developmental periods (Bayly et al., 2022; English et al., 2005; Jaffee and Maikovich-Fong, 2011) and sometimes by aggregating ordinal measures of ACEs into aggregate trauma scores (Steine et al., 2017). But the evidence for such claims remains limited (i.e., a relatively small number of studies using smaller or local samples), and at least one study has found that dosage effects are attenuated once other sources of adversity are taken into account (Jonson-Reid et al., 2012). Adjusting for the co-occurrence of other ACEs, we do not find evidence of linear dose-response relationships between maltreatment and midlife health. However, supplemental analyses that pool maltreatment counts into broader categories find that frequently maltreated respondents still faced greater risks of experiencing depression and obesity, compared to less frequently maltreated respondents. In line with prior studies on multi-victimization (Felitti et al., 1998; Goncalves Soares et al., 2021; Hughes et al., 2017; Miller and Lacey, 2022; Turner et al., 2013), we also find that midlife health risks were particularly common among those who experienced more than one type of maltreatment.

Our analyses do not allow us to identify the mechanisms that generated such vulnerabilities. But our findings—in particular, the combination of detecting health consequences for multi-victimized and frequently maltreated individuals, coupled with null findings about linear dose-response relationships—are consistent with models that emphasize resilience mechanisms or the protective benefits of social support systems among individuals whose maltreatment exposure is more isolated (O'Shields et al., 2024) and with theories that highlight homeostasis in physiological systems until a critical threshold is exceeded and adaptive capacities are overwhelmed (Romero et al., 2009). Alternative explanations include recall bias favoring more severe experiences or other nonlinear biological pathways where multiple hits are required to trigger pathological cascades. We consider studies that explore such nonlinear biological pathways and ascertain the importance of potentially moderating social/environmental contexts as particularly promising.

5.1. Limitations

It is possible that Add Health respondents were still too young at W5 to experience the full cumulative physiological consequences of risky behaviors, given that those consequences often accrue at older ages (Burns, 2003). Our analyses therefore reveal health risks during early midlife but cannot speak to the prevalence of those risks later in the life course. As already stated above, retrospectively self-reported maltreatment data may also be subject to recall bias; and respondents may have been able to recall the fact of maltreatment more accurately than the frequency of maltreatment, especially if such maltreatment occurred very early during childhood. Prior validations of Add Health maltreatment measures (Currie and Tekin, 2012; Hussey et al., 2006; Shin and Miller, 2012)—as well as the fact that our core findings are replicated in supplemental models that only rely on alternative maltreatment measures from W3 (which were collected in closer proximity to childhood than the W4 data used in the analyses shown above)—increase our confidence in these data. However, retrospectively self-reported maltreatment data can sometimes yield stronger associations with mental health outcomes than prospectively recorded maltreatment data (Baldwin et al., 2019, 2024). We cannot rule out this possibility for Add Health data but reiterate that all plausible alternative data sources are beset with considerable problems that complicate their use in large-n or nationally representative studies.

6. Conclusion

We provide robust evidence of the midlife health consequences of maltreatment, which are detectable even after adjusting for demographic factors, health risks due to low birth weight or lack of healthcare access during critical childhood developmental periods, and a wide range of ACEs that have independently been found to matter for adult health outcomes (Cheong et al., 2017; Felitti et al., 1998; Hughes et al., 2017; Miller and Lacey, 2022; Negriff, 2020; Pereira and Zhu, 2025). Specifically, we document multiple health vulnerabilities—most prominently, vulnerabilities due to depression and smoking—among victims of physical neglect, sexual abuse, and multi-victimization. Reducing exposure to maltreatment therefore has the potential to yield durable health benefits across the life course, with implications for both individual well-being and population health.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.ssmmh.2026.100625.

Data statement

This study used restricted-use data from the National Longitudinal Study of Adolescent to Adult Health (Add Health). Researchers can apply for data access through the Carolina Population Center at the University of North Carolina (UNC).