Stress timing, trauma exposure, and family resilience differentially affect internalizing and externalizing symptoms at 3, 5, and 7 years of age

Abstract

Mental health disorders are associated with decreased quality of life, economic productivity loss, and increased mortality. The association between stressful experiences and psychopathology is well documented. However, studies are needed to understand the impact of timing of stressful events, types of traumatic experiences, and of family resilience on internalizing and externalizing symptoms in early childhood. The present study used a longitudinal design towards this end. Parents (N = 456) completed study measures at infancy, 2 years, 3 years, 5 years, and 7 years. At 3 years, greater stressful events during the prenatal period, 1-2 years, and 2-3 years (B = 0.833–0.369, p = 0.028–0.046) predicted internalizing symptoms for female participants only. For externalizing symptoms at 3 years, every time point assessed was significantly associated with more symptoms across both sexes (B = 1.071–0.414, p < 0.001). At 5 years, both internalizing and externalizing symptoms were associated with a greater number of stressful events at every time point and across sexes (B = 1.372–0.465, p < 0.001–0.002). There was evidence for timing effects, including cumulative effects, sensitive periods, and recency effects. Exposure to interpersonal trauma associated with greater internalizing symptoms (B = 2.120, p = 0.002), whereas both interpersonal (B = 1.879, p = 0.005) and non-interpersonal (B = 1.223, p = 0.032) traumatic experiences were associated with greater externalizing symptoms. Aspects of family resilience including higher levels of family commitment, ability to face challenges, and sense of control reduced risk for internalizing symptoms (B = –0.496, p = 0.004) while only greater sense of control (B = –0.838, p = 0.040) reduced risk for externalizing symptoms at age 7 years, including in the context of trauma.

Across the first 7 years of life, greater exposure to stressful events predicted higher internalizing and externalizing symptoms for children. This pattern varied with sex. Higher family resilience in terms of commitment, ability to face challenges, and control provided some protection.

Introduction

Mental health problems have a significant burden globally, causing reduced quality of life, loss to economic productivity, and loss of life. The most recent estimates suggest that 418 million disability-adjusted life years (DALYs) were the result of mental health disorders in 2019, accounting for approximately 16% of the global DALYs, and the economic value lost as a result was estimated to be nearly 5 trillion USD¹. Notably, the global burden of mental health disorders has continued to increase rapidly in the context of growing and emerging stressors (e.g., the COVID-19 pandemic). The diathesis-stress model suggests that the emergence of mental health disorders is a consequence of the interaction between an individual’s in-born vulnerability towards the disorder and the experience of stressful events². Although the onset of disorder is believed to be triggered by stress, the amount or intensity of stress required to trigger a particular disorder may vary according to individual vulnerability².

Existing research has documented a clear link between stressful experiences and psychopathology, beginning in middle childhood (approximately 5 years of age) and continuing through adulthood. This trend is observed across both internalizing and externalizing disorders. For instance, Herbison, et al. ³ reported an association between chronic stress exposure measured longitudinally from the prenatal period to 17 years of age and depression/anxiety symptoms at follow-up in young adulthood (20 years of age). The same findings have also been replicated in adolescents. For example, Slavich, et al.⁴ found that early adversity and interpersonal stress were significantly associated with several mental health outcomes (depression, anxiety, anhedonia severity, general mental and physical health complaints, risky behavior engagement, number of interviewer-based psychiatric diagnoses) in adolescents at 15 years of age. McMullin, et al. ⁵ found links between lifetime stress exposure and addictive behaviors, such as binge eating and alcohol use, in adults. Similarly, Saccaro, et al. ⁶ synthesized existing evidence, concluding that chronic exposure to stress, inflammation, and anxiety contribute to the development of Attention-Deficit/Hyperactivity Disorder (ADHD).

Researchers have proposed a “life course theory” for the association between stress and health outcomes^7–9. One model is the “accumulation of risk model,” which suggests that additional time exposed to stressors is associated with increased risk in a dose-response manner for poorer mental and physical health outcomes. Greater cumulative stress exposure in childhood has been linked to higher risk and more severe outcomes in young adults, regardless of the timing of exposure^7,10,11. Another model, the “sensitive period model,” posits that the timing of exposure is crucial, with stress exposures during windows of vulnerability like the prenatal period and early childhood having the greatest impact on neurodevelopmental, endocrine, immune, and metabolic systems (i.e., biological embedding)^7,12–15. Finally, the “recency” model suggests that the development of psychopathology is linked most robustly to proximal (recent) stress exposures, as opposed to distal or accumulated exposures^7,16.

Existing studies have found inconsistent support for all three life course models (cumulative effects, sensitive periods/biological embedding, and recent/proximal effects of stress) for psychopathology^7,17. Thus, studies with repeated measure designs that can investigate the impact of timing of stressful experiences on the development of psychopathology are needed to examine evidence for all three models jointly. Internalizing and externalizing problems should be examined separately, as the timing of stress exposures could have differential impact on these domains of mental health. Existing work suggests that internalizing symptoms are characterized by enhanced cortisol output in the context of acute stressors in artificial settings (i.e., in a laboratory stress task) while externalizing symptoms are characterized by reduced cortisol output^18,19. The differential effects of ongoing stress exposure and timing in naturalistic settings continues to be a gap in the research.

Most existing research has focused on middle childhood, adolescence, and adulthood, with relatively limited longitudinal research examining outcomes in early childhood. For instance, a recent meta-analysis (N = 62) on the association between early life stress and depression in childhood and adolescence ( < 18 years) only included four longitudinal studies, six studies of children 7–10 years (of which only one was longitudinal), and no studies of children under 7 years²⁰. Therefore, identifying sensitive periods and the impact of cumulative stress within the first five years of life is important for understanding the impact of stress at earlier developmental stages. A more nuanced understanding of the effects of stress exposure timing over the first five years of life may facilitate more targeted interventions earlier in development, potentially improving mental health outcomes for young people exposed to stressors early in life. Research investigating psychopathology in children and adolescents (5 to 15 years of age) suggests that, once symptoms develop, they tend to be stable and do not spontaneously remit, which highlights the profound and long-lasting effects that stress can have on lifetime mental health²¹.

An additional limitation of the existing literature is the conflation of stress, adversity, and trauma, as it has been difficult to establish clear conceptual boundaries between the three²². Existing conceptualizations suggest that normative stress and adversity may function on one dimensional continuum, whereas trauma may function categorically ²². That is, stress may range from mild (e.g., relocating residences by choice) to more severe experiences (e.g., experiencing homelessness)²³. Trauma is defined more narrowly as events that would meet diagnostic criteria in the Diagnostic and Statistical Manual of Mental Disorders (DSM), such as: “death, threatened death, actual or threatened serious injury, or actual or threatened sexual violence”²⁴. However, others suggest that they function on the same dimensional continuum, and that trauma should be conceptualized by the subjective experience of an event, rather than a more narrowly defined set of experiences²³. Newer models advocate for hybrid conceptualizations, whereby experiences of stress are functioning on a continuum of severity, and in combination with one’s self-regulatory capacity, an individual can experience normative stress, pathogenic stress, or traumatic stress.

Given the conflated boundaries across these constructs, examining both stressful events and traumatic experiences is important to determine whether they may have the same or distinct effects on outcomes. The nature of traumatic experiences is also important to consider. Existing research has found that traumatic experiences with an interpersonal component are associated with an increased risk for meeting diagnostic criteria for Post-Traumatic Stress Disorder (PTSD) in adolescents aged 13 to 17 years^25,26. However, more research is needed to better understand the role of non-interpersonal traumatic experiences, to examine their contributions to both internalizing symptoms (beyond the single diagnostic category of PTSD) and externalizing symptoms, and to study this link in childhood. Such findings may be useful to identify the optimal timing for screening, prevention, and intervention efforts. For the purposes of the current study, stress and trauma exposures were conceptualized as total counts and along two dimensions (interpersonal vs. non-interpersonal trauma)¹⁰, given that past work has found inconsistent findings across types of maltreatment (e.g., physical versus sexual abuse) or other dimensions (i.e., threat versus deprivation)²⁷. Although the precise sensitive periods identified in the current literature for type of stress exposure vary, sensitive period effects nonetheless are observed and may need to be disentangled in future work²⁷.

Relatedly, researchers have started to explore factors that may confer resilience for child and adolescent psychopathology symptoms, particularly in the context of stress exposures, often with the goal of developing interventions that may be able to promote these factors^28,29. Vulnerability processes (such as those proposed by the diathesis-stress model) are believed to capture only one dimension of developmental psychopathology. Understanding developmental processes involved in resilience may further inform prevention and treatment approaches³⁰. Resilience has also been conceptualized using a multisystem framework encompassing individual, family, school, community, and organizational level factors²⁸. A recent review of the literature showed that longitudinal research on this topic is limited, but the evidence that exists has found that higher levels of resilience are related to fewer mental health problems²⁸. Much of the existing research has focused on pre-adolescence and adolescence (only two studies reviewed included children under a mean age of 12 years), and often do not consider the joint roles of adversity and resilience factors.

The current study aims to address some of these gaps by 1) elucidating the periods of stress exposure, from the prenatal period through age 3 years (i.e., prenatally, birth-1 year, 1-2 years, and 2-3 years) that most robustly contribute to the development of internalizing and externalizing symptoms in children by age 3 years; 2) determining the periods of stress exposure from the prenatal period to age 5 years that most robustly contribute to internalizing and externalizing symptoms in children by age 5 years; 3) examining whether different types of traumatic experiences (i.e., interpersonal vs. non-interpersonal trauma) over the first 7 years of life are associated with internalizing and externalizing symptoms at age 7 years; and 4) testing whether family resilience and adaptation characteristics protect against the development of child internalizing and externalizing symptoms by age 7 years, both independently and in the context of trauma exposures.

Methods

Participants

Participants were recruited from a registry of local births in the Greater Boston area comprising families who had indicated willingness to participate in developmental research. Families in the current analyses are part of a longitudinal study originally designed to examine the early development of emotion processing. Exclusion criteria in the parent study included known prenatal or perinatal complications, maternal use of medications during pregnancy that may have significant impact on fetal brain development (i.e., anticonvulsants, antipsychotics, opioids), pre- or post-term birth ( ± 3 weeks from due date), developmental delay, uncorrected vision difficulties, and neurological disorder or trauma. After enrollment, families were no longer followed, and their data were excluded from analyses if the child was diagnosed with an autism spectrum disorder, or a genetic or other condition known to influence neurodevelopment (n = 29). By design, families were enrolled in the parent study when the children were 5, 7, or 12 months old (infancy), with a smaller subsample to be followed when the children were 2 years, 3 years, 5 years, and 7 years of age. The infancy visits started in April 2013 and ended in April 2017. Data collection for the 7-year visits continued through September 2023. The current study is not pre-registered.

To be included in the current analyses, participants needed to have parent-reported child psychopathology data at one of the time points assessed: at 3 years (n = 456 for internalizing scores; n = 441 for externalizing scores), 5 years (n = 435 for both internalizing and externalizing scores), or 7 years of age (n = 388 for both internalizing and externalizing scores). A total of 246 participants had internalizing symptoms reported across all three time points, and 235 participants had externalizing symptoms reported across all three time points. A measure of family resilience and adaptation was added to the protocol of the current longitudinal study during data collection (mean completion date April 2022). Participants who had not already participated in the 7-year visit when the measure was added to the protocol (n = 107) provided data for this measure, which was used for the fourth aim in the current paper. Study procedures were approved by the Institutional Review Board at Boston Children’s Hospital, and the child’s parent or legal guardian provided written informed consent prior to the initiation of study activities.

Measures

Questionnaires, described below, were administered via REDCap, an online platform, to the child’s parent, primarily the child’s mother (97.8%). All measures were self-completed by parents.

Sociodemographics

Sociodemographic information was collected via parent-report at infancy to characterize the sample. This included: child age at each visit, sex assigned at birth (hereafter “sex”; male/female), ethnicity (non-Hispanic/Latinx, Mexican, Puerto Rican, Cuban, Other Hispanic/Latinx, or Mixed Hispanic/Latinx), race (White, Black or African American, American Indian or Alaska Native, Asian Indian, East Asian, Pacific Islander, Mixed Race), parent education (8th grade or less, some high school, high school/GED, associate’s degree, bachelor’s degree, master’s degree, doctoral degree), parent marital status (single, married, cohabitating, divorced), and annual household income.

Stressful events

Stressful life events were measured at each assessment time point using a 30-item version of the Recent Life Events Questionnaire (RLEQ)³¹. The RLEQ was developed to identify common life events that a high proportion of respondents report as having marked or moderate long-term threat (as opposed to mild or no long-term threat). The RLEQ includes items pertaining to serious injury or illness, death of relative or friends, separation (relationship or marriage), serious problems with close friends or relatives, serious abuse/attack/threats, unemployment or job loss, financial crisis, problems with police or court appearance, burglary/mugging, miscarriage/stillbirth, moving homes (through choice or not), and housing difficulties. The parent was asked to rate whether they had experienced the event (yes/no) in the time period of interest. A sum of the endorsed events, i.e., the number of stressful experiences reported over the time period assessed was calculated. Stressful experiences were calculated from pregnancy to birth and for each subsequent year of the child’s life (0-1 year, 1-2 years, 2-3 years, 3-4 years, and 4-5 years). In validation studies, the RLEQ has shown high levels of concurrent validity (SN = 1.0, SP = 0.88) in capturing stressful events in the past three months relative to more extensive interview data, and high inter-rater agreement across informants (k = 0.88)³².

Traumatic events

Potentially traumatic events experienced by the child were measured via the Traumatic Events Screening Inventory (TESI)³³, a retrospective structured clinical interview completed by the child’s parent at the 7-year visit. The TESI measures exposure to 24 potentially traumatic events that can be categorized into non-interpersonal and interpersonal events. For example, some of the non-interpersonal events assessed included accidents, illnesses, and disasters. Some of the interpersonal events assessed included abuse, neglect, and violence. Sum scores for interpersonal and non-interpersonal events are generated by adding the number of endorsed events within those respective categories. Total sum scores can also be generated by adding the number of endorsed events across both interpersonal and non-interpersonal categories. Validation studies with both parent and child report suggest that the TESI had mean inter-rater agreements of 97% (k = 0.88–1), and provide evidence of both convergent validity with other trauma measures and predictive validity for PTSD more broadly³⁴.

Family resilience and adaptation

Family resilience and adaptation was measured using the Family Hardiness Index (FHI) at the 7-year visit. The FHI is a 20-item parent-report scale that measures the extent to which each of the provided statements describes their family on a 4-point scale (0=False, 1=Mostly False, 2=Mostly True, 3=True)³⁵. For example, statements provided include: “we strive together and help each other no matter what”, “being active and learning new things are encouraged”, “we realize our lives are controlled by accidents and bad luck (reverse coded)”. Total scores were obtained by calculating the sum of all items. In addition, three subscales can be calculated. Internal consistency coefficients are provided for the current sample. The commitment subscale (α = 0.70) captures a family’s sense of internal strengths; the challenge subscale (α = 0.45) captures a family’s efforts to be innovative, active, to experience new things, and to learn; the control subscale (α = 0.71) captures a family’s sense of being in control of their family life as opposed to being shaped by outside events and circumstances.

Child Internalizing and externalizing symptoms

The Infant-Toddler Social and Emotional Assessment (ITSEA) was administered at 3 years of age to assess for child internalizing and externalizing symptoms³⁶. The ITSEA is a parent-report questionnaire that assesses socioemotional problems in early childhood. Internal consistency coefficients are provided for the current sample. The ITSEA provides composite scores for internalizing symptoms (α = 0.84) based on the following scales: depression/withdrawal, general anxiety, separation distress, inhibition to novelty. The ITSEA also provides composite scores for externalizing symptoms (α = 0.81) based on the following scales: activity/impulsivity, aggression/defiance, peer aggression. The respondent rates individual items on a 3-point scale (0 indicating “Not True/Rarely,” 1 indicating “Somewhat True/Sometimes,” 2 indicating “Very True/Often”) that are then summed to calculate raw scores. T-scores are calculated from norms based on age and sex and have a mean of 50 and a standard deviation of 10.

The Child Behavior Checklist (CBCL) for ages 1.5 to 5 years was administered at age 5 years, and the CBCL for ages 6 to 18 years was administered at age 7 years to measure child psychopathology symptoms^37–40. The CBCL 1.5–5 comprises 99 items and the CBCL 6–18 comprises 113 items³⁸. Internal consistency coefficients are provided for the current sample. The CBCL can generate composite scores for internalizing symptoms (α = 0.82 at both 5 and 7 years) based on the following scales: anxiety, depression, withdrawal, and somatic complaints⁴¹. The CBCL also provides composite scores for externalizing symptoms (5 years α = 0.90; 7 years α = 0.89) based on the aggressive and rule-breaking behavior scales⁴¹. The respondent rates how true each item was for their child in the prior 6 months (0 = “Not True,” 1 = “Sometimes/Somewhat True,” 2 = “Very True/Often True”). These ratings are used to calculate raw scores, which are then normed based on age and sex. T-scores on the CBCL have a mean of 50 and a standard deviation of 10.

Statistical analyses

Data analyses were conducted using IBM SPSS Statistics for Macintosh Version 29 (IBM Corp, Armonk, NY, USA) and Python Version 3.12.3. Descriptive statistics for sociodemographic variables were calculated to characterize the sample. The main study variables were evaluated for normality and multicollinearity using univariate procedures and tolerance statistics. For all analyses, a p-value of less than 0.05 at the two-tailed level was considered statistically significant. Bivariate analyses were conducted to identify potential covariates that should be included in models to limit confounding. If any covariates were significantly associated with the outcome variable, they were included in the main analyses.

To test Aims 1 and 2, focused on determining the time points of stress exposures most strongly associated with internalizing and externalizing symptoms, Linear Mixed Models (LMMs) with maximum likelihood (ML) estimation were used⁴². This method allows for modeling of individual changes over time by creating two-level hierarchical models that are able to nest time within individuals^43,44. LMMs can account for observations that tend to be correlated (e.g., because the observations are the result of repeated measures)⁴⁵. Data were prepared in a stacked format, whereby each wave of data was “stacked” vertically, and a time variable was created to differentiate between time points. LMMs for internalizing symptoms at 3 and 5 years of age accounted for sex as a fixed effect, given observed associations at the bivariate level between sex and internalizing symptoms at those time points. LMMs included fixed effects for stressful events in long form from pregnancy to the outcome time point assessed (i.e., 3 years or 5 years of age). Finally, all models also included a time variable (e.g., prenatal, birth to 1 year) to assess whether certain time points of stress exposure are more robustly associated with psychopathology symptoms.

To test Aim 3, focused on the contribution of interpersonal vs. non-interpersonal traumatic experiences on child psychopathology, an OLS regression model was first built. Within this model, collinearity statistics (tolerance and variance inflation factor) were calculated to assess for correlation between TESI categories that could impact the model findings. Multicollinearity can occur when two or more of the variables input in OLS as regressors have overlapping or redundant information⁴⁶. The linear dependencies between regressors can then affect the regression coefficients and cause larger variances in parameter estimates, making parameters less interpretable as multicollinearity increases⁴⁶. Typically, values of VIF ( = 1/tolerance) exceeding 10 are believed to indicate severe multicollinearity; near 2.5 may still be a cause for concern and indicate moderate levels of collinearity. The closer values are to a VIF of 1, the less concern there is for multicollinearity⁴⁷. Given these general heuristics from Monte Carlo simulations, the current analyses utilized 2.5 as a hard cutoff for collinearity tolerance in OLS models, and interpreted values closer to 1 as less affected by multicollinearity.

A supplemental analysis for this aim was conducted using a Ridge Regression (RR) model to attain additional parameter estimates that better account for multicollinearity⁴⁸. RR is a modified version of the least squares method that relies on shrinkage, also referred to as regularization, to bias coefficient estimates towards 0 and reduce the variance caused by multicollinearity⁴⁶. RR estimates have negligible increases in bias and are believed to better reflect true parameter values, improving model accuracy⁴⁶. However, a limitation of RR models is that, although they enhance the overall performance and reliability of linear models, they can limit the interpretability of individual coefficients. RR models do not allow for variable selection (e.g., through significance and hypothesis testing), but do allow for estimations of regression coefficients. Consequently, results from both OLS and RR models were run for this aim to enhance both the interpretability of findings and model accuracy by limiting the effects of any potential collinearity.

Aim 4 sought to examine whether family resilience and adaptation may protect against the development of internalizing and externalizing symptoms at age 7 years, both independently and in the presence of traumatic experiences. Given moderate to high levels of collinearity (r = 0.252, p = 0.004; r = 0.415, p < 0.001; r = 424, p = 0.004) at the bivariate level between composite scores across the three categories of family resilience and hardiness (commitment, challenge, control), RR models were used to identify which subscales of the FHI may be most salient for internalizing vs. externalizing symptoms. Then, OLS linear regression models were built to identify potential protective effects of family resilience and adaptation on the association between traumatic experiences and psychopathology at age 7 years. Scores were used, informed by prior RR models, to build OLS linear regression models for both internalizing and externalizing symptoms at age 7 years.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

Preliminary analysis

Descriptive statistics for the sample’s sociodemographic characteristics are presented in Table 1. Descriptive statistics for the outcome variables are reported in Table 2 and depicted in Fig. 1. Bivariate analyses of the main study variables with all sociodemographic variables (child sex, child race, child ethnicity, parent education, annual household income) were conducted to identify potential covariates. We found statistically significant evidence for only one of the sociodemographic variables tested, child sex, which was associated with psychopathology symptoms, specifically greater internalizing symptoms at 3 years (r = 0.152; 95% CI = 0.061, 241; p = 0.001) and at 5 years (r = 0.118; 95% CI = 0.024, 209; p = 0.014) among female participants. Thus, child sex was included as a covariate in models with internalizing symptoms at 3 or 5 years as the outcome variable.

Table 1.

Sample Demographics Collected at Infancy (n = 456)

Variables	n	%
Child sex
Male	242	53.07
Female	214	46.93
Child race
White	364	79.82
Black/African American	5	1.10
Asian American or Pacific Islander	12	2.63
Mixed race	69	15.13
Not reported	6	1.32
Child ethnicity
Non-Latino/a, Spanish, or Hispanic	410	89.91
Mexican	10	2.19
Puerto Rican	5	1.10
Cuban	1	0.22
Other Latino/a, Spanish, or Hispanic	22	4.82
Mixed Latino/a, Spanish, or Hispanic	3	0.66
Not reported	5	1.10
Maternal educational attainment
Less than high school	0	0
High school/GED	16	3.51
Associate degree	4	0.88
Bachelor’s degree	128	28.07
Master’s degree	216	47.37
M.D., Ph.D., J.D., or equivalent	90	19.74
Not reported	2	0.44
Paternal educational attainment
Less than high school	0	0
High school/GED	36	7.89
Associate degree	21	4.61
Bachelor’s degree	135	29.61
Master’s degree	142	31.14
M.D., Ph.D., J.D., or equivalent	118	25.88
Not reported	4	0.88
Parental marital status
Single	6	1.3
Married	434	95.2
Cohabitating	11	2.4
Divorced	1	0.2
Not reported	4	0.9
Annual family household income
Less than $35,000	12	2.63
$35,000-$49,000	12	2.63
$50,000-$74,999	44	9.65
$75,000-$99,000	59	12.94
$100,000 or more	294	64.47
Not reported	35	7.68

Table 2.

Descriptive Statistics for Internalizing and Externalizing Symptom Scores

Variable	M (SD)
Infant-Toddler Social and Emotional Assessment (ITSEA) T-scores (at 3 years)
Internalizing	48.98 (9.51)
Externalizing	47.81 (7.43)
Child Behavior Checklist (CBCL 1½-5) T-scores (at 5 years)
Internalizing	46.30 (9.74)
Externalizing	43.24 (9.51)
Child Behavior Checklist (CBCL 6-18) T-scores (at 7 years)
Internalizing	48.39 (9.71)
Externalizing	47.89 (9.64)

Fig. 1. Internalizing and externalizing symptoms over time.

Dotted line represents the reference population mean score for the ITSEA and CBCL. Sample sizes were n = 456 for internalizing scores and n = 441 for externalizing scores at 3 years, n = 435 for both internalizing and externalizing scores at 5 years, and n = 388 for both internalizing and externalizing scores at 7 years.

Aims 1 and 2: Stressful Events and Psychopathology Symptoms at ages 3 and 5 years

Results from the LMMs models for stressful life events and internalizing/externalizing symptoms are shown in Table 3 for the 3-year time point. For stressful events from the prenatal period to 3 years of age, there was a significant time x sex interaction. Specifically, stressful events were not associated with internalizing symptoms at 3 years of age for male participants at any time point assessed (refer to Table 3). However, for female participants, a greater number of stressful events during the prenatal period (B = 0.833; 95% CI = 0.092, 1.574; p = 0.028), 1 to 2 years (B = 0.478; 95% CI = 0.008, 948; p = 0.046), and 2 to 3 years (B = 0.369; 95% CI = 0.022, 715; p = 0.037) were associated with more internalizing symptoms at 3 years. A greater number of stressful events was associated with more externalizing symptoms at every time point (prenatal through 3 years of age) for both sexes.

Table 3.

Linear mixed model (LMM) for Stressful events & psychopathology symptoms at 3 years of age

Dependent variable: Internalizing symptoms (ITSEA) at 3 years (nested by sex)
	Estimate	Std. Error	t	Sig.	95% Confidence Interval
					Lower Bound	Upper Bound
Intercept	48.650	0.332	146.407	<0.001	47.999	49.302
Male - Stressful events (prenatal)	−0.310	0.476	−0.652	0.515	−1.244	0.624
Male - Stressful events (birth to 1 year)	−0.301	0.340	−0.887	0.375	−0.968	0.365
Male - Stressful events (1 to 2 years)	−0.099	0.248	−0.400	0.689	−0.585	0.387
Male - Stressful events (2 to 3 years)	−0.050	0.184	−0.274	0.784	−0.412	0.311
Female - Stressful events (prenatal)	0.833	0.378	20.205	0.028	0.092	10.574
Female - Stressful events (birth to 1 year)	0.530	0.351	1.509	0.132	−0.159	1.220
Female - Stressful events (1 to 2 years)	0.478	0.240	1.993	0.046	0.008	0.948
Female - Stressful events (2 to 3 years)	0.369	0.177	2.088	0.037	0.022	0.715

Dependent variable: Externalizing symptoms (ITSEA) at 3 years (both sexes)
	Estimate	Std. Error	t	Sig.	95% Confidence Interval
					Lower Bound	Upper Bound
Intercept	46.418	0.255	182.055	<0.001	45.918	46.918
Stressful events (prenatal)	1.071	0.236	4.545	<0.001	0.609	1.534
Stressful events (birth to 1 year)	0.866	0.194	4.463	<0.001	0.485	1.246
Stressful events (1 to 2 years)	0.595	0.139	4.281	<0.001	0.323	0.868
Stressful events (2 to 3 years)	0.414	0.104	3.965	<0.001	0.209	0.619

Bold formatting indicates statistically significant findings (two-sided p-values < 0.05).

Results from the LMMs for stressful life events and internalizing/externalizing symptoms at the 5-year time point are shown in Table 4. A greater number of stressful events was associated with more internalizing symptoms at every time point (prenatal through 5 years of age). No interaction effect of sex on the association between stressful events and internalizing symptoms was observed (refer to Table 4). A greater number of stressful events was associated with more externalizing symptoms at 5 years for every time point (prenatal through 5 years of age).

Table 4.

Linear mixed model (LMM) for stressful events & psychopathology symptoms at 5 years of age

Dependent variable: Internalizing symptoms (CBCL 1½-5) at 5 years (with sex as a main effect)
	Estimate	Std. Error	t	Sig.	95% Confidence Interval
					Lower Bound	Upper Bound
Intercept	46.104	0.340	135.445	<0.001	45.437	46.772
Sex (male)	–2.265	0.380	–5.958	<0.001	−3.011	−1.520
Stressful events (prenatal)	1.343	0.297	4.524	<0.001	0.761	1.925
Stressful events (birth to 1 year)	0.872	0.256	3.408	<0.001	0.370	1.373
Stressful events (1 to 2 years)	0.540	0.172	3.150	0.002	0.204	0.877
Stressful events (2 to 3 years)	0.482	0.143	3.362	<0.001	0.201	0.763
Stressful events (3 to 4 years)	0.916	0.203	4.518	<0.001	0.518	1.313
Stressful events (4 to 5 years)	0.941	0.172	5.484	<0.001	0.605	1.278

Dependent variable: Externalizing symptoms (CBCL 1½-5) at 5 years (both sexes)
	Estimate	Std. Error	t	Sig.	95% Confidence Interval
					Lower Bound	Upper Bound
Intercept	41.734	0.265	157.407	<0.001	41.214	42.254
Stressful events (prenatal)	1.372	0.294	4.672	<0.001	0.796	1.947
Stressful events (birth to 1 year)	1.028	0.253	4.062	<0.001	0.532	1.524
Stressful events (1 to 2 years)	0.653	0.170	3.847	<0.001	0.320	0.985
Stressful events (2 to 3 years)	0.465	0.142	3.280	0.001	0.187	0.743
Stressful events (3 to 4 years)	0.866	0.200	4.321	<0.001	0.473	1.259
Stressful events (4 to 5 years)	0.990	0.170	5.831	<0.001	0.657	1.323

Bold formatting indicates statistically significant findings (two-sided p-values < 0.05).

Aim 3: Traumatic experiences (from Birth to 7 years) and psychopathology symptoms at 7 years

An OLS linear regression model was used to determine the relative contributions of traumatic experience categories (interpersonal, non-interpersonal) on internalizing and externalizing symptoms at age 7 years (Table 5). Collinearity statistics (tolerance and variance inflation factor) were run for this model, given that both traumatic event categories were collected using the same measure and thus may be correlated, affecting the regression model’s findings. Tolerance statistics were 940, and VIF statistics were 1.064. Given that all VIFs were well below 2.5, variables in the current linear regression model likely did not have significant multicollinearity. Supplemental RR models run for both outcomes at this time point are presented in S1.

Table 5.

Ordinary least squares (ols) linear regression models for traumatic experiences (total events from birth to 7 years) & psychopathology symptoms at 7 years of age

Dependent variable: Internalizing Symptoms (CBCL) at 7 years
	Unstandardized Coefficients		Standardized Coefficients			Collinearity Statistics
	B	Std. Error	Beta	t	Sig.	Tolerance	VIF
(Constant)	47.294	0.712		66.404	<0.001
Traumatic experiences (interpersonal)	2.120	0.675	0.163	3.143	0.002	0.940	1.064
Traumatic experiences (non-interpersonal)	0.248	0.577	0.022	0.429	0.668	0.940	1.064

Dependent variable: Externalizing Symptoms (CBCL) at 7 years
	Unstandardized Coefficients		Standardized Coefficients			Collinearity Statistics
	B	Std. Error	Beta	t	Sig.	Tolerance	VIF
(Constant)	46.025	0.703		65.508	<0.001
Traumatic experiences (interpersonal)	1.879	0.665	0.145	2.824	0.005	0.940	1.064
Traumatic experiences (non-interpersonal)	1.223	0.569	0.111	2.150	0.032	0.940	1.064

Bold formatting indicates statistically significant findings (two-sided p-values < 0.05).

In the OLS linear regression model for internalizing symptoms at age 7 years, we found statistically significant evidence for traumatic experiences related to interpersonal events being associated with internalizing symptoms (β = 0.163, B = 2.120, SE = 0.675, p = 0.002) (refer to Table 5 for non-interpersonal events and internalizing symptoms at 7 years). Both traumatic experiences related to interpersonal events (β = 0.145; B = 1.879; 95% CI = 1.214, 2.544; p = 0.005) and non-interpersonal events (β = 0.111, B = 1.223; 95% CI = 0.654, 1.792; p = 0.032) were associated with externalizing symptoms. RR models also suggest a greater contribution of traumatic experiences related to interpersonal events relative to non-interpersonal events for internalizing symptoms at 7 years, and approximately equal contribution of trauma types to externalizing symptoms at 7 years.

Aim 4: Lifetime trauma exposures, family resilience/adaptation, and psychopathology symptoms at 7 years

RR models were run to examine the associations between family resilience/adaptation and psychopathology symptoms at age 7 years (S2). Each category of family resilience and adaptation, i.e., adaptation (B = –1.566), challenge (B = –1.137), and control (B = –2.089), was independently associated with internalizing symptoms. Higher levels of family commitment (family’s sense of internal strengths), challenge (family’s efforts to be innovative, active, to experience new things and to learn), and control (family’s sense of being in control of their family life as opposed to being shaped by outside events and circumstances) were associated with lower levels of internalizing symptoms. We found statistically significant evidence for only the control category being associated with externalizing symptoms (B = –1.336) (refer to S2 for other categories of family resilience).

Findings from RR models (in this Aim and Aim 3) were used to build OLS regression models to determine the potential protective effects of family resilience and adaptation (factor score or subscale) on the link between traumatic experiences (factor score or subscale) and psychopathology symptoms (Table 6). There was no longer evidence for a significant association between interpersonal traumatic experiences and internalizing symptoms at age 7 years once models accounted for the effects of total family resilience and adaptation (β = –0.295, B = –0.496, SE = 0.169, p = 0.004). Higher levels of family resilience and adaptation across all categories continued to be associated with lower levels of internalizing symptoms at age 7 years (Fig. 2). No interaction effect was observed between interpersonal traumatic experiences and total family resilience on internalizing symptoms. Similarly, we did not find evidence for a statistically significant association between traumatic experiences (total) and externalizing symptoms at age 7 years once models accounted for the effects of family resilience and adaptation (control category; β = –0.202, B = –0.838, SE = 0.403, p = 0.040) (Fig. 2). Families reporting a greater sense of being in control of their family life was associated with fewer externalizing symptoms in children, even after accounting for the effects of traumatic experiences. No interaction effect was observed between total traumatic experiences and family resilience (control category) on externalizing symptoms.

Table 6.

Cross-sectional OLS linear regression models for traumatic experiences, family resilience/adaptation, and child psychopathology symptoms at 7 years of age

Dependent variable: Internalizing Symptoms (CBCL) at 7 years
	Unstandardized coefficients		Standardized coefficients			Collinearity statistics
	B	Std. Error	Beta	t	Sig.	Tolerance	VIF
(Constant)	72.548	8.263		8.780	<0.001
Traumatic experiences (interpersonal)	1.368	1.254	0.106	1.091	0.278	0.945	1.058
Family resilience (total)	–0.496	0.169	–0.295	–2.940	0.004	0.877	1.141
Interaction term	0.252	0.284	0.087	0.885	0.378	0.923	1.083

Dependent variable: Externalizing Symptoms (CBCL) at 7 years
	Unstandardized coefficients		Standardized Coefficients			Collinearity statistics
	B	Std. Error	Beta	t	Sig.	Tolerance	VIF
(Constant)	47.949	0.936		51.217	<0.001
Traumatic experiences (total)	1.242	0.729	0.166	1.704	0.092	0.963	1.038
Family resilience (control)	–0.838	0.403	–0.202	–2.081	0.040	0.966	1.035
Interaction term	0.115	0.382	0.029	0.302	0.763	0.997	1.003

Bold fonts indicate statistically significant findings (two-sided p-values < 0.05).

Fig. 2. Relationship between family resilience and internalizing/externalizing symptoms at 7 years.

Dashed lines represent 95% confidence intervals for the observed trend. The sample size at the 7-year time point was n = 107.

Discussion

In the current study, we aimed to investigate periods of adversity exposure that may be most robustly associated with child psychopathology symptoms in early childhood. We further sought to examine whether different types of lifetime traumatic experiences may be differentially associated with internalizing and externalizing symptoms in middle childhood. Finally, we sought to determine whether family resilience and adaptation may protect against the development of internalizing and externalizing symptoms, both independently and when accounting for traumatic experiences.

Results showed a significant sex effect when examining internalizing symptoms at age 3 years in relation to stress exposures. Stressful events were associated with more internalizing symptoms in female participants but not in male participants, suggesting that the impact of stressful events on internalizing symptoms may differ by sex in early childhood. By age 5 years, stress exposures from the prenatal period through age 5 years was associated with greater internalizing symptoms across sexes. Some systematic review work has found evidence for sex-specific programming of the hypothalamic-pituitary-adrenal (HPA) axis⁴⁹. Specifically, the placenta in pregnancies of female fetuses may have increased permeability to maternal glucocorticoids following stress exposure due to differences in enzymatic activity, with evidence for continued increased reactivity and sensitivity in female offspring⁴⁹. As such, the amount and chronicity of accumulated stress exposure needed to produce risk for internalizing symptoms may be greater in male participants than in female participants. Greater externalizing symptoms at ages 3 and 5 years were associated with greater exposure to stressful events at every time point from the prenatal period through age 5 years. These findings expand on existing work in older children and young adults, which reports that stressful life events function as a transdiagnostic environmental exposure for psychopathology in youth (10–24 years of age), across both internalizing and externalizing domains^50,51.

From the perspective of life course theory, the results provide support for multiple models. The “accumulation of risk model” suggests that additional time exposed to stressors is associated with increased risk for psychopathology symptoms^10,52. The current results provide evidence for accumulation of risk in two ways. First, although exposure to stressful events from the prenatal period to age 3 years did not increase risk for internalizing symptoms at age 3 years for male participants, by the 5-year time point, male participants’ accumulated stressors (from the prenatal period to age 5 years) were associated with greater internalizing symptoms. Second, greater exposure to stressful events during every interval assessed was associated with greater internalizing symptoms for female participants at age 3 years and for the full sample at age 5 years and with externalizing symptoms for the full sample at both 3 and 5 years.

The “sensitive period model” suggests that the timing of exposure is most crucial, with stress exposures during windows of vulnerability (e.g., the prenatal period and early childhood) having the greatest impact^7,12–14. In the current study, psychopathology symptoms at ages 3 and 5 years were associated with stressful events at most time points assessed from the prenatal period through 5 years of age, which could be interpreted as evidence against the sensitive period model. However, the effect sizes for the association between stressful events and psychopathology symptoms were strongest in the prenatal period and steadily declined with age through 3 years. This pattern of effects suggests that there may be greater sensitivity to stress exposure during the prenatal period.

The “recency” model suggests that psychopathology is most closely tied to proximal (recent) stress exposures, as opposed to distal exposures during sensitive periods or accumulated exposures^7,16. Despite stressful events being consistently associated with psychopathology symptoms regardless of their timing, the effect of stressful events on symptoms at 5 years was greater at the 3-4 year and 4-5 year time points than earlier postnatal time points (0–3 years), although not as strong as effect sizes for prenatal exposures. This may suggest that as children age, postnatal exposures to stress that are occurring at more proximal time points may be more related to psychopathology symptoms, although prenatal exposures continue to have the largest effect.

The current study also identified differential effects of interpersonal vs. non-interpersonal lifetime traumatic experiences on internalizing and externalizing symptoms at 7 years. Interpersonal traumatic experiences (e.g., abuse, neglect, violence) but not non-interpersonal traumatic experiences (e.g., accidents, illnesses, disasters) were associated with internalizing symptoms. Both interpersonal and non-interpersonal traumatic experiences were associated with externalizing symptoms. Prior work investigating traumatic experiences and psychopathology has not consistently found evidence for a specific stressors (e.g., abuse) being linked with internalizing/externalizing outcomes (i.e., specificity), and instead report evidence for equifinality (i.e., many pathways leading to the same outcome) and multifinality (i.e., a shared risk factor leading to various outcomes)⁵³.

The current study’s findings on traumatic experiences and psychopathology symptoms provide evidence for both equifinality and multifinality, as well as some degree of specificity. In terms of equifinality, there was an observed association between both interpersonal and non-interpersonal traumatic experiences and externalizing symptoms. In terms of multifinality, interpersonal traumatic experiences were associated with both internalizing and externalizing symptoms. Finally, for internalizing symptoms there appeared to be some degree of specificity, as traumatic experiences that were interpersonal in nature were associated with internalizing symptoms, but those that were non-interpersonal in nature were not.

These findings further suggest that traumatic experiences may influence symptom development beginning in early childhood. Internalizing disorders (e.g., depressive disorders, anxiety disorders, obsessive-compulsive and related disorders, trauma and stressor-related disorders, and dissociative disorders) tend to reflect an individual’s internal expression of distress^54–56. Interpersonal traumatic experiences are characterized by victimization, or harm resulting from other human actors, which may increase risk for internalizing distress more than those experiences that are non-interpersonal and do not violate elements of beneficence, trust, justice, and morality in the same way. Externalizing disorders (e.g., disruptive, impulse control, and conduct disorders, and the substance-related and addictive disorders) are often associated with overt behaviors that may bring the individual into conflict with others^54–56. The experience of both types of traumatic experiences (interpersonal and non-interpersonal) may increase the likelihood of traumatic experiences being expressed in behaviors that increase external conflict.

The final aim of the current study was to examine the potential protective effects of family resilience and adaptation against psychopathology symptoms in childhood, both independently and in the context of traumatic experiences. Masten, et al. ³⁰ provided a comprehensive review of existing frameworks for understanding resilience, which was used to inform the current work. First, Masten, et al. ³⁰ asserted that the study of resilience relies on two essential components: 1) identifying the risk or threat to the individual and 2) elucidating adaptation and adaptive functioning in the context of risk. Second, Masten, et al. ³⁰ asserted that resilience is a multisystem process that lies both within the person (individual traits) and outside of the person (e.g., family, school, community, and organizational).

The current findings suggest that certain components of family resilience and adaptation were differentially related to internalizing and externalizing symptoms at age 7 years. All aspects related to family resilience and adaptation (i.e., the family’s sense of internal strengths, adaptability in the face of challenge, and sense of control over circumstances shaped by external factors) appeared to reduce risk for the development of internalizing symptoms, whereas only sense of control reduced risk for externalizing symptoms. A greater sense of control within the family specifically may be associated with a decreased likelihood of a child engaging in overt behaviors that would create conflict with others (i.e., externalizing behaviors). Importantly, in models accounting for family resilience, there was no longer evidence for a statistically significant association between traumatic experiences and psychopathology outcomes. These findings suggest that even in the presence of traumatic experiences, family resilience may protect against the development of psychopathology outcomes.

Importantly, resilience factors are likely functioning across multiple systems. Additional research is needed to understand how each factor may interact to influence psychopathology outcomes. For instance, some prior research in adolescents suggests that individual resilience characteristics differentially provide protection for internalizing versus externalizing symptoms. Specifically, low levels of negative cognitions (i.e., a lower tendency towards worry, ruminative thinking, or pessimism) were shared protective factors for all psychopathology in the context of sexual trauma, whereas high levels of social skills and confidence were uniquely protective for internalizing symptoms, and high levels of connectedness and empathy/tolerance were protective for externalizing symptoms⁵⁷. Additional research is needed at younger ages and with more types of stressful/traumatic experiences captured to determine how different forms of resilience may interact to differentially confer protection⁵⁸.

Limitations

The current study has several strengths including its longitudinal design, large sample size, assessment of variables of interest across the first 7 years of life, inclusion of granular timing intervals for stressful event exposures, consideration of both stressful and traumatic experiences in the same sample, operationalization of traumatic experiences across two types (i.e., interpersonal, non-interpersonal), repeated measures of psychopathology (at ages 3, 5, and 7 years), and inclusion of a measure of family resilience. However, some study limitations should be considered when interpreting the study’s findings.

The sample comprised predominantly middle to upper income, well-educated families living in an urban area of the United States. Existing meta-analytic work in the United States suggests that low family income, low Hollingshead index (education, occupation, sex, and marital status), low subjective socioeconomic status, low parental education, poverty status (being below the poverty line), and receipt of public assistance are all associated with higher levels of child psychopathology from 5 to 18 years of age (average age of 12)⁵⁹. Consequently, future research is needed to test these models in more diverse samples and determine whether the findings are comparable or may be more pronounced in the context of higher stress and trauma exposure^60–62.

Other limitations include a smaller sample size for the last aim, at times imperfect overlap in the timing of measure administration (e.g., stress assessed at repeated intervals from the prenatal period to 5 years; traumatic experiences measured cross-sectionally at 7 years to retrospectively capture events from birth to 7 years), and measurement differences for some constructs assessed (e.g., ITSEA at 3 years and CBCL at 5 and 7 years). Additionally, given that this work was conducted in early childhood, parent-report measures were used, which appears to be consistent with other informant reports (e.g., teachers/childcare providers) at younger ages. Existing evidence suggests that, as children become more independent and parents receive less information about their child’s experiences at school (e.g., after elementary school), parent ratings can become discrepant with other informants (e.g., teachers, child self-report) if there are differences in behavior across settings⁶³. Finally, family resilience may only be one process, among many possible processes, through which the effects of stress on psychopathology may be buffered and further research on resilience is necessary.

As such, future longitudinal work examining the current models in adolescence should include respondents who can capture behaviors across settings (e.g., teacher, parent, child). Future longitudinal work using consistent measures for a given construct, capturing multiple constructs at every time point (e.g., both stress and trauma; multiple resilience factors), and incorporating additional behavioral data to corroborate parent report may be useful. More repeated measures for psychopathology outcomes (e.g., annually) would also be beneficial to map trajectories of internalizing and externalizing symptoms, as well as to capture directions of temporal associations between stress, trauma, resilience, and psychopathology symptoms.

Conclusions

Taken together, the current findings highlight the importance of three factors on psychopathology symptoms in early childhood: the timing of stressful events, types of traumatic experiences, and specific facets of family resilience and adaptation. These findings are particularly important in the context of prevention and translational intervention efforts in early childhood. Specifically, the prenatal period, recent periods of stress, and accumulated stress appear to be associated with internalizing and externalizing symptoms at ages 3 and 5 years. Thus, screening, prevention, and intervention efforts that target very young children could have beneficial long-term impact on child mental health. Further, interventions that promote all aspects of family resilience may reduce risk for internalizing symptoms, whereas those that increase a family’s internal sense of control may reduce risk for externalizing symptoms. Promotion of resilience may be particularly beneficial for children who have experienced trauma. Future work is needed to determine whether the current findings generalize to samples who have been exposed to more chronic or severe stress and trauma.

Author contributions

C.A.N. and M.B.E. contributed to the funding acquisition and study design of the prospective cohort study used for this report. V.V. conceptualized the current project and led the data analysis. D.D.S. provided support for data analysis. V.V. wrote the first draft of the manuscript. All authors (D.D.S., C.A.N., M.B.E.) contributed to the revision of the manuscript.

Peer review

Peer review information

Communications psychology thanks Jonathan Schaefer and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Jennifer Bellingtier. A peer review file is available.

Data availability

The data for this manuscript are stored on a secure server and available from the corresponding author. Data is also uploaded to The National Institute of Mental Health Data Archive (https://nda.nih.gov/, NDA Collection C2655).

Code availability

Analyses for models included in this manuscript were run using IBM SPSS Statistics for Macintosh Version 29 (IBM Corp, Armonk, NY, USA). Packages for Linear Mixed Models, Linear Regression Models, and Ridge Regression Models (supported by extensions in Python Version 3.12.3) were used. Outputs are available at Harvard Dataverse 10.7910/DVN/TFQWO3.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s44271-024-00151-z.