The Long Arm of Childhood Experiences on Longevity: Testing Midlife Vulnerability and Resilience Pathways

Abstract

Adverse early experiences have been associated with higher mortality risk, but evidence varies by type of experiences, and relatively little is known about the role of favorable early experiences on health in later life. This study evaluated the independent contributions to longevity of favorable and unfavorable early experiences, including psychosocial stressors, childhood socioeconomic status (SES), and close relationships. We also examined four midlife psychosocial factors as vulnerability and resilience pathways potentially mediating these associations. The sample included 1,042 men from the VA Normative Aging Study. Early experiences were assessed retrospectively in 1961-70 and 1995. Midlife psychosocial factors were measured in 1985-91, and included stressful life events (SLEs), negative affect, life satisfaction, and optimism. Mortality was assessed through 2016. In multiple mediator structural equation models which account for the overlap among pathways, higher number of SLEs in midlife mediated the effect of having more childhood psychosocial stressors on reduced longevity, supporting stress continuity as a vulnerability pathway. Higher optimism in midlife also mediated the effects of higher childhood SES on greater longevity. In single mediator models, higher life satisfaction in midlife transmitted the benefits of higher childhood SES and presence of close relationships onto longevity. Higher optimism also mediated the association of fewer childhood psychosocial stressors to longevity. However, these indirect effects were attenuated when accounting for shared variance among mediators, suggesting overlapping pathways. Findings offer novel evidence on unique and shared pathways linking specific dimensions of early experiences to longevity.

Childhood adversity is associated with greater mortality risk in later life (Brown et al., 2009; Galobardes, Lynch, & Smith, 2004, 2008), yet evidence for this association varies by type of adverse experiences. Childhood socioeconomic deprivation has been linked consistently to reduced lifespan (Galobardes et al., 2008). In the psychosocial domain, early adversities – most commonly represented by a composite index of severe stressors such as abuse, parental mental illness, and household criminality (Felitti et al., 1998) – are related to higher risks of adult diseases and health-compromising behaviors (Hughes et al., 2017). A few studies have examined mortality as outcome (Brown et al., 2009; Chen, Turiano, Mroczek, & Miller, 2016; Kelly-Irving et al., 2013), but researchers have not directly evaluated the underlying pathways. Early supportive relationships are key determinants of developmental outcomes (Shonkoff, Boyce, & McEwen, 2009). In particular, attachment styles have been linked to later-life health outcomes (Puig, Englund, Simpson, & Collins, 2013), with stress processes potentially mediating the associations (Pietromanaco & Powers, 2015). However, surprisingly few studies have examined the association of early supportive relationships to mortality (Demakakos, Pillas, Marmot, & Steptoe, 2016).

Methodological issues, particularly different definitions of adversity (e.g., socioeconomic deprivation, composite index, or a sole focus on severe stressors such as abuse) across studies, are barriers to research on pathways underlying these associations. An emerging literature suggests that the common “composite index” approach may cloud specificity in the association between unique dimensions of early experiences and outcomes (Bush, Lane, & McLaughin, 2016).

In this study, we address these literature gaps by examining three dimensions of early experiences – socioeconomic deprivation, psychosocial stressors, and close relationships – in their unique associations with longevity; we also tested four midlife psychosocial factors – stressful life events (SLEs), negative affect (NA), optimism, and life satisfaction (LS) – as potential risk and resilience pathways underlying these associations. We considered the unique contribution of each pathway to longevity, as well as potential overlap among the pathways. We also adjusted for adult demographics and diseases as potential confounds or possible intermediate variables.

In defining the three dimensions of early experiences, we sought to optimize ecological validity by moving beyond the typical focus on severe events (e.g., abuse), including subtler forms of adversity exposure, and considering positive experiences. We defined socioeconomic deprivation using commonly used indicators of parental education, occupation, and home ownership status (Elo, 2009). For psychosocial stressors, we incorporated a broad range of experiences across the severity spectrum, from sexual molestation and important life events (e.g., parental divorce, moves), to parental problem drinking and parental discipline. We also considered supportive relationships as a positive dimension of early experiences given its key roles in numerous developmental outcomes (Bowlby, 1988; Shonkoff et al., 2009). Below, we summarize evidence on the associations on each type of early experiences to longevity, and review the literature on the mediating role of candidate pathways.

Early Experiences and Longevity

While substantial evidence has linked childhood socioeconomic deprivation to premature death in adulthood (Galobardes et al., 2004, 2008), only three studies have considered the impact of multiple childhood psychosocial stressors on longevity (Brown et al., 2009; Chen et al., 2016; Kelly-Irving et al., 2013). All three studies evaluated the cumulative impact of having more categories of early psychosocial stressors in large epidemiologic samples (N>6000), and reported curtailed lifespan among a small subset of individuals with extreme levels of early adversity. For example, in the landmark Adverse Childhood Experiences (ACE) Study, adults who endorsed six or more categories of early stressors (e.g., abuse, domestic violence) had elevated mortality risk compared with those who did not endorse any early stressors (Brown et al., 2009). The high-adversity group represented less than 3% of the sample, and it is unknown if the observed associations will hold if stressors are defined to encompass more generalizable experiences. These associations were maintained after adjusting for adult SES and lifestyle factors in two studies (Chen et al., 2016; Kelly-Irving et al., 2013), but mediation was not formally evaluated.

A burgeoning literature has examined the adult health sequelae of early adversity (Hughes et al., 2017), but few studies have considered whether favorable early experiences protect against poor health and/or promote positive outcomes (An & Cooney, 2006; Carroll et al., 2013; Huppert, Abbott, Ploubidis, Richards, & Kuh, 2010; Miller et al., 2011; Rothrauff, Cooney, & An, 2009). Early supportive relationships are an important type of favorable early experiences with known associations with developmental outcomes and later-life health (Russek & Schwartz, 1997; Shaw, Krause, Chatters, Connell, & Ingersoll-Dayton, 2004; Shonkoff et al., 2009); they are typically defined in terms of perceived emotional support, parenting style, or relationship quality (e.g., warmth) with key caregivers. Only one study to date has focused on its relation to mortality: Demakakos et al. (2016) reported that parenting style characterized by low levels of care and high levels of overprotection was associated with greater mortality risk in a large cohort of older adults.

Pathways from Early Experiences to Curtailed Longevity

This study focuses on psychosocial pathways linking early experiences to longevity. Our hypothesized pathways represent environmental influences (stressful life events), one’s transaction with the environment (optimistic attributional style), and psychological well-being (life satisfaction and negative affect). Conceptually, we define psychological well-being to include both positive feelings and cognitions as well as the absence of distress as indicators of optimal psychological functioning and experience (Ryan & Deci, 2001). As described below, the candidate vulnerability and resilience pathways have been linked to both early experiences and mortality, but their mediating role has seldom been evaluated.

Stressful life events.

Capturing the notion “stress begets stress”, stress continuity theory (Hammen, Hazel, Brennan, & Najman, 2012) has received support from prospective studies that followed children into young adulthood. For example, adverse experiences up to age five predicted higher levels of past-year stressors at ages 15 and 20. Children exposed to interparental violence were more likely to suffer from and perpetrate intimate partner violence as adults (Naraya, Labella, Englund, Carlson, & Egeland, 2017). Aldwin et al. (2011) reported that older men with chronically elevated levels of SLEs had a 37% greater hazard of dying compared to men with consistently lower SLE exposure, highlighting the value of considering stress exposure as a long-term process.

Adult differences in stress processes may also mediate the impact of early socioeconomic disparities on reduced longevity (Ferraro & Shippee, 2009). For example, in a national sample of Dutch adults, lower parental socioeconomic position was linked to lower likelihood of successful aging via lower adult socioeconomic position and greater SLE exposure (Kok, Aartsen, Deeg, & Huisman, 2017). However, few studies have tested the mediating role of stress processes in SES-health associations. Findings have been mixed and often indirect, in part due to methodological limitations such as different operationalization of SES and stress processes across studies (Matthews & Gallo, 2011).

Optimism.

Researchers have defined optimism in two ways: as an attributional style, or as a general expectancy of favorable outcomes (Carver, Scheier, & Segerstrom, 2010). We adopt the first approach, defining optimism as an adaptive attributional style that considers the cause of negative events as situational or external, transient in nature, and specific to single events, and the causes of positive events as internal or self-driven, temporally stable, and generalizable across situations (Malinchoc, Offord, & Colligan, 1995). Chronic or repeated stressors in childhood can shape the formation of unhealthy attributional styles, which amplifies the negative impact of subsequent stressors (Chorpita & Barlow, 1998). For example, having more recent SLEs preceded an increase in children’s tendency to make internal, stable, and global attributions about negative events (Garber & Flynn, 2001; Gibb et al., 2006). Children reared in unsupportive homes might lack proper adult guidance to form socioemotional competencies, learn healthy behaviors, and develop psychosocial resources, such as optimism, to deal with stressors (Jessor, 1991; Repetti, Taylor, & Seeman, 2002; Taylor, Way, & Seeman, 2011).

The reserve capacity model (Gallo & Matthews, 2003) posits that higher SES environments can promote positive cognition, such as an optimistic attributional style, and positive emotions, through the accrual of tangible, interpersonal, and intrapersonal resources. For example, higher-SES children have greater access to safe and friendly neighborhoods, which can foster a sense of interpersonal trust and safety (Cohen et al., 2010); whereas, the stigma of receiving welfare or living in decrepit housing is linked to lower self-esteem and lower sense of hope (Wiltfang & Scarbecz, 1990). Teachers tend to have higher academic expectations and provide more positive attention to higher-SES students (McLloyd, 1998). These resources may promote positive cognitive styles and emotions, and dampen the effects of stressors on negative cognition and emotions (Gallo & Matthews, 2003). Higher childhood SES has been prospectively associated with higher optimism in young adulthood (Heinonen et al., 2006). An emerging literature further supports optimism as a psychosocial asset that protects against premature death (Kim et al., 2017) and adverse health outcomes (Rasmussen, Scheier, & Greenhouse, 2009).

Psychological well-being.

A large literature documents the association between childhood psychosocial adversity – particularly abuse and neglect – with lower levels of psychological well-being in adulthood. For example, older adults recalling greater exposure to early psychosocial adversity also reported worse psychological well-being, as indicated by higher levels of anxiety, depression, and angry hostility (Wilson et al., 2006). Childhood psychosocial adversity has also been linked to higher incidence and prevalence of psychopathology (Clark, Caldwell, Power, & Stansfeld, 2010; McLaughlin et al., 2012; Schaefer et al., 2018; Widom, Dumont, & Czaja, 2007). Fewer studies have considered early supportive relationships as a predictor, but findings also suggest an association with adulthood well-being, such as greater LS and fewer depressive symptoms (Amato, 1994; Huppert et al., 2010; Shaw et al., 2004).

The reserve capacity model (Gallo & Matthews, 2003) also informs the conceptual linkage between childhood SES and adulthood psychological well-being. Children raised in low-SES environments encounter more barriers against tangible resources (e.g., safe playgrounds), interpersonal resources (e.g., higher-quality teachers, healthy relationships), and intrapersonal resources (e.g., self-efficacy). These barriers create additional stressors and strains, which can erode psychological well-being over time and precipitate psychopathology (Johnson, Cohen, Dohrenwent, Link, & Brook, 1999). Although markers of psychological well-being, including higher LS and lower NA, have been associated with lower mortality risk (e.g., Boehm, Winning, Segerstrom, & Kubzansky, 2015; Wilson, Bienias, Mendes de Leon, Evans, & Bennett, 2003), mediational models linking childhood SES via adult psychological well-being to longevity have not been tested.

Present Study

This study extends research linking childhood experiences to longevity in two ways. First, we examined the unique contributions of early psychosocial stressors, socioeconomic deprivation, and supportive relationships to longevity. In particular, we tested four vulnerability (SLEs, NA) and resilience (optimism, LS) psychosocial factors in midlife as mediators of these associations. Figure 1 is a conceptual diagram illustrating our theoretical model. From a cumulative advantage/disadvantage viewpoint (Dannefer, 2003), we expected positive early experiences to exert protective effects on longevity by promoting optimism and LS in midlife, and negative early experiences to reduce lifespan by increasing SLE exposure and NA, thereby resulting in differential mortality.

Figure 1.

Conceptual figure representing key pathways evaluated in the current study.

Second, our approach to conceptualizing and operationalizing early experiences produces findings with stronger external validity than common approaches (e.g., focusing on single type of adversity or severe stressors). We considered both favorable and unfavorable early experiences, and included indicators from across the severity spectrum and used a dimensional approach when possible. We modeled concurrent paths from early experiences via midlife factors to longevity, thus reflecting the co-occurring nature of these pathways in real life. To inform knowledge on causal mechanisms, we also compared results with models including one mediator at a time to gauge the extent of overlap among the hypothesized pathways.

METHOD

Sample

The sample was drawn from the Veterans Affairs (VA) Normative Aging Study (NAS), a longitudinal investigation of normal and pathological aging processes in men founded at the Boston VA Outpatient Clinic. Between 1961 and 1970, over 6,000 men were screened for chronic or major physical and mental illnesses, and for geographic stability, defined as kinship ties in the Boston area and stated intentions to remain local; 2,280 men were enrolled. Childhood SES was assessed retrospectively at NAS entry in the screening survey (1961-70). The Childhood Experiences Scale (Aldwin, Levenson, Cupertino, & Spiro, 1998) was administered as part of a 1995 mail survey (N=1,076, response rate=73%). Midlife psychosocial factors and socioeconomic position were assessed by mail surveys between 1985 (mean age=59.5, SD=7.2) and 1991 (mean age=65.4; SD=7.2). Midlife chronic disease status was assessed during in-person examinations occurring every 3-5 years since NAS entry. Mortality data through 2016 were used. The analytic sample included 1,042 men (97% of CES respondents) with data on at least one childhood variable and one midlife variable. All participants consented to study procedures and the research was approved by the relevant institutional review boards.

Measures

Early psychosocial stressors.

Early psychosocial stressors were assessed retrospectively in a 1995 mail survey containing the CES (Aldwin et al., 1998) and items on parental alcohol consumption. The CES was designed for use with the general population and measures occurrence of negative, neutral, and positive early experiences from birth to age 19. In an independent validation sample aged 22-66, item responses showed good stability over a 5-year period and response consistency was unrelated to age or depression status on both occasions (Yancura & Aldwin, 2009). Early psychosocial stressors were indicated by endorsement of CES items representing separation from parents (three items), loss involving siblings and key adults (three items), serious injury by others or sexual molestation (two items), residential change (two items), and harsh parental verbal and physical discipline (five items collapsed into two variables to indicate physical (e.g., slapped) and non-physical (e.g., saracasm) discipline). To increase the scope of experiences covered, we added two items on maternal and paternal problem drinking, respectively, also assessed in the 1995 mail survey. Each item was coded positive if respondents endorsed one of the three following conditions for the corresponding parent: Reporting (1) “mother” or “father” on an item assessing if the respondent experienced problem drinkers among family members; (2) a drinking binge by that parent when the respondent was growing up and a binge frequency of “several times a year” or greater; (3) one or more negative effects of alcohol (e.g., “became hostile when he/she drank,” “had the shakes”) for that parent at a frequency of “sometimes” or “often” when the respondent was growing up. Total count of childhood psychosocial stressor was coded as missing if respondents completed < 80% items. The count score was top-coded at six due to a positively skewed distribution (2.5% observations with scores >6).

Childhood SES.

Childhood SES was assessed via questionnaire at NAS entry (1961-70). As described in more detail in our earlier work (Lee et al., 2015; Peters et al., 2011), we used four indicators: Paternal and maternal education (0=did not complete grammar school to 6=beyond college); paternal occupation (0=unskilled to 5=professional/managerial/proprietary); and parental home ownership during childhood (0=no, 1=yes). The first principal component of these indicators was used as a composite index of childhood SES using a z-score metric, with higher scores indicating higher SES.

Early close relationships.

Early close relationships were assessed with four binary CES items asking whether one felt close to an adult relative, other adults (e.g., teacher), siblings, or a friend. We coded a binary variable indicating the presence of any close relationships in childhood, as indicated by an affirmative response to at least one item. We used a dichotomous approach because the four items do not have sufficient coverage to adequately reflect the quantity or quality of early support. It also reflects the idea that the presence of at least one supportive relationship is critical to the long-term health and adaptation of vulnerable children (Werner & Smith, 2001).

Stressful life events.

Past-year SLEs were assessed in 1985 using the Elders Life Stress Inventory (ELSI; Aldwin, 1990), which comprises 30 major events tailored for middle-aged and older adults. Examples include retirement and institutionalization of a parent. We excluded two items on one’s own health to reduce concerns about confounding. Item responses were coded dichotomously to indicate occurrence and summed. Total score was coded as missing if respondents completed <80% items. The total score (observed range: 0-15) was skewed (skewness=1.5) and therefore square-root transformed.

Negative affect.

The NA subscale of the Positive and Negative Affect Schedule (PANAS; Watson, Clark, & Tellegen, 1988; Cronbach’s alpha=0.90) was assessed in 1991. The scale comprises ten items assessing the extent to which respondents experienced negative emotions, such as “irritable” and “distressed”, in the past three months. Items were rated on a 5-point scale (0=very slightly or not at all to 4=extremely). After excluding respondents who completed <80% of items, mean substitution was used to fill in missing items and item scores were summed to create a total score.

Life satisfaction.

LS was measured in 1987 with Liang’s (1984) 11-item version of the Life Satisfaction Inventory: Form A (Neugarten, Havighurst, & Tobin, 1961; Cronbach’s alpha=0.74). Items include, “I am just as happy as when I was younger” and “As I look back on my life, I am fairly well satisfied”. Respondents indicated whether they agreed with each item (1=agree, 0=disagree). Negative-worded items were reverse coded. After excluding respondents who completed <80% of items, mean substitution was used to fill in missing items and the total score was computed. To facilitate interpretation, LS scores were recoded into standard deviation (SD) units.

Optimism.

Attributional style optimism was assessed in 1986 with the Revised Optimism-Pessimism scale (PSM-R; Malinchoc et al., 1995). Malinchoc et al. (1995) applied the Content Analysis of Verbatim Explanations (CAVE) technique to Minnesota Multiphasic Personality Inventory-II items (Butcher, Dahlstrom, Graham, Tellegen, & Kaemmer, 1989) and identified 263 which are weighted to yield a bipolar score on a continuum from optimistic to pessimistic. We created maximum likelihood estimates of the missing items based on available responses; however, the total score was set to missing if more than 5% of the items were missing (<2% of respondents). Total scores (originally on a T-score metric) were reverse-coded and divided by 10, such that each additional unit corresponds roughly to one SD higher in optimism level (Kuder-Richardson 20 =.87). The PSM-R has good stability over a 5-year period in the NAS (r=.87, p<.001).

Longevity.

Vital status information is collected via routine searches of Department of Veterans Affairs and the Social Security Administration (Death Master File) records, regular mailings to NAS participants, and notification of participant deaths from next-of-kin or postal authorities. Longevity was quantified by survival time between CES administration in 1995 and death from any cause through 12/31/2016 (deceased = 666, 63.9%; mean follow-up time: 14.7 years, SD=6.4). Men who survived beyond the cut-off date were right-censored.

Covariates.

All models were adjusted for age at the time of CES administration (1 unit=10 years). Marital status (1=married, 0=otherwise) and education (highest level attained, in years) were assessed by mail survey in 1986.

For midlife chronic disease status, we coded four binary variables to indicate the presence of coronary heart disease, Type II diabetes, stroke, and cancer prior to 1986. Disease diagnosis and year of onset were determined from physical examination and medical record review by the study physician every 3-5 years. Due to the low base rate of Type II diabetes (<5%) and stroke (<2%), they were combined with coronary heart disease (11%) and coded into a single binary variable indicating presence of cardiometabolic diseases prior to 1986.

Statistical Analyses

In preliminary analyses, we examined the main effects of each childhood experience on the risk of all-cause mortality using Cox regression with time-to-event (death or right-censoring) as outcome and adjusted for age. Next, we tested the primary hypothesis that early experiences are associated with longevity via midlife vulnerability and resilience pathways using structural equation models (SEMs). The main SEM included three childhood measures as predictors, four midlife vulnerability and resilience factors as mediators, time-to-event (death or right-censoring) as the outcome, and age as a covariate. Midlife factors were regressed on childhood variables and covariates using ordinary least square regression (a paths). Time-to-event was regressed on midlife factors (b paths), childhood variables (c’ paths, i.e., direct effects), and covariates using Cox regression. Time-to-event was coded as number of years from CES administration to death or right-censoring. Because death is coded as the event of interest, we describe findings in terms of mortality risk in the Results section, and interpret them in terms of longevity (indicated by lower mortality risk) in the Discussion. Covariances among all exogenous variables (i.e., childhood measures and any covariates) were freely estimated, as were covariances among the mediators. Mediators and time-to-event were adjusted for age in all models. We checked the proportional hazards assumption of Cox regression using the Supremum test and visually inspecting survival functions associated with all childhood variables, psychosocial mediators, and demographic variables. Analyses were conducted with Mplus version 8 (Muthén & Muthén, 1998-2017).

In two successive models, we first added marital status and education at midlife as demographic covariates, and then adjusted for midlife chronic disease status. As these covariates could be confounders or potentially also lie on pathways from childhood experiences to longevity, indirect effects from these models are conservative estimates.

Statistically significant indirect effects provide support for our hypothesis. Following best practice in mediation testing (Preacher, 2015), each indirect effect was computed as the product of raw coefficients from its composite paths. An indirect effect (a*b) is interpreted as the risk of dying associated with each midlife factor (b path), weighted by the strength of association between the midlife factor and an early experience (a path). Because the outcome in Cox regression is a logged hazard, the exponentiated indirect effect estimate (i.e., exp[a*b]) is a HR representing the unique influence of one dimension of early experience on mortality that is transmitted through a midlife factor, holding other childhood variables, mediators, and covariates constant. We used bootstrapping to approximate the empirical distribution of indirect effects; a 95% bootstrap confidence interval (95%CI) not overlapping with zero (or one for a HR) indicates a statistically significant indirect effect. Turiano and colleagues (2012) used this approach to test mediation involving mortality outcome in SEMs. We do not consider our results in terms of “partial” vs. “full” mediation, as this traditional approach is problematic when indirect effects are in opposite directions (e.g., they cancel each other out when summed to calculate the total effect; see Hayes, 2013).

Missing data estimation.

All SEMs were estimated using the full information maximum likelihood (FIML) robust method, which assumes missing-at-random (MAR; Rubin, 1976). For model estimates to be unbiased by missingness, the MAR assumption requires inclusion of correlates of missingness in the model and use of contemporary methods such as FIML for model estimation (Little, Jorgensen, Lang, & Moore, 2014). We considered three sources of data missingness in our design, including (1) missingness from NAS participants not in the analytic sample, (2) missingness due to non-response to a 1995 survey containing the CES (as the analytic sample was primarily defined by participation in this survey); and (3) missingness on specific measures within the analytic sample. In sensitivity analyses, we compared men with vs. without each type of missing data. Results are described in Supplemental Document 1. Briefly, correlates of data missingness include age, mortality status, education, marital status, childhood SES, and midlife levels of life satisfaction and optimism. Importantly, as we included relevant correlates of missing data and used FIML estimation in our analytic models, our design meets the MAR assumption.

Multicollinearity among mediators.

During the analysis, we observed that optimism was moderately correlated with negative affect (r=−.46) and life satisfaction (r=.48). All other correlations between mediators were under ∣r∣=0.30. Because multicollinearity inflates sampling variance of model estimates (Hayes, 2013), we ran two separate sets of models: one using SLEs, LS, and NA as correlated mediators, and another using SLEs and optimism as correlated mediators. We ran supplemental analyses involving one mediator in each model and compare findings against the main SEMs to gauge the extent of overlap among pathways.

RESULTS

Descriptive statistics are shown in Table 1 and correlations among variables are shown in Table 2. In this sample of men born 1903 to 1945, average parental education was less than high school. Men reported experiencing three childhood stressors on average, and over 87% reported having at least one close relationship during childhood. When assessed in 1986 (mean age=60), they had completed some college on average, and most (89%) were married. Over the next thirty years of follow-up (through 2016), 64% died. They had lived to an average of 84 years old (SD=7.6), likely reflecting a healthy selection effect at NAS entry.

Table 1.

Descriptive Statistics (N=1,042).

	N	M / n (%)	SD	Range
Childhood Experiences
Childhood socioeconomic status:
Father's education	991	1.57	1.55	0-6
Mother's education	994	1.63	1.44	0-6
Father's occupation	998	1.94	1.43	0-5
Parental home ownership (1=Y, 0=N)	1024	61.8%	--	--
Early psychosocial stressors (count)	1042	2.84	1.69	0-10
Early close relationships (1=any, 0=none)	1021	87.7%	--	--
Midlife Vulnerability and Resilience Factors
Stressful life events (count)	964	1.84	1.81	0-15
Negative affect	803	6.139	6.33	0-36
Life satisfaction	810	4.25	1.00	0.6-5.6
Optimism	952	5.44	1.04	2.0-8.3
Covariates
Age (at CES administration; 12/1995-8/1997)	1042	69.34	7.25	50-92
Education (years)	835	14.17	2.70	4-20
Married (1=Yes, 0=No)	919	88.7%	--	--
Cardiometabolic disease (1=Yes, 0=No)	957	15.7%	--	--
Cancer (1=Yes, 0=No)	945	1.8%	--	--
Mortality Status as of 12/2016
Number deceased	666		--	--
Time-to-death since CES administration (years)	666	11.13	5.52	0.3-21

Note. CES=Childhood Experiences Scale. Parental education was coded as: 0=did not complete grammar school, 1=completed grammar school, 2=some high school, 3=completed high school, 4=some college, 5=completed college, 6=beyond college. Father’s occupation was coded as: 0=unskilled, 1=semi-skilled, 2=skilled/foreman, 3=white collar, 4=semi-professional, 5=professional / managerial / proprietary.

Table 2.

Correlations among childhood experiences variables (1-3), midlife risk and resilience factors (4-8), covariates (9-12), mortality status and follow-up duration (13-14).

		1	2	3	4	5	6	7	8	9	10	11	12	13
Childhood experiences
1	Early psychosocial stressors
2	Childhood SES	.01
3	Early close relationships	.11	.02
Midlife risk and resilience factors
4	Stressful life events	.13	.04	.07
5	Negative affect	.08	−.03	.02	.15
6	Life satisfaction	−.06	.12	.11	−.13	−.29
7	Optimism	−.10	.12	.01	−.21	−.46	.48
Covariates
8	Age (at CES administration)	−.02	−.13	.01	−.06	−.12	.07	−.01
9	Education	.01	.32	.06	.10	.01	.01	.10	−.04
10	Married	.06	−.02	.03	−.06	.03	.11	.02	−.01	−.02
11	Cardiometabolic disease	.05	−.02	−.04	.04	<.01	−.08	−.07	.19	−.04	.04
12	Cancer	.03	<.01	<.01	−.03	−.03	−.08	.03	.06	−.04	.03	.13
Time-to-event variables
13	Dead (1=yes, 0=no)	.01	−.05	−.02	.05	−.02	−.03	−.05	.50	−.05	<.01	.24	.04
14	Follow-up duration	−.01	.03	.05	−.04	.02	.05	.06	−.48	.06	.06	−.26	−.05	−.73

Note. Bold: p < .05; Italics: .05 ≤ p < .10

Cox regression assumes that the effect of each predictor on the risk of dying is equivalent across the follow-up period. Results from the Supremum test indicated that the proportionality assumption was met by all variables except age (p=.03). We further dichotomized the sample by age (≤70 vs. >70 at CES administration) to inspect the survival functions, which showed a steeper mortality drop-off over time for older men than younger men, thus violating the proportionality assumption. Therefore, we repeated the main effects analysis and age-adjusted mediation models for the two age groups and report the results in the Supplemental Materials. Supplemental Table 2 provides descriptive statistics for the two age groups. Of note, the 95%CI for most parameters showed substantial overlap between the two groups (Supplemental Tables 3-5), even when a given path was statistically significant in one group but not another. This suggests that our sample size is too small for the age group comparisons to be reliable and meaningful. We conducted additional sensitivity analyses to determine whether parameter estimates involving the variables of interest are biased by violating the proportionality assumption: We first ran a Cox regression of mortality risk on the childhood variables, psychosocial mediators, and age. Results were compared against a model that added an age × follow-up time interaction term, which accounts for uneven age differences in mortality risk over follow-up time. Parameter estimates for the childhood variables and psychosocial mediators were nearly identical between the two models, suggesting minimal bias due to violating the proportionality assumption. Therefore, we present results based on the entire sample below.

Main Effects of Childhood Experiences on Mortality Risk

In main effect models, the three childhood experiences were not associated with all-cause mortality risk when they were entered simultaneously as predictors, and when each childhood experience was examined separately (Supplemental Table 3). Of note, a significant main effect is not a prerequisite for mediation analyses (Hayes, 2013), because indirect and direct effects in opposite directions may cancel each other when summed to compute the main effect.

Results with Stressful Life Events, Negative Affect, and Life Satisfaction as Mediators

Figure 2 shows path estimates from the age-adjusted multiple mediator SEM with three childhood experiences as predictors, three midlife vulnerability (SLEs, NA) and resilience factors (LS) as mediators, and time-to-death as the outcome; bold lines represent statistically significant paths. Table 3a shows the corresponding indirect effect estimates. Supplemental Table 4 shows the age-stratified results. For the paths from childhood experiences to midlife psychosocial factors, having more childhood psychosocial stressors was associated with more past-year SLEs. Higher childhood SES and endorsing any close relationships in childhood were associated with higher LS in midlife. Childhood experiences were unrelated to NA in midlife. As for the paths from midlife factors to mortality, higher SLE exposure predicted greater mortality risk. The hazard ratio for the square-root transformed SLE variable (HR=1.12) translates into 18% greater risk of dying across the follow-up period for men with 2 vs. 0 past-year SLEs (roughly corresponding to M vs. M-1SD in raw SLE count), and 26% greater risk of dying for men with 4 vs. 0 past-year SLEs (roughly corresponding to M+1SD vs. M-1SD in raw SLE count). To place these effect sizes into context, having a history of cancer was associated with a 25% greater risk of dying, and each additional year of education was marginally associated with a 2.8% lower risk of dying in this sample.

Figure 2.

Age-adjusted structural equation model on the effects of early experiences on longevity via midlife stressful life events, negative affect, and life satisfaction. Paths are labeled with raw regression coefficients or hazard ratios (HR; specific to paths predicting mortality) and 95% bootstrap confidence intervals in parentheses. Corresponding indirect effect estimates are shown in Table 3(a).

Table 3.

Indirect effects of early experiences via midlife psychosocial factors to longevity.

(a) Model with SLEs, NA, and LS as mediators
	Mediator = SLEs		Mediator = NA		Mediator = LS
	HR	95% CI	HR	95% CI	HR	95% CI
Early psychosocial stressors	1.01	(1.001, 1.02)	1.00	(1.00, 1.01)	1.00	(0.999, 1.01)
Childhood SES	1.00	(0.996, 1.01)	0.996	(0.99, 1.00)	0.99	(0.98, 1.001)
Early close relationships	1.01	(0.996, 1.05)	1.00	(0.98, 1.02)	0.98	(0.94, 1.01)
(b) Model with SLEs and optimism as mediators
	Mediator = SLEs		Mediator=Optimism
	HR	95% CI	HR	95% CI
Early psychosocial stressors	1.01	(1.001, 1.02)	1.01	(1.00, 1.01)
Childhood SES	1.00	(0.996, 1.01)	0.99	(0.98, 0.999)
Early close relationships	1.01	(0.996, 1.05)	1.00	(0.97, 1.02)

Note: Mediators and mortality were adjusted for age. HR=hazard ratio; SES=socioeconomic status; SLEs=stressful life events; NA=negative affect; LS=life satisfaction. Bold indicates a hazard ratio for which the 95%CI does not overlap with one. Italics indicates a hazard ratio for which the 95% UCI or LCI was estimated at one.

As for our mediational hypotheses, the indirect effect linking early psychosocial stressors via more midlife SLEs to greater mortality risk was significant. Based on the model estimates, comparing men with 5 vs. 1 childhood psychosocial stressors (roughly corresponding to M±1SD), those with more childhood psychosocial stressors had 3% greater risk of dying that was uniquely attributable to having more adulthood SLEs, net of the effect of midlife LS, NA, and age. Other indirect effects, as well as the direct effects between early experiences and mortality, were nonsignificant.

Adding demographics and diseases as covariates to the model did not alter the pattern of findings, although some associations became attenuated. After adjusting for education and being married in midlife, the indirect path from early psychosocial stressors via midlife SLEs to mortality remained significant (indirect effect: HR=1.01, 95%CI: 1.001, 1.02), but this path became marginally significant when further adjusting for midlife cardiometabolic disease and cancer (HR=1.01, 95%CI: 1.000, 1.014). In this fully-adjusted model, childhood psychosocial stressors remained significantly associated with midlife SLEs (B=0.06, 95%CI: 0.03, 0.09), but the association of SLEs to mortality became slightly weaker (HR=1.10, 95%CI: 1.00, 1.22). Other indirect paths remained nonsignificant (ns). In terms of the associations of covariates to the mediators and to mortality, more educated men reported more SLEs. Older men reported lower NA and higher LS. Being married was associated with higher LS and lower mortality risk. Older age and having any cardiometabolic disease at midlife were associated with higher mortality risk.

Results with Stressful Life Events and Optimism as Mediators

Figure 3 shows path estimates from the age-adjusted multiple mediator model with SLEs and optimism as mediators. The corresponding indirect effect estimates are shown in Table 3(b). Model estimates for the two age groups are shown in Supplemental Table 5. Results for paths mediated by SLEs are identical to those described in the previous section and not repeated. Having fewer childhood psychosocial stressors and higher childhood SES were associated with higher levels of optimism in midlife. In turn, higher optimism protected against mortality. Each SD higher in optimism was associated with 8% lower risk of dying across the follow-up period. The indirect path from higher childhood SES via higher midlife optimism to lower mortality risk was significant, and the indirect path from more childhood psychosocial stressors via lower optimism to greater mortality risk was marginally significant. For example, comparing men who scored 1 SD below vs. 1 SD above the mean on childhood SES, those with higher childhood SES had 2.1% greater risk of dying over the follow-up period that was uniquely attributable to higher optimism in midlife.

Figure 3.

Age-adjusted structural equation model on the effects of early experiences on longevity via midlife stressful life events and optimism. Paths are labeled with raw regression coefficients or hazard ratios (HR; specific to paths predicting mortality) and 95% bootstrap confidence intervals in parentheses. Corresponding indirect effect estimates are shown in Table 3(b).

In models that further adjusted for demographics and diseases, the overall pattern of findings remained similar, although associations were attenuated. When we added education and marital status, the two indirect paths originating from early psychosocial stressors (via SLEs and optimism to longevity) were unchanged. The indirect path from childhood SES via optimism to mortality became marginally significant (HR=0.99, 95%CI: 0.98, 1.00); this was because the paths from childhood SES to optimism (B=0.10, 95%CI: 0.04, 0.17) and from optimism to mortality (HR=0.93, 95%CI: 0.85, 1.00) became slightly weaker. When we further adjusted for cardiometabolic disease and cancer, the indirect path from early psychosocial stressors via midlife SLEs to mortality became marginally significant (HR=1.01, 95%CI: 1.00, 1.01), and the two indirect paths mediated by optimism became ns (from early psychosocial stressors: HR=1.00, 95%CI: 1.00, 1.01; from childhood SES: HR=0.99, 95%CI: 0.98, 1.00). In this fully adjusted model, optimism remained significantly associated with childhood SES (B=0.10, 95%CI: 0.03, 0.18) and childhood psychosocial stressors (B=−0.06, 95%CI: −0.11, −0.02), but its association to mortality became ns due to a wider CI (HR=0.93, 95%CI: 0.86, 1.01). In terms of the association between covariates and optimism, more years of education were related to higher optimism levels, and having a history of cardiometabolic disease was marginally associated with lower optimism.

Supplemental Analyses: Single Mediator Models

To better understand the impact of adjusting for multiple indirect effects in the same model, we conducted supplemental analyses wherein each model comprised the three childhood predictors, one midlife mediator, time-to-death as the outcome, and was adjusted for age. Table 4 presents indirect effect estimates from the four models. Path estimates are not shown, but available upon request.

Table 4.

Indirect effects from early experiences to longevity via adulthood stressful life events (Model A), negative affect (Model B), life satisfaction (Model C), and optimism (Model D). Each model had one mediator and was adjusted for age only.

	HR	95% CI
Model A: Stressful life events as mediator
Early psychosocial stressors	1.01	(1.001, 1.02)
Childhood SES	1.00	(0.995, 1.01)
Early close relationships	1.02	(0.99, 1.05)
Model B: Negative affect as mediator
Early psychosocial stressors	1.01	(0.999, 1.01)
Childhood SES	0.99	(0.98, 1.003)
Early close relationships	1.00	(0.98, 1.03)
Model C: Life satisfaction as mediator
Early psychosocial stressors	1.004	(1.00, 1.01)
Childhood SES	0.99	(0.97, 0.999)
Early close relationships	0.97	(0.93, 0.998)
Model D: Optimism as mediator
Early psychosocial stressors	1.01	(1.001, 1.01)
Childhood SES	0.99	(0.98, 0.997)
Early close relationships	1.00	(0.97, 1.02)

Note. HR=hazard ratio; SES=socioeconomic status. Bold indicates a hazard ratio for which the 95% asymmetric (bootstrap) confidence interval (CI) does not overlap with one. Italics indicates a hazard ratio for which the upper or lower bound of the 95%CI was estimated at one.

For SLEs as mediator (Table 4, Model A), the indirect effect estimates were identical in the single vs. multiple mediator models. For NA as mediator (Table 4, Model B), as in the multiple mediator model, all indirect paths via NA in the single mediator model was nonsignificant.

Whereas none of the indirect effects via LS was significant in the multiple mediator model, higher levels of LS accounted for the protective effects of higher childhood SES and having early close relationships against mortality in the single mediator model (Table 4, Model C). Path estimates indicated that while higher LS protected against mortality in the single mediator model such that each SD higher in LS was linked to 9% lower mortality risk (HR=0.91, 95%CI: 0.84, 0.99), this association was ns when we adjusted for its overlap with SLEs and NA in the multiple mediator model (HR=0.93, 95%CI: 0.85, 1.01; Figure 2).

For optimism, the pattern of findings was similar between the single mediator model (Table 4, Model D) and the multiple mediator model. The only difference was that, whereas the indirect path from early psychosocial stressors via lower optimism to higher mortality risk was marginally significant in the multiple mediator model that adjusted for its overlap with the SLE pathway, this indirect path became significant when optimism was the sole mediator. This was due to a slightly stronger association between optimism and lower mortality risk in the single mediator model (HR=0.90, 95%CI: 0.84, 0.97) than in the multiple mediator model (HR=0.92, 95%CI: 0.85, 0.99), suggesting the association of optimism to mortality is partially shared with SLEs.

DISCUSSION

In this study, we examined the contributions of three dimensions of early experiences on longevity, and identified mediating vulnerability and resilience pathways using longitudinal data spanning two decades (Figure 1). Regarding vulnerability pathways, evidence was most robust for stress continuity as a risk transmission mechanism: greater exposure to psychosocial stressors in childhood was associated with more SLEs in midlife, which in turn reduced longevity. Regarding resilience pathways, higher levels of optimism in midlife mediated the association from higher childhood SES to greater longevity. These two pathways were independent from the effects of other childhood experiences and midlife psychosocial factors. Additionally, results suggested higher life satisfaction in midlife as a resilience pathway conveying the benefits of higher childhood SES and close relationships onto greater longevity. Optimism also carried the benefits of fewer childhood psychosocial stressors onto longevity. These three pathways were apparent only without adjustment for other midlife factors, suggesting that the mediated effects were also transmitted through other psychosocial factors under consideration. Our findings add to the vast literature on early adversity and health by identifying differential pathways linking specific dimensions of childhood experiences to longevity, and illustrating the degree of overlap among these pathways.

Vulnerability Pathways via Adulthood Stress Exposure

Consistent with stress continuity theory, experiencing more SLEs in midlife transmitted the negative influence of having more childhood psychosocial stressors to reduced longevity. While stress continuity is typically considered in the etiology of psychopathology and in younger populations (e.g., Hammen et al., 2012), our findings are novel in demonstrating its impact on longevity. Stressful childhood environments, as captured by the “risky families” model (Repetti et al., 2002), can set off life trajectories characterized by impaired academic and socioemotional development, risky peers, and limited access to resources, thus creating additional life challenges and sustaining chronic exposure to stressors. While the association between childhood psychosocial stressors and midlife SLEs was small in magnitude, it is worth noting that they were assessed 10 years apart, and the association was robust to adjustment for midlife demographics and chronic diseases, suggesting a small yet enduring impact of early stress exposure on adulthood circumstances. In contrast, the effect of midlife SLEs on longevity was larger, with men endorsing 4 vs. 0 past-year events having 26% greater risk of dying. It would be fruitful to further investigate intermediate factors in the SLE-longevity association.

Resilient Pathways via Midlife Optimism and Life Satisfaction

Optimism conveyed the benefits of higher SES and fewer psychosocial stressors in childhood onto longevity. The finding regarding childhood SES offers developmental evidence to substantiate the reserve capacity model (Gallo & Matthews, 2003), which has been studied primarily in adults. Higher childhood SES likely imparts resources, such as better education, role models, and stable living environments, which can foster adaptive response to difficult situations (Cohen et al., 2010). Consistent with this notion, we observed that the association from childhood SES to optimism is partially attributable to adult education. On the other hand, childhood stressors such as parental substance use and frequent residential moves may suggest a lack of appropriate role models and routines for developing effective problem-solving and emotion regulation skills, and fewer opportunities to develop an optimistic attributional style. In turn, optimism may bring about good health through better self-regulation in the affective (e.g., emotion regulation), cognitive (e.g., goal engagement), and behavioral (e.g., delaying gratification) domains (Kubzansky, Boehm, & Segerstrom, 2015).

Comparing the multiple vs. single mediator models suggested some overlap in SLEs and (lower) optimism in conveying the effects of early psychosocial stressors onto longevity, as the indirect path via optimism was weakened when adjusted for its shared variance with the SLE pathway. One possibility is that repeated exposure to stressors could erode one’s ability to appraise the causes of these negative events as external, transient, and circumscribed, which render individuals more vulnerable to the detrimental health impact of the stressors.

In models that adjusted for cardiometabolic diseases at midlife, the effects of SLEs and optimism on mortality became attenuated, suggesting cardiometabolic diseases as potential confounds and/or intermediate variables of these associations. Although cardiometabolic diseases were assessed at a single occasion, the underlying pathophysiological processes develop over decades (which precede and follow our assessment of SLEs and optimism), thus we were unable to tease apart the direction of causality between biological and psychosocial factors mediating the associations of interest. Evidence suggests bidirectional influences between these psychosocial and physiological pathways (Taylor et al., 2011). For example, greater SLE exposure has been linked to subsequent increase in inflammation (Glei, Goldman, Wu, & Weinstein, 2013); being diagnosed with chronic diseases could trigger stressful events, such as financial hardship.

Our results suggest that LS could serve as a resilience pathway. When LS was the only mediator examined, higher LS conferred the benefits of higher childhood SES and presence of close relationships onto longevity. These indirect effects were no longer significant after adjusting for concurrent pathways via midlife SLEs and NA. Given that LS and NA are both indicators of psychological well-being (despite measuring different facets of the construct; Diener, Suh, Lucas, & Smith, 1999), to the extent that early experiences influence longevity via mechanisms common to both LS and NA, the indirect effects via LS were concealed when we removed its shared variance with NA. These findings lend support to the notion that socioeconomic and interpersonal resources in childhood could potentially extend lifespan by promoting a general sense of psychological well-being in adulthood.

Our findings did not support the hypothesis that disposition to experience negative emotions in midlife mediated the effects of early experiences on longevity. NA was unrelated to all the childhood experiences and to mortality. It might be that the “past three month” time frame of the PANAS did not adequately capture a trait-like disposition; replication using a trait measure of NA would help clarify this issue. Future work may also explore the role of affect dynamics, such as affective reactivity to stressors (e.g. Mroczek et al., 2013), in these associations.

Of the childhood experiences examined, early supportive relationships had the weakest associations with midlife psychosocial factors, and most indirect effects originating from this dimension were non-significant (except for one path via LS in the single mediator model). These findings are inconsistent with studies reporting a protective effect of early support against later health and well-being (Demakakos et al., 2016; Huppert et al., 2010; Shaw et al., 2004). One possibility is that our measure of early support did not adequately capture the quality and quantity of early relationships. Some studies have suggested a stress-buffering effect of early support on later health (Carroll et al., 2013; Miller et al., 2011), but we were limited in statistical power to test moderation because most men (88%) endorsed the presence of any supportive relationships. It would be worthwhile for studies with more refined measures of early experiences to consider their additive and multiplicative effects.

Alternative Explanations

There are several alternative explanations for the observed findings. Genetic influences is one example. Genetic propensity can influence how individuals select or enter their environments. For example, children genetically predisposed to higher intelligence are more academically competitive and likely to attain higher levels of education. Genetically-influenced attributes can evoke responses from the environment, as in caretakers being more likely to respond aggressively to defiant children. Parents’ genetic predisposition can also influence the environments they create for their children. However, heritability estimates for childhood SES, childhood maltreatment, and parenting styles are small to moderate in magnitude (Kendler & Baker, 2007; Schulz-Heik et al., 2009; South, Schafer, & Ferraro, 2015; Trzaskowski et al., 2014). Genetic influences account for 25% to 39% of variance in our mediators (Baker, Cesa, Gatz, & Mellins, 1992; Kendler & Baker, 2007; Plomin et al., 1992; Stubbe, Posthuma, Boomsma, & de Geus, 2005), and about 10% of variance in human lifespan (Ruby et al., 2018). Thus, genetic mechanisms cannot entirely account for the shared variance of early experiences with psychosocial factors, or with longevity. Nonetheless, genetically informative designs are useful in understanding the role of genetics in these associations.

Health behaviors, such as diet and physical activity, may also lie on the explanatory pathways from early experiences to midlife psychological well-being and to longevity. As the relationship between health behaviors and our psychosocial mediators are likely reciprocal (Azevedo Da Silva et al., 2014; Koivumaa-Honkanen et al., 2000; Lappan, Thorne, Long, & Hendricks, 2018), studies considering the role of childhood exposure on the interrelated developmental trajectories of health behaviors and psychological well-being are highly relevant.

Considerations and Challenges in Mediation-Testing of Lifespan Associations

The indirect effect estimates are small in magnitude, but nonetheless important, for several reasons. We propose that the “long arm of childhood” (Hayward & Gorman, 2004) is more aptly thought of as a bundle of varying-sized chains, each comprising a different number of interlocking segments covering the entire lifespan. Severe childhood stressors might have the strongest and most penetrant effect on adult health, similar to a thick, unsegmented chain directly connecting childhood to adulthood factors. A unique aspect of this study is that we examined more generalizable forms of early adversity of varying severity, rather than focusing solely on severe stressors. The effects of more commonplace stressors are expectedly smaller, more susceptible to modification by external forces, transmitted through a series of mediating factors, and diluted over decades of aging. These associations resemble multiple thin chains made of small, interlocking segments. Using this chain-of-risk analogy, the effects of severe childhood stressors on mortality manifest as large direct effects in analytic models, whereas the effects of commonplace early stressors are more aptly represented by indirect effects comprising a string of mediators. Consistent with this analogy, we observed a small but robust association between early psychosocial stressors and midlife SLEs in this study.

Statistically, as indirect effects are by definition a subset of the overall predictor-outcome association, they tend to be smaller in magnitude than the overall association. Because indirect effects are computed by multiplying the regression coefficients of the composite paths, this can sometimes result in small point estimates (e.g., Pudrovska & Anikputa, 2014). Joint modeling of multiple mediators also produces conservative indirect effect estimates by removing shared variance among overlapping pathways; nonetheless, this approach is useful for evaluating the degree of overlap. Importantly, dismissing indirect effects based on small magnitude can result in false rejection of true associations. Alternative approaches, such as using bootstrap confident intervals to determine the precision and robustness of small indirect effects, and focusing on the substantive significance of parameter estimates, are strongly recommended (Walters, 2018).

Limitations

Two key limitations of this study are the reliance on retrospective assessments of childhood experiences and the restriction it imposes on making causal inferences. Our measure of early experiences, the CES, was administered when the sample was on average 69 years old, thus the recall of early experiences was susceptible to the influence of later-life circumstances. It should also be noted that the CES was administered 4-10 years apart from the midlife factors. Nonetheless, we are limited in our ability to make causal inferences regarding the observed associations between early experiences and midlife factors, and our findings should be interpreted with this important consideration in mind.

Other known limitations of retrospective measures of early experiences include potential bias by inaccurate or unreliable recall, current mood, or a lack of endorsing negative events. An earlier study of the CES provided evidence for the temporal stability of item response in adults (Yancura & Aldwin, 2009). Inclusion of items on positive, neutral, and negative experiences likely reduced respondents’ discomfort in endorsing adverse experiences. Measures of childhood SES were obtained at NAS entry in the 1960s, which alleviates some concern regarding memory bias. A recent meta-analysis found modest overlap between individuals who endorsed childhood maltreatment prospectively vs. retrospectively (Baldwin, Reuben, Newbury, & Danese, 2019), highlighting the importance of considering potential differences in risk and resilience pathways for these individuals.

Our sample was all-male, predominantly Caucasian, relatively healthy, and born in the first half of the 20^th century. Thus, the distribution of their early experiences and midlife psychosocial factors might not generalize to women, ethnic minorities, more disadvantaged groups, and other birth cohorts. Furthermore, we observed differential survival functions across the follow-up period by age, but our study lacks statistical power to detect age differences in the associations of interest. More work is needed to evaluate the reproducibility of the current findings in larger and more diverse samples, and to test potential moderators.

Importantly, selection and survival effects in our study design might result in underestimating the associations of interest. Because NAS men were selected at entry for good health, this likely restricts variance in longevity. Men whose lifespan was significantly curtailed by early adversity might have died or dropped out prior to CES administration. Left-censoring can diminish statistical power to detect the effects of early experiences on longevity.

Conclusion

Despite these limitations, our findings contribute to a burgeoning literature on how early experiences cast a long shadow over lifespan health. Research on childhood precursors of health disparities has grown tremendously in the last two decades, yet questions on the underlying mechanisms remains largely unanswered (Shalev, Heim, & Noll, 2016). We were able to advance knowledge on pathways by which specific dimensions of early experiences contribute to longevity in later life.

Substantively, our finding regarding stress continuity as a precursor to reduced lifespan informs theory on how the effects stress exposure across developmental stages might combine to influence later-life health (Ferraro & Morton, 2018). Findings on optimism and life satisfaction as mediators also add to a growing literature on interventions aimed at reversing or limiting the deleterious impact of early stressors by boosting adulthood well-being (e.g., Mohammadi et al., 2018). While we only assessed mediators at single occasions and our model represents a simplified version of the sequenced chains linking early experiences to later-life health, our approach clearly illustrates some of the substantive questions and a viable analytic approach to conduct this type of inquiry. As more researchers are able to investigate the processes by which early circumstances influence health across the lifespan, we will build a more complete understanding of these issues and provide tools for more effective and targeted prevention strategies.