A longitudinal study shows stress proliferation effects from early childhood adversity and recent stress on risk for depressive symptoms among older adults

Abstract

We evaluated whether the effects of recent stressful life events (SLEs) and early childhood adversities (ECAs) on depressive symptoms are consistent between men and women and across older age, and whether there was evidence for: stress sensitization, whereby the psychological impact of SLEs is greater for individuals with ECAs compared to those without; or stress proliferation effect, whereby those with ECAs are more likely to report more SLEs than those without ECAs to effect depressive symptoms. ECAs, SLEs in the past two years, and current depressive symptoms through a modified CES-D were obtained from 11,873 individuals participating in a population representative study of older adults, yielding 82,764 observations. Mixed effects regression models on depressive symptoms were constructed to control for multiple observations per participant and evaluate within person effects over time, thereby reducing bias from reverse causation. Results suggest a stress proliferation effect, and do not support stress sensitization. ECAs contribute to vulnerability for depressive symptoms, with a dosage effect for each additional ECA. Recent SLEs result in greater depressive symptom risk, with stable effects over age and dosage effects for each additional SLE that were smaller than effects of ECAs among men, but not women. Belonging to an ethnic minority group, having less education, and less household income at baseline were associated with greater depressive symptom risk. Findings suggest the importance of addressing early childhood adversity and sociodemographic factors, among at-risk older adults to mitigate lifecourse stress proliferative processes and thereby reduce disparate risk for depression in older age.

Introduction

Depressive symptomatology among older adults garners short-term and long-term consequences, including earlier onset of physical disability (Mendes de Leon & Rajan, 2014), immediate social and cognitive impairment, earlier mortality (Chapman & Perry, 2008), and increased societal costs in terms of lost productivity and healthcare expenditures (Katon et al., 2003; Vasiliadis et al., 2013). Across all age groups, the positive relationship between recent stress and more depressive symptoms has been relatively consistent in the literature (Blazer, 2003; Djernes, 2006; Kraaij et al., 2002). Across the lifecourse, stressors and adversities experienced in childhood and adolescence have also strongly associated with increased risk for depression in adulthood (S.R. Dube et al., 2003; Green et al., 2010; Hammen et al., 2000; Kessler, 1997; Kessler et al., 2010; Li et al., 2016). At the same time, early life adversities may predispose individuals to experience more stress in adulthood and may have profound effects on the way individuals react to stress as adults (Hammen et al., 2000; McLaughlin et al., 2010). Such complexities have compelled many researchers to engage a lifecourse perspective in examining how these early life adversities set the stage for both stress and depression later in life, especially because coping and reactivity to stress may change with increasing age. Older age is also a time when there are increasing insecurities related to declining health and functionality, limited financial resources, and a diminishing social circle. Therefore, potential differences in age-related circumstances and stress reactivity could place older individuals at even greater risk for affective disorders and depression. In the present study, we incorporate a lifecourse approach to evaluate mechanisms by which early childhood adversity and later life stress events interrelate and are associated with depressive symptomatology in older age.

Typically, to be diagnosed with depression, one’s degree of symptomatology needs to meet a threshold of clinical criteria. Yet, it has been shown that depressive symptoms at the subthreshold level are associated with factors similar to clinical depression, including impairment in physical functioning, sleep, and concentration; poor appetite or overeating; fatigue; lower self-esteem, and poorer self-rated health (Chapman & Perry, 2008; Hybels et al., 2001). Furthermore, there are similar increases in costs and utilization of healthcare services for those with diagnosed depressive disorders and subthreshold depressive symptomatology (Katon et al., 2003). Thus, the presence of subthreshold symptoms warrants attention when characterizing the affective experiences of older adults.

Depressive Symptomatology with Aging

Trends for depression and depressive symptomatology through adulthood have been characterized previously wherein some studies show no age trends; however, more consistently, studies have indicated there is an increase in symptoms with older age (see review by Djernes, 2006). For example, when comparing prevalence rates of depressive disorders for individuals aged 75 to 79 to those in the 85+ age group, rates tend to increase by 20 to 25%, and continue to increase in the 90+ age group by 30 to 50% (Luppa et al., 2012). Some degree of the higher prevalence with age can be explained by factors associated with general aging, such as more physical and cognitive impairment, decreased income, loss of close social contacts, a higher proportion of women (since women generally experience later mortality), changes in marital and employment status, and sleep disturbance (see reviews by Blazer, 2003; Cole & Dendukuri, 2003; Djernes, 2006; Jorm, 2000). The majority of studies in this area have conducted correlational analyses (see meta-analysis by Kraaij et al., 2002) with few longitudinal examinations (A. Fiske et al., 2003; Amy Fiske et al., 2009). Single time-point evaluations, particularly of younger samples, may not capture later life manifestations of earlier life stressors. For example, military veterans exposed to combat-related trauma are at increased risk to develop mental health problems in older age even if they show no ill-effects at earlier ages (Davison et al., 2016).

With regard to age and gender differences, there is an extensive literature associating greater risk of depression with women versus men. This was reported among samples age 65 and older from a systematic review of 122 studies, published between 1993 and 2006 (Djernes, 2006). The report is consistent with research on younger adults, between the ages of 18 to 65, as reported in a systematic review of 32 studies, published between 1993 and 2004 (Cavanagh et al., 2017). Whether the gender differences in older adulthood are partially attributable to effects of early childhood adversity is unclear. For example, a prior meta-analysis among adults age 18 and older found that while there were stronger associations between adult depression and childhood maltreatment among women than men, the differences between gender were not statistically significant (Gallo et al., 2018). A closer examination of childhood adversity as a potential contributor to differential outcomes by gender for depression in older age is still needed.

Early Life and Later Life Stress May Combine to Effect Later Life Depressive Symptoms

There is substantial evidence in the literature finding that early childhood adversity associates with a wide range of outcomes in later life, from mortality and health outcomes (Blackwell et al., 2001; S.R. Dube et al., 2003; Hayward & Gorman, 2004; McCrory et al., 2015; Montez & Hayward, 2014) to alterations in gene expression in older adulthood (Cecil et al., 2016; Levine et al., 2015). As such, early adversity may serve as a general diathesis for elevated depression risk later in life (S.R. Dube et al., 2003; Green et al., 2010; Hammen et al., 2000; Kessler, 1997; Kessler et al., 1997; Kessler et al., 2010; Li et al., 2016). Given these long-arm effects, to better understand potential disparities on health outcomes in older ages, it is necessary to take a life-course perspective. This encapsulates the idea that later life health results from a long-arm effect of a range of childhood experiences that act in combination with adverse experiences in adulthood (Ben-Shlomo & Kuh, 2002).

Taking a life-course approach means there are many possible mechanisms through which earlier and later life adversity act in combination to cumulatively influence depressive symptoms, as there could be additive or multiplicative effects. Furthermore, when viewing stress as a process, because the process may unfold over long periods of time, it can be difficult to distinguish how the stress proliferation effect manifests over the life-course. Thus, we are guided by two main hypotheses on mechanisms for stress processing.

One hypothesis proposes an amplification effect from early life stress on reactivity to later life stress. This Stress Sensitization Hypothesis (SSH) was originally formulated to propose that experiences with stressors and depressive episodes could leave lingering effects at the level of gene expression, thereby eliciting neurobiological changes, and leave individuals more vulnerable to subsequent affective response (Post, 1992). SSH has been adapted to propose more generally that profound early life stress serves as a pre-existing vulnerability such that an individual’s reaction to later life stress would multiplicatively amplify their risk for depression compared to those without that vulnerability (Hammen, 2005; Pesonen et al., 2007). This effect of early life adversity on subsequent depression has been studied predominantly, and found support, among samples studied in adolescence or middle adulthood (Gillespie et al., 2009; McCrory et al., 2015; Pesonen et al., 2007; Taylor, 2010). Little work has been done to test whether the SSH applies across the life-course into older adulthood.

An alternative hypothesis to explain how earlier life adversity can affect depression is through the stress proliferation effect. Stress proliferation describes a process by which secondary stressors arise from having experienced primary stressors (Pearlin et al., 1981). This phenomenon was observed initially as a mechanism by which initial stress that was related to an individual’s social role gave rise to additional, or proliferated stress in different social domains (Pearlin et al., 1997; Pearlin et al., 1981). The concept has since been further adapted for relevance to health-related disparities such that the exposure to proliferated stress and primary stress need to bear some association with inequality in the health outcome (Pearlin, 2010; Pearlin et al., 2005). This hypothesis can therefore be applied to examine the dynamic in which individuals exposed to earlier life adversity are at greater risk for exposure to later subsequent adversity (Hammen, 2005; Pearlin et al., 2005; Pearlin & Skaff, 1996). An example of this in context from prior research is from a longitudinal study where individuals who had experienced childhood abuse and neglect, and were followed-up two decades later, were significantly more likely to have experienced deleterious life events in adulthood compared to those who did not have those childhood experiences (Horwitz et al., 2001). Thus, if true, this stress proliferation phenomenon may place individuals with trauma histories at greater risk for adverse outcomes, including depression, due to more occasions of stressful events through the lifecourse, from childhood to older adulthood (Hammen, 2005).

There is a fine-grained distinction between the two hypotheses. The stress sensitization hypothesis rests on the notion that early childhood adversity creates a vulnerability whereby an individual’s reaction to a subsequent stressor would be more pronounced in the outcome compared to those without the early experience. The stress proliferation hypothesis posits that the vulnerability is manifested by the individual experiencing more subsequent stressors to influence the outcome through a cumulative process. Longitudinal data is required to evaluate these hypotheses about lifecourse stress effects.

Reciprocal Bias for Depressive Symptoms

One major challenge in assessing etiological processes for depression within the same individuals is the potential reciprocal pattern of effect has been observed over time, whereby depressive symptoms themselves can influence exposure to and the occurrence of subsequent stressors (A. Fiske et al., 2003; Meeks et al., 2000). Most studies on this dynamic (Hammen, 1992) found that the stress effect on depression did not arise from the occurrence of discrete events, rather it was attributable to the occurrence of prior life interpersonal events (see review by Hammen, 2005). Such reporting bias particularly plagues studies with cross-sectional designs, which can only answer the question: Are people who experience more of more stressful life events the same people who have higher depressive symptoms? By using a longitudinal design, we can isolate the impact of recent stress on depressive symptoms by measuring fluctuations in depressive symptoms over time within the same individual, holding constant many relevant unmeasured factors (genetics, socioeconomic status, childhood adversity). Although it does not provide evidence of a causal relationship, this approach allows us to answer the question: After experiencing distressing life events, do individuals have increased depressive symptomatology compared to their baseline levels?

Current Study

With a lifecourse approach, we evaluate alternative hypotheses on how early life adversity and later life stress combine to effect risk for depressive symptoms; and whether relationships between stress and depressive symptoms are consistent across age for the same individuals. To evaluate stress sensitization, we ask the question: Is the effect of subsequent stress modified by the effect of early childhood adversity such that individuals experience differentially more depressive symptoms compared to those who did not experience early childhood adversity? To evaluate stress proliferation, we ask the two-part question: Do those who experience early childhood adversity also experience more secondary stressors over time; and do both independently contribute to more depressive symptoms? We address these by taking advantage of the availability of prospectively collected data from a large U.S. population-representative cohort of ethnically diverse, men and women aged 50 and older, with recent stress and depressive symptoms assessed every two years across 16 years.

Materials and Methods

Participants

Data are from the Health and Retirement Study (HRS, Juster & Suzman, 1995), a nationally representative sample of households of Americans aged 50 and over in the contiguous United States (Kapteyn et al., 2006). HRS has a panel design, with new cohorts enrolled every few years. The present study uses data collected in HRS waves 2 (1994) through 10 (2010) from participants aged 50 and older (i.e., excluding data from a small number of spouses younger than 50). Wave 1 (1992) data were excluded because the depressive symptom measure differed from that used subsequently. Depressive symptomatology and SLE data were obtained at each wave by interview; childhood adversity items were administered in self-report surveys given in 2006, 2008 or 2010. The analyses include 11,873 participants, for which DepSx and SLE data were available for between 6–9 occasions per individual (median = 7) with a total of 82,764 data points. The age range was 50 to 93 at the first assessment (mean = 58.3; SD = 7.5), with 60.3% being female. Through self-identification, 78.2% of the sample was Non-Hispanic White, 11.7% African American, 8.0% Hispanic, and 2.1% in other groups including American Indian, Alaskan Native, Asian, and Pacific Islander.

Measures

Depressive Symptoms (DepSx).

Participants were asked to rate their recent depressive symptoms using a modified version of the well-known Center for Epidemiologic Studies-Depression (CES-D) inventory (Radloff, 1977; Zimmerman & Coryell, 1994). The HRS uses 8 of the original items selected for their psychometric properties and to assess the continuum of depressive symptoms (Kohout et al., 1993). Scores demonstrate similar construct and external validity to those based on the original inventory (Steffick, 2002; Turvey et al., 1999). The distribution of the sum score, calculated by summing the binary endorsement for each item similar to previous studies (Demakakos et al., 2013; Langa et al., 2004; Zivin et al., 2010), is highly positively skewed so our analyses use ordinal categories to represent levels of symptomatology. It is common to define clinically significant depressive disorders using a threshold cut-off, such as 4+ symptoms in the HRS. However, prior work suggests that subthreshold symptomatology is an important indicator of impairment and disability (Chapman & Perry, 2008; Hybels et al., 2001; Katon et al., 2003). Thus, similar to prior studies (Arpawong et al., 2016; Cuijpers et al., 2004; Poulin et al., 2005), we use four ordered categories to indicate lower to more elevated symptom levels based on 0, 1–2, 3–4, and 5–8 items endorsed. We then model these depressive symptom levels using a continuous variable.

Recent Stressful Life Experiences (SLEs).

SLEs selected for this study were chosen because of the prior research on their depressogenic effects (Kendler et al., 1999; Kessler, 1997; Monroe et al., 2009) and that they were repeatedly assessed in the HRS such that we could characterize their occurrence within each two-year assessment period. Participants were asked at each wave if they had experienced a list of 14 specific life events since the last wave, two years previously. These included three interpersonal SLEs: the respondent’s divorce or separation, spouse death, and parent death. Eight medical SLEs code for whether the respondent had experienced a nursing home stay or moved to a nursing home, acute medical event (i.e., heart attack, stroke, cancer, lung disease, diabetes), had become disabled, or was hospitalized. Assessment of disability was based on reporting the date and month when the respondent had become disabled. Three financial SLEs included income shock, asset shock and unemployment. Income shock was defined as a reduction in total household income of 30% or more since the prior wave. Asset shock was reduction of 30% or more in household wealth (excluding secondary residence) since the prior wave. Unemployment status was defined as not working but actively seeking work. Detailed descriptions on variables and calculations for financial items are available (RAND, 2014). For each assessment wave, SLEs were summed to create a total count variable representing all SLEs that had occurred in the prior two years as well as count variables for each of the three types of SLEs. Because few respondents experienced more than 2 SLEs in a given period, primary analyses were based on ordered categories of 0, 1, 2, and 3 or more. We also conducted sub-analyses using count variables for the three types of SLEs.

Early Childhood Adversity (ECA) items were selected for several reasons: they captured major childhood events occurring prior to the age of 18 that indicate challenges within the home environment, have been widely studied for their relationship with adulthood psychiatric and health problems (Shanta R Dube et al., 2001; S.R. Dube et al., 2003; Felitti, 2002; Felitti et al., 1998), and for evidence of their long-term cumulative effects on adult psychiatric problems, which was part of the rationale for their inclusion in the HRS (Turner & Lloyd, 1995). ECAs were indexed by six binary variables reflecting parent death, physical abuse by parents, parent substance abuse problems, maternal neglect, displacement from home, and financial hardship. Parent death prior to respondent’s age 18 was coded using the respondent birthdate and the month and year parents were reported to have died. Physical abuse and parent substance abuse were assessed with yes/no items (e.g., “Before you were 18 years old, were you ever physically abused by either or your parents?” or “…did either of your parents drink or use drugs so often that it caused problems in the family?”) (Krause et al., 2004). Maternal neglect was coded from three items asking about time and attention from the mother; effort the mother put into making sure one had a good upbringing; and how much their mother taught them about life (Rossi, 2001). Response options for these items were: a lot, some, a little, not at all. Individuals were coded as positive for maternal neglect if they responded “not at all” to all three questions. In the HRS, specific items were asked for their occurrence prior to the age of 16. Displacement from home was based on a positive response to one item on having to move to a different place due to financial difficulties prior to age 16. Financial hardship was based on a yes response to either of two items asking about whether prior to age 16, the family received help due to financial difficulties or whether there was a period of several months or more when the father had no job. Prior reports have shown good reliability for this approach of retrospective reporting of childhood adversity and have shown that such retrospective reports tend to have high concordance over time and with informant reports (Bernstein et al., 1994; Dill et al., 1991). Because the 98.9% of men and 98.7% of women reported between 0 to 3 ECAs of the 6 possible, we recoded ECAs to a range of 0 to 3+.

Covariates.

Because this paper was aimed at characterizing within-person effects (vs. comparisons between different people), covariates were selected for inclusion based on whether there was a priori evidence for their expected effect on the relationship between SLEs and DepSx over time, within individuals. For instance, prior studies have indicated that mental health outcomes differ across age for ethnic minorities compared to Non-Hispanic Whites (Jackson et al., 2010; Keyes, 2009). Covariates for the study included age at each assessment wave, gender, race/ethnicity, medication use, history of childhood depression prior to age 16, respondent’s baseline socioeconomic status (education in years attained, and total household income categorized into quintiles because of the wide range reported of $0 to over $2 million), marital status (including married and non-married live-in partner), smoking history (ever/never), and physical functioning, including activities of daily living (ADLs) and instrumental ADLs (IADLs). Race/ethnicity was dummy coded for self-identification as Non-Hispanic White (reference), Black/African American, Hispanic/Latinx, or Other (as coded by HRS to capture other groups with smaller sample sizes). Medication is included this as a time-varying covariate for each timepoint, as a binary (yes=1, no=0) variable. This was coded from an item at each wave in HRS that asks whether the respondent takes tranquilizers, antidepressants, or pills for nerves. ADLs were calculated as a sum of five yes/no items asking about difficulty with bathing, eating, getting dressed, walking across a room, getting out of bed. IADLs were calculated as a sum of three yes/no items asking about difficulty with using the telephone, taking medications, and handling money. Both ADLs and IADLs are widely used assessment indexes for physical functioning in older age (Chan et al., 2012).

Statistical Analyses

Study hypotheses were tested using mixed effects multiple regression models for the within-person, repeated measures design (PROC MIXED; SAS, 2013). Mixed effects models allowed us to include both person-level (e.g., childhood adversity, education, race/ethnicity) and time varying (e.g., recent stress, age at assessment, medication status) variables. Quadratic age was included to adjust for non-linear effects of age on DepSx. First, we tested for the evidence of differing relationships between stress on DepSx by gender by entering two interaction terms between gender*ECAs (interaction b=−0.006, p=0.78) and gender*SLEs (interaction b=−0.014, p=0.006) into the model and found evidence for differential effects of SLEs by gender. Based on this and prior literature regarding DepSx differences between men and women, final models were stratified by gender. The SSH was tested by including an interaction between SLEs (total number or type) and ECA into models stratified by gender.

Because it was unclear whether the effect of each additional SLE or ECA, in the model conferred an equal effect on the dependent variable, we implemented a modeling approach that would enable us to evaluate proportionality, or whether the effect of a predictor was equal across all levels of the dependent variable. To also retain the ordinal structure of the independent variables, we used thermometer coding (Masters, 1993), which has been used similarly for count variables (Gyurak et al., 2013) and in evaluating the effect of a number of stress events on depression (Kendler et al., 2004). For this study, we coded ECAs using 3 dummy codes, eca1-eca3, to indicate whether respondents experienced 0, 1, 2, or 3+ ECAs. If respondents experienced no ECAs, all three were coded as zero. If one ECA was reported, then eca1 was coded as 1; if two ECAs were reported, then eca2 was also coded as 1; if three or more ECAs were reported then eca3 was also coded as 1. Thus, we could calculate the degree to which each additional stressor (e.g., zero ECAs to one, one ECA to two, two ECAs to three or more) contributed, proportionally or non-proportionally, to greater risk for DepSx. We implemented the same approach for recent SLEs.

Results

Sample Characteristics

Table 1 displays sample characteristics based on each individual’s first assessment, including the distribution of levels of DepSx, number of recent SLEs, and number of ECAs, and all covariates, stratified by gender. Women and men had similar levels of childhood adversity and recent SLEs, but women reported more DepSx (χ²=134.7, p<0.0001).

Table 1.

Baseline Characteristics of the Study Sample (age≥50 between 1994–2010, n=11,873)

	Men	Women
	N=4,715 (39.7%)	N=7,158 (60.3%)
Variable	mean(SD) or %	mean(SD) or %
Sociodemographic variables
Age: M (SD)	58.85 (7.28)	57.93 (7.58)
Race/ethnicity
Non-Hispanic White	80.6	76.6
African American	9.8	12.9
Hispanic/Latinx	7.6	8.3
Other	2.1	2.1
Education in Years	12.96 (3.19)	12.62 (2.88)
Household Income	2.30 (1.37)	1.92 (1.44)
Married (or live-in partner)	4.7	4.3
Health Covariates
Medication Use	3.0	5.8
Depression in Childhood	1.7	2.9
Smoking Status (ever)	68.0	47.7
ADL Difficulty	0.08 (0.42)	0.12 (0.51)
IADL Difficulty	0.05 (0.27)	0.05 (0.27)
Depression in Childhood	1.7	2.9
Depressive Symptoms (DepSx)
0	56.8	48.4
1–2	30.0	31.7
3–4	7.7	10.1
5+	5.6	9.9
Early Childhood Adversity (ECA)
Any	41.4	41.5
Neglect by parents	1.9	2.1
Parent died	10.2	10.9
Financial hardship	8.2	7.6
Physical abuse	6.1	7.8
Displacement from home	19.1	16.4
Parent substance abuse	15.8	16.7
Recent Stressful Life Events (SLEs)
Any	41.2	42.4
Interpersonal	10.4	11.6
Medical	19.1	15.6
Financial	36.8	40.9

Note. M=Mean. SD=standard deviation. Age range was 50–93 years at baseline (1994–2010). The range for early childhood adversities was 0–6. The range for recent SLEs was 0–14 for total number, 0–3 for interpersonal, 0–8 for medical, 0–3 for financial SLEs. ADL=Activities of Daily Living. IADL=Instrumental ADLs.

It is important to have sufficient within-person, between-occasion variation in SLEs to isolate their impact from person-level characteristics. The intra-class correlation (ICC) across waves for recent SLEs was r=0.06, meaning only 6% of variability is accounted for by person-level factors (or, between person), while 94% is accounted for by occasion-specific factors (including measurement error); thus, the sample provides ample variation for our analyses. The across-wave ICC for DepSx was r=0.48, indicating about half the variation is person-specific (e.g., could be due to early childhood experiences, gender, race/ethnicity, education) and half is occasion-specific, including wave-specific factors of stressful life events.

Main Effects of Childhood Adversity and Recent Stress on Depressive Symptoms

As shown in Table 2, having ECAs was associated with increased risk for more DepSx for both men and women, with each additional ECA reported being associated with a 0.07 and 0.06 increase in DepSx level, respectively, accounting for all covariates. Each additional interpersonal, medical, and financial SLE associated with an increase in 0.16, 0.06, and 0.03 DepSx level for men, respectively, and 0.21, 0.08, and 0.02 increase in DepSx level for women, respectively. From the model variance-covariance matrices, we calculate the ICC and found that 40% and 37% of the variability in DepSx was accounted for by person-level factors (between person), for men and women respectively, and 60% and 63% was accounted for by wave-specific factors (within person).

Table 2.

Effects of the Number of Early Childhood Adversities (ECAs) and Recent Stressful Life Experiences (SLEs) on Depressive Symptoms for 11,873 individuals with 82,764 observations across nine waves of data (1994–2010)

	Men			Women
Variable	b	SE	p-value	b	SE	p-value
Intercept	0.531	0.017	<.0001	0.714	0.012	<.0001
Age	−0.002	0.001	0.0004	−0.002	0.001	0.0035
Age*Age	0.001	0.00005	<.0001	0.0003	0.00004	<.0001
Medication Use (N=0,Y=1)	0.423	0.024	<.0001	0.390	0.016	<.0001
African American (vs. ref)	0.123	0.030	<.0001	0.124	0.023	<.0001
Hispanic/Latinx (vs. ref)	0.125	0.035	0.0003	0.276	0.030	<.0001
Other (vs. ref)	0.215	0.063	0.0006	0.188	0.054	0.0005
Education	−0.069	0.007	<.0001	−0.090	0.007	<.0001
Household Income	−0.065	0.007	<.0001	−0.075	0.006	<.0001
Married (N=0,Y=1)	0.057	0.045	0.2069	0.084	0.041	0.0417
Smoke (N=0,Ever=1)	0.037	0.019	0.0492	0.065	0.015	<.0001
Difficulty with ADLs	0.229	0.022	<.0001	0.243	0.016	<.0001
Difficulty with IADLs	0.272	0.034	<.0001	0.237	0.030	<.0001
Childhood Depression (N=0,Y=1)	0.538	0.068	<.0001	0.469	0.047	<.0001
Number of ECAs	0.066	0.010	<.0001	0.063	0.009	<.0001
No. of Recent SLEs: Interpersonal	0.163	0.014	<.0001	0.207	0.012	<.0001
No. of Recent SLEs: Medical	0.056	0.007	<.0001	0.081	0.006	<.0001
No. of Recent SLEs: Financial	0.032	0.006	<.0001	0.024	0.006	<.0001

Notes. Number of observations for N=4,715 men is 32,232; for N=7,158 women is 50,532.

b = standardized coefficients. SE = Standard error of the coefficient.

Race/Ethnicity reference is Non-Hispanic White. ADLs = Activities of Daily Living. IADLs = Instrumental ADLs.

In addition to the main effects of ECAs and SLEs, all covariates except for marital status significantly contributed to greater risk for DepSx among both men and women. All socioeconomic factors, including belonging to any of the race/ethnic minority groups when in comparison to being Non-Hispanic White, lower educational attainment, and lower household income at baseline were associated with greater risk for more DepSx. Having had a history of depression, prior to the age of 16 was also significantly correlated with more DepSx in older age as was having a history of smoking, baseline marriage status, and more difficulties performing activities of daily living. Taking medication was positively associated with more DepSx possibly due to the question capturing use of medications for other purposes, not exclusively for depression.

When using thermometer-coding to test whether the effect of each additional stressor, ECA or SLE, associated with progressively more DepSx, we found a consistent pattern. As shown in Figure 1, each additional recent SLE associated with a marginal increase in DepSx level. Similarly, each additional ECA associated with a marginal increase in DepSx level. However, there was a significantly greater magnitude of effect for each additional ECA compared to each additional SLE particularly for men. Generally, the marginal increase in DepSx levels for women correlated with each additional ECA or SLE was higher than for men when experiencing 1 to 3 additional stressors, and was similar to men when experiencing 4 or more stressors.

Figure 1.

Dosage effects were found for each additional early childhood adversity (ECA) and recent stressful life event (SLE) and risk for higher depressive symptom (DepSx) level. ECAs had larger effects for each additional ECA reported compared to recent SLEs, with evidence for non-linear polynomial effects on DepSx for both types of stressors. Marginal effects of additional stressors on DepSx were generally higher for women than men, and significantly higher for men experiencing additional ECAs compared to SLEs.

Age Differences in the Effects of Childhood Adversity and Recent Stress

We evaluated whether increasing age moderated the effects of recent stress and childhood adversity by repeating the analyses, with models adjusted for all covariates, for individuals stratified in 5-year age bands (50–54, 55–59, 60–64, 65–69, 70–74, 75–79, 80–84). We restricted analyses to participants with at least 10 years of follow-up data and excluded those 85 and older to reduce potential confounding of age and attrition. Because the effect of age (i.e., whether it magnifies or reduces the effect of stress) might be non-linear, we treated age group as a nominal rather than an ordered variable.

For both genders, the effect of childhood adversity on DepSx levels was constant (men: b=0.08, p<0.0001; women b=0.05, p=0.01) across age bands with no evidence for an interaction between age and childhood adversity (interaction for men: b=−0.001, p=0.90; for women: b=0.005, p=0.19). The effects of recent SLEs on DepSx did not differ by age band for women (interaction: b=−0.002, p=0.43). There was some indication that the effects of recent SLEs on DepSx differed across age bands for men (interaction: b=−0.005, p=0.04), such that there would be an overall decrease in risk for DepSx with age attributable to experiencing more SLEs; however, the result did not survive the multiple testing correction criteria for testing four interactions (cut-off of p=0.013).

Stress Sensitization: Is Childhood Adversity Associated with Greater Reactivity to Recent Stress?

When modeling the effect of the total number of recent SLEs—combining interpersonal, medical, and financial SLEs—each additional recent SLE significantly associated with a 0.04 and 0.06 increase in DepSx level for men and women, respectively (p’s <0.001). The interaction between childhood adversities and number of recent SLEs was not associated with DepSx for either men (b=0.0016, p=0.69) or women (b=0.0006, p=0.88).

Because specific types of recent SLEs (those involving bereavement, loss, or separation) are more relevant for eliciting DepSx, particularly for women (Hammen, 2005; Tennant, 2002), we also tested whether there was a sensitization effect dependent upon the SLE type. There was no evidence of sensitization for recent interpersonal SLEs (men: b=0.030, p=0.29; women: b=−0.036, p=0.12) or financial SLEs (men: b=0.003, p=0.78; women: b=−0.017, p=0.13), but some indication of a sensitization effect to recent medical SLEs among men (b=0.026, p<0.05) and women (b=0.025, p<0.05). This finding did not surpass a Bonferroni corrected p-value for testing three interactions within each gender (at p=0.017).

Because other studies that found an SSH effect conducted evaluations among younger samples (Hammen et al., 2000; McLaughlin et al., 2010), we tested whether the SSH holds true for younger individuals in the HRS. We constructed models by restricting the age range to those who were between 50 to 70 years old (mean=60.84, SD=5.45), adjusted for all covariates. The mixed-effect models were run separately for 6,400 women with 35,667 observations and 4,154 men with 22,324 observations. The interaction test for any ECA*SLEs was not significant for women (b=0.002, p=0.82) or men (b=0.008, p=0.36).

Does Earlier Life Stress Beget Later Life Stress?

To assess the stress proliferation effect, we ran a mixed-effects regression analysis testing whether the presence of ECAs predicted recent SLEs (coded as 0, 1, 2, 3+ SLEs) using all waves of data, adjusting for age at each wave, age*age, and race/ethnicity group. With gender-stratified models, we found a significant relationship between ECA history and the number of recent SLEs reported among men (b=0.015, p=0.04) and women (b=0.027, p<0.001). To assess whether this relationship was the same for all types of recent SLEs, we opted to use chi-square tests because with the distribution of counts for each type of SLE per individual varying over time, we could preserve the count structure and assess overall differences with all available information per person. We found that those with a history of ECAs report significantly more recent SLEs than those without for medical SLEs among both genders (men χ²(5)=19.55, p=0.002; women χ²(5)=26.14, p<0.0001), for interpersonal SLEs for men only (men χ²(2)=12.27, p=0.002; women χ²(2)=4.80, p=0.09), and for financial SLEs women only (men χ²(3)=4.37, p=0.22; women χ²(3)=7.91, p=0.05).

Discussion

This longitudinal study using repeated measures from a large population representative study of ethnically diverse, older adults in the U.S. provides novel insights about how the long arm effect of early life adversity combines with later life stress to effect depression among older Americans. We evaluated whether effects of both types of stress on depressive symptoms are consistent between men and women and across older age, and whether there was evidence for stress proliferation or sensitization effects. Major findings of this study include that the dose, or number of early adverse experiences is highly relevant for depressive symptoms; also, early childhood adversity and recent stressful life events have unique effects on depressive symptoms in adulthood. They do not interact to amplify risk for more depressive symptoms, yet there is some indication of a stress proliferation dynamic. Among those who experienced one or more early childhood adversities, we were more likely to observe a trajectory in which they report more stressful events through later life than those without such early adversity. Additionally, while early childhood adversity contributes to vulnerability for DepSx, an effect that persists from mid- into later life to the same degree for women, that effect diminishes over age for men. The effect of each additional recent SLEs on DepSx was weaker compared to each additional childhood adversity for men although both were similar in effect on DepSx for women. Implications of this study are that addressing exposure to early childhood adversity even in older adulthood is relevant for preventing more depressive symptoms.

Effects of Childhood Adversity and Recent Stress do not Interact, but are not Independent

The present study did not find support for an overall amplification effect. One of the largest studies supporting the SSH was based on 34,653 individuals from the National Epidemiological Survey of Alcohol and Related Conditions (NESARC), with a broad age range (20 to 70+) (McLaughlin et al., 2010). Childhood adversity was assessed using items similar to those in the present study, but recent stressors included broader types of events, and the outcome was past-year major depression rather than depressive symptoms. The study found the association between recent stressors and major depression was doubled for individuals with three or more childhood adversities, compared to individuals reporting no early adversities. To evaluate whether the difference in findings were due to study design nuances in characterizing childhood adversities, we conducted post hoc analyses comparing three or more childhood adversities to none reported, but still found no interactions of adversity status with total recent SLEs or types of SLEs. It is possible that divergent findings are due to other design differences, for instance that the NESARC study included the broader and younger age range and was constructed to assess a wider range of stressors of which the HRS does not assess (e.g., changes in work responsibilities, serious problems with a neighbor, damage to one’s property, legal problems). Both the focus on more acute childhood adversity and inclusion of younger adults in the NESARC is thus likely to increase the correlation and possibly strength of interaction among reported recent life events and ECAs on DepSx.

Prior work in the HRS showed evidence for selective survival, or that adults with early adversity lived fewer total years of life compared to those who had not experienced the same adversity (Montez & Hayward, 2014). When restricting our analysis to those under the age of 70, we still found no evidence for the SSH. Thus, our lack of finding for SSH was not an artifact of age range rather, our findings taken together suggest alternative mechanisms by which early life and later life stress work relate, with respect to DepSx in our older adult sample.

Results of the present study support a stress proliferation phenomenon because we found that having ECA predicted more instances of recent SLEs, accounting for intra- and inter-individual variability that occurs across time and for both genders. Of note, both men and women who had experienced earlier life adversity also experienced more medical stressful life events in particular. This corroborates research in other samples that found exposure to earlier childhood adversity predicts more diagnoses with chronic and acute disease in adulthood (S.R. Dube et al., 2003; Felitti et al., 1998). Implications of these findings are that it may be worthwhile to examine how to steer those who have been inequitably exposed to ECAs away from the continued trajectory of adversity throughout adulthood to mitigate greater morbidity in later life.

Within Person Recent Stress Effects on Depressive Symptoms Related to Age and Gender

Our finding that recent SLEs associate with greater DepSx is consistent with prior research on this topic (see reviews by Blazer, 2003; Djernes, 2006; and meta-analysis by Kraaij et al., 2002) and extends it by capitalizing on the long follow-up period to characterize affective responses to stress within the same individuals (vs. between), from middle age and older adulthood, the repeated-measures design to account for fluctuations in depressive symptoms over many years (vs. a single-timepoint snapshot), and by examining the effect of stress within age bands among adults aged 50–93. An extension to prior findings was that in our study, there was no general trend of an attenuated effect of SLEs on DepSx with increasing ages among the same people, from age 50 to 84. Prior research has reported trends of reduced negative affect in older ages (Charles et al., 2001), partially due to the recognition of the finality of life in older age, which can compel individuals to alter social behavior to selectively optimize opportunities for more emotionally supportive and fulfilling interactions (Carstensen, 1992), which can buffer the effects of stressors that previously induced negative mood. Although data from this population-based HRS sample supports a declining number of depressive symptomatology reported with increasing age (depressive symptoms mean=1.58, SD=1.75 for ages 50–54 v. mean=1.25, SD=1.86 for ages 80–84), findings from the present study show that the observed decrease in prevalence is not attributable to differential effects of SLEs on depressive symptoms by age.

Effects of Childhood Adversity Related to Age and Gender

Across both genders, effects of having experienced childhood adversity were consistently and strongly depressogenic, consistent with prior work in other population-based samples (e.g., Kessler et al., 1997; Kessler et al., 2010; McCrory et al., 2015). An extension to prior findings was that in our study, these findings reflect a large sample of ethnically diverse individuals followed-up from middle to older adulthood and show the robust and consistent, long-arm effects of childhood adversity on DepSx through older age, similarly for both genders. Early childhood adversity did contribute to different etiologies for depression for men and women. Also, similar to recent stress, the present study found that the decrease in DepSx typically observed with older age is not attributable to any change in effect of ECAs on depressive symptoms with age.

Limitations and Causal Interpretations

This study represents a novel examination of the long-arm effect of childhood adversity on later life DepSx, assessed up to nine times over 16 years. Our within-person longitudinal design has strengths for evaluating causal effects of recent stressful events on depressive symptomatology, and the assessment of SLEs in the two years prior to assessing depressive symptoms helps diminish reverse causation (depressive symptoms causing stressful events), but our design cannot rule out indirect effects or design artifacts. To fully evaluate indirect and potentially causal effects whereby early childhood adversity influences depressive symptoms through stressful life events, longitudinal mediation analysis would be useful.

It is possible there are unmeasured common risk factors that lead to both stress exposure and DepSx, such as via correlated environments or correlated genetic liability (e.g., Kendler & Prescott, 2006) with personality or behaviors. We would expect correlated environments of childhood adversity and DepSx, to be more pertinent for studies of young adults than for our sample. The economic and family circumstances of individuals exposed to ECAs may not have changed substantially by the time they are young adults, but by late life more change has occurred. Correlated genetic liability could also produce the childhood adversity–depression association, such as if a parent’s genetic predisposition to depression results in neglectful or abusive parenting and is transmitted to their offspring. A design which can evaluate direct causal effects while holding constant genetic factors and correlated environments is studies of twins discordant for exposure to ECAs. For example, among adult female twins discordant for reporting childhood sexual abuse, the abused twin was at greater risk for adult depression, substance abuse and other psychopathology (Bulik et al., 2001), supporting a causal interpretation.

Having more severe depressive symptoms are a predictor of study attrition (e.g., suicide or other cause for mortality, or elective drop-out). Also, loss of residential autonomy (i.e., movement to institutionalized-type settings or other care facilities) results in reduced retention for this community-based study. Thus, it is possible that related to effects of a stress proliferation cascade, there is greater attrition from older individuals who succumb to health consequences from the trajectory of diminished health and well-being. Although this would result in the inclusion of healthier and less depressed individuals, including those who experience fewer SLEs, there is evidence that such attrition bias would need to be severe to impact results (Liu et al., 2010). Furthermore, we limit this issue by implementing a design that focuses on within person change, rather than age-related change between individuals.

Lastly, it is possible that the issue of reverse causation has far reaches back in time such that individuals experiencing negative affective states in older adulthood report an inflated perception of early childhood adversities retrospectively. However, if this were a significant problem in the present study, it is likely we would have found evidence for a stress sensitization effect as there would have been amplified correlations among those with both childhood adversity and more late life depressive symptoms.

Conclusions

A life course perspective in stress research that considers early life history is critical for understanding the etiology of mood disorders in later life. Such phenomenologic approaches to studying etiological factors for DepSx provides crucial insights about mechanisms by which stress affects the aging population. Clearly, there are other influences that contribute to the manifestation of depressive symptomatology, or buffer against them. Further work is needed to evaluate whether there is clearer evidence for a continued trajectory of health disparities for those exposed to early childhood adversity, or evidence for the SSH found in other studies of older adults that assess a broader range of stressor types. Those likely to experience a continued trajectory of health disparities with respect to depressive symptomatology also includes those with a history of childhood depression, ethnic minorities, individuals with lower education and those who have lower household income in mid-life. This was true for both men and women. Thus, this study emphasizes the importance of addressing early childhood adversity and important socioeconomic factors among older adults to mitigate possible life-course stress proliferation processes and address the needs of those at greatest risk for more depressive symptoms in older age.