Examining the Influence of Perceived Stress on Developmental Change in Memory and Perceptual Speed for Adopted and Non-Adopted Individuals

Abstract

The present study prospectively evaluated cumulative early life perceived stress in relation to differential change in memory and perceptual speed from middle childhood to early adulthood. We aimed to identify periods of cognitive development susceptible to the effects of perceived stress among both adopted and non-adopted individuals. The sample consisted of participants in the Colorado Adoption Project (CAP, N = 690). Structured latent growth curves were fit to four memory outcomes as well as one perceptual speed outcome, which described nonlinear change between ages 9 and 30. Both adoption status and cumulative perceived stress indices served as predictors of the latent curves. The perceived stress indices were constructed from the Brooks-Gunn Life Events Scale for Adolescents, and reflected ‘upsettingness’ ratings associated with the occurrence of particular life events during middle childhood (ages 9 to 12) and adolescence (ages 13 to 16). For memory and perceptual speed, cumulative perceived stress did not predict differential cognitive gains. However, differences in perceptual speed trajectories between non-adopted and adopted individuals were observed, with adopted individuals showing smaller gains. Although these findings provide no evidence that emergent variability in memory and perceptual speed trajectories by age 30 are explained by cumulative perceptions of stress in childhood and adolescence, further investigations regarding potential vulnerability across the lifespan are warranted.

Early life stress has the potential to impact developmental pathways toward cognitive outcomes in both adulthood and late life. Indeed, previous research examining both human and animal models suggests that chronic stress and a variety of specific stressors experienced both early in ontogeny and later in life predict poorer cognitive performance (e.g., learning, memory, and perceptual speed; Korten, Pennix, Pot, Deeg, & Comijs, 2014; Korten, Sliwinski, Comijs & Smyth, 2014; McCormick & Mathews, 2010; Naninck et al., 2015; Pesonen, et al., 2013; Stawski, Sliwinski, & Smyth, 2006; Suri et al., 2012; Vallee et al., 1999). However, human studies have been methodologically limited with respect to their use of cross-sectional designs (Kremen, Lachman, Preussner, Sliwinski, & Wilson, 2012), or reliance on retrospective recall of stress and adversity (e.g., Lovallo et al., 2013; Korten et al., 2014a). Moreover, the majority of research has focused on momentary stressors and their immediate impact on cognitive performance (e.g., changes in cognitive functioning following exposure to a particular momentary stressor). Further examination of the impacts of early life stress on changes in cognitive functioning are critical, as such changes may be associated with a variety of socio-emotional, behavioral, and health outcomes (e.g., Davidson & McEwen, 2012; Kiang & Buchanan, 2014; Korosi et al., 2012). In particular, the extent to which early life stress is prospectively associated with growth and maintenance of perceptual speed and memory abilities from childhood into early adulthood has yet to be examined. The findings from such studies may illuminate whether cumulative stress effects on processing speed and memory are general or specific throughout the lifespan, especially in regard to relative timing in light of work suggesting that declines in perceptual speed may drive declines in other domains including memory (Finkel, Reynolds, McArdle, Gatz, & Pedersen, 2003; Finkel, Reynolds, McArdle, & Pedersen, 2007; Salthouse, 2000).

Stress and Cognition

Although acute stress may be an adaptive response to environmental challenges or learning paradigms (see Bisaz, Conboy, & Sandi, 2009; Joels et al., 2006 for reviews), the long-term experience of stress negatively impacts most physiological systems (McEwen, 2002, 2007). Chronic or persistent stress has been linked to fluctuations in glucocorticoid levels (e.g., cortisol; Sapolsky, Romero & Munck, 2000; Sapolsky et al., 1990). Fluctuations in glutocorticoid levels and increased stress over prolonged periods of time have been associated with Hypothalamic-Pituitary-Adrenocortical (HPA) axis dysregulation (Sapolsky, 1999) with subsequent neurobiological consequences (see as a review Gunnar & Quevedo, 2007). Furthermore, long-term stress has the potential to negatively impact both brain structures and brain function. In particular, stress experienced in childhood and adolescence might be especially detrimental to brain regions with long postnatal development or organization periods, high density of glucocorticoid receptors, and/or brain regions that undergo neurogenesis during these developmental periods (e.g., the hippocampus, amygdala, prefrontal and frontal cortices; Joels et al., 2004; Korosi et al., 2012; Lupien, McEwen, Gunnar, & Heim, 2009; McEwen, Gray, & Nasca, 2015; Teicher et al., 2003). One recent study in particular identified negative and widespread changes to grey matter, utilizing MRI and voxel-based morphometry (VBM), in adolescent brains exposed to moderate and chronic childhood adversity (Walsh et al., 2014). Moreover, stress has been found to have negative consequences on gene expression and behavior (e.g., see Korosi et al., 2012; Navailles, Zimnisky, & Schmauss, 2010). Hence, stress experienced during these sensitive periods of development may influence cognitive abilities dependent on these regions as well as changes in these cognitive abilities later in life.

Perceptions of Stress

Empirical studies have reported complex relationships between objective (e.g., biomarkers) and subjective measures of stress in response to lab-evaluated stressors (Aschbacher et al., 2013; Fujimaru et al., 2012; Rehm et al., 2013) and in relation to individual predictors of reactivity to stress (e.g., aspects of temperament such as positive emotionality and behavioral inhibition; Mackrell et al., 2014). Nevertheless, the past decade has seen an increase in psychological research focused on perceived stress and its impact on well-being, health, and cognitive functioning (e.g., Boals & Banks, 2012; Leng et al., 2013; Scott, Sliwinski & Fields, 2013; VanKim & Nelson, 2013). The perception of stress can invoke a host of involuntary responses associated with negative outcomes, including but not limited to intrusive thoughts, rumination and physiological responses (e.g., increased cortisol levels in response to social stressors; Dickerson & Kemeny, 2004; Sontag & Graber, 2010).

Stress and Memory Associations

Currently, the majority of human research conducted regarding stress effects on memory revolves around acute (i.e., momentary) stressors and their associations with increases in memory performance (e.g., Domes, Heinrichs, Rimmele, Reichwald & Hautzinger, 2004; Payne et al. 2007). However, the limited findings that exist examining the differential impacts of stress on different memory processes, suggest that stress may differentially impact encoding and recall (Payne et al., 2007). In particular, findings from the animal literature and computational models suggest that chronic stress may interfere with, and decrease, encoding of hippocampal-dependent tasks (e.g., Aimone, Wiles, & Gage, 2009). The literature on recent events (e.g., those experienced in the last 12 months) by middle aged and older adult samples suggests that self-report of recent stress may not be predictive of delayed recall performance (e.g., Korten et al., 2014b). However, this work suggests that the types of stressors experienced, as well as the perception of earlier stress versus perceptions of stress at event occurrence may matter (e.g., Comijs, van den Kommer, Minnaar, Penninx, & Deeg, 2011; Korten et al., 2014b). Furthermore, the few published findings that do address the relationship between early life stress and memory performance tend to utilize only a single or cross-sectional measurement of memory performance. The literature has yet to address the prospective effects of early life stress on longitudinal memory trajectories despite evidence that episodic memory (i.e., explicitly remembering times, places, and events) worsens with age (Baltes & Lindenberger, 1997; Dixon et al., 2004, Schaie, 2000; Reynolds et al., 2005). Moreover, results obtained from both cross-sectional and longitudinal studies suggest gradual declines for recall memory emerge in the 30s (Salthouse, 2009).

Stress and Perceptual Speed Associations

Previous research also provides support for a stress-speed relationship, although as for memory, findings are mixed. Poor perceptual speed performance in older adults has been associated with the occurrence of particular stressful life events in the prior year, such as a substantial drop in income, increases in crime victimization, and changes in social networks (e.g., friend relocating; Rosnick et al., 2007), although illness or injury of friends or a parent, respectively, predicted better performance. This is largely consistent with a later study of older Dutch adults, suggesting a distinction between more severe experiences (e.g., death of child, grandchild) as contributing to accelerating cognitive declines versus chronic experiences (e.g., illness of relative or partner) that may contribute to better performance across time via an arousal or stimulative process (Comijs, et al., 2011). On the other hand, a recent longitudinal study of the same sample across 10 years, observed that experiencing a higher number of stressful life events, including death or illness of relatives and crime victimization, predicted poorer overall speed performance (Korten et al., 2014a). Moreover, retrospective reports of adverse childhood events (before age 18 years) predicted faster speed declines in late life, but only for those reporting higher depressive symptoms (Korten et al., 2014a). Much like the limited body of literature pertaining to perceived stress and memory, research evaluating the relationship between stress and perceptual speed has typically focused on immediate short-term stressors (e.g., Conrad, 1951; Kremen et al., 2012; Sliwinski, Smyth, Hofer & Stawski, 2006), or relied on retrospective reports. The samples in this research are also limited in size and participant populations (e.g., Öhman, Nordin, Bergdahl, Birgander & Neely, 2007), or focus on traits related to stress such as well-being (e.g., Gerstorf, Lövdén, Röcke, Smith, & Lindenberger, 2007). In order to gain a fuller life-course perspective, it is necessary to investigate the links between prospectively measured stress in early life and changes in both perceptual speed and memory, considering the relative timing of both.

Adoption Status and Cognitive Trajectories

One early life experience that may be associated with differential speed and memory trajectories is adoption, wherein both resiliency and vulnerability may be possible. Research investigating cognitive performance in both children reared by adoptive parents and by biological parents has found similar IQ scores across these two populations between the ages of 4 and 18 years (Coon, Fulker, DeFries, & Plomin, 1990; Plomin et al., 1997; van IJzendoorn, Juffer, & Klein Poelhuis, 2005), although adopted children, especially those adopted after 1 year of age, are more likely to experience learning problems and demonstrate poorer school performance (van IJzendoorn, Juffer, & Klein Poelhuis, 2005). In the Colorado Adoption Project, in which the average age of placement is 29 days, analyses of reading achievement from ages 7 to 16 suggest only slight mean advantages for children reared by biological parents as compared to the adopted children (Wadsworth et al., 2002).

Murine studies suggest that early-life separations or adoptions from family of origin may pose an increased risk for altered stress functioning and cognitive performance in adulthood (c.f., Barbazanges et al., 1996; Koehl et al., 2001). Moreover, periods of early-life separations predict poorer cognitive performance in humans as well, and to a lessor extent cognitive change between early and late adulthood (Pesonen et al., 2013). However, age at separation or adoption may be a moderator, such that very early adoption may mitigate or lessen risk of altered stress and cognitive functioning (e.g., Barbazanges et al., 1996; Koehl et al., 2001; van IJzendoorn, Juffer, & Klein Poelhuis, 2005), wherein HPA alterations returns to normative functioning with very early adoption and positive maternal behaviors are increased above what would be expected (Barbazanges et al., 1996; Koehl et al., 2001; Musazzi & Marrocco, 2016). HPA functioning in international adoptees who were adopted earlier (prior to 8 months) versus later (more than 1 year) were compared to non-adopted US children (Gunnar et al., 2009). Among these groups of children, aged 10–12 years, no differences were observed in reports of stress, deemed in the normative range; however, early adoptees showed lower HPA activity relative to non-adoptees even while accounting for growth patterns possibly reflecting processes that relate to resilience rather than vulnerability. In the same adopted and control children, at ages 12–14, smaller hippocampal volumes were observed for early and late adoptees versus non-adoptees, and was most salient for late adoptees; but even within the early adoptees, age at adoption predicted differential hippocampal volumes.

In addition, prefrontal cortex volume differences were observed for early and late adoptees versus non-adoptees, and early and late adoptees were not significantly different from one another (Hodel et al., 2015). Apart from the relatively well-known changes in hippocampus associated with normative cognitive aging (e.g., Lister & Barnes, 2009), age related change in processing speed across adulthood has been associated with structural and functional changes in the prefrontal cortex (e.g., Bartzokis et al., 2010; Eckert et al., 2010; Ekert, 2011). Such differences in hippocampal and prefrontal cortex volumes may suggest the potential for differential trajectories of memory and perceptual speed functioning. Given the previous research, we examine trajectories of cognitive performance into adulthood by adoption status, evaluating resiliency or vulnerability, which has not yet been examined.

The Present Investigation

The primary goal of this investigation was to evaluate the long-term impacts of cumulative perceived stress on cognitive change from childhood into early adulthood in the Colorado Adoption Project. Although we anticipated that adopted and non-adopted individuals would have similar cognitive performance levels, particularly in childhood, we assessed difference in cognitive change based on adoption status of the child (reared by adoptive or biological parents). We also examined the impacts of prospectively measured annual self-reports of perceived stress associated with life events on both perceptual speed and memory performance longitudinally from childhood into early adulthood, investigating differences or similarities in the timing of effects. We hypothesized that stress experienced in middle childhood and early adolescence (ages 9 to 12) would predict changes in perceptual speed and memory performance differentially compared to stress experienced later in adolescence (ages 13 to 16). We also hypothesized that higher perceived stress in early life would predict poorer perceptual speed and memory performance during emerging and early adulthood.

Method

Participants

The sample was drawn from the longitudinal Colorado Adoption Project (CAP), one of the two samples included in the Colorado Adoption/Twin Study of Lifespan behavioral development and cognitive aging (CATSLife). The CAP sample consists of 245 adoptive families and 245 non-adoptive families that were matched on age, education and occupational status of the father, sex of the adopted child, and number of children in the family (Rhea et al., 2013a,b; Petrill et al., 2004; Plomin & DeFries, 1983). Adopted individuals did not leave the hospital with their birth parents and were placed with their adopting parents on average at 29 days after birth (range = 2 to 172 days). Until placement, adopted children were in foster care (Rhea et al., 2013a). Assessments began in 1977 when the CAP probands (i.e., adopted and biological children) were one year of age and continued through early adulthood. Assessments of the full CAP sample were conducted up through the 21-year assessment. At the 16-year assessment, there was a concerted effort to retest all participants which resulted in nearly 90% of the full CAP sample continuting participation into early adulthood (Rhea et al., 2013a). However, due to funding constraints the 30-year assessment included a randomly selected subset of particpants. The analysis sample for the present study includes both adopted and biologically reared probands and their enrolled unrelated and related siblings (N = 690), and was comprised of 47.0% females and 55.9 % biological children (29 of whom are the biological children of parents in the adoptive CAP families). Participants with cognitive data and stress measures available from at least one assessment point between 9 and 30 were included. On average, the sample was 9.48 years of age (SD = 0.37) at the 9-year assessment point (see Table 1).

Table 1.

Descriptive Statistics for each Cognitive Task

Assessment Point	N	Age		CPS		PMI		PMD		NAFI		NAFD
		M	SD	M	SD	M	SD	M	SD	M	SD	M	SD
9	625	9.48	0.37	16.26	4.36	9.63	3.89	7.25	3.87	3.59	2.10	3.35	2.15
10	629	10.45	0.37	19.79	4.78	10.80	3.81	8.14	3.88	4.60	2.12	4.34	2.26
12	628	12.47	0.40	26.17	5.85	12.90	3.02	9.91	3.55	3.68	2.17	3.13	2.09
14	597	14.50	0.38	30.24	6.28	12.57	3.42	10.09	3.72	5.35	3.06	4.86	3.10
16	649	16.42	0.61	36.34	7.30	13.67	3.25	10.84	3.94	6.59	2.94	5.77	3.05
21	550	21.54	0.74	39.92	7.69	13.45	3.21	10.74	3.71	7.08	3.07	6.49	3.19
30	244	31.82	1.23	41.42	7.88	13.46	3.61	10.35	3.73	7.35	2.99	6.73	3.26

Note. Seven assessment points were included in these analyses; assessment point name corresponds roughly to the age at which participants were tested. Number of participants included at each assessment point, as well as mean and standard deviation are presented for age and all five cognitive outcomes; Colorado Perceptual Speed (CPS), Picture Memory Immediate (PMI), Picture Memory Delayed (PMD), Names and Faces Immediate (NAFI), and Names and Faces Delayed (NAFD).

Cognitive Measures

Cognitive performance was measured seven times between the 9-year and the 30-year assessments. At assessment points that corresponded with ages 9, 10 and 14, cognitive performance was measured over the telephone. Cognitive performance was measured in-person at assessment points that corresponded with ages 12, 16, 21, and 30. Analyses by Kent and Plomin (1987) suggested no evidence of cheating or parental assistance in the telephone-based protocol administrations. Three tasks from the Colorado Battery of Specific Cognitive Abilities (Kent & Plomin, 1987) were analyzed for the present study; one measure of perceptual speed, the Colorado Perceptual Speed Test (CPS), and two memory measures, Picture Memory and Names and Faces (see Appendix A). Both memory tasks were administered immediately following an encoding session as well as approximately 15 minutes later, producing 4 scores; Picture Memory Immediate (PMI), Picture Memory Delayed (PMD), Names and Faces Immediate (NAFI), and Names and Faces Delayed (NAFD). These memory and speed tasks were assessed longitudinally at assessment points 9, 10, 12, 14, 16, 21, and 30. Of the 690 participants, 26.8% had cognitive data at all 7 assessments and at least one measurement of stress, while only 2.6% had just a single assessment point of cognitive data (Table A1). The numbers of participants at each assessment along with descriptive statistics for the cognitive measures are reported in Table 1. Although the smaller samples sizes at later ages are due in part to attrition, siblings of probands were not invited to participate at some of the assessments (Rhea et al. 2013b), and the 30-year assessment included only a randomly selected subset of participants due to funding constraints. Prior to statistical analyses, three data points from CPS, four data points from PMI, two data points from PMD, and one data point from NAFI were removed as outliers based on within-assessment (±3 SD from assessment mean) and within-person criteria (±3 SD from individual mean) on their respective outcomes.

Stress Measures

The introduction of the Brooks-Gunn Life Events Scale for Adolescents (LESA; Brooks-Gunn & Petersen, 1984; Saudino, Gagne, & Becker, 2003) to the CAP survey instrument occurred at the 9-year assessment and was administered through the 16-year assessment. The LESA consists of 54 events, with both positive and negative occurrences in the family, peer, and school domains, which can be conceptualized as stressors when they are reported as upsetting (e.g., broke up with boyfriend, physical fight with friend, home broken into, sick for more than 1 week). Participants indicated whether or not an event occurred in the past year, and how upset they were in response to the event on a 7-point scale (1 = “very upset”, 4 = “neutral”, 7 = “very pleased”). For the purposes of the current study, ratings were reverse scored to reflect perceived stressfulness (“very upset” = 3, “somewhat upset” = 2, “a little upset” = 1), and events rated neutral to positive were coded as 0. Events reported at each assessment point were weighted by perceived stressfulness and summed (see Table B1 for correlations between all assessment points). Assessment points from ages 9 to 12 were summed to construct the middle childhood cumulative stress variable (STRESS_9–12; M = 85.80, SD = 39.94) and assessment points from ages 13 to 16 were summed to construct the adolescent cumulative stress variable (STRESS_13–16; M = 79.36, SD = 40.96). Due to large standard deviations and skew at individual assessments, these variables were rank-normalized and standardized (M = 0, SD = 1.00; see Table B2) to reduce any non-normality before being entered into the models as predictors. In order to account for differential numbers of stress data assessment points (one to four occasions between 9 and 12 years and between 13 to 16 years), the rank-normalization was done by number of assessment points completed (see Table B3).

Control Variables

Participant sex, parental divorce history, and the highest reported parental education and parental occupational prestige, were examined as potential control variables. At the initial CAP assessment parents provided data on child’s sex, parental education reported as years of school completed, and parental occupational prestige was calculated using National Opinion Research Center (NORC) scores (Davis, Smith, Hodge, Nakao, & Treas, 1991). Parental education and parental NORC scores were collected again at the 7-year assessment point. The highest reported parental education and parental NORC scores at baseline or the 7-year assessment was taken for adoptive or control parents; coding of the highest reported values were done at the level of the family and the same value was used for all siblings in the current analysis (N = 688 for parental education and parental NORC). At the 21-year assessment, participants were asked to report retrospectively on whether parental divorce ever occurred while they were growing up (no = 0, yes = 1). In order to maximize the sample for attrition analyses, a reported divorce from one sibling was applied to all siblings within a family, who may or may not have participated at the 21-year assessment; this procedure increased the sample size from 587 to 643. However, by employing this coding a divorce may have occurred beyond childhood for some participants and is otherwise censored for 6.81% of participants in the full sample of 690.

Statistical Procedures

Attrition analyses compared those who participated during cognitive testing in middle childhood (10-year and/or 12-year assessments), adolescence (14-year and/or 16-year assessments), and adulthood (21-year and/or 30-year assessments), with those who did not participate during each developmental period on control variables (see Table C1). Additionally, bivariate correlations between all potential control variables, adoption status, and the substantive stress predictors were also examined (see Table C2). Analyses revealed that those who participated during middle childhood had more educated parents (t(686) = 2.14, p = .033), higher parental NORC occupational prestige scores (t(686) = 2.50, $p=.013$), and were less likely to have parents that divorced (χ²(1) = 11.23, p < .001, N = 643). During adulthood those who participated had more educated parents (t(686) = 2.00, $p=.046$) and higher parental NORC occupational prestige scores (t(686) = 2.00, p = .046). Given the previously mentioned effort to retest all participants at the 16-year assessment, it was not surprising that there were no differences between participation groups during adolescence (all p > .628). Further, there were no sex differences between participation groups at any age period (all p > .226). Parental divorce history and adoption status were significantly correlated such that adopted individuals were less likely to report that their parents divorced (r = −.143, p < .001, N = 643). Additionally, reported parental divorce history and the substantive stress predictors were significantly correlated such that those with divorced parents were more likely to report higher stress (Stress_9–12: r = .168, p < .001, N = 643; Stress_13–16: r = .160, p < .001, N = 643). Participant sex was significantly correlated with Stress_9–12 (r = −.078, p = .042, N = 690) such that females were more likely to report lower stress, but sex was not significantly correlated with the other control variables, adoption status or Stress_13–16 predictors (all p > .110). Based on the attrition analyses, coupled with the limitations of the retrospective divorce history variable as mentioned above, participant sex, parental education and parental NORC occupational prestige were selected as control variables for subsequent conditional growth models.

Latent growth models were fit to the individual cognitive test data across seven assessment points for the full sample of N = 690 using Mplus 7.4 (Muthén & Muthén, 2012) employing full maximum likelihood estimation with robust standard errors that accounted for familial clustering of data (MLR). A series of age-based polynomial models were fit to the data including: (1) a linear model (centered on age 16), (2) a quadratic model (centered on age 16), (3) a spline model evaluating two linear rates over age (before and after age 16), and (4) an increment-decrement model evaluating linear gains throughout the testing ages, centered at the baseline age 9, with a second linear rate starting at age 16 that defined deviation from the first linear rate (c.f., Raudenbush and Bryk, 2002). Except as stated above, polynomial models were centered at age 16, as the last year that stress data were collected, and 16 years was the approximate age where fit was minimized when exploring spline models with centering age as a free parameter. In addition, we evaluated two sigmoid models, logistic and Gompertz, adapting existing models to incorporate individual age at each assessment point centered at age 9 years (Grimm & Ram, 2009; Grimm, Ram, & Estabrook, 2010; Newsom, 2015; Sterba, 2014). All models assumed homoscedastic Level 1 residuals (c.f. Singer & Willet, 2003). A likelihood ratio test (LRT) for nested models and AIC and BIC fit criteria for non-nested competing models were used to choose the best-fitting unconditional model. Based on LRT and fit criteria, the best polynomial model was the two-rate increment-decrement model (see Appendix Table D1); however, the Gompertz or Logistic sigmoid models fit best overall based on AIC and BIC criteria. As the Gompertz and logistic models share some similar features, we describe the Gompertz model and discuss the constraints of the logistic model.

The Gompertz indexes individual variation via three features of change, including change from a lower to an upper asymptote, age at accelerating change (maximum growth rate), and the rate of approach to the asymptote. For interpretability of the parameters, age was centered at 9 years. Given that Y is the repeated cognitive measurement across one to seven occasions for n individuals at assessment point t, the Gompertz model is defined as follows (cf. Grimm, Ram, & Estabrook, 2010; Sterba, 2014):

$$Y_{nt}=i+a_n\ast \exp (-\exp (-r_n(\mathit{cAG}{E}_{nt}-d_n)))+e_{nt}$$ [1]

where i is the lower asymptote fixed to be the same for all individuals (representing age 9 performance), a_n represents the individual total change from lower to upper asymptote which is a function of the rate of approach to the asymptote (r_n) and the age at which accelerated changes is observed (d_n) for a given individual. cAge represents individual-specific age at assessment point t minus the centering age of 9 years. Last, e_nt represents the residual at age t. The rate of growth is slower at the asymptotes and more rapid at the inflection point (d), which for the Gompertz model represents the point at which 37% of growth has occurred. Figure 1 represents a Gompertz path model (cf. Grimm, Ram, & Estabrook, 2010; Sterba, 2014) where the following weights of the a, r, d factors are constrained such that:

$$L{a}_{nt}=exp(-1\ast \exp (-M_r(\mathit{cAG}{E}_{nt}-M_d)))$$ [1a]

$$L{d}_{nt}=-M_a\ast M_r\ast \exp (-\exp (-M_r\ast \mathit{cAG}{E}_{nt})\ast ((M_r\ast \mathit{cAG}{E}_{nt}-M_r\ast M_d)\ast \exp (M_r\ast \mathit{cAG}{E}_{nt})+\exp (M_r\ast M_d)))$$ [1b]

$$L{r}_{nt}=(M_a\ast \mathit{cAG}{E}_{nt}\ast \exp (M_r\ast M_d)-M_a\ast M_d\ast \exp (M_r\ast M_d))\ast \exp (-\exp (-M_r\ast \mathit{cAG}{E}_{nt})\ast (M_r\ast \mathit{cAG}{E}_{nt}\ast \exp (M_r\ast \mathit{cAG}{E}_{nt})+\exp (M_r\ast M_d)))$$ [1c]

where M_a, M_r, and M_d reflect the fixed effects or mean a, r, and d values (see Grimm, Ram & Estabrook, 2010 for further details).

Figure 1.

Structured latent growth model of cognitive change: Gompertz. This path diagram incorporates five latent variables: i (the lower asymptote representing age 9 performance, given age is centered at age 9), a (the total change from lower to upper asymptote), r (rate of approach to the asymptote), d (age at which accelerated change is observed) and Tele (change associated with the mode of estimation, with in person testing set at 0 and testing over the telephone set at 1). The observed cognitive tasks are shown as Y9, Y10, Y12, Y14, Y16, Y21, and Y30 for the seven assessment points. The paths from the slopes to the observed scores are the age basis coefficients, (c.f., Equations 1a–1c for the Gompertz model; see also Grimm, Ram, & Estabrook, 2010; Sterba, 2014). M_i = average performance at age 9; M_a= average total change from lower to upper asymptote; M_r = average rate of approach to the asymptote; M_d = average age at which accelerated change is observed; M_Tele= average difference in performance associated with telephone administration. V_a reflects variability in the total change from lower to upper asymptote. V_r reflects variability in rate of approach to the asymptote. V_d reflects variability in age at which accelerated changes is observed. V_e reflects assessment wave variability in performance not accounted for by the growth model. C_ad represents the covariance between change from lower to upper asymptote and age at which accelerated change is observed. C_dr represents the covariance between age at which accelerated changes is observed and rate of approach to the asymptote. C_ar represents the covariance between change from lower to upper asymptote and the rate of approach to the asymptote.

The Logistic model similarly indexes change from a lower to upper asymptote (a_n) as an individual difference feature, but the rate of approach to the asymptote (r) and the age at acceleration of accelerated changes (d) or inflection, is fixed across individuals. Moreover, d represents the age at which 50% of growth has occurred. The logistic model allows for individual differences in the lower asymptote (i), representing performance at age 9. The logistic model is defined as follows (c.f. Newsom, 2015; Sterba, 2014):

$$Y_{nt}=i_n+a_n\ast (1/(1+\exp -(\mathit{cAG}{E}_{nt}-d)\ast r))+e_{nt}$$ [2]

where the loadings of factor a were constrained as follows:

$$L{a}_{nt}=1/(1+\exp (-(\mathit{cAG}{E}_{nt}-d)\ast r))$$ [2a]

Estimated parameters included fixed effects, reflecting mean growth parameters, and random effects, reflective of deviations about the mean growth parameters. All models accounted for between family variability using the CLUSTER option to obtain robust standard errors. The baseline unconditional model for participant sex, adoption status, and stress variables included time-varying level 1 predictors of mode of administration effects (in person or telephone), and practice (NAFI and NAFD only, as assessment 9 was the first wave of administration for these tasks). These time-varying level 1 variables were estimated as having a fixed effect only.

Conditional models adjusted for participant sex, parental education, and parental NORC occupational prestige. For the conditional models the analysis sample was N = 688. This covariate-adjusted model was used as a comparison against models including adoption status, and in turn adding middle childhood cumulative stress (STRESS_9–12) and adolescent cumulative stress (STRESS_13–16). Note that for the Gompterz model, STRESS_9–12 was entered as predictors of a, d, and r and subsequently STRESS_13–16 for a and r, given the temporal precedence of d in models described below. For the logistic model, neither STRESS_9–12 nor STRESS_13–16 were entered as predictors of i, given the temporal precedence of i reflecting age 9 performance. Nested models were compared using chi-square difference tests using the scalar correction factors and loglikelhoods obtained under MLR estimation (Muthén & Muthén, 2012; see Appendix D).

Results

As suggested by the descriptive statistics reported in Table 1, on average, memory and speed performance increase with age, with the pace of gains lessening into early adulthood. For all memory tasks variability in performance was relatively uniform across assessment points according to the descriptive statistics; however, variability increased with age for perceptual speed.

Unconditional Growth Models

The best fitting growth models for these data were nonlinear sigmoid models centered at age 9 (see Equations 1 and 2, Figure 1), Gompertz for CPS and PMI (see Table D1) and logistic for PMD, NAFI and NAFD. For CPS, the average score at age 9 (i) was 13.92 ​(p < .001), with an average total gain between the lower to upper asymptote (a) of 26.99 points ( $p<.001$). Moreover, the rate of approach to the asymptote (r) was .31 points (p < .001), and the age at accelerated changes (d), where 37% of the growth in CPS performance occurred, was 2.50 years from the baseline age of 9 years (p < .001), or 11.50 years (see Table 2). The adjustment for Telephone was significant and suggested that mode of administration via telephone penalized participants’ scores by 1.57 points (p < .001). The random effect variances for the change parameters a, d, and r were all significant (p < .001), suggesting further prediction of individual differences in change was appropriate.

Table 2.

Fixed And Random Effects For Best Fitting Unconditional Growth Models Across Tasks

	CPS		PMI		PMD		NAFI		NAFD
Model Parameters	Gompertz	se	Gompertz	se	Logistic	se	Logistic	se	Logistic	se
Fixed Effects
i	13.92	0.69	9.43	0.45	6.93	0.61	2.73	0.12	2.21	0.12
a	26.99	0.77	4.06	0.46	3.75	0.61	3.33	0.12	3.21	0.12
r	0.31	0.01	0.78	0.10	1.82	0.47	6.14	0.11	6.36	0.11
d	2.50	0.15	0.84	0.24	0.81	0.12	1.68	0.21	1.72	0.28
9+d	11.50		9.84		9.81		10.68		10.72
Telephone	−1.57	0.17	−0.87	0.12	−0.51	0.14	0.84	0.10	1.13	0.11
Practice	–	–	–	–	–	–	0.95	0.10	0.92	0.10
Variances
i	–	–	–	–	6.96	1.54	1.68	0.18	1.89	0.17
a	50.90	3.43	3.46	0.41	6.84	2.22	2.32	0.35	2.88	0.37
d	1.01	0.12	4.07	0.96	–	–	–	–	–	–
r	0.003	0.001	0.74	0.49	–	–	–	–	–	–
Telephone	–	–	–	–	–	–	–	–	–	–
Practice	–	–	–	–	–	–	–	–	–	–
Ve	11.14	0.48	7.44	0.30	8.90	0.27	3.75	0.14	3.75	0.14
Covariances
i with a	–	–	–	–	−4.10	0.02	1.00	0.21	1.12	0.21
r with a	−0.20	0.05	−0.42	0.32	–	–	–	–	–	–
d with a	−2.77	0.50	−1.29	0.44	–	–	–	–	–	–
r with d	−0.02	0.01	0.86	0.37	–	–	–	–	–	–
Fit Criteria
LL	−11276.14		−9964.74		−10255.32		−8693.29		−8727.49
Scale Correction c	1.11		1.14		1.00		1.28		1.30
Parameters	12		12		9		10		10
AIC	22576.27		19953.48		20528.64		17406.58		17474.99
BIC	22630.71		20007.92		20569.48		17451.95		17520.35
N-Adjust BIC	22592.61		19969.82		20540.90		17420.20		17488.60

Bold = p < .05

Italic = p < .10

Note. N = 690, Colorado Perceptual Speed (CPS), Picture Memory Immediate (PMI), Picture Memory Delayed (PMD), Names and Faces Immediate (NAFI), and Names and Faces Delayed (NAFD). Parameter i is the lower asymptote fixed to be the same for all individuals (representing age 9 performance), a represents the total change from lower to upper asymptote. Parameter r is the rate of approach to the asymptote and d is the age at which accelerated changes is observed. The rate of growth is slower at the asymptotes and more rapid at the inflection point (d), which for the Gompertz model represents the point at which 37% of growth has occurred. 9+d represents the average age at acceleration, adding back the centering age. Telephone (Mode of Administration) effects represent the change associated with testing over the telephone (In Person was the reference group dummy coded as 0). Practice represents the change associated with practice (baseline at age 9 was the reference group dummy all other assessments coded as 1).

The best unconditional Gompertz model for PMI suggested that the average score at age 9 (i) was 9.43 points (p < .001), with an average total gain from lower to upper asymptote (a) of 4.06 points (p < .001); the rate of approach to the asymptote (r) was .78 points (p < .001), and the age at accelerated changes where 37% of growth occurred (d) was .84 years from the baseline age of 9 years (p < .001), or 9.84 years (see Table 2). Mode of administration via telephone penalized participants’ scores by .87 points (p < .001). The random effect variances for the change parameters, a and d, but not r, were significant (p < .001), suggesting further prediction of individual differences in total change between the lower and upper asymptotes and age at acceleration (inflection) were appropriate, but that rate of approach (r) was effectively the same for all individuals at .78 points.

The best unconditional logistic model for PMD suggested that the average score at age 9 was 6.93 points (p < .001) with total change from lower to upper asymptote (a) of 3.75 points (p < .001); the rate of approach to the asymptote (r) was 1.82 points (p < .001), and the age at accelerated changes (d) where 50% of growth occurred was .81 years from the baseline age of 9 years (p < .001), or 9.81 years (see Table 2). Telephone was significant, suggesting that administration via telephone penalized participants’ scores by .51 points (p < .001). The random effect variances for the change parameters, i and a were significant (all p ≤ .005), suggesting further prediction of individual differences in baseline performance or total change between the lower and upper asymptotes was appropriate.

The best unconditional Logistic model for NAFI suggested that the average score at age 9 was 2.73 points ( $p<.001$) with total change from lower to upper asymptote (a) of 3.33 points ( $p<.001$); the rate of approach to the asymptote (r) was 6.14 points ( $p<.001$), and the age at accelerated changes (d) was 1.68 years from the baseline age of 9 years ( $p<.001$), or 10.68 years (see Table 2). Mode of administration via telephone advantaged participants’ scores by .84 points ( $p<.001$) and the practice effect boosted scores by .95 points ( $p<.001$). The random effect variances for the change parameters, i and a were significant ( $p<.001$), suggesting further prediction of individual differences in baseline performance or total change between the lower and upper asymptotes was appropriate.

The best unconditional Logistic model for NAFD suggested that the average score at age 9 was 2.21 points ( $p<.001$) with total change from lower to upper asymptote (a) of 3.21 points ( $p<.001$); the rate of approach to the asymptote (r) was 6.36 points ( $p<.001$), and the age at accelerated changes (d) was 1.72 years from the baseline age of 9 years ( $p<.001$), or 10.72 years (see Table 2). Mode of administration via telephone advantaged participants’ scores by 1.13 points ( $p<.001$) and the practice effect boosted scores by .92 points ( $p<.001$). The random effect variances for the change parameters, i and a were significant ( $p<.001$), suggesting further prediction of individual differences in baseline performance or total change between the lower and upper asymptotes was appropriate.

Conditional Models

As described in the methods, the variables Adoption Status, STRESS_9–12, and STRESS_13–16, were sequentially evaluated as predictors of the memory and speed trajectory parameters. None were predictive of cognitive change in PMI, PMD, NAFI, or NAFD (all p > .069; see Appendix Table D2). Consequently, the findings below resulted from model fits that include adoption status and cumulative perceived stress indices only for CPS (see Table 3).

Table 3.

Evaluating Adoption Status and Stress Predictors in Conditional Growth Models for CPS.

Model Parameters	Gompertz	se	+Covs	se	+Adopt	se	+Stress9–12	se	+Stress13–16	se
Fixed Effects
i	13.95	0.70	13.78	0.72	13.97	0.72	13.97	0.72	13.98	0.72
a	26.97	0.77	27.51	0.85	28.76	0.89	28.77	0.89	28.67	0.89
Participant Sex	–	–	−0.95	0.70	−0.87	0.70	−0.90	0.70	−0.77	0.70
Parental Education	–	–	0.23	0.21	0.19	0.21	0.19	0.21	0.19	0.21
Parental NORC	–	–	0.03	0.04	0.04	0.04	0.04	0.04	0.04	0.04
Adoption Status	–	–	–	–	−3.51	0.70	−3.49	0.70	−3.42	0.70
Stress9–12	–	–	–	–	–	–	−0.20	0.31	0.02	0.36
Stress13–16	–	–	–	–	–	–	–	–	−0.48	0.38
r	0.31	0.01	0.29	0.01	0.28	0.01	0.28	0.01	0.28	0.01
Participant Sex	–	–	0.03	0.01	0.02	0.01	0.02	0.01	0.02	0.01
Parental Education	–	–	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00
Parental NORC	–	–	0.00	0.00	−0.001	0.001	−0.001	0.001	−0.001	0.001
Adoption Status	–	–	–	–	0.02	0.01	0.02	0.01	0.02	0.01
Stress9–12	–	–	–	–	–	–	0.00	0.01	0.00	0.01
Stress13–16	–	–	–	–	–	–	–	–	0.004	0.005
d	2.51	0.15	2.73	0.16	2.79	0.16	2.79	0.16	2.79	0.16
Participant Sex	–	–	−0.52	0.12	−0.51	0.12	−0.52	0.12	−0.52	0.11
Parental Education	–	–	−0.04	0.03	−0.05	0.03	−0.04	0.03	−0.04	0.03
Parental NORC	–	–	0.00	0.01	0.00	0.01	0.00	0.01	0.00	0.01
Adoption Status	–	–	–	–	−0.04	0.11	−0.04	0.11	−0.04	0.11
Stress9–12	–	–	–	–	–	–	−0.06	0.05	−0.06	0.05
Stress13–16	–	–	–	–	–	–	–	–	–	–
Telephone	−1.57	0.17	−1.57	0.17	−1.58	0.17	−1.58	0.17	−1.58	0.17
Variances
Va	55.00	3.83	55.08	3.97	53.54	3.97	53.64	3.97	53.37	3.90
Vd	1.01	0.12	1.08	0.14	1.03	0.14	1.02	0.14	1.02	0.14
Vr	0.003	0.001	0.003	0.001	0.003	0.001	0.003	0.001	0.003	0.001
Ve	11.10	0.47	11.07	0.47	11.05	0.47	11.05	0.47	11.05	0.47
Covariances
r with a	−0.24	0.06	−0.21	0.06	−0.19	0.05	−0.19	0.05	−0.19	0.05
d with a	−2.79	0.53	−3.15	0.54	−3.16	0.52	−3.17	0.52	−3.20	0.51
r with d	−0.02	0.01	−0.01	0.01	−0.01	0.01	−0.01	0.01	−0.01	0.01
Fit Criteria
LL	−11262.68		−11246.62		−11233.12		−11231.90		−11231.05
Scale Correction c	1.10		1.08		1.08		1.07		1.06
Parameters	12		21		24		27		29
AIC	22549.36		22535.24		22514.25		22517.79		22520.10
BIC	22603.77		22630.45		22623.06		22640.21		22651.58
N-Adjust BIC	22565.67		22563.77		22546.86		22554.48		22559.50
cd	–		1.06		1.05		0.99		0.98
TRd	–		30.40		25.82		2.47		1.73
df	–		9		3		3		2
p	–		3.76E-04		1.04E-05		.480		.420

Bold = p < .05

Italic = p < .10

Note. N = 688, parameter i is the lower asymptote fixed to be the same for all individuals (representing age 9 performance), a represents the total change from lower to upper asymptote. Parameter r is the rate of approach to the asymptote and d is the age at which accelerated changes is observed. The rate of growth is slower at the asymptotes and more rapid at the inflection point (d), which for the Gompertz model represents the point at which 37% of growth has occurred. Telephone (Mode of Administration) effects represent the change associated with testing over the telephone (In Person was the reference group dummy coded as 0).

For CPS, the inclusion of Adoption Status significantly increased fit [Δχ² (3) = 25.82, p = 1.04E-05], while controlling for mode of administration, participant sex, parental education, and parental occupational prestige (see Table 3). According to this model, both adopted and non-adopted individuals had an average performance level at age 9 (i) of 13.97 points ( $p<.001$), but with an average total gain (a) of 3.51 points less for adoptees than non-adoptees ( $p<.001$), and acceleration in rate of approach of .021 higher than non-adoptees (p = .032); adoptees and non-adoptees did not differ in age at acceleration (p = .695). The random effect variances for the change parameters, a, d, and r remained significant (all $p\le .007$ ), suggesting further prediction of individual differences in change were appropriate. Entering STRESS_9–12 [Δχ² (3) = 2.47, p = .480] as a predictor of a, d, and r and subsequently STRESS_13–16 as a predictor of a and r [Δχ² (2) = 1.73, p =.420] did not improve the model fit for CPS trajectories. However, models including STRESS_9–12, and both STRESS_9–12 with STRESS_13–16 predictors, suggest smaller total gains for participants with higher reports of stress, albeit non-significant (see Table 3). Figure 2 depicts the expected curves for adoptees and non-adoptees showing an emergent and increasing divergence of CPS trajectories into early adulthood.

Figure 2.

Trajectories of Colorado Perceptual Speed (CPS) between assessment points 9 through 30. Solid lines with open circles depict the raw trajectories for non-adoptees; dashed lines with dots reflect the raw trajectories of adoptees. Colored lines represent the expected trajectories estimated by the Gompertz model for adopted individuals (orange) and biologically reared individuals (red), adjusted for mode of administration, participant sex, parental education, and parental NORC occupational prestige.

Discussion

Findings from the current prospective study provide no evidence that cumulative perceived stress in early life impact memory and speed trajectories from childhood into early adulthood. Specifically, the findings suggest that when accounting for mode of administration, participant sex, parental education, parental occupational prestige and adoption status, neither cumulative perceived stress in middle childhood nor in adolescence were predictive of differential growth in perceptual speed or memory between the ages of 9 and 30. However, differential trajectories for perceptual speed were observed for adopted individuals compared to individuals reared by their biological parents.

We found no difference in the long-term effects of cumulative perceived stress experienced in middle childhood or in adolescence for memory or speed tasks. Our prospective findings are not inconsistent with those of Korten and colleagues who found no relationship between retrospectively-reported past or current severity ratings of stressful life events occurring in the last 12 months and episodic memory in adults (Korten et al., 2014b). Interestingly they did observe associations with primary and working memory that were strongest with current severity ratings. Relevantly, Korten and colleagues suggest that if there is a lasting impact of chronic daily stressors, it is likely that higher stress may lead individuals to adopt “stress-related intrusive and avoidant thinking” that impact primary and working memory processes, which are not addressable in these data.

Although perceived stress in middle childhood and adolescence did not predict change in perceptual speed (CPS), differences in trajectories of perceptual speed were observed between adopted and non-adopted individuals. Specifically, we see increasing divergence in average trajectories particularly in adolescence and into early adulthood. Some work on international adoptees has conceptualized early (prior to 8 months) versus late adoptee (after 1 year) groups as exposed to moderate versus severe stress, respectively, finding some differences in HPA functioning that may be indicative of resilience (Gunnar et al., 2009) but also observing smaller hippocampus and prefrontal cortex volumes in early adolescence (Hodel et al., 2015). The latter finding of smaller prefrontal cortical volumes may be pertinent to our current findings showing smaller gains in CPS performance in adolescence and early adulthood (c.f., Bartzokis et al., 2010; Eckert et al., 2010; Ekert, 2011). The extent to which differential speed trajectories is a consequence of the adoption experience, or is due to other correlated experiences or vulnerabilities, is questionable without further follow-up. Moreover, we did not observe differential patterns for the memory traits evaluated. That perceptual speed may be an important influence on later life memory change suggests that downstream consequences may be important to consider in terms of aging processes (Finkel, Reynolds, McArdle, & Pedersen, 2003; Finkel, Reynolds, McArdle, & Pedersen, 2007; Salthouse, 2000).

Chronic or prolonged stress have been reported to have negative outcomes in cross-sectional human studies as well as in animals (e.g., Hedges & Woon, 2011; Marin et al., 2011; McCormick & Mathews, 2010; Stawski, Sliwinski, & Smyth, 2006), and may negatively impact social interactions (Kremen et al., 2012) and cognitive aging (e.g., glucocorticoid cascade hypothesis; Sapolsky, Krey & McEwen, 1986). However, long-lasting negative impacts of cumulative perceptions of stress were not observed in the present study. Future research should investigate whether objective and subjective measures of stress do indeed differentially relate to trajectories of cognitive functioning like memory and perceptual speed. Further, having considered cumulative perceptions of stress across two developmental periods, it may be important to assess the proximal influences of perceived stress and any lasting carryover. For example, recent work suggests that adolescents may be more susceptible to immediate effects of heightened stress, conceptualized as short duration demands or pressures, in terms of cognitive control and neural functioning compared to adults (Rahdar & Galvin, 2014), but longer followups are warranted.

Although no significant impacts of cumulative perceived stress were found for memory or perceptual speed, the non-significant but negative impact on perceptual speed suggests that the temporal emergence of stress effects on performance changes in perceptual speed and memory might not be similar. This suggests that stress influences on perceptual speed might precede effects on memory. These findings warrant further investigation of how stress may impact the relationship between speed and other cognitive domains such as memory that has been observed in later cognitive change (Finkel et al., 2007; Finkel et al., 2009).

Some limitations of the current study must be noted. Although the multiple waves of prospectively measured stress during childhood are one of the greatest strengths of this study, perceived stress was not measured at cognitive assessment points beyond age 16, limiting our ability to test for the dynamic transactional relationship between stress and cognition from age 9 to 30. Additionally, these cumulative stress variables, tabulated based on perceived upsettingness of various life events, may not fully capture severity of stressful events. Finally, while our data suggest that there might be differences in trajectories of perceptual speed for adopted individuals, additional work is needed to investigate the underpinnings of this differentiation. For example, differences among the adoptees and non-adoptees may arise from differences in genetic, epigenetic, or uterine-environmental sources of influence that reveal themselves in different levels of stress reactivity, but may not necessarily be related to adoption per se.

Conclusions

While stress impacts on cognitive functioning have been documented in a variety of studies, many questions remain unanswered, including how these effects are manifested over the human lifespan and when the experience of stress might be most detrimental. The current study contributes to this line of research in multiple ways. First, cumulative perceptions of stress experienced during middle childhood and adolescence were not predictive of cognitive trajectories from childhood into early adulthood. Second, we found that adoption status was associated with differential perceptual speed across age. The causes of this association, and any lasting vulnerability on cognitive functioning into midlife should be assessed. Perceptions of stress during middle childhood and adolescence may not be predictive of memory and speed trajectories when considered in a cumulative form; whether perceptions of specific events, as well as finer examinations of individual occurrences of stress and development timing, pose any lasting impacts may be important to consider. Further, future studies should consider both objective biological measures of stress and subjective self-ratings of perceived stress simultaneously. Such a multi-method multi-level approach would facilitate a more explicit examination of the effects of stress on cognitive performance over the lifespan, and would elucidate whether different indices of stress affect cognitive performance differentially or dynamically.