Lifecourse Traumatic Events and Cognitive Aging in the Health and Retirement Study

Abstract

Introduction:

Much of the heterogeneity in the rate of cognitive decline and the age of dementia onset remains unexplained, and there is compelling data supporting psychosocial stressors as important risk factors. However, the literature has yet to come to a consensus on whether there is a causal relationship, and if there is, its direction and strength. This study moves the literature toward a causal understanding by estimating the relationship between lifecourse traumatic events (TEs) and cognitive trajectories and predicted dementia incidence.

Methods:

Using data on 7,785 participants aged 65+ from the Health and Retirement Study, this study estimated the association between lifecourse experience of ten TEs (e.g. losing a child) and trajectories of Telephone Interview for Cognitive Status (HRS-TICS) from 2006-2016 using linear mixed-effects models, and predicted incident dementia from 2006-2014 using cumulative incidence functions (data analysis 2020-2022). Inverse probability weights accounted for loss-to-follow-up and confounding by sex, education, race/ethnicity, and age.

Results:

Experiencing one or more TEs over the lifecourse was associated with accelerated decline compared to experiencing no events (e.g. β=−0.05 (95%CI: −0.07, −0.02) HRS-TICS units/year; 1 vs. 0 events). In contrast, experiencing TEs was associated with better cognitive function cross-sectionally. Furthermore, the impact of trauma on cognitive decline, was of greater magnitude when it occurred over age 64. However, the magnitude and direction of association varied by specific TE. There were no associations with predicted incident dementia.

Conclusions:

These results suggest that researchers and clinicians should not aggregate traumatic events for understanding risk of accelerated cognitive decline.

INTRODUCTION

While there are several established demographic and clinical risk factors for Alzheimer’s Disease and Related Dementias (ADRD), including chronological age, ApoE-e4 allele, sex, and education,^1–3 there is a growing body of research supporting a link between exposure to stressors and cognitive impairment or dementia risk. However, existing data on the impact of psychosocial stressors (e.g. traumatic events) on ADRD risk is missing crucial information. For example, there are little data on importance of timing of exposures. This matters because studies have shown that traumas occurring specifically during childhood may influence later-life cognition.^4–7 Further, the existing evidence of mixed results is primarily derived from cross-sectional studies in specialized populations (e.g., Holocaust survivors, former child laborers, or Aboriginal Australians), who have experienced above-average numbers of stressors in their lives.^5–8 The cross-sectional nature of these studies makes it impossible to rule out whether cognitive scores are associated with selection into the exposures. Moreover, it is unclear whether stressors impact cognitive decline over time. Lastly, none of the existing studies have used population-based data, making it difficult to generalize findings.

The purpose of this study was to examine the effects of experiencing traumatic events at different periods across the lifecourse on cognitive trajectories and dementia incidence in late life over 10 years of follow-up and to test whether the timing or total accumulation of trauma exposures impacted cognitive trajectories and incident dementia.

METHODS

Study Population

Data for this study came from the Health and Retirement Study (HRS), the largest on-going nationally-representative longitudinal survey of older adults in the U.S. HRS began in 1992 with >22,000 adults over 50 years old at baseline. Follow-up occurs every two years. HRS survey design and methods have been described previously.^9–11 This analysis used data from the 2006-2016 waves, with baseline psychosocial data from the 2006 and 2008 leave-behind Psychosocial and Lifestyle Questionnaires. Of the 9,950 participants who completed the baseline questionnaires, the cognitive trajectories analysis included 7,785 individuals who were ≥65 years old at baseline (so that the full, 35-point HRS-TICS was assessed), had ≥2 cognitive data points, and had full covariate data. The predicted incident dementia analysis was further restricted to 4,780 individuals without predicted baseline dementia and dementia identification available from the Gianattasio algorithm.¹²

Measurements

Participants were asked a 7-item list developed for an ongoing study of health consequences of trauma in older adults from several sources.^13,14 1) Has a child of yours ever died? 2) Have you ever been in a major fire, flood, earthquake, or other natural disaster? 3) Have you ever fired a weapon in combat or been fired upon in combat? 4) Has your spouse, partner, or child ever been addicted to drugs or alcohol? 5) Were you the victim of a serious physical attack or assault in your life? 6) Did you ever have a life-threatening illness or accident? 7) Did your spouse or a child of yours ever have a life-threatening illness. Death of a spouse was added by classifying those whose marital status is ‘widowed’ as an affirmative response to an eighth question. If participants indicated that they had experienced any of these events, they were subsequently asked in what year the event occurred.

Participants were also asked about experiences before age 18: 1) did either of your parents drink or use drugs so often that it caused problems in the family? And 2) were you ever physically abused by either of your parents. Supplemental Figure 1 depicts the number of participants who experienced each event and the 50 most common combinations of events experienced.

Using participant responses, two measures were created: a) an overall continuous traumatic events (TEs) experienced across the lifecourse and b) a life-period-specific dichotomous TEs experience. For overall, scores from both sets of questions were combined. For the life-period-specific measure, variables for childhood (ages 0 – 17), early adulthood (ages 18 – 34), mid-life (ages 35 – 64), and late-life (ages 65 and up) were created based on year participants reported the event to occur. If participants were missing data on the timing of their event, they were excluded from lifecourse period-specific analyses (sample in Supplemental Table 2).

In HRS, global cognition is based on performance on an abbreviated version of the Telephone Interview for Cognitive Status (HRS-TICS). This continuous measure has a possible range of 0-35, and includes immediate and delayed recall, serial 7s, counting backwards, object and person naming, and orientation to time. The psychometric properties of this HRS-TICS score have been assessed and determined to “display satisfactory psychometric properties”.¹⁵

Predicted Dementia.

HRS has no formal clinician-adjudicated diagnosis of dementia. Therefore, an approach previously described first by Hurd et al.¹⁶ and subsequently modified by Gianattasio et al.¹² which uses existing data collected from each wave of HRS to categorize participants’ dementia status was used. The Hurd paper took the biannually collected cognition data (HRS-TICS) and combined that with data on functional limitations, age, education, sex, and change in functional limitations and HRS-TICS in a model to predict dementia, training the algorithm in the ADAMS dataset. This method also incorporated proxy data via the Informant Questionnaire on Cognitive Decline in the Elderly.^17,18 Gianattasio then defined race- and ethnicity-specific cutpoints, improving sensitivity and specificity within these groups. Someone was defined as having predicted incident dementia only if they remained below this threshold for the remainder of follow-up.

Sociodemographic characteristics including educational attainment, sex, age, HRS birth cohort, existing comorbidities (HRS comorbidities index: high blood pressure, diabetes, cancer, lung disease, heart disease, stroke, psychiatric problems, arthritis), and race/ethnicity were collected in the 2006 or 2008 core interview, depending on the participant’s respective baseline. Educational attainment was categorized as <high school diploma, high school diploma, some college, or a college diploma and higher. Race/ethnicity was categorized as non-Hispanic White, non-Hispanic Black, Hispanic, or other, and biological sex was self-reported as male or female.

All statistical analyses were conducted in SAS 9.4 (SAS Institute, Inc., Cary, North Carolina) and figures were created in R¹⁹ using the ggplot2 package. Unweighted descriptive statistics were used to characterize the study population. To estimate the relationship between lifetime TEs and cognitive level and decline, linear mixed effects models with a random slope and intercept were used. To assess the relationship between the exposures and predicted incident dementia, stratified cumulative incidence functions were estimated using the Aalen-Johansen estimator, with death treated as a competing event. Inverse probability weights were used to account for informative censoring (based on exposure, educational attainment, age, age², baseline comorbidities, and change in cognition between two prior waves) and confounding (baseline age, age², birth cohort, educational attainment, race/ethnicity, and biological sex). Additionally, for lifecourse period-specific exposures, indicators (yes/no) of events in other lifecourse periods were included in confounding weights. While events occurring after the exposure (e.g., mid-life events occur after childhood events) are likely mediators of the relationship, for this paper were particularly interested in the direct effects of events during each period, independent of subsequent events, so this mediating path was blocked. This minimally sufficient adjustment set was determined by DAG analysis.²⁰ In order to further unravel the relationships and to check the built-in assumption in a summary score that each event matters the same amount to health, a sensitivity analysis was completed to assess the association between each individual TE and cognitive trajectories and predicted dementia.

RESULTS

The 7,785 adults who met eligibility criteria for the cognitive trajectories analysis were 59% female and 74 years old at baseline, on average. One quarter reported experiencing ≥3 TEs in their lifetime, while 22.4% reported never experiencing a TE. Mean baseline number of comorbidities was 2.2(SD=1.4) and was higher with each consecutive TE category. The mean baseline HRS-TICS score was 22.1(SD=4.6) and did not differ by number of events. The 4,741 with data on exposure timing were similar demographically to the full sample. The 4,780 adults who met eligibility criteria for the predicted incident dementia analysis were older (mean=77;range=67-99). Complete demographic and social characteristics of the populations for the cognitive decline analyses and predicted incident dementia analyses are shown in Table 1 and Supplemental Tables 1&2, respectively.

Table 1.

Baseline Sociodemographic and Cognitive Characteristics of Eligible Health & Retirement Study Participants, N = 7,785

Variable	Overall		0 Traumatic Events		1 Traumatic Event		2 Traumatic Events		3+ Traumatic Events
	N	%	N	%	N	%	N	%	N	%
Age, Mean (range)	74	(65, 104)	73	(65, 94)	74	(65, 99)	74	(65, 96)	74	(65, 104)
Sex (N, % Female)	4578	(58.8)	935	(53.7)	1331	(59.3)	1123	(60.8)	1189	(60.9)
Education
Less than High School	1702	(21.9)	365	(21.0)	501	(22.3)	423	(22.9)	413	(21.1)
High School Diploma	2998	(38.5)	706	(40.6)	887	(39.5)	689	(37.3)	716	(36.6)
Some College	1584	(20.4)	320	(18.4)	403	(18.0)	385	(20.8)	476	(24.4)
College Diploma +	1501	(19.3)	349	(20.1)	453	(20.2)	350	(19.0)	349	(17.9)
Race/Ethnicity
Non-Hispanic White	6182	(79.4)	1350	(77.6)	1783	(79.5)	1492	(80.8)	1557	(79.7)
Non-Hispanic Black	948	(12.2)	208	(12.0)	262	(11.7)	233	(12.6)	245	(12.5)
Hispanic	535	(6.9)	155	(8.9)	156	(7.0)	102	(5.5)	122	(6.2)
Other	120	(1.5)	27	(1.6)	43	(1.9)	20	(1.1)	30	(1.5)
TICS Score, Mean (SD)	22	(4.6)	22	(4.5)	22	(4.8)	22	(4.5)	22	(4.6)
Lifetime Traumatic Events
0 Events	1740	(22.4)
1 Event	2244	(28.8)
2 Events	1847	(23.7)
3+ Events	1954	(25.1)
Lfetime Traumatic Events (N, % 1 or more)
Early Life	708	(9.1)	0	(0.0)	197	(8.8)	219	(11.9)	292	(14.9)
Young Adulthood	634	(8.1)	0	(0.0)	144	(6.4)	204	(11.0)	286	(14.6)
Mid Life	1484	(19.1)	0	(0.0)	506	(22.6)	447	(24.2)	531	(27.2)
Late Life	1534	(19.7)	0	(0.0)	593	(26.4)	468	(25.3)	473	(24.2)
Missing Timing Data	3044	(39.1)	801		998		1245		692
Comorbidities, Mean (SD)	2.2	(1.35)	1.9	(1.25)	2.1	(1.28)	2.3	(1.32)	2.6	(1.46)

Figure 1 and Supplemental Table 4 depict the association between the total number of TEs experienced across the lifecourse (accumulation of risks model) and population mean cognitive level, before and after weighting. IP-weighted models showed that those experiencing any TEs (i.e.1,2,or3+) had higher mean HRS-TICS scores compared to those who experienced none. Those who experienced 2 events had HRS-TICS scores 0.47(95%CI:0.18,0.75) and those with 3+ events had scores 0.43(95%CI:0.14,0.71) points higher than those who experienced no events. Figure 2 also shows the association between TEs experienced during a specific life period and cognitive function (Supplemental Table 5). While there was no significant association between TEs experienced during mid-life and HRS-TICS, there were significant associations for each other life period (sensitive periods model). Compared to those who experienced no TEs during early-life, and controlling for TEs experienced during other life periods in order to estimate direct effects of each life period, those who experienced one or more TEs before 19 had a mean HRS-TICS score 0.59(95%CI:0.24,0.94) points higher, and those who experienced events during young-adulthood had a mean HRS-TICS score 0.49(95%CI:0.11,0.87) points higher. The association was largest among those who experienced TEs after the age of 65; these participants had a mean HRS-TICS score 1.16(95%CI:0.87,1.46) points higher than those who experienced no events in later-life.

Figure 1. Fixed Effects Associations Between Lifecourse Experiences of Traumatic Events and Mean Population Cognitive Function and Cognitive Decline.

This figure shows the fixed effects beta estimates and 95% confidence intervals for the exposure (main effect) and exposure*age (slope) terms from linear mixed effects models assessing the association between lifecourse traumatic events (accumulation and sensitive period models, specified on the y-axis) and cognitive function (HRS-TICS, specified on the x-axis; higher score is equivalent to higher cognitive function). E.g. Compared to those who experienced no traumatic events during early life, those who experienced one or more traumatic events before age 19 had a mean HRS-TICS score 0.59 (95%CI: 0.24, 0.94) points higher. For each year of age, experiencing one or more traumatic events after age 64 (Late Life) was associated with 0.07 (95%CI: 0.04, 0.10) fewer HRS-TICS points indicating an accelerated rate of cognitive decline compared to those who did not experience any traumatic events later in life

Figure 2. Cumulative Predicted Dementia incidence by Lifecourse Traumatic Events.

This figure depicts the cumulative predicted dementia incidence, IP-weighted to account for censoring and confounding, for each form of the exposure. Panels show incidence stratified by (a) cumulative lifecourse traumatic events (0, 1, 2, or 3+ events); (b) early-life events; (c) young-adulthood events; (d) mid-life events; and (e) late-life events. Cumulative incidence functions were estimated using the Aalen-Johansen estimator. There is significant difference between the predicted incidence of dementia by number of lifecourse traumatic events overall or during specific life periods, except for later life events. In small window (ages 84 to 97), there is a slightly lower predicted incidence of dementia for those who experience traumatic events in late-life.

Supplemental Tables 2&3 and Figure 1 also show the association between the total number of TEs experienced across the lifecourse and cognitive decline over the course of follow-up. Experiencing any events over the entire lifecourse was statistically associated with rate of change in HRS-TICS score in the study population. For each year of age, experiencing one or more TEs was associated with 0.04(95%CI:0.01,0.07) fewer HRS-TICS points, indicating an accelerated rate of cognitive decline compared to those who did not experience any TEs during their life. Figure 1 demonstrates that this association is largely driven by those experiencing events later in life. For each year of age, experiencing one or more TEs after age 64 was associated with 0.07(95%CI:0.04,0.10) fewer HRS-TICS points indicating an accelerated rate of cognitive decline compared to those who did not experience any TEs later in life. However, experiencing TEs during other life periods was associated with slightly slower decline or not associated with rate of decline.

Figure 2 depicts the IP-weighted cumulative predicted dementia incidence, stratified by experience of TEs. Panel a demonstrates no meaningful difference in risk of predicted dementia by the number of events experienced across the lifecourse. Panels b - e depict the cumulative incidence functions for each life period, showing no meaningful difference in the predicted dementia incidence by experiencing a TE during early-life or midlife. Experiencing an event in young-adulthood is associated with a slightly higher risk of predicted dementia from 72-88 years old, while experiencing a late-life event appears to initially be protective, with the association disappearing around age 80.

Analysis of each individual TE and its association with cognition and rate of decline indicate that type of event does matter. Those who experienced the loss of a child or were the victim of a serious physical attack during their life had significantly lower HRS-TICS scores than those who did not experience these types of events (−0.95,95%CI:−1.22,−0.68; −0.68,95%CI:−1.15,−0.21 respectively). On the other hand, those who had ever been in a major natural disaster or whose spouse, child, or other close relation had ever had a life-threatening illness or accident had higher HRS-TICS scores than those who had not experienced that (0.51,95%CI:0.24,0.79; 0.32,95%CI:0.49,0.97 respectively). The association of TEs with rate of cognitive decline appears to be driven primarily by experiencing the death of a spouse, which for each five years of age was associated with 0.44(95%CI:−56,−0.32) more HRS-TICS points of decline compared to those who had not experienced the death of a spouse. These associations are depicted in Figure 3 This pattern was not borne out with predicted dementia incidence. However, where an association only when comparing those who were physically abused by either parent during childhood was found (Supplemental Figure 2).

Figure 3. Fixed Effects Associations Between Individual Traumatic Events and Mean Population Cognitive Function and Cognitive Decline.

This figure shows the fixed effects beta estimates and 95% confidence intervals for the exposure (main effect) and exposure*age (slope) terms from linear mixed effects models assessing the association between each specific traumatic event (specified on the y-axis) and cognitive function (HRS-TICS, specified on the x-axis).

DISCUSSION

We found that overall, lifetime experience of TEs was associated with faster cognitive decline in a US population-based survey of adults over 65. When investigating life-period specific occurrence of these events, however, several notable trends were identified. The associations with both cognitive level and rate of decline appear to be largely driven by events experienced later in life. In sensitivity analyses of specific TEs, the death of a child and being the victim of a serious physical assault were associated with lower cognitive function. On the other hand, being in a major natural disaster or having a close relative experience a life-threatening illness or accident were associated with better cognition. Moreover, lifetime TE exposure was associated with higher population mean estimates of cognitive function; however, the magnitudes of these associations were relatively small. A model regressing cognition on age estimates cognition decreases by 0.32 HRS-TICS units/year; the present mean cognition estimates are roughly equivalent to 1-3 fewer years of cognitive aging, while rate of change estimates are even smaller but may be clinically meaningful when accumulated over decades. Finally, lifetime TEs were not associated with predicted dementia incidence over a 10-year period. However, the results for all analyses should not be overinterpreted, due to several limitations discussed below.

Prior research has found mixed associations between TEs and cognitive function and decline. Several studies have found that an accumulation of stressful or TEs is associated with lower performance in specific domains of cognitive function or MCI.^21,22 In the Irish Longitudinal Study on Ageing, researchers found that history of childhood sexual abuse was associated with better global cognition, memory, executive function, and processing speed among adults 50 and older.²³ The authors acknowledged that this finding was counterintuitive and hypothesize that it may be a reflection of a chronic lowering of cortisol-secretion (rather than heightening) or a state of hyperarousal causing upregulation of noradrenergic activity and promotion of cognitive reserve.

Similar to these results for specific TEs, an older study by Grimby et al. found that cognitive decline was not associated with stressful life events, except in the case of bereavement, where experiencing the death of a spouse or child was associated with larger declines in cognitive abilities suggesting that graver TEs may be more impactful.²⁴ Finally, Tschanz et al. found that stressful life events effects on cognition over seven years vary by educational attainment and type of stressful life events.²⁵ The results from this study are consistent with these findings, as they show associations of TEs overall with better cognition, depending on the event. There are many analogies where a small amount of stress can activate the immune system and be protective for health in the short term but detrimental at high levels for extended periods (e.g. exercise). This model of brief stressor-induced health protection could help explain these findings as they also related to differences by event type and timing. However, it is also impossible to know whether these unexpected cross-sectional associations are spurious or due to selection biases (discussed in limitations).

Our study has several strengths, including the use of a large, U.S.-population based longitudinal cohort and up to 10 years of follow-up cognitive data. Furthermore, HRS provided rare details on 10 TEs and when they occurred. However, this study also has several limitations. There is measurement error in the cognitive measures; HRS-TICS is a limited measure of global cognition, and the algorithm used to diagnose dementia has limited sensitivity and specificity and was initially designed to diagnose prevalent dementia, not incident. While the length of follow-up is longer than most existing studies, it is possible that the maximum time in study of 12 years and the average follow-up of 9 years (4.5 waves) in this study contributes to the null findings with predicted dementia incidence: the progression of brain pathophysiology that leads to dementia can occur over decades. There is also the potential for bias in these results due to unmeasured confounding, and measure of perceived stress was not included. Of particular concern is the potential for recall bias; that is, those with a better memory (resulting in a higher global cognition score) would likely more accurately report their exposure, as it is measured by self-report. In this case, those with better cognitive scores would be more likely to accurately report being exposed, although the events were very significant and might be less susceptible to recall bias than others. Moreover, a limited set of events was examined; it is possible that other TEs may be linked to cognitive outcomes. Further research on other the impact of significant stressors, such as structural racism, discrimination, and immigration experiences, among older black, indigenous, and people of color, is needed. Along with other TEs, the modifying and/or mediating nature of factors such as social support and coping skills on these relationships may be importantly influencing these results.

One of the most important limitations of this research is selection bias. The population investigated includes only individuals who survived to age 65 and were healthy enough to enroll. This leaves a large amount of time between exposure and study enrollment, particularly for early-life and young-adulthood TEs. TEs may have had unobserved health impacts prior to study enrollment; those who experienced such events may have been less likely to enroll, or may have been less likely to survive to eligibility. This bias would affect both the mean cognition and cognitive decline results. While there is little to be done about this issue in the existing cohort, including these exposures and outcomes in future longitudinal cohort studies of younger individuals would help mitigate these limitations. Therefore, it is important to note that these exposures should not be interpreted as brain protective, for the reasons described above.

In conclusion, these findings suggest that later-life traumas may impact the rate of cognitive decline, but the results should be interpreted cautiously. Studies examining the impact of treatment and support for coping with traumatic experiences in later-life may help identify interventions and methods for buffering the potential impact of trauma. This study highlights that use of composite scores of negative or stressful experiences may mask true effects on health outcomes and a more nuanced approach to modeling timing and impact of stressors is warranted. While these results are not causal, knowing patients’ histories with respect to TEs, understanding how these experiences may indicate higher risk of accelerated cognitive aging, and incorporating that information into monitoring and care plans for patients could improve clinical practice.