Education, Occupational Environment, and Cognitive Function in Later Life

Abstract

Objectives

Education is among the most robust predictors of cognitive health outcomes in later life. However, few studies have comprehensively evaluated whether and how much of this effect is explained by occupational exposures. This study aims to determine if and how much pre-retirement occupational exposures (occurring before age 60) mediate the association between education and cognitive function at age 65+.

Methods

We use data drawn from the Health and Retirement Study (HRS) and Occupation Information Network (O*NET) data. Informed by previous research and theory, we conducted Confirmatory Factor Analyses of occupation-level exposure measures using a longitudinal HRS-O*NET linked data set we created, and we identified 2 latent factors: occupational hazards and occupational complexity. Among initially employed adults (age 51–60 at baseline), we used Structural Equation Modeling (SEM) to evaluate the association between education and cognitive function at age 65+, and the role of our 2 occupational factors in mediating this association.

Results

The measurement and structural models both had good model fit (TLI, CFI ≥ 0.95, SRMR < 0.08). We found (a) that education remained a critical predictor of cognitive outcomes in later life even when accounting for occupational exposures, and (b) only hazardous exposures mediated the association between education and cognitive function in later life (a₂b₂=0.02, p = .01), explaining about 17% of the effect of education.

Discussion

These findings suggest interventions designed to decrease exposure to hazardous occupational exposures could reduce some of the cognitive disadvantages in later life associated with lower levels of education.

Alzheimer’s Disease and Related Dementias (ADRDs) are among the fastest-growing, most disabling, and most expensive health conditions affecting approximately seven million adults age 65+ in the United States (Alzheimer’s Association, 2024). Although genetic factors do influence ADRD risks, life course experiences significantly affect cognitive outcomes through a range of health behaviors and environmental exposures (e.g., education, smoking, obesity, depression, physical activity, social connections, and air pollution). Researchers argue that modifying these exposures has the potential to ameliorate approximately 40% of dementia cases in the United States and globally (Lee et al., 2022; Livingston et al., 2020; Wood et al., 2024).

Cognitive enrichment fostered through educational attainment has been highlighted as one of the most robust protective factors in reducing the cognitive loss that precedes ADRD (Liu, et al., 2022; Reuter-Lorenz et al., 2023). However, the protective effect of education on cognitive function may be partly explained by its role in shaping career pathways. Growing evidence suggests that occupational exposures may influence cognitive outcomes in later life (Livingston et al., 2020). On the one hand, our jobs could preserve cognitive resources through regular exposure to protective, cognitively enriching activities (Carr et al., 2020) On the other hand, our jobs could erode cognitive resources through exposures that harm brain health directly through harmful environmental exposures or through reduced exposure to cognitively protective activities (Oosterhuis, 2023). For instance, higher levels of education select people into occupations (e.g., chief executive officers, psychotherapists) requiring mental skills such as creative thinking and decision-making, characteristics that have been argued to be cognitively enriching (Edwin et al., 2024; Reuter-Lorenz et al., 2023). However, there is much less research evaluating how higher exposure to occupational hazards (e.g., noise and air pollution) that has been shown to erode cognitive resources (Paul et al., 2019; Weuve et al., 2021) are involved in explaining the association between educational attainment and cognitive outcomes in later life. This study comprehensively investigates whether occupational exposures that might promote cognitive enrichment (i.e., occupational complexity) or erode cognitive function (i.e., hazardous exposures; Livingston et al., 2020) mediate the association between education and cognitive function in later life. Findings have the potential to inform which kinds of work-based interventions could help reduce risks of poor cognitive function and development of dementia in later life.

Cognitive Reserve and Scaffolding Theory of Aging and Cognition

Compared with traditional cognitive aging theories that focus primarily on brain structures (e.g., the posterior-anterior shift in aging), cognitive reserve theory (CRT) and scaffolding theory of aging and cognition (STAC) aim to explain how aging-related changes in both cognitive function and brain structure are influenced by lifestyle and environmental factors (Oosterhuis, 2023). According to CRT, cognitively enriching activities (e.g., reading, solving puzzles, playing musical instruments) and physical exercise enhance cognitive reserve, sustain cognitive function, and promote cognitive capacity by facilitating the generation and activation of neural pathways. Furthermore, the cognitive reserve gained through neural enrichment also improves neurological flexibility by creating alternative neural pathways to compensate for aging- and disease-related cognitive decline, a process known as cognitive scaffolding in STAC (Oosterhuis, 2023; Reuter-Lorenz et al., 2014, 2023). Relying on the neuroplasticity of the brain, cognitive scaffolding preserves cognitive function when faced with health events (e.g., depression, traumatic brain injury [TBI]) that have the potential to reduce cognitive function (Oosterhuis, 2023; Reuter-Lorenz et al., 2023).

Growing evidence demonstrates that life course exposures accumulate to influence cognitive reserve and cognitive losses associated with normal aging (Reuter-Lorenz et al., 2014, 2023). For instance, although formal education is typically completed in early life, its benefits on one's cognitive function persist in later adulthood, protecting against cognitive impairment and delaying the onset of ADRDs (Lövdén et al., 2020). Even though education does not seem to directly modify the pace of aging-related cognitive decline (Lövdén et al., 2020; Tucker-Drob et al., 2019), higher levels of education cultivate people's cognitive skills (i.e., fluid and crystallized cognitive abilities), and select people into cognitive enriching occupations that might further bolster cognitive reserve, facilitate cognitive scaffolding, and preserve high levels of cognitive function in later life (Oosterhuis, 2023; Reuter-Lorenz et al., 2023).

Moreover, longer education might preserve cognitive function by protecting people from harmful exposures associated with limited education, such as unemployment, financial stress, and hazardous occupational environments in early or middle adulthood (Lager et al., 2017; Lövdén et al. 2020). Environmental hazards, such as exposure to toxins and noises, are known to deplete neural/cognitive resources and lead to reduced cognitive function in later life (Reuter-Lorenz et al., 2023). People with less education are more likely to be exposed to these environmental hazards in their occupations (e.g., construction workers, roofers). These occupational environments often involve the aforementioned hazardous exposures, but the impact of these occupational hazards on later-life cognition is not well understood.

Occupational Complexity and Cognitive Function in Late Life

Consistent with CRT and STAC, cognitive complexity of work (also known as occupational cognitive enrichment or mental demand) has been shown to benefit cognitive health. For instance, Carr et al. (2020) used longitudinal Health and Retirement Study (HRS) data, linked with data drawn from the Occupation Information Network (O*NET), the Department of Labor measures associated with occupations in the United States. They measured occupational complexity as the sum of five variables: making decisions and solving problems; thinking creatively; coaching and developing others; frequency of decision-making; and freedom to make decisions. They found that occupational cognitive complexity was protective of cognitive function in association with retirement transitions (Carr et al., 2020). Other researchers have examined different types of occupational complexity such as “Language & knowledge,” “Pattern detection,” “Information processing,” and “Service” demands (Rodriguez et al., 2020), and measures based on specific domains such as complexity with data, complexity with people, and complexity with things (Vélez-Coto et al., 2021). Evidence suggests that exposure to cognitively complex environments benefits later-life cognition (e.g., Andel et al., 2019; Calatayud et al., 2022; Thoma et al., 2020) and reduces the risk of cognitive impairment later in life by half for those who worked in more cognitively complex occupations compared with their counterparts in less cognitively complex occupations (Andel et al., 2019).

Occupational Hazards and Cognitive Function in Late Life

Exposure to environmental hazards through occupation (e.g., air pollution, noise, high places) has been associated with cognitive impairment and ADRD through inflammatory, vascular, and pulmonary mechanisms (Berr and Letellier, 2020). Compared with research examining the potential benefits of cognitive complexity in work, the cognitive health impact of occupational hazards has been less studied in gerontological literature. Among the few studies examining the cognitive impact of environmental and occupational hazards on cognitive health, most studies have examined individual factors such as noise and air pollution (Paul et al., 2019; Weuve et al., 2021). Only a few studies have systematically evaluated occupational hazards to identify exposures that jointly comprise certain occupational environments. For example, also using an O*NET data linkage, this time merged with the MIDUS (Midlife in the United States Study), Grzywacz and colleagues (2016) measured occupational physical hazards that escalate the risk for injury, accidents, and illness. They created a measure combining 12 variables: radiation, disease or infections, high places, hazardous conditions, hazardous equipment, minor cuts/bites/stings, noises, very hot or cold temperatures, extremely bright or inadequate lighting, exposures to contaminants, cramped workspace, and whole-body vibrations (Grzywacz et al., 2016). They noted that higher scores on this hazardous work measure were associated with cognitive consequences (Grzywacz et al., 2016).

Other research evaluating exposures to individual hazardous conditions has shown similar findings. For instance, occupational injuries such as falls from high places have been shown to affect cognitive health negatively (Zotcheva et al., 2023). Similarly, TBI resulting from traffic accidents has been associated with reduced cognitive function (Reuter-Lorenz et al., 2023). Prolonged exposure to occupational hazards, such as noise and extreme lighting, has been associated with elevated chronic stress, sleep disruptions, and risks of sensory impairment in late life (e.g., hearing or vision loss), which has implications for cognitive function (Paul et al., 2019; Weuve et al., 2021). Additionally, exposure to toxins and chemicals (e.g., lead, mercury) common in construction and manufacturing jobs has been shown to increase the risk of developing neurodegenerative conditions through direct damage to brain cells and indirect mechanisms such as disruptions to neurotransmitters critical for cognition (Reuter-Lorenz et al., 2023).

One of the key limitations of these studies, however, is that occupations with higher levels of hazardous exposures may be more likely to be simultaneously low in cognitive complexity. Although all jobs involve varying levels of exposure to occupational complexity and hazards on a regular basis, educational attainment often shapes occupational pathways in ways that result in occupations being sorted such that high scores in one exposure result in low exposures in the other (Yuma-Guerrero et al., 2018). Consequently, studies that only evaluate hazardous occupational exposures might merely be capturing the consequences of lower levels of cognitive enrichment in occupational environments. Alternatively, research evaluating only the effects of cognitively complex occupations may be capturing the benefits of individuals having low hazardous exposures. We are aware of no studies that have comprehensively evaluated both types of occupational environments together in association with cognitive outcomes in later life.

The Current Study

The current study is designed to address the following research gaps. Although existing studies suggest that cognitive complexity at work in midlife complements cognitive enrichment through education in early life (Livingston et al., 2020), few studies have attempted to evaluate the mediating role of occupational environment in adulthood on cognitive outcomes in later life. This study examines whether and how much of the cognitive outcomes associated with educational attainment are explained by occupational environment, focusing specifically on occupational hazards and occupational complexity. We conduct mediation analysis using nationally representative longitudinal data from the HRS, leveraging a linked data set using data our team created and released through the HRS website, based on Occupation Information Network (O*NET) data cross-walked to the detailed Census job level. We evaluate occupational environment exposures for individuals 51–60 on cognitive function at age 65+. We use these data to address our overarching research question: Is the association between education and cognitive function in later life mediated by midlife occupational environments (i.e., occupational hazards and cognitive complexity)?

Method

Data and Sample

This study used data drawn from the 2004 to 2016 waves of biennial longitudinal HRS, a cohort-based nationally representative study of adults aged 51+. We created a data set using individual-level data drawn from the (a) RAND HRS Longitudinal and Core data, and (b) data on cognitive function based on newly released standardized cognition and ADRD probability measures that adjust for subgroup differences (e.g., gender, race) among respondents age 65+ (Hudomiet et al., 2022; data only currently available through 2016 due to changes in the cognitive assessment protocol starting in 2018). We also used a recently released occupation-level exposure data set, providing variables for O*NET occupational exposure measures based on each respondent's detailed Census Job classification. Occupational Information Network (O*NET) data are ideal for comprehensively examining protective and hazardous occupational environments. Occupational Information Network (O*NET) data were developed by the U.S. Department of Labor (DOL), a gold standard source of objective measures for work exposures, with over 350 measures linked to over 1000 detailed occupations, about 500 of which are represented in the HRS data set. Members of the study team helped develop and release an HRS-O*NET data linkage in 2022, cross-walked to all study waves with detailed Census 2000 and Census 2010 job classification data, representative of occupations around the mid-2000s (Carpenter et al., 2022). The study procedures were reviewed and approved by the Institutional Review Board (IRB) of the Florida State University, USA (Study ID STUDY00003329).

Using these data, our sample was selected using the following inclusion criteria: (a) currently employed and between 51 to 60 years old at baseline (between 2004 and 2012), and (b) at least 65 or older at their last wave of cognition data (no later than 2016). This design is related to the restrictions imposed by the data availability of our study variables. This design also ensures that occupational exposures are measured before likely retirement ages, and cognitive outcome measures are drawn during an age period when cognitive function variations associated with dementia are likely to be observable. To minimize missingness and ensure sufficient data for accurate full information maximum likelihood (FIML) estimation, we include individuals who have at least one non-missing O*NET variable and non-missing cognitive outcome data in our sample. We use the FIML method to address missing data among endogenous variables (variables that are predicted by other variables in the model) in our analytic sample (Bowen & Guo, 2011). Our final sample has N = 1694 community-dwelling older adults.

Measures

Educational Attainment

Educational attainment is measured based on the number of years of formal education completed. All individuals with more than a college degree are top coded at 17 (range is 0–17).

Cognitive Function

Later life cognitive function is captured by a race/ethnicity-adjusted measure using participants’ last wave of cognition data in HRS (available for the 2000–2016 waves). The cognitive measures developed by Hudomiet et al. (2022) enhanced the accuracy of measuring cognitive function and dementia prevalence over time across socio-demographic subgroups among HRS participants 65 years and older. Dementia and cognitive impairment (not dementia) diagnoses in the Aging, Demographics and Memory Study (ADAMS) in the full HRS were used to calibrate these cognition measures. The expected value of cognition (ECog) is modeled as an unobserved continuous latent variable predicting their performance on the HRS cognition measures in each survey wave. The values in the continuous Ecog measure can be interpreted as follows: values below 0 indicate dementia; values between 0 and 1 indicate cognitive impairment, not dementia; and values above 1 indicate normal cognition.

Because the Hudomiet measures are only available for participants 65 years and older, we control for baseline cognitive function using a valid measure of overall cognitive function for those under age 65—a 27-point scale of cognitive function that accounts for memory, working memory, and processing speed (Langa et al., 2020). The score is based on the sum of immediate and delayed 10-noun free recall tests to measure memory (0–20 points); a serial sevens subtraction test to measure working memory (0–5 points); and a counting backward test to measure the speed of mental processing (0–2 points; Crimmins et al., 2011). This measure comes from cognitive data files that have been cleaned and imputed for missing values for non-proxy respondents by HRS. Because the answers from proxies were excluded to ensure response accuracy, this measure reflects the cognitive function of community-dwelling older adults.

Occupational Environment Measures

Cognitive complexity and occupational hazards are two distinct but connected components of work environments thought to influence late-life cognition (Grzywacz et al., 2016). We used confirmatory factor analysis (CFA) to create our occupational environment measures, leveraging all occupational data available for those who were working and met our age criteria (N = 1,800). Drawing from a previous cognitive complexity measure (Carr et al., 2020)we tested the same five items described in the introduction section of our study through CFA. However, based on recent work showing the benefits of occupational social complexity for brain health and reduced ADRD risks (Vélez-Coto et al., 2021), we tested the inclusion of occupational measures capturing interpersonal interactions (i.e., resolving conflicts and negotiating with others; Edwin et al., 2024; Reuter-Lorenz et al., 2023). To develop our latent occupational environment factors, items were removed from the model one by one according to their factor loadings, R²_, and standard error variance (Bowen & Guo, 2011). Our final measurement model showed the following items maximized model fit: thinking creatively, the freedom to make decisions, resolving conflicts and negotiating with others, making decisions and solving problems, coaching and developing others.

For occupational hazards, we examined the 12 measures Grzywacz and colleagues (2016) have previously identified as potentially important for cognitive function. Using the same approach as described for the complexity measure, our final hazard measure included the following five variables: using hazardous equipment, working at high places, extremely bright or inadequate lighting, very hot or very cold temperatures, and cramped workspace. Each O*NET variable in cognitively complex and cognitively hazardous jobs ranges from 1 to 5 continuously (see Table 1).

Table 1.

Descriptive Statistics for the Study Sample

Variables	Mean	Proportion	Variance
Years of education	13.49		8.50
Occupational hazards
Exposed to high places	1.36		0.28
Exposed to hazardous equipment	1.81		0.87
Extremely bright or inadequate lighting	1.82		0.35
Cramped workspace, award positions	1.81		0.32
Very hot or cold temperatures	2.03		0.63
Occupational complexity
Freedom to make decisions	4.13		0.18
Coaching and developing others	2.64		0.62
Making decisions and solving problems	3.52		0.48
Thinking creatively	2.87		0.66
Resolving conflicts and negotiating with others	2.92		0.65
Score at last wave of cognition measures	1.66		0.32
Statistical controls
Age at the last cognitive measure	67.28		4.39
Female		0.55	0.25
Race and ethnicity
White		0.69	0.22
Black		0.17	0.14
Hispanic		0.12	0.10
Baseline cognitive function	16.84		14.62
Baseline depressive symptoms	1.27		3.51

Notes: N = 1,694. Baseline depressive symptoms were measured by the Center for Epidemiological Studies Depression (CESD). A higher CESD score indicates higher levels of depression. Per Health and Retirement Study's restricted data access policies, the actual min and max values for each variable are not reported in the descriptive table due to confidentiality concerns for participants. Proportions were presented for categorical variables whereas means were presented for continuous variables.

Control Variables

Informed by existing empirical evidence (e.g., Reuter-Lorenz et al., 2023). Race/ethnicity is coded: (1) Non-Hispanic White, (2) Non-Hispanic Black, and (3) Hispanic. We also control for gender (female=1), and the age at the last wave of cognitive measure. Finally, we control for the number of depressive symptoms using the 8-item Center for Epidemiology Studies Depression Scale (CES-D; range is 0–8).

Analysis

Structural Equation Modeling (SEM) was performed using Mplus. Along with the default Maximum Likelihood (ML) Estimator, a robust estimator (MLR) was also tested to account for non-normality in our outcome and independent variables while correcting chi-square model fit statistics (Bowen & Guo, 2011; Li, 2021). Findings from these two estimators were similar, so we report the ML results because they produce intuitive measures of the direct and indirect effects of mediation analysis. We assessed our model fit with standard overall fit indices including the χ² test of overall fit, Comparative Fit Index (CFI), Tucker Lewis Index (TLI), Standardized Root Mean Squared Residual (SRMR), and the Root Mean Square Error of Approximation (RMSEA). A nonsignificant χ² test indicates a good model fit. However, the χ² statistic is biased by large samples, often reaching significance in large data sets (Bowen & Guo, 2011). A good model fit is indicated by TLI and CFI values of 0.95 or higher and a nonsignificant RMSEA value of less than.06 (Bowen & Guo, 2011). Additionally, the SRMR should be below 0.08 (Bowen & Guo, 2011). Missing data is addressed using a Full Information Maximum Likelihood (FIML), allowing cases missing on some O*NET measures to be included in analyses.

According to the two-step analysis procedure for estimating SEM models (Anderson & Gerbing, 1998), we first conducted confirmatory factor analysis (CFA), a measurement model, to identify latent occupational environment measures based on existing literature suggesting cognitively hazardous and cognitively enriching work environments serve as separate factors (Grzywacz et al., 2016). When testing items for the measurement model, items with standardized factor loadings below 0.4, accompanied by relatively small R²s and large residual variances were removed for model fit and parsimony (Bowen & Guo, 2011). After obtaining a good fit in our measurement model, we ran the full model including structural paths to examine the potential mediating effect of cognitive hazards and cognitive complexity, respectively, in the association between education and cognitive function in later life. Socio-demographic characteristics thought to influence cognition were accounted for as controls in the structural model.

Results

As presented in Table 1, participants in our sample were relatively young at the time of cognitive assessment, with an average age of 67.28 (Variance = 4.39). Approximately 55% of participants identified as female. In terms of race and ethnicity, 69% of the sample were non-Hispanic white, 17% identified as non-Hispanic Black, and 12% identified as Hispanic. Their mean baseline CES-D was 1.27 (ranging from 0 to 8). Participants had an average of 13.49 years of education. Participants scored an average of 16.84 on the 27-point cognition scale at baseline. Using the race/ethnicity adjusted Hudomiet measure, participants’ average cognition function score in their last wave of data using the Hudomiet measure was 1.66. Detailed socio-demographic information is presented in Table 1.

Findings from Confirmatory Factor Analysis (CFA)

A confirmatory factor analysis (CFA) was conducted to detect latent variables based on empirical studies on both occupational complexity and occupational hazards, which are thought to be jointly related to cognitive function in later life. The Cronbach’s alpha for items initially chosen for the hazardous and cognitive complexity measures were 0.90 and 0.84 respectively, demonstrating good internal consistency.

The two-factor CFA/measurement model (separate but correlated factors for cognitively hazardous and cognitively complex environments) yielded a good model fit for our analytic sample, X² (df=25, p = .00) = 304.69, CFI = 0.98; TLI = 0.96; RMSEA = 0.07 (95% CI [0.07, 0.09]; SRMR = 0.05 (Bowen & Guo, 2011). Both standardized and unstandardized factor loadings from our CFA are presented in Table 2. All items had standardized factor loadings (λ) above the recommended cutoff of 0.40 (Bowen & Guo, 2011; Watkins, 2021) . All items in the measurement model have standardized factor loadings of 0.70 and above, indicating that the latent constructs correlated well with their observed indicators. In establishing model fit, some items were allowed to covary across factors because specific hazardous and cognitively complex measures are often inversely associated (Zotcheva et al., 2023) above and beyond their latent factors.

Table 2.

Factor Loadings in the Confirmatory Factor Analysis (CFA)

Variables	Unstandardized factor loading	p [95% CI]	Standardized factor loading	p [95% CI]
Exposure to occupational hazards
Exposed to high places	1.00	0.00 [1.00, 1.00]	0.79	0.00 [0.77, 0.81]
Exposed to hazardous equipment	1.52	0.00 [1.43, 1.60]	0.70	0.00 [0.68, 0.73]
Extremely bright or inadequate lighting	1.30	0.00 [1.24, 1.36]	0.91	0.00 [0.90, 0.92]
Very hot or cold temperatures	1.67	0.00 [1.59, 1.76]	0.86	0.00 [0.85, 0.88]
Cramped work space, awkward positions	1.12	0.00 [1.06, 1.18]	0.82	0.00 [0.80, 0.84]
Cognitive complexity at work
Freedom to make decisions	1.00	0.00 [1.00, 1.00]	0.83	0.00 [0.81, 0.86]
Coaching and developing others	1.75	0.00 [1.64, 1.87]	0.85	0.00 [0.83, 0.87]
Making decisions and solving problems	1.58	0.00 [1.49, 1.66]	0.86	0.00 [0.85, 0.88]
Thinking creatively	1.68	0.00 [1.58, 1.78]	0.78	0.00 [0.76, 0.80]
Resolving conflicts and negotiating with others	1.82	0.00 [1.72, 1.92]	0.86	0.00 [0.84, 0.87]
Covariance and correlations
Cognitive complexity with Hazardous exposure	−0.03	0.00 [−0.04, −0.02]	−0.20	0.00 [−0.24, −0.15]

Notes: CFI = Comparative Fit Index; CI = Confidence Interval; RMSEA = Root Mean Square Error of Approximation; SEM = Structural Equation Modeling; SRMR = Standardized Root Mean Square Residual; TLI = Tucker-Lewis Index. N = 1,800. X²(df = 25, p = .00) = 304.69, RMSEA = 0.07 (95% CI [0.07, 0.09], CFI = 0.98, TLI = 0.96, SRMR = 0.05). Unlike Pearson’s correlations based on raw data, correlation and covariance in the CFAs under the SEM framework adjust for factor loadings and measurement errors, etc. The missing data for each correlation and covariance is below 10%. Covariance and correlations among individual items not reported for brevity.

Findings From the Mediation Analysis

Our full structural model presented in Table 3 examined whether occupational exposures (occupational hazards and occupational complexity) in various work environments mediate the association between education and later-life cognitive function, accounting for cognitive function at baseline (between ages 51-60). The mediation model results predicting cognitive function also had good model fit, X² (df = 101, p = .00) = 607.18, CFI = 0.96, TLI = 0.95, RMSEA = 0.05 (95% CI [0.05, 0.06]).

Table 3.

Mediation Model Results Predicting Cognitive Function at 65+

Variables	Unstandardized estimates	p [95% CI]	Standardized estimates	p [95% CI]
Cognitive complexity	0.01	0.84 [−0.06. 0.07]	0.01	0.23 [−0.04, 0.05]
Exposure to hazards	−0.09	0.01 [−0.15, −0.02]	−0.06	0.02 [−0.11, −0.02]
Statistical controls
Years of education	0.02	0.001 [0.01, 0.03]	0.09	0.01 [0.04, 0.14]
Female	0.05	0.08 [−0.01, 0.10]	0.04	0.06 [−0.01, 0.09]
Race/ethnicity
Hispanic	0.17	0.00 [0.09, 0.25]	0.09	0.00 [0.05, 0.14]
Black	0.01	0.93 [−0.06, 0.07]	0.002	0.55 [−0.04, 0.05]
Depressive symptoms	−0.06	0.00 [−0.07, −0.04]	−0.19	0.00 [−0.23, −0.15]
Baseline cognition	0.06	0.00 [0.06, 0.07]	0.42	0.00 [0.38, 0.47]
Age at the last wave of cognition measures	−0.02	0.00 [−0.03, −0.01]	−0.08	0.00 [−0.12, −0.04]
Indirect effects predicting cognitive complexity
Years of education	0.05	0.00 [0.04, 0.06]	0.38	0.00 [0.34, 0.42]
Indirect Effects predicting Exposure to Occupational Hazards
Years of education	−0.04	0.00 [−0.05, −0.03]	−0.28	0.00 [−0.32, −0.24]
Female	−0.34	0.00 [−0.37, −0.30]	−0.41	0.00 [−0.45, −0.37]

Notes: CFI = Comparative Fit Index; CI = Confidence Interval; RMSEA = Root Mean Square Error of Approximation; TLI = Tucker-Lewis Index. N = 1,694. Depressive symptoms were measured by Center for Epidemiological Studies Depression (CESD). A higher CESD score indicates higher levels of depression. White is the reference category for race/ethnicity. X² (df=101, p = .00) = 607.18, RMSEA = 0.05 (95% CI [0.05, 0.06]), CFI = 0.96, TLI = 0.95.

The standardized indirect effect of education through cognitive complexity was not significant (a₁b₁ = 0.004, p = .84). Although one standard deviation (SD) increase in years of education was associated with a 0.38 SD increase in work complexity (a₁ = 0.38, p = .00), the standardized coefficient from occupational cognitive complexity to cognitive function was not significant in our model (b₁ = 0.01, p = .23). See Figure 1 for more information. This indicates that although higher levels of education increase the opportunity for people to be in a cognitively complex occupational environment, the cognitive complexity of the work environment did not explain any additional effects of education on cognitive function in later life.

Figure 1.

Structure model. Only standardized coefficients are reported in the figure. The garnet arrows indicate positive correlations whereas the gold arrows indicate negative correlations. Covariates (gender, race, depressive symptoms, baseline cognitive, and age at the last wave of cognition measure were accounted for) in the mediation analysis. *p < .05. **p < .001.

In contrast, the standardized indirect effect of education through exposure to occupational hazards was significant (a₂b₂ = 0.02, p = .01), even when accounting for gender, race/ethnicity, depressive symptoms, baseline cognition, and age. One SD increase in education was associated with a 0.30 decrease in the SD of cognitive hazards (a₂ = −0.28, p = .00), and cognitive hazards were associated with a 0.06 SD reduction in cognitive function in later life (b₂ = −0.06, p = .01). Moreover, cognitive hazards reduced the effect of education on cognitive function by 17%. This suggests that although education remains an important predictive factor in shaping cognitive outcomes in later life, cognitively hazardous work environments explain almost one-fifth of this effect.

Sensitivity Analyses

To ensure the robustness of our findings, we conducted several sensitivity analyses. We initially tested occupational complexity and hazard as separate mediators when building the structural model piecewise. These preliminary results show that the direction and significance of the two mediators were similar to those of our final model. We also completed multiple group analyses in Mplus to examine whether the indirect effect of the occupational environments differs by racial/ethnic identity. Findings suggest that the indirect effect of occupational hazards is largely consistent across racial and ethnic groups. We also examined indirect effects using alternative cognitive outcome measures, including the 27-point cognitive scale, the probability of dementia, and non-dementia cognitive impairments. However, very few direct or indirect effects were detectable with these alternate measures, likely because of the relatively young age of the respondents at cognitive assessment or the relatively short time frame between the outcomes and baseline cognition (making decline more difficult to detect).

Additionally, we analyzed a mediation model with a dichotomous educational attainment variable (high school degree or not). Results show that the direction and size of the direct and indirect effect remained about the same as the model using years of education. To understand the impact of occupational and socioeconomic status on cognitive function, we also tested the household poverty ratio as a socioeconomic control variable in our model. Higher household earnings were positively significantly associated with later life cognitive function (β = 0.07, p = .00) in our mediation model. However, its inclusion has no impact on the direction and significance of our direct and mediation effect. To isolate the effect of occupational environments regardless of household earnings, we decided not to include household income in our mediation model for this exploratory study considering its high correlation with occupational environments.

Discussion

Our findings support that education is a well-established nongenetic predictor of cognitive outcomes in later life (Liu et al., 2022; Reuter-Lorenz et al., 2023). Education is completed during the early phases of the life course, with cognitive implications related to a variety of life course exposures. A key explanation for the persisting effects of education on cognitive function in later life is that learning during early life is thought to help people develop more sophisticated cognitive skills that promote cognitive function in adulthood and cognitive resilience as people age (Liu et al., 2022; Lövdén et al., 2020; Reuter-Lorenz et al., 2023). However, more years of education are often associated with more privileged “downstream” conditions such as better occupational environment, job status, income, health behavior, etc. (Lövdén et al., 2020). To be specific, educational attainment also influences the types of occupational exposures people have over the many decades they spend working. Occupational exposures have been shown to have associations with cognitive outcomes in later life, accounting for (i.e., above and beyond) educational attainment (Carr et al., 2020; Grzywacz et al., 2016). This study specifically focused on the effect of two types of occupational exposures—cognitively complex occupations and occupational hazards—as potential explanatory mechanisms (i.e., mediators) in the association between education and cognitive function in later life. Our results show that although cognitively complex occupational exposures are beneficial to cognitive function, they do not explain the effect of education on cognitive outcomes after age 65. On the other hand, hazardous occupational exposures have a significant negative association with cognitive function, and they explain almost one-fifth (17%) of the effect of education on cognitive function in later life.

To contextualize our findings, we identified the top occupations for each of the occupational hazards and complexity measures using O*Net data. Specifically, we weighted each occupational measure based on their CFA factor loadings, and for the overall occupational hazard and complexity measures we summed them to approximate the latent constructs we created. We show the top 10 high scores across each of the occupational hazard and cognitively complex measures (shown in Supplementary Tables 1 and 2, respectively). We find informative examples to help interpret our results. For instance, electrical power-line installers and repairers were among the top 10 occupations with exposure to high places, hazardous equipment, extreme lighting, and relatively low rankings on occupational complexity measures, contributing to their high ranking in total exposure to occupational hazards. In contrast, although chefs and head cooks were among the top 25 occupations that had exposure to very hot or very cold temperatures, they ranked 3rd for coaching and developing others, making them high in overall occupational complexity.

Our findings suggest that reducing hazardous occupational exposures has the potential to mitigate some of the negative effects of poor educational outcomes. Although one's exposure to occupational hazards is influenced by educational attainment during earlier phases of the life course, prioritizing occupational health and safety for all adult workers in hazardous occupations may help disrupt the association between educational disadvantages and cognitive decline. More studies are needed to explore potential inequitable exposure to occupational hazards among population subgroups (e.g., racial and ethnic minorities, women, etc.) with limited access to education. These findings also suggest that targeted interventions for those in hazardous jobs may reduce ADRD risks and promote better cognitive health in later life. Besides continuing to promote cognitive health through equitable early life education (Liu et al., 2022; Reuter-Lorenz et al., 2023), future research should also explore the feasibility of interventions at the individual, organizational, state, and federal level policy and program levels, such as ensuring access to personal protective equipment, strengthening human–robot interaction in hazardous occupational environments over the life course.

Our results should be interpreted in the context of our study limitations. Although the mediating effect of hazardous jobs on education is robust, our results may be more conservative due to the limitations imposed by our data available. First, we measure occupational exposure in the period leading up to retirement, but these occupational exposures may not represent the life course exposures that individuals have incurred over the course of their occupational history. Although for most people the occupation they spend the majority of their working lives is consistent with the job they have in the period leading up to retirement (A. Sonnega, Personal Communications, June 18, 2024), individuals in more hazardous jobs may be more likely to change jobs or leave the workforce before reaching 51–60 due to the physical demands of the work. With our study design, only the most robust workers are likely to still be engaged in hazardous jobs by the time they reach later phases of their working lives. We are aware of no research that has specifically evaluated if and when transitions in work occur over the life course to avoid long-term harm to physical or cognitive health. The HRS recently developed a new retrospectively collected life history data set that includes detailed occupational histories (HRS, 2024). Future research should explore if accounting for the full spectrum of occupational exposures across the adult life course, especially accounting for those who change jobs or leave the workforce before full retirement ages, shows more significant mediation effects for educational attainment on cognitive outcomes in later life.

In addition, it is also important to note that we were unable to account directly for work and non-work lifestyle factors that may be correlated with both education and occupational environments. Our occupations likely influence the activities we engage in before and after work, such as going to the gym versus drinking alcohol to cope with work stress. In addition, O*NET variables are linked to individual jobs based on average scores calculated at the occupational level. Consequently, our measures do not capture individual variations in exposures within each job, masking potentially much larger effects for certain individuals relative to others within the same occupation. Finally, because the newly developed Hudomiet measures for cognition and ADRD measures were only available in HRS from 2000 to 2016, participants in our sample were likely too young to show significant cognitive decline or for dementia to be detectable. Future research should explore longer-term changes in cognitive function to assess ADRD risks associated with these occupational exposures.

Despite these limitations, our study finds evidence that exposure to hazardous occupational environments between ages 51 and 60 years old is a significant mechanism linking limited educational attainment and poor late-life cognition. Our findings add to the growing body of research that suggests that lifestyle exposures over the life course may be the most promising strategy for reducing cognitive risks as we age (Lee et al., 2022; Livingston et al., 2020). Our findings tentatively show that prioritizing improvements in occupational health and safety may help mitigate some cognitive consequences of limited educational attainment. Policies and interventions that address occupational hazards could help disrupt the association between educational disadvantages and cognitive decline, particularly among workers in occupations with high levels of exposure to hazards.

Funding

This work is supported by the Network on Education, Biosocial Pathways, and Dementia in Diverse Populations through the National Institute on Aging (R24AG077433).

Conflict of Interest

The authors have no conflict of interest to disclose.

Data Availability

The data used in this study is available through the Health and Retirement Study virtual desktop infrastructure (VDI) system. The study findings were reviewed by HRS to ensure the confidentiality of their respondents. The study was not preregistered.

Author Contributions

Q. C. conceptualized the paper, selected the sample, coded variables, conducted statistical analysis, produced result tables and figures, and drafted the manuscript. D. C. contributed to conceptualization, sample section, variable coding, and contributed to revising the paper. M. T. provided guidance on statistical analysis and methodological design, software application, and manuscript revision.