Pathways explaining racial/ethnic disparities in incident all-cause dementia among middle-aged US adults

Abstract

INTRODUCTION:

Racial disparities in dementia incidence exist, but less is known about their presence and drivers among middle-aged adults.

METHODS:

We used time-to-event analysis among a sample of 4,378 respondents (age 40–59y at baseline) drawn from the third National Health and Nutrition Examination Surveys (NHANES III) with administrative linkage--spanning the years 1988–2014—to evaluate potential mediating pathways through socioeconomic status (SES), lifestyle, and health-related characteristics.

RESULTS:

Compared with Non-Hispanic White (NHW) adults, Non-White adults had higher incidence of AD-specific (HR=2.05, 95%CI: 1.21, 3.49) and all-cause dementia (HR=2.01, 95%CI: 1.36, 2.98). Diet, smoking, and physical activity were among characteristics on the pathway between race/ethnicity, SES and dementia, with health mediating effects of smoking and PA on dementia risk.

DISCUSSION:

We identified several pathways that may generate racial disparities in incident all-cause dementia among middle-aged adults. No direct effect of race was observed. More studies are needed to corroborate our findings in comparable populations.

INTRODUCTION

Alzheimer’s disease (AD), the 7^th leading cause of death and a leading cause of old age disability in the United States (US) in 2021[1], is considered among the most common causes of dementia [2–5]. AD is characterized by progressive episodic memory deterioration which is then followed by impairment across other cognitive domains.[3] Despite unclear etiology, AD is likely trigged by age-related and progressive Aβ-amyloid brain deposition,[6] and by a second pathological hallmark namely neurofibrillary tangles (NFT), which arise from hyper-phosphorylated tau proteins.[7] An estimated 6.5 million US adults aged 65 years and older are living with AD, with estimates suggesting an increase of up to 13.8 million by 2060[1]. Less frequently occurring dementia sub-types include vascular dementia (VaD), mixed dementias, dementia with Lewy bodies (DLB) and Parkinson’s disease with dementia (PD-D). In the absence of effective therapeutics, identification of modifiable risk and protective factors is critical to mitigate the expected rising burden of morbidity and mortality associated with AD and dementia as well as disparities therein.

The Lancet Commission on dementia prevention, intervention and care stated that addressing modifiable risk factors might prevent or delay up to 40% of dementia cases [8]. In contrast, genetic endowment explains a modest fraction of its prevalence. In fact, about half of AD cases not carrying the APOEε4 risk gene allele and 20–30% of US individuals carrying at least one copy of this risk allele [8, 9]. Modifiable risk factors include but may not be limited to educational attainment, health behaviors (e.g., smoking, physical activity, diet), social and mental engagement, and comorbidities (e.g., hypertension, type 2 diabetes) [1]. There may be other risk factors more common among Non-Hispanic Blacks (NHB), Hispanic Americans, Asian/Pacific Islanders, and Native Americans, in comparison to non-Hispanic Whites (NHW), but they are yet to be identified due to the lack of inclusion in previous research [10]. In the US, older NHB and Hispanic Americans are more likely to have AD and other dementias in comparison to older NHW adults [1]. Yet, older NHW live with AD and related dementias than any other US racial group [11]. For older adults, heterogeneity within and between racial/ethnic groups play a role in our understanding of modifiable risk factors [10, 12–16].

Racial/ethnic disparities are well documented in both AD and all-cause dementia risk as well as their associated risk factors [12, 17–35].For instance, based on findings from a longitudinal study by Xiong and colleagues, major AD risk factors modified racial differences [36]. The racial differences, when statistically significant, occurred only with older age among APOE ε4 negative individuals, but also with much younger age among APOE ε4 positive individuals. The racial differences, while statistically significant, were observed at older ages among APOE ε4 negative individuals but younger APOE ε4 positive adults [36]. They concluded that racial disparities in the risk of developing AD dementia depended on multiple AD risk factors in an interactive manner[36]. Understanding the drivers of these differences is critical to advance population health equity in light of an increasing percentage of older adults in the US comprised of racial/ethnic minority groups [37, 38].

We expand on the extant literature by examining racial and ethnic disparities in all-cause and AD-specific dementia, and for the first time, some of their underlying pathways among middle-aged adults. We use measures from the National Health and Nutrition Survey (NHANES) on socioeconomic, lifestyle, health, and cognition factors linked to Medicare claims data and mortality data from the National Death Index. Whereas prior work has typically focused on older adults, an important contribution is that we focus on middle-aged adults which allows us to evaluate the presence of and parse disparities at midlife which may help inform the disparities observed in later adulthood.

MATERIALS AND METHODS

Database

We use data from the NHANES III, a national probability survey conducted by the National Center for Health Statistics (NCHS). Since 1988, the NHANES investigators have assessed population health and nutrition among US adults using cross-sectional surveys by way of in-home interviews and in-depth examinations using mobile examination centers (MEC) [39]. The NHANES data were linked with National Death Index (NDI) and Centers for Medicare & Medicaid Services (CMS)-Medicare records using the approach described in Appendix I. The current study was approved by the Institutional Review Board of the National Institute on Aging, Intramural Research Program for ethical treatment of participants.

Study sample

Figure 1 details participant selection and numbers of incident AD and DEMENTIA cases. We selected NHANES III (1988–1994) participants aged 40–59y, with complete CMS-Medicare linkage data, and with HMO exclusion. Thus, of the initial 33,199 participants (aged 1–90 y) recruited in NHANES III (1988–1994), our final sample consisted of 4,378 participants. No other exclusions were applied because of multiple imputation (% missing<10%).

FIGURE 1. Participant Flowchart.

Abbreviations: AD=Alzheimer’s Disease; CMS=Centers for Medicare and Medicaid; NHANES=National Health and Nutrition Examination Surveys

Incident AD and all-cause dementia

Incident AD and DEMENTIA cases were ascertained using data from the CMS Chronic Condition Data Warehouse Categories. We classified persons with AD as those with ICD-9 code 331.0. We classified persons with DEMENTIA as those with at least one ICD-9 code of 331.0 or a list of others provided in Appendix I.

Exposure and effect modifier

Self-reported race/ethnicity and sex were the primary effect modifiers in our analyses. We defined a series of racial/ethnic contrasts (RACE_ETHN), with NHW as the reference group in most analyses.

Mediators

Socio-economic status

We obtained a measure of socioeconomic status (SES) by combining for each respondent their poverty income ratio (PIR, computed as percentage of poverty line) and level of education into an unweighted, standardized score with mean 0 and standard deviation 1.

Lifestyle and social support factors

We detail our operationalization of lifestyle and social support factors (SMOKING, ALCOHOL, DIET, NUTR [nutrition], PA [physical activity], and SS [social support]) in Appendix II. In brief, we used principal components analysis (PCA) to identify and combine positively correlated measured variables within each construct. We then combined these variables into the aforementioned constructs by taking the unweighted mean of their standardized values.

Health

We constructed a “HEALTH” marker composed of self-reported health (Excellent, very good, good, fair, poor), a co-morbidity index (unweighted sum of the self-reported presence of arthritis, congestive heart failure, stroke, asthma, chronic bronchitis, emphysema, hay fever, cataracts, goiter, thyroid disease, lupus, gout, skin cancer, other cancer), body mass index (BMI), and an allostatic load (AL; details in Appendix II [40].) score. We coded each of the four inputs such that higher scores reflected poorer health and standardized with mean 0 and standard deviation 1. The average of the four z-scores for each respondent was taken as their HEALTH score.

Cognitive performance tests and Poor cognition (COGN) summary PCA score

Three cognitive tests were administered to participants aged 20–59yo, simple reaction time (SRT), symbol-digit substitution test (Errors: SDS-E and Latencies: SDS-L) and serial digits learning (Trials to completion: SDL-TTC and total errors: SDL-TE) from which five cognitive test scores were computed (Appendix III). We restricted this sample to those aged 40–59 years at examination. The PCA score obtained from these 5 scores, reflecting poor cognitive performance, was tested as a predictor for all-cause dementia in the sub-sample with complete cognitive test scores (N=1,863), as a means to validate the dementia outcome in the sample of middle-aged adults followed for up to 26 years.

Covariates

We accounted for age at examination in years, marital status (Never married, married, divorced, widowed, other), household size, and whether the respondent resided in an urban or rural area at the time of the study.

Statistical methods

Using multiple imputed covariate and mediator data with chained equations (mi impute, 5 imputations, 10 iterations), we first characterized the sample using population means and proportions and used regression models to examine differences in characteristics across racial/ethnic groups. Imputed variables in the final selected sample had on average <10% missing data. We calculated incidence rates (IRs) of all-cause dementia in the overall sample and compared cumulative incidence across racial groups. We then estimated a series of Cox proportional hazards (PH) models for which we sequentially adjusted for SES, lifestyle and health factors. Age was used as the time metric, with age at entry as the start point and age at AD/DEMENTIA or censoring as the exit time. Proportionality of the hazards was tested in part of the analysis using Schoenfeld residuals.

We used discrete-time survival analysis within a GSEM framework to evaluate the potential mediating effects of each factor. Specifically, we used a structured approach to evaluate mediating pathways between RACE_ETHN and incident dementia with covariate adjustment. Specifically, 6 specific pathways were tested, allowing for all direct effects:

1. RACE_ETHN → SES → DEMENTIA;

2. RACE_ETHN → SES →LIFESTYLE → DEMENTIA;

3. RACE_ETHN → SES → LIFESTYLE → HEALTH → DEMENTIA;

4. RACE_ETHN → SES→HEALTH→DEMENTIA;

5. RACE_ETHN→LIFESTYLE→HEALTH→DEMENTIA;

6. RACE_ETHN→ HEALTH→DEMENTIA; with (C) hypothesized to be the main pathway, as described in Figure 2. Indirect effects across imputations were also combined using Rubin’s rule[41]. Type I error was set at 0.05.

FIGURE 2. GSEM full model and hypothesized pathway.

Abbreviations: AD=Alzheimer’s Disease; ALCOHOL= alcohol consumption, z-score; DIET/NUTR=diet and nutritional biomarkers z-score variable (2 dietary quality measures and 4 nutritional biomarkers); HEALTH=Health-related factors as mean of z-scores for allostatic load, self-rated health, co-morbidity index and body mass index; LIFESTYLE=Lifestyle-related factors composed of social support, physical activity, diet/nutritional biomarkers, smoking and alcohol consumption using means of z-scores for related measured variables; N’=number of observations; MA=Mexican American; NHANES III=Third National Health and Nutrition and Examination Survey; NHB=Non-Hispanic Blacks; NHW=Non-Hispanic Whites; PA=Physical activity z-score variable (3 measured variables); RACE_ETHN=racial/ethnic contrast; SES=Socio-economic status mean of z-scores composed of poverty income ratio and education (years); SMOKING=smoking z-score variable (2 measured variables); SS=Social Support z-score variable (5 measured variables). See Methods section for more details.

Notes: Plain arrows are statistically significant associations (p<0.05) within the hypothesized pathway; Dashed arrows are statistically significant associations (p<0.05) outside the hypothesized pathway.

We used Stata 16.0 (StataCorp, College Station, TX) [42] in all analyses. We incorporated sampling weights, primary sampling units, and strata calculated by the NHANES researchers to account for sampling design complexity [43].

RESULTS

Overall, a final sample of N=4,378 represented a population of 54,629,806 individuals aged 40–59y, with 77.7% being NHW in the weighted sample, and notable differences indicating lower SES, poorer diet quality, lower levels of nutritional biomarkers and poorer health among others in minority groups compared with NHW middle-aged adults. More specifically, the mean age for NHW (48.1y±0.2) was not statistically different than that of Non-White adults (47.8y±0.2). We observed racial differences in the distribution of several characteristics including sex (54.0% female among Non-White vs. 50.5% among NHW, p<0.05), percentage of adults living in urban vs rural areas (p<0.05), household size (p<0.05) and marital status (p<0.05), with greater percentages living in urban areas, never married and with larger household sizes among non-White adults compared to NHW. Racial differences, with a poorer profile observed among non-White participants, were also detected across socioeconomic characteristics (poverty income ratio, educational attainment, and socioeconomic status) as well as health and lifestyle markers including dietary habits, nutritional biomarkers, physical activity, self-rated health, co-morbidity index, allostatic load, and BMI. Compared to Non-White adults, NHW had, on average, lower levels of interaction with friends and neighbors and lower church attendance, but greater attendance of meetings within clubs and organizations. In this sample of middle-aged adults (40–59y), DEMENTIA had an IR of 2.5/1,000 person-years (P-Y) with a 95% CI: 2.1–3.0, while cumulative incidence was significantly higher among Non-White vs. NHW participants (7.5% vs. 4.5%). The average follow-up time to dementia incidence or censoring was 19.9 years, while among individuals diagnosed with dementia, the average age at diagnosis in this sample was 70.5 years.

Table 2 shows the quantified differences in incident all-cause and Alzheimer’s disease dementia for Non-White relative to NHW adults. In Model 1, which adjusts for age, Non-White adults had an increased risk of all-cause dementia compared with NHW (Model 1: HR=2.01, 95%CI: 1.36, 2.98) as well as Alzheimer’s disease dementia (Model 1: HR=2.05, 95%CI: 1.21, 3.49) suggesting that Non-White adults are approximately twice as likely to experience all-cause or AD-specific dementia at any given time point, relative to NHW adults. However, after sequential covariate adjustment, the disparities were attenuated for both all-cause (Model 4: HR=1.15, 95%CI: 0.70, 1.88) and AD-specific dementia (Model 4: HR=1.51, 95%CI: 0.92, 2.55).

TABLE 2.

Racial/ethnic disparities in incident all-cause and Alzheimer’s Disease dementia among middle-aged adults (Unweighted N=4,378; Weighted N=54,629,806): Cox proportional hazards models; NHANES III, 1988–1994^a

	Loge(HR)	(SE)	P
All-cause dementia
Model 1	+0.70	(0.20)	0.001
Model 2	+0.16	(0.25)	0.53
Model 3	+0.18	(0.24)	0.45
Model 4	+0.14	(0.25)	0.59
Alzheimer’s Disease dementia
Model 1	+0.72	(0.27)	0.011
Model 2	+0.34	(0.26)	0.20
Model 3	+0.46	(0.25)	0.074
Model 4	+0.41	(0.25)	0.11

Abbreviations: AD=Alzheimer’s Disease; ALCOHOL= alcohol consumption, z-score; DIET/NUTR=diet and nutritional biomarkers z-score variable (2 dietary quality measures and 4 nutritional biomarkers); HEALTH=Health-related factors as mean of z-scores for allostatic load, self-rated health, co-morbidity index and body mass index; HR=Hazard Ratio; LIFESTYLE=Lifestyle-related factors composed of social support, physical activity, diet/nutritional biomarkers, smoking and alcohol consumption using means of z-scores for related measured variables; MA=Mexican American; N=Number of participants; N’=number of observations; NHANES III=Third National Health and Nutrition and Examination Survey; NHB=Non-Hispanic Blacks; NHW=Non-Hispanic Whites; PA=Physical activity z-score variable (3 measured variables); RACE_ETHN=racial/ethnic contrast; SES=Socio-economic status mean of z-scores composed of poverty income ratio and education (years); SMOKING=smoking z-score variable (2 measured variables); SS=Social Support z-score variable (5 measured variables). See Methods section for more details.

^aValues are β ± SE (Log_e(HR)), considering sampling design complexity (PSU and strata), across 5 imputations with 10 iterations.

Model 1: adjusted for age; Model 2: adjusted for demographic factors other than age and SES score; Model 3: Model 2 further adjusted for lifestyle-related factors (average of z-scores of measured variables for SMOKING, ALCOHOL, DIET, NUTR, SS and PA); Model 4: Model 3 + health-related factors (HEALTH score); Model 5: Full model with cognitive test PCA score. Findings from “Other ethnicity” among men was not presented due to small number of events for AD and disclosure risk.

^bP<0.05 for sex×RACE_ETHN interaction in unstratified model.

^*P<0.05

^**P<0.01

^***P<0.001 for null hypothesis of Log_e(HR)=0.

As shown in Table 3 and Figure 3, the total effect (TE) of race/ethnicity on all-cause dementia incidence was statistically significant (TE = +0.5430±0.2446, P=0.031), suggesting a greater hazard of all-cause dementia among Non-White adults compared with NHW. The TE was generated by several pathways, including: (1) Non-White vs. NHW (−)→ SES(−)→ DEMENTIA; (2) Non-White vs. NHW (−)→ SES (−)→ HEALTH (+)→ DEMENTIA; (3) Non-White vs White (+)→ HEALTH (+)→DEMENTIA. Those pathways were formally tested, and their indirect effect was deemed statistically significant at a type I error of 0.05. It is worth noting, that about half of the total effect was explained by Pathway (1), (IE=+0.2952±0.0810, P<0.001; TE=+0.5430±0.2446, p=0.031), while the remaining two pathways explained about 13% of the TE. While pathways through LIFESTYLE factors were not statistically significant as a whole in explaining racial disparities in dementia, whether directly or through HEALTH, Figure 3 shows in more detail how each of these LIFESTYE factors were at play. Based on this Figure and quantitative results in Table 3, the total effect of RACE_ETHN on DEMENTIA was mediated by the following lifestyle-related pathways that would explain part of the dementia risk racial disparity: RACE_ETHN (−)→SES (−)→SMOKING (+)→HEALTH (+)→DEMENTIA; RACE_ETHN (−)→SES (+)→PA (−)→HEALTH (+)→DEMENTIA; RACE_ETHN (−)→SES (+)→DIET (−)→DEMENTIA. Pathways with smoking and PA are part of the main hypothesized pathway in Figure 2, involving HEALTH as a more proximal mediator to dementia risk. In contrast, DIET had a direct inverse relationship with dementia incidence. SS and ALCOHOL were not among mediators explaining these racial disparities. Moreover, a pathway going directly from RACE_ETHN into SMOKING, followed by HEALTH and DEMENTIA had a net effect of reducing the risk for dementia. The pathway [RACE_ETHN→SES→NUTR→HEALTH→DEMENTIA] also had a similar net effect. Furthermore, other pathways were uncovered that only included HEALTH and SES among mediators. There was no direct effect of RACE_ETHN on dementia, although the direct inverse effect of SES on dementia was statistically significant. In the sub-sample with complete cognitive test scores on this age sub-group in NHANES III (N=1,863), the poor cognitive performance PCA score was directly associated with incident dementia, with 1 SD increase being associated with 40% greater hazard of dementia, (HR=1.40, with a 95% CI of 1.06–1.86, p=0.020).

TABLE 3.

Pathways from race/ethnicity (Non-White vs. NHW) to all-cause dementia through of modifiable risk factors and cognitive performance among middle-aged adults (Age_base: 40–59 y); NHANES III, 1988–1994^a

Unweighted N’ (both phases)	(N’=21,701)
	β	(SE), p
Main pathway
RACE_ETHN→SES (β12)	−0.547 ***	(0.057), p<0.001
SES→SS (β23)	+0.044 **	(0.016), p<0.01
SES→PA(β24)	+0.098 **	(0.034), p<0.01
SES→DIET(β25)	+0.243 ***	(0.033), p<0.001
SES → NUTR (β26)	+0.170 ***	(0.029), p<0.001
SES → SMOKING (β27)	−0.204 **	(0.044), p<0.01
SES → ALCOHOL (β28)	+0.066	(0.037), p=0.082
SS → HEALTH (β39)	0.041	(0.026), p=0.121
PA → HEALTH (β49)	−0.154 ***	(0.018), p<0.001
DIET → HEALTH (β59)	−0.009	(0.017), p=0.628
NUTR → HEALTH (β69)	+0.040 *	(0.018), p<0.05
SMOKING → HEALTH (β79)	+0.072 **	(0.021), p<0.01
ALCOHOL → HEALTH (β89)	−0.052 **	(0.014), p<0.01
Selected direct effects on final outcomes
RACE_ETHN→DEMENTIA (β110)	+0.151	(0.255), p=0.558
SES → DEMENTIA (β210)	−0.540 ***	(0.136), p<0.001
SS → DEMENTIA (β310)	0.074	(0.069), p=0.286
PA → DEMENTIA (β410)	−0.161	(0.164), p=0.329
DIET → DEMENTIA (β510)	−0.265 *	(0.119), p<0.05
NUTR → DEMENTIA (β610)	0.243	(0.159), p=0.134
SMOKING → DEMENTIA (β710)	−0.018	(0.095), p=0.85
ALCOHOL → DEMENTIA (β810)	−0.028	(0.101), p=0.781
HEALTH → DEMENTIA (β910)	+0.352 *	(0.147), p<0.05
Other effects between endogenous variables
SES→HEALTH (β29)	−0.171 ***	(0.024), p<0.001
Other direct effects of race
RACE_ETHN→SS (β13)	+0.069 *	(0.026), p<0.05
RACE_ETHN→PA(β14)	0.050	(0.039), p=0.212
RACE_ETHN→DIET(β15)	−0.080	(0.045), p=0.083
RACE_ETHN→NUTR(β16)	−0.084 *	(0.037), p<0.05
RACE_ETHN→SMOKING(β17)	−0.348 ***	(0.046), p<0.001
RACE_ETHN→ALCOHOL(β18)	−0.010	(0.042), p=0.816
RACE_ETHN→HEALTH(β19)	+0.115 **	(0.032), p<0.01
Selected Indirect effects
RACE_ETHN → SES → DEMENTIA (βA)
1	+0.294 ***	(0.081), p<0.001
2	+0.307 ***	(0.081), p<0.001
3	+0.294 ***	(0.079), p<0.001
4	+0.284 ***	(0.081), p<0.001
5	+0.297 ***	(0.083), p<0.001
Rubin’s rule	+0.2952 ***	(0.0810)
RACE_ETHN → SES → LIFESTYLE → DEMENTIA (βB)
1	+0.016	(0.02), p=0.4
2	+0.015	(0.022), p=0.487
3	+0.023	(0.021), p=0.272
4	+0.019	(0.022), p=0.383
5	+0.017	(0.022), p=0.445
Rubin’s rule	+0.018	(0.0214)
RACE_ETHN → SES → LIFESTYLE → HEALTH → DEMENTIA (βC)
1	+0.006	(0.003), p=0.061
2	+0.005	(0.003), p=0.087
3	+0.005	(0.003), p=0.099
4	+0.005	(0.003), p=0.076
5	+0.006	(0.003), p=0.071
Rubin’s rule	+0.0054	(0.0030)
RACE_ETHN → SES → HEALTH → DEMENTIA (βD)
1	+0.033 *	(0.014), p<0.05
2	+0.033 *	(0.014), p<0.05
3	+0.033 *	(0.015), p<0.05
4	+0.033 *	(0.014), p<0.05
5	+0.032 *	(0.014), p<0.05
Rubin’s rule	+0.0328 *	(0.0142)
RACE_ETHN → LIFESTYLE → HEALTH → DEMENTIA (βE)
1	−0.013	(0.007), p=0.062
2	−0.010	(0.006), p=0.078
3	−0.009	(0.006), p=0.102
4	−0.011	(0.006), p=0.06
5	−0.014	(0.008), p=0.068
Rubin’s rule	−0.0114 *	(0.0066)
RACE_ETHN → HEALTH → DEMENTIA (βF)
1	+0.042 *	(0.021), p<0.05
2	+0.038	(0.02), p=0.055
3	+0.038	(0.02), p=0.06
4	+0.040	(0.021), p=0.053
5	+0.044 *	(0.023), p<0.05
Rubin’s rule	+0.0404 *	(0.0210)
TOTAL EFFECT OF RACE_ETHN	+0.5430	(0.2446), p=0.031

Abbreviations: AD=Alzheimer’s Disease; ALCOHOL= alcohol consumption, z-score; DIET/NUTR=diet and nutritional biomarkers z-score variable (2 dietary quality measures and 4 nutritional biomarkers); HEALTH=Health-related factors as mean of z-scores for allostatic load, self-rated health, co-morbidity index and body mass index; LIFESTYLE=Lifestyle-related factors composed of social support, physical activity, diet/nutritional biomarkers, smoking and alcohol consumption using means of z-scores for related measured variables; MA=Mexican American; N=Number of participants; N’=number of observations; NHANES III=Third National Health and Nutrition and Examination Survey; NHB=Non-Hispanic Blacks; NHW=Non-Hispanic Whites; PA=Physical activity z-score variable (3 measured variables); RACE_ETHN=racial/ethnic contrast; SES=Socio-economic status mean of z-scores composed of poverty income ratio and education (years); SMOKING=smoking z-score variable (2 measured variables); SS=Social Support z-score variable (5 measured variables). See Methods section for more details.

^aValues are path coefficients β ± SE or non-linear combinations of path coefficients to compute selected indirect effects, considering sampling design complexity (PSU and strata), across 5 imputations with 10 iterations. For indirect effects, 1 through 5 represent estimates for each extracted imputation. Rubin’s rule refers to pooled estimate across the 5 imputations using Rubin’s rule for point estimates and standard errors.

^*P<0.05

^**P<0.01

^***P<0.001 for null hypothesis of β=0.

FIGURE 3. GSEM model findings for Non-White vs. NHW racial/ethnic contrast vs. DEMENTIA, NHANES III (1988–1994): Final eligible sample (N=4,378; N’=21,701 observations).

Abbreviations: ALCOHOL= alcohol consumption, z-score; DIET/NUTR=diet and nutritional biomarkers z-score variable (2 dietary quality measures and 4 nutritional biomarkers); HEALTH=Health-related factors as mean of z-scores for allostatic load, self-rated health, co-morbidity index and body mass index; LIFESTYLE=Lifestyle-related factors composed of social support, physical activity, diet/nutritional biomarkers, smoking and alcohol consumption using means of z-scores for related measured variables; MA=Mexican American; N=Number of participants; N’=number of observations; NHANES III=Third National Health and Nutrition and Examination Survey; NHB=Non-Hispanic Blacks; NHW=Non-Hispanic Whites; PA=Physical activity z-score variable (3 measured variables); RACE_ETHN=racial/ethnic contrast; SES=Socio-economic status mean of z-scores composed of poverty income ratio and education (years); SMOKING=smoking z-score variable (2 measured variables); SS=Social Support z-score variable (5 measured variables); TE=Total effect; See Methods section for more details.

Notes: Plain arrows are statistically significant associations (p<0.05) within the hypothesized pathway; Dashed arrows are statistically significant associations (p<0.05) outside the hypothesized pathway; Red arrows are for positive (+) associations; Blue arrows are for inverse (−) associations.

Abbreviations: AD=Alzheimer’s Disease; IR=Incidence Rate; LCL=Lower Confidence Limits; NHANES III=Third National Health and Nutrition and Examination Survey; MA=Mexican American; NHB=Non-Hispanic Blacks; NHW=Non-Hispanic Whites; UCL=Upper Confidence Limits. See Methods section for more details.

Note: Groups 0, 6 and 12: All race/ethnicities combined; Groups 1, 7, 13: NHW, Groups 2, 8, 14: NHB, Groups 3, 9, 15: MA.

DISCUSSION

The present study tested racial/ethnic and socio-economic differences in dementia incidence among middle-aged adults, using national data on 4,378 respondents (age 40–59y at baseline) drawn from the third National Health and Nutrition Examination Surveys (NHANES III) with administrative linkage--spanning the years 1988–2014—to evaluate potential mediating pathways through socioeconomic status (SES), lifestyle, and health-related characteristics. Our key findings indicated that compared with Non-Hispanic White (NHW) adults, Non-White adults had higher incidence of AD-specific (HR=2.05, 95%CI: 1.21, 3.49) and all-cause dementia (HR=2.01, 95%CI: 1.36, 2.98). Diet, smoking, and physical activity were among characteristics on the pathway between race/ethnicity, SES and dementia, with health mediating effects of smoking and PA on dementia risk.

We found that the relationship between race and dementia was mediated by socioeconomic status (SES). Racial residential segregation is compounded with economic residential segregation as reflected by geographic socioeconomic variables like Area Deprivation Indices (ADI) when understanding health outcomes like dementia[44]. Comparing older Caribbean-born African American individuals with US-born African American individuals, a study detected cognitive variation according to socioeconomic status modifying race[45]. Differences in neuropsychological test performance between the two groups was explained by higher quality of education (resulting in higher earnings) among the Caribbean-born African American cohort[45]. According to the Canadian Community Health Survey (N=20,646 participants aged ≥60 years), socioeconomic status was found to mediate racial disparities in cognitive functioning [46]. This national level study demonstrates the importance of understanding that social determinants like race and SES have a combined negative effect on cognitive function [46].

Previous studies suggest that low SES may be associated with increases in risky behaviors which include smoking and substance use, poor dietary quality, a more sedentary lifestyle and the lack of access to quality resources [23–25, 30, 46–50]. Many of these behaviors were linked to increased dementia risk and age-related cognitive decline [12, 17–32, 46–50]. Reduced access to resources is a structural determinant that links low SES with dementia and especially among historically marginalized groups[12]. Low SES coupled with lack of social support have also been linked to additive chronic stress [51]. Accumulation of allostatic load is a mechanism by which chronic stressors such as low SES are thought to cause cognitive dysfunction[51]. Additionally, with low SES, chronic stress may trigger maladaptive responses, resulting in neuroendocrine, autonomic, as well as behavior modifications [51]. These modifications are thought to be associated with poor cognitive function [51]. Specifically, researchers have found the prefrontal cortex to be negatively affected by chronic stress, a byproduct of low SES[52]. These cortical changes can also be attributed to cognitive dysfunction. Thus, low SES can be attributed to a complex interplay of biological, physiological, and environmental factors which, in turn, results in cognitive dysfunction. Our finding that smoking status and low physical activity may mediate socio-economic disparities in dementia, particularly through poor cardio-metabolic and general health supports this general framework. Given that SES had a direct residual effect on dementia risk, suggests that other pathways may be at play that were not taken into consideration in this study.

Evidence for potential mediation of socio-economic and racial disparities in dementia by dietary quality is further demonstrated through previous studies[44, 46, 48, 53]. For instance, there was a racial gap in dietary quality in urban dwelling favoring White adults over African American adults within the HANDLS cohort [44]. The evidence leads credence to the theory that dietary quality compounded with existing disparities in race result in increased cognitive dysfunction. Nevertheless, unlike smoking and physical activity, dietary quality in our study had a direct impact on dementia risk without the mediating effect of poor cardio-metabolic and general health.

Nevertheless, our study indicated that in general, poor cardio-metabolic and general health explained part of the racial and socio-economic disparity in dementia incidence in this sample of middle-aged adults, with several mediating pathways working to explain the excess risk among Non-White adults over their NHW counterparts [e.g. Non-White vs. NHW (−)→ SES (+/−)→ PA/SMOKING (−/+)→ HEALTH (+)→ DEMENTIA; Non-White vs. NHW (−)→ SES (−)→ HEALTH (+)→ DEMENTIA]. This finding, while expected, is at odds with a finding from a previous study that including NHANES III participants who were older adults at baseline and followed-up for a similar amount of time (≤26 years). In this recent study, incidence of all-cause dementia among older adults in the US was significantly increased among Non-Hispanic Black women compared to NHW women, whereas Mexican-American women were at reduced AD risk compared with their NHW counterparts, especially upon further adjustment for SES and upstream factors[22]. SES markedly mediated the NHB-NHW women disparity in dementia, notably along with diet and physical activity[22]. Nevertheless, and unlike in our present study of middle-aged adults at baseline, the poor cardio-metabolic and general health (HEALTH) factor was not among the key mediators explaining racial/ethnic or socio-economic disparities in dementia risk alone or through lifestyle factors. This underscores age group differences in pathways explaining those disparities. Income group differences in pathways between race/ethnicity and dementia risk were observed in another national study of older adults using a comparable sample (i.e. 60+y at baseline), highlighting the importance of social support in reducing dementia risk within the lowest income category [54]. This was also a finding that was not observed among middle-aged adults, since social support did not have a major impact on dementia risk in our present study.

The current study has several strengths. First, the large sample size sufficiently powered our analyses to detect mediation effects across strata defined by the intersection of demographic characteristics. We used a nationally representative sample with Medicare linkage which allowed us to combine detailed information about respondents with their medical diagnoses. Studies strictly using claims data rely on accurate demographic reporting during patient encounters[55] and typically exclude micro-level, non-medical information. Classifying respondents with cognitive impairment using cognitive tasks is prone to measurement bias due to varying thresholds among demographic subgroups with different educational attainment and literacy. Thus, by combining a large, population-based survey with Medicare linkage, we were able to overcome limitations typically experienced in prior work that relies strictly on population-based survey or medical claims record data. In addition, due to the depth of the NHANES, we were able to test a variety of pathways that span multiple domains of risk. Our study is not without limitations. Despite the use of claims record to ascertain dementia status, studies have reported differential rates of persons receiving a dementia diagnosis by race/ethnicity as well as access to healthcare [56]. Further limitations include those typically reported in observational studies, including residual confounding, measurement error (including recall and social desirability bias), and potential selection bias due missing data on cognitive performance. However, the breadth of available covariates helps mitigate concern related to residual confounding.

Nevertheless, it remains important to contextualize these findings from the social determinants of health framework providing the impetus for further research and to inform clinical practice. Due to intersectionality, many of the characteristics such as social support and diet are crucial. For diet, some of the areas that are important to consider are access to healthy foods and access to grocery stores with fresh produce[57]. The composite effect of each of these social determinants are much more complex than each of the social determinants acting individually.

In summary, this work builds on existing literature reporting racial/ethnic and socio-economic disparities in dementia risk by identifying mediating factors between race/ethnicity and time to incident dementia. Despite the lack of disease-modifying strategies and cures, compressing all-cause dementia risk closer towards the end of life can have marked individual- and population-level benefits, particularly when preventive efforts are applied at mid-life [58]. We provide evidence for potentially modifiable risk factors that may delay onset of dementia and may explain the relationship between socio-economic status and dementia. Our findings underscore the importance of lifestyle factors such as diet, smoking and physical activity as mediators to be targeted in future intervention studies.

TABLE 1.

Baseline characteristics of selected participants by race/ethnicity among middle-aged adults (40–59 years), NHANES III, 1988–1994 (Unweighted N=4,378; Weighted N=54,629,806)^a

	Race/ethnicity
Selected participant characteristics	NHW N=1,770	Non-White N=2608	Prace
Weighted population %	77.7	22.3
Socio-demographic characteristics
Sex, % female	50.5±1.1	54.0±1.2	0.033
Age (years)	48.1±0.2	47.8±0.2	0.24
Urban/rural area of residence
Urban	45.0±5.0	64.1±5.3	<0.001
Rural	55.0±5.0	35.9±5.3
Household size	2.89±0.04	3.57±0.08	<0.001
Marital status
Never married	4.7±0.6	9.8±1.1	__
Married	76.3±1.3	59.4±1.7	<0.001
Divorced	12.4±1.0	12.9±1.1	0.001
Widowed	2.3±0.2	5.0±0.8	0.837
Other	4.3±0.6	12.9±0.8	0.15
Socio-economic status
Poverty income ratio	3.91±0.10	2.41±0.09	<0.001
Education, years	13.2±0.1	10.9±0.3	<0.001
SES z-score	+0.51±0.04	−0.15±0.05	<0.001
Dietary quality
1995-HEI total score	63.9±0.4	62.4±0.4	0.015
MAR total score	75.0±0.4	68.9±0.6	<0.001
DIET z-score	+0.17±0.02	−0.07±0.03	<0.001
Nutritional biomarkers
Folate, ng/mL	7.16±0.27	5.63±0.19	<0.001
Vitamin A, μg/dL	60.9±0.5	55.8±0.7	<0.001
Total carotenoids, μg/dL	81.8±1.3	84.5±1.7	0.26
Vitamin E, μg/dL	1,273±25	1,122±18	<0.001
NUTR z-score	+0.153±0.032	−0.062±0.023	<0.001
Physical activity 0=Less, 1=Same, 2=more
Compare activity for past mo to past yr
Less	27.7±1.3	21.6±1.6	__
Same	55.7±1.5	64.2±1.6	<0.001
More	16.6±1.2	14.3±1.5	0.55
Active compared with men/women your age
Less	21.6±1.2	22.2±1.5	__
Same	40.6±1.6	46.5±1.5	0.34
More	37.7±1.6	31.3±1.6	0.10
Active now compared with self 10 yrs ago
Less	52.0±1.7	54.7±1.6	__
Same	32.5±1.5	29.6±1.4	0.11
More	15.5±1.1	15.6±1.7	0.80
PA z-score	+0.03±0.02	+0.01±0.03	0.53
Smoking
# cigarettes/day	10.74±0.44	6.23±0.33	<0.001
Years smoked cigarettes	8.0±0.3	6.5±0.3	0.002
SMOKING z-score	+0.161±0.031	−0.100±0.025	<0.001
Alcohol consumption (g/d)	+9.65±0.84	8.18±0.42	0.12
ALCOHOL z-score	+0.022±0.037	−0.042±0.019	0.13
Social support
(1) In a typical week, how many times do you talk on the telephone with family, friends, or neighbors?	9.6±0.4	10.1±0.6	0.43
(2) How often do you get together with friends or relatives; I mean things like going out together or visiting in each other's homes? (per year)	95.5±3.8	113.7±8.4	0.010
(3) About how often do you visit with any of your other neighbors, either in their homes or in your own? (per year)	48.7±3.8	69.1±10.5	0.029
(4) How often do you attend church or religious services? (per year)	31.3±1.9	39.1±2.2	0.003
(5) Altogether, how often do you attend meetings of the clubs or organizations (per year)	15.1±1.1	10.0±1.2	0.002
SS z-score	−0.010±0.016	+0.040±0.027	0.062
Health-related factors
Self-rated health
Excellent/Very Good	56.0±2.0	32.0±2.0	__
Good	31.6±1.2	40.7±1.6	<0.001
Fair/Poor	12.5±1.2	27.4±1.7	<0.001
Co-morbidity index	0.75±0.03	0.61±0.04	0.012
Allostatic load, AL score	1.95±0.05	2.35±0.06	<0.001
BMI	27.4±0.2	28.2±0.2	0.003
HEALTH z-score	−0.167±0.024	+0.025±0.027	<0.001
All-cause dementia	4.5±0.5	7.5±1.0	0.009
Alzheimer’s Disease	2.2±0.4	3.3±0.7	0.16

Abbreviations: AD=Alzheimer’s Disease; ALCOHOL= alcohol consumption, z-score; DIET/NUTR=diet and nutritional biomarkers z-score variable (2 dietary quality measures and 4 nutritional biomarkers); HEALTH=Health-related factors as mean of z-scores for allostatic load, self-rated health, co-morbidity index and body mass index; LIFESTYLE=Lifestyle-related factors composed of social support, physical activity, diet/nutritional biomarkers, smoking and alcohol consumption using means of z-scores for related measured variables; MA=Mexican American; N=Number of participants; N’=number of observations; NHANES III=Third National Health and Nutrition and Examination Survey; NHB=Non-Hispanic Blacks; NHW=Non-Hispanic Whites; PA=Physical activity z-score variable (3 measured variables); RACE_ETHN=racial/ethnic contrast; SES=Socio-economic status mean of z-scores composed of poverty income ratio and education (years); SMOKING=smoking z-score variable (2 measured variables); SS=Social Support z-score variable (5 measured variables).

^aValues are weighted means ± SEM or percent ± SEP, considering sampling design complexity (PSU and strata), across 5 imputations with 10 iterations.

^bDesign-based F-test accounting for design complexity in terms of sampling weights, PSU and stratum, using multinomial logit models for categorical variables and linear regression for continuous variables, taking NHW as the referent category.

Research in context:

1. Systematic review: Racial disparities in dementia incidence and its underlying risk factors are well-documented[1]. Whether Alzheimer’s Disease (AD) or all-cause dementia tend to occur earlier in minority groups when compared to Non-Hispanic Whites (NHW) remains uncertain, as do mediating pathways.

2. Interpretation: Among a sample of 4,378 adults (40–59y at baseline) drawn from the third National Health and Nutrition Examination Surveys with linkage to Medicare claims data and ≤26 follow-up years we identified several characteristics–including diet, smoking, and physical activity–along the pathway between race/ethnicity, socioeconomic status, and dementia.

3. Future directions: These findings highlight the potential for modifiable risk factors to delay or prevent the onset of dementia. More studies are needed to corroborate our findings in comparable as well as expanded populations with broader capture of racial/ethnic groups.

ABBREVIATIONS

AD

Alzheimer’s Disease

AL

Allostatic Load

ALCOHOL

alcohol consumption, z-score

BDNF

Brain Derived Neurotrophic Factor

BMI

Body Mass Index

CI

Confidence Interval

CMS

Centers for Medicare and Medicaid Services

CPT4

Common Procedural Terminology

CSF

Cerebrospinal Fluid

DIET/NUTR

diet and nutritional biomarkers z-score variable (2 dietary quality measures and 4 nutritional biomarkers)

DX

Diagnosis

GSEM

Generalized Structural Equations Modeling

HCPCS

Healthcare Common Procedural Coding System

HHA

Home Health Agency

HOP

Health Options Program

HEALTH

Health-related factors as mean of z-scores for allostatic load, self-rated health, co-morbidity index and body mass index

HR

Hazard Ratio

IR

Incidence Rate

ICD-9

International Classification of Diseases, 9^th revision

ICD-10

International Classification of Disease, 10^th revision

LCL

Lower Confidence Limit

LIFESTYLE

Lifestyle-related factors composed of social support, physical activity, diet/nutritional biomarkers, smoking and alcohol consumption using means of z-scores for related measured variables

LOAD

Late-Onset Alzheimer’s Disease

MA

Mexican American

MEC

Mobile Examination Center

MRI

Magnetic Resonance Imaging

N

number of participants

N’

number of observations

NCHS

National Center for Health Statistics

NHANES III

Third National Health and Nutrition and Examination Survey

NHB

Non-Hispanic Blacks

NHW

Non-Hispanic Whites

OTHER

Other Race/ethnicity

PA

Physical activity z-score variable (3 measured variables)

RACE_ETHN

racial/ethnic contrast

SES

Socio-economic status mean of z-scores composed of poverty income ratio and education (years)

SMOKING

smoking z-score variable (2 measured variables)

SNF

Skilled Nursing Facility.

SS

Social Support z-score variable (5 measured variables).

UCL

Upper Confidence Limi

VEGF

Vascular endothelial growth factor

WMH

White Matter Hyperintensity