Geographic Patterns of Dementia in the United States: Variation by Place of Residence, Place of Birth, and Subpopulation

Abstract

Objectives

The prevalence of dementia varies geographically in the United States. However, the extent to which this variation reflects contemporary place-based experiences versus embodied exposures from earlier in the life course remains unclear, and little is known regarding the intersection of place and subpopulation. This study, therefore, evaluates whether and how risk for assessed dementia varies by place of residence and birth, overall and by race/ethnicity and education.

Methods

We pool data from the 2000 to 2016 waves of the Health and Retirement Study, a nationally representative panel survey of older U.S. adults (n = 96,848 observations). We estimate the standardized prevalence of dementia by Census division of residence and birth. We then fit logistic regression models of dementia on region of residence and birth, adjusting for sociodemographic characteristics, and examine interactions between region and subpopulation.

Results

The standardized prevalence of dementia ranges from 7.1% to 13.6% by division of residence and from 6.6% to 14.7% by division of birth, with rates highest throughout the South and lowest in the Northeast and Midwest. In models accounting for region of residence, region of birth, and sociodemographic covariates, Southern birth remains significantly associated with dementia. Adverse relationships between Southern residence or birth and dementia are generally largest for Black and less-educated older adults. As a result, sociodemographic disparities in predicted probabilities of dementia are largest for those residing or born in the South.

Discussion

The sociospatial patterning of dementia suggests its development is a lifelong process involving cumulated and heterogeneous lived experiences embedded in place.

Cognition, Education, Health disparities, Life course, Race/ethnicity

Dementia is a devastating syndrome characterized by progressive loss of cognitive function and memory, leading to disability, dependence, and death (Alzheimer’s Association, 2022). In the United States, an estimated 6.1 million older adults suffered from dementia in 2020, approximately 11% of those 65 and older and 35% of those 85+ (Rajan et al., 2021). Due to population aging, the number of U.S. adults with dementia is expected to rise in the coming years, possibly doubling to 12.7 million by 2050 (Rajan et al., 2021). Beyond those directly affected, the health consequences of dementia extend to the 11 million people in the United States who provide unpaid care, as dementia caregiving has been linked to poor mental and physical health (Alzheimer’s Association, 2022).

Although dementia and its broader impacts are widespread, not all are equally affected. In particular, this study is motivated by evidence that dementia prevalence varies across the United States, resulting in significant geographic disparities. The prevalence of diagnosed dementia among older U.S. adults ranged from 6.0% to 9.6% across states in 2008, with rates lowest in the Midwest and West and highest in the South (Koller & Bynum, 2015). Studies examining other years and data likewise find substantial geographic variation in the burden of dementia, with above-average rates among residents of the South (Ailshire et al., 2022; Akushevich et al., 2021; Ho & Franco, 2022; Kirson et al., 2020; Taylor et al., 2017).

The reasons for the geographic patterning of dementia in the United States are not fully understood. Life course perspectives posit that health reflects biological endowments in combination with a lifetime of cumulated experiences and circumstances that have unfolded in specific physical, structural, and social contexts, which are embedded in time and place (Ben-Shlomo & Kuh, 2002; Jones et al., 2019; Krieger, 2021; Short & Mollborn, 2015). If contextual factors linked to place influence the development of dementia, it follows that risk for dementia should vary geographically. Such variation might be due to contemporaneous place-based exposures and experiences or to previous place-based exposures and experiences, including those from early life, that in turn shape subsequent exposures and outcomes through biological or socially mediated pathways. Prior research has established relationships between current place-based characteristics, such as policy contexts, labor market conditions, structural racism, environmental hazards, and adult health and mortality (Case & Deaton, 2020; Diez Roux, 2001; Garcia et al., 2021; Montez et al., 2020). Studies have also established the importance of early-life residence and related contextual factors for later health and mortality, even when accounting for place of residence in older age (Case & Paxson, 2009; Xu et al., 2020, 2021). However, whether and how residence in early and later life together structure geographic patterns of dementia remains unclear.

Prior research has examined place of birth as a risk factor for dementia, with mixed results. One study finds that dementia mortality rates are higher for those born in a set of Southern states in the “Stroke Belt,” regardless of where they lived at death (Glymour et al., 2011). Studies likewise show that birthplace, and Southern birth in particular, is independently associated with diagnosed dementia and reduced cognitive functioning in older adults (George et al., 2021; Gilsanz et al., 2017; Lamar et al., 2020). Other studies, however, suggest that geographic variation in dementia mortality by birthplace is largely explained by place of residence in older age (Topping et al., 2021a, 2021b).

Interpreting these findings is complicated by the reliance of prior work on nonrepresentative data and measures of dementia diagnoses and mortality rather than assessed dementia. Many studies of dementia-related outcomes draw data from residents of a handful of states (Topping et al., 2021a, 2021b) or a single locale (George et al., 2021; Gilsanz et al., 2017; Lamar et al., 2020), shedding important light on the role of the birthplace, but possibly obscuring the independent or joint effects of residence in older age. Relatedly, few studies explore regional variation in dementia-related outcomes comprehensively. Many take interest primarily in the effect of Southern residence or birth (George et al., 2021; Gilsanz et al., 2017; Glymour et al., 2011; Topping et al., 2021b), leaving variation across other regions comparatively understudied. In addition, studies of dementia diagnoses and mortality likely underestimate its prevalence, as dementia is frequently undiagnosed and unrecorded on death certificates, particularly among Black, Hispanic, and less-educated older adults (Chen et al., 2019; Lang et al., 2017; Lin et al., 2020; Stokes et al., 2020). The extent of this underestimation may vary geographically due to differences in health care access, utilization, practice, and other factors (Bradford et al., 2009; Ganguli et al., 2017; Low et al., 2019). Geographic variation in dementia diagnoses and mortality, and related disparities, may, therefore, reflect variation in health care or administrative practice rather than the true spatial patterning of disease.

In addition, although geographic variation in dementia may reflect the influences of current and former place-based exposures, such exposures and their implications may vary across sociodemographic subpopulations. Research on place-based or “neighborhood” effects on health demonstrate that lived experiences embedded in place are heterogeneous (Sharkey & Faber, 2014). Risk for dementia is considerably higher for Black and Hispanic than White U.S. adults, and for those with lower levels of education, overall (Crimmins et al., 2018; Farina et al., 2020; Mayeda et al., 2016). However, the magnitude of these sociodemographic disparities may vary geographically if the consequences of place-based factors for dementia differ across subpopulations. Studies show, for example, that racist structures that proliferated in the South through the mid-twentieth century (e.g., school segregation) contribute to racial disparities in cognitive outcomes among current cohorts of older adults (Liu et al., 2015; Walsemann et al., 2022). Prior research also demonstrates that morbidity and mortality rates vary geographically primarily for less-educated Americans, with poorer outcomes in the South (Montez et al., 2017, 2019).

The current study addresses existing gaps in the literature by exploring geographic variation in risk for assessed dementia by place of residence, place of birth, and subpopulation in a nationally representative sample of older U.S. adults in the years 2000–16. We begin by estimating the standardized prevalence of dementia by place of residence and birth. Then, we examine associations between place of residence or birth and dementia in logistic regression models, adjusting for sociodemographic characteristics. Finally, we consider the intersection of place and subpopulation, namely self-identified race/ethnicity and education, by fitting logistic regression models with interaction terms. Sensitivity analyses explore the robustness of results to variation over time and to regional mobility between birth and older age.

Method

Data and Sample

We use data from the Health and Retirement Study (HRS), a biennial panel survey of older U.S. adults that is funded by the National Institute on Aging (grant U01AG009740) and conducted by the University of Michigan (Health and Retirement Study, 2021; RAND, 2021). The HRS began in 1992 with a sample of people born in the years 1931–41 and their spouses, and additional birth cohorts have since been incorporated. Since 1998, the HRS has been nationally representative of U.S. adults over age 50, surveying approximately 20,000 people each wave.

The HRS is an exemplary dataset for research on dementia and place. Respondents are followed up over time even if they move, and at each wave, dementia is assessed and current place of residence is recorded. Although only community-dwelling adults are eligible for initial sampling, follow-up is attempted for those who transition into nursing homes. Relatedly, those who are unable to participate due to a health limitation can elect a proxy (e.g., spouse or relative) to respond on their behalf.

We pool data from the years 2000–16 and restrict to respondents ages 65 and older, yielding a potential sample of 99,429 person–period observations from 21,443 respondents. Given our U.S. focus, we exclude observations from respondents living outside the United States or whose division of residence is unknown (n = 2,144 observations). We also exclude those with no recorded birthplace (n = 406 observations) or missing race/ethnicity or education (n = 31 observations). The final analytic sample includes 96,848 observations from 21,257 respondents.

Measures

Cognitive state and dementia

Our dependent variable is a binary measure of assessed dementia. Dementia is one of three cognitive state classifications developed by Langa and colleagues for use with HRS survey assessments, along with cognitive impairment, no dementia (CIND), and no impairment (Langa et al., 2020). These classifications have been validated against expert panel diagnoses, and their accuracy is similar to alternative measures (Crimmins et al., 2011; Gianattasio et al., 2019). For self-respondents, classifications are based on tests of immediate and delayed word recall, subtraction, and backward counting. For proxy respondents, classifications are determined by the proxy’s assessment of the respondent’s memory and activity limitations, and the interviewer’s assessment of the respondent’s cognitive functioning. Score cut-points were designed to ensure that the prevalence of each classification was similar to the Aging, Demographics, and Memory Study, for which a subsample of HRS respondents underwent more extensive assessments. We compare those with dementia to those with either CIND or no impairment.

Place of residence and birth

We create place of residence and birth measures using respondents’ current states of residence, which are recorded at each survey wave, and states of birth, which are collected at study entry. We group states into nine Census divisions (U.S. Census Bureau, 2021): New England, Middle Atlantic, East North Central, West North Central, South Atlantic, East South Central, West South Central, Mountain, and Pacific. To enhance statistical power, for some analyses we collapse divisions into regions: Northeast (New England and Middle Atlantic), Midwest (East North Central and West North Central), South (East South Central, West South Central, and South Atlantic), and West (Mountain and Pacific). Division and region of birth include an additional category for non-U.S. birth.

Other variables

We use measures of survey year and age (top-coded at 100), both expressed as continuous variables, a binary measure of female sex/gender (reference category: male), and categorical measures of racial/ethnic identification (non-Hispanic White [reference]; non-Hispanic Black; Hispanic; Other) and education (less than high school [reference]; high school or GED; Associate’s or some college; Bachelor’s or higher).

Analysis

The analysis proceeds in five parts. First, we calculate descriptive statistics across observations (all person–periods) and respondents (final person–period only). Second, we estimate dementia prevalence across divisions of residence and birth, standardizing estimates to the 2010 U.S. population by age and sex (Howden & Meyer, 2011).

Third, we estimate adjusted associations between place of residence, place of birth, and dementia using logistic regression. Models 1 and 2 regress dementia on region of residence. Model 1 adjusts for survey year, age, sex/gender, and the interaction between age and sex/gender. Model 2 also includes race/ethnicity and education as covariates. Models 3 and 4 parallel Models 1 and 2, regressing dementia on region of birth and covariates. Finally, Model 5 regresses dementia on both region of residence and region of birth, as well as all covariates, to evaluate their independent effects. Although we designate “South” as the reference region in these models, for Model 5, we assess all pairwise comparisons to explore differences among other regions as well. We consider two-sided p values below .05 to be statistically significant.

Fourth, we examine whether and how associations between dementia and region of residence or birth differ across subpopulations by adding interaction terms to Model 5. To enhance statistical power, in these models we collapse region of residence into two categories (non-South vs. South) and region of birth into three (non-South, South, or non-U.S.), specifying “non-South” as the reference region. In separate models, we examine five sets of interaction terms: region of residence by region of birth; region of residence by race/ethnicity; region of birth by race/ethnicity; region of residence by education; and region of birth by education. We consider sets of interaction terms to be statistically significant when adjusted Wald tests of joint significance return p values below .05. To facilitate interpretation, following each of the five models, we estimate the average marginal effects of region of residence or birth by subpopulation and plot corresponding predicted probabilities of dementia. We then estimate a model incorporating all five sets of interaction terms simultaneously to determine which are independently associated with dementia.

Finally, we conduct sensitivity analyses exploring robustness to variation over time and to regional mobility. Dementia prevalence has declined in recent decades, and somewhat unevenly across regions (Ailshire et al., 2022). To confirm that pooling data from 2000 to 2016 yields geographic patterns similar to that of any given year, we estimate the standardized prevalence of dementia across divisions in individual survey years and re-estimate Model 5 using data from 2016 only. Studies also suggest that those who move interstate over the life course are select in complex ways, which may explain geographic variation in adult health (Xu et al., 2020). We explore this possibility by re-estimating Model 5, and models with interaction terms, including as a covariate a binary measure of regional mobility that compares those residing in a different region than that in which they were born to those currently residing in their birth region.

All analyses apply probability weights (wave-specific respondent-level weights) and estimate standard errors that account for complex sampling and clustering with HRS-provided primary sampling unit and strata variables. Analyses were performed in Stata 17.0 (StataCorp, 2021).

Results

Descriptive Statistics

Table 1 shows the weighted descriptive statistics across observations and respondents. Here, we describe the analytic sample, which includes all person–period observations. The weighted prevalence of dementia is 9.6%. The percentage of observations from residents of each Census division ranges from 5.1% (New England) to 21.3% (South Atlantic), and the percentage by division of birth ranges from 3.5% (Mountain) to 18.2% (East North Central). The average age is 75.0 years (standard deviation = 7.6), and 57.3% are female. Four in five (82.7%) observations are from respondents who identify as non-Hispanic White, 8.5% non-Hispanic Black, 6.7% Hispanic, and 2.2% “Other” race/ethnicity. By education, observations are distributed as follows: less than high school: 22.1%; high school or GED: 35.9%; Associate’s or some college: 20.7%; Bachelor’s or higher: 21.3%.

Table 1.

Weighted Descriptive Statistics

	Observations (n = 96,848)	Respondents (n = 21,257)
	Mean (SD) or %	Mean (SD) or %
Cognitive state
No impairment or CIND	90.42	83.08
Dementia	9.58	16.92
Division of residence
New England	5.08	4.95
Middle Atlantic	12.57	12.40
East North Central	16.18	16.18
West North Central	9.21	8.56
South Atlantic	21.28	22.25
East South Central	5.91	6.08
West South Central	9.87	10.18
Mountain	5.95	6.06
Pacific	13.95	13.33
Division of birth
New England	5.54	5.46
Middle Atlantic	15.75	15.85
East North Central	18.17	18.43
West North Central	11.68	10.87
South Atlantic	13.10	13.64
East South Central	7.76	7.83
West South Central	9.72	9.53
Mountain	3.47	3.36
Pacific	5.91	6.15
Non-U.S.	8.91	8.88
Survey year	2008.62 (5.19)	2012.35 (4.86)
2000	9.44	5.04
2002	9.87	4.58
2004	9.90	4.50
2006	10.27	4.77
2008	10.71	5.89
2010	11.36	4.90
2012	12.04	6.15
2014	12.74	8.90
2016	13.67	55.28
Age	74.97 (7.59)	76.72 (8.04)
Sex/gender
Male	42.75	44.60
Female	57.25	55.40
Race/ethnicity
White, non-Hispanic	82.66	81.64
Black, non-Hispanic	8.52	9.00
Hispanic	6.65	6.98
Other	2.16	2.38
Education
Less than high school	22.10	21.78
High school or GED	35.94	34.23
Associate’s or some college	20.66	21.48
Bachelor’s or higher	21.30	22.52

Notes: Observation-level descriptive statistics reflect all person–periods. Respondent-level descriptive statistics reflect the final person–period from each respondent. Data are from Health and Retirement Study respondents ages 65 and older in years 2000–16. CIND = cognitive impairment, no dementia; SD = standard deviation.

Supplementary Table 1 shows additional descriptive statistics demonstrating how place, race/ethnicity, and education are intertwined. Across observations, a much larger percentage of Black compared to White older adults reside in the South (Black: 57.0%; White: 34.4%) or were born in the South (Black: 80.2%; White: 26.3%). Hispanic older adults and those identifying “Other” race/ethnicity are more likely than others to live in the West or to have been born outside the United States. Concerning education, those who did not complete high school disproportionately reside in the South and were born in the South or outside the United States. Supplementary Table 1 also shows that regional mobility is common. One in three (34.0%) observations is from an older adult residing in a different region than that in which they were born. Regional mobility is especially common among those who identify as Black, Hispanic, or “Other” race, and among more educated older adults.

Standardized Dementia Prevalence Across Divisions

Figure 1 maps the standardized prevalence of dementia over the 2000–16 period by division of residence and birth (see also, Supplementary Table 2). Geographic variation in dementia prevalence is striking. By division of residence, estimates range from 7.1% in West North Central to 13.6% in East South Central. The highest divisional prevalence is, therefore, 91% higher than (nearly double) that of the lowest. The second highest prevalence is in West South Central (13.3%), followed by the South Atlantic (10.7%). The prevalence of dementia is, therefore, highest in the three Census divisions that comprise the U.S. South.

Figure 1.

Standardized prevalence of dementia by division of residence and division of birth. Notes: Data are from HRS respondents ages 65 and older in years 2000–16. HRS-constructed weighting and clustering variables are applied. Estimates are age- and sex-standardized to the 2010 U.S. population. Both maps use identical scales, with scale labels corresponding to the minimum and maximum plotted estimates. ENC = East North Central; ESC = East South Central; HRS = Health and Retirement Study; MA = Middle Atlantic; MTN = Mountain; NE = New England; PAC = Pacific; SA = South Atlantic; WNC = West North Central; WSC = West South Central. See Supplementary Table 2 for numeric results.

Dementia prevalence by division of birth follows similar patterns. Prevalence is below 7% for those born in any Northeastern or Midwestern division, with a low of 6.6% in the Middle Atlantic. Meanwhile, prevalence exceeds 13% in all three Southern divisions and for those born outside the United States, with a maximum of 14.7% in East South Central. The maximum divisional prevalence of dementia by place of birth is, therefore, 124% greater than (more than double) the minimum.

Adjusted Associations Between Dementia and Region of Residence and Birth

Table 2 presents results from logistic regression models of dementia on region of residence and birth. Models 1 and 2 show statistically significant associations between region of residence and odds of dementia both before and after including race/ethnicity and education as covariates. Models 3 and 4 show similar patterns for region of birth. Across Models 1–4, Southern residence or birth is associated with elevated odds of dementia compared to virtually any other region.

Table 2.

Odds Ratios and 95% Confidence Intervals from Logistic Regression Models of Dementia on Region of Residence and Region of Birth

	Region of residence		Region of birth		Region of residence + Region of birth
	Model 1	Model 2	Model 3	Model 4	Model 5
	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
Region of residence	(p < .001)	(p < .001)			(p = .344)
Northeast	0.68 (0.59, 0.79) ***	0.78 (0.68, 0.89) ***	—	—	1.07 (0.91, 1.25)
Midwest	0.63 (0.56, 0.72) ***	0.78 (0.71, 0.87) ***	—	—	0.99 (0.87, 1.12)
South	Ref.	Ref.	—	—	Ref.
West	0.64 (0.54, 0.76) ***	0.81 (0.73, 0.90) ***	—	—	0.91 (0.80, 1.04)
Region of birth			(p < .001)	(p < .001)	(p < .001)
Northeast	—	—	0.39 (0.35, 0.44) ***	0.59 (0.52, 0.68) ***	0.57 (0.49, 0.65) ***
Midwest	—	—	0.41 (0.36, 0.47) ***	0.62 (0.55, 0.70) ***	0.63 (0.54, 0.74) ***
South	—	—	Ref.	Ref.	Ref.
West	—	—	0.52 (0.41, 0.66) ***	0.76 (0.64, 0.90) **	0.81 (0.65, 1.01)
Non-U.S.	—	—	0.94 (0.81, 1.09)	0.77 (0.66, 0.91) **	0.77 (0.65, 0.92) **
Survey year	0.98 (0.97, 0.98) ***	0.99 (0.98, 1.00) **	0.98 (0.97, 0.98) ***	0.99 (0.98, 1.00) **	0.99 (0.98, 1.00) **
Age	1.11 (1.10, 1.12) ***	1.11 (1.10, 1.12) ***	1.11 (1.11, 1.12) ***	1.11 (1.10, 1.12) ***	1.11 (1.10, 1.12) ***
Female	1.01 (0.92, 1.10)	0.95 (0.86, 1.04)	0.98 (0.89, 1.07)	0.94 (0.86, 1.03)	0.94 (0.85, 1.03)
Age × female	1.02 (1.01, 1.03) ***	1.02 (1.01, 1.03) ***	1.02 (1.01, 1.03) ***	1.02 (1.01, 1.03) ***	1.02 (1.01, 1.03) ***
Race/ethnicity		(p < .001)		(p < .001)	(p < .001)
White, NH	—	Ref.	—	Ref.	Ref.
Black, NH	—	2.63 (2.36, 2.92) ***	—	2.17 (1.93, 2.45) ***	2.16 (1.91, 2.44) ***
Hispanic	—	1.80 (1.54, 2.11) ***	—	1.72 (1.47, 2.01) ***	1.74 (1.48, 2.05) ***
Other	—	1.76 (1.22, 2.55) **	—	1.59 (1.09, 2.31) *	1.60 (1.11, 2.32) *
Education		(p < .001)		(p < .001)	(p < .001)
<HS	—	Ref.	—	Ref.	Ref.
HS/GED	—	0.40 (0.37, 0.43) ***	—	0.41 (0.38, 0.44) ***	0.41 (0.38, 0.44) ***
AA/SC	—	0.29 (0.25, 0.32) ***	—	0.30 (0.26, 0.34) ***	0.30 (0.27, 0.34) ***
BA+	—	0.20 (0.17, 0.23) ***	—	0.21 (0.18, 0.24) ***	0.21 (0.18, 0.24) ***
Constant	0.12 (0.11, 0.14) ***	0.18 (0.16, 0.20) ***	0.15 (0.13, 0.17) ***	0.20 (0.18, 0.23) ***	0.21 (0.18, 0.23) ***
Model fit	F(7,51) = 492.58 ***	F(13,45) = 372.58 ***	F(8,50) = 437.83 ***	F(14,44) = 449.08 ***	F(17,41) = 370.85 ***

Notes: N = 96,848 observations (person–periods). Data are from Health and Retirement Study respondents ages 65 and older in years 2000–16. HRS-constructed weighting and clustering variables are applied. Survey year is centered at 2000. Age is centered at 75. AA/SC = Associate’s or some college; BA+ = Bachelor’s or higher; CI = confidence interval; HRS = Health and Retirement Study; <HS = less than high school; HS/GED = high school or GED; NH = non-Hispanic; OR = odds ratio. Italicized p values are from adjusted Wald tests of joint significance.

Two-sided significance tests: ***p < .001. **p < .01. *p < .05.

Model 5 shows that, when accounting for birthplace, the association between region of residence and dementia is nonsignificant (overall p = .344). Region of birth, however, remains significantly associated with dementia (overall p < .001). Compared to Southern birth, birth in the Northeast (odds ratio [OR] = 0.57, p < .001), Midwest (OR = 0.63, p < .001), outside the United States (OR = 0.77, p = .004), or, marginally, the West (OR = 0.81, p = .059) is associated with lower odds of dementia. Several other pairwise comparisons are also significant. Relative to birth in the West, the odds of dementia are lower for those born in the Northeast (OR = 0.70, p = .004) or Midwest (OR = 0.78, p = .015). Similarly, relative to birth outside the United States, odds are lower for those born in the Northeast (OR = 0.74, p = .001) or, marginally, the Midwest (OR = 0.82, p = .075). Adjusting for sociodemographic characteristics and place of residence in older adulthood, the odds of dementia are, therefore, highest for those born in the South, followed by the West or outside the United States, and lowest for those born in the Northeast or Midwest. Model 5 also shows that the odds of dementia are lowest for non-Hispanic White older adults and those with higher education.

Geographic Variation in Dementia Across Subpopulations

Supplementary Table 3 presents results from logistic regression models that incorporate interactions to explore whether and how relationships between region of residence (non-South [reference] or South) or birth (non-South [reference], South, or non-U.S.) and dementia differ by birthplace, race/ethnicity, and education. All five sets of interaction terms are statistically significant overall (p < .05). Average marginal effects and corresponding predicted probabilities likewise reveal substantial heterogeneity in relationships between Southern residence or birth and risk for dementia.

Results from the first model in Supplementary Table 3 show that Southern residence amplifies the adverse association between Southern birth and odds of dementia (Southern birth OR = 1.31, p = .001; Southern residence × Southern birth OR = 1.42, p = .002). Figure 2 clarifies that the average marginal effect of Southern residence on the predicted probability of dementia is nonsignificant for those born outside the South or outside the United States (Supplementary Table 4). However, for those born in the South, Southern residence is associated with a significantly higher probability of dementia (p = .003; Southern residence: 12.1%; non-Southern residence: 10.4%).

Figure 2.

Predicted probability of dementia by region of residence and region of birth. Notes: Figure presents predicted probabilities of dementia (i.e., average predictive margins, multiplied by 100) from a logistic regression model of dementia on region of residence, region of birth, and their interaction, adjusting for survey year, age, sex/gender, the interaction between age and sex/gender, race/ethnicity, and education. Data are from HRS respondents ages 65 and older in years 2000–16. HRS-constructed weighting and clustering variables are applied. See Supplementary Table 4 for numeric results. HRS = Health and Retirement Study. *The average marginal effect of region of residence (reference: non-South) is statistically significant (p < .05).

The second model shows that the relationship between Southern residence and the odds of dementia varies by racial/ethnic identification. Southern residence is not significantly associated with dementia for the reference group of older adults who identify as non-Hispanic White (Southern residence OR = 0.96, p = .484). However, relationships between Southern residence and odds of dementia are significantly larger in magnitude for those who identify as non-Hispanic Black (Southern residence × Black OR = 1.49, p < .001) or Hispanic (Southern residence × Hispanic OR = 0.69, p = .003) compared to those who identify as White, in opposing directions. Average marginal effects and corresponding predicted probabilities in Figure 3 confirm that, for White older adults, Southern residence is not significantly associated with the probability of dementia (Southern residence: 8.2%; non-Southern residence: 8.5%; see also, Supplementary Table 5). However, among Black older adults, the average marginal effect of Southern residence is significant (p < .001), such that the predicted probability of dementia is higher for residents of the South (16.5%) than for those living elsewhere (12.8%). The racial disparity in Black and White older adults’ predicted probabilities of dementia is, therefore, larger among residents of the South (8.2 percentage points [pp]) than for those living outside the South (4.3pp). The average marginal effect of Southern residence is also significant and distinct for Hispanic older adults (p < .001), such that the predicted probability of dementia is lower for Southern residents (11.1%) than for those living elsewhere (14.9%).

Figure 3.

Predicted probability of dementia by region, race/ethnicity, and education. Notes: Figure presents predicted probabilities of dementia (i.e., average predictive margins, multiplied by 100) from logistic regression models of dementia on interactions between region of residence and race/ethnicity; region of birth and race/ethnicity; region of residence and education; and region of birth and education. Models adjust for survey year, age, sex/gender, the interaction between age and sex/gender, race/ethnicity, education, region of residence, and region of birth. Data are from HRS respondents ages 65 and older in years 2000–16. HRS-constructed weighting and clustering variables are applied. The figure omits predicted probabilities for those who self-identify as “Other” race/ethnicity due to small sample size. AA/SC = Associate’s or some college; BA+ = Bachelor’s or higher; HRS = Health and Retirement Study; <HS = less than high school; HS/GED = high school or GED; NH = non-Hispanic. See Supplementary Tables 5 and 6 for numeric results. *The average marginal effect of region of residence or birth (reference: non-South) is statistically significant (p < .05).

Interactions between region of birth and race/ethnicity reveal less straightforward patterns. Model results show that the adverse relationship between Southern birth and odds of dementia, relative to birth outside the South, is similar in magnitude for the reference group of non-Hispanic White older adults (Southern birth OR = 1.58, p < .001) and Black older adults (Southern birth × Black OR = 1.00, p = .993). The average marginal effect of Southern birth on the predicted probability of dementia is also significant for both White (p < .001) and Black older adults (p < .001). However, it appears to be somewhat larger for those who identify as Black (Southern birth: 18.2%; non-Southern birth: 13.3%; difference = 5.0pp) than White (Southern birth: 10.4%; non-Southern birth: 7.2%; difference = 3.1pp; Figure 3, Supplementary Table 5). These seemingly contradictory patterns may reflect the fact that dementia prevalence is higher among Black older adults overall, as ORs may mask disparate marginal effects on probabilities when “base rates” differ. Other model results and predicted probabilities show that, among White older adults, risk for dementia is also significantly higher for those born outside the United States than outside the South, whereas for Hispanic older adults, risk for dementia does not differ significantly by region of birth.

Interactions with education and corresponding predicted probabilities show that adverse relationships between both Southern residence and Southern birth and dementia are largest for the least educated older adults (Figure 3, Supplementary Table 6). Among those who did not complete high school, the average marginal effect of Southern residence is significant (p = .039), such that the predicted probability of dementia is higher for residents of the South (17.4%) than for those living elsewhere (15.7%). The average marginal effect of Southern residence on dementia is nonsignificant for those who completed high school or a GED and for those with an Associate’s or some college education. Meanwhile, for those with a Bachelor’s or higher, Southern residence is associated with a significantly lower probability of dementia (p = .046; Southern residence: 4.0%; non-Southern residence: 5.0%). The educational disparity in the predicted probability of dementia (i.e., the difference between the least and most educated groups) is, therefore, larger in the South (13.4pp) than elsewhere (10.7pp). Patterns are even more dramatic by place of birth. Although the average marginal effect of Southern birth on the predicted probability of dementia relative to birth outside the South is significant for all education groups (all p < .001), it is largest for the least educated. Among those who did not complete high school, the predicted probability of dementia is 7.8pp higher for those born in the South (21.5%) than outside the South (13.7%), whereas, among college graduates, the predicted probability is just 1.3pp higher for the Southern born (5.3% vs 4.0%). The educational disparity in the predicted probability of dementia is, therefore, larger for those born in the South (16.2pp) than outside the South (9.7pp). Patterns for those born outside the United States mirror those born outside the South.

The model incorporating all five sets of interaction terms simultaneously shows that the interaction between Southern residence and race/ethnicity is independently significant (overall p = .003; Supplementary Table 7). Southern residence in older adulthood confers risk for dementia on Black older adults only, and it is associated with reduced risk for dementia among Hispanic older adults.

Sensitivity Analyses

We conduct two sets of sensitivity analyses. First, we assess the robustness of our findings to variation over time. Supplementary Tables 8 and 9 present the standardized prevalence of dementia across divisions of residence and birth, respectively, in each survey year from 2000 to 2016. Although the prevalence of dementia declined over this period, geographic patterns changed little. In all years, the prevalence of dementia ranges considerably across divisions and is notably higher in the South than elsewhere in the United States. Furthermore, as shown in Supplementary Table 10, adjusted geographic patterns in 2016 are similar to those for 2000–16. Region of birth is significantly associated with dementia even when accounting for region of residence in older age, and the odds of dementia are significantly higher for those born in the South than in any other region, with the exception of the West.

Second, we examine the extent to which our results are explained by regional mobility between birth and older age. Results in Supplementary Table 11 show that residing outside one’s region of birth is associated with reduced odds of dementia (OR = 0.89, p = .019). However, adjusting for mobility does not alter conclusions about regional variation, as birthplace remains significantly associated with dementia (overall p < .001). Similarly, as shown in Supplementary Table 12, four of the five sets of interaction terms we estimate remain statistically significant when adjusting for regional mobility. Only the interaction between Southern residence and region of birth is no longer significant (overall p = .096), likely because the interaction itself reflects mobility.

Discussion

This study investigates geographic patterns of assessed dementia by place of residence, place of birth, and subpopulation in a nationally representative sample of older U.S. adults. Our analyses return three main findings. First, the standardized prevalence of dementia, pooled across the years 2000–16, is highest in the U.S. South, both by place of residence in older adulthood and by place of birth, and lowest in the Northeast and Midwest. Estimates range from 7.1% among residents of the West North Central division to 13.6% in East South Central, a relative difference of 91%. By place of birth, the relative difference is even larger (124%), ranging from 6.6% for those born in the Middle Atlantic to 14.7% in East South Central.

Second, we find that the relationship between place of birth and dementia remains significant when accounting for place of residence in older age as well as sociodemographic characteristics. Southern birth is associated with elevated odds of dementia relative to birth in any other region or outside the United States. Odds of dementia are lowest for those born in the Northeast or Midwest, and fall in the middle for those born in the West or outside the United States. Meanwhile, place of residence in older adulthood is not significantly associated with dementia when accounting for birthplace, suggesting that early-life contexts may matter more than contemporaneous place-based factors for the development of dementia. These results are consistent with prior research documenting adverse relationships between Southern birth and diagnosed dementia, dementia mortality, and cognitive functioning in older age (George et al., 2021; Gilsanz et al., 2017; Glymour et al., 2011; Lamar et al., 2020). Our results differ, however, from studies finding that place of residence in older adulthood is a stronger independent predictor of dementia mortality than place of birth (Topping et al., 2021a, 2021b). Sensitivity analyses suggest that the relationship between birthplace and dementia is unlikely due to regional mobility patterns. It may instead reflect the enduring influence of early-life place-based exposures and experiences.

Third and finally, exploring the intersection of place and subpopulation, we find that links between both Southern residence and Southern birth and dementia vary. Although Southern residence is not independently associated with dementia overall, it does appear to confer a significantly higher risk for dementia on specific subgroups, including those born in the South, Black older adults, and those with low education, whereas, among Hispanic older adults, Southern residence is associated with reduced dementia risk. We also find that Southern birth is associated with a significantly elevated risk for dementia among both White and Black older adults while it is unrelated to dementia among those who identify as Hispanic, and that its adverse association with dementia is strongest for those with the least schooling.

These results build on a large body of literature documenting elevated rates of dementia among Black and Hispanic adults and those with low education overall (Crimmins et al., 2018; Farina et al., 2020; Mayeda et al., 2016) by underscoring that disparities in dementia are not uniform across the country. Results demonstrate that risk for dementia varies more widely across place for adults who identify as Black or Hispanic than for White adults. Similarly, risk for dementia varies more across place for less-educated older adults, consistent with prior work on morbidity and mortality across U.S. states (Montez et al., 2017, 2019). Like earlier studies, our results suggest that contextual factors—especially those that characterize the U.S. South (Lamar et al., 2020; Peterson et al., 2021; Walsemann et al., 2022)—contribute significantly to geographic as well as sociodemographic health disparities.

These findings are troubling given the large proportion of older Black Americans who were born or reside in the South. In our weighted sample, approximately 78.5% of non-Hispanic Black older adults were born in the South and 58.7% reside there in older adulthood. Among non-Hispanic White respondents, Southern birth (26.7%) and residence (35.7%) are less common. This suggests that a larger fraction of Black than White older Americans will be impacted by any adverse effect of Southern birth on dementia, despite many moving out of the South over the life course. Black Americans’ risk for dementia is also associated with Southern residence in older adulthood, a pattern not observed among White older adults.

The strengths of this study include its use of nationally representative data on assessed dementia. Although prior studies shed valuable light on the geographic patterning of dementia, many use non-national data, limiting the ability to parse the independent or joint effects of place in early versus later life (George et al., 2021; Gilsanz et al., 2017; Lamar et al., 2020; Topping et al., 2021a, 2021b). Many also examine dementia diagnoses or mortality, which are not only underestimated, but also bound up with factors related to health care access, utilization, and practice that may vary spatially and across sociodemographic groups (Bradford et al., 2009; Chen et al., 2019; Ganguli et al., 2017; Lang et al., 2017; Lin et al., 2020; Low et al., 2019; Stokes et al., 2020).

This study has several limitations. Our measure of assessed dementia is based on a small set of cognitive tests and survey questions, rather than a full clinical evaluation (Langa et al., 2020). Resulting dementia classifications are imperfect, although their concordance with diagnoses by a panel of experts is high and similar to alternative measures (Crimmins et al., 2011; Gianattasio et al., 2019). Also, the geographic scales used in our analyses comprise multiple states with distinct laws, health care institutions, economies, and more. Although there is likely substantial variation in dementia prevalence within these areas, HRS sample sizes are insufficient to generate state- or lower-level estimates with precision. Finally, although we recognize that aging is a dynamic process that occurs over the life course, we do not explore individual trajectories of place and cognitive aging, a process related to, but distinct from, dementia. Future research could model how trajectories of cognitive aging align with detailed residential histories.

Despite these limitations, the results of the current study underscore the importance of situating people in places. Individuals age in place, and the characteristics of these places—physical, structural, and social—can promote or diminish health, including risk for dementia. Furthermore, our results suggest that places leave lasting imprints. Contextual factors rooted in place, such as policy, health care, and educational opportunities, affect individuals beginning in early life, and these effects may be sustained or even compounded over time to influence risk for dementia. This finding highlights the need for data resources, including residential histories, that extend back to early life and link to detailed contextual information, to support future research on the place-based mechanisms underlying dementia. Finally, our results indicate that sociodemographic disparities in dementia differ across U.S. regions. Research into how and why place matters differently for dementia across subpopulations is needed. In particular, we urge future researchers to investigate the contextual factors that distinguish the South from other U.S. places that are relevant to health and health disparities. Understanding the social and environmental drivers of dementia may aid in predicting, preventing, and managing its devastating consequences.

Acknowledgments

We thank those who attended an early presentation of this work at the 2022 Annual Meetings of the Population Association of America for their constructive feedback.

Funding

This work was supported by the Population Studies and Training Center at Brown University, which receives funding from the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health (grant numbers P2CHD041020, T32HD007338); the Interdisciplinary Network on Rural Population Health and Aging, which receives funding from the National Institutes of Health (grant number R24AG065159); and the Network on Life Course Health Dynamics and Disparities in 21st Century America, which receives funding from the National Institutes of Health (grant number R24AG045061). M. Zacher was supported by the National Institute on Aging of the National Institutes of Health (grant number K01AG078435). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflict of Interest

None declared.

Author Contributions

M. Zacher, S. Brady, and S. Short planned the study. M. Zacher and S. Brady organized the data. M. Zacher analyzed the data and drafted the paper. All authors revised the paper and approved the final version.