Rural versus Urban Residence in Adulthood and Incident Cognitive Impairment

Cognitive performance in adulthood is related to social determinants of health including socioeconomic factors, discrimination, segregation, migration, and group resources [1, 2]. In the USA, social determinants of cognitive health vary geographically, and prevalence of and mortality from dementia vary across counties, states, and regions [3‒7]. Cognitive impairment and dementia disproportionately afflict people who currently live in or previously lived in southeastern USA stroke belt states [8‒10].

Rural versus urban living may be an important social determinant of cognitive health, as suggested by findings from studies outside the USA [11, 12]. However, few investigations in the USA have considered rural versus urban residence. USA studies have shown rural residence associated with a higher incidence of and mortality from dementia [7, 13, 14]; and with a higher prevalence of dementia when assessed by clinical diagnostic criteria [15, 16] but lower prevalence when assessed by Medicare claims [14, 17]. Explanations or mechanisms for rural/urban differences in dementia or cognitive impairment have scarcely been explored.

With sparse literature on rural versus urban residence and cognition in the USA, we lack critical knowledge about place of residence as a determinant of cognitive health in the USA population. To address this gap, we studied over 20,000 middle-aged and older adults living in rural and urban areas throughout the 48 contiguous states to estimate the magnitude of and examine reasons for the rural/urban disparity in incident cognitive impairment (ICI).

The Reasons for Geographic and Racial Differences in Stroke Study (REGARDS) is a population-based prospective cohort study of 30,239 Black and white adults residing in the 48 contiguous states who were aged 45 years and above at study entry (2003–2007). Sampling was stratified by age, sex, race, and geographic region [18, 19]. For this analysis, 25,393 participants (84% of the cohort) met eligibility criteria at baseline by answering ≥5 items correctly on the Six-Item Screener [20], which indicated that their global cognitive ability at baseline was intact, and reporting no history of clinical stroke at baseline. The last follow-up of the cohort for this analysis was October 2020. Attrition from the eligible cohort, leading to inability to ascertain ICI status and therefore exclusion from data analysis, was 17.8% overall, 17.9% among urban residents, 17.8% among large rural residents, and 16.5% among small rural residents, leading to retention of 20,878 eligible participants (69% of the cohort) in this analysis. The similar degree of attrition across rural/urban categories suggests that selection bias would be minimal and not invalidate our results.

Longitudinal cognitive assessments, conducted in telephone interviews beginning in 2006 and repeated at 2-year intervals, included word list learning, word list delayed recall, and animal naming (semantic fluency) [21, 22]. In word list learning, a participant attempts to memorize a list of 10 words over 3 trials; scores range from 0 to 30. In word list delayed recall, a participant attempts to recall the 10 words after a delay in the interview filled with noncognitive questions; scores range from 0 to 10. In animal naming, a participant generates the names of as many animals as possible in 60 s. Using participants’ most recently recorded assessments at the time of this analysis, we identified participants with ICI as those who scored ≥1.5 standard deviations below the mean on at least 2 of the 3 cognitive tests [23‒25].

We used Rural-Urban Commuting Area (RUCA) codes applied at the census tract level to classify each participant’s home address at baseline into one of the following categories: urban area with population ≥50,000, large rural city/town with population 10,000–49,999, small rural town/area with population 2,500–9,999, or isolated small rural area with population ≤2,499 [19, 26]. Census tract-level RUCA coding provides more accurate characterization of each participant’s neighborhood than county-level RUCA coding [19]. We consolidated small rural town/area and isolated small rural area categories into a single category that we refer to as “small rural.” We refer to the large rural city/town category as “large rural.”

Sociodemographic factors included age (years), self-reported sex (female, male), self-reported race (Black, white), census region (Midwest, Northeast, South, West), attained education (college graduate, some college, high school graduate, less than high school), and annual household income (≥$75,000, $35–74,000, $20–34,000, <$20,000). Health behaviors included self-reported alcohol consumption (none, moderate, heavy), smoking (never, former, current), exercise frequency (≥4 times/week, 1–3 times/week, none), and Mediterranean diet score (high, medium, low) based on a self-administered semiquantitative Block food frequency questionnaire [27]. Clinical characteristics ascertained via ECG, blood assays, medication audits, anthropometric measurements, and self-report included coronary heart disease, atrial fibrillation, diabetes, hypertension, body mass index (normal or lower, overweight, obese), dyslipidemia, depressive symptom score based on the 4-item Centers for Epidemiologic Studies Depression Scale [28], and self-rated health (excellent, very good, good, fair, poor). We ascertained incident clinical stroke occurring after baseline but before most recent follow-up cognitive assessment, first identified by participant report and then confirmed through medical record review and adjudication by a physician committee using standard criteria [29].

Of 20,878 participants eligible for analysis, 2,008 (10%) were missing RUCA classification and another 7,346 (35%) were missing 1 or more covariate values (Table 1). To improve precision and reduce bias that could arise from omitting eligible participants who had missing values, we included all variables listed in Table 1 in multiple imputation by chained equations, created 25 imputation datasets, and used Rubin’s rules to combine estimates across datasets.

We summarized participant characteristics by baseline RUCA classification and by ICI status. We used logistic regression to determine the association of ICI with large rural and with small rural residence at baseline, each compared with urban. To analyze reasons for rural/urban differences in ICI, we used 2 modeling strategies, as follows.

First, we assessed independence of the association of rural residence with ICI by adjusting for sociodemographics, health behaviors, and clinical characteristics. We considered age, sex, race, and attained education as confounders because they were defined earlier in the life course than place of residence at baseline, and census region as a confounder because place of residence at baseline exists within a defined census region. We considered baseline income, health behaviors, and clinical characteristics as plausibly being mediators or plausibly being on confounding paths because these factors are time-varying across the adult life course and may be influenced more strongly either by rural/urban residence at baseline or by earlier life rural/urban residence or other unmeasured factors (online suppl. Figure; for all online suppl. material, see https://doi.org/10.1159/000530961). We considered incident stroke as a mediator because it occurred after place of residence at baseline was defined.

Second, we assessed effect heterogeneity using interaction terms for baseline rural residence with sociodemographic, behavioral, and clinical risk factors for ICI. For each risk factor, we defined a reference condition as unexposed and urban, then estimated ORs corresponding to the main effect of rural residence, the main effect of risk factor exposure, and the joint effect of rural residence and risk factor exposure. We calculated a measure of effect heterogeneity on the additive scale, the relative excess risk due to interaction (RERI), defined as the joint effect minus the sum of main effects plus 1 [30]. RERI >0 indicates that the joint effect of rural residence and risk factor exposure is greater than would be expected from summing the main effects. In addition to RERI, we reported a measure of effect heterogeneity on the multiplicative scale, the OR for rural residence × risk factor interaction, defined as the joint effect divided by the product of main effects.

At baseline, home addresses of our 20,878 participants were 79.8% urban, 11.7% large rural, and 8.5% small rural, which is similar to the USA national population distribution [26]. Compared with urban dwellers, those who resided in small rural areas were lower percentage Black, higher percentage living in the South, had lower education and income, less alcohol use and smoking, more frequent exercise, lower Mediterranean diet score, more heart disease and dyslipidemia, and lower self-rated health (Table 1). Compared with participants who were not cognitively impaired at follow-up, those who experienced ICI were older, higher percentage male, Black, and living in the South, had lower education and income, less alcohol use, less frequent exercise, lower Mediterranean diet score, more heart disease, diabetes, hypertension, dyslipidemia, and depressive symptoms, lower self-rated health, and more incident stroke (Table 1). The average length of follow-up was 9.4 years (range: 0.4, 15.8). ICI at follow-up occurred in 7.8% of participants residing in urban areas at baseline, 7.8% of those in large rural areas, and 9.2% of those in small rural areas (Table 2).

Residing in a small rural area relative to urban was independently associated with ICI (Table 3). All adjusted ORs for small rural were well above 1.00 and 95% CIs were compatible with adjusted ORs in the population being no lower than 1.02 and as high as 1.79, depending on the set of adjustment variables. Small rural dwellers had an estimated 24% higher odds of ICI than urban dwellers (95% CI: 2% higher to 53% higher) adjusted for age, sex, race, census region, attained education, baseline income, health behaviors, clinical characteristics, and incident stroke (Model 6). Residing in a large rural area relative to urban was not independently associated with ICI (Table 3).

Across 19 sociodemographic, behavioral, and clinical risk factors we investigated for effect heterogeneity with small rural residence, RERI point estimates were positive for several risk factors, indicating that the adjusted association of the risk factor with ICI may be stronger for small rural dwellers than for urban dwellers. The 95% CI of RERI fell completely above zero for former smoking, and largely above zero for nondrinking versus light alcohol drinking, no exercise versus ≥4 times/week, CES-D depressive symptom score of 2 versus 0, and fair versus excellent self-rated health (Table 4). For large rural residence versus urban residence, RERI estimates tended to be close to zero or had wide 95% CIs. However, Black race, hypertension, and depressive symptoms each had RERI with 95% CI mostly below zero, suggesting that ICI occurred less commonly than expected among those groups in large rural areas. In addition, multiplicative scale interaction showed that the combination of heavy alcohol use with large rural residence was associated with three times the odds of ICI compared with what would be expected if there were no interaction (Table 5).

In this study of ICI based on a large and diverse nationwide sample of middle-aged and older adults in rural and urban areas in the USA, residing in a small rural area at baseline was associated with higher odds of ICI over a mean of 9.4 years of follow-up when compared with residing in an urban area at baseline. We observed this association when we adjusted for demographics, attained education, baseline income, health behaviors, clinical characteristics, and incident stroke. We also observed that former smoking, nondrinking, lack of exercise, higher depressive symptoms, and low self-rated health may be associated with ICI more strongly among small rural residents than among urban residents, though many 95% CIs for RERI were too wide to draw firm conclusions. In contrast, we found no association of large rural residence with ICI when compared with urban residence. However, multiplicative interaction analysis showed higher than expected ICI among those with heavy alcohol use in large rural areas, and RERI analysis suggested lower than expected ICI among Black individuals, those with hypertension, and those with depressive symptoms in large rural areas.

Most comparisons of adult cognition in rural versus urban settings have been conducted outside the USA, with mixed findings perhaps arising from methodological heterogeneity [11, 31‒39]. Country-specific contextual factors likely modify the association of rural residence with cognitive outcomes, and therefore, studies such as ours conducted in the USA are essential to better understand social determinants of cognitive aging in the USA population. Although studies have clearly identified broad regional variability in dementia mortality, dementia prevalence, and ICI in the USA [3‒7], rural versus urban residence has been sparsely investigated [7, 13‒17]. In a small study of older adults in Indianapolis in the 1990s, rural residence before adulthood was associated with higher odds of prevalent Alzheimer's disease (AD) and incident AD relative to urban residence before adulthood [13, 15]. In the nationwide Health and Retirement Study in 2010, residents of rural census tracts had higher odds of prevalent dementia and prevalent cognitive impairment without dementia relative to residents of urban census tracts [16]. In a nationwide study of death certificates in 2014, mortality rates for AD were higher in counties classified as nonmetro rural than in counties classified as large central metro [7]. In contrast with other studies, in a county-level study of Kentucky and West Virginia Medicare beneficiaries in 2013, rural counties had a lower prevalence of diagnosed dementia relative to urban counties [17]. Finally, in a nationwide county-level study of Medicare beneficiaries in 2015 [14], prevalence of dementia was lower in rural counties than in metropolitan counties; however, incidence of dementia was higher in rural counties.

Our study has limitations. First, RUCA classification and covariate data were missing for some eligible participants. We mitigated this through multiple imputation to increase precision and reduce selection bias that could arise from excluding eligible participants. Second, RUCA classification and covariates largely reflected participants’ exposure profiles at a single point in time even though place of residence, social class indicators, health behaviors, and clinical history change over the life course. We mitigated this by considering whether certain covariates were more likely confounders or mediators (online suppl. Figure). Third, participants who scored lower on cognitive tests at baseline would more readily cross the threshold to be identified as cases of ICI during follow-up, even if their degree of decline during follow-up was similar to that of participants who scored higher at baseline. We determined ICI at the end of follow-up using the most recent cognitive assessment rather than estimating longitudinal trajectories of mean cognitive decline, modeling fluctuations over time in cognitive impairment status, or conducting survival analysis of time-to-ICI. Nevertheless, our ICI outcome has clinical and practical relevance because people who drop below a certain level of cognition will struggle to maintain independence and quality of life. Fourth, rural and urban dwellers may respond differently to cognitive tests used in REGARDS due to, for example, rural/urban differences in education. However, published age-, sex-, race-, and education-adjusted norms on these measures were derived from mixed rural and urban samples in the USA, in which associations of attained education with verbal fluency, word list learning, word list recall, and other cognitive measures were similar across samples from rural and urban areas [40]. Fifth, despite our large sample, we had low precision for detecting joint effects of risk factors for ICI plus rural residence, precluding strong conclusions about the effect heterogeneity we explored.

Small rural residence is a predictor of ICI among USA adults. Our findings expand the knowledge of social determinants of cognitive aging in the USA. Understanding determinants of cognitive aging may be even more important for rural populations than for urban populations given limited infrastructure and resources for caregiving in rural areas. Further research to better understand why small rural residents are at higher risk as well as mechanisms to ameliorate that risk and buffer the consequences of ICI will support future efforts to improve rural public health.

The authors thank the other investigators, the staff, and the participants of REGARDS for their valuable contributions. A full list of participating REGARDS investigators and institutions can be found at https://www.uab.edu/soph/regardsstudy/.

REGARDS was approved by the Institutional Review Boards of the University of Alabama at Birmingham (approval number IRB-020925004) and collaborating institutions. Verbal informed consent was obtained from participants during the telephone survey and written informed consent during the in-home visit.

The authors have no conflicts of interest to declare.

REGARDS is supported by cooperative agreement U01 NS041588 cofunded by the National Institute of Neurological Disorders and Stroke (NINDS) and the National Institute on Aging (NIA), National Institutes of Health, Department of Health and Human Services. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NINDS or the NIA. Mr. Harris was supported by a College Undergraduate Research Award from the College of Life Sciences, Brigham Young University. Representatives of NINDS were involved in the review of the manuscript but were not directly involved in the collection, management, analysis, and interpretation of the data. Representatives from the BYU College of Life Sciences did not have any role in the design and conduct of the study, the collection, management, and interpretation of the data, or the preparation or approval of the manuscript.

Dr. Thacker had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Harris, Bennion, Howard, Wadley, McClure, Levine, Manly, Glymour, Wisco, and Thacker. Acquisition, analysis, or interpretation of data, and critical revision of the manuscript for important intellectual content: all authors. Drafting of the manuscript: Harris, Bennion, Magnusson, and Thacker. Statistical analysis: Bennion, Magnusson, Avila, and Thacker. Obtained funding: Howard, Wadley, McClure, and Manly. Supervision: Howard, Manly, and Thacker.

Deidentified data will be shared by REGARDS investigators upon request from any qualified investigator.