Data for this project were drawn from participants enrolled in the Nevada Exploratory Alzheimer’s Disease Research Center (NVeADRC)¹ and the Nevada Center for Neurodegeneration and Translational Neuroscience (CNTN).² Both studies have been reviewed and approved locally by the Cleveland Clinic Institutional Review Board and all participants gave written, informed consent prior to participation.

Details of these cohorts have been previously reported (Ritter et al., 2018; Sabbagh et al., 2021; Miller et al., 2023). Briefly, the NVeADRC is actively enrolling community-dwelling adults over the age of 50 that maintain a primary and current residency in a non-metropolitan area surrounding Las Vegas, NV. CNTN is a longitudinal, natural history study that is actively enrolling the clinical population at the Cleveland Clinic Nevada in Las Vegas. In both studies, following the aligned protocols, annual visits were conducted at the same single site, including a clinical examination, neuropsychological assessment based on the Uniform Data Set (v3) (Weintraub et al., 2018), and brain magnetic resonance imaging (MRI) acquisition. Demographic information including age, sex, years of education (YOE), race and ethnicity were self-reported and collected during each clinical visit.

All statistical analyses were conducted in matrix laboratory (MATLAB) 2022b.³ Confidence intervals for effect-sizes were computed using the MBESS package in R (Kelley, 2007). If not otherwise stated, statistical significance levels were established at uncorrected p ≤ 0.05.

Participants’ demographic information is summarized in Table 1. Briefly, we assessed 71 rural-dwelling (62.0% women, 71.13 ± 6.45 years old, primarily non-Hispanic (88.5%) White (93.1%), 56.3% clinically normal) and 87 urban-dwelling (43.7% women, 72.16 ± 6.88 years old, primarily non-Hispanic (97.6%) White (88.7%), 52.1% clinically normal) older adults. The ADI state decile was significantly higher in rural than urban cohort, with distributions skewed toward 3–10 in the rural and 1–2 in the urban cohorts, respectively (Supplementary Figure 1).

In all participants, there were no significant differences in age, YOE, or race (1st column in Table 1). Rural cohort had less Hispanic participants ( χ 2 1 = 6.14, p = 0.01) and more women ( χ 2 1  = 5.24, p = 0.02) than urban cohort. Clinical impairment status, MoCA and CDR sum of boxes scores did not differ between rural and urban participants. However, the rural cohort demonstrated a better memory performance on RAVLT immediate [t(156) = 2.45, p = 0.02] and delayed recall [t(156) = 3.12, p = 0.002].

After stratifying by impairment status, there were no significant rural–urban differences for any demographic or cognitive variables expect the RAVLT scores. Rural dwelling order adults still demonstrated better verbal memory performances than urban-dwelling participants (2nd and 3rd column in Table 1).

Table 2 summarizes the residency, clinical impairment and interaction effects in the ANCOVA model for 70 CT measures. Significant FDR-corrected p-values (p_FDR ≤ 0.05) are listed (1st big column). Post-hoc effect-sizes (Cohen’s d) for residency (rural–urban, 2nd big column) and clinical diagnoses (normal-impaired, 3rd big column) are listed for all participants, and for participants in each group, respectively.

For the five regions with a significant interaction effect, a significantly negative correlation was observed between the CT of right para-hippocampal gyrus and ADI state decile in normal participants (partial correlation r, [95% confidence intervals (CI)] = −0.37, [−0.54, −0.17], p < 0.001, Figure 2A, blue), and this correlation was significantly different from the one in impaired participants (p = 0.003, Figure 2A). Similar patterns of associations were observed for all 18 temporal regions (Supplementary Figure 5A), with six regions demonstrating statistical significances (Supplementary Figure 5B).

In contrast, a significant positive correlation between the CT of right precuneus and ADI was observed in impaired participants (r = 0.25 [0.01, 0.45], p = 0.03, Figure 2B, orange).

Given the putative risk of dementia that rural participants are facing (Wiese et al., 2023), older adults from rural areas exhibiting lower CT measures than their urban peers in these regions (blue in Figure 1A) align with our hypotheses, suggesting disadvantages on brain measures for rural-dwelling older adults.

These rural–urban disparities tend to diminish once clinical impairment onset (red arrows in Supplementary Figure 2), as (1) both rural and urban groups showed a significantly lower CT measures in these regions in impaired participants (red arrows in Figure 1), and (2) these impairment-effect seemed to be less pronounced in rural participants (red arrows in Supplementary Figure 2). These interaction effects between residency and impairment reached statistical significance (p_FDR ≤ 0.05) in three regions from this category (left parsopercularis, right lateral-orbito-frontal cortex, and right parahippocampal gyrus, Figure 1C and Supplementary Figure 4), further consolidating that rural cohort tended to have a less pronounced differences between normal and impaired participants in these regions, as compared to urban participants.

Given the implication of these temporal and frontal regions in memory processes, our findings might imply that rural-dwelling older adults who are cognitively normal may have thinner cortex than their urban peer; and decreases in their CT measures to a lesser degree could notably affect the clinical impairment status, as compared to urban participants.

Additionally for CT measures in this category, a negative association with ADI state deciles was evident in only clinically normal participants (Figure 2A and Supplementary Figure 5), indicating that for cognitively normal individuals, living in a more disadvantaged neighborhood in Nevada was associated with thinner cortex in regions that were mostly involved in memory and vulnerable to AD. These results are consistent with previous reports on ADI associations with thinner cortex in AD signature regions (mostly temporal lobe) and smaller hippocampal volumes in younger unimpaired participants from another US state (Hunt et al., 2020, 2021). Our results additionally suggested that once clinical impairment emerged, impairment effects would be more dominant than the neighborhood effects and masked these associations.

Interestingly, our exploratory analysis with hippocampal volume did not find any association with ADI (Supplementary Figure 6B). Given the vulnerability of hippocampus to aging or AD, and our cohorts were in their 70s [participants in previous studies were in their 60 s (Hunt et al., 2020, 2021)], clinical impairment effect might be again more dominant for these regions and thus masked the neighborhood effect in our analyses.

Rural-dwelling participants show thicker cortex in these regions than urban-dwelling participants (red in Figure 1A). Considering the same enhanced dementia risks in rural participants, these rural advantages in brain CT measures were less expected, indicating that a more complex interplay of residency neighborhood contexts with neurobiology may exist. A potential compensatory mechanism might partially explain these observations, as rural participants could recruit other brain regions to supplement the thinner temporal and inferior frontal cortices.

In this study, rural–urban status and neighborhood disadvantage index were used in conjunction with separate analyses. We acknowledge that RUCA codes and ADI rankings are to some degree parallel such that more rural areas are likely to be more disadvantaged. However, our interpretation of the underlying components used to derive each metric suggests that they are sufficiently distinct to be used in conjunction. In our analyses, we first evaluated the rural–urban differences in whole-brain CT measures without any prior knowledge and examined the association between CT measures and ADI in a post-hoc manner.

Since ethnicity was significantly different between rural and urban groups, we have repeated our ANCOVA with ethnicity as an additional covariate. All main findings remained the same in this repeated analysis, as revealed by the overall concordance of significance levels in ANCOVA with and without ethnicity as covariates, and the two categories of rural–urban differences on CT measures (Supplementary Figure 7).

Several limitations should be considered when interpreting our results.

First, our cross-sectional setting and reliance on single measurement without baseline data could only infer associations between CT measures and residency and/or clinical impairment status. The small sample has further posed a threat to the generalizability of our results. However, we consider our preliminary findings as a starting point to address the underrepresentation of rural-dwelling older adults in neuroimaging research. In addition, our study is specifically designed as a longitudinal study. Future follow-up studies are planned once we have sufficient data available, to evaluate whether a significant cortical thinning exists in urban or rural participants once disease manifests, depending on different regions.

Sex was treated as a covariate in our analyses despite that brain structures could differ between men and women. The relatively small sample sizes have limited our statistical power to stratify our analyses by sex (Supplementary Figure 8). Due to the same reason of limited sample size, the association between ADI and CT measures were also analyzed in a post-hoc manner, focusing on regions with significant interaction effect in ANCOVA. Future analyses with increased sample-sizes might better evaluate the sex-stratified association between neighborhood context and brain structures.

The NVeADRC is a community-dwelling cohort whereas the CNTN recruits urban-dwelling order adults seeking active clinical care. Although these two cohorts did not differ in clinical, cognitive functioning or functional status, the differences in recruitment strategy may bias our findings. However, we observed thinner cortices in memory-related regions in community- and rural-dwelling older adults, which provided confidence to our interpretations that the status of rural-living and neighborhood disadvantage might negatively impact the neurobiology in brain regions involved in memory and vulnerable to AD. Nevertheless, future studies following individuals that have already established care in a rural-based clinical setting could further validate our findings.

In addition, the RUCA code and ADI state deciles in NVeADRC and CNTN were all based on participants’ current and primary residences, thus did not consider participants’ time of settlement. Moreover, besides composite RUCA or ADI scores, other detailed measures such as economic differences and living habits of each participant could also affect their brain structures. In the present study, we chose to focus on the core socio-demographic and socio-economic factors of community, and have not controlled these relevant factors due to the lack of information and the limited sample-size. Future studies evaluating full residency history with updated RUCA codes and detailed economic factors might further consolidate and clarify our results.

More importantly, the average RUCA score in our rural cohort is 4.27, and most participants live 1–2 h away from Las Vegas. Though this RUCA score and this distance confer some disadvantages in receiving healthcare, our cohort can only represent the rural and non-metropolitan areas locally in Nevada, and the disparities experienced by more rural and isolated areas representing by higher RUCA codes are not represented by our cohort. Furthermore, even under similar RUCA codes, detailed rural area characteristics and resource allocations could differ and differentially affect the brain structures of older adults who reside in them. Therefore, additional caution is advised generalizing our results to other rural areas with similar RUCA codes. Nevertheless, our study serves as an important initial step in exploring the impact of community background on brain structures.