Neighborhood socioeconomic deprivation is associated with worse cognitive performance and in vivo amyloid accumulation

Abstract

Socioeconomic disadvantage (SED) has been associated with poorer brain health, yet its underlying pathology remains incompletely understood. We examined whether neighborhood-level SED, measured using the Area Deprivation Index (ADI), relates to amyloid deposition assessed with amyloid positron emission tomography (PET). Participants (n = 1110) underwent cognitive assessment using the mini mental state examination (MMSE) and PET scanning with amyloid-specific tracers. Associations between national and state ADI and MMSE and global amyloid burden were evaluated using linear models adjusting for age, sex, and APOE-ε4 carrier status. Higher ADI was associated with lower MMSE scores, with both national and state measures showing significant inverse associations independent of age and sex (all p < 0.001). Higher ADI was also significantly associated with greater amyloid burden among cognitively unimpaired participants (β = 0.18, p = 0.006, d = 0.27). No mediation effect with centiloids was observed between ADI and MMSE in the analysis, however. Neighborhood socioeconomic disadvantage is associated with worse cognitive performance and for the first time were shown to be associated with amyloid accumulation. These findings underscore the need to consider socioeconomic context in early-stage risk assessment and may help inform targeted prevention strategies aimed at reducing disparities in dementia outcomes.

Introduction

Socioeconomic disadvantage is consistently linked to poorer health outcomes, including increased rates of cardiovascular disease, diabetes, and cognitive decline^1,2. The Area Deprivation Index (ADI) is a validated, census-based composite measure that captures neighborhood-level socioeconomic disadvantage through indicators such as income, education, employment, and housing quality³. Prior studies using ADI and related metrics show that dementias disproportionately affect individuals from economically disadvantaged communities⁴ and higher ADI scores have been associated with lower cognitive performance^1,5–7.

Alzheimer’s dementia (AD), characterized by multidomain cognitive decline⁸, currently affects an estimated 6.9 million Americans⁸. Although genetic factors such as the apolipoprotein E ε4 (APOE-ε4) allele are well-established contributors to AD risk and pathology, growing evidence suggests that environmental and socioeconomic exposures may also shape vulnerability to AD^9,10 potentially through mechanisms such as chronic stress, dietary patterns, toxin exposures, and infectious processes^9–11.

Despite these emerging links, the neurobiological pathways connecting socioeconomic disadvantages to cognitive impairment remain poorly understood. Prior studies by our group have examined the associations between brain structure and cognitive performance and have shown that neighborhood disadvantage is linked to widespread cognitive vulnerability and structural brain differences¹². Additionally, another study found that people from lower-income households exhibited patterns consistent with accelerated white matter aging, including higher mean diffusivity, reduced neurite density, decreased myelination, and lower iron levels¹³. However, the association between neighborhood disadvantage and the accumulation of dementia-related proteins has not been well characterized, underscoring the need for biomarker-based approaches such as amyloid positron emission tomography (PET) imaging.

Here, we examined whether neighborhood-level disadvantage is associated with Alzheimer’s-related pathology by assessing the relationship between ADI, brain amyloid-β (Aβ) burden using PET imaging, and serum biomarkers as a comparator in cognitively normal (CN) and cognitively impaired (CI) participants using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database.

Results

Participant characteristics

We included 1110 participants (Table 1) with a mean age of 69.45 ± 7.20 years, of whom 58% were female. The mean MMSE score was 28.11 ± 2.26. Median ADI values indicated moderate neighborhood disadvantage (state decile: 3 [IQR 1–6]; national percentile: 31 [IQR 11–55]). In multivariable linear regression models higher ADI was significantly associated with lower global cognitive performance as measured by the MMSE. Specifically, higher national ADI was associated with lower MMSE scores (β = −0.013, SE = 0.002, p < 0.001), and a similar association was observed for state ADI (β = −0.100, SE = 0.023, p < 0.001). In both models, older age was independently associated with lower MMSE performance (p < 0.001), while sex was significantly associated with MMSE scores (p < 0.001). The overall models explained a modest but significant proportion of variance in MMSE scores (national ADI model: R² = 0.066; state ADI model: R² = 0.060, Table 2).

Table 1.

Demographic, Cognitive, and Neighborhood Socioeconomic Characteristics of the Study Cohort

Characteristic		Overall Sample (N = 1110)	Imaging Subgroup		p-value
			CN	CI
Age (years)		69.45 ± 7.20	71 ± 8	71 ± 7	0.50
Sex	Male	464 (42%)	186 (44%)	127 (43%)	0.8
	Female	650 (58%)	240 (56%)	171 (57%)
MMSE		28.11 ± 2.26	29 (1.23)	26.6 (2.89)	<0.001
ADI (State)		3.00 (1, 6)	3 (4)	3 (4)	0.96
ADI (National)		31.00 (11, 55)	32 (40)	31 (37)
APOE ε4 carrier		270 (40.11%)	157 (40%)	113 (40%)	>0.90
Amyloid positive (Centiloid ≥ threshold)		200 (27.62%)	106 (25%)	94 (32%)	0.049

Table 2.

Associations of National and State Area Deprivation Index (ADI) With MMSE Performance in Models Adjusted for Age and Sex

N = 1110	Model / Predictor	β	SE	t value	p value	R²	Adj. R²	F statistic
National ADI	ADI (national)	−0.012	0.002	−5.10	<0.001	0.066	0.064	26.12
	Age (years)	−0.054	0.009	−5.88	<0.001
	Sex	0.581	0.136	4.27	<0.001
State ADI	ADI (state)	−0.100	0.023	−4.28	<0.001	0.060	0.057	23.45
	Age (years)	−0.054	0.009	−5.83	<0.001
	Sex	0.573	0.137	4.19	<0.001

ADI and plasma biomarkers

In additional analyses examining potential biological pathways, ADI was not significantly associated with plasma biomarkers of AD or neurodegeneration, including PT217, Aβ42/Aβ40 ratio, NfL, or GFAP (all p > 0.05). In contrast, age was significantly associated with all biomarkers (all p < 0.001, Supplementary Table 1).

ADI and amyloid burden

PET sub-analysis included a total of 737 participants (CN n = 518; CI n = 219, Fig. 1). CN participants were younger than those with CI (p = 0.001) and had a lower proportion of males (CN: 189 males, 329 females; CI: 108 males, 111 females; p = 0.001). The frequency of APOE ε4 carriers did not differ significantly between groups (CN: 56.7%; CI: 66.7%; p = 0.41). Overall amyloid positivity in our total study population with a centiloids cutoff of 30 was 27.62%. Centiloids, ADI-National, and ADI-State values were not normally distributed (Centiloids: W = 0.82, p < 0.001; ADI-National: W = 0.94, p < 0.001; ADI-State: W = 0.90, p < 0.001). Centiloid values showed a right-skewed distribution with a substantial proportion of low and negative values, consistent with the inclusion of amyloid-negative participants.

Fig. 1. Centiloids distribution and the Area Deprivation Index (ADI).

A Histogram and density plot of Centiloids values in the study sample. The distribution shows a predominance of low and negative values, reflecting a substantial proportion of amyloid-negative individuals in the ADNI cohort (mostly phase 4); B Distribution of MMSE scores across binned ADI (National) levels; C National ADI and amyloid burden (Centiloids) in cognitively unimpaired (navy) and cognitively impaired (orange) participants.

Linear regression analyses assessed the relationship between ADI and global amyloid burden (centiloids), controlling for age, sex, and APOE ε polymorphism. In the CN group, both state and national ADI were significantly associated with higher centiloids values (β = 1.38, p = 0.01; β = 0.18, p = 0.006, respectively). Effect sizes for both state and national ADI were moderate (Cohen’s d = 0.23, 0.27, respectively). In contrast, these associations were not significant in the CI group for state (β = –1.45, p = 0.24) or national (β = –1.14, p = 0.21) ADI (Table 3, Fig. 1). No significant interactions were found between ADI and APOE-ε4 carrier status (p = 0.09).

Table 3.

Results with ADI as the predictor, centiloids as the outcome and age, gender, and APOE ε polymorphisms as covariates

Group	Variable	Estimate (β)	SE	t-value	p-value	FDR-p	Cohen’s d
CN (n = 518)	Intercept	78.05	14.75	5.29	0.002	NA	0.30
	ADI-State	1.38	0.68	2.05	0.016	0.032	0.23
	Model	75.44	14.80	5.10	0.004	NA	0.28
	ADI-National	0.18	0.07	2.62	0.006	0.024	0.27
CI (n = 219)	Intercept	72.52	22.24	3.26	0.740	NA	0.04
	ADI-State	−1.45	0.82	−1.76	0.242	0.242	−0.17
	Intercept	68.18	22.54	3.02	0.807	NA	0.03
	ADI-National	−0.14	0.08	−1.69	0.210	0.242	-0.18

ADI Area Deprivation Index, CN Cognitively Normal, CI Cognitive Impairment (mild cognitive impairment or mild dementia), SE Standard Error.

Lastly, mediation analysis showed no evidence of an indirect effect of ADI on MMSE through amyloid burden (ACME = −0.00022, 95% CI − 0.00135 to 0.00, p = 0.70).

Discussion

We investigated the associations between socioeconomic disadvantage and cognition and brain amyloid accumulations. Our findings showed that ADI is significantly associated with worse cognition and with amyloid burden among CN individuals, even after adjusting for covariates.

Previous studies have addressed the associations between neighborhood deprivation and worse cognition and showed higher odds of cognitive impairment in those participants living in worse neighborhoods¹⁴. In agreement with previous findings and using the national level ADNI data, we found significant associations between higher state and national ADI and worse MMSE scores.

These results extend prior work demonstrating that socioeconomic disparities contribute to increased risk of cognitive decline and dementia, including neuropathological studies that have reported higher odds of AD pathology with increasing neighborhood disadvantage^2,4,6. Our findings suggest that these associations are detectable through amyloid PET imaging, at least in the preclinical phase of disease. The lack of association in the CI group may reflect that other disease associated processes exert stronger effects when the clinical symptoms have emerged and as disease processes advance (i.e., pathologic factors may dominate when amyloid positive, attenuating the observable influence of socioeconomic context).

Interestingly, we did not observe significant associations between ADI and plasma biomarkers of amyloid, tau, neurodegeneration, or astroglial activation. This may suggest that the relationship between neighborhood disadvantage and cognitive outcomes is not mediated by systemic differences in these biological pathways, at least as captured by circulating biomarkers. Alternatively, these findings may reflect the limitations of plasma biomarkers in capturing early or subtle neuropathological changes.

In addition, although the APOE-ε4 allele is a well-established genetic risk factor for AD and is associated with greater amyloid accumulation in the brain¹⁵, we did not find any significant interactions between APOE-ε4 carrier status and ADI. Future research should examine potential interactions between polygenic susceptibility and environmental disadvantage and employ longitudinal designs to characterize differential trajectories of amyloid accumulation and the combined effects of gene–environment interactions on disease progression.

Overall, these findings reinforce the multifactorial nature of AD risk and underscore the importance of incorporating social determinants of health into models of neurodegeneration. Although the observed effect sizes were small, they remain meaningful, as neighborhood deprivation represents one of the few potentially modifiable factors associated with dementia risk at the population level. As such, even modest associations may have substantial public health implications. Future studies integrating environmental exposures, biological markers, and advanced analytic approaches, including machine learning, may further elucidate the pathways through which social disadvantage contributes to AD risk and progression.

Lastly, one important consideration is the lack of an indirect effect of ADI on cognitive performance through amyloid burden. While this finding may suggest that amyloid pathology does not mediate the relationship between socioeconomic context and cognition in our sample, it should be interpreted with caution. The MMSE, although widely used, is relatively nonspecific for AD–related cognitive changes and has limited sensitivity, particularly among cognitively normal individuals. As a result, subtle cognitive differences or early disease-related changes may not have been adequately captured, potentially attenuating our ability to detect mediation effects. Future studies incorporating more sensitive and domain-specific cognitive measures, as well as longitudinal assessments, may be better positioned to clarify whether socioeconomic factors influence cognition through AD-related biological pathways.

Our main limitation was the cross-sectional study design, which precludes causal inference. Additionally, larger and more demographically diverse samples are needed to enhance statistical power and explore subgroup-specific effects. Future studies should leverage longitudinal ADNI data, as they become available, to examine temporal relationships between neighborhood deprivation and AD–related biomarkers, including amyloid burden, CSF measures, and cognitive decline.

Methods

Participants

ADNI participants with complete ADI, amyloid PET data, and clinical diagnosis were included (Fig. 2A). Written informed consent was obtained from all participants, and study procedures were approved by the institutional review boards of all participating institutions. The data selection process is illustrated in Fig. 2. For the present study, we used data from ADNI phase 4, as it is the first phase of ADNI with available ADI data.

Fig. 2. Study population selection and distribution of Area Deprivation Index (ADI) scores.

A Data selection and quality control flow diagram; B Distribution of the national ADI; C Distribution of the state ADI.

Cognitive assessment

Participants were categorized into cognitive groups based on ADNI clinical diagnoses, derived from a combination of standardized cognitive assessments, including the Alzheimer’s Disease Assessment Scale (ADAS), Clinical Dementia Rating (CDR), and Mini-Mental State Examination (MMSE), along with clinician evaluation by trained ADNI physicians following established ADNI protocols. In addition, MMSE scores were used as an outcome measure in analyses examining the association between neighborhood disadvantage (ADI) and cognitive performance. CN individuals had no memory complaints, demonstrated normal memory performance, and showed no cognitive or functional impairment. The CI group consisted of individuals who reported memory concerns, exhibited abnormal memory performance ranging from mild to more pronounced deficits, but retained functional independence¹⁶. Clinical diagnoses were obtained from ADNI. Participants were grouped into CN and CI based on global cognitive function that was assessed using the Mini-Mental State Examination (MMSE)¹⁷, a widely used screening tool that evaluates orientation, attention, memory, language, and visuospatial abilities. MMSE scores range from 0 to 30, with higher scores indicating better cognitive performance and was administered according to standardized procedures by trained study personnel. MMSE scores were treated as a continuous outcome variable in all analyses. Associations between neighborhood socioeconomic deprivation and cognitive performance were examined using multivariable linear regression models, adjusting for age and sex. Participants with missing MMSE or covariate data were excluded from the respective analyses.

Area deprivation index

The ADI data was extracted from ADNI. ADI was calculated using the United States Census indicators of poverty, education, housing, and employment and neighborhood socioeconomic status was ranked by disadvantage at the state and national level (https://www.neighborhoodatlas.medicine.wisc.edu/). National ADI ranks neighborhoods on a percentile scale (1–100) relative to all U.S. census block groups, whereas state ADI ranks neighborhoods within each state using deciles (1–10). In our study, ADI values were obtained directly from the ADNI dataset (Phase 4), where both national percentile rankings and state decile rankings were provided based on participants’ residential census block groups. Lower values indicate higher neighborhood disadvantage (Fig. 2)¹⁸.

APOE genotyping

APOE genotyping was performed using baseline blood samples in accordance with ADNI protocols. Participants were categorized based on APOE ε4 carrier status. Individuals with at least one ε4 allele were classified as APOE ε4 positive.

Plasma biomarker assessment

Plasma biomarkers were quantified using established immunoassay platforms. The biomarkers of interest included phosphorylated tau at threonine 217 (p-tau217), the amyloid-β 42 to amyloid-β 40 ratio (Aβ42/Aβ40), neurofilament light chain (NfL), and glial fibrillary acidic protein (GFAP). These markers were selected to capture key domains of AD biology, including tau pathology, amyloid deposition, axonal injury, and astroglial activation. Blood samples were collected following standardized protocols, processed to obtain plasma, and stored at −80 °C until analysis. All assays were performed in accordance with manufacturer protocols, with appropriate calibration and quality control procedures. Samples were analyzed in batches, and internal standards and controls were included to ensure assay reliability and reproducibility. For proteomic-related measurements, standardized workflows consistent with large-scale biomarker studies were applied, including normalization procedures across batches using internal standards and scaling approaches to minimize technical variability. Values below detection thresholds or with technical failure were treated as missing.

Imaging data acquisition and preprocessing

Imaging data processed using the ADNI pipelines (https://adni.loni.usc.edu/data-samples/adni-data/neuroimaging/pet/) were downloaded for these analyses (UC Berkley-Amyloid PET 6 mm Res Analysis, version March 13, 2026). In brief, two amyloid tracers of [¹⁸F]florbetapir (FBP) and [¹⁸F]florbetaben (FBB) were used. PET images were motion corrected before averaging all frames into a single static image. The MRI was segmented and parcellated with Free Surfer v7.1.1 to define a global cortical measure that is made up of frontal, anterior/posterior cingulate, lateral parietal, lateral temporal regions. To generate the standardized uptake value ratios (SUVRs), each amyloid PET scan was co-registered to the corresponding segmented MRI and normalized by the whole cerebellum. SUVR was then calculated for the cortical summary region.

Regarding quality control, we excluded scans with significant motion artifacts, failed PET-to-MRI registration, or structural MRI scans that did not pass quality assessment. In addition, centiloid outliers defined as ±3 standard deviations from the mean were excluded, and 112 participants were removed due to structural scan quality control failure. Figure 2 illustrates the data selection flowchart.

Lastly, centiloid values were calculated from cortical composite standardized uptake value ratios (SUVRs) following the ADNI centiloid processing pipeline¹⁹. Briefly, amyloid PET scans (including both florbetapir [FBP] and florbetaben [FBB] tracers) were first preprocessed and co-registered to each participant’s structural MRI, followed by spatial normalization to a standard template. SUVRs were then computed using a predefined cortical target region (including frontal, parietal, temporal, and cingulate regions) and a reference region (whole cerebellum), consistent with the centiloid standard methodology¹⁹.

These SUVRs were subsequently linearly transformed into centiloid units using tracer-specific calibration equations established within the ADNI framework, enabling harmonization across tracers and acquisition protocols. The centiloid scale provides a standardized metric for amyloid burden, anchored such that 0 corresponds to the mean uptake in young amyloid-negative controls and 100 corresponds to the typical uptake observed in patients with AD.

Both FBP- and FBB-derived centiloid values were included in the present analysis after harmonization. Additional details on the ADNI PET acquisition and processing pipeline are available (https://adni.loni.usc.edu/data-samples/adni-data/neuroimaging/pet/). Consistent with prior literature and expert consensus, a centiloid threshold >30 was used to define amyloid positivity²⁰.

Statistics analysis

All analyses were conducted in R (version 4.4.2). Outliers in amyloid centiloid values were defined as ±3 standard deviations from the mean and were excluded from analyses. Normality of centiloid, ADI-National, and ADI-State distributions was assessed using the Shapiro–Wilk test and visual inspection of histograms and Q–Q plots. All variables demonstrated deviations from normality. Accordingly, Spearman rank correlations were used for bivariate analyses. Multivariable associations were assessed using linear regression models, which are robust to moderate deviations from normality in large samples. Sensitivity analyses using non-parametric approaches yielded consistent results. Descriptive statistics were computed to summarize participant characteristics and key study variables. Continuous variables are reported as mean ± standard deviation or median (interquartile range), as appropriate, and categorical variables as counts and percentages. To examine the association between neighborhood disadvantage and amyloid burden, linear regression models were constructed with amyloid centiloid values as the dependent variable and ADI as the primary independent variable. Models were performed using both national percentile and state decile ADI measures, independently. Separate models were also constructed to evaluate the association between ADI and cognitive performance, using MMSE scores as the outcome. All models were adjusted for age, sex, and apolipoprotein E (APOE) ε4 carrier status. Model assumptions, including linearity, homoscedasticity, and normality of residuals, were evaluated. To account for multiple comparisons, false discovery rate (FDR) correction was applied using the Benjamini–Hochberg method. Statistical significance was defined as a two-sided p-value < 0.05.

Ethics statement

Data used in this study were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. The ADNI study was approved by the Institutional Review Board of the University of Southern California, which served as the central coordinating site, as well as by the institutional review boards of all participating sites. A full list of participating institutions is available at http://adni.loni.usc.edu. All procedures were conducted in accordance with the Declaration of Helsinki and its later amendments. Written informed consent was obtained from all participants or their legally authorized representatives prior to participation. As this study involved secondary analysis of de-identified data, no additional institutional review board approval was required. Specific approval numbers for the multi-center ADNI study are not publicly available.

Author contributions

M.M. conceived and designed the study, performed data analysis,and drafted the manuscript. R.S. contributed to manuscript writing. A.P.M. contributed to data interpretation and critically revised the manuscript. D.M. supervised the study, contributed to study design and interpretation, and critically revised the manuscript. All authors reviewed and approved the final manuscript.

Data availability

Data used in this study were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.usc.edu). The datasets analyzed during the current study are publicly available to qualified researchers upon application and approval of a data use agreement. Additional data processing scripts are available from the corresponding author on reasonable request.

Code availability

Code used for data processing and analysis is available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s44400-026-00121-8.