Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Jun 16.
Published in final edited form as: Ethn Health. 2026 Apr 8;31(6):583–599. doi: 10.1080/13557858.2026.2655623

Assessing Racial Disparities in Dementia-Related Vulnerabilities: A Machine Learning Analysis of Social, Environmental, and Climate Factors

Solmaz Amiri 1, Wyatt P Bensken 2, Dedra Buchwald 3
PMCID: PMC13265239  NIHMSID: NIHMS2179215  PMID: 41952418

Abstract

Introduction

Racial differences in exposure to area-level social determinants of health (SDoH) may contribute to distinct vulnerability profiles among people living with dementia.

Methods

This cross-sectional study used data from the OCHIN Community Health Equity Database to examine SDoH factors among patients with dementia and mild cognitive impairment. SDoH measures encompassed social, environmental, and climate vulnerability factors. We trained Random Forest classifiers to identify the most influential SDoH features distinguishing racial groups.

Results

Common SDoH factors — those ranking among the top features across multiple racial groups — included percentage of minority populations, limited English proficiency, exposure to diesel particulate matter or fine particulate matter, impaired watershed, and frequency of extreme heat and wildfire smoke days. Distinct SDoH factors — those identified as top-ranked for only one or two groups — included unemployment and disability rates in ZIP codes where American Indian and/or Alaska Native patients lived; lack of walkability in areas where Asian patients resided; hurricane occurrences in ZIP codes where Black patients lived; low educational attainment in Native Hawaiian or other Pacific Islander communities; and housing cost burden, lack of internet access, and older housing stock in areas where Multiracial patients lived.

Discussion

These findings highlight the importance of incorporating multidimensional SDoH measures when assessing structural contexts in dementia research.

Keywords: Machine learning, Environmental Justice Index, Dementia, Race

Sustainable Development Goals (SDG) Keywords: SDG 3: Good health and well-being, SDG 10: Reduced inequalities

1. Introduction

Dementia is a widespread condition characterized by cognitive impairments that usually emerge in older adulthood and interfere with a person’s capacity to carry out daily tasks independently.1,2 Mild cognitive impairment refers to a clinical stage in which individuals exhibit cognitive decline greater than expected for their age but not severe enough to meet criteria for dementia.3 The 2016 Harmonized Cognitive Assessment Protocol, a nationally representative substudy of the Health and Retirement Study, provides updated national estimates of cognitive impairment.2 Based on these data, the point prevalence of dementia among United States (US) adults aged 65 years and older is approximately 10% and that of mild cognitive impairment is approximately 22%.2 Dementia poses a major burden on the society and healthcare system. In 2022, the total cost of health care and long-term care for Medicare beneficiaries with dementia was almost three times higher than same aged patients without this condition ($43,444 versus $14,593 per person).4

Area-level social determinants of health (SDoH) play a critical role in shaping cognitive health.5 These include a range of conditions such as socioeconomic characteristics, access to healthcare, environmental exposures, and features of the built environment.6 Higher risk of cognitive impairment is observed among residents in areas characterized by low socioeconomic status,6–8 limited access to green space,9–14 limited access to healthy food,15 lower density of primary care physicians,16 greater exposure to air pollutants,17–20 and greater exposure to transportation noise21,22 compared to other populations. However, many of these studies have focused on a limited set of SDoH, often examining them in isolation. Advances in machine learning have facilitated the assessment of a comprehensive range of SDoH in relationship to health outcomes. One US national study using machine learning ranked SDoH associated with prevalence of dementia.23 This study showed that the percentage of adults without a high school diploma, the percentage of Black residents, and the percentage of individuals living below the poverty line were among the top ranked predictors of dementia prevalence at the county level.23

In this study, we applied machine learning to characterize how multidimensional area-level social, environmental, and climate-related determinants of health are patterned across racial groups among patients with dementia and mild cognitive impairment. This design describes structural distributions of vulnerability and does not estimate determinants of disease onset, progression, or outcomes. The findings underscore the value of incorporating multidimensional SDoH measures when characterizing structural contexts in dementia research.

2. Materials & Methods

2.1. Data

This cross-sectional retrospective study used data from OCHIN (the name is not an acronym).24,25 This is a network of community-based health centers in the US that uses the OCHIN Epic electronic health record. This network represents the largest collection of community-based health data in the country. OCHIN member health centers primarily serve patients with limited access to care.24,25 This study included adults 45 years and older who had an encounter diagnosis of dementia or mild cognitive impairment between 2016 and 2022. Patients were identified using International Classification of Diseases (ICD-10) codes of F01, F02, F03, G30, and G31.84.26

This study was determined to be Exempt Research by the Washington State University Institutional Review Board. Data were accessed through Amazon Web Services under an executed Data Use Agreement.

2.2. Measures

Race.

We created a categorical variable to classify patients as American Indian and/or Alaska Native (AIAN), Asian, Black, Native Hawaiian or Pacific Islander (NHOPI), White, or Multiracial. Data on Hispanic origin was not available.

Area-level SDoH

These factors were from the Centers for Disease Control and Prevention’s Environmental Justice Index.27,28 This index summarizes vulnerability under three broad domains of social, environmental, and climate vulnerability.27,28 Social and environmental vulnerability data were from the 2022 Environmental Justice Index, and climate vulnerability data were from the 2024 Environmental Justice Index. To derive indicators at the ZIP code level, we linked the census tract data to the US Housing and Urban Development census tract-to-zip code crosswalk data. This crosswalk provides population-based weights for census tracts and zip codes. We applied these weights to calculate weighted Environmental Justice Index values for each census tract which were aggregated from the census tract level to the zip code level and merged with the patient data.29

Social vulnerability indicators are summarized under four themes of minority status (Theme I), socioeconomic status (Theme II), household characteristics (Theme III), and housing type (Theme IV).27 Theme I included the percentage of minority populations. Theme II encompassed percentage of populations with income below 200% of federal poverty line, percentage without a high school diploma, unemployment rate, percentage of renter-occupied housing units, percentage of households burdened by housing cost, percentage of population without health insurance, and percentage of households with no internet access. Theme III included percentage of populations with a disability and with limited English language proficiency. Theme IV included percentage of population living in group quarters and in mobile homes.27

Environmental burden modules include five themes of air pollution (Theme I), potentially hazardous and toxics sites (Theme II), built environment (Theme III), transportation infrastructure (Theme IV), and water pollution (Theme V).27 Theme I included the average annual number of days with ozone exceeding the National Ambient Air Quality Standard, the number of days with fine particulate matter (PM2.5) exceeding the same national standard, and diesel PM concentrations. Theme II encompassed proportion of census tracts within a one-mile of National Priority List (NPL) sites, Toxic Release Inventory (TRI) sites, Treatment, Storage, and Disposal Facilities (TSDF), and Risk Management Plan (RMP) sites, coal mines, and lead mines. Theme III included the proportion of census tracts not located within a one-mile buffer of recreational parks, the percentage of occupied housing units built prior to 1980, and lack of walkability. Theme IV encompassed proportion of census tracts within a one-mile buffer of high-volume roads, railways, and airports. Theme V included percent of watershed area classified as impaired.27

Climate vulnerability modules comprise three themes of heat (Theme I), wildfire (Theme II), and extreme events (Theme III). Theme I included the annual mean number of extreme heat days, defined using a threshold at the 95th percentile of the temperature distribution specific to each census tract. Theme II included the average annual frequency of wildfire smoke days and the average annual area burned by wildfires. Theme III encompassed several types of extreme weather events, including annual mean number of coastal flooding occurrences, drought occurrences, riverine flooding occurrences, recorded hurricane occurrences, tornadoes, and strong wind occurrences.28

2.3. Analysis

Descriptive statistics were used to summarize patients’ and SDoH characteristics. We applied T-Distributed Stochastic Neighbor Embedding (t-SNE) to visualize SDoH by race in a two-dimensional space.30 Three t-SNE plots were generated incorporating social, environmental, and climate vulnerability indicators. These resulting plots allow visualizing how certain SDoH cluster by race. SNE is a non-linear dimensionality reduction technique, based on stochastic neighbor embedding, and designed for visualizing high-dimensional data by preserving local structures. The resulting visualization reveals district groupings or “islands” of data points,30 each representing clusters of observations that share similar in the original high-dimensional space. T-SNE was applied to the normalized SDoH data, producing two components that correspond to the x and y axes in the final visualization.

We trained supervised machine learning models using a Random Forest classifier31 to identify the most influential SDoH features characterizing vulnerability patterns across racial groups. The dataset was randomly split into training (80%) and testing (20%) subsets with stratification to preserve the distribution of race groups. The Random Forest classifier included 50 estimators and balanced class weights. Model performance was evaluated on the test set using accuracy, precision, recall, and F1 score, calculated with weighted averaging to account for class imbalance. Feature importance was assessed using the mean decrease in impurity metric inherent to the Random Forest algorithm. Additionally, a one-vs-rest classification approach was implemented to identify SDoH most relevant to each racial group. Three models were generated incorporating social, environmental, and climate vulnerability indicators. All analyses were conducted using Python 3.10 and the scikit-learn, pandas, seaborn, and matplotlib libraries.32,33

3. Results

3.1. Characteristics of the study population

The sample included 26,101 patients with dementia and mild cognitive impairment between 2016 and 2022 from 42 states and District of Columbia (Table 1). Racial minority (non-White) patients constituted 34% of the sample. The average age of the study sample was 72 (SD = 11). Most patients were female (62%) and lived in urban areas (74%). Over half of the sample (55%) had incomes at or below the federal poverty line. An additional nine percent had incomes between 101-150% federal poverty line, four percent between 151-200% federal poverty line, and eight percent above 200% federal poverty line. Data on income was missing for 24% of participants.

Table 1.

Characteristics of patients with dementia and mild cognitive impairment (n=26,101)

Characteristics N %
Age (mean (SD)) 75.25 11.35
Sex
 Female 16,156 61.9
 Male 9,945 38.1
Race
 AIAN 182 0.70
 Asian 3,655 14.00
 Black 4,867 18.65
 NHOPI 136 0.52
 White 17,130 65.63
 Multi-racial 131 0.50
Geography
 Rural 6,703 5.68
 Urban 19,398 74.32
Income: Federal poverty line
 <=100 14,346 54.96
 101-150 2,242 8.59
 151-200 999 3.83
 >200 2,168 8.31
 Unknown 6,346 24.31
Open in a new tab

Note: American Indian and/or Alaska Native (AIAN); Native Hawaiian or Pacific Islander (NHOPI)

3.2. T-SNE

The t-SNE visualizations revealed overlaps in social, environmental, and climate vulnerability profiles across racial groups (Figure 1). While indicators were broadly intermixed in the two-dimensional space, some local clustering by race is observed. Examination of individual SDoH indicators (Table 2; Appendix A) reveals that specific vulnerabilities are more pronounced in certain groups. AIAN participants had one social indicator with the highest mean vulnerability. Asian participants had the highest mean vulnerability for three social indicators, nine environmental indicators, and one climate indicator. Black participants had the highest mean vulnerability for five social indicators, one environmental indicator, and three climate indicators. NHOPI participants had the highest mean vulnerability for three social indicators, two environmental indicators, and one climate indicator. White participants had one social, one environmental, and two climate indicators with the highest mean vulnerability. Multiracial participants had two environmental and two climate indicators with the highest mean vulnerability. These patterns suggest that elevated vulnerabilities are distributed differently across racial and ethnic groups and domains, with each group showing higher mean vulnerability in specific areas.

Figure 1.

Figure 1.

Open in a new tab

Visualization of social, environmental, and climate vulnerability indicators by race using the t-distributed stochastic neighbor embedding (T-NSE).

Note: AIAN: American Indian and/or Alaska Native; NHOPI: Native Hawaiian or Other Pacific Islander

Table 2.

Summary statistics (mean and standard deviation) of social, environmental, and climate vulnerability indicators by thematic domain and race/ethnicity.

Indicator Theme Indicator AIAN Asian Black Multiracial NHOPI White
Social vulnerability (12 variables) Minority status (%) Racial and ethnic minority 23.3 (17.1) 38.1 (16.8) 35.4 (18.3) 28.2 (19.6) 35.3 (18.2) 20.5 (17.9)
Socioeconomic Status (%) Income below 200% FPL 17.9 (8.5) 17.3 (8.1) 22.4 (10.5) 20.9 (10.4) 20.8 (8.8) 19.5 (10.1)
No high school diploma 7.3 (5.5) 8.8 (5.5) 9.3 (5.7) 8.8 (6.7) 9.7 (5.3) 7.9 (7.2)
Unemployed 3.2 (1.5) 3.4 (1.4) 4.5 (2.5) 3.9 (2.3) 4.1 (2.0) 3.4 (2.0)
Rented occupied housing 23.2 (11.6) 32.0 (13.1) 27.9 (13.1) 27.2 (13.2) 29.0 (12.3) 22.4 (12.5)
Housing cost-burdened households 17.6 (6.9) 18.8 (6.7) 20.3 (8.5) 20.3 (8.0) 21.4 (7.3) 17.9 (8.2)
Lack of health insurance 3.8 (2.3) 3.6 (2.2) 4.6 (3.4) 4.5 (3.1) 4.9 (3.7) 4.3 (3.0)
Lack of broadband access 8.9 (4.7) 8.6 (4.3) 12.3 (6.6) 9.6 (4.9) 9.8 (4.6) 9.8 (5.3)
Household Characteristics (%) Disability 7.3 (2.7) 7.1 (2.3) 8.2 (3.1) 8.2 (3.3) 7.8 (2.9) 8.1 (3.8)
Speak English “less than well” 3.3 (3.7) 6.4 (4.4) 3.5 (3.9) 3.6 (3.9) 4.7 (3.4) 2.9 (4.4)
Housing Type (%) Living in group quarters 1.1 (1.5) 1.1 (1.4) 1.6 (2.0) 1.4 (1.9) 1.6 (2.6) 1.3 (1.9)
Living in mobile homes 2.5 (3.1) 0.9 (1.5) 2.2 (4.0) 2.7 (3.7) 2.0 (3.0) 3.6 (4.2)
Environmental vulnerability (16 variables) Air Pollution (n) Days ozone over NAAQS 1.3 (4.6) 0.6 (2.7) 1.3 (4.6) 2.0 (6.0) 1.9 (5.0) 1.7 (6.0)
Days PM2.5 over NAAQS 4.4 (1.7) 5.3 (1.5) 5.0 (1.7) 5.1 (2.0) 5.1 (1.6) 4.5 (1.9)
Diesel particulate matter (μg/m3) 0.3 (0.2) 0.4 (0.3) 0.3 (0.2) 0.3 (0.2) 0.3 (0.2) 0.2 (0.2)
Potentially Hazardous and Toxic Sites (% area within 1-mile) NPL sites 1.2 (3.6) 3.0 (9.3) 1.5 (5.1) 0.9 (3.4) 1.8 (7.4) 0.9 (3.7)
TRI sites 24.3 (20.4) 33.5 (22.2) 30.6 (22.7) 26.9 (22.4) 27.9 (22.2) 18.9 (19.9)
TSDF 1.8 (5.7) 1.5 (5.1) 2.7 (7.8) 1.7 (7.6) 1.4 (4.1) 1.3 (5.0)
RMP sites 8.6 (11.3) 10.3 (13.0) 9.7 (13.3) 8.8 (13.2) 11.0 (12.8) 7.1 (11.8)
Coal mines 0.001 (0.01) 0.001 (0.001) 0.001 (0.001) 0.001 (0.001) 0.001 (0.001) 0.001 (0.1)
Lead mines 0.1 (0.8) 0.004 (0.2) 0.02 (0.8) 0.004 (0.03) 0.07 (0.6) 0.2 (1.3)
Built Environment Lack of parks (% area not within 1-mile) 41.8 (22.1) 59.5 (15.5) 46.7 (24.6) 48.2 (22.3) 53.5 (19.5) 38.5 (23.5)
Houses built pre-1980 (%) 33.4 (15.8) 42.2 (15.8) 40.2 (17.7) 37.9 (17.3) 38.9 (16.6) 34.1 (16.9)
Lack of walkability 6.4 (2.9) 8.7 (2.4) 6.8 (3.1) 7.4 (2.8) 7.6 (2.5) 6.1 (3.0)
Transportation Infrastructure (% area within 1-mile) High-volume roads 28.3 (19.7) 44.2 (18.6) 32.7 (22.3) 32.2 (20.3) 31.9 (21.0) 26.9 (20.8)
Railways 2.5 (5.7) 2.9 (8.2) 2.0 (6.3) 3.8 (7.2) 4.1 (7.3) 3.5 (6.8)
Airports 2.50 (5.71) 2.90 (8.25) 1.99 (6.33) 4.08 (7.25) 3.47 (6.81) 3.83 (7.18)
Water Pollution Impaired surface water (% areas) 31.1 (19.3) 30.5 (18.5) 33.0 (23.7) 36.4 (21.6) 33.3 (22.3) 31.1 (22.3)
Climate vulnerability (9 variables) Heat Extreme heat days (day/year) 8.2 (2.9) 8.1 (2.3) 7.1 (3.2) 9.1 (3.1) 9.0 (2.9) 8.5 (3.7)
Wildfire Wildfire smoke days (day/year) 8.1 (4.5) 8.5 (2.8) 5.6 (3.5) 9.6 (4.8) 7.8 (4.4) 8.8 (5.3)
Burned area from wildfires (area/year) 0.02 (0.1) 0.01 (0.1) 0.002 (0.03) 0.03 (0.1) 0.01 (0.04) 0.04 (0.2)
Extreme Events (n/year) Coastal flooding occurrence 0.3 (0.6) 0.1 (0.4) 0.2 (0.6) 0.3 (0.6) 0.3 (0.6) 0.4 (0.9)
Drought occurrences 10.2 (14.9) 20.0 (15.2) 9.5 (13.0) 14.9 (16.2) 15.6 (16.4) 13.5 (15.5)
Riverine flooding occurrences 0.5 (0.7) 0.6 (0.7) 0.7 (0.8) 0.6 (0.7) 0.8 (0.9) 0.6 (0.8)
Hurricane occurrences 0.02 (0.04) 0.01 (0.03) 0.05 (0.06) 0.02 (0.04) 0.04 (0.05) 0.02 (0.04)
Tornado occurrences 0.002 (0.01) 0.001 (0.002) 0.01 (0.02) 0.001 (0.003) 0.002 (0.01) 0.004 (0.01)
Strong wind occurrence 0.4 (0.9) 0.2 (0.5) 0.9 (1.1) 0.4 (0.8) 0.6 (1.0) 0.5 (0.9)
Open in a new tab

Note: American Indian and/or Alaska Native (AIAN); Native Hawaiian or Pacific Islander (NHOPI); Federal poverty level (FPL); National Ambient Air Quality Standards (NAAQS); National priority list (NPL); Toxic release inventory (TRI); Treatment, storage, and disposal facilities (TSDF); Risk management plan (RMP)

Detailed examination across social vulnerability indicators showed that Asian participants lived in ZIP codes with higher rates of minority populations, higher percentage of renter-occupied housing units, and higher rates of people with limited English proficiency. Black participants lived in areas with higher rates of individuals with income below the federal poverty line, higher unemployment rate, higher percentage of households with broadband, higher percentage of people with disability, and higher percentage of population living in group quarters. NHOPI participants lived in ZIP codes with higher percentage of individuals without a high school diploma, higher percentage of households burdened by housing costs, and higher percentage of population without health insurance. White participants lived in ZIP codes with higher percentage of population living in mobile homes.

Environmental vulnerability indicators showed that Asian participants lived in ZIP codes with higher average annual number of days with PM2.5 concentrations exceeding national standards, higher diesel PM concentrations, and greater land area within one mile of NPL sites, TRI sites, high volume roads, and railways. They also lived in areas not accessible to recreational parks and with lower walkability. Black participants resided in ZIP codes with greater land area located within one mile of TSDF. NHOPI participants lived in ZIP codes with greater land area within one mile of RMP sites and airports. White participants resided in areas with greater land area within a one mile of lead mines. Multiracial participants lived in areas with a higher average annual number of days with ozone levels exceeding national standards and impaired watershed.

Climate vulnerability indicators showed Asian participants lived in ZIP codes with greater exposure to drought occurrences. Black participants lived in areas with greater exposure to hurricanes, tornado, and wind occurrences. NHOPI participants lived in areas with greater risk of riverine flooding occurrences. White participants lived in ZIP codes with greater exposure to burned areas due to wildfires and higher frequency of coastal flooding occurrences. Multiracial individuals lived in ZIP codes with a greater number of extreme heat days and smoky days resulting from wildfire.

3.3. Random Forest results

Random Forest classification models demonstrated modest overall performance in distinguishing social, environmental, and climate vulnerability profiles, with a balanced accuracy of 0.35 across all models. Macro-averaged precision, recall, and F1 scores were 0.34, 0.35, and 0.33, respectively, reflecting variability in model performance across racial groups (Table 3). Per-class performance was highest for Asian, Black, and White participants, with F1 scores ranging from 0.56–0.73 across vulnerability domains, and lowest for AIAN, NHOPI, and Multiracial populations, for whom both precision and F1 scores were below 0.04. To account for class imbalance, we compared our Random Forest model to a majority-class baseline. The baseline achieved a balanced accuracy of 0.17 and a macro F1 of 0.13. The Random Forest model improved these imbalance-aware metrics (balanced accuracy = 0.35, macro F1 = 0.33), indicating that it captures meaningful patterns across both majority and minority groups.

Table 3.

Classification performance metrics for social, environmental, and climate vulnerability indicators by race.

Vulnerability | Per class performance Social Environmental Climate
precision recall f1-score precision recall f1-score precision recall f1-score
AIAN 0.020 0.139 0.035 0.020 0.139 0.035 0.020 0.139 0.035
Asian 0.554 0.673 0.608 0.553 0.670 0.606 0.555 0.674 0.609
Black 0.538 0.602 0.568 0.536 0.589 0.561 0.534 0.601 0.565
Multiracial 0.003 0.038 0.006 0.003 0.038 0.006 0.003 0.038 0.006
NHOPI 0.003 0.037 0.006 0.003 0.037 0.006 0.003 0.037 0.006
White 0.904 0.618 0.734 0.899 0.621 0.734 0.905 0.617 0.734
Open in a new tab

Note: American Indian and/or Alaska Native (AIAN); Native Hawaiian or Pacific Islander (NHOPI)

Table 4 shows the relative importance of SDoH by race. We defined common SDoH factors as variables that appeared among the top three most important features across multiple racial groups. Distinct factors were those that ranked among the top three features for only one or two racial groups. Among the social vulnerability variables, the percentage of minority populations was among the top three features in ZIP codes where Asian, Black, NHOPI, White, and Multiracial patients lived. Limited English proficiency was a prominent feature in areas where AI/AN, Asian, Black, NHOPI, and White patients resided. The percentage of individuals living in mobile homes was among the top three features in areas where Asian, Black, and White patients resided. For AI/AN patients, the percentage of the population with a disability and unemployment rate were additional key ZIP code characteristics. For NHOPI patients, the percentage of the population without a high school diploma was an additional key feature. Among Multiracial patients, ZIP code-level percentages of households burdened by housing costs and without internet access were additional key ZIP code features.

Table 4.

Feature importance of social, environmental, and climate vulnerability indicators by race.

Top AIAN Asian Black Multiracial NHOPI White
Social Vulnerability
Racial and ethnic minority (%) 0.081 0.141 0.189 0.090 0.112 0.268
Living in mobile homes (%) 0.085 0.109 0.122 0.075 0.080 0.142
Speak English “less than well” (%) 0.090 0.228 0.098 0.087 0.148 0.101
With a disability (%) 0.094 0.057 0.072 0.085 0.070 0.069
No high school diploma (%) 0.069 0.046 0.077 0.087 0.090 0.066
Lack of health insurance (%) 0.081 0.067 0.057 0.089 0.067 0.066
Rented occupied housing units (%) 0.081 0.068 0.066 0.080 0.061 0.063
Housing burdened low-income households (%) 0.084 0.046 0.063 0.096 0.088 0.049
Income below 200% of federal poverty level (%) 0.089 0.084 0.063 0.076 0.073 0.047
Lack of broadband access (%) 0.079 0.063 0.079 0.092 0.074 0.047
Unemployed (%) 0.091 0.058 0.065 0.073 0.061 0.046
Living in group quarters (%) 0.076 0.033 0.049 0.069 0.077 0.036
Environmental Vulnerability
Diesel particulate matter (μg/m3) 0.083 0.193 0.158 0.091 0.114 0.224
Impaired surface water (% areas) 0.094 0.086 0.118 0.111 0.078 0.095
Lack of parks (% area not within 1-mile) 0.078 0.161 0.073 0.077 0.091 0.095
Toxic release inventory sites (% area within 1-mile) 0.075 0.055 0.079 0.065 0.066 0.084
Days PM2.5 over the national standard (n) 0.092 0.060 0.065 0.086 0.085 0.066
Days ozone over the national standard (n) 0.060 0.085 0.090 0.045 0.100 0.062
Lack of walkability 0.077 0.094 0.077 0.089 0.078 0.059
High-volume roads (% area within 1-mile) 0.097 0.072 0.055 0.080 0.084 0.057
Railways (% area within 1-mile) 0.073 0.039 0.062 0.084 0.061 0.055
Airports (% area within 1-mile) 0.051 0.030 0.048 0.051 0.047 0.054
Risk management plan sites (% area within 1-mile) 0.065 0.037 0.056 0.069 0.065 0.051
Houses built pre-1980 (%) 0.079 0.048 0.057 0.091 0.064 0.050
Treatment, storage, and disposal facilities (% area within 1-mile) 0.043 0.021 0.039 0.033 0.040 0.027
National priority list sites (% area within 1-mile) 0.029 0.019 0.021 0.025 0.028 0.020
Lead mines (% area within 1-mile) 0.001 0.001 0.003 0.003 0.001 0.001
Coal mines (% area within 1-mile) 0.001 0.000 0.000 0.000 0.000 0.000
Climate Vulnerability
Annual frequency of strong wind occurrences 0.184 0.273 0.278 0.148 0.137 0.154
Wildfire smoke days (day/year) 0.141 0.121 0.134 0.189 0.160 0.148
Annual frequency of drought occurrences 0.135 0.119 0.101 0.118 0.116 0.128
Burned area from wildfires (area/year) 0.033 0.051 0.067 0.039 0.042 0.120
Extreme heat days (day/year) 0.151 0.101 0.091 0.161 0.150 0.109
Annual frequency of flooding occurrences 0.147 0.094 0.093 0.153 0.150 0.102
Annual frequency of coastal flooding occurrence 0.073 0.075 0.049 0.066 0.067 0.086
Annual frequency of hurricane occurrences 0.070 0.044 0.134 0.041 0.074 0.083
Annual frequency of tornadoes 0.065 0.117 0.049 0.080 0.100 0.065
Open in a new tab

Note: American Indian and/or Alaska Native (AIAN); Native Hawaiian or Pacific Islander (NHOPI)

Top three features for each race are highlighted in gradients of orange.

Among the environmental vulnerability variables, exposure to PM2.5 (for AI/AN patients) or diesel PM (for Asian, Black, Multiracial, NHOPI, and White patients) emerged as key features for patients. Ozone was notably important in areas where Black and NHOPI patients lived. The proportion of impaired watershed was a key feature in ZIP codes where AI/AN, Black, White, and Multiracial patients resided. The proportion of ZIP codes not located within a one-mile buffer of recreational parks was an important feature for Asian, NHOPI, and White patients. For AI/AN patients, the proportion of high-volume roads was an additional key factor. Lack of walkability was an important attribute for Asian patients and the percentage of occupied housing units built prior to 1980 emerged as a relevant feature in areas where Multiracial patients resided.

Among the climate vulnerability variables, strong wind occurrences emerged as a key feature in areas where AI/AN, Asian, Black, and White patients lived. Smoky days resulting from wildfire smoke were important across all areas except in ZIP codes where AI/AN patients lived. The frequency of extreme heat days and riverine flooding occurrences were key features in ZIP codes where AI/AN, NHOPI, and Multiracial patients resided. Drought occurrences were prominent in ZIP codes of Asian and White patients. Hurricane occurrences were notably important in areas where Black patients lived.

3.4. Sensitivity Analysis

We conducted a sensitivity analysis excluding percentage of minority populations (Appendix B). Overall model performance (balanced accuracy, macro F1) was largely unchanged, indicating that predictions are not solely driven by racial composition. Feature importance rankings shifted, with indicators such as the percentage of renter-occupied housing units, populations with limited broadband access, unemployment rate, and populations without a high school diploma emerging as key drivers across racial groups.

4. Discussion

This study ranked area-level SDoH among patients with dementia and mild cognitive impairment, revealing both common and distinct patterns of vulnerability across racial groups. Common factors included percentage of minority populations, percentage of people with limited English proficiency, exposure to diesel PM or PM2.5, impaired watershed conditions, and frequency of smoke and heat days. District factors included unemployment and disability rates in areas where AI/AN patients lived; lack of walkability in areas where Asian patients resided; hurricane occurrences in areas where Black patients lived; educational attainment in areas where NHOPI patients lived; and housing burden, lack of internet access, and housing age in area where Multiracial people resided.

The percentage of minority populations and those with limited English proficiency emerged as the top ranked common indicators of social vulnerability. Living in areas with a higher percentage of minority populations often correlates with low socioeconomic status, low insurance coverage, limited healthcare access, and fewer community resources.34 Previous research has linked residence in low socioeconomic status areas to brain atrophy and low cognitive performance.6–8 Furthermore, higher odds of cognitive decline have been observed in aeras with lower density of primary care physician.16 Language barriers may further restrict access to healthcare resources and lead to issues such as incomplete understanding of patients’ condition and prescribed treatment, misdiagnosis, and delayed treatment.35,36

Exposure to diesel PM or PM2.5 and impaired watershed were among the top ranked common indicators of environmental vulnerability. This is concerning given that the Lancet Commission identified air pollution as a risk factor for dementia.37 Studies from England, Mexico, Sweden, Taiwan, and US have consistently linked higher exposure to air pollution with increased risk of dementia and mild cognitive impairment.17–20,38 Impaired watershed is another critical indicator of environmental health, shaping water quality and soil contamination in surrounding areas.39 A review of studies on drinking water composition found that exposure to aluminum increased the risk of cognitive decline.40

Among climate vulnerability indicators, frequency of smoky days resulting from wildfire smoke and heat days emerged as key common features. Similar to exposure to diesel PM or PM2.5, wildfire smoke has been linked with increased risk of dementia incidence.41 People with dementia are vulnerable to extreme heat42 as they may have difficulty recognizing symptoms of extreme heat exposure or responding appropriately once the symptoms are identified.43,44 Not surprisingly, exposure to extreme heat has been associated with increased risk of hospitalization among patients with dementia.45

Among the distinct vulnerability indicators, Asian people lived in areas with lower walkability despite evidence linking walkability with reduced risk of cognitive impairment.46 Hurricane occurrence emerged as a key factor among Black patients, aligning with research showing that individuals with dementia have increased mortality risk following natural disasters such as hurricanes.47 AIAN, NHOPI, and Multiracial people lived in areas marked by socioeconomic disparities. In 2020, 14% of the minority populations in the United States lived in low socioeconomic status areas compared to four percent of the non-Hispanic White people.48 This is concerning given that lower socioeconomic status areas often lack health-promoting resources49–51 and face unequal burden of disease and mortality.7,52

Taken together, although social, environmental, and climate vulnerability profiles overlapped across racial groups, Asian participants exhibited the most pronounced environmental vulnerabilities, Black participants had the highest social vulnerabilities, and other groups showed smaller but domain-specific vulnerabilities across domains. These common and distinct vulnerabilities highlight the need for public health strategies that combine broad structural interventions with targeted approaches responsive to specific contextual patterns. Policies aimed at reducing socioeconomic, environmental, and climate stressors at the community level may benefit the broader population, while more tailored efforts can address distinct structural conditions affecting particular groups. Such balanced approaches may promote more equitable and healthier communities for all people.53

Certain limitations in the context of this study should be acknowledged. First, SDoH were assessed using ZIP code, which represents relatively large geographic areas that may contain substantial socioeconomic and environmental heterogeneity. As a result, this level of analysis may obscure within-area variation and potentially attenuate or misrepresent vulnerability patterns. Analyses conducted at smaller geographic scales, such as census tracts or block groups, might capture more localized contextual differences and yield different patterns of vulnerability. Additionally, given that the measures reflect area-level conditions, the findings describe structural contexts rather than individual-level exposures or lived experiences and should not be interpreted as directly representing individual risk or experience. Race-specific vulnerability profiles should therefore be interpreted as reflecting differences in the distribution of contextual conditions across groups within this sample, rather than implying that individual patients directly experienced or were uniformly exposed to these conditions.

Second, model performance differed substantially across racial groups. ML models performed well for Asian, Black, and White participants but demonstrated poor accuracy for AIAN, NHOPI, and Multiracial participants. This likely reflects class imbalance, small sample size for these groups, and the high dimensionality of SDoH indicators relative to available observations. Although we explored synthetic data generation techniques to oversample underrepresented racial groups, the application of these methods did not meaningfully improve model performance for these groups, suggesting SDoH patterns across groups or limited discriminatory signal for smaller classes. Importantly, in the presence of class imbalance, overall accuracy can obscure poor performance in minority groups. Therefore, balanced accuracy and macro-averaged precision, recall, and F1 scores were reported to provide a more appropriate summary of model performance. Given the low classification performance for smaller racial groups, feature importance for these groups should be interpreted cautiously, as rankings may be unstable when classification accuracy is low. Future work may benefit from alternative strategies, including group-balanced sampling frameworks, dimensionality reduction prior to classification, ensemble approaches, or reframing analyses toward descriptive group-level comparisons rather than predictive classification for smaller populations. Third, our study was conducted using data from the OCHIN member health centers, so our findings may not generalize to other populations in the US.

In conclusion, this study characterized structural patterns of vulnerability among patients with dementia and mild cognitive impairment, revealing both common and group-specific patterns of social, environmental, and climate-related vulnerability by race. These findings highlight the need to move beyond a narrow set of indicators and integrate a comprehensive range of SDoH into dementia research. This will improve our understanding of the multifaceted drivers of cognitive decline and inform the development of targeted, equity-focused aging initiatives. Future research should examine how combinations of SDoH shape dementia onset, progression, and outcomes.

Supplementary Material

Appendix

Funding source:

Research reported in this publication was supported by the Office of the Director, National Institutes of Health Common Fund under award number 1OT2OD032581. The research reported in this work was powered by PCORnet®. PCORnet has been developed with funding from the Patient-Centered Outcomes Research Institute® (PCORI®) and conducted with the Accelerating Data Value Across a National Community Health Center Network (ADVANCE) Clinical Research Network (CRN). ADVANCE is a Clinical Research Network in PCORnet® led by OCHIN in partnership with Health Choice Network, Fenway Health, University of Washington, and Oregon Health & Science University. ADVANCE’s participation in PCORnet® is funded through the PCORI Award RI-OCHIN-01-MC.

Funding statement

The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of funding agencies.

Footnotes

Conflicts of Interest: The authors declare they have no conflicts of interest related to this work to disclose

Declaration of generative AI in scientific writing: No AI or AI-assisted technologies was used in the scientific writing process.

Contributor Information

Solmaz Amiri, Elson S Floyd College of Medicine, Washington State University, USA.

Wyatt P. Bensken, OCHIN, Inc., Department of Population and Quantitative Health Sciences, School of Medicine, Case Western Reserve University, USA.

Dedra Buchwald, Department of Neurological Surgery, University of Washington, USA.

Data availability:

The authors do not have permission to share data.

References

  • 1.Rajan KB, Weuve J, Barnes LL, McAninch EA, Wilson RS, Evans DA. Population estimate of people with clinical Alzheimer’s disease and mild cognitive impairment in the United States (2020–2060). Alzheimer’s & dementia. 2021;17(12):1966–1975. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Manly JJ, Jones RN, Langa KM, et al. Estimating the Prevalence of Dementia and Mild Cognitive Impairment in the US: The 2016 Health and Retirement Study Harmonized Cognitive Assessment Protocol Project. JAMA Neurology. 2022;79(12):1242–1249. doi: 10.1001/jamaneurol.2022.3543 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Albert MS, DeKosky ST, Dickson D, et al. The diagnosis of mild cognitive impairment due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. May 2011;7(3):270–9. doi: 10.1016/j.jalz.2011.03.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.2023 Alzheimer’s disease facts and figures. Alzheimers Dement. Apr 2023;19(4):1598–1695. doi: 10.1002/alz.13016 [DOI] [PubMed] [Google Scholar]
  • 5.Alderwick H, Gottlieb LM. Meanings and misunderstandings: a social determinants of health lexicon for health care systems. The Milbank Quarterly. 2019;97(2):407. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Zuelsdorff M, Larson JL, Hunt JFV, et al. The Area Deprivation Index: A novel tool for harmonizable risk assessment in Alzheimer’s disease research. Alzheimers Dement (N Y). 2020;6(1):e12039. doi: 10.1002/trc2.12039 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Hunt JFV, Buckingham W, Kim AJ, et al. Association of Neighborhood-Level Disadvantage With Cerebral and Hippocampal Volume. JAMA Neurology. 2020;77(4):451–460. doi: 10.1001/jamaneurol.2019.4501 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Hunt JFV, Vogt NM, Jonaitis EM, et al. Association of Neighborhood Context, Cognitive Decline, and Cortical Change in an Unimpaired Cohort. Neurology. 2021;96(20):e2500. doi: 10.1212/WNL.0000000000011918 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Klompmaker JO, Laden F, Browning MHEM, et al. Associations of Greenness, Parks, and Blue Space With Neurodegenerative Disease Hospitalizations Among Older US Adults. JAMA Network Open. 2022;5(12):e2247664–e2247664. doi: 10.1001/jamanetworkopen.2022.47664 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Besser LM, Lovasi GS, Michael YL, et al. Associations between neighborhood greenspace and brain imaging measures in non-demented older adults: the Cardiovascular Health Study. Social Psychiatry and Psychiatric Epidemiology. 2021/09/01 2021;56(9):1575–1585. doi: 10.1007/s00127-020-02000-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Besser LM, Hirsch J, Galvin JE, et al. Associations between neighborhood park space and cognition in older adults vary by US location: The Multi-Ethnic Study of Atherosclerosis. Health & Place. 2020/11/01/ 2020;66:102459. doi: 10.1016/j.healthplace.2020.102459 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.De Keijzer C, Tonne C, Basagaña X, et al. Residential surrounding greenness and cognitive decline: a 10-year follow-up of the Whitehall II cohort. Environmental health perspectives. 2018;126(7):077003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Zhu A, Yan L, Shu C, Zeng Y, Ji JS. APOE ε4 Modifies Effect of Residential Greenness on Cognitive Function among Older Adults: A Longitudinal Analysis in China. Scientific Reports. 2020/01/09 2020;10(1):82. doi: 10.1038/s41598-019-57082-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.de Keijzer C, Gascon M, Nieuwenhuijsen MJ, Dadvand P. Long-Term Green Space Exposure and Cognition Across the Life Course: a Systematic Review. Current Environmental Health Reports. 2016/12/01 2016;3(4):468–477. doi: 10.1007/s40572-016-0116-x [DOI] [PubMed] [Google Scholar]
  • 15.Tani Y, Suzuki N, Fujiwara T, Hanazato M, Kondo K. Neighborhood Food Environment and Dementia Incidence: the Japan Gerontological Evaluation Study Cohort Survey. Am J Prev Med. Mar 2019;56(3):383–392. doi: 10.1016/j.amepre.2018.10.028 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Mullins MA, Bynum JPW, Judd SE, Clarke PJ. Access to primary care and cognitive impairment: results from a national community study of aging Americans. BMC Geriatrics. 2021/10/20 2021;21(1):580. doi: 10.1186/s12877-021-02545-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Wu YC, Lin YC, Yu HL, et al. Association between air pollutants and dementia risk in the elderly. Alzheimers Dement (Amst). Jun 2015;1(2):220–8. doi: 10.1016/j.dadm.2014.11.015 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Oudin A, Forsberg B, Adolfsson AN, et al. Traffic-Related Air Pollution and Dementia Incidence in Northern Sweden: A Longitudinal Study. Environ Health Perspect. Mar 2016;124(3):306–12. doi: 10.1289/ehp.1408322 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Chen H, Kwong JC, Copes R, et al. Exposure to ambient air pollution and the incidence of dementia: A population-based cohort study. Environ Int. Nov 2017;108:271–277. doi: 10.1016/j.envint.2017.08.020 [DOI] [PubMed] [Google Scholar]
  • 20.Peters R, Ee N, Peters J, Booth A, Mudway I, Anstey KJ. Air Pollution and Dementia: A Systematic Review. J Alzheimers Dis. 2019;70(s1):S145–s163. doi: 10.3233/jad-180631 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.van Kempen E, van Kamp I, Lebret E, Lammers J, Emmen H, Stansfeld S. Neurobehavioral effects of transportation noise in primary schoolchildren: a cross-sectional study. Environmental Health. 2010/06/01 2010;9(1):25. doi: 10.1186/1476-069X-9-25 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Jafari MJ, Khosrowabadi R, Khodakarim S, Mohammadian F. The Effect of Noise Exposure on Cognitive Performance and Brain Activity Patterns. Open Access Maced J Med Sci. 2019;7(17):2924–2931. doi: 10.3889/oamjms.2019.742 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Ciciora D, Vásquez E, Valachovic E, et al. Social and Behavior Factors of Alzheimer’s Disease and Related Dementias: A National Study in the U.S. American Journal of Preventive Medicine. 2024;66(4):573–581. doi: 10.1016/j.amepre.2023.11.017 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Bensken WP, McGrath BM, Gold R, Cottrell EK. Area-level social determinants of health and individual-level social risks: Assessing predictive ability and biases in social risk screening. J Clin Transl Sci. 2023;7(1):e257. doi: 10.1017/cts.2023.680 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Bensken WP, Navale SM, McGrath BM, et al. Variation in multimorbidity by sociodemographics and social drivers of health among patients seen at community-based health centers. J Multimorb Comorb. Jan-Dec 2024;14:26335565241236410. doi: 10.1177/26335565241236410 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Duan R, Chen Z, Tong J, et al. Leverage Real-world Longitudinal Data in Large Clinical Research Networks for Alzheimer’s Disease and Related Dementia (ADRD). AMIA Annu Symp Proc. 2020;2020:393–401. [PMC free article] [PubMed] [Google Scholar]
  • 27.CDC. Technical Documentation for the Environmental Justice Index. Centers for Disease Control and Prevention. https://www.atsdr.cdc.gov/placeandhealth/eji/technical_documentation.html [Google Scholar]
  • 28.CDC. Technical documentation for the 2024 environmental justice index. https://www.atsdr.cdc.gov/place-health/media/pdfs/2024/10/EJI_2024_Technical_Documentation.pdf [Google Scholar]
  • 29.Gibbons RD, Olfson M, Saulsberry L, et al. Social Vulnerability and Prevalence and Treatment for Mental Health and Substance Use Disorders. JAMA Psychiatry. Oct 1 2024;81(10):976–984. doi: 10.1001/jamapsychiatry.2024.1870 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Van der Maaten L, Hinton G. Visualizing data using t-SNE. Journal of machine learning research. 2008;9(11) [Google Scholar]
  • 31.Rigatti SJ. Random forest. Journal of Insurance Medicine. 2017;47(1):31–39. [DOI] [PubMed] [Google Scholar]
  • 32.Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: Machine learning in Python. the Journal of machine Learning research. 2011;12:2825–2830. [Google Scholar]
  • 33.Oliphant TE. Python for scientific computing. Computing in science & engineering. 2007;9(3):10–20. [Google Scholar]
  • 34.Williams DR, Mohammed SA, Leavell J, Collins C. Race, socioeconomic status, and health: complexities, ongoing challenges, and research opportunities. Ann N Y Acad Sci. Feb 2010;1186:69–101. doi: 10.1111/j.1749-6632.2009.05339.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Al Shamsi H, Almutairi AG, Al Mashrafi S, Al Kalbani T. Implications of Language Barriers for Healthcare: A Systematic Review. Oman Med J. Mar 2020;35(2):e122. doi: 10.5001/omj.2020.40 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.de Moissac D, Bowen S. Impact of Language Barriers on Quality of Care and Patient Safety for Official Language Minority Francophones in Canada. Journal of Patient Experience. 2019/03/01 2018;6(1):24–32. doi: 10.1177/2374373518769008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Livingston G, Huntley J, Sommerlad A, et al. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. The Lancet. 2020;396(10248):413–446. doi: 10.1016/S0140-6736(20)30367-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Oudin A Short review: Air pollution, noise and lack of greenness as risk factors for Alzheimer’s disease-epidemiologic and experimental evidence. Neurochem Int. Mar 2020;134:104646. doi: 10.1016/j.neuint.2019.104646 [DOI] [PubMed] [Google Scholar]
  • 39.Kolok AS, Beseler CL, Chen XH, Shea PJ. The watershed as a conceptual framework for the study of environmental and human health. Environ Health Insights. Feb 18 2009;3:1–10. doi: 10.4137/ehi.s1925 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Wasick A, Kim Y. Association between the Composition of Drinking Water and Cognitive Function in the Elderly: A Systematic Review. Int J Environ Res Public Health. Mar 19 2024;21(3)doi: 10.3390/ijerph21030362 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Elser H, Frankland TB, Chen C, et al. Wildfire Smoke Exposure and Incident Dementia. JAMA Neurology. 2025;82(1):40–48. doi: 10.1001/jamaneurol.2024.4058 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Sisodiya SM, Gulcebi MI, Fortunato F, et al. Climate change and disorders of the nervous system. Lancet Neurol. Jun 2024;23(6):636–648. doi: 10.1016/s1474-4422(24)00087-5 [DOI] [PubMed] [Google Scholar]
  • 43.Van Hoof J, Kort HSM, Hensen JLM, Duijnstee MSH, Rutten PGS. Thermal comfort and the integrated design of homes for older people with dementia. Building and environment. 2010;45(2):358–370. [Google Scholar]
  • 44.Jung CG, Kato R, Zhou C, et al. Sustained high body temperature exacerbates cognitive function and Alzheimer’s disease-related pathologies. Sci Rep. Jul 18 2022;12(1):12273. doi: 10.1038/s41598-022-16626-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.Delaney SW, Stegmuller A, Mork D, et al. Extreme Heat and Hospitalization Among Older Persons With Alzheimer Disease and Related Dementias. JAMA Internal Medicine. 2025;185(4):412–421. doi: 10.1001/jamainternmed.2024.7719 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Da L, Song X, Jia Z, et al. Objectively measured environmental features and their association with cognition and dementia: A systematic review and meta-analysis. Ageing Research Reviews. 2025/02/01/ 2025;104:102630. doi: 10.1016/j.arr.2024.102630 [DOI] [PubMed] [Google Scholar]
  • 47.Bell SA, Miranda ML, Bynum JPW, Davis MA. Mortality After Exposure to a Hurricane Among Older Adults Living With Dementia. JAMA Network Open. 2023;6(3):e232043–e232043. doi: 10.1001/jamanetworkopen.2023.2043 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Atlas NE. Neighborhood poverty: All neighborhoods should be communities of opportunity. Accessed May 17, 2024. https://nationalequityatlas.org/indicators/Neighborhood_poverty
  • 49.Wang MC, Kim S, Gonzalez AA, MacLeod KE, Winkleby MA. Socioeconomic and food-related physical characteristics of the neighbourhood environment are associated with body mass index. J Epidemiol Community Health. Jun 2007;61(6):491–8. doi: 10.1136/jech.2006.051680 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Amiri S, Espenschied JR, Roll JM, Amram O. Access to Primary Care Physicians and Mortality in Washington State: Application of a 2-Step Floating Catchment Area. J Rural Health. Dec 15 2019;doi: 10.1111/jrh.12402 [DOI] [PubMed] [Google Scholar]
  • 51.Nieuwenhuis J, Kleinepier T, van Ham M. The Role of Exposure to Neighborhood and School Poverty in Understanding Educational Attainment. J Youth Adolesc. May 2021;50(5):872–892. doi: 10.1007/s10964-021-01427-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Powell WR, Buckingham WR, Larson JL, et al. Association of Neighborhood-Level Disadvantage With Alzheimer Disease Neuropathology. JAMA Network Open. 2020;3(6):e207559–e207559. doi: 10.1001/jamanetworkopen.2020.7559 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Arcaya MC, Ellen IG, Steil J. Neighborhoods And Health: Interventions At The Neighborhood Level Could Help Advance Health Equity. Health Aff (Millwood). Feb 2024;43(2):156–163. doi: 10.1377/hlthaff.2023.01037 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Appendix
NIHMS2179215-supplement-Appendix.docx (757.3KB, docx)

Data Availability Statement

The authors do not have permission to share data.

RESOURCES