Dementia, dementia's risk factors and premorbid brain structure are concentrated in disadvantaged areas: National register and birth‐cohort geographic analyses

Abstract INTRODUCTION Dementia risk may be elevated in socioeconomically disadvantaged neighborhoods. Reasons for this remain unclear, and this elevation has yet to be shown at a national population level. METHODS We tested whether dementia was more prevalent in disadvantaged neighborhoods across the New Zealand population (N = 1.41 million analytic sample) over a 20‐year observation. We then tested whether premorbid dementia risk factors and MRI‐measured brain‐structure antecedents were more prevalent among midlife residents of disadvantaged neighborhoods in a population‐representative NZ‐birth‐cohort (N = 938 analytic sample). RESULTS People residing in disadvantaged neighborhoods were at greater risk of dementia (HR per‐quintile‐disadvantage‐increase = 1.09, 95% confidence interval [CI]:1.08‐1.10) and, decades before clinical endpoints typically emerge, evidenced elevated dementia‐risk scores (CAIDE, LIBRA, Lancet, ANU‐ADRI, DunedinARB; β’s 0.31‐0.39) and displayed dementia‐associated brain structural deficits and cognitive difficulties/decline. DISCUSSION Disadvantaged neighborhoods have more residents with dementia, and decades before dementia is diagnosed, residents have more dementia‐risk factors and brain‐structure antecedents. Whether or not neighborhoods causally influence risk, they may offer scalable opportunities for primary dementia prevention.


BACKGROUND
Fifty million individuals worldwide are currently living with Alzheimer's disease and related dementias (ADRD), a number expected to triple within 30 years as the global population ages. 1 With few interventions that can effectively halt or delay disease progression, 2 primary prevention of dementia has become a global goal.
Collectively, this existing evidence suggests that neighborhoodbased interventions could offer a new avenue for primary dementia prevention 16 -one that may leverage existing resources outside the healthcare sector, influence whole communities at once, and operate without necessarily requiring individual behavior change.For example, interventions targeting individuals, such as neighborhood mobility voucher programs, have shown efficacy for reducing obesity and diabetes, 17 while interventions targeting whole neighborhoods, such as vacant-lot greening initiatives, have been found to influence area-level trends in criminal behavior, 18 diets, 19 and mental health. 20e nature of neighborhood-based risk for dementia remains undercharacterized, however. 16Critically, it is unclear when in the lifespan such risk emerges.The lion's share of the evidence (∼70% of published studies) on neighborhood-ADRD associations arrives from studies of older adults who have either received diagnoses, donated their brains for post mortem study, or been observed over the last years of their life (Table S1).This limits causal inference and the identification of intervention opportunities for four reasons.First, it does not rule out reverse causation, whereby individuals in the long premorbid phase of ADRDs involuntarily migrate to less desirable neighborhoods as a consequence of illness (e.g., cognitive decline forces early outflow from the labor market).Second, it does not rule out the accumulation within disadvantaged neighborhoods of individuals at risk for ADRDs due to pre-existing shared risk factors (e.g., low educational attainment earlier in life).Third, older adult samples are biased by healthy survivor bias.

Fourth, it does not indicate when neighborhood interventions would
need to be delivered to be effective.
To help address this limitation in the literature, we first investigated the full New Zealand (NZ) population to test the hypothesis that dementia diagnoses follow neighborhood socioeconomic ("disadvantage") gradients across the entire country.Then, we turned to a deeply phenotyped population-representative NZ-birth cohort followed to age 45 (the Dunedin Study) to test the hypothesis that the geographic patterning of dementia is preceded by geographic gradients in dementia's antecedent risk factors and brain-integrity differences by midlife, decades before ADRD endpoints typically emerge.

Dementia in the NZ population: The integrated data infrastructure (NZ-IDI)
2][23] The study population included the 1,695,447 individuals aged 31-90 years who were born in NZ between 1929 and 1968 and resided in NZ for any time between July 1999 and June 2019.We selected this age range to capture the period of risk for both early-onset and later-onset dementia during the 20-year period.We divided the population into age bands (born between 1929-1938, 1939-1948,   1949-1958, and 1959-1968)

Outcome measure: Dementia diagnoses
We collected information about ADRD diagnoses using a previously published scheme. 23,25

Dementia risk factors and antecedents in a New Zealand birth cohort: The Dunedin study
Birth-cohort participants were members of the Dunedin Study.The full cohort comprises all individuals born between April 1972 and March 1973 in Dunedin, NZ, who were eligible based on residence in the province and who participated in the first assessment at age 3. The cohort represents the full range of socioeconomic status in the general population of NZ's South Island. 26On adult health, the cohort matches the NZ-National Health and Nutrition Survey on key indicators (e.g., body mass index, smoking, physical activity, visits to a physician) 26 and the NZ-Census of citizens of the same age on educational attainment. 27The cohort is primarily white; 7.5% self-identify as having M āori ancestry, which matches the ethnic distribution of the South Island.Assessments were carried out at birth and ages 3, 5, 7, 9,   11, 13, 15, 18, 21, 26, 32, 38, and, most   Outcome Set 1: Dementia risk-factor index included five indexes which are suitable for use in midlife (Supplementary Table S3):  S4). 3 These indexes are checklists that aggregate risk factors known to increase dementia risk because they are either robustly predictive of dementia (e.g., hearing impairment, a history of depression) or are believed to mechanistically precipitate or enhance disease processes (e.g., apolipoprotein E [APOE] genotype status, heart disease).Figure 1 and Supplementary Figure S1 present the risk factors captured in the DunedinARB and the four other indexes, respectively.

Statistical analysis
The full project premise and analysis plan were preregistered and stored at https://tinyurl.com/5ytbax52.Findings were checked for reproducibility by an independent data analyst, who recreated the code based on the manuscript and applied it to a fresh dataset.This report follows STROBE reporting guidelines. 38Analyses conducted in SAS v7.1 and Stata v16.1.

Analyses in the New Zealand-IDI study population
Poisson regression models with relative risks (RRs) and Cox proportional hazard models with hazard ratios (HRs) (with censoring for out-migration or death from causes other than dementia) were used to (1) estimate the association of neighborhood disadvantage with dementia per quintile increase in disadvantage, and ( 2 All models controlled for ethnicity (European, M āori, Pacific, and Asian ethnicity); models using the total population also controlled for sex and birth year.Per the confidentiality rules of Statistics-NZ, reported frequencies/counts were randomly rounded to a base of three.

Analyses in the New Zealand birth cohort
Ordinary least-squares regression models were used to estimate the association of cumulative neighborhood disadvantage with the ADRD risk indexes and the brain structural and functional integrity measures, adjusted for sex.Second-stage models were adjusted for sex and individual socioeconomic status at age 45 years (Supplementary Appendix 3).

Do dementia diagnoses follow neighborhood socioeconomic gradients in the New Zealand population?
We observed 1,695,447 individuals (full study population, 842,028   identified as having dementia.Similar percentages of men and women and more older than younger individuals were identified as having dementia (Supplementary Table S6).Neighborhood-dementia associations increased modestly when all residential addresses across the observation period were taken into account during the 20-year window (dementia RR per-quintileincrease in cumulative disadvantage = 1.13, 95%CI: 1.12-1.14).Sex and age-band differences followed the same general trend of higher Pairwise comparisons of higher disadvantage quintiles against the lowest quintile did not identify significant statistical-effect thresholds: Neighborhood-dementia associations increased linearly with increasing disadvantage (Table 1).Comparisons of NZ-population neighborhood-dementia HRs to those reported from specific samples in the United States and the United Kingdom using similar area-level disadvantage indexes are presented in Table 1; the NZ population findings are similar these pooled HR estimates from previous studies.

(A) (B)
F G U E Neighborhood disadvantage associations with dementia diagnoses in the New Zealand study population, by sex and age cohort.

Are neighborhood-disadvantage gradients in dementia diagnoses preceded by gradients in dementia risk factors and structural and functional brain antecedents by midlife?
California, 8 Minnesota, 10 and Ohio, 12 and among US Veterans Health Administration patients. 13Details on these studies are provided in Supplementary Table S1.Similar studies with other samples that utilized tertiles and quartiles of disadvantage are not included here but are detailed in Supplementary Table S1. of genetic risk (β = 0.02, p = 0.504) (Supplementary Table S8).This pattern of associations remained after adjustment for individual-level socioeconomic status, although effect sizes were attenuated, one to non-significance ("harmful events and exposures" risk) (Supplementary Table S8).Overall, Study members from the most disadvantaged neighborhoods (5 th quintile) exhibited DunedinARB scores a full standard deviation higher (1.06 SD, 95%CI: 0.86, 1.27; t = 10.377,p < 0.001) than members from the least disadvantaged neighborhoods

DISCUSSION
This study produced three findings.First, dementia diagnoses were found to follow linear neighborhood socioeconomic gradients in the NZ population.Risk of dementia was over 20% greater in the most disadvantaged neighborhoods compared to the least.This replicates past  to Opportunity Study identified significant reductions in the prevalence of dementia risk factors of obesity and diabetes after several years among women who received the treatment. 17It has now been nearly three decades since enrollment in that study began-potentially enough time to identify whether the prevalence of dementia or dementia antecedents are also lower among members of the treatment group.
Notably, individual-level neighborhood interventions could be effective while remaining entirely agnostic about any causal mechanisms that may or may not underlie the neighborhood-dementia association.
The second avenue for neighborhood-based dementia interventions would be to improve neighborhood conditions directly (a neighborhood-level intervention), targeting conditions known to influence the development of the dementia risk factors identified in the current study as aggregating in disadvantaged neighborhoods (Supplementary Table S8).Efforts to target such conditions, extensively investigated in the urban design literature, 16 include altering and enhancing pedestrian infrastructure, natural and recreational amenities, the availability of fresh produce, and the accessibility of healthcare services.They also include efforts to reduce nervous system stressors such as heat, noise, and air pollution through altered fuel, air quality, and zoning codes and standards, cooling centers, tree plantings, roaddiets, and road-space rationing.Investigation of just one such initiative, a randomized-controlled trial of vacant-lot greening in Philadelphia, found that low-cost improvements to neighborhood spaces result in significant reductions in neighborhood-level crime 18 and poor mental health. 20,45Although mental disorder is a risk factor for dementia, 23 to the best of our knowledge, such neighborhood-level interventions

CONCLUSIONS
recently, at age 45 years (data collection completed April 2019) when 94% (N = 938) of the cohort members still alive participated.Participants gave written informed consent, and Study protocols were approved by the NZ-Health and Disability Ethics Committee.

2. 2 . 1
Exposure measure: New Zealand index of deprivation Residential neighborhood disadvantage was again assessed via the NZDep, described above.Neighborhood disadvantage scores were available for Study waves at ages 26, 32, 38, and 45 years (from 1998 to 2019).A cumulative adult neighborhood disadvantage score was generated via confirmatory factor analysis producing a single unitary factor from the four Study waves via maximum likelihood estimation with robust standard errors in MPlus (Version 7.11), with factor loadings set to one for each age to equalize disadvantage contributions across time (χ 2 = 95.607,p < 0.001, CFI = 0.834, TLI = 0.815, RMSEA = 0.096, SRMR = 0.092).Sensitivity tests utilizing simple mean-disadvantage scores did not change the pattern of results.

3 a.
The Cardiovascular Risk Factors, Aging, and Incidence of Dementia (CAIDE) index 28 ; b.The LIfestyle for BRAin health (LIBRA) index 29 ; c.The Australian National University Alzheimer's Disease Risk Index (ANU-ADRI) 30 ; and d.Modifiable risk factors selected by the Lancet Commission on Dementia (Lancet) 31 ; and e.A comprehensive midlife index, the Dunedin ADRD Risk Benchmark (DunedinARB), comprised of 48 putative ADRD risk indicators organized into 10 conceptually distinct risk domains (Figure 1; Supplementary Table ) and three measures of brain functional integrity (objective cognitive function assessed via cognitive tests; subjective cognitive function assessed via self-and informant-report; and cognitive decline assessed as change in cognitive test scores from childhood).
) compare higher disadvantage quintiles against the lowest quintile.We used the neighborhood-disadvantage score at individuals' first registered residential address in the 20-year study period (primary exposure) for both RRs and HRs and the mean disadvantage score across up to 20 addresses during the study period for RRs (sensitivity tests).Data were weighted for Poisson models based on time alive and in NZ to account for differences between individuals in observation time due to death or out-migration.No measures were excluded at baseline.Associations were estimated within the total study population with present neighborhood data (analytic sample), and by age band and sex.
[49.7%] female; aged 31-70 years at baseline) who were born in NZ between 1929 and 1968 and resided in NZ for any period between July 1999 and June 2019; 231,567(13.7%)were born between 1929 and 1938, 361,200 (21.3%) between 1939 and 1948, 504,372 (29.7%) between 1949 and 1958, and 598,308 (35.3%) between 1959 and 1968.During the 20-year period, 36,753 individuals (2.2%) wereF I G U R E 1 Schematic of the Dunedin Alzheimer's Disease and Related Dementias Risk Benchmark (DunedinARB).The comprehensive Dunedin Alzheimer's Disease and Related Dementias Risk Benchmark (DunedinARB) is comprised of 48 risk indicators grouped into 10 conceptually distinct domains.Genetic risk includes family history of dementia and apolipoprotein E (APOE) ε4 allele status.Lifestyle risk includes physical activity, diet, tobacco smoking, alcohol consumption, folic acid supplementation, and regular prophylactic non-steroidal anti-inflammatory drug (NSAID) use.Socioeconomic risk includes occupational and educational attainment.Psycho-somatic Function risk includes chronic pain, history of migraine, history of depression, social isolation, sleep quality, neuroticism, and conscientiousness.Physical and Sensory Function risk includes balance, gait, hearing acuity, and subjective hearing function, objective and subjective vision function, and sense of smell.Cardio-Metabolic Status risk includes hypertension, obesity, and diabetes status, total cholesterol, triglycerides, and retinal vascular health.Inflammatory risk includes C-reactive protein (CRP), interleukin 6 (IL-6), and soluble urokinase plasminogen activator receptor (suPAR)levels and history of rheumatoid arthritis.Epigenetic Cellular Aging risk includes four separate DNA methylation "aging" clocks (Horvath, Hannum, PhenoAge, and GrimAge).Harmful Events and Exposures risk includes childhood lead exposure, occupational exposure to neurotoxicants, and history of traumatic brain injury.Subjective Overall Health risk includes self, informant, and research-worker ratings of Study member overall health.Figure from Reuben et al. 3 Details on the individual risk factors and indicators are provided in Supplementary Table

Figure 2
Figure 2 displays the geographic distribution of neighborhood disadvantages in NZ.Neighborhood disadvantage data were available for 1,408,812 individuals (the analytic sample, 83.1% of the study population; 652,581 [46.3%] female) (Supplementary TableS7).After ).After adjustment for covariates, New Zealanders living in more disadvantaged neighborhoods at baseline were modestly more likely to develop dementia across the 20-year observation period (RR perquintile-increase in disadvantage = 1.06, 95% confidence interval [CI]: 1.05-1.06;HR = 1.09, 95%CI: 1.08-1.10).In general, neighborhooddementia associations were modestly greater among men than women and observably greater among more recently born, younger agebands.The youngest age-band demonstrated risk associations 21.2% (for females) to 26.0% (for males) greater than the oldest age band (Figure 3, Panel A).

F I G U R E 2
Geographic distribution of neighborhood socioeconomic disadvantage in New Zealand and the city of Dunedin, New Zealand (2018).The New Zealand Index of Deprivation ranks all populated small areas (encompassing approximately 100-200 residents each on average) from least (quintile 1) to most (quintile 5) socioeconomically disadvantaged.Disadvantaged areas are distributed across both the South and North Island, with the North Island having a slightly larger share of the most disadvantaged areas.Unpopulated areas (Fiordland) pictured in gray.risk associations among men and among more recently born age bands (Figure 3, Panel B).
By age 45 years, Study members living in more disadvantaged neighborhoods across adulthood demonstrated significantly greater accumulation of midlife risk factors for later ADRD (Table 2) (β's of 0.31 to 0.39 for associations with the CAIDE, LIBRA, Lancet, and ANU-ADRI risk indexes and the DunedinARB, p's < 0.001).Figure 4 (Panel A) presents the distribution of DunedinARB scores at each quintile of neighborhood disadvantage.Risk-index-associations remained significant after adjustment for individual-level socioeconomic position (Table 2, column 2), although statistical effect sizes were attenuated (adjusted β's of 0.19 to 0.24, p's < 0.001).Examination of the 10 domains of risk comprising the DunedinARB indicated that no single risk domain accounted for the geographic patterning of dementia riskindex scores (Figure 4, Panel B).All domains of risk were elevated among Study members who had lived in disadvantaged neighborhoods across adulthood (β's of 0.08 to 0.35, p's < 0.05), with the exception TA B L E 1 Risk of dementia (hazard ratios, HR) by quintile of neighborhood disadvantage in New Zealand, the United Kingdom, and the United States.

F I G U R E 4
Association of cumulative adult neighborhood disadvantage with Alzheimer's disease and related dementias (ADRD) risk factors in the New Zealand Dunedin Study at age 45 years.Panel A presents box plot and underlying data points for Dunedin ADRD Risk Benchmark scores at each quintile of neighborhood disadvantage.Boxes display means and 25 th and 75 th percentile scores.Orange line presents the unadjusted trend for the association, with the association's standardized beta coefficient reported in the orange box above the trend line.Panel B: all p < 0.001 except for cardio-metabolic risk (p = 0.005), harmful events and exposures risk (p = 0.009), and genetic risk (p = 0.504).Dark lines present 95% confidence intervals.reports of similar gradients in specific communities in the United States and United Kingdom (Supplemental- have yet to be investigated in relation to dementia or dementia antecedents.This study has limitations.First, it is observational and cannot establish causation.Second, we did not have access to individual-level socioeconomic information for the entire NZ-IDI population; effect estimates would likely be attenuated after adjustment for individual-SES, although the attenuation observed in the Dunedin Study suggests associations would remain significant.Third, administrative-register data are known to undercount dementia cases.Fourth, brain-integrity measures in the Dunedin Study were largely, although not exclusively, cross-sectional.Continued follow-up will be required to further estimate ADRD risk (e.g., using plasma biomarkers) as well as to document neighborhood-associations with longitudinal declines in brain structure.Fifth, neighborhood disadvantage was only assessed across adulthood, leaving open the possibility that neighborhood-brain-integrity associations could be present earlier in the lifespan.
midlife Dunedin Study included the 938 (49.5% female) Study members who attended the age-45 assessment wave (94.1% of the Association of cumulative adult neighborhood socioeconomic disadvantage with ADRD risk factors and brain integrity antecedents by midlife in the Dunedin Study cohort.Note: Cumulative adult neighborhood disadvantage spans ages 26-45 years.Fully adjusted associations include the covariates of sex and individual-level socioeconomic status at age 45 years.Associations with the midlife ADRD indexes are presented as standardized regression beta coefficients.Associations with the midlife brain integrity measures represent unit change in outcomes per one standard deviation increase in neighborhood disadvantage.

Table S1
7,10,40,41ion, poor hearing).The notable exception was genetic risk.APOE-ε4 status and a family history of dementia were entirely unassociated with neighborhood disadvantage in the cohort-a finding matched in older adult studies, where neighborhooddementia association estimates tend to be unchanged by adjustment for genetic risk.7,10,40,41Thissuggests that neighborhood-dementia gradients are not generated by the selective migration of individuals at taged neighborhoods across adulthood.Such risk included all putative ADRD risk factors, from lifestyle risks such as tobacco and alcohol use, to metabolic, psychiatric, and sensory dysfunction (e.g., high cholesterol, history