Abstract

1. Introduction

Alzheimer's Disease and Related Dementia (ADRD) significantly impacts numerous American families. Approximately 5 million Americans over age 65 have Alzheimer's Disease – a subset of those with ADRD – and these numbers are expected to swell to nearly 14 million by 2050 (Hebert et al., 2013). Roughly one in five American women or one in ten American men are anticipated to face dementia in their lifetimes (Chêne et al., 2015). Fifteen percent of Americans over the age of 70 have dementia, with an estimated financial burden of just under $30,000 annually per person, excluding the expenses related to informal caregiving (Hurd et al., 2013). Individuals shouldering an average of over $6000 in out-of-pocket expenses (Hurd et al., 2013). The substantial economic and health burdens underscore the urgency to understand the fundamental causes of ADRD, one of the primary causes of mortality in the US.

Numerous studies have identified specific individual risk factors that contribute to the risk of ADRD mortality. Individual characteristics including lower educational attainment (Korhonen et al., 2020), poorer health (e.g., high blood pressure, physical inactivity, obesity, diabetes, etc.) (Chen et al., 2023; Omura et al., 2022), rural residence (Cross & Warraich, 2021), genetic predispositions (Bellenguez et al., 2022), and social and marital relationships (Kim & Hwang, 2024a, Kim & Hwang, 2024b, Kim & Kwon, 2023, Kim & Park, 2023). are associated with elevated risks of ADRD mortality. However, some of these individual characteristics could systematically vary by place, suggesting local environments could play a role in ADRD mortality. For example, although individuals choose their education attainment, compulsory schooling laws in the late 19th century and early 20th century in the US varied by state and meaningfully altered individual attainment (Clay et al., 2012, 2021; Lleras-Muney, 2002; Stephens & Yang, 2014). Moreover, health behaviors and healthcare systems are subject to significant geographic disparities (Tsui et al., 2020), which indicates that regional variations in healthcare access and health behaviors may contribute to an increased risk of ADRD mortality.

In fact, the local environment itself likely also matters for dementia risk, especially Alzheimer's disease and mortality. Case and Paxson (2009) found that lower local infant mortality rates around age two in the US, indicative of the local health environment, were linked with improved cognition at older ages. Similarly, Topping et al. (2023) showed that early life state-level infant mortality rates were associated with ADRD mortality, especially among those under 65 years of age. On the policy front, Medicaid, primarily a state-administered health insurance program, has been linked to several health outcomes. Miller et al. (2019) found recent ACA-related expansions reduced mortality. Earlier expansions also appear to improve population health (Boudreaux et al., 2016; Miller & Wherry, 2019). Access to Food Stamps or the Supplemental Nutrition Assistance Program also improved health years later and reduced mortality rates (Heflin et al., 2019; Hoynes et al., 2016; Jones, 2020). While improved health could decrease dementia prevalence, lower mortality could increase prevalence by eliminating the competing risk. In any case, policies at the state or local levels can theoretically influence both the risk and prevalence of dementia.

The relative importance of early-, mid-, or late-life environments for ADRD mortality risk remains uncertain. Supporting the case for the importance of early-life environments, lower educational attainment is one of the most widely recognized risk factors for dementia (Cook & Fletcher, 2015; Harrison et al., 2015; Sattler et al., 2012; Wilson et al., 2009), as well as ADRD mortality (Garcia et al., 2021; Korhonen et al., 2020). It is hypothesized that higher education is linked with lower dementia prevalence and its related mortality via increases in cognitive reserve (Stern, 2009). Importantly, primary and secondary education is largely determined in early-life, and historic state policies in the late 19th and early 20th altered attainment in cohorts currently at risk for dementia (Clay et al., 2012, 2021; Lleras-Muney, 2002; Stephens & Yang, 2014). More directly, a few studies found that while both residing in and being born in the US “stroke belt” states is a strong predictor of dementia, birth location may dominate (Glymour et al., 2011; Topping et al., 2021a).

A growing body of research also underscores the role of early-life environments in shaping health outcomes (Topping et al., 2024). Cognition is related to the local disease environment when an individual was two years old (Case & Paxson, 2009). Exposure to food stamps in early childhood (i.e., ages 0–5) decreased metabolic syndrome (Hoynes et al., 2016), while early-life exposure to Medicaid around these same ages improved adult health (Boudreaux et al., 2016) and decreased mortality (Goodman-Bacon, 2021). In the US, however, most major social safety-net expansions started in the 1960s. This suggests that many individuals impacted by these policies during childhood might be too young to observe dementia onset and mortality. One exception is the Mother's Pension Program—a cash transfer program for mothers between 1911 and 1935 that varied state-by-state—that meaningfully altered longevity, educational attainment, and income for beneficiaries' sons (Aizer et al., 2016). In the early 20th century, poverty was widespread in the US, exacerbated by the Great Depression. However, the impact of the Great Depression varied across the US, with the South Atlantic states enduring a milder downturn and the Mountain states suffering more, influenced by industrial composition and long-term regional trends in employment (Margo, 1993; Rosenbloom & Sundstrom, 1999).

In addition to early exposure, individuals are likely to be exposed to different environments during middle and old age. Approximately one-third of people in the US reside in a different state than where they were born (Kahn & Pearlin, 2006). Therefore, to better understand the contribution of early-life environments to ADRD mortality risk, it is required to consider the role of later-life environments simultaneously (Fletcher et al., 2022). For example, Miller et al. (2019) found that ACA Medicaid expansions reduced mortality among individuals aged 55–64. Expansions of the Earned Income Tax Credit (EITC) in the 1990s increased employment and mental health among mothers (Evans & Garthwaite, 2014; Hoynes & Patel, 2018) and improved the overall health status (Lenhart, 2019). The fact that (i) many US safety-net programs started after 1960, (ii) exposure to the social safety-net in later-life offers material and health advantages, and (iii) data limitations on early-life exposure to dementia risks, all curb researchers’ ability to detect early-life exposure effects. This would lend support to the notion that later-life environments may matter more for determining ADRD mortality, particularly among individuals currently at risk of dementia. Moreover, regional variation in competing causes of death (deaths from other causes such as infection or cardiovascular disease) during the 1970s–1990s, potentially attributable to varying levels of regional public investment (Mensah et al., 2017), may reflect in regional variation in ADRD mortality later in life.

Although there are a few studies on geographic variation in mortality, evidence on the relative importance of state of residence versus state of birth is mixed (Finkelstein et al., 2021; W. Xu et al., 2020). Xu et al. (2021) found that the state of birth accounts for almost twice the variation in cardiovascular disease mortality compared to the state of residence. Conversely, another study revealed that state of birth and state of residence contributed the similar amount of variation in all-cause mortality (Xu et al., 2020). With respect to Alzheimer's disease mortality specifically, one study documented the association between local-level infant mortality rates in early life and death from ADRD in the US (Topping et al., 2023). However, Topping et al. (2021b) reported that an individual's state of birth explains relatively little variation in Alzheimer's disease mortality—providing evidence against early-life factors affecting Alzheimer's disease mortality. The findings suggested that almost all the variation in Alzheimer's disease mortality related to the state of birth (about 4%) is accounted for by the state of residence in later life.

Our study investigates the relative importance of early-versus late-life environment contexts in explaining ADRD mortality risk. This research holds critical implications for policymakers endeavoring to discern the optimal life course stages for policy interventions and for researchers or practitioners seeking insights into how ADRD mortality risk progresses over the lifespan. Specifically, our study considers whether birthplace and/or place of residence are most important when modeling ADRD mortality. Our study offers several key improvements. First, we broaden the definition of dementia to include all dementia types, not just Alzheimer's disease. Second, we use a more representative sample from across the United States. Finally, we have approximately four times as many ADRD deaths recorded. With these improvements, our results offer continued support for the notion that later-life contexts matter more than early-life contexts in ADRD risk for current cohorts at risk of dementia.

2. Data and methods

3. Results

Our sample descriptive statistics (Table 2) demonstrate that ADRD mortality is primarily among the oldest Americans. Women and individuals from advantaged backgrounds appear more likely to survive to an age when ADRD can be the main underlying cause of death. Survivors are the youngest sample members—on average being interviewed at age 65 and last observed at age 78—while the deceased are significantly older. Individuals dying from non-ADRD causes on average are interviewed at age 71 and die 8 years later. However, people dying of ADRD are significantly older. On average they are interviewed at age 74 and die at age 85.

Nearly two-thirds of survivors are women, while closer to half of the deceased from non-ADRD causes are women. ADRD deaths though are predominantly among women. ADRD deaths are also primarily among non-Hispanic White individuals. While 88% of survivors are non-Hispanic White, 92% of ADRD deaths are among non-Hispanic Whites. Non-Hispanic Black individuals make up a slightly larger share of the deceased from non-ADRD causes relative to their share of survivors, but they make up a significantly smaller share of ADRD deaths relative to their share of survivors. The distribution of educational attainment among the deceased skews toward lower education levels. For example, half of the deceased from non-ADRD causes have less than a high school degree while just 28% of survivors have less than a high school degree. People dying of ADRD however, appear slightly more educated than those deceased from other causes. Just 44% of ADRD deaths are among people with less than a high school education.

We begin with observing spatial variation in ADRD mortality across the United States in Fig. 1. Similar to Akushevich et al. (2021), our figure highlights significant spatial variation in dementia mortality. The northwest, and to a lesser degree southeast, appears to have higher rates of ADRD mortality while the midsection tends to have lighter shades of blue representing lower ADRD mortality rates. Despite these broad similarities, the side-by-side maps of spatial ADRD mortality rates by state of birth and state of residence are clearly not the same. ADRD mortality is higher among individuals born in Idaho or New York relative to individuals who live in in those states. Conversely, ADRD mortality is lower among individuals born in Washington or Louisiana relative to those who live in those states.

Open in a separate window

Fig. 1

Spatial variation in ADRD mortality by state of residence and state of birth.

Notes: Arizona and Nevada excluded for state of birth illustration in accordance with Census Bureau rounding rules. ADRD = Alzheimer's Disease and Related Dementias.

Source: Authors' calculations using NLMS data at a Federal Statistical Research Data Center under FSRDC Project Number 2338 (CBDRB-FY23-P2338-R10359).

Regression results (Table 3) suggest that state of residence explains relatively more variation in ADRD mortality than either state of birth alone or a combination of state of birth and state of residence. Column 1 contains baseline estimation excluding random effects. Column 2 includes state of residence random effects while column 3 includes state of birth random effects. Column 4 includes a combination of state of residence and state of birth random effects. Both the AIC and BIC are minimized when including only state of residence random effects (column 2) while ${{\sigma }_u}^2$ and $\rho$ are maximized in this specification. Table 3 shows that state of residence explains 1.19% of the total variation in ADRD mortality (column 2), whereas state of birth explains only 0.6% (column 3). Column 4 suggests that both state of residence and state of birth explain only 1.05% of the total variation in ADRD mortality.

To assess the robustness of our findings, we conducted a series of sensitivity analyses. First, we examined whether our results are sensitive to the inclusion of age as a quadratic continuous variable. In Table S1 of the online supplementary file, substituting a quadratic continuous age covariate confirms the robustness of our main conclusions. Second, to address the potential information misclassification bias caused by including non-deceased participants—where participants who may develop dementia before death are classified as non-cases—we estimated models that exclude survivors from the sample. The results, presented in Table S2 of the online supplementary file, remain substantially similar to the original findings. We have also used alternative specifications that use Stata's “melogit” command that nest state of residence and state of birth random effects instead. In this case, the variance explained by state of residence is still the dominating feature.

4. Discussion

ADRD is a pressing public health concern, the impacts of which reverberate through individual lives, families, and the broader socio-economic fabric of the US. This study investigates the relative importance of early-versus late-life environment contexts in explaining ADRD mortality risk. The findings of this study support the notion that later-life environments could matter more than early-life environments for the likelihood of dying from ADRD. Leveraging a large sample of ADRD deaths available in restricted-use NLMS data, we find that state of birth consistently explains less variation in ADRD deaths relative to observed state-of-residence at ages 52–93. These results corroborate an earlier study using a smaller sample and more narrow definition of dementia mortality (Topping et al., 2021b); however, we doubt this is to be a definitive or final verdict on the question. While our results are interesting, informative, and a necessary first step to understanding the relative importance of environment contexts during the life course, future research with younger birth cohorts aging into ADRD mortality and incorporating alternative frameworks to modeling competing risks of mortality could be better equipped to address optimal life course stages for policy intervention.

These findings echo and expand upon the current discourse in the field. While early-life factors such as educational attainment, influenced by state-level policies and local environments, undoubtedly play a role in cognitive outcomes in later life, our findings suggest that where an individual spends their later years might be just as, if not more, influential. Delving deeper into potential mechanisms, several plausible explanations emerge. One of the foremost considerations is the exposure to state-specific health policies and infrastructures (Osypuk et al., 2014). Every state has its own health mandates (e.g., Affordable Care Act/Medicaid), insurance structures, and established guidelines for treating older adults. These policies, whether they pertain to preventive care, early detection, or disease management, can significantly shape the health trajectories of its older residents. Furthermore, states that have robust healthcare infrastructures, with easy access to geriatric care specialists, neurologists, and mental health professionals, might offer residents a better chance at early detection and management of ADRD (Bradford et al., 2009).

That said, it is essential to approach these findings with caution. While we have made strides in our understanding, the magnitudes of explained variation by state of residence or birth are modest. The substantial remainder of unexplained variation suggests that there are other significant factors, beyond state of residence or birth, that influence ADRD outcomes. There are undoubtedly many risk factors that remain unknown and several known risk factors lack a causal link to dementia. These might include individual genetic predispositions, lifestyle factors, or other unmeasured environmental exposures. An asset of our study design is that we can remain agnostic on what specific early- and late-life factors contribute to our observed patterns. It implicitly incorporates both known and unknown environmental risk factors into determining whether early- or late-life matters more for ADRD mortality. In this way, the study can retain applicability even after new discoveries are made. Thus, our study serves as a foundational step, opening doors for more nuanced and multi-dimensional research into the geographical distribution of ADRD.

This study has several limitations. First, we are examining ADRD mortality, not the presence of cognitive impairment, decline, or even the presence of dementia leading up to death. To be classified as an individual with ADRD mortality, this condition must have been listed as the underlying cause of death on the death certificate. It is unlikely that all individuals with underlying dementia indications or pathology at the end of life will be listed with dementia as an underlying cause of death (Todd et al., 2013), which could bias our results. Thus, although the populations of individuals with dementia and those dying from dementia overlap, they are distinct. NLMS is not equipped to comment on the presence of cognitive impairment or pathology, only the underlying cause of death as recorded. It is likely that some individuals with respiratory or circulatory diseases listed as the underlying cause of death may have had ADRD (Brunnström & Englund, 2009). When considering the link between pollution and dementia, a review found higher exposure is almost always linked with dementia, but the link with cognitive decline is more mixed (Peters et al., 2019). In their analysis of geographic disparities of Alzheimer's Disease mortality, Akushevich et al. (2021) conclude that differences in filling out death certificates are an unlikely cause of geographic disparities. Their conclusion suggests this limitation may not meaningfully contribute to our results.

Second, mortality selection is an important factor that could bias results. Bishop et al. (2022) suggest that individuals who died would have been more likely to develop ADRD (as measured by Medicare claims data while alive) had they lived longer. We see evidence of this pattern in our own sample (Table 2). Individuals dying during follow-up without ADRD are typically younger. There's also a larger portion with lower educational attainment, and individuals who are non-Hispanic Black among those dying of non-ADRD causes relative to those dying of ADRD. Thus, ADRD mortality is concentrated among the oldest old: those who have survived long enough to acquire and die from ADRD. At the same time, one of the most consistent risk factors for cognitive impairment or developing ADRD is lower educational attainment (Cook & Fletcher, 2015; Harrison et al., 2015; Sattler et al., 2012; Wilson et al., 2009). However, lower socioeconomic status, and education in particular, is importantly also a significant risk factor for higher mortality rates (Chetty et al., 2016; Montez et al., 2019) and reiterates the fact that we are studying the presence of ADRD mortality, not cognitive impairment, decline, or ADRD indications near the end of life.

Third, we limit our sample to individuals born in 1930 or earlier. While this ensures ample time to observe mortality, no one in our sample was exposed to the social programs forming today's safety-net during their young childhood. Instead, individuals from these cohorts primarily encountered the safety-net in adulthood. Two potential exceptions to this generalization are the Mothers' Pension program and the Aid to Families with Dependent Children (AFDC) program, which was introduced in 1935. In the case of the AFDC, the youngest in our sample might have been exposed during childhood, but never before reaching five years of age. Later safety-net programs like Food Stamps, the predecessor of SNAP, began as a pilot in 1961 (Hoynes et al., 2016), by which time our youngest cohort was already in their 30s. Medicaid started shortly after, and the Earned Income Tax Credit did not begin until 1975. Many programs comprising the US social safety-net differ from state to state. To the degree that improved health and wellbeing from these programs is linked with ADRD mortality, in our sample it would likely only be detected by adulthood state of residence. As individuals exposed to the social safety-net as children age, future research may need to revisit this important question.

In conclusion, our study sheds light on the often-overlooked geographical nuances in the understanding of ADRD. While early-life factors and state of birth may be influential, our findings emphasize the importance of considering where one resides in their later years, at least for current cohorts at risk of ADRD. It is crucial to consider the broader implications of these findings. If the state of residence in later life is indeed important, then interventions targeted at improving the quality of life for older adults, enhancing healthcare access, or reducing environmental risk factors in states with higher ADRD prevalence could be strategic starting points. Additionally, understanding the influence of state-specific policies or practices could guide more targeted, state-level interventions, potentially leading to reduced ADRD incidence or improved care outcomes. As an effort to address ADRD progress, such insights can guide more effective, targeted strategies, bringing us one step closer to a world with a deeper understanding and, hopefully, better management of this debilitating condition.

Ethical statement

This research was approved by the Institutional Review Boards at the University of Wisconsin-Madison (2016-0856).

Disclaimer

Any views expressed are those of the authors and not those of the U.S. Census Bureau. The Census Bureau has reviewed this data product to ensure appropriate access, use, and disclosure avoidance protection of the confidential source data used to produce this product. This research was performed at a Federal Statistical Research Data Center under FSRDC Project Number 2338 (CBDRB-FY23-P2338-R10359 and CBDRB-FY24-P2338-R11203).

CRediT authorship contribution statement

Jason Fletcher: Writing – review & editing, Supervision, Methodology, Investigation, Funding acquisition, Conceptualization. Katie Jajtner: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Conceptualization. Jinho Kim: Writing – review & editing, Writing – original draft, Supervision, Methodology, Investigation.

Declaration of competing interest

The authors declare no conflict of interest.

Acknowledgements

References