8

Key Takeaways and Future Directions

OVERVIEW

Hector González, Professor of Neurosciences in the School of Medicine at the University of California, San Diego, introduced the final session and commented that one of the benefits of the workshop was bringing together the perspectives of experts from multiple disciplines. As a population neuroscientist, his typical discussions involving dementia are very different from those provided by the social scientists participating in the workshop. Following is his summary of the workshop.

As a medical professional, González noted the importance of using correct and consistent terminology. For example, cognitive function is a cross-sectional measure while cognitive decline is longitudinal.

González said it is important to take a life course approach. Figure 8-1 illustrates how the various factors affecting cognitive function (focusing on AD) apply across the life course. Both individual and societal factors are important, with health-related symptoms and precursors of dementia appearing at different levels of aging.

González said greater attention needs to be given to early life, including epigenetics and the first 1,000 days. Placental banks exist in some states, and blood samples are available that could be used for tapping into micro-nutrients in the first 1,000 days. These could provide clues as to some of the earliest impacts on health and later cognitive decline.

Education is very important as a potential tool for predicting and lowering the risk of cognitive decline, but echoing what Eric Grodsky discussed, González said more nuanced measures are needed than simple counts of the

number of years completed. Furthermore, knowing the statistical association between education and cognitive decline does not establish the causal connections: the level of education also may act as a proxy for many other variables, including social deprivation in early life, and cognitive stimulation with the mother, father, and family unit. Beyond education, González said, other forms of cognitive stimulation, such as social interactions, should also be measured. For example, the Lancet report discussed the importance of social isolation but for later in life, not in early life (Livingston et al., 2024). A new factor not yet well understood is the degree to which screen time constitutes cognitive stimulation and social connections.

Factors relating to cognitive functioning may have differing levels of importance when taking a global perspective, González said. For example, leaded gasoline was banned in the United States earlier than in some other countries, and the same is true of some neurotoxicants. In fact, some of these substances that were banned in the United States were shipped abroad at discounts. Since food is not necessarily local, some impacts from differences in international practices may circulate back to this country as well. Also, education levels vary from one country to another, so countries will vary in the extent to which education may be a source of cognitive reserve.

Still, given that there are areas internationally with little education but also little dementia, the precise role of education relative to other factors is not yet fully understood.

González said cardiovascular health is clearly on the pathway to a lot of dementias and one that disproportionately affects people of color. Hypertension rates are high among African American people, and diabetes rates are high among Hispanic/Latino people. These factors occur or arise in midlife and, as Priya Palta indicated, our interventions may arrive too late.

González said diversity is an area that deserves greater attention. While a diverse group of researchers is engaged, key databases often include only small numbers of minority groups, resulting in only limited research on non-White people. Cognitive decline manifests differently in midlife for people of different heritages and ethnic/racial backgrounds. One relevant study is the Study of Latinos–Investigation of Neurocognitive Aging, which is similar to the ARIC Neurocognitive Study; González said more such studies are needed.

DATA AND METHODS

Katrina Walsemann, Roger C. Lipitz Distinguished Chair in Health Policy and Professor at the University of Maryland’s School of Public Policy and a Multiple Principal Investigator of the Network on Education, Biosocial Pathways, and Dementia across Diverse Populations, summarized the workshop, focusing on the data and methods. Following are her comments.

The ability to answer the questions raised in the workshop is tightly linked to the data that are collected, the quality of those data, and the quality of the measures used to assess these pathways and these relationships, Walsemann said. Much of what was discussed in the workshop was the need to get greater precision in both the measurement of predictors and outcomes, and the ability to obtain significant findings for smaller populations. Diverse samples that are broadly representative are going to naturally have more variability, making it harder to pick up on the association for small groups having few members in the sample.

Walsemann said precision and error depend on three key areas: (a) sampling (with statistical inference typically based on the assumption of simple random samples); (b) statistical power, based on the size of the sample; and (c) measurement. One idea that has been posited is that nonrepresentative, non-probability types of samples that have really large sample sizes might actually increase statistical power and therefore generate more useful information. Walsemann cautioned that such approaches also increase both sampling error and omitted variable bias; a key question is why people choose to participate in such studies and whether that creates some unmeasured bias. One statistician compared the estimates from

a simple random sample of 400 with a non-random sample of 2 million, finding that the standard errors were equivalent (Meng, 2018).

Walsemann concluded that the real key is measurement precision, whether of exposure levels or of the endpoint (i.e., dementia). The associations that researchers find between exposures and non-dementia-related outcomes may not persist when estimating associations with dementia, given that the disease etiology is so different. This makes connecting the research to theory very important.

Walsemann said there also is a need to balance cost with the information gained. That is, does the extra work required to develop more precise measures produce substantially different results? It would be too costly in large studies like the Wisconsin Longitudinal Study (WLS), Add Health, and HRS to perform clinical assessments—the strength of clinical studies—but there may be cost-effective ways of measuring dementia more precisely than they currently do, using blood-based biomarkers, measures of depression and physical functioning, and proxy reports of memory.

Walsemann concluded by stressing the importance of NIA’s continued investment in probability samples like the WLS, HRS, Add Health, and NLSY. Increasing the sample size will not necessarily reduce the error significantly; however, it could be a viable strategy if these increases are targeted at the sampling of key subpopulations. Walsemann also commented that it is worthwhile to invest in improved measures of exposure, pathways, and outcomes, and to ensure that the research questions are theoretically motivated and reflect a deep understanding of a disease process linking exposures to dementia in later life.

DIRECTIONS, RESOURCES, AND METHODS

Robert Hummer, Howard W. Odum Distinguished Professor of Sociology and Fellow of the Carolina Population Center at the University of North Carolina at Chapel Hill, offered closing comments about (a) where work should be headed, (b) what data resources will be needed, and (c) what promising methods are worth investing in.

Where Work Should Be Headed

Hummer concluded that current knowledge is limited about how midlife exposures modify the effects of early life disadvantages, in part because even the best data sources have not been designed for this task. Nonetheless, some researchers have found that there are midlife modifications that can ameliorate or exacerbate early life influences. Hummer listed the following steps identified by workshop participants to be taken to improve our understanding of these relationships:

• Education: Researchers need more detail on education, beyond simply the number of years of education completed by an individual.
• Early life: Researchers need to give more attention to true early life, including the first 1,000 days, early life nutrition, adverse childhood experiences, family context, and rapidly evolving technologies.
• Jobs: A greater focus on work, and on jobs more specifically, is needed. The work environment is a rapidly changing part of the life course, especially after COVID-19.
• Experiments and policy research: Researchers can conduct experiments such as what the Work, Family & Health Network did in terms of health—for example, looking at how to make work environments more stimulating and how to give people more flexibility. Hummer said there is much potential for policy work; for example, the results for SNAP showed that many are eligible but not participating. Similarly, changes in the child tax credit may have important long-term implications, along with changes in policing and law enforcement.
• Caregiving: Researchers need to look at how midlife caregiving roles are impacting the health and cognition of people in that age range, how those are going to influence their later life, and what modifiable impacts can be made in those very difficult roles.
• Minoritized groups: Hummer said a great deal of work needs to be done on population heterogeneity in the context of inequities and oppression. Characteristics such as race and ethnicity, immigrant status, sex and gender, identity, veteran status, and sexual orientation are broadly systemic in how our society has attached meaning to them. Jennifer Ailshire added that cultural differences affect the frequency of different types of exposures and of perceptions of them. For example, neighborhood-based noise might be considered detrimental by some groups, while other groups may consider it a normal aspect of the neighborhood. Intergenerational living, which could have a positive impact on cognitive stimulation, has much higher prevalence in Asian, Latino, and Black families than in White families. Also, some diagnoses may get missed among people with limited access to or use of medical centers.

Hummer also commented that much has been learned about how physical health affects cognition, but more knowledge is needed.

• Environmental exposures: Hummer said excellent work has been performed on the impact of environmental exposures on cognitive functioning.

• Cardiovascular pathways: Hummer said researchers need to focus more on cardiovascular pathways. Early midlife death rates are starting to go back up for cardiovascular causes of mortality. The Add Health study participants are age 45, on average, and some have been obese for 30 years; this is a new phenomenon that is worthy of study for ameliorating later life cognitive decline and risk of ADRD.

Data Sources

Hummer continued his summary of the workshop proceedings, turning to a discussion of data resources, both those that exist and those that are needed.

• Representative studies: As indicated by the workshop’s participants, Hummer said population-representative cohort studies have been and will continue to be an important resource. They have the advantage of telling the story of the full population.
• Big data: Hummer said big data, such as electronic health records (EHR), mortality records, and Medicare claims reports, can offer different ways to test hypotheses about policies related to space, time, and subgroups. They have shortcomings as well, particularly with regard to measurement, and could not fully replace NIA-supported surveys; still, they offer valuable information. Jason Fletcher commented that some studies require more data than are available from surveys, and this is an area where big data can help.
• State data: Hummer said state-based educational data could be another valuable tool for unpacking educational processes.
• Qualitative data: Hummer said there also is value in complementing survey-based work with qualitative data projects and targeted local and group-specific samples. These may be particularly valuable for understanding hard-to-study populations.

Types of Data

• Minoritized groups: In many cases, Hummer said, there is value in increasing the sample sizes of minoritized groups. For example, HRS increased the number of Hispanic Latinos entering into its new cohorts. Add Health is not adding cohorts at this point, but it is focused on getting response rates as high as possible and improving understanding of the midlife for minoritized groups.
• Biomarkers: Hummer said biological samples are needed, and they should be assayed now and stored for future assays as well, as new biomarkers come into play.

• Precise measurement: Hummer said measurement of both explanatory variables and outcome variables needs to become more precise, whether for factors such as education or for cognition itself. Interdisciplinary work can be helpful in this regard. Measurement is also important both at the individual level and at the contextual level—for example, relating to the timing of exposures to the environment, the timing of structural racism, structural sexism, and policies about immigrants. González emphasized that it is important to think of measuring cognition in terms of there being multiple etiologies. Deborah Carr added that there may be value in preparing a position paper with recommendations about what are conceptually and statistically distinct operationalizations for particular subtypes of cognitive functioning.
• Early life: Hummer said researchers and policymakers should consider the potential of early life cohort studies: The Future of Families, and the Early Childhood Longitudinal Studies birth and kindergarten cohorts. These will help provide data on a full life course rather than starting at an arbitrary age such as 15, 30, or 50.

Data Approaches

• Harmonizing data: Hummer said harmonization of nationally representative cohort studies is hard and does not always make sense. It may be more useful to harmonize small subsets of measures, with crosswalks across studies.
• Coordination: NHLBI also supports important data collections, and Hummer said there should be better coordination between those working with NHLBI and those working with NIA.

Methods

As the last part of his workshop summary, Hummer reviewed what prominent methods are worth investing in.

• Synthetic cohorts: The work to pull multiple datasets together and create synthetic cohorts has great promise, Hummer said, and is more practical than creating a new birth cohort study and waiting 50 years. Hummer said researchers and policymakers need to know more about the methods used, the pitfalls, the promises, the computing environments needed, the software, weighting, and the big assumptions.
• Experimental and quasi-experimental designs: Hummer said quasi-experiments and real interventions also can be valuable, exploring

• topics such as income changes, the child tax credit, Ozempic, workplace interventions, tree planting, caregiving, hearing aids, blood pressure, and other behavior interventions. Researchers can look at both who benefits most from the interventions and how to best impact cognitive functioning now and in the future. Jacqueline Torres added that it is valuable to think cross-nationally, as many global settings offer quasi-experiments that may not be possible in the United States, either because of a lack of variation or a lack of experimentation. Adina Zeki Al Hazzouri added that there are advances in quasi-experimental designs that are underused and that will help in making use of policy changes or cutoffs in how policies are being shaped.
• Modeling: Fletcher commented that it is important to consider competing risks—for example, a large event that no one survives—so that risks of dementia cannot be measured. This is something that Hummer said needs to be considered on the modeling side.