to consider what kinds of policy interventions, such as those related to pensions, long-term care, and formal and informal care, can influence the health and the well-being of older populations in LMICs.

CONTEXTUAL DETERMINANTS OF FUNCTIONAL HEALTH AND DISABILITY

There is a huge amount of heterogeneity in LMICs, said Collin Payne (Australian National University). This heterogeneity means that there is an opportunity to learn how different contexts, economic situations, and policies can affect health and aging. However, it also presents a challenge because it can be difficult to draw strong conclusions from scattered evidence and to understand how evidence from one country can translate to another. Payne focused his presentation on several themes that he has identified in recent literature:

• functional health across different contexts;
• development and trends in health over time;
• met and unmet needs for care;
• moving beyond population-level approaches—social inequalities in LMICs;
• policy changes and health; and
• determinants of disability.

He discussed each of these in turn.

There is a clear relationship between economic development and mortality, Payne said, but the relationship between economic development and functional health is less clear. Recent studies in LMICs have found no clear relationship between economic development and disability-free life expectancy (DFLE; Prina et al., 2020). Payne’s own research found that DFLE is not substantially different between the United States and a set of Latin American countries, despite wide economic differences (Payne, 2018). While the absolute number of older adults with limitations in LMICs will grow massively in the next 30 years, he said, it is unclear whether and how economic development in these countries will affect disability and DFLE. There is substantial variation among LMICs that cannot be explained by national-level measures of economic development; more research is needed to understand how social context and policies might be influencing these differences.

Payne noted that many presenters have already touched on the issue of early-life conditions and health later in life. More research is needed to understand how life-course exposure to rapid social, contextual, and policy changes shapes health among older adults in LMICs. Research has found that changes in life conditions may lead to changes in successive birth cohorts’ risk of obesity, diabetes, and mortality (Beltrán-Sánchez et al., 2022;

Palloni & Beltrán-Sánchez, 2017). One “troubling” finding is that cohorts that grow up in relatively more developed conditions end up with worse health than older cohorts. Studies have found an expansion of morbidity occurring over cohorts in Mexico (Payne & Wong, 2019), as well as among less advantaged groups in the United States (Payne, 2022). In China, research found that the socioeconomic status of adults plays a substantial role in shaping healthy longevity. It is critical, said Payne, to look at the full life course and the exposures that happen over time in order to better understand their impact on older age health, disability, and mortality.

As older adults age in LMICs, they will need care. The question, said Payne, is how these needs for functional health support will match up with available care. Unmet needs are determined by complex factors that are connected to development, family structure, and institutional supports. When institutional support is not available, the availability and capacity of others is hugely important. Who cares for adults and the quality of this care matter for meeting care needs. There are some indications, said Payne, that unmet needs may be higher among “younger-old” people with physical limitations (Brinkmann et al., 2021; Harling et al., 2020). These individuals might not “fit the mold” of who is expected to need care, but attention is needed to ensure that this population is not left behind.

Inequalities within and among LMICs are an interesting area that has often been overlooked, said Payne. Some research has shown that inequalities in health are greater in high-income countries than in low-income countries. For example, inequality in disability by income is much greater in high-income countries (Li et al., 2023), and mortality gradients by education, wealth, and occupation were lower in middle-income countries as compared with high-income ones in the Asia-Pacific area (Xu et al., 2023). Payne noted that effective health care systems may affect inequalities in LMICs (Rosero-Bixby & Dow, 2016).

One policy change that has been studied quite a bit in LMICs is the introduction of public health insurance. Many countries have implemented national health care schemes in the last several decades, with varying levels of success. However, said Payne, there is lots of heterogeneity in how national health insurance reforms are designed, which makes direct comparisons across countries difficult (Lagomarsino et al., 2012). The available evidence does suggest that insurance has led to increased utilization of care; however, the effects on treatment and health outcomes are unclear (Erlangga et al., 2019; Limwattananon et al., 2017; Parker et al., 2018; Rivera-Hernandez et al., 2016).

Payne said that one area that is ripe for further research is the determinants of disability in LMICs. Rather than simply looking at cross-national differences of functional health, one needs to understand why these differences exist. Research has found that chronic disease plays a large role in disability. One cross-sectional analysis in 11 countries found dementia to

be a major factor, and two-thirds of disability was associated with chronic diseases (Sousa et al., 2009). Diabetes is also a major contributor: studies in Mexico and South Africa found large losses in life expectancy and DFLE among people with diabetes (Andrade, 2010; Payne et al., 2023). In fact, a study in South Africa found that the losses of life expectancy and DFLE were larger for diabetes than for HIV. These findings are especially pertinent given the rising rates of diabetes and the poor state of diabetes care in many LMICs, he said.

In terms of improving and expanding research on aging in LMICs—and ensuring that research findings are helpful for developing policy—Payne identified five priority areas for action:

1. measurement and comparability;
2. the role of policy in shaping later life health;
3. noncommunicable diseases and care as a determinant of population health;
4. life-course exposures at micro and macro levels; and
5. translating findings to other contexts.

He discussed each of these in turn.

Researchers who are working across countries often seek to use the same measures so that results can be compared. Payne said that while comparability is beneficial, they need to consider whether measures developed in high-income countries are serving the needs of LMICs. For example, subjective and objective measures of health often do not closely align in LMICs (Capistrant et al., 2014; Payne, 2018). One study found that accounting for physical performance rather than subjective measures leads old-age dependency ratios to increase in LMICs and decline in high-income countries (Kämpfen et al., 2020). Payne said there is a constant balancing of interpretability and measurement and simplicity and complexity. One way to address this tension may be to approach “health” from multiple angles rather than looking at each facet in isolation.

There is relatively limited research on the relationship between macro-level policy changes and micro-level well-being in later life. There is evidence on how policies have affected health behaviors and health care access, but far less evidence about how they have affected health itself, he said. It is challenging to study policies cross-nationally because of the complexities of policies and the contexts in which they are implemented. Payne noted that the Social PoLicy Archive for SHARE (SPLASH) database¹ for social policy researchers in Europe is an “amazing resource”; a similar shared database

___________________

¹ SPLASH: https://share-eric.eu/data/data-set-details/social-policy-archive-for-share-splash

would be a “huge boon” for researchers working in LMICs who want to understand the role of policy in their work.

Noncommunicable diseases are a major determinant of health in aging in LMICs, said Payne, but there is a question as to whether the right data exist to inform policy. Studies on disease in LMICs tend to ask questions in slightly different ways, which makes comparability very difficult. Noncommunicable diseases are increasingly a major determinant of late-life well-being in LMICs, he said, so it is essential that one can measure disease and outcomes in order to assess how health systems are (or are not) keeping up with the challenges, and that measures are comparable across contexts. There is a need to collect both micro- and macro-level data about individual health conditions and how individuals are interacting with health care systems.

Studies on older adults tend to collect information about education, wealth, housing status, and other factors, and look at how it relates to health. However, Payne said, it is critical to look at the entire life courses of older adults in order to understand how changes and development have affected population health. For example, he said, in looking at disability, one could compare obesity with changing food systems and diets in LMICs. Life history surveys should be incorporated into more Health and Retirement Study (HRS) International Family of Studies; this would provide a “huge wealth” of detailed information on people’s lives as they reach older age. One use of this information would be to allow for retroactive investigation of how policies affect long-term health. In addition to expanded research, Payne said, better theories are needed of how social inequalities emerge and change alongside economic and political development.

As other presenters have noted, Payne said, research in LMICs may be able to inform our knowledge of aging in other contexts by improving understanding of what relationships are contextual, and what relationships are causal. For example, social inequalities in health in many LMICs do not align with those in high-income countries. There are likely many relationships that have been established in high-income countries that do not translate to LMICs, and further research is needed to understand how and why. Payne suggested that understanding whether LMIC findings are generalizable to other contexts will require replicating findings in other geographical areas, social contexts, and levels of development in order to build the evidence base.

GENERALIZATION AND TRANSPORTABILITY

There is a great deal of really interesting research on aging happening in LMICs, including research on biomarkers, the influence of life-course exposures, and the impact of policy, said David Canning (Harvard Univer-

sity). Most of this research has good internal validity: studies with randomization, regression discontinuity, and long follow-up times all help ensure that findings are accurate. However, there are challenges when it comes to external validity, that is, the ability to generalize to the population and transport findings to other populations and contexts. This is problematic, he said, because international comparisons can be very useful for learning more about a given factor and its impact on health. For example, research on the effects of pensions on health in South Africa has found positive effects on health; similar effects were found in rural China. Finding a causal relationship requires a large sample size and a lot of variability in exposure, he said. More data with more variation allows for more information about potential relationships; looking across multiple different countries, with their different contexts and policies, increases the chances of finding a true relationship between a factor and a health outcome. Some countries, such as India, China, and the United States, have a lot of variation in policy across provinces and states, which can be useful for studying the impact of different policies in similar contexts.

There are three assumptions needed for external validity, said Canning: (1) conditional exchangeability, (2) selection positivity, and (3) stable unit treatment value. Conditional exchangeability is necessary for generalizability, or the ability to apply the findings from a sample to the larger population. Conditional exchangeability means that the study sample and the target population have the same distribution of unobserved effect modifiers. In a random sample, the expected distribution would be the same across the sample and the population. This assumption is testable, said Canning. While it may not be possible to identify and correct for all effect modifiers, it is possible to test for parameter equality across a sample and the population. One way of doing so is to make predictions from the sample data and then check if they hold true in the population data. For example, based on mortality data from an internally valid study, is the predicted mortality rate in the population borne out in the macro data?

Selection positivity means that every type of person who is in the population has to be in the sample, that is, have a “positive probability of selection,” Canning said. If some segments of a population are not represented, “you have really no hope of generalizing to them.” With a random sample, selection positivity is straightforward. With a nonrandom sample, there are ways to test the sample to ensure it is representative and adjust the sample if not.

Canning explained that stable unit treatment value assumption means that what happens in the study sample and in the general population is the same, and that the effects are the same. Harmonization across countries is possible if the measured exposures and outcomes are the same. However, there is an assumption being made that being in the study does not affect

the outcome. This is “incredibly problematic,” said Canning, because there is evidence that there can be effects from being in a study. For example, people that learn health information due to participation in a study may change their behaviors or health care decisions. Even just being asked questions in a study may have an effect, he said, by making people more sensitive to and aware of health issues. Many studies use a panel design, in which the same group of participants is measured over time. While this design has many benefits for internal validity, it presents challenges for external validity if the panel becomes more dissimilar from the population over waves. This can be accounted for, however, by either comparing the panel with new cross sections over time or by creating a control group to compare with the treatment group. The control group is made up of individuals who are eligible for the study but randomly not selected.

For transportability, or the ability to apply the findings to a new population, Canning said that it is important to look at effect heterogeneity and effect modifiers. Effects are heterogeneous across individuals, and while average population effects may not be the same in different settings, individual effects could be similar. Canning noted that transporting findings from LMICs to the United States as a whole may be too difficult. However, it may be possible to make comparisons with populations that are similar. For example, low-income people in the United States are wealthier than low-income people in LMICs, but the effects of a policy or intervention could be similar. Other groups for which findings could potentially be transportable include HIV-positive populations, migrants from LMICs, and people who experienced early-life adversities.

RESEARCH INFORMING SOCIAL POLICY

Social protection policies that support the economic health and well-being of the most vulnerable populations are critically important, said Lindsay Kobayashi (University of Michigan). These policies have been expanding across the world in recent decades, with more than 130 countries using cash transfers as part of their social protection policies. However, evidence on the relationship between these policies and health remains nascent, she said, for four reasons. First, there is a lack of high-quality data: large population-based samples and longitudinal studies are needed in order to measure pre- and post-policy implementation outcomes. Second, there needs to be variation across different variables so that one can measure heterogeneous treatment effects and subgroup effects. The HRS International Family of Studies are a great starting place, said Kobayashi, but more is needed. The third reason is the challenge of causal identification. Policy change is “very messy” in the real world, and it is difficult to find a valid instrument or source of exogenous variation in exposure to a policy change. Finally, trans-

portability is a big challenge: transporting effect estimates across LMICs or between LMICs and high-income countries is difficult because of big contextual differences. However, she said, she agreed with Canning that there may be opportunities for group transportability.

Despite all these challenges, there is some promising empirical evidence from South Africa and other settings on the effects of social policies on older adults’ health. Kobayashi shared data from her work in this area. South Africa has a “really robust” set of social protection policies; the largest are the old-age pension and the child support grant. Between 2008 and 2010, the pension part of the program expanded for men aged 65 and older to those aged 60 and older. Between 2003 and 2012, the upper age limit for the child part of the program, which is targeted at caregivers of young children (the vast majority of whom are mothers), was expanded from 7 to 18.

In investigating the impact of these policy expansions on cognitive health among older adults, Kobayashi and her colleagues (Kobayashi et al., 2021) found that men who benefited from the expansion of the old-age pension had better cognitive function than predicted, based on trends in the cohort of men who did not benefit from the expansion, said Kobayashi. Depending on their age at the time of expansion, different cohorts had an additional 1–5 years of pension over the transition time period. The data revealed a dose–response relationship: men with 5 years of additional eligibility benefited the most. Researchers tested these findings against a negative control group of women of the same age who were not eligible for pension expansion. Kobayashi explained that this comparison was done in order to ensure that the observed effect was not the result of birth cohort or other differences. There was no association, she said, which enhanced the robustness of the original finding.

Women who benefited from the expansion of the child support grant also showed improved cognitive health. Researchers compared eligible women with women who were not eligible but who had children of similar ages. She noted that the positive effect was not seen in women who had five or more children, but that this could be because of imprecise estimation (Kobayashi et al., 2021).

Kobayashi shared another study (Rosenberg et al., 2023) that was designed to look at the impact of cash transfers on HIV incidence among young women; the unique design of the study allowed for an examination of the impact of cash transfers on older adults’ cognitive health. She explained that the study provided cash to a randomized sample of households, and a large proportion of these households included an older adult member who eventually was enrolled in Health and Aging in Africa: A Longitudi-

nal Study in South Africa (HAALSI).² The overlap between the HIV study and HAALSI provided a unique opportunity to look at the relationship between cash transfers and memory function. On average, individuals from households that were randomized to the cash transfer arm of the study had a slower rate of memory decline than individuals from non–cash transfer households. The effect was strong, said Kobayashi, with a difference of about 0.15 standard deviations over the 6-year period. This provides “really robust evidence” for a casual effect of improved income on protecting memory as individuals age.

In closing, Kobayashi identified three areas to prioritize for future research. First, there is a need for improved data infrastructure on aging in LMICs. This will require longitudinal population-based data on health outcomes. It is “incredibly important” to continue to support the HRS International Family of Studies network, she said. Data are needed to support evaluations of external validity to general populations—without a well-enumerated sampling frame, it is difficult to establish external validity for the purposes of translating findings into policy. Kobayashi agreed with Payne that researchers need comprehensive and publicly available data on social policies (e.g., SPLASH). This is “really high-priority, low-hanging fruit” that would greatly enhance research. A publicly available database with comprehensive information on different types of social protection policies across LMICs would be very valuable and is a “tractable action step.”

Kobayashi’s second priority area for future research is to focus on key health outcomes, including dementia, cardiometabolic outcomes, disability, and functional capacity. It is critical to look at inequalities in these outcomes, which may play out differently across settings, and how policies can affect inequalities.

Finally, Kobayashi said there is a need for methodological work to strengthen causal identification, external validity, and transportability. This work is “absolutely essential” for appropriate policy translation, so it is important to get it right. Kobayashi noted, however, that transportability is difficult, requiring data on effect-modifying variables in both samples in a study, which is sometimes not possible or desirable. Causal identification and internal validity should “come first.”

DISCUSSION

Following the presentations, Minki Chatterji (National Institute on Aging [NIA]) moderated a question-and-answer session. She began by emphasizing her concern about transportability of findings, noting that one purpose for NIA-funded research is to apply evidence from one area to an-

___________________

² https://haalsi.org/

other. Kobayashi responded that transportability is not as big an issue as it may seem. There are many analytical purposes for data that do not require transportability, she said, and research helps to build a cohesive evidence base. There are many general lessons that can be learned from research in one setting that can likely be applied to many other settings—for example, the effectiveness of cash transfers.

Chatterji also commented on the idea for a social policy research database, similar to SPLASH. She wondered about the scale of the task and where and how it could be appropriate to start. Kobayashi suggested that beginning with countries in the HRS International Family of Studies network could be a good starting place and that it could be important to partner with governments in order to fully understand the policies and how they are implemented. Payne agreed with the need for partnerships, saying that partnering with government statistical organizations and other agencies would “go a long way.” He said that there is no need to reinvent the wheel, but, instead, researchers can build on networks that already exist. Lisa Berkman (Harvard University) added that while cross-country policy comparisons can be quite complicated, there are many opportunities to compare policies within countries. For example, data can be collected before and after the implementation of a policy, and some policies vary across states or regions in a country. A workshop participant suggested that, in addition to studying the effects of intentional government policies, it could be worthwhile to study how indirect actions or events can affect population health. For example, government regulation of markets and subsidization of certain products could have big implications for health (e.g., increase in obesity and diabetes associated with subsidization of high-calorie foods).