Resilience, Accelerated Aging and Persistently Poor Health: Diverse Trajectories of Health in Malawi

Abstract

Individuals age at vastly different rates resulting in significant within-population heterogeneity in health and aging outcomes. This diversity in health and aging trajectories has rarely been investigated among low-income aging populations that have experienced substantial hardships throughout their lifecourses. Utilizing 2006–2018 data from the Malawi Longitudinal Study of Families and Health (MLSFH) and estimating group-based trajectory models (GBTM), our analyses identified three distinct lifecourse health trajectories: (1) comparatively good initial mental and physical health that persisted throughout the lifecourse (“resilient aging”); (2) relatively good initial mental and physical health that started to deteriorate during mid-adulthood (“accelerated aging”); and (3) poor initial mental and physical health that further declined over the lifecourse (“aging with persistently poor health”). For both physical and mental health, men were more likely to enjoy resilient aging than women. Predictors other than gender of trajectory membership sometimes confirmed, and sometimes contradicted, hypotheses derived from high-income country studies. Our analyses highlight the long arm of early life conditions and gender in determining aging trajectories and show that a non-trivial sub-population is characterized by aging with persistently poor health. The study uncovers widening gaps in health outcomes between those who age with resilience and those who experience accelerated aging.

Introduction

Individuals can age at vastly different rates (Ailshire et al. 2015; Kunkel et al. 2019; Levine and Crimmins 2018), and significant within-population heterogeneity can exist in the onset of diseases and morbidity as individuals get older (Belsky et al. 2020; Schrempft et al. 2021; Vaupel and Yashin 1985; Wrycza and Baudisch 2014). While differential aging trajectories have been extensively studied in high-income populations (e.g., Hoang et al. 2023; Walsemann and Ailshire 2020; Zheng et al. 2021), including linking lifecourse disparities in health to early life contexts or socioeconomic status (Barker 1990; Link and Phelan 1995), the diversity of aging patterns has rarely been investigated among poor aging populations who experience substantial hardships throughout their lifecourses (Baulch and Hoddinott 2000; Castaneda et al. 2018). This research gap between high and low income populations, which is importantly due to a lack of global aging studies covering low-income populations (Figure 1), is noteworthy as aging in low-income countries (LICs) is often accelerated: substantial declines in physical, mental and cognitive health emerge at younger ages (i.e., during mid-adulthood) compared to higher-income countries (HICs) and persist to older ages (Aboderin and Beard 2015; He et al. 2020; Sudharsanan and Bloom 2018; WHO 2015). Evidence increasingly indicates that earlier onsets and faster paces of health decline are distinctive hallmarks of aging in LICs where older adults are often exposed to multiple recurrent adversities throughout their lifecourses (GBD 2019 Disease Collaboration 2020; Kobayashi et al. 2021; Kohler et al. forthcoming, 2017). Little research, however, has addressed the diversity in aging trajectories in LICs. For example, are there groups of individuals who, against the odds and despite extensive lifecourse adversities, retain high and stable levels of physical and mental health into old age? And if so, what are predictors of this “resilient aging” in LICs, and do they differ from predictors of resilience in HICs? Similarly, are other groups of individuals particularly vulnerable to the adverse contexts that characterize LICs, and what are their trajectories of “accelerated aging”? Understanding this potential heterogeneity of health trajectories in LICs is critical for assessing the implications of global aging with rapid growth of older individuals in LIC contexts (NASEM 2019), and it is important for devising health policies that facilitate healthy aging among the global poor, while also providing critical support for vulnerable aging populations in LICs (National Academy of Medicine 2022).

FIGURE 1.

Population-based aging studies by Human Development Index (HDI) and annual growth of the population aged 50+ years: Most aging studies are in relatively high-income contexts where the older populations are not growing very rapidly in global comparison.

Notes: Each point represents a major (and mostly publicly available) population-based global aging study (see Table A.1 for list of studies). Annual growth rate for the 50+ population pertains to the period 2020– 40, based on United Nations World Population Prospects (UN Population Division 2019).

This study is among the first to focus on the diversity of health trajectories in an aging LIC population. Malawi represents an important context in which to conduct these analyses. The country shares common hallmarks with other sub-Saharan African (SSA) LICs such as largely a subsistence-based economy, wide-spread poverty, and a fragile healthcare system that is to a large extent unprepared to address the rapidly growing burden of non-communicable diseases (NCDs) and the needs of the fast-growing older population. The country is predominantly rural with currently only 16% of the population living in urban areas (Malawi NSO 2019). About 15% of the Malawian population is considered “ultra-poor,” i.e., with an estimated food consumption below the minimum level of dietary energy requirements (UNDP 2010). In rural areas, where the study population of the Malawi Longitudinal Study of Families and Health (MLSFH) is based, the majority of individuals engage primarily in home production of crops, complemented by some market and small-scale business activities. Currently, remaining life expectancy at age 45 is ≈ 28 years (UN Population Division 2019) and healthy life expectancy is ≈ 22 years (GBD 2019 Disease Collaboration 2020). Older adults aged 45+ years can expect to live a large proportion of their remaining life expectancy subject to physical and mental health limitations impacting their daily activities and overall wellbeing (Kohler et al. 2022a, forthcoming, 2017, 2022b; Payne et al. 2013). HIV/AIDS remains widespread, with a prevalence of about 7.4% among rural adults 15–49 years old (Malawi DHS 2017). Similar to other SSA countries, the Malawian population is also experiencing a double burden of infectious diseases and increasing NCDs, with the latter accounting for 38% of total deaths, and 59% of deaths above age 45 years (GBD 2019 Disease Collaboration 2020).

Our analyses are based on the MLSFH (Kohler et al. 2015, 2020), thus focusing on a single country. But it is important to emphasize that the MLSFH study population reflects lifecourse experiences and adversities that are typical for older persons in SSA and other LICs characterized by extreme-to-moderate poverty. For example, they have lived most of their lives with per capita incomes of less than $1/day (Malawi NSO 2018). Our older study participants were born when under-5 mortality was almost 30% (UN Population Division 2022) and have survived through frequent adversities as a result of sustained poverty, repeated famines and epidemics, including the HIV/AIDS epidemic that devastated much of sub-Saharan Africa (Ciancio et al. 2022). Fertility of these older individuals was high (Kohler et al. 2012). The status of women was generally low and women often experienced worse mental health at adult and older ages than men (Kohler et al. 2017; Stieglitz et al. 2014). The health-care system continues to be generally underfunded, particularly with respect to prevention, treatment and care for poor mental health and and other non-communicable diseases (Bollyky et al. 2017; Ciancio et al. 2021; Kohler et al. 2017). These socioeconomic contexts experienced by our and other global-poor populations are vastly different from the lifecourse contexts experienced by the populations used for the overwhelming majority of aging studies in Figure 1 (Baulch and Hoddinott 2000; Castaneda et al. 2018).

We applied group-based trajectory models (GBTM; Nagin 2009; Nagin et al. 2018) to the 2006–19 waves of the MLSFH and focused on three pertinent health outcomes—SF12 mental and physical health scores that have been validated in Malawi (Ohrnberger et al. 2020) and body mass index (BMI) that at both high and low levels represents a key risk factor for poor health and mortality in LICs (GBD 2019 Disease Collaboration 2020; Hendriks et al. 2012). Our analyses identified three distinct lifecourse health trajectories: (1) comparatively good initial mental and physical health that persisted throughout the lifecourse (“resilient aging”); (2) relatively good initial mental and physical health that deteriorated during the lifecourse at younger ages (i.e., during mid-adulthood) compared to higher- income countries (HICs) (“accelerated aging”); and (3) poor initial mental and physical health and particularly BMI that possibly further declined over their lifecourses (“aging with persistently poor health”).

These trajectories identified in our MLSFH GBTM analyses are related to important current themes in the literature. Accelerated aging, for example, was initially proposed through Geronimus’ (2006) weathering hypothesis arguing that African Americans experience health declines related to aging younger than Whites due to the cumulative impacts of social, economic, and political disadvantages and exclusion. This notion of accelerate aging has been broadened to denote the substantial declines in physical, mental and cognitive health that occur at younger ages in low-income and/or disadvantaged populations as compared to high-income and/or more-advantaged populations (Noren Hooten et al. 2022). In our analyses, similar to many other studies, we measured accelerated aging by examining disease onsets or aging trajectories (e.g., Aboderin and Beard 2015; Sudharsanan and Bloom 2018), and not via allostatic load, telomere length and/or epigenetic age as in recent biosocial-aging studies (Geronimus et al. 2006; Horvath and Raj 2018; Mather et al. 2011).

In contrast, resilient aging is receiving increasing attention as a key component of “successful” aging in disadvantaged populations. Resilient aging refers to the ability to deal with adversity and maintain (relatively) good health due to an ability to adapt and regenerate in response to health or socioeconomic shocks. Individuals who face adversity or experience functional declines can therefore exhibit resilient aging, as we document in our analyses of health trajectories in Malawi, while these individuals would often not be classified as healthy aging. Specifically, the notion of healthy aging requires maintaining good health and functioning across many domains (Aldwin and Igarashi 2015; Cosco et al. 2017), which is rarely the case for older individuals in Malawi and other LICs. Resilient aging is thus a more realistic aspiration for aging in LICs where high overall disease burdens and frequent lifecourse adversities almost inevitably are associated with lasting deleterious effects on the health and well-being of older persons (Merchant et al. 2022).

Background, Data, and Methods

Background:

A biosocial-lifecourse framework, which has been widely adopted for studies of human development and aging (Crimmins 2020; Glass and McAtee 2006; Harris and McDade 2018) provides the overarching theoretical framework for how lifecourse contexts, behaviors, adversities and shocks affect health as individuals age and reach older ages (Figure 2). Harris and McDade (2018) for instance define this biosocial approach as “a broad concept referencing the dynamic, bidirectional interactions between biological phenomena and social relationships and contexts, which constitute processes of human development over the lifecourse.” Yet, within current global aging research, it is mostly higher-income contexts that have informed our understanding of this biosocial lifecourse framework of health at older ages (Figure 1).

FIGURE 2.

Social and biological, or “biosocial,” influences affecting health and aging trajectories across the lifecourse and into old age

Notes: Adapted from Harris and McDade (2018) and Glass and McAtee (2006)

For our analyses of lifecourse trajectories of health from middle to older ages, this biosocial framework has two important implications: First, aging in rich and poor populations is likely to differ not just in terms of life expectancies and health at various ages (GBD 2019 Disease Collaboration 2020; NASEM 2019). Instead, the biosocial aging process is likely to start diverging in early life, with differences accumulating into old age as a result of the interactions of behaviors, social contexts and the biological embedding of lifecourse adversities (Glass and McAtee 2006; Harris and McDade 2018). Aging inequities and gender differences in LMICs therefore tend to be more acute than in HICs (Economist Impact 2023). Second, the social processes having the largest impact on health trajectories from middle to old ages in LICs might be different from those documented in higher-income contexts (Bell et al. 2019; Halfon et al. 2018). Distinct LIC social contexts, including some that might increase resilience (e.g., social integration of the old), are likely to entail that trajectories of health differ across the middle to late lifecourse, and that associations between these trajectories and socioeconomic predictors potentially differ between LICs and HICs (Castro Torres et al. 2021).

Our analyses can test these hypotheses directly. Prior research using GBTM has shown that there is diversity in aging patterns and health trajectories while mostly studying HICs. For example, Zheng et al. (2021) found seven distinct health trajectories for BMI for individuals between ages 31–80 years in the Framingham Heart Study, including one (with 24% of individuals) that maintained normal weight throughout adulthood. A scoping review by De Rubeis et al. (2021) summarizes the results of 59 studies that estimate BMI trajectories using GBTM. None of these studies focused on LICs let alone sub-Saharan Africa. While the studies differed in the number of trajectories from 2–9, most commonly 4 trajectories were identified as optimal. Most studies found one trajectory to exhibit persistent overweight and some studies identified one trajectory exhibiting low BMI at younger age and increasing to overweight/obesity in later life. None of those studies identified an underweight trajectory. With respect to our study sample, we may expect underweight to be more prevalent since malnutrition and poverty remain key challenges in Malawi.

Musliner et al. (2016) provide a literature review of 25 studies on depressive symptoms trajectories using GBTM. All studies used general population samples. The most common trajectory was characterized by stable low symptoms. Most of the studies also identified a small group that followed a trajectory with persistently high depressive symptoms. Trajectories with increasing or decreasing symptoms were also identified in some studies with a large range in group sizes. Consistently, studies found gender and education as well as income to be significant predictors of group membership: females and individuals with lower education and/or lower income were associated with higher depressive symptoms trajectories. Liang et al. (2011) examined trajectories for depressive symptoms among adults aged 50+ years in the United States Health and Retirement Study (HRS) over an 11-year period, identifying individuals with minimal (16%) and low (26%) initial depressive symptoms that persisted as individuals aged. Many other studies have identified groups of individuals with stable physical- and mental-health trajectories in HICs (Andreescu et al. 2008; Byers et al. 2012; Mallett et al. 2022; Musliner et al. 2016; Song et al. 2018; Yang et al. 2019; Zheng et al. 2013).

The distinct health trajectories identified in HICs, however, may not generalize to LICs as vastly different socioeconomic and epidemiological contexts affect biosocial aging processes across all stages of the lifecourse (Figure 2). But studying global poor populations is important because the world’s aging populations are increasingly concentrated in such contexts and the different socioeconomic contexts may have implications for health across the lifecourse. For example, in our study population, high blood pressure is common among older individuals despite the lack of conventional risk factors such as obesity or access to Western-type diets (Kohler et al. 2020, 2022b; NCD Risk Factor Collaboration (NCD-RisC) 2021). Declines in physical and mental health occur on average in this cohort at younger ages as compared to individuals in HICs (Kohler et al. 2022a, forthcoming), and similar patterns of early health declines have been documented in other populations exposed to significant lifecourse adversities (Geronimus et al. 2006; Phelan and Link 2015; Si- mons et al. 2021). Global aging research is also only starting to document the determinants and correlates of trajectories of later-life health and aging in LICs (NASEM 2019; National Academy of Medicine 2022), with LIC aging studies—such as the MLSFH—allowing important new perspectives to aging research that so far has focused primarily on high and middle income countries (Figure 1).

The Malawi Longitudinal Study of Families and Health (MLSFH):

The MLSFH is one of a few long-standing publicly available cohort studies in the low-income context of sub-Saharan Africa (Kohler et al. 2015, 2020). The MLSFH increasingly serves as an LIC aging study (e.g., Ciancio et al. 2022, forthcoming, 2021; Kämpfen et al. 2020; Kohler et al. 2022a, forth- coming, 2022b; Payne et al. 2018) given increasing age of the MLSFH study population and the expansion of the MLSFH study instruments to include aging-related measures and outcomes (Kohler et al. 2020). The MLSFH is conducted in rural areas in three districts (Mchinji in the Central, Rumphi in the Northern and Balaka in the Southern regions). Between 1998–2021, MLSFH has collected 12 rounds of data with extensive life-course, socioeconomic and health information about the study participants. MLSFH respondents who migrated to urban centers and other locations in Malawi are also followed up longitudinally by MLSFH. Although not drawn as a nationally representative sample in 1998, MLSFH broadly represents the rural population in Malawi, where about 85% of all Malawians reside. The study closely matches the rural sub-sample in the 2010 nationally representative Integrated Household Survey 3 (Malawi IHS 2017) in key observed characteristics. Cohort profiles (Kohler et al. 2015, 2020) provide detailed information about sampling, study instruments, and attrition. Most of the MLSFH cohort members living in these rural areas engage in manual, intensive physical labor such as home production of crops, complemented by some market activities.

While the broad demographic, socioeconomic and epidemiological conditions are fairly similar across the three MLSFH study regions, and also across other parts of rural Malawi, some noteworthy differences across the MLSFH regions include the following. Rumphi (Northern Region) is inhabited primarily by Tumbukas, who are predominantly Protestant, and have a patrilineal system of kinship and lineage where residence is primarily patrilocal. Mchinji (Central Region) follows a less rigid matrilineal system whereby residence may be matrilocal or patrilocal or neither (among MLSFH participants in Mchinji, about 75% follow patrilocal traditions). It is primarily inhabited by Chewas, with almost equal proportions of Catholics and Protestants. Balaka (Southern Region) is primarily inhabited by Lomwes and Yaos and has the highest proportion of Muslims. This region follows a matrilineal system of kinship and residence is mostly matrilocal, although it is not uncommon for wives to live at least some time in their husbands’ villages. The Balaka region also exhibits lower ages of sexual debut and larger number of lifetime sexual partners than the other MLSFH study regions, and residents tend to be less schooled and poorer than those living in the North (see cohort profiles in Kohler et al. 2015, 2020, for additional details).

Strengths of the MLSFH for the analyses in this study include the availability of longitudinally measured health indicators that cover important aspects of physical and mental health. This extensive health information spans over two decades and provides valuable insights from a lifecourse perspective on how individuals age in this context. Moreover, MLSFH also reflects considerable heterogeneity of social, demographic and cultural contexts across rural Malawi that reflect much of the diversity that exists more broadly in sub-Saharan Africa.

Health Measures:

We used the Short Form 12-item (SF12) mental- and physical-health scales, which range from 0 to 100 (worst to best health). The SF12 scales are widely used and validated measures of overall physical and mental health (Gandek et al. 1998; Ohrn- berger et al. 2020). The original SF12 scores were standardized to have a mean score of 50 and a standard deviation of 10 in the general U.S. Population (Ware et al. 2001). In the overall MLSFH population (2010 round), the means are similar with 50.1 (physical; SD = 9.2) and 52 (mental; SD = 9.7). A score of 50 or less on SF12 physical has been recommended as a cut-off to determine the presence of a physical/health condition, while a score of 42 or less on the SF12 mental-health scale may be indicative of clinical depression (Ware et al. 2001). We also used BMI (kg/m2) as another measure of physical health. Both height and weight were measured by trained MLSFH study personnel.

Correlates of Group Membership:

We examined demographic and socioeconomic factors that correlated with the probability of belonging to a particular health trajectory. We focused on predictors of trajectories that are determined in early life, as adult characteristics are endogenous to health and may themselves be the result of health trajectories. The demographic variables included in our models therefore consisted of region of birth and gender. Limited information is available in the MLSFH about childhood contexts, and we are limited to measuring childhood socioeconomic status via an indicator variable for childhood poverty that equals 1 if the primary reason for not attending or completing primary school was that parents could not afford the schooling fees. Otherwise, the indicator variable is 0 including all cases when primary education was completed. Adult socioeconomic status was measured by schooling attainment.

Analytic Sample:

The analysis was restricted to (a) the 2006–2018 MLSFH waves because the SF12 scales were collected from 2006 onward, (b) individuals aged between 20 and 70 years and (c) individuals who had at least two observations on the health measures. Across all waves, the full sample includes 6,498 individuals out of which 5,245 (4,446) provide at least one SF12 (BMI) measurement. Once we restrict the sample to individuals with at least two measurements within the age range 20–70, we obtain an analytic sample of 3,712 individuals for SF12 physical and mental health and 2,366 individuals for BMI. The SF12 (BMI) analytic sample had 3.9 (3.2) observations on average per individual, yielding a total of 14,618 (7,474) person-year observations. Supplemental Figure A.1 shows how our analytic sample differs from the full sample. The second bar in these figures shows the sample size for respondents who have at least one measurement within the ages 20–70. The third bar shows how many individuals have at least two measurements in the relevant age range and the fourth bar additionally restricts to those with non-missing predictors. Figure A.2 in the Supplemental Materials shows the number of observations by age.

Table 1 reports summary statistics of the correlates for the sample used to estimate trajectories of the SF12 physical- and mental-health scores. The summary statistics for the BMI sample are reported in the Supplemental Material (Table A.2). 14% of the analytic SF12 sample and 3% of the BMI sample have missing data on childhood poverty. We use Multiple Imputation, a stochastic imputation based on a logit regression of childhood poverty on all other predictors using non-missing observations. We obtain 100 datasets with imputed values and perform the main estimation with each dataset yielding 100 sets of estimates. These estimates are then averaged to produce our main results that we present here. In the case of SF12 physical and mental health, we impute for 500 individuals whereas for BMI we impute for 99 individuals. Comparing the sample for SF12 physical and mental health with the sample for BMI, the two samples are very similar. 42% of the sample is male and born in one of the three regions in Malawi with roughly equal proportions. Overall, more than half of the sample reported having experienced poverty during childhood: 61% of men and 49% of women grew-up in poor households. A quarter of individuals has less than primary schooling and close to 90% has no more than primary schooling. 30% of women have less than primary schooling and 8% of women have completed secondary schooling or more. In contrast, 17% of men have less than primary schooling and 20% have secondary schooling or more.

TABLE 1.

Summary Statistics for SF12 mental/physical health analyses sample (N = 3, 712)

	Women		Men		Total
	Mean	SD	Mean	SD	Mean	SD	N
Male	0.00	0.00	1.00	0.00	0.42	0.49	3712
Born North	0.30	0.46	0.29	0.45	0.29	0.46	3712
Born Central	0.34	0.47	0.36	0.48	0.35	0.48	3712
Born South	0.35	0.48	0.34	0.47	0.35	0.48	3712
Born Abroad	0.01	0.12	0.01	0.12	0.01	0.12	3712
Childhood Poverty	0.49	0.50	0.60	0.49	0.54	0.50	3212
Less than Primary Schooling	0.31	0.46	0.16	0.37	0.25	0.43	3711
Primary Schooling	0.61	0.49	0.64	0.48	0.62	0.49	3711
Secondary Schooling or more	0.08	0.28	0.20	0.40	0.13	0.34	3711
Mean SF-12 Physical Health Age 20–70	50.30	5.74	52.41	4.87	51.19	5.49	3712
Mean SF-12 Mental Health BMI Age 20–70	52.70	5.99	55.19	5.18	53.75	5.79	3712

Statistical Methods:

The analysis consisted of two parts. We first used GBTM—a specialized application of finite-mixture modeling—to identify latent-health trajectories (Nagin 2009; Nagin et al. 2018). GBTM assumes that the population is composed of a mixture of distinct groups of individuals defined by their developmental trajectories. It identifies groups of individuals following distinct trajectories across ages and estimates trajectory parameters (e.g., shape of trajectory and probability of group membership) separately for each group. This is in contrast to growth curve modeling (Duncan et al. 2013), the most common approach for studying trajectories of health during the aging process, which generally assumes that all individuals in the population follow a similar trajectory that varies around a single mean path. However, the biosocial lifecourse framework in Figure 2 and the related recent literature on the pace of aging suggest that this assumption of growth curve models may be inadequate in capturing the diverse trajectories that characterize the human-aging process and age-related changes in health (Ailshire et al. 2015; Belsky et al. 2020; Levine and Crimmins 2018; Schrempft et al. 2021; Vaupel and Yashin 1985; Wrycza and Baudisch 2014). GBTM avoids this limitation by allowing for different aging trajectories across empirically identified subgroups in a study population, thereby allowing analyses to document more complex heterogeneities in aging patterns that would remain undiscovered by growth-curve-modeling approaches. Additional details about GBTM, including our detailed analytic specification, are provided in the Supplemental Materials.

GBTM uses maximum likelihood to identify groups of individuals with statistically similar trajectories. In our application to the MLSFH, we modelled trajectories with a polynomial function of age (quadratic/cubic) and used the normal distribution for all outcomes. GBTM then provides (1) a predicted shape for each trajectory, (2) group membership probabilities, and (3) for each individual the estimated probability of belonging to each group given realized outcomes (posterior probabilities).

Basic GBTM assumes independence of probabilities of group membership and attrition, leading to biased trajectory group membership when attrition is nonrandom. To overcome this limitation, we used the generalized-enhanced GBTM application developed by Haviland et al. (2011), which models the joint estimation of trajectories and the probability of nonrandom dropout. The dropout model calculates a trajectory-specific dropout probability based on prior wave observations and adjusts group-specific membership probabilities. We defined nonrandom dropout as any instance of survey attrition due to death or permanent migration.¹ We modelled the probability of dropout as a logit function of gender and the last observed outcome prior to dropout. In the sample for SF12 physical and mental health, we find 6% attrition due to death and 4% as a result of permanent migration. In the BMI sample, these numbers are 4% and less than 1% respectively.

To assess the model fit, we estimated various models with two, three and four trajectories with each trajectory as either a quadratic or cubic polynomial in age. We found three trajectories with a quadratic polynomial to deliver the best fit among all specifications, considering jointly all standard criteria such as the Bayesian Information Criterion, the discrepancy between estimated group sizes and assigned group sizes according to the maximum-posterior-probability rule, meaningful group size as well as high average posterior probabilities and odds of correct classification. The criteria for inference about the optimal number of trajectories are laid out in more detail in Nagin (2009). Our model-fit statistics are presented in the Supplementary Materials. The GBTM analysis was carried out in Stata 15 with the traj plugin (Jones and Nagin 2013).

After estimating the trajectories, we computed for each individual the posterior probabilities of belonging to each of the trajectory groups where the probabilities for each individual sum to 1. In contrast to some of the existing GBTM literature, we did not assign individuals to a single trajectory according to the maximum posterior probability rule as this rule ignores the information contained in the posterior distribution. Our approach recognizes that individuals have different propensities of belonging to each of the trajectories, and we utilize this information. Specifically, we estimated fractional-multinomial-logit (FML) regressions to study the correlations of demographic factors and socioeconomic status with the posterior probabilities. The FML is an extension of the multinomial logit to responses that are fractional (between 0 and 1), instead of binary, and sum to one (Mul- lahy 2015). This approach uses the complete information contained in the posterior distributions and avoids the loss of information entailed by assigning individuals to a single trajectory. A further advantage of FML is that the marginal effects sum to zero across the outcomes (in any discrete distribution, a increase in the probability of one outcome needs to be associated with a corresponding decline in the probabilities of other outcomes). Our FML regressions were implemented in Stata 15 with the fmlogit plugin (Buis 2008).

Results

Mental-Health and Physical-Health Trajectories:

Our analyses identified three distinct trajectories for each health measure. For SF12 mental health (Figure 3a), two trajectory groups had relatively good initial mental health, indicated with scores above 55. For 62% of individuals the SF12 mental-health score was stable through adulthood, whereas it declined sharply over the whole age range for another group to which 32% of individuals belonged. We refer to these two trajectories as resilient aging (yellow in Figure 3a) and accelerated aging (blue), respectively. The lowest trajectory group only had 6% of individuals belonging to it, and was less precisely estimated with a wider 95% confidence interval. This group had an initial SF12 score of about 42, which remained low with slight declines over the whole age range; we therefore refer to it as the trajectory group with persistently poor health (red in Figure 3a).

FIGURE 3.

SF12 Mental and Physical Health: Trajectories of health across the lifecourse: resilient aging (yellow), accelerated aging (blue) and persistently poor health (red)

Notes: Trajectories were estimated using Group-based Trajectory Modeling (GBTM; see Supplemental Materials for additional details and specifications). Each trajectory is modelled as a quadratic polynomial in age. The legend labels each trajectory along with the estimated trajectory group size in parentheses.

Trajectories for SF12 physical health (Figure 3b) mirrored those for SF12 mental health. Individuals that followed the trajectory of resilient aging (54%; yellow in Figure 3b) and accelerated aging (39%; blue) both started with good physical health with scores above 50. This relatively good physical health persisted for the resilient-aging group until about age 50, and thereafter declined slightly. For the accelerated-aging group, the SF12 physical health score followed a steep decline as individuals aged and progressed from mid-adulthood to old ages (age 30–70). 8% of respondents followed the persistently poor health trajectory (red in Fig. 3b) that is characterized by already fairly poor health in early/mid adulthood and then subsequent further declines during late adulthood and old age.

For BMI (Figure 4) the two lowest-trajectory groups both started with BMI in the normal range (18.5≤BMI<25) at age 20. BMI for the trajectory group with 55% of individuals steadily declined with age (persistently poor health; red in Fig. 4), whereas it increased with age but remained in the normal range for the trajectory group with 36% of individuals (resilient aging; yellow). The third trajectory group had 9% of individuals. At age 20, BMI was just below 25 and thus in the normal range. There was a sharp increase in BMI between ages 20–50 and individuals on this trajectory were classified as obese (BM≥30) after age 50 (accelerated aging; blue in Figure 4).

FIGURE 4.

Body Mass Index (BMI): Trajectories of health across the lifecourse: resilient aging (yellow), accelerated aging (blue) and persistently poor health (red)

Note: Trajectories were estimated using Group-based Trajectory Modeling (GBTM; see Supplemental Materials for additional details)). Each trajectory is modelled as a quadratic polynomial in age. The legend labels each trajectory along with the estimated trajectory group size in parentheses.

The parameters of the dropout model are reported in Supplemental Table A.6. For SF12 mental and physical health, non-random subject attrition is the highest on the persistently poor health trajectory (13% and 14%) whereas for BMI we find attrition to be the highest on the resilient aging trajectory with 4%. Most estimates of the dropout model are not significant at conventional levels with two exceptions. Higher SF12 mental and physical health significantly decreases the likelihood of dropout on the persistently poor health trajectory.

Predictors of Trajectory-Group Membership:

Marginal effects (evaluated at the means) from FML regressions are presented in Table 2. The predictive power of demographic factors for trajectory-group membership of the SF12 mental- and physical-health components were similar. Men were 11% (13%) more likely to belong to resilient-aging trajectories for SF12 mental (physical) health than women. Region of birth did not predict trajectory-group membership for mental health, but being born in the North Region was associated with a significant 3.4% increase in the probability of following the trajectory of accelerated aging and a 2.7% decrease in the probability of belonging to the trajectory group with persistently poor health for SF12 physical health. Birth region had similar associations with BMI group membership.

TABLE 2.

Associations of Demographic Characteristics and Socioeconomic Status with Trajectory Group Membership

Outcome	SF12 Mental Health			SF12 Physical Health			BMI
Trajectory Group	Resilient Aging	Accelerated Aging	Persistently Poor Health	Resilient Aging	Accelerated Aging	Persistently Poor Health	Resilient Aging	Accelerated Aging	Persistently Poor Health
Group Membership	62%	32%	6%	54%	39%	8%	36%	9%	55%
Male	0.110*** (0.009)	−0.058*** (0.007)	−0.053*** (0.005)	0.126*** (0.009)	−0.062*** (0.008)	−0.064*** (0.007)	−0.106*** (0.017)	−0.102*** (0.012)	0.208*** (0.019)
Born South	0.003 (0.012)	−0.006 (0.009)	0.003 (0.005)	0.010 (0.010)	−0.006 (0.009)	−0.004 (0.008)	−0.032 (0.021)	0.003 (0.013)	0.029 (0.024)
Born North	−0.002 (0.011)	0.001 (0.008)	0.001 (0.006)	−0.007 (0.011)	0.034*** (0.010)	−0.027*** (0.009)	0.014 (0.022)	0.036*** (0.011)	−0.050** (0.024)
Born Abroad	−0.018 (0.046)	0.012 (0.040)	0.005 (0.024)	−0.022 (0.035)	0.002 (0.031)	0.021 (0.025)	−0.028 (0.063)	0.055** (0.028)	−0.027 (0.063)
Primary Schooling	−0.010 (0.010)	−0.006 (0.009)	0.015*** (0.006)	0.031** (0.013)	−0.028** (0.012)	−0.003 (0.008)	0.071*** (0.017)	0.031** (0.013)	−0.102*** (0.020)
Secondary Schooling or more	0.001 (0.015)	−0.014 (0.012)	0.013 (0.010)	0.055*** (0.016)	−0.032** (0.015)	−0.023 (0.016)	0.086*** (0.030)	0.065*** (0.020)	−0.152*** (0.032)
Childhood Poverty	−0.023*** (0.009)	0.014** (0.007)	0.009 (0.006)	0.009 (0.010)	−0.008 (0.009)	−0.001 (0.007)	0.023 (0.018)	−0.003 (0.010)	−0.020 (0.018)
Observations	3712	3712	3712	3712	3712	3712	2366	2366	2366

Note: Marginal effects evaluated at the mean from FML regressions where the probability of individual i belonging to trajectory group j is regressed on gender, region of birth indicators, year of birth indicators, and indicators for child poverty and schooling attainment. Standard errors clustered at the village level.

^***p<0.01

^**p<0.05

^*p<0.10.

The influence of childhood poverty and schooling attainment differed across SF12 outcomes. Childhood poverty was associated with a 2% decrease in the probability of belonging to the resilient-aging trajectory for mental health, and a 1% higher probability of belonging to the accelerated-aging trajectory. Childhood poverty did not statistically significantly predict trajectory-group membership for physical health. More schooling was associated with a higher likelihood of being on the physical-health resilient-aging trajectory, and lower likelihoods of being on the accelerated-aging and persistently poor-health trajectories. For example, individuals who completed primary schooling were 3% more likely to be on the resilient-aging trajectory, and 3% less likely to be on the accelerated-aging trajectory compared to individuals who did not complete primary schooling. For mental health, counter intuitively, individuals who completed primary schooling were 2% more likely to belong to the persistently poor-health trajectory compared to individuals who did not complete primary schooling.

For BMI, men were more likely to belong to the trajectory group of persistently poor health with BMI in the range 19–21 throughout adulthood. Childhood poverty had no significant association with group membership in any of the BMI trajectories. Higher schooling attainment was associated with a higher probability of belonging to the resilient-aging trajectory, and a lower probability of belonging to the persistently poor-health trajectory. In particular, individuals who completed primary school were 7% (10%) more (less) likely to belong to the resilient-aging (persistently poor-health) trajectory than those who did not complete primary schooling. Individuals who completed primary schooling though were also 3% more likely to be on the accelerated-aging trajectory compared to individuals who did not complete primary schooling. In the Supplemental Material Table A.7, we provide a robustness check using only the subsample of individuals with non-missing predictors and without performing multiple imputations. Reassuringly, the results are very similar.

Conclusions

Though on average health deteriorates with age almost universally as a basic feature of the human lifecourse (Kunkel et al. 2019), studies documenting the heterogeneity and diversity of aging patterns have concentrated on HICs (e.g., Hoang et al. 2023; Walsemann and Ailshire 2020; Zheng et al. 2021). A key insight emerging from these HICs studies has been the existence of distinct aging trajectories, wherein some groups of individuals experience worsening health over their lifecourses, while others are resilient and have (relatively) stable health trajectories and thus exhibit healthy—or resilient—aging (Ailshire et al. 2015; Newman and Murabito 2013).

Yet, with most aging studies being conducted in high- and middle-income countries (Figure 1), critical evidence is missing for low-income populations (NASEM 2019) that face vastly different socioeconomic and epidemiological contexts and are affected by high disease burdens across the lifecourse (Castaneda et al. 2018; GBD 2019 Disease Collaboration 2020). Biosocial-lifecourse approaches (Figure 2), which underpin many recent studies of health and aging (Crimmins 2020; Harris and McDade 2018), suggest that trajectories of health and aging in low-income contexts might potentially differ significantly from those that have been documented in HICs. Ultimately, to understand and improve aging among the global poor, and to understand patterns of aging across the spectrum of economic development, analyses need to include studies in low-income contexts. Yet, so far, only very few aging studies focus on the global poor (Figure 1). This paper illustrates some of the novel findings that can emerge from such studies.

Our analyses extended investigations of the heterogeneity and diversity of aging patterns to a LIC to identify if resilient groups of individuals with relatively stable health trajectories also exist in contexts where poverty and adversities are common across the life- course. Whether this is the case is unclear from the previous literature as prior evidence from HICs may not generalize to LICs where individuals face vastly different economic and social contexts, and where accelerated aging entails that declines in physical health tend to occur on average at younger ages compared to HICs. To fill this gap, our study employed GBTM to identify latent mental- and physical-health trajectories from ages 20– 70 for individuals in the MLSFH.

Our analysis indicated three trajectory groups–resilient aging, accelerated aging and aging with persistently poor health—that followed distinctive patterns of health and aging across the adult and older-age lifecourse (For comparison with results from studies in other context see the Background section). For SF12 mental and physical health, the two most-common groups both had trajectories that started with relatively high SF12 scores in early/mid adulthood, with subsequently stark differences emerging during midlife to old age. In a trajectory that we label resilient aging, individuals maintained comparatively good health—measured by physical/mental health SF12—throughout theie life-courses into old age. This pattern was in sharp contrast to the accelerated-aging trajectory on which relatively good early/mid adulthood health was followed by rapidly declining health as individuals aged. For physical- and mental-health (SF12), about 1.6–1.8 times as many individuals in this LIC population belonged to the group following the resilient-aging trajectory as compared to the group with the accelerated-aging trajectory (62% vs 32% for mental, and 54% vs. 39% for physical health). Our finding that the majority of individuals exhibit high stable mental health is in line with studies in HICs (Musliner et al. 2016). The final trajectory, which we label aging with persistently poor health, started with low mental/physical SF12 scores and was characterized by further declines thereafter as individuals got older. Yet, for the SF12 mental and physical health, the group exhibiting this trajectory had the smallest group membership in our analyses (6–8%) for the self-reported outcomes (SF12 mental/physical health).

Similar trajectory patterns were also identified for BMI. However, for BMI, only 9% of individuals were on the accelerated-aging trajectory that was characterized by an increasing BMI with the average reaching obesity by about age 50; 36% were on the resilient aging trajectory with a BMI that remained centered in the normal range throughout the lifecourse. As one might expect given poverty levels and often insufficient nutrition (Malawi NSO 2018), a large fraction (55%) of individuals were on the persistently poor health trajectory for BMI that was characterized by low BMI in early/mid adulthood followed by further declines towards undernutrition as as individuals aged. This fraction is likely to decrease in future cohorts, while the proportion of individuals on accelerated BMI aging trajectories is likely to increase as obesity further spreads in rural LICs (Ford et al. 2017). Our results are partly consistent with results from HIC studies but contrast in some respects (De Rubeis et al. 2021)². For example, many prior studies identified trajectories with persistent overweight, which we did not find. Other studies find one trajectory exhibiting normal weight at younger ages and increasing to overweight/obesity at older ages similar to our results.

Our analyses has limitations. First, we cannot speak to heterogeneity in trajectories over time. Younger individuals are aging in environments that are different from those of older cohorts. While there have been many different changes in economic, environmental and social conditions, it is worthwhile elaborating on the roll-out of antiretroviral therapy (ART) starting around 2008 as a response to the HIV/AIDS epidemic in Malawi. Payne and Kohler (2017) document mortality declines associated with the scale-up of ART. In addition to the direct health effects of ART, it is possible that investments in health also may have responded to ART leading to better health trajectories for the younger population. For instance, a study by Baranov and Kohler (2018) documents increases in savings and human-capital investments after the roll-out of ART driven by reduced perceptions of mortality risk.

Second, we note that BMI is the only objectively measured outcome that we study while SF12 mental and physical health are self-reported. There is a long literature documenting that self-reports have significant predictive power for longer-run health outcomes such as mortality (Idler and Benyamini 1997; Torres-Collado et al. 2022; Wuorela et al. 2020). One concern is that reporting biases could be larger in LICs potentially due to lack of knowledge in identifying health issues and a lack of health-care engagement. Moreover, the three health measures used in our analyses capture only a subset of the health trajectories that are critical for understanding the aging process in LICs. Disease, disability and cognition also are critical dimensions of aging, but lack of longitudinal data covering sufficiently long periods across various stages of the lifecourse prevent us from conducting such analyses at this point.

Third, we do not have physical and mental health data starting in early life and individuals in our sample provide on average between 3–4 measurements within a span of 9–10 years. For this reason, we set the threshold to be included in our analysis sample at 2 measurements minimum. We also have limited demographic and SES measures from early life for predicting trajectory group membership in adulthood. However, no other data from a sub-Saharan African LIC would allow analyses by combining data on early life influences with longitudinal health measures stretching from early adulthood into old age. As more MLSFH and other global aging data becomes available, future research will be able to investigate heterogeneous trajectories—for example, analyzing different trajectories for older and more recent cohorts, as well as to investigate a more comprehensive set of health measures to paint a more complete picture of the aging processes unfolding in LICs.

Fourth, while our analyses documented some distinctive—as well as some similar— patterns of aging and health between LICs and HICs, a conceptual limitation of this paper is that our analyses are correlational and cannot identify the biosocial mechanisms related to why aging trajectories in LICs sometimes differ from those in HICs, and sometimes they do not differ; nor do our analyses identify the mechanisms (Figure 2) underlying why some—but not all—socioeconomic determinants of health in LICs have different associations with lifecourse trajectories of health than in HICs. Such an analysis is beyond the scope of this specific paper, but will be made possible as part of forthcoming MLSFH biosocial research that will integrate measures of biological (epigenetic) aging within the rich socioeconomic data available in the MLSFH.³

Despite these limitations, our analyses make important contributions to our understanding of health and aging trajectories over long segments of the lifecourse for the first time in a low-income context. Specifically, three key insights emerge from the GBTM analyses that identify three distinctive aging trajectories based on the MLSFH (Table 3). First, and noteworthy given the widespread and frequent lifecourse adversities that this LIC population has experienced, a sizable subset—and for SF12 mental and physical health, the majority—of individuals followed the resilient-aging trajectory with comparatively good health from early/mid-life into older ages, indicating a capacity for many to recover from, and/or adjust to, the inevitable lifecourse shocks and other adversities that were common for older cohorts in Malawi and other LICs. Second, accelerated aging was also widespread in this LIC population in terms of self-reported physical and mental health, while in terms of BMI, it was (still) relatively rare as a result of poor average nutrition and the absence of widespread Western diets in this rural Malawian population. Third, only a relatively small fraction of individuals followed the aging with persistently poor health trajectory for self-reported physical and mental health, while for BMI, this trajectory was common.

TABLE 3.

GBTM Health trajectories ages 20–70 identified in the Malawi Longitudinal Study of Families and Health (MLSFH) 2006–19

GBTM Health Trajectory	SF12 Mental Health	SF12 Physical Health	BMI
Resilient Aging	62%	54%	36%
Accelerated Aging	32%	39%	9%
Aging with persistently poor health	6%	8%	55%

Note: Bold/gray background indicates the most common trajectory for each outcome

Hence, not unlike the HICs populations studied in prior trajectory analyses, our LIC study population from Malawi is characterized by distinct and heterogeneous aging trajectories that were identified in all three outcomes analyzed here. Importantly, this diversity of aging patterns implies that persistently poor health or accelerated aging are not inevitable predicaments in this LIC population; in contrast, our findings suggest a sizable subset of individuals enjoyed resilient (“relatively healthy”) aging despite poverty and frequent adversities throughout the lifecourse (Table 3).

In addition, our analyses of trajectory-group membership highlight the long arm of early life conditions and gender in determining aging trajectories (Table 4). The associations we document in this LIC population sometimes confirm, but sometimes contradict, the hypothesized associations based on HIC studies (Column 1 in Table 4). Evidence from HICs generally documents women to have greater longevity but greater morbidity at all ages (Austad 2006, 2011), and based on this HIC experience one might thus expect men to be more likely to experience resilient aging, and less likely to follow the accelerated-aging or persistently poor-health trajectories. Childhood poverty or adversity in HICs are generally found to be risk factors for poor health in later life (Geronimus et al. 2006; Monaghan and Haussmann 2015), thereby increasing the risk of being on the accelerated-aging or persistently poor-health trajectories. Schooling is generally found to be protective (Link and Phelan 1995; Walsemann and Ailshire 2020), thus decreasing the risk of being on these trajectories, while increasing the likelihood of resilient aging.

TABLE 4.

Predictors of Trajectory Group Membership

		MLSFH Trajectory Analyses
	Hypothesized Association (HIC studies) (1)	SF12 Mental Health (2)	SF12 Physical Health (3)	BMI (4)
Resilient Aging
Male	+	+	+	−
Childhood Poverty	−	−
Schooling	+		+	+
Accelerated Aging
Male	−	−	−	−
Childhood Poverty	+	+
Schooling	−		−	+
Aging with persistently poor health
Male	−	−	−	+
Childhood Poverty	+
Schooling	−	+		−

Notes: Color coding indicates if associations in Columns 2–4 are in expected/unexpected direction based on HIC research (Column 1): Blue: association in expected direction based on HIC research. Red: association in unexpected direction based on HIC research

Our analyses documented marked gender differences in group membership and aging trajectories consistent with expectations based on HIC studies; we found that health of men was often better and/or declined less rapidly with age (Columns 2–4 in Table 4). For physical and mental health, for example, men were less likely to be on the persistently poor health or accelerated aging trajectories; for BMI, men were less likely to follow the accelerated-aging trajectory. Related findings document men to have better health than women at older ages in other MLSFH studies of mental health and cognition (Kohler et al. 2020, 2017; Payne et al. 2018). While we cannot identify the specific mechanisms underlying these gender patterns in this paper, lifecourse models of health (Figure 2) suggest that these are likely related to socioeconomic gender inequalities that disadvantage women more in LICs than in HICs and/or due to consequences of high fertility and often poor reproductive health in these cohorts compared to HICs.

In terms of schooling and childhood poverty, our analyses indicate that some well- established associations with later-life health from HIC studies do indeed generalize to LICs (Columns 2–4 in Table 4). In particular, for SF12 mental health, childhood poverty was associated with a lower likelihood of the resilient-aging trajectory. Similarly, for SF12 physical health and BMI, higher adult schooling attainment was associated with increased probabilities of following the resilient-aging trajectories. These associations are consistent with the notion that that SES and schooling are fundamental causes of health disparities (Link and Phelan 1995), and our analyses show that these factors were associated with lifecourse health and aging trajectories also in this LIC population.

However, not all of our findings are in the direction that might be expected based on HIC studies (Columns 2–4 in Table 4). Importantly, for mental health we found that schooling was associated with a higher—rather than a lower—probability of belonging to the persistently poor-health trajectory. For BMI, our analyses show that schooling is predictive of accelerated aging. This finding is also contrary to expectations based on HIC studies, but it is consistent with recent comparative analyses showing that in LICs higher levels of schooling tend to be associated with increased risks of obesity, while in HICs, the schooling–obesity relationship is the opposite (Frankenberg et al. 2016).

The diverse aging trajectories we find are relevant for policymakers and researchers, as they uncover widening gaps in health outcomes between those who age with resilience and those who experience accelerated aging. Our findings also highlight that a non-trivial sub-population is characterized by aging with persistently poor health, that is, experiences relatively poor health from mid-adulthood and never recovers as they get older. In terms of policy-relevance, these observations call for potentially gender-specific policies enabling resilient aging, as well as the strengthening of institutions that aim to improve health during early life to mid-adulthood with the goal that these improvements will persist until older ages.

Data Availability:

The data supporting the findings of this study are available on request from the corresponding author. The data will be made publicly available as per commitment to funders later.