Cognitive ageing across the lifecourse in a low-income context: evidence from Malawi

Abstract

Background:

We investigate heterogeneity in cognitive ageing trajectories in rural Malawi, a low-income country (LIC) in sub-Saharan Africa (SSA), where current understanding of cognitive ageing is limited.

Methods:

We use longitudinal data on 1,214 adults aged 45+ years from the Mature Adults Cohort of the Malawi Longitudinal Study of Families and Health (MLSFH-MAC) from 2012 – 2022 and employ latent class trajectory models to identify distinct trajectories of cognition. We estimate associations between sociodemographic variables and group membership probabilities using fractional multinomial logit regression models.

Results:

Three cognitive trajectory groups were identified: 36% of the respondents belonged to the ‘high’ cognition group characterised by a stable relative decline in cognition overtime, 46% belonged to the ‘medium’ cognition group experiencing intermediate relative decline by age 80, and 18% of respondents belonged to the ‘low’ cognition group experiencing the largest relative decline in cognition by age 80. Individuals on higher cognition trajectories were more likely to be men, have completed primary or secondary level of schooling, lived in a house with a metal roof, and were born in younger cohorts.

Conclusions:

Heterogeneity exists in cognitive ageing trajectories in Malawi—while some individuals experience healthy cognitive ageing, a nontrivial subgroup also experience accelerated cognitive decline over the lifecourse. The results provide guidance for policymakers and practitioners in developing targeted programs to prevent or reduce the burden of cognitive decline among high-risk groups and seek to enhance healthy cognitive ageing among older adults in Malawi where resources for treatment and care are limited.

Background

The burden of Alzheimer’s Disease and Related Dementias (ADRD) is expected to increase in low-income countries (LICs) in the coming decades and countries in eastern sub-Saharan Africa (SSA) alone are projected to see a 357% increase in dementia cases by 2050.^1–3 This burden poses serious risks to population health and health systems in resource-constrained countries such as Malawi where nearly 70% of the population lives below the poverty line⁴ and experiences significant lifecourse adversity.⁵ Cognitive ageing—defined as the gradual age-related changes in cognition functions⁶—has been poorly understood in rural LIC context as evidence has been missing due to a lack of longitudinal ageing studies in SSA LICs and the fact that findings from high-income countries (HICs) may not apply to LICs given the vastly different socioeconomic and epidemiologic contexts.^4,7 Examination of cognitive ageing trajectories over the lifecourse is critical for distinguishing normal cognitive decline from accelerated cognitive decline that is a precursor to ADRD.⁸ Therefore, a comprehensive understanding of the heterogeneity in cognitive ageing trajectories across the lifecourse in SSA LICs is critical to informing research for facilitating healthy cognitive ageing in this vulnerable population.

Lifecourse literature suggests that cognitive ageing is a lifelong process—cognitive capabilities in older ages are shaped by exposures and experiences accumulated from childhood, throughout adulthood, and into older ages.⁹ Specifically, the cognitive reserve hypothesis proposes that brain functions developed early in life and consolidated in mid-life may buffer against neurodegenerative disease in later life and delay the onset of dementia.¹⁰ In other words, greater cognitive reserve helps maintain cognitive abilities at older ages and slow cognitive decline, which offers a potential explanation for the inter-person heterogeneity in cognitive ageing.

Research examining the heterogeneity in cognitive ageing trajectories using group-based trajectory modelling (GBTM)^11,12 to identify distinct developmental trajectories in cognition is mostly limited to HICs. These studies have identified latent trajectories of cognition over the lifecourse that varied in the proportion of individuals assigned to each class.^13–16 Fewer studies in SSA that examine cognitive impairment and ADRD are limited due to the use of cross-sectional data or data with short-term longitudinal follow-up (with exceptions primarily provided by the HAALSI study in South Africa).^17–20 A related strand of research using the publicly released Harmonized Cognitive Assessment Protocol (HCAP)²¹ battery data for a set of eight countries so far (the United States, Chile, England, Mexico, China, India, South Korea and South Africa), has examined cognitive decline and risk of dementia.^22,23 However, none of the ageing studies so far have examined cognitive ageing trajectories in SSA LICs over the lifecourse partly because of the lack of longitudinal data required for this investigation.

Understanding the heterogeneity in cognitive ageing in SSA LICs such as Malawi is critical for generating context-specific evidence to identify subgroups at risk for ADRD and to develop targeted strategies for ADRD prevention and reduction that would greatly improve quality of life and reduce the economic burden of ADRD in this region. Extant research from HICs suggests that cumulative lifetime adversity and hardship elevates the risk of cognitive impairment and ADRD in later life, thus the magnitude of this risk could be more pronounced for older adults in SSA LICs such as Malawi.^24,25 Prior research from rural Malawi has found that age-related cognitive decline accelerates with age for adults and is already detectable in middle-ages.²⁶ Our study advances this line of research by investigating – for the first time – 10-year trajectories of cognition using longitudinal cognitive assessments with repeated observations using GBTM that provides a unique opportunity to understand the heterogeneity in cognitive trajectories in an ageing SSA population.

Leveraging novel longitudinal data on mature adults aged 45 years and older from 2012–2022 in rural Malawi, our study addresses three specific objectives: 1) analyse the trajectories of cognitive ageing by estimating GBTM for the first time in an African low-income context using an ageing cohort that has been followed up for 10 years, 2) analyse the variation in the pace of cognitive decline across the different cognition trajectories, and 3) estimate the correlates of trajectory group membership probabilities in this study population.

Methods

Study Sample

The study used data from four waves of the Mature Adults Cohort of the Malawi Longitudinal Study of Families and Health from 2012 to 2022 (MLSFH-MAC).^27,28 The MLSFH is a longitudinal population-based cohort study implemented in rural communities in three districts in Malawi-Mchinji in the Central region, Rumphi in the Northern region, and Balaka in the Southern region (for details on regional characteristics see MLSFH cohort profile²⁷ and Supplementary Section 2). The study was initiated in 1998 to investigate fertility and social networks and gradually expanded to cover a broad array of social and contextual determinants of health across the lifecourse. The MLSFH-MAC is a subset of the MLSFH study population that includes adults aged 45 and above that were surveyed since 2012 with an ageing focus with subsequent follow-up waves conducted in 2013, 2017 and 2022. Although not nationally representative, the MLSFH-MAC provides a broad representation of individuals aged 45 years and older living in rural Malawi, where 85% of the national population resides (for details and comparative analyses with nationally representative samples, see the MLSFH cohort profiles^27,28). Supplementary Table S1 summarises the analytic sample of the current study.

The analytic sample was constructed using the following criteria. Inclusion was restricted to participants aged 45–80 years in the MLSFH-MAC baseline year of 2012. Cognition was measured at baseline, and in the few instances where we had missing values in 2012 (52 respondents), the 2013 cognition score was used. Overall, 1,320 participants were enrolled at the baseline of which 1,220 respondents met the age cut-off criteria. After removing 6 respondents with missing cognition scores, the final analytic sample consisted of 1,214 respondents. Figure 1 summarises the sample selection criteria along with the observations dropped at each stage of selection.

Figure 1:

Flowchart of sample selection

Note: Baseline sample was enrolled in 2012. For 52 respondents, cognition score measurements from 2013 were used as their cognition scores were missing in 2012. Out of 1,320 participants who were enrolled at the baseline, 1,220 respondents were between ages 45–80. While respondents above age 45 were targeted, some younger respondents were enrolled, and they have been excluded from the study sample.

Measures

MLSFH Cognitive Assessment (MCA):

The main outcome of interest in the study, the MCA, was adapted to the local cultural context and is based on widely used measures of cognition such as the Mini-Mental State Examination²⁹ and the Montreal Cognitive Assessment³⁰ which are implemented in existing studies on cognition that utilise the HCAP instrument.²¹ MCA covers domains such as orientation, language, episodic memory, executive functioning, and attention and working memory (for details about the MCA see Kohler et al.^26,28). The total score of the MCA ranges from 0–30 with a higher score indicating better performance on the cognitive assessment.

Sociodemographic predictors of trajectory group membership:

We examine the association of key predictors of cognition and cognitive change identified by prior studies with the probability of the trajectory group membership.^13,31 Specifically, our predictors include gender, region of birth, schooling, marital status, as well as whether a respondent lives in a house with a metal roof (a widely-used indicator of wealth for this study population), and birth cohort.

Statistical analysis

Our analyses use a two-step approach. We first use an extended version of GBTM to examine latent patterns in cognitive ageing over time that takes into account non-random attrition due to mortality.^12,13,32 The extended version of GBTM jointly estimates the probability of trajectory group membership and the probability of non-random dropout based on last observed outcome in any preceding wave and adjusts the group membership probabilities accordingly (see Supplementary Section 1 for details on estimation).³² The optimal model was evaluated based on a combination of the Bayesian information criterion (BIC), Bayes factor, average posterior probability of group membership, and the odds of correction classification based on the posterior probabilities (see Supplementary Table S2 for model fit diagnostics).¹² The GBTM analysis was performed using the ‘traj’ package in Stata 18.³³

Next, after estimating the cognition trajectories, fractional multinomial logit (FML) regression models were used to investigate associations between sociodemographic predictors and probability of trajectory group membership.^34,35 Marginal effects evaluated at means were reported to examine the association of characteristics in baseline year 2012 with trajectory group membership probabilities.¹³ The marginal effects of FML model exhibit the zero-sum property—an increase in the probability in one outcome corresponds to an identical decrease in the probability in other outcomes—which makes them a desirable choice in this context.³⁵

Results

Trajectories of cognitive ageing in Malawi

Our GBTM analyses identified three cognition trajectories (high, medium, and low cognition) as the optimal model to delineate unique developmental trajectories of cognition over a decade of observations, shown in Figure 2. Around 36% of the respondents belonged to the ‘high’ cognition group characterised by a cognition score of 25.3 at age 45. The ‘medium’ cognition group contained 46% of the respondents and had a cognition score of 22.6 at age 45. The remaining 18% of respondents belonged to the ‘low’ cognition group with a score of 16.9. Overall, all three groups experienced age-related decline in cognition, however, the pace of cognitive decline – or relative decline – varied significantly across the groups as further illustrated below.

Figure 2:

Cognition trajectories of rural Malawians from 2012–2022 (n = 1,214)

The baseline characteristics of the study sample by trajectory group membership from Table 1 reveals that respondents in the ‘low’ cognition group were mostly women (84%), mostly born in the Southern region (54%) and Central region (32%), had no schooling (73%), and mostly lived in houses without a metal roof (54%). On the contrary, the ‘high’ cognition group consisted of 60% men, were mostly born in the Northern region (51%) and had relatively more schooling (15% had secondary education or higher, 74% attended primary school while only 11% had no schooling).

Table 1:

Baseline characteristics of respondents by trajectory group membership (n = 1,214)

Trajectory	Low Cognition	Medium Cognition	High Cognition	P-value
Group Membership %	18%	46%	36%
Women, n (%)	168 (83.6%)	351 (61.4%)	177 (40.1%)	<0.001
Men, n (%)	33 (16.4%)	221 (38.6%)	264 (59.9%)
Born Central, n (%)	65 (32.3%)	177 (30.9%)	122 (27.7%)	<0.001
Born North, n (%)	27 (13.4%)	166 (29.0%)	224 (50.8%)
Born South, n (%)	109 (54.2%)	229 (40.0%)	95 (21.5%)
No schooling, n (%)	147 (73.1%)	251 (44.0%)	49 (11.1%)	<0.001
Primary Schooling, n (%)	54 (26.9%)	310 (54.3%)	328 (74.4%)
Secondary or above, n (%)	0 (0.0%)	10 (1.8%)	64 (14.5%)
Wealth Indicator: Metal Roof, n (%)	92 (45.8%)	330 (57.7%)	321 (72.8%)	<0.001
Cohort 1949 or before, n (%)	75 (37.3%)	213 (37.2%)	99 (22.4%)	<0.001
Cohort 1950–1959, n (%)	67 (33.3%)	176 (30.8%)	167 (37.9%)
Cohort 1960–1969, n (%)	59 (29.4%)	183 (32.0%)	175 (39.7%)

Note: For factor variables, the P-values from Pearson χ2 tests are reported.

Pace of cognitive decline over the lifecourse

The pace of cognitive decline across the cognition groups is further explicated in Figure 3 and Table 2. Derived from the trajectory plot, Figure 3 examines the relative change in cognition levels across the three trajectories from the cognition level at age 45 for each trajectory. The pace of decline is similar for individuals across all three trajectory groups through age 50, after which they diverge. The heterogeneity in the pace of cognitive decline becomes evident with age: respondents in the low cognition group experience a 39% decline in cognition by age 80 relative to their cognition level at age 45; while respondents in the high cognition group experience a modest 15% decline in their cognition by age 80 relative to their cognition level at age 45.

Figure 3:

Relative change in cognition score relative to cognition at age 45 across the trajectories (n = 1,214)

Note: The y-axis plots the relative change in cognition score at each age j for the i^th trajectory from cognition score at age 45: $\frac{\left(MC{A}_{ij}-MC{A}_{i45}\right)}{MC{A}_{i45}}*100$. Details of the calculations are provided in Supplementary Section 1.

Table 2:

Age at which cognition declines by 5% - 35% from cognition level at age 45 for each trajectory

Trajectory	Low Cognition	Medium Cognition	High Cognition
Group Membership %	18%	46%	36%
% Change in cognition from cognition score at age 45
5%	53	57	65
10%	59	63	73
15%	64	69	80
20%	68	73	-
25%	71	77	-
30%	75	80	-
35%	78	-	-

Note: the percentage change and age reported have been rounded to the nearest whole number. Detailed estimates are provided in Supplementary Table S3.

Table 2 provides a different perspective of this early onset in cognitive decline by showing the ages at which cognition declines by 5%,10%,15%,20%,25%, 30%, 35% from cognition level at age 45. For example, individuals in the low cognition trajectory experience a 25% decline in cognition scores by age 71, this corresponding 25% decline for the medium trajectory group is reached six years later at 77 years old; whereas individuals in the high cognition group never experience a 25% decline even by age 80 (the decline is 15% at age 80). Detailed age profiles of the relative change in cognition are provided in Supplementary Table S3.

Sociodemographic predictors of trajectory group membership probabilities

The marginal effect estimates from the FML regression model of the predictors of trajectory group membership probabilities are shown in Table 3. Women were more likely to belong to the low cognition group (β=0.13, 95% CI: 0.10 to 0.17, P < 0.01) and less likely to belong to the high cognition group compared to men (β=−0.18, 95% CI: −0.24 to −0.12, P < 0.01). Respondents born in the North were more likely to belong to the high cognition group compared to respondents born in the Central (β=0.06, 95% CI: 0.00 to 0.12, P < 0.10). Completing primary school or higher was positively associated with the probability of belonging to the high cognition group compared to respondents who had no schooling (β=0.28, 95% CI: 0.23 to 0.33, P <0.01). Residing in a house with a metal roof was an important predictor of trajectory group membership—those living in a house with a metal roof were more likely to belong to the high cognition group (β=0.13, 95% CI: 0.08 to 0.18, P <0.01) and less likely to belong to the medium (β=−0.06, 95% CI: −0.10 to −0.01, P <0.05) or low cognition group (β=−0.07, 95% CI: −0.10 to −0.04, P <0.01). A null association was observed between baseline marital status and trajectory group membership probabilities. Younger respondents (1960s cohort) were more likely to belong to the high cognition group (β=0.14, 95% CI: 0.08 to 0.20, P <0.01) and less likely to belong to the medium (β=−0.09, 95% CI: −0.15 to −0.03, P <0.05) and low cognition groups (β=−0.05, 95% CI: −0.09 to −0.00, P <0.10) compared to respondents born in 1949 or before.

Table 3:

Fractional multinomial regression analysis for the associations of sociodemographic characteristics at baseline with trajectory group membership probabilities, (n=1,214)

Trajectory	Low Cognition	Medium Cognition	High Cognition
Group Membership %	18%	46%	36%
	Coeff (95% CI)	Coeff (95% CI)	Coeff (95% CI)
Women	0.13 [0.10,0.17]	0.05 [−0.01,0.11]	−0.18 [−0.24,−0.12]
Birth region (Ref: Born Central)
Born North	−0.02 [−0.06,0.02]	−0.03 [−0.09,0.02]	0.06 [0.00,0.12]
Born South	0.03 [−0.01,0.06]	−0.01 [−0.06,0.05]	−0.02 [−0.08,0.04]
Schooling (Ref: No schooling)
Primary Schooling	−0.18 [−0.22,−0.13]	−0.10 [−0.15,−0.05]	0.28 [0.23,0.33]
Secondary or above	−0.26 [−0.30,−0.21]	−0.38 [−0.47,−0.29]	0.63 [0.54,0.73]
Currently Married (Ref: Not married)	0.00 [−0.04,0.04]	−0.05 [−0.10,0.00]	0.05 [−0.01,0.11]
Wealth Indicator: Metal Roof	−0.07 [−0.10,−0.04]	−0.06 [−0.10,−0.01]	0.13 [0.08,0.18]
Birth Cohort (Ref: Born 1949 or before)
Cohort 1950–1959	−0.02 [−0.06,0.03]	−0.08 [−0.14,−0.02]	0.10 [0.04,0.15]
Cohort 1960–1969	−0.05 [−0.09,−0.00]	−0.09 [−0.15,−0.03]	0.14 [0.08,0.20]

Note: Marginal effects evaluated at the mean from fractional multinomial regression models where the probability of individual i belonging to trajectory group j is regressed on gender, region of birth, schooling, marital status, wealth indicator and cohort dummy variables. 95% CI reported in parentheses.

Discussion

Our study makes important contributions to the emerging research on cognitive ageing in a LIC in SSA by estimating cognitive trajectories – for the first time – using a decade of longitudinal data from rural Malawi. Our analyses identify three groups with distinct cognitive ageing patterns in the sample. The ‘high’ cognition group starts with higher cognition, experiences relatively stable ageing and are largely buffered from substantial age-related cognitive decline. Our analyses also identify a nontrivial subgroup in this rural sample that experiences accelerated cognitive decline with age and may be a high-risk group for cognitive impairment and ADRD. Importantly, our research indicates that this early onset of cognitive deficits can already be detected in mid-life. Notably we also find that women, individuals without formal schooling, those living in houses without a metal roof, and those born in older cohorts, are more likely to belong to this ‘low’ cognition group characterized by accelerated cognitive decline with age.

The existence of distinct cognition trajectories in Malawi aligns with results from HICs that have found inter-person heterogeneity in cognitive ageing. A study using Health and Retirement Study (HRS) data found six distinct cognitive classes for adults in the United States³⁶ while another study¹⁴ identified four to seven trajectories per cognitive test in their sample of healthy older adults from Australia and the United States. These findings from HICs, alongside our results from Malawi, underscore that cognitive ageing is a highly heterogeneous phenomenon. Despite socioeconomic contexts that are vastly different from their high-income counterparts, over one third of our rural Malawian sample experiences (relatively) healthy cognitive ageing. Also, consistent with the Flynn effect observed in studies from HICs,³⁷ younger cohorts have a higher probability to belong to the high cognition group compared to older cohorts.

Even though schooling levels are low and school quality is often poor,³⁸ the protective role of schooling on individuals’ cognition are illustrated through the results consistent with the cognitive reserve hypothesis. Even in Malawi, basic schooling provides meaningful protection against later-life cognitive impairment. These findings align with research highlighting education as a key protective factor and underscore low educational attainment as an important modifiable risk factor for later-life dementia even in a LIC setting where schooling levels have been historically low.^31,38,39

While gender differences in cognitive ageing are widely documented, our results reveal a distinctive pattern for this SSA LIC: men were more likely to belong to ‘high’ and ‘medium’ cognition trajectories while women were more likely to belong to the ‘low’ cognition trajectory. This finding differs from research on HICs that suggest women are cognitively better off compared to men.⁴⁰ This may be reflective of gender inequality in this context, particularly for an older cohort whose women have endured fewer education opportunities, early marriage, the burden of high fertility and high maternal mortality, all of which potentially affect cognitive abilities.^26,41 Additionally, being married at baseline was not a significant predictor of cognition trajectory membership compared to those that were not married at baseline—a finding that stands in contrast from HICs.⁴²

The findings from this study contribute to the literature on cognitive ageing in LICs in two distinct ways. First, we provide evidence on heterogeneity in cognitive ageing in a rural SSA LIC. Identifying potential early risk factors to cognitive impairment before pathological symptoms appear, is critical in tackling the rise in ADRD burden among older persons. Rural areas in Malawi have little to non-existent neuroscience infrastructure or healthcare to address cognitive decline or dementia.^4,43 Given Malawi’s limited healthcare resources for ageing-related diseases, high-risk groups may experience worsening cognitive outcomes and not receive timely interventions which may have devastating social and economic consequences in the long run.

Second, the longitudinal design of the MLSFH enhances the relevance of our findings for ageing populations in other LICs in Eastern and Southern African countries where ageing is poorly understood, and as a result, there is poor targeting as resources are not allocated to meet the identified needs of the older population. The absence of longitudinal data on cognition has often limited adding a robust evidence base for designing effective policies to mitigate the consequences of rising dementia burden in this part of the world. A recent World Bank report emphasises that investments and policies that facilitate the maintaining of healthy cognitive functioning into older ages, along with other dimensions of physical and mental health, are central to the economic development of LICs in the next decades.⁴⁴ As some of the first longitudinal evidence on the diversity of cognitive ageing in a SSA LIC, this study provides a glimpse into the future burden of cognitive impairment that is projected to increase in this region. Expanding research on ageing and aligning it with health service planning is essential to ensure evidence-based care for older populations in Malawi.⁴⁵

Some limitations of this study need to be acknowledged. First, while the study population broadly represents the rural population in Malawi (> 85% of the Malawi population is rural),^27,28 results do not capture cognitive trajectories of adults living in urban areas in Malawi. Second, although our estimation model accounts for non-random attrition due to mortality, selection and panel attrition may have biased the generalisability of our findings to some extent.²⁶ Third, latent classes are sample-specific statistical approximations and reflect the patterns of the underlying data. These classes are not universal and may vary across samples. Fourth, the MLSFH-MAC does not yet provide a clinical or research diagnosis of dementia. The MLSFH provides a comprehensive measure of cognitive function rather than an assessment of dementia itself (the latter is forthcoming based on future MLSFH waves).

Key Messages.

• Our study fills a critical evidence gap by providing rare longitudinal analysis on heterogeneity in cognitive ageing trajectories in a sub-Saharan African (SSA) low-income country (LIC) where such insights have been limited by the absence of repeated cognitive assessments.

• Three distinct cognitive trajectories were identified and with a nontrivial share of the older population experiencing accelerated cognitive decline that is detectable in middle-ages.

• Parsing of heterogeneity in trajectories of cognitive ageing helps to identify vulnerable subgroups at risk of experiencing accelerated cognitive decline in this resource-poor context which provides important guidance for policymakers in developing targeted programs to prevent or slow cognitive decline among high-risk groups in Malawi as the country prepares for a surge in dementia cases in the coming decades.

Data availability

Data for replicating the study is available from the authors and will be made publicly available as part of future MLSFH public data releases.