Subjective social position and cognitive function in a longitudinal cohort of older, rural South African adults, 2014–2019

Background The relationship between subjective social position (SSP) and cognitive ageing unclear, especially in low-income settings. We aimed to investigate the relationship between SSP and cognitive function over time among older adults in rural South Africa. Methods Data were from 3771 adults aged ≥40 in the population-representative ‘Health and Ageing in Africa: A Longitudinal Study of an INDEPTH Community in South Africa’ from 2014/2015 (baseline) to 2018/2019 (follow-up). SSP was assessed at baseline with the 10-rung MacArthur Network social position ladder. Outcomes were composite orientation and episodic memory scores at baseline and follow-up (range: 0–24). Mortality- and attrition-weighted linear regression estimated the associations between baseline SSP with cognitive scores at each of the baseline and follow-up. Models were adjusted for age, age2, sex, country of birth, father’s occupation, education, employment, household assets, literacy, marital status and health-related covariates. Results SSP responses ranged from 0 (bottom ladder rung/lowest social position) to 10 (top ladder rung/highest social position), with a mean of 6.6 (SD: 2.3). SSP was positively associated with baseline cognitive score (adjusted β=0.198 points per ladder rung increase; 95% CI 0.145 to 0.253) and follow-up cognitive score (adjusted β=0.078 points per ladder rung increase; 95% CI 0.021 to 0.136). Conclusion Independent of objective socioeconomic position measures, SSP is associated with orientation and episodic memory scores over two time points approximately 3 years apart among older rural South Africans. Future research is needed to establish the causality of the observed relationships, whether they persist over longer follow-up periods and their consistency in other populations.

According to the ISCO-08, we classified the 30 different types of occupations reported for the respondent's father's occupation into the ISCO-08 major occupation groups, and then into the four skill levels, as follows: We created inverse probability weights that jointly accounted for potential selection bias due to nonresponse to the wave 2 HAALSI interview for two key reasons: 1) mortality between waves 1 and 2, and 2) attrition due to reasons other than mortality between waves 1 and 2 (e.g., refusals to be interviewed or not found for contact).
We first used logistic regression to predict survival between waves 1 and 2 based on the following covariates measured at wave 1: age (continuous), sex (male; female), country of birth (South Africa; Mozambique or other), years of education (continuous), literacy (can read and write; cannot read or write), marital status (never married, currently married or living with a partner, separated/deserted, divorced, or widowed), employment status (employed full-or parttime; not working; homemaker), household per capita consumption quintiles, cognitive function score (composite of orientation, immediate word recall, delayed word recall, and two numeracy items), CES-D depression scale score (continuous), grip strength (continuous; the maximum grip strength value recorded over four measurement sessions), average 2.5 meter walk time (continuous), HIV status based on dried blood spot measures (positive; negative or indeterminate/missing), whether respondent was missing HIV data (yes; no), HIV viral load (0; <100; 100-400; 400-1000; 1000-10,000; >10,000 copies/mL), and whether the respondent had a proxy interview (yes; no). Mean or mode imputation was used to impute missing values for a small number of individuals with missing data on certain variables. The full logistic regression model that was used to predict survival is shown in Supplementary Table 1.
We then used logistic regression to predict non-attrition between waves 1 and 2 based on the following covariates measured at wave 1: age (categorical), sex (male; female), country of birth (South Africa; Mozambique or other), years of education (continuous), literacy (can read and write; cannot read or write), marital status (never married, currently married or living with a partner, separated/deserted, divorced, or widowed), employment status (employed full-or parttime; not working; homemaker), household per capita consumption quintiles, cognitive function score (composite of orientation, immediate word recall, delayed word recall, and two numeracy items), whether the respondent had a proxy interview (yes; no), migration status (measured at wave 2, capturing whether the respondent had moved out of Agincourt but remained in Mpumalanga, migrated to another province in South Africa, or migrated to another country), participation in other local research studies (yes; no), month of first contact for the wave 2 interview, and time of day of first contact for the wave 2 interview. Mean or mode imputation was used to impute missing values for a small number of individuals with missing data on certain variables. The full logistic regression model that was used to predict non-attrition is shown in Supplementary Table 2. The final inverse probability weight was calculated by taking the inverse of each of the survival and non-attrition probabilities estimated by the two logistic regression models and multiplying them together. We considered truncating the weight at the 99 th percentile, but we did not do so because there were no extreme outlying individual weights. This final joint mortality and attrition weight was applied to all linear and quantile models in the present study.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) The weighted distributions are presented as percentages and 95% confidence intervals around the percentages. The weighted estimates were generated using the NIDs wave 4 panel weights, which correct for non-response to the original wave 1 survey, panel attrition between waves 1 and 4, and are calibrated to provincial population totals and to gender-age group-race cell totals.
b Categories for the current employment status variable differed between HAALSI and NIDS. Variable categories that did not apply within each study are indicated as "N/A" cells in the table.

Subjective social position and cognitive function in a longitudinal cohort of older, rural South African adults, 2014-19
11 Note: All models incorporate IPWs for mortality and attrition a Adjusted for Model 1 covariates, plus socioeconomic and social factors (father's occupation, education, literacy, marital status, employment status, household asset quintile) b Adjusted for Model 1 and 2 covariates, plus health-related factors (self-rated health today compared to one year ago, alcohol intake frequency, number of depressive symptoms, diabetes, hypertension) BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)