Being overweight or obese prior to, and during, pregnancy can have adverse consequences for women and their babies(Reference Catalano and Shankar1). Among women, the risks include reduced fertility, gestational diabetes or impaired glucose tolerance, lipid disorders, hypertension and pre-eclampsia(Reference Catalano and Shankar1,Reference Catalano2) . Among the offspring, there are risks of fetal macrosomia and obesity in the short and longer term, respectively, as well as of developing cardiometabolic disorders during adolescence and adulthood(Reference Catalano2). Rates of overweight and obesity among women of reproductive age are increasing globally and particularly in middle- and upper middle-income countries(Reference Chen, Xu and Yan3). In South Africa, there has been an increase in female obesity from approximately 33 % in 2005 to 38 % in 2014(Reference Chen, Xu and Yan3). In Soweto, the combined prevalence of overweight and obesity has been estimated at approximately two-thirds(Reference Wrottesley, Pisa and Norris4). A recent Lancet series reported the double burden of malnutrition as a public health problem which has emerged in South Africa since the 1990s(Reference Popkin, Corvalan and Grummer-Strawn5). Specifically, a large proportion of women who are currently of reproductive age would have been stunted, wasted or underweight in childhood and are now susceptible to, or experiencing, overweight and obesity. Early undernutrition, followed by over nutrition in adult life, is the least favourable trajectory for cardiometabolic disease risk across the life course(Reference Wells, Sawaya and Wibaek6).
A food security–obesity paradox has been described in the USA whereby both food insecurity and obesity have been increasing since 1999, particularly among women(Reference Dinour, Bergen and Yeh7). The term ‘food insecurity’ is often understood to mean ‘insufficient access to food’ or ‘hunger’(Reference Jones, Ngure and Pelto8). However, it is also possible to consume sufficient, or even excess, energy yet be unable to consume a diverse, high-quality diet due to lack of availability, access to, or affordability of healthier foods, as well as a lack of time and resources to prepare them. This lack of access to a healthy, diverse diet is sometimes termed ‘nutrition insecurity’ and is characterised by a double burden of overnutrition (overweight and obesity) alongside persisting micronutrient deficiencies such as anaemia(Reference Perez-Escamilla, Bermudez and Buccini9).
In low- and middle-income countries, the likelihood that food insecurity will lead to obesity, rather than undernutrition, is driven by the interplay of several complex mechanisms, including the quantity and diversity of food consumed, as well as the affordability of high energy, processed foods and the level of access to nutritious food(Reference Farrell, Thow and Abimbola10). Thus, the relevance of the food security–obesity paradox to low- and middle-income settings, where the capacity of health systems is least equipped to respond to the consequences, requires further investigation(Reference Farrell, Thow and Abimbola10). Central to the growing burden of obesity in middle-income settings, including South Africa, is the nutrition transition, whereby traditional, indigenous diets are replaced with frequent consumption of processed high fat, salt and sugar foods and drinks(Reference Popkin, Corvalan and Grummer-Strawn5,Reference Popkin11–Reference Steyn and McHiza14) . Such foods are often cheaper and more accessible than healthier foods such as fruit, vegetables, dairy foods, pulses, fish, lean meats, nuts and seeds. Also, they often require less time and resources for preparation.
In terms of the measurement of overweight and obesity, these are typically assessed using body mass index (BMI). This approach is convenient in large-scale studies, but does not consider the distribution of fat, or differentiate between fat and lean mass. For example, a waist circumference ≥ 80 cm can be indicative of central adiposity whereby fat is stored around the abdominal organs and is associated with cardiometabolic risk(Reference Lee, Huxley and Wildman15). Alternatively, a person with a high proportion of lean mass could also have a high BMI, for example, an athlete. Metabolically and functionally, the consequences of either of these would be very different, which is why BMI is limited. The proportion of an individual’s adiposity and lean mass is also associated with disease risk(Reference Wells16).
The objective of the present study was to determine whether food security, diet diversity and diet quality are associated with anthropometric measurements and body composition among non-pregnant women of reproductive age enrolled in the Healthy Life Trajectories Initiative (HeLTI) study in Soweto, South Africa(Reference Grobler, Visser and Siegfried17). The association between food security and anaemia prevalence in these women was also tested.
Methods
Study setting
The HeLTI study was initiated in 2016 in South Africa, Canada, China and India in response to the need for preconception interventions in low- and middle-income countries. It aims to implement interventions during preconception, pregnancy, infancy and childhood, to optimise women’s physical and mental health, reduce childhood obesity and the risk for cardiometabolic disease, and improve child development(18).
The HeLTI site in South Africa is in Soweto, Johannesburg. Soweto is a large urban area situated in the mining belt of Johannesburg, consisting of both formal and informal housing (shacks). The population was 1·27 million in 2011 with a population density of 6357 people per km2(19). Over the past decade, economic change and a rapid emergence of fast-food outlets have transformed the landscape of food availability and choice.
Participants
Non-pregnant women aged 18–25 years were recruited between June 2018 and June 2019 from randomly selected clusters in Soweto through home visits conducted by trained fieldwork teams. Women with a medical history of type 1 diabetes, cancer or epilepsy were not eligible for the study.
Data collection
The present study uses data from the Soweto Young Women’s Health Survey which was the screening and baseline data collection tool for the HeLTI study in Soweto. Household-level data collection was conducted in participants’ homes. The young women were then invited for individual-level data collection at the South African Medical Research Council (SAMRC)/Wits Developmental Pathways for Health Research Unit, located within the Chris Hani Baragwanath Academic Hospital in Soweto (the largest hospital in the southern hemisphere). Food security and dietary data were collected by an interviewer-administered questionnaire. Body size and composition data were collected using anthropometry and dual-energy x-ray absorptiometry (DXA). Anthropometric data collection followed World Health Organization (WHO) anthropometry standards, and all research assistants were trained by trainers from WHO. Haemoglobin (Hb) levels (g/dl) were assessed using a HemoCue, and anaemia was classified as a Hb level of <12·5 g/dl based on the WHO’s altitude-adjusted threshold(20,Reference Silubonde, Baumgartner and Ware21) .
Food security
Food security was assessed using a questionnaire based on the tool developed by the Community Childhood Hunger Identification Project which has been described in detail elsewhere(Reference Wehler22–Reference Coates, Swindale and Bilinsky24). Specifically, three questions were asked regarding whether the respondent’s household experienced any occurrence of being unable to buy food as follows:
-
‘Does your household ever run out of money to buy food?’
-
‘Do you ever cut the size of meals or skip meals because there is not enough money for food?’
-
‘Do you go to bed hungry because there is not enough money to buy food?’
Women who answered ‘no’ to all three questions were categorised as ‘food secure’. If a woman answered ‘yes’ to one of these three questions, they were categorised as ‘at risk’, and if they answered yes to two or more of the three questions, they were categorised as ‘food insecure’.
Diet diversity and diet quality
Diet diversity and quality were assessed using a dietary practices questionnaire. This questionnaire asked women about the occurrence and frequency of consumption of specified food or beverage groups during the previous day and month. Specifically, diet diversity was assessed by asking whether the women consumed foods from the following fourteen groups on the day before the interview (grains, orange vegetables, white roots and tubers, dark green leafy vegetables, orange fruit, other fruit, other vegetables, organ meat, other meat or poultry, eggs, fish or seafood, beans or peas, nuts or seeds, milk or milk products). Food groups were based on the Minimum Dietary Diversity Score for Women produced by the Food and Agriculture Organization (FAO) and United States Agency for International Development (USAID’s) Food and Nutrition Technical Assistance III Project, which is designed primarily to derive a diet diversity score as an indicator of dietary macro- and micronutrient adequacy(25). While this document is based on ten mutually exclusive food groups, the food group list was adapted into fourteen food groups for this population in order to consider context-specific characteristics as recommended by the FAO(26) and more accurately capture diversity within core food groups. For example, extensive research in this setting using quantitative food frequency questionnaires shows that intake of refined grains is common, while availability and intake of vitamin-A-rich vegetables and white roots and tubers are less so(Reference Wrottesley, Pisa and Norris4,Reference Wrottesley, Micklesfield and Hamill27,Reference Zingoni, Norris and Griffiths28) . Thus, distinguishing between these categories is important in describing dietary diversity and micronutrient adequacy in this setting. All items consumed on the previous day were coded ‘1’, with a code of ‘0’ given for all items that had not been consumed. A diet diversity score was then calculated by summing the responses(Reference Steyn, Nel and Nantel29). A maximum score of fourteen indicated maximum diet diversity, and women were divided into tertiles based on their diet diversity score.
Diet quality was assessed according to the frequency of consumption of the above fourteen food and beverage groups during the past month. In addition, women were asked about the frequency of consumption of processed meat, fried snacks, savoury snacks, bakery items, sweets, fizzy drinks and condiments. The possible responses for frequency of consumption were: ‘every day’, ‘2–4 times per week’, ‘5–6 times per week’, ‘once per week’, ‘less than once per week’, ‘never’. A diet quality score was created based on the method described by Imamura et al. (Reference Imamura, Micha and Khatibzadeh30). Briefly, we computed frequency scores based on responses to eighteen of the food items. For the following ‘healthy’ foods, more frequent intakes scored more highly: orange vegetables, dark green leafy vegetables, orange fruit, other fruit, other vegetables, organ meat, other meat or poultry, eggs, fish or seafood, beans or peas, nuts or seeds, milk or milk products. For the ‘unhealthy’ foods, more frequent intakes scored low: processed meat, fried snacks, savoury snacks, bakery items, sweets and fizzy drinks. The maximum diet quality score of 126 would be achieved if a woman consumed all of the ‘healthy’ foods on a daily basis and none of the ‘unhealthy’ foods. If a woman consumed all of the ‘unhealthy’ foods on a daily basis and consumed the healthy foods less than once per week, she would have a score of 0. For data analysis, the diet quality score was used as a continuous variable.
Body size and composition
All measurements were taken by trained members of research staff. Height was measured to the nearest 0·1 cm using a Holtain wall-mounted stadiometer (Holtain Ltd), and weight was measured to the nearest 100 g using SECA scales (SECA). BMI was calculated as: weight (kg)/height (m2). Measurement equipment was calibrated daily. Mid-upper arm circumference (MUAC) was measured half way between the shoulder (acromion) and elbow (olecranon), and waist circumference was measured halfway between the iliac crest in the mid-axillary plane and the lowest rib margin according to standard protocols. All measurements were taken in triplicate, and the mean values were calculated for use in analyses. Fat and lean mass were measured using DXA whole body scans (Hologic Inc.) and were analysed as ‘whole body less head’. DXA scanning was conducted by trained radiographers on a Hologic machine and following daily quality control procedures. % fat mass was calculated using the DXA values in kg divided by total weight. Fat mass index (FMI) and lean mass index (LMI) were calculated by dividing the mass in kg by height in metres squared.
Covariates
Socio-demographic data on age, duration of education in years, parity (number of live births) and household socio-economic status were collected by questionnaire. Socio-economic status was assessed using an asset score which summed the number of assets owned in the household from the following options: TV, car, washing machine, fridge, phone, radio, microwave, cell phone, DVD/Video, DSTV (cable channel), computer, internet access and medical aid. This asset score was based on standard measures used in the Demographic and Health Surveys household questionnaire (available at: www.measuredhs.com) and has been extensively utilised in this setting(Reference Kagura, Adair and Pisa31,Reference Griffiths, Johnson and Cameron32) .
Data analysis
Variables that were not normally distributed were log-transformed (weight, BMI, waist circumference, fat mass, lean mass, % fat mass, % lean mass, FMI, LMI) for analysis. These were all dependent variables, and therefore, the results relating to these variables in regression models should be interpreted in terms of percentage points. Descriptive statistics were computed for participant characteristics and dietary intakes. Food security and diet diversity scores were derived and categorised. A diet quality index was constructed and analysed as a continuous variable. The relationships between food security and diet diversity were examined using χ 2 tests, and the associations between food security and diet quality using univariate linear regression. Associations between food security category and body composition outcomes, as well as between diet quality and body composition outcomes, were examined by creating dummy variables and using univariate and multivariate linear regression models, the latter adjusted for age, parity, education status and socio-economic status. Analyses were completed using SPSS v25 (IBM Corp) and StataSE version 16.0 (StataCorp).
Results
Table 1 presents the characteristics of women included in the analysis. The flow of participants through the study to reach the final sample size is depicted in Supplemental Figure 1. The median duration of women’s education was 12 years. Of the thirteen assets used to assess socio-economic status, the median score was 8. The median BMI was within the normal range, but approximately 44 % of women were overweight or obese, and almost 9 % were underweight. The median (IQR) % fat mass of normal-weight women was 38 (35, 42) % compared with 32 (29, 34) % in underweight, 46 (43, 48) % in overweight and 51 (48, 53) % in obese women (data not shown). Half of the women were nulliparous and approximately 30 % were anaemic.
Table 1 Characteristics of non-pregnant women of reproductive age enrolled in the Healthy Life Trajectories Initiative (HeLTI) study
BMI, body mass index; FMI, fat mass index; IQR, interquartile range; LMI, lean mass index; MUAC, mid-upper arm circumference.
* Asset score out of a maximum of 13 items.
† Indicates those variables described as mean (sd).
‡ Whole body less head.
Table 2 presents data on the percentage of women who consumed any item from each of the food groups on the day before the interview, as well as data on the frequency of consumption over the past month. The majority of women reported consuming grains on the previous day. Approximately two-thirds of women consumed ‘other than vitamin-A rich’ fruit, milk and fizzy drinks, and about half of the women ate orange vegetables, tubers, other vegetables, processed meat, fried snacks, savoury snacks and sweets on the previous day. The median (IQR) diet diversity score was 6 (4, 9) with the maximum score being 14.
Table 2 Dietary intake, diversity and quality of study participants
The majority of women ate grains and condiments on a daily basis. Approximately half of the women consumed orange fruit, beans and fish/seafood less than once per week, and two-thirds consumed nuts and seeds less than once per week. The majority ate non-processed and processed meat at least twice per week. Organ meat was consumed at least once per week by over two-thirds of the women, and approximately one-third of women consumed fizzy drinks at least once per day. The median (IQR) diet quality score was 48·0 (44·0, 52·0), where a score of 126 would be the maximum.
Table 3 shows that women who were classified as food insecure and at risk of food insecurity tended to consume less diverse diets (i.e. have diet diversity scores in the lowest tertile). In addition, based on a univariate linear regression model, food security category was positively associated with diet quality score; that is, food insecure women had the lowest quality diets and food secure women had the highest quality diets (B = –0·35, 95 % CI –0·70, –0·01, P = 0·049; data not shown).
Table 3 Diet diversity scores according to food security categories
* P value relates to χ 2 test for difference between diet diversity groups.
The relationships between anaemia and food insecurity prevalence rates according to BMI categories are presented in Table 4. While the prevalence of anaemia was highest in underweight women (34 %) and lowest in obese women (28 %), there were no significant differences in anaemia prevalence according to BMI status. Irrespective of their nutritional status, over half of the women were either at risk of, or experiencing, food insecurity in their household.
Table 4 Anaemia and household food insecurity prevalence rates according to body mass index (BMI) categories
Underweight, BMI < 18·5 kg/m2; normal weight, BMI 18·5–24·9 kg/m2; overweight, BMI 25·0–29·9 kg/m2; obese, ≥ 30·0 kg/m2.
* P value relates to χ 2 test for difference between BMI categories.
† Includes mild to severe anaemia; n 1513.
Table 5 presents the associations between food security category and body size and composition using log regression models. The coefficients should be interpreted in terms of percentage points. Food security was not associated with measures of size including weight, MUAC and waist circumference. During univariate analyses, there was a significant association between food security and lean mass, such that women who were at risk of food insecurity had a lower absolute lean mass than those who were food secure. However, relative fat mass was not associated with food security nor was indices of fat mass. There was a borderline association between food security and lean mass index such that women at risk of food insecurity had a lower LMI than those who were food secure. There were no associations between food security and BMI, fat mass index or LMI in the adjusted multivariate models (adjusted for age, parity, education status and socio-economic status) (data not shown).
Table 5 Univariate associations between food security category* and measures of body size and composition
BMI, body mass index; FMI, fat mass index; LMI, lean mass index; MUAC, mid-upper arm circumference.
* Independent variable was food security in 3 categories, the reference category was ‘food secure’.
† Coefficients represent associations with log-transformed variables and should therefore be interpreted in terms of percentage points.
‡ Whole body less head.
There were no statistically significant associations between either diet diversity score or diet quality score and measures of body size or composition in the univariate analyses (data not shown). In addition, multivariate models (adjusted for age, parity, education status and socio-economic status) showed no association between diet quality and body size or composition outcomes (BMI, fat mass index or LMI).
Discussion
The present study aimed to investigate associations between food security, diet diversity and diet quality and anthropometric measurements and body composition among women of reproductive age in the urban township of Soweto, South Africa. The association between food security and anaemia prevalence was also tested. We found that women who were the least food secure had the least diverse diets, with 46 % of food insecure women consuming diets in the lowest tertile of dietary diversity compared with 39 % of those who were food secure. Food insecurity was also associated with lower diet quality. Irrespective of their BMI, food insecurity and anaemia affected approximately a third of women. In addition, women who were the least food secure had a lower absolute lean mass, but there were no associations between food security or diet and measures of adiposity in models adjusted for socio-demographic factors.
To date, there has been little research in urban South Africa on the links between food security and diet quality(Reference Symington, Baumgartner and Malan33). However, there is evidence that in poorer settings, both adults and children consume monotonous diets mainly comprised processed cereals and lacking in fruit, vegetables and other nutritious foods(Reference Wrottesley, Micklesfield and Hamill27,Reference Walsh and Van Rooyen34–Reference Kesa and Oldewage-Theron36) . The paradox is that these diets are often high in energy, refined carbohydrates and sugar and therefore may be linked with obesity, pregnancy disorders such as gestational diabetes and long-term cardiometabolic risk among women and their children(Reference Prioreschi, Wrottesley and Draper37). Furthermore, in some settings, there are issues around intra-household allocation of food with women and girls having reduced access to any nutrient-rich foods that are available(Reference Hanson, Bardsley and De-Regil38). Our study shows that women of reproductive age in Soweto consume poor-quality diets with grains likely to comprise a dietary staple for most households due to their affordability and wide availability. Findings also indicate a tendency towards frequent consumption of highly processed high sugar and fat foods, as well as low intakes of nutrient- and protein-rich foods such as vegetables, fruit, seafood, pulses and nuts/seeds.
Our data also suggest that underweight and obesity may have common underlying causes, both resulting from the consumption of poor-quality diets and often co-existing with micronutrient deficiencies and conditions such as anaemia. While undernutrition may be the outcome of energy- and nutrient-inadequate diets, obesity is likely caused by high intakes of inexpensive, energy-dense, but micronutrient-poor, foods. Both of these scenarios are the result of poverty and limited financial resources which restrict the ability to afford a healthy and diverse diet. This provides the potential for introducing common interventions, such as those focused on improved diet quality, irrespective of body weight. However, while poverty and unstable incomes are likely to play a large part in the link between food security and diet diversity and quality, there may be other confounding factors among those who are food insecure. These could include: time and resources available to prepare food; increased convenience and availability of ultra-processed foods; taste preference, particularly if any ‘healthier’ options such as fruit and vegetables are not fresh or have been poorly stored; marketing and advertising of ultra-processed foods or an unhealthy food environment(Reference Kehoe, Dhurde and Bhaise39,Reference Wrottesley, Bosire and Mukoma40) (Ersze et al., Public Health Nutrition, in press). The food environment in settings such as Soweto has to date been under-studied and will be a focus of our future research. However, a previous qualitative study in this setting showed that young women living in Soweto feel that, while unhealthy, energy-dense foods are easily accessible, affordability and availability, limit their access to healthier food items(Reference Ware, Prioreschi and Bosire41).
In terms of the link between food security and body size and composition, our findings were somewhat unexpected. Our data indicated that food security may be associated with greater lean mass; however, these associations were no longer evident after accounting for overall body size. In addition, there was no association between diet quality and any of the outcomes. Previous research findings are inconsistent which may be explained by the extent of the nutrition transition in different settings(Reference Gowda, Hadley and Aiello42–Reference Gulliford, Mahabir and Rocke44). For example, in lower income settings where the quantity of energy consumption is inadequate, it is more likely that chronic energy deficiency will result from food insecurity. In contrast, in higher income settings where energy intakes are plentiful, but nutrition security may be an issue, overweight and obesity are more likely to be a problem. A cross-sectional study in Connecticut, USA, among parents and children found that 50 % of households were food insecure and adults in these households were more likely to be obese than those who were food secure. In addition, having an obese parent in the household significantly increased the risk of being an overweight child – suggesting that the relationships between food insecurity and obesity may track into the next generation(Reference Martin and Ferris45). In South Africa, where two-thirds of women enter pregnancy overweight or obese, targeting this intergenerational cycle of obesity through improving household-level food and nutrition security and optimising body size and composition prior to conception is critical.
In order to assess cardiometabolic risk, it is important to study longitudinal associations between food security and body size and composition(Reference Dinour, Bergen and Yeh7). While not possible from our cross-sectional data, a previous National Health and Nutrition Examination Survey (NHANES) study in the USA assessed food security and followed participants up for 1 year. Women who were food insecure were more likely to be obese at baseline than those who were food secure. In addition, women in food insecure households were likely to gain approximately 4·5 kg more weight than those from food secure households(Reference Wilde and Peterman46). This highlights the need to understand how the observed relationships between food security and diet track over time, as well as how these factors may be associated with body size and composition in the longer term.
Strengths and limitations
A major strength of our approach was our access to a large, high-quality data set. The availability of detailed body size and composition measurements enabled us to study body size, composition and distribution of body fat as outcomes. This level of detail is limited in large-scale studies and is likely to be important in understanding the mechanisms associated with cardiometabolic disease in transitioning societies. However, the data we analysed were cross-sectional, and thus it is not possible to determine the direction of the associations between the exposure and outcome variables in this study. It will be of interest to determine whether there are longitudinal associations between food security, diet and body composition outcomes in the future and the HeLTI study platform provides an opportunity to do so. The age range in this study was relatively narrow with women recruited between 18 and 25 years. Although almost half of the women were overweight, the variability in body composition outcomes is likely to be lower than among the full range of reproductive age, since age is positively associated with adiposity.
In future research, it would be beneficial to collect more detailed quantitative dietary data in order to calculate energy and nutrient intakes. In addition, assessing other elements of diet, for example, calculating a dietary inflammatory index, may be useful in assessing the associations between diet and cardiometabolic risk(Reference Phillips, Chen and Heude47).
Public health implications and future research
It has been previously estimated that over a quarter of South African households are at risk of food insecurity(Reference Shisana, Labadarios and Rehle48). Specifically, research suggests a shift in poverty and poor nutrition from rural to urban areas, with greater prevalence of food insecurity in urban settings(Reference Walsh and Van Rooyen34). In the recently published Lancet series on the double burden of malnutrition, it appears that the prevalence of undernutrition has declined in South Africa but that of overweight and obesity is increasing. The authors argue that it is important to understand where the burden of overweight is occurring and how it is likely to affect the health of those living in poverty now and in the future. Particularly in South Africa, the proportion of women and children who are undernourished has reduced in the past 20 years, but maternal overweight and obesity are a major public health problem(Reference Symington, Baumgartner and Malan33). Given the link between food security and diet diversity and quality, strategies aimed at improving household food security in urban areas such as South Africa may be important in establishing healthier dietary habits, as well as nutritional and metabolic profiles in the long term. Food environments in particular should benefit from public policies aimed at restricting the availability and affordability of energy-dense, ultra-processed foods and increase access to minimally processed and fresh foods, especially in low-income areas such as Soweto. In addition, it is important to understand how other aspects of lifestyle – for example, women’s changing roles which require them to work outside the home and place greater demands on their time - influence demands for ultra-processed energy-dense foods(Reference Popkin, Corvalan and Grummer-Strawn5).
Future research aimed at reducing the burden of cardiometabolic disease in low- and middle-income countries would benefit from focusing on the link between food security and diet and aiming to fully understand the determinants of dietary diversity and diet quality. Complex long-term interventions that are both nutrition-specific and sensitive are likely required to improve maternal and child health outcomes.
Acknowledgements
Acknowledgements: We would like to thank the participants for their time and interest in the study. We also thank the project staff and the Improved Nutrition Pre-conception, Pregnancy and Post-Delivery (INPreP) collaboration for its support. Financial support: This study was funded by South African Medical Research Council and Canadian Institute for Health Research. This research was funded by the National Institute for Health Research (NIHR) (17\63\154) using UK aid from the UK Government to support global health research. The views expressed in this publication are those of the authors and not necessarily those of the NIHR or the UK Department of Health and Social Care. The funding agencies were not involved in the study design, data analysis or writing of this article. Conflict of interest: None. Authorship: S.A.N. and S.L. designed the study. S.A.N., S.L., A.P., L.W. and C.D. developed the research methods and implemented the study. S.H.K. and S.V.W. analysed the data and all co-authors assisted in the analysis plan and/or interpretation of the findings. S.H.K. and S.V.W. wrote the manuscript which has been reviewed and approved by all co-authors. Ethics of human subject participation: This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all research involving study participants was approved by the University of the Witwatersrand’s Research Ethics Committee (Medical; approval number M1811111). Written informed consent was obtained from all participants.
Supplementary material
For supplementary material accompanying this paper visit https://doi.org/10.1017/S136898002100046X
