Body composition and physical activity as mediators in the relationship between socioeconomic status and blood pressure in young South African women: a structural equation model analysis

Objectives Varying hypertension prevalence across different socioeconomic strata within a population has been well reported. However, the causal factors and pathways across different settings are less clear, especially in sub-Saharan Africa. Therefore, this study aimed to compare blood pressure (BP) levels and investigate the extent to which socioeconomic status (SES) is associated with BP, in rural and urban South Africa women. Setting Rural and urban South Africa. Design Cross-sectional. Participants Cross-sectional data on SES, total moderate and vigorous physical activity (MVPA), anthropometric and BP were collected on rural (n=509) and urban (n=510) young black women (18–23 years age). Pregnant and mentally or physically disabled women were excluded from the study. Results The prevalence of combined overweight and obesity (46.5% vs 38.8%) and elevated BP (27.0% vs 9.3%) was higher in urban than rural women, respectively. Results from the structural equation modelling showed significant direct positive effects of body mass index (BMI) on systolic BP (SBP) in rural, urban and pooled datasets. Negative direct effects of SES on SBP and positive total effects of SES on SBP were observed in the rural and pooled datasets, respectively. In rural young women, SES had direct positive effects on BMI and was negatively associated with MVPA in urban and pooled analyses. BMI mediated the positive total effects association between SES and SBP in pooled analyses (ß 0.46; 95% CI 0.15 to 0.76). Conclusions Though South Africa is undergoing nutritional and epidemiological transitions, the prevalence of elevated BP still varies between rural and urban young women. The association between SES and SBP varies considerably in economically diverse populations with BMI being the most significant mediator. There is a need to tailor prevention strategies to take into account optimising BMI when designing strategies to reduce future risk of hypertension in young women.

The use of structural modelling allowed us to explore direct and indirect (mediation) effects of social economic status, physical activities and body mass index on elevated blood pressure from representative sample of rural and urban population of South African young women. 2. Although the urban and rural cohorts were from two different studies, the same research unit conducted both studies and, therefore, the methodology was harmonized between the two sites, thereby allowing for accurate comparison.

1.
Other unmeasured data, such as undernutrition in infancy, and dietary patterns were not included in the current analyses. We are currently working on research to address this limitation. 2. The low reliability of self-report data on physical activity could introduce bias. Thus, there is need for more precise, objective measures of physical activity to strengthen the results of our analysis. 3. There is need to do comparison on longitudinal data, especially as the socioeconomic environment is changing rapidly due to rural-urban labor migration and other factors would be helpful to examine these associations over time. High blood pressure (hypertension) is a leading risk factor contributing to the global disease burden, accounting for 7% of global disability-adjusted life years (DALYs) and contributing to the 34.5 million non-communicable disease (NCD) related deaths in 2010 [1,2]. A recent global meta-analysis, involving 19.1 million individuals, reported that on average there has been a decrease in blood pressure globally, but the low-to middle-income countries (LMICs) have seen an increase in hypertension [3]. The prevalence of high blood pressure in LMICs is estimated at 30% [4,5] and it is the most significant risk factor for cardiovascular disease, most notably stroke [6]. In 2000, hypertension was estimated to have caused 9% of all deaths and over 390 000 DALYs in South Africa. Further, hypertension contributed to 50% of all strokes and 42% of ischaemic heart disease (IHD), signifying a substantial public health burden [7]. A systematic review of sub-Saharan African (SSA) data shows prevalence rates of hypertension of up to 41% with higher prevalence rates noted in urban compared to rural populations [8,9]. A recent study in men and women aged 40 to 60 years of age in six sites across four SSA countries, including South Africa, showed the same trend with South African urban and rural cohorts having the highest prevalence (41.6 to 54.1%) [10].
Low and middle-income countries are experiencing both epidemiological and nutritional transitions with urban populations further along the transition as demonstrated by the higher prevalence of obesity and NCDs [4,5,8,[10][11][12][13][14][15]. Some evidence has shown that there are differences in the levels of blood pressure between rural and urban settings [8], while other studies have found no significant differences [16]. According to Glass and McAtee, internal biological systems are sculpted by an interaction between genes and prolonged exposure to particular external environments, a principle they call embodiment [17]. Thus the differences in built and social environments between rural and urban settings may explain the differences in disease prevalence. A Ghanaian study showed that both systolic and diastolic blood pressure were significantly lower in rural participants compared to urban participants [18]. However, a similar study in adolescents found that blood pressure levels were only lower in rural boys, with no difference in the girls [19]. Pediatric and adolescent hypertension have been reported to track into adulthood in a South African urban population [20]. Results on elevated blood pressure from studies in rural South African children have reported prevalence rates varying from 1.0% to 25.4% [21][22][23][24]. The factors explaining these differences have not been fully studied in LMICs.
Socioeconomic factors such as education, household income and household assets have been associated with blood pressure levels [25][26][27]. In a US cohort of young adults, a higher household income remained associated with lower systolic blood pressure (SBP) even after controlling for all potential covariates including age, sex and bio-behavioral factors [28]. Similarly, in a French sample of 30-79 year olds, SBP independently increased and was inversely associated with both individual education and residential neighborhood education [29]. Studies in African countries have also found varying associations between SES and blood pressure patterns, with both positive and negative associations reported [8,30,31]. Some studies have speculated that the association between SES and body mass index (BMI), physical activity levels, diet, smoking, alcohol intake and malnutrition may influence blood pressure patterns [18,28,31,32]. Physical activity has been inversely associated with blood pressure and BMI directly associated with BP in more advanced economies, but inconsistent associations have been reported in LMICs [25,[33][34][35][36][37].
There is a need to examine blood pressure and its determinants in young South African adults given the high rates of overweight and obesity and hypertension observed in this age group [20,38]. Recent South African reports also indicate that the highest pregnancy rates occur in the age range of 20-24 years, with 26.2% of births reported, followed closely by the 25-29 year age group (25.7%) [39], and therefore targeting young adult women would also reduce adverse health outcomes in their children. To better target policies or programmes in future to address hypertension and obesity in the different settings, it is important to examine more closely rural-urban differences in hypertension due to differences in the epidemiology of obesity, SES divergence in the South African context [23,26,30,[40][41][42][43]. Therefore, this study aims to compare blood pressure between rural and urban young adult South African women, and to determine whether there is an association between SES and blood pressure and whether it is mediated physical activity and BMI.

Study sample and site
The rural Agincourt site, 2016 potential the female participants between the ages of 18 and 23 years were in the existing Agincourt Health and Socio-demographic Surveillance System database [44]. Only 996 were located during the data collection period and were invited to participate, of these 509 female participants were recruited after giving consent to participate. The urban sample consisted of 510 young women between the ages of 21 and 24 years who were randomly selected from the sample of 720 females who were part of the Birth-to-Twenty plus (BT20+) Young Adult Survey [45,46]. Young women (n=51) who were pregnant at the time of the study were excluded.
Measurements and questionnaires were completed by trained research assistants and nurses, and were standardised between both sites, to eliminate biases. The study protocols were approved by the Human Research Ethics Committee of the University of the Witwatersrand (Clearance certificates M120138 for the Ntshembo-Hope Cross Sectional Survey in Agincourt and M111182 for the BT20+ survey). A written consent to participate was provided by participants and mentally or physically disabled women were exluded from the study.

Blood pressure
Blood pressure (mm Hg) was the outcome variable and it was measured using an Omron 6 automated machine (Kyoto, Japan). A five minute seated rest was observed before taking the blood pressure measurements. Participants' seated blood pressure was measured three times on the right side, with a 2 min interval between each measurement.
The mean for the second and third readings was recorded for the current analysis.
According to the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and   Table 1. These cut-offs were utilized in the current study. Prehypertension and hypertension were combined to create a new variable called elevated BP. Systolic Blood Pressure was used in structural equation models (SEM) as it is more relevant in adults, and a good predictor of adverse health outcomes later in life [48], such as CVDs.

Anthropometry
At both sites, participants' height and weight were measured by trained research assistants using standard techniques [49,50] . Weight was measured in light clothing and barefoot to the nearest 0.1 kg using a digital scale (Tanita model TBF-410; Arlinghton Heights; USA). Height was measured barefoot to the nearest 0.1 cm using a stadiometer (Holtain, Crymych, UK). Waist circumference was measured with a non-stretchable fibreglass tape at the level of the umbilicus. Body mass index (BMI) was calculated as weight/height 2 (kg/m 2 ).

Socio-economic status (SES)
Physical assets owned in the participants' household were used as a proxy for socio-economic status index [51]. It was generated by summing the number of assets owned in the household from the following: television, car, washing machine, fridge, phone, radio, microwave, cell phone, DVD/Video, DSTV (cable channel), computer, internet, medical aid. Previous studies in this population have shown that the sum of physical assets (household assets) is closely related to the household per capital expenditure and household income [51][52][53]. The household SES is regarded as a good measure of accumulated household wealth so it is a more reflective wealth index than income of a household's wealth over time.

Statistical analyses
Analysis of variance and student's t test, and Chi-squared tests and Wilcoxon rank sum test for non-parametric variables, were conducted to compare study characteristics between urban and rural young women. Structural equation modeling (SEM), with missing data option, was used to test and estimate the direct and indirect associations between variables, most especially the mediation roles of physical activity (MVPA) or sedentary time (sitting), and body composition (BMI and waist circumference), in the association between SES and blood pressure (systolic blood pressure).
Direct, indirect and total effects were computed and recorded, and the proportion of the total effect mediated was calculated. To evaluate the best fitting model for our data, we calculated different goodness of fit indices including Chi-squared test, Root mean squared error of approximation (RMSEA), Comparative fit index (CFI), Tucker-Lewis index (TLI) and Standardized root mean squared residual (SRMR) [55]. Though the Chi-squared test has been popularly used as a goodness of fit index, it has been reported to be biased and not reliable as the only goodness of fit index. It is also highly sensitive to sample size [56,57] , and often inflated with non-normal data such as physical activity data and we therefore employed the Hu and Bentler's Two-Index Presentation Strategy (1999) combination rule, with cut off values depending on the fitness index, to determine the best model fit [55,58].
If the direct and indirect effects had opposite signs (negative or positive effects) the proportion mediated was assessed using the absolute values for all indirect and direct effects [59]. All the analyses were conducted using STATA (version 13.0; STATA Corp., College Station, TX, USA).

Study characteristics
Descriptive statistics for the non-pregnant study participants (urban, n=492; rural, n=476) are presented in Table 2.
There was no difference in BMI or waist circumference between the urban and rural participants, but the prevalence of overweight and obesity was significantly higher in the urban (46.5%) compared to the rural young women (38.8%). Household SES was significantly higher in the urban compared to the rural group. Self reported physical activity (total MVPA) was significantly higher in the rural than urban women (p<0.001), and the urban women spent significantly more time sitting than their rural counterparts (p<0.001). Systolic and diastolic BP were significantly higher in the urban group, as was the prevalence of elevated BP (27.0 vs. 9.3%).

Structural equation models for BMI and waist circumference
Results from the SEMs for SES associations with SBP via MVPA and BMI are presented in Tables 3a, 3b and 3c for urban, rural and combined analyses respectively, and also shown in Figures 1,2,3. No significant direct or indirect effects via (MVPA or BMI) of SES on SBP were observed in either the urban or rural women, but there were significant direct effects of SES on MVPA. Results showed that individuals with a higher SES index were less likely to be physically active in pooled data and urban women. In rural women, a one-unit increase in total household assets was associated with a decrease of 0.65 mmHg (-1.19 to -0.10) in SBP and an increase of 0.27 kg/m 2 in BMI (0.1 to 0.53) (Tables 3a, 3b and Figures 1, 2). The SEM for the combined sample showed a significant indirect effect of household SES on SBP via BMI, with 50% of the total effect being mediated by BMI (Table 3c and Figure 3). Direct positive effects of BMI on SBP were observed in both settings and the pooled sample with a 1 kg/m 2 increase in BMI being associated with an increase of 0.37 mmHg (0.21 to 0.53) and 0.33 (0.12 to 0.54) mmHg SBP in urban and rural young women, respectively. Similar results were observed when including waist circumference as the body composition indicator (data not shown).

Discussion
A rising prevalence of hypertension has been reported in South Africa, with Peer et al. reporting a higher prevalence in 2008 (35.6%) compared to 1990 (21.6 %) in men and women aged 25-74 years in an urban black community in Cape Town, South Africa [40]. We have shown in young adult women from urban and rural South Africa, an overall elevated BP prevalence of 18.4 % (27.0 % in urban and 9.3 % in rural). We have also shown a direct effect of BMI on SBP in the urban and rural women separately, as well as when pooled, thereby providing further evidence of an association between overall adiposity and blood pressure. The total effects of SES on SBP were the same in both settings.
Prevalence data on elevated BP and hypertension from other countries in sub-Saharan Africa have shown conflicting results when comparing urban and rural communities. In Malawi, a higher prevalence of hypertension in urban compared to rural communities has been reported and attributed to differences in lifestyle as rural communities participate in subsistence based agricultural activities while the urban community has a more westernized lifestyle with higher salt intake and physical inactivity [9]. Similarly, data from Ghana have shown a higher mean SBP and DBP and a higher prevalence of hypertension in urban communities [18,60] Our study showed significant differences in SES between the urban and rural samples, as well a big variation in SES between these two settings. The social patterning of CVD risk factors, including hypertension, in SSA and LMICs has in part been attributed to differences in countries' socioeconomic development. Previous results from five countries, (two high income and three LMICs), reported that hypertension and other CVD risk factors were substantially associated with education and wealth status; individuals with less education and lower wealth generally showing higher prevalence of CVD risk factors [62] . The effect of SES in this study is most evident in the rural women for whom household SES was lower (compared to urban) and who may be transitioning faster (both nutritionally and economically) than the urban women. Though SES is positively associated with BMI in rural young women, it is negatively associated with SBP. There may be other factors, such as physical activity due to agricultural activities or dietary patterns, which were not recorded. In addition, the weight gain observed might not be due to fat mass but rather to muscle mass and bone mass, which has been reported to be associated with SBP before [63].
In Mexico, women in rural and upper SES categories were likely to have a higher SBP, while we have reported that a higher SES was associated with a decrease in SBP in rural communities. At population level, there is a need to consider different SES categories and monitor the effect of transitioning from one category to another on hypertension, since these categories may respond differently to an increase or a decrease in their SES. Kagura and colleagues tracked SES in South African children and reported that moving from the low SES in infancy to a higher SES in adolescence had a protective effect on SBP level in young adulthood [26]. Our results have shown that this could be more pronounced in rural areas.
We observed a positive association between SES and BMI in the rural sample and the same direction of effects was observed in the urban (though not significant), which is in line with results reported in many LMICs including South Africa, but in contrast with those reported in higher income populations [33,34,62] Africa and within the SSA region [11,33,40,42,67,68]. This link was consistent in rural, urban and combined data sets, indicating the importance of BMI in the aetiology of blood pressure. Munthali et al reported that the link between obesity and hypertension could be observed as early as five years of age. Children with early onset of obesity were at higher risk of developing hypertension in late adolescence [38].
In this study, using SEM models to explore the mediation role of BMI and PA helped quantify potential contributions of these variables to the effect of SES on SBP. The results show that PA was not a significant mediator in the association between SES and BP in the urban or the rural samples. SES was negatively associated with MVPA in urban and pooled samples, indicating that as individuals transition from low to higher SES, they reduce their physical activity level. We speculate that these differences in the association between SES and SBP in both our rural and urban results and in those from high-income countries are due to differences in levels of nutritional and epidemiological transition in these regions [69,70]. Those with low SES in high-income countries are likely to consume cheaper, more energy dense foods, participate in less leisure time physical activity and be more sedentary [71,72] In LMICs, agricultural activities remain a part of everyday life and a day-to-day activity in rural living, while those with higher SES in the same settings rapidly adopt the westernized life style with less PA, fewer agricultural activities and home grown food. However, this speculation is not supported by the data on PA in this study despite the rural participants having a higher PA. Our understanding of the Agincourt rural economy is that agriculture is quite a minor aspect though very useful to augment the household income.
The limitations of this study are that other unmeasured data, such as undernutrition in infancy, which is a known risk factor for high blood pressure later in life [73], and dietary patterns were not included in the current analyses. We are currently working on research to address this limitation. We can also not rule out the role of genetics. Secondly, the low reliability of self-report data on physical activity could introduce bias. Thus, there is need for more precise, objective measures of physical activity to strengthen the results of our analysis. Lastly, longitudinal data, especially as the socioeconomic environment is changing rapidly due to rural-urban labor migration and other factors would be helpful to examine these associations over time.

Conclusions
Though the prevalence of overweight or obesity is relatively higher in both rural and urban than those reported in other SSA countries, women in the urban setting were at more risk for elevated blood pressure than their rural counterparts. The link between socioeconomic status and SBP varies in a more economically diverse population, as

Conflict of interest
Authors have no conflicts of interest to disclose.

Consent for publication
Not applicable

Par 3
Generalisability 21 Discuss the generalisability (external validity) of the study results Discussion: Page 9 Par 3 -Page 11 Other information Funding 22 Give the source of funding and the role of the funders for the present study and, if applicable, for the original study on which the present article is based Funding: Page 12 Par 1 *Give information separately for cases and controls in case-control studies and, if applicable, for exposed and unexposed groups in cohort and cross-sectional studies.

90
Saharan Africa. Therefore, this study aimed to compare blood pressure (BP) levels, and investigate the extent to 91 which socioeconomic status (SES) is associated with blood pressure, in rural and urban South Africa women.

93
Setting Rural and urban South Africa.

97
Participants Cross-sectional data on SES, total moderate-vigorous physical activity (MVPA), anthropometric and 98 blood pressure were collected on rural (n=509) and urban (n=510) young black women (18-23 years age). Pregnant 99 and mentally or physically disabled women were excluded from the study.   The use of structural equation modelling allowed us to explore direct and indirect (mediation) effects of social economic status, physical activity and body mass index on elevated blood pressure from a representative sample of rural and urban populations of South African young women. 2. Although the urban and rural cohorts were from two different studies, the same research unit conducted both studies and, therefore, the data collection and management process were consistent between the two sites, thereby allowing for accurate comparison.

1.
Other unmeasured data, such as undernutrition in infancy, and dietary patterns were not included in the current analyses. 2. The low reliability of self-report data on physical activity could introduce bias. Thus, there is need for more accurate, objective measures of physical activity to strengthen the results of our analysis. 3. There is a need to do comparison on longitudinal data, especially as the socioeconomic environment is changing rapidly due to rural-urban labor migration and other factors would be helpful to examine these associations over time.   Blood pressure (mm Hg) was the outcome variable and it was measured using an Omron 6 automated machine 214 (Kyoto, Japan). A five minute seated rest was observed before taking the BP measurements. Participants' seated BP 215 was measured three times on the right side, with a 2-minute interval between each measurement. The mean for the 216

286
Descriptive statistics for the non-pregnant study participants (urban, n=492; rural, n=476) are presented in Table 2.

287
There was no difference in BMI or waist circumference between the urban and rural participants, but the prevalence 288 of overweight and obesity was significantly higher in the urban (46.5%) compared to the rural young women 289 (38.8%). Household SES was significantly higher in the urban compared to the rural group. Self-reported MVPA 290 was significantly higher in the rural than urban women (p<0.001), and the urban women spent significantly more 291 time sitting than their rural counterparts (p<0.001). Systolic and diastolic BP were significantly higher in the urban 292 group, as was the prevalence of elevated BP (27.0 vs. 9.3%).

295
Results from the SEMs for SES associations with SBP via MVPA and BMI are presented in Tables 3a, 3b and 3c for   296 urban, rural and pooled analyses respectively, and also shown in Figures 1, 2

331
Our study showed significant differences in SES between the urban and rural samples, as well a big variation in SES 332 within these two settings.

350
We observed a positive association between SES and BMI in the rural sample and the same direction of effects was 351 observed in the urban, though not significant. This is in line with results reported in many LMICs including South

Other unmeasured data, such as undernutrition in infancy, and dietary patterns
were not included in the current analyses.
2. The low reliability of self-report data on physical activity could introduce bias.
Thus, there is need for more accurate, objective measures of physical activity to strengthen the results of our analysis.

204
Recent South African reports also indicate that the highest pregnancy rates occur

328
Results showed that individuals with a higher SES index were less likely to be physically active 329 in pooled data and urban women. In rural women, a one-unit increase in total household 330 assets was associated with a decrease of 0.65 mmHg (95% CI: -1.19 to -0.10) in SBP and an 331 increase of 0.27 kg/m 2 in BMI (95% CI: 0.1 to 0.53) (Tables 3a, 3b and Figures 2, 3). The  (Table 3c and

369
Our study showed significant differences in SES between the urban and rural samples, as well               Interpretation 20 Give a cautious overall interpretation of results considering objectives, limitations, multiplicity of analyses, results from similar studies, and other relevant evidence Page 9 Par 3 -Page 11