Neighbourhood socioeconomic characteristics and blood pressure among Jamaican youth: a pooled analysis of data from observational studies

Introduction

Neighbourhood characteristics, in particular conditions of the physical and social environment, have been shown to be associated with a number of health-related conditions (Arcaya et al., 2016; Diez Roux, 2016; Diez Roux & Mair, 2010; Duncan & Kawachi, 2018; Oakes et al., 2015; Pemberton & Humphris, 2016; Ribeiro, 2018). Research in this area has grown exponentially in the last twenty years, with obesity and depression being among the conditions most often studied (Arcaya et al., 2016; Diez Roux, 2016; Diez Roux & Mair, 2010; Oakes et al., 2015). Other outcomes studied include cardiovascular health, allostatic load, diabetes, dietary practices, physical activity, alcohol consumption and tobacco use (Airaksinen et al., 2015; Assari et al., 2016; Diez Roux, 2016; Ribeiro et al., 2019; Unger et al., 2014). Neighbourhood characteristics assessed include poverty, deprivation, walkability, food environment, healthy food availability, fast food density, air pollution, traffic noises, social cohesion, crime, victimization, fear of neighbourhood violence, and levels of urbanization, among others (Airaksinen et al., 2015; Arcaya et al., 2016; Assari et al., 2016; Diez Roux & Mair, 2010; Jivraj et al., 2019; Ribeiro et al., 2019; Unger et al., 2014). In general, these research studies have found that poorer neighbourhood conditions are associated with worse health outcomes (Diez Roux, 2016).

Underlying mechanisms possibly include interactions between adverse environmental exposures, reduced sense of safety due to potential exposure to crime and violence, increase levels of adverse health behaviours and increased psychosocial stress (Diez Roux & Mair, 2010). The effects of neighbourhood on health can be conceptualized as being either compositional or contextual (Macintyre & Ellaway, 2003). The compositional effects refer to the differences in the characteristics of the people living in the neighbourhood, while the contextual effects are due to differences between the places (Macintyre & Ellaway, 2003). Contextual effects include collective social functioning and social practices (Macintyre, Ellaway & Cummins, 2002). Overall, McIntyre and colleagues have proposed that features of the local area that influence health include: physical features of the environment, availability of healthy environments, services provided to support people in their daily lives, sociocultural features of the neighbourhood and the reputation of the area (Macintyre, Ellaway & Cummins, 2002).

With regards to blood pressure, associations have been reported for systolic blood pressure, diastolic blood pressure and hypertension (Coulon et al., 2016; Fan et al., 2015; Liu et al., 2013), using indicators such as community level income, poverty, social and environmental characteristics and ‘perceived neighbourhood crime and satisfaction’. Additionally, associations have been reported between attenuation of the normal nocturnal decrease in blood pressure (dipping) and higher levels of perceived neighbourhood problems (Euteneuer et al., 2014). Studies have also shown associations between neighbourhood characteristic and blood pressure in youth (McGrath, Matthews & Brady, 2006). Possible mechanisms contributing to the effects of neighbourhood on blood pressure include poor dietary behaviour, increased obesity, higher consumption of alcohol and increased psychosocial stress (Chaix et al., 2010; Chaix et al., 2008; Euteneuer et al., 2014; Mujahid et al., 2011).

In Jamaica, research on the association between neighbourhood and health is limited. Cunningham-Myrie and colleagues found associations between characteristics of the neighbourhood environment and physical activity and obesity, with significant differences by sex (Cunningham-Myrie et al., 2015), while Mullings and colleagues found neighbourhood associations with depressive symptoms, again with significant sex differences (Mullings et al., 2013). Two other studies were identified, one reporting an association between neighbourhood crime and childhood malnutrition (Thompson et al., 2017) and the other reporting associations between perceived neighbourhood characteristics and depressive symptoms among adolescents in Jamaica and other Caribbean islands (Lowe et al., 2014). More recently, associations have been reported between neighbourhood disorder and cumulative biological risk and substance use (Cunningham-Myrie et al., 2018; Felker-Kantor et al., 2019).

To our knowledge, no previous study on the relationship between blood pressure and neighbourhood characteristics has been conducted in Jamaica. Research on the neighbourhood effects on blood pressure is also limited in developing countries, among black populations and among youth. Studies have shown that social determinants of health often vary between countries and in different social context (Leng et al., 2015; Mackenbach et al., 2008), suggesting that findings are often not generalizable across countries or regions. Given the high burden of hypertension in developing countries and black populations (Gakidou et al., 2017; NCD Risk Factor Collaboration, 2017), and the fact that blood pressure tracks across the life course from childhood into adulthood (Chen & Wang, 2008), studies on the neighbourhood effects of blood pressure in these settings and among youth will further enhance our understanding and inform public health decisions. In light of the foregoing, this study forms part of a larger body of work, exploring social and biological determinants of blood pressure in adolescents and young adults in a predominantly black population from a developing country (Ferguson et al., 2015; Ferguson et al., 2018). Previous studies evaluated individual and household socioeconomic status but did not address the neighbourhood context.

This study therefore aimed to:

1. Evaluate the relationship between neighbourhood characteristics and systolic and diastolic blood pressure among Jamaican youth, 15–24 years old.

2. Evaluate whether the presence of elevated blood pressure or hypertension (BP ≥ 120/80 mmHg) is associated with neighbourhood characteristics.

Estimates will adjust for potential confounders including age, sex and household socioeconomic status and will also evaluate the effect of possible intermediary variables and independent covariates. We hypothesized that lower neighbourhood SES will be associated with higher systolic and diastolic blood pressure and higher prevalence of elevated blood pressure or hypertension. We chose to assess elevated blood pressure or hypertension as our high-risk category since the study involved adolescents and young adults and prevalence of hypertension would be relatively low in this age-group.

Methods

Results

Final analyses included 2,556 participants (1,446 females; 1,110 males) from 306 communities. Mean age was 17.9 years (standard deviation 2.0). There was evidence for significant clustering of both systolic BP and diastolic BP within communities, with intra-class correlation coefficients of 10.7% (95% CI [6.8–16.6%]) and 10.9% (95% CI [7.0–16.6%]), respectively.

PCA yielded two components with adjusted eigenvalues >1.0 and these were used to derive two neighbourhood SES variables. These two variables were standardized using z-scores for analyses. The first component, PCA-SES1, had an adjusted eigenvalue of 3.21 and explained 54% of the variance; this component loaded highly positive for tertiary education and larger house size (higher value = higher SES). The second component, PCA-SES2, had an adjusted eigenvalue of 1.40 and explained 24% of the variance; this component loaded highly positive for unemployment and population density (higher value = lower SES). Figure S1 shows the plot for the observed and adjusted eigenvalues for the principal components, while Table S3 shows the eigenvectors (factor loadings) for the individual variables included in PCA-SES1 and PCA-SES2. The investigators assessed the PCA-SES components as having good face validity, in that communities with high scores for PCA-SES1 were those generally accepted as high SES in the local context and those with high scores on PCA-SES2 were those accepted as having lower SES.

Table 1 shows summary statistics for characteristics of all study participants combined and for each of the three studies included in the analysis. Mean age ranged from 16.5 to 19.8 years and was highest for JHLS-II. While there were statistically significant differences for participant characteristics between studies, most of these differences were small, except for waist circumference which showed an approximately six-point difference, with mean value of 77.1 cm in JHLS-II compared to 71.6 in JYRRBS. Mean z-score for PCA-SES 1 was highest in the 1986-JBCS (0.37) and lowest in the JYRRBS (-0.12), suggesting that participants from the 1986-JBCS had higher overall SES. This was supported by the distribution of individual community SES characteristics as shown in Table S2. Participants in the 1986-JBCS came from communities with larger populations, greater population density, less poverty, and lower dependency ratio. Mean z-score for PCA-SES 2 was also highest in the 1986-JBCS. Prevalence of elevated blood pressure or hypertension was 29% overall and ranged from 21% in the 1986-JBC to 37% in JHLS-II (see Table 1). There were also significant differences in the distribution of persons in BMI categories, thirds of household possessions, fast food consumption categories and physical activity levels. Additional data with summary statistics for participant characteristics, stratified by study and sex, are shown in Table S6. There were significant sex differences for most of the participant characteristics.

Table 2 shows the means or percentages for participant characteristics within blood pressure categories, defined as normal (<120/80 mmHg) and elevated BP or hypertension (≥120/80 mmHg). Participants with elevated BP or hypertension were taller and heavier and had higher mean BMI and waist circumference. For the categorical variables, elevated BP was significantly associated with sex, BMI category, and physical activity level.

As stated before, we assessed whether the relationship between systolic BP and the PCA-SES variables showed evidence for non-linear effects by adding a quadratic term to the model. This was non-significant for PCA-SES1, but significant for PCA-SES2 (p < 0.001); we therefore used linear splines in the bivariate and multivariable analyses for PCA-SES2. We also assessed for interaction by sex and study. Tests for sex interaction were significant for the relationship between systolic BP and some covariates (age, BMI, PCA-SES2 at spline 4, and fast food consumption category 3–4 times/week), therefore we present sex-specific analyses for bivariate and multivariable models. Details of the sex interaction effects are shown in Table S5. There was no evidence for interaction by study in the relationship between systolic BP and PCA-SES1. For PCA-SES2, there was some evidence for interaction at spline 1 and 4 for females and spline 3 for males (see Table S7). Given these interactions with study, in addition to our sex-specific models in our primary results, we show study-specific estimates in Supplemental Information 1.

Bivariate analyses using mixed effects linear regression with systolic BP as the outcome are shown in Table 3. PCA-SES1 showed a statistically significant inverse association with systolic BP among both males and females. A one standard deviation unit increase in PCA-SES1 score was associated with a 1.45 mmHg lower systolic BP among males (p < 0.001) and 1.30 mmHg lower systolic BP among females (p = 0.003). For PCA-SES2 we report results for linear splines, given the evidence for a non-linear effect mentioned above. Among males, a significant association was seen only for spline 2 (z-score - 1 to 0) where a one standard deviation increase in PCA-SES2 (lower SES) was associated with a 2.6 mmHg higher systolic blood pressure. Among females, significant associations were seen for spline 1 (z-score <−1), spline 2 (z-score −1 to 0), and spline 4 (z-score >1). Coefficients were 2.7 for spline 1, −3.5 for spline 2 and −0.5 at spline 4. Age and BMI were directly associated with systolic BP among both males and females, while fasting glucose showed significant direct association among females only. Household possession categories was inversely associated with systolic BP for the middle third among males and the upper third among females. There was an inverse association with fast food consumption for the 3-4 times/week category compared to the to the ≤2 times/week category for females only. No association was seen for physical activity levels in these models.

Results for bivariate analyses for diastolic BP are shown in Table 4. There were no significant associations between diastolic BP and PCA-SES1, however inverse associations were seen for males at spline 1 and females at spline 3 for PCA-SES2. When elevated BP or hypertension was used as the outcome variable in bivariate analyses (Table S8), only PCA-SES1 was significantly associated with elevated BP or hypertension, and this was seen only among males in the upper third of PCA-SES1 scores (odds ratio 0.65, p = 0.015).

Sequential multivariable analyses with systolic BP as the outcome are shown in Table 5. These results were derived from sex-specific mixed effects linear regression models, with cluster variables being community and parish. Model 1 includes adjustments for age, household SES and study. BMI is added to model 2 and glucose, physical activity and diet added in model 3. Among males, PCA-SES1 remained inversely associated with systolic BP in all models. The coefficient was slightly attenuated (i.e., less negative) when adjustments were made in model 1, changing from −1.45 (p < 0.001) in the unadjusted model to −1.31 (p < 0.001) in model 1; however, addition of BMI to the model resulted in the magnitude of the coefficient increasing (i.e., becoming more negative) to −1.49 (p < 0.001). There was no appreciable change in the coefficient when fasting glucose, physical activity and diet were added to the model. Among females, the significant association seen in the bivariate analysis was attenuated and no longer significant when age, household possessions and study were added in model 1. The coefficient reduced in magnitude from −1.30 (p = 0.003) in the unadjusted model to −0.46 (p = 0.423) in model 1. The addition of BMI (model 2) and other covariates (model 3) resulted in only very small changes in the coefficient. For PCA-SES2, the association seen at spline 2 in the unadjusted model among males was no longer significant in any of the sequential models, but a significant positive association was seen at spline 4 (β = 0.94, p = 0.042) in model 2 only. Among females, PCA-SES2 was directly associated with systolic BP at spline 2 (z-score -1 to 0) with little change across the models. At this level a one standard deviation increase in PCA-SES2 (lower SES) was associated with a 4.09 mmHg increase in systolic BP in the fully adjusted model. At spline 3 (z-score >0 to 1), higher PCA-SES2 score was associated with lower blood pressure (β = −2.81, p = 0.013 in model 3).

Table 6 shows the sequential multivariable models with diastolic blood pressure as the outcome. There were no significant associations seen for PCA-SES1 in any of the models for males or females. Among males, a significant inverse association was seen at spline 1 for PCA-SES2 (z-score <−1) (β = −4.10, p < 0.001 in the fully adjusted model). There was a significant direct association at spline 2 (z-score −1 to 0) (β = 2.85, p = 0.002). Among females, there was a significant positive association between diastolic BP and PCA-SES2 at spline 1 (β = 1.06, p = 0.023).

Table 7 shows the odds ratio for elevated blood pressure or hypertension derived from sex-specific multi-level logistic regression models adjusted for age category, household SES, study, BMI categories, glucose category, physical activity and fast food consumption. Among males, the odds ratio for elevated blood pressure or hypertension for persons living in communities in the upper third of PCA-SES1 was 0.67 (p = 0.051), suggesting that higher SES may be associated with lower odds of hypertension or elevated blood pressure, but did not achieve statistical significance in this study. There were no significant associations among females for thirds of PCA-SES1. Thirds of PCA-SES2 was not associated with elevated blood pressure in males or females.

In order to assess whether our results could have been influenced by the multiple imputation, we re-ran the full models (model 3) with complete cases only (i.e., no imputations) for systolic and diastolic blood pressure. These models are shown in Tables S9 and S10. The associations seen were generally very similar to that seen in the models with multiple imputation but given the smaller numbers (1,019 for males and 1,362 for females) estimates were often less precise. We also assessed whether the observed associations differed according to study by re-running the full final model for each study separately. These results are shown in Tables S11 and S12. While there were some differences in association in the different studies, associations were generally similar in direction to the findings in the pooled analysis, however there was some variation in the magnitude of the estimates; additionally, estimates were again imprecise with wide confidence intervals and often not statistically significant.

Discussion

We have shown from these analyses that lower community SES is associated with higher systolic BP among youth in Jamaica, particularly among males. This association persisted after adjusting for possible confounding by age and household SES and after accounting for the effects of BMI, fasting glucose, physical activity and diet (fast food consumption). Among females, the association with the first PCA-SES component was no longer significant after adjusting for confounders. Associations for the second PCA-SES component demonstrated non-linear effects. Diastolic BP also showed some non-linear associations for the second PCA-SES variable. However, while lower community SES may also be associated with higher odds of elevated blood pressure or hypertension among males, this did not achieve statistical significance.

The findings from this study adds to the body of literature showing an important contribution of neighbourhood conditions to health and underscores the importance of including the neighbourhood context when evaluating the social determinants of health. It also highlights that the relationship between neighbourhood SES and blood pressure may be complex, with both linear and non-linear effects, thus necessitating a nuanced approach to our interpretation of these associations. A relationship between neighbourhood SES and blood pressure in adults is fairly well established, with several studies showing that lower neighbourhood SES was associated with higher BP (Coulon et al., 2016; Fan et al., 2015; Liu et al., 2013; Morenoff et al., 2007; Riva, Larsen & Bjerregaard, 2016; Sprung et al., 2019). Additionally, Cho and colleagues found that lower neighbourhood SES was associated with increased all-cause mortality among persons with newly diagnosed hypertension in Korea (Cho et al., 2016). The relationship between neighbourhood SES and BP among adolescents and young adults is less clear with some studies showing an inverse association and others no significant association. Murakami and colleagues found that neighbourhood socioeconomic disadvantage was associated with higher blood pressure among Japanese dietary students at age 18–22 years, with those in the highest quartile of neighbourhood socioeconomic disadvantage having 3 mmHg higher SBP (Murakami et al., 2010). In another study, McGrath and colleagues found that neighbourhood income, measured as percent at or below the poverty line, was significantly associated with ambulatory systolic BP among 14-year old adolescents in Pittsburgh (McGrath, Matthews & Brady, 2006). In that study, each 1% increase in the percentage of community members at or below the poverty line was associated a 0.29 mmHg increase in systolic BP. Hofelmann and colleagues, in a study conducted in Brazil among persons 20–29 years old, found that area level income was inversely associated with both SBP and hypertension; being in the upper third of community income was associated with 5.7 mmHg lower SBP (Hofelmann et al., 2012). Data from the Birth to Twenty Cohort in South Africa also found significant associations with a number of neighbourhood SES indices in bivariate analyses at age 16 years, however only one of these remained statistically significant in the fully adjusted model (Griffiths et al., 2012). Kwok and colleagues found that relative household income, but not neighbourhood income, was inversely associated with blood pressure in Chinese adolescents at age 13 (Kwok et al., 2015); while Chen and Paterson found no significant association between neighbourhood SES and blood pressure (Chen & Paterson, 2006). In the Hong Kong Growth Study of children 6-19 years, Ip and colleagues reported a significant association between maternal education and hypertension, but no significant association between community income and hypertension (Ip et al., 2016). Community income was however significantly associated with obesity. The study did not report associations with SBP or DBP as continuous variables; such analyses may have been more likely to show an association given the higher statistical power associated with continuous variable (Altman & Royston, 2006).

In this study, the association between neighbourhood SES and diastolic BP was seen only for the second PCA-SES variable and showed non-linear effects. We have previously reported weaker associations between diastolic BP and socioeconomic variables in a longitudinal analysis from the 1986 Birth Cohort (Ferguson et al., 2015). Additionally, McGrath and colleagues found that while neighbourhood SES was associated with systolic BP among adolescents in Pittsburgh, there was no significant association with diastolic BP (McGrath, Matthews & Brady, 2006). On the other hand, Riva and colleagues in a study from Greenland found that while neighbourhood SES was associated with both systolic BP and diastolic BP, the association was inverse U-shaped (Riva, Larsen & Bjerregaard, 2016), while Coulon also found significant associations for both systolic BP and diastolic BP among African-American adults (Coulon et al., 2016). Taken together these studies suggest that there is some variability in the relationship between neighbourhood SES and diastolic BP and that unraveling these relationships will require further studies.

The finding of sex differences in the association between community SES and health have been previously reported in studies from Jamaica (Cunningham-Myrie et al., 2015; Mullings et al., 2013). Cunningham-Myrie and colleagues (Cunningham-Myrie et al., 2015) found significant sex interaction in the relationship between neighbourhood infrastructure score and the prevalence of obesity, while Mullings and colleagues (Mullings et al., 2013) found that significant sex differences in the relationship between prevalent depressive symptoms and neighbourhood infrastructure and settlement pattern. In a study from Greenland, Riva and colleagues found that the associations between neighbourhood SES and blood pressure appeared to be stronger in men compared to women (Riva, Larsen & Bjerregaard, 2016). These findings are consistent with several other studies from our group that have shown sex-differences in the relationship between individual or household level socioeconomic status and health outcomes (Ferguson et al., 2010a; Ferguson et al., 2015; Ferguson et al., 2018; Ferguson et al., 2017; Ferguson et al., 2010b). This suggests that sex may be an important effect modifier in the relationship between social factors and health and therefore should be explored in studies evaluating these relationships. The associations found in this study could be interpreted to suggest that education and wealth (high loadings in PCA-SES1) may be protective for men, whereas poverty and crowding (as represented by unemployment and population density) may have important effects for women. This is supported by the findings of Mullings and colleagues who found that for depressive symptoms in Jamaica, poor community infrastructure was associated with increased risk among males, whereas, among females, residing in informal unplanned (usually crowded) communities was associated with increased risk (Mullings et al., 2013). It is possible that increased stress from living in poorer, tension-prone environments, as well as higher levels of alcohol and tobacco use may contribute to the higher blood pressure from lower SES for males, whereas household stressors related to crowding, lack of financial resources contribute to higher blood pressure in females. These hypotheses would need to be further explored in future studies.

The mechanisms underlying the relationship between neighbourhood SES and blood pressure has not yet been fully elucidated. Broadly speaking, neighbourhood SES may influence health through compositional effects (based on aggregate effect of individual risk factors), contextual effects (group level effects beyond those due to risk factors) or environmental effects (Diez Roux & Mair, 2010). This involves the process of embodiment where social and psychological factors in one’s environment result in biological changes and are ultimately expressed as biological characteristics (Krieger, 2005). With regards to blood pressure, it has been suggested that persistent vigilance in a setting of neighbourhood disorder or poor neighbourhood social circumstances may induce changes in vascular reactivity which results in higher blood pressure (Ewart et al., 2004). Another study found that agonistic striving (induced by challenges to interpersonal influence and control) and subordination (due to social denigration, rejection and mistreatment) are potential mechanisms through which neighbourhood stress may result in hypertension (Ewart, Elder & Smyth, 2014). Other studies have found associations between adverse neighbourhood circumstances and reduced nocturnal blood pressure dipping (Euteneuer et al., 2014; Mellman et al., 2015). Taken together these studies suggest that neural regulation of blood pressure may be altered by psychosocial stress induced by living in neighbourhoods with adverse social and economic conditions.

Conclusion

We have demonstrated that neighbourhood SES was inversely associated with BP among Jamaican youth and that there were sex differences in these associations. Findings from this study suggest that the neighbourhood context may be an important factor in the aetiology of hypertension and that interventions to address the growing public health challenges resulting from hypertension should include evaluation of neighbourhoods. Further research in Jamaica and similar populations is warranted to improve our understanding of the aetiology of elevated blood pressure and to help in the development of appropriate interventions.

Supplemental Information