Study design and participants

This cross-sectional analysis uses baseline data from individuals enrolled in the PURE cohort study, which seeks to determine the relationship between the burden of NCDs and a range of individual, household, and community characteristics. Its methods have been described elsewhere [7, 8]. Briefly, participants were recruited from communities in 21 countries, purposefully selected to capture wide variation in sociocultural diversity and economic development, as defined by the World Bank in 2006 [9]. Low-income countries include Bangladesh, India, Pakistan, Tanzania, and Zimbabwe; lower middle-income countries include China, Colombia, Iran, Occupied Palestinian Territory (OPT), and the Philippines; upper middle-income countries include Argentina, Brazil, Chile, Malaysia, Poland, South Africa, and Turkey; and high-income countries were Canada, Saudi Arabia, Sweden and the United Arab Emirates (UAE). The timing of data collection in each country is shown in Additional file 1.

Within each country, participants were selected from communities in both rural and urban areas. Communities were defined as groups of people who reside within a specific geographic area and who were generally expected to have similar characteristics (e.g. culture, socioeconomic status, use of amenities, goods, and services). Existing administrative boundaries, such as village limits or postal code areas, or physical features (e.g. area bounded by selected streets) were used to define urban communities, while rural communities were defined as villages or postal code areas located at least 50 km away from an urban center. Within strata, communities were drawn from areas of varying income levels. Sampling within participating communities sought to recruit an unbiased sample of households (Additional file 2 provides supplemental information on sampling).

Eligible households had at least one member aged 35 to 70 years intending to stay at that address for another 4 years. For practical reasons the goal was not to undertake proportionate sampling, but rather, to show economic and sociocultural diversity and include sites where investigators were committed to collecting high-quality data at a modest budget and following up participants over a number of years. However, those included are broadly comparable to the populations of the countries in which they reside in terms of risk factors and mortality (Additional file 3 provides supplemental information on the characteristics of the PURE samples) [10, 11]. Eligible participants between the ages of 35 and 70 years who provided written informed consent completed a standardized questionnaire on medical history, individual- and household-level risk factors, and underwent a basic physical examination as described in the INTERHEART study [12] (Additional file 4 provides supplemental information on these procedures). The PURE study was approved by research ethics committees in the participating countries.

Measurement of health-related outcomes and participant characteristics

Following the Demographic and Health Surveys (DHS), the questionnaire collected an extensive range of socio-demographic data, including household possessions: electricity supply, automobile, computer, television, motorbike, livestock, fridge, other four-wheel vehicle, washing machine, stereo, bicycle, kitchen mixer, telephone, land/real estate and kitchen window [16]. Within each country, these were used to generate an asset-based wealth index, based on that used in the DHS and using principal components analysis, which places surveyed households on a continuous scale of relative wealth from poorest to richest [17]. This approach is designed to permit meaningful comparisons of observed inequalities across countries at different levels of development [18]. The continuous score was divided into country-specific wealth quintiles that reflect the distribution of household wealth within each of the country cohorts. Additional file 5 provides further details on the asset-based wealth index. Age was calculated in years from date of birth, or self-reported age in years when date of birth was unknown.

Analysis of wealth-related differences

For each country, we used the continuous household wealth score to generate a summary measure of the degree of wealth-related differences for each hypertension outcome. We estimated the Wagstaff concentration index indirectly standardized for age, gender and urban/rural location derived using the convenient regression approach to produce point estimates and 95% confidence intervals corrected for within-cluster correlation at community level obtained using the delta method [19–21]. As the binary outcomes were coded as 0 for no and 1 for yes, index values were normalized by dividing by 1 minus the mean [22].

The value of the index ranges from –1 to 1, where a negative value indicates a disproportionate concentration of the outcome among the poor, and a positive value indicates a disproportionate concentration of the outcome among the rich. The index is zero when there is no wealth-related inequality. Thus, for the different outcomes, a positive index means that the probability of having hypertension and being aware of it, or having it treated or controlled, is higher among the rich (i.e. the outcomes are ‘pro-rich’). The magnitude of the concentration index can be used to give a relative sense of the degree of inequality in either direction (i.e. closer to 0 = more equal).

We present the concentration index as our core measure of relative wealth-related inequality; however, we also illustrate absolute inequalities for each outcome in two ways. Within each country, we used mixed-effects logistic regression to analyze data on those with hypertension, as defined above, and generated estimates of three prevalence outcomes: being aware of one’s hypertension diagnosis, being treated for hypertension, and having controlled hypertension. Although basic logit models without random effects produced a better fit in some countries with few communities, the mixed-effects model enabled us to apply a consistent approach, thus enabling comparability. The estimates were adjusted for age, gender, urban/rural location, and a random effect for community. These were plotted by wealth quintile to show the patterns at each stage of the treatment pathway, and the difference in prevalence between the poorest and richest quintiles, with 95% confidence intervals and p-values, to quantify absolute inequalities. Although this approach is intuitive, we also present the slope index of inequality for each outcome generated using the same mixed-effects logistic regression framework above. Similar to the concentration index, this alternative summary measure uses data across the entire sample, but examines absolute, rather than relative, inequality [23]. We used Stata v.13 for all analyses [24].

Finally, to identify potential correlates of the magnitude of wealth-related inequalities we used log-linear regression trend lines to compare hypertension management outcomes with economic development of each country, measured by its 2012 per capita Gross National Product (in US dollars) [9].

Participant characteristics

Gaps in the hypertension management pathway within countries

In all countries, a substantial number of hypertensive subjects were unaware of their condition (Table 3) and, of those who were aware, only a proportion was receiving treatment. For example, in Canada, an estimated 55.2% of patients were aware they had hypertension, and, among those, 97.9% were treated. Of these, only 45.8% were controlled, revealing that, in total, an estimated 24.8% of hypertensive patients were adequately managed. In India, the corresponding numbers were 42.1, 79.7, and 40.8%, respectively, so that only 13.7% of those with hypertension were adequately managed. The highest level of control was seen in Saudi Arabia, with the lowest in Tanzania.

Wealth-related inequalities in the hypertension management pathway within countries

The concentration indices, standardized for age, sex and urban-rural location (Fig. 1), show that those with hypertension from richer households were significantly more likely to be aware than those from poorer households in most lower-income countries. In contrast, awareness was greater among poorer households in Canada, Sweden, Argentina, Brazil and Poland, although the magnitude of these inequalities was relatively small. The inequalities in hypertension awareness were not statistically significant in the remaining countries. Treatment rates were higher among richer households in South Africa, China, the Philippines, Colombia, Bangladesh, India, Pakistan and Zimbabwe. They were higher among poorer households in Canada, Sweden and Poland.

Hypertension management and economic development

It might be expected that hypertension detection, treatment and control would be better in richer countries, which can afford strong health systems. Figure 2 examines the association of hypertension management outcomes with Gross National Product, with a trend line based on the log values of the variables. The lowest levels of awareness were seen in the poorest countries; yet this is not inevitable as in some poor countries, levels of awareness were high. Awareness was lower than expected given national income in Sweden and South Africa, but was higher than expected in Brazil, Saudi Arabia, OPT and the Philippines. A similar pattern is seen with levels of treatment, although there was a particularly large gap between awareness and treatment in low-income countries and in China and the Philippines. Control rates were low everywhere, but again some of the lowest levels are seen in the poorest countries, with Sweden performing unexpectedly poorly.

Wealth-related inequalities and economic development

It might also be expected that inequalities would be greater in poor countries, which tend to have weaker health systems. Figure 3 summarizes the association between concentration indices for each hypertension management outcome and Gross National Product. For each outcome there is more variation in the levels of inequality among poorer countries, with some, such as the Philippines, having a very unequal distribution on all measures while others are much more equal. In Tanzania this is because awareness, treatment and control are poor for all groups. However, this is not the case everywhere and some, such as Iran and OPT, achieve similar outcomes for all. In rich countries the distribution is fairly similar for all groups or even pro-poor.

Summary of findings

We have previously shown that there is sub-optimal control of hypertension in countries at all economic levels [3]. This paper confirms that, while there are gaps along the entire management pathway, the greatest problems in most high- and middle-income countries lie in awareness and control. Once a patient is diagnosed, they are very likely to be treated but not necessarily controlled. In contrast, in most of the low-income countries there tends to be steady attrition from awareness through treatment to control. What this study adds is that the gaps from detection through treatment to control vary among people with different levels of material resources within each country and within groups of countries at similar levels of development. This emphasizes the importance of country-specific information that can inform the national policies needed to implement universal health coverage, now a target in the 2015 Sustainable Development Goals [25].

Some countries, such as Iran or OPT, exhibit relatively equitable outcomes. Others, such as the Philippines, Colombia, India and Pakistan, do not. However, both levels and distribution of outcomes are important; Tanzania and Zimbabwe achieve very poor levels of control, but for everyone. Finally, Canada, Sweden and Poland exhibit distributions that are actually pro-poor.

Limitations

Other international comparative studies of hypertension management outcomes exist [4, 26], but PURE is unique in its geographical reach, both in including countries at all levels of development, and in the detail collected on individuals and the communities they live in. However, it is subject to a number of limitations. In addition to the recognized biases associated with using self-reported data, PURE’s design as a cohort study necessitated inclusion of subjects in communities who could feasibly and affordably be followed up over many years. Hence, the sampling framework in each country was not designed to be nationally representative, and sample sizes were relatively modest in some countries, such as Zimbabwe. Whilst analysis of the PURE household sample has showed good concordance with the national age, sex, urban/rural, education, and mortality profiles of the study countries, suggesting that there was no systemic bias in data collection [10], caution is still warranted when interpreting and extrapolating our findings to national level.

Also, the individual element reported here was limited to subjects aged 35 to 70, as they are at highest risk of cardiovascular disease. This means that our estimates of hypertension prevalence, its detection, treatment and control will be higher than those derived from individual surveys of people at all ages. Some of the most marginalized groups, such as migrant workers, will also have been missed. As with all research on inequalities, it is important to recognize the scope for artefacts where sampling methods exclude those at the extremes of the distribution, although considerable care was taken when recruiting communities to achieve representation and avoid this problem.

The asset-based wealth index, while facilitating comparisons across such a diverse range of countries, is only one approach to measuring inequalities and is subject to certain limitations. We have also adopted a parsimonious approach when estimating concentration indices, avoiding over-adjustment for other variables. Although such variables may be important determinants of inequalities, examinations of their effects would be best studied when limited to a single country or to a few similar ones as any determinant of inequality is, to some extent, context specific, such as membership of a particular ethnic group.

Notwithstanding these limitations, the data presented here provide important new information on the scale and nature of inequalities at each stage of the hypertension management pathway, presenting the data by individual country rather than in groups within the same category of economic development. This provides a much more finely grained picture, highlighting important differences in otherwise similar countries. This is critical, given that health policy development takes place at the national level.

Comparisons with other research

Our findings are broadly consistent with the existing, albeit somewhat fragmentary research, looking primarily at differences in prevalence of hypertension or adherence to treatment, with few international comparative studies and none, to our knowledge, that look at inequalities using comparable measures, such as the concentration index. Thus, we found 92% of Canadians aware of hypertension were being treated, compared with 82% in the Canadian Community Health Survey [27]. While the rates of awareness, treatment and control in Sweden were substantially lower than the other high-income PURE countries, they were similar to those in an earlier study that used data from the WHO MONICA (Multinational MONItoring of trends and determinants in CArdiovascular disease) Project [28]. In Iran, data from the National Surveillance of Risk Factors for NCDs also showed a similarly large attrition in the pathway of hypertensive individuals from detection through treatment to control (33% treated and 12% controlled, respectively, compared with 51 to 18% in our study) [29]. A recent paper combined data from SAGE (WHO Study on global AGEing and adult health) and COURAGE (Collaborative Research on Ageing in Europe), with data on hypertension prevalence by household wealth [5]. Among the countries in common with PURE, similar levels of hypertension awareness and control were reported: in China, 43% were aware and 8% controlled vs. 42 and 8% in our study; in India, 38% aware and 14% controlled vs. 42 and 14% in our study; and in South Africa, 38% aware and 8% controlled vs. 31 and 6% in our study.

Studies examining wealth-related inequalities are much fewer in number, however the pro-rich inequalities observed for detection and control observed in our study were consistent with previous findings in China, India and South Africa [5]; as were other study findings regarding hypertension treatment rates in Brazil [30], China [31], Iran [29] and Tanzania [32]. Finally, a recent Canadian study of adults using national data found no evidence for income-related inequality with respect to hypertension detection or control [33], while we observed a small, albeit significant, pro-poor distribution for detection but not control.

Interpretation

The most important message from this study is that generalizations based on research in a few countries must be cautious. We describe considerable diversity in the patterns observed even in countries at similar levels of development, both at what point on the clinical pathway the problems are greatest and in how detection, treatment and control vary by household wealth. Consequently, further research within individual countries is needed to understand better how characteristics of national health systems influence treatment seeking behavior for hypertension. Studies such as this one can only be a first step in developing appropriate policies.

Nonetheless, the observed gaps and inequalities still highlight the importance of ensuring universal access to affordable, efficient and contextually appropriate methods to detect hypertension in middle age, and when detected, to provide continuous access to skilled health workers and treatment, including medications. As such, good population-level hypertension control is unlikely to be achieved unless approaches to universal health care include mechanisms to ensure access to both primary care and essential drugs for all.

As alluded to above, further research is being taken forward within PURE in the form of detailed health system analyses. Those completed so far reveal complex pictures relating to both demand for treatment, influenced by cost of drugs and by how hypertension – a chronic, asymptomatic, but hazardous condition – is conceptualized, and by supply of trained health professionals and medicines [34, 35]. Other data from the Environmental Profile of a Community’s Health study (EPOCH), nested within PURE, also provide insights. Among the upper-middle income countries, treatment rates were highest in Brazil and Poland, where all pharmacies surveyed in PURE communities had four common CVD medicines in stock and the median cost, as a percentage of ability, to pay was 3 and 1% respectively. On the other hand, the lowest treatment rates were in Malaysia and South Africa, where availability was, respectively, 37 and 33% and the cost as a percentage of ability to pay was 9 and 32% [36]. This has clear consequences for inequalities in treatment and control, particularly as effective blood pressure control typically requires use of two or more medicines [37]. Thus, wealth-related inequality in hypertension control was pro-rich in Malaysia and South Africa, and in many other lower middle- and low-income countries. In contrast, it was relatively equitable in Iran, where access to affordable medicines is good [38].

Some other observations are possible based on knowledge of the health systems involved. In general, the percentage of participants aware of hypertension and on treatment tends to be higher in countries with universal health care, but only if it renders long-term medication affordable. Universal health care should also reduce inequalities where levels of coverage are high. Brazil has implemented wide-ranging healthcare reforms, including provision of common medications for free, with one survey finding that over 90% of users of the public system, the Family Health Strategy, who were aware of hypertension were taking medicine [39]. Yet, while China has three insurance schemes that also provide some degree of coverage for most of the population, medicines are paid for out of pocket and can be expensive [40]. In contrast, though India has yet to implement universal health coverage, it has a thriving domestic generics industry with high levels of availability, but affordability is poor [36].