Within LICs and MICs, wealth- and education-related inequalities of variable magnitudes and direction were quantified for the five NCDs. In most populations, regular inequalities in terms of wealth and education were reported for angina, arthritis, asthma and depression, with the strongest associations for angina, asthma and comorbidity. For all NCDs, additional adjustments for confounding factors in Model 2 (marital status, urban/rural area and wealth/education) tended to decrease the magnitude of inequality, and may thus help to explain disparities in NCD prevalence.
These findings are in accordance with previous reports from a variety of settings, which also reported inverse associations between socioeconomic position and prevalence of angina [10
], arthritis [12
], asthma [57
], and depression [23
]. Previously, European data (including eight higher-income countries) from the 1990s demonstrated education-related inequality in 14 of 17 studied NCDs, most notably stroke, diseases of the nervous system and diabetes. There was a tendency for stronger inequality in adults aged 25–59 than those aged 60–79 [9
]. Goyal et al. (2010) highlighted differences based on country income group, reporting that education had more of a protective effect against cardiovascular events in high-income countries than LMICs, and also among men [27
The connection between socioeconomic status and health is complex, and shaped by diverse circumstantial factors as well as political, social and economic forces [63
]. For example, people living in poverty may experience material deprivation and high stress levels, which may lead to constrained choices and a higher likelihood of engaging in risky health behaviours, increasing the risk of disease; following disease onset, reduced access to care hinders opportunities to prevent complications [64
]. It has been estimated that up to 80% of cases of cardiovascular disease or type 2 diabetes and 40% of cancer cases are preventable based on current knowledge, however, prevention initiatives may not adequately reach vulnerable populations where disease risk factors cluster [65
Unlike the other four NCDs, we reported higher diabetes prevalence among the wealthier and more educated, especially in LICs. These findings conflict with trends reported by previous studies conducted in higher-income countries [25
]. Epidemiological studies in lower-income countries are less-forthcoming, however diabetes was reported to be positively associated with affluence in the Dominican Republic [66
], and metabolic syndrome was positively associated with affluence among adolescents in India [67
]. Nations at different levels of development may realize different stages of disease epidemiological transitions [67
]. Noting methodological differences in determining diabetes prevalence, it is also possible that our findings may be subjected to bias stemming from a methodology issue whereby cases with a lower wealth or education were more likely to be under-diagnosed and therefore prevalence rates were underestimated. Populations in less-developed nations may have limited access to medical professionals [6
], which could result in under-diagnosis of diabetes, particularly among populations of lower socio-economic status; for example, better educated individuals may be more aware of diabetes as a health condition. Alternatively, this finding may reflect a complex relationship between wealth, overweight, obesity, other risk factors (such as physical inactivity), and diabetes [68
]. Ideally, future surveys may integrate objective indicators of disease, such as HbA1C diagnostic testing for diabetes [69
Comorbidity significantly lowers quality of life, affecting physical, social and psychological well-being [70
]. Our findings showed that comorbidity was more prevalent among the poor and less educated, in all sex-income groups. We reported overall comorbidity rates of up to 10%, with even greater prevalence in some poorer wealth quintiles and least educated subpopulations. In a multinational study of high-income countries 30.2% of adults over 18 reported more than one chronic condition, although the study included seven diseases whereas ours included five [71
]. Consistent with the present study, previous research has reported inverse associations between comorbidity and markers of socioeconomic status [30
Like other studies, NCD prevalence tended to be higher in women than men [30
], and the greatest burden was reported for a cardiovascular-related condition [3
]. Overall, study NCDs tended to be more prevalent in the MIC group, with the exception of depression. That depression rates were higher in LICs than MICs was not expected. Previously, depression prevalence was reported to be lower in a less-developed setting, although cultural willingness to report depressive symptoms may bias outcomes [72
]. According to epidemiological transition models, LICs may be expected to carry a lower-- albeit increasing-- burden of NCDs than higher-income country groups, as risk factors for infectious diseases are progressively replaced by risk factors for NCDs [73
]. Projections for 2005–2030 forecasted a 10% increase in the deaths due to chronic disease in LMICs (from 61% to 71%) [74
]. Monitoring trends in NCD prevalence in LMIC groups will help to characterize the nature of the modern epidemiological transition, and identify populations that are most at risk for NCDs.
Strengths, limitations and implications
Data for five NCDs were collected systematically in a large sample of LMICs that participated in the WHS, allowing for comparisons of standardized data across pooled data sets. Consistent diagnostic criteria in WHS data facilitated broad-scale analyses and comparisons across several countries, minimizing limitations associated with variable measurement tools and disease classifications. However, the use of pooled data from geographically- and culturally-diverse settings inevitably masks problems of comparability between countries [54
]. Nine studies were excluded from analysis due to insufficient data or high item non-response rates. There is no reason to believe that the excluded countries would have changed the main findings on socioeconomic inequality in LMICs. The non-response was not selective, including countries of both low and middle income groups. We included a country variable in our multivariate analysis in order to control for any potential confounding effect of the individual countries. We did not aim to explore the interaction effects of our study’s independent variables with each of the countries.
Wealth and education levels were determined nationally, and pooled across LICs and MICs. We acknowledge that patterns of wealth distribution vary between countries, however, quintile classification provided a widely accepted method to compare respondents based on relative wealth position within their country. Levels of education were standardized to be comparable across countries.
The use of symptom-based diagnoses for angina, arthritis, asthma and depression was a strength, as other methods that rely on medical charts or self-reported diagnoses may introduce biases related to health system access. Self-reported data could reflect systematic over- or under-reporting, which may vary by socioeconomic status [75
]. A tendency for under-reporting of symptoms by people with low levels of education [76
] raises the possibility that our data may underestimate prevalence in low education classes, and show weaker-than-actual inverse associations. As a result, our data may underestimate true NCD rates in socioeconomic disadvantaged populations.
It is possible that a selection bias may have occurred in the sampling process, especially in countries with lower response rate, although we are not aware of evidence to suggest that this had occurred. The main reasons for household non-response included inability to locate the selected household, or household refusal to participate even before a roster could be obtained.