The magnitude and direction of socioeconomic inequalities showed different patterns across risk factors, sex and country income group. Historically, the adoption of risky health behaviours tends to transition from higher to lower socioeconomic groups as countries grow richer: new behaviours are adopted earlier by higher socioeconomic groups who, over time, largely abandon these behaviours upon learning of the associated detrimental health effects (for example, due to health promotion efforts). Lower socioeconomic groups tend to engage in risky health behaviours later in the course of a country’s economic development . Acknowledging that the timing and patterning of this transition are risk-factor dependent, these characteristic transition patterns may help to explain variance in our findings.
The strongest regular relative inequalities were reported for current daily smoking, particularly for education-related inequality. Other studies from developed,  and developing countries [39, 40] have also reported significant education-based disparities in smoking rates, and the importance of tobacco control in LMICs has been widely recognized as an immediate public health priority [5, 41, 42]. These findings underscore the importance of integrating equity considerations into widespread tobacco control efforts, including planning, implementation, monitoring and surveillance activities. The WHO Framework Convention on Tobacco Control,  the primary tool developed by and available to countries to control tobacco use and exposure, noted the serious burden that the production and consumption of tobacco products place on the poor. While tobacco reduction efforts have achieved considerable success in high-income countries, [44, 45] tobacco companies have intensified marketing strategies to target vulnerable populations of LMICs, such as women and adolescents [46–48]. Interventions at national and international levels stand to benefit by adopting equity focused approaches to reduce smoking prevalence, bearing in mind that population groups may differ in their ability to participate in such initiatives and/or experience intended health benefits [49–51].
Low fruit and vegetable consumption demonstrated regular inequality, with consistently higher prevalence among the disadvantaged across all study groups. Hall et al.  were the first to quantify socio-demographic prevalence data for fruit and vegetable consumption in 52 WHS countries, also noting a trend for decreased risk factor prevalence with increased household wealth status, and prevalence rates of low fruit and vegetable consumption in excess of 70%. The widespread nature of low fruit and vegetable consumption indicates an opportunity for population-wide interventions. At the same time, observed regular inequality calls attention to the need for equity-based approaches targeting populations of lower socioeconomic standing. Given the shortage of data about fruit and vegetable consumption in many LMICs,  additional research is warranted at national and international levels to further characterize these trends, with intensified efforts to collect these data with the necessary precision across all socioeconomic strata.
Our measure of physical inactivity did not differentiate between physical activity at work, commuting or during leisure time, distinctions which may demonstrate diverging patterns according to socioeconomic position. Leisure-time physical activity has been reported to increase with socioeconomic position, while work-time physical activity tended to decrease [53, 54]. Because leisure-time activity has been found to constitute a relatively small proportion of overall physical activity in LMICs, [55, 56] the socioeconomic gradients observed in the present study may reflect high levels of physical activity associated with work or commuting among those of lower education or household income levels. The promotion of physical activity as an aspect of NCD prevention in LMICs has been largely neglected; increased advocacy and partnerships across health and non-health sectors, however, can help to create and maintain physical, social and cultural environments that encourage active lifestyles .
For heavy episodic alcohol drinking, data indicated a mixed pattern of inequality, as has been reported by other cross-national studies in countries of all income levels . Harmful use of alcohol is the third leading risk factor for global disease burden,  with complex determinants including cultural practices, lifestyle and social status influences on one hand, and health risk considerations on the other . NCD prevention efforts should work both nationally and globally to target the general population as well as subpopulations with a greater chance of engaging in heavy episodic drinking. The WHO global strategy to reduce the harmful use of alcohol outlines action-oriented strategies for implementation at global, regional, sub-regional and national levels .
Overall, wealth- and education-related inequalities showed similar directional trends for each risk factor, with varying magnitude across sex-country income groups, indicating a concerted need for educational and poverty-reduction oriented policies. Poverty-reduction strategies may contribute to the success of policies addressing risk factors that displayed prominent within-country regular, wealth-based inequality, such as current daily smoking, low fruit and vegetable consumption and, within LICs, heavy episodic alcohol drinking. Kim et al.  highlighted the importance of using education to establish cultural norms that reinforce healthier lifestyles. This approach may be particularly effective in situations demonstrating strong regular education-based inequality. While education grouping was on an absolute basis-- i.e. the highest level of education attainment, irrespective of country of residence-- wealth quintiles were derived from the asset index, which scored the relative position of each household comparing to others within each country. We showed fairly similar patterning of risk factors across education and wealth levels, despite that these stratifiers were defined based on different terms (absolute or relative). In other words, wealth- and education-related inequalities in health exist in countries at all levels of development. The present study did not explore possible interactions between wealth- and education-related health inequalities. Thus, it remains unknown whether a reduction in education-based inequality, for example, would impact wealth-based health inequalities.
Strengths and limitations
This study quantified and compared wealth-and education-based inequality in four health risk factors, stratified by sex and country-income group. The resolution of the United Nations High-level Meeting on NCDs recognized the disaggregation and analysis of data as an important component of strengthening national policies and health systems. Combining national level datasets facilitated broad-scale analyses of socioeconomic distribution of leading NCD risk factors, contributing international evidence from understudied LMIC regions. Wealth classification by quintiles is an accepted method of making relative comparisons among respondents, although patterns of wealth distribution may vary between countries. Likewise, education levels were standardized to be comparable across countries.
Pooling datasets inherently masks individual country differences; however, a country variable was included in the multivariate model in order to control for any potential confounding effect related to the individual countries.
It is possible that a selection bias may have occurred in the sampling process especially in countries with lower response rates, although we are not aware of evidence to suggest that this had occurred. The main reason for household non-response was inability to locate the selected households, or the households refusing to participate even before a roster could be obtained.
The study countries were not probabilistically selected and therefore are not necessarily representative of all LMICs.
During analysis and discussion the health impact of the variable level of risk by each risk factor was not explored. A comparable level of alcohol consumption has been shown to result in higher levels of alcohol-attributed harm among poorer populations,  and within certain geographical regions . Measuring prevalence alone may not fully represent inequalities in the impact of the studied health risks.
Finally, this is a cross-sectional study that examined inequalities in the distribution of risk factors across a range of countries at a point in time. Patterns of distribution of risk factors, however, may fluctuate over time, with variable increases, decreases, or periods of stagnation across sections of a population (as was evident for example, with smoking rates in high income countries ). Thus, longitudinal studies are needed to track these changes over time and to understand the drivers of these trends. Other NCD risk factors, such as blood glucose or lipid levels, were not included in the present study, but may show inequalities in their distribution. Future studies may include a broader range of risk factors to provide a more comprehensive distribution of risk profiles across socioeconomic groups.