Cancer Causes & Control

, Volume 23, Supplement 1, pp 11–25

Global socioeconomic inequalities in tobacco use: internationally comparable estimates from the World Health Surveys

Authors

    • Department of Epidemiology, Biostatistics and Occupational HealthMcGill University
  • Brittany McKinnon
    • Department of Epidemiology, Biostatistics and Occupational HealthMcGill University
Original paper

DOI: 10.1007/s10552-012-9901-5

Cite this article as:
Harper, S. & McKinnon, B. Cancer Causes Control (2012) 23: 11. doi:10.1007/s10552-012-9901-5
  • 331 Views

Abstract

Objective

To produce internationally comparable estimates of socioeconomic differences in tobacco exposure within low and middle-income countries.

Methods

We used data from 50 countries that participated in the World Health Surveys in 2002–2003. We measured two aspects of smoking: current smoking prevalence and accumulated pack-years of smoking. We used an asset-based approach to estimate permanent income. We measured absolute inequalities, separately by gender, across the entire socioeconomic distribution by using the concentration index and summarized the results and explored heterogeneity by meta-analysis.

Results

The overall prevalence of current smoking was highest in Southeast Asia, the Western Pacific, and Europe, and lowest in Africa. Pack-years among current male smokers were highest in Europe. Wealthier men were generally less likely to be current smokers in all regions. However, there was substantial heterogeneity within each region, and in some countries (Georgia, Mexico, Mauritania) current smoking was greater among the more advantaged. Among currently smoking men socioeconomic differences for pack-years of smoking were generally much weaker than for smoking prevalence. Among women the concentration index in current smoking was largest and favored the poor in Europe (1.4, 95% CI 0.8, 2.1) but favored the rich in Southeast Asia and the Western Pacific. National income was generally not associated with the magnitude of socioeconomic gradients.

Conclusions

In low and middle-income countries there is substantial between and within-region heterogeneity in socioeconomic inequality in tobacco exposure that is not explained by national income. Our results imply that the relationship between socioeconomic position and smoking in poorer countries is dynamic and may not reflect the historical pattern in wealthier countries.

Keywords

TobaccoHealth inequalitiesSocioeconomic positionSmokingLow- and middle-income countries

Introduction

Consumption of tobacco remains one of the world’s leading contributors to premature mortality and a range of disabilities. Tobacco use is currently estimated to account for 18% of deaths and 11% of disability-adjusted life years in high-income countries, compared with 7% and 3%, respectively, for low and middle-income countries (LMICs) [1]. However, because of the approximately 30-year time lag between peak tobacco consumption and peak tobacco-related mortality, this global pattern largely reflects past tobacco use in high-income countries, and the future burden of tobacco will largely fall on LMICs. Previous surveys suggest the prevalence of tobacco use has been increasing rapidly in LMICs, notably in populous countries such as India, China, and Indonesia; without intervention it is estimated that over 1 billion deaths may be caused by tobacco during the twenty-first century [25].

Estimates of the impact of tobacco in poorer countries has mostly relied on a variety of techniques of indirect estimation, for example smoking-impact ratios (observed over expected lung cancer deaths) that use mortality data to measure the accumulated impact of tobacco [6, 7]. Although such indirect measures provide a more comprehensive summary of the impact of tobacco, there is also a need to understand the distribution and determinants of tobacco use among current smokers, given that 50% of future deaths will come from current smokers [2, 8, 9]. Additionally, smoking impact ratios are typically only estimated for countries as a whole or, in some cases, for gender groups. Other characteristics, for example urban/rural location, education, or income, are also relevant correlates of tobacco consumption.

In addition to concerns about the overall impact of the spread of tobacco to LMICs, the widespread existence of socioeconomic differences in smoking that have been observed in higher-income nations [1012] has raised concerns that similar patterns may be likely to develop in LMICs, adding additional burdens to already disadvantaged populations. However, information about the socioeconomic distribution of tobacco consumption in LMICs has been severely limited by a lack of routine sources of data, the use of non-comparable measures of socioeconomic position (education, literacy, income, occupation) [13, 14], and the wide range of tobacco products used within and between countries.

More recent attempts to obtain comparable estimates of tobacco use across countries, for example the WHO’s MPOWER report [15], typically do not provide estimates of tobacco use by urban/rural, socioeconomic, or marital status, and there remains little information about the extent of variation across countries in the effects of demographic factors and smoking. Additionally, virtually all previous studies have focused primarily on the prevalence of smoking, with less emphasis on measures of accumulated exposure.

The objective of this paper is to contribute to the evidence base on global inequalities in tobacco consumption in four ways.

  • First, we utilize cross-population-comparable measures of both tobacco use and of socioeconomic position.

  • Second, rather than simply compare prevalence of smoking among the most disadvantaged with the prevalence among the least disadvantaged [13, 14], we evaluate the socioeconomic pattern of smoking across the entire range of socioeconomic position.

  • Third, we go beyond simple measures of prevalence to measure accumulated tobacco exposure across socioeconomic groups.

  • Fourth, we make some attempts to quantify and explain heterogeneity in socioeconomic gradients across regions and countries, because some previous work suggests socioeconomic differences are largest in poorer countries [13].

Materials and methods

Data and sample

We used data from 53 countries that participated in the World Health Surveys (WHS), a series of large cross-sectional studies conducted by the World Health Organization (WHO) in 2002–2003. The WHS collected information covering a wide range of topics related to population health, including socio-demographics, adult and child morbidity and mortality, risk factors, and health-care expenditures [16]. Comprehensive information about the surveys can be found on the WHS website [17], and the specific countries included in the survey are given in Table 3, in the “Appendix”, according to WHO geographic region and 2003 World Bank income classification. Figure 3, in the “Appendix”, shows the WHO geographic regions. We excluded three countries (Slovenia, Spain, United Arab Emirates) that were classified as high-income countries in 2003. Briefly, the survey was administered in-person to adults ages 18 and over who were living in private households. Interviewers were trained and all questionnaires were translated into local languages and modified for cultural appropriateness according to standard WHO procedure. The WHS’s sampling framework covered 100 percent of a country’s eligible population, and random national samples were obtained in all countries except China, Comoros, the Republic of Congo, Côte d’Ivoire, India, and the Russian Federation, by using a multistage cluster design. The target population included any men or women adult aged 18 or over, present in the country and residing in a private household during the survey period. Survey weights were available for all countries except Guatemala.

Measures

Individuals were asked, “Do you currently smoke any tobacco products such as cigarettes, cigars, or pipes?”, were queried about other tobacco types, and responded as either daily, non-daily, or non-smokers. We defined current smokers (yes/no) as daily or non-daily smokers of any tobacco products (cigarettes, cigarillos, cigars, cheroots, chuttas, bidis, goza/hookah, pipes, or other local tobacco products). The WHS also asked individuals about average daily consumption of each tobacco product (manufactured cigarettes, hand-rolled cigarettes, pipefuls of tobacco, other tobacco products) and for how many years individuals considered themselves daily smokers. From this information we calculated a measure of accumulated exposure among current smokers. We estimated pack-years among current smokers from the total number of all types of cigarettes and other forms of tobacco smoked daily, plus twice the number of pipefuls (if applicable), divided by 20 [18]. Because pack-years are partially a function of age, we used age-adjusted estimates of pack-years of smoking to produce estimates of pack-years at age 40.

To estimate socioeconomic position, we used an asset-based approach to compute an index of permanent household income [19]. This approach assumes economic status is an unobserved latent variable, and it is estimated by use of a random-effects probit model using measures of household ownership of country-specific assets (e.g., refrigerator, radio, car, etc.), access to services (e.g., drinking water), and known predictors of income (e.g., age and education). Table 4, in the “Appendix”, contains a list of assets used in calculating the index. This model produces asset thresholds on the latent income scale, and if a household’s estimated permanent income is greater than the asset threshold, there is a greater than 0.5 probability they own the item. This asset scale is then applied to each household to estimate permanent income. The validity of this approach has been examined in several countries, with studies finding moderate to high correlations between estimated permanent income and both reported household income and expenditures [1921]. For descriptive analyses, estimates of household permanent income were categorized into quintiles within each country.

Given the strong global patterning of tobacco consumption by gender [15, 22], we present estimates of tobacco exposure separately for men and women.

Missing data

The total sample size was 247,421, and information on smoking variables and permanent income was missing for 4.8 and 3.8% of the sample, respectively. To account for missing data, we performed multiple imputation using the ice procedure in Stata 11, which uses an iterative multivariable regression procedure to generate distributions for each variable with missing data that are conditional on all other variables in the imputation models [23]. All variables with missing data were imputed using smoking and other demographic predictors and a total of ten imputed datasets were generated. Linear, logistic, ordered logistic, multinomial logistic, and negative binomial regressions were used to model the distributions of the imputed variables, as necessary. Where possible, analysis and pooling of results across the imputed datasets was done using the mi estimate procedures in Stata 11 [24].

Statistical methods

Measuring health inequalities across multiple groups (such as for socioeconomic position) involves making judgments about absolute versus relative inequality, whether to weight social groups by population size, and reference points for measuring departures from equality, among other issues [25, 26]. For this study we summarized the magnitude of health inequality using the concentration index [27], which may be measured on the absolute or relative scale, and measures differences between each individual and the average prevalence of smoking among the total population. Because measures of relative inequality may reach extreme values when baseline values are very low (e.g., prevalence of smoking among women in some LMICs), and because measures of absolute effect are typically of more interest from a population health perspective, we focus on measuring absolute inequality.

The absolute concentration index (ACI) is derived from an absolute concentration curve, which plots the cumulative percentage of the population, ranked by socioeconomic position, against their cumulative contribution to the prevalence of smoking among the total population. The ACI is calculated as \( ACI = 1 - \frac{\nu }{n}\sum\nolimits_{i = 1}^{n} {y_{i} (1 - R_{i} )^{\nu - 1} } \), where n is the sample size, i indexes individuals, y is the tobacco variable, R represents each individual’s relative rank in the cumulative socioeconomic distribution (i.e., from 0 to 1), and ν is a weighting factor that decreases with increasing socioeconomic rank. We used the “traditional” value of 2 for ν, which assigns weights of 2.0, 1.5. 1.0, 0.5, and 0 to the health of individuals at the 0th, 25th, 50th, 75th, and 100th percentiles of the cumulative distribution of increasing socioeconomic position [28]. When the outcome is negative (i.e., something to be avoided, as with current smoking prevalence or accumulated exposure), negative values of the ACI indicate greater concentration of tobacco use among poorer individuals, and positive values indicate greater concentration among richer individuals. To facilitate comparison with more traditional estimates, we also calculated prevalence differences and ratios.

We summarized estimates across the 50 country surveys by using the meta-analysis commands in Stata [29]. Because we take the particular countries included in the WHS as a potentially random sample from a larger population of studies across all countries, we used a random-effects meta-analysis, which does not assume there is a single, common estimate of socioeconomic differences in smoking across all countries [30]. Heterogeneity across studies was assessed using the I2 statistic, which quantifies the amount of variation in results across studies beyond that expected by chance and is calculated as 100% × (Q − df)/Q, where Q is Cochran’s heterogeneity statistic and df the number of degrees of freedom [31]. We also investigated whether heterogeneity in the magnitude of socioeconomic differences across studies was related to country income, by using random-effects meta-regression and each country’s gross domestic product per capita (GDPpc) in 2002 (in 2005 purchasing power parity-adjusted international dollars), obtained from the World Bank’s World Development Indicators database [32]. GDPpc was not available for Zimbabwe, and meta-regression estimates were weighted by the inverse of the standard error of the estimated ACI in each country [29].

Results

Table 1 shows descriptive information for the WHS sample. Among the WHS countries the overall prevalence of smoking was highest in Bangladesh (45.8%) and lowest in Ethiopia (3.9%). Table 1 also includes estimates of global permanent income, which is measured on a continuous (latent) scale, and each country’s rank for the average of the income variable. Although our analysis focuses on country-specific estimates of permanent income, the estimates of global permanent income accord reasonably well with external estimates of national income, with the richest countries being largely European and the poorest being African. The correlation coefficient between the estimates of global permanent income and GDPpc was 0.6.
Table 1

Descriptive characteristics, World Health Survey

Country

Obs

Mean age (SD)

Male (%)

Current smoking (%)

Urban residence (%)

Global permanent income (rank)

Africa

 Burkina Faso

4,821

37.0

47.4

18.0

16.1

−2.71 (50)

 Chad

4,652

36.5

50.0

11.4

21.8

−0.53 (39)

 Comoros

1,758

40.7

49.1

22.5

30.6

−0.75 (44)

 Congo

2,488

35.3

47.7

10.0

94.4

−0.44 (34)

 Côte d’Ivoire

3,178

34.7

57.8

13.5

71.1

−0.3 (30)

 Ethiopia

4,937

35.6

50.3

4.0

12.9

−1.94 (49)

 Ghana

3,932

40.1

45.5

6.0

42.4

−0.71 (42)

 Kenya

4,346

35.3

49.6

14.3

17.9

−0.59 (41)

 Malawi

5,300

35.5

44.0

14.8

9.5

−0.28 (29)

 Mali

4,262

40.1

65.9

17.8

22.0

−1.56 (47)

 Mauritania

3,772

37.5

44.0

14.4

49.5

−1.15 (46)

 Mauritius

3,888

41.2

49.6

22.6

43.8

0.56 (18)

 Namibia

4,250

37.4

42.6

19.6

38.2

−0.72 (43)

 Senegal

3,021

37.9

55.5

14.3

48.1

−0.39 (32)

 South Africa

2,352

37.6

47.6

25.2

60.1

0.61 (17)

 Swaziland

3,069

38.9

47.0

12.0

26.7

0.52 (19)

 Zambia

3,811

35.4

47.3

14.2

34.7

−1.68 (48)

 Zimbabwe

4,065

37.1

39.9

12.9

40.8

−0.45 (35)

Americas

 Brazil

5,000

41.1

45.5

22.2

84.6

0.68 (15)

 Dominican Republic

4,534

40.0

46.5

15.1

63.0

0.36 (20)

 Ecuador

4,614

40.7

41.8

18.0

82.5

0.25 (22)

 Guatemala

4,767

40.0

38.4

11.7

42.2

0.24 (23)

 Mexico

38,618

40.6

42.5

24.5

78.4

0.35 (21)

 Paraguay

5,132

39.0

46.4

26.9

53.0

0.14 (26)

 Uruguay

2,976

44.5

46.6

33.5

77.3

0.74 (10)

Eastern Mediterranean

 Morocco

4,472

40.1

50.9

15.9

62.5

0.20 (25)

 Pakistan

6,106

37.6

62.0

27.6

30.9

−0.41 (33)

 Tunisia

5,069

40.1

48.7

26.9

64.2

0.21 (24)

Europe

 Bosnia and Herzegovina

1,028

43.5

48.9

44.2

42.4

0.92 (7)

 Croatia

990

49.5

41.9

26.6

67.2

1.23 (1)

 Czech Republic

935

46.5

47.6

30.5

79.7

1.14 (2)

 Estonia

1,012

49.7

35.8

35.7

68.7

1.04 (4)

 Georgia

2,749

45.9

43.7

29.0

51.7

0.71 (12)

 Hungary

1,419

49.8

40.6

34.0

67.2

0.84 (8)

 Kazakhstan

4,496

42.6

34.1

25.9

80.9

0.92 (6)

 Latvia

856

50.5

32.4

34.8

67.2

0.64 (16)

 Russian Federation

4,422

51.4

35.6

27.6

87.6

0.75 (9)

 Slovakia

2,518

34.3

40.7

36.7

87.2

0.93 (5)

 Turkey

11,220

41.3

43.6

32.9

61.9

1.11 (3)

 Ukraine

2,503

46.3

35.7

25.6

71.3

0.71 (13)

Southeast Asia

 Bangladesh

5,552

38.9

49.0

45.8

19.7

−0.75 (45)

 India

9,723

38.4

52.8

36.1

10.2

−0.48 (37)

 Myanmar

5,886

40.6

44.6

30.8

26.2

−0.53 (38)

 Nepal

8,688

38.8

46.1

40.7

15.2

−0.58 (40)

 Sri Lanka

6,698

41.4

48.3

21.3

14.5

−0.47 (36)

Western Pacific

 China

3,993

45.1

49.0

30.0

30.8

0.70 (14)

 Lao PDR

4,889

38.2

47.5

39.3

26.6

−0.39 (31)

 Malaysia

6,038

39.9

45.3

25.6

62.8

0.72 (11)

 Philippines

10,078

38.9

47.5

34.7

51.3

−0.27 (28)

 Vietnam

3,491

39.2

47.2

25.8

15.4

−0.25 (27)

For descriptive purposes, and to provide an indication of exposure to tobacco among different socioeconomic groups, Table 2 presents (unweighted) average estimates of current smoking and pack-years among current smokers by WHO region, gender, and permanent income quintile, along with measures of inequality. We present two traditional measures of inequality, the rate difference between the poorest and richest quintiles (Q1–Q5) and their ratio (Q1/Q5), and our summary estimates of the absolute (ACI) and relative (RCI) concentration index, which is the ACI divided by average prevalence. Not surprisingly, in all regions men were both more likely to be current smokers and to have accumulated greater numbers of pack-years of smoking. Regionally, prevalence of current smoking tended to be highest for men in Southeast Asia, the Western Pacific, and European regions, and lowest in the African region. This pattern was generally true for women also, with the notable difference being particularly low levels of reported current smoking among women in the Eastern Mediterranean region.
Table 2

Average prevalence of current smoking and pack-years among current smokers according to permanent income quintile, and measures of inequality, by WHO region, and gender

 

Permanent income quintile (1st = poorest)

Measures of inequality

1st

2nd

3rd

4th

5th

ACI

RCI

Q1–Q5

Q1/Q5

Current smoking (%)

Women

 Africa

6.5

6.0

5.9

5.4

5.2

−0.30

−0.05

1.32

2.34

 Americas

16.5

14.8

13.3

13.9

15.2

−0.31

0.02

1.30

1.29

 Eastern Mediterranean

3.3

3.1

3.1

3.5

3.3

0.05

0.08

0.01

0.97

 Europe

18.0

20.6

21.5

21.0

23.6

1.10

0.09

−5.64

0.72

 Southeast Asia

25.7

19.4

20.3

16.5

9.6

−3.43

−0.19

16.11

2.77

 Western Pacific

11.3

6.8

8.1

5.5

3.4

−1.77

−0.20

7.94

4.03

Men

 Africa

27.8

24.1

23.2

22.7

20.7

−1.59

−0.07

7.17

1.56

 Americas

36.6

30.4

30.2

28.4

27.1

−2.22

−0.08

9.52

1.46

 Eastern Mediterranean

46.3

39.6

47.3

35.7

29.7

−3.19

−0.09

16.63

1.72

 Europe

55.5

47.6

50.4

46.3

48.0

−1.71

−0.04

7.47

1.22

 Southeast Asia

63.8

51.3

51.0

48.3

39.2

−5.19

−0.10

24.61

1.66

 Western Pacific

67.6

55.8

58.2

53.3

45.6

−4.79

−0.08

21.97

1.50

Pack-years among current smokers

Women

 Africa

5.3

5.8

6.1

6.2

5.6

0.37

0.06

−0.38

1.25

 Americas

9.2

9.4

9.7

11.8

8.8

0.23

0.02

0.35

1.04

 Eastern Mediterranean

7.3

10.1

20.8

12.5

4.9

0.15

−0.01

2.39

1.44

 Europe

12.5

12.6

9.8

10.6

9.5

−0.19

−0.02

2.98

1.40

 Southeast Asia

6.9

8.3

7.1

7.2

6.1

−0.45

−0.05

0.88

1.18

 Western Pacific

12.9

11.6

13.6

11.0

10.9

−0.58

−0.06

2.07

1.28

Men

 Africa

7.6

7.6

6.9

6.3

6.3

−0.04

0.00

1.26

1.24

 Americas

12.7

12.8

13.6

15.4

12.5

0.17

0.01

0.21

1.05

 Eastern Mediterranean

13.5

13.4

14.4

14.5

13.4

−0.01

−0.01

0.06

1.08

 Europe

24.7

20.2

19.7

18.3

15.3

−0.92

−0.04

9.39

1.84

 Southeast Asia

13.5

11.9

11.0

9.2

8.3

−1.10

−0.09

5.18

1.53

 Western Pacific

14.3

14.1

14.0

13.1

12.6

−0.52

−0.04

1.63

1.12

ACI absolute concentration index (for current smoking ACI is multiplied by 100 for ease of interpretation); RCI relative concentration index; Q1, permanent income quintile 1; Q5, permanent income quintile 5

In general, simple comparison of the poorest and richest quintiles paralleled the magnitude and direction of the summary measures of inequality. The Spearman correlations comparing summary measures with extreme quintile comparisons for absolute and relative inequality were greater than −0.9 for current smoking for both genders (the correlations are negative because more negative values of the concentration index corresponded with higher quintile differences/ratios). For accumulated exposure, the correlations for absolute and relative measures were −0.68 and −0.71 for men, and −0.54 and −0.61 for women (all p < 0.0001). Despite this general concordance, there were important discrepancies. For example, among women in the Eastern Mediterranean region both the average risk difference and the average risk ratio comparing pack-years for the lowest quintile (7.3 years) with those for the highest quintile (4.9) would suggest that accumulated tobacco exposure is greater among the poor (Q1/Q5 = 1.44). However, if one looks systematically across all levels of permanent income, pack-years of exposure generally increases with increasing permanent income, leading to a positive average absolute concentration index (ACI = 0.15). Similarly, comparing only extremes of the socioeconomic distribution for current smoking among women in Africa versus Southeast Asia suggests both have similar levels of relative inequality, because women in the poorest quintile were approximately 2.5 times more likely than those in the richest quintile to be current smokers (Q1/Q5 ~ 2.5). However, the average RCI is more than three times as high for Southeast Asia (−0.19) as for Africa (−0.05), because in the latter current smoking does not vary much across the middle income quintiles whereas in the former smoking prevalence declines steadily as one moves up the socioeconomic ladder.

Overall we found, for both men and women, substantial heterogeneity in the magnitude of socioeconomic inequalities in tobacco exposure. The overall random-effects estimate of the ACI for current smoking among men was −2.50 (95% CI = −3.2, −1.8) and for women was −0.45 (95% CI = −0.86, −0.04). The negative value of the ACI indicates that, for both men and women, current smoking is more concentrated among individuals with lower permanent income (though for women this estimate was not distinguishable from zero). However, for both men and women the I2 statistic suggested high heterogeneity [33] across estimates (men I2 = 92%, 95% CI = 90–93%; women I2 = 96%, 95% CI = 95–97%), with both statistics suggesting that virtually all of the overall variation in inequality estimates across the WHS is because of between-country rather than within-country variation. To determine whether similar heterogeneity was seen within regions, we conducted WHO region-specific meta-analyses.

Figure 1a, b provides estimates of the magnitude of the socioeconomic gradient in current smoking for men and women derived from region-specific meta-analysis, with estimates of the ACI for each country. Overall, absolute socioeconomic differences in current smoking are larger for men than for women in each region. For men, socioeconomic inequalities favored wealthier individuals in all regions and most countries, and were largest in Southeast Asia (ACI = −5.20, 95% CI = −6.7, −3.7) and the Western Pacific (ACI = −4.82, 95% CI = −6.3, −3.4) and much smaller in Africa (ACI = −1.59, 95% CI = −2.4, −0.7). However, in all regions there was substantial evidence of heterogeneity, with the exception of the Eastern Mediterranean region (I2 = 65.5%, 95% CI = 0, 90%) which is likely to be underpowered given that only three countries in this region participated in the WHS. Notably, although the overall estimate of inequality for European men favored the rich, positive socioeconomic gradients in current smoking (more smoking among the rich) were observed for the Ukraine, Slovakia, and Georgia, for Mauritania in Africa, and for Mexico in the Americas region.
https://static-content.springer.com/image/art%3A10.1007%2Fs10552-012-9901-5/MediaObjects/10552_2012_9901_Fig1a_HTML.gif
https://static-content.springer.com/image/art%3A10.1007%2Fs10552-012-9901-5/MediaObjects/10552_2012_9901_Fig1b_HTML.gif
Fig. 1

Random effects meta-analysis of absolute socioeconomic inequality in current smoking among a men, b women, World Health Surveys, 2002–2003

Absolute socioeconomic inequalities in current smoking were generally smaller for women than for men, although just as variable (Fig. 1b). Overall estimates suggested higher smoking prevalence among the poor in Southeast Asia (ACI = −3.43, 95% CI = −5.5, −1.3), the Western Pacific (ACI = −1.75, 95% CI = −3.3, −0.2) and, marginally, in Africa (ACI = −0.24, 95% CI = −0.5, 0.0). In Europe, however, positive associations between permanent income and current smoking were observed for many countries, leading to an overall positive ACI of 1.44 (95% CI = 0.8, 2.1). As for men, however, in every region except the Eastern Mediterranean, there was substantial statistical evidence of heterogeneity, particularly in the Americas, where ACI estimates ranged from −3.4 in the Dominican Republic to 3.05 in Mexico.

In contrast with the pattern of greater smoking prevalence among poorer men in all regions, among current smokers there were generally weaker associations between socioeconomic position and pack-years of smoking (Fig. 2a). In Africa, the Americas, and the Eastern Mediterranean regions the ACI estimates were near zero, whereas poorer individuals had more pack-years of exposure in Europe, Southeast Asia, and the Western Pacific. The largest ACI was for the Southeast Asian region at −1.03 (95% CI = −1.7, −0.4). However, it should still be noted that in Europe, Southeast Asia, and the Western Pacific regions the I2 statistics were close to 50% or higher, and generally different from zero, suggesting there is substantial heterogeneity within regions in the relationship between income and pack-years of smoking. For women (Fig. 2b), given their lower prevalence of smoking, estimates of socioeconomic differences in pack-years of exposure were less precise than for men, and few countries had ACIs different from zero. The overall relationship between income and pack-years of smoking was positive among women in the African region (ACI = 0.28, 95% CI = 0.0, 0.6) and negative in Europe (ACI = −0.25, 95% CI = −0.5, −0.1) but both effects were small in magnitude. Estimates in other regions were not statistically distinguishable from zero.
https://static-content.springer.com/image/art%3A10.1007%2Fs10552-012-9901-5/MediaObjects/10552_2012_9901_Fig2a_HTML.gif
https://static-content.springer.com/image/art%3A10.1007%2Fs10552-012-9901-5/MediaObjects/10552_2012_9901_Fig2b_HTML.gif
Fig. 2

Random effects meta-analysis of absolute socioeconomic inequality in accumulated pack-years of smoking among a men who are current smokers, b women who are current smokers, World Health Surveys, 2002–2003

With regard to the relationship between socioeconomic differences and average prevalence, among women there was little correlation between average smoking prevalence and the magnitude of absolute inequality in current smoking (Pearson’s r = −0.09, p = 0.50); among men, however, countries with higher smoking prevalence were somewhat more likely to have greater socioeconomic differences in current smoking, i.e., more negative ACIs (r = −0.29, p = 0.04).

Finally, we attempted to explain heterogeneity in estimates of socioeconomic inequality across countries by using country GDPpc (in $10,000 units, logged). With the exception of a weak positive relationship between log GDPpc and socioeconomic inequality in current smoking among women (β = 0.90, 95% CI = 0.3, 1.5), suggesting slightly larger gradients in poorer countries, we could not detect any other association between national income and the extent of socioeconomic inequalities in tobacco use (results not shown). Meta-regressions that included region fixed effects produced similar results. Figure 4, in the “Appendix”, shows scatter plots of the four separate meta-regressions.

Discussion

The primary objective of this study was to estimate cross-nationally comparable measures of socioeconomic inequalities for two important aspects of smoking—current smoking prevalence and accumulated pack-years of exposure among current smokers. Using similar measures of socioeconomic position we found that poorer men were more likely to be current smokers in most WHO regions, especially in Southeast Asia and the Western Pacific, but that there was substantial heterogeneity in the extent of the social gradient in smoking in all regions. However, similar estimates in the accumulated amount of smoking (pack-years) among currently smoking men did not reveal a similar inverse gradient. This suggests that estimates of socioeconomic differences based solely on smoking prevalence may not reflect intensity of smoking among income groups. Gradients among women were generally weaker, though no less heterogeneous within and across regions.

Although the lack of previous comparable systematic surveys across LMICs largely limits our ability to compare our estimates with previous studies, our results nevertheless have some bearing on previous efforts. Our results are in some respects consistent with the summary of evidence on poverty and smoking presented in the review by Bobak et al. [13]. They also found substantial heterogeneity in socioeconomic gradients across countries, and in most countries more disadvantaged men were more likely to smoke. In other respects, however, our results are less consistent with previous research. Bobak et al. found that relative differences in smoking prevalence in poorly and highly educated males were largest in low-income countries. We found limited evidence of this pattern among women, but virtually no relationship between country income and socioeconomic inequalities in either smoking prevalence or accumulated exposure among men. The fact that the estimates of Bobak et al. used a different indicator of socioeconomic position, were from (largely) different countries, and were from the mid-1990s could potentially explain the differences between their conclusions and the WHS estimates.

In addition to estimating the magnitude of socioeconomic inequalities in current smoking, we also found substantially smaller inequalities in accumulated tobacco exposure in all regions, and virtually no inequality among men in Africa and the Americas. There have been far fewer studies looking at socioeconomic inequalities in measures of accumulated smoking in LMIC, although some have looked beyond simply current or daily smoking [34]. Hu and Tsai found more highly educated rural Chinese less likely to smoke overall, but no educational differences in the amount of daily consumption. A systematic review of smoking surveys in sub-Saharan Africa by Townsend et al. [14] also considered measures of smoking intensity, but concluded that the use of inconsistent measures of socioeconomic position made cross-country comparisons difficult. We hope our estimates may provide useful reference values for future assessment of socioeconomic differences in multiple measures of tobacco consumption.

Limitations

Our estimates of socioeconomic inequalities in tobacco use are subject to some limitations. First, the objective of the WHS was to provide a broad survey of several aspects of individual health and health system interactions. As such, the measures of smoking we describe are both self-reported and unable to capture potentially important socioeconomic differences in other tobacco-related phenomena, for example quitting behavior and exposure to passive smoke. Second, we focused primarily on estimating absolute inequalities using the concentration index. It is well known that the use of different measures of inequality may lead to different conclusions [3537]. Third, we only studied the sample of countries that participated in the WHS and grouped countries according to WHO regions. We used WHO geographic regions because they correspond to WHO program offices. However, although they are geographically based, they are not synonymous with geographic areas. Thus, regional estimates using different definitions such as those of the World Bank or the United Nations may differ from ours. While this survey was designed to be broadly representative of all global regions and levels of economic development, estimates from other countries may show different patterns. Given the extent of heterogeneity in our estimates, this seems a likely possibility.

We also note that our estimates of pack-years of smoking among current smokers are subject to uncertainty with regard to calculating tobacco equivalents. We could not separate cigars from other types of tobacco, and did not have information on the amount of loose tobacco used for hand-rolled cigarettes, which may vary widely. The current literature on tobacco equivalents in poorer countries is diverse, with some studies equating hand-rolled cigarettes and pipes with manufactured cigarettes [38] whereas others equate one manufactured cigarette with up to one hand-rolled cigarettes [39, 40]. Given that the use of non-manufactured cigarettes is greater in poorer countries, we may have underestimated accumulated smoking in these areas. Similarly, if poorer individuals are less likely to use manufactured cigarettes, we may have underestimated socioeconomic gradients in accumulated tobacco exposure.

A more important limitation is the fact that the WHS surveys are now almost 10 years old, and may not accurately reflect the current state of the tobacco epidemic in LMICs. However, we looked at the most recent data on smoking prevalence from the WHO’s Global Public Health Observatory, which generally date from 2006 for the countries we studied. In most cases our estimates were similar to WHO’s 2006 estimates, but in some cases, particularly for some European countries, for example the Russian Federation and Ukraine, our prevalence estimates are much lower. On the other hand, our estimates are higher than recent surveys from WHO’s 2009–2010 Global Adult Tobacco Surveys (e.g., for Bangladesh, China, Brazil, and Mexico) [41]. Whether this is a consequence of different survey designs or actual secular changes requires additional investigation. As a final point, the main advantage of the WHS surveys, despite being somewhat dated, is their ability to provide comparable estimates of socioeconomic differences in smoking across countries, which is currently not possible with the GATS data. Understanding whether socioeconomic differences in LMICs are increasing or decreasing over time more rapidly in some regions will require new sources of harmonized data.

Conclusions

Our findings suggest there is substantial heterogeneity within geographic regions in the association between socioeconomic position and tobacco exposure. Both positive and negative socioeconomic gradients in smoking are evident for both men and women, suggesting there is unlikely to be any universal relationship between socioeconomic factors and smoking behavior. The extent to which the heterogeneity in socioeconomic gradients we observed may be explained by other individual-level characteristics or the extant tobacco policy environment and more general determinants of socioeconomic inequality should be the subjects of future research.

Acknowledgments

We thank the Institute for Health Metrics and Evaluation (http://www.healthmetricsandevaluation.org) for providing us with the estimates of permanent income that were used in these analyses. This work was supported by the Canadian Institutes for Health Research (191612). Sam Harper was supported by a Chercheur-boursier from the Fonds de la Recherche en Sante du Quebec (FRSQ). The funders had no role in the study design, data gathering and analysis, interpretation of data, decision to publish, or preparation of the manuscript. The corresponding author had full access to all data that were analyzed and had final responsibility for the decision to submit the report for publication.

Copyright information

© Springer Science+Business Media B.V. 2012