Four indicators of socioeconomic position: relative ... - SAGE Journals

190 downloads 16 Views 169KB Size Report
between the four indicators was observed, with causes of death known to be ... education, and causes of death which were likely to be determined by adult ...
Scandinavian Journal of Public Health, 2005; 33: 215–221

ORIGINAL ARTICLE

Four indicators of socioeconomic position: relative ranking across causes of death ØYVIND NÆSS1, BJøRGULF CLAUSSEN1, DAG S. THELLE2 & GEORGE DAVEY SMITH3 1

Institute of General Practice and Community Medicine, University of Oslo, Norway, 2Centralkliniken, Sahlgrenska Universitetssjukhuset, Gothenburg, Sweden, 3Department of Social Medicine, University of Bristol, UK

Abstract Objective: A study was undertaken to examine the relative ability of occupational class, education, household income, and housing conditions to discriminate all cause and cause-specific mortality-risk in Oslo, and to see if this relative ability is consistent across the 12 most common causes of death. Design and setting: Census records of inhabitants in Oslo 1990 aged 45 to 64 were linked to death records 1990–98 (n588,159). All inhabitants were included except those who lacked census data on the independent variables. The relative index of inequality (RII) for each indicator was calculated. Main results: Education, occupation, and housing conditions had similar RIIs for all-cause mortality in both sexes. Household income had low RIIs, particularly in men. For the 12 most common causes of death some heterogeneity in the relative ranking between the four indicators was observed, with causes of death known to be related to early-life social circumstances (stomach cancer, cardiovascular disease, chronic obstructive pulmonary disease) being particularly strongly related to education, and causes of death which were likely to be determined by adult social circumstances (violence, sudden unexpected death) being particularly strongly related to occupation and housing conditions. Conclusions: Education, occupational class, and housing conditions all seemed to discriminate all-cause mortality to a similar degree. However, the cause-specific analysis revealed a heterogeneous pattern.

Key Words: cause-specific mortality, socioeconomic position, heterogeneity

Introduction A large number of studies on social inequalities in health are characterized by lack of desirable samples and valid variables [1]. Recommendations suggest collection of multiple indicators of social position but there is limited empirical knowledge on which indicator to use [2,3]. In a Scottish study, occupation and education predicted all-cause mortality to a similar degree [4]. One US study found income to be the best discriminator of mortality rates, followed by education and occupation [5]. Another recent US study looking at all-cause mortality found similar discriminative power of occupation, education, and

economic resources [6]. In a Swedish study, housing tenure and education were most associated with increased mortality risk [7]. These studies have aimed at identifying optimal indicators but none of them has studied causespecific patterns in detail. Depending on how much heterogeneity there is between disease outcomes, it may be misleading to search for ‘‘optimal’’ indicators of social inequality as these may differ between populations and cohorts [8,9]. It is likely that the indicators of social position tap into different constructs on the causal pathway to various health outcomes [10,11]. When specific diseases are investigated, this may vary depending on the indicator chosen. Measures of occupational class

Correspondence: Ø. Næss, Institute of General Practice and Community Medicine, PO Box 1130 Blindern, N-0317 Oslo, Norway. Tel: +47 2285 0550. Fax: +47 2285 0610. E-mail: [email protected] (Accepted 5 May 2004) ISSN 1403-4948 print/ISSN 1651-1905 online/05/030215-7 # 2005 Taylor & Francis Group Ltd DOI: 10.1080/14034940410019190

216

Ø Næss et al.

may be of importance in measuring occupational hazards [12]. Education may be related to early lifecourse factors, as education is usually set at an early age [13,14]. Also, it could indicate a capacity to adopt a healthier lifestyle [15]. The effects of income on health could reflect purchasing power for healthy food, good quality housing, health services and so forth. This implies that empirical studies should compare the indicators with important diseasespecific outcomes and relate them to likely causal pathways. The aim of this analysis is to determine which is the best discriminator of socioeconomic differentials in mortality, both alone and controlled for each of the other variables. Second, we wanted to see if this relative ability to discriminate mortality is consistent across the most common causes of death. For this purpose we chose to link a census file with subsequent mortality data from Oslo.

Subjects and methods A cohort of all 88,159 inhabitants aged 45 to 64 years who lived in the municipality of Oslo on 1 January 1990 was chosen, as death in this age group is regarded as premature, and often used in social inequality studies. Data were available on a linked file from Statistics Norway. Variables Dependent variables were deaths, categorized as death from any cause, and the 12 most common causes of death in this group of men and women during the years 1990–98. All death certificates are registered with Statistics Norway, without any missing cases. The 12 most common causes of death were, according to ICD-9 (codes): stomach cancer (151), large-bowel cancer (153–154), pancreatic cancer (157), lung cancer (162), breast cancer (174), prostate cancer (185), alcohol-related diseases (291, 303, 305, 571), other cardiovascular diseases (390–409, 415–429, 440–459) coronary heart disease (410–414), stroke (430–438), chronic obstructive pulmonary disease (490–496), sudden unexpected deaths (798–799) and violent deaths (800–999). Independent variables were age and socioeconomic position determined separately by occupation, education, household income, and housing conditions registered in 1980. A period of 10 years before death was assumed to be sufficient for morbidity prior to death not to affect any of the explanatory variables. Each variable was grouped in

five hierarchical strata, with as similar distributions as possible. Occupation was classified according to the Eriksson–Goldthorpe scheme in five groups of employment (see Table I) [16]. Eriksson and Goldthorpe explicitly reject hierarchical ordering of occupational strata. Such an ordering is used in this analysis in order to present a measure across all categories that is comparable to the other measures of social position. Self-employed in 1980 could not be ordered in this scheme. Information on occupational class from 1970 or 1960 was therefore used when this was missing for 1980. Individuals who otherwise did not report any occupation were outside the labour force, such as students (0.1%) and those staying at home or working for wages for less than 10 hours a week (2.3%). Individuals who reported being retired in 1980 were classified according to their last occupation from 1960 and 1970 if that was known leaving only 0.1% as pensioners at all censuses. Married women were placed in their own occupational class. Education was defined as the highest obtained education in 1980 derived from the educational register collected by Statistics Norway. Length of education was ordered in five groups: first primary school (7–9 years of education), middle school (10– 11 years), secondary school (12 years), college (12– 16 years) and university (over 16 years). There were 7.0% missing values, mostly from Oslo inhabitants in 1990 not living in Norway in 1980. Household income was the taxable earnings in the household minus claimed tax according to the Revenue Authorities, deflated to 1990 NKr, and divided by the number of consumer units in the household. Consumer units were 1.0 for the first adult, 0.5 for each child and 0.7 for the married adult and each child of 16 or older living at home without income. Zero or negative income was treated as missing, this applied to 4.0% of the cohort in 1980. Housing conditions was an index based on the Census recording of six aspects of housing (values in parentheses): Living in blocks (0), attached houses (1), or flats (2); renting (0) or owning the dwelling (1); less than one room per person (0), 1–2 rooms per person (1), and more than 2 rooms per person (2); having telephone (1) or not (0); having WC in the dwelling (1) or not (0); and having bathroom (1) or not (0). This Index ranged from 0 (least favourable) to 8 (most favourable), but given small numbers in the least favourable two groups (0 and 1) and the most favourable four groups (5 to 8) these were collapsed, giving an index running from 1 to 5.

Four indicators of socioeconomic position

217

Table I. Employment status, education, net household income per consumer unit, and Housing Index in 1980 (%), by number of deceased and alive (1990–98) and odds ratio for all causes (95% CI) among 45- to 64-year-old inhabitants in Oslo, 1 January 1990. Women (n541,877) Inequality measures Occupation Class VII unskilled workers Class V/VI skilled workers Class III low level employees Class II medium level employees Class I high level employees

Deceased

Alive

Men (n537,644) Odds ratio

Alive

Odds ratio

1,092 949 748 807

5,264 6,002 5,371 7,635

2.15 1.66 1.52 1.22

798

8,978

1.00

513 443 825 545

5,202 5,358 10,426 8,976

1.62 1.40 1.38 1.10

508

9,081

1.00

1,303 852 338 255 85

12,546 13,534 4,880 5,629 2,540

2.20 1.49 1.57 1.15 1.00

1.77–2.75 1.19–1.86 1.24–1.99 0.90–1.47

1,677 849 1,065 430 373

8,220 5,944 8,879 4,245 5,962

2.59 1.95 1.68 1.46 1.00

2.32–2.90 1.73–2.21 1.50–1.90 1.27–1.68

Net household income 1–50.000 50–75.000 75–100.000 100–125.000 125.001+

217 607 993 755 262

2,191 10,045 14,024 9,248 3,535

1.93 1.19 1.14 1.13 1.00

1.61–2.32 1.02–1.37 0.99–1.30 0.98–1.30

318 969 1,474 1,133 500

2,015 9,185 10,931 7,439 3,680

1.57 1.00 1.03 1.05 1.00

1.36–1.80 0.89–1.11 0.93–1.14 0.94–1.16

Housing Index 1 (poor housing) 2 3 4 5 (good housing)

408 682 851 359 534

3,993 8,884 10,332 5,596 10,238

1.92 1.45 1.45 1.19 1.00

1.69–2.18 1.30–1.63 1.30–1.61 1.04–1.36

796 1,064 1,174 498 862

4,220 7,415 7,957 4,562 9,096

2.02 1.50 1.49 1.14 1.00

1.84–2.23 1.37–1.64 1.36–1.63 1.02–1.27

Education Primary school Middle school Secondary school College University

1.43–1.83 1.23–1.59 1.24–1.54 0.97–1.24

Deceased

1.97–2.36 1.51–1.82 1.38–1.68 1.10–1.34

Missing data

Statistics

The 11% of men and 7% of women who did not have a value for any of the independent variables or were self-employed were excluded from the analysis. The age-adjusted death rates were 143.0 per 10,000 person years among the included men and 155.8 among the excluded. The corresponding figures in women were 83.5 and 93.9, respectively.

Age-adjusted hazard ratios by the four ordinal social position measures were calculated using Cox proportional hazards regression (Table I). Relative indices of inequality (RII) were computed (Table II). The four different measures of social inequality have different proportions of the population in each of the five groups, and thus ratio

Table II. Age-adjusted relative index of inequality, RII (95% CI) of death crude and mutually adjusted by the four different measures of social position among 45- to 64-year-old inhabitants in Oslo, 1 January 1990. Inequality measures Occupation RII (Crude) RII (Mutually adjusted) Education RII (Crude) RII (Mutually adjusted) Net household income RII (Crude) RII (Mutually adjusted) Housing Index RII (Crude) RII (Mutually adjusted)

Women (n541,877)

Men (n537,644)

1.76 1.22

1.54–2.00 1.08–1.41

2.58 1.57

2.32–2.86 1.38–1.78

3.12 2.22

2.52–3.87 1.76–2.79

3.11 1.99

2.76–3.51 1.71–2.30

1.40 1.19

1.22–1.61 1.03–1.37

1.15 1.03

1.03–1.28 0.92–1.15

2.01 1.60

1.75–2.30 1.39–1.85

2.26 1.53

2.03–2.51 1.36–1.72

218

Ø Næss et al.

measures cannot be directly compared. The RII is constructed in order to avoid this problem. For each indicator and sex, a value between 0 and 1 was assigned according to the proportion of participants with a higher position than the midpoint of each group within the hierarchy in which that individual lay, starting with the best-off group. In the case of education a proportion of 0.17 of all men in 1980 were in the highest educational group; the mid-point individual would have a proportion of 0.085 of the population, giving a socioeconomic position (SEP) score of 0.085 for this group. In the next most favourable group were 12% of the population. All 17% in the better educational group are taken to be in more favourable circumstances than this group, and for the mid-point individual a proportion of 0.06 of the population would be above them (0.12/2). The SEP score for this group is therefore 0.17+0.06. This procedure was then continued for the rest of the groups, and the method applied to all the indicators in each sex. The SEP scores were related to mortality in Cox proportional hazards regression: The larger the RII, the greater the degree of inequality of deaths across the socioeconomic hierarchy. Here, RII is a measure of trend across the five groups.

Results The ordered occupational groups had increasing age-adjusted hazard ratios of death (see Table I). The education groups showed greater differences, whereas net household income per consumer unit showed a substantial difference only between the small bottom group and the others. The Housing Index demonstrated a similar step-wise mortality pattern to occupation. Occupational class, housing index, and education were correlated to some extent in both sexes whereas household income was weakly correlated with each of them (Table III). Each indicator was first investigated adjusting only for age. Occupation, education, and the Housing Index had rather similar estimates of RIIs in both sexes (see Table II). Household income showed a minor effect in men and women. The small RII in women corresponded to a large odds ratio in the lowest income group. When controlling each measure of social position for the other three, all differences in RIIs of death were reduced (see Table II). The RIIs in education remained somewhat larger than in the other measures. In women, the gradient for occupational class almost disappeared.

Table III. Correlation matrix class (1), level of education Housing Index (4) in both inhabitants in Oslo, 1 January

(Spearman’s rho) of occupational (2), household income (3), and sexes among 45- to 64-year-old 1990.

Women: Occupational class (1) Education (2) Household income (3) Housing index (4) Men: Occupational class (1) Education (2) Household income (3) Housing index (4)

(1) 1.00 0.44 0.11 0.33

(2)

(3)

(4)

1.00 0.11 0.29

1.00 0.08

1.00

1.00 0.59 0.08 0.36

1.00 0.07 0.36

1.00 0.03

1.00

We ran a separate analysis on married working women in our population (n523,094) depending on weather they were coded according to own or husband’s occupation. RIIs of all-cause mortality were, for own occupation, 1.66 (1.40–1.96) and for husband’s occupation 1.83 (1.50–2.23). The RIIs for specific causes of death showed a heterogeneous tendency in the ranking between the four indicators (Table IV). Household income was most poorly associated with many but not all causes of death. Breast cancer is notable, as low income was associated with increasing risk, and low education with decreasing risk. For some causes of death, such as coronary heart disease, stroke, stomach cancer, and chronic obstructive pulmonary disease, education was the strongest predictor closely followed by occupational class and housing conditions. Lung cancer and chronic obstructive pulmonary disease in men were associated with both occupational class and education. This was parallel to the all-cause pattern, partly reflecting the relative importance of these death categories as part of the total number of deaths. The pattern was less striking for some other causes such as violence in men, stroke in women, sudden deaths in both sexes, and alcoholrelated deaths in both sexes. Discussion Occupation, education, and housing conditions are all powerful discriminators of mortality risk in this population. Surprisingly, low household income did not discriminate mortality as well, particularly in men. We found no strong support for big differences between education, occupation and housing conditions as socioeconomic indicators related to all-cause mortality. This was similar to the British and Swedish but not the US situation. The latter found somewhat stronger support for income and education.

Four indicators of socioeconomic position

219

Table IV. Relative index of inequality comparing the 12 most common cause-specific death groups (deaths registered 1990–98) with those alive for four different indicators of social inequality unadjusted (M1) and mutually adjusted (M2) measured in1980 in women and men aged 45–64 years in Oslo, 1 January 1990. Socioeconomic indicator Occupational class Causes of death (n5number of deaths) Women Stomach cancer (n539) Large bowel cancer (n5151) Pancreatic cancer (n578) Lung cancer (n5148) Breast cancer (n5254) Alcohol-related (n582) Other cardiovascular (n5156) Coronary heart disease (n5339) Stroke (n5146) Chronic obstructive pulmonary disease (n5146) Sudden unexpected deaths (n582) Violent (n5164) Men Stomach cancer (n570) Large bowel cancer (n5218) Pancreatic cancer (n577) Lung cancer (n5222) Prostate cancer (n5) Alcohol-related (n5246) Other cardiovascular (n5319) Coronary heart disease (n51,084) Stroke (n5238) Chronic obstructive pulmonary disease (n5153) Sudden unexpected deaths (n5209) Violent (n5282)

M1

M2

Level of education

Household income

Housing conditions

M1

M2

M1

M2

M1

M2

12.18** 1.10 8.54** 5.79*** 0.60* 2.86 5.93*** 13.25*** 5.76*** 44.21***

11.74* 1.28 6.74* 3.41* 0.60 2.79 3.12* 7.11*** 2.77 20.00***

2.21 0.95 2.07 0.91 1.86** 1.85 1.59 1.81*** 2.55** 8.62**

1.51 0.96 1.70 0.69 2.07** 1.70 1.21 1.35 2.03* 5.01

3.28 1.18 1.11 1.71** 0.84 1.51 3.68*** 2.87*** 2.60** 6.05***

2.28 1.32 0.68 1.10 0.96 1.34 2.76** 1.83** 1.71 2.67**

1.47 0.71 2.30* 3.09*** 0.70 1.06 2.26** 2.65*** 2.74*** 4.36***

0.53 0.61 1.57 2.38** 0.76 0.70 1.28 1.44 1.70 1.27

1.59

0.88

3.77*

2.48

1.30

1.06

3.81**

3.36**

1.24

1.09

1.33

1.19

0.86

0.81

1.45

1.38

3.11** 0.83 1.44 4.79*** 1.15 4.46*** 1.50* 2.76***

1.27 0.74 1.02 3.07*** 0.83 1.57 0.71 1.50**

5.57*** 1.00 1.81 4.73*** 1.58 8.96*** 2.87*** 3.85***

3.17 1.14 1.63 2.04* 1.69 4.45** 2.65*** 2.53***

2.10 0.89 0.86 0.99 1.58 1.69*** 1.07 1.20

1.82 0.89 0.82 0.83 1.54 1.41 0.98 1.06

4.10** 1.08 1.59 2.52*** 1.20 5.10*** 2.62*** 2.36***

2.53 1.15 1.34 1.37 1.04 2.73*** 2.12*** 1.50**

3.05*** 11.04***

1.89* 5.37***

3.58*** 12.77***

2.17* 3.24**

1.53 1.74

1.37 1.39

1.95* 4.24***

1.19 1.69

3.17***

2.05*

3.07***

1.50

1.06

0.94

2.96***

2.02**

2.75***

2.13**

2.17***

1.08

0.76

0.69

2.76***

2.09**

*p-valuev0.05, **p-valuev0.01, ***p-valuev0.001.

With respect to cause-specific analyses, education was strongly associated with stomach cancer and chronic obstructive pulmonary disease risk, and to a lesser extent with cardiovascular disease. This is strikingly similar to associations seen with childhood social circumstances [4,19]. Therefore it is likely that education is serving in part as an indicator of social conditions in childhood. With respect to stomach cancer, the acquisition of H pylori infection in childhood is an important aetiological factor and is related to overcrowded living conditions. Families that come from such backgrounds are also likely to obtain less education for their children. The study thus provides some support for the notion that childhood social circumstances are of importance with regard to generating inequalities in health for some specific causes.

Sudden unexpected deaths in both sexes and violent deaths in men were better or similarly well predicted by occupational class or housing conditions as by education. These causes of death are perhaps less likely to be influenced by childhood social conditions and more by occurrences proximal to the event. Lung cancer mortality – which is triggered by cigarette smoking – was strongly related to both education and occupation. In a Scottish study, occupational social class was more strongly related to smoking behaviour than education [4]. Chronic obstructive pulmonary disease, which is associated with childhood social conditions as well as smoking behaviour in adult life, is associated with both education and occupation in men but not in women. This could indicate a direct influence of education on the ability to avoid taking up smoking

220

Ø Næss et al.

or on quitting once the habit has been established. It also suggests that the socioeconomic determinants of smoking may be context-specific. The strength of the study is that it comprises nearly the whole population in the city of Oslo in a homogeneous age group. Deaths are registered 10– 18 years after the measures of social position. This avoids health-related downward social drift generating associations between social position and health [17]. Missing data were mostly due to individuals not living in Norway in 1980 or previously as the population was defined in 1990. The low predictive ability of household income as opposed to the US National Longitudinal Mortality Study could be a reflection of low-income inequality in Norway, in accordance with a recent Danish report [22]. However based on European income surveys, income inequality in Norway measured with the Gini coefficient was comparable with West Germany and Sweden in 1985 [20]. Lack of an association could also be a spurious finding due to attempts at reducing tax liability. Legal options for such reductions were substantial in 1980 before new tax legislation was purposely set up to prevent this. We reanalysed the data comparing our results with household income in 1990. This gave RII values for all causes of death of 3.84 in men and 3.43 in women. Even after taking the possible effect of reverse causality into account, this suggests that misreporting is the most probable reason for a weak association with 1980 income. Housing conditions, used as an additional indicator in this study, are a composite measure. Several studies have demonstrated an independent effect of housing on health, which suggests that it is not simply a proxy indicator of income or education but rather may have an independent healthdamaging impact [23]. This is a strong indication that studies on health inequalities should go beyond the classical triad of occupational class, education, and income. Occupational class may reflect distinct dimensions of occupational exposures, which is supported by our data as the effect of husband’s occupational class in women almost disappears when adjusting for the other indicators. This adjusted effect is also strongly attenuated when women are coded by their own occupation, indicating that women’s and men’s occupational exposures may differ. The relation between education and mortality may be explained by several possible mechanisms: the material and cultural resources of the family of origin, the fact that education is strongly related to occupation and income in adult age, that education

may influence receptivity to health education messages, that it may be linked to common background factors influencing both capacity to complete education and maintain health, and finally that ill health in childhood may limit educational attainment and predispose towards later outcomes. Disentangling the independent contribution of these processes empirically remains largely to be investigated [24]. Our main findings in this study are that education, occupation, and housing all seem to predict all-cause mortality risk to a similar degree. These three indicators seem to have separate associations and their relative rank across the most common causes of death varies. Combined measures that are locally validated and positioned appropriately in the causal pathway of particular outcomes should be used in future aetiological research on the effect of social circumstances on health. Acknowledgement This research has been funded by Health and Rehabilitation. Thanks are due to Professor Haakon Gjessing for statistical advice. References [1] Mackenbach JP, Kunst AE. Measuring the magnitude of socio-economic inequalities in health: an overview of available measures illustrated with two examples from Europe. Soc Sci Med 1997;44:757–71. [2] Kunst AE, Mackenbach JP. Measuring socioeconomic inequalities in health. Copenhagen: World Health Organization, 1995. [3] Syme SL, Moss N, Krieger N. Recommendations of the conference ‘‘Measuring Social Inequalities in Health’’. Int J Health Serv 1996;26:521–7. [4] Davey Smith G, Hart C, Hole D, et al. Education and occupational social class: which is the more important indicator of mortality risk? J Epidemiol Community Health 1998;52:153–60. [5] Sorlie PD, Backlund E, Keller JB. US mortality by economic, demographic, and social characteristics: The National Longitudinal Mortality Study. Am J Public Health 1995;85:949–56. [6] Daly MC, Duncan GJ, McDonough P, Williams DR. Optimal indicators of socioeconomic status for health research. Am J Public Health 2002;92:1151–7. [7] Sundquist J, Johansson SE. Indicators of socio-economic position and their relation to mortality in Sweden. Soc Sci Med 1997;45:1757–66. [8] Braveman P, Cubbin C. Optimal SES indicators cannot be prescribed across all outcomes. Am J Public Health 2003;93:12–3. [9] Wadsworth ME. Health inequalities in the life course perspective. Soc Sci Med 1997;44:859–69. [10] Najman JM. Theories of disease causation and the concept of a general susceptibility: a review. Soc Sci Med 1980;14A:231–7.

Four indicators of socioeconomic position [11] Davey Smith G, Leon D, Shipley MJ, Rose G. Socioeconomic differentials in cancer among men. Int J Epidemiol 1991;20:339–45. [12] Karasek R, Theorell T. Healthy work. New York: Basic Books, 1990. [13] Liberatos P, Link BG, Kelsey JL. The measurement of social class in epidemiology. Epidemiol Rev 1988;10:87–121. [14] Davey Smith G, Hart C, Blane D, Hole D. Adverse socioeconomic conditions in childhood and cause specific adult mortality: prospective observational study. Br Med J 1998;316:1631–5. [15] Lynch JW, Kaplan GA, Salonen JT. Why do poor people behave poorly? Variation in adult health behaviours and psychosocial characteristics by stages of the socioeconomic lifecourse. Soc Sci Med 1997;44:809–19. [16] Eriksson R, Goldthorpe JH. The constant flux. Oxford: Oxford University Press, 1999. [17] Dahl E. Social inequality in health – the role of the healthy worker effect. Soc Sci Med 1993;36:1077–86. [18] Blane D, Davey Smith G, Bartley M. Social selection: what does it contribute to social class differences in health? Sociol Health Illn 1993;15:1–15.

221

[19] Davey Smith G, Gunnel D, Blane D. Life-course approaches to socio-economic differentials in cause-specific adult mortality. In: Leon D, Walt G, editors. Poverty, inequality and health. Oxford: Oxford University Press, 2000: 88–124. [20] Atkinson A, Rainwater L, Smeeding T. Income distribution in European countries. Luxembourg Income Study. Working paper No. 121. Luxembourg: CEPS/INSTEAD, 1995. [21] Wessel T. Losing control? Inequality and social division in Oslo. European Planning Studies 2001;9:889–906. [22] Osler M, Prescott E, Gronbaek M, Christensen U, Due P, Engholm G. Income inequality, individual income, and mortality in Danish adults: analysis of pooled data from two cohort studies. Br Med J 2002;324:13–16. [23] Macintyre S, Hiscock R, Kearns A, Ellaway A. Housing tenure and health inequalities: a three-dimensional perspective on people, homes and neighbourhoods. In: Graham H, editor. Understanding health inequalities. Buckingham; Philadelphia: Open University Press, 2000. [24] Blane D. Commentary: Explanations of the difference in mortality risk between educational groups. Int J Epidemiol 2003;32:355–6.