METHODS

All Norwegians were assigned a Personal Identification Number in 1964 that was projected back to the census of 1960 for those living in Norway at the time. Thus it was possible to link census data from 1960, the 1990 Personal Register, and the mortality register 1990–94 in order to study social conditions over a span of 30 years and subsequent mortality.

The data included all 128 723 persons aged 31 to 50 years who lived in the municipality of Oslo on 1 January 1990. This sample was chosen because they were children and teenagers in 1960, and thus the 1960 census data provided information on childhood social circumstances. Some 1655 deaths occurred in the 1990–1994 period.

Mortality

All death certificates are registered with Statistics Norway. Causes of death were classified as cardiovascular diseases (ICD-9 390–459), cancer (140–208), psychiatric (290–319), accidents/violence (800–999) and others. Cancers were also divided into smoking related, (lip (140), tongue (141), mouth and pharynx (143–149), oesophagus (150), pancreas (157), respiratory tract (160–163), and urinary tract (188–189)), and non-smoking related (other cancers).

Housing conditions

The 1960 census recorded details on six aspects of housing (table 1), which are known to be related to the living standards of the households in Oslo.⁹ These variables were summarised according to the values in table 1 into a Housing Index, ranging from 0 (least favourable) to 8 (most favourable). Given small numbers in the least favourable two groups (0 and 1) and the most favourable two groups (7 and 8) these were collapsed, giving an index from 1 to 7 (table 2).

Income

A census was not held in 1990, but was replaced by a Personal Register in Statistics Norway, which does not contain housing information. Instead we used taxable earnings from the taxation authorities. These were yearly wages, social security benefits, and other earnings minus a personal allowance of about 25 000 NKr and other tax deductible expenditures. The taxable incomes were added for married couples and taxes were subtracted, giving net household income. This amount was divided by the figure of “consumer units” in the household, which is 1.0 for the first adult person, 0.7 for spouse and each child of 16 or older living at home without taxable income, and 0.5 for each child under 16 years of age. If older children had taxable earnings, they were registered as a one person household.

To compare the middle aged standard of living with the childhood/adolescence standard of housing conditions, net household income per consumer unit was grouped in seven categories, with similar proportions of the population falling into the categories as is seen with the Housing Index: (1) 1–25 000 NKr, (2) 25 001–50 000 NKr, (3) 50 001–75 000 NKr, (4) 75 001–100 000 NKr, (5) 100 001–125 000 NKr, (6) 125 001–150 000 NKr, (7) 150 001+ NKr. ($1 was about 6.25 NKr in 1990).

Missing data

Completing the census is a legal obligation. Those who did not return the form by post were visited at home by officials who collected the data. Although there were no missing forms some people did not answer all questions on the forms. Some 18% of the sample did not have data for any or all the housing questions in 1960. The 1990 Tax Register recorded zero or negative income for 6% of the sample, which were regarded as missing values. Responders were those both answering the 1960 housing issues and having an income in 1990 above zero. A total of 21% had incomplete data. The most common reason was not living in Norway in 1960. Some 11% were born outside Norway and 2% were Norwegians living abroad in 1960, both groups therefore not participating in the 1960 census. Sixty one per cent of missing cases were attributable to these reasons. The relative rates of mortality for those excluded from full analysis compared with those included were 1.05 (95% CI 0.92 to 1.20) for men and 1.04 (0.84 to 1.30) for women, indicating small differences in mortality risk.

Statistics

Crude mortality rates by housing variables and income were directly age standardised with four five-year age bands in the statistical package Stata, with the whole cohort as standard population (tables 1–3).

To compare mortality rates by the seven social positions, a measure called the relative index of inequality (RII) was used.^10,11 The measures of socioeconomic position in 1960 and 1990 have different (although similar) proportions of the population in each of the seven groups. Therefore a direct comparison of rate ratios is not possible. The RII is constructed to avoid this problem.

For the 1960 and 1990 socioeconomic position measures, each group was assigned a value between 0 and 1 according to the proportion of participants with a higher socioeconomic position than the midpoint of each group within the hierarchy which that person lay, starting with the best off group (group 7). In the case of the 1960 housing index a proportion of 0.12 of all men in 1960 were in the highest Housing Index group, the midpoint person would have a proportion of 0.06 of the population having more favourable socioeconomic circumstances, giving a socioeconomic position (SEP) score of 0.06 for this group. In the next most favourable group (group 6), there were 11% of the population. All 12% in the better housing conditions group (group 7) were taken to be in more favourable circumstances than this group, and for the midpoint person a proportion of 0.055 of the population in group 6 would be above them (0.11/2). The SEP score for this group was therefore 0.12 plus 0.055=0.175. This procedure was then continued for the rest of the groups, and the method applied both to 1960 housing index scores and the 1990 income groups. The SEP scores were related to mortality in logistic regressions:

where D is the odds of death, and Agecat is the five year age bands, providing age adjustment for the RII. Exponentiating b yields the RII, which is the odds ratio of mortality for the hypothetically poorest individual compared with the hypothetically richest individual; SEP score =1 compared with SEP score =0. The larger the RII, the greater is the degree of inequality of deaths across the socioeconomic hierarchy.

RESULTS

Age adjusted mortality when the cohort was from 31 to 55 years of age in 1990–94, was 34 per 10 000 person years for men and 18 for women (table 1). Male mortality was lower among those with better housing conditions in 1960. Mortality was particularly related to not having a telephone, bath or toilet inside the dwelling, but also to crowding and type of dwelling for men. Female mortality followed the same pattern, but the differences tended to be smaller.

Indices of socioeconomic position

A Housing Index with seven steps was constructed using the values in table 1. The age adjusted mortality rate for men was 2.4 times higher in the group in least favourable housing circumstances than the group in the most favourable circumstances during childhood (table 2). The corresponding figure for women was 1.6.

Mean net household income per consumer unit in 1990 was 81 203 NKr (1990 $12 992) for men and 78 799 NKr ($12 608) for women. The age adjusted mortality rate for men was 3.3 times higher in the lowest income group compared with the highest; among women the equivalent difference was 3.6 (table 2).

The correlation between the two indices was low, at 0.08 for both men and women. In table 3 mortality rates across a compressed form of the indices of social position in 1960 and 1990 are presented. For men in general, the 1960 index predicted mortality within strata of the 1990 index and vice versa. For women the patterns were less clear, perhaps because of the low number of deaths in some strata.

Relative indices of inequality for mortality

Table 4 shows cause specific and all cause death rates by childhood/adolescence (1960) and adulthood (1990) socioeconomic position. Socioeconomic differences were seen for most of the mortality categories that have been examined; the main exceptions being cancers not related to smoking (for which sizeable gradients were seen for neither men nor women), and cancers related to smoking (for which sizeable effects were not seen for women, although they were seen for men). Psychiatric mortality was almost exclusively made up of deaths attributed to alcohol or drug dependence.

Cardiovascular mortality was more strongly related to childhood socioeconomic circumstances than to adulthood circumstances for both men and women (table 4). Other causes of death tended to be more strongly associated with income in adulthood than housing conditions in childhood.

In table 5, the 1960 and 1990 socioeconomic position indicators are related to all cause and cause specific mortality after mutual adjustment. In line with the simple associations, childhood social circumstances seem particularly important with respect to cardiovascular disease mortality in both men and women, and mortality within the “other causes” group for men. Interaction terms indicated that mortality for people in adverse social circumstances both in 1960 and 1990 were above the levels predicted by a simple multiplicative model for all cause and cardiovascular mortality among men and women, accidental and violent deaths among men, and psychiatric mortality among women.

DISCUSSION

This study confirms the finding that socioeconomic deprivation in early life is associated with adulthood mortality.^1–4 The greater relative importance of childhood social circumstances than adulthood social circumstances for cardiovascular disease mortality, reported previously for men,⁴ was also seen in this study. We have extended previous findings to show that the same pattern is seen for women. For other cause of death groups, adulthood social position tended to be of greater relative importance. This study is based on an entire population sample and thus has advantages over most previous investigations of this issue.

We found strong associations of both childhood and adulthood social circumstances and all cause mortality in this study, in line with previous investigations.^3,12 A striking inverse association between adulthood income and mortality was seen in the present data, as in previous studies from other countries,^13,14 and previous studies from Norway.^15,16 While it is sometimes suggested that ill health in adulthood leads to lower income, and thus generates the association between income and mortality, it is clear that adulthood health cannot influence childhood social circumstances. Thus the observation of a marked gradient between childhood socioeconomic position and all cause mortality is a clear demonstration that socioeconomic differentials in mortality are not primarily caused by such reverse causation, adding to previous evidence in this regard.^17,18

Income data were not available for households in 1960, therefore housing condition information has been used. In Norway, housing conditions have been shown to correlate strongly with other measures of socioeconomic position.⁹ Norwegian families spend much of their income on their dwelling, making this a particularly sensitive indicator of social circumstances.

The highest risk of mortality among men and women was observed in the group who were poor both in childhood and in adulthood, but clear gradients were found through the steps of the 1960 and 1990 social position measures. This demonstrates a clear cumulative influence of social circumstances across the lifecourse on mortality risk.^4,19–21 In this study, there were significant interactions such that people in adverse socioeconomic circumstances in both childhood and adulthood had higher mortality than would be expected on the basis of a simple multiplicative model. This lifecourse cumulative influence seems especially important for the multifactorial course of cardiovascular disease risk, but also for female psychiatric deaths and male violent deaths suggesting an additional health burden for people who experience economic hardship right the way across their lives.

Clear childhood socioeconomic position effects have also been seen for coronary heart disease in previous studies.^6,7,22,24 The accumulation of the influences of socially patterned risk factors for cardiovascular disease acting across the lifecourse—from low birth weight,^25,26 suboptimal growth in childhood,²⁷ smoking, excessive (binge) alcohol consumption²⁸ and high saturated fat/low fruit and vegetable diet in adulthood, and job insecurity in adulthood²⁹ would lead the observed pattern—with a contribution of both childhood and adulthood social position to cardiovascular mortality to be anticipated. In previous studies particularly strong influences of childhood social circumstances have been demonstrated for stroke in comparison with coronary heart disease.^3,30,31 In this study, however, there were only 39 stroke deaths, reflecting the comparatively young age at death in this cohort.

Health related behaviours, such as smoking and binge drinking, tend to be more strongly socially patterned by adulthood rather than childhood social circumstances,³² although some influence of childhood social position can be seen,³³ especially with respect to smoking in women.³⁴ This greater influence of adulthood social environment on the maintenance of health related behaviours would lead to the expectation that smoking related and alcohol related causes of death, such as smoking related cancer, psychiatric causes, and accidents/violence, would be more strongly related to adulthood social position, as was observed in our data. This is in keeping with evidence that the influence of childhood social circumstances on adulthood smoking is considerably greater in women than in men.²⁷

In this study, we are largely interpreting the associations with housing conditions in early life and income in later life as reflecting socioeconomic circumstances in general. It is also possible that they could be closer correlates of direct causal factors related to disease risk. For example, deprivation in childhood is related to stomach cancer risk.^3,35 This is probably attributable to acquisition of Helicobacter pylori infection in childhood, which would be directly related to overcrowded living conditions and ability to maintain adequate hygiene practices (reflected in possession of a toilet within the dwelling or a bath). In this study we do not have enough events to study stomach cancer mortality, but it is also possible that cardiovascular disease risk is influenced by infections acquired in childhood. The dual nature of measures of social position—both as proxies of underlying general social circumstances, and also as factors that may have a more direct influence on disease risk is a general—but, underdiscussed, aspect of research into socioeconomic inequalities in health.

There are three principal conclusions from our study. Firstly, consideration of the social distribution of adult mortality risk must evaluate the influence of socioeconomic circumstances acting across the entire lifecourse. If data covering only one point in time are used—whether this is early life or adulthood—there are serious limitations to the ability to determine the key socioeconomic influences on health, or to appreciate the magnitude of these influences. Secondly, the influence of socioeconomic circumstances at different stages in life varies between particular causes of death. Cardiovascular disease mortality is particularly influenced by childhood social circumstances, whereas mortality related to adulthood health related behaviours is more strongly influenced by adulthood social position. The specificity of these associations reinforces the limited degree of explanation that can be provided by models of health inequalities based on a notion of a heightened general susceptibility to diseases among those in poor social circumstances.^35,36 Finally, current changes in mortality rates for cardiovascular disease reflect in part social changes that occurred some decades ago. Current increases in child poverty that have been seen in many countries—including Norway—over the past two decades³⁷ could herald unfavourable future trends in adult health.