Trends in educational inequalities in mortality, seven types of cancers, Norway 1971–2002

Abstract

Background: Knowledge about educational disparities in deaths from specific cancer sites is incomplete. Even more scant is information about time trends in educational patterns in specific cancer mortality. This study examines educational inequalities in Norway 1971–2002 for mortality in lung and larynx, colorectal, stomach, melanoma, prostate, breast and cervix uteri cancer. Methods: A data file encompassing all Norwegian inhabitants registered some time during 1971–2002 while aged 45–74 was constructed with linked information from administrative registers. During an exposure of more than 40 millions person-years, about 87 000 deaths in the analysed cancer types were registered. Absolute and relative inequalities during three periods were analysed by age-standardized deaths rates, hazard regression odds ratios and Relative Index of Inequality. Results: Educational inequalities in lung and related cancer mortality widened considerably from the 1970s to the 1990s for both sexes. The moderate educational gradient for stomach and cervix uteri cancer persisted, as did the weak gradient for colorectal cancer. No educational differences in prostate cancer were observed in any of the time periods. The modest inverse educational gradients in deaths from breast cancer and melanoma remained at the same level. Conclusion: Among the seven cancer types examined in this study, only lung cancer mortality showed a clear widening in educational disparities. As lung cancer mortality constitutes a large proportion of all cancer deaths, this increase may result in larger disparities for overall cancer mortality. Some explanations for the observed patterns in cancer mortality are suggested.

Introduction

Cancers cause a large proportion of the premature deaths in Europe,¹ and cancers contribute to the patterns of socio-economic mortality inequalities, but in intricate and partly contradictory ways. Mortality from lung cancer in particular, but also from stomach and (among women) cervix uteri cancer have been found to be higher among low-educated persons.^2–5 Mortality in breast cancer, a major cause of death among middle-aged women, is on the other hand usually found to be higher among the highly educated.^2–4,6 Moreover, the socio-economic and educational gradient varies in several ways. Social differences in male lung cancer mortality appear large in Central and Eastern Europe and considerable in the Nordic countries. In several South European countries, however, studies have found a reversed educational gradient in women's lung cancer mortality.^3,4,7,8 The educational gap in alcohol-related cancers seems on the other hand clearly larger in some South European countries than in North Europe.⁹ Adding all cancer deaths, educational disparities in overall cancer mortality during the 1990s among men were substantial in the North and West of Europe and even larger in Central and Eastern Europe,³ but in Slovenia and some Spanish provinces, overall cancer mortality among women was actually higher among the highly educated.^2–4

A further addition to these complexities is that the socio-economic gradient changes over time, but not necessarily in a uniform manner. Increasing socio-economic disparities in lung cancer mortality have been observed in the USA,¹⁰ France,¹¹ Norway¹² and Australia,¹³ but not in Finland 1981–95^1,4 nor in Barcelona 1992–2003.¹⁵ As for breast cancer, a levelling off as regards the previous higher mortality among highly educated women has been observed in Finland¹⁶ and France.¹⁷ Among French men, increasing occupational inequalities occurred in mortality from upper aerodigestive tract cancers.¹¹ A study from Barcelona found, on the other hand, only small changes over time in educational inequalities in mortality for a number of cancer sites.¹⁵ Several studies have suggested that socio-economic disparities in overall cancer mortality are on the increase,^10,13,18,19 but a recent study from The Netherlands found actually a narrowing socio-economic gap in overall cancer incidence during 1996–2008.²⁰

All in all, few studies have analysed how socio-economic inequalities in cancer mortality have evolved in recent decades. The aim of the present study is to increase knowledge about developments in educational inequalities in mortality from specific cancer sites. In Norway, overall cancer mortality among men increased moderately from the early 1970s to the mid-1980s, followed by a slight decline during the 1990s, while the overall cancer mortality rate among women has been quite stable during this period.²¹ Less is known about time trends in educational disparities for specific cancer types. Here, we analyse the three largest cancer types: lung and related sites, colorectal and prostate among men (53% of new cancer cases in 2005), and breast, colorectal and lung among women (47% of new cases in 2005).²² Furthermore, two cancers with declining incidence in recent decades (stomach, cervix uteri) and one with rising incidence (melanoma of the skin) are examined.

Methods

Data

For research purposes, a data file with anonymized individual information was constructed by Statistics Norway. Administrative registers covering the entire Norwegian population were linked by means of the personal identification number. Data utilized for this study encompass all inhabitants who were both registered as residing in Norway sometimes during 1971–2002 and aged 45–74 years sometime within that period. The Norwegian population in this age range numbered about 600 000 men and 650 000 women in the early 1970s, gradually increasing to about 640 000 men and 670 000 women in the late 1990s. For each individual, a series of 1-year observations was created, starting in 1971 or the year the person turned 45 years and ending either with the year of death, or with emigration, or by the year the person turned 75 years, or in 2002, whichever came first. Each observation included variables about the situation at the beginning of the year and an outcome variable indicating whether the person died that year from one of the seven analysed cancer types. The oldest individuals in the data were born in 1897 (i.e. age 74 in 1971), while the youngest were born in 1957 (and therefore 45 in 2002).

Measurements

Within each category, number of deaths for each cancer type and number of person-years were summarized for 1971–79, 1980–89 and 1990–2002. The classification of causes of death followed the European shortlist, using the ninth or tenth version of the International Classification of Disease (ICD).²³ The following ICD codes were used: Cancer in larynx and lung, bronchus and trachea ICD-9 161–162 and ICD-10 C32–C34; colorectal ICD-9 153–154 and ICD-10 C18–C21; stomach ICD-9 151 and ICD-10 C16; melanoma of the skin ICD-9 172 and ICD-10 C43; prostate ICD-9 185 and ICD-10 C61; breast ICD-9 174–175 and ICD-10 C50; and cervix uteri ICD-9 180 and ICD-10 C53.

In addition to sex and age, information about highest educational level was obtained from the educational register after 1980 and from censuses of 1970 and 1980 with fairly comparable information for earlier years. A large expansion in education has occurred in Norway. University and similar higher education was quite rare among the oldest cohorts, especially among women, but became much more available among those born around the mid-20th century. For the present analyses, educational information was simplified into a three-level hierarchy—basic, secondary and tertiary education—ensuring sufficient number of deaths within each category and making our findings comparable to other studies.¹² Basic (compulsory) education implies up to 7 years of schooling for the older cohorts and usually 9 years for those born since the late 1940s (9-year compulsory schooling was introduced in Norway during the 1960s). Secondary education comprises additional education above the compulsory level, up to 12 years in total length of education. Tertiary education comprises all levels above secondary, often college or university education, making total education duration about 13 years and more. Those very few (less than 0.1 %) lacking codifiable educational information were excluded from the analyses.

Statistical analysis

Following common practice,^9,12,15,24 both absolute and relative educational inequalities are analysed. As to absolute differences, directly age-standardized mortality, i.e. deaths per 100 000 person-years, was estimated for each time period by summing number of person-years and number of deaths during the respective periods. Calculations were made for each gender and each educational level, separately for the seven cancer sites. The European standard population²³ was used for age standardization. The age span 45–74 implies six 5-year categories, which were given the following weights, from the youngest (45–49) to the oldest (70–74): 0.219, 0.219, 0.188, 0.156, 0.125 and 0.093. The size of absolute educational disparities for each time period was calculated by subtracting mortality rates for basic and secondary education, respectively, from the mortality rate of tertiary education.

As to relative disparities, odds ratios (ORs) for deaths from one of the studied cancer types were calculated for secondary and basic education, relative to the reference category of tertiary education. Calculations were performed by discrete time hazard regression analysis (i.e. for each time period, logistic models were estimated from all 1-year observations with cause-specific deaths as the outcome variable and age, calendar year and education as independent variables).²⁵

Furthermore, the Relative Index of Inequality (RII)²⁴ was calculated for each gender and time period. The RII takes into account not only the relative mortality differences between educational levels, but also the relative size of the educational categories. This is particularly relevant for judging changes in relative disparities when the educational distributions change considerably over time, as is the case in this study. The RII is calculated by transforming the educational variable, an ordinal scale, into a ranked interval scale with values for each educational level indicating the mean proportion with higher education than that particular educational level. Given that 20% of the population have the highest, tertiary education, 30% have secondary and 50% have basic education, those with tertiary education will be coded 0.20/2 = 0.10; those with secondary education obtain the value (0.20 + 0.30/2 =) 0.35, and those with basic education get the value (0.20 + 0.30 + 0.50/2 =) 0.75.²⁶ This transformed education variable assumes that educational resources can be ranked from zero to unity (from 0 to 1), and the underlying assumption is that the educational hierarchy is linearly associated with the outcome. When the new education variable is used as a predictor in the hazard regression models, the exponential of the coefficient [exp(b)] will indicate the relative difference in death risk between those with the hypothetical position of being at the very lowest end of the educational scale, compared to the hypothetical position of being ‘on the very top’ of the educational hierarchy.¹¹

Results

All in all, the material comprises nearly 50 000 male deaths and close to 37 000 female deaths in the seven cancer types during 1971–2002 (table 1). Total exposure time was more than 20 million person-years for each gender. Among men, deaths from lung cancer and related sites were by far the most common among the studied cancers. Among women, deaths from breast cancer were the most frequent (about one-third of the studied deaths), followed by colorectal cancers, but by the 1990s, lung cancer deaths had become more frequent than colorectal cancer deaths among women. As regards age-standardized mortality, also shown in table 1, the most conspicuous changes from the 1970s to the 1990s are the increase in death rates from lung cancer among women, from 15 deaths per 100 000 person-years in 1971–79 to 48 in the 1990s, and the steady decline in death rates from stomach cancer during the same period from 48 to 23 for men and 23 to 11 among women. The increase during the 1970s in lung and larynx cancer death rates among men seemed to level off after the 1980s. The rates among men for colorectal and prostate cancers suggest a slight increase (from 41 to 51 and from 33 to 39, respectively). Several main cancers among women had a considerable stability in mortality rates during the observed period: Around 35–38 deaths per 100 000 person-years for colorectal cancer, and 54–57 for breast cancer. As for the lesser cancer types among women, the death rate went slightly up for melanoma and declined somewhat for cervix uteri cancer. In addition to the cancer deaths data, table 1 also shows how the population's educational level moved ‘upwards’ during these three decades.

Tables 2 and 3 report the time trends in educational inequalities. The most conspicuous change from the 1970s to the 1990s is the growing educational gradient in mortality for lung cancer (including larynx, bronchus, trachea). This can be seen for both sexes, both in absolute (table 2) and relative terms (table 3). In the 1970s, men with basic education had 22 more deaths per 100 000 person-years in lung and related cancers than men with tertiary education, a gap which increased to 59 in the 1990s. In the first time period, absolute inequalities in women's lung cancer deaths were negligible, but in the 1990s, women with basic education had a clearly higher death rate than women with higher education. The growing inequalities in lung cancer deaths are also reflected in relative differences. ORs for basic education were, for men and women, respectively, 1.37 and 1.12 in the 1970s and 2.02 and 2.17 in the 1990s. The RII increased from 1.17 to 2.78 for men and from 0.76 to 3.18 for women.

As regards the other cancer sites, however, changes in educational patterns were more modest. Educational differences in colorectal cancer deaths were quite small across the entire period, with a tendency, especially in the latter period, to lower death rates among tertiary educated, most visible among women. For prostate cancers, fairly stable in total death rate, educational differences were practically absent during the entire observation period and for breast cancer, the mortality disadvantages of higher educated women remained fairly unchanged. The decline in mortality from stomach cancer occurred in all educational categories. The absolute inequalities, especially for men, were slightly reduced, but the relative disadvantages of the lower educated continued by and large unchanged, indicated by the RII, which was 1.82 for men and 2.13 for women in the 1970s and 2.14 and 2.21, respectively, in the 1990s.

Estimations for cervix uteri and melanoma of the skin may be troubled by the relatively small number of deaths. During the entire period, lower educated women had higher death rates from cervix uteri cancer; the relative disparities seemed fairly stable, but a slight tendency to reduced absolute inequalities can be observed. A tendency to more unfavourable death rates from melanoma of the skin among highly educated men was observed during the entire observation period, but among women, the reversed educational gradient for melanoma which was vaguely noticeable in the 1970s, had disappeared by the 1990s.

Discussion

The most distinct finding in this study is that educational inequalities in mortality from lung and related cancer types increased substantially, both among men and women, from the 1970s to the late 1990s. For the six other analysed types of cancer, educational patterns in death rates changed comparatively little. As to stomach cancer for both sexes and cervix uteri cancer among women, the educational gradients in death rates continued without drastic changes. The weak educational gradient for colorectal cancer in the 1970s persisted fairly unchanged among men but with a slight widening among women. No definite educational differences for prostate cancer deaths could be established in any of the observation periods. Finally, a weak tendency to a reversed educational gradient, with higher death rates among the highly educated, was observed for breast cancer among women and for melanoma of the skin among men. Also this pattern remained fairly constant during the observation period.

Thus, only for lung and related cancer mortality could a substantial widening of inequalities be observed, but since lung cancer deaths constitute a considerable and, for women, increasing part of all cancer mortality, the growing educational gap for this death cause will tend to result in increasing educational gradients in overall cancer mortality.

The findings concur in many ways with recent results from comparable cross-sectional studies in Western Europe: very marked educational differences in lung and related cancer mortality (exception: women in parts of South Europe),^2–4,7,8,14 moderate, but clear, disparities in deaths from stomach cancer and cervix uteri cancer,^2,4,15 a quite weak educational gradient in colorectal cancer mortality,^2,4,26 no inequalities in deaths from prostate cancer,^2,4,15,26 and a modest reversed gradient for deaths in breast cancer.^2–4,6 The present study found also a slight reversed educational gradient in deaths from melanoma of the skin, similar to the pattern observed in a Spanish study.²⁷

The specific contribution of the present study is the examination of time trends. The finding of a marked widening of educational differences in lung cancer mortality is parallel to previous Norwegian¹² and French observations,¹¹ but somewhat at odds with Finnish comparisons between the 1980s and the 1990s^1,4 and a Barcelona study for the 1992–2003 period.¹⁵ Our study displayed a rather stable, but modest, reversed gradient for breast cancer mortality, similar to the Finnish¹⁴and Barcelona¹⁵ studies, but deviating slightly from the French study suggesting a disappearance during 1968–96 of the reversed gradient.¹⁷ It seems that only two studies in Western Europe, from France¹¹ and Barcelona,¹⁵ have fairly comparable data about trends in socio-economic differentials in mortality for the other cancer sites studied here. Our study confirms, by and large, their results of a considerable stability in the educational pattern of stomach, colorectal, prostate and cervix uteri mortality.

Interpretations

The main explanation for the trends in lung and related cancer mortality inequalities will most certainly be the trends in tobacco use. Since the 1970s, smoking has been reduced in Norway, but the decline started later among women than among men and has been much larger among the highly educated than among lower educated.²⁸ The slight overall mortality increase from melanoma of the skin, as well as the modest reversed educational gradient, could well be related to the rising affluence in Norway in recent decades, which has led to more unhealthy sunbathing, first occurring among upper socio-economic strata.²⁹ The decline in stomach cancer mortality, fairly parallel in all educational strata, could be related to historical changes in dietary factors affecting practically all social strata in the latter half of the 20th century.

The full explanation of the changes in and persistence of educational patterns in specific cancer mortality will certainly be complex. Cancer mortality is a function of incidence influenced by exposures to risk factors and perhaps some inherited factors,³⁰ and survival that depends, among other things, on health service factors.³¹ Differences between educational groups in lifestyle trends may contribute to changing differentials in cancer incidence with subsequent effects on changes in mortality disparities. With respect to survival, studies from Norway³² as well as from other countries^33,34 have shown socio-economic differentials, even after controls for cancer site and diagnostic severity. A possibility exists that improvements in cancer diagnostics and treatment, resulting in higher survival rates, tend to favour the more socio-economically advantaged because they are better positioned to acquire knowledge and gain access to improved medical opportunities.³⁵ However, several patterns observed in the present study do not easily fit into such an explanation. For instance, 5-year survival in Norway for prostate cancer improved from 52% in the early 1970s to 78% in the late 1990s,²² but the present study did not find any increasing educational disparities in prostate cancer mortality during these decades.

Mass screening for cancers could potentially influence the patterns of educational inequalities in cancer deaths. In Norway, nation-wide cervix cancer screening started in 1995 and breast cancer screening was gradually implemented 1995–2004, but no other mass cancer screening programme has been introduced.³⁶ Therefore, it is not likely that screening programmes have had any large impact on the developments of educational differentials in cancer deaths during the analysed period 1971–2002.

Strengths and limitations

This study is among the few analyses of trends since the 1970s in educational differences in specific cancer mortality, and it sheds light on some mortality patterns for which previous information is scant. The study has had access to an unusually large and unbiased material, analysing about 87 000 deaths and using register information, which covers practically every person in the selected age categories living in Norway during 1971–2002. Generally, data reliability is high and missing information is insignificant, but a possible source of error is the diagnostic information. To establish the main underlying cause of death is difficult when the disease history is complex. Sometimes, causes of death are ambiguously registered. Cervix uteri cancer deaths, for instance, poses a problem as cancers deaths classified as ‘uterus, unspecified’ are relatively frequent²¹ and an unknown proportion of them are probably cervix uteri. Moreover, low autopsy rates may lead to underreporting or misclassification of cancer deaths among persons who were not diagnosed when alive. In addition, diagnostic criteria will hardly remain constant across three decades.³⁷ To what extent such bias impacts on results is difficult to estimate.

Funding

The present study has been financed by the authors' employers Norwegian Social Research (NOVA), Oslo, Norway; Department of Sociology and Human Geography, University of Oslo, Norway; and Department of Economics, University of Oslo, Norway.

Conflicts of interest: None declared.

Acknowledgements

The topic of this study was addressed by R.T. in her master thesis, supervised by J.I.E. and T.H.L. and based on aggregate data prepared by Ø.K.

References