• Open Access

Socioeconomic factors and cancer incidence, mortality, and survival in a metropolitan area of Japan: A cross-sectional ecological study

Authors

  • Kimiko Ueda,

    Corresponding author
      To whom correspondence should be addressed. E-mail: kueda@xb4.so-net.ne.jp
    Search for more papers by this author
  • Hideaki Tsukuma,

    1. Department of Cancer Control and Statistics, Osaka Medical Center for Cancer and Cardiovascular Diseases, 3-3 Nakamichi 1-chome, Higashinari-ku, Osaka 537-8511, Japan
    Search for more papers by this author
  • Wakiko Ajiki,

    1. Department of Cancer Control and Statistics, Osaka Medical Center for Cancer and Cardiovascular Diseases, 3-3 Nakamichi 1-chome, Higashinari-ku, Osaka 537-8511, Japan
    Search for more papers by this author
  • Akira Oshima

    1. Department of Cancer Control and Statistics, Osaka Medical Center for Cancer and Cardiovascular Diseases, 3-3 Nakamichi 1-chome, Higashinari-ku, Osaka 537-8511, Japan
    Search for more papers by this author

To whom correspondence should be addressed. E-mail: kueda@xb4.so-net.ne.jp

Abstract

Cancer mortality is generally high in people of low socioeconomic status compared with people of high socioeconomic status (SES). Although these differences in mortality may be caused by differences in cancer incidence and survival, analysis of these factors has rarely been conducted. The objective of our cross-sectional ecological study was to analyze socioeconomic differences in cancer incidence, mortality and survival in a metropolitan area of Japan. The age-adjusted cancer incidence rates, age-adjusted mortality rates, relative 5-year survival, and proportions of early stage cancer were calculated for 67 municipalities in Osaka, Japan. For area-based socioeconomic variables, we used the percentages of male unemployment, college or graduate school graduates, home ownership, households receiving government assistance, and households below the subsistence habitation level in each municipality. We performed linear regression taking each municipality's population as weight to examine the relationships between measurements relating cancer and socioeconomic variables. Factor analysis of socioeconomic variables was carried out to determine whether a particular socioeconomic variable tended to be associated with another. Cancer incidence, cancer mortality, 5-year cancer survival, and proportion of early stage cancer were highly correlated with each socioeconomic variable at the municipality level. Five area-based socioeconomic variables could be explained by three factors: economic status, housing characteristics and educational attainment. Despite the major limitation of a lack of individual information about socioeconomic characteristics and outcomes related to cancer, we hypothesize that a municipal area's socioeconomic status might be a predictor of individual incidence, mortality, and survival of cancer. (Cancer Sci 2005; 96: 684 – 688)

Abbreviations:
DCO

death certificate only registrations

OCR

Osaka Cancer Registry

SES

socioeconomic status

SSDS

System of Social and Demographic Statistics.

Cancer is the major cause of death in both developed and developing countries. More than 10 million people are diagnosed with cancer every year.(1) It is estimated that there will be 15 million new cases every year by 2020.(1) Cancer causes 6 million deaths worldwide each year, or 12% of all deaths.(1) Numerous studies about the occurrence and causes of cancer have been conducted at the molecular, lifestyle, environment, and socioeconomic factors levels. Just as a variety of health effects occur in various organ systems at the level of the individual, a variety of individual exposures may have common socioeconomic causes at the population level.(2)

Socioeconomic differences in mortality have been studied for a variety of causes of death.(3–12) Cancer mortality is generally high in people of low socioeconomic status compared with people of a high socioeconomic status.(13) These differences may be caused by differences in cancer incidence and survival.(14) However, analysis of socioeconomic differences in incidence, mortality and survival at the same time has been conducted only in a limited fashion.

Additionally, socioeconomic differences in health, including cancer mortality in Japan, have rarely been identified. For example, in light of income structure, Shibuya et al. indicated that individual income levels may have more influence on self-rated health than regional income inequality in Japan, and suggested that the disparity between their findings and other studies in other countries can be explained by differences in social and political characteristics specific to place and cultural norms.(15) More studies about socioeconomic status and health in Japan are required to confirm these hypotheses.(16)

We have conducted an ecological study to analyze the socioeconomic differences in cancer incidence, mortality and survival in one metropolitan area of Japan.

Materials and Methods

Measurement of municipality variations related to cancer

The age-adjusted cancer incidence rates, age-adjusted mortality rates, relative 5-year survival, and proportions of early stage cancer were calculated for 67 municipalities in Osaka, Japan. We treated International Classification of Diseases (ICD) (10th revision), codes C00–97 as cancer. Cancer incidence rates from January 1995 through to December 1999 were directly age standardized to the Japanese model population of Japan in 1985 with data obtained from vital statistics compiled by the Ministry of Health, Labour and Welfare and expressed as the number of cancer incidents per 100 000 persons. Cancer mortality rates from January 1995 through to December 1999 were also directly age standardized to the Japanese model of Japan in 1985 and expressed as the number of deaths per 100 000 persons. The data were obtained from the Osaka Cancer Registry (OCR), one of the oldest and largest population-based cancer registries in Japan. Procedures and the validity of the OCR have been described elsewhere.(17) The relative 5-year survival after diagnosis of cancer was calculated among subjects, adjusting for differences in the probability of dying from causes other than cancer, as the ratio of observed survival to expected survival. The expected survival was estimated by the Ederer II(18) method using the survival probability in the general population of Japan with respect to sex, age, and calendar year at diagnosis. Subjects of relative 5-year survival were cases diagnosed with primary cancer from January 1990 through to December 1994 and registered in the OCR, excluding carcinoma in situ and mucosal cancer in the colorectum. Death certificate only (DCO) registrations were excluded. The proportion of early stage cancer was the proportion of cases with a localized cancer stage at diagnosis among newly registered cases from January 1995 through to December 1999.

Measurement of municipality variations in socioeconomic status

The area-based socioeconomic status (SES) measurements we used were occupational, educational, and economic: the percentages of male unemployment, college or graduate school graduates, home ownership, households receiving government assistance, and households below the subsistence habitation level in each municipality. All area-based SES measurements were obtained from the System of Social and Demographic Statistics (SSDS) provided by the Ministry of Internal Affairs and Communications. The percentages of unemployment in 1995 and college or graduate school graduates in 1990 were originally from the Japan census; percentages of home ownership and subsistence habitation households were from the 1998 Housing and Land Survey, and the percentage of households receiving government assistance was from 1998 prefecture reports.

Data analysis

We performed linear regression, weighting according to each municipality's population, to examine the relationships between measurements relating cancer and area-based SES measurements. The population size of each municipality was from the 1995 Japan census. Factor analysis with promax rotation of socioeconomic variables was also carried out in order to examine whether particular socioeconomic variables tended to be associated with each other. To confirm the relationships between scores for each factor obtained by factor analysis and measurements relating cancer, we conducted weighted regression again. STATA (version 7.0) was used for statistical analyses (StataCorp, College Station, TX).

Results

Each age-adjusted cancer mortality rate among the male population only, the female population only, and the total population was highly correlated with each socioeconomic variable. A strong inverse relationship was found between the percentages of home ownership and college or graduate school graduates, and age-adjusted cancer mortality (Table 1). As Figure 1 shows, a strong relationship was found between the age-adjusted cancer mortality rate among the total population and unemployment (r = 0.87, P < 0.0001). Each 1% increment in unemployment was associated with an increase in cancer mortality of 6.0 deaths per 100 000 people (95% CI = 5.1–6.8). Each age-adjusted cancer incidence rate among the male population only, the female population only, and the total population was also positively related to the percentages of unemployment, households receiving government assistance and subsistence habitation households, and inversely related to home ownership and the percentage of college or graduate school graduates, even though the relationship between the age-adjusted cancer incidence rate among the female population only and the percentage of college or graduate school graduates was not significant (r = –0.12, P = 0.315; Table 1).

Table 1. Correlations between socioeconomic variables and cancer incidence, mortality, proportion of early stage cancer and relative 5-year survival
 % Unemployment% Households receiving government assistance% Subsistence habitation households% Home ownership% College or graduate school graduates
  • *

    P < 0.01,

  • P < 0.05.

Age-adjusted mortality rate
 All 0.87* 0.71* 0.77*−0.62*−0.58*
 Male 0.75* 0.54* 0.69*−0.60*−0.59*
 Female 0.77* 0.56* 0.72*−0.66*−0.44*
Age-adjusted incidence rate
 All 0.58* 0.42* 0.61*−0.62*−0.25
 Male 0.47* 0.28 0.54*−0.59*−0.28
 Female 0.51* 0.34 0.55*−0.60*−0.12
Proportion of early stage cancer
 All−0.56*−0.49*−0.48* 0.34 0.45*
 Male−0.61*−0.53*−0.58* 0.45* 0.41*
 Female−0.24*−0.18−0.11 0.03 0.31*
Relative 5-year survival
 All−0.77*−0.53*−0.75* 0.69* 0.53*
 Male−0.79*−0.56*−0.75* 0.68* 0.58*
 Female−0.68*−0.45*−0.70* 0.66* 0.45*
Figure 1.

Relationship between percentage male unemployment and age-adjusted cancer mortality ratex (both sexes).

Each socioeconomic variable is related to relative 5-year cancer survival in an opposite way from the age-adjusted cancer mortality rate (Table 1). A strong positive relationship was found between relative 5-year cancer survival among the total population and home ownership (r = 0.69, P < 0.0001; Fig. 2). Each 10% rise in home ownership was associated with an increase in 5.0 points of cancer survival (95% CI = 3.5–6.4).

Figure 2.

Relationship between percentage home ownership and relative 5-year cancer survival (both sexes).

As with relative 5-year cancer survival, proportions of early stage cancer among the total population and the male population only were correlated with socioeconomic variables. The proportion of early stage cancer in the female population only was not associated with the percentages of households receiving government assistance, subsistence habitation households and home ownership, but was associated with percentages of unemployment and college or graduate school graduates (Table 1).

Factor analysis with promax rotation showed that variables loading on the first factor included percentages of unemployment and households receiving government assistance (Table 2). The direction of the coefficients suggested that this factor captured the effect of less poor households in municipalities with stable occupational status. Variables loading on the second factor included percentages of subsistence habitation households and home ownership (Table 2), which captured the effect of housing characteristics in municipalities. There were fewer households below the subsistence habitation level in municipalities that had households owning their own homes. The variable loading on the third factor was percentage of college or graduate school graduates (Table 2), which captured the effect of educational structure in municipalities. As same as each socioeconomic measurement, reasonably strong relationships were observed between scores for each factor obtained by factor analysis and measurements relating cancer (Table 3).

Table 2. Factor analysis with promax rotation of socioeconomic variables
VariableFactor 1Factor 2Factor 3
% Unemployment 0.64776−0.29412−0.19057
% Households receiving government assistance 0.90387 0.0114 0.0677
% Subsistence habitation households 0.25586−0.71950−0.05673
% Home ownership−0.00241 0.85489−0.08435
% College or graduate school graduates−0.29100−0.05334 0.48012
Explained variance 1.38673 1.33800 0.28175
Table 3. Correlations between scores for each factor obtained by factor analysis and cancer incidence, mortality, proportion of early stage cancer and relative 5-year survival
 Factor 1Factor 2Factor 3
  1. Economic status for factor 1, housing characteristics for factor 2 and educational attainment for factor 3. P < 0.01. P < 0.05.

Age-adjusted mortality rate
All 0.81−0.57−0.44
Male 0.68−0.53−0.45
Female 0.67−0.55−0.33
Age-adjusted incidence rate
All 0.51−0.49−0.16
Male 0.40−0.47−0.18
Female 0.42−0.45−0.06
Proportion of early stage cancer
All−0.52 0.29 0.40
Male−0.57 0.42 0.31
Female−0.21−0.02 0.35
Relative 5-year survival
All−0.70 0.59 0.39
Male−0.71 0.59 0.44
Female−0.62 0.59 0.31

Discussion

The objective of this study was to analyze socioeconomic differences in cancer incidence, mortality and survival in a metropolitan area of Japan. The results suggest that the age-adjusted cancer mortality rate, as well as the age-adjusted cancer incidence rate, the relative 5-year cancer survival, and the proportion of early stage cancer were highly correlated with each socioeconomic measurement at the municipality level. These results are supported by several studies.(13,19) Because our data refer to areas rather than individuals, such ecological data should be interpreted cautiously. Ecological relationships should be taken only as pointers to possible individual level relationships.(8)

Osaka prefecture has the second largest population density in Japan,(20) and tends to feel the effects of changes in the economy due to the large number of medium and small companies there. The populations of 67 municipalities in our study ranged from approximately 7000 to 800 000.(20) Compared with the US and Europe, Japan is said to be socially, economically and racially uniform.(21) Our findings, however, indicated that there are at least some socioeconomic differences at smaller aggregate levels that result in differences in cancer incidence, mortality and survival.

Much research shows that behavioral risk-factors alone can explain only a small proportion of health outcomes,(11) especially in adulthood when socioeconomic status seems to play an important independent causal role.(22) Socioeconomic status is determined by those social and economic factors that influence the status that any given groups and individuals hold within the structure of their society.(23)

We used five broad area-based SES measurements from the occupational, educational, and economic structures available from SSDS. These area-based SES measurements were used not only as proxies for individual-level data but also to assess contextual socioeconomic effects on individuals.(23,24) For example, the percentage of unemployment within a municipality not only denotes something about the individuals who live there, but it may also present other information about the municipality that may have conditioned the health risks of all inhabitants there, not just the unemployed people.

Theoretically, these SES measurements have different meanings and, at the same time, correlate. Educational structure is usually measured at one key point in the life course, the transition from childhood and adolescence into adulthood and the world of work, but it also provides information about the likelihood of future success.(23) Occupational structure is the major structural link between education and income (economics).(23,25) For instance, educational experiences are important in determining what sorts of employment are available to a person, and this employment then determines the amount of economic return.(23) Economic structure relates directly to the material conditions that influence health. Our study indicated that five different SES measurements correlated with each other (P < 0.01), except for relationships between percentage of college or graduate school graduates and percentage of home ownership. At the same time, these variables could be explained entirely in terms of a much smaller number of variables, or three factors: economic status, housing characteristics, and educational attainment, as shown in Table 2.

There might be variations in the relationships between incidence, mortality and survival, and socioeconomic factors among different primary sites as well as different age groups according to characteristics of the cancer. Further studies will be required considering the prognostic factors of each cancer site such as age, cancer stage, treatment and histology.

Bias may have occurred if socioeconomic status associated with residence at the time of diagnosis or death differed from the status at exposure to cancer risk. To minimize the bias, we used 1995 or 1998 socioeconomic variables. These were more likely to accurately characterize the socioeconomic characteristics of municipalities during 1995–1999. The percentage of college or graduate school graduates used was from the 1990 census because the data could be obtained only every 10 years.

Our study indicated that the age-adjusted mortality rate had a strong correlation with socioeconomic variables at the municipality level (Table 1, Fig. 1). In general, socioeconomic differences in cancer mortality are mainly caused by differences in incidence and survival.(14) Socioeconomic differences in cancer survival are caused by factors such as cancer stage at diagnosis, access or barriers to medical care, consistency of treatment, and so on.(26) Our study showed differences in age-adjusted incidence rate, relative 5-year survival, and proportion of early stage cancer according to socioeconomic differences.

A major limitation of our study was the lack of individual information about socioeconomic characteristics, as well as outcomes related to cancer. It was not possible to control adequately for confounders, effect modifiers and mediators at the individual level, the ‘Ecological Fallacy’.(27) Variables could be measured only at the municipality level; that is, the unit of our analyses based on an administrative definition was much more likely than a larger administrative area, such as the prefecture level, to represent a neighborhood effect, because it is the closest administrative unit to the citizen. To prevail over this limitation, multilevel analysis operating simultaneously at the level of individuals and at the level of contexts will be required.(28)

Another limitation was that cross-sectional analyses could not point out the causality. We cannot know whether lower socioeconomic status caused higher mortality or if, conversely, higher cancer mortality caused lower socioeconomic status. Indisputably, there was a reciprocal relationship between socioeconomic status and cancer mortality. Future research will need to clarify which SES measurements, including a full range of variables, are best suited to detect variations by municipality in cancer incidence, mortality and survival.

Despite these limitations, our findings suggested that economic, as well as educational and occupational, status were strongly correlated with cancer mortality rates by municipality in a metropolitan area of Japan. The myth that ‘all people in Japan belong to the middle class’ has collapsed,(29) at least at the municipal level in Osaka. Additionally, and what may be worse in terms of public health, is that existing socioeconomic differences clearly have some relationship with variations in cancer mortality, incidence and survival. These results regarding the relative weight of the material and cultural dimensions of the socioeconomic structure, as well as the biological risk factors of cancer, need to be investigated with individual data. However, on the basis of our analyses, we hypothesize that area-based economic, educational, and occupational status are predictors of individual incidence, mortality, and survival of cancer. In Japan, individual socioeconomic data for larger populations cannot be obtained easily.(16) But our study suggests that socioeconomic data at the ecological level, which are available from the government, may have effective policy implications for planning cancer control. That is to say, socioeconomic differences in cancer incidence among municipalities might suggest some interventions in terms of primary prevention. Additionally, differences in cancer mortality and survival might suggest interventions for secondary prevention or treatment.

Acknowledgments

This work was supported in part by a Grant-in-Aid for Cancer Research (14-2) from the Ministry of Health, Labour and Welfare of Japan. We would like to thank all who work in the Osaka Cancer Registry.

Ancillary