Background Reducing disparities in cancer outcomes is a major priority for cancer-control agencies. The authors examine the relationships between geographic remoteness, area disadvantage and risk of advanced breast cancer among women.
Methods Multilevel models were used to assess the area- and individual-level contributions to the risk of advanced breast cancer among women aged 30–79 years diagnosed as having breast cancer in Queensland, Australia between 1997 and 2006 (n=18 658).
Results Women who resided in the most socio-economically disadvantaged areas were significantly more likely (OR 1.21, 95% CI 1.07 to 1.37) than residents of the most advantaged areas to be diagnosed as having advanced breast cancer after adjustment for individual-level factors. When geographic remoteness and area-disadvantage (and all the individual-level factors) were simultaneously adjusted, the rates of advanced breast cancer were significantly higher for women residing in Outer Regional areas (OR 1.13, 95% CI 1.02 to 1.24) and those who lived in the most disadvantaged areas (OR 1.16, 95% CI 1.02 to 1.32). There was no statistically significant interaction between geographic remoteness and area disadvantage.
Conclusions A woman's risk of being diagnosed as having advanced breast cancer depends on where she lives, separate from the individual characteristics of the woman herself. Both the rurality and socio-economic characteristics of the geographical area in which women lived were important. The socio-economic factors contributing to advanced breast cancer, existing in both urban and rural environments, need to be investigated.
- Breast cancer
- multilevel modelling
- inequalities SI
- epidemiology ME
Statistics from Altmetric.com
- Breast cancer
- multilevel modelling
- inequalities SI
- epidemiology ME
Breast cancer is the most common invasive cancer among Australian women, accounting for about 28% (or 12 614) of all cancers diagnosed among women in 2006 and is the second leading cause of cancer mortality (2680 deaths, or 16% of cancer deaths).1
Owing to the strong relationship between breast cancer stage and survival,2 detecting breast cancer early in its disease pathway increases the opportunity for effective treatment to reduce the morbidity and mortality associated with this disease, and to improve long-term prognosis.3 A better understanding of the population subgroups at greatest risk of being diagnosed as having breast cancer at an advanced stage would inform the development of interventions to address these inequalities.
Since the early diagnosis of breast cancer requires access to appropriate primary care and screening services, higher rates of advanced breast cancer might be expected among women living in rural areas, who have poorer access to services. However, the evidence for an urban–rural gradient in breast cancer outcomes is mixed, with some studies in Australia4 5 and the USA6 showing no difference. Other studies have shown an urban–rural gradient, but the direction of this gradient is not consistent; some report a higher risk of advanced breast cancer in more rural areas, including from New South Wales (Australia),3 USA7 and Denmark,8 while for others such as Illinois (USA)9 and Massachusetts (USA),10 the risk of advanced breast cancer was highest in urban areas.
Complicating the urban–rural comparison is the correlation between geographical location and socio-economic status. Most studies have consistently demonstrated that socio-economically disadvantaged populations are more likely to be diagnosed as having late-stage breast cancer,11–16 but this effect is not universal.17 18 When combined with the reported urban–rural gradients, there is the potential for very high rates of late-stage breast cancer in rural minority women.12 However, this interaction between rurality and socio-economic status is different for different countries. In England, for example, rural environments are generally synonymous with a healthier and wealthier lifestyle,19 while in Queensland, Australia, the opposite is the case, with 39% of people living in remote areas also live in areas classed as the lowest socio-economic quintile, compared with 8% of people living in major cities.20
Previous Australian studies examining area variation in breast-cancer outcomes have used aggregate ecological design21 22 or have not specifically accounted for the multilevel relationship between area- and individual-level factors.3 5 While sometimes ecological studies are the only option owing to the availability of data, their limitations have been well documented.23 They cannot quantify the variation in breast-cancer outcomes between geographical areas, and then, more importantly, indicate whether the variation is due to the clustering of individuals (ie, a composition effect) or the environmental characteristics of the areas per se (ie, a context effect). Even when studies report significant differences in rates of advanced breast cancer by socio-economic disadvantage or geographic remoteness, this does not mean that areas per se are important in terms of influencing the probability of the area's residents experiencing advanced breast cancer. Ecological studies leave open the possibility that geographic variations in advanced breast cancer are an artefact of varying population compositions, and unless these are taken into account (which ecological studies cannot do), individual- and area-level sources of variation remain confounded. It is therefore an open question as to whether areas are important (independent) determinants of advanced breast cancer in Australia.
To date, no Australian study has employed multilevel analytical methods to investigate links between geographic remoteness, area disadvantage, individual-level factors and advanced breast cancer. A multilevel perspective, however, is increasingly being advocated24 as a way of advancing understanding about area-level inequalities in health and, by extension, improving attempts to reduce the inequalities.25 26 Queensland is the most decentralised state in Australia, and as such represents an ideal opportunity to investigate these issues in detail. In this present paper, we use multilevel analytical methods to examine the relationships between geographic remoteness, area disadvantage and risk of advanced breast cancer among women aged 30–79 years who were living in Queensland during 1997–2006. Specifically, we investigate four inter-related questions:
Do areas in Queensland differ in their average rates (probabilities) of advanced breast cancer before and after adjustment for within-area variation in individual-level factors (ie, year of diagnosis, age, indigenous status, occupation and marital status) and between-area variation in area-disadvantage and geographic remoteness?
What is the association between individual-level factors and the probability of experiencing advanced breast cancer?
What is the association between area-level socio-economic disadvantage and advanced breast cancer, and geographic remoteness and advanced breast cancer (independent of individual-level factors)?
Do area-disadvantage and geographic remoteness interact to modify the probability of experiencing advanced breast cancer?
Materials and methods
Ethical approval to conduct this study was obtained from the University of Queensland Social and Behavioural Sciences Ethical Review Committee.
With appropriate approvals, data were extracted from the population-based Queensland Cancer Registry (QCR). All women diagnosed as having invasive breast cancer (ICD-0: C50) in Queensland between 1 January 1997 and 31 December 2006 (inclusive) and aged 30–79 at diagnosis (n=19 544) were eligible for the study. As with all invasive cancers, public and private hospitals, pathology services and residential care facilities are required by law to notify all confirmed cases of breast cancer to the QCR. In situ breast cancer was not considered.
The groupings for stage of cancer were based on the International Union Against Cancer classification of breast cancer stage, as outlined in the Australian Clinical Practice Guidelines for early breast cancer.27 Tumour size, nodal involvement and presence of metastases (the TNM system) are used to classify the breast-cancer stage. While the QCR does not collect this detailed information, information has been recorded on tumour diameter and lymph-node involvement since 1997.
Localised breast cancer was defined as a tumour of not more than 20 mm diameter, with no evidence of lymph-node involvement or distant metastases. The only component of the Stage I breast-cancer definition which could not be confirmed in the QCR was the absence of metastasis (M0). Although it would be unlikely that these cases had metastasised, this could not be definitively ruled out.
Since it was not possible to distinguish between Stages II, III or IV with the available information, these were collectively categorised as ‘Advanced’ breast cancers in one grouping. Cancers which were diagnosed as a result of metastatic disease were included in this category. Similar categorisation has been used previously,28 but other studies have used a narrower definition of advanced breast cancer, such as the presence of nodal or metastatic spread.11
Tumours of unknown size or with unknown lymph-node involvement when the tumour size was 20 mm or less were excluded from the study cohort.
Age at diagnosis was collapsed into 5-year age groups from 30–34 to 75–79 years.
We converted occupation at time of diagnosis to the Australian Standard Classification of Occupations Second Edition. The occupations were then collapsed into ‘Blue Collar’ (including tradespersons, plant and machine operators and drivers, and labourers and related workers), ‘White collar’ (including clerks, salespersons and personal service workers), ‘Professional’ (including managers and administrators, professionals and para-professional), ‘Not in workforce’ (including retired, students, unemployed and home duties) and ‘Unknown’ (no information available). Similar categories have been used previously.29
Women's marital status was recorded as single, married, widowed, divorced, separated or not stated.
The QCR obtains information about indigenous status of patients with cancer through the process of cancer notification from Queensland hospitals. Since this is based on self-assessment, not all indigenous cases may have been identified, and there is potential misclassification of true indigenous status. However, this is thought to be small.30
Period of diagnosis
The period of diagnosis was collapsed into three time periods; 1997–2000, 2001–2003 and 2004–2006.
Statistical Local Areas (SLAs) were used as the area-level geographical definition for this analysis. SLAs are deemed to be relatively homogeneous in terms of the socio-economic characteristics of the populations they contain. SLAs are often based on the incorporated bodies of local governments and councils, and these are responsible for service and infrastructure provision at the local and regional level; hence SLAs are spatial entities that are likely to be socially and economically relevant to their residents.
In 2006, there were 478 SLAs in Queensland with a median population of 5810. Owing to boundary changes in the SLA definitions over time, we used concordance files provided by the Australian Bureau of Statistics to allocate all location information from 1997 onwards to the 2006 SLA definitions.
Remoteness of residence when diagnosed as having breast cancer was categorised using the ARIA+ classification,31 which is a purely geographical measure of remoteness, categorising it into Major City, Inner Regional, Outer Regional, Remote and Very Remote areas based on road distance from a locality to the closest service centre in each of five classes of population size.
Area-level socio-economic disadvantage
This was measured using the Index of Relative Socio-economic Disadvantage (IRSD) calculated by the Australian Bureau of Statistics.32 The IRSD provides a general measure of disadvantage and considers factors such as the percentage of residents in each SLA with low income, low educational attainment, high unemployment and jobs in relatively unskilled occupations.
Multilevel logistic modelling was used to assess whether geographic remoteness and area-level disadvantage were associated with advanced breast cancer after controlling for individual-level socio-demographic characteristics. The data were analysed using MLwiN version 2.15 (University of Bristol, UK).
The analyses were conducted in three stages. First, we specified a null model that comprised individuals (level 1) nested in SLAs (level 2) with no individual- or area-level variables in the fixed part of the model. Substantive interest for the null model focuses on the SLA-level random term, which, if significant (indicated using Wald χ2),33 34 suggests between-SLA variation in advanced breast cancer. Second, the null model was subsequently extended to include individual-level fixed-effects for year of diagnosis, age (5 year categories), indigenous status, occupation and marital status (Model 2), and then geographic remoteness (Model 3), neighbourhood disadvantage (Model 4), and then all variables simultaneously (Model 5). Third, interactions between area-disadvantage, geographic remoteness and advanced breast cancer were assessed by including second-order terms that reflected different combinations of the categories of disadvantage and remoteness (Model 6). The substantive focus of the interaction analysis was on whether the association between area-disadvantage and advanced breast cancer was different at varying levels of geographic remoteness, and whether the relationship between remoteness and advanced breast cancer differed depending on the extent of area disadvantage. The fit of the interaction model was tested against its main-effect counterpart (Model 5) using the Deviance test (−2 log likelihood); we also used the Wald χ2 test to assess the statistical significance of each of the interaction coefficients. The fixed-effect results for all models are presented as ORs, and their 95% CIs and joint χ2 tests were performed to evaluate the contribution of each variable to model fit.
Table 1 presents the number of women aged 30–79 years in Queensland who experienced breast cancer between 1997 and 2006, and the proportion who were diagnosed as having advanced breast cancer. Rates of advanced breast cancer were higher for residents of Outer Regional and Remote/Very Remote SLAs and those living in socio-economically disadvantaged areas. The diagnosis of advanced breast cancer was highest for the period 2004 to 2006, and rates were highest for women aged 30–44 years and women who were indigenous, employed in white- and blue collar occupations, and who never married.
Table 2 presents the multilevel association between geographic remoteness, area disadvantage, individual-level socio-demographic factors and the likelihood of experiencing advanced breast cancer. Results for the null model (Model 1) show that the probability of experiencing advanced breast cancer differed significantly across the SLAs (p=0.002). The differences were successively reduced with adjustment for the individual-level factors (by 30.4% for model 2 relative to the null model), geographic remoteness (39.1%), area-disadvantage (47.8%) and all variables (60.9%). In the latter (fully adjusted) model (Model 5), there was no longer any significant between-SLA variation in the probability of experiencing advanced breast cancer (p=0.111).
Independent of year of diagnosis and age, indigenous status, occupation and marital status (Model 3), women who lived in Outer Regional (OR 1.18, 95% CI 1.07 to 1.30) and Remote/Very Remote (OR 1.19, 95% CI 1.00 to 1.41) areas had significantly higher rates of advanced breast cancer compared with their counterparts living in Major Cities. Women who resided in the most socio-economically disadvantaged areas (Model 4) were significantly more likely (OR 1.21, 95% CI 1.07 to 1.37) than residents of the most advantaged areas to be diagnosed as having advanced breast cancer after adjustment for each of the individual-level factors. When geographic remoteness and area-disadvantage (and all of the individual-level factors) were simultaneously adjusted (Model 5), the effects of geographic remoteness (χ2=10.648, df=3, p=0.0137) and area-disadvantage (χ2=12.155, df=4, p=0.0162) were both significant. Rates of advanced breast cancer were significantly higher for women residing in Outer Regional areas (OR 1.13, 95% CI 1.02 to 1.24) compared with women living in Major cities, and higher for women who lived in the most disadvantaged areas (OR 1.16, 95% CI 1.02 to 1.32) compared with the least disadvantaged areas. There was no longer a significant difference in the rate of advanced breast cancer between women in Major Cities and Remote/Very remote areas after adjusting for area-disadvantage.
Tests of interaction between area disadvantage and geographic remoteness on the risk of advanced breast cancer were not statistically significant based on the Deviance test (p=0.294); nor did any of the interaction terms reach significance at p≤0.10 using the Wald χ2 test (results not shown).
Between 1997 and 2006, the probability of experiencing advanced breast cancer was highest for the period 2004 to 2006 (all models) compared with the period 1997 to 2000. The rate of advanced breast cancer across all areas in Queensland (irrespective of their degree of remoteness or extent of disadvantage) was significantly higher for women aged 30–44 relative to those aged 50–54; indigenous women compared with their non-indigenous counterparts; white- and blue-collar workers compared with professionals; and the never married and widowed compared with married women.
Reducing disparities in cancer outcomes is a major priority for cancer-control agencies.35 36 Using a large, population-based cohort of women diagnosed as having breast cancer in Queensland, this study has demonstrated that the characteristics of where a woman lives, specifically, rurality and area-level socio-economic disadvantage, are important contributor to the likelihood of her being diagnosed as having advanced disease, even after adjusting for individual-level variables.
A recent Australian study found strong evidence that detection by a screening mammogram was the main health-system factor that reduced the odds of women being diagnosed as having a large (≥2 cm) breast cancer.5 This, when combined with the observed association between early detection and travel distance to mammography screening services,37 could suggest that rural women have reduced access to mammography screening services. However, in Queensland, data from BreastScreen Queensland, the sole provider of population-based mammography screening in Queensland, show that participation rates in 2006–2007 were higher among women aged 50–69 years in rural (60%) and remote areas (57%) than among women in urban areas (54%), so access to population-based screening does not appear to explain the apparent increased risk of advanced breast cancer in outer regional areas.38
Women who lived in areas of relative socio-economic disadvantage were more likely to be diagnosed as having later-stage breast cancer than women living in more affluent areas. This area-level socio-economic effect has also been reported in international studies,11 13 28 39 although specific reasons for this effect have not been proposed. It is unlikely to be due simply to lower rates of breast screening, since Australian women living in areas of high SES are least likely to participate in population-based screening programmes.40 However, the extent to which these women access screening in private practice is not known. That the effect is found after adjusting for the rurality gradient suggests that there are disadvantaged populations within urban areas with a higher likelihood of being diagnosed as having larger breast cancers, so it is also unlikely that access issues related to distance can account for all the socio-economic effect. We also found no evidence of any interaction between area-level disadvantage and rurality. While numbers of women across the socio-economic categories in remote areas were small, limiting our power to detect any interaction, this null finding suggests that the increased risk of advanced breast cancer associated with socio-economic disadvantage is the same for women living in major cities as for women in rural and remote areas. It is possible that there are other unmeasured variables that differ strongly between low and high SES areas that also impact on the diagnosis of breast cancer, and these need to be identified.39
We found that the likelihood of being diagnosed as having advanced breast cancer was highest for younger women, and then generally decreased with increasing age up to age 70–74 years, consistent with that reported elsewhere.28 37 There is biological evidence to suggest that breast cancer among younger women (under the age of 35–40 years at diagnosis) is a distinct disease,41 with breast cancers in these younger women tending to be larger, less well differentiated and more likely to metastasise compared with breast cancers diagnosed in older women.41–43 In addition, since routine screening is generally not recommended for women under the age of 40 because of a lower incidence and mammography being less effective owing to their denser breast tissue,42 43 the majority of younger women with breast cancer present to a doctor with symptoms indicative of a more advanced stage.42
Given the relatively small numbers of indigenous women diagnosed as having breast cancer (1% of the total cohort), the fact that the indigenous effect remained after adjusting for rurality and measures of individual-level SES and area-level socio-economic disadvantage highlights both the strength of this association and the complexity of the reasons underlying it. A recent study in the Northern Territory (Australia)44 also found that of women diagnosed as having breast cancer, indigenous women were significantly more likely to be diagnosed as having advanced disease compared with non-indigenous women, suggesting this was due to later diagnosis.
Reasons for more advanced disease among indigenous women are varied and could include a reduced awareness of early symptoms of breast cancer, delay in seeking medical advice, poorer access and quality of screening and diagnostic services, and nihilistic beliefs about cancer and the chance of its cure causing reluctance to seek medical attention44 (personal communication, BreastScreen Queensland).
With mortalities for breast cancer decreasing in Queensland, as is the case internationally, the increasing risk of advanced cancer over time is somewhat surprising. While other studies have reported decreases in breast-cancer tumour size,14 a recent study in New South Wales3 reported that the percentage of breast cancers with distant spread was highest in the early 1980s, decreased up to the late 1990s, but then started to increase again up to 2003. This is consistent with our Queensland patterns, and could be related to the converging trends in incidence rates, with incidence rates of localised breast cancer decreasing sharply since 2001 (−2.6% per year) while rates of advanced breast cancers have remained relatively stable (+0.4% per year). We also found no evidence that the percentage of cases with unknown stage changed over time (p=0.391, data not shown). While studies have shown that female patients with breast cancer who used screening were less likely to be diagnosed at a late stage,5 45 this is unlikely to explain these trends, and the participation rate in population-based mammography screening has remained relatively stable over this time period in Queensland.38 Other possible explanations could be related to increased sensitivity in detecting positive nodes or distant metastases through changes in clinical priorities with greater attention to regional lymph-node investigation, advances in imaging technologies or a reduction in mammography sensitivity through increased use of hormone-replacement therapy.3
The strengths of this study include the large, unselected, population-based cohort of all female patients with breast cancer diagnosed in Queensland between 1997 and 2006, including information about the spread of the disease at diagnosis. All information used in this study has been collected prospectively for administrative purposes independently of the study hypotheses, thus removing recall or information bias. The use of administrative data is not without problems, however, since the individual-level socio-economic measure was limited to occupation; the QCR does not collect information about income,46 body mass index,5 private insurance status47 or other personal or clinical factors known to be associated with increased risk of advanced disease such as menopausal status8 or previous mammography screening history.5 Moreover, the occupation measure used in this study lacked both sensitivity and specificity, as it was not possible to disaggregate the ‘Not stated’ category into more homogenous groupings such as ‘home duties,’ ‘retired’ or ‘unemployed.’ Further, our use of a single individual-level measure of SES meant that we could not account for other relevant sources of socio-economic variation (eg, education, income) within each area. Hence, while our multilevel study extends and improves on previous ecological designs, there remains the possibility that the area-level geographic and socio-economic differences in advanced breast cancer reported here were due to the confounding effects of unmeasured socio-economic factors.
While the focus of this study has been on the area-level inequalities, we have specifically used a multilevel approach to adjust appropriately for the individual-level variables, including measures of socio-economic status such as occupation. These data are not readily available in other population-based cancer registries.39
A women's neighbourhood of residence and her personal social and economic characteristics separately influence the probability of experiencing advanced breast cancer; hence policies and interventions to reduce cancer inequalities should be targeted at places as well as people.
What is already known on this subject
Detecting breast cancer early in its disease pathway increases the opportunity for effective treatment to reduce the morbidity and mortality associated with this disease, and to improve long-term prognosis. Ecological studies in Australia have suggested that the risk of advanced breast cancer varies across population and geographical subgroups. A multilevel perspective is increasingly being advocated as a way of advancing understanding about area-level inequalities in health and, by extension, improving attempts to reduce the inequalities.
What this study adds
This is the first Australian study to demonstrate that a woman's risk of being diagnosed as having advanced breast cancer depends on where she lives, and on the individual characteristics of the woman herself. Both the rurality and socio-economic characteristics of the geographical area in which women lived were important. While distance and access issues associated with rurality are well recognised, although not easily addressed, the socio-economic factors contributing to advanced breast cancer are less tangible, existing in both urban and rural environments, and need to be investigated.
Funding This study was partially supported by a grant from the (Australian) National Health and Medical Research Council (NHMRC) (ID: 561700). GT is supported by a NHMRC Senior Research Fellowship (ID 390109).
Competing interests None.
Ethics approval Ethics approval was provided by the University of Queensland Social and Behavioral Sciences Ethical Review Committee.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.