SEARCH

SEARCH BY CITATION

Keywords:

  • breast cancer;
  • relative survival;
  • comparative study;
  • Australia;
  • England

Abstract

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgements
  8. References

Survival from breast cancer in the UK is lower than in other countries in Western Europe, the USA and Australia. However, these international differences have not yet been examined in relation to tumor characteristics, treatment, screening history or other prognostic factors. We calculated relative survival by age, period of diagnosis, category of unemployment and extent of disease for women diagnosed with breast cancer during the period 1980–2002 in New South Wales (Australia) and West Midlands (England). National cancer registry data for each country for the period 1990–1994 were also examined. The excess hazard ratio was modeled as a function of prognostic covariables. Survival in Australia and New South Wales was higher than in England and West Midlands, respectively. In both regions, survival was lower for more deprived women and for the elderly. These differences were greater in West Midlands. Survival from localized and regional disease in New South Wales was higher than in West Midlands, but survival from metastatic disease was similar. Differences in breast cancer survival are unlikely to be entirely due to differences in data quality or to limitations of the analyses, although the measure of extent of disease used may not have been adequate to elucidate the effect of stage fully. One possible causal explanation is that the management of breast cancer differs between these regions. Further research should acquire better data on stage and investigate the effect of comorbidity and of patterns of care upon the difference in breast cancer survival between England and Australia. © 2008 Wiley-Liss, Inc.

International comparative studies of breast cancer patients have shown that there are substantial differences in the relative survival of women diagnosed in the UK and other equally developed nations in Europe.1–4 In the most recent European study, 5-year relative survival of women diagnosed in England during the period 1995–1999 was 78%, somewhat lower than Norway (83%), Finland (84%) and Sweden (85%).4 Higher 5-year survival from breast cancer has also been reported in Australia than in England.5 Here, women diagnosed in New South Wales, Australia displayed more than a 7% survival advantage over women diagnosed in Yorkshire, England.

These international differences have not yet been examined in relation to tumor characteristics, treatment, screening history or other covariates, which may influence the probability of death. In particular, social deprivation, a factor associated with increased risk of survival in England and Wales,6, 7 was found to be associated with better relative survival in an early analysis of breast cancer patients in New South Wales, Australia.8 The only other examination of this relationship conducted for an Australian population found poorer survival amongst women living in more deprived areas,9 as has commonly been observed for cancer patients in England and Wales.

Our objective was to quantify and investigate the reasons for international differences in breast cancer survival between Australia and England. Using regional cancer registry data, we have examined the temporal consistency of the international differences over a 23-year period. We have assessed survival in relation to social deprivation, extent of disease at diagnosis and tumor histology, to elucidate further the nature of these differences and their possible cause.

Material and methods

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgements
  8. References

Tumor data

All invasive breast cancers diagnosed in women aged 15–99 years in West Midlands or New South Wales during the period 1980–2002 were eligible for inclusion. Data were obtained directly from the West Midlands Cancer Intelligence Unit (WMCIU, England) and the New South Wales Central Cancer Registry (NSWCCR, Australia). These 2 registries cover populations of a similar size, 5.3 and 6.4 million, respectively.3, 4 Both register around 3,000 cases of invasive breast cancer in women per year, and over 1,000 deaths amongst women due to breast malignancy.5–7 The published survival rates for breast cancer from WM and NSW are similar to the published national rates for England and Australia, respectively.2, 8–10 Tumor records included information on date of diagnosis, date of death, age, extent of disease and tumor histology. Information was provided for 133,101 eligible women with a first invasive breast cancer: 68,396 in the West Midlands and 64,705 in New South Wales. All these women were followed up to the end of 2002.

We obtained data on all invasive breast cancers diagnosed amongst women aged 15–99 years in England or Australia during the period 1990–1994. These data were originally collected as part of the CONCORD study.10 Information on date of diagnosis, date of death and age at diagnosis was available for 182,337 eligible women with a first invasive breast cancer: 143,526 in England and 38,811 in Australia. All these women were followed up to the end of 1999.

Women were excluded if their first breast cancer occurred after a previous invasive malignancy at a different anatomic site, if the only registration of their cancer was via the cause of death information recorded on their death certificate (“DCO” cases for whom survival time is null) or if the sequence of dates provided was illogical. More than 96% of eligible women were analyzed. Missing days or months of diagnosis were imputed in a comparable fashion. The data on extent of disease provided by the West Midlands Cancer Intelligence Unit were recoded according to the rules used by the New South Wales Central Cancer Registry in consultation with the coding staff of both registries. This resulted in a comparable variable consisting of 4 categories: localized (confined to the organ of origin), regional (spread to adjacent muscle, organ, fat, connective tissue or regional lymph nodes), distant (distant metastasis) and unknown stage. Histology codes were grouped into 8 separate categories as follows: adenocarcinoma, ductal, lobular or other, carcinoma NOS, other carcinoma, other histology, non-specific histology and missing. Women with breast cancer in New South Wales were younger and diagnosed with less advanced disease than women in the West Midlands throughout 1980–2002 (Table I).

Table I. Distribution of Cases by Age at Diagnosis, Extent of Disease at Diagnosis and Histological Type by Region and Period of Diagnosis: Women Diagnosed with Breast Cancer in New South Wales, Australia and the West Midlands Region of England 1980–2002
 Period of diagnosis
1980–19871988–19951966–2002
New South WalesWest MidlandsNew South WalesWest MidlandsNew South WalesWest Midlands
N%N%N%N%N%N%
Age at diagnosis
 15–391,4009.11,2316.31,6797.61,3575.81,5846.31,3085.6
 40–492,86418.63,03715.54,53120.53,74016.14,74919.03,51715.0
 50–541,5169.81,7669.02,44011.02,59811.23,33713.43,12913.4
 55–591,78511.62,19111.22,29210.42,54811.03,14012.62,81012.0
 60–641,93212.52,40212.32,51311.42,97712.82,82811.32,60511.1
 65–691,76811.52,36912.12,66012.02,49510.72,56010.31,9848.5
 70–792,77418.04,23621.74,00418.14,62619.94,48318.04,64819.9
 80–991,3748.92,32011.91,9708.92,91812.52,2879.23,37014.4
Extent of disease at diagnosis
 Localised7,23046.98,64444.211,03950.011,56249.713,33253.411,33348.5
 Regional4,93632.07,31137.47,04231.96,98930.08,13532.67,77733.3
 Distant8885.81,6868.68553.91,1735.01,0984.48653.7
 Unstaged2,35915.31,9119.83,15314.33,53515.22,4039.63,39614.5
Histological group
 Adenocarcinoma: Ductal8,09152.57,82740.015,19468.812,84955.218,55474.314,07260.2
 Adenocarcinoma: Lobular6284.11,2116.21,8078.22,0979.02,82111.33,21113.7
 Adenocarcinoma:Other3,62123.51,3617.01,4496.61,1264.85112.07393.2
 Carcinoma NOS7394.83,29716.91,1495.23,47514.99803.92,50210.7
 Other carcinoma1,85312.03,24116.62,20310.02,1699.31,8577.42,1479.2
 Other histology610.4610.3610.3650.3850.3650.3
 Non-specific histology4192.72,54613.02211.01,4776.41520.66282.7
 Missing or invalid1<0.18<1.05<0.11<0.18<0.17<0.1

Deprivation data

In each region, patient records were linked to an ecologically defined deprivation category based upon the quintile of unemployment of their small area of residence. We used the unemployment rate because it was a temporally and internationally comparable measure of deprivation. We used the smallest geographic areas available for each census to maximize the accuracy of the ecological data. We used data from the 1991 and 2001 census in each country. We did not use data from the 1981 census because the very small-area data were not obtainable for Australia nor did we use data from either the 1986 or 1996 Australian census because there was no corresponding survey conducted in England. Unemployment rates from the 1991 census were calculated for each Collection District (CD, mean population size 544, s.d. 248) in New South Wales and each Enumeration District (ED, mean population size 469, s.d. 153) in the West Midlands. These were linked to women diagnosed 1980–1995. Unemployment rates from the 2001 census were calculated for each CD in New South Wales (mean population size 539, s.d. 254) and each Lower-Level Super-Output Area (LL-SOA, mean population size 1,513, s.d. 194) in West Midlands. These were linked to women diagnosed 1996–2002. Despite the larger size of LL-SOAs, we have shown previously that they are comparable to EDs when used to examine socio-economic differences in cancer survival in England.11 In New South Wales, linking women's individual tumor records to unemployment data necessitated geocoding the cancer registry data to a precise latitude and longitude to establish the exact location of each residential address. Less than 1% of records in the West Midlands and 1.5% of records in New South Wales failed to match due to missing address data. These records were excluded from all deprivation-specific analyses.

Survival analyses

In analyzing the survival of a group of cancer patients, it is preferable to describe the survival which is related directly to the disease rather than the observed (crude) survival of the patient group. This concept is known as net survival, the survival that would occur if mortality from other causes of death were removed. Relative survival is one method of estimating net survival. When estimating relative survival, the mortality that would be expected had the patients not had cancer (expected or background mortality) is derived from the mortality observed in the general population with the same age and sex as the cancer patients (for the same calendar period in the same region). This expected or background mortality is then removed from the mortality observed in the cancer patients to derive the relative survival rate.12 Relative survival is the most defensible method of estimating net survival in population-based studies, since it does not rely upon accurate reporting of cause of death.13

Period- and region-specific life tables

We required life tables for each region to estimate expected survival. National population counts in 1991 and 2001 by region, age and sex were obtained from the Office for National Statistics for England and Wales and the Australian Bureau of Statistics. Counts of deaths in Australia by state or territory, sex and 5-year age group occurring during 1990–1992 and 2000–2002 were obtained from the Australian Institute of Health and Welfare. The Office for National Statistics provided the equivalent counts of deaths for each Government Office for the Region in England for the period 2000–2002. The mean numbers of deaths per year for the periods 1990–1992 and 2000–2002 were combined with the population data for 1991 and 2001, respectively. Observed annual age-specific death rates were then calculated for each area in each year. The resulting abridged life tables were translated into smooth, complete (single year of age) mortality rates up to 100 years of age using a reducible four-parameter model life table system,14 constrained to 3 independent parameters.6 The Government Actuary's Department life tables 1991 and 2001 in each country were used as the reliable standards. No data on deaths were available for England for the period 1990–1991. A set of published regional life tables for England15 was used instead.

Descriptive analyses

We used the maximum-likelihood approach for individual records16 as implemented by the algorithm strel to estimate the excess hazard of death from breast cancer.17 Estimates of relative survival were then derived up to 10 years after diagnosis using the cohort or complete approach by age group, calendar period of diagnosis, and within each region by deprivation category and extent of disease. We also applied the more recently described period approach18 to predict relative survival in the near future using the survival experienced by women alive during 2002.

We did not have the full date of diagnosis for women in Australia (only month and year), and therefore, we could not distinguish between those who died on the same day as their diagnosis (cases with zero survival) and those who died in the same month as their diagnosis. Comparing short-term survival between England and Australia was therefore not possible. Comparable estimates of survival were obtained by estimating conditional survival for women who survived 12 months from their diagnosis using the same maximum-likelihood approach.16

Background mortality for women dying during the period 1980–1995 was represented by the 1990–1992 life tables whilst the 2000–2002 life tables were used for women dying during the period 1996–2002. Estimates of relative survival were age-standardized using the direct approach. We used the combined age structure of all women diagnosed with breast cancer in the West Midlands and New South Wales during the period 2000–2002 divided into 8 groups as the standard population.

The survival “gradient” across the 5 categories of deprivation was estimated using simple least-squares regression, weighted by the variance of each of the relative survival estimates.19 Both linear and quadratic models were fitted, and the Pearson chi-squared statistic used to assess which was more parsimonious. The difference between the fitted relative survival for the highest and lowest deprivation categories is described as the “deprivation gap”. Where survival is lower amongst women living in more deprived areas, the deprivation gap is negative.

Multivariable analyses

We modeled the excess hazard of death from breast cancer in West Midlands relative to that in New South Wales using a GLM approach with Poisson error structure for individual survival times.20A priori, models were fitted separately to each of 3 broad periods of diagnosis (1980–1987, 1988–1995 and 1996–2002) and for each of 3 segments of follow-up time (up to 1 year, 2nd–5th year and 6th–10th year). In all models, follow-up time was divided into 11 intervals as follows: the first and second 6 months, and then annual intervals up to 10 years after diagnosis.

The initial model included only follow-up and region to estimate the crude excess hazard ratio of breast cancer death between West Midlands and New South Wales. Age-adjusted models including each of the other 4 covariates (year of diagnosis, deprivation category, extent of disease and tumor histology) were then examined. For age and year of diagnosis, both linear and categorical forms of the model were assessed. Interactions between region and age, region and extent of disease and region and deprivation were also explored. A full multivariable model was then fitted. This model included all variables which found to be statistically significant in the univariable analyses.

Data were missing for 2 of the explanatory variables used in the analyses; extent of disease was unknown for 13% of women, whilst the histological classification was either “carcinoma NOS” or “non-specific histology” for 15% of women (Table I). A 10-fold multiple imputation by chained equations was used to generate these missing values. Multiple imputation is an unbiased method of imputing plausible values, using an imputation model, when data are missing at random (MAR).21–23 Even when data are not missing at random (NMAR), multiple imputation has been shown to perform well.24 Where imputed variables were used (and therefore 10 separate model likelihoods were produced) the Wald test replaced the likelihood ratio test to evaluate the significance of the additional parameters in the model. The p-value of the Wald test was derived using rules for multidimensional estimands.23

Results

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgements
  8. References

International and temporal trends

Although survival increased dramatically in both the West Midlands and New South Wales between 1980 and 2002 (Fig. 1) large, persistent and increasing differences in age-standardized survival were evident between the 2 regions throughout this period (Table II). At any one time, age-standardized relative survival for women in West Midlands was approximately equal to that observed in New South Wales 8 years earlier. Period analysis suggested that these international differentials may increase in the near future. Differences in survival were larger between England and Australia than between New South Wales and West Midlands and greater during 1992–1994 than 1990–1991 (Table III).

thumbnail image

Figure 1. Relative survival up to 23 years since diagnosis in New South Wales, Australia and the West Midlands region of England by period of diagnosis: women diagnosed with breast cancer 1980–2002.

Download figure to PowerPoint

Table II. Age-Standardized Relative Survival (95% CIs) in New South Wales, Australia and the West Midlands Region of England, and Absolute Differences in Survival (%) by Period of Diagnosis: Women Diagnosed With Breast Cancer 1989–2002
Period of diagnosisAge-standardized relative survival (95% Cl)Difference (%)
New South WalesWest Midlands
  1. Relative survival derived using strel and period- and regional-specific life tables, directly age-standardized using eight age groups. The combined population of breast cancer patient in West Midlands and New South Wales during the period 1999–2002 was used as the standard.

1-year survival
 1980–198392.6 (91.9; 93.3)86.8 (86.0; 87.5)5.8
 1984–198794.4 (93.8; 95.0)88.7 (88.0; 89.4)5.7
 1988–199195.2 (94.7; 95.8)91.2 (90.6; 91.8)4.1
 1992–199596.2 (95.8; 96.7)92.2 (91.6; 92.7)4.1
 1996–199996.6 (96.2; 97.0)93.2 (92.7; 93.6)3.5
 2000–200297.6 (97.1; 98.0)94.5 (94.0; 95.1)3.0
 2002: period analysis98.1 (97.4; 98.7)96.0 (95.2; 96.8)2.1
5-year survival
 1980–198370.8 (69.4; 72.2)60.8 (59.7; 62.0)10.0
 1984–198772.7 (71.4; 74.0)65.3 (64.2; 66.5)7.4
 1988–199177.9 (76.8; 79.0)70.8 (69.8; 71.9)7.1
 1992–199583.2 (82.3; 84.2)74.2 (73.2; 75.1)9.1
 1996–199986.8 (85.9; 87.7)77.6 (76.7; 78.5)9.2
 2002: period analysis91.3 (89.8; 92.8)79.6 (77.9; 81.4)11.6
10-year survival
 1980–198358.5 (56.7; 60.4)48.0 (46.5; 49.4)10.6
 1984–198761.4 (59.7; 63.1)53.4 (51.9; 54.9)8.0
 1988–199167.3 (65.8; 68.7)60.1 (58.8; 61.5)7.1
 1992–199576.8 (75.3; 78.2)64.3 (63.1; 65.6)12.4
 2002: period analysis86.6 (84.2; 89.1)70.6 (68.3; 72.9)16.0
Table III. Age-Standardized Relative Survival in New South Wales, in Australia, in the West Midlands and in England and Absolute Differences in Survival (95% CIs) by Period of Diagnosis Restricted to Women Who Were Diagnosed With Breast Cancer 1990–1994 and Who Survived at Least 12 Months from Diagnosis (Conditional Survival)
 Age-standardized relative survival (95% CI)Differences (%)
AustraliaEnglandNew South WalesWest MidlandsAUS-ENGNSW-WM
  1. Relative survival derived using strel and period- and regional-specific life tables, directly age-standardized using eight age groups. The combined population of breast cancer patients in West Midlands and New South Wales during the period 1999–2002 was used as the standard.

Conditional 5-year survival
 1990–199184.2 (83.3; 85.1)77.6 (77.1; 78.1)82.5 (81.1; 84.0)79.4 (77.9; 80.9)6.63.1
 1992–199486.9 (86.3; 87.5)79.4 (79.0; 79.7)85.4 (84.3; 86.5)79.8 (78.6; 80.9)7.55.7
Conditional 10-year survival
 1990–199174.9 (73.6; 76.2)65.1 (64.4; 65.8)71.5 (69.5; 73.5)68.3 (66.3; 70.4)9.83.2
 1992–199480.4 (79.0; 81.7)66.8 (66.1; 67.5)78.2 (76.5; 79.9)69.6 (68.0; 71.2)13.68.6

Age at diagnosis

Patterns in survival by age were very different in the 2 regions. Women aged over 80 years at diagnosis in West Midlands had poor survival and only small increases in survival over time (Fig. 2). By contrast, in New South Wales, survival was less variable by age and improvements in survival were observed fairly equally for all age groups. The largest improvements in survival were observed amongst the screened age groups: women aged 50–69 years in New South Wales and aged 50–64 years in West Midlands. For these women, screening was available from the late 1980s onwards.

thumbnail image

Figure 2. Age-specific 5-year relative survival in New South Wales, Australia and the West Midlands region of England by period of diagnosis: women diagnosed with breast cancer 1980–2002. Relative survival is plotted at the mid-point of each age group.

Download figure to PowerPoint

Deprivation

The “deprivation gap” in survival was persistently greater in West Midlands than in New South Wales for 1-, 5- (Fig. 3) and 10-year survival. International differences in survival were evident even when comparing the most affluent women in each country. In fact, the best survival in the West Midlands (amongst the most affluent) was usually lower than the worst survival rate in New South Wales (amongst the most deprived).

thumbnail image

Figure 3. 5-year relative survival in New South Wales, Australia and the West Midlands region of England by unemployment category: women diagnosed with breast cancer 1980–2002.

Download figure to PowerPoint

Extent of disease

There were large differences in survival amongst women with localized disease. Predicted 10-year survival of women diagnosed with localized disease was 89% in the West Midlands and 94% in New South Wales. By contrast, survival differences for women diagnosed with distant disease were small.

Multivariable analyses

In the first 12 months following diagnosis, the age-adjusted excess hazard of breast cancer death in West Midlands was around double that of New South Wales. After the first year of diagnosis, the excess mortality in West Midlands was 30 or 60% higher than in New South Wales. These differences increased over time (Table IV).

Table IV. Ratios (95% CIs) of the Excess Hazard of Death from Breast Cancer Between West Midlands and New South Wales (Reference) in the Baseline Model and in Models Adjusted for Age, Deprivation and Extent of Disease by Period of Diagnosis and Time Since Diagnosis: Women Diagnosed with Breast Cancer 1980–2002
 Up to 1 year after diagnosis
1980–19871988–19951996–2002
  1. Hazard ratios from the models which include extent of disease are derived from a ten-fold multiple imputation (see text).

Follow-up (11 intervals) and region2.06 (1.89;2.24)2.24 (2.03;2.47)2.36 (2.07;2.69)
+ age group1.89 (1.75;2.05)2.01 (1.84;2.19)2.09 (1.87;2.33)
+ age group and deprivation1.88 (1.73;2.04)2.01 (1.84;2.19)2.14 (1.91;2.39)
+ age group and extent of disease1.58 (1.46;1.71)1.75 (1.61;1.91)2.17 (1.95;2.41)
+ age group, deprivation and extent of disease1.57 (1.44;1.70)1.74 (1.60;1.90)2.18 (1.96;2.43)
 2nd–5th years after diagnosis
1980–19871988–19951996–2002
Follow-up (11 intervals) and region1.28 (1.22;1.35)1.33 (1.26;1.40)1.59 (1.46;1.72)
+ age group1.28 (1.22;1.34)1.34 (1.27;1.41)1.58 (1.46;1.71)
+ age group and deprivation1.27 (1.21;1.34)1.34 (1.28;1.41)1.60 (1.48;1.73)
+ age group and extent of disease1.21 (1.15;1.27)1.38 (1.31;1.45)1.60 (1.48;1.73)
+ age group, deprivation and extent of disease1.20 (1.15;1.26)1.38 (1.32;1.46)1.60 (1.48;1.73)
 6th–10th years after diagnosis
1980–19871988–1995 
Follow-up (11 intervals) and region1.31 (1.21;1.41)1.29 (1.18;1.40) 
+ age group1.28 (1.19;1.38)1.31 (1.20;1.42) 
+ age group and deprivation1.28 (1.19;1.37)1.31 (1.21;1.43) 
+ age group and extent of disease1.26 (1.17;1.35)1.39 (1.28;1.51) 
+ age group, deprivation and extent of disease1.26 (1.17;1.35)1.39 (1.28;1.51) 

Part of the difference in survival between the 2 regions in the first year of diagnosis during 1980–1987 and 1988–1995 was attributable to more advanced extent of disease amongst women diagnosed in the West Midlands. However, this was not the case during the period 1996–2002. Differences in survival following the first anniversary of diagnosis could not be attributed to differences in extent of disease at diagnosis. None of the differences in survival examined were explained by deprivation category or by histology (data not shown).

Significant interactions were observed between region and deprivation, region and age, and region and extent of disease. These were incorporated into the full multivariable models. Figure 4 displays the excess hazard ratios derived from these 3 models, 1 for each calendar period. In each case, the excess hazard ratio (vertical scale) represents the increased hazard of excess death in the West Midlands compared with New South Wales for the particular group (horizontal scale) during the first 12 months after diagnosis. So, the left-hand upper-most point on Figure 4a indicates that in 1980–1987 the excess hazard of death during the first year after diagnosis amongst affluent women was almost 5 times higher in West Midlands than in New South Wales. This estimate is adjusted for age, extent of disease, histology and year of diagnosis.

thumbnail image

Figure 4. Excess hazard ratios of death during the first 12 months in the West Midlands of England relative to New South Wales, Australia (reference) for (a) categories of deprivation (b) age group and (c) extent of disease at diagnosis. Hazard ratios are derived from 3 separate multivariable models, 1 for each period of diagnosis. Each model includes follow-up time (11 intervals), and interaction terms between (a) unemployment category and region (b) age group and region and (c) extent of disease and region. All hazard ratios are adjusted for year of diagnosis (using a linear form), extent of disease and tumor histology (using multiple imputation for records with missing data: see text).

Download figure to PowerPoint

Adjusted differences in survival during the first year were of a broadly similar magnitude for each of the 5 deprivation categories within each of the periods of diagnosis (Fig. 4a). This pattern was also observed for the second to fifth and sixth to tenth years of follow-up, although the differences themselves were of a smaller magnitude (excess hazard ratios of less than 2.0, data not shown). There was a much greater adjusted survival difference amongst older compared with younger women during the period 1996–2002 (Fig. 4b). Here, after adjustment for all other variables, women 80–99 years of age diagnosed in the West Midlands experienced an estimated excess mortality rate more than 11 times that of the same aged women in New South Wales. The same pattern was also observed in the second to fifth year after diagnosis (data not shown). Substantial differences in survival were observed for both localized and regional disease in the first year of diagnosis (Fig. 4c). By contrast, the survival amongst women with distant disease was similar.

Discussion

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgements
  8. References

We have shown that throughout 1980–2002, women diagnosed with breast cancer in New South Wales experienced a large, increasing and significant survival advantage over women diagnosed in West Midlands. Similarly, large differences in survival were observed between England and Australia, suggesting that the differences between the 2 regions we examined in detail are also present at a national level.

Differences in survival between West Midlands and New South Wales were most marked amongst older women and amongst women diagnosed with localized disease, but were of a similar magnitude across the socio-economic spectrum. The patterns could not be accounted for by differences in the extent of disease at diagnosis, or by variations in histological type, insofar as the measured variables used in this study represent the true values of these parameters for the women analyzed.

These international survival differences are consistent with the only other comparative study of breast cancer survival between an English and an Australian population.5 Poorer survival amongst older women is also consistent with several detailed analyses of EUROCARE data, which show that in countries with lower overall survival the differences by age are more marked.25–28 Our observation that extent of disease does not help explain differences in survival between Australia and England contrasts with 2 previous international analyses within Europe and between Europe and the USA.29–35 These studies have concluded that a large proportion of the differences in survival between the UK and other developed nations can be attributed to differences in stage at diagnosis.

Data quality

The patterns observed could result from differences in data quality between Australia and England.

It is possible that the registry in New South Wales selectively under-counts cancer cases with poorer prognosis or fails to register the deaths of more short-term survivors than West Midlands. Either of these biases would artificially inflate survival in New South Wales. We calculated the additional number of deaths needed in New South Wales to equalize 1-year survival by multiplying the excess hazard ratios (Fig. 4) by the number of excess deaths for each age group. This showed that during the period 1980–2002, around 4,500 additional breast cancer deaths would be needed in New South Wales to nullify the differences in 1-year survival, the majority being for women aged 70–99 years. If we assume that these 4,500 deaths occurred solely amongst the cases registered, this would imply that only one-third of the deaths within the first year due to breast cancer amongst registered cases are actually known to the New South Wales Central Cancer Registry and only one-quarter amongst the elderly. We consider this implausible, particularly given the relatively low proportion of death-certificate-only (DCO) cases in New South Wales (an accepted marker of completeness of registration). The alternative interpretation that there were 4,500 women with breast cancer who were never registered, and who also died during the first year after their diagnosis, also seems unlikely. The incidence of breast cancer was indeed higher in West Midlands than in New South Wales during this period, particularly amongst the elderly (Woods et al., submitted). If we assume that the incidence in New South Wales is artefactually low, 2,000 additional cases of breast cancer would have been needed in New South Wales during the period 1980–2002 to equalize the incidence rates. Since this represents less than half of the 4,500 cancer deaths needed to equalize survival for the first year only, under-registration of cancer cases cannot explain the survival differences.

It is also possible that the New South Wales Central Cancer Registry is less successful at matching deaths with cancer registrations because their target population includes a large proportion of recent migrants, who possibly return to their country of birth following a diagnosis of cancer. This phenomenon also seems unlikely to explain the survival differences, however, since the largest differences are observed in the first 12 months after diagnosis when women are receiving first-line treatment for their cancer, a time at which they are unlikely to emigrate. Further, survival differences are greatest amongst the elderly. Emigration from Australia, in contrast, is largely confined to those under 35 years.36

It could be that a higher proportion of the registered tumors are in situ in New South Wales than in West Midlands, but are erroneously registered as localized invasive cancers. To examine this hypothesis, we randomly deleted women with localized cancer in New South Wales, who did not die during the observed follow-up, to establish the proportion of cases that would need to have been in situ to equalize the relative 5-year survival. For all 3 periods of diagnosis, similar 5-year relative survival was reached only after the removal of more than half of these women with localized cancers. It cannot be that so great a proportion of the cancers in New South Wales are in situ, particularly given that the proportion microscopically verified (i.e., confirmed as invasive tumors) during 1980–2002 was more than 80%.

Estimation of survival is sensitive to both the accuracy and the comparability of the rules used to establish the date of diagnosis. This is unlikely to have influenced our results, however, because both New South Wales and West Midlands apply the same rules to determine when a diagnosis took place where more than 1 date is available. Further, the excess hazard of breast cancer death was persistently lower in New South Wales for the entirety of the first 2 years following diagnosis, rather than simply “shifted” to the left (data not shown). If differences in survival were due simply to a difference in the way date of diagnosis had been determined, such a shift in the excess hazard would be expected.

We chose to use a temporally and internationally comparable measure: the unemployment rate to categorize women into socio-economic groups. This is a relatively crude measure of socio-economic status. However, we observed very similar survival patterns for women diagnosed from 1996 to 2002 when using quintiles of 2 highly-validated and locally defined deprivation measures; for West Midlands, the IMD income domain score 2004,37 and for New South Wales, the index of disadvantage of the Socio-Economic Indexes For Areas (SEIFA) 2001.38

In applying a life table derived from deaths in 1990–1992 to the entire period 1980–1995, it was assumed that background mortality from the early 1990s was a good estimate of the true background mortality for each of these 16 years. Although region-specific life tables for 1981, 1991 and 2001 were available for the West Midlands, it was not possible obtain either life tables or deaths data for the years 1980–1982 for New South Wales. It was desirable to use the same approach in each region in order to ensure comparability of the results. Consequently, it was assumed that regional life tables for 1991 provided a good estimate of age-specific mortality in 1980 and more particularly for women older than 50, who represent the majority of breast cancer patients. In fact, the increase in life expectancy of a 50-year old woman was greater in Australia than in England between the periods 1980–1982 and 1990–1992 (1.64 years39 compared with 1.34 years40). This implies that the differences in survival that we present for the earliest years are slight overestimates of the true differences because our estimate of expected survival is likely to be relatively higher for New South Wales in comparison to the West Midlands. The only alternative approach, using national rather than regional life tables, would have biased our results to a greater degree.

Possible causal explanations

Delay in diagnosis

The interpretation that delays in diagnosis are responsible for the international survival differences is not strongly supported by these data because extent of disease at diagnosis does not display much explanatory power. Adjustment for this variable would tend to reduce the estimated excess hazard ratio of death between the 2 countries if the international differences in survival were due to greater delays in diagnosis, longer waiting times for hospital consultation or less effective screening in West Midlands than in New South Wales. Residual confounding by extent of disease may partially account for this. This implies that the accuracy of the extent of disease variable might be lower in West Midlands, so that within each stage grouping, the true (unknown) stage of disease is more advanced in West Midlands than in New South Wales (stage migration).41 This would lead to better extent-adjusted survival in New South Wales since better staging practice tends to lead to more accurate allocation of women into the higher staging categories.

It is also possible that the imputation model used to accommodate missing data on extent of disease was not accurate enough to account for the missing observations. In particular, the proportion of tumors which were unstaged was greatest amongst women aged over 80 years at diagnosis in West Midlands during the years 1988–2002. If these women's disease was more often regional or distant than estimated by the imputation model the impact of extent of disease on the international difference in survival would have been underestimated.

To measure the impact of disease stage upon these differences in survival more accurately, detailed information would need to be re-extracted at source for both populations, using a comparable protocol and similarly trained personnel, as has been done in the EUROCARE high-resolution studies.29 This would enable us to better understand whether the overall differences in survival that we observe here are, in fact, partially due to women in West Midlands being diagnosed later than women in New South Wales.

Cancer management

A second possible explanation for lower survival is different cancer management in West Midlands than in New South Wales. Receipt of treatment by women with breast cancer is influenced by the availability of different treatments, of the decisions made by healthcare professionals, of policies set by the healthcare provider(s) or of differential compliance with treatment by the women themselves.

Survival differences were observed for both localized and regional disease, whereas the survival of women with distant metastases was more similar. Although it is possible that we have overestimated the magnitude of this difference for localized disease, such extent-specific differences still suggest that the overall outcome of treatment with curative intent may be less good in West Midlands than in New South Wales, since women diagnosed with distant disease receive only palliative care. In cancer populations where treatment has been equalized, differences in cause-specific survival between social or racial groups are small or nonexistent.42–46 However, whether patterns of treatment can explain the international survival differences for breast cancer has yet to be examined.

Treatment differences might also be the underlying reason for the age-specific differences and, in turn, the international differences. Differential treatment for breast cancer by age has been observed.47 Prevalent comorbidity is likely to play an important part in accelerating death in the first year after diagnosis because of the contraindication of effective treatment or more severe complications of treatment amongst patients with comorbid conditions. This is one of the explanations that has been given for the fact that older women receive less surgery compared with younger women in West Midlands. Women who have not had surgery are almost always women who have comorbidity (mostly heart conditions) which contraindicate treatment (Personal communication, West Midlands Breast Screening Quality Assurance Reference Centre). This could explain some of the observed international differences in survival, if the management of breast cancer in the presence of comorbidity were substantially different in West Midlands than in New South Wales. It is also possible that treatment may be rationed for elderly patients or that compliance amongst the elderly is particularly low in West Midlands. It is known that during the late 1980s and the early 1990s, there was a tendency to treat elderly women with tamoxifen alone, without surgery.48 If this was occurring to a greater extent in West Midlands, it could partly explain the age-specific differences observed here. However, the largest differences amongst elderly women were observed during the period 1996–2002, suggesting that treatment with tamoxifen only cannot entirely account for these differences. Access to medical services may be more limited for elderly women than younger women, due either to lack of social support or to physical disability, which could also explain a part of this association.

Since no comparative data on treatment were examined in these analyses, in particular on adjuvant endocrine therapy, the treatment modality which has had the greatest impact on breast cancer survival amongst postmenopausal women in the last 30 years,49 it is difficult to draw any firm conclusions regarding the possible role of treatment differentials in these results. Comorbidity has not, so far, been examined in relation to international differences in survival. Although it has been considered as a possible explanation for differences in survival between socio-economic groups,50–52 it has received relatively little attention to date. The investigation of patterns of care in relation to preexisting comorbidity should be a priority for future research.

Conclusions

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgements
  8. References

This study has shown large, persistent differences in breast cancer survival between West Midlands and New South Wales. These differences are most unlikely to be due to differences in data quality or to limitations of the analyses. It is possible that greater delay in diagnosis in the West Midlands explains some of the observed difference in survival, but that the measure of extent of disease used in this study was inadequate to elucidate the impact of disease stage fully. A further possible causal explanation may be that treatment for breast cancer is less effective for women in West Midlands than those in New South Wales, particularly amongst the elderly. Differences in the effectiveness of treatment could in turn be due to differences in patient compliance or in the way patients are treated in the presence of comorbidity, as well as to differences in treatment protocols. Further research should reassess the importance of stage of disease at diagnosis by deriving more detailed information for each woman. Studies should also seek to examine the importance of treatment differences and comorbidity in explaining international differences in breast cancer survival between England and Australia.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgements
  8. References

This work was supported by a Medical Research Council UK (MRC) studentship, 2003–2006.

References

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgements
  8. References
  • 1
    Berrino F,Sant M,Verdecchia A,Capocaccia R,Hakulinen T,Estève J. Survival of cancer patients in Europe: the EUROCARE study. Lyon: International Agency for Research on Cancer, 1995.
  • 2
    Berrino F,Capocaccia R,Estève J,Gatta G,Hakulinen T,Micheli M,Sant M,Verdecchia A. Survival of cancer patients in Europe: the EUROCARE-2 study. Lyon: International Agency for Research on Cancer, 1999.
  • 3
    Berrino F,Capocaccia R,Coleman MP,Estève J,Gatta G,Hakulinen T,Micheli M,Sant M,Verdecchia A. Survival of cancer patients in Europe: the EUROCARE-3 study. Ann Oncol 2003; 14 ( Suppl 5): v1v155.
  • 4
    Berrino F,De AR,Sant M,Rosso S,Bielska-Lasota M,Coebergh JW,Santaquilani M. Survival for eight major cancers and all cancers combined for European adults diagnosed in 1995–99: results of the EUROCARE-4 study. Lancet Oncol 2007; 8: 77383.
  • 5
    Yu XQ,O'Connell DL,Forman D. Comparison of cancer survival in UK and Australia: rates are higher in Australia for three major sites. Br J Cancer 2004; 91: 16635.
  • 6
    Coleman MP,Babb P,Damiecki P,Grosclaude P,Honjo S,Jones J,Knerer G,Pitard A,Quinn MJ,Sloggett A,De Stavola BL. Cancer survival trends in England and Wales 1971–1995: deprivation and NHS region. Series SMPS No. 61. London: The Stationery Office, 1999.
  • 7
    Coleman MP,Rachet B,Woods LM,Mitry E,Riga M,Cooper N,Quinn MJ,Brenner H,Estève J. Trends and socioeconomic inequalities in cancer survival in England and Wales up to 2001. Br J Cancer 2004; 90: 136773.
  • 8
    Supramaniam R,Smith D,Coates M,Hayes L,Armstrong BK.Breast cancer survival in NSW in 1973 to 1995. Sydney: Cancer Epidemiology Research Unit and NSW Central Cancer Registry Cancer Control Information Centre, NSW Cancer Council, 1998.
  • 9
    Australian Institute of Health and Welfare (AIHW), Australasian Association of Cancer Registries (AACR). Cancer survival in Australia 1992–1997: geographic categories and socioeconomic status. Report No.: 22. Canberra: Australian Institute of Health and Welfare, 2003.
  • 10
    CONCORD Steering Committee. Cancer survival in Europe and North America: the CONCORD study protocol. London School of Hygiene & Tropical Medicine. Accessed on January 10, 2001. Available at: http://www.lshtm.ac.uk/ncdeu/cancersurvival/concord/index.htm.
  • 11
    Woods LM,Rachet B,Coleman MP. Choice of geographic unit influences socioeconomic inequalities in breast cancer survival. Br J Cancer 2005; 92: 127982.
  • 12
    Berkson J,Gage RP. Calculation of survival rates for cancer. Proc Staff Meet Mayo Clin 1950; 25: 27086.
  • 13
    Ederer F,Axtell LM,Cutler SJ. The relative survival: a statistical methodology. Natl Cancer Inst Monogr 1961; 6: 10121.
  • 14
    Ewbank DC,Gomez de Leon JC,Stoto MA. A reducible four-parameter system of model life tables. Popul Stud 1983; 37: 10527.
  • 15
    Coleman MP,Babb P,Mayer D,Quinn MJ,Sloggett A. Cancer survival trends in England and Wales 1971–1995: deprivation and NHS region (CDROM). London: Office for National Statistics, 1999.
  • 16
    Estève J,Benhamou E,Croasdale M,Raymond L. Relative survival and the estimation of net survival: elements for further discussion. Stat Med 1990; 9: 52938.
  • 17
    CRUK Cancer Survival Group. strel computer program version 5.8 and life tables for cancer survival analysis. Non-communicable Disease Epidemiology Unit, London School of Hygiene & Tropical Medicine, UK. Accessed on November 19, 2006. Available at: www.lshtm.ac.uk/ncdeu/cancersurvival/tools.htm.
  • 18
    Brenner H,Gefeller O. Deriving more up-to-date estimates of long-term patient survival. J Clin Epidemiol 1997; 50: 2116.
  • 19
    Grizzle JE,Starmer CF,Koch GG. Analysis of categorical data by linear models. Biometrics 1969; 25: 489504.
  • 20
    Dickman PW,Sloggett A,Hills M,Hakulinen T. Regression models for relative survival. Stat Med 2004; 23: 5164.
  • 21
    Rubin DB. Multiple imputation for non-response in surveys. New York: Wiley, 1987.
  • 22
    Little RL,Rubin DB. Statistical analysis with missing data. New York: Wiley, 1987.
  • 23
    Schafer JL. Analysis of incomplete multivariate data. New York: Chapman & Hall/CRC, 1997.
  • 24
    Collins LM,Schafer JL,Kam CM. A comparison of inclusive and restrictive strategies in modern missing data procedures. Psychol Methods 2001; 6: 33051.
  • 25
    Quinn MJ,Martinez-Garcia C,Berrino F,EUROCARE Working Group. Variations in survival from breast cancer in Europe by age and country, 1978–1989. Eur J Cancer 1998; 34: 220411.
  • 26
    Sant M,Capocaccia R,Verdecchia A,Estève J,Gatta G,Micheli A,Coleman MP,Berrino F. Survival of women with breast cancer in Europe: variation with age, year of diagnosis and country.The EUROCARE Working Group. Int J Cancer 1998; 77: 67983.
  • 27
    Vercelli M,Capocaccia R,Quaglia A,Casella C,Puppo A,Coebergh JW,The EUROCARE Working Group. Relative survival in elderly European cancer patients: evidence for health care inequalities. Crit Rev Oncol Hematol 2000; 35: 16179.
  • 28
    Vercelli M,Quaglia A,Casella C,Parodi S,Capocaccia R,Martinez-Garcia C,The EUROCARE Working Group. Relative survival in elderly cancer patients in Europe. Eur J Cancer 1998; 34: 226470.
  • 29
    Sant M,Allemani C,Capocaccia R,Hakulinen T,Aareleid T,Coebergh JW,Coleman MP,Grosclaude P,Martinez C,Bell J,Youngson J,Berrino F. Stage at diagnosis is a key explanation of differences in breast cancer survival across Europe. Int J Cancer 2003; 106: 41622.
  • 30
    Christensen LH,Engholm G,Ceberg J,Hein S,Perfekt R,Tange UB,Andersson M,Mouridsen HT,Moller T,Storm HH. Can the survival difference between breast cancer patients in Denmark and Sweden 1989 and 1994 be explained by patho-anatomical variables?—a population-based study. Eur J Cancer 2004; 40: 123343.
  • 31
    Jensen AR,Garne JP,Storm HH,Ewertz M,Cold S,Alvegaard T,Overgaard J. Stage and survival in breast cancer patients in screened and non-screened Danish and Swedish populations. Acta Oncol 2003; 42: 7019.
  • 32
    Karjalainen S,Aareleid T,Hakulinen T,Pukkala E,Rahu M,Tekkel M. Survival of female breast cancer patients in Finland and in Estonia: stage at diagnosis an important determinant of the difference between countries. Soc Sci Med 1989; 28: 2338.
  • 33
    Kliukiene J,Andersen A. Survival of breast cancer patients in Lithuania and Norway, 1988–1992. Eur J Cancer 1998; 34: 3727.
  • 34
    Aareleid T,Staneczek W,Rahu M,Mohner M. On the background of differences in survival of female breast cancer patients in the German Democratic Republic and the Estonian SSR. Neoplasma 1990; 37: 58794.
  • 35
    Sant M,Allemani C,Berrino F,Coleman MP,Aareleid T,Chaplain G,Coebergh JW,Colonna M,Crosignani P,Danzon A,Federico M,Gafa L, et al. Breast carcinoma survival in Europe and the United States: a population based study. Cancer 2004; 100: 71522.
  • 36
    Australian Bureau of Statistics. Migration, Australia 2004–2005. Report No.: Cat. no. 3412.0. Canberra: 2006.
  • 37
    Neighbourhood Renewal Unit—Office for the Deputy Prime Minister. The English indices of deprivation 2004 (revised). London: Neighbourhood Renewal Unit, Office for the Deputy Prime Minister, 2004.
  • 38
    Australian Bureau of Statistics. Socio-economic indexes for areas: Australia. Technical report. Report No.: 2039.0.55.001. Canberra: 2004.
  • 39
    Australian Bureau of Statistics. Table 49: expectation of life at single ages (0–100), females, Australia, 1881 onwards. Report No.: 3105.0. 65.001. 2004.
  • 40
    Interim life tables. Government Actuary's Department 1991. Accessed on July 8, 2008. Available at: http://www.gad.gov.uk.
  • 41
    Feinstein AR,Sosin DM,Wells CK. The Will Rogers phenomenon: stage migration and new diagnostic techniques as a source of misleading statistics for survival in cancer. N Engl J Med 1985; 312: 16048.
  • 42
    Greenwald HP,Polissar NL,Borgatta EF,McCorkle R,Goodman G. Social factors, treatment, and survival in early-stage non-small cell lung cancer. Am J Public Health 1998; 88: 16814.
  • 43
    Rabeneck L,Souchek J,El Serag HB. Survival of colorectal cancer patients hospitalized in the Veterans Affairs Health Care System. Am J Gastroenterol 2003; 98: 118692.
    Direct Link:
  • 44
    Farley JH,Hines JF,Taylor RR,Carlson JW,Parker MF,Kost ER,Rogers SJ,Harrison TA,Macri CI,Parham GP. Equal care ensures equal survival for African-American women with cervical carcinoma. Cancer 2001; 91: 86973.
  • 45
    McCollum AD,Catalano PJ,Haller DG,Mayer RJ,Macdonald JS,Benson AB,III,Fuchs CS. Outcomes and toxicity in African-American and Caucasian patients in a randomized adjuvant chemotherapy trial for colon cancer. J Natl Cancer Inst 2002; 94: 11607.
  • 46
    Brawley OW,Freeman HP. Race and outcomes: is this the end of the beginning for minority health research? J Natl Cancer Inst 1999; 91: 19089.
  • 47
    Lavelle K,Todd C,Moran A,Howell A,Bundred N,Campbell M. Non-standard management of breast cancer increases with age in the UK: a population based cohort of women > or =65 years. Br J Cancer 2007; 96: 1197203.
  • 48
    Ramesh HAJ,Pope D,Gennari R,Audisio RA. Optimising surgical management of elderly cancer patients. World J Surg Oncol 2005; 3: 17.
  • 49
    Early Breast Cancer Trialists' Collaborative Group. Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet 2005; 365: 1687717.
  • 50
    Prout GR,Wesley MN,McCarron PG,Chen VW,Greenberg RS,Mayberry RM,Edwards BK. Survival experience of black patients and white patients with bladder carcinoma. Cancer 2004; 100: 62130.
  • 51
    Polednak AP. Poverty, comorbidity, and survival of colorectal cancer patients diagnosed in Connecticut. J Health Care Poor Underserved 2001; 12: 30210.
  • 52
    Munro AJ,Bentley AHM. Deprivation, comorbidity and survival in a cohort of patients with colorectal cancer. Eur J Cancer Care 2004; 13: 25462.