Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Breast cancer five–year survival, by New South Wales regions, 1980 to 1991

Breast cancer five–year survival, by New South Wales regions, 1980 to 1991 Abstract: Breast cancer five-year relative survival was calculated for 16 urban and rural regions in New South Wales (NSW) for cases incident in 1980-1991. Survival analysis employed cancer registry data linked with the death register, and age- and periodmatched regional mortality of NSW women. Proportional hazard regression analysis was used to compare excess mortality in breast cancer cases in each region. The effect of region was significant ( P < 0.05) in the analysis, after age and the follow-up variable (and their interaction) were adjusted for, although no region was significantly different from the referent group (chosen because of average relative five-year survival). When degree of spread and its interactions were entered into the model, the effect of region became nonsignificant. A significant linear trend ( P < 0.05) in the adjusted relative risk for excess mortality in breast cancer cases was noted when regions were divided into quartiles based on socioeconomic status, with higher relative risk in low-socioeconomic-status groups; this effect also disappeared with adjustment for degree of spread at diagnosis. There was no general effect of rurality versus capital city or other metropolitan centres. This study demonstrates a small effect of region of residence and implied socioeconomic status on breast cancer survival in NSW women, but this becomes nonsignificant when the data are adjusted for degree of spread at diagnosis. This suggests that earlier diagnosis would be of benefit in reducing minor inequalities in breast cancer survival in NSW women. (Aust N ZJPublic Health 1997; 21: 206-10) URVIVAL. is an important outcome measure in cancer control. It is influenced by health promotion and health service factors (including screening) that affect the stage at which the cancer is diagnosed, and by access to and quality of medical services, which determine promptness and efficacy of treatment. The two main methods of measuring cancer survival in population or clinical cohort studies are: relative survival (which compares the survival of cases with the age-, period- and sex-matched population from which they arise, and which does not require cause-of-death information) ;I,* and cancerspecific survival (which involves censoring of noncancer deaths in the survival analysis, and requires precise cause-ofdeath data) .3 Similar results, particularly for five-year survival, are obtained from both meth~ds.~ Differential survival by area of residence is useful for surveillance of outcomes of health services. Populations that have particularly poor outcomes require more detailed investigation to ascertain then rectify problems; areas with particularly good outcomes can profitably be studied to define characteristics that could be adopted usefully in other areas. Survival data can be used to establish benchmarks for achievable outcomes, and for target-setting within the planning process. This article examines differential five-year relative survival for breast cancer cases in New South Wales (NSW) by 16 areas of residence for 1980-1991. Data for NSW as a whole have been published in this issue (pp. 199-205) .4 Since different population subgroups may have different levels of underlying mortality and this may affect the relative survival analysis, this report uses regionspecific life tables in the calculations. Methods Cancer data Correspondence to Associate Professor Richard Taylor, Department of Public Health and Community Medicine, LJniversity of Sydney, NSW 2006. Fax (02) 9351 4179. The NSW Central Cancer Registry, a populationbased registry, began data collection in January 1972. Notification of malignant neoplasms has been a statutory requirement for all NSW public and private hospitals, radiotherapy departments and nursing homes since 1972, and for pathology and outpatients departments since 1985. For the period 1972 to 1991, the date of diagnosis was defined as the date of first definitive treatment for cancer. Information concerning completeness of enumeration3,6and histological confirmation of breast cance? has been published.* Data on degree of spread of tumour was recorded from notifiers as: 1. localised to tissue of origin (localised) ; 2. invasion of adjacent organs or regional lymph nodes (regional); and 3. distant metastasis (metastatic). Degree of spread at diagnosis was 82 per cent complete for 1978-1982, and improved to 88 per cent complete for 198&1991.5 Cancer cases were geocoded by a computer program by postcode and locality name (and street NO AUSTRALIAN AND NEW ZEALAND JOURNAL O PUBLIC HEALTH 1997 vot. 21 F BREAST CANCER SURVIVAL BY REGION address if necessary) to local government area of residence at diagnosis; these were aggregated to conform to NSW Health Service Areas and Regions as they existed during most of the latter part of the period to which the data apply. An exception was for Central and Eastern Sydney Health Service Areas, where the border transected Sydney and South Sydney local government areas; geocoding to the Area Health Service used exact addresses in this instance. For analysis of survival by socioeconomic status, a marker was generated for each region from census data from the Australian Bureau of Statistics ( A B S ) for adults (aged over 15 years) which consisted of the ratio of the proportion with university qualifications to the proportion with no postschool qualificat i o n ~Such a marker takes into account some of the .~ social heterogeneity within regions. Quartiles were established from the data; there was no change in the socioeconomic status quartile for each region over the period 1981-1991 based on the information from successive censuses. The 25 793 cancer cases included in this analysis were those diagnosed in NSW women from 1980 to 1991 inclusive; survival follow-up extended to 1992. These data apply only to invasive breast cancer. Ductal carcinoma in situ and other in situ breast neoplasms are not included. Survival of breast cancer cases was determined by passive follow-up. All women with breast cancer not known to be dead by the registry were matched against the death records from the NSW Registrar of Births, Deaths, and Marriages, enhanced by information obtained from the A B S . The matching techniquesx-'' have been d e ~ c r i b e d . ~ For the purposes of this survival analysis, it was assumed that those cases not known to be dead were alive. While this assumption can lead to an overestimation of breast cancer survival (because of matching difficulties and out-of-state noncancer deaths), this would be minimal at five years since most deaths of cases during this period would be attributable to the disease, or would have breast cancer mentioned on the death certificate, and ascertainment of these deaths is likely to be high. instances where the abridged tables could not adequately be modelled, unabridgement was undertaken by estimating single-year mortality from 0 to 5 years from the <1, 1-4 and 5-9 year mortality and then fitting a series of curves and straight lines to the mx values for the various age ranges, using least squares regression. Following unabridgment of the period life tables, annual q, and ex values were obtained by linear interpolation, using the middle years of the quinquennia and triennum under study; and 1991 data were replicated for 1992. Population mortality data The expected survival for NSW women for 19801992 by 16 regions was not available from published sources and required calculation. Because of small numbers of deaths in some age groups in some regions, abridged life tables were constructed" for each region from unit record mortality data and population data (from the A B S ) , aggregated from local government areas, for the quinquennia 1980-1984 and 1985-1989, and the triennium 1990-1992, with age groups <1, 0-4, 5-9, ... 285 years. The central death rate (mx),the probability of surviving during the interval (q,), and the life expectancy at particular ages (ex)were calculated in the usual way." The life tables were then unabridged to single years of age and extrapolated to age 99 years by fitting a three-component In the few Analysis Relative survival was calculated based on the period between diagnosis (date of 'definitive' treatment) and death (from any cause) for every case, and the age- and period-matched survival for all women in each region.',' The relative survival analysis was carried out with a program designed by the Finnish Cancer Registry and Newcastle Univer~ity.'~~'' Calculations were made from each region's own particular set of life tables. The annual excess risk of death (for the first five years) due to breast cancer, which is the difference between the absolute mortality of cases and expected mortality of the age-, period-, and regionmatched populations, was modelled with proportional hazards regre~sion'~,'~,'' various predictor on variables: age group, degree of spread and region of residence at diagnosis. Southern Sydney was chosen as the referent region because it had average breast cancer survival. Adjustment for the year of follow-up was accomplished by introduction of a variable with five categories for each year of survival. For this analysis (which employed regional life tables), the 16 aggregate data files put out by the program (one for each region) were concatenated prior to regression analysis. No period effect was included in order to contain the number of strata. Statistical significance was assessed by change in deviance and degrees of freedom evaluated as a chi-squared with the conventional level of 0.05. The year-of-follow-up variable was included in all regression analyses. The predictor variables were modelled in combination and including statistically significant first-order interaction terms. Models examining differences in relative risk (RR) for excess mortality by regon and aggregations of regions were developed, with adjustment for other determinants. Results Although there was variation in five-year relative survival for breast cancer by region during the period 1980-1991, the difference between the areas with the highest survival (76 per cent in North Coast and New England) and lowest survival (70 per cent in Central West) were not statistically significant (Table I). Although there was significant = 27.4, 15 df, P < 0.05) variation in RR for excess mortality by region (adjusted for five-year follow-up and age), no one regon was statistically different from Southern Sydney (referent group). Adjusting in addition for degree of spread of the breast cancer at diagnosis (x2 AUSTRALIAN AND NEW ZEAIAND JOURNAL O PUBLIC HEALTH 1997 F VOL. NO. TAYLOR Table 1 : Breast cancer five-year relative survival, New South Wales females 1980 to 1991 (follow-up to 1992) and relative risk for excess mortality, by region a Relative risk (RR) Adjusted for 5-year follow-up, agec, staged and interactions 5-vear relative survival New South Wales region Southern Sydney Eastern Sydney Central Sydney South West Sydney Western Sydney Wentworth (Sydney) Northern Sydney Central Coast Hunter (including Newcastle) Central West South East New South Wales Illawarra (including Wollongong) North Coast New England Orana-Far West South West New South Wales Adjusted for 5-year follow-up, age and interactions Mean CI RR 1 .oo 0.90 0.95 0.97 1.04 1 .oo 0.88 1.04 0.99 1.15 0.90 1.10 0.86 0.87 0.96 1.03 CI RR 1 .oo 0.87 0.91 0.96 1.07 0.96 0.93 1.08 0.94 1.08 0.92 1.05 0.90 0.98 0.94 0.93 CI 71 to75 72 to 78 71 to77 71 to77 70 to 76 70 to 78 73 to 77 68 to 74 70 to 76 66 to 74 71 to79 68 to 74 73 to 79 72 to 80 69 to 79 70 to 78 0.69 to 0.73 to 0.76 to 0.82 to 0.72 to 0.71 to 0.79 to 0.78 to 0.83 to 0.64 to 0.84 to 0.66 to 0.64 to 0.65 to 0.75 to 0.75 to 1.01 0.78 to 1.06 0.83 to 1.1 1 0.94 to 1.23 0.80 to 1.16 0.82 to 1.05 0.92 to 1.27 0.81 to 1.08 0.89 to 1.30 0.76to1.12 0.90 to 1.23 0.76 to 1.05 0.82 to 1.17 0.76 to 1.18 0.78t01.11 Notes: [a) Region-specific life tables used for underlying mortality. (b) Referent group. Ic) Five categories of age: 0-39, 40-49, 50-59, 6C-69, 270. (d) Four Categories of stage: localised, regional, metastatic, unknown. (e) CI = 95% confidence interval. (f) Range 0.86 to 1.15 , residual deviance for model = 4928, 1527 df; (9) Range 0.87 to 1.08, residual deviance for model = 1659, 1500 df; age, stage x follow-up x2 x2 for region = 27.4, 15 df, P c 0.05; significant interaction: age x follow up. for region = 19.2, 15 df, P > 0.05; significant interactions: age x follow up, stage (and interactions) resulted in the region variable becoming nonsignificant. The 16 regions were grouped into quartiles according to a socioeconomic status marker based on postschool education. Regression analysis adjusting for age and year of follow-up (and interaction) indicated that socioeconomic status was significant = 8.9, 3 df, P < 0.05), with a significant linear trend = 5.1, 1 df, P < 0.05) for higher RR for excess mortality in the lower socioeconomic groups. The RR for excess mortality was 1.11 (95 per cent confidence interval (CI) 0.97 to 1.26) for the lowest quartile compared with the highest (referent). However, socioeconomic status was not significant when it was added to a model including degree of spread of cancer (and interactions). Analysis by three geographic areas, capital city (Sydney), other metropolitan centres (Hunter and Illawarra), and rural, showed no significant effect; an analysis of border regions versus others also showed no significant difference. (x2 (x2 Discussion This analysis covers a near-complete enumeration of breast cancer cases in NSW women from 1980 to 1991. As outlined in the companion paper,f data completeness is likely to be high for breast cancers recorded by the NSM’ Cancer Registry. Relative survival calculations are subject to several kinds of bias and confounding; these are more evident with differential analysis by region of residence. The major possible bias arises from problems of ascertainment of noncancer deaths in these cancer cases, since the predominant method of follow-up is passive (through matching with death certificates). There were a few instances in which inquiries were made, and responses indicated migration out of NSW, these cases were censored from the analysis at that date. Interstate or international noncancer deaths in this cohort of NSW breast cancer cases which occurred within five years of diagnosis may have been missed. The effect of possible underascertainment o f deaths in the breast cancer cases will lead to inflation ofthe relative survival estimate, especially in the elderly. However, detailed analysis of this database has produced findings in the opposite direction to expected biast Ascertainment bias could be important in regional analysis because it may be that residents of border areas have a greater propensity to die in another state, lowering the death rates in the border regions. However, there was no evidence that border regions had higher survival rates than other areas. While the age and period categories were reasonably able to describe the variability in the data, the categories of degree of spread at diagnosis (localised, regional and metastatic) probably con- AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 1997 vot. 21 NO. 2 BREAST CANCER SURVIVAL BY REGION tain considerable heterogeneity within them, and there was an undesirably large proportion of cases with unknown spread. This method of classification of cancer spread is used by several major cancer registries around the world," such as the Surveillance Epidemiology and End-results program in the United States (US) (which grouped nine US population-based registries) and the Danish Cancer Registry.'"-21 Although the degree-of-spread classification is crude, it has been found possible to use it for a high proportion of cancers on a populationwide basis." Its effect in explaining variability in the data in regression analyses is many orders of magnitude greater than that of age or period of diagnosis.' Breast survival data using this classification are generally comparable to those from other Australian registries that have published data according to V"" UICC TNM stages I, 11-111 and I . - ' Various confounders of the relationship between excess mortality from breast cancer and region of residence were dealt with in the proportional hazards regression: in particular, age and degree of spread (and their interaction) were controlled for. The significant effects of these variables and their interactions on RR for excess mortality are described in the companion paper..' Another potential confounder in relative survival analysis is the underlying mortality included in the calculations. Because there exists heterogeneity in mortality between regions, separate life tables were used in the analysis to obviate this problem. This study has documented a small statistically significant variation in relative survival of breast cancer in NSW women by region of residence (adjusted for age and year of follow-up), which became nonsignificant when the data were also adjusted for degree of spread at diagnosis (and interactions). Survival or excess mortality from breast cancer by region (adjusted for age and year of f o l l o ~ w p ) , while providing information on the overall functioning of health services, does not distinguish between delay in diagnosis and the effectiveness of treatment. Although the degree-of-spread data are relatively crude, they can be used to adjust, at least partly, for the effect of variation in stage at diagnosis by region. Inclusion of degree of spread (and its interactions with other VdriableS) in the proportional hazards regression model of excess mortality made the effect of region nonsignificant. This suggests that part of the variation between regions is due to variation in degree of spread of disease at diagnosis, and this is supported by data on variation in degree of spread by region.' The implication is that there was no significant variation in the effectiveness of treatment for breast cancer by region of residence for NSW women during from 1980 to 1991. Bonett et al. documented significantly lower fiveyear relative survival for breast cancer in nonmetropolitan areas of South Australia compared with Adelaide, particularly in older The poorer outcome in the rural area was maintained when the data were adjusted for tumour size and nodal spread. However, some of this variation may have been due to differences in underlying mortality between Adelaide and rural women, because state life tables were used for both groups. New South Wales data do not show worse overall outcomes in rural areas than in Sydney metropolitan and regional centres (Hunter and Illawarra). Analysis of variation in five-year relative survival for breast cancer in Finland during the 1970s revealed a range from 59 per cent to 76 per cent for 21 health service districts; appropriate underlying mortality rates were used." Most of the differences were due to chance and confounding by age and degree of spread at diagnosis (localised or not). However, there was evidence that persons residing in districts near university teaching hospitals had better survival than others, after the effects of confounding and random variation were allowed for. In the \Vest Midlands, United Kingdom (UK), variation in five-year relative survival for breast cancer cases incident in 1981-1985 among the 22 health districts was 55 per cent to 73 per cent." In Northwest England during 1985-1989, five-year relative survival varied from 58 per cent to 69 per cent among 18 health districts.'9 For Southeast England, relative five-year sunival for breast cancer varied from 63 per cent to 81 per cent for 1986-1988 among 32 Thames health districts."" In Southwest England, five-year survival varied from 60 per cent to 73 per cent among 11 health districts for cancers diagnosed from 1987 to 1989.'" In Quebec province (Canada) variation in relative five-year survival for breast cancer between 12 health districts (with over 200 cases) for 1984-1986 was 64 per cent to 75 per cent."? However, it is unclear to what extent these differences were confounded by differences in underlying mortality (since no indication was given that district-specific life tables were used), particularly because socioeconomic differentials (which imply mortality differentials) were noted in some studies."' Furthermore, confidence intervals were often not provided, and therefore the statistical significance of differences cannot be assessed. Regional analysis by socioeconomic status quartile revealed a small but significant linear trend for increasing RR for excess mortality with decreasing socioeconomic status, when it was adjusted for age and five-year follow-up. No socioeconomic status effect was observed when data were also adjusted for degree of spread at diagnosis, suggesting that the socioeconomic status effect is mediated through a pattern of more advanced cancers in lower socioeconomic status groups. This is likely to be the minimum socioeconomic status effect, since regions (compared with local government areas or postcode areas) consist of large socially heterogeneous populations, analysis of which would provide better estimates of socioeconomic status effects. Bonett et al. found some evidence of social stratification in breast cancer survival in Adelaide, but this was no longer evident when the data were adjusted for tumour diameter and nodal status."' International studies of breast cancer survival in relation to socioeconomic status (using both individual and spatial measures of socioeconomic status) NO. AUSTRALIAN AND NEW ZEALAND JOURNAL O PUBLIC HEALTH 1997 vot. 21 F TAYLOR have generally found higher survival rates in upper than in lower socioeconomic status groups. In The Netherlands, differentials disappeared when the data were adjusted for stage at diagnosis,'3'' in sevbut eral other studies, in Finland,:'4USy5and UK,""."' differentials remained-although usually to a lesser degree. New South Wales regions were aggregated into capital city (Sydney), other metropolitan centres (Hunter and Illawarra) and others (rural), which may reflect differential access to more sophisticated medical services for cancer treatment. No effect in the regression analysis was noted with this variable. This study demonstrates a small effect of region of residence and implied socioeconomic status on breast cancer survival in NSM' women, but this becomes nonsignificant when the data are also adjusted for degree of spread at diagnosis. This suggests that earlier diagnosis would be of benefit in reducing minor inequalities in breast cancer survival, but there appears no reason to suspect that there were significant variations in efficacy of treatment for breast cancer in NSM7 women by region of residence during 1980-1991. 1 1 HakulilleIi T, Gibberd R, Abe)ivickrama K, S6dcrman B. '4 computrrpr-ogmmpnckrigejir rancp1 survianl studirs. Version 1.O. (:ancer Soc. Finland Puhl. n o . 39, Newcastle, NSU': Cniversity of Newcastle, and Finnish (hncer Registv, 1988. 15 Hakcllinen T, Abeyickrama KH. A computer program package for relative s u n i \ d analysis. Cornput Prop Bioinrd 1985; Acknowledgments The data upon which this analysis was based were provided by the NSM' Central Cancer Registry. Special thanks are due to Ms Marylon Coates for data quality and extraction. The NSW Central Cancer Registry is managed by the NSM' Cancer Council, under contract to and with financial support from NSM' Health Department. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Australian and New Zealand Journal of Public Health Wiley

Breast cancer five–year survival, by New South Wales regions, 1980 to 1991

Download PDF
5 pages

Loading next page...
 
/lp/wiley/breast-cancer-five-year-survival-by-new-south-wales-regions-1980-to-pP7osQy6RC

References (34)

Publisher
Wiley
Copyright
Copyright © 1997 Wiley Subscription Services, Inc., A Wiley Company
ISSN
1326-0200
eISSN
1753-6405
DOI
10.1111/j.1467-842X.1997.tb01684.x
Publisher site
See Article on Publisher Site

Abstract

Abstract: Breast cancer five-year relative survival was calculated for 16 urban and rural regions in New South Wales (NSW) for cases incident in 1980-1991. Survival analysis employed cancer registry data linked with the death register, and age- and periodmatched regional mortality of NSW women. Proportional hazard regression analysis was used to compare excess mortality in breast cancer cases in each region. The effect of region was significant ( P < 0.05) in the analysis, after age and the follow-up variable (and their interaction) were adjusted for, although no region was significantly different from the referent group (chosen because of average relative five-year survival). When degree of spread and its interactions were entered into the model, the effect of region became nonsignificant. A significant linear trend ( P < 0.05) in the adjusted relative risk for excess mortality in breast cancer cases was noted when regions were divided into quartiles based on socioeconomic status, with higher relative risk in low-socioeconomic-status groups; this effect also disappeared with adjustment for degree of spread at diagnosis. There was no general effect of rurality versus capital city or other metropolitan centres. This study demonstrates a small effect of region of residence and implied socioeconomic status on breast cancer survival in NSW women, but this becomes nonsignificant when the data are adjusted for degree of spread at diagnosis. This suggests that earlier diagnosis would be of benefit in reducing minor inequalities in breast cancer survival in NSW women. (Aust N ZJPublic Health 1997; 21: 206-10) URVIVAL. is an important outcome measure in cancer control. It is influenced by health promotion and health service factors (including screening) that affect the stage at which the cancer is diagnosed, and by access to and quality of medical services, which determine promptness and efficacy of treatment. The two main methods of measuring cancer survival in population or clinical cohort studies are: relative survival (which compares the survival of cases with the age-, period- and sex-matched population from which they arise, and which does not require cause-of-death information) ;I,* and cancerspecific survival (which involves censoring of noncancer deaths in the survival analysis, and requires precise cause-ofdeath data) .3 Similar results, particularly for five-year survival, are obtained from both meth~ds.~ Differential survival by area of residence is useful for surveillance of outcomes of health services. Populations that have particularly poor outcomes require more detailed investigation to ascertain then rectify problems; areas with particularly good outcomes can profitably be studied to define characteristics that could be adopted usefully in other areas. Survival data can be used to establish benchmarks for achievable outcomes, and for target-setting within the planning process. This article examines differential five-year relative survival for breast cancer cases in New South Wales (NSW) by 16 areas of residence for 1980-1991. Data for NSW as a whole have been published in this issue (pp. 199-205) .4 Since different population subgroups may have different levels of underlying mortality and this may affect the relative survival analysis, this report uses regionspecific life tables in the calculations. Methods Cancer data Correspondence to Associate Professor Richard Taylor, Department of Public Health and Community Medicine, LJniversity of Sydney, NSW 2006. Fax (02) 9351 4179. The NSW Central Cancer Registry, a populationbased registry, began data collection in January 1972. Notification of malignant neoplasms has been a statutory requirement for all NSW public and private hospitals, radiotherapy departments and nursing homes since 1972, and for pathology and outpatients departments since 1985. For the period 1972 to 1991, the date of diagnosis was defined as the date of first definitive treatment for cancer. Information concerning completeness of enumeration3,6and histological confirmation of breast cance? has been published.* Data on degree of spread of tumour was recorded from notifiers as: 1. localised to tissue of origin (localised) ; 2. invasion of adjacent organs or regional lymph nodes (regional); and 3. distant metastasis (metastatic). Degree of spread at diagnosis was 82 per cent complete for 1978-1982, and improved to 88 per cent complete for 198&1991.5 Cancer cases were geocoded by a computer program by postcode and locality name (and street NO AUSTRALIAN AND NEW ZEALAND JOURNAL O PUBLIC HEALTH 1997 vot. 21 F BREAST CANCER SURVIVAL BY REGION address if necessary) to local government area of residence at diagnosis; these were aggregated to conform to NSW Health Service Areas and Regions as they existed during most of the latter part of the period to which the data apply. An exception was for Central and Eastern Sydney Health Service Areas, where the border transected Sydney and South Sydney local government areas; geocoding to the Area Health Service used exact addresses in this instance. For analysis of survival by socioeconomic status, a marker was generated for each region from census data from the Australian Bureau of Statistics ( A B S ) for adults (aged over 15 years) which consisted of the ratio of the proportion with university qualifications to the proportion with no postschool qualificat i o n ~Such a marker takes into account some of the .~ social heterogeneity within regions. Quartiles were established from the data; there was no change in the socioeconomic status quartile for each region over the period 1981-1991 based on the information from successive censuses. The 25 793 cancer cases included in this analysis were those diagnosed in NSW women from 1980 to 1991 inclusive; survival follow-up extended to 1992. These data apply only to invasive breast cancer. Ductal carcinoma in situ and other in situ breast neoplasms are not included. Survival of breast cancer cases was determined by passive follow-up. All women with breast cancer not known to be dead by the registry were matched against the death records from the NSW Registrar of Births, Deaths, and Marriages, enhanced by information obtained from the A B S . The matching techniquesx-'' have been d e ~ c r i b e d . ~ For the purposes of this survival analysis, it was assumed that those cases not known to be dead were alive. While this assumption can lead to an overestimation of breast cancer survival (because of matching difficulties and out-of-state noncancer deaths), this would be minimal at five years since most deaths of cases during this period would be attributable to the disease, or would have breast cancer mentioned on the death certificate, and ascertainment of these deaths is likely to be high. instances where the abridged tables could not adequately be modelled, unabridgement was undertaken by estimating single-year mortality from 0 to 5 years from the <1, 1-4 and 5-9 year mortality and then fitting a series of curves and straight lines to the mx values for the various age ranges, using least squares regression. Following unabridgment of the period life tables, annual q, and ex values were obtained by linear interpolation, using the middle years of the quinquennia and triennum under study; and 1991 data were replicated for 1992. Population mortality data The expected survival for NSW women for 19801992 by 16 regions was not available from published sources and required calculation. Because of small numbers of deaths in some age groups in some regions, abridged life tables were constructed" for each region from unit record mortality data and population data (from the A B S ) , aggregated from local government areas, for the quinquennia 1980-1984 and 1985-1989, and the triennium 1990-1992, with age groups <1, 0-4, 5-9, ... 285 years. The central death rate (mx),the probability of surviving during the interval (q,), and the life expectancy at particular ages (ex)were calculated in the usual way." The life tables were then unabridged to single years of age and extrapolated to age 99 years by fitting a three-component In the few Analysis Relative survival was calculated based on the period between diagnosis (date of 'definitive' treatment) and death (from any cause) for every case, and the age- and period-matched survival for all women in each region.',' The relative survival analysis was carried out with a program designed by the Finnish Cancer Registry and Newcastle Univer~ity.'~~'' Calculations were made from each region's own particular set of life tables. The annual excess risk of death (for the first five years) due to breast cancer, which is the difference between the absolute mortality of cases and expected mortality of the age-, period-, and regionmatched populations, was modelled with proportional hazards regre~sion'~,'~,'' various predictor on variables: age group, degree of spread and region of residence at diagnosis. Southern Sydney was chosen as the referent region because it had average breast cancer survival. Adjustment for the year of follow-up was accomplished by introduction of a variable with five categories for each year of survival. For this analysis (which employed regional life tables), the 16 aggregate data files put out by the program (one for each region) were concatenated prior to regression analysis. No period effect was included in order to contain the number of strata. Statistical significance was assessed by change in deviance and degrees of freedom evaluated as a chi-squared with the conventional level of 0.05. The year-of-follow-up variable was included in all regression analyses. The predictor variables were modelled in combination and including statistically significant first-order interaction terms. Models examining differences in relative risk (RR) for excess mortality by regon and aggregations of regions were developed, with adjustment for other determinants. Results Although there was variation in five-year relative survival for breast cancer by region during the period 1980-1991, the difference between the areas with the highest survival (76 per cent in North Coast and New England) and lowest survival (70 per cent in Central West) were not statistically significant (Table I). Although there was significant = 27.4, 15 df, P < 0.05) variation in RR for excess mortality by region (adjusted for five-year follow-up and age), no one regon was statistically different from Southern Sydney (referent group). Adjusting in addition for degree of spread of the breast cancer at diagnosis (x2 AUSTRALIAN AND NEW ZEAIAND JOURNAL O PUBLIC HEALTH 1997 F VOL. NO. TAYLOR Table 1 : Breast cancer five-year relative survival, New South Wales females 1980 to 1991 (follow-up to 1992) and relative risk for excess mortality, by region a Relative risk (RR) Adjusted for 5-year follow-up, agec, staged and interactions 5-vear relative survival New South Wales region Southern Sydney Eastern Sydney Central Sydney South West Sydney Western Sydney Wentworth (Sydney) Northern Sydney Central Coast Hunter (including Newcastle) Central West South East New South Wales Illawarra (including Wollongong) North Coast New England Orana-Far West South West New South Wales Adjusted for 5-year follow-up, age and interactions Mean CI RR 1 .oo 0.90 0.95 0.97 1.04 1 .oo 0.88 1.04 0.99 1.15 0.90 1.10 0.86 0.87 0.96 1.03 CI RR 1 .oo 0.87 0.91 0.96 1.07 0.96 0.93 1.08 0.94 1.08 0.92 1.05 0.90 0.98 0.94 0.93 CI 71 to75 72 to 78 71 to77 71 to77 70 to 76 70 to 78 73 to 77 68 to 74 70 to 76 66 to 74 71 to79 68 to 74 73 to 79 72 to 80 69 to 79 70 to 78 0.69 to 0.73 to 0.76 to 0.82 to 0.72 to 0.71 to 0.79 to 0.78 to 0.83 to 0.64 to 0.84 to 0.66 to 0.64 to 0.65 to 0.75 to 0.75 to 1.01 0.78 to 1.06 0.83 to 1.1 1 0.94 to 1.23 0.80 to 1.16 0.82 to 1.05 0.92 to 1.27 0.81 to 1.08 0.89 to 1.30 0.76to1.12 0.90 to 1.23 0.76 to 1.05 0.82 to 1.17 0.76 to 1.18 0.78t01.11 Notes: [a) Region-specific life tables used for underlying mortality. (b) Referent group. Ic) Five categories of age: 0-39, 40-49, 50-59, 6C-69, 270. (d) Four Categories of stage: localised, regional, metastatic, unknown. (e) CI = 95% confidence interval. (f) Range 0.86 to 1.15 , residual deviance for model = 4928, 1527 df; (9) Range 0.87 to 1.08, residual deviance for model = 1659, 1500 df; age, stage x follow-up x2 x2 for region = 27.4, 15 df, P c 0.05; significant interaction: age x follow up. for region = 19.2, 15 df, P > 0.05; significant interactions: age x follow up, stage (and interactions) resulted in the region variable becoming nonsignificant. The 16 regions were grouped into quartiles according to a socioeconomic status marker based on postschool education. Regression analysis adjusting for age and year of follow-up (and interaction) indicated that socioeconomic status was significant = 8.9, 3 df, P < 0.05), with a significant linear trend = 5.1, 1 df, P < 0.05) for higher RR for excess mortality in the lower socioeconomic groups. The RR for excess mortality was 1.11 (95 per cent confidence interval (CI) 0.97 to 1.26) for the lowest quartile compared with the highest (referent). However, socioeconomic status was not significant when it was added to a model including degree of spread of cancer (and interactions). Analysis by three geographic areas, capital city (Sydney), other metropolitan centres (Hunter and Illawarra), and rural, showed no significant effect; an analysis of border regions versus others also showed no significant difference. (x2 (x2 Discussion This analysis covers a near-complete enumeration of breast cancer cases in NSW women from 1980 to 1991. As outlined in the companion paper,f data completeness is likely to be high for breast cancers recorded by the NSM’ Cancer Registry. Relative survival calculations are subject to several kinds of bias and confounding; these are more evident with differential analysis by region of residence. The major possible bias arises from problems of ascertainment of noncancer deaths in these cancer cases, since the predominant method of follow-up is passive (through matching with death certificates). There were a few instances in which inquiries were made, and responses indicated migration out of NSW, these cases were censored from the analysis at that date. Interstate or international noncancer deaths in this cohort of NSW breast cancer cases which occurred within five years of diagnosis may have been missed. The effect of possible underascertainment o f deaths in the breast cancer cases will lead to inflation ofthe relative survival estimate, especially in the elderly. However, detailed analysis of this database has produced findings in the opposite direction to expected biast Ascertainment bias could be important in regional analysis because it may be that residents of border areas have a greater propensity to die in another state, lowering the death rates in the border regions. However, there was no evidence that border regions had higher survival rates than other areas. While the age and period categories were reasonably able to describe the variability in the data, the categories of degree of spread at diagnosis (localised, regional and metastatic) probably con- AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 1997 vot. 21 NO. 2 BREAST CANCER SURVIVAL BY REGION tain considerable heterogeneity within them, and there was an undesirably large proportion of cases with unknown spread. This method of classification of cancer spread is used by several major cancer registries around the world," such as the Surveillance Epidemiology and End-results program in the United States (US) (which grouped nine US population-based registries) and the Danish Cancer Registry.'"-21 Although the degree-of-spread classification is crude, it has been found possible to use it for a high proportion of cancers on a populationwide basis." Its effect in explaining variability in the data in regression analyses is many orders of magnitude greater than that of age or period of diagnosis.' Breast survival data using this classification are generally comparable to those from other Australian registries that have published data according to V"" UICC TNM stages I, 11-111 and I . - ' Various confounders of the relationship between excess mortality from breast cancer and region of residence were dealt with in the proportional hazards regression: in particular, age and degree of spread (and their interaction) were controlled for. The significant effects of these variables and their interactions on RR for excess mortality are described in the companion paper..' Another potential confounder in relative survival analysis is the underlying mortality included in the calculations. Because there exists heterogeneity in mortality between regions, separate life tables were used in the analysis to obviate this problem. This study has documented a small statistically significant variation in relative survival of breast cancer in NSW women by region of residence (adjusted for age and year of follow-up), which became nonsignificant when the data were also adjusted for degree of spread at diagnosis (and interactions). Survival or excess mortality from breast cancer by region (adjusted for age and year of f o l l o ~ w p ) , while providing information on the overall functioning of health services, does not distinguish between delay in diagnosis and the effectiveness of treatment. Although the degree-of-spread data are relatively crude, they can be used to adjust, at least partly, for the effect of variation in stage at diagnosis by region. Inclusion of degree of spread (and its interactions with other VdriableS) in the proportional hazards regression model of excess mortality made the effect of region nonsignificant. This suggests that part of the variation between regions is due to variation in degree of spread of disease at diagnosis, and this is supported by data on variation in degree of spread by region.' The implication is that there was no significant variation in the effectiveness of treatment for breast cancer by region of residence for NSW women during from 1980 to 1991. Bonett et al. documented significantly lower fiveyear relative survival for breast cancer in nonmetropolitan areas of South Australia compared with Adelaide, particularly in older The poorer outcome in the rural area was maintained when the data were adjusted for tumour size and nodal spread. However, some of this variation may have been due to differences in underlying mortality between Adelaide and rural women, because state life tables were used for both groups. New South Wales data do not show worse overall outcomes in rural areas than in Sydney metropolitan and regional centres (Hunter and Illawarra). Analysis of variation in five-year relative survival for breast cancer in Finland during the 1970s revealed a range from 59 per cent to 76 per cent for 21 health service districts; appropriate underlying mortality rates were used." Most of the differences were due to chance and confounding by age and degree of spread at diagnosis (localised or not). However, there was evidence that persons residing in districts near university teaching hospitals had better survival than others, after the effects of confounding and random variation were allowed for. In the \Vest Midlands, United Kingdom (UK), variation in five-year relative survival for breast cancer cases incident in 1981-1985 among the 22 health districts was 55 per cent to 73 per cent." In Northwest England during 1985-1989, five-year relative survival varied from 58 per cent to 69 per cent among 18 health districts.'9 For Southeast England, relative five-year sunival for breast cancer varied from 63 per cent to 81 per cent for 1986-1988 among 32 Thames health districts."" In Southwest England, five-year survival varied from 60 per cent to 73 per cent among 11 health districts for cancers diagnosed from 1987 to 1989.'" In Quebec province (Canada) variation in relative five-year survival for breast cancer between 12 health districts (with over 200 cases) for 1984-1986 was 64 per cent to 75 per cent."? However, it is unclear to what extent these differences were confounded by differences in underlying mortality (since no indication was given that district-specific life tables were used), particularly because socioeconomic differentials (which imply mortality differentials) were noted in some studies."' Furthermore, confidence intervals were often not provided, and therefore the statistical significance of differences cannot be assessed. Regional analysis by socioeconomic status quartile revealed a small but significant linear trend for increasing RR for excess mortality with decreasing socioeconomic status, when it was adjusted for age and five-year follow-up. No socioeconomic status effect was observed when data were also adjusted for degree of spread at diagnosis, suggesting that the socioeconomic status effect is mediated through a pattern of more advanced cancers in lower socioeconomic status groups. This is likely to be the minimum socioeconomic status effect, since regions (compared with local government areas or postcode areas) consist of large socially heterogeneous populations, analysis of which would provide better estimates of socioeconomic status effects. Bonett et al. found some evidence of social stratification in breast cancer survival in Adelaide, but this was no longer evident when the data were adjusted for tumour diameter and nodal status."' International studies of breast cancer survival in relation to socioeconomic status (using both individual and spatial measures of socioeconomic status) NO. AUSTRALIAN AND NEW ZEALAND JOURNAL O PUBLIC HEALTH 1997 vot. 21 F TAYLOR have generally found higher survival rates in upper than in lower socioeconomic status groups. In The Netherlands, differentials disappeared when the data were adjusted for stage at diagnosis,'3'' in sevbut eral other studies, in Finland,:'4USy5and UK,""."' differentials remained-although usually to a lesser degree. New South Wales regions were aggregated into capital city (Sydney), other metropolitan centres (Hunter and Illawarra) and others (rural), which may reflect differential access to more sophisticated medical services for cancer treatment. No effect in the regression analysis was noted with this variable. This study demonstrates a small effect of region of residence and implied socioeconomic status on breast cancer survival in NSM' women, but this becomes nonsignificant when the data are also adjusted for degree of spread at diagnosis. This suggests that earlier diagnosis would be of benefit in reducing minor inequalities in breast cancer survival, but there appears no reason to suspect that there were significant variations in efficacy of treatment for breast cancer in NSM7 women by region of residence during 1980-1991. 1 1 HakulilleIi T, Gibberd R, Abe)ivickrama K, S6dcrman B. '4 computrrpr-ogmmpnckrigejir rancp1 survianl studirs. Version 1.O. (:ancer Soc. Finland Puhl. n o . 39, Newcastle, NSU': Cniversity of Newcastle, and Finnish (hncer Registv, 1988. 15 Hakcllinen T, Abeyickrama KH. A computer program package for relative s u n i \ d analysis. Cornput Prop Bioinrd 1985; Acknowledgments The data upon which this analysis was based were provided by the NSM' Central Cancer Registry. Special thanks are due to Ms Marylon Coates for data quality and extraction. The NSW Central Cancer Registry is managed by the NSM' Cancer Council, under contract to and with financial support from NSM' Health Department.

Journal

Australian and New Zealand Journal of Public Health – Wiley

Published: Apr 1, 1997

There are no references for this article.