Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Are international differences in breast cancer survival between A ustralia and the UK present amongst both screen‐detected women and non‐screen‐detected women? survival estimates for women diagnosed in W est M idlands and N ew S outh W ales 1997–2006

Are international differences in breast cancer survival between A ustralia and the UK present... IJC International Journal of Cancer Are international differences in breast cancer survival between Australia and the UK present amongst both screen-detected women and non-screen-detected women? Survival estimates for women diagnosed in West Midlands and New South Wales 1997–2006 1 1 2 3 1 Laura M. Woods , Bernard Rachet , Dianne L. O’Connell , Gill Lawrence and Michel P. Coleman Cancer Research UK Cancer Survival Group, Department of Non-Communicable Disease Epidemiology, London School of Hygiene and Tropical Medicine, London, WC1E 7HT Cancer Research Division, Cancer Council NSW, NSW 1340, Australia Breast Cancer Audit Consultant and Former Director, West Midlands Cancer Intelligence Unit, Public Health Building, University of Birmingham, Birmingham, B15 2TT We examined survival in screened-detected and non-screen-detected women diagnosed in the West Midlands (UK) and New South Wales (Australia) in order to evaluate whether international differences in survival are related to early diagnosis, or to other factors relating to the healthcare women receive. Data for women aged 502 65 years who had been eligible for screen- ing from 50 years were examined. Data for 5,628 women in West Midlands and 6,396 women in New South Wales were linked to screening service records (mean age at diagnosis 53.7 years). We estimated net survival and modelled the excess hazard ratio of breast cancer death by screening status. Survival was lower for women in the West Midlands than in New South Wales (5-year net survival 90.9% [95% CI 89.9%291.7%] compared with 93.4% [95% CI 92.6%-94.1%], respectively). The difference was greater between the two populations of non-screen-detected women (4.9%) compared to between screen-detected women, (1.8% after adjustment for lead-time and over-diagnosis). The adjusted excess hazard ratio of breast cancer death for West Midlands compared with New South Wales was greater in the non-screen-detected group (EHR 2.00, 95% CI 1.702 2.31) but not significantly different to that for women whose cancer had been screen-detected (EHR 1.72, 95% CI 0.872 2.56). In this study more than one in three breast cancer deaths in the West Midlands would have been avoided if survival had been the same as in New South Wales. The possibility that women in the UK receive poorer treatment is an important potential explanation which should be examined with care. We have previously shown a difference of 6% in 5-year breast tors or to other factors relating to the healthcare women receive. We have previously identified these as possible explan- cancer survival between Australia and England for women in ations for socioeconomic differences, but they also may the target age group for screening and diagnosed during the explain international variations in survival. period 19962 1999. Examining survival by screening status The trials that led to the implementation of mammo- has the potential to shed further light on whether international graphic screening worldwide were evaluated by examining differences are more likely to be due to tumour or patient fac- the reduction in breast cancer mortality amongst the popula- tions of women screened. In this context, a reduction in the Key words: breast cancer, net survival, excess mortality, UK, Aus- number of breast cancer deaths in the screened population tralia, New South Wales, West Midlands, cancer screening, can be interpreted as the number of cancer deaths avoided or mammography deferred by the intervention. This outcome is helpful in eval- This is an open access article under the terms of the Creative Commons uating the public health impact and economic value of the Attribution License, which permits use, distribution and reproduction screening programme as a whole. in any medium, provided the original work is properly cited. Other studies have examined the impact of mammo- DOI: 10.1002/ijc.29984 History: Received 12 Aug 2015; Accepted 30 Nov 2015; Online 12 graphic screening upon individual patient survival. Analyses 4,5 Jan 2016 of survival include examinations of interval cancers (can- Correspondence to: Laura M. Woods, Cancer Research UK Cancer cers diagnosed following a normal mammogram but prior to Survival Group, Department of Non-Communicable Disease Epide- the next screening invitation), comparisons of women in miology, London School of Hygiene an Tropical Medicine, Keppel dichotomous groups (attenders vs. never-attenders and those Street, London, WC1E 7HT, Tel: 144 20 7612 7849, Fax: 144 20 with screen-detected vs. non-screen-detected cancers ) and 8 9 7436 4230, E-mail: [email protected] spatial analyses. A review conducted in the UK in 2003 V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Woods et al. 2405 What’s new? Breast cancer patients in Australia are known to have higher survival than those in the UK. In this study, the authors found that these international differences in survival persist in both screen-detected and non-screen-detected groups, even after adjustment for both lead-time bias and over-diagnosis. These results suggest that it is essential that the mechanisms underly- ing these differences be understood, including potential variations in effective treatment between the two regions. concluded that a better understanding of the effect of screen- screening programme was being established and expanded. detection required more detailed data. In particular, the We excluded women aged over 65 years at diagnosis because review identified the importance of linkage of mortality data the target age for screening was up to age 64 years in the UK to screening invitations so that the outcome for tumours during this period. The eligibility criteria resulted in a cohort diagnosed after the introduction of screening might be which built up over time (median month of diagnosis August examined. 2003 in West Midlands and November 2003 in New South Examination of survival by screening status enables us to Wales). All women were followed up to 31 December 2008 establish, at the population level, the survival benefit afforded (at least 2 years following diagnosis). Data were obtained to women whose cancers were screen-detected compared to from the West Midlands Office of the English National Can- women whose cancers were detected symptomatically. The cer Registration Service (WMNCRS, England) and the New disadvantage of this approach is that it is susceptible to lead- South Wales Central Cancer Registry (NSWCCR, Australia). time bias and to over-diagnosis. Lead-time is the additional These two registries cover populations of 5.6 and 6.9 million, 12,13 observation time credited to women who are screen-detected respectively. by virtue of the fact that they are asymptomatic. Breast Information was obtained from each cancer registry on tumours considered to be “over-diagnosed” are those detected each woman’s age at diagnosis (completed years), the month by screening mammography but which would not have been and year of their diagnosis and death (if dead), the sub-site, diagnosed during the patient’s lifetime in the absence of grade, histology and behaviour of the tumour, and all infor- screening. These biases together lead to apparently better mation pertaining to the extent of disease at diagnosis survival, even if the actual time of death is not deferred. This (stage). Staging information for cases in the West Midlands skews estimates of survival in favour of screening, resulting was recoded according to the rules used by the New South in statistics which appear to show a survival advantage Wales Central Cancer Registry: localised (confined to the amongst women who have been screened, even when none organ of origin), regional (spread to adjacent muscle, organ, might exist. Recently, methodological advances have been fat, connective tissue or regional lymph nodes), distant (dis- made into ways to account for lead-time bias in the analysis tant metastasis) and unknown stage. of survival so that the underlying differences in survival can The cancer registry data were linked to the population- be assessed. This involves correcting the observed survival based mammographic screening service records in each local- time to account for the additional follow-up observed in the ity to establish each woman’s screening status at diagnosis (the cohort as a result of screen-detection. National Health Service Breast Screening Programme for the In this article, we examine net survival for breast cancer in West Midlands and BreastScreen NSW for New South Wales). screen-detected and non-screen-detected women diagnosed in We defined four categories for the screening status at diagno- the West Midlands (UK) and New South Wales (Australia), sis: (1) women whose cancer was detected at a routine screen, applying a correction for lead-time bias and over-diagnosis. We (2) women who presented with cancer following a negative use the results to discuss the extent to which the international screen but before being invited to their next routine screen differences in breast cancer survival between the UK and Aus- (interval cancers), (3) women who presented with cancer after tralia may be explained by tumour or patient factors or to other at least one negative screen but who had not attended their factors relating to the healthcare women receive. most recent appointment (lapsed attenders), and (4) women who presented with cancer who had never attended screening. Materials We also compared women in the screen-detected group (Cate- The cohort of interest consisted of women who were invited gory 1) to all those with non-screen-detected cancer (Catego- to attend for screening mammography in a fully-functioning, ries 2, 3 and 4). This broadly corresponded to comparing mature screening programme during a defined calendar those with asymptomatic disease identified via routine screen- period. Women diagnosed with a primary invasive breast ing to women presenting with symptomatic disease. cancer at ages 50–65 years during the period 1 January 1997 to 31 December 2006 and aged 51 years or younger on 1 Jan- Methods uary 1997 were considered eligible (Fig. 1). We thus excluded Net survival estimation women who were first invited to be screened at ages over 50 Net survival is defined as the survival from the disease of years, as well as women invited during the years when the interest. It is derived by adjusting the overall survival in the V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology 2406 International differences in breast cancer survival Figure 1. (a) Schematic diagram of women eligible for the study alongside (b) a histogram showing the total number of women included in New South Wales (Australia) and the West Midlands (UK) by year of diagnosis (1997–2006). patient group for their expected survival in the absence of observed survival time in order to obtain corrected survival the disease. We estimated net survival using the non- time (Fig. 2, Patients A and B). parametric Pohar-Perme estimator, which has been imple- We considered tumours to be over-diagnosed if they mented in Stata. The Pohar-Perme estimator is an unbiased would not have been detected symptomatically during the estimator of net survival with respect to informative censor- study period or during the predicted lifetime of the patient. To account for over-diagnosis we excluded tumours in ing (defined as the tendency for the estimates to reflect the survival of patients with lowest expected mortality as time instances where the value of E(s) , E(s) .. . E(s) exceeded 1 2 10 16,17 since diagnosis increases) for population-based data. the woman’s actual observed survival time, either because the We estimated expected survival from region-specific life predicted date of diagnosis was after 31st December 2008, or tables provided by the Office for National Statistics for Eng- before her death. (Fig. 2, Patients C and D). land and Wales and the Australian Bureau of Statistics for We used the corrected survival times to estimate non- each calendar year of follow-up. parametric net survival for each of these ten separate data sets for the screen-detected group. We used the rules estab- lished by Rubin for the re-combination of estimates in a Adjustment for lead-time and over-diagnosis multiple-imputation setting to derive an overall estimate of To account for the potential effect of lead-time bias, we cal- net survival and its variance, adjusted for lead-time bias and culated additional survival time due to screening, E(s), for over-diagnosis (Fig. 3a). the screen-detected group, as proposed by Duffy et al. and assuming a mean sojourn time (time from carcinogenesis to symptomatic cancer in the absence of screening) in both Missing data regions of 4 years. We applied 10 separate simulations to Data on extent of disease were missing for 8.9% of women obtain a range of possible values, E(s) , E(s) .. . E(s) ,by diagnosed in West Midlands and 5.3% of those diagnosed in 1 2 10 assuming that survival times were exponentially distributed New South Wales. We used a 10-fold hot-deck approach to with a mean of E(s). Values of E(s) were subtracted from take account of these missing values for extent of disease. V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Woods et al. 2407 Figure 2. Schematic diagram demonstrating the exclusion of women in order to adjust for lead-time bias and over-diagnosis. The hot-deck approach involves identifying ‘donor groups’ indicate a better fit. We examined non-linearity of age by the for each woman with missing information on extent of dis- inclusion of restricted cubic splines and tested for time- ease. The donor group for each woman comprised women varying effects for region, age at diagnosis, and extent of dis- diagnosed in the same period (1997–2000, 2001–2006) and ease. We examined interactions between region and age, and region (West Midlands, New South Wales), at a similar age between region and extent of disease. (2 groups: 50–53 years [prevalent screening round], 541 For the screen-detected group, we applied the model with years [incident screening rounds]) who had been followed for the smallest number of parameters to each unique combina- a similar amount of time (6 groups: up to 1 year, 1-1.9 years, tion of E(s) and extent (100 separate combinations of i j 2-3.9 years, 4-5.9 years, 6-7.9 years and 81 years), and with results). For non-screen-detected women we refitted the the same vital status at the end of follow-up (dead, alive), model found to fit best to using values of extent to the data and screening status (screen-detected, not screen-detected). for extent , extent ... extent (10 sets of results). 2 3 10 For each woman with missing data, ten separate values of We predicted from the final models estimates of crude extent of disease (extent , extent ... extent ) were obtained mortality due to breast cancer and crude mortality due to 1 2 10 by randomly and independently selecting values of extent of other causes for the whole cohort. Crude mortality can be disease from the donor group. derived directly from the net survival models, and allows Combining these two procedures resulted in data sets with the mortality observed during follow-up to be partitioned a set of 10 imputed values for the variable extent for both into mortality due to the cancer itself and due to other the screen-detected group and non-screen-detected group, causes. Estimates of crude mortality were derived for each of and a set of 10 imputed values for the variable E(s) for the the covariate patterns in the sample and a weighted average screen-detected only (where i5 1–10). of deaths due to breast cancer across all patterns was calcu- lated by region and screening. Estimates were derived sepa- Modelling rately for screen-detected women and non-screen-detected We fitted flexible non-parametric regression models for net women in West Midlands and New South Wales. 19 18 survival to estimate the excess hazard ratio associated with We used Rubin’s rules to re-combine the 100 separate being diagnosed with breast cancer in the West Midlands estimates of the excess hazard ratio of breast cancer death compared to New South Wales. We fitted 10 models for and crude mortality from breast cancer for screen-detected women with screen-detected cancer using the values E(s) to women and the 10 separate estimates for non-screen-detected E(s) combined with extent and one model for the non- women. This resulted in separate estimates for screen- 10 1 screen-detected cancer using observed survival times and val- detected and non-screen-detected women of the relative ues of extent . A priori, we included age at diagnosis, region change in the excess hazard of death due to breast cancer for and extent of disease in the models. We used a reduction of women living in West Midlands compared to women in New 3 or more in the AIC (Akaike Information Criterion) to South Wales, as well as the crude probability of death from V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology 2408 International differences in breast cancer survival Figure 3. Schematic diagram illustrating (a) net survival estimation correcting for lead-time bias and over-diagnosis and (b) the modelling strategy taking into account missing values for extent of disease. breast cancer and other causes, and their associated variances. in New South Wales). Fewer than one in ten women died These final estimates took into account lead-time bias and during follow-up: 10.8% in West Midlands and 7.6% in New over-diagnosis in the screen-detected group and were also South Wales. Overall, net survival in the cohort was high (Table 1). adjusted for age and extent of disease at diagnosis (Fig. 3b). The estimates of crude mortality were used to establish Consistent with our previous findings, net survival overall the number of cancer deaths that could have been avoided in was significantly lower in the West Midlands than in New the hypothetical situation in which survival was equalised South Wales (5-year net survival 90.9% [95% CI 89.9%291.7%] and 93.4% [95% CI 92.6%294.1%], respec- between the two regions. This provides an estimate of the public health impact of survival differentials in the net sur- tively). Women diagnosed with interval cancers in New vival setting. South Wales had lower survival than screen-detected women (5-year net survival 93.5% compared to 98.5%), but better Results survival than women who had never attended screening We analysed data for 5,628 women in West Midlands (98.5% (89.5%) and those who had attended previously but lapsed in of those eligible, mean age at diagnosis 53.7 years) and 6,396 attendance prior to diagnosis (86.8%; Table 1, Fig. 4a). In women in New South Wales (99.9% of those eligible, mean West Midlands, however, the survival of women diagnosed age at diagnosis 53.8 years). Those excluded were the very with interval cancers was not dissimilar to that of lapsed small number of women who were known to the registry attenders, whilst those who had never attended had the worst only because breast cancer had been mentioned on their survival (Table 1, Fig. 4b). The difference in net survival death certificate (DCOs) or because the sequence of dates between West Midlands and New South Wales was greater provided was illogical. The proportion of tumours that were among non-screen-detected women (4.9% five years after screen-detected was greater in West Midlands (44.8% com- diagnosis) than among screen-detected women in the two pared to 36.5%, Table 1). The majority of women were diag- regions (1.8%; 1.0% before adjustment for lead-time bias, nosed with localised disease, (54.1% in West Midlands, 53.9% Table 1). V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Table 1. Net survival estimates at 1 and 5 years after diagnosis by mode of presentation and extent of disease at diagnosis: women aged 50–65 (mean age 53.7 years) diagnosed with invasive breast cancer 1 Jan- uary 1997–31 December 2006 and followed up to 31 December 2008 in New South Wales (Australia) and the West Midlands (UK) New South Wales West Midlands Deaths (% of N) within Net Survival, % (CI) Deaths (% of N) within Net Survival , % (CI) (a) Mode of presentation N (%) 1 year 5 years 1-year 5-year N (%) 1 year 5 years 1-year 5-year Screen-detected 2,335 (36.5) 11 (0.5) 54 (2.3) 99.8 (99.2,99.9) 98.5 (97.5,99.1) 2,524 (44.8) 11 (0.2) 90 (1.4) 99.9 (98.8,100.0) 97.5 (96.4,98.3) adjusted for lead-time 1,390 (21.7) 10 (0.7) 48 (3.5) 98.9 (98.3,99.5) 96.5 (95.2,97.9) 1,534 (27.3) 10 (0.7) 81 (5.3) 98.6 (97.9,99.3) 94.7 (93.2,96.2) Lapsed-attender 129 (2.0) 4 (3.1) 16 (12.4) 97.2 (92.0,99.0) 86.8 (78.2,92.2) 175 (3.1) 6 (0.1) 17 (0.3) 96.9 (92.7,98.7) 89.8 (82.6,94.2) Interval cancer 1,028 (16.1) 4 (0.4) 64 (6.2) 99.8 (98.4,100.0) 93.5 (91.3,95.2) 1,537 (27.3) 34 (0.5) 157 (2.5) 98.1 (97.2,98.7) 90.3 (88.4,92.0) Never-attender 2,904 (45.4) 86 (3.0) 297 (10.2) 97.3 (96.6,97.8) 89.5 (88.1,90.7) 1,392 (24.7) 97 (1.5) 280 (4.4) 93.3 (91.8,94.5) 79.8 (77.4,82.0) All groups 6,396 (100.0) 105 (1.6) 431 (6.7) 98.6 (98.3,98.9) 93.4 (92.6,94.1) 5,628 (100.0) 148 (2.3) 544 (8.5) 97.7 (97.2,98.1) 90.9 (89.9,91.7) Non-screen-detected Screen-detected Non-screen-detected Screen-detected Deaths (% of N) Net Survival, Deaths (% of N) Net Survival, Deaths (% of N) Net Survival, Deaths (% of N) Net Survival, (b) Extent of within % (CI) within % (CI) within % (CI) within % (CI) disease at diagnosis N (%) 1 year 5 years 1-year 5-year N (%) 1 year 5 years 1-year 5-year N (%) 1 year 5 years 1-year 5-year N (%) 1 year 5 years 1-year 5-year Localised 1,955 7 71 99.9 96.9 1,490 3 16 99.9 99.9 1,351 6 59 99.9 96.8 1,702 4 36 100.0 99.1 (48.1) (0.1) (4.1) (98.8,100.0) (95.7,97.8) (64.1) (0.1) (1.1) (99.0,100.0) (12.5,100.0) (44.1) (0.1) (4.4) (98.4,100.0) (95.3,97.9) (67.1) (0.1) (2.1) (100.0,116.3) (97.4,99.7) adjusted for N/A N/A N/A N/A N/A 875 2 10 99.7 99.2* N/A N/A N/A N/A N/A 1,008 3 27 99.4 97.3 lead-time (37.1) (0.1) (1.1) (99.1,100.2) (98.0,100.5) (40.1) (0.1) (3.1) (98.8,100.0) (95.7,98.8) Regional 1,644 19 162 99.1 89.5 693 7 28 99.2 96.3 1,319 39 241 97.3 80.9 634 6 49 99.4 92.6 (40.1) (1.1) (10.1) (98.4,99.5) (87.6,91.1) (30.1) (1.1) (4.1) (98.0,99.7) (93.9,97.8) (42.1) (3.1) (18.3) (96.3,98.1) (78.3,83.2) (25.1) (1.1) (8.1) (97.9,99.8) (89.6,94.8) Distant 224 56 99 75.1 51.2 52 1 6 98.3 - 115 54 83 53.1 19.5 9 0 1 - 89.7 (6.1) (3.1) (6.1) (68.9,80.3) (43.6,58.3) (2.1) (2.1) (12.1) (85.8,99.8) (4.1) (47.1) (72.2) (43.6,61.7) (11.1,29.6) (0.1) (0.0) (0.1) (44.5,98.6) Unknown 238 12 45 95.2 79.9 100 0 4 - 96.4 319 38 71 88.3 77.7 179 1 4 99.7 99.3 (6.1) (1.1) (3.1) (91.5,97.3) (73.5,85.0) (4.1) (0.0) (4.1) (87.0,99.0) (10.1) (12.1) (22.3) (84.2,91.4) (72.2,82.2) (7.1) (1.1) (2.1) (84.8,100.0) (81.9,100.0) All stages 4,061 94 377 97.9 90.4 2,335 11 54 99.8 98.5 3,104 137 454 95.9 85.5 2,524 11 90 99.9 97.5 (100.0) (2.1) (9.1) (97.4,98.3) (89.3,91.5) (100.0) (0.5) (2.3) (99.2,99.9) (97.5,99.1) (100.0) (4.1) (14.6) (95.1,96.6) (84.1,86.9) (100.0) (0.2) (1.4) (98.8,100.0) (96.4,98.3) adjusted for N/A N/A N/A N/A N/A 1,390 10 48 98.9 96.5 N/A N/A N/A N/A N/A 1,534 10 81 98.6 94.7 lead-time (59.5) (0.7) (3.5) (98.3,99.5) (95.2,97.9) (60.8) (0.7) (5.3) (97.9,99.3) (93.2,96.2) Net survival estimate at the time of previous event before 1st or 5th anniversary of diagnosis. Where no estimate is given (-) no event occurred in the first 12 months after diagnosis (1 year estimates) or between the third and fifth years after diagnosis (5 year estimates) Cases are excluded due to imputed follow-up being greater than observed follow-up (see text). Values are the mean of the 10 imputed data sets with the exception of * which is the mean of 8 estimates. Not adjusted for lead-time. Cancer Epidemiology 2410 International differences in breast cancer survival Figure 4. Net survival estimates for women aged 50–65 (mean age 53.7 years) diagnosed with breast cancer 1 January 1997–31 December 2006 and followed up to 31 December 2008. (a) by screening status, New South Wales, (b) by screening status, West Midlands, (c) screen- detected compared to non-screen-detected, New South Wales, (d) screen-detected compared to non-screen-detected, West Midlands. The final models were adjusted for age and extent of dis- Amongst the cohort of women we examined, an estimated ease at diagnosis. For screen-detected women all effects total of 236 deaths, 38.1% of those due to breast cancer, (excess hazard ratios of breast cancer death) were constant would have been avoided in the West Midlands had their over follow-up time and followed a log-linear form. The survival been the same as those diagnosed in New South effect of age upon survival amongst non-screen-detected Wales; 200 (40.2%) amongst non-screen-detected women and women was non-linear. The effect of both age and extent of 36 (29.5%) amongst those whose cancer was screen-detected disease were found to change over follow-up time amongst (Table 2). non-screen-detected women. The excess hazard of death from breast cancer within five years of diagnosis in the base- Discussion line model was 57% higher among women diagnosed in the Breast cancer survival for the women included in this study was West Midlands than women in New South Wales (95% CI significantly lower in West Midlands (UK) than New South 35%-80%, Table 2). The baseline (age-adjusted) disadvantage Wales (Australia), which is fully consistent with our previous 1,23–27 was slightly greater for women with non-screen-detected can- findings. Our results further show the extent and persist- cer (EHR 1.65, 95% CI 1.402 1.89) than for women whose ence of this difference amongst a cohort of peri-menopausal cancer had been screen-detected (EHR: 1.46, 95% CI women who were invited for screening in a mature, fully func- 0.732 2.20). After additional adjustment for extent of disease tioning population-based screening programme. these differentials increased (EHR 2.00, 95% CI 1.70-2.31 in the non-screen-detected and 1.72, 95% CI 0.872 2.56 for Survival differences screen-detected cancer). In the West Midlands, 5-year survival amongst women who Crude mortality due to breast cancer 5 years after diagno- had never attended for screening was 4.9% lower (absolute sis was correspondingly much higher in the West Midlands. difference) than amongst the never-attenders in New South V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Woods et al. 2411 Table 2. Numbers of deaths, excess hazard ratios of breast cancer death and estimates of avoidable mortality within five years of diagnosis: women aged 50–65 (mean age 53.7 years) diagnosed with invasive breast cancer 1 January 1997–31 December 2006 and followed up to 31 December 2008 in New South Wales (Australia) and the West Midlands (UK) Non-screen-detected Screen-detected New South West New South West Wales Midlands Wales Midlands Number of women Total T 4,061 (100.0) 3,104 (100.0) 2,335 (100.0) 2,524 (100.0) Excluded when correcting for E N/A N/A 945 (40.5) 990 (39.2) lead-time and over-diagnosis Included in analyses I5 T – E 4,061 (100.0) 3,104 (100.0) 1,390 (59.5) 1,534 (60.8) Excess Hazard Ratios (EHR) Overall EHR, adjusted 1.57 (1.35-1.80) only for age (95% CI) [NSW reference] Baseline EHR, adjusted only 1.00 1.65 (1.40-1.89) 1.00 1.46 (0.73-2.20) for age (95% CI) Screening-specific EHR, 1.00 2.00 (1.70-2.31) 1.00 1.72 (0.87-2.56) adjusted (95% CI) Avoidable mortality 5 years after diagnosis Crude mortality due to breast cancer (%) CM 9.5 16.0 5.6 7.9 Corresponding number of deaths due D 5 I * CM 388 496 77 121 actual to breast cancer If excess hazard of death due to breast cancer in West Midlands was equal to New South Wales Deaths due to breast cancer D 5 I *CM N/A 296 N/A 85 equal WM NSW Deaths due to breast cancer that D 5 D 2 D N/A 200 (40.2) N/A 36 (29.5) avoid actual equal could be avoided (% of deaths due to breast cancer) Wales. For women whose cancer was screen-detected, this women in West Midlands would have been avoided had their difference was 1.7% after adjustment for lead-time bias. survival been the same as the women in New South Wales. The 5-year adjusted excess hazard ratio of breast cancer death for the non-screened group indicates a substantial and Bias and artefact Taken together, our results suggest that differences in screen- significant survival disadvantage for West Midlands. This is striking because these estimates are adjusted for differences ing practice and extent of disease at diagnosis do not explain in age and extent of disease at diagnosis, and so one might the overall difference in survival between West Midlands and expect survival to be much more similar. Even among New South Wales for this age group, and that women with screen-detected women the survival disadvantage is distinct breast cancer in West Midlands have a higher risk of excess death from their cancer than women in New South Wales, which is particularly striking because these are women diag- nosed with asymptomatic cancers. Their tumours are pre- whether they are screened or not. dominantly localised, and as such they would almost all be treated surgically and with curative intent and have a high The role of ‘de facto’ screening chance of long-term survival. These differences in survival are likely to be in part due to Although the overall number of deaths is relatively modest the differences in the way screening is delivered in the West in this cohort of cancer patients, with only 9.1% of all Midlands and in New South Wales. In the UK, the National Health Service is free at the point of delivery for the whole women dying during follow-up, the impact of these differen- ces is important. The increased excess hazard of breast cancer population and private mammography is rare. In contrast, in death 5 years after diagnosis in the West Midlands is double Australia, mammography is obtained through BreastScreen that of New South Wales amongst non-screen-detected Australia but also through private radiology clinics. Mammo- women and 72% greater amongst those with a screen- grams conducted privately for diagnostic purposes, rather than in asymptomatic women, may be refunded via the Med- detected cancer. Overall we estimated that more than a third of the deaths attributable to breast cancer observed for icare Benefits Scheme (MBS). A substantial proportion of V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology 2412 International differences in breast cancer survival those conducted in private clinics is likely to constitute de diagnosed at an earlier stage, with better prognosis. However, facto screening, (regular diagnostic mammography not the distribution by extent of disease was similar in both recorded by BreastScreen Australia), but it is unknown to regions; the proportion of localised disease was in fact slightly what degree this occurs. This is likely to be the reason for higher in the West Midlands than in New South Wales the higher proportion of tumours in the West Midlands that (67.1% versus 64.1%, Table 1). A shorter screening interval were apparently screen-detected, despite a shorter screening will lead to detection of a greater number of slower-growing interval in New South Wales. It also implies that women in tumours, but also greater numbers of aggressive, faster- New South Wales whom we defined as ‘never-attenders’ growing tumours, which will also be identified at an earlier includes a sub-group of women who had, in fact, been stage than would otherwise be the case. In our data, the dis- screened outside of the national screening programme. This tribution of tumours by extent of disease amongst interval interpretation is supported by the observation that a signifi- cancers was fairly similar in both regions (localised tumours cantly larger proportion of these women classified as ‘never- representing 51.8% in New South Wales and 50.0% in West attenders’ in New South Wales were diagnosed with localised Midlands, Chi p value 0.07) This supports the interpretation tumours (50.8% compared to 46.6% in West Midlands). that the breast cancer survival differences between New South Although it is probable that we incorrectly allocated some Wales and West Midlands cannot be fully explained by the women to the never-attender group who were actually screen- shorter screening interval in New South Wales. detected, especially in New South Wales, information on their We have made adjustment for lead-time and over- personal characteristics and the features of their cancer would diagnosis in our analysis, and demonstrated that the survival not have been compromised since these data items were col- differences observed are robust to these biases. Adjustment lected from the Cancer Registry, rather than via the screening involved a ten-fold simulation where both the individual sur- service. It is possible, however, that this may have biased our vival times were shortened and the number of women estimates of net survival. We therefore performed a sensitivity included in the cohort was reduced. On average, the survival analysis to examine the potential for de facto screening to time of screen-detected women was reduced by 1.5 years and explain the difference in survival for the non-screen-detected 40% were excluded (Table 2). This latter proportion does not group. We randomly reallocated women in New South Wales represent the percentage of tumours over-diagnosed, but with localised disease from the non-screen-detected to the rather the probability that a screen-detected cancer would screen-detected group, for selected proportions ranging from not have been detected symptomatically during the period of 1% to 95%, and then re-estimated the net survival function. time between the actual date of diagnosis and 31st December Over 100 iterations the five-year net survival estimates for the 2006 (the mean of which was 3.4 years). The number of non-screen-detected group in New South Wales became simi- tumours over-diagnosed might be reasonably obtained by lar to those for West Midlands only when an implausible 90% estimating the probability that the cancer would not have of the localised cancers (43% of all non-screen-detected can- been detected symptomatically during the woman’s remain- cers, c. 1800 women) were reallocated (data not shown). This ing expected life time (the mean of which was 30.6 years). level of reallocation would require that the true proportion of cancers screen-detected in New South Wales in the cohort Other non-causal explanations was in excess of 64%, in comparison to the 36.5% actually We have previously summarised the possible non-causal observed (and the 44.8% observed in West Midlands, that explanations for this difference. The first of these that may probably reflects the order of magnitude one might expect for be applicable here is the possibility that the NSWCCR Regis- New South Wales, since private mammography is very rare in try more often fails to link a woman’s death to the record of West Midlands). For smaller, but substantial proportions of her cancer registration than the WMNCRS, leading to appa- reallocation the reduction in the survival difference was rela- rently inflated survival. This explanation is very unlikely to tively small. We thus consider it very unlikely that de facto apply in this younger age group during this period of time – screening can fully explain the difference in survival between these are young women among whom death is a relatively non-screen-detected women. This analysis also served to illus- rare event, and who were followed up during a period of reli- trate the robust nature of the difference for the screen- able death registration. The second possible explanation is detected group: there remained a survival advantage for New that a higher proportion of in situ tumours registered in New South Wales, albeit very small, even when 95% of localised South Wales were misclassified as invasive breast cancers (apparently symptomatic) tumours were reallocated to the than in the West Midlands. Again, we do not consider that screen-detected group. this could be an explanation for the differences observed in this study, since >99% of the tumours analysed were micro- Screening-specific biases scopically verified and in situ cancers were excluded. We We may consider whether the longer screening interval in have also previously considered the accuracy and consistency the West Midlands compared with New South Wales (3 years of date of diagnosis as a potential mechanism by which sur- versus 2 years) might contribute to these differences. Screen- vival in New South Wales might be extended relative to West detected cancers in New South Wales could perhaps be Midlands. Again, however, this explanation has very little V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Woods et al. 2413 credibility here, since the date of diagnosis is established in treated with similar effectiveness. We consider a much more the same manner for both screen-detected and non-screen- likely explanation for our findings is that the tumours them- detected women. selves are not substantially different, but that something dif- ferent happens once the woman is diagnosed. This is of Potential Explanations greater concern: we have examined a group of young women Examining possible explanations operating before diagnosis, with predominantly localised disease. Almost all of them the differences in breast cancer survival between West Mid- would have had treatment with curative intent. lands and New South Wales could arise from (a) greater Treatment may vary with comorbidity, and is subject to delays in diagnosis in West Midlands, (b) longer waiting patient compliance, but there is no particular reason to times for hospital consultation for non-screen-detected can- assume that these would persistently differ between New cers or (c) less effective screening in West Midlands than in South Wales and West Midlands, particularly in this age New South Wales. The fact that differences persist after group. The alternative explanation is that the treatment in adjustment for extent of disease at diagnosis does not support West Midlands is not as effective as those in New South any of these explanations, however. It is theoretically possible Wales leading to poorer stage-specific survival. that residual confounding may partially account for this lack of explanatory power. Residual confounding may have arisen Conclusions due to the fact that the screening interval for women in West Although overall survival was high for this cohort of women, Midlands is longer than in New South Wales, combined with our data suggest that more than one in three breast cancer a tendency for the accuracy of the ‘extent of disease’ variable deaths within five years of diagnosis in the West Midlands for non-screen-detected cancers to be lower in West Mid- would be avoidable if five-year survival were the same as in lands. Together, this would imply that within each stage New South Wales. The women we analysed here are rela- grouping, the true (unknown) stage of disease is more advanced in West Midlands than in New South Wales (stage tively young. They are therefore less likely to be suffering migration). This would have the effect of better extent- from other serious illnesses and more likely to be economi- adjusted survival in New South Wales. Although this expla- cally and socially active. In order to improve the prognosis nation is possible, we consider that in the context of this for women diagnosed with breast cancer in the UK during study it is not very likely. This is because two mechanisms their early 50s it is essential that we understand better the would both need to apply: delays for non-screen-detected mechanisms that underlie these international differences in cancers matched with less effective screening for screen- stage-adjusted survival. Differences in the effectiveness of detected cancers leading to differences in extent of a similar treatment are an important possibility and they now deserve magnitude in both groups. At the very least, the fact that to be examined with great care. It is not possible to dismiss there is a significant difference in survival amongst women differences in breast cancer survival between the UK and with screen-detected disease in the two regions refutes the other countries such as Australia as artefactual. hypothesis that international differences are entirely due to practitioner delay in referral (since all these women were Acknowledgements diagnosed through routine screening) or differences in The authors gratefully acknowledge the staff of the West Midlands Office of patient delay in seeking medical diagnosis following the the English National Cancer Registration Service, the New South Wales Cen- detection of breast cancer symptoms. tral Cancer Registry, and Professor Richard Taylor (BreastScreen New South These findings thus tend to refute the idea that breast Wales) for their assistance in obtaining and analysing these data. We also cancers in West Midlands and New South Wales are very wish to thank Dr Ula Nur and Dr Melanie Morris (Cancer Research UK different at the point of diagnosis, but are subsequently Cancer Survival Group) for helpful comments on the manuscript. References 1. Woods LM, Rachet B, O’Connell D, et al. Large 5. Brekelmans CT, Peeters PH, Deurenberg JJ, et al. cancer detection methods. Australian New Zea- differences in patterns of breast cancer survival Survival in interval breast cancer in the DOM land J Stat 2013;55:351–67. between Australia and England: a comparative screening programme. Eur J Cancer A 1995;31: 9. Sasieni P. Evaluation of the UK breast screening study using cancer registry data. Int J Cancer 1830–5. programmes. Annals Oncol 2003;14:1206–8. 2009;124:2391–9. 6. Olivotto IA, Mates D, Kan L, et al. Prognosis, 10. Zackrisson S, Andersson I, Janzon L, et al. Rate of over-diagnosis of breast cancer 15 years after 2. Woods LM, Rachet B, Coleman MP. Origins of treatment and recurrence of breast cancer for socio-economic inequalities in cancer survival: a women attending or not attending the Screening end of Malmo mammographic screening trial: review. Ann Oncol 2006;17:5–19. Mammography Program of British Columbia. follow-up study. Br Med J 2006;332:689–92. 3. Nystrom L, Rutqvist LE, Wall S, et al. Breast cancer Breast Cancer Res Treat 1999;54:73–81. 11. Duffy SW, Nagtegaal ID, Wallis M, et al. Correct- screening with mammography: overview of Swed- 7. Joensuu H, Lehtimaki T, Holli K, et al. Risk for ing for lead time and length bias in estimating ish randomised trials. Lancet 1993;341:973–8. distant recurrence of breast cancer detected by the effect of screen detection on cancer survival. 4. Collins S, Woodman CBJ, Threlfall A, et al. Sur- mammography screening or other methods. Am J Epidemiol 2008; 168:98–104. vival rates from interval cancer in NHS JAMA 2004;292:1064–73. 12. Census 2011. Office for National Statistics, 2013. breast screening programme. Br Med J 1998;316: 8. Hsieh JC-F, Cramb SM, McGree JM, et al. Bayes- 13. Australian census 2011. Australian Bureau of 832–3. ian spatial analysis for the evaluation of breast Statistics, 2014. V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology 2414 International differences in breast cancer survival 14. Pohar-Perme M, Stare J, Este`ve J. On Estimation other causes: a crude analogue of relative survival. 25. Allemani C, Weir HK, Carreira H, et al. Global in Relative Survival. Biometrics 2011. Stat Med 2000;19:1729–40. surveillance of cancer survival 1995-2009: analysis 15. StataCorp. STATA statistical software, ed. 12.1 of individual data for 25 676 887 patients from 21. Lambert PC, Dickman PW, Nelson CP, et al. College Station, Texas: Stata Corporation, 2011. 279 population-based registries in 67 countries Estimating the crude probability of death due to 16. Danieli C, Remontet L, Bossard N, et al. (CONCORD-2). Lancet 2014;385:977–1010. cancer and other causes using relative survival Estimating net survival: the importance of allow- 26. Coleman MP, Quaresma M, Berrino F, et al. models. Stat Med 2010;29:885–95. ing for informative censoring. StatMed 2012;31: Cancer survival in five continents: a worldwide 22. Abdel-Rahman M, Stockton D, Rachet B, et al. 775–86. population-based study (CONCORD). Lancet What if cancer survival in Britain were the same 17. Roche L, Danieli C, Belot A, et al. Cancer net Oncol 2008;9:730–56. as in Europe: how many deaths are avoidable? Br survival on registry data: use of the new unbiased 27. Yu XQ, O’Connell DL, Forman D. Comparison J Cancer 2009;101:S115–S24. Pohar-Perme estimator and magnitude of the of cancer survival in UK and Australia: rates are 23. Coleman MP, Forman D, Bryant H, et al. Cancer bias with the classical methods. Int J Cancer higher in Australia for three major sites. Br J survival in Australia, Canada, Denmark, Norway, 2013;132:2359–69. Cancer 2004;91:1663–5. Sweden, and the UK, 1995-2007 (the Interna- 18. Rubin DB. Multiple imputation for non-response 28. Morrell S, Taylor R, Roder D, et al. Mammog- tional Cancer Benchmarking Partnership): an in surveysed. New York: Wiley, 1987. raphy screening and breast cancer mortality in analysis of population-based cancer registry data. 19. Royston P, Parmar MK. Flexible parametric Australia: an aggregate cohort study. J Med Lancet 2011;377:127–38. proportional-hazards and proportional-odds Screen 2012;19:26–34. 24. Walters S, Maringe C, Butler J, et al. Breast can- models for censored survival data, with applica- 29. Feinstein AR, Sosin DM, Wells CK. The Will cer survival and stage at diagnosis in Australia, tion to prognostic modelling and estimation of Rogers phenomenon: stage migration and new treatment effects. StatMed 2002;21:2175–97. Canada, Denmark, Norway, Sweden and the UK, diagnostic techniques as a source of misleading 20. Cronin KA, Feuer EJ. Cumulative cause-specific 2000-2007: a population-based study. Br J Cancer statistics for survival in cancer. N Engl J Med mortality for cancer patients in the presence of 2013;108:1195–208. 1985;312:1604–8. V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png International Journal of Cancer Wiley

Are international differences in breast cancer survival between A ustralia and the UK present amongst both screen‐detected women and non‐screen‐detected women? survival estimates for women diagnosed in W est M idlands and N ew S outh W ales 1997–2006

Download PDF
11 pages

Loading next page...
 
/lp/wiley/are-international-differences-in-breast-cancer-survival-between-a-JrLAVGigDh

References (42)

Publisher
Wiley
Copyright
© 2016 UICC
ISSN
0020-7136
eISSN
1097-0215
DOI
10.1002/ijc.29984
pmid
26756306
Publisher site
See Article on Publisher Site

Abstract

IJC International Journal of Cancer Are international differences in breast cancer survival between Australia and the UK present amongst both screen-detected women and non-screen-detected women? Survival estimates for women diagnosed in West Midlands and New South Wales 1997–2006 1 1 2 3 1 Laura M. Woods , Bernard Rachet , Dianne L. O’Connell , Gill Lawrence and Michel P. Coleman Cancer Research UK Cancer Survival Group, Department of Non-Communicable Disease Epidemiology, London School of Hygiene and Tropical Medicine, London, WC1E 7HT Cancer Research Division, Cancer Council NSW, NSW 1340, Australia Breast Cancer Audit Consultant and Former Director, West Midlands Cancer Intelligence Unit, Public Health Building, University of Birmingham, Birmingham, B15 2TT We examined survival in screened-detected and non-screen-detected women diagnosed in the West Midlands (UK) and New South Wales (Australia) in order to evaluate whether international differences in survival are related to early diagnosis, or to other factors relating to the healthcare women receive. Data for women aged 502 65 years who had been eligible for screen- ing from 50 years were examined. Data for 5,628 women in West Midlands and 6,396 women in New South Wales were linked to screening service records (mean age at diagnosis 53.7 years). We estimated net survival and modelled the excess hazard ratio of breast cancer death by screening status. Survival was lower for women in the West Midlands than in New South Wales (5-year net survival 90.9% [95% CI 89.9%291.7%] compared with 93.4% [95% CI 92.6%-94.1%], respectively). The difference was greater between the two populations of non-screen-detected women (4.9%) compared to between screen-detected women, (1.8% after adjustment for lead-time and over-diagnosis). The adjusted excess hazard ratio of breast cancer death for West Midlands compared with New South Wales was greater in the non-screen-detected group (EHR 2.00, 95% CI 1.702 2.31) but not significantly different to that for women whose cancer had been screen-detected (EHR 1.72, 95% CI 0.872 2.56). In this study more than one in three breast cancer deaths in the West Midlands would have been avoided if survival had been the same as in New South Wales. The possibility that women in the UK receive poorer treatment is an important potential explanation which should be examined with care. We have previously shown a difference of 6% in 5-year breast tors or to other factors relating to the healthcare women receive. We have previously identified these as possible explan- cancer survival between Australia and England for women in ations for socioeconomic differences, but they also may the target age group for screening and diagnosed during the explain international variations in survival. period 19962 1999. Examining survival by screening status The trials that led to the implementation of mammo- has the potential to shed further light on whether international graphic screening worldwide were evaluated by examining differences are more likely to be due to tumour or patient fac- the reduction in breast cancer mortality amongst the popula- tions of women screened. In this context, a reduction in the Key words: breast cancer, net survival, excess mortality, UK, Aus- number of breast cancer deaths in the screened population tralia, New South Wales, West Midlands, cancer screening, can be interpreted as the number of cancer deaths avoided or mammography deferred by the intervention. This outcome is helpful in eval- This is an open access article under the terms of the Creative Commons uating the public health impact and economic value of the Attribution License, which permits use, distribution and reproduction screening programme as a whole. in any medium, provided the original work is properly cited. Other studies have examined the impact of mammo- DOI: 10.1002/ijc.29984 History: Received 12 Aug 2015; Accepted 30 Nov 2015; Online 12 graphic screening upon individual patient survival. Analyses 4,5 Jan 2016 of survival include examinations of interval cancers (can- Correspondence to: Laura M. Woods, Cancer Research UK Cancer cers diagnosed following a normal mammogram but prior to Survival Group, Department of Non-Communicable Disease Epide- the next screening invitation), comparisons of women in miology, London School of Hygiene an Tropical Medicine, Keppel dichotomous groups (attenders vs. never-attenders and those Street, London, WC1E 7HT, Tel: 144 20 7612 7849, Fax: 144 20 with screen-detected vs. non-screen-detected cancers ) and 8 9 7436 4230, E-mail: [email protected] spatial analyses. A review conducted in the UK in 2003 V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Woods et al. 2405 What’s new? Breast cancer patients in Australia are known to have higher survival than those in the UK. In this study, the authors found that these international differences in survival persist in both screen-detected and non-screen-detected groups, even after adjustment for both lead-time bias and over-diagnosis. These results suggest that it is essential that the mechanisms underly- ing these differences be understood, including potential variations in effective treatment between the two regions. concluded that a better understanding of the effect of screen- screening programme was being established and expanded. detection required more detailed data. In particular, the We excluded women aged over 65 years at diagnosis because review identified the importance of linkage of mortality data the target age for screening was up to age 64 years in the UK to screening invitations so that the outcome for tumours during this period. The eligibility criteria resulted in a cohort diagnosed after the introduction of screening might be which built up over time (median month of diagnosis August examined. 2003 in West Midlands and November 2003 in New South Examination of survival by screening status enables us to Wales). All women were followed up to 31 December 2008 establish, at the population level, the survival benefit afforded (at least 2 years following diagnosis). Data were obtained to women whose cancers were screen-detected compared to from the West Midlands Office of the English National Can- women whose cancers were detected symptomatically. The cer Registration Service (WMNCRS, England) and the New disadvantage of this approach is that it is susceptible to lead- South Wales Central Cancer Registry (NSWCCR, Australia). time bias and to over-diagnosis. Lead-time is the additional These two registries cover populations of 5.6 and 6.9 million, 12,13 observation time credited to women who are screen-detected respectively. by virtue of the fact that they are asymptomatic. Breast Information was obtained from each cancer registry on tumours considered to be “over-diagnosed” are those detected each woman’s age at diagnosis (completed years), the month by screening mammography but which would not have been and year of their diagnosis and death (if dead), the sub-site, diagnosed during the patient’s lifetime in the absence of grade, histology and behaviour of the tumour, and all infor- screening. These biases together lead to apparently better mation pertaining to the extent of disease at diagnosis survival, even if the actual time of death is not deferred. This (stage). Staging information for cases in the West Midlands skews estimates of survival in favour of screening, resulting was recoded according to the rules used by the New South in statistics which appear to show a survival advantage Wales Central Cancer Registry: localised (confined to the amongst women who have been screened, even when none organ of origin), regional (spread to adjacent muscle, organ, might exist. Recently, methodological advances have been fat, connective tissue or regional lymph nodes), distant (dis- made into ways to account for lead-time bias in the analysis tant metastasis) and unknown stage. of survival so that the underlying differences in survival can The cancer registry data were linked to the population- be assessed. This involves correcting the observed survival based mammographic screening service records in each local- time to account for the additional follow-up observed in the ity to establish each woman’s screening status at diagnosis (the cohort as a result of screen-detection. National Health Service Breast Screening Programme for the In this article, we examine net survival for breast cancer in West Midlands and BreastScreen NSW for New South Wales). screen-detected and non-screen-detected women diagnosed in We defined four categories for the screening status at diagno- the West Midlands (UK) and New South Wales (Australia), sis: (1) women whose cancer was detected at a routine screen, applying a correction for lead-time bias and over-diagnosis. We (2) women who presented with cancer following a negative use the results to discuss the extent to which the international screen but before being invited to their next routine screen differences in breast cancer survival between the UK and Aus- (interval cancers), (3) women who presented with cancer after tralia may be explained by tumour or patient factors or to other at least one negative screen but who had not attended their factors relating to the healthcare women receive. most recent appointment (lapsed attenders), and (4) women who presented with cancer who had never attended screening. Materials We also compared women in the screen-detected group (Cate- The cohort of interest consisted of women who were invited gory 1) to all those with non-screen-detected cancer (Catego- to attend for screening mammography in a fully-functioning, ries 2, 3 and 4). This broadly corresponded to comparing mature screening programme during a defined calendar those with asymptomatic disease identified via routine screen- period. Women diagnosed with a primary invasive breast ing to women presenting with symptomatic disease. cancer at ages 50–65 years during the period 1 January 1997 to 31 December 2006 and aged 51 years or younger on 1 Jan- Methods uary 1997 were considered eligible (Fig. 1). We thus excluded Net survival estimation women who were first invited to be screened at ages over 50 Net survival is defined as the survival from the disease of years, as well as women invited during the years when the interest. It is derived by adjusting the overall survival in the V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology 2406 International differences in breast cancer survival Figure 1. (a) Schematic diagram of women eligible for the study alongside (b) a histogram showing the total number of women included in New South Wales (Australia) and the West Midlands (UK) by year of diagnosis (1997–2006). patient group for their expected survival in the absence of observed survival time in order to obtain corrected survival the disease. We estimated net survival using the non- time (Fig. 2, Patients A and B). parametric Pohar-Perme estimator, which has been imple- We considered tumours to be over-diagnosed if they mented in Stata. The Pohar-Perme estimator is an unbiased would not have been detected symptomatically during the estimator of net survival with respect to informative censor- study period or during the predicted lifetime of the patient. To account for over-diagnosis we excluded tumours in ing (defined as the tendency for the estimates to reflect the survival of patients with lowest expected mortality as time instances where the value of E(s) , E(s) .. . E(s) exceeded 1 2 10 16,17 since diagnosis increases) for population-based data. the woman’s actual observed survival time, either because the We estimated expected survival from region-specific life predicted date of diagnosis was after 31st December 2008, or tables provided by the Office for National Statistics for Eng- before her death. (Fig. 2, Patients C and D). land and Wales and the Australian Bureau of Statistics for We used the corrected survival times to estimate non- each calendar year of follow-up. parametric net survival for each of these ten separate data sets for the screen-detected group. We used the rules estab- lished by Rubin for the re-combination of estimates in a Adjustment for lead-time and over-diagnosis multiple-imputation setting to derive an overall estimate of To account for the potential effect of lead-time bias, we cal- net survival and its variance, adjusted for lead-time bias and culated additional survival time due to screening, E(s), for over-diagnosis (Fig. 3a). the screen-detected group, as proposed by Duffy et al. and assuming a mean sojourn time (time from carcinogenesis to symptomatic cancer in the absence of screening) in both Missing data regions of 4 years. We applied 10 separate simulations to Data on extent of disease were missing for 8.9% of women obtain a range of possible values, E(s) , E(s) .. . E(s) ,by diagnosed in West Midlands and 5.3% of those diagnosed in 1 2 10 assuming that survival times were exponentially distributed New South Wales. We used a 10-fold hot-deck approach to with a mean of E(s). Values of E(s) were subtracted from take account of these missing values for extent of disease. V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Woods et al. 2407 Figure 2. Schematic diagram demonstrating the exclusion of women in order to adjust for lead-time bias and over-diagnosis. The hot-deck approach involves identifying ‘donor groups’ indicate a better fit. We examined non-linearity of age by the for each woman with missing information on extent of dis- inclusion of restricted cubic splines and tested for time- ease. The donor group for each woman comprised women varying effects for region, age at diagnosis, and extent of dis- diagnosed in the same period (1997–2000, 2001–2006) and ease. We examined interactions between region and age, and region (West Midlands, New South Wales), at a similar age between region and extent of disease. (2 groups: 50–53 years [prevalent screening round], 541 For the screen-detected group, we applied the model with years [incident screening rounds]) who had been followed for the smallest number of parameters to each unique combina- a similar amount of time (6 groups: up to 1 year, 1-1.9 years, tion of E(s) and extent (100 separate combinations of i j 2-3.9 years, 4-5.9 years, 6-7.9 years and 81 years), and with results). For non-screen-detected women we refitted the the same vital status at the end of follow-up (dead, alive), model found to fit best to using values of extent to the data and screening status (screen-detected, not screen-detected). for extent , extent ... extent (10 sets of results). 2 3 10 For each woman with missing data, ten separate values of We predicted from the final models estimates of crude extent of disease (extent , extent ... extent ) were obtained mortality due to breast cancer and crude mortality due to 1 2 10 by randomly and independently selecting values of extent of other causes for the whole cohort. Crude mortality can be disease from the donor group. derived directly from the net survival models, and allows Combining these two procedures resulted in data sets with the mortality observed during follow-up to be partitioned a set of 10 imputed values for the variable extent for both into mortality due to the cancer itself and due to other the screen-detected group and non-screen-detected group, causes. Estimates of crude mortality were derived for each of and a set of 10 imputed values for the variable E(s) for the the covariate patterns in the sample and a weighted average screen-detected only (where i5 1–10). of deaths due to breast cancer across all patterns was calcu- lated by region and screening. Estimates were derived sepa- Modelling rately for screen-detected women and non-screen-detected We fitted flexible non-parametric regression models for net women in West Midlands and New South Wales. 19 18 survival to estimate the excess hazard ratio associated with We used Rubin’s rules to re-combine the 100 separate being diagnosed with breast cancer in the West Midlands estimates of the excess hazard ratio of breast cancer death compared to New South Wales. We fitted 10 models for and crude mortality from breast cancer for screen-detected women with screen-detected cancer using the values E(s) to women and the 10 separate estimates for non-screen-detected E(s) combined with extent and one model for the non- women. This resulted in separate estimates for screen- 10 1 screen-detected cancer using observed survival times and val- detected and non-screen-detected women of the relative ues of extent . A priori, we included age at diagnosis, region change in the excess hazard of death due to breast cancer for and extent of disease in the models. We used a reduction of women living in West Midlands compared to women in New 3 or more in the AIC (Akaike Information Criterion) to South Wales, as well as the crude probability of death from V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology 2408 International differences in breast cancer survival Figure 3. Schematic diagram illustrating (a) net survival estimation correcting for lead-time bias and over-diagnosis and (b) the modelling strategy taking into account missing values for extent of disease. breast cancer and other causes, and their associated variances. in New South Wales). Fewer than one in ten women died These final estimates took into account lead-time bias and during follow-up: 10.8% in West Midlands and 7.6% in New over-diagnosis in the screen-detected group and were also South Wales. Overall, net survival in the cohort was high (Table 1). adjusted for age and extent of disease at diagnosis (Fig. 3b). The estimates of crude mortality were used to establish Consistent with our previous findings, net survival overall the number of cancer deaths that could have been avoided in was significantly lower in the West Midlands than in New the hypothetical situation in which survival was equalised South Wales (5-year net survival 90.9% [95% CI 89.9%291.7%] and 93.4% [95% CI 92.6%294.1%], respec- between the two regions. This provides an estimate of the public health impact of survival differentials in the net sur- tively). Women diagnosed with interval cancers in New vival setting. South Wales had lower survival than screen-detected women (5-year net survival 93.5% compared to 98.5%), but better Results survival than women who had never attended screening We analysed data for 5,628 women in West Midlands (98.5% (89.5%) and those who had attended previously but lapsed in of those eligible, mean age at diagnosis 53.7 years) and 6,396 attendance prior to diagnosis (86.8%; Table 1, Fig. 4a). In women in New South Wales (99.9% of those eligible, mean West Midlands, however, the survival of women diagnosed age at diagnosis 53.8 years). Those excluded were the very with interval cancers was not dissimilar to that of lapsed small number of women who were known to the registry attenders, whilst those who had never attended had the worst only because breast cancer had been mentioned on their survival (Table 1, Fig. 4b). The difference in net survival death certificate (DCOs) or because the sequence of dates between West Midlands and New South Wales was greater provided was illogical. The proportion of tumours that were among non-screen-detected women (4.9% five years after screen-detected was greater in West Midlands (44.8% com- diagnosis) than among screen-detected women in the two pared to 36.5%, Table 1). The majority of women were diag- regions (1.8%; 1.0% before adjustment for lead-time bias, nosed with localised disease, (54.1% in West Midlands, 53.9% Table 1). V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Table 1. Net survival estimates at 1 and 5 years after diagnosis by mode of presentation and extent of disease at diagnosis: women aged 50–65 (mean age 53.7 years) diagnosed with invasive breast cancer 1 Jan- uary 1997–31 December 2006 and followed up to 31 December 2008 in New South Wales (Australia) and the West Midlands (UK) New South Wales West Midlands Deaths (% of N) within Net Survival, % (CI) Deaths (% of N) within Net Survival , % (CI) (a) Mode of presentation N (%) 1 year 5 years 1-year 5-year N (%) 1 year 5 years 1-year 5-year Screen-detected 2,335 (36.5) 11 (0.5) 54 (2.3) 99.8 (99.2,99.9) 98.5 (97.5,99.1) 2,524 (44.8) 11 (0.2) 90 (1.4) 99.9 (98.8,100.0) 97.5 (96.4,98.3) adjusted for lead-time 1,390 (21.7) 10 (0.7) 48 (3.5) 98.9 (98.3,99.5) 96.5 (95.2,97.9) 1,534 (27.3) 10 (0.7) 81 (5.3) 98.6 (97.9,99.3) 94.7 (93.2,96.2) Lapsed-attender 129 (2.0) 4 (3.1) 16 (12.4) 97.2 (92.0,99.0) 86.8 (78.2,92.2) 175 (3.1) 6 (0.1) 17 (0.3) 96.9 (92.7,98.7) 89.8 (82.6,94.2) Interval cancer 1,028 (16.1) 4 (0.4) 64 (6.2) 99.8 (98.4,100.0) 93.5 (91.3,95.2) 1,537 (27.3) 34 (0.5) 157 (2.5) 98.1 (97.2,98.7) 90.3 (88.4,92.0) Never-attender 2,904 (45.4) 86 (3.0) 297 (10.2) 97.3 (96.6,97.8) 89.5 (88.1,90.7) 1,392 (24.7) 97 (1.5) 280 (4.4) 93.3 (91.8,94.5) 79.8 (77.4,82.0) All groups 6,396 (100.0) 105 (1.6) 431 (6.7) 98.6 (98.3,98.9) 93.4 (92.6,94.1) 5,628 (100.0) 148 (2.3) 544 (8.5) 97.7 (97.2,98.1) 90.9 (89.9,91.7) Non-screen-detected Screen-detected Non-screen-detected Screen-detected Deaths (% of N) Net Survival, Deaths (% of N) Net Survival, Deaths (% of N) Net Survival, Deaths (% of N) Net Survival, (b) Extent of within % (CI) within % (CI) within % (CI) within % (CI) disease at diagnosis N (%) 1 year 5 years 1-year 5-year N (%) 1 year 5 years 1-year 5-year N (%) 1 year 5 years 1-year 5-year N (%) 1 year 5 years 1-year 5-year Localised 1,955 7 71 99.9 96.9 1,490 3 16 99.9 99.9 1,351 6 59 99.9 96.8 1,702 4 36 100.0 99.1 (48.1) (0.1) (4.1) (98.8,100.0) (95.7,97.8) (64.1) (0.1) (1.1) (99.0,100.0) (12.5,100.0) (44.1) (0.1) (4.4) (98.4,100.0) (95.3,97.9) (67.1) (0.1) (2.1) (100.0,116.3) (97.4,99.7) adjusted for N/A N/A N/A N/A N/A 875 2 10 99.7 99.2* N/A N/A N/A N/A N/A 1,008 3 27 99.4 97.3 lead-time (37.1) (0.1) (1.1) (99.1,100.2) (98.0,100.5) (40.1) (0.1) (3.1) (98.8,100.0) (95.7,98.8) Regional 1,644 19 162 99.1 89.5 693 7 28 99.2 96.3 1,319 39 241 97.3 80.9 634 6 49 99.4 92.6 (40.1) (1.1) (10.1) (98.4,99.5) (87.6,91.1) (30.1) (1.1) (4.1) (98.0,99.7) (93.9,97.8) (42.1) (3.1) (18.3) (96.3,98.1) (78.3,83.2) (25.1) (1.1) (8.1) (97.9,99.8) (89.6,94.8) Distant 224 56 99 75.1 51.2 52 1 6 98.3 - 115 54 83 53.1 19.5 9 0 1 - 89.7 (6.1) (3.1) (6.1) (68.9,80.3) (43.6,58.3) (2.1) (2.1) (12.1) (85.8,99.8) (4.1) (47.1) (72.2) (43.6,61.7) (11.1,29.6) (0.1) (0.0) (0.1) (44.5,98.6) Unknown 238 12 45 95.2 79.9 100 0 4 - 96.4 319 38 71 88.3 77.7 179 1 4 99.7 99.3 (6.1) (1.1) (3.1) (91.5,97.3) (73.5,85.0) (4.1) (0.0) (4.1) (87.0,99.0) (10.1) (12.1) (22.3) (84.2,91.4) (72.2,82.2) (7.1) (1.1) (2.1) (84.8,100.0) (81.9,100.0) All stages 4,061 94 377 97.9 90.4 2,335 11 54 99.8 98.5 3,104 137 454 95.9 85.5 2,524 11 90 99.9 97.5 (100.0) (2.1) (9.1) (97.4,98.3) (89.3,91.5) (100.0) (0.5) (2.3) (99.2,99.9) (97.5,99.1) (100.0) (4.1) (14.6) (95.1,96.6) (84.1,86.9) (100.0) (0.2) (1.4) (98.8,100.0) (96.4,98.3) adjusted for N/A N/A N/A N/A N/A 1,390 10 48 98.9 96.5 N/A N/A N/A N/A N/A 1,534 10 81 98.6 94.7 lead-time (59.5) (0.7) (3.5) (98.3,99.5) (95.2,97.9) (60.8) (0.7) (5.3) (97.9,99.3) (93.2,96.2) Net survival estimate at the time of previous event before 1st or 5th anniversary of diagnosis. Where no estimate is given (-) no event occurred in the first 12 months after diagnosis (1 year estimates) or between the third and fifth years after diagnosis (5 year estimates) Cases are excluded due to imputed follow-up being greater than observed follow-up (see text). Values are the mean of the 10 imputed data sets with the exception of * which is the mean of 8 estimates. Not adjusted for lead-time. Cancer Epidemiology 2410 International differences in breast cancer survival Figure 4. Net survival estimates for women aged 50–65 (mean age 53.7 years) diagnosed with breast cancer 1 January 1997–31 December 2006 and followed up to 31 December 2008. (a) by screening status, New South Wales, (b) by screening status, West Midlands, (c) screen- detected compared to non-screen-detected, New South Wales, (d) screen-detected compared to non-screen-detected, West Midlands. The final models were adjusted for age and extent of dis- Amongst the cohort of women we examined, an estimated ease at diagnosis. For screen-detected women all effects total of 236 deaths, 38.1% of those due to breast cancer, (excess hazard ratios of breast cancer death) were constant would have been avoided in the West Midlands had their over follow-up time and followed a log-linear form. The survival been the same as those diagnosed in New South effect of age upon survival amongst non-screen-detected Wales; 200 (40.2%) amongst non-screen-detected women and women was non-linear. The effect of both age and extent of 36 (29.5%) amongst those whose cancer was screen-detected disease were found to change over follow-up time amongst (Table 2). non-screen-detected women. The excess hazard of death from breast cancer within five years of diagnosis in the base- Discussion line model was 57% higher among women diagnosed in the Breast cancer survival for the women included in this study was West Midlands than women in New South Wales (95% CI significantly lower in West Midlands (UK) than New South 35%-80%, Table 2). The baseline (age-adjusted) disadvantage Wales (Australia), which is fully consistent with our previous 1,23–27 was slightly greater for women with non-screen-detected can- findings. Our results further show the extent and persist- cer (EHR 1.65, 95% CI 1.402 1.89) than for women whose ence of this difference amongst a cohort of peri-menopausal cancer had been screen-detected (EHR: 1.46, 95% CI women who were invited for screening in a mature, fully func- 0.732 2.20). After additional adjustment for extent of disease tioning population-based screening programme. these differentials increased (EHR 2.00, 95% CI 1.70-2.31 in the non-screen-detected and 1.72, 95% CI 0.872 2.56 for Survival differences screen-detected cancer). In the West Midlands, 5-year survival amongst women who Crude mortality due to breast cancer 5 years after diagno- had never attended for screening was 4.9% lower (absolute sis was correspondingly much higher in the West Midlands. difference) than amongst the never-attenders in New South V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Woods et al. 2411 Table 2. Numbers of deaths, excess hazard ratios of breast cancer death and estimates of avoidable mortality within five years of diagnosis: women aged 50–65 (mean age 53.7 years) diagnosed with invasive breast cancer 1 January 1997–31 December 2006 and followed up to 31 December 2008 in New South Wales (Australia) and the West Midlands (UK) Non-screen-detected Screen-detected New South West New South West Wales Midlands Wales Midlands Number of women Total T 4,061 (100.0) 3,104 (100.0) 2,335 (100.0) 2,524 (100.0) Excluded when correcting for E N/A N/A 945 (40.5) 990 (39.2) lead-time and over-diagnosis Included in analyses I5 T – E 4,061 (100.0) 3,104 (100.0) 1,390 (59.5) 1,534 (60.8) Excess Hazard Ratios (EHR) Overall EHR, adjusted 1.57 (1.35-1.80) only for age (95% CI) [NSW reference] Baseline EHR, adjusted only 1.00 1.65 (1.40-1.89) 1.00 1.46 (0.73-2.20) for age (95% CI) Screening-specific EHR, 1.00 2.00 (1.70-2.31) 1.00 1.72 (0.87-2.56) adjusted (95% CI) Avoidable mortality 5 years after diagnosis Crude mortality due to breast cancer (%) CM 9.5 16.0 5.6 7.9 Corresponding number of deaths due D 5 I * CM 388 496 77 121 actual to breast cancer If excess hazard of death due to breast cancer in West Midlands was equal to New South Wales Deaths due to breast cancer D 5 I *CM N/A 296 N/A 85 equal WM NSW Deaths due to breast cancer that D 5 D 2 D N/A 200 (40.2) N/A 36 (29.5) avoid actual equal could be avoided (% of deaths due to breast cancer) Wales. For women whose cancer was screen-detected, this women in West Midlands would have been avoided had their difference was 1.7% after adjustment for lead-time bias. survival been the same as the women in New South Wales. The 5-year adjusted excess hazard ratio of breast cancer death for the non-screened group indicates a substantial and Bias and artefact Taken together, our results suggest that differences in screen- significant survival disadvantage for West Midlands. This is striking because these estimates are adjusted for differences ing practice and extent of disease at diagnosis do not explain in age and extent of disease at diagnosis, and so one might the overall difference in survival between West Midlands and expect survival to be much more similar. Even among New South Wales for this age group, and that women with screen-detected women the survival disadvantage is distinct breast cancer in West Midlands have a higher risk of excess death from their cancer than women in New South Wales, which is particularly striking because these are women diag- nosed with asymptomatic cancers. Their tumours are pre- whether they are screened or not. dominantly localised, and as such they would almost all be treated surgically and with curative intent and have a high The role of ‘de facto’ screening chance of long-term survival. These differences in survival are likely to be in part due to Although the overall number of deaths is relatively modest the differences in the way screening is delivered in the West in this cohort of cancer patients, with only 9.1% of all Midlands and in New South Wales. In the UK, the National Health Service is free at the point of delivery for the whole women dying during follow-up, the impact of these differen- ces is important. The increased excess hazard of breast cancer population and private mammography is rare. In contrast, in death 5 years after diagnosis in the West Midlands is double Australia, mammography is obtained through BreastScreen that of New South Wales amongst non-screen-detected Australia but also through private radiology clinics. Mammo- women and 72% greater amongst those with a screen- grams conducted privately for diagnostic purposes, rather than in asymptomatic women, may be refunded via the Med- detected cancer. Overall we estimated that more than a third of the deaths attributable to breast cancer observed for icare Benefits Scheme (MBS). A substantial proportion of V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology 2412 International differences in breast cancer survival those conducted in private clinics is likely to constitute de diagnosed at an earlier stage, with better prognosis. However, facto screening, (regular diagnostic mammography not the distribution by extent of disease was similar in both recorded by BreastScreen Australia), but it is unknown to regions; the proportion of localised disease was in fact slightly what degree this occurs. This is likely to be the reason for higher in the West Midlands than in New South Wales the higher proportion of tumours in the West Midlands that (67.1% versus 64.1%, Table 1). A shorter screening interval were apparently screen-detected, despite a shorter screening will lead to detection of a greater number of slower-growing interval in New South Wales. It also implies that women in tumours, but also greater numbers of aggressive, faster- New South Wales whom we defined as ‘never-attenders’ growing tumours, which will also be identified at an earlier includes a sub-group of women who had, in fact, been stage than would otherwise be the case. In our data, the dis- screened outside of the national screening programme. This tribution of tumours by extent of disease amongst interval interpretation is supported by the observation that a signifi- cancers was fairly similar in both regions (localised tumours cantly larger proportion of these women classified as ‘never- representing 51.8% in New South Wales and 50.0% in West attenders’ in New South Wales were diagnosed with localised Midlands, Chi p value 0.07) This supports the interpretation tumours (50.8% compared to 46.6% in West Midlands). that the breast cancer survival differences between New South Although it is probable that we incorrectly allocated some Wales and West Midlands cannot be fully explained by the women to the never-attender group who were actually screen- shorter screening interval in New South Wales. detected, especially in New South Wales, information on their We have made adjustment for lead-time and over- personal characteristics and the features of their cancer would diagnosis in our analysis, and demonstrated that the survival not have been compromised since these data items were col- differences observed are robust to these biases. Adjustment lected from the Cancer Registry, rather than via the screening involved a ten-fold simulation where both the individual sur- service. It is possible, however, that this may have biased our vival times were shortened and the number of women estimates of net survival. We therefore performed a sensitivity included in the cohort was reduced. On average, the survival analysis to examine the potential for de facto screening to time of screen-detected women was reduced by 1.5 years and explain the difference in survival for the non-screen-detected 40% were excluded (Table 2). This latter proportion does not group. We randomly reallocated women in New South Wales represent the percentage of tumours over-diagnosed, but with localised disease from the non-screen-detected to the rather the probability that a screen-detected cancer would screen-detected group, for selected proportions ranging from not have been detected symptomatically during the period of 1% to 95%, and then re-estimated the net survival function. time between the actual date of diagnosis and 31st December Over 100 iterations the five-year net survival estimates for the 2006 (the mean of which was 3.4 years). The number of non-screen-detected group in New South Wales became simi- tumours over-diagnosed might be reasonably obtained by lar to those for West Midlands only when an implausible 90% estimating the probability that the cancer would not have of the localised cancers (43% of all non-screen-detected can- been detected symptomatically during the woman’s remain- cers, c. 1800 women) were reallocated (data not shown). This ing expected life time (the mean of which was 30.6 years). level of reallocation would require that the true proportion of cancers screen-detected in New South Wales in the cohort Other non-causal explanations was in excess of 64%, in comparison to the 36.5% actually We have previously summarised the possible non-causal observed (and the 44.8% observed in West Midlands, that explanations for this difference. The first of these that may probably reflects the order of magnitude one might expect for be applicable here is the possibility that the NSWCCR Regis- New South Wales, since private mammography is very rare in try more often fails to link a woman’s death to the record of West Midlands). For smaller, but substantial proportions of her cancer registration than the WMNCRS, leading to appa- reallocation the reduction in the survival difference was rela- rently inflated survival. This explanation is very unlikely to tively small. We thus consider it very unlikely that de facto apply in this younger age group during this period of time – screening can fully explain the difference in survival between these are young women among whom death is a relatively non-screen-detected women. This analysis also served to illus- rare event, and who were followed up during a period of reli- trate the robust nature of the difference for the screen- able death registration. The second possible explanation is detected group: there remained a survival advantage for New that a higher proportion of in situ tumours registered in New South Wales, albeit very small, even when 95% of localised South Wales were misclassified as invasive breast cancers (apparently symptomatic) tumours were reallocated to the than in the West Midlands. Again, we do not consider that screen-detected group. this could be an explanation for the differences observed in this study, since >99% of the tumours analysed were micro- Screening-specific biases scopically verified and in situ cancers were excluded. We We may consider whether the longer screening interval in have also previously considered the accuracy and consistency the West Midlands compared with New South Wales (3 years of date of diagnosis as a potential mechanism by which sur- versus 2 years) might contribute to these differences. Screen- vival in New South Wales might be extended relative to West detected cancers in New South Wales could perhaps be Midlands. Again, however, this explanation has very little V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology Woods et al. 2413 credibility here, since the date of diagnosis is established in treated with similar effectiveness. We consider a much more the same manner for both screen-detected and non-screen- likely explanation for our findings is that the tumours them- detected women. selves are not substantially different, but that something dif- ferent happens once the woman is diagnosed. This is of Potential Explanations greater concern: we have examined a group of young women Examining possible explanations operating before diagnosis, with predominantly localised disease. Almost all of them the differences in breast cancer survival between West Mid- would have had treatment with curative intent. lands and New South Wales could arise from (a) greater Treatment may vary with comorbidity, and is subject to delays in diagnosis in West Midlands, (b) longer waiting patient compliance, but there is no particular reason to times for hospital consultation for non-screen-detected can- assume that these would persistently differ between New cers or (c) less effective screening in West Midlands than in South Wales and West Midlands, particularly in this age New South Wales. The fact that differences persist after group. The alternative explanation is that the treatment in adjustment for extent of disease at diagnosis does not support West Midlands is not as effective as those in New South any of these explanations, however. It is theoretically possible Wales leading to poorer stage-specific survival. that residual confounding may partially account for this lack of explanatory power. Residual confounding may have arisen Conclusions due to the fact that the screening interval for women in West Although overall survival was high for this cohort of women, Midlands is longer than in New South Wales, combined with our data suggest that more than one in three breast cancer a tendency for the accuracy of the ‘extent of disease’ variable deaths within five years of diagnosis in the West Midlands for non-screen-detected cancers to be lower in West Mid- would be avoidable if five-year survival were the same as in lands. Together, this would imply that within each stage New South Wales. The women we analysed here are rela- grouping, the true (unknown) stage of disease is more advanced in West Midlands than in New South Wales (stage tively young. They are therefore less likely to be suffering migration). This would have the effect of better extent- from other serious illnesses and more likely to be economi- adjusted survival in New South Wales. Although this expla- cally and socially active. In order to improve the prognosis nation is possible, we consider that in the context of this for women diagnosed with breast cancer in the UK during study it is not very likely. This is because two mechanisms their early 50s it is essential that we understand better the would both need to apply: delays for non-screen-detected mechanisms that underlie these international differences in cancers matched with less effective screening for screen- stage-adjusted survival. Differences in the effectiveness of detected cancers leading to differences in extent of a similar treatment are an important possibility and they now deserve magnitude in both groups. At the very least, the fact that to be examined with great care. It is not possible to dismiss there is a significant difference in survival amongst women differences in breast cancer survival between the UK and with screen-detected disease in the two regions refutes the other countries such as Australia as artefactual. hypothesis that international differences are entirely due to practitioner delay in referral (since all these women were Acknowledgements diagnosed through routine screening) or differences in The authors gratefully acknowledge the staff of the West Midlands Office of patient delay in seeking medical diagnosis following the the English National Cancer Registration Service, the New South Wales Cen- detection of breast cancer symptoms. tral Cancer Registry, and Professor Richard Taylor (BreastScreen New South These findings thus tend to refute the idea that breast Wales) for their assistance in obtaining and analysing these data. We also cancers in West Midlands and New South Wales are very wish to thank Dr Ula Nur and Dr Melanie Morris (Cancer Research UK different at the point of diagnosis, but are subsequently Cancer Survival Group) for helpful comments on the manuscript. References 1. Woods LM, Rachet B, O’Connell D, et al. Large 5. Brekelmans CT, Peeters PH, Deurenberg JJ, et al. cancer detection methods. Australian New Zea- differences in patterns of breast cancer survival Survival in interval breast cancer in the DOM land J Stat 2013;55:351–67. between Australia and England: a comparative screening programme. Eur J Cancer A 1995;31: 9. Sasieni P. Evaluation of the UK breast screening study using cancer registry data. Int J Cancer 1830–5. programmes. Annals Oncol 2003;14:1206–8. 2009;124:2391–9. 6. Olivotto IA, Mates D, Kan L, et al. Prognosis, 10. Zackrisson S, Andersson I, Janzon L, et al. Rate of over-diagnosis of breast cancer 15 years after 2. Woods LM, Rachet B, Coleman MP. Origins of treatment and recurrence of breast cancer for socio-economic inequalities in cancer survival: a women attending or not attending the Screening end of Malmo mammographic screening trial: review. Ann Oncol 2006;17:5–19. Mammography Program of British Columbia. follow-up study. Br Med J 2006;332:689–92. 3. Nystrom L, Rutqvist LE, Wall S, et al. Breast cancer Breast Cancer Res Treat 1999;54:73–81. 11. Duffy SW, Nagtegaal ID, Wallis M, et al. Correct- screening with mammography: overview of Swed- 7. Joensuu H, Lehtimaki T, Holli K, et al. Risk for ing for lead time and length bias in estimating ish randomised trials. Lancet 1993;341:973–8. distant recurrence of breast cancer detected by the effect of screen detection on cancer survival. 4. Collins S, Woodman CBJ, Threlfall A, et al. Sur- mammography screening or other methods. Am J Epidemiol 2008; 168:98–104. vival rates from interval cancer in NHS JAMA 2004;292:1064–73. 12. Census 2011. Office for National Statistics, 2013. breast screening programme. Br Med J 1998;316: 8. Hsieh JC-F, Cramb SM, McGree JM, et al. Bayes- 13. Australian census 2011. Australian Bureau of 832–3. ian spatial analysis for the evaluation of breast Statistics, 2014. V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology 2414 International differences in breast cancer survival 14. Pohar-Perme M, Stare J, Este`ve J. On Estimation other causes: a crude analogue of relative survival. 25. Allemani C, Weir HK, Carreira H, et al. Global in Relative Survival. Biometrics 2011. Stat Med 2000;19:1729–40. surveillance of cancer survival 1995-2009: analysis 15. StataCorp. STATA statistical software, ed. 12.1 of individual data for 25 676 887 patients from 21. Lambert PC, Dickman PW, Nelson CP, et al. College Station, Texas: Stata Corporation, 2011. 279 population-based registries in 67 countries Estimating the crude probability of death due to 16. Danieli C, Remontet L, Bossard N, et al. (CONCORD-2). Lancet 2014;385:977–1010. cancer and other causes using relative survival Estimating net survival: the importance of allow- 26. Coleman MP, Quaresma M, Berrino F, et al. models. Stat Med 2010;29:885–95. ing for informative censoring. StatMed 2012;31: Cancer survival in five continents: a worldwide 22. Abdel-Rahman M, Stockton D, Rachet B, et al. 775–86. population-based study (CONCORD). Lancet What if cancer survival in Britain were the same 17. Roche L, Danieli C, Belot A, et al. Cancer net Oncol 2008;9:730–56. as in Europe: how many deaths are avoidable? Br survival on registry data: use of the new unbiased 27. Yu XQ, O’Connell DL, Forman D. Comparison J Cancer 2009;101:S115–S24. Pohar-Perme estimator and magnitude of the of cancer survival in UK and Australia: rates are 23. Coleman MP, Forman D, Bryant H, et al. Cancer bias with the classical methods. Int J Cancer higher in Australia for three major sites. Br J survival in Australia, Canada, Denmark, Norway, 2013;132:2359–69. Cancer 2004;91:1663–5. Sweden, and the UK, 1995-2007 (the Interna- 18. Rubin DB. Multiple imputation for non-response 28. Morrell S, Taylor R, Roder D, et al. Mammog- tional Cancer Benchmarking Partnership): an in surveysed. New York: Wiley, 1987. raphy screening and breast cancer mortality in analysis of population-based cancer registry data. 19. Royston P, Parmar MK. Flexible parametric Australia: an aggregate cohort study. J Med Lancet 2011;377:127–38. proportional-hazards and proportional-odds Screen 2012;19:26–34. 24. Walters S, Maringe C, Butler J, et al. Breast can- models for censored survival data, with applica- 29. Feinstein AR, Sosin DM, Wells CK. The Will cer survival and stage at diagnosis in Australia, tion to prognostic modelling and estimation of Rogers phenomenon: stage migration and new treatment effects. StatMed 2002;21:2175–97. Canada, Denmark, Norway, Sweden and the UK, diagnostic techniques as a source of misleading 20. Cronin KA, Feuer EJ. Cumulative cause-specific 2000-2007: a population-based study. Br J Cancer statistics for survival in cancer. N Engl J Med mortality for cancer patients in the presence of 2013;108:1195–208. 1985;312:1604–8. V C Int. J. Cancer: 138, 2404–2414 (2016) 2016 The Authors International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC Cancer Epidemiology

Journal

International Journal of Cancer – Wiley

Published: May 15, 2016

Keywords: ; ; ; ; ; ; ; ;

There are no references for this article.