Original Article
Prognosis and Pathology of Screen-Detected Carcinomas How Different Are They? Iris D. Nagtegaal, MD, PhD1; Prue C. Allgood, PhD3; Stephen W. Duffy, MSc3; Olive Kearins2; E. O. Sullivan, BSc2; Nancy Tappenden, MPH2; Matthew Wallis, MB, ChB , FRCR4; and Gill Lawrence, PhD2
BACKGROUND: It has been observed that screen-detected breast cancers have a better prognosis than symptomatic tumors, even after taking pathological tumor attributes into account. This has led to the hypothesis that screendetected tumors are substantially biologically different from symptomatic cancers. METHODS: The pathology and survival by detection mode was investigated in 21,382 breast cancers diagnosed in women aged 50-64 years in the West Midlands, United Kingdom, between 1988 and 2004. Tumor attributes were compared using chi-square tests and logistic regression. Survival was analyzed using Cox regression. RESULTS: Screen-detected cancers were significantly smaller, better differentiated, and less likely to be node-positive than symptomatic cancers (P < .001 in all cases). In addition, a higher proportion of screen-detected cancers were hormone receptor-positive, and a higher proportion were tubular carcinomas (P < .001). Survival was substantially better in screen-detected breast cancers (86% at 10 yearsvs 70% for interval cancers and 58% for cancers in women unexposed to screening). Adjustment for age, tumor size, nodal status, grade, histological type, and year of diagnosis accounted for 64% (interval cancers) and 68% (unexposed women) of these survival differences, respectively. Overall survival improved with time. Approximately half of this improvement was due to the increase over time in the proportion of tumors that were screendetected. CONCLUSION: The majority of the difference in prognosis between screen-detected and symptomatic breast cancers is due to the differences in routinely measured pathological features (size, type, grade, and nodal status), leaving a small residual difference to be accounted for by other biological differences. Cancer 2011;117:1360–8. C 2010 American Cancer Society. V KEYWORDS: breast cancer, prognosis, pathology, screen-detected, survival.
The introduction of population-wide breast cancer screening programs has been accompanied by a decrease in breast
cancer-related mortality1,2 and by an increase in the incidence of breast cancer, notably of early-stage tumors.3 It has been suggested that the latter may constitute overdiagnosis (ie, the detection of tumors that would not have caused problems during the lifetime of the affected women) or possibly the less extreme phenomenon of length bias, whereby slow-growing tumors have a longer window of opportunity for screen detection and are therefore overrepresented in screen-detected cancers.4,5 Both hypotheses are consistent with biological differences between screen-detected and symptomatic cancers.5,6 It is clear that for breast screening to work, it must detect cancers at an earlier stage.7 Several reports indicate that breast cancers detected by screening have more favorable pathological features compared with symptomatic breast cancers, some of which may be due to a stage shift, and some to underlying biological differences.8-10 In addition to stage shift and length bias or overdiagnosis, there is an artificial increase in survival due to lead time (if screening detects a tumor on average 3 years earlier than symptoms would have arisen, 3 years are added to the survival time regardless of the eventual outcome). Between screening rounds, breast cancers occur symptomatically in women with a previous negative screen. These tumors, usually called interval cancers, have a worse prognosis than screen-detected carcinomas, although it is sometimes better than that of subjects unexposed to screening.11,12 They are likely to constitute a mixture of aggressive tumors that Corresponding author: Prue C. Allgood, PhD, CR-UK Centre for EMS, Wolfson Institute of Preventive Medicine, Charterhouse Square, London EC1M 6BQ, United Kingdom; Fax: (011) 44-2078823890;
[email protected] 1 Department of Pathology, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands; 2West Midlands Cancer Intelligence Unit, Birmingham, United Kingdom; 3CR-UK Center for Epidemiology, Mathematics and Statistics, Wolfson Institute of Preventive Medicine, London, United Kingdom; 4Cambridge Breast Screening Unit and National Institute for Health Research, Cambridge Biomedical Research Center, Addenbrooke’s Hospital Cambridge, United Kingdom
We thank Rosie Day at the West Midlands Cancer Intelligence Unit for extracting the data from the cancer registration database. DOI: 10.1002/cncr.25613, Received: May 20, 2010; Accepted: July 28, 2010, Published online November 2, 2010 in Wiley Online Library (wileyonlinelibrary.com)
1360
Cancer
April 1, 2011
Breast Cancer Differences/Nagtegaal et al
have developed rapidly to the symptomatic phase and less aggressive tumors that have been missed at the previous screening. In the present study, we used a database of more than 21,000 breast cancers recorded on the West Midlands Cancer Intelligence Unit’s cancer registration database to address the following issues to a level of detail and precision that has not been possible in smaller tumor series: 1. What are the typical pathological features of screen-detected breast cancers and how do these differ from interval cancers and tumors in women unexposed to screening? 2. What are the implications of these differences for prognosis? 3. To what extent can the better survival of screendetected breast cancers be attributed to shifts in nodal status, tumor size, and other routinely measured histological factors, and what remains to be explained by biological and other differences? 4. What changes in tumor attributes and in survival have occurred over the period of observation, and to what extent are these changes attributable to screening or to other factors?
MATERIALS AND METHODS Patients In collaboration with the National Health Service Breast Screening Programme, the West Midlands Cancer Intelligence Unit collected screening history, pathological data, and follow-up data on cancers diagnosed in women aged 50-69 years in the West Midlands, United Kingdom, from 1988 to 2001. From 2002 to 2004, the dataset was expanded to include women aged 50-74 years, and with an extension of the upper age limit for invitation to 70 years, nonnegligible screening was being performed in women aged 70-74 years. In the present study, to avoid complications in screening history classification caused by the staggered rollout of the upper age limit extension, only data for the 21,382 breast cancers diagnosed in women aged 50-64 were included. The data were available for 9259 screen-detected breast cancers, 5413 interval cancers, and 6710 cancers in women not recently exposed to screening. In the early years of the program, the latter group mainly comprised women who had not yet been invited, whereas in the later years it was mainly made up
Cancer
April 1, 2011
of nonattenders. The group as a whole is referred to as ‘‘unexposed’’ for purposes of brevity. Follow-up was available through April 12, 2007. Pathology Data Invasive status (invasive, microinvasive, in situ), differentiation grade, vascular invasion, nodal status, tumor size, and tumor type (according to ICD-O3 morphology classification) were recorded from the pathology reports. In addition, estrogen receptor (ER) and progesterone receptor (PR) status were collected, although the majority of these data were not available until 2002 and 2004, respectively. Statistics Data were analyzed using the Stata software package.13 Associations among the various parameters were analyzed using chi-square tests and logistic regression.14 Trends over time were also analyzed using logistic regression. Survival analyses were performed using Cox regression, with the time of diagnosis as the entry date and breast cancerrelated death as outcome,15 yielding relative hazards of breast cancer death and 95% confidence intervals on these. The extent to which screening effects on survival could be attributable to other factors was calculated by comparison of adjusted and unadjusted Cox regression analyses using the method of Freedman et al.16 Cumulative survival was calculated using Kaplan-Meier survival curves.15 P .05 was considered statistically significant.
RESULTS Tumor Attributes and Mode of Detection A total of 21,382 breast cancers diagnosed in women aged 50-64 years were studied: 43.3% were screen-detected, 25.3% were interval cancers, and 31.4% occurred in women unexposed to screening. Between the 3 groups, there were substantial differences in tumor attributes (Table 1). Screen-detected breast cancers were lower grade (better differentiated), less likely to be node-positive, more likely to be ER- or PR–positive, and smaller than tumors in the 2 symptomatic groups (P < .001 in all cases). The women unexposed to screening were more likely to have larger tumors than those with interval cancers and were more likely to have nodal status, histological grade, and undetermined ER and PR status. More in situ carcinomas were present in the screen-detected group (15.5% vs 3.6% and 3.2%, respectively, P < .001). The histological type of tumors differed significantly (P
20 to £50 mm >50 mm
50 mm
50 mm Lobular carcinoma Tubular carcinoma
6%a 3%a 4%a þ2%a 3%a þ3%a þ2%a
10%a 7%a 10%a þ3% 7%a þ4%a þ8%a
5%a 0.8% þ1% þ1% þ4% % 4%a
a b
Figure 1. Proportion of all breast tumors (in situ and invasive) diagnosed over time for women screened and women not screened is shown.
.001) by detection mode. Most tumors were invasive ductal carcinoma without further specification. In the screendetected group, there was a higher proportion of tubular carcinoma (8.4% vs 2.0% and 2.1% respectively) and a lower proportion of ductal carcinoma. Rare tumor types included acinar cell carcinoma (n ¼ 2), clear cell adenocarcinoma (n ¼ 4), and inflammatory carcinoma (n ¼ 15). Inflammatory carcinoma was mostly diagnosed symptomatically (n ¼ 1, n ¼ 9, and n ¼ 5 for screendetected cancers, interval cancers, and cancers in women unexposed to screening, respectively). The sarcomas con-
Cancer
April 1, 2011
Trend significant with p < .05. Only patients with known nodal status were considered.
sisted of fibrosarcomas (n ¼ 2), leiomyosarcomas (n ¼ 1), osteosarcomas (n ¼ 1), and sarcomas not otherwise specified (n.o.s.) (n ¼ 5). The in situ carcinomas that were screen-detected were smaller (Table 2) and more often ductal carcinoma in situ. Although there was a trend towards a higher cytonuclear grade for all modes of detection there was no significant difference between detection modes (P ¼ .6). In situ interval cancers had the largest proportion of intraductal papillary tumors. Trends Over Time Changes in detection mode with time are shown in Figure 1. In 1988, 5.9% of all breast cancers were diagnosed in women exposed to screening (screen-detected or interval
1363
Original Article
Figure 2. Mode of detection data are shown for (A) node positivity in invasive breast tumors, (B) grade 3 invasive breast tumors, (C) lobular invasive breast tumors, (D) invasive breast tumors 10 mm, (E) tubular invasive breast tumors, and (F) invasive breast tumors >50 mm.
cancers), whereas in 1997, 78.1% of all breast cancers were diagnosed in women who had been exposed to screening. This figure rose slightly each year and was 81.9% for all breast cancers in 2004. This was contempo-
1364
raneous with considerable changes in the distribution of pathological features in invasive cancers (Table 3). We determined the trends in time during the starting phase of the program (1988-1996) and during the continuation
Cancer
April 1, 2011
Breast Cancer Differences/Nagtegaal et al
Figure 3. Kaplan-Meier survival curve for women aged 50-64 years diagnosed with invasive breast tumors is shown.
phase (1997-2004) using logistic regression with year as a continuous variable. Overall, there were reductions with time in the proportion of invasive breast cancers, node-positive cancers, grade 3 cancers, and large (pT3) invasive tumors >50 mm; these were much more pronounced during the starting phase of the program. Overall and during the early years there was a 2% and 8% annual increase, respectively, in the proportion of tubular carcinomas, whereas during the later years there was a 4% annual decrease. When evaluating trends in time for the different detection modes (screen-detected, interval, women unexposed to screening; Figure 2), changes in interval cancers and cancers in women unexposed to screening were remarkably similar for node positivity, poor differentiation, and small tumors (50 mm were consistently more often present in the unexposed group compared with both interval and screen-detected cancers. Tubular carcinoma was consistently higher in the screen-detected group, but with time an increase in lobular carcinoma was present in all 3 groups. Within the detection modes, proportions of small tumors and tubular carcinomas remained constant, indicating that the overall increases were due to the increase in the proportion of screen-detected tumors. Survival Women with screen-detected cancers had a better prognosis than women with interval cancers and cancers in
Cancer
April 1, 2011
women with no screening (Figure 3). Ten-year survival rates for invasive breast cancer were 85.5%, 69.8%, and 57.7%, respectively (P < .001), with an overall rate of 71.6%. These differences are also reflected in the unadjusted Cox regression model (Table 4) for invasive breast tumors only, in which the hazard ratios for interval cancers and cancers in women unexposed to screening were 2.3 and 3.4, respectively, for breast cancer-related death. In the adjusted model, after correction for nodal status, tumor size, differentiation grade, tumor type, age, and year of diagnosis, this difference was smaller but was still present, with hazard ratios of 1.26 and 1.54, respectively. When adjusted for age, tumor size, and node status only, the hazard ratios for interval cancers and cancers in women who had not been screened were 1.56 and 2.04, respectively, compared with screen-detected cancers (complete results are available from the authors). The Freedman statistics for the proportion of the survival differences for invasive cancers explained by age, tumor size, and node status were 47% for the difference in survival between screen-detected and interval cancers and 42% for the difference between screen-detected cancers and cancers in women unexposed to screening. Additional adjustment for grade, tumor type, and year of diagnosis accounted for a further 17% and 26%, respectively, bringing the total proportions explained to 64% and 68%, respectively. When evaluating screening status and year of diagnosis only, the Freedman proportions explained for interval cancers and cancers in women unexposed to screening were 0.03% and 4.8%, respectively, suggesting that the better prognosis of screen-detected cancers was not simply due to the strong confounding with year of diagnosis. On the other hand, for year of diagnosis, the Freedman statistic for the proportion of the year of diagnosis effect explained by confounding with detection mode was 34%, which means that a substantial part of the increase in survival by year of diagnosis is explained by the increase in the proportion of screen-detected cancers. This finding is consistent with the predicted 10-year survival rates from the Cox regression model, including year of diagnosis and detection mode in 1990 and 1997, respectively (Table 5). Expected survival rates increased during these 7 years for all detection modes, but most of the change in overall survival was due to the increased proportion of screendetected cases (a 2% increase) and the improved survival of the symptomatic cancers (a 3% increase). Ten-year survival for women with screen-detected cancers and women unexposed to screening increased over time, but the survival of women with interval cancers decreased by 1.5%.
1365
Original Article Table 4. Unadjusted and Adjusted Cox Regression Model for Breast Cancer Survival After Invasive Carcinoma
Factor
Unadjusted HR n519,411 HR
95% CI
Adjusted HR n510,245 P
HR
95% CI
Histological type Screen-detected Interval Unexposed
1.00 2.3 3.4
— 2.1 3.2-3.7
—
