Abstract
Background
Many breast cancer survivors experience anxiety related to dying from their disease even if it is detected at an early stage. We aimed to increase knowledge about fatal and non-fatal breast cancer by describing how histopathological tumour profiles and detection modes were associated with 10-year breast cancer-specific survival.
Methods
This cohort study included data from women targeted by BreastScreen Norway (aged 50–69) and diagnosed with invasive breast cancer during 1996–2011. Breast cancer was classified as fatal if causing death within 10 years after diagnosis and non-fatal otherwise. We described histopathologic characteristics of fatal and non-fatal cancers, stratified by mode of detection. Recursive partitioning identified subgroups with differing survival profiles.
Results
In total, 6.3% of 9954 screen-detected cancers (SDC) were fatal, as were 17.4% of 3205 interval cancers (IC) and 20.9% of 3237 cancers detected outside BreastScreen Norway. Four to five subgroups with differing survival profiles were identified within each detection mode. Women with lymph node-negative SDC or Grade 1–2, node-negative IC without distant metastases had the highest 10-year survival (95–96%).
Conclusions
Two subgroups representing 53% of the cohort had excellent (95–96%) 10-year breast cancer-specific survival. Most women with SDC had excellent survival, as did nearly 40% of women diagnosed with IC.
Subject terms: Risk factors, Breast cancer, Epidemiology, Population screening, Breast cancer
Background
Breast cancer is the leading cause of cancer deaths in women worldwide [1]. Mammographic screening aims to reduce breast cancer mortality by detecting the disease at an early and curable stage. This effect is best documented for average-risk women aged 50–69, where those who attend organized screening have about 30% lower breast cancer mortality than those who do not attend [2, 3].
The association between prognostic histopathological factors and breast cancer mortality is well described, particularly in relation to stage and molecular subtypes [4–6]. Regarding the effect of screening on breast cancer mortality, observational studies often stratify analyses for “ever” versus “never” attenders or use an “intention to screen” approach based on whether a woman was invited to screening [3, 7]. Less research has explored the effect of the women’s individual screening history and whether cancers were screen-detected, interval, or detected outside of a screening program [8, 9].
A 2021 meta-analysis reported that as many as 34% of women diagnosed with breast cancer—including those diagnosed as a direct result of screening—experience anxiety associated with their diagnosis, and as many as 39% experience symptoms of non-specific distress [10]. Increased knowledge about the histopathology and survival profiles associated with fatal versus non-fatal outcomes stratified by mode of detection can improve targeted information provided to women who are invited to organized screening. This could potentially reduce some of the psychological consequences of receiving a breast cancer diagnosis.
The aims of this population-based cohort study were to (a) describe the histopathological tumour characteristics of fatal and non-fatal cancers and (b) describe 10-year breast cancer-specific survival, stratified by mode of detection.
Methods
Study setting
BreastScreen Norway targets women aged 50–69 for biennial mammographic screening, based on national and international recommendations [11–13]. The program started in four counties in 1996 and became nationwide in 2005. As of 2023, the target group consists of roughly 650,000 women. The program is described in detail elsewhere [14].
Study sample
This retrospective cohort study included all women in the target group of BreastScreen Norway diagnosed with invasive breast cancer after receiving at least one invitation to attend the program between January 1, 1996 and December 31, 2011. For women diagnosed with multiple primary breast cancers, we included their first cancer. For women with multifocal disease, the largest tumour was included. We excluded women who: were ≥72 years old at diagnosis; had a history of ductal carcinoma in situ (DCIS) or invasive breast cancer prior to their first invitation to BreastScreen Norway; were diagnosed at death; or participated in the Oslo age trial, which invited women aged 45–49 to mammographic screening [15].
Data sources
Study data were obtained from the Cancer Registry of Norway. The Cancer Registry registers incidence and histopathology information about invasive breast cancer diagnoses in Norway for all women of all ages. Information about breast cancer diagnoses is nearly complete and over 99% of cases are morphologically verified [16]. The Cancer Registry administers BreastScreen Norway and therefore also captures information about screening invitations, attendance, and associated outcomes. The Norwegian Cause of Death Registry provides the Cancer Registry with information about the date and cause of death.
Variables
We here defined breast cancer as invasive tumours. Cancer was described as “fatal” if it was listed as a woman’s cause of death within 10 years after her diagnosis. The mode of detection was categorized as screen-detected, interval, or outside the screening program. Screen-detected cancers were defined as those diagnosed within 6 months of a screening examination that led to a recall for further assessment. Interval cancers were those diagnosed within 24 months of a negative screening examination or 6–24 months after a screening examination with a false-positive result. Cancers detected outside of screening were defined as those diagnosed among women who had not attended BreastScreen Norway in the 24 months prior to their diagnosis.
Person-level variables included age at and year of diagnosis, and birth country (Norway or other). For women with screen-detected cancer, attendance at the screening examination that led to their diagnosis was categorized as “prevalent screen”, “regular screen” or “irregular screen”. Prevalent screens were defined as a woman’s first screening examination in BreastScreen Norway. Regular screens were those occurring within 24 ± 6 months of the previous screening examination, and irregular screens were those performed >30 months after the previous screening examination.
Histopathologic variables included histologic type (invasive carcinoma of no special type, lobular, or other), histologic grade (1, 2, 3), and pathologic TNM categories described by the American Joint Committee on Cancer (pT-categories: T1a/b, T1c, T2, T3, T4; pN-categories: N0, N1, N2, N3; and pM categories: M0, M1) [4]. Estrogen receptor (ER) and progesterone receptor (PR) status were also included and classified as positive, negative, or missing. Due to changes in national guidelines, ER status was recorded as positive if there was ≥10% ER expression (study start to Jan 2010) or ≥1% ER expression (Jan 2010 onwards) [4]. PR status was recorded as positive if there was ≥10% PR expression (entire study period).
Statistical analysis
We described the variables outlined above using means and standard deviations for continuous variables, and frequencies and proportions for categorical variables. The proportion of fatal and non-fatal cancers within each mode of detection was presented for histopathological variables. In addition, we evaluated the proportion of fatal and non-fatal cancers for various pT and pN combinations of non-metastatic screen-detected cancers. 95% confidence intervals (CIs) for proportions were calculated using the Wilson score interval [17].
Defining subgroups with differing survival profiles
For each mode of detection, we used a recursive binary partitioning algorithm to create a tree diagram (survival tree) that identified patient and histopathological tumour characteristics that could be used to classify patients into subgroups with distinct 10-year breast cancer-specific survival profiles.
The recursive binary partitioning algorithm used information about 11 person and histopathological tumour characteristics (“input variables”) to repeatedly split a group of women (parent group) into two subgroups (child groups) with more homogeneous survival outcomes than the initial (parent) group. The input variables are described later. All values for all input variables were candidates for creating each binary split. In this closed cohort study, women were followed from their breast cancer diagnosis until death from the disease. Their follow-up time was censored if they died from other causes or emigrated, or 10 years after diagnosis, whichever occurred first. Our approach assumed an exponential survival (i.e. constant hazard) model and variable/value selection was based on the likelihood ratio test for two variables with Poisson distributions [18, 19]. This splitting process stopped when any resulting child group would include <30 women or when only marginal gains in survival homogeneity were obtained from creating additional splits. Although the outcome of interest was time to death from breast cancer, this approach did not explicitly consider the competing risk of death from other causes. The majority of women diagnosed with breast cancer in Norway have a good overall prognosis and we assumed that any bias resulting from this approach would not substantially affect the results of the recursive partitioning algorithm.
Recursive binary partitioning can produce very large trees with many splits that can describe the survival of very specific groups but lack generalizability. Indeed, the aim of this descriptive study was to identify broad groups with differing survival profiles. Therefore, to prune potentially large trees, we used tenfold cross-validation to select the minimum number of splits in our decision trees for which the cross-validated error was less than the minimum observed error plus the associated standard error [20]. The resulting survival trees outlined a simple set of classification rules to define subgroups of women with similar survival profiles.
The 11 input variables we considered were: year of diagnosis, age at diagnosis, birth country, attendance pattern (screen-detected cancers only), histopathological type and grade, T, N and M categories, and ER and PR status. The levels and ranges for these variables are described in Supplementary Table S1. Missing information for input variables (summarized in Tables 1 and 2) was handled automatically using surrogate splits where a suitable non-missing variable was used to classify an observation instead of the missing variable [19, 20]. There was no missing information for the outcome of interest (survival).
Table 1.
Patient characteristics of 16,396 women in the target group of BreastScreen Norway and diagnosed with breast cancer during 1996–2011, stratified by mode of detection and whether the cancer was fatal or non-fatal within 10 years of diagnosis.
| Screen-detected cancers | Interval | Outside screening | Total | |||||
|---|---|---|---|---|---|---|---|---|
| Fatal | Non-fatal | Fatal | Non-fatal | Fatal | Non-fatal | Fatal | Non-fatal | |
| n = 630 (6.3%) | n = 9324 (93.7%) | n = 559 (17.4%) | n = 2646 (82.6%) | n = 675 (20.9%) | n = 2562 (79.1%) | n = 1864 (11.4%) | n = 14,532 (88.6%) | |
| Age at diagnosis (years), mean (SD) | 60.2 (5.6) | 60.1 (5.7) | 59.9 (5.4) | 60.0 (5.6) | 61.6 (6.2) | 61.1 (6.2) | 60.6 (5.8) | 60.3 (5.8) |
| Age at diagnosis (years), [range] | [49, 71] | [49, 71] | [50, 71] | [49, 71] | [49, 71] | [48, 71] | [49, 71] | [48, 71] |
| Birthplace, n (%) | ||||||||
| Norway | 602 (6.3) | 8895 (93.7) | 529 (17.3) | 2520 (82.7) | 630 (20.7) | 2419 (79.3) | 1761 (11.3) | 13,834 (88.7) |
| Other | 21 (6.1) | 322 (93.9) | 24 (22.6) | 82 (77.4) | 34 (22.5) | 117 (77.5) | 79 (13.2) | 521 (86.8) |
| Unknown | 7 (6.1) | 107 (93.9) | 6 (12.0) | 44 (88.0) | 11 (29.7) | 26 (70.3) | 24 (11.9) | 177 (88.1) |
| Year of diagnosis, n (%) | ||||||||
| 1996–1999 | 112 (9.7) | 1040 (90.3) | 56 (19.6) | 229 (80.4) | 52 (20.1) | 207 (79.9) | 220 (13.0) | 1476 (87.0) |
| 2000–2004 | 226 (7.1) | 2941 (92.9) | 189 (20.7) | 723 (79.3) | 200 (22.1) | 703 (77.9) | 615 (12.3) | 4367 (87.7) |
| 2005–2009 | 218 (5.5) | 3710 (94.5) | 253 (17.8) | 1170 (82.2) | 300 (21.4) | 1102 (78.6) | 771 (11.4) | 5982 (88.6) |
| 2010–2011 | 74 (4.3) | 1633 (95.7) | 61 (10.4) | 524 (89.6) | 123 (18.3) | 550 (81.7) | 258 (8.7) | 2707 (91.3) |
| Attendance prior to diagnosis, n (%) | ||||||||
| Prevalent screena | 238 (7.4) | 2995 (92.6) | – | – | – | – | 238 (7.4) | 2995 (92.6) |
| Regular screenb | 352 (5.7) | 5789 (94.3) | – | – | – | – | 352 (5.7) | 5789 (94.3) |
| Irregular screenc | 40 (6.9) | 540 (93.1) | – | – | – | – | 40 (6.9) | 540 (93.1) |
aPrevalent, first screening examination in BreastScreen Norway.
bRegular screening examinations performed within 24 ± 6 months of the previous screen.
cIrregular screening examinations performed more than 30 months after the previous screen.
Table 2.
Histopathology of breast cancers among 16,396 women in the target group of BreastScreen Norway and diagnosed with breast cancer during 1996–2011, stratified by mode of detection and whether the cancer was fatal or non-fatal within 10 years of diagnosis.
| Screen-detected cancers | Interval | Outside screening | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Fatal | Non-fatal | Fatal | Non-fatal | Fatal | Non-fatal | |||||||
| n = 630 (6.3%) | n = 9324 (93.7%) | n = 559 (17.4%) | n = 2646 (82.6%) | n = 675 (20.9%) | n = 2562 (79.1%) | |||||||
| n | % (95% CI) | n | % (95% CI) | n | % (95% CI) | n | % (95% CI) | n | % (95% CI) | n | % (95% CI) | |
| Histologic type | ||||||||||||
| Invasive NSTa | 561 | 6.6 (6.1, 7.2) | 7904 | 93.4 (92.8, 93.9) | 465 | 17.8 (16.4, 19.3) | 2144 | 82.2 (80.7, 83.6) | 560 | 21.5 (19.9, 23.1) | 2050 | 78.5 (76.9, 80.1) |
| Lobular | 61 | 6.2 (4.8, 7.9) | 927 | 93.8 (92.1, 95.2) | 80 | 19.2 (15.7, 23.3) | 336 | 80.8 (76.7, 84.3) | 75 | 21.0 (17.1, 25.5) | 282 | 79.0 (74.5, 82.9) |
| Other | 8 | 1.6 (0.8, 3.1) | 493 | 98.4 (96.9, 99.2) | 14 | 7.8 (4.7, 12.6) | 166 | 92.2 (87.4, 95.3) | 15 | 9.9 (6.1, 15.6) | 137 | 90.1 (84.4, 93.9) |
| Not available | 0 | – | 0 | – | 0 | – | 0 | – | 25 | – | 93 | – |
| Histologic grade | ||||||||||||
| 1 | 87 | 2.6 (2.1, 3.2) | 3207 | 97.4 (96.8, 97.9) | 26 | 4.7 (3.2, 6.7) | 532 | 95.3 (93.3, 96.8) | 30 | 5.7 (4.0, 8.0) | 498 | 94.3 (92.0, 96.0) |
| 2 | 293 | 6.3 (5.6, 7.0) | 4349 | 93.7 (93.0, 94.4) | 230 | 16.0 (14.2, 18.0) | 1207 | 84.0 (82.0, 85.8) | 230 | 16.9 (15.0, 19.0) | 1130 | 83.1 (81.0, 85.0) |
| 3 | 233 | 13.1 (11.6, 14.8) | 1540 | 86.9 (85.2, 88.4) | 266 | 24.7 (22.2, 27.4) | 811 | 75.3 (72.6, 77.8) | 248 | 28.4 (25.5, 31.5) | 626 | 71.6 (68.5, 74.5) |
| Not available | 17 | – | 228 | – | 37 | – | 96 | – | 167 | – | 308 | – |
| pT-category | ||||||||||||
| T1a/b: > 0–10 mm | 81 | 2.2 (1.8, 2.8) | 3529 | 97.8 (97.2, 98.2) | 30 | 6.3 (4.5, 8.9) | 444 | 93.7 (91.1, 95.5) | 19 | 4.1 (2.7, 6.3) | 442 | 95.9 (93.7, 97.3) |
| T1c: 11–20 mm | 239 | 5.5 (4.8, 6.2) | 4134 | 94.5 (93.8, 95.2) | 123 | 10.5 (8.9, 12.4) | 1047 | 89.5 (87.6, 91.1) | 75 | 7.8 (6.2, 9.6) | 890 | 92.2 (90.4, 93.8) |
| T2: 21–50 mm | 229 | 14.4 (12.7, 16.2) | 1364 | 85.6 (83.8, 87.3) | 205 | 19.6 (17.3, 22.1) | 840 | 80.4 (77.9, 82.7) | 178 | 20.3 (17.7, 23.1) | 700 | 79.7 (76.9, 82.3) |
| T3: > 50 mm | 47 | 21.1 (16.2, 26.9) | 176 | 78.9 (73.1, 83.8) | 116 | 34.8 (29.9, 40.1) | 217 | 65.2 (59.9, 70.1) | 44 | 57.1 (46.0, 67.6) | 33 | 42.9 (32.4, 54.0) |
| T4: Direct extension to the chest wall | 13 | 23.6 (14.4, 36.3) | 42 | 76.4 (63.7, 85.6) | 23 | 44.2 (31.6, 57.7) | 29 | 55.8 (42.3, 68.4) | 96 | 60.8 (53.0, 68.0) | 62 | 39.2 (32.0, 47.0) |
| Not available | 21 | – | 79 | – | 62 | – | 69 | – | 263 | – | 435 | – |
| pN-category | ||||||||||||
| pN0 | 257 | 3.5 (3.1, 3.9) | 7094 | 96.5 (96.1, 96.9) | 151 | 8.7 (7.4, 10.1) | 1590 | 91.3 (89.9, 92.6) | 112 | 7.5 (6.3, 8.9) | 1385 | 92.5 (91.1, 93.7) |
| pN1 | 232 | 11.5 (10.2, 13.0) | 1787 | 88.5 (87.0, 89.8) | 218 | 21.6 (19.2, 24.3) | 790 | 78.4 (75.7, 80.8) | 356 | 30.7 (28.1, 33.4) | 804 | 69.3 (66.6, 71.9) |
| pN2 | 62 | 23.6 (18.8, 29.1) | 201 | 76.4 (70.9, 81.2) | 84 | 37.8 (31.7, 44.4) | 138 | 62.2 (55.6, 68.3) | 24 | 35.8 (25.4, 47.8) | 43 | 64.2 (52.2, 74.6) |
| pN3 | 52 | 47.3 (38.2, 56.5) | 58 | 52.7 (43.5, 61.8) | 54 | 50.9 (41.6, 60.3) | 52 | 49.1 (39.7, 58.4) | 14 | 35.9 (22.7, 51.6) | 25 | 64.1 (48.4, 77.3) |
| Not available | 27 | – | 184 | – | 52 | – | 76 | – | 169 | – | 305 | – |
| PM-category | ||||||||||||
| M0 | 542 | 5.8 (5.4, 6.3) | 8752 | 94.2 (93.7, 94.6) | 421 | 14.6 (13.4, 15.9) | 2463 | 85.4 (84.1, 86.6) | 378 | 14.1 (12.9, 15.5) | 2298 | 85.9 (84.5, 87.1) |
| M1 | 54 | 71.1 (61.0, 80.0) | 22 | 28.9 (20.0, 40.0) | 109 | 80.7 (73.3, 86.5) | 26 | 19.3 (13.5, 26.7) | 239 | 76.4 (71.3, 80.7) | 74 | 23.6 (19.3, 28.7) |
| Not available | 34 | – | 550 | – | 29 | – | 157 | – | 58 | – | 190 | – |
| ERb status | ||||||||||||
| Positive | 449 | 5.4 (5.0, 5.9) | 7831 | 94.6 (94.1, 95.0) | 319 | 14.1 (12.7, 15.5) | 1951 | 85.9 (84.5, 87.3) | 135 | 17.1 (14.7, 19.9) | 653 | 82.9 (80.1, 85.3) |
| Negative | 148 | 13.5 (11.6, 15.6) | 949 | 86.5 (84.4, 88.4) | 207 | 27.3 (24.2, 30.5) | 552 | 72.7 (69.5, 75.8) | 45 | 23.8 (18.3, 30.4) | 144 | 76.2 (69.6, 81.7) |
| Not available | 33 | – | 544 | – | 33 | – | 143 | – | 495 | – | 1765 | – |
| PRc status | ||||||||||||
| Positive | 320 | 4.9 (4.4, 5.5) | 6167 | 95.1 (94.5, 95.6) | 222 | 13.0 (11.5, 14.7) | 1480 | 87.0 (85.3, 88.5) | 91 | 15.1 (12.5, 18.2) | 511 | 84.9 (81.8, 87.5) |
| Negative | 272 | 9.7 (8.7, 10.9) | 2520 | 90.3 (89.1, 91.3) | 300 | 23.1 (20.9, 25.5) | 996 | 76.9 (74.5, 79.1) | 85 | 23.4 (19.4, 28.0) | 278 | 76.6 (72.0, 80.6) |
| Not available | 38 | – | 637 | – | 37 | – | 170 | – | 499 | – | 1773 | – |
aInvasive carcinoma of no special type.
bEstrogen receptor.
cProgesterone receptor.
Describing subgroups with differing survival profiles
Using the subgroups defined by the recursive binary partitioning algorithm and the same time-to-event data, we calculated women’s 10-year breast cancer-specific survival as one minus the cumulative incidence of breast cancer-specific death. Corresponding 95% CIs were also presented. These results were presented graphically using survival curves that can be interpreted as the real-world proportion of women who did not die from their breast cancer.
Recursive partitioning was performed using R (v4.1.3) with the rpart (v4.1.19), and rpart.plot (v3.1.1) packages [21–23]. All other analyses were performed using Stata (v18.0); cumulative incidences were calculated using the stcompet package [24].
Results
In total, 18,890 women were diagnosed with breast cancer after receiving at least one invitation to BreastScreen Norway between January 1, 1996 and December 31, 2011. After applying the exclusion criteria, 16,396 women were included in the study sample: 9954 (60.7%) with screen-detected cancer, 3205 (19.6%) with interval cancer, and 3237 (19.7%) with cancer detected outside of the screening program (Fig. 1). Among screen-detected cancers, 6.3% (n = 630) were fatal, whereas 17.4% (n = 559) of interval cancers, and 20.9% (n = 675) of cancers detected outside of screening were fatal (Table 1).
Fig. 1. Number of women included and excluded in this study.
Women were excluded sequentially using the given criteria.
The mean (SD) age at diagnosis was 60 (5.8) years and was similar across all modes of detection and for fatal and non-fatal cancers (Table 1). Overall, 95% of women in the cohort were born in Norway. The incremental roll-out of organized screening was reflected in an increasing number of cancers diagnosed later in the study period. Moreover, the proportion of fatal cancers was somewhat higher earlier in the study period versus later in the study period for all modes of detection (Table 1). The majority (64.3%) of screen-detected cancers were associated with regular screens and the proportion of fatal cancers was higher for prevalent screens (7.4%, n = 238) than regular screens (5.7%, n = 352).
Across all modes of detection, the proportion of fatal cancers increased among tumours with less favourable histopathology: increased tumour diameter, histopathologic grade, or lymph node involvement, and negative ER or PR receptor status (Table 2). The most pronounced difference in the proportions of fatal cancers was for women with versus without distant metastases at diagnosis, however only a small proportion of women were diagnosed with distant metastases (0.8% of screening-detected cancers, 4.2% of interval cancers, and 9.7% of cancers detected outside screening). Some of the lowest proportions of fatal cancers were observed among women with screen-detected cancers that were Grade 1 (2.6% fatal), T1a/b (>0–10 mm; 2.2% fatal), or that were node-negative (3.5% fatal; Table 2).
When evaluating the proportion of fatal screen-detected cancers among various combinations of pTNM-classifications and grade (Supplementary Table S2), the highest proportions were observed among T1c or T2, node-positive, and M0 cancers (10.7% among Grade 1 or 2 cancers and 23.4% among Grade 3 cancers). The lowest proportions of fatal screen-detected cancers were observed among T1a/b, N0, and M0 cancers (1.5% among Grade 1 or 2 cancers and 3.5% among Grade 3 cancers).
Survival
The 9954 women with screen-detected cancers contributed a total of 93,403 person-years to the analysis. The recursive partitioning algorithm identified four subgroups with distinct survival profiles among these women: node-negative (SDC Subgroup 1); N1 and M0 (SDC Subgroup 2); N2 or N3, and M0 (SDC Subgroup 3); and node-positive and M1 (SDC Subgroup 4; Fig. 2a). The 10-year breast cancer survival was 96.3% (95% CI: 95.9–96.7) for women classified as belonging to SDC Subgroup 1; 89.3% (95% CI: 87.9–90.6) for SDC Subgroup 2; 72.6% (95% CI: 67.9–77.1) for SDC Subgroup 3; and 21.6% (95% CI: 12.6–35.6) for SDC Subgroup 4 (Fig. 2b).
Fig. 2. Survival profiles for women diagnosed with screen-detected cancer through BreastScreen Norway.
a Breast cancer-specific survival tree, % indicates the proportion of women who died of breast cancer within each subgroup. b 10-year breast cancer-specific survival, stratified using the subgroups created by the survival tree.
The 3205 women with interval cancer contributed a total of 27,623 person-years to the analysis. Five subgroups were identified by the recursive partitioning algorithm among women with interval cancer: Grade 1 or 2, node-negative and M0 (IC Subgroup 1); Grade 3, node-negative and M0 (IC Subgroup 2); N1 and M0 (IC Subgroup 3); N2 or N3, and M0 (IC Subgroup 4); and M1 (IC Subgroup 5; Fig. 3a). IC Subgroups 2 and 3 had similar survival profiles and were therefore combined for the survival curves. The 10-year breast cancer survival for IC Subgroups 1, 2/3, 4 and 5 were: 94.9% (95% CI: 93.6–96.0); 82.7% (95% CI: 80.8–84.6%); 60.7% (95% CI: 55.3–66.2); and 19.2% (95% CI: 13.5–26.9) (Fig. 3b). Survival estimates for each of the five subgroups are presented in the supplemental material (Supplementary Fig. S1).
Fig. 3. Survival profiles for women diagnosed with interval cancer through BreastScreen Norway.
a Breast cancer-specific survival tree, % indicates the proportion of women who died of breast cancer within each subgroup. b Ten-year breast cancer-specific survival, stratified using the subgroups created by the survival tree.
The 3237 women with breast cancer detected outside of screening contributed 25,845 person-years to the analysis and four subgroups were identified for these women: T1 and M0 (OS Subgroup 1); T2 and M0 (OS Subgroup 2); T3 or T4, and M0 (OS Subgroup 3); and M1 (OS Subgroup 4; Fig. 4a). The corresponding 10-year breast cancer survival for OS Subgroups 1 through 4 were 91.8% (95% CI: 90.5–93.1); 79.3% (95% CI: 76.8–81.7); 52.2% (95% CI: 44.9–60.0); and 23.4% (95% CI: 19.1–28.5) (Fig. 4b).
Fig. 4. Survival profiles for women diagnosed outside of BreastScreen Norway.
a Breast cancer-specific survival tree, % indicates the proportion of women who died of breast cancer within each subgroup. b Ten-year breast cancer-specific survival, stratified using the subgroups defined by the survival tree.
Discussion
This closed cohort study described the histopathology of 16,396 breast cancers diagnosed in women targeted by BreastScreen Norway and used recursive partitioning to identify survival profiles for subgroups of women attending and not attending the screening program. We found that 61% of cancers were screen-detected, 19% were interval cancers, and 20% were detected outside screening. Ten years after diagnosis, 6.3% of women with screen-detected cancers had died from breast cancer, compared to 17.4% of those with interval cancer and 20.9% of women with cancer detected outside screening.
Univariable analyses revealed a higher proportion of fatal cancers among prevalent screen-detected cancers than “subsequent regular” screen-detected cancers (7.4% vs 5.7%, respectively). Although the relative risk between these groups is roughly 30%, the risk difference is small (1.7 percentage points). The proportion of fatal cancer was particularly low for women with histologic Grade 1, T1a/b, or node-negative screen-detected cancers. The recursive partitioning algorithm identified that women with node-negative screen-detected cancer had excellent 10-year breast cancer-specific survival (96%). This group represented 76% of women with screen-detected cancer. Women with interval cancer whose disease was histologic Grade 1 or 2, node-negative, and did not present with distant metastases also had excellent 10-year breast cancer-specific survival (95%); this group represented 38% of women diagnosed with interval cancer. Combined, these groups represented 53% of the cohort, suggesting that many women diagnosed with breast cancer had an excellent prognosis when offered the best available treatment according to national guidelines.
Across all modes of detection, the proportion of fatal cancers increased with increasing histologic grade, tumour diameter, and lymph node involvement. The highest proportion of fatal cancers was observed among women with distant metastases at diagnosis, and all subgroups with distant metastases at diagnosis had <25% 10-year breast cancer-specific survival. However, the proportion of women with distant metastases at diagnosis was low, particularly for women with screen-detected cancers.
The variables selected by the recursive partitioning algorithm to define women with differing survival profiles are well-established prognostic factors for breast cancer, underlining the validity of the approach used [2, 4, 25]. Nonetheless, the present approach did not select T-category to define prognostic subgroups for women with screen-detected or interval cancer. This does not imply that tumour diameter is not an important prognostic factor—the opposite is well-established [2, 4, 25]. It rather implies that the T-category did not provide more information about 10-year breast cancer-specific survival than lymph node involvement/histologic grade for the women with screen-detected/interval cancers included in our cohort. Indeed, nodal status and histologic grade have been shown to be stronger prognostic factors than tumour diameter in the Nottingham Prognostic Index and univariate analyses in our study showed a larger difference in the proportion of fatal cancers across N-categories than across T-categories for these modes of detection [25].
This closed cohort study used 10-year breast cancer-specific survival to evaluate non-fatal and fatal cases. Using an open cohort with longer follow-up and more contemporary cases may have improved external validity, but the closed cohort design helped ensure that the proportion of non-fatal cancers generally mirrored the 10-year breast cancer-specific survival. This improves the interpretability and face validity of the results produced by the recursive partitioning algorithm, which we consider important since this data-driven approach is novel in descriptive epidemiology. A limitation of using single trees, such as those presented in our study, is that small changes in the data used to create the trees can lead to different variables included in the final models. This instability is inherent to the recursive partitioning method and is the “cost” of the intuitive output that trees offer [26]. More advanced methods called ensemble methods can reduce this instability but offer less intuitive interpretation than single trees [20, 26]. Our descriptive study prioritized the simple interpretation of single trees, but future research may consider using ensemble methods such as random forests to evaluate whether a more advanced approach may be suitable for predicting survival outcomes for individuals. Such research could also include information from a more contemporary cohort where not all women have 10 years of follow-up.
Another limitation of this study is the absence of data on human epidermal growth factor receptor 2 (HER2) status and tumor cell proliferation (by Ki67), since this information is largely missing in the Cancer Registry of Norway’s databases for cases diagnosed ≤2009. Such data may be useful for further stratification of survival groups for women with screen-detected breast cancer, particularly ER-positive cases [6]. Indeed, HER2 and Ki67 are currently used as surrogate markers for molecular breast cancer subtypes, which are known to have different survival profiles, and HER2 is also used to define the AJCC prognostic stage groups [4, 6, 27]. The focus of the current study was on women’s prognosis at diagnosis. Information on treatment was not included and this is a limitation since survival is influenced by treatment, however, treatment is determined by tumour histopathology and not explicitly by mode of detection. In terms of external validity, women diagnosed after 2011 would have increased access to neoadjuvant therapy and newer therapies such as immunotherapy than women included in this study [28]. Moreover, breast cancer mortality in Norway has continued to decrease over time and, therefore, the breast cancer-specific survival estimates we presented are likely to be conservative estimates for patients diagnosed today [29]. Nonetheless, as noted above, the variables selected by the recursive partitioning algorithm are well-established prognostic factors for breast cancer and have been shown to be relevant for modern cohorts [6].
Lastly, we did not have information on whether an interval cancer or cancer detected outside of screening was asymptomatic. Opportunistic screen-detected cancers likely have a prognosis similar to screen-detected cancers in BreastScreen Norway but were classified as interval cancers or cancers detected outside screening in our study because reporting information about opportunistic (private) screening is not mandatory in Norway and this information was unavailable for our study. The proportion of fatal cancers among symptomatic interval cases and tumours detected outside of screening is therefore likely somewhat higher than we observed.
A strength of our study is the use of recursive partitioning: this relatively simple algorithm evaluated many linear combinations of patient and histopathologic variables against breast cancer survival to produce simple and applicable tree diagrams describing survival-based subgroups. We used survival curves to further describe the profiles of the subgroups identified by the algorithm; this presentation is familiar to researchers and clinicians and supports the validity of our results. Presenting 10-year breast cancer-specific survival adjusted for the competing risk of death from other causes represents another strength of our study because it represents the real-world proportion of women who survived their breast cancer. (It should be noted, however, that some women who survived their breast cancer will have died of other causes during the follow-up period.) On the other hand, the often-used Kaplan–Meier estimate of breast cancer-specific survival is biased in the presence of competing risks [30–33]. In this context, the Kaplan–Meier method would have estimated the probability of surviving from breast cancer in a hypothetical world where it is not possible to die from other causes (so-called net survival) [33]. Although such hypothetical estimates can be useful for making comparisons over time or between countries, real-world estimates are most relevant for communicating with patients.
Conclusion
Our cohort study included data from women diagnosed with breast cancer in 2011 and earlier and considered person and tumour characteristics to define subgroups with differing survival profiles, stratified by mode of detection. Only tumour characteristics (histologic grade, tumour diameter, lymph node involvement, and distant metastasis at diagnosis) were ultimately selected to describe these subgroups. Approximately half of women were classified into subgroups with excellent (95–96%) 10-year breast cancer-specific survival. The proportion of fatal breast cancer 10 years after diagnosis was 6.3% among women with screen-detected cancers, versus 17.4% among women with interval cancer and 20.9% among women with cancer detected outside screening. The results of our study are descriptive, but clinicians may use them to inform and potentially reassure women diagnosed with breast cancer about 10-year outcomes.
Disclaimer
Data from the Cancer Registry of Norway (CRN) have been used in this publication. The authors are solely responsible for the interpretation and presentation of the results in this paper.
Supplementary information
Acknowledgements
The authors would like to thank Luca Pestarino for sharing his insights on recursive partitioning.
Author contributions
Conceptualization: SH, SRH and LAA. Methodology: SH, SRH, LAA and KMT. Software: KMT. Validation: KMT, SH, SRH and LAA. Formal analysis: KMT. Investigation: all authors. Resources: SH. Data curation: KMT and SH. Writing—original draft preparation: KMT, SH and SRH. Writing—review and editing: KMT, SH, SRH and LAA. Visualization: KMT. Supervision: not relevant. Project administration: SH. Funding acquisition: not relevant.
Funding
The author(s) received no specific funding for this work.
Data availability
The data underlying this article cannot be shared publicly due to patient privacy. The data can be shared for research purposes on request to the Cancer Registry of Norway’s data delivery unit via Helsedata.no (https://helsedata.no/).
Competing interests
SH is the head of BreastScreen Norway. The remaining authors declare no competing interests.
Ethics approval and consent to participate
This study has been reviewed by the privacy ombudsman at the Oslo University Hospital (PVO 20/12601) and was performed in accordance with the Declaration of Helsinki. It has a legal basis in accordance with Articles 6 (1) (e) and 9 (2) (j) of the GDPR. The data were disclosed with legal basis in the Cancer Registry Regulations section 3-1 and the Personal Health Data Filing System Act section 19 a to 19 h.
Consent for publication
Not applicable.
Footnotes
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
The online version contains supplementary material available at 10.1038/s41416-023-02512-7.
References
- 1.International Agency for Research on Cancer (IARC). Estimated age-standardized incidence and mortality rates (World) in 2020, World, females, ages 20-84 (excl. NMSC). 2020. https://gco.iarc.fr/today/online-analysis-multi-bars?v=2020&mode=cancer&mode_population=countries&population=900&populations=900&key=asr&sex=2&cancer=39&type=0&statistic=5&prevalence=0&population_group=0&ages_group%5B%5D=4&ages_group%5B%5D=16&nb_items=10&group_cancer=0&include_nmsc=0&include_nmsc_other=1&type_multiple=%257B%2522inc%2522%253Atrue%252C%2522mort%2522%253Atrue%252C%2522prev%2522%253Afalse%257D&orientation=horizontal&type_sort=0&type_nb_items=%257B%2522top%2522%253Atrue%252C%2522bottom%2522%253Afalse%257D. Accessed February 1, 2023.
- 2.Breast Cancer Screening. IARC handbook of cancer prevention. Vol. 15. Lyon, France: International Agency for Research on Cancer; 2016.
- 3.Dibden A, Offman J, Duffy SW, Gabe R. Worldwide review and meta-analysis of cohort studies measuring the effect of mammography screening programmes on incidence-based breast cancer mortality. Cancers. 2020;12:976. doi: 10.3390/cancers12040976. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Badve SS, Beitsch PD, Bose S, Byrd D, Chen VW, Connolly JL, et al. Part XI, breast. In: Amin MB, et al., editors. AJCC Cancer Staging Manual, 8th edn. New York, USA: Springer International Publishing; 2017. p 587–636.
- 5.Tabar L, Duffy SW, Vitak B, Chen H-H, Prevost TC. The natural history of breast carcinoma. Cancer. 1999;86:449–62. doi: 10.1002/(SICI)1097-0142(19990801)86:3<449::AID-CNCR13>3.0.CO;2-Q. [DOI] [PubMed] [Google Scholar]
- 6.Johansson ALV, Trewin CB, Fredriksson I, Reinertsen KV, Russnes H, Ursin G. In modern times, how important are breast cancer stage, grade and receptor subtype for survival: a population-based cohort study. Breast Cancer Res. 2021;23:17. doi: 10.1186/s13058-021-01393-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Broeders M, Moss S, Nystrom L, Njor S, Jonsson H, Paap E, et al. The impact of mammographic screening on breast cancer mortality in Europe: a review of observational studies. J Med Screen. 2012;19:14–25. doi: 10.1258/jms.2012.012078. [DOI] [PubMed] [Google Scholar]
- 8.Duffy SW, Tabar L, Yen AM, Dean PB, Smith RA, Jonsson H, et al. Beneficial effect of consecutive screening mammography examinations on mortality from breast cancer: a prospective study. Radiology. 2021;299:541–7. doi: 10.1148/radiol.2021203935. [DOI] [PubMed] [Google Scholar]
- 9.Shen Y, Yang Y, Inoue LY, Munsell MF, Miller AB, Berry DA. Role of detection method in predicting breast cancer survival: analysis of randomized screening trials. J Natl Cancer Inst. 2005;97:1195–203. doi: 10.1093/jnci/dji239. [DOI] [PubMed] [Google Scholar]
- 10.Fortin J, Leblanc M, Elgbeili G, Cordova MJ, Marin MF, Brunet A. The mental health impacts of receiving a breast cancer diagnosis: a meta-analysis. Br J Cancer. 2021;125:1582–92. doi: 10.1038/s41416-021-01542-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Cancer Registry of Norway. Kvalitetsmanualen i Mammografiprogrammet [Quality manual for BreastScreen Norway]. Oslo, Norway: Cancer Registry of Norway; 2003.
- 12.European Commission Initiative on Breast Cancer. European guidelines on breast cancer screening and diagnosis. 2022. https://healthcare-quality.jrc.ec.europa.eu/ecibc/european-breast-cancer-guidelines. Accessed November 8, 2022.
- 13.Lauby-Secretan B, Scoccianti C, Loomis D, Benbrahim-Tallaa L, Bouvard V, Bianchini F, et al. Breast-cancer screening–viewpoint of the IARC Working Group. N. Engl J Med. 2015;372:2353–8. doi: 10.1056/NEJMsr1504363. [DOI] [PubMed] [Google Scholar]
- 14.Hofvind S, Tsuruda K, Mangerud G, Ertzaas AK, Holen Å, Pedersen K, et al. The Norwegian Breast Cancer Screening Program, 1996-2016: celebrating 20 years of organised screening in Norway. Oslo, Norway: Cancer Registry of Norway; 2017.
- 15.Skaane P, Skjennald A. Screen-film mammography versus full-field digital mammography with soft-copy reading: randomized trial in a population-based screening program—the Oslo II Study. Radiology. 2004;232:197–204. doi: 10.1148/radiol.2321031624. [DOI] [PubMed] [Google Scholar]
- 16.Larsen IK, Smastuen M, Johannesen TB, Langmark F, Parkin DM, Bray F, et al. Data quality at the Cancer Registry of Norway: an overview of comparability, completeness, validity and timeliness. Eur J Cancer. 2009;45:1218–31. doi: 10.1016/j.ejca.2008.10.037. [DOI] [PubMed] [Google Scholar]
- 17.Lydersen S, Fagerland MW, Laake P. Categorical data and contingency tables. In: Veierød MB, Lydersen S, Laake P, editors. Medical statistics in clinical and epidemiological research, 1st edn. Oslo, Norway: Gylendal Norsk Forlag; 2012. p 48–89.
- 18.LeBlanc M, Crowley J. Relative risk trees for censored survival data. Biometrics. 1992;48:411–25. doi: 10.2307/2532300. [DOI] [PubMed] [Google Scholar]
- 19.Therneau T, Atkinson E. An introduction to recursive partitioning using the RPART Routines. 2022. https://cran.r-project.org/web/packages/rpart/vignettes/longintro.pdf.
- 20.Zhang H, Singer BH. Recursive partitioning and applications. 2nd edn. New York NY, USA: Springer; 2010.
- 21.R Core Team. R: a language and environment for statistical computing v. 4.1.3. Vienna, Austria: R Foundation for Statistical Computing; 2022.
- 22.Therneau T, Atkinson B. rpart: Recursive partitioning and regression trees. R package version 4.1.19 (2022). https://cran.r-project.org/package=rpart.
- 23.Milborrow S. rpart.plot: Plot ‘rpart’ models: an enhanced version of ‘plot.rpart’. R package version 3.1.1 (2022). https://cran.r-project.org/package=rpart.plot.
- 24.Coviello V, Boggess M. Cumulative incidence estimation in the presence of competing risks. Stata J. 2004;4:103–12. doi: 10.1177/1536867X0400400201. [DOI] [Google Scholar]
- 25.Haybittle JL, Blamey RW, Elston CW, Johnson J, Doyle PJ, Campbell FC, et al. A prognostic index in primary breast cancer. Br J Cancer. 1982;45:361–6. doi: 10.1038/bjc.1982.62. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Hastie T, Tibshirani R, Friedman J. Chapter 9: additive models, trees, and related methods. In: Hastie T, Tibshirani R, Friedman J. The elements of statistical learning, 2nd edn. New York, NY, USA: Springer; 2009. p 295–336.
- 27.Goldhirsch A, Winer EP, Coates AS, Gelber RD, Piccart-Gebhart M, Thurlimann B, et al. Personalizing the treatment of women with early breast cancer: highlights of the St Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2013. Ann Oncol. 2013;24:2206–23. doi: 10.1093/annonc/mdt303. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Debien V, De Caluwe A, Wang X, Piccart-Gebhart M, Tuohy VK, Romano E, et al. Immunotherapy in breast cancer: an overview of current strategies and perspectives. NPJ Breast Cancer. 2023;9:7. doi: 10.1038/s41523-023-00508-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Larønningen S, Ferlay J, Beydogan H, Bray F, Engholm G, Ervik M, et al. NORDCAN: period, age-specific rate per 100 000, mortality, females, age [40-74] (Norway, Breast). Association of the Nordic Cancer Registries, Cancer Registry of Norway. 2023. https://nordcan.iarc.fr/en/dataviz/cohorts?cancers=180&sexes=2&populations=578&age_start=8&years_available=1943_2020&types=1&cohort=period (Data version 9.2 - June 23, 2022). Accessed August 26, 2023.
- 30.Andersen PK, Geskus RB, de Witte T, Putter H. Competing risks in epidemiology: possibilities and pitfalls. Int J Epidemiol. 2012;41:861–70. doi: 10.1093/ije/dyr213. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Satagopan JM, Ben-Porat L, Berwick M, Robson M, Kutler D, Auerbach AD. A note on competing risks in survival data analysis. Br J Cancer. 2004;91:1229–35. doi: 10.1038/sj.bjc.6602102. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Austin PC, Lee DS, Fine JP. Introduction to the analysis of survival data in the presence of competing risks. Circulation. 2016;133:601–9. doi: 10.1161/CIRCULATIONAHA.115.017719. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Gooley TA, Leisenring W, Crowley J, Storer BE. Estimation of failure probabilities in the presence of competing risks: new representations of old estimators. Stat Med. 1999;18:695–706. doi: 10.1002/(SICI)1097-0258(19990330)18:6<695::AID-SIM60>3.0.CO;2-O. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data underlying this article cannot be shared publicly due to patient privacy. The data can be shared for research purposes on request to the Cancer Registry of Norway’s data delivery unit via Helsedata.no (https://helsedata.no/).