Interval Breast Cancer Rates and Tumor Characteristics in the Prospective Population-based Malmö Breast Tomosynthesis Screening Trial

Introduction

Digital breast tomosynthesis (DBT) has been shown to have higher sensitivity in screening than digital mammography (DM) (1–5). The additional invasive cancers detected at DBT seem to have similar or more favorable tumor characteristics than cancers detected at DM, such as higher proportions of small, spiculated, low-grade, luminal A–like tumors (4,6–8). To determine whether the additional cancers detected at DBT constitute a potential screening benefit and do not simply contribute to overdiagnosis, interval cancer rates in DBT screening should be compared with those in DM screening (9). Interval cancer reporting is required in many screening programs as an indicator of effectiveness (10). If there is a decrease in the interval cancer rate when DBT is used, it might be attributed to improved detection of rapidly growing cancers with poorer prognosis, possibly contributing to lower breast cancer mortality. The interval cancer rates found in previous prospective DBT trials have been the same as or slightly lower than rates shown by DM (11,12), but those trials were not designed and powered to primarily evaluate interval cancer rates. Retrospective trials have shown similar results with no statistically significant differences between interval cancer rates (13,14). To our knowledge, randomized controlled screening trials of DBT with interval cancer rates as end point have not yet published such data (15–17). In general, interval cancers are more aggressive than screen-detected cancers because they are similar to symptomatic breast cancer in women who are not screened (18). Nonfavorable tumor characteristics, such as node-positive and nonluminal A cancers, have also been reported for interval cancers when screening with DBT (11,13,19).

The prospective population-based Malmö Breast Tomosynthesis Screening Trial (MBTST) was designed to assess the sensitivity and specificity of one-view DBT in screening and to compare it with two-view DM (1). We hypothesized that the interval cancer rate in DBT screening would be lower than that in DM screening.

The purpose of our study was to compare the interval cancer rate in the MBTST with that in a contemporary DM screened control group, and to describe the characteristics of interval cancers.

Materials and Methods

The prospective, population-based, single-arm MBTST compared one-view DBT (mediolateral oblique) with two-view DM (mediolateral oblique and craniocaudal) in breast cancer screening. A random sample of 21 691 women selected from the screening registry in the city of Malmö, Sweden were invited by letter between January 2010 and February 2015. Exclusion criteria were pregnancy, non-Swedish speakers, or non-English speakers. Written informed consent was obtained. The trial was approved by the regional ethics review board at Lund University (trial protocol at https://www.ClinicalTrials.gov: NCT01091545). The tomosynthesis equipment was provided by Siemens Healthcare. The authors had control of the data and all information submitted for publication, and none of the authors was employed by Siemens Healthcare.

The women participating in the MBTST underwent one-view DBT and two-view DM by using a wide-angle (50°) system (Mammomat Inspiration, Siemens Healthcare) at one screening occasion. The images were read in two separate reading groups, the DBT reading group (reading step 1, one-view DBT; step 2, with current DM craniocaudal view; step 3, with any previous DM images) and the DM reading group (Fig 1), with two readers in each group. No synthetic images or computer-aided detection were used. Women were screened with DM before and after the trial.

Figure 1:

Download as PowerPoint

Results from the MBTST have been reported in previous publications regarding screening accuracy (1,20), tumor characteristics, radiographic appearance (8), false-positive recalls (21,22), and density measurements (23,24). The number of interval cancers and some tumor characteristics have also been reported briefly (1), whereas our study presents final detailed interval cancer data, including comparison with a control group.

Results

In the MBTST, 14 848 women (mean age, 57 years ± 10 [standard deviation]) were screened. A total of 137 women in the MBTST had screen-detected cancers. There were 43 769 women screened with DM in the contemporary screened population (mean age, 53 years ± 11). In the contemporary screening population, 259 women were found to have screen-detected cancer. Two unique age- and screen-date matches were found for 13 639 women in the MBTST (mean age at screening, 56 years ± 10), of which 21 women were diagnosed with interval cancer. Seventy-six interval cancers were diagnosed in the 26 738 women in the matched control group (mean age at screening, 56 years ± 10) (Table 1). The characteristics of cancers detected at screening in the contemporary population and in the control group are in Tables E1 and E2 (online).

Table 1: Population Characteristics and Cancer Detection Rates in the MBTST and in the Age-matched Control Group

The interval cancer rate in the MBTST was 1.6 per 1000 screened women (21 of 13 639; 95% CI: 1.0, 2.4), and in the control group it was 2.8 per 1000 screened women (76 of 26 738; 95% CI: 2.2, 3.6). The age-adjusted odds ratio was 0.6 (95% CI: 0.3, 0.9; P = .02), meaning that the odds of interval cancer in the MBTST is 40% lower compared with the odds of interval cancer in the matched control group. The interval cancer rate in women aged younger than 55 years at screening was 1.3 per 1000 screened women (eight of 6289; 95% CI: 0.6, 2.5) in the MBTST compared with 2.6 per 1000 screened women in the control group (33 of 12 541 [95% CI: 1.8, 3.7]; conditional odds ratio, 0.5 [95% CI: 0.2, 1.1]; P = .07). In women aged 55 years or older at screening, the interval cancer rate in the MBTST was 1.8 per 1000 screened women (13 of 7089; 95% CI: 1.0, 3.1) compared with 3.0 per 1000 screened women in the control group (43 of 14 197 [95% CI: 2.2, 4.1]; conditional odds ratio, 0.6 [95% CI: 0.3, 1.1]; P = .1). There was no difference in DM screen-detected cancer rate in the MBTST compared with the control group (6.5 per 1000 screened women [87 of 13 369; 95% CI: 5.2, 8.0] vs 6.6 per 1000 screened women [176 of 26 738; 95% CI: 5.6, 7.6], respectively; conditional odds ratio, 1.0 [95% CI: 0.8, 1.3]; P = .93) (Table 1). Seven of 21 (33%) interval cancers in the MBTST and 36 of 76 (47%) interval cancers in the control group were diagnosed within 1 year (ie, less than 365 days) after screening.

Two sensitivity analyses were performed in addition to the matched analyses by using the contemporary DM screening occasions (n = 43 769) for comparison. Age-adjusted logistic regression with the contemporary screens as reference showed a lower interval cancer rate in the MBTST (1.5 per 1000 screened women; 22 of 14 848; 95% CI: 0.9, 2.2) compared with the contemporary screens (2.4 per 1000 screened women; 105 of 43 769 [95% CI: 2.0, 2.9]; odds ratio, 0.6 [95% CI: 0.3, 0.9]; P = .02). Age-weighted analysis with the contemporary screens as reference also showed a lower interval cancer rate in the MBTST compared with the contemporary screens. There were 1.5 per 1000 screened women (22 of 14 848; 95% CI: 0.9, 2.2) versus 2.5 per 1000 screened women (108 of 43 769 [95% CI: 2.0, 3.0]; odds ratio, 0.6 [95% CI: 0.5, 0.7]; P ≤ .001).

Interval Cancer Characteristics

Of the 21 interval cancers diagnosed in the MBTST, 90% (19 of 21) were invasive compared with 95% (72 of 76) in the control group. Neoadjuvant chemotherapy was administered to 16% (three of 19) of women with invasive interval cancers in the MBTST and to 10% (seven of 72) of women with invasive interval cancer in the control group. A large proportion of the invasive interval cancers consisted of invasive ductal carcinomas in both the MBTST and the control group (90% [17 of 19] and 80% [58 of 72], respectively). The mean invasive tumor size, after excluding tumors administered neoadjuvant treatment, was 15 mm ± 7 in the MBTST and 20 mm ± 10 in the control group. The mean tumor size before neoadjuvant treatment, assessed at imaging, was 21 mm ± 4 and 32 mm ± 10 in the MBTST and in the control group, respectively. The proportions of lymph node–positive invasive interval cancers diagnosed in the MBTST and in the control group were 37% (seven of 19) compared with 44% (32 of 72) (Table 2).

Table 2: Histopathologic Characteristics of Interval and Screen-detected Cancers in the Malmö Breast Tomosynthesis Screening Trial and in the Age-matched Control Group

A slightly higher proportion of positive progesterone receptor status (63% [12 of 19] vs 53% [38 of 72], respectively), and a slightly lower proportion of high proliferation cancers (ie, Ki-67 ≥20) (63% [12 of 19] vs 75% [54 of 72], respectively) were observed in interval cancers in the MBTST compared with the control group (Table 3). Positive estrogen receptor status (84% [16 of 19] vs 78% [56 of 72], respectively) and human epidermal growth factor 2 gene amplification (10% [two of 19] vs 15% [11 of 72]) showed similar distributions in the two groups.

Table 3: Immunohistochemical Profiles and St Gallen Subtypes of Age-matched Invasive Interval Cancers

The proportions of luminal A–like subtype cancer in interval cancers were low, both in the MBTST and in the control group (26% [five of 19] vs 17% [12 of 72]). The most common subtype in both groups was luminal B–like HER2 negative subtype in the MBTST group (47%; nine of 19) and 54% (39 of 72) in the control group. Three (16%; three of 19) and 10 (14%; 10 of 72) interval cancers were triple-negative subtype in the MBTST and in the control group, respectively. Two image examples are provided in Figures 2 and 3.

Figure 2a:

Download as PowerPoint

Figure 2b:

Download as PowerPoint

Figure 2c:

Download as PowerPoint

Figure 2d:

Download as PowerPoint

Figure 2e:

Download as PowerPoint

Figure 3a:

Download as PowerPoint

Figure 3b:

Download as PowerPoint

Figure 3c:

Download as PowerPoint

Figure 3d:

Download as PowerPoint

Figure 3e:

Download as PowerPoint

Figure 3f:

Download as PowerPoint

The one interval cancer in the MBTST with no matches in the control group was an invasive node-negative triple-negative ductal carcinoma that was undergoing neoadjuvant therapy.

Discussion

Interval cancer rates can be measured to determine whether the additional cancers depicted at digital breast tomosynthesis (DBT) contribute to a screening benefit. The relevance of addressing the evidence gap on the effect of DBT screening on interval cancer rates has been emphasized (27,28). We provided additional data by calculating the interval cancer rate in a prospective, population-based, one-view DBT screening trial and in a contemporary digital mammography (DM) screened control group. The interval cancer rate in the Malmö Breast Tomosynthesis Screening Trial (MBTST) was 1.6 per 1000 screened women (95% CI: 1.0, 2.4) compared with 2.8 per 1000 screened women (95% CI: 2.2, 3.6) in the control group. The age-adjusted odds ratio was 0.6 (95% CI: 0.3, 0.9; P = .02). The lower interval cancer rate after screening with DBT and DM compared with DM might translate into additional benefits, such as earlier detection and further reduction in breast cancer mortality. Considerable proportions of invasive interval cancers in the trial were lymph node–positive subtype (37%; seven of 19), high Ki-67 (63%; 12 of 19), and luminal B–like HER2-negative subtype (47%; nine of 19); features associated with a less favorable prognosis.

The MBTST was not specifically designed to assess interval cancer rates. Therefore, the interval cancer rate reduction with wide confidence intervals for the odds ratios should be interpreted with caution. However, the results of the sensitivity analyses are consistent with the age-matched approach, indicating a robustness of the data. The results differ from those reported in previous screening studies, where no statistically significant differences in interval cancer rates have been reported (Table 4). Because, to our knowledge, our study is the only prospective DBT trial that evaluated one-view wide-angle DBT, we cannot exclude that this screening approach could partially explain the different results. Furthermore, our trial includes women ages 40–49 years and to our knowledge the interval cancer rate in our control group is the highest in the trials published thus far.

Table 4: Overview of Interval Cancer Rates in Digital Breast Tomosynthesis Studies

The characteristics of the interval cancers in the MBTST and in the control group were in general similar but the tendency toward smaller size and less lymph node involvement in the MBTST could indicate an effect of earlier detection with DBT. Larger studies or meta-analyses are needed to confirm this (27). Previous prospective studies report similar results as ours. In the Oslo Tomosynthesis Screening Trial (12), 38.0% (43 of 113) invasive interval cancers were node positive. Bahl et al (19) reported 139 interval cancers in DBT screening, 44 of which had positive nodes (31.6%). Of the eight invasive interval cancers found in the Screening with Tomosynthesis or Standard Mammography trial (11), 75% (six of eight) had high Ki-67 status.

Our study had several limitations. The number of interval cancers in the MBTST were few, which makes it difficult to study rate differences, especially in subgroups. The age matching led to an even smaller number of interval cancers. The age stratification was on the basis of the different screening intervals in the age groups because screening intervals affect the interval cancer rate, and therefore were deemed appropriate even though it could split the matching in the subgroup analyses. To our knowledge, age and cancer detection rates have not been reported in control groups of other prospective trials, and we do not know if they exhibited similar differences (12,27,29). Women who chose not to participate in the MBTST or who were excluded from the trial underwent regular DM screening and were included in the control group, which could imply selection biases. The lack of breast density data, an interval cancer risk factor (30), was a limitation. Breast density and menopausal status are not routinely collected in the Swedish screening program. These parameters may partly explain the reduction in interval cancer rate. However, age matching may have compensated to some extent because both breast density and menopausal status are correlated with age. Another limitation was the different number of readers in the MBTST (ie, four readers) because of the paired trial design, compared with the control group (ie, two readers). More readers increase sensitivity, which likely contributed to the lower interval cancer rate in the MBTST. Eight cancers were detected in the DM reading group only and one or several of those could have appeared as interval cancers, possibly resulting in a lower interval cancer rate in the trial. Furthermore, the MBTST was a single-center, single-vendor, one-view, wide-angle DBT trial in a Swedish setting with screening intervals and age inclusion that may differ from those in other screening programs. A prespecified matched control group and risk factor assessment would have limited potential biases and should be considered when designing future studies on the screening benefits of DBT.

In conclusion, a statistically significant reduction in interval cancer rate was observed in our prospective, population-based, one-view digital breast tomosynthesis (DBT) screening trial compared with a large contemporary control group screened with digital mammography. Our study suggests that there is an effect of DBT on interval cancer rate in population screening and that it could translate into additional screening benefit.