Introduction

Aims

A breast cancer (BC) that emerges following a negative mammographic screen is referred to as an interval BC.1 In this overview, we describe the epidemiology, radiology and biological characteristics of interval BCs in population mammography screening, highlighting published research from the most recent decade. The aims of the review were to provide an update on interval BCs that extends both our work on radiological surveillance of interval BCs1 and that from other researchers that have quantified interval BC rates,2,3,4 to elucidate evidence on interval BCs following transition to digital mammography screening, and to identify knowledge gaps that warrant further research.

Background and definitions

An interval BC refers to a cancer that presents after a ‘normal’ screening mammogram and before the next scheduled mammogram, in other words a BC that arises or is diagnosed in the inter-screening interval [see also Fig. 1].1 This definition may be qualified by specifying an interval case as an invasive BC,5 given that the vast majority of interval cases are invasive malignancies and much of the routinely reported data on interval BC rates is based on invasive BC. In addition, some qualify the definition further by specifying that interval BCs are those that arise clinically2, 5 in the inter-screening interval—although that would be the likely presentation for almost all interval BCs it should be noted that a BC identified in the inter-screening interval would still be classified as such irrespective of how it came to be diagnosed. Factors that have been associated with increased risk of an interval BC in screened women include high mammographic breast density,6,7,8 current use of hormone replacement therapy,8, 9 young relative to older age (partly reflecting confounding from breast density), however, absolute incidence rates are higher in older women given higher underlying BC rates,8 previous false-positive mammography,10 and a family history of BC.8, 10, 11

Fig. 1
figure 1

Summary of background definitions and themes

Full size image

Because interval BCs are representative of the sensitivity of population breast screening, and given that they are an adverse outcome for women partaking in screening, surveillance of interval BCs is routinely practiced in many screening programs. Surveillance comprises epidemiologic measures (such as interval BC rates)1, 3 to monitor the frequency of interval cases, and may be complemented by radiological surveillance as part of quality assurance in organised screening programs.1 Various methodological and analytic parameters can substantially influence estimates of interval BC rates and other epidemiologic measures of interval BCs, as highlighted by several investigators.3, 12 These include variability in the definition of an interval BC (whether based on invasive BC or whether ductal carcinoma in-situ (DCIS) cases are included); whether false-negative assessment cases and lapsed attenders are counted or excluded; the adequacy of ascertainment of interval cancers, hence also the adequacy of cancer notification and registration; and the duration of follow-up for ascertainment of cases.3, 12 Importantly, underlying BC rates or burden in the population also affects interval BC rates. For these reasons, epidemiologic measures of interval BCs are best suited for monitoring within screening services or programs because comparisons between screening programs and countries is limited by heterogeneity in the above-described variables that affect quantitative estimates of interval BCs.

Radiological surveillance is a qualitative form of surveillance that defines and measures the extent that interval BCs represent screen-reading ‘errors’ as opposed to being non-detectable cancers at mammography screening. Radiological surveillance entails review of the mammograms taken at the time the interval BC is identified (usually at clinical presentation, hence the diagnostic mammograms) and the pre-diagnosis mammographic screen (the ‘negative’ index screen) and an interpretative judgement to classify each case into pre-defined categories.1, 13,14,15 These categories may vary in definition, however, most include a ‘true interval’ category (where the cancer is not visible at the index screen but becomes visible at the diagnostic mammogram) and a ‘missed’ interval BC being the equivalent of a false-negative (where the cancer is visible on the index mammogram but is not recalled or is misinterpreted) and is at times referred to as screening error.1, 13,14,15 Various methods have been used to perform radiological surveillance, as described in a review by Houssami et al.1 with potential biases inherent in the review strategy and the extent that readers are informed that they are evaluating interval BC cases.1 Notwithstanding the methodological limitations of radiological surveillance, it provides insights into screening quality and on how screening could be improved.

In the present review, we consider both epidemiological and radiological aspects of interval BCs, and complement these with information on tumour prognostic characteristics of interval BCs, to define common themes as well as evidence gaps, to enhance our understanding of interval BCs and inform research priorities.

Results

Epidemiologic surveillance

Table 1 presents a summary of epidemiologic measures for interval cancers including interval BC rates, which were the most commonly reported estimates for routine screening monitoring.4, 13, 15,16,17,18,19,20,21,22,23,24,25,26,27,28,29 The table highlights the broad variability in both interval BC rates and cancer detection rates at screening, both of which are partly driven by underlying cancer rates in the populations reported in these studies. There is wide variability in the overall interval BC rates, ranging between 7.0 and 49.3 per 10,000 screens, partly explained by data shown for screening rounds (initial and repeat screens) and the duration and year of the inter-screening intervals; where reported, data for the inter-screening interval are presented by yearly rates for biennial or triennial screening. If restricted to studies of annual screening or to year 1 data from biennial screening programs, there is evidence that interval BC rates are consistently <8/10,000 screens. The majority of studies in Table 1 report data for biennial screening programs: the interval BC rates spanning the two years between screens are within the range of 8.4 to 21.1 per 10,000 screens, with the larger proportion of the estimated interval BC rates occurring in the second year of a two-yearly interval. The evidence also consistently shows that interval BC rates are higher at repeat (incident) rounds than initial (prevalent) screening rounds.

Table 1 Epidemiological measures of the frequency of interval breast cancers in population mammography screening
Full size table

For the majority of studies (based on biennial screening) interval BCs represent around 17–30% of the cancers occurring in screening participants as summarised in the simple proportion in the last column of Table 1; that proportion is relatively lower for annual (14.7% based on one study) and higher for triennial (32–38%) screening intervals. The proportional interval BC rate, also shown in Table 1, is not reported by all studies because this measure requires estimating the expected underlying incidence in the absence of screening, so may not be feasible to calculate in contemporary screening practice.2

Mammography screening interventions associated with reduction in interval BC rates

A study conducted in the United Kingdom’s screening program, based on women aged 50–64 screened between 2003 and 2005, reported that two-view mammography (at the last routine screen) was associated with a reduction in interval BC rates of 6.8/10,000 screens compared with one-view mammography.30 The investigators concluded that this suggests that two-view (instead of one-view) mammography at incident screening was accompanied by a 15–20% reduction in interval BC rates. An earlier study by Seigneurin et al.27 reported similar evidence from the French population breast screening program, based on women aged 50–69, in a comparison of two time frames reflecting transition from one-view to two-view mammography: a reduction of 8.6 interval BCs per 10,000 screens was associated with two-view mammography, with an estimated 36% difference in relative risk of interval BC at 24 months for two vs. one view mammography (Table 1). Weber et al.13 reported a study of the Southern screening region of the Dutch program, in which the use of digital (compared to film-screen) mammography was shown to be associated with a modest but significant reduction in interval BC rates of 3/10,000 screens (p = 0.02; Table 1).

Radiological surveillance

Contextual background for radiological review and classification of interval BCs, along with definitions of the categories of interval BCs, have been outlined in the introduction of the paper; methodological issues have been comprehensively explained in our previous review.1 Table 2 summarises findings from the literature search on radiological surveillance including the methods used to conduct mammographic review;10, 13,14,15, 18, 26, 31,32,33,34,35,36,37,38,39,40 the latter substantially influences the distribution of radiological categories and can bias estimated proportions.1, 33, 41 For example, a pilot study examining radiological review methods showed that informed vs. blinded (uninformed) review of interval BC leads to bias in classification whereby informed reviewers (aware they were reviewing mammograms containing interval cases) more frequently classified some interval BCs as positive, compared to reviewers who were unaware they were reading mammograms of interval BCs that had been added into routine screen-reading practice (‘uniformed’ review).41 To the extent that ‘blinding’ is possible in evaluation of interval BCs, this methodology sometimes referred to interval cases being interspersed with screen-reading as part of the routine screening workflow, or more frequently partial blinding was achieved by interval cases being mixed with normal screening mammograms (Table 2). Semi-informed radiological review methods involved knowledge that interval BC cases were being reviewed, without information on the side and location of the interval BC. In general, studies of radiological classification were based on an initial review of the index screen (the screen preceding the subsequently diagnosed interval BC) with provisional identification and classification of the interval BC, followed by review of both the screening and diagnostic mammograms to enable definitive classification or sub-classification. However, not all studies provided these details and some studies did not specify whether the diagnostic mammograms were available for classification. Interpretation was generally performed by experienced mammography readers, and varied from one expert screen-reader performing classification, to panels of several radiologists with classification based on reaching consensus or derived from majority reads. Table 2 footnote provides further definitions on radiological review methods and classification terminology.

Table 2 Radiological surveillance of interval breast cancers: methods and results of mammographic review and classification
Full size table

As shown in Table 2, radiological categories varied slightly across studies, however, most studies reported the distribution for true interval BC and also for missed or false-negative cases (defined in the introduction); some studies reported the additional categories of ‘occult’ and ‘minimal-signs’ interval BCs (see Table 2 footnote). The evidence table shows that the vast majority of interval BCs were not missed at screen-reading but were true interval BCs (range 40% to 77%) or occult interval BCs (7% to 32%), meaning that they were not visible on the index screen even in hindsight. Of note, some of the high proportions reported for true interval BCs (>60%) appear to have included the occult cases among the true interval cases. The proportion of missed (false-negative) interval BCs ranged between 13.6% and 35%, with the majority reporting a frequency of 20–25% based on radiological surveillance. A study from Ciatto and colleagues33 used a multi-methods evaluation of the same set of mammograms, and showed that increasing the information available to screen-readers significantly increased the proportion of interval BCs classified as missed (Table 2), highlighting the impact of review methods on radiological classification.

Additional findings summarised in Table 2 describe radiological findings (where present) for the index screen, which were frequently masses or asymmetric densities. They also highlight study-specific data showing differences in the variables associated with interval BCs, and in the tumour characteristics of interval BCs, across radiological categories derived from mammographic review.

Radiological review of interval BCs following screening with digital or film-screen mammography was reported by Knox et al.42 showing that a similar proportion of cancers were classified as missed cases at digital and film-screen (10.5% and 8.1%, respectively, p = 0.77). However, fewer interval BCs were depicted as microcalcifications alone or in association with another imaging abnormality following digital than film-screen mammography (16% and 32%, respectively, p = 0.02) (ref. 42). Nederend and colleagues38 investigated interval BCs in a population-based study of regional screening units in the Netherlands, and showed that significantly more interval BCs were classified as true-negative or true interval cases (not visible at the index screen) at radiological review of prior digital than prior film-screen mammography (65.3% vs. 47.1%, p = 0.02) as shown in Table 2; otherwise, there were no differences between interval BCs at digital or film-screen in terms of mammographic abnormalities at the prior screen or in tumour characteristics. Generally similar findings were reported by Weber et al.13 also from the Dutch screening program, who additionally observed that interval BCs emerging in year 2 of the inter-screen interval for digital mammography were more frequently true (than missed) interval cancers compared to those for film-screen mammography (p = 0.03; Table 2).

Biological characteristics and prognosis

Table 3 summarises biological findings including tumour characteristics and biomarker profile for interval BCs, and outlines the comparison reported in each study because that accounts for some of the apparent heterogeneity in results.11, 13, 14, 26, 31, 35,36,37, 39, 43,44,45,46,47,48 For the majority of studies, which compared interval BCs with screen-detected cancers, there were consistent findings that interval BCs had worse prognostic features, such as larger tumour size, higher frequency of node metastases, higher histologic grade, and more advanced disease compared to screen-detected BC (Table 3). Although biomarker data were not consistently reported in these studies, where reported there was evidence that interval BCs had a higher frequency of triple-negative or HER2-positive cancers and a lower frequency of hormone receptor-positive cancers than screen-detected BC (Table 3).

Table 3 Biological characteristics of interval breast cancers
Full size table

Some studies compared prognostic features between radiological categories of interval BCs, with variable findings (and limited statistical comparisons once study data were examined in subsets), however, some differences were noted between radiological subgroups. These differences by radiological subgroup are detailed in Table 3, and suggest that the ‘missed’ group had worse prognostic features than the true interval and occult interval cancers, except for tumour grade, which was reported to be more frequently higher for true interval and occult interval BCs than missed cases. Additional findings suggest that these outcomes may differ slightly between dense and non-dense breasts but density-specific findings were reported in very few studies (Table 3).

In the limited number of studies comparing interval BCs with clinically presenting cancers in non-screened women,44, 46 there was evidence that interval cancers were similar in terms of prognostic features to the BCs occurring in non-screened women, however, one study reported that interval BCs had slightly higher proportions of larger tumours (>20 mm) than BCs in non-screened women.46

Prognosis of interval BC

A population-based cohort study found similar survival for women who had an interval BC in the Norwegian screening program (hazard ratio (HR) 0.98; 95% confidence interval 0.84–1.15) as those who had BC diagnosed in the same time frame but had not yet been invited to mammography screening (non-screened women).46 A study from the Malmo mammography screening program49 showed that interval BCs from the first decade of service screening had similar stage distributions and survival as the BCs diagnosed in non-attenders to screening, whereas the screen-detected cancers in that time frame had more favourable stage distributions and survival than the interval cases. In this same study, there was also evidence that the prognosis of women with interval BC had improved over a 20-year period, as may be expected from overall improvements in BC prognosis over time.

Domingo and colleagues14, 50 conducted several studies examining the characteristics of interval BCs; one of these evaluated 2245 invasive BCs and clearly showed that interval BCs had more advanced tumours than screen-detected BCs (additional details by interval BC category shown in Table 3).14 In an earlier study of 228 invasive BCs diagnosed among Barcelona women aged 50–69 years, Domingo et al.50also found that disease-free survival rates (at 5 year follow-up from diagnosis) for screen-detected, true interval, and symptom-detected BC were 87.5%, 64.1%, and 79.4%, respectively, and overall survival rates were 94.5%, 65.5%, and 85.6%, respectively.50 In keeping with these findings, they concluded that clinically-detected BC especially where these are true interval cancers had worse prognosis and poorer survival than screen-detected BC even after adjustment for clinical-pathological variables.50 Porter and colleagues48 compared the features of interval BCs by radiological classification, and although they observed differences in the histological characteristics (shown in Table 3) there was no significant survival difference between interval BC radiological types (p = 0.64).

Some studies have examined the prognostic characteristics of interval BCs by breast tissue density.11, 45 Holm et al.45 showed that interval BCs occurring in non-dense breasts (defined by<20% density) had poorer prognostic features than screen-detected BC (Table 3), whereas interval BCs in dense breasts (≥50% density) were phenotypically more similar to screen-detected cancers. Eriksson et al.51 compared survival in interval and screen-detected BC allowing for mammographic density in women aged 50 years and older; they showed that hazard rates for BC-specific survival were significantly higher for interval than for screen-detected cancers, independent of density. In addition, interval BC in women with non-dense breasts had increased 5-year survival HR (2.43, p = 0.001) compared to screen-detected BC in non-dense breasts, but this was not the case in women with dense breasts, in whom a difference in survival was not statistically evident between interval and screen-detected BC (5-year survival HR 1.41, p = 0.49) after adjustment for cancer size.51

Discussion

This overview of the epidemiological, radiological, and tumour characteristics of interval BCs—cancers that emerge following a negative mammographic screen—highlights key themes on interval BCs, which are summarised in Fig. 2. Interval BCs are an important consideration in population BC screening because they are indicative of screening quality hence evaluating these cancers may help identify areas for potential improvement, and because they represent a failure of screening to detect a BC that subsequently progresses to presentation. It is clear from radiological surveillance data summarised in our work that these cancers are not necessarily missed at mammography screening, with most studies reporting around 20–25% of interval BCs to be missed (false-negative) cases. While radiological surveillance has limitations inherent in retrospective re-interpretation of imaging, and radiological classification of interval BCs is affected by the methodology used to perform mammographic review (Table 2), quantitative evidence shows that the majority of interval cancers are true interval or occult interval BCs that were not visible on the index screen. It should be acknowledged that radiological surveillance is not practiced in all screening settings, and the aim of summarising the evidence is not to advocate this form of surveillance, rather its findings can inform practice. For example, it seems likely that enhancing screen-reader skills would have relatively less effect in controlling interval BCs than, say, enhancing mammography technology or using alternate or additional technology to address the majority of cases that are not visible at the index mammography screen even in hindsight.

Fig. 2
figure 2

Key findings of the review

Full size image

Epidemiological monitoring of interval BC rates is a more widely performed surveillance in population mammography screening programs. Variability in the overall interval BC rates shown in Table 1 reflects the underlying BC risk in the population (which is also glimpsed in study-specific BC detection rates), the mix of initial and repeat screen rounds, and the length of the inter-screening interval. The variability due to the inter-screening interval is particularly evident where data are presented by yearly rates for biennial or triennial screening: our summary shows consistent evidence that rates in year 2 are at least twice those in year 1 (Table 1). The majority of studies in the evidence table report data from biennial screening, and show that interval BCs represent roughly one quarter of the cancers occurring in screening participants—that proportion is lower for annual, and higher for triennial, screening intervals. These findings should not be taken to infer that annual screening has better population outcomes than biennial screening (in fact biennial screening reduces some screening harms compared to annual screening), they merely highlight that many interval BCs are identified in the second year of a biennial screening round, and is commensurate with the pattern of findings from radiological surveillance specifically that many interval BCs are true interval cancers. Additional data from radiological surveillance from Weber and colleagues13 indicates that the proportion of interval cancers that are true interval BCs increases in year 2 (relative to year 1) at biennial screening. The evidence table also shows that interval BC rates are higher at repeat (incident) rounds than initial (prevalent) screening rounds within the same screening setting and population. This finding has not been thoroughly explained in the reviewed studies but is presumably due to age increase and a tendency for lower recall rates at repeat screening of the same women but warrants further research.

Although there is a substantial body of knowledge on interval BC rates in mammography screening, as shown in Table 1, there was little direct evidence on mammography-based interventions that reduce interval BC rates. A limited number of studies identified in this review reported that two-view (vs. one-view) mammography was associated with significant reductions in interval BC rates in population-based programs.27, 30 There was little evidence on the effect of digital mammography on interval BC rates, limited to one study showing that the use of digital (compared to film-screen) mammography was associated with a small but significant reduction in interval BC rates.13

Evidence on radiological features of interval BCs following transition to digital mammography was also limited to few studies13, 38, 42 with one study suggesting that implementation of digital mammography modified the mammographic pattern of interval BCs, with fewer interval BCs depicted as microcalcifications.42 However, two studies did not find substantial differences in interval BCs, in terms of the pattern of mammographic findings, following transition to digital mammography.13, 38

Studies providing data on the biological characteristics of interval BCs have mostly compared them to screen-detected BCs from the same screening setting and population, and have shown that interval BCs have relatively worse tumour prognostic characteristics, and worse survival outcomes, than screen-detected BCs. Interval BCs were consistently reported to be at a more advanced stage when diagnosed compared to screen-detected BC in terms of larger tumour size, higher frequency of node metastases, higher histologic grade, and had less favourable biomarker profile including a higher frequency of triple-negative cancers. This does not mean that interval BCs are an aggressive group of cancers, in fact they have tumour characteristics and survival outcomes that approximate those of BCs diagnosed in non-screened women, based on data from the few studies reporting that comparison.44, 46 One study reported that interval BCs had higher proportions of larger tumours than BCs in non-screened women but did not find any difference in survival between these groups.46 Further insights were provided by studies comparing prognostic features between radiological categories of interval BCs, with findings suggesting that the ‘missed’ group had worse prognostic features for tumour size and node status (possibly due to the delay in detection) than the true interval and occult interval BCs although the latter were more likely to have higher tumour grades (suggesting these to be more rapidly growing cancers representing new events on the mammogram).

Because mammographic breast density is an established risk factor for interval BC in screened women,6, 7 some of the studies summarised in this review evaluated interval BCs in relation to mammographic density.10, 11, 14, 39, 45, 51 The detailed findings, summarised in our results, are complex but reveal some common findings. Mammographic density was consistently associated with occult BCs (followed by true interval BCs) rather than missed interval cases,10, 14 highlighting a likely masking effect. In addition, there was a suggestion that interval BCs occurring in non-dense breasts may be associated with worse prognostic features and outcomes,45, 51 perhaps reflecting that interval BCs in non-dense breasts were more likely to be newly arising cases associated with rapid growth—however, the data were from few studies.

Epidemiological and radiological surveillance of interval BCs, complemented by an understanding of the biology of these cancers, provide insights into ‘how often’ and ‘why’ screening may not detect a BC that is subsequently diagnosed. Evidence shows that quantitative data on interval BCs are very heterogeneous and are influenced by several factors including the length of the inter-screening interval. Most published studies have reported data from biennial screening practice and in that context interval BCs represent roughly 17–30% of the cancers occurring in screening participants. Radiological surveillance highlights that the majority of interval BCs represent true interval or occult interval BCs that were not visible on the index mammographic screen, with only around 20–25% of interval BCs reported to have been missed cases on mammographic review. Biological characteristics of interval BCs show that they have relatively worse tumour prognostic characteristics and survival outcomes than screen-detected BCs, but similar characteristics and prognosis to BCs occurring in non-screened women. There was limited evidence of the effect on interval BC frequency and outcomes following transition from film to digital mammography screening.

Methods

This is a descriptive review based on a literature search and extraction of relevant information into evidence tables for each of three themes: epidemiologic measures, radiological surveillance and biology of interval cancers in population mammography screening.

Literature search

Relevant publications were identified through a Medline literature search: we exploded the term ‘breast neoplasms’ to August 2016, and combined this with title-searching for ‘interval cancer$’ or ‘interval breast cancer$’. Study identification focused on published work from 2006 onwards, given the above-stated aims of the review and the time frame from our previous evidence review,1 however, earlier studies were considered where data were reported in more recent publications.2, 4 Studies that provided information on population mammography screening allowing contribution into the evidence tables were included. Additional relevant studies were included in descriptive text if they provided key information on interval BCs that was not captured in the evidence table format. Studies that screened groups at increased risk of BC were not within the scope of the present review. Appendix 1 shows a flow diagram of the literature search and study inclusion process.

Evidence tables

Each evidence table provided a summary of key findings from each study contributing information on interval BCs into at least one of three themes: epidemiologic measures, radiological surveillance and biologic features. For the evidence table on epidemiological measures (Table 1), studies were included if they reported data on interval BC rates (overall, or by year of inter-screen interval or by screening round) and also cancer detection rates because the latter provide complementary information about the study population and screening sensitivity. For the table on radiological surveillance (Table 2), studies were included if a radiological review and categorisation was performed allowing reporting of data on the frequency of one or more categories of interval BCs, and at minimum reporting data on false-negative (missed) interval cases. The evidence table on biological characteristics of interval BCs (Table 3) summarised data on tumour prognostic features, and also biomarkers where reported.