Mammography and Population Screening

Mammography, from the Latin prefix ‘mamma-’ for female breasts and ‘-graphy’ meaning something written or represented in a specific way, is the use of X-rays to image breasts and identify abnormalities, especially cancer. It was first developed as a technique for mastectomy specimens in 1913 (Gold et al., 1990). Today, mammography is used to screen asymptomatic women for breast cancer in many countries as well as to further assess screen-detected and symptomatic abnormalities. Typically, eligible women are invited to screen every 2 years. One examination consists of two different views per breast (see Figure 4.1). The first trial of screening asymptomatic women for breast cancer with physical examination and mammography started in 1956 (Gershon-Cohen, 1961), detecting 92 benign lesions and 23 malignancies (including six non-palpable cancers) after screening 1,312 women. This demonstrated that screening could reveal asymptomatic cancers, allowing earlier treatment. At the time, some of these cancers were undetectable by any other method. Debate continues on the effectiveness of mammographic screening programs (Autier & Boniol, 2018). Largely citing observational and cohort studies, some authors examine the impact of screening programs on incidence and question whether reductions in rates of advanced breast cancer and rates of breast cancer deaths over time are more attributable to improvements in treatment and population demography (Moller et al., 2019). However, numerous randomized controlled trials have demonstrated significant reductions in breast cancer mortality for women aged 50-69years (Strax et al., 1973; Tabar et al., 2011; Nystrom et al., 2002; Bjurstam et al., 2003) and in women from age 40 (Moss et al., 2015). This

Figure 4.1 Four images demonstrating mammography views and appearance of cancer. Images reproduced with permission from Ines Domingues, (a) Right craniocaudal (CC) view in a patient who went on to have further investigation and biopsy-proven cancer, (b) Left CC view in a separate patient assessed as normal, for routine followup. (c) Right mediolateral oblique (MLO) view in another patient with biopsy-proven cancer, (d) Left MLO of normal breast (separate patient). (Images reproduced from the INbreast database with permission from Ines Domingues).

relative reduction in the risk of dying from breast cancer is on the order of 20%-30% over various follow-up periods ranging from 3 to 29 years (Tabar et al., 2011; Nystrom et al., 2002; Bjurstam et al., 2003))- It is possible that longer studies may demonstrate a larger reduction over time. The World Health Organisation (WHO) has methodically examined the evidence of harms and benefits for organized population-based screening mammography and made recommendations relative to the availability of resources and quality of the screening program, and by age group (WHO, 2019). For women aged 50-69 in well-resourced settings, WHO recommends population-based screening (with conditions regarding cost, quality and assurance, shared decision-making, and respect for patient values and preferences, among others). Likewise, for women aged 40-49 and 70-75 with the proviso that the screening be in the context of rigorous research, monitoring and evaluation (WHO, 2019). The benefits of mammography screening programs include diagnosis of asymptomatic breast cancer, reduction in the risk of dying from breast cancer, and reduced risk of progressing to advanced breast cancer (Breeders et al., 2018; Moss et al., 2015). Benefits outweigh disadvantages including rates of overdiagnosis due to screening mammography and the anxiety of recall due to an abnormal examination (Moss et al., 2015). Estimates of number-needed-to-treat and relative risk reduction vary but can be approximated as follows. In women aged 50-74, eight deaths are prevented for every 1,000 women screened every 2 years. Two to four episodes of screening in women aged 40-74 can reduce the risk of breast cancer death by 25%—31% (Tabar et al., 1985; Shapiro et al., 1985).

Abnormalities on mammograms often constitute a very small portion of the total image, approximately 2% of pixels. Around 0.5%-l% of asymptomatic women attending screening programs are found to have breast cancer (Lehman et al., 2017). The detection of cancer from screening mammography therefore is a doubly imbalanced problem (rare in the cohort, rare in the image) and requires years of specialty training for human detection. The appearance of abnormalities representing cancer can be very subtle (see Figure 4.1a). Some cancers are mammographically occult, with no visible abnormalities at all. These can currently only be detected by other imaging modalities (e.g., ultrasound, magnetic resonance imaging) and comprised 13-2% of all cancers in one study (Wu et al., 2019). The contour of masses found on mammography is highly correlated with the underlying pathology, with clearly defined, well-circumscribed masses associated with benign abnormalities and indistinct or spiculated contours often representing cancer (Berment et al., 2014).

Assessment of human radiologists’ performances for detecting signs of breast cancer varies widely in methodology and results (Carney et al., 2010) and is confounded by myriad variables. The experience of the radiologist likely affects performance metrics, with some data suggesting that subspecialist breast fellowship radiologists are more sensitive but less specific (Elmore et al., 2009). Barlow et al. (2004) in contrast found that neither a greater volume of examinations nor experience at interpreting mammograms was associated with improved performance. Studies assessing human performance may exclude radiologists that do not meet an arbitrary threshold of volume of examinations (Lehman et al., 2017) and this, while necessary for adequate statistical power, may bias studies towards more experienced readers (when in actuality, other less experienced radiologists are assessing mammograms). Other assessments have relied on participation of breast radiologists, which may also select for more confident, experienced radiologists (Carney et al., 2010). The number and qualifications of readers differs geographically (USA single reader, Europe double reader) and effects screening performance. The incidence of abnormalities in the given population affects the cancer detection rate (Evans et al., 2013).

The reported sensitivity of screening mammography for breast cancer detection ranges widely from 51.5% (Duijm et al., 2009) to 92.5% with specificity generally >88% up to 99-5% and significant interobserver variability (Elmore et al., 2009; Hofvind et al., 2012). The most comprehensive estimated performance metrics are consistent with these ranges (see Table 4.2).

Table 4.2 Estimates of Human Screening Performance

Examinations conducted 2007-2013 (Lehman et al., 2017).

Table 4.3 Patient Characteristics Affecting Mammographic Appearance (Boyd et al., 2002; Stomper et al., 1996)

Variability in cancer detection performance ultimately influences the ability of screening programs to reduce breast cancer mortality (Carney et al., 2010). Regarding image characteristics, the mammography system vendor and the specifications of the reading station introduce variability into the appearance of a mammogram. Patient characteristics bring a source of variation to the mammographic appearance of breasts (see Table 4.3). Variables such as the type, contrast, and lighting of the radiologists’ monitors and details of ambient light are stipulated in screening protocols.