Over-Diagnosis and Breast Cancer Screening

Over the past decade, there has been increased awareness and recognition of over-diagnosis as an important public health problem and potential source of harm with implications for overtreatment and other adverse consequences [1–5]. Over-diagnosis is a contemporary phenomenon closely linked with screening efforts, where the latter can result in the detection of indolent forms of cancer that would never have progressed sufficiently to be of any clinical consequence during a patient’s lifetime. By definition, over-diagnosis refers to detection of either in situ or early stages of invasive cancer that have low rates of proliferation and growth without clinical manifestation prior to death from other causes—indeed, there may be some cancers that would have spontaneously regressed at a time point beyond screen-detection when specific treatments with attendant harms and toxicities have already been initiated. Screening tests should ideally prevent, reduce or delay the clinical impact of a target disease and essentially aim to detect only those cancers that if left alone would continue to grow and earlier intervention could yield benefit in terms of mortality reduction and not necessarily increased survival per se. The natural history of this target disease should be “well understood” with a recognisable early stage for which treatment outcomes are improved compared to treatment at a later stage. For screening to produce a genuine improvement in survival and reduction in cause-specific mortality, there must be an event in the natural history beyond which prognosis is adversely affected with a threshold effect reflected in the size of a detectable lesion. If there were a progressive increase in size without any such concomitant event, then it would be of no consequence whether a cancer was detected by screening or when symptomatic. Two possible events occur in the natural history and progression of malignant breast lesions that account for the efficacy of screening strategies; firstly, early haematogenous dissemination with formation of micro-metastases at distant sites and secondly phenotypic progression. Should one or both events occur at some stage in the neoplastic continuum dependent on tumour size, then earlier detection and intervention may pre-empt formation of micro-metastases and/or a more biologically aggressive tumour. Screening tests should distinguish between indolent cancers where there is a lower chance of these two events occurring, and more aggressive lesions that are likely to have a lower threshold for haematogenous dissemination [1–6]. Unfortunately most screening tests tend to detect slow-growing or non-growing cancers or are non-specific and associated with incidental findings of non-growing cancers. Hence screening is a major driver of over-diagnosis for several common cancers including breast, lung and prostate with rates of over-diagnosis of 50%, 13–25% and 50–60%, respectively, using standard screening modalities for these cancer types [1, 7, 8]. Estimates of over-diagnosis are notoriously variable with a range of methodologies employed that each has limitations; over-diagnosis can only be calculated at the population level based on disparities between incidence and mortality and not for any individual patient—screen-detected cancers usually have some form of treatment and their unperturbed natural history is unknown. Genomic analysis may offer opportunities for assessing risk of progression for screen-detected lesions at the individual level but these assays currently have limited specificity and hence clinical application in terms of guiding intensity of treatments. There have been attempts to estimate the proportion of non-progressive versus progressive forms of ductal carcinoma in situ (DCIS) based on breast cancer incidence data in prevalence and first incident screening rounds. Yen and colleagues used a continuous-time “six-state” Markov model to estimate the incidence and relative proportions of non-progressive (over-diagnosis) and progressive DCIS. The average incidence of non-progressive DCIS was 1.11 per 100, 000 per year compared with a figure of 2.1 per 1000 per year for progressive DCIS with 37% of DCIS estimated to be non-progressive in a prevalence screen (range 20–50%). Hence a woman attending a prevalence screen has a 19-fold greater chance of having progressive DCIS or invasive cancer diagnosed than non-progressive DCIS. The chance of being diagnosed with non-progressive DCIS was 1 in 3300 that is 19 times less than a 1 in 175 chance of progressive DCIS or invasive cancer. Interestingly, at the incident screen, only 4% of DCIS was calculated to be non-progressive and this accords with concepts of a reservoir of indolent (non-progressive) DCIS within the population that is tapped into and essentially exposed by screening [9]. Esserman and colleagues have explored the use of a multi-gene classifier to determine indolent character amongst invasive carcinomas treated loco-regionally with mastectomy or lumpectomy (plus irradiation) and receiving tamoxifen or no systemic therapy [10]. A new ultralow-risk MammaPrint 70-gene threshold has been defined with a threshold based on no breast cancer cause-related death for node-negative patients at 15 years. These tumours were associated with hormone receptor positivity, human epidermal growth factor receptor 2 (HER2) negativity, low proliferation index (Ki-67) and a luminal A phenotype. However, only 20–25% of tumours with these conventional characteristics were ultralow-risk, and other factors are required to identify a high proportion of non-progressive tumours and avoid overtreatment with systemic therapies.