Start here: what breast cancer is
Breast cancer is a malignancy arising from breast tissue — usually from the cells lining the ducts or the milk-producing lobules. It is the most commonly diagnosed cancer in women worldwide and a leading cause of cancer death; the GLOBOCAN 2022 estimates put it among the largest contributors to the global cancer burden across 185 countries [1], and projections of the current and future burden run out to 2040 [2].
The single most important thing to understand is that it is not one disease. Two tumours that look similar under a microscope can behave completely differently, respond to completely different drugs, and carry completely different prognoses, depending on which molecules they express. Since the early 2000s the field has organised itself around intrinsic molecular subtypes — luminal A, luminal B, HER2-enriched, basal-like — identified from gene-expression portraits and later reduced to a practical classifier [3] [4] [5] [6]. In everyday practice those subtypes are approximated by three receptors measured on the biopsy:
- ER — the estrogen receptor — and PR, the progesterone receptor. Together these define hormone-receptor-positive disease, the largest group, which can be treated by depriving the tumour of oestrogen.
- HER2 — the receptor tyrosine-protein kinase erbB-2 — amplified and overexpressed in 25–30% of breast cancers, which increases the aggressiveness of the tumour [7], and which can be targeted directly.
- Tumours negative for all three are triple-negative, historically the group with the fewest options.
The WHO classification of breast tumours codifies the pathology behind this [8]. The practical consequence is that "breast cancer" names a category of decisions, not a single diagnosis, and every section below splits along those lines.
Why it is in the news, perennially. Two reasons, and they are different in kind. The first is a genuine, unresolved public argument about mammographic screening — how much benefit it delivers, at what cost in overdiagnosis, and at which ages and intervals. That argument has been going for decades and is quantified rather than settled [9] [10] [11]. The second is that the treatment side has moved very fast: antibody–drug conjugates have just redrawn the boundary of who counts as "HER2-positive" at all [12].
Its sibling in this series. The colorectal cancer review covers the other great screening-programme cancer, and the logic there — lead time, sojourn time, overdiagnosis, interval cancers — is the same logic that runs through this review's centrepiece. Obesity appears here as a risk and prognostic factor [13] [14] [15].
Three pillars follow — measurements, medicines, and progress — with a tumour-growth model between the first two that explains exactly what screening can and cannot catch.
Pillar 1: measurements and diagnosis
Mammographic screening, and the actual size of the argument
The case for screening is that finding a cancer before it declares itself allows treatment to work better. The case against is not that screening finds nothing — it is that screening finds too much, including cancers that would never have harmed the woman in her lifetime.
The most careful public accounting is the UK independent panel convened under Michael Marmot, and its numbers are worth stating precisely because they are so often mangled. From meta-analysis of 11 randomised trials with 13 years of follow-up, screening invitation was associated with a 20% reduction in breast cancer mortality (95% CI 11–27%). Applied to UK absolute risks, that means about one breast cancer death prevented for every 235 women invited, or 180 women actually screened. On the harm side, using the three RCTs that did not systematically screen the control group at the end, the frequency of overdiagnosis was "of the order of 11% from a population perspective, and about 19% from the perspective of a woman invited to screening" [9].
Both of those are real and both are large. The Cochrane review reaches a more sceptical reading of the same trials [10]; the American Cancer Society guideline reaches a more favourable one for average-risk women [11]. An analysis of three decades of US screening found that the incidence of early-stage disease roughly doubled while the incidence of late-stage disease fell only slightly — the signature of substantial overdiagnosis [16]. Estimates vary widely depending on study features and methods, and that variation has itself been studied [17].
The honest summary is that screening does reduce breast cancer mortality, that it also diagnoses cancers that would never have mattered, and that the ratio between the two depends on age, interval and the assumptions used. The model below is about why those two things travel together.
Diagnosis and staging
A suspicious finding on imaging leads to biopsy, which establishes whether cancer is present, its histological type and grade, and — critically — the receptor status above. Disease extent is described by TNM staging: the size and local extent of the tumour, whether regional lymph nodes are involved, and whether there are distant metastases. Nodal status is assessed surgically, and the extent of axillary surgery has itself been progressively de-escalated as evidence accumulated [18].
The measurements that choose the treatment
Three tiers of measurement now sit between diagnosis and a treatment plan.
Receptor status, as above, sorts patients into the three therapeutic worlds. Grade and Ki-67, a proliferation marker, further separate the slow-growing luminal A pattern from the more proliferative luminal B pattern, which carries a worse prognosis at the same receptor status [19].
Genomic recurrence scores answer a narrower and more agonising question: for a woman with hormone-receptor-positive, node-negative disease, will adding chemotherapy to endocrine therapy actually help? The 21-gene recurrence score was validated prospectively — in the low-risk group assigned to endocrine therapy alone, outcomes at five years were excellent without chemotherapy [20], and real-world cohorts have reported consistent results [21]. Professional bodies have issued guidance on which biomarkers should and should not drive adjuvant decisions [22] [23] [24]. The same logic has now been extended to omitting radiotherapy in women with low clinical and genomic risk [25].
Hereditary risk. Pathogenic variants in BRCA1 and BRCA2 confer substantially elevated lifetime risk and change both screening and treatment decisions; beyond those two genes, association analysis has identified 65 further breast cancer risk loci, so common polygenic risk is real but individually small [26]. BRCA1 mutation and young age together predict faster tumour growth on MRI screening [27] — a fact that matters directly for the model below.
Centerpiece: a simple simulatable model of tumour growth and the screening window
Screening arguments usually get conducted in words. They are much clearer as arithmetic, and the arithmetic needs only one idea: tumours do not grow exponentially — they decelerate.
Fitting exponential, Gompertz and seven generalised logistic equations to serial mammographic measurements in 448 patients, all the decelerating equations fit better than the exponential [28]. Norton's Gompertzian model of human breast cancer growth is the parsimonious version of that idea, and it makes explicit clinical projections about "the estimated duration of silent growth prior to diagnosis" [29]. Write it in cell number:
V(t) = V_max · exp( −ln(V_max/V₀) · exp(−b·t) )
with V₀ = a single cell and V_max = 2⁴⁰ cells, both the assumptions used in the fitting study [28]. That leaves one rate constant b, and it does not have to be guessed: it can be solved so the tumour takes exactly the time a population study measured.
What the model explains. Four things, and each is one side of the screening argument.
First, why most of a cancer's life is invisible. On this curve a tumour takes about six years to reach 5 mm — where mammographic sensitivity is only 26% — and it reaches 10 mm, where sensitivity is 91%, at about seven years [30]. The silent phase is longer than the entire detectable phase. Screening is not catching disease "early" in any absolute sense; it is catching it a year or two earlier than symptoms would.
Second, why growth deceleration matters for what screening buys. The same eightfold increase in volume takes 1.3 years from 5→10 mm but 1.7 years from 10→20 mm. Under exponential growth those would be equal. Because the tumour slows as it enlarges, the detectable window is wider than exponential intuition suggests — which helps screening — but the same deceleration means that a tumour detected small may have been present for years, which is part of why earlier detection does not translate one-for-one into lives saved.
Third, why the screening interval is the whole design question. The transit-time distribution is enormously wide: 5% of tumours cross from 10 mm to 20 mm in under 1.2 months, and another 5% take more than 6.3 years [30]. Fitted to those two tails, 57% of tumours cross that window in under a year, 77% in under two, 85% in under three. Tumours in the fast tail can appear and grow past the window entirely between two screening rounds — they present symptomatically as interval cancers, which have systematically worse characteristics and outcomes than screen-detected ones [32]. Shortening the interval catches more of them, at the cost of more screens and more overdiagnosis.
Fourth, why screening preferentially finds the tumours that matter least. This is the uncomfortable corollary. A slow tumour spends years in the detectable window and will be caught by almost any screening programme; a fast one may never be caught at all. Screening therefore over-samples indolent disease — which is the mechanism of the 11–19% overdiagnosis the Marmot panel quantified [9], and the reason early-stage incidence doubled without a matching fall in late-stage disease [16]. The two headline findings of the screening debate are not in tension; they are the same curve seen from two ends.
What the model deliberately does not do. It has one curve where the sources insist heterogeneity is intrinsic [29], and growth rate demonstrably varies with age and genotype — younger women and BRCA1 carriers have faster-growing tumours [30] [27] [33]. It selects on size alone, when real screening also detects DCIS and is affected by breast density. It contains no treatment, so it cannot say how much of the mortality benefit is earlier detection versus better therapy — a distinction the Marmot panel flagged as central, noting that treatment advances have had "a demonstrably larger influence on mortality trends than does screening" [9]. And it says nothing about the woman's experience of a diagnosis she did not need.
Pillar 2: medicines
Treatment is local plus systemic. Local control is surgery — breast-conserving or mastectomy — plus radiotherapy, whose effect is well quantified: the EBCTCG meta-analysis of trials of radiotherapy after mastectomy and axillary surgery measured its effect on 10-year recurrence and 20-year breast cancer mortality [18]. Systemic therapy is where the subtypes diverge completely.
Hormone-receptor-positive disease: starve it
If a tumour depends on oestrogen, remove the signal. Two mechanisms do this.
Tamoxifen blocks the estrogen receptor directly. The EBCTCG overview of the randomised trials established its effect on recurrence and mortality in early breast cancer [34], and it is effective enough that it reduced breast cancer incidence in high-risk women in the NSABP P-1 prevention trial [35].
Aromatase inhibitors — anastrozole, letrozole, exemestane — instead block aromatase, the enzyme that synthesises oestrogen from androgens, which in postmenopausal women is the main remaining source. ATAC compared anastrozole against tamoxifen and against the combination as adjuvant therapy [36]; BIG 1-98 compared letrozole with tamoxifen [37]; MA.17 tested letrozole after five years of tamoxifen and found further benefit, establishing extended adjuvant therapy [38] [39]. ASCO issued a technology assessment on their adjuvant use [40].
Resistance eventually develops, and its mechanism is partly known: acquired mutations in ESR1 itself can be detected in plasma and predict outcome in ER-positive advanced disease [41]. Adding an mTOR inhibitor, everolimus, to endocrine therapy extended progression-free survival in that setting [42].
HER2-positive disease: target it
This is the field's clearest demonstration that a molecular target can be turned into a drug. Trastuzumab is a monoclonal antibody against HER2. In metastatic disease, adding it to chemotherapy lengthened time to progression (7.4 vs 4.6 months), raised response rate (50% vs 32%), extended median survival (25.1 vs 20.3 months) and reduced the risk of death by 20% — with cardiac dysfunction as the important toxicity, in 27% of those also receiving an anthracycline [7].
Moved into the adjuvant setting the effect was larger. The joint analysis of NSABP B-31 and NCCTG N9831 reported a hazard ratio of 0.48 for recurrence, second primary or death, a 12-percentage-point absolute improvement in three-year disease-free survival, and a 33% reduction in the risk of death [43]; HERA confirmed benefit for trastuzumab given after adjuvant chemotherapy [44]. Later agents extended the approach: lapatinib, a small-molecule kinase inhibitor, for trastuzumab-refractory disease [45] and combined with letrozole in HR-positive/HER2-positive disease [46]; and trastuzumab emtansine (T-DM1), the first antibody–drug conjugate in this space, in previously treated advanced disease [47]. The whole trajectory has been reviewed as a model for targeted oncology [48] [49].
HR-positive/HER2-negative advanced disease: block the cell cycle
The newest pillar in this group inhibits CDK4 and CDK6, the kinases that drive cells past the G1 restriction point, and it works specifically in combination with endocrine therapy. PALOMA-1 first showed longer progression-free survival for palbociclib plus letrozole [50]; PALOMA-2 confirmed it in phase 3, with median progression-free survival of 24.8 versus 14.5 months (hazard ratio 0.58, 95% CI 0.46–0.72), at the cost of substantial neutropenia — grade 3–4 in 66.4% versus 1.4% [51]. PALOMA-3 established benefit with fulvestrant after progression on endocrine therapy [52] [53], later with an overall survival readout [54], and MONARCH 3 showed the same for abemaciclib as initial therapy [55].
Triple-negative disease: chemotherapy, and now immunotherapy
Lacking all three targets, this subtype depended on cytotoxic chemotherapy alone for decades. Checkpoint inhibition changed that. Early signals came from KEYNOTE-012 [56], and KEYNOTE-522 established the approach in early disease: adding pembrolizumab to neoadjuvant chemotherapy raised the pathological complete response rate from 51.2% to 64.8% [57], with event-free survival benefit confirmed on longer follow-up [58]. In advanced disease, KEYNOTE-355 showed benefit for pembrolizumab plus chemotherapy [59] [60]. Not every trial in the class succeeded — IMpassion131 was negative for atezolizumab with paclitaxel [61] — which is a useful reminder that the chemotherapy backbone and the biomarker definition both matter.
Management of advanced disease across all subtypes has been codified in international consensus guidelines [62], and survivorship care in its own guideline [63] — including the cardiovascular consequences of the treatments above [64].
Pillar 3: progress
Antibody–drug conjugates, and the erasure of a category
The most consequential recent change is conceptual as much as pharmacological. Trastuzumab deruxtecan (T-DXd) couples the anti-HER2 antibody to a potent topoisomerase inhibitor with a high drug-to-antibody ratio and a membrane-permeable payload, so it kills neighbouring cells as well as the one it binds. It showed striking activity in previously treated HER2-positive disease [65] [66] [67], and then beat T-DM1 head-to-head in DESTINY-Breast03 [68] [69].
The category-breaking result was DESTINY-Breast04. Among breast cancers without HER2 amplification or overexpression, a large proportion express low levels of HER2 — defined as an immunohistochemistry score of 1+, or 2+ with negative in-situ hybridisation — and "currently available HER2-directed therapies have been ineffective" in them. T-DXd was tested against physician's-choice chemotherapy in 557 patients with HER2-low metastatic disease, 88.7% of them hormone-receptor-positive, and improved progression-free survival [12]. Activity has since been shown across HER2-expressing solid tumours generally [70].
The significance is that "HER2-positive" was never a property of tumours; it was a threshold on an assay, chosen because that was the level at which the old drugs worked. A better drug moved the threshold, and a large group of patients previously classified as HER2-negative became treatable.
Risk-stratified screening
If the model's message is that a single interval applied to everyone is a blunt instrument, the response is to vary it. Risk-stratified screening — using age, density, family history, and polygenic risk to set each woman's interval and modality — has been assessed for cost-effectiveness and harm–benefit ratio [71] [72], and the WISDOM trial was designed to test personalised against annual screening, with simulation work assessing its power in advance [73]. Deep learning applied to screening mammograms improves detection [74], which changes the sensitivity term in the same arithmetic.
De-escalation
Progress is not only additive. Because the measurements in Pillar 1 identify low-risk disease reliably, it is now possible to remove treatment: omitting chemotherapy on the basis of a genomic recurrence score [20] [22], and omitting radiotherapy after breast-conserving surgery in women with low clinical and genomic risk [25]. For a disease whose treatments carry lasting cardiac and other costs [64], subtracting therapy safely is as much a gain as adding it.
Risk factors that are actually modifiable
Body weight is the clearest. A systematic review and meta-analysis of 82 follow-up studies examined body mass index and survival in women with breast cancer [13], with a separate meta-analysis reaching the same direction [14], and national cancer statistics have been reported specifically featuring cancers associated with excess weight and insufficient activity [15]. Risk factors and prevention have been reviewed broadly [75] [76], as has public awareness of them [77].
Dig deeper in lmmol
Breast cancer sits alongside the other screening-programme cancer and the metabolic risk factor it shares with much of this collection:
- Colorectal cancer — the oncology and screening sibling. Lead time, sojourn time, interval cancers and overdiagnosis are the same machinery there, applied to a different organ and a different test.
- Obesity — a risk and prognostic factor here, quantified across 82 follow-up studies [13].
- The health reviews index collects the rest of the series.
Then move down into lmmol's graph, to the receptors and enzymes that decide the treatment:
- Estrogen receptor — what ER-positive means, what tamoxifen blocks, and where ESR1 resistance mutations arise [34] [41].
- Progesterone receptor — the second hormone receptor in the standard panel.
- Aromatase — the oestrogen-synthesising enzyme that anastrozole, letrozole and exemestane inhibit [36] [37].
- Receptor tyrosine-protein kinase erbB-2 — HER2, amplified in 25–30% of breast cancers, and the target of trastuzumab, pertuzumab, T-DM1 and T-DXd [7] [12].
- CDK4 and CDK6 — the cell-cycle kinases palbociclib, ribociclib and abemaciclib inhibit [51] [55].
- BRCA1 and BRCA2 — the hereditary susceptibility proteins [26] [27].
- PI3K catalytic subunit alpha and the androgen receptor — two further axes under active investigation in this disease.
- For entities without a linked static page here, use the graph index, all proteins, or all diseases rather than guessing an entity URL.