Five Years that Changed Breast Cancer: Advances in Prevention, Diagnosis, and Treatment (2021–2026), and the Rise of Artificial Intelligence in Oncology
Adrián Hunis MD *
*Correspondence to: Adrián Hunis MD, School of Medicine, Universidad de Buenos Aires, Honorary Member AMA, Emeritus Member ASCO, Emeritus Member ESMO.
Copyright
© 2026 Adrián Hunis MD is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Received: 12 October 2026
Published: 01 November 2026
DOI: https://doi.org/10.5281/zenodo.23911188
Fifty years ago, five-year survival for breast cancer was 74.8%; today it is 91% overall and exceeds 99% for localized disease, and mortality has fallen 44% since 1989. The last five years accelerated this trajectory across every domain. Prevention moved toward risk-adapted strategies, validated by the WISDOM trial, while polygenic scores and population genetics refined who should be screened and how intensively. Diagnosis sharpened with tomosynthesis, abbreviated MRI, and contrast-enhanced mammography, and circulating tumor DNA began to anticipate recurrence more than a year in advance. Surgery and radiotherapy embraced de-escalation — from sentinel-node validation to the omission of axillary surgery (SOUND) and of radiotherapy itself in low-risk disease (PRIME II, LUMINA) — while hypofractionation compressed treatment into a single week (FAST-Forward). Systemic therapy delivered antibody–drug conjugates that created the HER2-low entity, adjuvant CDK4/6 inhibitors, immunotherapy with proven overall survival in triple-negative disease, PARP inhibition for BRCA carriers, and new endocrine options. Underpinning all of it, genetic and molecular studies turned treatment into a series of biomarker-driven decisions, and artificial intelligence moved from promise to practice in risk prediction, image interpretation, pathology, and response forecasting. This review connects these advances, draws on five decades of perspective, and looks at what comes next.
Objectives
Introduction
Fifty years ago, a diagnosis of breast cancer carried a markedly different prognosis than it does today. For women diagnosed in 1975–1977, five-year relative survival was 74.8% [1]; today it stands at 91% overall and exceeds 99% for localized disease [2], while the breast cancer death rate has fallen 44% since its 1989 peak — a decline that has averted an estimated 517,900 deaths in the United States alone [2,3] (Figure 1). This transformation was not the product of a single breakthrough but the compound interest of five decades: screening, surgery, radiotherapy, systemic therapy, and, increasingly, molecular understanding.
In Argentina, where the author has practiced oncology for more than forty years, this same arc is clearly visible, refracted through the realities of a middle-income health system [4] — even as the global burden of cancer continues its alarming rise, in the world and in our country [5]. The last five years, however, deserve a telling of their own, because the pace of change has unmistakably accelerated. Screening recommendations in the United States were rewritten: the US Preventive Services Task Force now recommends biennial mammography from age 40 to 74 [6], while the American College of Radiology and the Society of Breast Imaging advise annual screening from age 40, with supplemental imaging for women at elevated risk or with dense breasts [7]. And artificial intelligence, long a promise on the horizon, has moved into the clinic as a working instrument, touching every step of the oncology pathway from drug discovery to treatment selection [8].
This review follows those threads — prevention, diagnosis, the quiet revolution in surgery and radiotherapy, the deepening precision of systemic therapy, and the rise of artificial intelligence — and connects them, because in modern breast oncology no advance stands alone.
Development
Every other advance in this review depends on a prior question: whom should we watch, and how closely? The last five years gave us sharper answers. The population-based CARRIERS consortium quantified, in tens of thousands of women, the breast cancer risks conferred by pathogenic variants beyond BRCA1 and BRCA2 — high risk with PALB2, moderate risk with CHEK2 and ATM, and increased risk of estrogen receptor–negative disease with BARD1, RAD51C, and RAD51D [9]. The 2024 ASCO–Society of Surgical Oncology guideline translated such knowledge into practice, recommending germline BRCA1/2 testing for every patient newly diagnosed with breast cancer at age 65 or younger, with a clear framework for multi-gene panel testing [10]. And risk estimation itself grew more granular: adding a 313-variant polygenic risk score to family-history–based BOADICEA predictions would have changed screening recommendations in up to one third of familial cases [11] — genetics, in other words, is no longer only about rare mutations but about the cumulative weight of common ones.
Prevention is also pharmacologic, and it has become gentler. In the TAM-01 trial, just 5 mg of tamoxifen daily for three years — a quarter of the conventional dose — reduced breast events, and the benefit was still visible at ten years of follow-up: invasive disease halved in women with ductal carcinoma in situ, fewer contralateral cancers, and no excess of serious toxicity [12]. Less drug, durable benefit — a theme that will recur throughout this review.
Nor can prevention ignore the body as a whole. In a controlled trial followed for nearly a quarter of a century, bariatric surgery was associated with a 32% lower incidence of breast cancer, an effect concentrated in women with baseline hyperinsulinemia [13]: metabolic health is oncologic health, and the insulin–cancer axis deserves a place in every prevention conversation.
The most consequential prevention trial of the period, however, asked a disarmingly simple question: does every woman need the same screening? WISDOM randomized more than 28,000 women to annual mammography or to risk-based screening integrating genetics, polygenic risk, breast density, and personal history. The risk-adapted strategy proved noninferior for the detection of stage IIB or higher cancers — while using fewer mammograms [14]. Personalized screening is no longer a concept debated in editorials; it is an evidence-based option. And that sets the stage for everything that follows, because better risk stratification is also what makes better diagnosis possible.
If prevention decides whom to watch, diagnosis decides how closely we can see. Here too, the arc bends toward precision. Digital breast tomosynthesis has consolidated its place: in TOSYMA, a randomized trial of nearly 100,000 screening participants, tomosynthesis plus synthesized mammography detected 48% more invasive cancers than conventional digital mammography — 7.1 versus 4.8 per 1,000 — without increasing recall rates [15]. For the large population of women with dense breasts and a negative mammogram, the BRAID trial provided the first randomized head-to-head comparison of supplemental modalities: abbreviated breast MRI and contrast-enhanced mammography each detected roughly three times as many invasive cancers as automated whole-breast ultrasound (17.4 and 19.2 versus 4.2 per 1,000) [16]. The message is consistent: we are finding disease that two-dimensional mammography leaves invisible.
A parallel revolution is unfolding in blood. In a prospective study of high-risk hormone receptor–positive, HER2-negative breast cancer, tumor-informed circulating tumor DNA testing identified molecular residual disease in 10% of patients — and preceded every distant recurrence by a median of 12.4 months [17]. Surveillance after curative treatment may soon be molecular as well as radiologic, and the boundary between diagnosis and monitoring is beginning to blur.
And genomics continues to protect women from overtreatment. With nearly nine years of follow-up, the MINDACT trial confirmed that the 70-gene MammaPrint signature identifies patients with high clinical but low genomic risk who can safely forgo chemotherapy — an effect most pronounced in women over 50 [18]. Diagnosis, in 2026, is no longer only about finding cancer; it is about characterizing it so precisely that we know when not to treat.
No discipline illustrates the philosophy of modern breast oncology better than surgery, whose history is a long retreat from the radical. Half a century ago, the Halsted mastectomy — chest-wall muscle included — was standard; then breast conservation proved that less could be equivalent, and the question moved to the axilla. The sentinel-lymph-node concept promised staging without the morbidity of full dissection, and NSABP B-32 validated the technique [19]. But validation was only the beginning: the real story of the last fifteen years is what happened after the sentinel node came back positive.
The ACOSOG Z0011 trial gave the first startling answer: in women with one or two positive sentinel nodes undergoing breast conservation with radiotherapy, omitting further axillary dissection did not compromise outcomes [20]. The AMAROS trial offered an alternative path — axillary radiotherapy instead of dissection, with less lymphedema [21]. And then, in 2023, the SOUND trial took the logic to its endpoint: in women with small tumors and a negative axillary ultrasound, omitting sentinel-node biopsy altogether proved noninferior [22]. From removing everything, to sampling, to — in carefully selected patients — not even looking: that is de-escalation, and it was achieved without sacrificing survival. This surgical restraint is inseparable from what follows, because every millimeter of surgery spared must be weighed against radiotherapy — the other half of locoregional treatment.
Radiotherapy has traveled the same road as surgery, and for the same reason: when systemic therapy and surgery do their jobs well, radiation can afford to do less. The FAST-Forward trial compressed adjuvant whole-breast irradiation into 26 Gy in five fractions over a single week, with outcomes noninferior to the traditional 40 Gy in fifteen fractions [23] — a transformation in convenience and access with particular meaning for working women and stretched health systems. IMPORT LOW showed that irradiating only part of the breast could be enough for selected low-risk patients [24].
And then came omission. PRIME II demonstrated that women aged 65 or older with low-risk, hormone receptor–positive disease could safely forgo radiotherapy after breast-conserving surgery [25]; LUMINA extended the concept to women 55 and older with low-risk luminal disease defined by clinical and biomarker criteria [26]. The through-line from the axilla to the radiation field is unmistakable: oncology is learning, trial by trial, how much treatment is enough — and having the courage to stop there (Table 2).
If locoregional therapy learned restraint, systemic therapy learned precision — and nowhere is that clearer than in the neoadjuvant setting, where treatment becomes its own diagnostic test. The defining result of the period is KEYNOTE-522: adding pembrolizumab to neoadjuvant chemotherapy, continued adjuvantly, in early triple-negative breast cancer, improved five-year overall survival from 81.7% to 86.6% [27]. Immunotherapy, once confined to metastatic disease, had crossed into cure-directed treatment — and the neoadjuvant platform is what made the signal visible early, through pathological complete response, years before survival data matured. That same platform now invites artificial intelligence to read response from tissue before the surgeon operates — a convergence we will return to [48,49]. But first, the adjuvant story, which the author has had occasion to review from CMF to the present day [28].
In a recent review, the author traced adjuvant systemic therapy from the CMF era to today's precision, risk-adapted strategies [28] — and the last five years supplied the newest chapters. For carriers of BRCA1/2 pathogenic variants with high-risk, HER2-negative early breast cancer, one year of adjuvant olaparib in OlympiA improved three-year invasive disease–free survival from 77.1% to 85.9% [29]: synthetic lethality, deployed as prevention of recurrence.
CDK4/6 inhibition, meanwhile, moved decisively into the adjuvant setting on two fronts. In NATALEE, three years of ribociclib plus endocrine therapy improved invasive disease–free survival in stage II–III hormone receptor–positive, HER2-negative disease [30]; in monarchE, two years of abemaciclib showed a sustained benefit at five years, with an absolute gain of 7.6% [31]. The author had flagged the early CDK4/6 signals — MONARCH 2 among them — nearly a decade ago when reviewing the highlights of ASCO 2017 [33]; what was then a metastatic story is now an adjuvant standard.
And alongside escalation came the wisdom of restraint. RxPONDER showed that postmenopausal women with one to three positive nodes and a 21-gene recurrence score of 25 or lower derive no benefit from adjuvant chemotherapy — while premenopausal women do [32]. Knowing whom not to treat is, in 2026, as much an advance as any new drug.
In metastatic disease, the period belongs to the antibody–drug conjugates. Sacituzumab govitecan opened the account in triple-negative breast cancer, improving both progression-free and overall survival versus chemotherapy in ASCENT [34]. Trastuzumab deruxtecan then redefined what HER2-targeting means: DESTINY-Breast03 showed a 72% reduction in the risk of progression or death versus trastuzumab emtansine in HER2-positive disease [35]; DESTINY-Breast04 became the first trial to show progression-free and overall survival benefits of a HER2-directed therapy in HER2-low metastatic breast cancer [36] — creating, overnight, an entirely new actionable entity comprising more than half of all breast cancers; and DESTINY-Breast06 extended the benefit to HER2-ultralow disease after endocrine therapy [37].
Endocrine resistance — the central problem of hormone receptor–positive metastatic disease — met two new answers. Elacestrant, the first oral selective estrogen receptor degrader, improved progression-free survival versus standard endocrine therapy in EMERALD, particularly in ESR1-mutated tumors [38]. The AKT inhibitor capivasertib, combined with fulvestrant, doubled median progression-free survival in CAPItello-291 [39]. And sacituzumab govitecan proved its worth beyond triple-negative disease, improving progression-free survival in heavily pretreated, endocrine-resistant hormone receptor–positive, HER2-negative cancer in TROPiCS-02 [40] (Table 1).
Running beneath every section above is a single instrument: the molecular characterization of tumor and host. Consider what the modern oncologist actually does: estimates germline risk with population data [9] and guideline-directed testing [10], refines it with polygenic scores [11], detects disease with imaging chosen by density and risk [15,16], watches for molecular relapse in blood [17], decides on chemotherapy with 70-gene and 21-gene signatures [18,32], selects PARP inhibition by BRCA status [29], chooses endocrine strategies by ESR1 mutation [38], and defines HER2-low and HER2-ultralow as therapeutic categories [36,37]. Not one of these decisions is made on anatomy alone anymore. The studies are no longer ancillary to treatment — they are treatment decisions, and the laboratory report has become the most-read page in the chart.
While breast oncology was rewriting its own rules, a parallel revolution was rewriting the tools. A comprehensive review mapped artificial intelligence across the entire oncology pipeline — target discovery, biomarker development, trial matching, and treatment selection [8]. The notion of foundation models for generalist medical AI was formally defined, with its validation and regulatory challenges honestly laid out [41]. In drug discovery, AlphaFold's atomic-accuracy protein structure prediction unlocked structure-based design [42]. Deep learning reached expert-level performance across radiology, pathology, and dermatology imaging [43], and large language models showed that clinical knowledge itself could be encoded: Med-PaLM answered USMLE-style clinical questions at 67.6% accuracy [44] — an early, careful step toward AI-assisted clinical reasoning. The question was no longer whether AI would enter oncology, but where it would matter most. In breast oncology, the answer came quickly.
In breast imaging, AI has already changed practice. Mirai predicts one-to-five-year breast cancer risk from a mammogram alone, holding its discrimination across seven international health systems and outperforming the Tyrer-Cuzick model [45] — risk prediction, in other words, can now be read from the image itself. In the prospective, randomized MASAI trial, AI-supported single reading matched the cancer detection rate of standard double reading while cutting screen-reading workload by 44% [46]. And a Danish simulation showed that an AI triage protocol could reduce radiologist workload by nearly two thirds and avoid a quarter of false-positive recalls at matched sensitivity [47] — efficiency that translates directly into access.
Beyond detection, AI is learning prognosis. Federated learning across centers predicted response to neoadjuvant chemotherapy in triple-negative disease from whole-slide images, without centralizing patient data [48]. A multi-omic machine-learning model — clinical, digital pathology, genomic, and transcriptomic data fused — predicted pathological complete response with an externally validated AUC of 0.87 [49]. And graph neural networks can now infer HER2 status directly from routine hematoxylin-eosin slides [50], triaging cases for confirmatory testing. From pixels to prognosis, in five years: that is the arc (Table 3).
Future
The next five years will be about integration rather than invention. Risk-adapted screening programs built on image-based models such as Mirai [45] and validated by trials such as WISDOM [14] will move from evidence into guidelines and practice. Circulating tumor DNA will guide escalation and de-escalation of adjuvant therapy, building on the lead time already demonstrated [17]. Antibody–drug conjugates will move earlier in the disease course, and new targets will follow the HER2-low precedent [36,37]. AI readers will face prospective validation and regulation following the MASAI example [46], and large language models will enter the clinic as supervised assistants [44].
The central challenge, however, will not be technological but human: equity. The author's view of oncology in Argentina [51] is a reminder that the advances reviewed here must travel — across income levels, health systems, and geographies — if they are to matter. The best test, the best drug, and the best algorithm are the ones that reach the patient.
Conclusion
Fifty years took breast cancer from 74.8% to 91% survival [1,2]; the last five accelerated every trend that made that possible. Prevention became personal [9–14]; detection became sharper [15–18]; surgery and radiotherapy learned the discipline of de-escalation [19–26]; systemic therapy became simultaneously more potent and more restrained [27–40]; molecular studies became the compass of every decision [9–11,17,18,29,32,36–38]; and artificial intelligence became a working colleague [41–50]. For the millions of women affected each year, this is not an abstract record — it is longer life, with better quality. Awareness, the theme of October 19, remains the first step: the best test is the one that gets done.
Disclosure on the Use of Artificial Intelligence
This review was prepared with the assistance of artificial intelligence tools for literature search, organization of epidemiological and statistical data, and drafting. All cited references were verified against primary sources (PubMed, journal websites, and DOI records), and all epidemiological and statistical figures were drawn from the sources cited. The author reviewed, edited, and approved the final manuscript and assumes full responsibility for its content.
References