A machine learning-based algorithm to eliminate breast and axillary surgery in patients with breast cancer and pathological complete response after neoadjuvant chemotherapy

Efforts have continually been made over the two past decades to de-escalate loco-regional and systemic treatments in patients with breast cancer (BC). Regarding loco-regional therapies, several trials demonstrated equivalent survival in de-escalating routine therapies such as omission of axillary dissection or radiotherapy in selected patients (1,2). Consistently, the modern approach to neoadjuvant chemotherapy (NACT), more and more used during the last decades, allows for downstaging of cancer and less mutilating breast and axillary surgery. Its efficacy in triple-negative (TN) and human epidermal growth factor receptor 2 (HER2)-positive BC continues to improve with incremental gains in the rate of pathological complete response (pCR), which reaches today 60% and, in several cases, leads to improvements in event-free survival. In this context, a natural next question is whether surgery can be omitted in selected patients thus allowing avoidance of complications from surgery and anaesthesia, improved cosmesis, increased patient satisfaction, and reduced cost and resource use. As oncologists, we often discuss with our BC patients the potential for pCR following NACT, and their first reaction is often to ask if this would allow them to avoid surgery. So far, our answer has always been “only surgery is able to confirm pCR”, but it might change in a close future. Indeed, in a recent article titled “Intelligent Vacuum-Assisted Biopsy to Identify Breast Cancer Patients With Pathologic Complete Response (ypT0 and ypN0) After Neoadjuvant Systemic Treatment for Omission of Breast and Axillary Surgery”, reported in the Journal of Clinical Oncology, Pfob et al. described a model that identifies BC patients with pCR response (ypT0 and ypN0) to NACT who may be able to avoid breast and axilla surgery (3). They retrospectively analyzed individual data from patients enrolled in two US prospective trials that evaluated the use of a minimally invasive vacuum-assisted biopsy (VAB) to reliably exclude residual cancer in the breast after NACT between 2016 and 2020. Their model is a machine learning algorithm-based VAB model combining both patient, imaging, tumor, and VAB variables. It was defined in a multicentric discovery set including 318 patients with a cT1–3N0–1M0, HER2-positive, TN, or highly proliferative luminal B-like BC who had VAB performed before surgery. The robustness of this intelligent VAB model was then tested in an independent unicentric validation set of 45 patients, and its ability to predict residual cancer was compared against the actual results of the pathological evaluation of breast and axillary surgery. In the validation set, this intelligent model showed a false-negative rate (FNR) of 0% [95% confidence interval (CI): 0–13.7%] for detecting residual cancer, a specificity of 40% (95% CI: 19.1–63.9%) and AUC of 0.91 (95% CI: 0.82–0.97). Importantly, it performed better than imaging after NACT alone, VAB alone, or combinations of both using narrow patient selection criteria. Therefore, the authors concluded that this intelligent VAB model can reliably exclude the presence of residual cancer after NACT and before surgery, thus paving the way for omission of breast and axillary surgery for these exceptional responders in future trials.

Until now studies evaluating the prediction of pCR were based on breast imaging (4), minimal tumor biopsy (5), and tumor biology (6). Although promising, none of these predictive models proved to be as robust as the model developed by Pfob et al. with a 0% FNR. In fact, the algorithm used in their study accumulated all these parameters, and the five most important ones were tumor cells in VAB, accompanying in situ disease, lesion diameter on imaging before and after NACT, and VAB needle size. Although the rate of 0% of missed cancer in an external validation set is very promising, further prospective confirmatory evidence needs to be obtained with the primary objective to prospectively demonstrate that pCR prediction allows to omit surgery safely, ie without loco-regional relapse (LRR) during the follow-up. Initial efforts to omit surgery and instead to deliver radiation alone after achievement of complete clinical response failed because of unacceptably high LRR rates, probably due to the absence of precise imaging or VAB approaches to predict pCR (7,8). However, prospective studies are underway to determine whether a subgroup of patients may forego surgery in the setting of clinical complete response after NACT. The MD Anderson-promoted multicenter trial for eliminating breast surgery for invasive BC in exceptional responders to NACT based on image-guided VAB evidence of a pCR was recently reported in the Lancet Oncology (9). Women aged 40 years or older, not pregnant, with unicentric cT1-2N0–1M0 TNBC or HER2-positive BC and a residual breast lesion less than 2 cm on imaging after standard NACT were eligible for inclusion. A minimum of twelve 9G image-guided VAB were required. If no invasive or in-situ disease was identified, breast surgery was omitted and patients received standard radiation therapy alone (whole-breast radiotherapy plus a boost). Of 50 enrolled patients (42% with TNBC and 58% with HER2-positive BC), VAB identified a pCR in 31 patients (62%, 95% CI: 47.2–75.4%). Very promisingly, after a median follow-up of 26.4 months [interquartile range (IQR), 15.2–39.6], no ipsilateral breast tumor recurrence occurred in these 31 patients, and no serious biopsy-related adverse event or treatment-related death occurred. One may suppose that such results might be further improved using the intelligent VAB model developed by Pfob et al. who showed better FNR than other models based on imaging after NACT alone, VAB alone, or combinations of both using narrow patient selection criteria.

Altogether, these data suggest that surgery may be avoided for exceptional responder patients, provided that these latter are properly identified. Intelligent VAB provides the perspective to overcome this main limitation of pCR prediction accuracy. The study’s strength is its 0% FNR. Indeed, on one hand, de-escalation by eliminating surgery may help reduce the treatment burden for patients and improve their quality of life (10,11), which is a principal concern in curative strategy. On the other hand, it is critical to avoid loco-regional or metastatic recurrence that would result from an inappropriate de-escalation strategy in patients whose tumor has not been entirely eradicated by chemotherapy. Like others (12), we reported a strong negative survival impact of residual nodal tumor burden (ypN1), and completion of axillary lymph node dissection is recommended for sentinel lymph node biopsy micro- and macro-metastases after NACT (13). Moreover, for cN1 patients, the highest axillary pCR rate was reported for ER-negative/HER2-positive tumor subtype, but without major differences in axillary pCR rates per tumor subtype (14). The use of radiological imaging is limited to predict small residual nodal disease (i.e., micrometastasis), as illustrated in the GANEA study with 8.1% of the 123 cN0 patients that were staged as ypN1mi after surgery (15). The rate of ypN1 in cN0 patients after NACT is estimated between 1 and 2.1% according to molecular subtype of the tumor (16). Moving forward, given that invasive BC currently displays a 50% or more chance of having a pCR in the breast and nodes following NACT (17-21), another challenge would be to identify all patients with pCR eligible for a de-escalation strategy. In the Pfob et al. study, the specificity of intelligent VAB model was only 40% and will need to be improved. But the potential for improvement exists, likely though more experience in performing and evaluating minimally invasive biopsies after NST, since 53.3% of VAB in the validation set were deemed to be unrepresentative by the biopsying physician and 24.4% by the pathologist.

Besides the needed validation of every new predictive model in prospective clinical trials, many other factors might prevent the applicability of this study’s results to clinical practice. The effective implementation of digital health tool remains influenced by various stakeholders, social expectations, and environmental contexts. Many promising technological innovations in health and social care are characterized by non-adoption or abandonment by individuals or by failed attempts to scale up locally (22). By examining de-escalation processes, introduction, testing, and implementation of recommendation to limit surgery have often taken several decades (23). On the patients point of view, it is controversial whether the stress and apprehension brought on by more imaging, additional biopsies, and the potential for increased rates of local recurrence is justified in foregoing a low-morbidity outpatient procedure (24). Indeed the 7.1% complication rate of VAB is clinically meaningful (25) with regards of the 1.9% morbidity estimated at 30-day for lumpectomy with sentinel node biopsy after a radiographical complete response (26). All these questions, including cost-effectiveness issues, will need to be clarified in future clinical trials. However, this study, via the development of a machine learning algorithm–based innovative tool, is a new step forward in the growing body of data identifying potential low-value breast cancer surgery to de-implement. A first prudent step could be to consider sentinel lymph node biopsy in patients identified as in-breast pCR, in order to avoid ignoring an absence of in-node pCR with a risk of undertreatment, while waiting for higher performances of radiological imaging to predict small residual nodal disease.

Acknowledgments

Funding: Our work was supported by Institut Paoli-Calmettes, la Ligue Nationale Contre le Cancer (Label Ligue EL2022 to FB), and Le Prix Ruban Rose 2020 (to FB). None of them had any role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript.

Footnote

Provenance and Peer Review: This article was commissioned by the editorial office, Annals of Translational Medicine. The article did not undergo external peer review.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://atm.amegroups.com/article/view/10.21037/atm-23-689/coif). AdN declares Consulting fees by Gilead, Seagen, Lilly, and Novartis, payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing, or educational events by Gilead, Daiichi Sankyo, and MSD, Support for attending meetings and/or travel by Gilead, Lilly, and Daiichi Sankyo. AG declares Support for attending meetings and/or travel by Menarini, Participation on a Data Safety Monitoring Board or Advisory Board by Novartis, MSD, Astra Zeneca and Daiichi Sankyo. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Brackstone M, Baldassarre FG, Perera FE, et al. Management of the Axilla in Early-Stage Breast Cancer: Ontario Health (Cancer Care Ontario) and ASCO Guideline. J Clin Oncol 2021;39:3056-82. [Crossref] [PubMed]
2. Shubeck SP, Morrow M, Dossett LA. De-escalation in breast cancer surgery. NPJ Breast Cancer 2022;8:25. [Crossref] [PubMed]
3. Pfob A, Sidey-Gibbons C, Rauch G, et al. Intelligent Vacuum-Assisted Biopsy to Identify Breast Cancer Patients With Pathologic Complete Response (ypT0 and ypN0) After Neoadjuvant Systemic Treatment for Omission of Breast and Axillary Surgery. J Clin Oncol 2022;40:1903-15. [Crossref] [PubMed]
4. Gampenrieder SP, Peer A, Weismann C, et al. Radiologic complete response (rCR) in contrast-enhanced magnetic resonance imaging (CE-MRI) after neoadjuvant chemotherapy for early breast cancer predicts recurrence-free survival but not pathologic complete response (pCR). Breast Cancer Res 2019;21:19. [Crossref] [PubMed]
5. van Loevezijn AA, van der Noordaa MEM, van Werkhoven ED, et al. Minimally Invasive Complete Response Assessment of the Breast After Neoadjuvant Systemic Therapy for Early Breast Cancer (MICRA trial): Interim Analysis of a Multicenter Observational Cohort Study. Ann Surg Oncol 2021;28:3243-53. [Crossref] [PubMed]
6. Feng K, Jia Z, Liu G, et al. A review of studies on omitting surgery after neoadjuvant chemotherapy in breast cancer. Am J Cancer Res 2022;12:3512-31.
7. Ring A, Webb A, Ashley S, et al. Is surgery necessary after complete clinical remission following neoadjuvant chemotherapy for early breast cancer? J Clin Oncol 2003;21:4540-5. [Crossref] [PubMed]
8. Mauriac L, MacGrogan G, Avril A, et al. Neoadjuvant chemotherapy for operable breast carcinoma larger than 3 cm: a unicentre randomized trial with a 124-month median follow-up. Institut Bergonié Bordeaux Groupe Sein (IBBGS). Ann Oncol 1999;10:47-52. [Crossref] [PubMed]
9. Kuerer HM, Smith BD, Krishnamurthy S, et al. Eliminating breast surgery for invasive breast cancer in exceptional responders to neoadjuvant systemic therapy: a multicentre, single-arm, phase 2 trial. Lancet Oncol 2022;23:1517-24. [Crossref] [PubMed]
10. Flanagan MR, Zabor EC, Romanoff A, et al. A Comparison of Patient-Reported Outcomes After Breast-Conserving Surgery and Mastectomy with Implant Breast Reconstruction. Ann Surg Oncol 2019;26:3133-40. [Crossref] [PubMed]
11. Gärtner R, Jensen MB, Nielsen J, et al. Prevalence of and factors associated with persistent pain following breast cancer surgery. JAMA 2009;302:1985-92. [Crossref] [PubMed]
12. Almahariq MF, Levitin R, Quinn TJ, et al. Omission of Axillary Lymph Node Dissection is Associated with Inferior Survival in Breast Cancer Patients with Residual N1 Nodal Disease Following Neoadjuvant Chemotherapy. Ann Surg Oncol 2021;28:930-40. [Crossref] [PubMed]
13. Moo TA, Pawloski KR, Flynn J, et al. Is Residual Nodal Disease at Axillary Dissection Associated with Tumor Subtype in Patients with Low Volume Sentinel Node Metastasis After Neoadjuvant Chemotherapy? Ann Surg Oncol 2021;28:6044-50. [Crossref] [PubMed]
14. Samiei S, Simons JM, Engelen SME, et al. Axillary Pathologic Complete Response After Neoadjuvant Systemic Therapy by Breast Cancer Subtype in Patients With Initially Clinically Node-Positive Disease: A Systematic Review and Meta-analysis. JAMA Surg 2021;156:e210891. [Crossref] [PubMed]
15. Classe JM, Loaec C, Gimbergues P, et al. Sentinel lymph node biopsy without axillary lymphadenectomy after neoadjuvant chemotherapy is accurate and safe for selected patients: the GANEA 2 study. Breast Cancer Res Treat 2019;173:343-52. [Crossref] [PubMed]
16. Barron AU, Hoskin TL, Day CN, et al. Association of Low Nodal Positivity Rate Among Patients With ERBB2-Positive or Triple-Negative Breast Cancer and Breast Pathologic Complete Response to Neoadjuvant Chemotherapy. JAMA Surg 2018;153:1120-6. [Crossref] [PubMed]
17. Schmid P, Cortes J, Pusztai L, et al. Pembrolizumab for Early Triple-Negative Breast Cancer. N Engl J Med 2020;382:810-21. [Crossref] [PubMed]
18. Hurvitz SA, Martin M, Symmans WF, et al. Neoadjuvant trastuzumab, pertuzumab, and chemotherapy versus trastuzumab emtansine plus pertuzumab in patients with HER2-positive breast cancer (KRISTINE): a randomised, open-label, multicentre, phase 3 trial. Lancet Oncol 2018;19:115-26. [Crossref] [PubMed]
19. van la Parra RF, Kuerer HM. Selective elimination of breast cancer surgery in exceptional responders: historical perspective and current trials. Breast Cancer Res 2016;18:28. [Crossref] [PubMed]
20. de Nonneville A, Houvenaeghel G, Cohen M, et al. Pathological complete response rate and disease-free survival after neoadjuvant chemotherapy in patients with HER2-low and HER2-0 breast cancers. Eur J Cancer 2022;176:181-8. [Crossref] [PubMed]
21. Houvenaeghel G, de Nonneville A, Cohen M, et al. Neoadjuvant Chemotherapy for Breast Cancer: Evolution of Clinical Practice in a French Cancer Center Over 16 Years and Pathologic Response Rates According to Tumor Subtypes and Clinical Tumor Size: Retrospective Cohort Study. J Surg Res (Houst) 2022;5:511-25. [Crossref] [PubMed]
22. Greenhalgh T, Wherton J, Papoutsi C, et al. Beyond Adoption: A New Framework for Theorizing and Evaluating Nonadoption, Abandonment, and Challenges to the Scale-Up, Spread, and Sustainability of Health and Care Technologies. J Med Internet Res 2017;19:e367. [Crossref] [PubMed]
23. Kuerer HM. Moving Forward with Omission of Breast Cancer Surgery Following Neoadjuvant Systemic Therapy. Ann Surg Oncol 2022;29:7942-4. [Crossref] [PubMed]
24. Heil J, Pfob A, Morrow M. De-escalation of breast and axillary surgery in exceptional responders to neoadjuvant systemic treatment. Lancet Oncol 2021;22:435-6. [Crossref] [PubMed]
25. Basik M, Costantino JP, De Los Santos JF, et al. NRG Oncology BR005: Phase II trial assessing accuracy of tumor bed biopsies (Bx) in predicting pathologic response in patients (Pts) with clinical/radiological complete response (CR) after neoadjuvant chemotherapy (NCT) in order to explore the feasibility of breast-conserving treatment (BCT) without surgery. J Clin Oncol 2018;36:TPS604.
26. Al-Hilli Z, Thomsen KM, Habermann EB, et al. Reoperation for Complications after Lumpectomy and Mastectomy for Breast Cancer from the 2012 National Surgical Quality Improvement Program (ACS-NSQIP). Ann Surg Oncol 2015;22:S459-69. [Crossref] [PubMed]