Axillary treatment options in clinically nodes-positive breast cancer whose nodes become pathologically nodes-negative after neoadjuvant chemotherapy: a pairwise and network meta-analysis.

1.
Introduction
Breast cancer (BC) is the most frequently diagnosed malignancy among women and remains the leading cause of cancer-related mortality worldwide. Even when both sexes are considered, BC accounted for the second most commonly diagnosed cancer and the fourth leading cause of cancer-related death globally in 2022 [1,2]. Multiple randomized controlled trials (RCTs) have demonstrated that neoadjuvant chemotherapy (NAC) provides survival outcomes comparable to those of adjuvant chemotherapy in BC [3–5]. After NAC, approximately 20–40% of patients initially presenting with clinically node-positive (cN+) disease achieve pathologically node-negative (ypN0) status [6–9]. A large retrospective analysis of the National Surgical Adjuvant Breast and Bowel Project (NSABP) trials B-18 and B-27 showed that patients with cN + and ypN0 disease (T1-3, N0-1, M0) experienced lower locoregional recurrence (LRR) rates than those with residual nodal disease (ypN+), irrespective of surgical approach, including breast-conserving surgery (BCS) or mastectomy [10–12]. Achieving a pathological complete response (pCR) in both the breast and axilla is strongly related to favorable outcomes and has become an important factor in tailoring adjuvant systemic therapy [13]. Despite these advances, controversy remains regarding the optimal management of the axilla in this setting.
A particular clinical dilemma arises in patients who present with cN + disease but convert to ypN0 following NAC. It remains unclear whether axillary management in this setting should be guided by the initial clinical stage or by post-treatment pathological findings. Historically, most locoregional treatment decisions have been informed by data from patients undergoing upfront surgery rather than NAC [14–16]. The growing use and effectiveness of NAC have therefore introduced uncertainty and variability in axillary management strategies. To address this issue, clinical trials have explored tailored axillary approaches aimed at de-escalation, including sentinel lymph node biopsy (SLNB) and targeted axillary dissection (TAD). These studies reflect a broader shift toward minimizing axillary surgery and refining the use of tailored regional radiation therapy (RT). Consequently, post-NAC SLNB [17–19] and TAD [20–25] are increasingly considered less invasive alternatives to axillary lymph node dissection (ALND) in patients with cN + and ypN0 disease. However, nodal downstaging also raises important questions regarding the indication for regional nodal irradiation (RNI). The role of RNI in patients with ypN0 disease remains particularly contentious. Several retrospective studies have evaluated its potential benefit, but the results have been inconsistent [26–35]. Data from French and Korean cohorts suggest that RNI may be safely omitted in patients with stage II-III disease who achieve ypN0 status [28,29]. Similarly, an analysis of the National Cancer Data Base (NCDB) by Kantor et al. found no relation of RNI to improved survival in this population [31]. In contrast, Liu et al. reported improved overall survival (OS) with RNI in patients with clinically stage IIIB or T3 or higher disease [36], and Rusthoven et al. also observed OS benefits associated with RNI [32]. Results from the American College of Surgeons Oncology Group (ACOSOG) Z1071 trial indicated that although RNI was associated with a trend toward improved locoregional recurrence-free survival (LRRFS), it did not result in significant improvements in OS, BC-specific survival (BCSS), or disease-free survival (DFS) [37].
Given these conflicting findings, several ongoing randomized trials aim to clarify the role of RNI in patients with cN + and ypN0 disease. The NSABP B-51/Radiation Therapy Oncology Group (RTOG) 1304 trial (NCT01872975), initiated in August 2013 and closed to enrollment in December 2020, was specifically designed to address this question. Until mature results are available, National Comprehensive Cancer Network (NCCN) guidelines recommend basing locoregional RT decisions after NAC on the maximum disease stage at diagnosis, in conjunction with pathological findings after surgery. Evidence from multiple studies in stage III disease suggests that postoperative RT improves local control even among patients achieving pCR in both the breast and axilla [35,38–40].
Preliminary five-year results from the B-51 trial, published online in June 2025, suggest that patients with cN + disease who convert to ypN0 after NAC may not derive additional benefit from RNI, regardless of whether they undergo BCS or mastectomy [41]. These findings highlight the growing importance of pathological response in guiding axillary management. Continued follow-up is ongoing to evaluate long-term outcomes.
Despite these developments, optimal axillary management in patients with cN + and ypN0 disease remains unsettled. Available evidence is limited by relatively short follow-up and a predominance of retrospective studies. Moreover, most studies have evaluated individual treatment strategies in isolation rather than in direct comparison, complicating clinical decision-making. Two previous meta-analyses examined the impact of RNI in ypN0 patients after NAC and found no clear survival benefit in the cN + subgroup [34,42]. The findings did not seem to confer survival benefits in the cN + and ypN0 subset of patients. Both analyses focused exclusively on patients who underwent mastectomy with RNI and did not compare or rank different axillary treatment strategies. To tackle these challenges, a network meta-analysis (NMA) offers a solution by enabling a thorough and quantitative comparison of the efficacy of various treatment options for this cN + and ypN0 subset of patients [43–45]. Therefore, an NMA provides a robust framework for quantitatively comparing the efficacy of multiple therapeutic strategies in this specific population [43–45]. Notably, no NMA has yet examined axillary management strategies in patients with cN + and ypN0 BC following NAC. Therefore, a pairwise meta-analysis and the first Bayesian NMA were conducted to systematically assess the comparative effectiveness of different axillary management approaches in this setting. The findings are expected to provide robust evidence to support clinical decision-making and to inform the development of individualized, evidence-based guidelines for axillary management in this subset of patients.

2.
Methods
2.1.
Protocol registration
This study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Protocols (PRISMA-P) [46] and NMA statements [47]. The study protocol was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO) to ensure transparency, methodological rigor, and originality (Registration No.: CRD420251111301). The protocol was developed in compliance with the PRISMA 2020 guidelines [48].

2.2.
Data sources and search strategy
PubMed, Web of Science, and Embase were thoroughly retrieved. In addition, the Cochrane Central Register of Controlled Trials (CENTRAL) within the Cochrane Library was searched. All database searches were conducted on July 3, 2025 (final search date). Studies published online ahead of print or in print by this date were eligible for inclusion; preprint servers were not searched. Search strategies were constructed using Medical Subject Headings (MeSH), Emtree terms, and entry terms, incorporating keywords like ‘breast neoplasm’, ‘breast tumor’, ‘neoadjuvant therapy’, ‘axilla’, ‘underarm’, ‘armpit’, and ‘lymph node’. The complete search strategies and corresponding results for each database are presented in Table S1.

2.3.
Eligibility criteria
The inclusion criteria were pre-defined as per the Population Intervention Comparative Outcomes Study (PICOS) design principle. Population (P): Eligible participants were patients with histologically confirmed BC diagnosed as cN+ (based on a positive percutaneous lymph node biopsy for malignancy before NAC) who achieved downstaging to ypN0 following NAC. Male patients and female patients with pregnancy-associated BC, bilateral disease, or distant metastases were excluded. Intervention (I): Various auxiliary treatment modalities following NAC. Comparison (C): Studies comparing the efficacy of different axillary management strategies were included. Outcomes (O): The primary endpoints were OS and DFS. Secondary endpoints included LRRFS, distant metastasis-free survival (DMFS), BCSS, and axillary event-free survival (AFS). Study design (S): Both RCTs and longitudinal observational studies were eligible. Studies were required to compare at least two independent treatment arms. No restrictions were applied regarding histological subtype or specific axillary management techniques. Definitions of each auxiliary treatment strategy and survival outcome are provided in Supplementary Table S2. RNI was treated as a non-uniform exposure in the primary analysis.
Exclusion criteria were: (1) Conference abstracts, summaries, letters, protocols, editorials, guidelines, reviews, commentaries, case reports, single-arm studies, and animal studies; (2) Duplicates; (3) No full text; (4) Lacking adequate data reporting or missing key outcomes of interest; (5) Non-English articles.
To ensure a comprehensive search, the references of included studies and prior meta-analyses were manually screened for pertinent publications.
Literature screening was independently conducted by two reviewers (YG and BBY). Retrieved records were imported into EndNote™ X20 (Clarivate, Philadelphia, PA, USA) for automatic de-duplication, followed by manual verification to remove residual or near-duplicate records. Titles and abstracts were then independently screened in a double-blind manner based on the predefined PICOS criteria. Full texts of potentially eligible studies were subsequently reviewed to confirm eligibility with respect to study population, intervention and comparator arms, outcomes, and study design. Discrepancies were resolved through discussion, and when necessary, adjudicated by senior investigators (STH and TW).

2.4.
Data extraction and quality assessment
A standardized data extraction form was developed to ensure consistency and accuracy. Extracted data included the first author, year of publication, country, study design, axillary management strategy, sample size (restricted to cN + patients achieving ypN0 after NAC), patient age, clinical stage, type of breast surgery, duration of follow-up, and survival outcomes (Table 1). Hazard ratios (HRs) with corresponding 95% confidence intervals (CIs) were extracted; when both unadjusted and adjusted estimates were reported, adjusted HRs were preferentially used. For comparative analyses, ALND served as the reference for surgical interventions (or SLNB when ALND was not reported). ‘No RNI’ was used as the reference for analyses involving RNI. When assessing RNI as an adjunct to axillary surgery, axillary surgery alone was used as the reference. Additionally, ALND plus RNI was designated as the reference comparator for analyses involving SLNB plus RNI, reflecting standard clinical practice in BC management. Effect estimates reported in different formats were standardized using established statistical methods [49,50] to facilitate pooled analysis. When survival data were available only as Kaplan-Meier (KM) curves, Engauge Digitizer 11.3 (© Mark Mitchell) was employed to extract time-to-event data. For studies lacking information on the number of patients at risk across time intervals, estimations were derived following the methods described by Williamson et al. [51] and Tierney et al. [52]. As digitization of KM curves may introduce minor measurement error, these reconstructed data were considered approximate and are acknowledged as a limitation.
Due to limited and heterogeneous reporting of adverse events (AEs), with only two studies providing quantitative toxicity data [41,53], a pooled analysis of treatment-related toxicities was not feasible.
All synthesized data were derived from published literature or supplementary materials. No individual patient-level data were collected. Study quality and risk of bias were independently assessed by two reviewers (YG and BBY). RCTs were evaluated using the National Institutes of Health Quality Assessment Tool (NIH-QAT) for Controlled Intervention Studies [NHLBI, 2013], which comprises 14 signaling questions. Based on responses (‘yes’, ‘no’, ‘other’, or ‘cannot determine’), studies were classified as having low (good), moderate (fair), or high (poor) risk of bias. Non-randomized studies were assessed using the Newcastle-Ottawa Scale (NOS) [54], which allocates up to nine stars across the domains of selection, comparability, and outcome. For RCTs, 0–5, 6–10, and 11–14 were suggested as high, moderate, and low risk, respectively. For non-RCTs, 0–3, 4–6, and 7–9 indicated high, moderate, and low risk, respectively. Dissents were resolved through discussion with senior investigators (STH and TW) to reach consensus.

2.5.
Statistical analysis
After data extraction, statistical analyses were conducted in R 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria) using RStudio. The R version was verified using session information at the time of analysis.
Between-study heterogeneity was assessed using the I2 statistic. A random-effects model was applied when I2 > 50%, whereas a fixed-effects model was used when I2 ≤ 50%. Model fit was further compared using the Deviance Information Criterion (DIC), with a difference <5 indicating comparable model performance [55]. For random-effects analyses, the Hartung-Knapp adjustment was applied to improve the robustness of CIs, particularly in the presence of limited study numbers or moderate heterogeneity [56]. The between-study variance (τ2) was estimated using the DerSimonian-Laird method [57] Forest plots and league tables were generated to summarize comparative efficacy from pairwise meta-analyses and NMA.

2.6.
Statistical methods for pairwise meta-analysis
Pairwise meta-analyses were performed to evaluate direct, head-to-head comparisons between axillary management strategies. Sensitivity analyses were conducted to examine the influence of individual studies on pooled estimates, and subgroup analyses were performed to assess the efficacy of specific axillary procedures. Publication bias was evaluated using Egger’s regression asymmetry test [58]. Further, as per the Cochrane Handbook 5.1.0 [59], potential publication bias was examined by funnel plots and Egger’s test when a sufficient number of studies (n ≥ 10) were found. The study data were visually examined for symmetry of point distribution along the vertical axis of treatment effect [60,61]. Publication bias was considered significant when funnel plot asymmetry was evident or when Egger’s test yielded p < 0.05. The trim-and-fill method was applied to assess the impact of publication bias on result stability. All analyses were two-sided, with statistical significance defined as p < 0.05.

2.7.
Statistical methods for NMA
Statistical methods for Bayesian NMA [62,63] were conducted in RStudio via gemtc and rjags, as most head-to-head comparisons provided limited direct evidence. Each node represents each treatment strategy, and the size of the node is proportional to the total number of subjects (N) included. The thickness of the line represents the number of trials in which each treatment pair was compared.
Posterior distributions were estimated using MCMC simulation with 5,000 burn-in iterations followed by 20,000 iterations across four independent chains, thinned every 10 iterations to reduce autocorrelation [64]. Model convergence was evaluated through trace plots and Gelman-Rubin diagnostics, adopting a cut-off value of 1.00 [65]. Treatment effects were reported as HRs with 95% credible interval (CrI). Local inconsistency was examined using the node-splitting method to compare direct and indirect evidence [66].
Treatment ranking was assessed using the surface under the cumulative ranking curve (SUCRA), with higher values indicating a greater probability of being among the most effective treatment options.

3.
Results
3.1.
Study characteristics and quality
3,275 records were identified from four databases. After the removal of 836 duplicates by automated screening and 163 duplicates manually, 2,276 records remained for title and abstract screening. Of these, 132 articles underwent full-text review. Ultimately, 20 publications [26,27,32,36,41,53,67–80], encompassing 19,870 patients with cN + disease downstaged to ypN0 after NAC and published between 2015 and 2025, met the eligibility criteria. The study selection process is illustrated in Figure 1. The baseline characteristics are summarized in Table 1. Among the included studies, one was an RCT, and nineteen were retrospective observational studies. Axillary management strategies were categorized into two groups: (i) axillary dissection-based approaches (ALND and SLNB); and (ii) axillary irradiation-based strategies involving RNI. In the latter category, the extent or method of upfront axillary surgery was variably reported or unspecified. As this information lay beyond the primary research objective, it was not analyzed further. Combined strategies consisted of either ALND or SLNB administered as adjuncts to RNI.
The RCT [41] was rated as good quality using the NIH Quality Assessment Tool, with limitations primarily related to its open-label design and incomplete allocation concealment. The nineteen non-randomized studies achieved NOS scores of 7–9 stars, reflecting well-defined study objectives, clearly specified eligibility criteria, adequate follow-up, and consistent outcome assessment (Table S3).

3.2.
Pairwise evaluations
Pairwise meta-analyses assessed irradiation versus no irradiation for OS, DFS, LRRFS, DMFS, and BCSS, as well as SLNB versus ALND for OS, DFS, and axillary failure-free survival (AFS). Detailed study-level information and pooled effect estimates are provided in Figure S1. No significant differences were observed between treatment categories for either primary or secondary outcomes. Heterogeneity was low across analyses of OS (I2=42.1%, p = 0.068 for irradiation vs no irradiation; I2=28.4%, p = 0.192 for SLNB vs ALND), AFS (I2=0.0%, p = 0.609), BCSS (I2=0.0%, p = 0.668), and DFS (I2=0.9%, p = 0.423), and fixed-effects models were applied accordingly. Random-effects models were used when I2 exceeded 50%. Sensitivity analyses confirmed the stability of pooled estimates (Figure S2). In studies comparing SLNB with ALND, substantial heterogeneity was detected for DFS (I2=79.9%, p = 0.001). Despite this, the pooled HR remained non-significant under a random-effects model, with 95% CIs crossing unity. Leave-one-out sensitivity analyses demonstrated overall stability. Nevertheless, exclusion of Park S et al. [53] reduced heterogeneity to 65.4%, suggesting this study as a major contributor to variability. Therefore, the pooled DFS estimate warrants cautious interpretation. Mild heterogeneity was also noted for LRRFS in SLNB versus ALND comparisons (I2=53%, p = 0.047). The pooled HR remained non-significant, and sensitivity analysis showed stability; exclusion of Wang Q et al. [77]reduced I2 to 0%, indicating an influence of small sample size. For DMFS, moderate heterogeneity was detected in irradiation versus no irradiation comparisons (I2=55.9%, p = 0.132). The non-significant P value suggested that variability was likely attributable to random error. Sensitivity analyses confirmed result robustness, although exclusion of Mamounas EP et al. [41] or Tan CF et al. [78] rendered I2 uncomputable, implying potential design-related heterogeneity. Subgroup analyses revealed no significant differences among individual axillary procedures, except that SLNB plus RNI was associated with inferior OS compared with SLNB alone (HR: 1.28, 95% CI 1.09–1.51; I2=0%; Figure S3).
All included studies were double-arm designs. Funnel plot analysis for OS in irradiation versus no irradiation comparisons (n = 11) showed no evidence of publication bias (Egger’s test: t= −1.20, df = 9, p = 0.259; Figure S4). Publication bias was not assessed for other outcomes due to limited study numbers.

3.3.
Network evaluations
Bayesian NMA evaluated four axillary strategies, ALND, ALND + RNI, SLNB, and SLNB + RNI, for OS and DFS. Network geometry is shown in Figure 2. For OS, 14 direct comparisons from 12 studies involving 11,955 patients were included [41,53,67–76]. For DFS, 11 direct comparisons from 8 studies involving 2,602 patients were included [41,67–71,75,76]. Model convergence was satisfactory, with Gelman-Rubin statistics of 1.00 for both outcomes (Figure S5). Node-splitting analyses revealed no significant inconsistency between direct and indirect evidence (p > 0.05; Figure S6). Statistical heterogeneity across treatment comparisons was minimal (I2=0% in most cases; Figure S6; Table S4).
Direct and indirect treatment comparisons were further evaluated using league tables (Table 2). Overall, no strategy was statistically superior to ALND, whereas SLNB was superior to SLNB+RNI for OS. Relative to ALND, none of the other three strategies, ALND + RNI (HR: 1.32 [95% CrI: 0.77–2.29]), SLNB (HR: 0.88 [95% CrI: 0.74–1.05]), or SLNB + RNI (HR: 1.14 [95% CrI: 0.90–1.44]), showed a significant OS advantage. However, SLNB demonstrated significantly longer OS than SLNB + RNI (HR: 1.29 [95% CrI: 1.10–1.52]). For DFS, no significant differences were observed among ALND, ALND + RNI, SLNB, and SLNB + RNI across all comparisons.
Based on the SUCRA analysis, SLNB (SUCRA = 0.95) ranked highest in OS, while ALND + RNI (SUCRA = 0.17) ranked the lowest. For DFS, the SUCRA values were closely clustered (range, 0.46–0.54), with ALND (SUCRA = 0.54) ranking highest and SLNB + RNI (SUCRA = 0.46) ranking lowest (Table 2).

4.
Discussion
This study integrates pairwise meta-analysis and the first Bayesian NMA to evaluate oncologic outcomes associated with different axillary management strategies in patients with cN + BC downstaged to ypN0 after NAC. The included studies reported a maximum median follow-up of 120 months. Among this subgroup of cN + and ypN0 patients, no significant differences were noted in OS or event-free survival (EFS) across different axillary treatment modalities, except that SLNB demonstrated favorable OS compared with SLNB combined with RNI. The Bayesian framework was employed for its ability to incorporate prior knowledge and address uncertainty. Within the NMA, SLNB ranked highest for OS, whereas ALND plus RNI ranked lowest. DFS rankings were closely aligned, suggesting comparable disease control among axillary management approaches. These findings align with previous research results [37,41,77,78].
Two previous meta-analyses addressing related questions reported concordant results. Wang et al. showed that post-mastectomy radiotherapy (PMRT), typically incorporating RNI, reduced locoregional recurrence in unselected cN patients converting to ypN0 after NAC, without conferring a survival benefit [34]. Likewise, pooled analysis by Munaser Alamoodi et al. reported that PMRT (including RNI if given) did not yield improved survival outcomes in those with cN1 and ypN0 for stages T1-3 [42]. However, both analyses focused primarily on mastectomy cohorts treated with RNI and did not compare or rank alternative axillary strategies. The present NMA addresses these gaps by jointly evaluating surgical- and irradiation-based approaches across varied clinical contexts. By simultaneously comparing multiple treatment options, this analysis provides a broader synthesis of available evidence and offers clinically relevant insights for individualized axillary management in cN + and ypN0 patients following NAC.
The more favorable of SLNB, probably related to patients’ characteristics and the relation to treatment. The two retrospective studies comparing SLNB with SLNB+RNI (size n 1148 and 1040, respectively) [71,75] primarily included patients with T0-2, cN1 or any T, cN1 disease, for whom multimodal treatment might have been associated with a higher frequency of AEs related to treatment.
Several factors may explain the overall equivalence observed among treatment strategies. The included studies spanned more than two decades, during which substantial advances occurred in radiotherapy, axillary surgery, and systemic therapy. Radiotherapy evolved from two-dimensional techniques to CT-based three-dimensional conformal radiotherapy and intensity-modulated radiotherapy, improving target conformity while reducing exposure to the heart, lungs, upper limb, and brachial plexus. Concurrently, axillary surgery underwent progressive de-escalation from ALND to SLNB or TAD. Irradiation fields also became more standardized, typically encompassing axillary levels I-III with supraclavicular coverage and selective inclusion of internal mammary nodes. Despite these advances, no additional survival benefit from routine RNI was identified in patients with cN + and ypN0 disease. The close similarity in DFS rankings supports the notion that modern NAC effectively eradicates micrometastatic disease, thereby limiting residual recurrence risk.
Parallel improvements in systemic therapy likely reinforced this trend. During the same period, NAC increasingly incorporated targeted and immunotherapeutic agents tailored to molecular subtype. In HER2-positive BC, trastuzumab- or pertuzumab-based regimens combined with taxanes or anthracyclines became standard, followed by adjuvant HER2-directed therapy. In triple-negative BC, immune checkpoint inhibitors such as pembrolizumab or atezolizumab were integrated into platinum-based neoadjuvant regimens, with some protocols extending immunotherapy postoperatively. These approaches enhance systemic disease control by eliminating circulating tumor cells and occult metastases, thereby complementing locoregional therapy. In this modern multimodal context, the lack of added benefit from extensive axillary surgery or irradiation likely reflects the effectiveness of contemporary systemic treatment.
However, the absence of detailed patient-level data precluded further stratification by clinically relevant variables like age, race, surgical type (BCS versus mastectomy), lymphovascular invasion (LVI), hormone receptor or HER2 status, achievement of pCR in the breast, and receipt of adjuvant or novel systemic agents. These factors are particularly relevant, as HER2-positive and triple-negative subtypes demonstrate higher response rates to NAC and may exhibit distinct patterns of locoregional control and survival [13,81,82]. The lack of adjustment for these parameters possibly has introduced residual heterogeneity, underscoring the need for future patient-level meta-analyses. In NMA, indirect comparisons rely on the assumption of transitivity, whereby key effect modifiers are similarly distributed across comparisons. Although baseline nodal stage was summarized at the study level (Table 1), heterogeneity in reporting of nodal burden, surgical approach, systemic therapy, and indications for RNI limited full verification of this assumption. Accordingly, indirect estimates should be interpreted with caution.
Following NAC for cN + disease, SLNB alone has been related to a false-negative rate exceeding 10% [17–19], unless dual tracers are used, at least three sentinel nodes are retrieved, and the clipped positive node is excised. TAD can reduce this rate to approximately 2% [20,83]. When reliable nodal staging is ensured, escalation of axillary treatment loses justification, a pattern reflected in the observed survival hierarchy. These findings support the oncologic safety of limited axillary surgery in appropriately selected patients. Although TAD is increasingly adopted in clinical practice, it cannot be evaluated separately due to insufficient comparative survival data. Therefore, SLNB-related findings should not be extrapolated to TAD without qualification.
Regarding RNI, the NCCN Guidelines [Breast Cancer. Version 4.2025. available from: https://www.nccn.org/professionals/physician_gls/pdf/breast.pdf] recommend comprehensive RNI (including any undissected at-risk axilla) in conjunction with whole-breast radiotherapy (WBRT) after BCS, or PMRT to the chest wall plus comprehensive RNI (Category 2 A). However, emerging evidence suggests that selected patients achieving ypN0 status may derive limited benefit from routine RNI [41]. The guidelines, therefore, emphasize individualized decision-making based on both initial clinical stage and final pathological findings. In clinical practice, RNI is commonly reserved for higher-risk features, including initial cN2-3 disease, medial or central tumor location, large tumor size, aggressive biology, younger age, or uncertain axillary mapping.
Although efficacy remains the primary endpoint, treatment safety is equally important. AE reporting was inconsistent across studies, precluding quantitative comparisons. Severe toxicities were uncommon, whereas mild to moderate radiation dermatitis and lymphedema were most frequently reported. Standardized and comprehensive toxicity reporting in future trials is essential to enable robust safety analyses and inform individualized treatment decisions.
There are several important limitations. First, radiotherapy dose prescriptions and techniques varied substantially. Patients were treated using two-dimensional, three-dimensional conformal, or intensity-modulated radiotherapy, with axillary doses typically ranging from 45 to 54 Gy in 1.8–2.0 Gy fractions. Variability in technique, field design, and fractionation may influence locoregional control and morbidity, thereby diluting estimated RNI effects and limiting generalizability. Neoadjuvant regimens and surgical criteria were also heterogeneous, further contributing to variability. Second, many included studies were nonrandomized and retrospective, and residual confounding cannot be fully excluded. Notably, the RNI category encompassed heterogeneous implementations (e.g. following ALND, following SLNB, or with unspecified axillary surgery), potentially obscuring clinically relevant distinctions. This heterogeneity, together with confounding by indication, may partly explain the variability in observed treatment effects. Substantial differences in sample size, study setting, and treatment protocols contributed to between-study heterogeneity. Therefore, any observed OS advantage, particularly those favoring SLNB over SLNB + RNI, should be interpreted as associative rather than causal and may reflect patient selection and treatment indication. Moreover, the transitivity assumption required for indirect comparisons cannot be fully verified using aggregated study-level data. Finally, SUCRA reflects probabilistic rankings rather than clinical effect sizes and does not account for confounding or heterogeneity; therefore, treatment de-escalation decisions should not rely on SUCRA rankings alone. Third, only one study [68] classified patients with isolated tumor cells (ITCs, pN0(i+), defined as malignant cell clusters ≤0.2 mm) as ypN0 and incorporated them into analyses of OS, DFS, and AFS. This classification aligns with current American Joint Committee on Cancer (AJCC) and NCCN staging guidelines, which consider pN0(i+) a low-risk subset within pN0 and distinct from micrometastatic disease (pN1mi, >0.2 mm to ≤2.0 mm). Accordingly, no additional treatment beyond that recommended for ypN0 disease is advised. Although ITCs represent minimal residual disease and may theoretically exert an adverse prognostic effect, a survival benefit from RNI, if present, would be expected to manifest as a directional improvement. However, no such effect was observed. Moreover, as this classification was applied in only a single dataset, it is unlikely to materially influence the pooled estimates. Therefore, any observed OS neutrality in the network analysis can reasonably be interpreted as a conservative and methodologically acceptable outcome rather than a source of bias. Fourth, between-study heterogeneity represents a major limitation, reducing the certainty of several estimates, particularly for DFS; pooled DFS results should therefore be interpreted cautiously. Sensitivity analyses indicated that heterogeneity was partly driven by individual studies, most notably Park et al. [53] in the SLNB versus ALND comparison for DFS. This variability likely reflects the limited number of available studies, small sample sizes, and heterogeneity in study design and treatment protocols. Lastly, partial overlap of study populations or clinically comparable subpopulations is common in real-world data and cannot be fully addressed in meta-analyses. The combined use of pairwise and network meta-analytic approaches in this study helps mitigate these influences and enhances the stability and robustness of the pooled estimates.
Therefore, advances in radiotherapy techniques, surgical management, and systemic therapies have substantially attenuated survival differences among axillary treatment strategies in the NAC era. Nevertheless, to validate these observations and to more precisely define the optimal axillary management for patients with cN + and ypN0 BC, particularly those achieving pathologic complete response in both the breast and axilla, future prospective, randomized studies with long-term follow-up and standardized radiotherapy, surgical, and systemic treatment protocols are warranted.
The foregoing results support tailored axillary management in carefully selected patients based on oncologic outcomes. However, oncologic noninferiority alone does not establish overall clinical benefit, as the net value of treatment de-escalation also depends on morbidity and patient-reported outcomes (PROs). These endpoints were infrequently and inconsistently reported across the included comparative studies (qualitatively summarized in Table S5), precluding quantitative synthesis. Moreover, survival estimates in real-world settings are strongly influenced by systemic treatment response and treatment selection, underscoring the importance of incorporating standardized morbidity and PRO endpoints when interpreting existing evidence and designing tailored axillary strategies [84].

5.
Conclusion
In patients with cN + and ypN0 BC after NAC, tailored axillary strategies achieved OS and DFS comparable to those of conventional or more extensive treatments in pairwise and Bayesian network meta-analyses. In these tailored axillary treatments, SLNB demonstrated noticeable benefits, supporting its use in appropriately selected patients. These findings highlight the implementation of individualized, less invasive axillary approaches, which warrant confirmation in large-scale prospective studies with extended follow-up.

Supplementary Material

Supplementary materials R1.docx

PRISMA 2020 Checklist.docx