Association of neoadjuvant therapy with locoregional recurrence according to recurrence definitions in breast cancer: a retrospective cohort study

Abstract

Background

The impact of neoadjuvant therapy (NAT) on locoregional recurrence (LRR) remains controversial, particularly due to differences in recurrence definitions and the presence of competing events such as distant metastasis. This study aimed to evaluate the association between NAT and LRR using distinct recurrence definitions and appropriate analytical approaches.

Methods

We conducted a retrospective cohort study of patients with invasive breast cancer who underwent surgery at a single institution. LRR was evaluated using two definitions: Any LRR, defined as locoregional recurrence regardless of distant metastasis, and True LRR, defined as isolated locoregional recurrence without distant metastasis. Multivariable Cox regression, propensity score–based inverse probability of treatment weighting (IPTW), and competing risk analysis using the Fine–Gray model were applied.

Results

A total of 1,670 patients were included, of whom 323 received NAT and 1,347 did not. During a median follow-up of approximately 52.8 months, 79 patients (4.7%) developed Any LRR and 51 patients (3.1%) developed True LRR. NAT was observed to be associated with a higher incidence of Any LRR in multivariable analysis. Although a similar trend was observed for True LRR, the association did not consistently reach statistical significance across models. Competing risk analysis yielded similar findings for True LRR when distant metastasis was treated as a competing event.

Conclusions

NAT was observed to be associated with a higher incidence of LRR in this retrospective cohort study; however, this association should be interpreted with caution due to substantial residual confounding and temporal heterogeneity. No causal relationship between NAT and LRR can be established. Instead, these findings highlight the importance of recurrence definitions and analytical approaches, including competing risk methods, in interpreting locoregional outcomes. These results should be considered hypothesis-generating and warrant further validation in well-designed studies.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12957-026-04488-z.

Background

Neoadjuvant therapy (NAT) has become an integral component of modern breast cancer management, providing well-established benefits in tumor downstaging, increased rates of breast-conserving surgery, and in vivo assessment of treatment response [1]. Both neoadjuvant chemotherapy (NAC) and neoadjuvant endocrine therapy (NET) are now routinely employed across a wide range of tumor subtypes and clinical stages in accordance with contemporary clinical guidelines [1]. While the survival benefit of NAT has been extensively evaluated in randomized trials and meta-analyses [2, 3], its impact on postoperative locoregional recurrence (LRR) remains incompletely understood.

The impact of neoadjuvant therapy on postoperative locoregional recurrence remains controversial, largely due to heterogeneity in recurrence definitions and analytical approaches across studies. As a result, it remains unclear whether reported differences in locoregional recurrence after neoadjuvant therapy reflect true impairment of locoregional control or differences in endpoint definition and methodology.

Several studies have suggested higher rates of locoregional or regional recurrence following NAT compared with upfront surgery, raising concerns regarding locoregional control [4]. Proposed explanations include residual microscopic disease in initially involved lymph nodes, challenges in accurately assessing nodal response after systemic therapy, and heterogeneity in surgical and radiotherapy strategies after NAT [4, 6]. In contrast, more recent series incorporating modern systemic therapy, optimized surgical techniques, and tailored radiotherapy have reported acceptable locoregional control after NAT [5, 6]. Interpretation of these findings is further complicated by substantial variability in the definition of LRR, particularly with respect to whether events accompanied by synchronous distant metastasis are included, which may lead to overestimation of the direct impact of NAT on locoregional failure [7]. Moreover, most available evidence is derived from Western populations, and data from Asian cohorts remain limited [8]. Against this background, we conducted a retrospective cohort study to evaluate the association between NAT and locoregional recurrence after definitive breast surgery by clearly distinguishing between True LRR (without distant metastasis) and Any LRR (regardless of synchronous distant metastasis), aiming to clarify how differences in recurrence definitions and analytical approaches influence the observed association between neoadjuvant therapy and locoregional recurrence.

Methods

Study design and patient selection

We conducted a retrospective single-center cohort study of patients with invasive breast cancer who underwent definitive breast surgery at our institution between January 2000 and September 2025. A total of 1,931 patients were initially identified.

Patients with ductal carcinoma in situ, distant metastasis at initial diagnosis, or missing key clinicopathological or follow-up data were excluded. The exclusion criteria were predefined and consistently applied throughout the study cohort. Ultimately, 1,670 patients with invasive breast cancer were included in the final analysis.

Patients who received any form of neoadjuvant therapy (NAT) were classified as the NAT group, while those who underwent upfront surgery were classified as the non-NAT group. Patients in the NAT group were further categorized according to treatment modality as neoadjuvant chemotherapy (NAC) or neoadjuvant endocrine therapy (NET). All clinical, pathological, and treatment-related data were extracted from electronic medical records. The study protocol was revised according to the institutional ethics committee’s recommendations prior to approval.

Definitions of recurrence

Locoregional recurrence (LRR) was defined as tumor recurrence occurring in the ipsilateral breast or chest wall (local recurrence) and/or in regional lymph nodes, including the axillary, supraclavicular, infraclavicular, interpectoral, and internal mammary lymph nodes.

In accordance with prior studies distinguishing isolated locoregional recurrence from locoregional recurrence accompanied by distant metastasis, LRR events were further classified into two categories [9–11]. True LRR was defined as local and/or regional recurrence occurring in the absence of distant metastasis, corresponding to the concept of isolated locoregional recurrence used in previous clinical trials and cohort studies [9, 10]. Any LRR was defined as local and/or regional recurrence regardless of the presence of synchronous or subsequent distant metastasis [5, 11].

Distant metastasis was defined as recurrence occurring outside the ipsilateral breast/chest wall and regional lymph node basins, including bone, lung, liver, brain, or other distant organs.

Patients who developed distant metastasis without evidence of locoregional recurrence were classified as having distant metastasis alone and were not included in the LRR categories. The date of recurrence was defined as the date of radiologic or pathological confirmation of the first recurrence event.

Variables and covariates

Clinicopathological variables included age, tumor size (continuous), nodal status (positive vs. negative), intrinsic subtype (hormone receptor–positive/HER2-negative, HER2-positive, and triple-negative), surgical procedure (breast-conserving surgery vs. mastectomy), postoperative radiotherapy (yes/no), and receipt of NAT (yes/no).

Clinical stage and reconstructive surgery status were not included in the primary multivariable models to minimize collinearity and potential overadjustment, as these factors were closely correlated with tumor size, nodal status, and surgical procedure.

Surgical decision policy

At our institution, mastectomy is generally selected for patients with true multifocal invasive disease, whereas extensive intraductal components are regarded as ductal extension of a single lesion and managed based on resectability and margin status. Consequently, multifocal residual disease was uncommon among patients undergoing breast-conserving surgery in this cohort, reflecting institutional surgical selection practices.

Radiotherapy policy

Adjuvant radiotherapy was administered according to institutional guidelines, including all patients undergoing breast-conserving surgery and selected high-risk mastectomy patients. Regional nodal irradiation was not routinely performed in the early study period but became more common in later years, reflecting temporal changes in radiotherapy practice. Treatment fields were determined through multidisciplinary discussion.

Statistical analysis

Continuous variables are presented as mean ± standard deviation or median (interquartile range), as appropriate, and were compared using the Student’s t test or the Wilcoxon rank-sum test. Categorical variables are expressed as number (percentage) and were compared using the chi-square test or Fisher’s exact test, as appropriate.

Time-to-event outcomes were analyzed using Kaplan–Meier methods, and differences between groups were assessed using the log-rank test. Locoregional recurrence (LRR)–free survival was calculated from the date of surgery to the date of recurrence or last follow-up. Median follow-up was calculated from the date of surgery to the date of last clinical follow-up or death.

Two distinct LRR endpoints were evaluated: True LRR, defined as local or regional recurrence occurring in the absence of distant metastasis, and Any LRR, defined as local or regional recurrence regardless of the presence of synchronous distant metastasis.

Univariate and multivariable Cox proportional hazards models were used to evaluate factors associated with True and Any LRR. Multivariable models were adjusted for tumor size, nodal status, and tumor subtype, which were selected a priori as clinically relevant baseline covariates. Surgical procedure and postoperative radiotherapy were not included in the primary multivariable models because these factors may act as downstream variables influenced by neoadjuvant therapy. Given the limited number of recurrence events, the number of covariates included in multivariable models was intentionally restricted to clinically relevant baseline variables selected a priori.

Ki-67 was not included in the multivariable Cox models owing to substantial missing data and the limited number of recurrence events. The proportional hazards assumption was assessed using Schoenfeld residuals. Hazard ratios (HRs) are presented with 95% confidence intervals (CIs).

Because patients who develop distant metastasis are no longer at risk of isolated locoregional recurrence, distant metastasis was treated as a competing event. Accordingly, competing risk analyses using the Fine–Gray subdistribution hazard model were performed to estimate the cumulative incidence of True LRR, as previously described [12].

To further address potential confounding due to non-random treatment allocation, a propensity score for receipt of NAT was estimated using a logistic regression model including age, tumor size, nodal status, and intrinsic subtype. Inverse probability of treatment weighting (IPTW) with stabilized weights was applied to generate a weighted pseudo-population, as described in previous methodological studies [13]. Covariate balance was assessed using standardized mean differences (SMD), with an absolute value < 0.1 considered acceptable. The distribution of stabilized weights was visually inspected to assess model stability. Additional sensitivity analyses were performed using Cox models that included postoperative radiotherapy to evaluate whether the association between NAT and LRR was influenced by differences in radiation treatment.

All statistical tests were two-sided, and a p-value < 0.05 was considered statistically significant. Statistical analyses were performed using R software (version 4.5.2; R Foundation for Statistical Computing, Vienna, Austria).

Results

Patient characteristics

A total of 1,931 patients were initially identified. After applying the predefined exclusion criteria, 1,670 patients with invasive breast cancer were included in the final analysis. Among these, 1,644 patients were eligible for time-to-event analyses after excluding those with missing data required for survival analysis (Fig. 1).

Fig. 1.

Patient selection and study flow. Flowchart of patient selection and classification according to neoadjuvant therapy (NAT) status. A total of 1,931 patients were initially identified. After applying exclusion criteria, 1,670 patients were included in the final cohort, and 1,644 patients were eligible for time-to-event analysis.Locoregional recurrence was categorized into Any LRR, defined as locoregional recurrence regardless of distant metastasis, and True LRR, defined as locoregional recurrence occurring in the absence of distant metastasis. All recurrence events were defined based on the first documented recurrence

Of the included patients, 323 (19.3%) received neoadjuvant therapy (NAT), including neoadjuvant chemotherapy (NAC) and neoadjuvant endocrine therapy (NET), while 1,347 (80.7%) underwent upfront surgery. Among those who received NAT, 273 underwent NAC and 50 underwent NET. Among the 50 patients who received NET, 4 developed Any LRR and 2 developed True LRR (Supplementary Table S7).

Baseline clinicopathological characteristics differed between the groups (Table 1). Patients in the NAT group had larger tumors, a higher frequency of nodal involvement, and more aggressive tumor subtypes compared with those in the non-NAT group. Detailed treatment regimens varied substantially during the study period and were therefore not incorporated into primary analyses.

Table 1.

Clinicopathological characteristics according to receipt of neoadjuvant therapy

Variable	Overall (n = 1,670)	Non-NAT (n = 1,347)	NAT (n = 323)	p-value
Age, years	61.4 ± 13.8	62.0 ± 14.0	59.0 ± 12.9	< 0.001
Tumor size, cm	2.44 ± 1.80	2.15 ± 1.52	3.59 ± 2.32	< 0.001
Node-positive	465 (28.6%)	356 (27.3%)	109 (34.3%)	0.016
Subtype				< 0.001
HR+/HER2−	1,119 (80.4%)	981 (85.9%)	138 (55.2%)
HER2+	106 (7.6%)	59 (5.2%)	47 (18.8%)
Triple-negative	167 (12.0%)	102 (8.9%)	65 (26.0%)
Surgery (mastectomy)	1,095 (65.6%)	869 (64.5%)	226 (70.0%)	0.074
Postoperative radiotherapy	595 (38.1%)	450 (36.1%)	145 (45.7%)	0.002

Values are presented as mean ± standard deviation or number (%)

P values were calculated using the t test for continuous variables and the chi-square test for categorical variables, as appropriate

Subtype data were available for 1,392 patients (Non-NAT, n = 1,142; NAT, n = 250)

Abbreviations: NAT neoadjuvant therapy

Recurrence events

During a median follow-up period of approximately 52.8 months (1588 days in the NAT group and 1615 days in the non-NAT group), 79 patients (4.7%) developed Any LRR and 51 patients (3.1%) developed True LRR. The distribution of recurrence patterns according to NAT status is summarized in Supplementary Table S2. Kaplan–Meier curves for Any LRR-free survival are shown in Fig. 2, and cumulative incidence curves for True LRR using competing risk analysis are presented in Fig. 3.

Fig. 2.

Kaplan–Meier curves for Any LRR-free survival according to neoadjuvant therapy (NAT) status.Kaplan–Meier curves showing Any LRR-free survival stratified by NAT status. Any LRR was defined as locoregional recurrence regardless of the presence of distant metastasis. Numbers at risk are shown below the x-axis

Fig. 3.

Cumulative incidence of True LRR according to neoadjuvant therapy (NAT) status. Cumulative incidence curves for True LRR, estimated using the Fine–Gray competing risk model with distant metastasis treated as a competing event. True LRR was defined as locoregional recurrence occurring in the absence of distant metastasis

Multivariable Cox analysis

In multivariable Cox proportional hazards models adjusted for tumor size, nodal status, and intrinsic subtype, NAT was associated with an increased risk of Any LRR (HR 1.91, 95% CI 1.01–3.60, p = 0.046; Table 2). For True LRR, NAT showed a similar direction of effect (HR 2.16, 95% CI 0.97–4.82), although the association did not reach statistical significance (p = 0.060; Table 2).

Table 2.

Cox proportional hazards models for locoregional recurrence

Outcome	Variable	Adjusted HR	95% CI	p-value
True LRR	NAT (vs. non-NAT)	2.16	0.97–4.82	0.060
	Tumor size (per cm)	0.96	0.75–1.22	0.715
	Node-positive (vs. node-negative)	0.60	0.27–1.33	0.208
	HER2 + vs. HR+/HER2−	2.04	0.86–4.85	0.106
	TNBC vs. HR+/HER2−	0.55	0.16–1.84	0.329
Any LRR	NAT (vs. non-NAT)	1.91	1.01–3.60	0.046
	Tumor size (per cm)	0.99	0.82–1.19	0.899
	Node-positive (vs. node-negative)	1.03	0.59–1.79	0.917
	HER2 + vs. HR+/HER2−	1.44	0.67–3.13	0.352
	TNBC vs. HR+/HER2−	0.72	0.30–1.71	0.456

Adjusted for tumor size, nodal status, and intrinsic subtype

Hazard ratios (HRs) were estimated using multivariable Cox proportional hazards models

True locoregional recurrence (True LRR) was defined as local or regional recurrence occurring in the absence of distant metastasis

Any locoregional recurrence (Any LRR) was defined as local or regional recurrence regardless of the presence of synchronous distant metastasis

Abbreviations: LRR locoregional recurrence, HR hazard ratio, CI confidence interval, NAT neoadjuvant therapy, TNBC triple-negative breast cancer

Competing risk analysis

In Fine–Gray competing risk models treating distant metastasis as a competing event, NAT was associated with an increased subdistribution hazard for True LRR (sHR 2.28, 95% CI 1.08–4.81, p = 0.03; Supplementary Table S4).

IPTW analysis

IPTW based on propensity scores was applied. Covariate balance improved after weighting. In IPTW-weighted Cox models, NAT remained significantly associated with an increased risk of Any LRR (HR 2.23, 95% CI 1.15–4.33, p = 0.017; Table 3), with consistent effect estimates compared with the primary multivariable models. A similar trend was observed for True LRR (HR 2.18, 95% CI 0.93–5.08), although statistical significance was not reached (p = 0.073; Table 3). After IPTW, all standardized mean differences were below 0.1 (Fig. 4).

Table 3.

IPTW-weighted Cox proportional hazards models for locoregional recurrence

Outcome	HR	95% CI	p-value
Any LRR	2.23	1.15–4.33	0.017
True LRR	2.18	0.93–5.08	0.073

Propensity scores were estimated using a logistic regression model including age, tumor size, nodal status, and intrinsic subtype

Models were weighted using stabilized inverse probability weights　in the IPTW analysis 

IPTW inverse probability of treatment weighting

Fig. 4.

Covariate balance before and after inverse probability of treatment weighting (IPTW).Absolute standardized mean differences for baseline covariates before (blue circles) and after IPTW (red triangles). The vertical dashed line indicates an absolute standardized mean difference of 0.1, representing acceptable covariate balance

Sensitivity analysis including postoperative radiotherapy

Additional Cox models including postoperative radiotherapy were performed to evaluate whether differences in radiation treatment influenced the observed associations.

In sensitivity analyses including postoperative radiotherapy, NAT remained significantly associated with an increased risk of True LRR (HR 2.51, 95% CI 1.16–5.41, p = 0.019; Table 4), whereas postoperative radiotherapy itself was not significantly associated with recurrence.

Table 4.

Sensitivity analysis using Cox models including postoperative radiotherapy

Variable	True LRR HR	95% CI	p-value
NAT	2.51	1.16–5.41	0.019
Tumor size	0.97	0.79–1.20	0.807
Node positive	0.55	0.24–1.27	0.159
Postoperative radiotherapy	0.97	0.50–1.89	0.935

Hazard ratios (HRs) were estimated using multivariable Cox proportional hazards models including postoperative radiotherapy

Subtype was not included due to model instability resulting from sparse events

Model stability

Models including detailed subtype classification in combination with postoperative radiotherapy were unstable due to sparse event distribution, and therefore were not used for primary interpretation.

In an exploratory era-based analysis, recurrence rates were lower among patients treated during 2011–2025 than among those treated during 2000–2010. True LRR occurred in 24 of 505 patients (4.8%) in the earlier era and in 27 of 1,139 patients (2.4%) in the later era. Similarly, Any LRR occurred in 34 patients (6.7%) and 41 patients (3.6%), respectively (Supplementary Table S8).

Discussion

In this retrospective cohort study, neoadjuvant therapy (NAT) was observed to be associated with a higher incidence of locoregional recurrence (LRR) in this cohort. However, this association should be interpreted with caution, as several methodological limitations may have influenced the observed relationship.

First, residual confounding by indication remains a major concern. Patients who received NAT generally had more advanced disease and more aggressive tumor biology at baseline, both of which are established risk factors for LRR [5]. Although multivariable adjustment and propensity score–based weighting were applied [13], important clinical variables such as initial clinical stage, clinical nodal burden, treatment regimens, pathological response, and details of locoregional treatment were not fully captured in this study. Therefore, the observed association may reflect underlying disease severity rather than the effect of NAT itself.

Second, the long study period (2000–2025) introduces potential temporal heterogeneity. During this period, substantial advances occurred in systemic therapy, including HER2-targeted therapies [1–3], as well as improvements in imaging modalities, surgical techniques, and radiotherapy strategies [4, 6]. These changes may have influenced both treatment selection and recurrence outcomes, thereby affecting the comparability between groups. In an exploratory era-based analysis, recurrence rates were lower among patients treated during 2011–2025 than among those treated during 2000–2010. This finding may partly reflect advances in systemic therapy, including the widespread adoption of HER2-targeted therapies, as well as improvements in imaging, surgery, and radiotherapy (Supplementary Table S8).

Third, the number of LRR events, particularly for True LRR, was limited. As a result, estimates derived from multivariable and competing risk models [12] may be subject to statistical instability, and the findings should be interpreted with caution. Confidence intervals in competing risk models remained relatively imprecise because of the limited number of True LRR events. In addition, some established prognostic factors did not show expected associations, possibly due to limited event numbers and model instability.

Importantly, our findings do not imply a causal relationship between NAT and LRR. Instead, the observed association is likely influenced by treatment selection and baseline disease characteristics. Therefore, NAT should primarily be interpreted as a marker of higher baseline recurrence risk rather than an independent determinant of locoregional recurrence. This interpretation is particularly relevant when considering how locoregional recurrence is defined and analyzed.

One notable aspect of this study is the distinction between Any LRR and True LRR. While Any LRR reflects all locoregional recurrence events regardless of distant metastasis, True LRR captures isolated locoregional failure. This distinction is clinically meaningful, as the interpretation of recurrence risk may differ depending on whether competing events such as distant metastasis are taken into account [11]. Among the two endpoints, True LRR may be the more clinically relevant measure of locoregional control because it excludes recurrence events occurring in the setting of distant metastatic progression. In contrast, Any LRR reflects the overall burden of locoregional recurrence but may be influenced by concurrent distant metastasis. Because True LRR events were less frequent, both endpoints should be interpreted as complementary measures of locoregional outcome.

Previous studies evaluating LRR after NAT have reported heterogeneous results, partly due to differences in tumor subtype, treatment strategy, and endpoint definition [5, 8]. Recent studies have also suggested that locoregional outcomes may vary according to biological subtype, baseline disease characteristics, and surgical management after NAT [8, 14]. In this context, the present study indicates that the apparent association between NAT and LRR may vary depending on how recurrence events are defined and analyzed.

Furthermore, the use of competing risk analysis provided additional insights into the risk of True LRR by accounting for distant metastasis as a competing event [12]. These findings indicate that both endpoint definition and statistical methodology play a crucial role in interpreting LRR outcomes.

NAC and NET were analyzed together because the limited number of recurrence events precluded reliable subgroup analyses. Nevertheless, these therapies differ substantially in indication and biological context, and future studies should evaluate them separately.

From a clinical perspective, our results do not suggest that NAT itself independently increases the risk of LRR. Rather, they highlight the importance of cautious interpretation of recurrence outcomes in observational settings, particularly in the presence of treatment selection bias and competing risks.

Therefore, these findings should be considered hypothesis-generating, and further studies incorporating detailed clinical staging, treatment variables, and contemporary cohorts are warranted to better clarify the relationship between NAT and locoregional recurrence.

Conclusion

Neoadjuvant therapy was observed to be associated with a higher incidence of locoregional recurrence in this retrospective cohort study. However, this association should be interpreted with caution due to substantial residual confounding and temporal heterogeneity. Given these limitations, no causal relationship between neoadjuvant therapy and locoregional recurrence can be established. Instead, our findings underscore the importance of recurrence definitions and analytical approaches, including competing risk methods, in interpreting locoregional outcomes. These results should be considered hypothesis-generating and require validation in well-designed prospective studies.

Abbreviations

Any LRR

Any locoregional recurrence

BCS

Breast-conserving surgery

CI

Confidence interval

ER

Estrogen receptor

HER2

Human epidermal growth factor receptor 2

HR

Hazard ratio

IQR

Interquartile range

LRR

Locoregional recurrence

NAC

Neoadjuvant chemotherapy

NAT

Neoadjuvant therapy

NET

Neoadjuvant endocrine therapy

PgR

Progesterone receptor

True LRR

True locoregional recurrence

Authors' contributions

Yoshiko Masuda: Conceptualization, data curation, formal analysis, writing – original draft, and supervision. Rina Kudo, Kyoko Eto, Yukari Koga, Shuko Ono, Masumi Tanaka: Investigation, resources, data collection, and review & editing. Yasuteru Yoshinaga, Toshihiko Sato: Supervision, validation, and critical review of the manuscript. All authors read and approved the final manuscript.

Funding

This study was not supported by any external sponsor or funding source.

Data availability

The data that support the findings of this study are included in this published article and its supplementary information files. Additional data are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Fukuoka University Hospital (approval number: H26-04-005).

Because this study used only existing clinical data and involved no intervention, the requirement for written informed consent was waived. An opt-out method was implemented.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.