Optimal surgical margin and the efficacy of adjuvant radiotherapy in borderline and malignant phyllodes tumors: a multicenter retrospective study

Abstract

Purpose

Optimal management of borderline and malignant phyllodes tumors (PT) is not well established, especially regarding margin width and the use of adjuvant radiotherapy (RT). This multicenter study aimed to identify prognostic factors and to assess the effectiveness of adjuvant RT.

Materials and Methods

We conducted a retrospective analysis of 183 patients (155 surgery-only, 28 adjuvant RT) diagnosed with borderline or malignant PT from 2009 to 2025. To address selection bias, whereby RT is preferentially given to higher-risk patients, we used overlap weighting based on propensity scores, which estimates the average treatment effect in the overlap population (ATO). Independent prognostic factors were identified through multivariable Cox regression with Firth's penalized likelihood.

Results

A resection margin <1 mm was identified as the only independent predictor of local recurrence in multivariable analysis (hazard ratio [HR], 2.07; 95% confidence interval [CI], 1.03 to 4.08; p = 0.043). Although conventional analysis did not show a benefit from RT, overlap-weighted analysis suggested an improved local recurrence-free survival with adjuvant RT, though this was not statistically significant (HR, 0.25; 95% CI, 0.05 to 1.19; p = 0.081). Distant metastasis events were rare (n = 3), and no significant difference in distant metastasis–free survival was observed after overlap weighting (HR, 2.27; 95% CI, 0.33 to 15.66; p = 0.404). In an exploratory subgroup analysis, no local recurrence (LR) was observed in the 12 borderline patients who received adjuvant RT (0/12 vs. 27/136 in surgery-only).

Conclusion

Margins <1 mm were independently linked to an increased risk of LR. After adjusting for selection bias, adjuvant RT showed a trend toward reduced LR, though it did not achieve statistical significance. Adjuvant RT might be considered for selected high-risk patients.

Introduction

First described by Johannes Müller in 1838, phyllodes tumors (PT) represent a rare spectrum of fibroepithelial breast neoplasms [1]. It accounts for approximately 0.3% to 1% of all breast tumors. Epidemiological studies have shown that Asian populations have a higher incidence and younger age at presentation compared to Caucasians [2]. Under the World Health Organization (WHO) classification system, tumors are categorized into three subtypes: benign, borderline, and malignant [3]. While clinically characterized by their rapid growth and palpable presentation, PTs display a heterogeneous behavioral profile rather than functioning as a uniform clinical entity [4].

Despite their rarity, the clinical burden of PT is significant, with reported rates of local recurrence and distant metastasis reaching 8%–9% and 8%–10%, respectively [3,5]. This presents a significant therapeutic challenge. While many prognostic markers—such as tumor size, resection margins (RMs), and histological features including stromal atypia, cellularity, overgrowth, and mitotic activity—have been proposed, their incorporation into standardized treatment protocols remains inconsistent [6]. Existing evidence primarily comes from either small single-institution studies or multi-institutional collaborations, which may suffer from variations in diagnostic criteria, treatment protocols, indications for adjuvant radiotherapy (RT), and follow-up practices [7-11].

The Catholic Medical Center (CMC) offers a valuable opportunity to overcome existing limitations in healthcare research. CMC is a cohesive healthcare system comprising multiple affiliated hospitals that operate under a unified hospital information system (CMC nU). This system facilitates the sharing of standardized clinical protocols and encourages regular academic exchanges among the departments of surgery, radiology, and pathology. Patient data are collected and stored in the CMC Clinical Data Warehouse (CDW), an integrated research platform that houses over 15 million fully anonymized electronic medical records. This infrastructure ensures data consistency while providing a sufficiently large sample size, thereby reducing the variability typically found in conventional multi-institutional studies.

This study aims to thoroughly analyze the clinicopathological features and treatment outcomes of borderline and malignant PTs, identify independent prognostic factors for recurrence, and offer evidence-based recommendations for surgical margin management and the role of adjuvant RT in the Korean population.

Materials and Methods

1. Study cohort and data acquisition

We conducted a retrospective analysis of patients diagnosed with borderline or malignant PTs who underwent surgical resection at five institutions within the CMC in South Korea, from January 2009 to October 2025. These hospitals were chosen based on their patient volume and their consent to share data. Clinical data were de-identified and obtained from the CMC CDW.

2. Selection criteria

Eligibility criteria were limited to patients who had comprehensive clinicopathological and treatment records. Specifically, data on tumor size, margin status, and histological parameters (including stromal atypia, mitotic count, cellularity, and stromal overgrowth) were required. Additionally, inclusion necessitated accessible surgical operative reports to confirm the extent of resection, as well as detailed RT records outlining the total delivered dose, fractionation schedules, and boost techniques. Exclusion criteria included: (1) benign histology, (2) presence of distant metastasis at diagnosis, (3) history of other malignancies, or (4) insufficient follow-up data.

3. Objectives

The primary endpoint was to evaluate the impact of adjuvant RT on local recurrence–free survival (LRFS). Secondary endpoints included distant metastasis–free survival (DMFS) and overall survival (OS). Our goal was to identify independent prognostic factors for LRFS and assess the effect of adjuvant RT while accounting for selection bias.

4. Statistical methodology

Baseline characteristics were compared using appropriate statistical tests, including Student's t-tests, Mann-Whitney U tests, chi-square tests, and Fisher’s exact tests. Survival analyses were conducted using the Kaplan-Meier method and Cox proportional hazards regression. For all time-to-event analyses, time zero was defined as the date of initial surgery for all patients, regardless of subsequent RT administration, to avoid immortal time bias.

To identify independent prognostic factors, we constructed a multivariable Cox regression model. With 33 local recurrence events, the model size was limited by the event-per-variable (EPV) rule (EPV ≥ 10) to approximately three covariates, in order to minimize overfitting. We therefore included the three covariates most directly relevant to our research questions: RM (primary exposure), tumor size (the most consistently reported prognostic factor in prior studies), and adjuvant RT (secondary exposure). To further mitigate small-sample bias, we applied Firth's penalized likelihood method.

To rigorously address selection bias, we utilized the average treatment effect in the overlap population (overlap weighting, ATO) method based on propensity scores. This approach optimizes the balance of baseline covariates between treatment groups and focuses on the population with the greatest overlap, thereby enhancing the precision of treatment effect estimation. The weighting model included age, tumor size, margin status, histologic diagnosis, type of surgery, stromal atypia, stromal cellularity, and stromal overgrowth. This complementary approach allowed histopathologic features that could not be entered into the parsimonious Cox model to be controlled for through the weighting procedure. Mitotic count and tumor border, although recognized components of the WHO grading system, were not entered as separate covariates because they are highly correlated with both the histologic diagnosis and the other histopathologic features already included in the model [12-14].

For the analysis of margin status, RM were specifically categorized into four subgroups (RM < 1 mm, 1 mm ≤ RM < 5 mm, 5 mm ≤ RM < 10 mm, and RM ≥ 10 mm) to evaluate the impact of margin width on local recurrence. If no incremental benefit was observed beyond 1 mm in pairwise comparisons, margins were planned to be analyzed as a dichotomized variable (<1 mm vs. ≥1 mm) in regression models. All statistical analyses were conducted using R software version 4.5.2 (R Foundation for Statistical Computing, Vienna, Austria), with significance defined as p < 0.05.

Results

1. Patient characteristics and selection bias

Between January 2009 and October 2025, a total of 1,208 patients with PTs were identified. After excluding 1,006 patients with benign histology and 19 with insufficient follow-up data, a final cohort of 183 patients was analyzed, consisting of 155 in the surgery-only (OP) group and 28 in the adjuvant RT group. The median follow-up duration was 56 months (95% confidence interval [CI], 49 to 63 months), estimated using the reverse Kaplan-Meier method. Before weighting, the RT cohort exhibited significantly more adverse baseline characteristics, including a higher frequency of malignant histology (57.1% vs. 12.3%, p = 0.001), larger tumor size (median, 8.8 cm vs. 4.2 cm; p < 0.001), and the presence of stromal overgrowth, marked stromal cellularity, marked nuclear atypia, and permeative tumor borders (Table 1). These findings confirm a substantial selection bias, indicating that clinicians preferentially utilized RT for high-risk patients. After applying overlap weighting (ATO), key baseline covariates were well balanced (absolute standardized mean difference < 0.001 for all variables) (Supplementary Fig. S1A), and the distribution of overlap weights remained within the bounded 0–1 range with no extreme values (Supplementary Fig. S1B), suggesting that baseline selection bias was effectively mitigated without compromising weight stability.

Table 1.

Baseline characteristics of the study population before and after adjustment for overlap weighting (ATO)

Characteristic	Cohort before weighting			Cohort after weighting
	Surgery-only, n (%) (n = 155)	Surgery + RT, n (%) (n = 28)	p-value	Surgery-only (%) (ESS = 59.6)a)	Surgery + RT (%) (ESS = 25.0)a)	SMDb)
Age, median (IQR)	43.0 (37.5–51.0)	51.0 (41.8–55.0)	0.203	46.0 (38.0–53.0)	50.0 (41.0–54.0)	<0.001
Type of surgery			0.001			<0.001
BCS	146 (94.2)	20 (71.4)		78.6	78.6
Mastectomy	9 (5.8)	8 (28.6)		21.4	21.4
Diagnosis			<0.001			<0.001
Borderline PT	136 (87.7)	12 (42.9)		55.0	55.0
Malignant PT	19 (12.3)	16 (57.1)		45.0	45.0
Tumor size (cm), median (IQR)	4.2 (3.0–6.0)	8.8 (4.2–11.3)	0.001	5.9 (4.0–10.5)	6.8 (3.5–11.3)	<0.001
Stromal atypia			0.002			<0.001
Mild	70 (45.2)	4 (14.3)		20.7	20.7
Moderate	75 (48.4)	18 (64.3)		63.0	63.0
Marked	10 (6.4)	6 (21.4)		16.3	16.3
Stromal cellularity			0.008			<0.001
Non-marked	136 (87.7)	19 (67.9)		76.7	76.7
Marked	19 (12.3)	9 (32.1)		23.3	23.3
Stromal overgrowth			<0.001			<0.001
Absent	122 (78.7)	9 (32.1)		41.6	41.6
Present	33 (21.3)	19 (67.9)		58.4	58.4
Resection margin (mm)			0.220			<0.001
<1	45 (29.0)	5 (17.9)		17.5	17.5
1–5	40 (25.8)	5 (17.8)		18.7	18.7
5–10	39 (25.2)	8 (28.6)		33.2	33.2
>10	31 (20.0)	10 (35.7)		30.6	30.6
Mitotic count, median (IQR)	7.0 (3.5–11.0)	12.0 (5.8–20.0)	0.001	N/A	N/A
Tumor border			0.007	N/A	N/A
Well-defined	45 (29.0)	5 (17.9)
Focally-permeative	94 (60.7)	14 (50.0)
Permeative	16 (10.3)	9 (32.1)

ATO, average treatment effect in the overlap population; RT, radiotherapy; ESS, effective sample size; SMD, standardized mean difference; IQR, interquartile range; BCS, breast-conserving surgery; PT, phyllodes tumor; IQR, interquartile range; N/A, not available.

^a)The effective sample size was calculated based on the ATO.

^b)An SMD < 0.2 indicates good balance between groups.

2. Treatment characteristics and toxicity

Reflecting the higher prevalence of malignant histology and greater tumor burden in the adjuvant RT group, the mastectomy rate was significantly elevated in this group compared to the OP group (28.6% vs. 5.8%, p = 0.001) (Table 1). Among the 28 patients who received adjuvant RT, eight underwent chest wall irradiation after mastectomy, while 20 received whole breast irradiation following breast-conserving surgery.

The median dose administered to the whole breast or chest wall was 50.4 Gy (interquartile range, 50 to 55). Ten patients (35.7%) were treated with a hypofractionated schedule, consisting of 40 Gy in 15 fractions (n = 5) and 50 Gy in 20 fractions (n = 5). The conventional schedules included 50 Gy in 25 fractions, 50.4 Gy in 28 fractions, and 54 Gy in 27 fractions. A tumor bed boost was administered to 10 patients (median boost dose, 10 Gy [interquartile range, 10 to 15]), with the majority (n = 9) receiving a sequential boost. The choice of fractionation and the decision to deliver a boost were made by the treating radiation oncologist based on institutional practice and clinicopathologic factors.

Regarding toxicity, adverse events were limited to radiation dermatitis, which occurred in 14 of the 28 patients (50.0%). The severity was generally mild, with 10 patients experiencing grade 1 dermatitis and four patients developing grade 2 dermatitis. No acute grade 3 or higher toxicities were reported. However, one patient with malignant PT developed post-RT angiosarcoma in the ipsilateral breast 1 year after completing adjuvant RT (54 Gy/27 fractions following breast-conserving surgery). Despite salvage mastectomy and chemotherapy, she died due to organ failure caused by lung metastasis 2 years after RT.

3. Survival outcomes and RT efficacy

During a median follow-up of 56 months, local recurrence was observed in 33 patients (18.0%). Among these, 19 patients (57.6%) experienced multiple recurrence events, averaging 2.4 episodes per patient. Distant metastasis occurred in three patients (1.6%), all of whom had malignant histology. The metastatic sites included isolated lung involvement in one patient, lung and chest wall metastasis in another, and lung with bone metastasis in the third. Notably, all distant metastasis events followed repeated local recurrences, suggesting that inadequate local control may precede systemic dissemination. Two of the three patients with distant metastasis died from the disease; no other deaths were reported during the follow-up period. Clinical outcomes and patterns of failure for the entire cohort are summarized in Table 2. Among the 19 patients with repeated local recurrence, 15 had borderline and four had malignant histology, with only two having received adjuvant RT, both of whom had malignant histology. When stratified by margin width, three had positive margins (defined as ink on tumor), three patients had margins < 1 mm, six had 1–5 mm, three had 5–10 mm, and four had ≥ 10 mm. Notably, three of these 19 patients (15.8%) subsequently developed distant metastasis, accounting for all distant metastasis events in the entire cohort.

Table 2.

Unadjusted comparison of patterns of failure and survival outcomes between surgery alone and adjuvant RT groups

	Surgery-only (n = 155)	Surgery + RT (n = 28)	p-value
Local recurrence	31 (20.0)	2 (7.1)	0.173
Borderline (n=148)	27/136 (19.9)	0/12 (0)	0.442
Malignant (n=35)	4/19 (21.1)	2/16 (12.5)	0.368
Distant metastasis	1 (0.6)	2 (7.1)	0.092
Borderline (n=148)	0/136 (0)	0/12 (0)	-
Malignant (n=35)	1/19 (5.3)	2/16 (12.5)	0.584
Death	1 (0.6)	1 (3.6)	0.702
Borderline (n=148)	0/136 (0)	0/12 (0)	-
Malignant (n=35)	1/19 (5.3)	1/16 (6.3)	>0.99

Values are presented as number (%).

RT, radiotherapy.

Unadjusted analysis indicated no significant difference in LRFS between the treatment arms (5-year LRFS: 92.3% vs. 77.7%, p = 0.135) (Fig. 1A). The RT group demonstrated significantly inferior DMFS in the unadjusted analysis (5-year DMFS: 100% vs. 91.7%, p = 0.005) (Fig. 1B). However, this finding reflects substantial selection bias rather than a deleterious effect of RT itself. As shown in Table 1, patients in the RT group had markedly more adverse baseline features, including malignant histology (57.1% vs. 12.3%) and larger tumors (median, 8.8 cm vs. 4.2 cm). After applying overlap weighting to balance these baseline differences, the apparent disadvantage in DMFS disappeared (hazard ratio [HR], 2.27; 95% CI, 0.33 to 15.66; p = 0.404) (Fig. 2B), confirming that the unadjusted difference was driven by confounding by indication rather than by RT exposure.

Fig. 1.

Unadjusted Kaplan-Meier estimates of survival outcomes for the entire cohort. (A) Local recurrence-free survival and (B) Distant metastasis-free survival are compared between the surgery alone and adjuvant radiotherapy (RT) groups before weighting adjustments. p-values were calculated using the unweighted log-rank test.

Fig. 2.

Survival outcomes after overlap weighting adjustment. (A) Weighted Kaplan-Meier curves for local recurrence-free survival, comparing the surgery alone and adjuvant radiotherapy groups. (B) Weighted Kaplan-Meier curves for distant metastasis-free survival, comparing the two groups. Adjusted hazard ratios (HRs) and p-values were derived from overlap-weighted Cox proportional hazards models with robust variance estimation. Numbers at risk represent weighted counts (i.e., based on overlap weights), not raw patient counts. CI, confidence interval.

After adjusting for baseline disparities using overlap weighting, adjuvant RT was associated with a potential reduction in local recurrence risk, though this did not achieve conventional statistical significance (HR, 0.25; 95% CI, 0.05 to 1.19; p = 0.081) (Fig. 2A). Systemic endpoints were underpowered due to the small number of events (distant metastasis, n = 3), warranting cautious interpretation of these findings. Regarding OS, no significant difference was noted between the two groups both before (p = 0.096) and after weighting (HR, 1.15; 95% CI, 0.17 to 8.05; p = 0.886).

In an exploratory subgroup analysis stratified by histologic grade, adjuvant RT showed varying effects across tumor subtypes. For borderline PT, the local recurrence rate was 19.9% (27/136) in the surgery-only group, while no local recurrences were observed in patients receiving adjuvant RT (0/12, 0%) (Table 2). In contrast, the benefit of RT for malignant PT was less significant, with local recurrence rates of 21.1% (4/19) in the surgery-only group and 12.5% (2/16) in the RT group. Within the RT cohort, when stratified by surgical extent, all local recurrences occurred in patients who underwent breast-conserving surgery followed by whole-breast irradiation (2/20, 10.0%), while no local failures were seen among patients who had a mastectomy followed by chest wall irradiation (0/8, 0%).

4. Prognostic factors for local recurrence

Univariable analysis identified RM (<1 mm) and mitotic count as factors associated with local recurrence (Table 3). To determine the optimal margin threshold, we initially categorized RMs into four categories (<1 mm, 1 to <5 mm, 5 to <10 mm, and ≥10 mm); however, no significant differences in LRFS were observed among these groups (log-rank p = 0.110). Pairwise comparisons also indicated no incremental benefit for margins beyond 1 mm (all p > 0.05). Consequently, we analyzed margins as a dichotomous variable (<1 mm vs. ≥1 mm), which revealed significantly poorer LRFS in patients with margins <1 mm (log-rank p = 0.022) (Fig. 3).

Table 3.

Univariable and Firth's multivariable Cox regression analysis for local recurrence in the unweighted group

Variable	Hazard ratio	95% CI	p-value
Univariable analysis
Age (continuous)	0.97	0.94–1	0.065
Diagnosis: malignant (vs. borderline)	0.98	0.38–2.17	0.969
Tumor size (cm, continuous)	0.95	0.84–1.05	0.375
Atypia
Mild	0.79	0.27–3.05	0.699
Moderate	0.89	0.32–3.37	0.840
Cellularity
Mild	1.14	0.31–4.02	0.836
Moderate	0.85	0.36–2.39	0.729
Stromal overgrowth: present	1.11	0.52–2.25	0.772
Mitotic count (continuous)	1.05	1–1.09	0.049*
Border
Permeative	0.65	0.17–1.80	0.441
Well defined	1.11	0.51–2.28	0.794
Margin width: <1 mm (vs. ≥1 mm)	2.22	1.10–4.36	0.026*
Adjuvant RT: RT (vs. no-RT)	0.44	0.09–1.31	0.153
Firth’s Multivariable analysis
Adjuvant RT: RT (vs. no-RT)	0.52	0.10–1.67	0.299
Margin: <1 mm	2.07	1.03–4.08	0.043*
Tumor size (cm, continuous)	0.98	0.86–1.09	0.757

CI, confidence interval; RT, radiotherapy.

^*p < 0.05.

Fig. 3.

Kaplan-Meier curves for local recurrence-free survival stratified by resection margin width (<1 mm vs. ≥1 mm). The p-value was calculated using the log-rank test.

In the multivariable analysis using Firth's penalized Cox regression, adjusting for tumor size and adjuvant RT, a margin <1 mm remained an independent predictor of local recurrence (HR, 2.07; 95% CI, 1.03 to 4.08; p = 0.043) (Table 3). In contrast, tumor size and adjuvant RT did not reach statistical significance in the conventional multivariable model (Table 3). However, a sensitivity analysis employing overlap weighting to control for treatment selection bias indicated a substantial shift in the point estimate for adjuvant RT toward a potential benefit (HR, 0.25; 95% CI, 0.05 to 1.19; p = 0.081) (Table 4), suggesting that traditional multivariable adjustment may inadequately account for significant baseline imbalances between treatment groups.

Table 4.

Impact of adjuvant radiotherapy on local recurrence: a sensitivity analysis across statistical models

Model	Adjusted variables	Hazard ratio (95% CI)	p-value
Model 1: Unadjusted	None	0.44 (0.09–1.31)	0.153
Model 2: Multivariable adjusteda)	Tumor size, margin width	0.52 (0.10–1.67)	0.299
Model 3: ATO adjustedb)	All weighted covariates	0.25 (0.05–1.19)	0.081

CI, confidence interval; ATO, average treatment effect in the overlap population.

^a)Adjusted for age, histologic diagnosis, tumor size, and resection margin status.

^b)Weighted using overlap weights (ATO) based on propensity scores to balance age, diagnosis, tumor size, margin, surgery type, and key pathologic features.

Discussion and Conclusion

This multicenter retrospective study aimed to clarify the prognostic landscape and the utility of adjuvant RT in borderline and malignant PT. Two principal findings emerged from our analysis. First, an RM <1 mm was associated with an increased risk of local recurrence and remained the only independent predictor in the multivariable analysis. Second, after addressing selection bias through overlap weighting (ATO), adjuvant RT was associated with improved LRFS in the overlap population; however, the confidence interval was wide, and the result did not reach statistical significance.

The optimal management of margin status to prevent LR remains a significant challenge, as it can dictate the need for adjuvant RT. However, previous studies investigating treatment outcomes have often been limited by the lack of detailed margin data and ongoing debates surrounding the ‘1 cm’ margin rule, which restricts their ability to refine indications for RT [15-17]. For instance, Choi et al. [16] and Li et al. [15] relied on relatively broad binary comparison (≥1 cm vs. <1 cm), leaving the efficacy of narrower clear margins unexplored. Conversely, other researchers have questioned this traditional threshold, arguing that achieving a ‘no ink on tumor’ status or a minimal margin of 1 mm is adequate for oncologic safety [3,17]. To address this gap, our study uniquely combines ATO analysis with intentionally stratified margin widths categorized into four groups (RM < 1 mm, 1 mm ≤ RM < 5 mm, 5 mm ≤ RM < 10 mm, and RM ≥ 10 mm). We found no statistically significant improvement in local control when margins exceeded 1 mm. Specifically, patients with margins <1 mm had a two-fold increase in the risk of local recurrence (HR, 2.07). Importantly, extending the margin beyond 1 mm—whether to 5 mm, 10 mm, or more—did not provide any additional significant benefit for local control. These findings challenge the traditional belief that “wider is better” and support a paradigm of surgical de-escalation. While major guidelines, including those from the National Comprehensive Cancer Network and UK recommendations, have promoted wider margins (≥1 cm), such extensive resections can be cosmetically harmful for large PTs, which often present as large, rapidly growing tumors [18,19].

Our findings suggest that a 1 mm margin strikes an optimal balance between oncologic safety and breast conservation, avoiding unnecessary wide excisions while maintaining local control. This aligns with a large-scale Dutch cohort study that emphasized clear margin status—rather than width—as the key predictor of recurrence, as well as a single-center study of 142 patients confirming that a 1 mm clearance is sufficient for local control [3,17]. Furthermore, in line with prior studies, our multivariable model identified the RM as the primary predictor of local recurrence, surpassing adverse histological features like stromal overgrowth [20-23].

The role of adjuvant RT in PTs remains unclear due to the lack of randomized trials. A recent international cross-sectional survey highlighted significant variability in management strategies, indicating an urgent need for consensus [24].

In our cohort, the apparent benefit of RT was obscured in conventional analyses because it was preferentially administered to patients with adverse features, such as malignant histology and larger tumors. This pattern mirrored the MD Anderson algorithm [25]. Overlap weighting was used to mitigate selection bias and estimate the treatment effect within the overlap (clinical equipoise) population. Following the weighting, the point estimate suggested a potential benefit of RT for local control (HR, 0.25), although the confidence interval was wide and the result did not achieve statistical significance (p = 0.081). These findings suggest a potential benefit of adjuvant RT for local control; however, definitive conclusions are limited due to the small size of the RT cohort and the low number of local recurrence events. Our results align with previous meta-analyses that report improved local control with RT in high-risk patients. Further validation in larger datasets is needed [26].

After applying overlap weighting, no clear differences in distant DMFS were observed between the groups. However, the analysis was underpowered due to the very small number of distant metastasis events (n = 3), resulting in wide confidence intervals. Consequently, our study lacks the power to determine whether adjuvant RT affects systemic outcomes. Notably, all distant metastases occurred following repeated local recurrences. This temporal pattern generates hypotheses and highlights the clinical significance of achieving durable local control, although it does not establish a causal relationship. Patients with repeated local recurrences may require closer systemic monitoring, and the potential roles of more extensive re-excision or salvage RT at recurrence should be explored in larger studies.

We recognize several limitations, primarily the retrospective design and the small sample size of the RT group, which limited statistical power. Although we applied overlap weighting to address baseline imbalances, unmeasured confounders may still exist; specifically, we could not account for temporal variations in surgical and RT techniques over the lengthy study period. Additionally, overlap weighting estimates the treatment effect in the overlap (clinical equipoise) population, which may limit generalizability to patients at the extremes of risk. Systemic outcomes (DMFS/OS) were underpowered due to the very small number of events, leading to wide confidence intervals. Furthermore, the lack of a central pathology review could introduce inter-observer variability. Despite these limitations, our multicenter data support a paradigm of surgical de-escalation aimed at achieving a 1 mm margin, complemented by the selective use of adjuvant RT to enhance local control in selected high-risk patients.

In conclusion, this study shows that surgical margins <1 mm are independently linked to a higher risk of local recurrence in borderline and malignant PT. In contrast, wider margins greater than 1 mm do not provide additional benefits. After carefully adjusting for selection bias, adjuvant RT demonstrated a trend toward reduction in local recurrence risk (HR, 0.25), although this result did not achieve statistical significance, likely due to the small sample size. Importantly, all instances of distant metastasis occurred after repeated local recurrences, highlighting the vital need for effective local control. These findings indicate that adjuvant RT may be a viable option for select high-risk patients, including those with borderline tumors or positive margins, to enhance local control and potentially reduce the risk of systemic progression.

Supplementary Materials

Supplementary materials can be found via https://doi.org/10.3857/roj.2026.00353.

Supplementary Fig. S1.

(A) Covariate balance before and after overlap weighting (ATO). (B) Distribution of overlap weights (ATO) by treatment group.