Risk of Lymphedema After Sentinel Node Biopsy in Patients With Breast Cancer

Abstract

Purpose

Although numerous studies have identified potential risk factors for ipsilateral lymphedema development in patients with breast cancer following axillary node dissection, the risk factors for lymphedema in patients undergoing sentinel node biopsy without axillary dissection remain unclear. In this study, we aimed to determine the real-world incidence and risk factors for lymphedema in such patients.

Methods

We conducted a single-center, retrospective review of medical records of patients with breast cancer who underwent sentinel node biopsy alone. The development cohort (5,051 patients, January 2017–December 2020) was analyzed to identify predictors of lymphedema, and a predictive model was subsequently created. A validation cohort (1,627 patients, January 2014–December 2016) was used to validate the model.

Results

In the development cohort, 49 patients (0.9%) developed lymphedema over a median follow-up of 56 months, with most cases occurring within the first three years post-operation. Multivariate analysis revealed that a body mass index (BMI) of 30 kg/m² or above, radiation therapy (RTx), chemotherapy, and more than three harvested lymph nodes significantly predicted lymphedema. The predictive model showed an area under the curve of 0.824 for systemic chemotherapy, with the number of harvested lymph nodes being the most significant factor. Patients were stratified into four risk groups, showing lymphedema incidences of 3.3% in the highest-risk group and 0.1% in the lowest-risk group. In the validation cohort, the incidences were 1.7% and 0.2% for the highest and lowest risk groups, respectively.

Conclusion

The lymphedema prediction model identifies RTx, chemotherapy, BMI ≥ 30 kg/m², and more than three harvested lymph nodes as significant risk factors. Although the overall incidence is low, the risk is notably influenced by the extent of lymph node removal and systemic therapies. The model’s high negative predictive value supports its application in designing tailored lymphedema surveillance programs for early intervention.

INTRODUCTION

Surgical staging of the axillary lymph nodes is essential for determining the type of adjuvant treatment and assessing the risk of disease recurrence in patients with early breast cancer. Historically, axillary staging primarily involved axillary lymph node dissection, which is linked to significant arm morbidities such as lymphedema, arm pain, and numbness. These complications can severely impact long-term quality of life [1,2]. As the most dreaded postoperative complication in patients with breast cancer, the development of ipsilateral lymphedema profoundly affects the quality of life and physical function, necessitating lifelong management [3,4,5,6]. The introduction of sentinel node biopsy as an alternative to axillary dissection has revolutionized axillary staging in patients with early breast cancer [7]. Since the advent of sentinel node biopsy, the prevalence of lymphedema has decreased compared to traditional axillary dissection [8,9,10,11]; however, a small proportion of patients still develop ipsilateral lymphedema of varying severities [4,12,13,14]. Most previous studies aimed at identifying lymphedema risk factors and included populations that underwent both axillary dissection and sentinel node biopsy. Consequently, the specific risk factors for lymphedema following only sentinel node biopsy have not been clearly established.

Recent studies have shown that early interventions, including various physical therapy options through active surveillance, can effectively prevent or mitigate lymphedema in high-risk patients [15,16,17]. Moreover, this strategy of active surveillance for lymphedema has proven to be more cost-effective [18,19,20]. To implement an efficient active surveillance program, it is crucial to assess each patient’s risk of developing lymphedema. Targeted surveillance based on risk stratification can help optimize clinical utilization and enhance patient compliance. Unfortunately, while the risk factors for lymphedema in patients who have undergone axillary dissection are well-documented [1,3,21,22], the risk factors following sentinel node biopsy alone remain under-characterized, possibly due to the low incidence of lymphedema cases.

In this study, we investigated the risk factors for developing lymphedema in early breast cancer patients who underwent sentinel node biopsy alone using a relatively large cohort of patients who were treated at a single institution. Based on our findings, we propose a lymphedema risk prediction model that can identify a high-risk group of patients who may benefit from active surveillance programs.

METHODS

Study design and ethics

For this retrospective study, we identified all women diagnosed with breast cancer who underwent breast cancer surgery at Seoul National University Hospital (SNUH) from January 1, 2014 to December 31, 2020, by accessing the electronic medical record database. We excluded patients who underwent axillary lymph node dissection, did not undergo axillary staging, or had a history of axillary surgery.

Data from patients treated between January 2017 and December 2020 (the development set) were used to identify potential risk factors for lymphedema and to construct risk prediction models based on these factors. The risk prediction model was validated with data from patients who underwent surgery between January 2014 and December 2016 (validation set). This study received approval from the Institutional Review Board (IRB) of SNUH (IRB No. H-2308-041-1456) and was conducted in accordance with the principles of the Declaration of Helsinki.

Data collection

Data on patient demographics, clinicopathological features, operative characteristics, and relevant outcomes were collected through the clinical data warehouse of the Seoul National University Hospital Patients’ Research Environment (SUPREME), by accessing the SNUH Breast Care Center Database [23] or by reviewing electronic medical records. Baseline patient factors included age at breast cancer surgery, type of breast surgery, and body mass index (BMI), which was categorized according to the World Health Organization classification: normal (< 25.0 kg/m²), overweight (25.0–29.9 kg/m²), and obese (≥ 30.0 kg/m²). We also considered the use of preoperative chemotherapy, adjuvant chemotherapy, and radiation therapy (RTx). The number of harvested lymph nodes, both sentinel and non-sentinel, was analyzed as a continuous variable. For categorical analysis, we used “more than three harvested lymph nodes,” as this cutoff was identified as the most clinically relevant in multivariate analysis.

Definition of lymphedema

Lymphedema was diagnosed if there was: 1) a difference of more than 200 mL in volume compared to the contralateral arm, as measured by perometry [24]; 2) a 10% increase in perometric limb volume; or 3) a difference of more than 2 cm in circumference compared to the contralateral limb in either the upper arm or the forearm [12]. In our study, perometry was primarily used. However, for patients with restricted range of motion due to postoperative adhesive capsulitis, a circumference difference of more than 2 cm between the affected and contralateral limbs was used to diagnose lymphedema. Patients diagnosed using the 2 cm circumference difference criteria constituted less than 5% of the study population. Upper limb circumference measurements to detect lymphedema were conducted when patients visited the rehabilitation clinic and reported symptoms. Subsequently, patients were monitored every three months during follow-up visits to the clinic. Perometric limb measurements were obtained at predefined anatomical points by a trained nurse specializing in lymphedema management. Patients whose lymphedema resolved or who were determined not to have lymphedema were transitioned to open follow-up at the rehabilitation clinic after six months.

Statistical analysis

All statistical analyses were conducted using R software (version 4.2.1; R Foundation for Statistical Computing, Vienna, Austria). Univariate logistic regression models were used to evaluate risk factors associated with the development of lymphedema. The multivariate logistic regression included variables with p-values < 0.05 in univariate analysis to identify independent predictors. The multivariate model included BMI ≥ 30 kg/m², RTx, chemotherapy, and more than three harvested lymph nodes. Based on these variables, the final predictive model stratified patients into four risk groups: low, intermediate-low, intermediate-high, and high risk, according to the risk score calculated from the logistic regression formula.

The performance of the model was evaluated using receiver operating characteristic (ROC) curve analysis to compare patients in the development and validation sets. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for all variables. The performance of the model was further assessed using ROC curve analysis, which plots the sensitivity (true positive rate) against 1-specificity (false positive rate) at various thresholds, with the area under the curve (AUC) providing a measure of diagnostic accuracy.

RESULTS

Patient information

In the development set, 5,051 patients were included in the analysis. The median age of the patients was 52 years (interquartile range [IQR], 45–60 years). The demographic and clinicopathological characteristics of patients are detailed in Table 1. Most patients had T1 or T2 tumors, with a median of five lymph nodes harvested. Approximately 65% of patients underwent breast-conserving surgery. In the validation set, 1,627 patients were included, with a median age of 51 years (IQR, 44–59 years). Regarding tumor stage, 1,068 patients (65.6%) had T1 tumors, and 327 patients (20.1%) had T2 tumors. The median number of lymph nodes harvested was also five. The median follow-up periods were 56.0 months (IQR, 43.0–67.0 months) and 64.0 months (IQR, 55.0–78.5 months) for the development and validation sets, respectively, as shown in Table 1.

Table 1. Clinicopathologic characteristics of all patients.

Characteristics		Development cohort (n = 5,051)	Validation cohort (n = 1,627)
Age at surgery (yr)		52 (45–60)	51 (44–59)
BMI (kg/m2)		22.93 (20.98–25.27)	22.95 (21.13–25.17)
Type of surgery
	BCS	3,243 (64.2)	1,131 (69.5)
	Mastectomy	774 (15.3)	210 (12.9)
	Immediate reconstruction*	1,034 (20.5)	286 (17.6)
Tumor subtype
	HR+/HER2−	2,959 (58.6)	892 (54.8)
	HR+/HER2+	621 (12.3)	182 (11.2)
	HR−/HER2+	576 (11.4)	221 (13.6)
	TNBC	895 (17.7)	332 (20.4)
T stage
	In situ	486 (9.6)	224 (13.8)
	1	3,110 (61.6)	1,068 (65.6)
	2	1,399 (27.7)	327 (20.1)
	3	51 (1.0)	8 (0.5)
	4	5 (0.1)	0 (0.0)
CTx
	No	3,074 (60.9)	1,144 (70.4)
	Yes	1,977 (39.1)	483 (29.6)
RTx
	No	2,237 (44.3)	523 (32.1)
	Yes	2,814 (55.7)	1,104 (67.9)
No. of positive lymph nodes
	0	4,550 (90.1)	1,509 (92.7)
	1	388 (7.7)	104 (6.4)
	2	113 (2.2)	14 (0.9)
No. of harvested lymph nodes		5 (4–8)	5 (4–7)

Values are presented as median value (interquartile range) or number of patients (%).

BMI = body mass index; BCS = breast-conserving surgery; HR+ = hormone receptor positive; HER2− = human epidermal growth factor receptor 2 negative; HER2+ = human epidermal growth factor receptor 2 positive; HR− = hormone receptor negative; TNBC = triple-negative breast cancer; CTx = neoadjuvant chemotherapy or adjuvant chemotherapy; RTx = radiation therapy.

^*Immediate reconstruction followed mastectomy.

Factors associated with lymphedema development

In the development set, 49 patients (0.9%) developed postoperative ipsilateral lymphedema during the follow-up period. The majority of these patients became symptomatic within the first three years after surgery (Supplementary Figure 1).

Univariate analysis revealed that the use of cytotoxic systemic chemotherapy (OR, 9.51; 95% CI, 4.54–23.19; p < 0.001) and RTx (OR, 3.12; 95% CI, 1.62–6.64; p = 0.001) were significantly associated with the development of lymphedema (Table 2). In our study, 2,814 patients received RTx, and 38 were diagnosed with objective lymphedema. Additionally, BMI was a significant predictor of lymphedema. A cutoff value of 30 kg/m² was selected based on its strong association with an increased risk of lymphedema (OR, 4.34; 95% CI, 1.83–9.14; p < 0.001). This cutoff was used in the final predictive model. The number of harvested lymph node, analyzed as a continuous variable, was significantly associated with lymphedema (OR, 1.14; 95% CI, 1.07–1.21; p < 0.001). After adjusting for RTx, chemotherapy, and BMI, patients with more than three harvested lymph nodes showed the highest risk of developing lymphedema (OR, 6.12; 95% CI, 1.88–37.63; p = 0.012) (Supplementary Table 1). Thus, a threshold of more than three lymph nodes was selected to categorize patients in the predictive model.

Table 2. Univariate and multivariate analyses of factors associated with lymphedema.

Variable		Univariate		Multivariate
		OR (95% CI)	p-value	OR (95% CI)	p-value
Age at surgery (yr)		0.99 (0.97–1.02)	0.84
Type of surgery
	BCS	1
	Mastectomy	1.14 (0.51–2.30)	0.721
	Immediate reconstruction*	0.66 (0.26–1.41)	0.325
Pathologic T stage
	In situ	1
	T1	0.89 (0.34–8.07)	0.842
	T2	1.65 (0.62–5.74)	0.359
	T3	7.53 (1.44–35.1)	0.009
	T4	0.00 (0.00–65.8)	0.986
BMI (kg/m2)
	Normal (< 25.0)	1		1
	Overweight (25.0–29.9)	1.16 (0.55–2.26)	0.668	1.21 (0.58–2.38)	0.578
	Obese (≥ 30.0)	4.34 (1.83–9.14)	< 0.001	4.34 (1.80–9.35)	< 0.001
RTx			0.001		0.028
	No	1		1
	Yes	3.12 (1.62–6.64)		2.21 (1.13–4.75)
CTx			< 0.001		< 0.001
	No	1		1
	Yes	9.51 (4.54–23.19)		7.37 (3.46–18.21)
No. of metastatic lymph nodes		1.65 (0.91–2.68)	0.068
No. of harvested lymph nodes		1.14 (1.07–1.21)	< 0.001	1.10 (1.03–1.18)	0.001

OR = odds ratio; CI = confidence interval; BCS = breast-conserving surgery; BMI = body mass index; RTx = radiation therapy; CTx = neoadjuvant chemotherapy or adjuvant chemotherapy.

^*Immediate reconstruction followed mastectomy.

Prediction model for lymphedema development

In this study, we developed a model to predict the risk of lymphedema development after sentinel node biopsy using logistic regression analysis with independent risk factors. As shown in Supplementary Table 2, both neoadjuvant chemotherapy (NAC) and adjuvant chemotherapy significantly increased the risk of developing lymphedema. In multivariate analysis, after adjusting for a BMI ≥ 30 kg/m², RTx, and more than three harvested lymph nodes, NAC had an OR of 1.98 (95% CI, 1.06–3.80; p = 0.035) and adjuvant chemotherapy had an OR of 5.03 (95% CI, 2.25–12.37; p < 0.001). To create a more practical predictive model, we combined NAC and adjuvant chemotherapy into a single category “CTx”.

The final model is presented in Table 3, where p represents the probability of lymphedema development. CTx, RTx, more than three harvested lymph nodes, and a BMI of ≥ 30 kg/m² were identified as significant predictors of lymphedema. This model predicted the risk of lymphedema with an AUC of 0.824 (95% CI, 0.780–0.869; p < 0.001) (Figure 1A). The model’s cutoff value for predicted probability was 0.56, resulting in a sensitivity of 87.8%, specificity of 66.8%, positive predictive value (PPV) of 2.5%, and negative predictive value (NPV) of 99.8%. Among the variables, the use of systemic chemotherapy was the most significant contributor to the model, followed by having more than three harvested lymph nodes.

Table 3. Multivariable logistic regression for probability calculator in the development set.

Variables	OR* (95% CI)	Coefficient	p-value
CTx	7.61 (3.59–18.72)	2.030	0.001
RTx	2.12 (1.08–4.56)	0.756	0.036
LN (> 3)	6.12 (1.88–37.63)	1.813	0.012
BMI (≥ 30 kg/m2)	4.01 (1.70–8.38)	1.390	< 0.001

Predicted probability using coefficient contribution of developing lymphedema:

= −8.185 + 1.390 × BMI + 0.756 × RTx + 2.030 × Chemotherapy + 1.813 × LN

OR = odds ratio; CI = confidence interval; CTx = neoadjuvant chemotherapy or adjuvant chemotherapy; RTx = radiation therapy; LN = the number of harvested lymph nodes; BMI = body mass index.

^*Adjusted multi-analysis with CTx, RTx, LN (> 3), BMI (< 30 kg/m² vs. 30 kg/m²).

Figure 1. Receiver operating characteristic curve.

Receiver operating characteristic curves for different coefficient contributions to lymphedema: (A) development set (AUC, 0.824; 95% CI, 0.780–0.869) and (B) internal validation set (AUC, 0.726; 95% CI, 0.597–0.895).

AUC = area under the curve; CI = confidence interval.

In the validation set, the area under the ROC for predicting lymphedema was 0.726 (95% CI, 0.597–0.895) (Figure 1B), with a sensitivity of 70.0%, specificity of 75.4%, PPV of 1.7%, and NPV of 99.8%. These results indicate that our prediction model is effective in identifying patients who are unlikely to develop lymphedema, as evidenced by the high NPV (99.8%). This suggests that the model has a strong ability to rule out the risk of lymphedema in low-risk patients.

Integrated model for lymphedema development

Figure 2A illustrates the stratification of patients into four risk groups (low, intermediate-low, intermediate-high, and high risk) based on the risk scores calculated from the predictive model presented in Table 3. Initially, we divided the patients into two groups (high vs. low risk), but this method did not reveal a significant difference in the incidence of lymphedema between the groups, particularly in the high-risk group. We divided the patients into four distinct risk groups to enhance differentiation between the groups and make the model more clinically applicable. This division provided clearer distinctions in the incidence of lymphedema, especially in the high-risk group, which exhibited a noticeably higher incidence than the other groups.

Figure 2. Predicting lymphedema risk.

(A) Intergrated model of predicting lymphedema risk. Incidence of lymphedema by risk groups (B) in development cohort and (C) in validation cohort (a comparison between the groups used for creating the lymphedema prediction development group, in validation group).

BMI↑ = body mass index ≥ 30 kg/m²; BMI↓ = body mass index < 30 kg/m²; RTx = radiation therapy; LN = the number of harvested lymph nodes; CTx = neoadjuvant chemotherapy or adjuvant chemotherapy.

As a result of this stratification, 1,098 patients (21.7%) in the development set were classified as high-risk out of the total 5,051 patients. The low-risk and intermediate-low-risk groups each had a lymphedema risk of less than 1.0%, whereas the high-risk group had a 3.3% risk of lymphedema (Figure 2B). In the validation set, our model successfully identified patients with varying risks of developing postoperative lymphedema. The low-risk and intermediate-low-risk groups had an incidence of lymphedema below 1%, while the high-risk group had a 1.7% risk of developing lymphedema (Figure 2C).

DISCUSSION

In this study, we developed a predictive model based on clinicopathological factors to identify patients with breast cancer at high-risk of developing lymphedema. The incidence of lymphedema development after sentinel node biopsy was 0.9% (49 of 5,051 patients) and that in the high-risk group was 3.3% (36 of 1,091 patients) in the development cohort. Clinicopathological factors, including the use of systemic chemotherapy, RTx, obesity, and more than three harvested lymph nodes, have been individually reported as risk factors in many studies [8,9,10,11]. In our study, the use of cytotoxic chemotherapy was as significant as the extent of axillary surgery in predicting lymphedema risk in patients with breast cancer undergoing sentinel node biopsy alone. Our prediction model incorporated these four risk factors, each contributing differently to the risk of lymphedema.

Sentinel node biopsy demonstrated a significantly lower incidence of lymphedema compared to axillary dissection, while maintaining comparable accuracy in axillary staging [3,10,25,26,27]. Although previous studies have reported the incidence of lymphedema ranging from 0% to 8% in patients undergoing sentinel node biopsy alone, our results confirmed a generally low incidence of lymphedema [3,10,25,26]. Because we defined lymphedema using objective measurements, we could not assess the subjective symptoms of patients. However, studies have shown that subjective symptoms of the ipsilateral arm are important in determining quality of life and carry diagnostic value for lymphedema in patients with breast cancer [28,29]. Future studies addressing the importance of subjective symptoms in patients undergoing sentinel node biopsy are required. Isik et al. [30] found that harvesting more than five lymph nodes during sentinel node biopsy increased the risk of lymphedema by two-fold in their review of 2,940 cases. However, this study did not explore the impact of systemic therapy as a potential risk factor for lymphedema. Furthermore, De Groef et al. [31] observed no significant association between chemotherapy and lymphedema during a longitudinal follow-up of 100 patients. In addition to well-known risk factors like obesity and RTx, our study found that cytotoxic chemotherapy contributed to lymphedema development similarly to axillary surgery in patients undergoing sentinel node biopsy alone. While the receipt of chemotherapy has been identified as a potential risk factor for lymphedema, its effect was relatively minor in analyses including patients who underwent axillary dissection [1,21,32]. These findings suggest that information on systemic therapy can further refine lymphedema risk assessment in patients undergoing sentinel node biopsy alone.

The low PPV of our model highlights its limitations in accurately predicting patients who will develop lymphedema. However, the high NPV suggests that the model is more effective in ruling out those who will not develop lymphedema, which can be valuable in clinical settings. This characteristic is beneficial for reassuring low-risk patients and reducing unnecessary anxiety and interventions. Nevertheless, the low PPV of the model underscores the need for further refinement of the model to enhance its predictive accuracy. Future research should focus on identifying additional risk factors and integrating more precise measurement techniques to improve overall model performance.

Our study has several limitations. First, as this was a retrospective study conducted at a single institution, it carries inherent risks, including selection bias and limited generalizability. Second, lymphedema cases were defined as patients who were referred to the clinic with symptoms of lymphedema confirmed by objective measurements. This approach may have missed asymptomatic cases or those managed outside our clinic. Finally, we primarily used perometry data for diagnosing lymphedema in our dataset [29,33]. Although perometry is a standard diagnostic tool, the optimal approach for lymphedema diagnosis remains controversial; alternatives include circumference tape and volume measurement. Future studies should aim to standardize diagnostic criteria and integrate baseline measurements to enhance diagnostic accuracy and reduce the risk of misdiagnosis.

In conclusion, we developed a lymphedema prediction model using logistic regression analysis that identified key risk factors such as RTx, chemotherapy, a BMI of 30 kg/m² or higher, and more than three harvested lymph nodes. Despite its low PPV, the model effectively ruled out patients who were unlikely to develop lymphedema, thus providing significant clinical utility for early intervention. Our findings enable the stratification of patients into high-risk and low-risk groups, facilitating timely interventions and allowing high-risk patients to seek treatment before the onset of lymphedema, thereby enhancing patient care.

SUPPLEMENTARY MATERIALS

Supplementary Table 1

Association between the number of harvested lymph nodes and lymphedema adjusted for other potential factors

Supplementary Table 2

Univariate and multivariate analyses of risk factors associated with lymphedema

Supplementary Figure 1

Cumulative incidence of lymphedema.