The weight of each node: Modeling and quantifying breast cancer lymphedema risk.

Abstract

e12750 Background: Lymphedema (LE) is a common morbidity following breast cancer surgery, affecting quality of life. While axillary surgery is a LE risk factor, prior studies have largely stratified risk by surgical procedure, resulting in wide-ranging risk estimates. Although procedure type likely contributes, data quantifying per-lymph node (LN) risk and other factors remain limited. This study was performed to identify risk factors for LE development, quantify the incremental risk per LN removed, and provide quantitative estimates not previously established. Methods: A cohort study identified from institutional medical records was performed of patients with nonmetastatic noninflammatory invasive breast cancer having axillary surgery at an NCI-designated comprehensive cancer center from 2013–2022. Axillary surgery was categorized as sentinel lymph node biopsy alone (SLNB), delayed completion axillary dissection after prior SLNB (dALND), or upfront axillary dissection with or without SLNB performed at the same procedure (ALND). Multivariable logistic regression with LASSO-based variable selection was used to identify independent predictors of LE using the training set, with validation in a held-out test set. Final model parameters were estimated using the full cohort. Results: A total of 3,969 patients were included, whose mean age was 59.6 ± 12.1 years. BMI averaged 30.0 ± 7.0 and mean follow-up was 5.0 ± 2.8 years. A mean of 4.4 ± 5.1 LNs was removed and LE occurred in 14.0% overall. SLNB was performed in 78.9%, dALND in 1.5%, and ALND in 19.6%. Independent predictors of arm LE included increasing number of regional LN removed (OR 1.06 per node, p = 0.046), higher BMI (OR 1.04 per kg/m², p < 0.001), younger age (OR 0.98 per year, p < 0.001), radiotherapy (OR 1.79, p < 0.001), and mastectomy without reconstruction (OR 2.40 vs lumpectomy, p < 0.001). Model-predicted risk rose as greater numbers of LNs were excised, rising to 29.9% when ≥24 LN were removed. The risk-prediction model demonstrated good discrimination (AUC = 0.73) and calibration (slope = 0.95). Conclusions: The increase in risks of LE conferred by each LN removed and other factors such as higher BMI, younger age, radiotherapy, and mastectomy are now estimable. With these contributors now quantified, physicians have better estimates than prior procedure-based ranges. Such data may better guide preoperative counseling, postoperative surveillance, and direct more tailored interventions for those at greatest risk.