Scholarlink

Background <p>Breast cancer-related lymphedema (BCRL) is a common and debilitating sequela of axillary lymph node dissection (ALND). Although machine learning (ML)-based prediction models have been proposed, few focus exclusively on patients undergoing ALND, and direct comparisons with traditional statistical models remain limited. This study aimed to develop accurate and clinically feasible prediction models for BCRL using supervised ML and multivariable logistic regression.</p> Methods <p>Demographic and clinical data were prospectively collected from women undergoing unilateral ALND for breast cancer at Memorial Sloan Kettering Cancer Center between 2016 and 2024. Supervised ML and multivariable logistic regression models to predict BCRL were trained and internally validated. Model performance was evaluated using area under the receiver operator characteristic curve (AUC), accuracy, sensitivity, specificity, and Brier score. Shapley additive explanations were used for model interpretability.</p> Results <p>A total of 474 eligible patients were included. BCRL developed in 113 (23.8%) patients at a mean ± standard deviation of 16.6 ± 7.5 months postoperatively. The highest-performing ML model (random forest) achieved an AUC of 0.83, whereas traditional multivariable logistic regression achieved an optimism-corrected AUC of 0.62. Key ML predictors of BCRL on Shapley additive explanations analysis included clinical cancer stage, body mass index, age, and neoadjuvant chemotherapy.</p> Conclusions <p>ML-based models outperformed traditional logistic regression in predicting BCRL among patients undergoing ALND. These models demonstrate the potential of ML for early BCRL identification and risk stratification but also highlight the difficulties in accurately predicting BCRL development. Further research is needed to improve model predictive performance and facilitate clinical implementation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Using Machine Learning to Predict Breast Cancer-Related Lymphedema Following Axillary Lymph Node Dissection

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Benjamin D. Wagner,
  • Jonlin Chen,
  • Lillian A. Boe,
  • Francis D. Graziano,
  • Arielle N. Roberts,
  • Andrea V. Barrio,
  • Michelle R. Coriddi,
  • Jonas A. Nelson,
  • Babak J. Mehrara,
  • Danielle H. Rochlin

摘要

Background

Breast cancer-related lymphedema (BCRL) is a common and debilitating sequela of axillary lymph node dissection (ALND). Although machine learning (ML)-based prediction models have been proposed, few focus exclusively on patients undergoing ALND, and direct comparisons with traditional statistical models remain limited. This study aimed to develop accurate and clinically feasible prediction models for BCRL using supervised ML and multivariable logistic regression.

Methods

Demographic and clinical data were prospectively collected from women undergoing unilateral ALND for breast cancer at Memorial Sloan Kettering Cancer Center between 2016 and 2024. Supervised ML and multivariable logistic regression models to predict BCRL were trained and internally validated. Model performance was evaluated using area under the receiver operator characteristic curve (AUC), accuracy, sensitivity, specificity, and Brier score. Shapley additive explanations were used for model interpretability.

Results

A total of 474 eligible patients were included. BCRL developed in 113 (23.8%) patients at a mean ± standard deviation of 16.6 ± 7.5 months postoperatively. The highest-performing ML model (random forest) achieved an AUC of 0.83, whereas traditional multivariable logistic regression achieved an optimism-corrected AUC of 0.62. Key ML predictors of BCRL on Shapley additive explanations analysis included clinical cancer stage, body mass index, age, and neoadjuvant chemotherapy.

Conclusions

ML-based models outperformed traditional logistic regression in predicting BCRL among patients undergoing ALND. These models demonstrate the potential of ML for early BCRL identification and risk stratification but also highlight the difficulties in accurately predicting BCRL development. Further research is needed to improve model predictive performance and facilitate clinical implementation.