American journal of translational research · 2026
Identification of chemotherapy-related risk factors for lymphedema in breast cancer patients using Lasso regression and model development.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Guo Yanxiang, An Aihu, Li Po, Li Jun, Zhang Baiming
Аннотация
OBJECTIVE: We aimed to identify chemotherapy-related predictors of upper-limb breast cancer-related lymphedema (BCRL) and to build and validate a clinically usable prediction model.
METHODS: Our multicenter study analyzed 670 breast cancer patients treated with chemotherapy from December 2018 through February 2024. We divided patients into training (381 patients, Gansu Provincial Cancer Hospital) and validation (289 patients, Zhangye Second People's Hospital) cohorts. The prediction model combined four machine learning algorithms - decision tree, random forest, support vector machine, and XGBoost-using Lasso regression as the final integrator. We employed K-fold cross-validation to prevent data leakage. Performance metrics included AUC, Brier score, calibration curves, and decision curve analysis. SHAP values helped us understand which factors mattered most. Variables showing P<0.10 in initial screening entered the multivariable model after checking for multicollinearity.
RESULTS: Eight factors showed preliminary associations with lymphedema by simple comparisons. However, comprehensive analysis accounting for overlapping effects identified six independent predictors: disease stage, complete versus limited axillary surgery, sentinel node procedure, total nodes removed, radiation treatment, and whether chemotherapy came before or after surgery. Statistical significance was strongest for stage (P<0.001) and weakest for chemotherapy timing (P=0.033). When we tested the model, discrimination remained good though slightly lower than development (AUC dropped from 0.773 to 0.713). Prediction errors stayed modest (Brier scores 0.136-0.146). Calibration was excellent since predicted probabilities matched observed rates. Clinical decision analysis suggested the model adds value when risk thresholds fall between 0-63%, peaking near 18-19% threshold. Feature importance analysis confirmed disease burden (stage, node count), and treatment aggressiveness (surgery type, radiation, chemotherapy sequence) jointly determined risk.
CONCLUSION: We identified six factors that independently raise lymphedema risk after breast cancer treatment. The machine learning model we developed discriminates risk well enough for clinical application. It may help doctors intensify monitoring for high-risk patients while avoiding unnecessary intervention for low-risk patients.
METHODS: Our multicenter study analyzed 670 breast cancer patients treated with chemotherapy from December 2018 through February 2024. We divided patients into training (381 patients, Gansu Provincial Cancer Hospital) and validation (289 patients, Zhangye Second People's Hospital) cohorts. The prediction model combined four machine learning algorithms - decision tree, random forest, support vector machine, and XGBoost-using Lasso regression as the final integrator. We employed K-fold cross-validation to prevent data leakage. Performance metrics included AUC, Brier score, calibration curves, and decision curve analysis. SHAP values helped us understand which factors mattered most. Variables showing P<0.10 in initial screening entered the multivariable model after checking for multicollinearity.
RESULTS: Eight factors showed preliminary associations with lymphedema by simple comparisons. However, comprehensive analysis accounting for overlapping effects identified six independent predictors: disease stage, complete versus limited axillary surgery, sentinel node procedure, total nodes removed, radiation treatment, and whether chemotherapy came before or after surgery. Statistical significance was strongest for stage (P<0.001) and weakest for chemotherapy timing (P=0.033). When we tested the model, discrimination remained good though slightly lower than development (AUC dropped from 0.773 to 0.713). Prediction errors stayed modest (Brier scores 0.136-0.146). Calibration was excellent since predicted probabilities matched observed rates. Clinical decision analysis suggested the model adds value when risk thresholds fall between 0-63%, peaking near 18-19% threshold. Feature importance analysis confirmed disease burden (stage, node count), and treatment aggressiveness (surgery type, radiation, chemotherapy sequence) jointly determined risk.
CONCLUSION: We identified six factors that independently raise lymphedema risk after breast cancer treatment. The machine learning model we developed discriminates risk well enough for clinical application. It may help doctors intensify monitoring for high-risk patients while avoiding unnecessary intervention for low-risk patients.