Abstract:

Objective To develop and preliminarily validate a risk prediction model for musculoskeletal symptoms in breast cancer patients undergoing endocrine therapy，providing a basis for early clinical prevention of endocrine therapy-related musculoskeletal symptoms. Methods This was a prospective study，and 262 patients with early-stage breast cancer who were undergoing postoperative endocrine therapy were consecutively enrolled in a tertiary hospital in Shanghai from April 2023 to January 2024. Sociodemographic data，clinical data，and information on musculoskeletal symptoms were collected at baseline. Then，we followed up the occurrence of musculoskeletal symptoms among these participants every 3 months，with a total period of one-year therapy. The samples were randomly divided into a training set and a test set at a ratio of 7∶3. Totally 6 survival machine learning algorithms were used to construct risk prediction models. Results There were 262 patients included in this study. The results showed that the XGBoost survival machine learning algorithm performed the best. The areas under the receiver operating characteristic（ROC） curve（AUC） for predicting the risk of musculoskeletal symptoms after 12 months of endocrine therapy were 0.812（training set），0.815（bootstrap training set），and 0.753（test set），respectively. The brier scores of the bootstrap calibration curves in the training set and test set were 0.184 and 0.219，respectively. In addition，XGBoost model could accurately classify the risk levels of the musculoskeletal symptoms. The P-values of the Log-rank test in the training set and the test set are <0.001 and 0.024 respectively. The importance of the predictors was ranked in the following order：bone mineral density，sleep quality，anxiety and depression，type of endocrine agent，progesterone receptor expression，financial stress，BMI，and history of arthritis. Conclusion This study developed a risk prediction model for musculoskeletal symptoms among breast cancer patients with endocrine therapy，showing a well predictive performance，which could contribute to providing nursing decision and making bases for the early identification and prevention of musculoskeletal symptoms.

Key words: Breast Cancer, Endocrine Therapy, Musculoskeletal Symptoms, Machine Learning, Prediction Model, Nursing Care

摘要：

目的 构建并初步验证乳腺癌内分泌治疗患者骨关节症状风险预测模型，为临床早期预防内分泌治疗相关骨关节症状提供依据。方法 采用前瞻性研究设计，从2023年4月—2024年1月，采用便利抽样法，选取在上海市某三级甲等医院连续入组262例乳腺癌术后内分泌治疗的患者为调查对象，在基线采集相关社会人口学资料、临床资料及骨关节症状信息，每3个月随访评估患者的骨关节症状发生情况，共计随访1年。按照7∶3的比例随机分为训练集和测试集，采用6种生存机器学习算法训练模型。结果 共纳入262例患者，发现XGBoost生存机器学习算法表现最佳，在训练集、Bootstrap训练集和测试集中预测内分泌治疗12个月骨关节症状发生的受试者操作特征曲线下面积分别是0.812、0.815和0.753，在训练集和测试集中Bootstrap校准曲线的Brier评分分别是0.184和0.219。此外，XGBoost能准确划分乳腺癌内分泌治疗患者12个月内骨关节症状发生风险的高低，在训练集和测试集中Log-rank检验P值分别为<0.001和0.024。预测因子的重要性排序依次为骨密度、睡眠质量、焦虑抑郁、内分泌药物类型、孕激素受体表达、经济压力、BMI和关节炎史。结论 该研究构建的乳腺癌内分泌治疗患者骨关节症状风险预测模型具有较好的预测性能，可为早期识别和预防骨关节症状提供决策依据。

关键词: 乳腺癌, 内分泌治疗, 骨关节症状, 机器学习, 预测模型, 护理