Abstract:

Objective To develop and evaluate a risk predictive model for high level of self-reported symptom cluster distress in breast cancer patients undergoing chemotherapy. Methods 647 patients who received chemotherapy and met the inclusion and exclusion criteria were selected at 4 tertiary hospitals in Shanghai，Shandong，Jiangsu，and Zhejiang from October 2019 to May 2021. Cases were randomly assigned to a modeling group and a validation group based on 5-fold cross-validation method at a ratio of 8：2. The random forest algorithm was used to develop the risk predictive model in the modeling group. The receiver operating characteristic curve，Hosmer-Lemeshow goodness-of-fit test，calibration curve，and decision curve were used to comprehensively evaluate the prediction performance of the model in the validation group. Risk factors were identified based on the order of the importance of each influencing factors. Results The incidence of high symptom cluster distress was 33.27% in the modeling group and 29.23% in the validation group. The area under the receiver operating characteristic curve of the prediction model was 0.91；the sensitivity was 65.8%；the specificity was 93.5%；Hosmer-Lemeshow goodness-of-fit test was insignificant（P=0.1365）；the decision curve was above the reference line. Body image，self-efficacy，and financial burden were the most important predictors. Conclusion The risk predictive model based on random forest algorithm has good predictive performance，which is of great significance to help identify breast cancer subgroups at high risk of symptom cluster distress，and will potentially promote symptom management.

Key words: Breast Neoplasm, Symptom Distress, Symptom Cluster, Random Forest Algorithms, Predictive Model

摘要：

目的 构建化疗期乳腺癌患者自我报告症状群困扰高风险预测模型并评价模型的预测性能。 方法 采用便利抽样法，选取2019年10月—2021年5月在上海市、山东省、江苏省、浙江省4所三级甲等医院接受化疗并符合纳入标准的乳腺癌患者647例，按5折交叉验证法以8 ∶ 2的比例随机分为建模组和验证组。在建模组中采用随机森林算法构建，在验证组中采用受试者操作特征曲线、Hosmer-Lemeshow拟合优度检验、校准曲线以及决策曲线综合评价模型的预测性能，最后对各影响因素进行重要性排序。 结果 乳腺癌患者高症状困扰发生率建模组为33.27%，验证组为29.23%。预测模型的受试者操作特征曲线下面积为0.91，灵敏度为65.8%，特异度为93.5%；Hosmer-Lemeshow拟合优度检验P=0.136；决策曲线显示高于参考线。身体心像、自我效能感、经济负担等为最主要的预测因子。 结论 基于随机森林算法构建的预测模型具有良好的预测性能，对识别症状困扰高风险的乳腺癌患者有重要意义。

关键词: 乳腺肿瘤, 症状困扰, 症状群, 随机森林算法, 预测模型