乳腺癌术后淋巴水肿风险预测模型的系统评价

Risk Prediction Models for Postoperative Lymphedema of Breast Cancer:A Systematic Review

目的系统评价乳腺癌术后淋巴水肿风险预测模型,为临床实践提供依据。方法计算机检索The Cochrane Library、PubMed、Web of Science、Embase、CINAHL、CBM、CNKI、WanFang Data和VIP数据库中有关乳腺癌术后淋巴水肿风险预测模型的研究,检索时限为建库至2021年11月29日。由2名研究者独立筛选文献、提取数据,并对纳入文献的偏倚风险和适用性进行评价。结果共纳入7项研究,包括6项模型开发研究和1项模型验证研究,其中:3项为COX比例风险模型、4项为Logistic回归模型,1项进行内部和外部验证、5项进行内部验证、1项进行外部验证。模型的受试者工作特征曲线下面积(area under curve,AUC)为0.68~0.88,涉及最多的预测因子是体质量指数(body mass index,BMI)和腋窝淋巴结清扫(axillary lymph node dissection,ALND)。7项研究适用性较好,但偏倚风险较高,主要来源统计分析。结论乳腺癌术后淋巴水肿风险预测模型研究偏倚风险较高,正处于发展阶段;未来研究应遵循模型开发与报告规范,选择或构建预测性能优良、偏倚风险低、便于我国临床实际应用的风险预测模型。

Objective To systematically review the risk prediction models for postoperative lymphedema of breast cancer,so as to provide references for clinical practice.Methods The databases of Cochrane Library,PubMed,Web of Science,Embase,CINAHL,CBM,CNKI,WanFang Data and VIP were searched for the literature on risk prediction models for postoperative lymphedema of breast cancer.The retrieval time frame was from inception to November 29,2021.Two researchers independently screened the literature,extracted relevant data,and evaluated the risk of bias and applicability of involved literature.Results A total of 7 studies were involved,including 6 model development studies and 1 model validation study.Three of them were COX proportional risk models,and four were logistic regression models;one study conducted internal and external validation,five underwent internal validation,and one adopted external validation.The AUC of receiver operating characteristic curve of the involved models ranged from 0.68 to 0.88,and the predictors with the highest frequency were BMI and ALND.Statistical analysis showed that seven studies reported good applicability,while with high risks of bias.Conclusions The literature on risk prediction models for postoperative lymphedema of breast cancer is developing and with high risks of bias.Future research should follow the standards of model development and reporting,choose or construct risk prediction models with good prediction performance,low risk of bias,and good clinical practical applicability.