Constructing a Risk Prediction Model of Breast Cancer-related Lymphedema Based on a Meta-analysis of Prospective Cohort Studies
SHEN Aomei,LU Qian,FU Xin,WEI Xiaoxia,BIAN Jingru,ZHANG Liyuan,QIANG Wanmin,PANG Dong
DOI: https://doi.org/10.12114/j.issn.1007-9572.2022.0827
2023-01-01
Chinese General Practice
Abstract:Background Lymphedema is a common chronic complication bothering breast cancer patients. Early assessment and prediction of the risk for developing breast cancer-related lymphedema (BCRL) is particularly important. However, there is still a lack of an authoritatively recognized and suitably promoted risk prediction model.Objective To construct and validate a risk prediction model for BCRL based on the results of a meta-analysis.Methods Electronic databases including PubMed, Embase, CINAHL, Scopus, Web of Science, Cochrane Library, CNKI, CBM, and Wanfang Data, were searched for prospective cohort studies on risk factors of BCRL from inception to November 2021. Two systematically trained researchers independently screened the literature, extracted data, and assessed the study quality using the Newcastle-Ottawa Scale. Stata 17.0 was used for meta-analysis. The risk factors with significant pooled effect size and their combined risk value were extracted to construct the Logistic risk prediction model. The Logistic and additive risk scoring models were constructed based on regression coefficients and pooled risk values, respectively. The data of 486 breast cancer patients recruited in the breast cancer research center of Peking University People's Hospital, from April 2017 to December 2018, were selected as the validation set. The area under the ROC curve (AUC) and the Hosmer-Lemeshow test were used to evaluate the prediction performance of the risk scoring model. Decision curve analysis was used to evaluate the clinical practicability.Results A total of 49 prospective cohort studies involving 32 543 breast cancer patients were included. Meta-analysis showed that the incidence of BCRL was 20.6%〔95%CI (17.9%, 23.3%) 〕. Among 49 studies, five risk factors with significant pooled effect sizes were reported more than five times, including body mass index (BMI) 〔RR=1.777, 95%CI (1.515, 2.085) 〕, type of breast surgery〔RR=1.320, 95%CI (1.125, 1.549) 〕, type of axillary surgery〔RR=3.058, 95%CI (2.325, 4.020) 〕, radiotherapy〔RR=1.620, 95%CI (1.214, 2.160) 〕, and postoperative complications〔RR=2.373, 95%CI (1.278, 4.405) 〕. The total score for the Logistic risk scoring model ranged from 0 to 34, and that for the additive risk scoring model ranged from 5 to11. The AUCs of Logistic and additive risk scoring models were 0.748〔95%CI (0.701, 0.794) 〕and 0.737〔95%CI (0.691, 0.784) 〕, respectively. The values of Hosmer-Lemeshow test were 0.185 and 0.763, respectively. The optimal cut-off value of the Logistic risk scoring model was 18, with a sensitivity of 81.7%, and a specificity of 43.1%. The optimal cut-off value of the additive risk scoring model was 8.5, the sensitivity was 80.9%, and the specificity was 42.8%. When the probability threshold ranged from 20% to 30%, the model achieved higher net clinical benefit. Conclusion The BCRL risk prediction model based on this meta-analysis has good predictive performance. It can be used as a risk assessment tool for BCRL to guide the hierarchical monitoring and management of BCRL. However, prediction performance and clinical practicability of the model still needs to be validated and optimized in future research.