基于随机森林模型的乳腺癌患者术后并发下肢深静脉血栓预测模型的初步研究

Preliminary study on the prediction model of postoperative deep vein thrombosis of lower extremity in breast cancer patients based on random forest model

目的探讨乳腺癌患者术后并发下肢深静脉血栓的危险因素,并建立预测乳腺癌患者术后并发下肢深静脉血栓的随机森林模型。方法收集2018年2月至2021年7月于潍坊市人民医院进行手术治疗的430例乳腺癌患者的临床资料,分析乳腺癌患者术后并发下肢深静脉血栓的危险因素,运用R软件建立预测乳腺癌患者术后并发下肢深静脉血栓的随机森林模型。结果乳腺癌患者术后下肢深静脉血栓的发生率为14.19%(61/430)。多因素分析结果显示,年龄≥60岁、肿瘤分期为Ⅲ~Ⅳ期、体重指数≥30 kg/m^(2)、手术时间≥2 h、新辅助化疗、糖尿病、术前D-二聚体水平>5.0 mg/L及术前甘油三酯(TG)水平>1.70 mmol/L均是乳腺癌患者术后并发下肢深静脉血栓的独立危险因素(P﹤0.05)。随机森林模型结果显示,乳腺癌患者术后并发下肢深静脉血栓影响因素的重要性排序依次是肿瘤分期、体重指数、糖尿病、术前TG水平、年龄、新辅助化疗、手术时间及术前D-二聚体水平。随机森林模型预测乳腺癌患者术后并发下肢深静脉血栓的曲线下面积(AUC)为0.785(95%CI:0.706~0.855),Logistic回归模型预测乳腺癌患者术后并发下肢深静脉血栓的AUC为0.773(95%CI:0.702~0.844)。结论肿瘤分期为Ⅲ~Ⅳ期、体重指数≥30 kg/m^(2)、糖尿病、术前TG水平>1.70 mmol/L、年龄≥60岁、新辅助化疗、手术时间≥2 h及术前D-二聚体水平>5.0 mg/L均是乳腺癌患者术后并发下肢深静脉血栓的危险因素,随机森林模型有助于临床医师快速、简便地分析乳腺癌患者术后并发下肢深静脉血栓的风险并尽早制定防治措施。

Objective To investigate the risk factors of postoperative deep vein thrombosis(DVT)of lower extremity in patients with breast cancer,and to establish a random forest model for predicting postoperative deep vein thrombosis of lower extremity in patients with breast cancer.Method The clinical data of 430 breast cancer patients who underwent surgical treatment in the Weifang People's Hospital from February 2018 to July 2021 were collected.The risk factors of postoperative deep vein thrombosis of lower extremity in breast cancer patients were analyzed,and the R software was used to establish a random forest model for predicting postoperative deep vein thrombosis of the lower extremity in breast cancer patients.Result The incidence rate of lower extremity DVT after breast cancer surgery was 14.19%(61/430).Multivariate analysis showed that age≥60 years old,stage Ⅲ-Ⅳ tumors,body mass index≥30 kg/m^(2),operation time≥2 h,neoadjuvant chemotherapy,diabetes,preoperative D-dimer>5.0 mg/L and preoperative triglyceride(TG)>1.70 mmol/L were risk factors for postoperative DVT of lower extremity in breast cancer patients(P<0.05).The results of random forest model showed that the order of importance of factors affecting postoperative DVT of lower extremity in breast cancer patients was tumor stage,body mass index,diabetes,preoperative TG level,age,neoadjuvant chemotherapy,operation time and preoperative D-dimer level.The random forest model predicted the area under the curve(AUC)of postoperative DVT in breast cancer patients was 0.785(95%CI:0.706-0.855),and the Logistic regression model predicted the AUC of postoperative DVT in breast cancer patients was 0.773(95%CI:0.702-0.844).Conclusion Stage Ⅲ-Ⅳ tumors,body mass index≥30 kg/m^(2),diabetes,preoperative TG>1.70 mmol/L,age≥60 years old,neoadjuvant chemotherapy,operation time≥2 h and preoperative D-dimer>5.0 mg/L are risk factors for postoperative DVT in patients after breast cancer surgery.The random forest model is helpful for clinicians to quickly and easily analyze the risk of postoperative DVT in patients after breast cancer surgery and to formulate the prevention and treatment methods of DVT early.