预测上皮性卵巢癌复发机器学习模型及列线图的建立

Development of machine learning models and nomogram to predict epithelial ovarian cancer recurrence

目的:本研究拟基于机器学习及Cox回归开发上皮性卵巢癌复发机器学习模型及列线图。方法:回顾性分析2010年01月至2020年12月于云南省肿瘤医院确诊739例Ⅲ-Ⅳ期EOC患者的医疗记录。收集患者的基本信息、手术、化疗细节和预后结果。使用单多因素逻辑回归及Cox回归筛选变量,使用5种机器学习算法基于单多因素逻辑回归的结果构建预测模型,采用10折交叉验证方法评估模型性能。基于Cox回归结果开发列线图。结果:739例患者中,399(54.0%)例最终发生了复发,340(46.0%)例未复发。复发患者分期以Ⅲ_(C)期为主,占59.1%,病理类型以浆液性癌为主,占91.0%。单多因素逻辑回归显示围手术期化疗周期、术后残余病灶、手术方式、新辅助化疗是与复发独立相关的4个变量,基于这些变量和FIGO分期建立5个机器学习模型,XGBoost在识别复发病例方面表现最佳,AUC为0.775。Cox回归分析显示,术前局部灌注化疗、残余病灶直径、围手术期化疗周期、手术方式是影响复发的独立危险因素,基于上述因素开发了晚期上皮性卵巢癌患者复发的预测列线图。结论:机器学习模型和列线图可早期识别卵巢癌复发,通过早期识别可改善晚期卵巢癌预后。

Objective:To develop machine learning models and a nomogram to predict epithelial ovarian cancer recurrence based on machine learning and Cox regression.Methods:The medical records of 739 patients diagnosed with stage Ⅲ-Ⅳ EOC in Yunnan Cancer Hospital from January 2010 to December 2020 were retrospectively analyzed.Basic information,surgery,chemotherapy details and survival results were collected.Univariate and multivariate logistic regression and Cox regression were used to screen variables.Five machine learning algorithms were used to construct models based on the results of multivariate logistic regression.The model performance was evaluated by 10-fold cross-validation.The nomogram was developed based on the Cox regression results.Results:A total of 739 patients entered the study,among them,399(54.0%)eventually recurred and 340(46%)were censored.StageⅢ_(C) predominates in relapsed populations,accounting for 59.1%.The pathological type was mainly serous carcinoma,accounting for 91.0%.Multivariate logistic regression showed that the perioperative chemotherapy cycle,residual tumorsize,surgical method,and neoadjuvant chemotherapy were four variables independently related to recurrence.Based on these variables and the FIGO stage,five machine learning models were established.The XGBoost-based model performed best with an AUC of 0.775.Cox regression showed that local infusion chemotherapy before surgery,residual tumor,perioperative chemotherapy cycle,and surgical approach were independent prognostic risk factors.A nomogram for predicting recurrence in patients with advanced epithelial ovarian cancer was developed based on the four factors.Conclusion:Machine learning models and nomogram could identify ovarian cancer recurrence early,demonstrating the potential to improve the prognosis of advanced ovarian cancer through early detection.