%0 Journal Article
%T 基于随机生存森林的低级别脑胶质瘤复发预测模型<br>Prediction Model of Recurrence of LGG Based on Random Survival Forest
%A 毛丹怡
%A 曾庆
%A 杨裕隆
%A 陈宇轩
%A 钱金山
%J Statistics and Applications
%P 151-163
%@ 2325-226X
%D 2024
%I Hans Publishing
%R 10.12677/SA.2024.131016
%X 背景：低级别脑胶质瘤(LGG)患者在治疗后仍可能面临复发的风险。本研究旨在考虑复发与死亡相关性的基础上分析影响LGG患者复发的因素，并建立复发预测模型。方法：收集来自TCGA数据库LGG患者的临床数据，最终纳入457例LGG患者。建立联合脆弱模型分析复发危险因素，采用随机生存森林(Random Survival Forest，RSF)的方法建立LGG复发预测模型，同时建立Cox模型作为比较。结果：主要治疗结局、年龄、ICD-O-3组织学编码、首次症状持续时间、术后、肿瘤组织学分级和放射治疗作为预测因子纳入LGG复发预测模型。与Cox模型相比，RSF模型具有较好的区分度和校准度，具体表现为，基于bootstrap重抽样数据集计算1、3和、5年RSF的C指数分别为0.813、0.748和0.745，Cox分别为0.824、0.724和0.727，RSF在1、3、5年的AUC值分别为0.824、0.746和0.754，而Cox分别为0.833、0.713和0.730；校准曲线也表明RSF模型表现更优。结论：复发和死亡事件不应独立看待，RSF在复杂生存数据的预测建模方面具有优势，未来期待更多探索预测脑胶质瘤患者复发的方法和工具，且这些方法能得到在临床上的实践应用。<br />
Background: Low-Grade Glioma (LGG) patients face risks of recurrence after treatment. Our study aims to analyze recurrence risk factors, considering the association between recurrence and death, and to develop an improved recurrence prognosis prediction model for LGG patients. Methods: We collected clinical data from TCGA, 457 LGG patients were finally included. We developed a Joint Frailty model to analyze recurrence risk factors. We employed the Random Survival Forest (RSF) model for recurrence prognosis prediction, with a Cox model for comparison. Results: Prognostic factors, including primary therapy outcome, age, ICD-O-3 histology, first presenting symptom dura-tion, postoperative, neoplasm histologic grade, and radiation therapy, were integrated into the LGG recurrence prediction model. The RSF model excelled in discrimination and calibration compared to the Cox model. RSF’s C-index for 1, 3, and 5 years on bootstrap validation was 0.813, 0.748, and 0.745, respectively, versus Cox’s 0.824, 0.724, and 0.727. RSF’s AUC values for 1, 3, and 5 years by bootstrapping were 0.824, 0.746, and 0.754, respectively, versus Cox's 0.833, 0.713, and 0.730. The calibration curve also favored the RSF model. Conclusions: Recurrence and death events should not be treated independently. RSF has advantages for predictive modeling of complex survival data. We foresee further research to enhance LGG recurrence prediction, with potential practical clinical ap-plications in the future.
%K 低级别脑胶质瘤，复发预测模型，随机生存森林，联合脆弱模型，TCGA<br>Low-Grade Glioma
%K Recurrence Prediction Model
%K Random Survival Forest
%K Joint Frailty Model
%K TCGA
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=81323