基于机器学习方法构建IDH野生型胶质母细胞瘤预测模型研究

目的建立异柠檬酸脱氢酶(IDH)野生型胶质母细胞瘤生存概率的列线图模型及随机生存森林模型。方法回顾性分析2017年1月—2020年12月在空军军医大学附属西京医院手术治疗的127例IDH野生型胶质母细胞瘤患者临床资料，进行预后因素分析并建立列线图模型及随机生存森林模型，通过C指数，校准曲线，决策曲线评价模型的区分度，校准度以及临床净获益率。结果使用Cox比例风险模型进行多因素分析发现，患者术前卡氏功能状态评分（KPS）、是否接受同步放化疗、年龄、O6-甲基鸟嘌呤甲基转移酶（MGMT）蛋白表达，是独立的预后因素（P<0.05）。通过Cox回归模型建立列线图预测模型；通过R软件建立随机生存森林模型，两个模型均具有良好的区分度和校准度，随机生存森林模型的临床净获益优于列线图模型。结论建立的列线图模型及随机生存森林模型有助于临床医生判断患者特定时间点的生存概率。

Objective To create nomogram prediction and random survival forest models for patients with isocitric dehydrogenase(IDH) wild-type glioblastoma to estimate their survival probabilities. Methods The clinical data of 127 patients diagnosed with IDH wild-type glioblastoma at Xijing Hospital Affiliated to Air Force Military Medical University from January 2017 to December 2020 were analyzed retrospectively. Prognostic factor analysis was conducted and a column chart model and a random survival forest model were established. The discrimination, calibration, and clinical net benefit rate of the model were evaluated through the C-index, calibration curve, and decision curve. Results Multivariate analysis using Cox proportional hazards model revealed that patients had preoperative Karnofsky performance scale（KPS）, acceptance of concurrent radiotherapy and chemotherapy, age, the expression of O6-methylguanine-DNA methyltransferase(MGMT) protein were an independent prognostic factor(P<0.05). The nomogram prediction model was developed using Cox regression, while the random survival forest model was established via the software R. Both models demonstrated excellent discrimination and calibration, with the random survival forests exhibiting a superior clinical net benefit compared to nomograms. Conclusion The established column chart model and random survival forest model help clinical doctors determine the survival probability of patients at specific time points.