基于Deepsurv模型预测非转移性前列腺癌患者的生存率

Prediction on Survival Rate of Patients with Non-metastatic Prostate Cancer Based on Deepsurv Model

目的建立Deepsurv深度神经网络模型与Cox比例风险回归模型并比较两种模型对非转移性前列腺癌症患者生存时间的预测性能。方法从SEER数据库中选择2014-2018年确定诊断为非转移性前列腺癌的男性患者,将患者数据集按照8∶2划分为训练集和测试集,构建Deepsurv深度神经网络模型的基本结构。利用随机超参数优化搜索算法获得预定义范围内的最优网络超参数,建立模型后在训练集上训练,并在测试集上测试。通过一致性指数(C-index)、ROC曲线下面积(AUC)和Brier分数(Brier Score)比较Deepsurv深度神经网络模型和Cox比例风险回归模型对非转移性前列腺癌症患者1、3年生存情况的预测性能。结果建立了以患者年龄、前列腺特异性抗原(PSA)水平、前列腺癌组织恶性程度(Gleason分级)、肿瘤分期(T分期)和活检核心阳性总数为预后因素的预测模型。Deepsurv深度神经网络模型的C-index为0.713,高于Cox比例风险回归模型的0.654;Deepsurv深度神经网络模型预测患者1、3年生存率的Brier Score为0.312、0.229,低于Cox比例风险回归模型的0.356、0.241;ROC曲线显示,Deepsurv深度神经网络模型预测患者1、3年生存率的AUC为0.680、0.652,高于Cox比例风险回归模型的0.631、0.649。结论Deepsurv深度神经网络模型在预测非转移性前列腺癌患者的生存方面的表现优于传统的Cox比例风险回归模型。

Objective To establish a Deepsurv deep neural network model and a Cox proportional hazard regression model and compare the predictive performance of the two models on the survival time of patients with non-metastatic prostate cancer.Methods Male patients diagnosed with non-metastatic prostate cancer from 2014 to 2018 were selected from the SEER database.The patient data set was divided into training set and test set according to 8∶2.The basic structure of the Deepsurv deep neural network model was constructed.The random hyperparameter optimization search algorithm was used to obtain the optimal network hyperparameters within the predefined range.After the model was established,it was trained on the training set and tested on the test set.The predictive performance of Deepsurv deep neural network model and Cox proportional hazard regression model for 1-year and 3-year survival of patients with non-metastatic prostate cancer was compared by consistency index(C-index),area under the ROC curve(AUC)and Brier score(Brier Score).Results A predictive model was established with patient age,prostate specific antigen(PSA)level,malignant degree of prostate cancer tissue(Gleason grade),tumor stage(T stage)and total number of positive biopsy cores as prognostic factors.The C-index of the Deepsurv deep neural network model was 0.713,which was higher than 0.654 of the Cox proportional hazard regression model.The Brier Scores of the Deepsurv deep neural network model for predicting the 1-year and 3-year survival rates of patients were 0.312 and 0.229,which were lower than 0.356 and 0.241 of the Cox proportional hazard regression model.The ROC curve showed that the AUC of the Deepsurv deep neural network model for predicting the 1-year and 3-year survival rates of patients was 0.680 and 0.652,which was higher than 0.631 and 0.649 of the Cox proportional hazards regression model.Conclusion Deepsurv deep neural network model is superior to the traditional Cox proportional hazard regression model in predicting the survival of patients with non-metastatic prostate cancer.