Prediction on Survival Rate of Patients with Non-metastatic Prostate Cancer Based on Deepsurv Model
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.
Deepsurv deep neural network modelNon-metastatic prostate cancerSurvival prediction
万方数据
维普
