Prediction of Overall Survival and Novel Classification of Patients with Gastric Cancer Using the Survival Recurrent Network
Artificial neural networks (ANNs) have been applied to many prediction and classification problems, and could also be used to develop a prediction model of survival outcomes for cancer patients.
The aim of this study is to develop a prediction model of survival outcomes for patients with gastric cancer using an ANN.
This study enrolled 1243 patients with stage IIA–IV gastric cancer who underwent D2 gastrectomy from January 2007 to June 2010. We used a recurrent neural network (RNN) to make the survival recurrent network (SRN), and patients were randomly sorted into a training set (80%) and a test set (20%). Fivefold cross-validation was performed with the training set, and the optimized model was evaluated with the test set. Receiver operating characteristic (ROC) curves and area under the curves (AUCs) were evaluated, and we compared the survival curves of the American Joint Committee on Cancer (AJCC) 8th stage groups with those of the groups classified by the SRN-predicted survival probability.
The test data showed that the ROC AUC of the SRN was 0.81 at the fifth year. The SRN-predicted survival corresponded closely with the actual survival in the calibration curve, and the survival outcome could be more discriminately classified by using the SRN than by using the AJCC staging system.
SRN was a more powerful tool for predicting the survival rates of gastric cancer patients than conventional TNM staging, and may also provide a more flexible and expandable method when compared with fixed prediction models such as nomograms.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
M-GC and SWS contributed to the concept of this study and revised it critically; SEO and SWS collected, analyzed the data and drafted the work; and TSS, JMB, and SK ensured that questions related to the accuracy or integrity of any part of the work were appropriately investigated and resolved. All authors gave approval for the final version to be published.
:Sung Eun Oh, Sung Wook Seo, Min-Gew Choi, Tae Sung Sohn, Jae Moon Bae, and Sung Kim report no proprietary or commercial interest in any product mentioned or concept discussed in this article.
- 4.Japanese Gastric Cancer Association. Japanese classification of gastric carcinoma: 3rd English edition. Gastric Cancer. 2011;14:101–112.Google Scholar
- 15.Anguita D, Ghelardoni L, Ghio A, Oneto L, Ridella S. The ‘K’ in K-fold Cross Validation. ESANN 2012 proceedings. ESANN 2012 proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. 25–27 April 2012; Bruges: pp. 441–446.Google Scholar