A commentary on "An artificial intelligence model to predict survival and chemotherapy benefits for gastric cancer patients after gastrectomy development and validation in international multicenter cohorts".

We read with great interest the article by Li et al1. This study included 255 gastric cancer (GC) patients who underwent surgical R0 resection from a Chinese institution as the training and internal validation cohort and included 587 patients with the same inclusion and exclusion criteria from 4 domestic centres and The Cancer Genome Atlas (TCGA) database as an external validation cohort. This study constructed a support vector machine (SVM) model based on clinicopathological factors. The model can be used to predict survival outcomes for GC patients after surgical R0 resection accurately and easily, and it presented to be a valuable addition to the current TNM staging system. Despite inspiring and thought-provoking, we have the following comments. First, the authors highlighted that the predictive performance of the SVM was better than that of the nomogram, and in the section Discussion, the authors mentioned: The SVM was proven to be a more powerful classification tool than the nomogram via Cox proportional risk regression analysis. From our perspective of view, for the modeling method, only the most suitable, not the best. In fact, SVM is suited to manage classification based on high-dimensional data with a limited number of training samples to select the most efficient of all available features2,3. When predicting the group for which a new observation is made, SVM models based on a single variable are a viable alternative to logistic regression. However, for the Poisson, Normal, and Exponential distributions, the polynomial SVM model is not recommended by previous research4. If the authors want to show that SVM can classify different types of patients better than logistic regression, they should construct two models, one based on SVM, and another Nomogram based on logistic regression. Then, this study confirmed that the area under the curve of the model based on SVM was greater than that of the nomogram to demonstrate that the SVM-based model might have better predictive performance. It is not too much convincing to cite other references as evidence. Second, the authors mentioned at the outset that they wanted to build a simple tool to predict survival in GC patients. However, pure computational approaches have been viewed as “black boxes” to which data sets are thrown in and calculations are implemented without a clear understanding of what occurs inside. This consequently limits their application. Then, from the point of view of practical clinical application, how can surgeons easily classify the risks of different patients without visual tools such as nomogram? Finally, we are still very appreciative of Li and colleagues for developing such a new model based on SVM to provide a personalized prediction of survival and greatly assist surgeons and GC patients. Ethical approval Not available. Sources of funding This study was supported by Chongqing Technology Innovation and Application Development Special Key Project (No. CSTC2021jscx-gksb-N0009). Authors’ contribution Z.-P.L., H.-S.D., and X.-Y.Y.: manuscript preparation. Y.-Q.Z. and Z.-Y.C.: critical revision. Conflicts of interest disclosure The authors declare that they have no financial conflict of interest with regard to the content of this report. Research registration unique identifying number (UIN) Not available. Guarantor Zhi-Yu Chen.