Application of machine learning algorithms and establishment of a web calculator in predicting distant metastasis of T2-T4 gastric cancer.

BACKGROUND

T2-T4 gastric cancer often has distant metastasis.The aim of this study is to establish and validate a prediction model for distant metastasis of T2-T4 gastric cancer using machine learning algorithms.

METHODS

We developed nine machine learning models using 17030 patients with T2-T4 gastric cancer in the Surveillance, Epidemiology, and End Results (SEER) database. 100 patients from a Chinese hospital were selected for external verification of the performance of the model. We evaluated the model using the area under the receiver operating characteristic curve (AUC), the area under the exact recall curve (AUPRC), decision curve analysis, calibration curve, accuracy, F1-score, precision, and specificity for the internal test set and the external validation cohort. We used Shapley's Additive explanation (SHAP) to explain the machine-learning models. Finally, the best model was applied to develop a network calculator for predicting the risk of distant metastasis of gastric cancer.

RESULTS

Multivariate logistic regression analysis showed that age, AJCC N, tumor size, tumor number, primary site, differentiation, and histology are independent risk factors for distant metastasis of gastric cancer. The GBDT model was the best model compared with the other 8 machine learning models in three sets. Finally, we constructed a network calculator using the GBDT model.

CONCLUSION

The GBDT model has a good predictive efficiency for predicting the risk of distant metastasis in patients with T2-T4 gastric cancer, and the construction of a network calculator can help clinicians make clinical decisions.