Application of routine test big data in early diagnosis of gastric cancer / 中华检验医学杂志

ABSTRACT

Objective:To evaluate the feasibility of a predictire model composed of non-specific test indexes in early diagnosis of gastric cancer.Methods:From the database of electronic medical record system of Shanghai Changhai Hospital, a total of 24 615 case records were included from January 1, 2010 to April 30, 2019, including 10 497 cases of gastric cancer, 5 198 cases of precancerous diseases, and 8 920 cases of health examination. Through stratified random sampling, the study population was divided into validation set, training set and test set. After data processing and quality control for all laboratory variables, the optimal machine learning algorithm and diagnostic efficiency grouping were selected through four machine learning algorithms, induding the gradient boosting decision tree, random forest, support vector machine, and artificial neural network, and the data were trained by backward stepwise regression method to build the best feature model.Result:In this study, a diagnostic model V22 consisting of 22 routine testing parameters was established. V22 could distinguish early gastric cancer from control group composed of healthy group and precancerous disease, AUC was 0.808, the sensitivity was 85.7%, and the specificity was 91.9%. For CEA negative gastric cancer, V22 also showed high diagnostic accuracy, AUC was 0.801.Conclusion:V22 was a valuable model for the diagnosis of gastric cancer. V22 was an auxiliary diagnostic model of gastric cancer with clinical application value, which could well distinguish early gastric cancer from the control group composed of healthy group and precancerous disease, and the detection rate of early gastric cancer was better than the traditional tumor marker CEA.