Development of a Routine Serological Test Index Panel for the Surveillance of Gastric Cancer Risk in a High-Risk Population.

Abstract

Objective: This study aims to develop a predictive model for the detection of gastric cancer risk utilizing non-invasive parameters and to assess the model's effectiveness in risk stratification for gastric cancer (GC).

Methods: A case-control study was conducted among inpatients with various gastric diseases. These individuals were categorized into two groups: the gastric cancer group (138 cases) and the chronic non-atrophic gastritis (CNAG) group (319 cases). We employed a comprehensive panel of hematological, biochemical, and coagulation parameters derived from routine blood tests. Random Forest and Logistic regression analysis was used for feature selection and model building. Statistical analyses were performed using R version 4.2.3.

Results: Logistic regression analysis was employed to establish risk prediction models for GC, incorporating variables such as D-dimer, carcinoembryonic antigen (CEA), carbohydrate antigen 724 (CA724), and hemoglobin (HGB). A visual nomogram was generated as the final prediction model. The area under the receiver operating characteristic curve (AUC) for the training and test sets were 0.8093 [95% confidence interval (CI), 0.7541-0.8644], and 0.8076 [95% CI 0.7237-0.8915], respectively. Furthermore, we have developed an HTML file, featuring the Logistic equation, which enables real-time assessment of GC risk scores.

Conclusion: The performance of this predictive model demonstrates its adequacy, making it a valuable and cost-effective noninvasive tool for identifying early gastric cancer (EGC) in patients. Consequently, this model may facilitate the implementation of targeted preventive and intervention strategies in clinical practice.