Serum Cystatin 4 combined with routine clinical laboratory indicators: A support vector machine diagnostic model for...

Gastrointestinal tumors lack non-invasive early screening tools, with most patients diagnosed at advanced stages. Serum tumor markers combined with machine learning show promise, but single markers have low specificity. This study aimed to construct a non-invasive diagnostic model using multiple laboratory variables. A retrospective study included 214 gastrointestinal tumor patients (observation group) and 130 non-tumor individuals (control group). Thirty-eight laboratory indicators were detected, and seven core variables (HCT, Age, TP, ALB, PLT, WBC, CST4) were identified via feature selection. Eleven machine learning algorithms were used to build models, with performance evaluated by AUC, sensitivity, specificity, calibration curves, and DCA. Serum CST4 levels were significantly higher in the observation group (P < 0.001, AUC = 0.706). The SVM model showed the best performance: AUC = 0.85 (95% CI: 0.78-0.92), sensitivity = 95.4% (95% CI: 0.87-0.98), specificity = 61.5% (95% CI: 0.52-0.71), PPV = 80.5% (95% CI: 0.74-0.86), NPV = 88.9% (95% CI: 0.81-0.94) in the test set. Calibration curves demonstrated high consistency (deviation < 5%), and DCA confirmed superior net benefits. SHapley Additive exPlanations (SHAP) analysis identified HCT, CST4, and Age as top contributors. This internally validated tool demonstrated promising preliminary performance for gastrointestinal tumor screening in high-risk populations (sensitivity 95.4%, specificity 61.5%). Due to its modest specificity, it is suitable as an auxiliary risk stratification tool for high-risk individuals but not for general population screening. External prospective multi-center validation is mandatory before clinical application.