Development and validation of a logistic regression model for the diagnosis of colorectal cancer

Abstract

Colorectal cancer (CRC) diagnosis is challenging due to generalized symptoms. Various biomarker models exist, but their clinical application is limited by low sensitivity and heterogeneous cutoff values. This study aimed to develop and validate a diagnostic model for CRC. Data from 489 patients—337 with CRC and 152 with benign disease—were included. Patients were randomly assigned to training (n = 342) and validation (n = 147) cohorts. Logistic regression identified age (OR 1.06), CA153 (OR 0.26), CEA (OR 4.49), CYFRA 21-1 (OR 5.88), ferritin (OR 0.15), and hs-CRP (OR 0.05) as independent risk factors. Sensitivity and specificity were 88.61% and 82.86% in the training cohort and 90.00% and 76.60% in the validation cohort. Cutoff values for the biomarkers were: CA199, 9.809 U/mL; CA125, 7.743 U/mL; CA153, 6.295 U/mL; CEA, 3.982 ng/mL; CYFRA 21-1, 1.769 ng/mL; ferritin, 163.361 mg/L; hs-CRP, 0.196 mg/L; and serum albumin, 55.966 g/L. The model showed higher sensitivity for early-stage CRC (95.45%, 95% CI 87.2–98.6%) than late-stage CRC (87.27%, 95% CI 76.4–93.5%; P = 0.08). AUCs were 0.907 (training) and 0.872 (validation). The model demonstrated higher sensitivity for early-stage CRC (95.45%) than late-stage CRC (87.27%), underscoring its utility in early detection.