中野 尚子

The gut microbiome is a potential source of non-invasive cancer biomarkers. We evaluated six fecal microbial markers and developed targeted qPCR-based logistic models for colorectal cancer (CRC) and pancreatic cancer (PC). Using LASSO with the 1-standard-error rule, four markers were selected for CRC (afb, nan, fsr, and 5ar) and three for PC (but, fsr, and saa). In post-selection leave-one-out cross-validation of fixed model structures, the CRC and PC models yielded AUCs of 0.824 and 0.780, respectively. Fixed-model application yielded AUCs of 0.716 for colorectal adenoma and 0.540 for the pancreatic high-risk group. In an exploratory Early PC versus high-risk comparison, the fecal qPCR score showed a higher AUC point estimate than CA19-9, while the difference was not statistically significant. Overall, the disease-specific model performance and fixed-model behavior across clinically related groups support further evaluation of model-derived cancer probability scores as exploratory cancer-assessment tools.