XGBoost-based urinary microbial signatures enable non-invasive diagnosis and prognosis for urothelial carcinoma

BACKGROUND: Urothelial carcinoma (UC) is the most common malignant tumor of the urinary system, characterized by high incidence and recurrence rates, posing a serious threat to human health. While previous studies have linked urinary microbiota alterations to bladder cancer, little is known about the broader spectrum of UC subtypes or their clinical implications. To address this gap, we analyzed microbiota profiles across multiple UC subtypes, incorporating microbial network analyses and machine learning to establish diagnostic models and identify prognostic biomarkers. METHOD: A total of 112 subjects were enrolled for 16 S rDNA sequencing of clean-catch midstream urine samples, including 63 patients with bladder cancer (BCA), 29 with Upper Tract Urothelial Carcinoma (UTUC), 9 with renal pelvis cancer (RPC), and 40 healthy controls (HC). Microbial diversity, community networks, and clinical associations were analyzed. An XGBoost-based diagnostic model was developed with a 70/30 train-test split, cross-validation, and external validation. Model interpretability was assessed with the SHAP algorithm. RESULTS: UC groups showed elevated α-diversity versus HC, with consistent enrichment of Streptococcus and Clostridium. Microbial structure networks significantly differed in tumors. A urinary microbiota-based diagnostic model achieved high accuracy for BCA detection (AUC = 0.927), and Lachnospiraceae family members showed potential prognostic value. CONCLUSION: Our study illuminates the microbial profiles in the UC and suggests that urinary microbiota signatures represent promising non-invasive independent biomarker for the diagnosis and prognosis of UC.