Plasma cfDNA fragmentomics enables non-invasive detection and preoperative risk stratification of bladder cancer in a...

Accurate non-invasive tools for bladder cancer detection and preoperative assessment of muscle invasion remain limited. We developed a plasma cell-free DNA (cfDNA) fragmentomics assay to support two clinically related tasks: detection of bladder cancer and preoperative estimation of muscle-invasive risk. Using low-pass whole-genome sequencing, we extracted fragment size coverage, copy-number variation, and nucleosome positioning features. The multicenter cohort included 656 participants and an external extension of 31 cancer cases (total, 687). An integrated classifier was first trained for cancer detection. Subsequently, a separate non-muscle-invasive bladder cancer (NMIBC) versus muscle-invasive bladder cancer (MIBC) classifier was developed using definitively staged cases from the training cohort with internal threshold selection and independent external validation. For cancer detection, the model achieved AUCs of 0.9484 (training), 0.9446 (internal validation), and 0.9390 (external validation). At the locked threshold of 0.592, selected in the internal validation cohort, sensitivities were 81.0%, 80.0%, and 84.3%, with specificities of 95.0%, 94.7%, and 91.8% against healthy controls and 93.3% and 91.1% against benign disease controls in internal and external validation, respectively. For invasion stratification, the NMIBC/MIBC classifier achieved AUCs of 0.9254, 0.9004, and 0.8845 in the training, internal validation, and expanded external validation cohorts, respectively, with 90.3% sensitivity and 77.4% specificity for MIBC detection externally. The assay demonstrated high reproducibility across storage conditions and sequencing batches. This blood-based framework supports multicenter bladder cancer detection and preoperative stratification of muscle invasion as an adjunctive translational approach, while prospective validation in defined clinical use settings remains necessary.