Dual-Biofluid Metabolomics Combined With Machine Learning Enables Noninvasive Diagnosis of Drug Resistance in Benign
Shilong Liu1,2, Yanbo Chen1, Ziwei Wang1
1Department of Urology School of Medicine, Shanghai Ninth People's Hospital, Shanghai Jiao Tong University, Shanghai, People's Republic of China.
Abstract:
With global population aging, benign prostatic hyperplasia (BPH) prevalence has risen, with one-quarter of patients showing inadequate treatment response or drug resistance (DR) and no reliable noninvasive diagnostic method currently available. Current methods rely on clinical experience, delaying precision management. We developed a nanoparticle-enhanced mass spectrometry platform to analyze serum and urine metabolomics in 224 BPH patients (104 DR, 120 drug-sensitive [DS]). Using gradient-boosted decision trees (GBDT), we integrated dual-biofluid metabolic fingerprints (serum and urine) to distinguish DR/DS subgroups. The platform enabled rapid analysis (<25 s/sample, 1 µL volume) with high reproducibility (CV < 10%). SMF yielded a five-feature panel including five putatively annotated DR-associated metabolites, achieving an AUC of 0.89 for DR detection. UMF analysis yielded 3 features (AUC = 0.81). Critically, combining SMF/UMF data (8-feature panel) enhanced diagnostic performance to AUC = 0.95 (95% CI: 0.91-0.96), outperforming single-biofluid models. This dual-biofluid metabolomic approach provides a noninvasive, rapid method for BPH-DR stratification, integrating systemic (serum) and local (urine) metabolic insights. The platform's scalability and 0.95 AUC highlight its potential for clinical translation, offering a foundation for personalized BPH management and reducing reliance on invasive procedures. Notably, multicenter external validation further supported the model's discriminative performance in an independent cohort.