A machine-learning powered liquid biopsy predicts response to paclitaxel plus ramucirumab in advanced gastric cancer: results from the prospective IVY trial

Paclitaxel plus ramucirumab (PTX + RAM) is a widely used second-line treatment for advanced gastric cancer, yet no validated biomarkers exist to predict therapeutic response. Identifying non-invasive predictors could enable patient stratification and optimize outcomes. We conducted a prospective observational multicenter study (IVY trial; NCT06490055) enrolling 115 patients with advanced gastric cancer treated with PTX + RAM. Serum was collected prior to the initiation of treatment. Small RNA sequencing identified differentially expressed exosomal microRNAs (exo-miRNAs) in patients with controlled disease versus those with progressive disease. Machine learning and logistic regression were employed to construct a predictive model, which was subsequently validated using quantitative real-time polymerase chain reaction (qRT-PCR) in the entire cohort. Ten candidate exo-miRNAs were initially discovered, and a five-miRNA panel (miR-10a-5p, miR-25-5p, miR-125a-5p, miR-139-5p, and miR-450a-5p) was selected via stepwise elimination. This 5-exo-miRNA model achieved high accuracy in distinguishing controlled disease patients from progressive disease patients (AUC = 0.84). When combined with body mass index (BMI), the composite model (EXEMPLAR) demonstrated enhanced predictive performance (AUC = 0.87). High-risk patients exhibited significantly shorter progression-free survival (PFS: median, 1.9 vs. 4.2 months, p = 0.019) and overall survival (OS: median, 1.1 vs. 1.7 years, p < 0.001). Decision curve analysis confirmed the clinical benefit of the model. A nomogram was developed to facilitate personalized risk assessment. This study identifies and validates a novel 5-exo-miRNA panel for predicting response to second-line PTX plus RAM therapy in gastric cancer. The combined exosomal signature and BMI risk model provides a clinically applicable, non-invasive tool for personalized treatment selection.