Modified RECIST submodels and ordinal regression model predict neoadjuvant chemoimmunotherapy response in locally...

This study evaluated predictive models integrating Computed Tomography(CT) and ultrasound(US) to assess neoadjuvant therapy response in locally advanced gastric cancer (LAGC). A prospective multicenter trial (n = 75) developed five RECIST 1.1-derived submodels (RECIST_Expansion, UC_RECIST, CT_RECIST, V-RECIST, U-RECIST) and an ordinal regression-based nomogram, using pathological tumor regression grade (TRG) as the reference standard. The geometric approximation model V-RECIST exhibited superior diagnostic performance for TRG 0-2 vs. 3 discrimination (area under the curve [AUC] = 0.951, sensitivity 96.8%/specificity 80.1%) and TRG 0-1 vs. 2-3 classification (AUC = 0.868, sensitivity 85.7%/specificity 80.0%), validated by calibration curves and decision curve analysis. The US-only U-RECIST demonstrated good diagnostic accuracy (AUC = 0.906/0.811), suitable for preliminary assessment due to its safety and cost-effectiveness. The remaining models-RECIST_Expansion (lymph node evaluation only), UC_RECIST (CT-US combined protocol with gastric lesions as measurable targets), and CT_RECIST (CT-only for no-US settings)-also showed satisfactory accuracy. The ordinal regression model achieved slightly higher AUC values (0.952/0.870) and cross-validation accuracy (0.73) compared to modified RECIST models. These findings highlight the clinical utility of revised RECIST 1.1 models across diverse scenarios, with V-RECIST demonstrating the highest performance, U-RECIST offering radiation-free assessment and cost-effectiveness, and the ordinal model enhancing visualization while preserving diagnostic superiority.