AI and Cancer Research ── Gastric ── 2026-11-05
Gastric cancer: treatment, trials and AI
- Abstract
Neoadjuvant chemotherapy (NAC) is a standard treatment for locally advanced gastric cancer (LAGC), but response varies greatly among patients. Existing predictive models often neglect host-related factors such as body composition. This study aimed to develop and validate a multimodal model integrating clinical, tumor radiomics, and body composition features to predict NAC response. A total of 1,568 LAGC patients were analyzed, including 1,150 from six centers for model development and validation, 352 in a real-world incremental cohort, and 66 from two prospective trials (ChiCTR2000030414 and NCT03718624). Radiomic features were extracted from pre-treatment CT scans of tumors and L3-level body composition. Feature selection included statistical testing and LASSO regression. A multimodal ensemble model (NeoGEM) was constructed by integrating clinical and radiomic features. NeoGEM achieved high performance in the training cohort (AUC = 0.888; accuracy = 0.809) and showed comparable discrimination across internal (AUC = 0.870–0.875), external (AUC = 0.859–0.877), and prospective cohorts (AUC = 0.864–0.867). The model demonstrated stable predictive performance across different chemotherapy regimens and treatment cycle subgroups. In the incremental real-world cohort, where radiological rather than pathological response was used, NeoGEM also maintained good discrimination (AUC = 0.840), providing supportive evidence of broader clinical applicability. NeoGEM achieved higher AUCs than the clinical, tumor-only, and body-composition-only models across the validation cohorts examined. NeoGEM score stratification was associated with 5-year OS (35.5% vs. 62.4%) and DFS (21.5% vs. 50.8%) in the low vs. high NeoGEM score groups ( P < 0.001). Exploratory transcriptomic analysis suggested immune-related pathway enrichment and higher infiltration of CD4+, CD8+, and FCGR3A+ cells in tumors from high NeoGEM score patients. NeoGEM showed consistently good discrimination across retrospective, external, and small prospective validation settings, suggesting potential utility for response stratification in patients with LAGC. The integration of body-composition radiomics may improve predictive performance and provide additional host-related biological context.
Journal IF-equivalent: 7.9 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-05; not the official Clarivate IF)Reference: Ding P, Guo H, Yang J, Chen S, Ma S, Yang S, et al. Multimodal prediction of neoadjuvant chemotherapy response in locally advanced gastric cancer: a multicenter cohort study. BMC Med. 2026 Nov 2 [Epub ahead of print]. doi:10.1186/s12916-026-05123-0.Checked: Abstract only