Abstract Background: Advanced gastric cancer (GC) responds poorly to systemic therapy, including Immune Checkpoint Blockade (ICB), yet how the tumor microenvironment (TME) drives resistance remains unknown. We assembled the largest spatial transcriptomic dataset in Immune Checkpoint Blockade (ICB)-treated GC, derived signatures encoding spatial composition and their cellular functions and phenotypes, and validated these as outcome predictors across bulk, single-cell, and independent spatial cohorts. Methods: Pre-treatment samples from advanced GC patients receiving ICB were profiled using spatial transcriptomics to derive spatial signatures that deconvolve immune, stromal, and tumor compartments and capture TME niches associated with ICB non-response. These spatial signatures were projected onto multiple external datasets, including four bulk RNA-seq cohorts, one single-cell RNA-seq cohort, two 10x Visium cohorts (27 GC samples), and one proteomic cohort, representing the largest integrated validation set for GC immunotherapy biomarkers to date. Results: Spatial immune signatures in non-responders showed Th2, regulatory Th17, and a homing-lymphocytes program marked by VLA-4 and CXCR4, confirmed in scRNA-seq. Stromal signatures in non-responders included inflammatory CAF programs, complement system activation (C1S, C3, A2M, IL6ST), growth-factor signaling, and MHC-II antigen-presenting states. When projected onto bulk RNA-seq and proteomic cohorts, these spatial immune and stromal modules were consistently enriched in clinical non-responders and associated with inferior survival. In 10x Visium data, non-responder stroma exhibited a strong complement system, aligned with ECM remodeling and myofibroblast properties, and showed reduced neighboring T-cell infiltration. Spatial bivariate analyses localized MHC genes and M2-like C1q+ macrophages to inflamed stromal niches. Additional non-responders, including an MSI-H case, retained Th2-dominant profiles, whereas responders revealed higher tumor-antigen expression and dendritic-cell activation. Conclusion: Spatially derived transcriptomic signatures that incorporate functional states of the GC immune microenvironment robustly distinguish ICB non-responders across bulk, single-cell, and independent spatial datasets. These data nominate the inflammatory stromal niches, C1q+ macrophage clusters, and skewed Th2 as actionable features of ICB resistance and potential therapeutic targets. Our results support incorporating spatial transcriptomics into pre-treatment assessment to refine patient selection, prioritize combination strategies that remodel the stromal-immune interface, and advance toward clinically deployable, GC-specific biomarkers for immunotherapy. AI was used for language editing only; authors are responsible for all content and approved the final version. Citation Format: Changjin Hong, Sunho Park, Jean R. Clemenceau, Minji Kim, Inyeop Jang, Seock-Jin Chung, Sung Hak Lee, Sam C. Wang, Tae Hyun Hwang. Spatially derived transcriptomic predictors of immune checkpoint blockade outcome in advance gastric cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 3957.

Importance:Incidence of early age-onset colorectal cancer (EOCRC) is increasing. Delays in initiation of definitive therapy are associated with worse outcomes in colorectal cancer (CRC), but their impact on EOCRC has not been comprehensively characterized. Objective:To evaluate incidence of EOCRC, identify patients affected by treatment delays, and determine targetable factors contributing to delayed therapy. Design, Setting, and Participants:This retrospective, population-based cross-sectional study analyzed data from patients diagnosed with CRC from January 1, 2004, to December 31, 2019, using the Texas Cancer Registry. Patients were classified as having either EOCRC (diagnosed age <50 years) or average age-onset colorectal cancer (AOCRC; diagnosed age ≥50 years). The data analysis was performed between August 2024 and November 2025. Main Outcomes and Measures:The main outcomes were EOCRC status and treatment delays, defined as more than 6 weeks from tissue diagnosis to initiation of definitive therapy. Overall survival (OS), prevalence, impact of treatment delays, and patient-level and system-level factors associated with delayed treatment were also assessed. Results:Among 112 672 patients with CRC (overall mean [SD] age, 65.4 [13.5] years; 61 570 [54.6%] male), 12 079 (11%) had EOCRC, and 100 593 (89%) had AOCRC. The cohort comprised 3111 Asian and Pacific Islander individuals (2.8%), 14 517 Black individuals (12.9%), 23 372 Hispanic individuals (20.7%), and 71 672 White individuals (63.6%). Mean (SD) age for the EOCRC cohort was younger (41.6 [5.9] years) compared to the AOCRC cohort (68.2 [11.2] years; P < .001). Compared to patients with AOCRC, patients with EOCRC were less likely to be of White race (6421 [53.2%] vs 65 251 [64.9%]; P < .001) and more likely to be of Hispanic ethnicity (3389 [28.1%] vs 19 983 [19.9%]; P < .001). Median OS for patients with EOCRC was not reached compared to patients with AOCRC at 80 months (hazard ratio [HR], 0.56; 95% CI, 0.56-0.60; P < .001). In multivariable analysis, higher Social Vulnerability Index (HR, 1.22; 95% CI, 1.19-1.26; P < .001) and treatment delays (HR, 1.29; 95% CI, 1.26-1.32; P < .001) were associated with worse OS. Median OS for patients with EOCRC was not reached in patients with or without treatment delay; however, it remained significant (HR, 1.35; 95% CI, 1.32-1.38; P < .001). After controlling for demographic and clinical factors, language barriers were associated with treatment delay in EOCRC (odds ratio, 1.45; 95% CI, 1.18-1.79; P < .001). Conclusions and Relevance:In this cross-sectional study, EOCRC was associated with improved OS compared with AOCRC; however, treatment delays were independently associated with worse survival among patients with EOCRC. Language barriers could be a potentially modifiable risk factor associated with delayed treatment and may provide an opportunity to improve timely care and outcomes in EOCRC.

Abstract Diffuse-type gastric adenocarcinoma (DGC) presents as the more invasive and aggressive gastric cancer subtype with poorer prognosis. Hereditary germline CDH1 mutations are known to drive DGC in 1-3% of cases in what is known as hereditary diffuse gastric cancer (HDGC). Despite the genomic characterization of HDGC, the mechanisms of onset are still poorly understood, given that evidence suggests it can bypass the classic cascade of gastric intestinal metaplasia (IM) to dysplasia to cancer. Additionally, HDGC presents a complex tumor microenvironment (TME) given by its highly infiltrative distribution with increased immune and stromal interactions, as well as IM and dysplastic marker gene expression. These factors present an opportunity to better understand the complex cellular dynamics to elucidate the etiology of this disease. New advances in spatial transcriptomics and fluorescence imaging have reduced per-sample costs of spatial biology assays, allowing the scale necessary for studying serial sections. HDGC samples were collected as formalin-fixed, paraffin embedded blocks. Samples were serially sectioned at 5µm thickness into G4x gel pads. Regions of interest (10mm x 10mm) were isolated from the gels pads and transferred to a G4x X2 spatial flow cell. Samples were processed using Singular Genomics G4x spatial multi-omic assay with a custom pre-gastric cancer panel consisting of 16 proteins, 341 transcripts, and fluorescent-based H&E images for every section of tissue. Data was processed following quality control, implementing cell type annotation, followed by registration of cell coordinates, incorporation of histopathological tissue annotations by a board-certified pathologist, and cell neighborhood analysis. Two models were produced with 7 and 9 serial sections (16 total) representing a tissue depth of 35 µm and 45 µm and populations of 1.6M and 3.4M cells, respectively. The models recapitulate 3D morphology of known tissue structures, such as non-neoplastic epithelial glands, vasculature, and tertiary lymphoid structures. We observed TFF2, a SPEM cell IM marker, expression in the tumor-adjacent gastric mucosa. We also found small clusters of TFF2+ cells within the superficial tumor-invasive area, only present in a few of the total tissue layers. We successfully built two 3D spatial multi-omic, subcellular resolution models for CDH1-mutant HDGC representing 35-45µm of tissue thickness that recapitulate the samples’ histological structures. We show the spatial distribution of metaplastic markers expressed in tumor-adjacent epithelium. Our 3D models allowed for the identification of rare-cell events, such as small TFF2+ cell clusters within tumor-infiltrated tissue regions. These results show the promise that high-resolution 3D models present for improving our understanding of complex TMEs, such as HDGC. Citation Format: Jean R. Clemenceau, Yunhe Liu, Idania Carolina Lubo Julio, Soyoung Im, Seock-Jin Chung, Sam C. Wang, Paul F. Mansfield, Luisa Maren Solis Soto, Linghua Wang, Tae Hyun Hwang. Subcellular 3D multi-omic models of CDH1-mutant diffuse gastric cancer improve recapitulation of tumor microenvironment structure and reveal precancer signature niches [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6890.

4073 Background: Germline pathogenic or likely pathogenic (P/LP) CDH1 variants increase the lifetime risk of developing diffuse gastric cancer (DGC). Given that the epidemiology and prognosis of gastric cancer vary based on patient ancestry, and Hispanic and Black/African-American patients have been underrepresented in clinical and translational research, we sought to determine the prevalence of CDH1 variants in a diverse population of gastric cancer patients. Methods: Germline whole exome sequencing (WES) data from gastric cancer patients were obtained as part of commercial ordering of personalized circulating tumor DNA testing (Signatera, Natera, Inc.) from March 2024 to June 2025. Ancestry was inferred by EthSeq and classified into predefined groups: European (EUR), admixed American (AMR), African (AFR), and East Asian (EAS). Germline single-nucleotide variants and short insertions/deletions were identified and annotated with Variant Effect Predictor (v.105), and the pathogenicity of variants in CDH1 was determined using classifications available in the ClinVar database. The prevalence of P/LP variants and variants of uncertain significance (VUS) were compared across ancestries. Potential pathogenicity of VUS was evaluated using in silico predictors SIFT (v5.2.2), Polyphen (v2.2.2), and AlphaMissense (hg19). Results: Among 2,567 patients, 55% (1,414/2,567) were male and the median age was 67 years. Ancestry inference identified 47% (n=1,201) as EUR, 22% (n=575) as AMR, 18% (n=457) as AFR, and 13% (n=334) as EAS. Patients with AMR ancestry were younger compared to EUR ancestry patients (median 62 years vs 69 years; adj p<0.0001), and less likely to be male (51% [295/575] vs. 58% [693/1201]; OR, 0.7; 95% CI, 0.63–0.94; p=0.015). No significant differences in age or sex were observed in patients with AFR (median 68 years, 55% male) or EAS (median 68 years, 52% male) ancestry compared to EUR. Overall, 0.7% (n=17) gastric cancer patients had a P/LP variant, and there was no difference across ancestries (0.8% of EUR, 0.5% of AMR, 0.4% of AFR, and 0.6% of EAS patients). We found that 1.1% (n=27) of the cohort had a CDH1 VUS, the rate for which did not vary across ancestries (1.2% of EUR, 0.52% of AMR, 0.84% of AFR, 0.60 of EAS). Of the 21 unique VUS, 33% (n=7) were each predicted to be pathogenic by all three in silico predictors. One unique VUS was observed in two unrelated patients: NM_004360.5:c.635G>T (p.Gly212Val). The predicted-pathogenic VUS were similarly distributed across ancestries. Conclusions: This real-world analysis highlights that CDH1 VUS were similarly distributed across ancestries. Notably, we found that one-third of VUS identified were predicted to have pathogenic function across three in silico prediction tools. This highlights the need for functional assays to determine the variant biology to improve gastric cancer risk assessment.

BACKGROUND:The increasing adoption of incomplete cholecystectomy in severe cholecystitis has created a growing population of patients who later develop recurrent symptoms from remnant gallbladder or cystic duct pathology. Reoperative cholecystectomy is definitive but technically challenging, and the role of robotic surgery in this setting remains poorly defined. METHODS:A retrospective cohort study was performed of adult patients undergoing robot-assisted completion or remnant cholecystectomy for recurrent gallbladder disease at a single tertiary institution between 2022 and 2024. Perioperative outcomes were evaluated, and institutional primary cholecystectomy outcomes were used as a contextual comparator. RESULTS:Twenty-one patients underwent robotic reoperative cholecystectomy. The median age was 41 years, and most patients were female and obese. All had radiographic evidence of remnant biliary pathology. The median operative time was 166 min, with no intraoperative complications or conversions to open surgery. The median postoperative length of stay was 0 days, with most patients discharged the same day. One patient experienced a 30-day major complication managed nonoperatively. There were no 90-day readmissions or mortalities, and all patients reported symptom resolution. DISCUSSION:In this single-institution series, robotic reoperative cholecystectomy was safe and effective, with minimal morbidity, very short hospital stays, and universal symptom resolution. When performed at experienced centers, robotic reoperative cholecystectomy represents an effective definitive strategy for recurrent gallbladder disease after incomplete cholecystectomy.

Abstract Background: The progress of spatial multimodal platforms has enabled high-resolution mapping of various types of molecular and morphological characteristics within a tissue. Proteins influence cellular phenotypes and the organization of the tumor microenvironment, with their abundance and patterns highly correlated in adjacent serial sections. Spatial Multimodal platforms such as Singular Genomics G4X, which capture H&E-stained images, multiplexed protein expression, and targeted RNA transcriptomics from the same tissue slide, now create a foundation for three-dimensional (3D) spatial multimodal profiling and 3D tumor microenvironment modeling. However, comprehensive multiplex protein imaging across entire tissue stacks is costly, labor-intensive, and impractical, leading to sparse proteomic sampling that limits accurate 3D reconstruction. Methods: We developed ProteoBridge, a deep learning-based framework that predicts protein expression patterns across serial tissue sections to densify the proteomic information needed for 3D molecular modeling. For each tissue stack, the model was trained on the first section, which contains H&E, multiplex protein images, and RNA transcriptomics. The inputs include H&E tiles and RNA transcriptome data, while the outputs are multi-channel protein maps for the same field of view. The combination of targeted RNA transcriptomics and H&E stained morphology helps define cell types and states, allowing the model to learn morphology-to-protein relationships based on transcriptional programs. The training focused on reducing mean absolute error across protein channels to establish histology-protein correspondences for unmeasured sections. Results: We evaluated ProteoBridge on tissue stacks with multiple profiled sections. Image-level similarity between predicted and measured protein maps was assessed. Using single-slide supervision, ProteoBridge accurately reproduced marker intensities and preserved cross-plane consistency, capturing both intensity values and spatial organization of proteins.. By filling in skipped planes, ProteoBridge generates a more continuous 3D representation of the tumor proteome suitable for downstream 3D molecular modeling and visualization. Conclusion: ProteoBridge requires only routine H&E staining on the remaining sections in a stack to infer protein intensities across serial planes, reducing cost and turnaround time while expanding effective proteomic coverage. Using the G4X platform's multimodal data and extending protein predictions to unmeasured areas allows for cost-effective 3D reconstruction of the tumor proteome, enhancing molecular modeling of the tumor microenvironment and its heterogeneity.Generative AI assisted only with language editing of this abstract. The authors retain sole responsibility for the scientific content and conclusions, having reviewed and approved the final version. Citation Format: Minji Kim, Sunho Park, Seock-Jin Chung, Inyeop Jang, Jean R. Clemenceau, Soyoung Im, Eric Sha, Hwanil Choi, Soonyoung Lee, Jongseong Jang, Sam C. Wang, Tae Hyun Hwang. ProteoBridge: Bridging skipped sections via histology-based protein prediction [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 4218.

ABSTRACT The role of ARID1A in cancer immune evasion remains uncertain, with prior studies reaching opposing conclusions. In addition, previous work has shown that the role of ARID1A in cell-autonomous tumorigenesis is context-dependent. Using isogenic murine gastric cancer models, we found that in vivo Arid1a loss in an autochthonous genetically engineered mouse model of gastric cancer conferred T cell-dependent immune evasion, while in vitro deletion did not. Mechanistically, tumor Arid1a loss reprogrammed the tumor microenvironment into an immune desert through suppression of GM-CSF secretion and interferon-γ responsiveness. These changes were not observed when Arid1a was deleted in vitro . In human gastric cancer, an immune-cold phenotype was restricted to ARID1A mutants in the genomically stable subtype, while ARID1A loss in the chromosomal instability subtype was associated with variable immune profiles. These results demonstrate that tumor ARID1A loss does not intrinsically confer pro- or anti-tumor immune properties and instead is determined by tissue context.

BACKGROUND:Guidelines recommend patient-facing material be written at or below a 6th- to 9th-grade reading level; however, surgical consent forms frequently exceed this standard. Artificial intelligence chatbots offer accessible tools to enhance readability, but their performance for non-English material remains unclear. STUDY DESIGN:We evaluated 3 strategies to improve consent readability. First, ChatGPT-4.0 was prompted to simplify preexisting surgical consent forms in English and Spanish. Second, a custom generative pretrained transformer (GPT) was trained using these outputs and tested on a separate validation cohort. Finally, both GPT's were prompted to translate English forms into Spanish, targeting a 6th-grade reading level. Readability was measured by the Fry Readability Formula, Simple Measure of Gobbledygook (SMOG) Index, Flesch-Kincaid Grade Level for English, and the Gilliam-Peña-Mountain (GPM) and SOL Readability Formulas for Spanish. RESULTS:ChatGPT-4.0 modestly improved English readability, with median Fry decreasing from 12.0 to 10.0, SMOG 12.3 to 11.7, and Flesch-Kincaid 9.2 to 8.6 (all p < 0.05). Spanish readability did not significantly improve (median GPM 9.5 to 10.0; SOL 9.6 to 9.1; both p > 0.05).Our custom GPT improved English readability, reducing median Fry from 12.0 to 7.0, SMOG 12.3 to 9.3, and Flesch-Kincaid 9.6 to 5.9 (all p < 0.001). Spanish readability gains were modest with median SOL decreasing from 9.5 to 7.3 (p < 0.01). GPM was unchanged (8.0, p > 0.99). English-to-Spanish translation outperformed direct Spanish simplification. ChatGPT-4.0 achieved median GPM of 8.0 (p = 0.25) and SOL of 7.6 (p < 0.001, compared with original Spanish consents). Translation via the custom GPT achieved the largest improvement (median GPM 7.0 and SOL 6.8; both p < 0.05, compared with original Spanish consents). CONCLUSIONS:Tailored artificial intelligence chatbots can improve readability of patient-facing material; however, diverse strategies are necessary to adapt existing models for multilingual populations.

e16086 Background: HER2 overexpression is present in 20 to 30% of advanced upper gastrointestinal (UGI) cancers and can be targeted in the first-line setting by combining chemotherapy with trastuzumab +/- pembrolizumab. However, the persistence of HER2 overexpression in progressive cases is unknown. Despite this, HER2 overexpression is often targeted in the second-line or more-advanced settings with trastuzumab-deruxtecan based on the initial pathology. Small international series (n = 48 and n = 7) demonstrated that up to 43% of UGI cancer patients lose HER2 overexpression in progressive tissue. We sought to characterize HER2 overexpression loss in subsequent biopsies in a US-based population. Methods: This study queried the US-based, electronic health record-derived, deidentified Flatiron Health Research Database from January 2011 to June 2025, which has longitudinal patient-level data from approximately 280 cancer clinics. Eligible patients had HER2 assessed prior to initiation of first-line therapy and a second assessment at least 7 weeks or later after first biopsy. HER2 positivity was defined by the ASCO/CAP guidelines. Demographics and baseline clinical and laboratory variables were extracted. We describe proportions with 95% confidence intervals (CI) using the Clopper-Pearson method. We used univariate logistic regressions to estimate odds ratios (ORs) for the binary outcome of loss of HER2 overexpression. Subgroup analyses were stratified by baseline HER2 status (IHC 3+ vs IHC 2+ with fluorescence in situ hybridization [FISH] amplification). Results: A total of 17,305 individuals were assessed for inclusion. In total, 696 patients had at least one HER2 assessment at ≥7 weeks from the initial assessment and 333 patients met criteria for inclusion. Of these, 83 patients (25%) were classified as HER2+ at baseline. HER2+ were more frequently male (73%), former or active smokers (70%), and were more likely to receive HER2-directed therapies in the front line (59%). The initial biopsy site was commonly the primary site (88%), as was the site of second biopsy (67%). Overall, HER2 overexpression was lost on subsequent biopsy 36% (95% CI 25.9-47.4%) of the time. Regardless of therapy received, loss of HER2 overexpression occurred in 83% (51.6-97.9%) of patients with baseline IHC2+/FISH amplified compared to 28% (18.1-40.1%) in patients with baseline IHC3+ (OR 12.8, 95% CI 2.6-63.4, p = 0.002). No other clinical or laboratory factors were associated with loss of HER2 overexpression on logistic regression. Conclusions: Loss of HER2 overexpression is a common occurrence in UGI cancers, particularly amongst patients with baseline HER2 IHC2+/FISH amplification. This is possibly due to clonal pressure from targeting HER2, tumoral heterogeneity, or other unknown mechanisms. Greater consideration should be given for repeat biopsies and biomarker assessments in these patients prior to targeting HER2 in the progressive setting.

Supplementary Figure 1. ACTA2 expression was elevated in the molecular subtype with the worst prognosis (Group 4). Supplementary Figure 2. ACTA2 expression stratified by microsatellite stability status in the pooled cohort. Supplementary Figure 3. Alternatively tested thresholds to stratify patients into ACTA2-High and ACTA2-Low groups.Supplementary Figure 4. ACTA2 expression stratified by microsatellite stability status in the immune checkpoint inhibitor cohort.

Determining tumor microsatellite status has significant clinical value because tumors that are microsatellite instability-high (MSI-H) or mismatch repair deficient (dMMR) respond well to immune checkpoint inhibitors (ICIs) and oftentimes not to chemotherapeutics. We propose MSI-SEER, a deep Gaussian process-based Bayesian model that analyzes H&E whole-slide images in weakly-supervised-learning to predict microsatellite status in gastric and colorectal cancers. We performed extensive validation using multiple large datasets comprised of patients from diverse racial backgrounds. MSI-SEER achieved state-of-the-art performance with MSI prediction by integrating uncertainty prediction. We achieved high accuracy for predicting ICI responsiveness by combining tumor MSI status with stroma-to-tumor ratio. Finally, MSI-SEER's tile-level predictions revealed novel insights into the role of spatial distribution of MSI-H regions in the tumor microenvironment and ICI response.

OBJECTIVES:There is an urgent need for effective second line treatments for advanced upper gastrointestinal (UGI) cancers. Ramucirumab and paclitaxel (Ram-Pac) has long been the standard therapy; however, this regimen is often complicated by cumulative neuropathy. FOLFIRI-Ramucirumab (FOLFIRI-Ram) may be an alternative second line option but real-world data to support its use are limited. METHODS:The deidentified Flatiron Health Research Database was queried for patients treated for unresectable or metastatic UGI cancers with second line Ram-Pac or FOLFIRI-Ram from January 2011 to June 2024. Study cohorts were propensity score matched 1:6 (FOLFIRI-Ram:Ram-Pac) from key clinical and laboratory characteristics. The endpoints of interest were overall survival (OS) and real-world time to treatment discontinuation (rwTTD). RESULTS:Of 15,908 patients with UGI cancer, 631 received second line Ram-Pac and 40 received second line FOLFIRI-Ram. After matching, 40 FOLFIRI-Ram and 240 Ram-Pac patients were included. Median OS was 9.7 months with FOLFIRI-Ram (95% CI: 6.9-12.3) and 7.7 months with Ram-Pac (95% CI: 6.2-8.8), with a hazard ratio (HR) for death of 0.74 with FOLFIRI-Ram versus Ram-Pac (95% CI: 0.50-1.11, P =0.14). The median rwTTD with FOLFIRI-Ram was 5.2 months (95% CI: 4.1-6.2), compared with 3.7 months with Ram-Pac (95% CI: 3.2-4.3), HR for treatment discontinuation =0.70 (95% CI: 0.48-1.00, P =0.048). CONCLUSIONS:In a real-world propensity-score matched analysis, no survival difference was noted with the combination of FOLFIRI-Ram compared with Ram-Pac; however, FOLFIRI-Ram was associated with a significantly longer rwTTD. Altogether, these data suggest FOLFIRI-Ram is a potential alternative for second line treatment of UGI cancers.

4058 Background: Given historically poor outcomes for advanced upper gastrointestinal (UGI) cancers, there is an urgent need for effective 2L treatments. Since the phase III RAINBOW trial, combination ramucirumab and paclitaxel (Ram-Pac) has filled this role; however, this regimen is plagued by dose-limiting toxicities, specifically neuropathy. The phase II RAMIRIS trial demonstrated the efficacy and tolerability of FOLFIRI-Ram as an alternative 2L, even if it did not improve survival over Ram-Pac. Real-world data to support its use, however, are lacking. Methods: The nationwide Flatiron Health electronic health record-derived de-identified database, which includes treatment data from around 280 cancer clinics across the United States, was queried for patients treated for unresectable or metastatic UGI cancers with 2L Ram-Pac or FOLFIRI-Ram from January 2011-June 2024. Demographics and lab values at time of treatment were extracted. Study cohorts were derived using a greedy match based on a logit model to predict propensity scores from key clinical and laboratory characteristics; patients were matched 1:6 (FOLFIRI-Ram:Ram-Pac) given expected imbalances in sample size. The endpoints of interest were overall survival (OS) and real-world time to treatment discontinuation (rwTTD), determined via Kaplan-Meier method, log-rank test, and Cox proportional hazards model. A hybrid approach was used to construct a multivariate Cox model. Results: Of 15,908 UGI cancer patients identified, 631 received 2L Ram-Pac and 40 received 2L FOLFIRI-Ram. After matching, 40 FOLFIRI-Ram and 240 Ram-Pac patients were included. Median OS from initiation of 2L therapy was 9.7 months with FOLFIRI-Ram (95% CI 6.9-12.3) and 7.7 months with Ram-Pac (95% CI 6.2-8.8), with a hazard ratio (HR) for death of 0.74 with FOLFIRI-Ram versus Ram-Pac (95% CI 0.50-1.11, p = 0.14). Similar results were seen in the multivariate model (HR 0.72, 95% CI 0.49-1.08, p = 0.114) after adjustment for albumin, neutrophil:lymphocyte ratio, and alkaline phosphatase. The median rwTTD with FOLFIRI-Ram was 5.2 months (95% CI 4.1-6.2), compared to 3.7 months with Ram-Pac (95% CI 3.2-4.3). The HR for treatment discontinuation was 0.70 with FOLFIRI-Ram versus Ram-Pac (95% CI 0.48-1.00, p = 0.048). The reduced hazard for treatment discontinuation with FOLFIRI-Ram persisted in the multivariate model (HR 0.67, 95% CI 0.46-0.97, p = 0.033) after adjustment for ECOG status, history of prior surgery, PDL1, albumin, and neutrophil:lymphocyte ratio. Conclusions: In a real-world propensity-score matched analysis, no survival difference was noted with the combination of FOLFIRI-Ram compared to Ram-Pac, however FOLFIRI-Ram was associated with a significantly longer rwTTD. Altogether, these data suggest FOLFIRI-Ram is a viable and tolerable alternative for 2L treatment of UGI cancers.

BACKGROUND AND OBJECTIVES:The presence of lymph node metastases in patients with gastric and gastroesophageal junction (GEJ) adenocarcinoma provides prognostic information and guides treatment decisions. We sought to determine the sensitivity of computed tomography (CT) imaging for clinical nodal staging in patients with resectable gastric and GEJ adenocarcinoma and determine a lymph node size cut-off to optimize diagnostic accuracy. METHODS:We performed a retrospective review of patients who underwent curative-intent resection for gastric or GEJ adenocarcinoma at our institution between 2010 and 2023. We reviewed CT scan images performed immediately before resection and measured lymph nodes in the short axis to identify patients with lymph nodes larger than the radiologic upper limit of normal. We compared histopathologic data from resection specimens to CT scans to determine pathologic concordance for metastatic involvement of lymph nodes and calculated the sensitivity and specificity of CT scans to identify nodal metastases. We used the largest lymph node measurement from each scan to construct a receiver operating characteristic (ROC) curve and calculated Youden's J Index to determine the optimal lymph node size cut-off. RESULTS:We identified 192 consecutive patients who underwent resection during the study period and had preoperative CT scans available for review. 72 patients (38%) had diffuse or mixed type tumors, and 85 patients (44%) had intestinal-type tumors. 157 patients (82%) underwent neoadjuvant chemotherapy or chemoradiation. 110 patients (57%) had pathologic node-positive disease and in this cohort, 27 patients (25%) had lymph nodes deemed radiographically enlarged. The sensitivity of preoperative CT scans for nodal metastases was 25%, and specificity was 83%. Based on the ROC curve, an optimal lymph node size cutoff of 6.5 mm was identified. At this cutoff, the estimated sensitivity was 47%, and the estimated specificity was 72%. When patients were stratified by Lauren histology, the AUC for intestinal-type tumors was significantly better than for diffuse or mixed-type tumors (p = 0.02). The area under the ROC curve for patients with diffuse or mixed type tumors was 0.51 indicating lymph node size on CT scan was no better than random chance for diagnosis of lymph node metastases. CONCLUSIONS:CT scans are not sensitive to identify nodal metastases in gastric and GEJ adenocarcinoma using current radiologic guidelines. While a lower lymph node size cutoff may improve sensitivity, this does not benefit patients with diffuse or mixed-type tumors. Since CT scans understage a large proportion of patients with gastric and GEJ cancers, techniques to improve clinical nodal staging in this population are needed.

Delay in gastric cancer diagnosis is associated with inferior outcomes. The effects of pre-existing mental health disorders (MHDs) on delays in gastric cancer diagnosis and treatment disparities are not well-understood. In this study, we evaluated the impact of MHDs on time to gastric cancer diagnosis and receipt of guideline-concordant treatment. We performed a retrospective review of patients diagnosed with gastric adenocarcinoma between 2015 and 2022. Patients with pre-existing diagnoses of mood, affective, and substance use disorders were classified as having an MHD. Univariable and multivariable regression were used to analyze the association between MHDs and delay in diagnosis. The association between MHD and receipt of guideline-concordant care was also evaluated. Overall, 460 patients diagnosed with gastric cancer were included in the analytic group. Seventy patients (15

Background: Compared to open pancreaticoduodenectomies (OPD), the robotic (RPD) approach decreases the rate of complication and the length of stay (LOS). However, it remains unknown if these benefits persist in octogenarians, who are at higher risk for perioperative morbidity and mortality. Methods: A retrospective analysis of the ACS-NSQIP database was performed to identify patients aged 80 years or older who underwent PD for pancreatic adenocarcinoma between 2015-2021. Patients who underwent RPD or OPD were compared using inversed probability weighting of the propensity score. readmission, return to the operating room, mortality, and clinically relevant postoperative pancreatic fistula. Results: Of 30,751 patients, 1720 were octogenarians. One thousand six hundred twenty-five patients (94 %) underwent OPD, and 95 (6 %) underwent RPD. RPD was significantly associated with a reduced incidence of major complications (32.6 % vs. 45.6 %; p <0.01) and a lower rate of non-home discharge (24.7 % vs. 34.3%; p < 0.05). However, RPD was associated with a longer operative time (438 min vs. 342 min; p < 0.0001). There was no difference in other assessed outcomes. Conclusion: RPD may reduce major postoperative complications and non-home discharges compared to the open approach for octogenarians.

Introduction Human tissue samples are essential for translational cancer research. However, less than 20% of current biobank and genomic samples were obtained from minority patients, which may lead to biased understanding of cancer biology. The objective of this study was to identify factors associated with patient enrollment in our institution's gastric cancer biobank. Methods Patients with suspected or confirmed gastric or gastroesophageal junction cancer undergoing surgical procedures at our institution were invited to enroll in a prospective gastric cancer biobank. We retrospectively reviewed patients who were invited to enroll from 2017 to 2023 at our safety-net and university hospitals. We compared patients who enrolled to those who declined to identify factors that predict enrollment. Results Hispanic patients had similar odds of enrollment as non-Hispanic White patients (odds ratio (OR): 1.22, 95% confidence interval (CI): 0.54-2.73, P = 0.63). Non-Hispanic minorities (Black/African Americans and Asians) were less likely to enroll when compared to non-Hispanic Whites (OR: 0.41, 95% CI: 0.18-0.95, P = 0.04). Minority patients treated at our safety-net hospital had higher odds of enrollment than those treated at our university hospital (OR: 2.62, 95% CI: 1.11-5.99, P = 0.02). Conclusions Efforts to improve diversity in biomedical research cannot consider minority patients as a monolithic cohort. Instead, targeted interventions that address diverse cultural concerns and improve access to enrollment at safety-net centers are requisite.

Supplementary Table 1. Univariable associations of the multivariable Cox proportional hazards model in the Yonsei cohort.Supplementary Table 2. Clinical characteristics of patients in the pooled validation cohort.Supplementary Table 3. Univariable associations of the multivariable Cox proportional hazards model of the pooled validation cohort.Supplementary Table 4. Univariable Cox proportional hazards regression of the association of individual genes of the 32-gene signature and their association with overall survival in the Yonsei cohort.Supplementary Table 5. Univariable Cox proportional hazards regression of the association of individual genes of the 32-gene signature and their association with overall survival in the pooled cohort.Supplementary Table 6. Multivariable analysis of the pooled cohort including microsatellite stability status.Supplementary Table 7. Multivariable analysis of the association of ACTA2 expression with recurrence-free survival in the pooled cohort.Supplementary Table 8. Clinical characteristics of patients treated with immune checkpoint inhibitors.Supplementary Table 9. ACTA2 expression in cells analyzed by single nuclear RNA sequencing cohort.Supplementary Table 10. Characteristics of patients analzyed by digital spatial transcriptomics.

Gastric cancer is a sexually dimorphic disease. Male patients have twice the incidence, are diagnosed at a younger age, and have worse survival outcomes than females. Whether there are molecular mechanisms that underlie these sex-based differences is unclear. We generated genetically engineered mice on C57BL/6 backgrounds with parietal cell-specific conditional deletions of Trp53 and Cdh1, two tumor suppressors commonly lost in gastric cancer (Atp4b-Cre; Trp53 flox/flox ; Cdh1 flox/flox , “APC” mice). APC mice developed gastric cancer with 100% penetrance and median overall survival was significantly longer in females (50.6 weeks, N = 20) than males (41.7 weeks, N = 23; p < 0.001). At the time of death, serum were collected and circulating cytokines were quantified using multiplex ELISA. Females had significantly higher levels of circulating IFN-γ (60.8 pg/mL vs. 10.9 pg/mL; p = 0.02) and CCL4 (257 pg/mL vs. 118.3 pg/mL; p = 0.001), suggesting enhanced cytotoxic lymphocyte and macrophage activity. Bulk RNA sequencing of tumors revealed significant enrichment of Hallmark Gene Sets associated with type I and II interferon responses and antigen presentation pathways in females. These findings suggested an immune-mediated basis for the survival difference seen between male and female APC mice. To further explore the immune microenvironment, we derived cell lines from APC tumors and performed flank injections into C57BL/6 mice. Tumor volumes at 6 weeks were not significantly different between sexes (645 mm2 in females vs. 951 mm2 in males; p = 0.35). However, flow cytometry revealed marked immune differences in total CD8+ T cells (2.4% in females vs. 0.63% in males vs.; p = 0.05), activated CD8+ T cells (23.0% vs. 9.2%; p < 0.001), dendritic cells (2.6% vs. 1.3% of myeloid cells; p = 0.05), and a trend toward increased M1 macrophages in females (1.6% vs. 0.9%; p = 0.06). We next evaluated the role of sex hormones in mediating sex-associated differences in gastric cancer. Females underwent oophorectomies (F-O) or sham operations (F-S), while males underwent orchiectomies (M-O) or sham operations (M-S). Median overall survival was significantly longer in F-S mice (47.1 weeks, N = 7) compared to F-O mice (44.1 weeks, N = 6; p=0.003). In contrast, median overall survival did not differ in M-S (44.3 weeks, N = 7) than M-O (44.7 weeks, N = 4; p=0.8). In conclusion, we demonstrate sex-based differences in survival in an autochthonous genetically engineered mouse model of gastric cancer, consistent with clinical observations that female human patients have improved overall survival than female patients. Our findings implicate the immune system as potential mediators of these differences. Preliminary data suggests hormone deprivation impacts overall survival of female mice, but not male mice. Future work is directed at identifying the molecular mechanism between female sex, sex hormones, and improved overall survival. Ryan T. Heslin, Shu Xiao, Morgan F. Pettigrew, Nafeesah Fatimah, Hsien-Tsung Lai, Suntrea T.G. Hammer, Sam C. Wang. Unraveling the sexually dimorphic immune microenvironment in gastric cancer [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Mechanisms of Cancer Immunity and Cancer-related Autoimmunity; 2025 Sep 24-27; Montreal, QC, Canada. Philadelphia (PA): AACR; Cancer Immunol Res 2025;13(9 Suppl):Abstract nr PR-14.