Citations

e20761 Background: The prognostic role of serial circulating tumor DNA (ctDNA) dynamics during osimertinib treatment has been described mainly in Asian and European populations; however, it remains poorly defined in Hispanic patients with advanced EGFR-mutant non–small cell lung cancer (NSCLC). Our aim was to determine whether serial plasma EGFR testing identifies patients at high risk of early mortality. Methods: This multicenter, retrospective study included 64 Hispanic patients with stage IIIB–IV EGFR-mutant NSCLC treated with first-line osimertinib between 2018 and 2023. Tumor tissue genotyping was performed using hybrid-capture next-generation sequencing. Plasma EGFR mutations were assessed at baseline, week 8, and week 24. Overall survival (OS) and progression-free survival (PFS) were analyzed using Cox models adjusted for p53 status, RAF-pathway alterations, and metastatic burden. Results: The cohort had a median age of 61 years, with 65.6% of patients being female and 56.2% being never-smokers. Sensitizing EGFR mutations consisted mainly of exon 19 deletion (59.4%) and L858R (39.1%). Relevant co-mutations were identified in TP53 (29.7%) and RAF-pathway genes (6.2%). Plasma ctDNA remained positive in 42.2% of patients at week 8 and 22.0% at week 24. Week-8 ctDNA persistence was strongly associated with an inferior OS (HR 9.36, 95% CI 3.80–23.03, p < 0.001) after multivariable adjustment. Week-24 positivity also predicted worse OS (HR 3.34, 95% CI 1.13–9.89, p = 0.02). ctDNA status at week 8 or 24 was not significantly associated with PFS. Patients with TP53 or RAF pathway mutations demonstrated higher rates of molecular persistence. Conclusions: In Hispanic patients receiving Osimertinib, ctDNA positivity at 8-24 weeks is able to identify patients with poor survival outcomes. ctDNA monitoring may provide an early, non-invasive biomarker to guide treatment intensification and trial enrollment in EGFR-mutant NSCLC.

e17134 Background: Despite being the most prevalent malignancy among men, prostate cancer (PC) lacks FDA-approved primary prevention therapies. Current options, such as 5-alpha reductase inhibitors (5-ARIs), offer only modest risk reduction—approximately 23–25%—as shown in the PCPT and REDUCE trials. Emerging preclinical data indicate that GLP-1 receptor agonists (GLP-1RAs) may offer a novel alternative by inhibiting oncogenic pathways and inducing apoptosis in PC cell lines. To investigate this potential, we conducted the first real-world comparative analysis of PC incidence among high-risk men receiving GLP-1RAs versus 5-ARIs. Methods: Using the TriNetX global collaborative network (150M patients), we identified men with elevated PC risk (genetic risk, family history, or high-risk demographics/comorbidities). GLP-1RA users (cohort A) were propensity-score-matched 1:1 with 5-ARI users (cohort B) on demographics, metabolic comorbidities, baseline PSA, BMI and HbA1c. A minimum of 2 month’s prescription was required. The primary endpoint was incident PC. The secondary endpoint was adverse events related to GLP-1RA and 5-ARI. Patients with outcomes before the study inclusion were excluded. Sensitivity and subgroup analyses were performed, adjusting for screening frequency and baseline BMI. Kaplan–Meier curves assessed cumulative incidence, Cox models estimated hazard ratios (HRs, 95% CI), and logistic regression analyzed binary outcomes. Results: After matching, 118,904 high-risk men were identified (59,452 patients per cohort). Cohorts had comparable demographics with a mean age of 48 years, 77% White, 9% Black, and 2% Asian. GLP-1RA users had higher mean A1C (6.3% vs 5.7%) and BMI (38 vs 31), while 5-ARI users had higher mean PSA (3.1 vs 1.3). Median follow-up was 7 years for both cohorts. PC incidence was 2.10% (1,245/59,189) in GLP-1RAs users versus 2.85% (1,689/59,164) in 5-ARI users (NNT 133). GLP-1RAs were associated with a 20% reduction in PC risk (HR: 0.803; CI: 0.746–0.864). Safety outcomes showed that GLP-1RAs were associated with an increased risk of erectile dysfunction and nausea/vomiting. In contrast, GLP-1RAs were associated with lower risks of abdominal pain and constipation. No significant associations were observed for gynecomastia or diarrhea. Conclusions: In this large-scale, head-to-head comparison, GLP-1RAs demonstrated superior efficacy over 5-ARIs, providing a 20% relative risk reduction in PC incidence among high-risk men. Given the global surge in GLP-1RA utilization, these findings suggest a profound public health opportunity: the ability to achieve systemic oncologic protection through existing metabolic therapies. If validated by prospective trials, GLP-1RAs could transition from metabolic blockbusters to a cornerstone of precision oncology and primary cancer prevention.

e16488 Background: Small bowel adenocarcinoma (SBA) and ampullary adenocarcinoma (ampullary AC) are rare gastrointestinal (GI) malignancies. Treatment strategies are largely extrapolated from colorectal cancer, where oxaliplatin-based regimens are the most common first-line therapy in advanced disease. Although bevacizumab is considered safe in metastatic SBA, its efficacy remains unproven, and evidence is limited to small retrospective studies. We assessed the association of adding bevacizumab to FOLFOX with survival and toxicity in a large real-world cohort. Methods: We conducted a retrospective cohort study using the TriNetX Global Collaborative Network (January 2000-December 2025), including data from 105 healthcare organizations. Adults (≥18 years) with SBA or ampullary AC treated with FOLFOX with or without bevacizumab were identified. Cohorts were propensity score-matched on age, sex, race, comorbidities, and metastatic sites. Outcomes included overall mortality, GI perforation, bleeding, venous thromboembolism (VTE), hospital admission, and emergency department (ED) visits. Kaplan-Meier and Cox proportional hazards models were used for time-to-event analyses. Results: After matching, 818 patients were included in each cohort. In the FOLFOX + bevacizumab cohort, mean age was 66.1 ± 12.0 years and 53.8% were male; 63.7% were White, 13.8% Black, 5.6% Asian, 12.0% Unknown/Other, and 4.4% Hispanic/Latino. Over a median follow-up of 13 months, FOLFOX + bevacizumab was associated with higher overall mortality compared with FOLFOX alone (55.3% vs 45.2%; HR 1.24 [95% CI 1.1-1.4]; p = 0.002) and shorter median OS (15.1 vs 19.5 months). Rates of GI perforation (15.5% vs 16.7%; HR 0.90 [95% CI 0.71-1.15]; p = 0.39), GI bleeding (13.1% vs 13.6%; HR 0.96 [95% CI 0.73-1.25]; p = 0.74), and VTE (19.1% vs 21.4%; HR 0.88 [95% CI 0.71-1.09]; p = 0.23) were similar between groups. Hospital admissions were higher in the FOLFOX + bevacizumab cohort (2.8% vs 1.3%; HR 2.1 [95% CI 1.0-4.3]; p = 0.04). Findings were consistent at 5-year follow-up. Conclusions: In this real-world cohort, the addition of bevacizumab to FOLFOX was not associated with improved survival and was associated with higher healthcare utilization. Prospective, disease-specific studies are needed to define its role in this population. Outcomes at median 13-month follow-up after propensity matching (FOLFOX + bevacizumab vs FOLFOX). Outcome % vs % HR p Overall mortality 55.3 vs 45.2 1.24 0.002 Hospital admission 2.8 vs 1.3 2.1 0.04 ED visit 45.2 vs 42.4 1.02 0.75 GI perforation 15.5 vs 16.7 0.90 0.39 GI bleeding 13.1 vs 13.6 0.96 0.74 VTE 19.1 vs 21.4 0.88 0.23

e16449 Background: Pancreatic cancer (PC) carries a dismal 5-year survival of < 12%, and no FDA-approved agents exist for primary prevention. PC risk is strongly linked to obesity, diabetes, and chronic inflammation. Glucagon-like peptide-1 receptor agonists (GLP-1RAs) directly target obesity-driven inflammatory pathways, reduce metabolic stress, and modulate signaling pathways (PI3K/AKT/mTOR) that drive cancer apoptosis. Based on this biological rationale, we conducted the first large-scale real-world study to evaluate the safety and efficacy of GLP-1RA for primary prevention of PC in high-risk patients. Methods: Leveraging a large real-world database (TriNetX), we identified patients aged 18-90 years with ≥1 PC risk factor (tobacco use, obesity, type 2 diabetes, alcohol use disorder, chronic pancreatitis, first-degree relative with PC, genetic susceptibility to PC) based on ICD-10 codes. Patients were stratified by GLP-1RA exposure into two cohorts: Cohort A (GLP-1RA users with ≥ 2 months of prescriptions) and Cohort B (controls). The primary endpoint was incident PC, and the secondary endpoints were adverse events related to GLP-1RAs. Patients with outcomes before the study window were excluded. Propensity score matching adjusted for confounders. Multivariate logistic regression was used to estimate odds ratios for categorical outcomes, and the Cox proportional hazards model was used to estimate hazard ratios for time-to-event outcomes. Subgroup and sensitivity analyses were employed to validate the strength of our results. Results: Median follow-up was 3,397 days for GLP-1RA users and 2,531 days for controls. Demographics were balanced between cohorts (mean age 42 years; 74% White, 15% Black; 72% female). Among 572,330 high-risk patients (286,165 per cohort), PC incidence was significantly lower among GLP-1RA users (0.05%; 134/286,165) than among non-users (0.13%; 362/286,097), with a 2.6-fold reduction in the absolute incidence of PC among high-risk patients (NNT of 1,255). GLP-1RA use was associated with a 71% reduction in PC incidence (HR 0.289; 95% CI 0.237–0.352). GLP-1RA use was associated with an increased risk of gastrointestinal adverse events, including nausea/vomiting (HR 1.108; 95% CI 1.096–1.121), abdominal pain (HR 1.081; 95% CI 1.066–1.097), diarrhea (HR 1.199; 95% CI 1.182–1.217), and constipation (HR 1.159; 95% CI 1.143–1.176). Conclusions: GLP-1RA use significantly reduced PC incidence by 71% in high-risk patients, the strongest real-world signal yet for a pharmacologic prevention of this lethal malignancy. Although the absolute incidence remained low, the magnitude of the risk reduction underscores the transformative potential of GLP-1RA in preventive oncology. These findings address a critical unmet need in pancreatic oncology. Prospective mechanistic and randomized studies are warranted to validate these associations.

e16335 Background: Splanchnic vein thrombosis (SVT) is a rare but serious complication in gastrointestinal (GI) cancer patients. This study investigated its features and outcomes in U.S. hospitalizations. Methods: Hospitalizations (≥18 years old) with a GI malignancy (ICD-10-CM C15–C26) and portal vein thrombosis (PVT) (I81), hepatic vein thrombosis (HVT) (I82.0), or mesenteric/splenic vein thrombosis (MSVT) (I82.890 + K55) in any of the 40 diagnosis fields were found in 2022 National Inpatient Sample. Exclusions: non-GI cancers (C00-C14, C27-C96), thrombophilia (D68.5, D68.6, D45, D47.1, D59.5), or cirrhosis (K74, K70.3). Demographics and outcomes (in-hospital mortality, length of stay [LOS], and total hospital charges [TOTCHG]) were compared between SVT and non-SVT groups using unweighted descriptive statistics (medians, IQRs), χ², and Mann-Whitney U tests. Multivariable regression of mortality, LOS, and TOTCHG were adjusted for age, sex, race, income quartile, insurance type, hospital region and bed size, and comorbidity burden (elixsum). Stata 17.0 was used. Results: Among 51,940 GI cancer hospitalizations (mean age 68.37 ± 12.79 years; 44.46% female), 833 (1.60%) had SVT. SVT group was younger (66.60 ± 11.76 vs. 68.40 ± 12.80 years; p < 0.001). Sex distribution was similar (58.34% vs. 55.50% male; p = 0.101). Common cancers in SVT: liver (n=427; 51.26%), pancreas (n=310; 37.21%), and colorectal (n=42; 5.04%). SVT prevalence was highest in liver (7.10%), pancreatic (3.14%), gallbladder (2.65%), and biliary (2.13%) cancers. PVT was the predominant subtype (n=788; 94.60%), followed by HVT (n=21; 2.52%), MSVT (n=4; 0.48%), and multiple SVT sites (n=20; 2.40%). SVT subtype distribution was: PVT (n=788; 94.60%), HVT (n=21; 2.52%), MSVT (n=4; 0.48%), and multiple sites (n=20; 2.40%). SVT had longer median LOS (5 [3–10] vs. 4 [3–8] days; z = -6.666, p < 0.001), similar median charges ($65,053 [I33,560–130,026] vs. $66,433 [35,897–121,038]; z = -0.76, p = 0.447), and higher mortality (9.48% vs. 4.41%; χ² = 49.042, p < 0.001). In adjusted analyses, SVT had higher mortality (aOR = 2.23, 95% CI 1.74–2.85, p < 0.001), longer LOS by 1.50 days (95% CI 0.99–2.02, p < 0.001), and 14.63% higher TOTCHG (coefficient 0.136, 95% CI 0.051–0.222, p = 0.002). Conclusions: SVT affects ~1.6% of GI cancer hospitalizations, is most prevalent in liver cancer, and associated with longer stays and higher mortality, underscoring the need for proactive monitoring. Multivariable regression analyses of SVT on clinical outcomes (unweighted sample). Outcome SVT Effect 95% CI P-value Interpretation In-hospital mortality OR 2.23 1.74–2.85 <0.001 SVT doubles odds of in-hospital death Length of stay (days) +1.50 0.99–2.02 <0.001 SVT associated with ~1.5 extra hospital days Total hospital charges ($) +0.136 (log scale) 0.051–0.222 0.002 SVT increases hospital costs by ~14.63%

e16334 Background: Liver cancer remains a leading cause of cancer-related mortality in the United States. Although overall mortality has increased over the past two decades, age-specific patterns and demographic disparities are not fully characterized. Here, we analyzed U.S. liver cancer mortality trends from 1999 to 2023, with a focus on age-related, racial and ethnic, and geographic differences. Methods: Mortality data for U.S. adults aged ≥25 years from 1999 to 2023 were obtained from the CDC WONDER database (ICD-10 codes C22.0–C22.9). Age-adjusted mortality rates (AAMRs) were standardized to the 2000 U.S. population. Joinpoint regression was used to estimate annual percent change (APC) and average annual percent change (AAPC) across age groups, sex, race/ethnicity, U.S. census regions, and levels of urbanization. Results: From 1999 to 2023, overall liver cancer AAMRs increased from 6.95 to 10.1 per 100,000. Marked age-specific differences were observed. Mortality among adults aged 45–54 and 55–64 years peaked in the mid-2010s and subsequently experienced significant declines through 2023 (recent APCs < −4.5% for both groups; P < 0.001). Specifically, while the overall AAPC for the 55–64 age group was +1.89%, mortality significantly reversed after 2016 (APC, −4.59%). In contrast, mortality continued to rise among older adults, including those aged 65–74 years (AAPC, +2.26%), 75–84 years (AAPC, +1.69%), and aged ≥85 years (AAPC, +1.77%) (all P < 0.001). Differences in mortality trends were also identified across racial and ethnic groups. Sustained increases were observed among Non-Hispanic White individuals (AAPC, +1.89%; P < 0.001), while Hispanic/Latino populations experienced declines after 2013 (APC, −0.67%; P = 0.003). Non-Hispanic Black individuals demonstrated a significant decrease in mortality between 2017 and 2020 (APC, −3.82%; P = 0.03). Regionally, liver cancer mortality declined in the Northeast after 2013 (APC, −0.66%; P < 0.001), whereas no significant post-peak declines were observed in the South or West. Conclusions: U.S. liver cancer mortality trends exhibit substantial heterogeneity by age, race and ethnicity, and geographic region. The recent, sharp declines among middle-aged adults (APCs < −4.5%) likely reflect the success of large-scale viral hepatitis screenings and highly effective antiviral therapies, whereas continued increases among older adults may relate to the cumulative impact of metabolic risk factors and delayed diagnosis. Persistent racial and regional disparities highlight the need for targeted, age-specific prevention and surveillance strategies.

e16102 Background: Esophageal cancer remains a significant cause of cancer mortality in the United States. While overall trends have been reported, less is known regarding contemporary long-term patterns and specific disparities across demographic and geographic strata. We aimed to characterize esophageal cancer mortality trajectories and identify populations with divergent outcomes by age, race/ethnicity, geographic region, and urbanization status. Methods: We analyzed esophageal cancer mortality data from 1999 to 2023 using the Centers for Disease Control and Prevention (CDC) WONDER database. Age-adjusted mortality rates (AAMRs) were standardized to the 2000 U.S. standard population. Joinpoint regression analysis was utilized to estimate the annual percent change (APC) and average annual percent change (AAPC) with 95% confidence intervals (CIs). Analyses were stratified by age, sex, race/ethnicity, census region, state, and urbanization level. Results: Overall esophageal cancer mortality in the U.S. declined significantly from 1999 to 2023 (AAPC, -0.81%; 95% CI, -0.97% to -0.65%; P < 0.001). Declines were observed across most age groups, with the steepest reductions among adults aged 45-54 years (AAPC, -1.68%; 95% CI, -2.01% to -1.35%) and 55-64 years (AAPC, -1.25%; 95% CI, -1.51% to -0.99%). Marked racial and ethnic disparities were identified. Non-Hispanic (NH) Black individuals experienced the most rapid decline (AAPC, -4.07%; 95% CI, -4.42% to -3.72%). Conversely, trends among NH White individuals remained stable overall (AAPC, -0.05%; P = 0.59), characterized by an initial increase from 1999 to 2005 (APC, 1.14%; P = 0.002) followed by a modest decline from 2005 to 2023 (APC, -0.44%; P < 0.001). Regionally, the West demonstrated the largest decline (AAPC, -1.13%; 95% CI, -1.26% to -0.99%), followed by the Northeast (AAPC, -1.05%). West Virginia was the only state to exhibit a significant increase in mortality (AAPC, 1.01%; 95% CI, 0.38% to 1.64%; P = 0.003). Mortality declined significantly in metropolitan areas (AAPC, -1.09%; P < 0.001), while trends in nonmetropolitan areas remained stagnant (AAPC, 0.48%; P = 0.12). Conclusions: Despite a national decline in esophageal cancer mortality, substantial disparities persist across racial, geographic, and urbanization groups. The dramatic improvement among NH Black populations contrasts with the relative stability in NH White populations and the lack of progress in nonmetropolitan areas. The significant rise in mortality in West Virginia warrants targeted investigation. These findings underscore the critical need for population-tailored strategies to ensure equitable progress in cancer prevention and care.

e16017 Background: Pancreatic ductal adenocarcinoma (PDAC) is the most common form of pancreatic cancer and remains one of the deadliest malignancies worldwide, largely due to late-stage diagnosis. Early-stage PDAC is frequently asymptomatic, limiting opportunities for timely detection and intervention. We aimed to develop prognostic models using routinely acquired pre-diagnostic clinical data to identify individuals at high risk for PDAC prior to clinical manifestation. Methods: We conducted a retrospective study of patients treated at Mayo Clinic sites in Rochester, MN; Phoenix, AZ; and Jacksonville, FL between January 2011 and February 2023. The cohort included 1,151 patients with PDAC and 3,745 noncancer controls. CT scans were standardized to axial abdominal acquisitions with slice thickness between 0.8–3.0 mm. Radiology reports were de-identified, normalized, and parsed into structured sections (Indications, Findings, Impression), with an additional pancreas-specific section aggregating pancreas-related sentences. Demographic and clinical variables, e.g., age, sex, race, etc., were encoded using a templated sentence representation. All textual inputs were embedded using a Sentence-BERT model. A multimodal deep learning survival model was developed to integrate CT imaging, radiology report embeddings, and demographic features. The model was trained using a negative log-likelihood loss to estimate patient-specific PDAC risk scores. Model performance was assessed using the concordance index (C-index). Internal validation included patients from MN and AZ sites, while external validation used the FL cohort. Results: We evaluated unimodal (imaging, text, and demographics) and multimodal inputs. Fusion of imaging, radiology report, and demographic features achieved the highest performance, with internal and external validation C-indices of 0.885 (95% CI, 0.856–0.908) and 0.751 (95% CI, 0.719–0.783), respectively. Subgroup analysis by tumor location demonstrated differential performance. For pancreatic head tumors, internal and external C-indices were 0.916 (95% CI, 0.841–0.960) and 0.778 (95% CI, 0.730–0.823). For pancreatic body tumors, C-indices were 0.954 (95% CI, 0.897–0.997) internally and 0.929 (95% CI, 0.897–0.953) externally. For pancreatic tail tumors, internal and external C-indices were 0.903 (95% CI, 0.830–0.974) and 0.789 (95% CI, 0.689–0.874). Conclusions: A multimodal deep survival model leveraging pre-diagnostic CT imaging, radiology report embeddings, and demographic data can effectively identify individuals at high risk for PDAC prior to clinical diagnosis. Differences in performance were observed across pancreatic head, body, and tail tumors. This approach highlights the potential of opportunistic cancer risk stratification from routine imaging to enable earlier detection and targeted intervention strategies.

e15557 Background: In metastatic colorectal cancer (mCRC), systemic chemotherapy can enable conversion surgery in a subset of patients with initially unresectable disease. Although R0 resection considered a key determinant of long-term survival, its clinical impact across heterogeneous trial populations and treatment strategies remains insufficiently defined. We aimed to clarify the survival benefit of R0 resection after first-line chemotherapy using large-scale individual patient data. Methods: Using the ARCAD database, which comprises 67 randomized clinical trials and approximately 47,000 patients, we conducted a pooled analysis of individual patient data from six first-line randomized clinical trials (COIN, COIN-B, 03-TTD-01, ATOM, SOFT and PARADIGM) for which data on R0 resection were available. Patients with initially unresectable mCRC and available surgical data were categorized into three groups: R0 resection, non-R0 resection, and no surgery. All R0 cases were included, comprising liver-only, lung-only, combined liver and lung metastases, and peritoneal dissemination or other metastatic sites. Overall survival (OS) and progression-free survival (PFS) were evaluated using multivariable Cox proportional hazards models adjusted for age, sex, and performance status. Pre-specified subgroup analyses examined the consistency of outcomes according to tumor response, metastatic site, primary tumor sidedness, RAS status, and targeted therapy class. Results: Among 4,445 eligible patients, 263 underwent R0 , 233 underwent non-R0 resection, and 3,949 received no surgical intervention. The R0 resection group comprised patients with liver-only metastases (n = 176), lung-only metastases (n = 23), and combined liver and lung metastases (n = 27). R0 resection was associated with significantly improved OS and PFS compared with both non-R0 resection and no surgery. These survival benefits were consistently observed across key clinical subgroups, including patients achieving complete or partial response to chemotherapy, those with liver or lung metastases, left- and right-sided primary tumors, and both RAS wild-type and mutant disease. Importantly, the benefit of R0 resection was maintained irrespective of anti-EGFR– or anti-VEGF–based first-line therapy. Conclusions: R0 resection following systemic chemotherapy is strongly associated with durable survival benefits in patients with initially unresectable mCRC across diverse clinical settings. These findings support the pivotal role of conversion strategies and highlight the importance of multidisciplinary evaluation to identify appropriate candidates for curative-intent resection.

e15509 Background: Large language models (LLMs) are increasingly explored for oncology decision support, yet comparative evidence evaluating medical-specific versus general-purpose LLMs using expert-reviewed clinical cases is limited. We compared a medical-specific LLM (Google MedGemma 27B) with a general-purpose LLM (ChatGPT-5) using gastrointestinal (GI) and genitourinary (GU) oncology vignettes. Methods: This cross-sectional study evaluated 32 de-identified oncology vignettes (26 GI, 6 GU) created by a U.S. board-certified medical oncologist and informed by real-world cases. Vignettes included text-only information on the clinical context, description of pathology, tumor staging, biomarkers, and relevant comorbidities. Both LLMs were prompted to provide National Comprehensive Cancer Network(NCCN) guideline-based treatment recommendations, monitoring strategies, maintenance therapy considerations, and references, along with a self-reported confidence score (0–100). Board-certified U.S. medical oncologists independently reviewed outputs using a structured instrument; two disease specific expert oncologists evaluated each case. Outcomes included correctness and comprehensiveness (1–10 scales), hallucinations (factually fabricated or unsupported statements), NCCN traceability, potential patient harm, and clinician trust. Associations between accuracy and comprehensiveness were assessed using correlation coefficients. Inter-rater reliability was assessed using a two-way random-effects intraclass correlation coefficient (ICC). Results: ChatGPT-5 demonstrated higher mean correctness (8.65 vs 4.68) and comprehensiveness (8.68 vs 5.22) scores compared with MedGemma 27B. Accuracy and comprehensiveness were strongly correlated overall (Pearson r = 0.93, p < 0.001), with significant model-specific correlations (ChatGPT-5 r = 0.86, p < 0.001; MedGemma r = 0.77, p < 0.001). Hallucinations occurred less frequently with ChatGPT-5 (12.3% vs 65%), and NCCN-traceable recommendations were more common (75.4% vs 36.7%). Responses judged as potentially harmful occurred in 18.5% of ChatGPT-5 outputs versus 70% of MedGemma outputs. Clinician trust favored ChatGPT-5 (84.6% vs 15%). Inter-rater reliability improved with rater averaging (ICC(A,k): correctness 0.72; comprehensiveness 0.64). Conclusions: In expert-reviewed GI and GU oncology cases, ChatGPT-5 demonstrated higher clinical accuracy, comprehensiveness, and guideline alignment than a medical-specific LLM; however, a substantial rate of hallucinations and potentially harmful recommendations persisted across models. These findings underscore that, despite performance differences, current LLMs require rigorous clinician oversight and should not be used as standalone decision-support tools in oncology.