Cohen, Steven J.

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.

AbstractBackgroundAdipocyte-derived adiponectin may play a role in the host inflammatory response to cancer. We examined the association of plasma adiponectin with the density of tumor-infiltrating lymphocytes (TILs) in colon cancers and with vitamin D, clinicopathological features, and patient survival.MethodsPlasma adiponectin and 25-hydroxyvitamin D [25(OH)D] were analyzed by radioimmunoassay in 600 patients with stage III colon cancer who received FOLFOX-based adjuvant chemotherapy (NCCTG N0147 [Alliance]). TIL densities were determined in histopathological sections. Associations with disease-free survival (DFS), time to recurrence, and overall survival were evaluated by multivariable Cox regression adjusting for potential confounders (ie, body mass index, race, TILs, and N stage). All statistical tests were 2-sided.ResultsWe found a statistically significant reduction in adiponectin, but not 25(OH)D, levels in tumors with high vs low TIL densities (median = 6845 vs 8984 ng/mL; P = .04). A statistically significant reduction in adiponectin was also observed in obese (body mass index &gt;30 kg/m2) vs nonobese patients (median = 6608 vs 12 351 ng/mL; P &lt; .001), in men vs women (median = 8185 vs 11 567 ng/mL; P &lt; .001), in Blacks vs Whites or Asians (median = 6412 vs 8847 vs 7858 ng/mL; P &lt; .03), and in those with fewer lymph node metastases (N1 vs N2: median = 7768 vs 9253 ng/mL; P = .01). Insufficiency of 25(OH)D (&lt;30 ng/mL) was detected in 291 (48.5%) patients. In multivariable analyses, neither adiponectin nor 25(OH)D were associated with a statistically significant difference in DFS, overall survival , or time to recurrence in models adjusted for potential confounders. We found a statistically significant association of TILs with prognosis, yet no such interaction was observed for the association of adiponectin with TILs for DFS.ConclusionsLower circulating adiponectin levels were associated with a statistically significant increase in TIL densities in colon cancers, indicating an enhanced antitumor immune response. In contrast to TILs, neither adiponectin nor 25(OH)D was independently prognostic.

3591 Background: Adiponectin is a peptide hormone exclusively secreted by adipocytes that plays a role in immune regulation and in the host inflammatory response to cancer. We examined postsurgical adiponectin levels in relationship to tumor infiltrating lymphocytes (TILs), clinicopathological features, vitamin D status, and patient survival in participants in a phase 3 trial of adjuvant chemotherapy. Methods: Plasma adiponectin and 25-hydroxyvitamin D [25(OH)D] were analyzed by radioimmunoassay in 600 patients with stage III colon carcinoma who received adjuvant FOLFOX +/- cetuximab. TIL densities were determined at light microscopy in routine histopathological sections. The associations between adiponectin and 25(OH)D, TILs, other factors were evaluated by Fisher’s Exact, Chi-squared, t-test, and Kruskal-Wallis tests where appropriate. The association between adiponectin or 25(OH)D with disease-free survival (DFS), time to recurrence (TTR) and overall survival (OS) were evaluated by multivariable Cox regression, adjusting for body mass index (BMI), race, T, N stage, performance status, tumor location, TILs, BRAF/KRAS, and mismatch repair status. Results: A statistically significant and inverse association between adiponectin level and BMI was observed with lower levels found with obesity (BMI > 30 kg/m2) [p < 0.001]. The level of adiponectin was significantly lower in men vs women (p < 0.001), in blacks vs whites or Asians (p < 0.032), and in patients with fewer regional lymph node metastases (N1 vs N2 stage, p = 0.011). A significantly lower level of adiponectin was found in patients whose tumors had high vs low TIL densities (p = 0.040), but was unrelated to 25(OH)D. Insufficiency of 25(OH)D ( < 30 ng/ml) was detected in 291 (49%) of patients and was not associated with TILs. By multivariable analysis, adiponectin was not associated significantly with patient DFS (HRadj= 0.98, 95% CI 0.74-1.29, padj= 0.88) nor with OS nor time-to-recurrence (TTR). TIL densities were significantly prognostic, but 25(OH)D was not (DFS: HRadj= 1.12, 95% CI 0.85-1.47, padj= 0.44). No significant interaction was observed for adiponectin with TILs for the association with DFS. Conclusions: Lower adiponectin levels were associated with significantly increased TIL densities in colon cancers, indicating an enhanced anti-tumor immune response. In contrast to TILs, adiponectin was not independently associated with patient outcome. Nearly one-half of stage III patients were vitamin D insufficient, although 25(OH)D was not prognostic.

Abstract Abstract 4723 Introduction: Combination G/A has been shown to be apromising investigational therapy for patients with metastatic pancreatic cancer. A randomized phase II trial was initiated to determine if ODSH, a low anticoagulant heparin derivative that blocks several pancreatic cancer cell survival pathways, would enhance the efficacy of G/A. Preliminary data from the first 10 patients administered ODSH demonstrated an unexpected and profound preservation of peripheral blood counts, despite the highly myelosuppressive G/A regimen they also received. Methods: Patients with previously untreated, metastatic pancreatic cancer were treated with G (1000 mg/m2) and A (125 mg/m2) on days 1, 8, and 15 every 28 days until either disease progression or unacceptable toxicities ensued. ODSH (4 mg/kg bolus followed by a 48 hr. continuous IV infusion at a dose of 0.375 mg/kg/hr for 48 hrs) was administered with each treatment. The first 10 patients received A/G plusODSH with plans for a randomized comparison between A/G with and without ODSH involving 50 patients. Patients were required to have ECOG 0–1 as well as adequate hematological and bilirubin level <1.5 ULN. G-CSF administration was permitted per investigator discretion. Here we present hematological data on the 10 patients treated with A/G plus ODSH. Results: 10 patients were administered up to 5 cycles of G/A plus ODSH. The patients had a median age of 66 years (range, 59 to 75 years) and 50% were male. The median platelet and neutrophil counts prior to each cycle are listed in Table 1. No patients had grade 3/4 thrombocytopenia during any of the cycles. Forty percent of patients had grade 3 neutropenia, but none had grade 4 during any of the cycles. Only 1 patient received 1 dose of G-CSF after day 15 in cycle #1. Conclusions: A previous trial of G/A in patients with previously untreated, metastatic pancreatic cancer reported grade 3/4 thrombocytopenia and neutropenia in 28% (19% grade 3 and 9% grade 4) and 75% (26% grade 3 and 49% grade 4) of patients, respectively (Von Hoff, J Clin Oncol, 2011). With the addition of ODSH to an equivalent chemotherapy regimen in a similar patient cohort, the current trial reported grade 3 thrombocytopenia and neutropenia in 0% and 40% of patients, respectively, with no patients having grade 4 thrombocytopenia or neutropenia. With successive cycles, median platelet counts were stable to improved and a trend existed for reduced neutropenia with ODSH treatment. An explanation for the apparent salutary effects of ODSH on marrow function may involve its interaction with platelet factor 4 (PF4). PF4 is released from the alpha-granules of injured megakaryocytes after chemotherapy delivery. Increased PF4 levels promote thrombocytopenia by inhibiting thrombopoietin production and megakaryopoiesis (Lambert, Blood, 2007; Lambert, Int J Radiation Oncology Biol Phys, 2011). The high molecular weight and strong negative charge of ODSH may result in avid binding to platelet factor 4 (PF4), possibly leading to decreased PF4 levels and more rapid post-chemotherapy platelet recovery (Joglekar, Thromb Haemost, 2012). PF4 may also facilitate neutrophil adhesion to endothelial cells, and reduced PF4 levels from ODSH binding may demarginate and preserve neutrophil values (Kasper, Blood, 2006). Further study of ODSH as a marrow protective agent in myelosuppressivechemotherapy regimens is warranted. Disclosures: Off Label Use: The use of nab-paclitaxel and ODSH in pancreatic cancer. Marcus:Paringenix: Employment, Equity Ownership. Quintana Diez:Paringenix, Inc.: Employment.

TPS475 Background: The prognosis of pancreas adenocarcinoma (PCA) remains poor. A chemotherapy backbone is the current standard of care in PCA. The choice of a chemotherapeutic backbone may impact the efficacy of anti-angiogenic therapy in PCA. Ramucirumab has increased activity with fluoropyrimidines (5FU) because 5FU increases VEGF expression. Methods: This is a double-blind, placebo controlled Phase II study. Subjects will receive either Ramucirumab or a placebo followed by mFOLFIRINOX every two weeks of a 28 day cycle until progression or discontinuation for other reasons. The primary endpoint of this clinical trial is nine month PFS defined as the time from enrollment to the time of progression or death. Among the key inclusion criteria, subjects must have recurrent or metastatic pancreas adenocarcinoma (PCA) with no prior first-line systemic treatment, ECOG PS 0-1, adequate organ function, no DVT, PE or other thromboembolism within three months of randomization. Total number of patient enrolled as of September 19, 2018 is 48 of 85 at eight sites; 27 male (56%), 42 Caucasians (87.5%), three African American (6.2%), one Asian (2%). Median age is 63 (40 - 76). Majority of patients (41) had de novo metastatic disease and six with recurrent disease after surgery. Regimen has been tolerated well with no unanticipated events. Clinical trial information: NCT02581215.

752 Background: Elderly patients (≥65 yrs) with metastatic colorectal cancer (mCRC) are more likely to have comorbid conditions than younger patients ( < 65 yrs), limiting their therapeutic options. As previously reported (Mayer et al. J Clin Oncol 2017;35:abs 3559), FTD/TPI confirmed its clinical profile in mCRC in a real-world setting. Here, we report the safety and tolerability profile of FTD/TPI from the expanded-access program (EAP) in patients aged ≥65 yrs. Methods: Patients (≥18 yrs) with mCRC following ≥2 regimens of anticancer therapy and an Eastern Cooperative Oncology Group performance status of 0 or 1 were enrolled to this open-label EAP. Patients received FTD/TPI 35 mg/m2 twice daily for 5 days followed by 2 days’ rest for 2 weeks, with a 14-day rest period (28-day treatment cycle). Data were collected for therapy duration, treatment discontinuation, and adverse events (AEs). Results: Of 549 patients, 377 (68.7%) were aged < 65 and 172 (31.3%) were ≥65. The treatment duration of FTD/TPI was similar between < 65 and ≥65 yrs (Table). A lower proportion of those aged ≥65 had fatigue and nausea compared with < 65 (Table). However, the frequency of neutropenia and diarrhea was slightly increased in those ≥65 compared with < 65 (Table). One patient (0.6%) in the ≥65 group developed febrile neutropenia compared with 9 (2.4%) in the < 65 group, with no associated deaths in either group. Conclusions: This analysis from the EAP showed that the safety profile and treatment duration of FTD/TPI in patients aged ≥65 with mCRC were similar to those in patients aged < 65. FTD/TPI is well tolerated and can be considered in patients aged ≥65 with mCRC. Clinical trial information: NCT-02286492. [Table: see text]

3559 Background: A Phase 3 clinical trial (RECOURSE) showed that trifluridine/tipiracil (FTD/TPI) was effective in the treatment of refractory metastatic colorectal cancer (mCRC) treated with standard chemotherapy and targeted-therapy drugs (Mayer et al. N Engl J Med 2015;372:1909-19). An expanded-access program (EAP) was started to provide access to FTD/TPI for patients with mCRC with similar eligibility and to further assess the safety of FTD/TPI in a “real-world” setting in US patients. Methods: Patients aged ≥18 years with refractory mCRC resistant to ≥2 regimens of standard chemotherapy with appropriate biologics and an Eastern Cooperative Oncology Group performance status of 0 or 1 were enrolled to this open-label EAP. Patients received FTD/TPI 35 mg/m2twice daily for 5 days followed by 2 days’ rest repeated twice followed by 14 days’ rest over a 28-day treatment cycle until drug discontinuation. Sites reported duration of therapy, discontinuation due to disease progression, and adverse events (AEs) for each patient. Results: 549 patients were enrolled; 53.2% were male. Median duration of FTD/TPI therapy was 9.7 weeks, which was similar to US patients from RECOURSE (8.9). 10.2% of patients died during the study period, 9.7% due to disease progression. 76.1% of patients discontinued treatment due to disease progression and 4% discontinued due to AEs. Drug-related clinically significant AEs (all grades) are summarized in the table. Conclusions: Patients with refractory mCRC in this EAP had a similar exposure duration to that reported in US patients from RECOURSE, with no unexpected safety concerns. While gastrointestinal toxicity was lower in this population, this may be due to reporting of “significant” events. The clinical profile of FTD/TPI is confirmed in this larger, non-randomized “real-world” experience. Sponsorship: Taiho Oncology, Inc. Editorial Assistance: Complete HealthVizion. Clinical trial information: NCT02286492. [Table: see text]

3516 Background: Epidemiological data suggest a protective effect of vitamin D on CC risk. However, association of serum vitamin D level with clinical outcome in pts with surgically resected CC remains unknown. We analyzed total circulating 25-hydroxyvitamin D [25(OH)D)] with clinical outcome in pts with resected stage III CC treated with FOLFOX +/- cetuximab in a phase III trial. Methods: Association of pre-treatment level of plasma 25(OH)D with disease-free survival (DFS) and time-to-recurrence (TTR) was analyzed with 25(OH)D as a continuous variable by a relative risk model with splines and as a binary variable (vitamin D deficient [<30 ng/mL]) using multivariable Cox regression. Results: Overall, 49% (291/600) of pts were vitamin D deficient. Prevalence of vitamin D deficiency was significantly higher in women vs men (52% vs. 48%, p=0.030) and in blacks vs whites (74% vs. 45%, p=0.0003). The continuous 25(OH)D level and pt sex were associated by the interaction test (pinteraction=0.044). There were no statistically significant associations with DFS/TTR for either continuous 25(OH)D level (DFS, p=0.22; TTR, p=0.26) nor for vitamin D deficiency (DFS, adjusted hazard ratio [HRadj]=1.01, 95% CI, 0.74-1.38; padj=0.94; TTR, HRadj=1.00; CI, 0.72-1.38; padj=0.99. A comparison of the highest vs lowest tertile showed a significant association between the highest 25(OH)D level and better outcomes in men (DFS, HRadj.=0.18, CI, 0.04-0.76, padj=0.020; TTR, HRadj.=0.12, CI, 0.02-0.64, padj=0.014), but not in women (DFS, pinteraction=0.039; TTR, pinteraction=0.033). Conclusions: Nearly one-half of pts in the clinical trial cohort were vitamin D deficient with lowest 25(OH)D levels found in women and blacks. While vitamin D deficiency was not associated with adverse outcome, high levels of vitamin D in men, but not women, were associated with longer survival. Support: U10CA180821, U10CA180882, U10CA180820, U10CA180863, U10CA180888, U10CA077202, CCSRI 021039. ClinicalTrials.gov Identifier: NCT00079274.

284 Background: ODSH, a low anticoagulant heparin derivative, interferes with several important pancreatic cancer cell survival pathways. A randomized phase II trial was initiated to determine if ODSH improves the activity of G/A, a promising pancreatic cancer investigational therapy. Preliminary data are presented here for the first 10 patients administered ODSH as part of an initial run-in phase prior to patient randomization. Methods: Patients with previously untreated, metastatic pancreatic cancer were treated with G (1,000 mg/m2) and A (125 mg/m2) plus ODSH (4 mg/kg bolus followed by a 48 hr. continuous IV infusion at a dose of 0.375 mg/kg/hr) on days 1, 8, and 15 every 28 days until disease progression or unacceptable toxicity. Patients were required to have ECOG ≤ 1 and appropriate blood and liver function levels. The primary endpoint was progression-free survival (PFS). Results: 10 patients (median age, 66 years) were administered up to 5 cycles of G/A plus ODSH. Per RECIST criteria, a partial response (PR) was achieved in 5 patients (50%) and the median PFS will exceed 6 months. Disease control (PR or stable disease for > 16 weeks) occurred in all 10 patients. Of 9 patients with baseline elevated CA 19-9 levels, 8 had > 60% reduction. ODSH was well tolerated and unexpectedly appeared to preserve marrow function as no patients had grade 3/4 thrombocytopenia or grade 4 neutropenia during any of the cycles. Conclusions: Although G/A and FOLFIRINOX are important advances, additional progress is necessary for metastatic pancreatic cancer patients. ODSH may interfere with pancreatic cancer invasion, metastasis, and chemo-resistance by inhibiting P- and L-selectins, heparanase, and the HMGB1-RAGE axis, respectively, suggesting that ODSH may enhance chemotherapy activity. This preliminary analysis in a small number of patients is encouraging and consistent with a previous report for this regimen (Von Hoff, J Clin Oncol, 2011). ODSH has also been very well tolerated and early data suggests that it may preserve platelet and neutrophil levels. Further follow-up of the ongoing randomized phase II trial is warranted. Clinical trial information: NCT01461915.