Citations

198 Background: The Hedgehog (Hh) signaling pathway regulates tissue patterning, growth and differentiation, including within gastrointestinal tract. Mammalian homologs include Sonic (SHH), Indian (IHH) and Desert (DHH) Hedgehog, which signal through PTCH and SMO and the GLI transcription factors. In the digestive tract, both SHH and IHH are expressed, but the functional role of Hh signaling in mCRC remains incompletely defined. Methods: We analyzed 433 mCRC patients treated with either bevacizumab (bev = 226) or cetuximab (cet = 207) plus first line chemotherapy in CALGB/SWOG 80405 (NCT00265850). RNA expression (Illumina HiSeq 2500) of 9 Hh pathway genes (SHH, IHH, DHH, PTCH1, PTCH2, SMO, GLI1, GLI2, GLI3) was evaluated. Median overall survival (mOS) and progression free survival (mPFS) in months (mo) were compared across expression tertiles (low [L], medium [M], high [H]). Likelihood ratio tests, hazard ratios and 95% confidence intervals were computed from multivariable Cox proportional hazards models, adjusting for age, sex, ethnicity, ECOG PS, tumor location, number of metastatic sites, KRAS , microsatellites instability status and treatment. Prognostic effects were modeled as marginal terms; predictive effects were modeled as multiplicative interactions. Results: Overall, IHH and GLI2 demonstrated prognostic value for OS after adjusting for multiple testing (Benjamini-Hochberg False Discovery Rate (FDR) < 0.05). High IHH expression (H group, n = 143) correlated with improved mOS (H: 36.0 mo vs M: 32.8 mo vs L: 23.5 mo; p= 0.031), whereas low GLI2 expression (L group, n = 145) was associated with longer mOS (H: 25.6 mo vs M: 33.9 mo vs L: 33.7 mo; p= 0.0016). Moreover, IHH showed predictive value for cet benefit in both OS (mOS H: 38.1 mo vs M: 34.0 mo vs L: 18.4 mo; p= 0.0032) and PFS (mPFS H: 12.6 mo vs M: 12.9 mo vs L: 8.8 mo; p= 0.0025) with significant treatment interaction for OS and PFS (p= 0.0046 and p= 0.0049, respectively; FDR< 0.05). No significant associations were found in bev treated pts. No other Hh genes showed significant prognostic or predictive effects after FDR correction. Conclusions: These results highlight a dual role of Hh pathway in mCRC. Consistent with prior preclinical evidence, reduced IHH expression appears to mark poor prognosis, likely reflecting the loss of IHH-mediated restraint on Wnt signaling, which drives proliferation and dedifferentiation. Conversely, low GLI2, a downstream transcription factor, was linked with favorable outcomes, underscoring context-specific effects of Hh signaling. The strong predictive association of IHH with cet efficacy supports its potential as a biomarker in RAS wild type mCRC. These results suggest that Hh pathway modulation, particularly IHH signaling, may influence disease biology and treatment response. Validation in independent cohorts and mechanistic studies is warranted to clarify implications.

179 Background: Neuroligins are a family of postsynaptic cell adhesion molecules that play a crucial role in synapse formation and function. Recent studies suggest Neuroligin family genes (NLGN1,2,3,4X) may promote cancer cell growth and metastasis in various cancers, including colorectal cancer. Here, we investigated the association between neuroligin family gene expression and first-line treatment outcomes in mCRC. Methods: We evaluated the expression of 4 genes in the neuroligin family (NLGN1, NLGN2, NLGN3 and NLGN4X) on treatment outcomes of 433 pts with mCRC treated with either bevacizumab (Bev, n = 226) or cetuximab (Cet, n = 207) in combination with first-line chemotherapy within the CALGB/SWOG 80405 trial. mRNA was isolated from FFPE tumor samples and sequenced on the HiSeq 2500 (Illumina). Overall survival (OS) and progression-free survival (PFS) were compared between groups of pts categorized by low expression (Q1) (≤25 percentile), medium expression (Q2-3) (25th-75th percentile) and high expression (Q4)(>75th percentile) of NLGN family genes. The associations between gene expression and clinical outcomes were assessed via the multivariable Cox model and plotted with Kaplan-Meier curve. Treatment-gene interactions were evaluated using the likelihood ratio test. False discovery rate (FDR) was controlled through the Benjamini-Hochberg procedure. Results: High NLGN2 expression was associated with shorter PFS (median 10.0 mos, 95% CI (8.3, 11.3)) and OS (median 25.2mos, 95%CI (20.9, 30.0)) compared to low NLGN2 expression (PFS: median 11.2mos, 95% CI (9.8,14.3))(HR 1.41, 95% CI (1.06, 1.89)), P = 0.029). (OS: (median 31.8mos, 95%CI (26.8, 37.5)( HR 1.49, 95% CI (1.10, 2.01) P =0.0095) in multivariate analysis. Patients treated in the cetuximab arm(n=207), high HLGN2 expression significantly associated with shorter PFS (median 8.0 vs 13.4, HR 2.02, 95% CI (1.35, 3.03), P = 0.0021) and OS (median19.6 vs 36.3, HR 2.30, 95% CI (1.50, 3.53), P =8.3e-05). High NLGN3 expression was associated with shorter OS (23.9 vs 36.0 mos, P =0.0083) in cetuximab treatment arms, no significance found in PFS. We found significant treatment interaction for OS with NLGN2 and NLGN3 favoring cetuximab. In pts treated with cetuximab, High NLGN2 expression was significantly associated with shorter OS (median 29.1vs 36.3) (p=0.024) compared to pts treated with bevacizumab. For NLGN3, pts treated with cetuximab, high NLGN3 expression was significantly associated with shorter OS (median 25.0 vs 36.0) (p=0.018) compared to pts treated with bevacizumab. No significant association was found in NLGN1 and NLGN4X. Conclusions: Our results suggest NLGN2/3 gene expression may predict mCRC treatment outcomes. Potential therapeutic drug target NLGN in combination with anti-EGFR treatment in mCRC is warranted.

175 Background: BRAF V600E mutations occur in ~8–12% of metastatic colorectal cancer (mCRC) and are linked to poor prognosis, therapy resistance, and limited response to anti-EGFR monotherapy. While the BEACON trial established encorafenib plus cetuximab as standard, the added value of triplet regimens remains unclear. Clarifying the efficacy and toxicity of doublet vs triplet therapies is essential to optimize care in this high-risk group. Methods: Studies published across PubMed, Embase, and Google Scholar from inception till May 2025 with reported outcomes of the desired targeted therapies for BRAF V600E-mutated mCRC were screened and finally 12 studies included in this review. Data on efficacy parameters such as overall survival (OS), progression-free survival (PFS), objective response rate (ORR), disease control rate (DCR) as well as safety indicators such as adverse events (AEs), and discontinuation rates were extracted and analyzed. Results: In the ≥2nd-line setting, doublet and triplet regimens (doublet plus binimetinib) demonstrated improved outcomes compared with cetuximab plus irinotecan/FOLFIRI, with higher OS (7.2–9.2 vs 5.6 months), PFS (4.2–4.6 vs 1.5 months), DCR (65–75% vs 31%), and a comparable incidence of severe AEs (3–5% vs 4%). Panitumumab plus vemurafenib, also in the ≥2nd-line setting, reported an OS of 7.6 months, PFS of 3.2 months, and a DCR of 27%. In the 1st-line setting, the doublet regimen plus mFOLFOX6 with longer treatment duration (~50 weeks) showed improved outcomes compared with the doublet regimen plus higher-dose cetuximab, with higher OS (30.3 vs 19.5 months), PFS (12.8 vs 6.8 months), ORR (65.7% vs 45.6%), but also a higher incidence of severe AEs (46% vs 30%). Similarly, the cetuximab + vemurafenib + irinotecan regimen outperformed conventional chemotherapy (mFOLFOX6, FOLFOXIRI, or CAPOX) ± bevacizumab, with improved OS (25.3 vs 10–15 months), PFS (11.9 vs 7.5 months), ORR (63.2% vs 37–40%), DCR (94.7% vs 75%), and a lower rate of grade ≥3 AEs (34.2% vs 65%). Conclusions: The survival benefit observed with the doublet regimen was comparable to that of the triplet, but with lower rates of adverse events and treatment discontinuation, supporting current evidence that the doublet remains the preferred option for patients with BRAF V600E–mutated mCRC who have received one or more prior lines of therapy. Although adding conventional agents such as mFOLFOX6 to the doublet in the first-line setting improved OS, PFS, ORR, and DCR, this approach was associated with higher toxicity and discontinuation rates, raising questions about its overall benefit. Emerging data on vemurafenib in combination with cetuximab and conventional agents show promise, but further studies directly comparing encorafenib- versus vemurafenib-based regimens are warranted to optimize outcomes in this patient population.

163 Background: Being physically inactive is an independent predictor of poor outcomes in patients with stage III colon cancer. Whether physical activity can mitigate the oncologic risk of postoperative MRD is unknown. Methods: We concluded a post hoc analysis of a phase III trial evaluating celecoxib vs. placebo and 3 vs. 6 months of adjuvant chemotherapy in patients with resected stage III colon cancer. Self-reported physical activity was assessed in standardized measures of metabolic equivalent hours per week (MET-h/wk) at 2 timepoints: 1) midway through and 2) 6 months after completing adjuvant chemotherapy, modeled as a time-weighted average. MRD was determined via a clinically validated, tumor-informed circulating tumor DNA (ctDNA) assay (Signatera, Natera, Inc.). Disease-free survival (DFS) and overall survival (OS) were estimated using flexible parametric survival models. Results: Among 763 patients (median follow-up 6.0 years), 24% were highly active (≥18 MET-h/wk) and 32% were inactive (<3 MET-h/wk); 18% were ctDNA positive. Adjusting for covariates, DFS improved in patients who were ctDNA negative and moderately active (Ref: inactive; 3-17.9 MET-h/wk; HR=0.62, 95% CI [0.40-0.97]) or highly active (HR = 0.53 [0.30-0.94]). No benefit was seen among patients who were ctDNA positive (Ref: inactive; moderately active: HR = 1.20 [0.75-1.92]; highly active: HR=0.83 [0.45-1.54]). The interaction was not significant (Pint=0.269). Physical activity did not impact OS regardless of ctDNA status. The interaction of physical activity and adjuvant celecoxib use is shown in Table. Conclusions: Among patients who were ctDNA negative, physical activity was associated with improved DFS, and may demonstrate increased survival benefit alongside adjuvant celecoxib. The additive effects of both interventions were inconclusive among those who were ctDNA positive. Support: U10CA180821, U10CA180882, U24CA196171; https://acknowledgments.alliancefound.org; Natera, Pfizer; NCT01150045. Disease-free survival by physical activity level and celecoxib use, stratified by ctDNA status. 3-yr event # at risk Adjusted HR (95% CI) P-value Pint ctDNA Negative 0.258 Inactive + placebo 23 182 Ref Active + placebo 17 132 0.74 (0.43-1.28) 0.283 Inactive + celecoxib 23 167 0.92 (0.57-1.48) 0.720 Active + celecoxib 10 146 0.40 (0.22-0.79) 0.007 ctDNA Positive 0.221 Inactive + placebo 26 33 Ref Active + placebo 15 20 0.96 (0.50-1.84) 0.903 Inactive + celecoxib 34 54 0.62 (0.37-1.05) 0.076 Active + celecoxib 14 29 0.34 (0.18-0.64) 0.001 ctDNA, circulating tumor DNA; HR, hazard ratio; CI, confidence interval. Active = ≥9 MET-h/week, inactive = <9 MET-h/week. Celecoxib use defined by assignment in CALGB/SWOG 80702. Models adjusted by T and N stage, age, sex, race, assigned cycles of fluorouracil, leucovorin, and oxaliplatin, KRAS and BRAF status, BMI, and Western diet.

162 Background: In preclinical studies, the tryptophan-metabolizing enzyme, Indoleamine 2, 3-dioxygenase 1 (IDO1), promotes resistance to radiation in colorectal cancer, and IDO1 inhibition with epacadostat improves tumor radiosensitivity. In a phase 1 trial, patients (pts) with locally advanced rectal cancer (LARC, n=17) were treated with epacadostat in combination with short course radiotherapy (SCRT) and CAPOX which was well-tolerated, and the recommended phase 2 dose (RP2D) of epacadostat was 400mg BID. Here we report efficacy and safety from the NCI-supported phase 2 (NCT03516708) multicenter, open-label trial (treatment cohort). Methods: Pts were treated with standard of care (SOC) SCRT, then SOC neoadjuvant chemotherapy (NAC) followed by SOC surgical resection or non-operative management (NOM). Epacadostat (400 mg BID) started with SCRT and continued until starting NAC (a minimum of 21 days). Eight pts were enrolled prior to discontinuation of epacadostat by Incyte in February/March 2025. The primary endpoint is Neoadjuvant Rectal (NAR) score. In patients who underwent NOM, we calculated MRI-based tumor regression grade (MR-TRG). Secondary endpoints are pathologic complete response (pCR) rate, complete clinical response (cCR) rate, and progression free survival (PFS). Results: Table 1 summarizes baseline characteristics and endpoints (primary and secondary). NOM was pursued in 7/8 (87.5%) pts, surgical resection was performed on 1/8 (12.5%). At median follow-up of 12 months, local tumor regrowth occurred in 1 pt initially treated with NOM (after tumor regrowth, he underwent salvage surgery and is currently on surveillance). Treatment with epacadostat was interrupted in 2/8 (25%) patients due to grade 3 maculopapular skin rash. Most frequent treatment-emergent adverse events (TEAEs) regardless of causality (all grades (G)/G3-4) were bloating (75%/0%), increased ALT (62.5%/0%), increased AST (50%/0%), nausea (50%/0%), and constipation (50%/0%). Grade 3 diarrhea and grade 3 non-cardiac chest pain occurred in 1 patient each (both unrelated to epacadostat). Conclusions: Epacadostat demonstrates promising efficacy and manageable safety in patients with LARC. The biomarker cohort and correlative studies are ongoing. Clinical trial information: NCT03516708 . Baseline characteristics and endpoints. Patient Age Sex Ethnicity Stage Surgery/NOM NAR MR-TRG cCR/pCR Recurrence 1 52 M Caucasian T3N1 NOM NA 2 Y/NA No 2 45 M Caucasian T3N1 NOM NA 3 Y/NA Local tumor regrowth 3 48 M Caucasian T3N2 NOM NA 3 Y/NA No 4 61 M Caucasian T4N2 NOM NA 3 Y/NA No 5 57 M Caucasian T3N0 NOM NA 1 Y/NA No 6 60 F Caucasian T3N1 Surgery 15 3 N/ypT3N0 No 7 67 M Caucasian T4N0 NOM NA 3 Y/NA No 8 51 M Caucasian T4N2 NOM NA P NA No

158 Background: PIK3CA mutations are common in CRC and may be a negative prognostic factor. INAVO is a potent and selective PI3Kα inhibitor that also promotes mut p110α degradation. BEV and CETUX are standard treatment (tx) components for RAS mut and RAS -wildtype (wt) mCRC, respectively. The Phase I/Ib INTRINSIC open-label umbrella study (NCT04929223) evaluates targeted therapies in biomarker-selected mCRC subpopulations. We report INAVO + BEV and INAVO + CETUX results. Methods: Patients (pts) with PIK3CA mut, RAS mut mCRC received INAVO (9 mg oral once daily) + BEV (intravenous [IV] 15 mg/kg every 3 weeks). Pts with PIK3CA mut, RAS wt mCRC received INAVO + CETUX (IV 400 mg/m 2 then 250 mg/m 2 weekly thereafter). Primary endpoint: confirmed objective response rate (cORR). Secondary endpoints included disease control rate (DCR), duration of response (DoR), and safety. Progression-free survival (PFS) and overall survival (OS) were exploratory. Results: At the clinical cut-off dates, 16 pts were enrolled in the BEV arm; 35, in the CETUX arm. Most had ≥2 metastatic sites at enrollment (n = 16 [100%] and 30 [85.8%], respectively) and ≥2 prior lines of metastatic therapy (16 [100%] and 28 [80.0%]). In the BEV arm, 3 pts (18.8%) had no prior BEV. In the CETUX arm, 25 pts (71.4%) had no prior CETUX; 17 (48.6%) had prior anti-EGFR therapy. All pts in the BEV arm and 34 in the CETUX arm were evaluated for efficacy and safety. Median follow-up was 9.4 months (mo; interquartile range: 5.5–14.4) and 12.6 mo (8.9–16.6), respectively. Efficacy is shown in the Table. All pts had adverse events (AEs); 15 (93.8%) and 34 (100%), respectively, had tx-related AEs (TRAEs); 5 (31.2%) and 12 (35.3%), grade 3–4 TRAEs; 2 (12.5%) and 11 (32.4%), serious AEs. One grade 5 AE (infusion-related reaction) occurred (CETUX arm). Dose reduction/interruption due to AEs occurred in 12 (75.0%) and 27 (79.4%) pts; tx withdrawal due to AEs, in 0 and 4 (11.8%) pts, respectively. Hyperglycemia was the most common AE, TRAE, and AE leading to dose reduction/interruption in both cohorts. Hypertension was observed more frequently with BEV; dermatitis acneiform and dry skin, with CETUX. Conclusions: INAVO + BEV and INAVO + CETUX showed encouraging efficacy in pts with PIK3CA mut mCRC, with no new or unexpected safety signals. Clinical trial information: NCT04929223 . INAVO + BEV(n = 16) INAVO + CETUX(n = 34) cORR, all PR, n (%) 1 (6.3) 7 (20.6)* SD, n (%) 10 (62.5) 17 (50.0) Progressive disease, n (%) 3 (18.8) 9 (26.5) Missing/NE, n (%) 2 (12.5) 1 (2.9) mDoR, mo (90% CI) 7.7 (NE) 4.4 (4.2–5.3) DCR, n (%) 11 (68.8) 15 (44.1) mPFS, mo (90% CI) 4.2 (2.8–6.8) 3.5 (2.8–4.4) mOS, mo (90% CI) 11.0 (8.5–14.5) 12.8 (10.7–13.8) CI, confidence interval; m, median; NE, not evaluable; PR, partial response; SD, stable disease. *3 pts with PRs received prior anti-EGFR therapy; 1 had a best prior response of SD (last dose 233 days before study tx) and 2 had best prior responses of PRs (last doses 134 and 171 days before study tx).

126 Background: NNT/NNH analyses can estimate the magnitude of benefit vs harm of therapeutic agents and can inform clinical decision making. We quantified the benefit/risk profile of fruquintinib (F; a highly selective, oral inhibitor of all 3 VEGFRs) vs other later-line mCRC monotherapies. Methods: Six phase 3 randomized (2:1), placebo (P)-controlled trials (F: FRESCO, FRESCO-2; regorafenib [rego]: CONCUR, CORRECT; TAS-102: TERRA, RECOURSE) were identified. NNT (average number of pts needing treatment to prevent 1 additional death/progression event) for overall survival (OS) and progression-free survival (PFS) was calculated as the inverse of difference in survival function between treatment and P. NNH (average number of pts needing treatment for 1 additional pt to have an adverse event; AE) for treatment-emergent AEs (TEAEs) was calculated as the inverse of absolute risk increase relative to P. Low NNTs and high NNHs suggest a more favorable treatment profile. Results: F had the lowest NNT (most favorable) for 6-month OS (FRESCO-2) and PFS (FRESCO) vs rego and TAS-102 (Table). F had the highest NNH (most favorable) for grade ≥3 TEAEs, palmar-plantar erythrodysesthesia (PPE; FRESCO-2), and fatigue (FRESCO). TAS-102 (TERRA) had the highest NNH for diarrhea, followed by F (FRESCO-2). Consistent with mechanism of action of anti-VEGFRs, F and rego had lower NNHs for hypertension (HTN) vs TAS-102 (TERRA). F (FRESCO) had the lowest NNH for discontinuation due to TEAEs. Conclusions: In this cross-trial comparison, F had the lowest NNT for OS (FRESCO-2) and PFS (FRESCO/FRESCO-2), indicating survival benefit. F had the highest NNHs for grade ≥3 TEAEs, PPE (FRESCO-2), and fatigue (FRESCO), indicating a low risk profile. The analysis highlights fruquintinib as a favorable treatment for pts with mCRC. Clinical trial information: NCT02314819 , NCT04322539 , NCT01584830 , NCT01103323 , NCT01955837 , NCT01607957 . NNT/NNH (95% CI) for mCRC treatments.* F Rego TAS-102 FRESCO NCT02314819 N=416 FRESCO-2 NCT04322539 N=691 CONCUR NCT01584830 N=204 CORRECT NCT01103323 N=760 TERRA NCT01955837 N=406 RECOURSE NCT01607957 N=800 6-month OS 6.5(5.5–8.1) 5.3(4.8–5.9) 5.9(NA) 9.8(NA) 14.1(NA) 7.2(NA) 6-month PFS 4.2(3.7–4.8) 4.4(3.9–5.1) 5.0(NA) 9.1(NA) 7.1(NA) 7.2(NA) Grade ≥3 TEAEs 2.4 (2.0–3.1) 8.1 (5.0–22.5) 2.6 (2.0–3.7) 2.5 (2.2–3.0) 2.8 (2.3–3.6) 5.9 (4.1–10.2) TEAE leading to discontinuation 10.8 (6.7–28.5) NA † (18.0–NA) 12.5 (6.2–NA) 17.5(11.3–39.2) 250.0 (15.3–NA) 50.0 (22.9–NA) PPE 2.2 (1.9–2.5) 6.0 (4.8–8.0) 1.4 (1.3–1.7) 2.6 (2.2–3.0) NA (NA) NA (46.0–NA) HTN 2.4 (2.0–3.0) 3.6 (3.0–4.5) 5.3 (3.6–9.5) 4.5 (3.7–5.8) 14.3 (10.0–25.2) NA (NA) Diarrhea 5.1 (3.8–7.4) 7.3 (5.2–12.3) 6.2 (4.3–11.4) 3.8 (3.2–4.8) 7.9 (5.7–13.0) 5.0 (3.9–6.9) Fatigue 32.5 (10.3–NA) 25.8 (10.1–NA) 10 (5.3–80.3) 5.3 (3.8–8.4) 7.4 (5.0–13.9) 8.3 (5.4–18.1) *All trials except TERRA included best supportive care in both arms. † Lower rate of pts discontinued F vs P due to TEAEs, thus NNH was not applicable (NA).

117 Background: Oxaliplatin remains a cornerstone of colorectal cancer (CRC) chemotherapy but is frequently complicated by chemotherapy-associated liver injury (CALI). This manifests as sinusoidal obstruction syndrome (SOS), nodular regenerative hyperplasia (NRH), or porto-sinusoidal vascular disorder (PSVD). These microvascular changes lead to portal hypertension (PH), splenomegaly, and thrombocytopenia, interrupting chemotherapy delivery and compromising survivorship. Despite increasing recognition, the incidence, spectrum, and management of oxaliplatin-induced portal hypertension (OIPH) remain undefined. Methods: We systematically reviewed PubMed, Embase, and Cochrane through August 2025 for prospective and retrospective cohorts of oxaliplatin-treated CRC patients reporting OIPH incidence, pathology, radiology, or interventional management. Case reports were excluded. Outcomes included incidence, latency, complications, misdiagnosis, and interventions such as bevacizumab and partial splenic embolization (PSE). Results: Eight studies including 730 patients were analyzed. OIPH developed in ~20–30% of patients, with 2–5% progressing to clinically overt disease requiring variceal banding, Transjugular Intrahepatic Portosystemic Shunt (TIPS), PSE, or management of liver failure. Latency to presentation was typically 4–10 years after oxaliplatin exposure. Common features included splenomegaly, thrombocytopenia, varices, and ascites. Histopathology consistently showed sinusoidal dilatation, congestion, perisinusoidal fibrosis, NRH, and microthrombi, confirming endothelial-mediated injury. Up to half of chronic cases were misclassified as metabolic dysfunction–associated steatotic liver disease (MASLD/NAFLD) or cryptogenic cirrhosis. Bevacizumab demonstrated consistent hepatoprotection, delaying splenic enlargement and reducing thrombocytopenia, likely via preservation of sinusoidal endothelial integrity and reduction of von Willebrand factor–mediated platelet microthrombi. Interventional radiology provided therapeutic options: in a prospective phase II study, PSE restored platelet counts >130×10⁹/L in 94% and enabled chemotherapy resumption within a median of 14 days, with no grade ≥3 adverse events. Conclusions: OIPH is an under-recognized complication of CRC chemotherapy, with latency spanning years after treatment. Splenomegaly and thrombocytopenia serve as early clinical signals, while misdiagnosis as MASLD or cirrhosis is common. Bevacizumab provides hepatoprotection, and PSE offers a safe and effective option for chemotherapy-limiting thrombocytopenia. Routine monitoring of spleen volume and platelet trends should be integrated into follow-up, and prospective validation of prophylactic and interventional strategies is urgently needed.

103 Background: Clinicians spend significant time extracting and synthesizing information to risk-stratify and make adjuvant chemotherapy (AC) recommendations for CRC. By automating parts of this process, LLMs could help reduce documentation burden and variability in decision-making. Methods: We selected 144 stage III, pMMR CRC patients treated at Mayo Clinic (2017–2021). HIPAA-compliant Gemini-2.0-Flash-001 and GPT-4o extracted clinical variables (age, neuropathy grade, pathological stage) from EHRs and generated AC recommendations. Two approaches were compared: (1) a rule-based algorithm with nested if/else statements translating variables into NCCN-based recommendations, and (2) a dynamic LLM agent generating recommendations using a structured prompt informed by NCCN guidelines. Concordance with treating physicians was assessed, with disagreements reviewed by a blinded GI oncologist. Results: The dataset included 144 patients with median age 67 (IQR 66–68), 48.6% female. The LLM pipeline extracted neuropathy grade with 100% accuracy and pathological T/N stages with 99.3% accuracy. Concordance with physician recommendations was 93.1% for the rule-based algorithm and 62.5% for the LLM agent. In discordant cases, blinded adjudication by a GI oncologist favored the algorithm and LLM agent over physicians in most cases (Table 1). Conclusions: This exploratory work shows LLM agents can reliably extract variables and generate evidence-based treatment recommendations for stage III CRC. The high rate of expert preference for LLM recommendations in discordant cases is intriguing but limited by small sample size and possible adjudicator bias. Discordant physician recommendations reflect the complexity of decision-making, including patient preferences, performance status, comorbidities, and subtle findings not fully captured in EHRs. Rather than replacing judgment, we envision LLMs as decision support tools—analogous to an advanced trainee reviewing charts and presenting evidence-based recommendations to the attending physician with ultimate authority. Larger multi-institutional studies with multiple adjudicators are needed to validate these findings and clarify AI’s role in oncology. Patient characteristics, LLM extraction accuracy, and concordance with physician recommendations. Metric Result Total patients 144 Median age (IQR) 67 (66-68) Female (%) 48.6% LLM extraction accuracy – Neuropathy grade 100% LLM extraction accuracy – Pathological T and N stages 99.3% Primary Concordance – Rule-based algorithm vs. physician 93.1% Adjudicator preference of rule-based algorithm over treating physician for discordant cases 100% Concordance – LLM agent vs. physician 62.5% Adjudicator preference of LLM agent over treating physician for discordant cases 82.0%

Abstract Aims Cardiogenic shock (CS) is often treated with catecholamines titrated to an adequate target mean arterial pressure (MAP) while minimizing adverse effects. We aim to assess the optimal catecholamine dose/MAP balance in heart failure-associated CS (HF-CS). Methods Patients with HF-CS were retrospectively enrolled from 16 tertiary centres in 5 European countries (2016–2021; NCT03313687). Dosage was quantified by inotropic scores (epinephrine, norepinephrine, and dobutamine). Associations of baseline and seven-day summarized dosage with intensive care unit (ICU) discharge (mixed-effects logistic regression) and 30-day mortality (Cox regression) were analysed. Potential catecholamine/MAP target ratios for optimized outcomes were assessed in models adjusted for age, sex, pH, lactate and prior resuscitation, stratified by centre. Results N = 704 patients: median age 63 years, 74% male, 34% post-resuscitation, median lactate 5.2 mmol/l. Of these, 53% were discharged from ICU, 48% died within 30 days. Higher inotropic scores independently predicted a lower probability of ICU discharge (baseline score: OR 0.78 [95%-CI 0.69–0.88]; summarized score: OR 0.46 [0.38–0.56]; both P &lt; .001) and higher risk of 30-day mortality (baseline score: HR 1.27 [1.15–1.40], summarized score HR 1.83 [1.60–2.09]; both P &lt; .001). A score/MAP ratio &lt;0.403 µg/kg/min/mmHg was associated with higher ICU discharge odds (ceiling effect); a &lt; 0.426 µg/kg/min/mmHg with lower 30-day mortality hazards (no ceiling effect). Lowering catecholamine doses by accepting reduced MAP targets was linked to better outcomes. Conclusion In HF-CS, higher catecholamine support independently associates with worse outcomes. Accepting lower blood pressure targets to reduce catecholamine dosage may improve outcomes. Validation in randomized controlled trials is urgently needed.