Citations

2652 Background: Circulating Tumor Cells (CTCs) are prognostic for poor outcomes in metastatic Breast Cancer (mBC), however CTCs are uncommon in mBC (<20%) and many pts without CTCs often progress. Cancer associated macrophage-like cells (CAML) are prognostic inflammatory pro-tumorigenic PD-L1 expressing macrophages common in mBC pt blood (>90%). In a previous randomized phase II trial, CTC & CAML decreases post Bria-IMT induction correlated with clinical benefit. Bria-IMT is an allogenic whole cell vaccine engineered to express tumor associated antigens & GM-CSF, promoting adaptive & innate immune responses. The ongoing Bria-ABC (NCT06072612) phase 3 study compares Bria-IMT to physician’s choice (TPC) in late stage mBC. We present interim results, without treatment arm comparison, for Progression Free Survival (PFS) by CTC & CAML changes as the exploratory part of the trial. Methods: This still blinded ongoing multicenter randomized open label Phase 3 trial evaluates Bria-IMT+ checkpoint inhibitor (CPI) vs TPC in mBC pts lacking approved therapies. Pts are randomized 1:1:1 to Bria-IMT+CPI, TPC, or Bria-IMT monotherapy (discontinued after 150 pts). The Bria-IMT consists of cyclophosphamide, irradiated SV-BR-1-GM cells, micro-dose pegylated α IFN at each inoculation site. CPI is administered day -3 to 3. TPC followed standard of care. Blinded anonymized blood was taken at baseline (BL), prior to therapy & 2nd (T1) taken at cycle 3 (~4 weeks post initiation). CTCs & CAMLs, and PD-L1 expressions, were quantified using LifeTracDx liquid biopsy with analysis of PFS by censored univariate analysis. Results: At time of analysis, >250 consented, >170 randomized, 119 had BL and 78 had T1. Median age 56 yrs [34–83], median 6 [2–13] prior lines of therapy, 31% TNBC, 62% ER+/PR+, & 15% HER2+. ≥1 CTCs were found in 25% (30/119) at BL & 22% (17/78) at T1. ≥1 CAMLs were found in 93% (111/119) at BL & 95% (74/78) at T1. At BL, ≥1 CTC was not significant for PFS (HR=1.7, CI95% 1.0-2.9, p=0.0513), but ≥2 CTCs was significant for worse PFS (HR=1.8 CI95% 1.1-3.1, p=0.0480). At T1, ≥1 CTCs nor ≥2 CTCs correlated with PFS (HR=0.9, p=0.8392) & (HR=1.6, p=0.1079), respectively. Further, a decrease in CTCs was seen in 11 pts but did not correlate with PFS (HR=0.7, p=0.6552). ≥1 CAML at BL nor T1 correlated with PFS (HR=1.2 CI95% 0.7-2.0, p=0.6811) or (HR=0.9 CI95% 0.7-2.5, p=0.5307), respectively. However, 51 pts (65%) had a decrease or stable CAML counts between BL & T1 which did significantly correlate with better PFS (HR=2.2 CI95% 1.2-3.9, p=0.0154). Conclusions: In an ongoing analysis of a heavily treated mBC pts, we observed that in the entire blinded population, 65% of pts had stability/drop in CAMLs significantly correlated with better PFS. Treatment arm specific comparisons will not be unblinded until completion of the designated milestone (144 mortalities). Clinical trial information: NCT06072612 .

2646 Background: Immune checkpoint inhibitors (ICIs) have transformed the management of many advanced cancers but identifying patients who are likely or unlikely to benefit from ICI treatment remains a critical challenge. Biomarkers, such as PD-L1, mismatch repair deficiency, and tumor mutational burden, capture only static, unidimensional features of the complex tumor microenvironment, and therefore inadequately predict response to immunotherapy in patients. In contrast, live tumor fragments (LTFs) preserve the full complexity of the tumor microenvironment and enable functional ex vivo assessments of ICI activity. Previous efforts have had limited clinical applicability because they required larger resection specimens often collected at a single academic center. Here, the predictive capability of an ex vivo platform using core needle and forceps biopsies from multiple clinical sites was assessed and compared to approved biomarkers. Methods: Patients with eligible solid tumors (where ICI has an indication, including renal, non-small cell lung, bladder, colorectal, and triple-negative breast cancer, and others) were enrolled in ongoing, prospective observational trials (NCT05478538, NCT05520099, NCT06349642). Biopsies were cut into LTFs, encapsulated in hydrogel, and treated sequentially with IgG antibody followed by ICI (αPD-[L]1 with or without αCTLA-4). Cytokine production was assessed by a multiplexed bead-based immunoassay at multiple time points during ex vivo treatment. Using receiver operating characteristic and precision recall analyses, we identified 9 predictive cytokines (including IFN-γ, granzyme B, and CXCL9/10), which were used to develop the Elephas score (ELP-score) to assess cytokine response to ex vivo ICI treatment. Results: Of 167 eligible patient specimens, 85% (n = 142) passed quality control metrics and 130 were used for model development. Hierarchical clustering of cytokine production revealed that 27% (n = 35) of patient specimens exhibited a cytokine response to ICI. In a validation set of 20 tumors, collected from patients who were subsequently treated with ICI, ELP-score positivity correctly identified 9 of 11 patients (82%) who had an objective response (PR/CR) to ICI, including 2 patients who were negative for FDA-approved standard-of-care ICI biomarkers. Furthermore, 5 of 5 patients (100%) with clinical progressive disease were correctly characterized as ELP-score negative. Conclusions: Using routine biopsy specimens, the ex vivo live tumor platform accurately identifies patients with clinical response to ICI therapy, including those incorrectly identified by conventional biomarkers. Ongoing efforts will test these findings in an additional validation set. If confirmed, the ex vivo LTF platform could enable the identification of patients likely to benefit from ICI therapy. Clinical trial information: NCT05478538 , NCT05520099 , NCT06349642 .

2090 Background: A subset of meningiomas has an aggressive clinical course requiring multimodal therapies. Yet, treatment options beyond surgery and radiation are limited. Somatostatin receptor 2 (SSTR2) is uniformly expressed in meningiomas and can be targeted by the radiopharmaceutical agent [ 177 Lu]Lu-DOTA-TATE ( 177 Lu-Dotatate). [ 68 Ga]Ga-DOTA-TATE ( 68 Ga-Dotatate) PET is emerging as a promising imaging modality to assess treatment response in SSTR2-expressing tumors. However, prospective clinical data supporting the use of 177 Lu-Dotatate and 68 Ga-Dotatate in treatment-refractory meningiomas is limited. Methods: In this multicenter, open-label, phase 2 clinical study (NCT03971461), adult patients with progressive intracranial meningiomas received 177 Lu-Dotatate at 7.4 GBq (200 mCi) every 8 weeks for 4 doses. The primary endpoint was 6-month progression-free survival (PFS-6). Up to 5 meningioma lesions per patient were assessed by gadolinium-enhanced T1 MRI every 12 weeks. 68 Ga-Dotatate PET was obtained before/after 177 Lu-Dotatate treatments. Lesional 68 Ga-Dotatate PET activity was assessed relative to liver/spleen or superior sagittal sinus (SSS). Results: 32 patients (female = 21, male = 11) with progressive meningiomas (WHO grade 1 = 7, grade 2 = 24, grade 3 = 1) were enrolled. Median age was 65.5 (range 41-78) years. All patients had at least one prior tumor surgery and at least one course of radiation. Overall, 177 Lu-Dotatate was well tolerated and the most common grade 3/4 adverse events were transient cytopenia in 15 patients (47%). One patient (3%) discontinued treatment due to grade 4 thrombocytopenia. PFS-6 was reached in 22 patients (69%), which includes 5 of 7 (71%) with grade 1 and 17 of 24 (71%) with grade 2 meningiomas. Best radiographic response by MRI was partial response (PR) in 5 (16%) and stable disease (SD) in 20 (63%) patients. Eight patients (16 lesions) were evaluable by paired MRI and 68 Ga-Dotatate PET pre- and post-treatment. Referenced to SSS, 68 Ga-Dotatate avidity reduced by ≥50% in 2 lesions (PR), 25-49% (minor response, MR) in 3 lesions and remained stable ( < 25% reduction and < 25% increase) in 11 lesions. Using liver/spleen as a reference (6 patients, 14 lesions), 68 Ga-Dotatate uptake was reduced in 9 lesions and stable in 5 lesions. 68 Ga-Dotatate PET assessments were generally concordant with MRI-based treatment response assessments. Conclusions: The primary endpoint (PFS-6) was met in this phase 2 study with 69% of patients achieving PFS-6, thereby surpassing established historical benchmarks. 177 Lu-Dotatate was generally well tolerated and deserves further study in patients with advanced meningiomas. Exploratory 68 Ga-DOTATATE PET studies showed decreased uptake in a subset of lesions and bears promise as an imaging biomarker to assess treatment response. Clinical trial information: NCT03971461 .

2089 Background: Central neurocytomas (CNs) are rare, WHO grade 2 brain tumors that can have relatively high recurrence rates following surgical resection. Owing to the rarity of CNs and limited robust data, the optimal use and timing of radiotherapy (RT) remain undefined, particularly in young adults with low mortality. Tumor proliferation index (Ki-67) may also predict CN recurrence rates, but there is no consensus on its utility in clinical practice. The present collaboration between Northwestern University and the Mayo Clinic describes long-term outcomes of CN as it relates to RT timing and Ki-67 level. Methods: Patients established between 1990-2025 with a tissue diagnosis of CN at Northwestern University and the Mayo Clinic were reviewed. Demographic, treatment, and tumor characteristics (histologic/molecular) were recorded. Recurrence free survival (RFS) and overall survival (OS) were calculated according to the Kaplan Meier-method. Results: Fifty-one patients with CN (51% male, 49% female)--with a median age of 26 years and 9.7 years median follow-up--were included. RFS at 2, 5, and 10 years for the cohort was 79%, 65%, and 59% respectively; OS at 10 years for the cohort was 94%, and no deaths were attributed to uncontrolled CN. Subtotal resection (STR) yielded 2- and 5-year RFS of 74% and 65%, while gross total resection (GTR) yielded 2- and 5-year RFS of 90% and 66%. Median Ki-67 of the cohort was 3%; RFS at 2 and 5 years for Ki-67 <3% was 82% and 76%, compared to 76% and 43% for Ki-67 ≥3% (p<0.02). In patients who underwent resection without adjuvant RT (n=35), those with STR had a 2- and 5-year RFS of 61% and 45%, and those with GTR had a 2- and 5- year RFS of 88% and 61%. Additionally, without adjuvant RT, RFS at 2 and 5 years for Ki-67 <3% was 73% and 64% versus 71% and 24% for Ki-67 ≥3%. Among patients without adjuvant RT and Ki-67 ≥3% (n=14), 2-year RFS was 60% after STR and 78% after GTR. Salvage RT after first recurrence (n=16) had a 5-year RFS of 81%, with a median follow-up of 4.1 years. Conclusions: We report one of the largest cohorts of CN with long-term follow-up. STR and/or Ki-67 ≥3% were associated with shorter RFS. Salvage RT at recurrence achieved durable long-term tumor control, and long-term (~10-year) mortality was low. In appropriately selected patients, observation after resection may be a reasonable initial strategy, with RT reserved for salvage. General cohort characteristics (n=51) with Ki-67 (n=39). Variable Number (%) Median Age (years) 26 (range: 5-57) Gender, Male:Female (M/F) 26:25 (51/49) Subtotal Resection 29 (57) Gross Total Resection 22 (43) Adjuvant Radiotherapy 16 (31) Salvage Radiotherapy 16 (31) No Radiotherapy 19 (38) Ki-67 ≥3 21 (56) Ki-67 <3 18 (44)

2028 Background: Patients with EGFR-mutated non–small cell lung cancer (NSCLC) frequently develop BM, and CNS progression is a common mode of failure on earlier-generation EGFR tyrosine kinase inhibitors (TKIs). After TKI failure in the setting of BM, optimal next-line systemic therapy remains uncertain, particularly regarding the comparative effectiveness of targeted escalation, immunotherapy (IO)-based regimens, and chemotherapy. Methods: We conducted a retrospective cohort study using the TriNetX Global Collaborative Network. We identified EGFR-mutated NSCLC patients treated with first- or second-generation EGFR TKIs who subsequently developed metachronous BM and initiated next-line systemic therapy between 2015 and 2025. Patients with leptomeningeal disease prior to earlier-generation TKI exposure were excluded. Treatment strategies were categorized as targeted escalation with osimertinib, IO-based therapy (immune checkpoint inhibitor with or without chemotherapy), or chemotherapy alone. The index date was defined as initiation of next-line systemic therapy. Propensity score matching was performed for pairwise comparisons. The primary endpoint was overall survival (OS) at 180 days. Restricted mean survival time (RMST) through 181 days was evaluated as a sensitivity analysis. Baseline covariate balance after matching was assessed using standardized mean differences, with values <0.10 indicating adequate balance. Results: After propensity score matching with adequate baseline balance, osimertinib was associated with significantly lower 180-day mortality compared with chemotherapy (24.5% vs 33.2%; hazard ratio [HR] 0.69, 95% CI 0.47–1.00). Comparisons between osimertinib and IO-based therapy showed numerically lower 180-day mortality with osimertinib (26.0% vs 32.6%; HR 0.79, 95% CI 0.54–1.15), but this difference did not reach statistical significance. Similarly, IO-based therapy was not associated with a statistically significant survival advantage compared with chemotherapy alone (31.7% vs 37.5%; HR 0.78, 95% CI 0.58–1.06). RMST analyses for death over 181 days were directionally consistent with hazard ratio–based results, with modest, non-significant differences in mean survival time. Conclusions: In EGFR-mutated NSCLC patients with metachronous brain metastases after earlier-generation TKI failure, targeted escalation with osimertinib was associated with significantly lower 180-day mortality compared with chemotherapy. IO-based approaches demonstrated intermediate, directionally favorable but non-significant outcomes relative to both osimertinib and chemotherapy. These real-world findings support guideline recommendations prioritizing next-generation TKIs and help inform systemic treatment sequencing in a CNS-enriched population where prospective data remain limited.

2016 Background: Bevacizumab therapy (Bev) for recurrent glioblastoma (rGBM) has not demonstrated improved overall survival (OS) in randomized clinical trials. However, single-center and preliminary multi-center studies suggest that relative cerebral blood volume (rCBV) measured with dynamic susceptibility contrast MRI (DSC-MRI) at baseline or shortly after treatment initiation may predict response to Bev. The primary aim of ECOG-ACRIN EAF151 was to evaluate whether early binary change in rCBV, assessed 2–3 weeks after initiation of Bev-containing therapy, identifies patients with an OS benefit of ≥4 mos. Pre-specified secondary aims assessed baseline rCBV; post-hoc exploratory analyses examined early post-Rx rCBV. Methods: A prospective, phase II multi-center trial without randomization enrolled subjects with rGBM receiving their first Bev-containing therapy. Progression was determined by local sites (RANO criteria; MRI within 28 days of registration, ≥42 days since end of chemoradiation). Baseline (S0) and follow-up (S1) DSC-MRI were performed before the first Bev infusion (within 3 days), and 12-25 days post-infusion but before the second infusion. Anatomic MRI and DSC-MRI (double-dose Gadavist injection, full-dose preload) complied with recommended protocols. Mean normalized (nRCBV) and standardized (sRCBV) rCBV were extracted from contrast-enhancing tumor ROIs. A 2-sided logrank test (α=0.1) was used to test the primary aim (change in mean nRCBV from S0 to S1, ≥0 vs. <0). Cox regression with restricted cubic splines was used to assess the association between continuous rCBV markers and OS. Results: 146 subjects were accrued from 33 sites. 134 completed S0 and 118 completed both S0 and S1. After exclusion of uninterpretable scans, 107 were evaluable for change in rCBV, and 124 for baseline rCBV. There was no statistically significant difference in OS between subjects with binary increase (n=40; median OS 8.0 mos [90% CI 5.5–9.2]) vs. decrease (n=67; 7.9 mos [90% CI 6.3–10.3]) in mean nRCBV (p=0.67). Based on the spline fits, we observed a strong nonlinear association between OS and continuous nRCBV and sRCBV at both S0 and S1: low marker values are associated with lower risk, but beyond a threshold, higher values are not associated with greater risk. As an exploratory analysis, we estimated optimal cut points (nRCBV: S0=1.11, S1=1.16; sRCBV: S0=1.13, S1=0.99) and found that subjects whose S1 rCBV decreased below threshold had a median OS 2.7 mos (nRCBV, HR 0.61 [90% CI 0.41–0.88]) and 5.2 mos (sRCBV, HR 0.43 [90% CI 0.28–0.63]) longer than subjects whose rCBV either increased or did not decrease below threshold. Conclusions: EAF151 prospectively evaluated whether DSC-MRI markers predict OS in patients with rGBM treated with Bev. Although the change in these markers was not predictive, both baseline and early post-Rx markers were predictive of OS. Clinical trial information: NCT03115333 .

2036 Background: Patients with oncogene-driven metastatic non–small cell lung cancer (NSCLC) treated with CNS-penetrant tyrosine kinase inhibitors (TKIs) frequently develop brain metastases (BM) while continuing targeted therapy, and outcomes of adding local CNS-directed therapy in this setting are not well defined. We evaluated outcomes associated with local CNS therapy after post-TKI BM diagnosis. Methods: We conducted a retrospective cohort study using the TriNetX Global Collaborative Network. Adults with NSCLC treated with osimertinib or lorlatinib who subsequently had a coded diagnosis of BM (ICD-10 C79.31) at least 1 month after TKI initiation and evidence of continued TKI use were included. Patients were stratified by receipt of CNS-directed local therapy (SRS/SBRT or neurosurgery) within 3 months of the index BM diagnosis versus no local therapy. The primary outcome was all-cause mortality within 180 days of the index event. Secondary outcomes included hospitalization and systemic corticosteroid use. Cohorts were balanced using 1:1 propensity score matching on demographics, comorbidities, extracranial disease, laboratory values, and TKI exposure. Survival analyses were performed, and proportional hazards assumptions were validated using Schoenfeld residuals (p>0.05). Results: Among 1,157 eligible patients, 1,053 matched pairs were generated. Within 180 days of the index brain metastasis diagnosis, mortality was lower among patients who received local CNS therapy compared with those who did not (10.9% vs 18.1%), corresponding to a hazard ratio (HR) for death of 0.54 (95% CI 0.43–0.68; p<0.001). Survival probabilities at 180 days were 88.7% in the local therapy cohort and 80.2% in the no local therapy cohort. Hospitalization within 180 days occurred in 20.3% of patients receiving local therapy and 22.4% of those without local therapy (p=0.24), with a nonsignificant trend toward delayed hospitalization (HR 0.84, 95% CI 0.69–1.01). Systemic corticosteroid use was common and similar between groups (61.3% vs 59.4%; p=0.40), with no significant difference in time to steroid initiation. In an osimertinib-only sensitivity analysis, 180-day mortality was 10.5% among patients receiving local CNS therapy compared with 19.4% among those without local therapy (HR 0.48, 95% CI 0.37–0.61; p<0.001). Conclusions: In this multi-institutional real-world cohort of patients with NSCLC receiving osimertinib or lorlatinib who developed BM, receipt of local CNS-directed therapy within 3 months was associated with lower 6-month mortality compared with continuation of targeted therapy alone, with consistent findings in an osimertinib-only sensitivity analysis. These results suggest a potential survival benefit of local CNS intervention in selected patients, while acknowledging residual confounding related to intracranial disease burden and clinical indication.

1673 Background: Survivorship care plans (SCPs) summarize cancer treatments and provide evidence-based recommendations for surveillance, screening, and management of treatment-related complications. Despite the availability of guidelines and templates, SCPs remain underutilized in routine practice due to the time-intensive and error-prone nature of manual creation. Empowering cancer survivors to use publicly available large language models (LLM)–based chatbots, such as ChatGPT, may offer a feasible solution to this problem. Methods: Standardized patient profiles were developed for six common cancers: breast, colorectal, non–small cell lung, small cell lung, prostate cancer, and diffuse large B-cell lymphoma. Profiles were reviewed by a primary care physician and an oncologist to ensure clinical accuracy and representativeness of common survivorship scenarios. Using lay language to simulate real-world patient input, ChatGPT (version 5.2, publicly available at the time of the study) was prompted with each profile using the question, “What follow-up care do I need?” Common survivorship symptoms for each cancer were also tested. Responses were evaluated across five domains: medical accuracy (concordance with current ASCO or NCCN guidelines), thoroughness, patient safety and risk framing, clarity and patient comprehension, and actionability. Five-point Likert scales were used (1 = not at all; 5 = very much). Scores were averaged across two independent physician raters. Readability was assessed using the Flesch Reading Ease Score. Results: ChatGPT-generated SCPs demonstrated high medical accuracy (mean score 4.25/5), with no factual errors or hallucinations identified on manual review. Patient safety and risk framing scored 3.75/5; symptom red flags and care escalation guidance were appropriately stated in most scenarios. Thoroughness was moderate (3.50/5), with key survivorship elements, such as smoking cessation counseling or genetic risk considerations; sometimes omitted unless explicitly prompted. Readability was limited, with a mean Flesch Reading Ease Score of 40.5, corresponding to a college reading level and exceeding recommended readability standards for patient education materials. Clarity (3.38/5) and actionability (3.38/5) were similarly constrained due to dense medical language. Conclusions: ChatGPT demonstrated high medical accuracy in generating survivorship care plans across six common cancers, supporting the feasibility of using publicly available LLMs to guide survivorship care. However, improvements in comprehensiveness, readability, and patient-centered actionability are necessary before such tools can be safely integrated into clinical survivorship workflows.

1633 Background: CAR-T cell therapy has transformed outcomes of hematologic malignancies and is poised to expand into solid malignancies. However, broader use of CAR-T remains constrained by the need for timely cytokine release syndrome (CRS) detection. Current episodic monitoring may miss early physiologic changes. Wearable digital health technologies (DHTs), offering continuous monitoring and real-time alerts, could enable earlier CRS detection, further support outpatient delivery, and expand access to CAR-T and potentially other immunotherapies associated with CRS. Methods: This single-center observational study (NCT05018208) enrolled patients receiving FDA-approved CAR-T products in the outpatient setting at Mayo Clinic, Rochester MN. Management of post-CAR-T complications (e.g. CRS) followed the institution’s standard of care. Patients were monitored from day -5 to day +30 post infusion with intermittent remote patient monitoring (SoC) and an arm-worn wearable device for continuous monitoring (DHT). Patients were included in this analysis if DHT data were available for ≥ 5h of data/day or ≥ 2h of data/night on ≥2 days or nights for baseline and ≥1 day or night within 72h prior to CRS, or at least 2h within 4h of CRS onset, or if DHT was available for ≥7 days/nights post infusion with no CRS event. CRS detection time by DHT was defined as the onset of pyrexia or threshold-based deviation from an individual’s baseline pulse distributions. Results: 32 patients were included in this report (median age 64 years, gender 12 F/20 M, diagnoses 18 MM/14 NHL). 25 patients experienced a CRS event, with maximum CRS grades ranging from 1-3 (16 G1, 8 G2, 1 G3). 24/25 patients with CRS required tocilizumab/steroid. Median CRS onset time was 3.4 days (range 0.4-18.4). 12/32 patients had sufficient DHT data (7 CRS, 5 no CRS) for analysis. 5 patients with CRS had a DHT detection time in advance of the SoC by a median of 13h (range 2.8 - 69.2). The DHT prediction of CRS by baseline HR change was earlier than the first episodic fever by a median of 14.5h (range 13-77.5). The remaining 2/7 patients who did not have CRS detected by the DHT had missing DHT data for more than 45h before a treated CRS event. DHT detected CRS in 2/5 patients without CRS according to SoC. One patient had fever caught with intermittent monitoring. The other was treated for neutropenia fever with suspicion of CRS and/or infection. Conclusions: These results offer evidence that continuous DHT can identify CRS events earlier than intermittent monitoring (SoC). Despite limitations related to compliance with wearable device use and data missingness, and variability in clinical management of CRS, we highlight the usability of DHT and its potential for future studies with optimized strategies to address these limitations. Clinical trial information: NCT05018208 .

1615 Background: Immune checkpoint inhibitors (ICIs) have revolutionized lung cancer treatment but cause adverse events, including immune-related colitis (ir-colitis). Accurate identification of ir-colitis is critical for improving patient outcomes and clinical research. However, current methods relying on ICD codes have substantial limitations: non-specific categories that fail to capture the cases or distinguish cases from infectious or ischemic etiologies, as well as incomplete coding that leads to missed cases. Large language models (LLMs), with advanced natural language understanding capabilities, offer a scalable approach for automated phenotyping using unstructured clinical notes. We aimed to develop and evaluate an LLM-based approach for identifying ir-colitis in patients with lung cancer receiving ICIs and compare it with ICD code based method. Methods: We identified 5,278 lung cancer patients with ICI administration at Mayo Clinic (2012-2025). To investigate ir-colitis occurrence within two years after ICI initiation, patients were stratified into three ICD-based cohorts: (1) IR-Colitis, with clinician-assigned ir-colitis-specific diagnosis codes; (2) Non-IR Colitis, with colitis diagnosis codes excluding ir-colitis; and (3) No-Colitis, absent of any colitis-related codes. We used three open-source LLMs to process clinical notes : Llama4, MedGemma, and Qwen. Prompts were designed to extract both gold standard evidence (explicit ir-colitis terminology, such as "immune-mediated colitis" and "checkpoint inhibitor colitis") and supportive evidence (such as pathology and lab results). Results: Llama4 demonstrated superior accuracy among three LLMs in manual evaluation and was thus selected for all cohort analyses. Among ICD-based IR-Colitis patients (n = 339), 82.0% had confirmatory evidence while 18% lacked evidence, suggesting potential false-positive coding. In the Non-IR Colitis cohort (n = 297), LLM-based extraction identified 43.8% patients in fact had ir-colitis, demonstrating substantial misclassification. In the No-colitis cohort (n = 4,642, random sample of 996 for analysis), the LLM identified 10.7% with documented ir-colitis, revealing substantial missing by administrative coding. Conclusions: LLM-based phenotyping demonstrates superior sensitivity for identifying ir-colitis compared to ICD codes alone, successfully uncovering cases missed by administrative coding. This scalable approach enables more comprehensive case identification for immunotherapy safety research and clinical surveillance. ICD-Based Cohorts LLM-Identified ir-Colitis Gold Standard Evidence Supportive Evidence Any Evidence ir-Colitis (n = 339) 277/339 (81.7%) 275/339 (81.1%) 278/339 (82.0%) Non-ir-Colitis (n = 297) 129/297 (43.4%) 126/297 (42.4%) 130/297 (43.8%) No Colitis (n = 996) 106/996 (10.6%) 104/996 (10.4%) 107/996 (10.7%)