Citations

e13683 Background: Immune checkpoint inhibitors (ICIs) have transformed cancer treatment but are frequently complicated by immune-related adverse events (irAEs), which may result in treatment interruption, hospitalization and mortality. Reliable pre-treatment tools to identify patients at higher risk for irAEs, particularly checkpoint inhibitor pneumonitis (CIP) remain limited. Artificial intelligence (AI)-based prediction models have emerged but their clinical performance and readiness for implementation are unclear. Methods: We systematically searched PubMed, Embase, Scopus and Web of Science (inception-January 2025) for studies developing AI-based models to predict irAEs or CIP in patients receiving ICIs. Eligible studies included radiomics, deep learning, natural language processing and clinical machine-learning approaches. Two reviewers independently extracted data on cancer type, toxicity endpoint, modality, algorithm, validation strategy and model performance. Study design and reporting characteristics were assessed. Due to substantial methodological heterogeneity, results were synthesized descriptively. Results: Fifteen studies met inclusion criteria, including mixed solid tumor cohorts and lung-predominant populations. Predictive targets included any-grade irAEs and organ-specific toxicities, most commonly pneumonitis. Model approaches ranged from clinical and electronic health record–based models to CT- and PET/CT-based radiomics and multimodal imaging-clinical frameworks. Reported discriminatory performance varied widely, with AUC values ranging from 0.64 to > 0.90 across model types. The most consistent performance was observed in deep learning and multimodal radiomics-clinical models for pneumonitis, with AUCs frequently between 0.75 and 0.92, whereas several clinical-only models demonstrated more modest performance. Most studies were retrospective and single-center, used heterogeneous irAE definitions and lacked standardized imaging protocols. External validation and calibration assessment were uncommon. Conclusions: AI-based models for predicting irAEs in ICI-treated patients demonstrate promising performance, particularly for pneumonitis using radiomics and multimodal approaches. However, current evidence is limited by heterogeneity, scarce external validation and uncertain generalizability. Prospective, multi-institutional validation and standardized outcome definitions are needed before AI-driven irAE risk prediction can be integrated into routine oncology practice to guide surveillance and patient counseling.

9021 Background: An escape room is a team-based activity in which participants collaborate to solve puzzles within a fixed time limit and escape. Prior studies demonstrate high learner satisfaction and improved teamwork and communication with medical education escape rooms. To date, no published reports describe an escape room designed specifically for hematology/oncology (hem/onc) fellows. As part of Hem/Onc Curriculum, we designed and implemented a hem/onc-focused escape room and evaluated learner satisfaction and perceived learning impact. Methods: The escape room was designed using evidence-based learning science principles (Table 1). Teams worked to solve sequential puzzles, centered on a patient case, addressing key hem/onc concepts, including transfusion medicine, peripheral blood smears, lymphoma classification, and coagulation disorders. Each solved puzzle revealed additional patient laboratory and pathology data. Successful completion required teams to diagnose the patient and select appropriate treatment. Following the activity, teams participated in a facilitator-led debrief focused on team dynamics, communication, and problem solving. Teams observed each other and shared feedback. A larger group debrief co-led by chief fellows and faculty reviewed learning objectives and key takeaways. Learner outcomes were assessed using a post-activity survey evaluating satisfaction, engagement, and impact on learning. Results: Eighteen learners participated, including 14 hem/onc fellows, 1 APP hem/onc fellow, and 3 internal medicine residents. Three of four teams escaped within 1 hour (mean completion time, 52 minutes). All participants provided positive verbal feedback. Fifteen learners completed the post-activity survey, and all respondents reported satisfaction with the activity and would recommend it for future fellows. One-word descriptors included “engaging,” “fun,” and “challenging.” Most respondents (93.3%) agreed that the activity reinforced key hem/onc concepts and improved confidence in clinical problem-solving. All learners felt the activity supported learning, and 93.3% reported that collaboration enhanced the learning experience. Participants identified team-based problem solving, clinical application, and interactive puzzles as the most effective components. Conclusions: A hem/onc escape room resulted in high learner satisfaction and reinforced hem/onc knowledge and clinical problem-solving. This feasible, fun, and innovative activity complements traditional fellowship education by leveraging adult learning principles through gamified, team-based learning and should be incorporated into training programs’ curriculum. Learning science principles applied in a hem/onc escape room. Active Learning Belonging Contrasting Cases Debriefing Deliberate Practice Elaboration Generation Listening and Sharing Reward Worked Examples

6501 Background: Hepcidin, a central regulator of iron homeostasis, is pathologically elevated in patients with myelofibrosis (MF) and anemia. Chronic hepcidin elevation limits iron availability for red blood cell (RBC) production and contributes to anemia onset and severity. DISC-0974 is an investigational, first-in-class, monoclonal antibody that blocks hemojuvelin, a co-receptor in the bone morphogenetic protein-signaling pathway driving hepcidin expression. Methods: RALLY-MF (NCT05320198) is an ongoing Phase 2, open-label study conducted in the US. The primary aim is to evaluate DISC-0974 efficacy in participants with MF and anemia. Secondary aims include evaluating safety, pharmacokinetics (PK), and pharmacodynamics (PD). Eligible participants are ≥18 years of age with primary or secondary MF and hemoglobin (Hgb) <10 g/dL or RBC transfusion requirement. Participants enroll into 3 cohorts based on RBC units transfused in the 84 days prior to screening: non-transfusion dependent (nTD, 0 units and Hgb <10 g/dL), low transfusion burden (TD low, 1-2 units), high transfusion burden (TD high, 3-12 units). Stable dose of concomitant Janus kinase inhibitor (JAKi) is allowed. DISC-0974 is administered subcutaneously at 50 mg monthly for up to 6 doses, with escalation to 75 mg for inadequate or lost response. Primary endpoints include major hematologic response defined as transfusion independence (TI) during any consecutive 16 weeks (TD low) or 12 weeks (TD high), or a mean Hgb increase of ≥1.5 g/dL from baseline for ≥12 weeks (nTD). Secondary endpoints include PK/PD markers of iron regulation and safety assessments. Data were summarized using descriptive statistics. Results: At data cut (16 Oct 2025), 47 participants enrolled in the nTD (n=30), TD low (n=10), and TD high (n=7) cohorts, with 53% on concomitant JAKi therapy. DISC-0974 led to sustained hepcidin reduction and iron mobilization across cohorts. DISC-0974 resulted in meaningful hematologic responses among evaluable participants: 50% of nTD participants achieved a mean Hgb increase of ≥1.5 g/dL for ≥12 weeks, 71% of TD low participants achieved TI for ≥16 weeks, and 67% of TD high participants experienced a ≥50% reduction in transfusion burden, with follow-up ongoing. Major hematologic response rates were 50% for participants on and off concomitant JAKi therapy (n=18 and n=16, respectively). DISC-0974 was associated with meaningful improvement in FACIT-Fatigue scores for nTD and TD low participants. Serious adverse events (AEs, n=9) and ≥Grade 3 AEs (n=16) were considered unrelated to DISC-0974. There were no early withdrawals due to AEs. Conclusions: DISC-0974 was well tolerated and shows a favorable safety profile. Anemia response and fatigue scores improved across patients with MF and anemia, including those receiving JAKi therapy. These data validate hepcidin reduction as a promising approach to treating anemia in MF. Clinical trial information: NCT05320198 .

4514 Background: Maintenance avelumab is a standard of care in patients (pts) with la/mUC who do not progress after first-line PBC. Cabozantinib (CABO) is an oral inhibitor of MET, VEGFR and TAM, active in multiple solid tumors. We hypothesized that an avelumab (Av)-CABO combination would improve outcomes in pts with mUC with an acceptable safety profile compared to avelumab maintenance alone. Methods: MAIN-CAV (NCT05092958) is a phase III randomized, multicenter, international trial for la/mUC pts who did not progress after 4-6 cycles of any PBC (gemcitabine-cisplatin, gemcitabine-carboplatin, MVAC, ddMVAC). Pts were randomized 1:1 to Av 800 mg IV every 2 weeks alone or with CABO 40 mg daily for ≤2 years. Stratification factors were response to first-line PBC (complete response [CR] vs partial response [PR] vs stable disease [SD]) and visceral metastases. The primary endpoint was overall survival (OS); secondary endpoints included progression-free survival (PFS), safety and quality of life. The target accrual was 654 pts; the trial closed early after enrollment of 68 pts following adoption of enfortumab vedotin plus pembrolizumab as the new preferred first-line standard in the US. Results: From 3/2022 to 3/2024, 68 pts were randomized (Av n = 33; CABO-Av n = 35). Median OS was 24.8 months (mo) with Av and 23.0 mo with Avb-CABO (Hazard ratio 1.41; 95% CI 0.66–3.01; p = 0.186). Median PFS was 4.6 mo with Av vs 6.0 mo with Av-CABO (p = 0.343). Objective response rate among pts with PR or SD after first-line PBC was 20.8% with Av vs 11.5% with Av-CABO (p = 0.424). Grade ≥3 treatment-related adverse events occurred in 24.2% pts on the Av arm vs 39.4% of pts on the Av-CABO arm and CABO dose reductions occurred in 63% of pts. Conclusions: Addition of CABO to Av did not improve OS or PFS compared with Av alone in this underpowered, prematurely closed phase III trial; safety was consistent with known profiles. U10CA180821,U10CA180882; U24CA196171, U10CA180863. (CCTG);U10CA180888 (SWOG); https://acknowledgments.alliancefound.org. Clinical trial information: NCT05092958 .

4559 Background: Baseline QOL has been linked to survival in mRCC therapy clinical trials (Cella, 2024); however, there is a paucity of prospectively collected real-world QOL data across front-line management strategies. We used data from ODYSSEY to examine the relationship between baseline QOL and early death. Methods: ODYSSEY is a multi-site, prospective observational study of 468 US pts with mRCC. Pts must have mRCC (any histology); have no prior systemic therapy (ST) or be within 3 months of starting ST; and follow up at a PCORnet study site. Exclusion criteria include treatment for cancer except mRCC. The primary objective is to determine patterns of change in QOL and symptom burden of pts with mRCC. Early death was defined as death within 12 months post-enrollment. All pts with non-missing FKSI-19 or FKSI-DRS at enrollment were included. Cumulative event probabilities as a function of time were stratified by either < or ≥ median QoL value and plotted as a step function vs time since enrollment to create cumulative event curves. To determine association between QoL at enrollment and early death, separate Cox proportional hazards models (unadjusted and adjusted) for time-to-first-event as a function of baseline QoL were fit. The adjusted model controlled for the following baseline covariates: IMDC risk (favorable vs intermediate or poor), histology (clear cell vs non). Results: As of 1/5/26, 468 pts were enrolled; 425 had non-missing FKSI-19 or FKSI-DRS at enrollment (329 immediate ST, 96 deferred ST [DST]). ST included 129 IO-IO, 108 IO-TKI, 92 Other. DST included 57 Active Surveillance, 22 Local Therapy Planned, 17 Other/Missing. 209 and 190 pts have < median FKSI-19 (score = 59) and FKSI-DRS (score = 29), respectively; 216 and 235 pts have ≥ median FKSI-19 and FKSI-DRS, respectively. Those below the median have lower rates of nephrectomy, worse IMDC risk and KPS, shorter time since RCC diagnosis, and higher rates of bone metastasis (all p<0.05). Cumulative event probabilities for death at 12 months are 21% and 7% in those < and ≥ median baseline FKSI-19, respectively; and 21% and 7% in those < and ≥ median baseline FKSI-DRS, respectively. For FKSI-19, the unadjusted hazard ratio (HR) for early death is 0.55 (95% CI: 0.43 – 0.70; p<0.001) and adjusted HR is 0.65 (0.49 – 0.85; p = 0.002) per 1 SD higher score. For FKSI-DRS, the unadjusted HR for early death is 0.57 (95% CI: 0.46 – 0.71; p<0.001) and adjusted HR is 0.66 (0.51 – 0.86; p =0.002) per 1 SD higher score (higher score indicates better QOL). Conclusions: Across management strategies in our ODYSSEY cohort of systemic therapy naïve mRCC patients, better baseline QOL was associated with better survival. This is to our knowledge the largest reported real world mRCC cohort with prospectively collected PROs. Future analyses will focus on association of outcomes with longitudinal PROs. Clinical trial information: NCT04919122 .

3014 Background: Macrocyclic next-generation tyrosine kinase inhibitors (TKIs) have transformed outcomes in ALK- and ROS1-positive NSCLC; such creatively designed medicines are similarly needed to improve outcomes in EGFR-mutant NSCLC. Using a novel, mutant-selective macrocyclic structure, BH-30643 was designed for super-potency against driver EGFRm, largely sparing wildtype (WT) EGFR, and with unique invulnerability to resistance from C797S with or without T790M. Here we report emerging first-in-human dose-escalation data from the SOLARA trial (NCT06706076). Methods: Phase 1 dose escalation used a BOIN design with backfill at potentially efficacious doses. Expansion cohorts targeted populations of interest. Patients (pts) had advanced NSCLC with EGFR and/or HER2 mutations detected on local testing (tissue or plasma). Measurable disease was required; asymptomatic untreated brain metastases (BM) were permitted. Objectives included safety, PK, and RECIST response rate (ORR). Plasma NGS was performed at baseline and on treatment. Results: As of 01/05/2026, 72 pts enrolled into dose escalation and backfill across 7 dose levels at a total daily dose of 20mg -160mg, median age 62, 44% Asian, 65% with history of BM, median of 3 prior lines of treatment (range 1-12). EGFR mutations included classical (38 pts), exon20ins (14 pts) and other/atypical (18 pts) or HER2 mutations (2 pts). Plasma exposures at doses of 40mg BID and above achieved > 10x coverage over target EC90 and exceeded exposure levels of most contemporary EGFR TKIs. Common TRAE ( > 10%) were WT EGFR-associated, including diarrhea (38%), rash (28%), fatigue (15%), stomatitis (13%), and were predominantly grade 1. Bilirubin elevation was seen in 26 pts (36%) related to off-target UGT1A1 inhibition (Gilbert’s-like). Grade 3 TRAE were seen in 15 pts (18%), most commonly bilirubin (8/15); DLTs included bilirubin increase and mucositis, occurring at doses ≥ 60mg BID. No pneumonitis was observed. Responses were observed across diverse EGFRm subsets, including in CNS. Including expansion pts, 28 pts enrolled with C797S on their most recent informative molecular testing (6 with concurrent MET+ or KRAS+), with 23 still on therapy. Of 17 with on-treatment scans, partial responses (confirmed or ongoing) were seen in 10 (59%); ORR was comparable with (50%) or without (66%) T790M. Of 9 pts with baseline C797S detected in ctDNA (3 with T790M), 7 had > 99% reduction of C797S VAF on treatment. Conclusions: Leveraging a novel macrocyclic scaffold, BH-30643 is the first TKI monotherapy to demonstrate clinical responses against C797S both with or without T790M. High plasma exposures combined with super-potency, CNS activity, and favorable safety offers potential for deep and durable EGFR inhibition. Enrollment into backfill and expansion cohorts is ongoing to better characterize activity in both TKI-naïve and resistant pts. Clinical trial information: NCT06706076 .

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%)

ABSTRACT Background Patients with end‐stage renal disease (ESRD) undergoing dialysis have an increased risk of pacemaker‐related complications due to limited vascular access and increased susceptibility to infection. Leadless pacemakers (LPMs) may mitigate these risks compared with conventional transvenous pacemakers (TPMs), but long‐term comparative data in this population remain limited. Methods We conducted a retrospective cohort study using the TriNetX Research Network, including adults (≥ 18 years) with ESRD on dialysis who underwent LPM or TPM implantation between May 2016 and December 2020. Propensity score matching (1:1) was performed using demographics and comorbidities. Outcomes assessed over up to 5 years included all‐cause hospitalization, heart failure hospitalization, device re‐intervention, device‐related complications, infection‐related hospitalization, pacemaker‐induced cardiomyopathy, and all‐cause mortality. Results After matching, 901 patients were included in each group with well‐balanced baseline characteristics. LPM implantation was associated with a significantly lower risk of device re‐intervention compared with TPM (4.1% vs. 9.3%; hazard ratio [HR] 0.44; 95% confidence interval [CI] 0.28–0.69; p < 0.01). There were no significant differences between groups in all‐cause hospitalization (65.5% vs. 68.3%; p = 0.22), heart failure hospitalization (26.9% vs. 30.2%; p = 0.13), all‐cause mortality (37.5% vs. 35.3%; p = 0.34), device‐related complications (14.5% vs. 14.3%; p = 0.94), infection‐related hospitalization (1.6% vs. 2.1%; p = 0.35), or pacemaker‐induced cardiomyopathy (9.0% vs. 8.6%; p = 0.66). Conclusion In a large, propensity‐matched cohort of patients with ESRD on dialysis, LPMs were associated with fewer device re‐interventions and comparable long‐term clinical outcomes compared with TPMs.

PURPOSE Integrating external control data into clinical trial designs and analyses has the potential to accelerate drug development processes. We reanalyzed the three experimental arms of the Individual Screening Trial of Innovative Glioblastoma Therapy (INSIGhT), a randomized phase II platform trial in newly diagnosed O 6 -methylguanine-DNA methyltransferase-unmethylated glioblastoma (ClinicalTrials.gov identifier: NCT02977780 ). To evaluate the validity of using external data sets, we compared treatment effect estimates based on internal INSIGhT control data and matched external control data. METHODS The three experimental arms of INSIGhT (abemaciclib [n = 72], neratinib [n = 80], and CC-115 [n = 12]) did not improve survival compared with internal controls (standard chemoradiation [n = 70]). We derived external control patient-level data from multiple real-world and clinical trial data sets. We applied propensity score matching and Cox proportional hazards models to estimate treatment effects with external controls. Additionally, using this glioblastoma (GBM) data collection, we specified simulation scenarios to evaluate trial designs that integrate external controls. RESULTS After matching to external controls, no survival benefit was observed for patients receiving abemaciclib (hazard ratio [HR], 1.00 [95% CI, 0.75 to 1.34]), neratinib (HR, 0.93 [95% CI, 0.70 to 1.24]), or CC-115 (HR, 0.88 [95% CI, 0.41 to 1.88]). Simulations, together with the INSIGhT data and a collection of GBM data sets, allowed us to examine efficiencies and risks of clinical trial designs that leverage external control data. CONCLUSION The use of carefully matched external controls, to replace or augment the internal controls of INSIGhT, produced treatment effect estimates that were similar to previously published analyses. Single-arm trial designs and hybrid randomized designs incorporating propensity score–matched external control data evaluated treatment effects in the early-phase testing of experimental therapies in newly diagnosed GBM. The validity of this approach and risks of bias depended on the availability of comprehensive and accurate data on all potential confounders, in the absence of unmeasured confounding.

ABSTRACT Early-life nutrition profoundly influences long-term metabolic health, and breast milk not only provides nutrients but also conveys maternal signals shaping infant metabolic development. While postpartum exercise by lactating women benefits maternal health, its impact on milk-borne signaling remains largely undefined. Small extracellular vesicles (sEVs) in breast milk are key mediators of maternal–infant communication because of their selectively packaged bioactive cargo and resistance to infant digestive enzymes and acids, enabling delivery of their cargo to peripheral tissues. Here, we show that a single session of moderate-intensity postpartum aerobic exercise robustly increases human breast milk sEV concentration, which persists for multiple post-exercise milk collections. Exercise enriches breast milk with sEVs containing regulatory metabolic cargo (proteins, miRNAs, and metabolites), which translates into enhanced mitochondrial capacity in neonatal-stage cells. These findings implicate sEVs as an exercise-responsive signaling compartment in breast milk capable of connecting postpartum maternal physical activity to beneficial infant metabolic programming. Highlights Acute moderate-intensity exercise increases human breast milk sEV concentration The exercise-mediated sEV increase lasts for multiple subsequent milk expressions Exercise coordinates a multi-omic enrichment of sEVs in breast milk Exercised breast milk sEVs enhance mitochondrial respiration in UC-MSCs