Browse Articles

Extreme persistence of polyclonal CD8⁺ CAR T cells as a driver of marrow failure, neurotoxicity, and fatal immune collapse after ciltacabtagene autoleucel.

2561 Background: BCMA-directed chimeric antigen receptor (CAR) T-cell therapies induce deep responses in relapsed/refractory multiple myeloma (MM), but increasing potency has expanded the spectrum of treatment-related toxicities. While cytokine release syndrome and immune effector cell–associated neurotoxicity are well characterized, the consequences of extreme and prolonged CAR T-cell persistence remain poorly understood, particularly as CAR T-cell strategies are extended to solid tumors. Methods: We performed longitudinal, multi-compartment immunomonitoring in a patient with high-risk MM treated with ciltacabtagene autoleucel (cilta-cel) who developed an unusual leukemia-like clinical course. Peripheral blood (PB), bone marrow (BM), and cerebrospinal fluid (CSF) samples were analyzed by multiparameter flow cytometry, functional cytokine secretion assays, multiplex cytokine profiling, single-cell RNA sequencing with paired T-cell receptor (TCR) repertoire analysis, and lentiviral CAR integration site mapping. Results: Following cilta-cel infusion, the patient developed massive and persistent expansion of non-malignant CD8⁺ effector-memory CAR T cells, comprising >90% of circulating lymphocytes and ~95% of BM lymphocytes, with absence of detectable non-transduced T cells. Despite achieving complete remission of MM, extensive BM infiltration by CAR T cells was associated with marked marrow hypocellularity, trilineage hypoplasia, and prolonged pancytopenia. A distinct trafficking-competent CD8⁺ effector-memory subset (TEM5) was selectively enriched in CSF and associated with severe neurotoxicity and a local pro-inflammatory cytokine milieu. Single-cell transcriptomics demonstrated a highly migratory, cytotoxic effector program with suppressed proliferation, MAPK/TCR signaling, and tissue-residency signatures. TCR repertoire and integration site analyses confirmed polyclonality and excluded malignant transformation or insertional oncogenesis. Sustained CAR T-cell dominance coincided with profound hypogammaglobulinemia, failure of immune reconstitution, recurrent life-threatening infections, and ultimately fatal sepsis. Conclusions: These findings define a previously underrecognized toxicity paradigm of BCMA CAR T-cell therapy characterized by pathologic immune dominance and extreme persistence of cytotoxic CAR T cells, leading to marrow failure, neurotoxicity, and lethal immunosuppression. Extended multi-parametric immunomonitoring may identify patients at risk and inform risk-adapted management strategies, and this principle has potential implications for the design of next-generation CAR T-cell therapies, including for solid tumors.

Comparative efficacy and safety of anti-HER2 neoadjuvant therapy in early-stage HER2-positive breast cancer: A real-world multicenter analysis.

e12668 Background: For patients with early-stage HER2-positive breast cancer, neoadjuvant therapy represents a cornerstone of modern management. The combination of trastuzumab and pertuzumab (HP) alongside chemotherapy constitutes the current evidence-based standard. Emerging evidence indicates that an alternative dual blockade strategy pairing trastuzumab with pyrotinib (HPy) also demonstrates promising efficacy with a manageable toxicity profile. In routine clinical practice, the selection between these two potent targeted regimens (HP versus HPy) remains a critical and unresolved decision point, necessitating further comparative evaluation. Methods: This prospective, multicenter, observational real-world study enrolled patients with stage II-III HER2-positive breast cancer. Treatment allocation to either the HP or HPy group was based on physician recommendation and patient preference. The administered chemotherapeutic backbones included: 6 cycles of TCbHP/TCbHPy, 4 cycles of EC followed by 4 cycles of THP/THPy, or 6 cycles of THP/THPy. The primary study endpoint was the total pathological complete response (tpCR) rate, defined as ypT0/is ypN0. Statistical analyses incorporated propensity score matching (PSM) to mitigate baseline imbalances. Results: Between January 2022 and June 2025, 687 patients from 10 participating hospitals were enrolled. As of the current analysis, 520 patients have completed neoadjuvant therapy and subsequent surgery with evaluable data, comprising 320 in the HP cohort and 200 in the HPy cohort. Initial comparisons revealed significant differences in chemotherapy regimens and Ki-67 expression levels between the groups. To address potential selection bias, a 1:1 PSM was performed, yielding 382 well-matched patients (191 per group) with balanced baseline characteristics. Prior to matching, the tpCR rates were 61.25% for the HP group and 53.00% for the HPy group, a difference that was not statistically significant ( P = 0.064). After PSM, the tpCR rates were 57.59% (HP) versus 53.93% (HPy), again showing no statistically significant difference ( P = 0.471). Subgroup analyses consistently demonstrated no significant interaction effects on tpCR across various patient strata (all P -interaction > 0.05). Safety analysis of the 520 patients revealed distinct toxicity profiles: the most frequent adverse event in the HPy group was diarrhea of any grade (83.0%), with a 13.5% incidence of grade ≥3 diarrhea. In contrast, the most common adverse event in the HP group was vomiting (53.6%). Conclusions: In this real-world study, the HP and HPy regimens achieved comparable pathologic complete response rates as neoadjuvant therapy for early HER2-positive breast cancer. The treatment choice can therefore be individualized based on their distinct toxicity profiles, with HPy associated more with diarrhea and HP with vomiting. Clinical trial information: ChiCTR2200056467.

Real-world outcomes of dual checkpoint inhibition (nivolumab + ipilimumab) versus anti-PD-1 monotherapy in metastatic uveal melanoma: Impact of local therapy and adjusted survival analysis—A multicenter retrospective study in Russia.

9548 Background: Metastatic uveal melanoma (mUM) remains a rare and aggressive malignancy. Immune checkpoint inhibitors (ICI) demonstrate modest activity, and real-world evidence comparing dual (nivolumab + ipilimumab) versus anti-PD-1 monotherapy, as well as the additive role of local therapy is limited. Methods: Multicenter retrospective observational study conducted in Russia (IRB-approved at participating centers). Outcomes assessed: median progression-free survival (mPFS), median overall survival (mOS), objective response rate (ORR), disease control rate (DCR), and grade ≥3 immune-related adverse events (irAEs). Stratified by treatment line, ICI type, and local therapy use. Univariable Cox proportional hazards regression was performed. Results: Patients with metastatic uveal melanoma treated between 2019 and 2025 were included (n=262). Of these, 25 patients (9.5%) who did not receive systemic treatment or received only palliative care were excluded. Among the full cohort: 158 (60.3%) were female, median age at primary diagnosis was 55.9 years, median time to metastatic disease was 2.42 years. Liver metastases were present in 90.8%. After exclusion, 237 patients were analyzed for survival. Results are presented in Table 1. Univariable Cox regression (after exclusion n=237): Dual ICI exposure (yes vs no): HR 0.60 (95% CI 0.44–0.84, p=0.003); Any ICI exposure (yes vs no): HR ≈0.72 (p=0.102) — non-significant trend. Conclusions: Dual ICI (nivolumab + ipilimumab) was associated with improved overall survival (HR 0.60, p=0.003), particularly with combination with local therapy (predominantly IHP, STRT, and TACE), which substantially enhanced ORR, DCR, PFS, and OS across ICI types. Patients who were unable to receive ICI in the first-line setting may be candidates for this option in the second-line; however, this requires further investigation through subgroup comparative analyses. Results. Treatment Line ICI Type Local Therapy n mPFS, mo (95% CI) mOS, mo (95% CI) ORR (%) DCR (%) Gr ≥3 irAEs (%) First-line Dual (nivo+ipi) Without 28 3.5 (3.0–6.6) 18.4 (13.4–33.5) 10.7 28.6 39.3 First-line Dual (nivo+ipi) IHP 60% STRT 35%TACE 5% 20 7.5 (6.2–18.2) 45.7 (23.4–NR) 45.0 80.0 45.0 First-line Mono anti-PD-1 Without 47 3.1 (2.8–3.7) 18.8 (11.9–27.5) 2.1 14.9 2.1 First-line Mono anti-PD-1 STRT 50%IHP 40%TACE 6.7%RFA 3.3% 30 7.53 (4.5–12.3) 27.9 (18.6–NR) 36.7 60.0 16.7 Second-line Dual Without 35 3.3 (2.5–5.5) 13.1 (9.3–18.8) 2.9 31.4 45.7 Second-line Dual STRT 100% 3 9.3 (8.7–NR) 39.1 (9.6–NR) 33.3 100.0 33.3 Second-line Mono anti-PD-1 Without 15 9.0 (5.2–21.6) 21.3 (14.8–NR) 13.3 46.7 0 Second-line Mono anti-PD-1 STRT 50%IHP 25%TACE 25% 8 8.6 (4.7–NR) 17.3 (13.3–NR) 0 75.0 0

Safety-net coverage instability during cancer care and association with overall survival.

11090 Background: Means-tested (contingent on income) financial programs for uninsured individuals, such as Medicaid and county charity/indigent care, require periodic eligibility re-verification. A recent systematic review found that insurance coverage disruptions are associated with worse cancer stage and survival, but also highlighted that studies suffered from a limited ability to distinguish administrative lapses from eligibility-driven disruptions, and relied on single time-point measures to define longitudinal coverage. Consequently, the dynamics and clinical implications of coverage instability during cancer care are incompletely characterized. We investigated the coverage portions of the electronic health record (EHR) to overcome these issues and characterize coverage instability. Methods: We conducted a retrospective cohort study linking tumor registry records (2010–2022) to coverage verification fields from the EHR of a large county safety-net health system (2010-2024). Renewal limbo was defined as any gap between verified coverage intervals occurring > 365 days after a first verification. Multivariable logistic regression estimated odds of renewal-limbo within two years after diagnosis. Cox models estimated time to first renewal limbo and negative-binomial models for limbo duration; Prentice–Williams–Peterson models examined recurrent limbo. Two-year overall survival was estimated using a 12-month landmark design, with survival follow-up beginning at 12 months after diagnosis to ensure comparable risk windows across coverage instability groups. Results: Among 24,186 patients (56% female; 78% < age 65; 6% Medicaid, 50% County charity care; 15% breast, 8% colorectal, and 8% lung cancer), 26% experienced renewal limbo in the first year after diagnosis and 36% by the second year. Among patients with charity coverage at diagnosis who experienced renewal limbo, nearly 90% returned to charity coverage, and over two-thirds of Medicaid enrollees returned to Medicaid. Median limbo duration was 32 days (IQR, 3–220). Compared to diagnoses in 2010–2019, those diagnosed in 2020–2022 had substantially reduced odds of renewal limbo (aOR 0.29, 95% CI: 0.25-0.32) in Medicaid. In landmark analyses, uninterrupted coverage during the first year was associated with the highest survival, while patients in limbo at the time of diagnosis and at renewal limbo had the lowest survival (Table 1). Conclusions: Coverage lapses at annual renewal were common among safety-net patients treated for cancer and was largely temporary, suggesting administrative rather than eligibility-driven disruptions. Although lapses were brief, they were associated with measurably lower overall survival even 2 years after cancer diagnosis. Strong calendar-year effects in 2020-2022 suggests the influence of suspended reverification during the COVID19 public health emergency.

Current threshold of measurable disease in AL amyloidosis clinical trials.

7547 Background: Eligibility for clinical trials in AL amyloidosis requires measurable disease to ensure that the hematologic response can be adequately assessed, usually set at 5mg/dL. However, there are validated response criteria in low burden disease (dFLC 2-5mg/dL) that correlate with clinical outcomes in the upfront setting (Milani, 2017). Despite validated criteria in low burden disease, most trials maintain a dFLC of 5mg/dL as measurable disease. Shorter survival is seen in patients with high-risk dFLC progression, defined as dFLC >2mg/dL, >20% of baseline value and >50% increase from nadir (Palladini, 2018). Patients with organ progression also have poor outcomes. We examined patients who were excluded from clinical trials to determine if the current measurable disease criteria adequately capture patients who require salvage therapy. Methods: We reviewed patients screened for three clinical trials enrolling patients with relapsed/refractory AL amyloidosis: phase I/II trial of carfilzomib (NCT01789242), phase III trial of ixazomib/dexamethasone versus physician’s choice (NCT01659658) and phase I trial of etentamig (NCT06158854) to identify patients who failed screening. All clinical trials required measurable disease, defined as dFLC at least 5mg/dL. High-risk dFLC progression (Palladini, 2018) and organ progression (Palladini, 2012/2014) were evaluated at the time of screen failure. Results: Among 47 patients screened, 14 (30%) were enrolled: 7 for carfilzomib, 6 on the ixazomib trial, 1 for etentamig (Figure 1). Of 33 excluded patients, 18 (54%) were due to dFLC < 5mg/dL. Among 18 patients excluded due to insufficient dFLC at the time of screening, 6 met criteria for organ progression (cardiac (n=4), renal (n=2)), 6 had high-risk dFLC progression (as defined in background) and 2 had both (Table 1), indicating 10 patients were at high risk of poor outcomes. Four patients were lost to follow-up within 6 months of screening. In the remaining 14 patients, 5 received alternative commercially available therapy within 6 months of screening, 5 enrolled in trials with lower dFLC eligibility, and 1 progressed to dFLC > 5, was rescreened and enrolled on the carfilzomib trial. Among the other 3 patients, none achieved a dFLC of 5mg/dL: 1 died within 12 months, 1 was treated after 12 months and one remains off treatment at 8 months. Conclusions: Two thirds of patients with AL amyloidosis failed screening with >50% due to dFLC<5mg/dL. Among those with dFLC<5mg/dL, the majority met criteria for high-risk dFLC progression or organ progression and were treated within 6 months. The current dFLC threshold for measurable disease excludes AL amyloidosis patients requiring therapy from clinical trials and results in trial patients not being representative of real-world practice. In the era of highly effective T cell redirecting therapies, we advocate a lower dFLC inclusion (>2mg/dL) to enable participation of patients with low-burden disease.

iMARIX for primary and secondary central nervous system lymphomas.

7086 Background: The Bruton tyrosine kinase inhibitor (BTKi) ibrutinib is used in relapsed/resistant central nervous system (CNS) lymphoma because it can cross the blood-brain barrier and is effective as a single drug. Its efficacy has been demonstrated in combination with rituximab and high-dose methotrexate. CNS lymphomas are aggressive lymphomas, and responses to single-agent therapies are generally poor or insufficient. Also, even well-established agents used in lymphoma treatment may sometimes be unavailable worldwide, and drug restrictions may be experienced worldwide. Thiotepa is one of these agents. In this case series, we summarized the iMARIX treatment in which we used ibrutinib in combination with high-dose methotrexate, cytarabine and rituximab instead of thiotepa for the treatment of primary/secondary CNS lymphomas. Methods: The participants diagnosed with primary/secondary CNS lymphoma and treated between August 2022 and December 2025, received rituximab 375 mg/ m2 (one day), 3.5 g/m2 methotrexate (one day), 2x2 g/ m2 cytarabine (two days), and ibrutinib 560 mg/day in each course. Ibrutinib was continued provided that platelet count is over 30.000/ mm3 and neutrophil count is over 500/ mm3. Patient data were collected retrospectively. Results: 17 patients were treated with iMARIX protocol during the study course. Male/ female ratio was 8 (47%)/ 9 (53%) and the median age was 53 (21- 71). 8 (47 %) of the participants had primary whereas 9 (53 %) had secondary CNS lymphoma. 4 (23.5%) participants died of either relapse or sepsis during the study period. The median overall survival was 16.3 months (4- 31.6). 13 (76,5 %) patients underwent consolidative autologous stem cell transplantation (ASCT). One patient experienced atrial fibrillation due to ibrutinib use. One patient experienced relapse after ASCT and was further treated with allogeneic stem cell transplantation. Conclusions: Although our patient series is small, the results suggest that the combination of ibrutinib with high-dose methotrexate, rituximab and cytarabine is a combination that may yield good clinical results with an acceptable tolerance and adverse effect profile.

Single-cell profiling of the tumor microenvironment in pMMR/MSS rectal cancer treated with neoadjuvant PD-1 inhibitor combined with FOLFOXIRI and radiotherapy: Insights from the PANFORTE trial.

3647 Background: Programmed cell death-1 (PD-1) inhibitor-based combination therapies have emerged as a promising neoadjuvant strategy for locally advanced rectal cancer (LARC). However, the underlying tumor microenvironment (TME) dynamics during multimodal chemoimmunoradiotherapy remain incompletely understood. This study leveraged single-cell RNA sequencing to systematically map the TME remodeling induced by neoadjuvant PD-1 inhibitor combined with chemoradiotherapy in the PANFORTE trial (NCT06099951), aiming to identify cellular correlates of therapeutic response and resistance in pMMR/MSS rectal cancer. Methods: Pre- and post-treatment tumor samples from patients with pMMR/MSS rectal cancer enrolled in the PANFORTE trial and treated with neoadjuvant FOLFOXIRI, PD-1 inhibitor, and radiotherapy were subjected to single-cell RNA sequencing. Following stringent quality control, 227,447 high-quality cells from 50 samples were retained and annotated into seven major lineages. Cellular compositional changes across response groups were quantified using the observed-to-expected (Ro/e) index. Fibroblast subpopulations were identified via non-negative matrix factorization (NMF), and pseudotime trajectory analysis of myeloid cells was performed with Monocle 2. Results: Single-cell profiling revealed marked TME remodeling after neoadjuvant therapy. Complete responders (CR) showed significant stromal reduction, while non-responders (NCR) retained higher proportions of myeloid cells, endothelial cells, and fibroblasts. CR tumors exhibited enrichment of effector populations (NK cells, CD8⁺ effector memory T cells), whereas NCR tumors expanded immunosuppressive subsets (regulatory T cells, naïve T cells). Fibroblast subclustering identified inflammatory cancer-associated fibroblasts (iCAFs) as enriched in NCR. Pseudotime analysis demonstrated that myeloid cells in CR differentiated toward M1-like states, while NCR samples were biased toward M2-like programs. Conclusions: This study delineates distinct TME landscapes underlying differential responses to neoadjuvant FOLFOXIRI, PD-1 inhibitor, and radiotherapy in pMMR/MSS rectal cancer. Responders are characterized by cytotoxic immune activation and stromal depletion, whereas non-responders maintain an immunosuppressive milieu rich in iCAFs and M2-like myeloid cells. These findings provide mechanistic insight into treatment resistance and highlight potential cellular targets for enhancing therapeutic efficacy. Clinical trial information: NCT06099951 .

Prevalence of microsatellite stable Lynch syndrome and universal germline testing: A pan-cancer analysis of 246,694 patients.

10613 Background: Lynch Syndrome (LS) is an inherited cancer syndrome that confers up to a 60% lifetime risk of colorectal (CRC), endometrial (EC), or other cancers, and up to a 40% risk of a second primary cancer. The current standard-of-care includes LS screening upon the diagnosis of CRC or EC (especially in patients < 50 years old or with a strong family history) via immunohistochemistry for mismatch repair (MMR) protein deficiencies or microsatellite instability (MSI) testing algorithms; germline genetic testing is subsequently performed in MMR-deficient cases for confirmation . To assess the concordance between MSI status and LS, we evaluated the proportion of patients with pathogenic or likely pathogenic germline variants in a LS gene (LS-PGVs) who had microsatellite stable (MSS) tumors. Methods: We analyzed a pan-cancer cohort (n = 246,694) of patients who underwent personalized ctDNA testing (Signatera, Natera, Inc.), which involved whole exome sequencing (WES) of tumor and matched normal samples. MSI status was assessed in the tumor WES data by the MSISensor2 algorithm (MSI-H was defined as score ≥6.5). Germline single-nucleotide variants and short insertions/deletions from germline WES were identified and annotated with Variant Effect Predictor (v.105). The pathogenicity of variants in LS genes ( EPCAM , MSH2 , MSH6 , MLH1 , and PMS2 ) was determined using classifications available in the ClinVar database. Results: The pan-cancer cohort largely consisted of patients with CRC (n = 107,907; 44% of cohort), breast (n = 55,330; 22%), and other cancer types (each ≤4% of cohort) including EC (n = 5519; 2%). MSI status was determined in 84% (207,536/246,694) of the cohort, of whom 93% (193,110/207,536) had MSS tumors. LS-PGVs were identified in 1% (2418/207536) of the MSI-scored pan-cancer cohort. Among all pan-cancer patients with LS-PGVs, 36% (862/2418) had MSS tumors. Most patients with EPCAM PGVs (91%; 32/35) had MSS tumors. Conversely, only 17% (101/588) of patients with MLH1 PGVs had MSS tumors. We next assessed concordance between LS-PGVs and MSI status in CRC and EC. Among CRC patients with an MSI score (n = 97,309), 12% (11,752/97,309) had MSI-H and 88% (85,554/97,309) had MSS tumors. Further, LS-PGVs were identified in 2% (1,759/97,309), of whom 25% (437/1,759) had MSS tumors. In EC patients with MSI scores (n = 4,625), 20% (926/4625) had MSI-H and 80% (3699/4625) had MSS tumors. Of these, LS-PGVs were identified in 2% (86/4625) among whom 44% (38/86) had MSS tumors. Conclusions: LS-PGV yield was 2% for both CRC and EC, and a substantial proportion of these patients had MSS tumors (CRC: 25%; EC: 44%). Given the high risk for secondary cancers associated with LS and the implications for cascade testing, these findings strengthen the existing recommendation for considering broader germline screening of all EC and CRC patients for LS.

Predictors of mortality in hospitalized patients with metastatic non-small cell lung cancer (mNSCLC) with or without idiopathic pulmonary fibrosis (IPF).

e20703 Background: mNSCLC is associated with poor survival, with 5-year rates < 15%. Concurrent IPF further worsens prognosis and increases the risk of hospitalization due to respiratory complications. However, factors associated with in-hospital mortality among patients with mNSCLC, especially those with IPF, remain poorly understood. We examined patient- and hospital-level predictors of in-hospital mortality among patients with mNSCLC with and without IPF. Methods: Retrospective cohort analyses were conducted using data from the National Inpatient Sample (NIS) from 2017 to 2021. Adult (≥18 years) inpatient hospitalizations with a diagnosis mNSCLC were identified using ICD-10 CM codes and stratified by the presence of a concurrent diagnosis of IPF. Descriptive analyses and standard tests of association were performed. Multivariable logistic regression models were used to examine the factors associated with in-hospital mortality among patients with mNSCLC, stratified by IPF status. Results: Of 1,564,000 hospitalizations with mNSCLC, 13.6% had a concurrent diagnosis of IPF. In-hospital mortality was higher among patients with IPF compared to those without IPF (58% vs 42%, p < 0.001). Among patients with IPF, being non-Hispanic Black (NHB) (Adjusted odds ratio [AOR] 1.80, 95% CI 1.20–2.60) or Hispanic (AOR 1.60, 95% CI 1.20–3.10) was associated with higher odds of in-hospital mortality compared to being non-Hispanic White (NHW). While age ≥65 years was associated with higher mortality odds relative to age < 65 (AOR 2.25, 95% CI 1.40–3.79), having private insurance was associated with lower odds compared to being on Medicare (AOR 0.43, 95% CI 0.35–0.78). Patients who received care at urban teaching hospitals had lower risk of mortality than those treated at rural hospitals (AOR 0.34, 95% CI 0.20–0.80). Notably, those with acute respiratory failure (ARF) were more likely to have in-hospital mortality than patients without ARF (AOR 1.86, 95% CI 1.24–2.85). However, among patients without IPF, NHB race was associated with lower odds of in-hospital mortality compared to NHW race (AOR 0.23, 95% CI 0.10–0.60). Across both cohorts, prolonged length of stay ( > 5 days), endotracheal intubation, and higher comorbidity burden were associated with increased mortality. Conclusions: Our findings demonstrate that hospitalized patients with mNSCLC and concurrent IPF experience high in-hospital mortality, with notable disparities identified by age, race and ethnicity, insurance status and hospital type. The differences in mortality outcomes by care setting suggest that health system-level factors may influence survival in this vulnerable population. Further investigation into the drivers of these disparities is needed to shape clinical interventions and inform policy decisions to improve outcomes.

Behavioral gains offset by metabolic drift: A 34-year global profile of breast cancer risk attribution.

e12765 Background: Breast cancer (BC) remains one of the most common malignancies globally and a leading cause of female mortality and disability. Although advancements in screening and treatment have improved outcomes in high- income countries, the burden attributable to modifiable risk factors—particularly behavioral risks (BRs) and metabolic risks (MRs) varies substantially over time, regions, and socio-demographic strata. Methods: We utilized the Global Burden of Disease (GBD) 2023 framework to quantify global, regional, and age-specific BC burden attributable to BRs and MRs across 204 countries from 1990 to 2023. Outcomes included deaths, years of life lost (YLLs), years lived with disability (YLDs), and disability-adjusted life years (DALYs). Results: Globally, BC deaths attributable to all risk factors increased from 120,728 (95% UI: 71,625–165,691) in 1990 to 232,506 (133,617–322,736) in 2023 (EAPC +1.81%). YLLs rose from 3.55 to 6.43 million (+1.58%), and YLDs from 226,705 to 509,110 (+2.38%). Age-standardized mortality rates (ASMRs) declined in HI (−1.61%), High-Socio-demographic Index (SDI) (−1.51%), SEA, EA, and Oceania (−0.66%), High-middle SDI (−0.50%), and CE, EE and CA (−0.37%); remained stable in Low SDI (−0.17%); and increased in SA (+1.84%), NEMA (+1.21%), SA (+1.01%), Low-middle SDI (+0.61%), Middle SDI (+0.55%), and LA and Caribbean (+0.16%). From 1990-2023, BC burden attributable to BRs declined across all metrics: Deaths (EAPC −1.06%), DALYs (−0.97%), YLLs (−1.02%), and YLDs (−0.34%). Conversely, MR-attributable burden remained stable or slightly increased: Deaths (−0.27%), YLLs (−0.26%), YLDs (+0.41%). Age-stratified analysis showed that BR- attributable DALY rates declined in: 20–54 years (−0.35%), and ≥55 years (−1.04%). In contrast, MR- attributable DALY rates increased in 20–54 years (+0.88%) and slightly declined in ≥55 years (−0.16%). Conclusions: The global burden of BC attributable to BRs consistently declined, while that due to MRs remained stable or rose—particularly among younger and middle-aged adults. These findings highlight the shifting landscape of modifiable exposures and underscore the urgent need for targeted MR reduction strategies, especially in low- and middle-income countries (LMICs) where BC mortality is on the rise. Region Diet high in red meat High alcohol use High body-mass index High fasting plasma glucose Low physical activity Secondhand smoke Smoking High-income (HI) -1.59 -1.86 -0.99 -0.9 -1.44 -3.14 -2.26 Latin America (LA) and Caribbean 0.31 0.92 1.46 0.44 0.45 -1.83 -1.79 North Africa and Middle East (NEMA) 1.08 -2.4 2.83 2.83 1.39 0.86 0.19 South Asia (SA) 2 0.46 5.56 2.64 1.55 0.35 0.29 Southeast Asia (SEA), East Asia (EA), and Oceania -0.62 -1.18 0.98 -1.04 -0.2 -0.95 -2.05 Sub-Saharan Africa (SA) 1.8 1.71 3.54 2.99 1.94 1.01 0.17 Central Europe (CE), Eastern Europe (EE), and Central Asia (CA) -0.85 -1.02 0.5 0.29 -0.34 -1.94 -0.36

Trends in pediatric and adolescent thyroid carcinoma incidence and survival in the United States, 2001-2022.

e18102 Background: Thyroid carcinoma incidence in children, adolescents, and young adults has changed substantially over recent decades. Understanding trends in incidence and survival is crucial for clinical management and healthcare planning. Methods: Data from the North American Association of Central Cancer Registries (NAACR) and SEER registries (representing up to 76% of U.S. children, adolescents, and young adults) were analyzed for patients aged <40 years diagnosed between 2001-2022. Age-adjusted incidence rates per 1,000,000 were calculated and standardized to the 2000 U.S. population. Five-year relative survival rates were assessed for diagnoses from 2001-2020. Annual Percent Change (APC) and Average Annual Percent Change (AAPC) were calculated using Joinpoint Trend Analysis Software. Results: Thyroid carcinoma incidence in patients aged <40 years increased substantially from approximately 44 per 1,000,000 in 2001 to a peak of approximately 80 per 1,000,000 in 2014-2015, representing an 82% increase over this period. Subsequently, incidence rates declined to approximately 68 per 1,000,000 by 2022, a 15% decrease from peak levels. Despite the rise and subsequent decline in incidence, 5-year relative survival rates remained consistently excellent throughout the study period, maintaining rates of approximately 98-100% for all patients diagnosed between 2001-2020, with no statistically significant trend in survival outcomes. Conclusions: This analysis demonstrates a biphasic trend in pediatric and adolescent thyroid carcinoma incidence, with a sustained increase until mid-2010s followed by a decline in recent years. The initial rise likely reflects increased detection through enhanced imaging and surveillance, while the recent decline may represent stabilization of diagnostic practices. The consistently excellent survival rates exceeding 98% underscore the favorable prognosis of thyroid carcinoma in younger populations and the effectiveness of current therapeutic approaches. These findings have important implications for clinical management strategies and suggest that the recent decline in incidence may reflect more judicious use of diagnostic imaging and biopsy practices without compromising patient outcomes.

Molecular characteristics and outcomes of a real-world cohort of carcinoma of unknown primary determined by tissue-of-origin assays.

3076 Background: Carcinoma of unknown primary (CUP) increasingly intersects with precision oncology as molecular profiling becomes integrated into diagnostic algorithms. However, published CUP cohorts and guideline-based series remain anchored to clinical and pathologic criteria and do not characterize patients whose tissue-of-origin (TOO) predictions remain low probability despite comprehensive molecular analysis. This population—where molecular data fail to resolve lineage—is poorly described. We evaluated clinical, molecular, and outcome features of CUP defined as < 90% probability for all predicted lineages. Methods: A retrospective chart review of TOO tests ordered at a single academic center between 2021-2025 was conducted. Demographics, pathology, molecular features including Tumor Mutational Burden (TMB), mutational profiles, predicted lineages, and treatments were abstracted from the TOO report and electronic health record. Survival status was assessed as of 12/31/2025. Results: Among 132 patients evaluated for CUP, 50 met the study definition (TOO probability <90% for a single lineage). Median age was 66 years, 56% were female and 76% were Black. Predicted lineages spanned gastrointestinal, pancreaticobiliary, lung, breast, melanoma, and genitourinary tissues, reflecting heterogeneity and lack of dominant lineage signal. Many tumors demonstrated overlapping immunophenotypes consistent with molecular ambiguity. High TMB (≥10 mut/Mb) was reported in 16% with PD-L1 positive in 4%, limiting eligibility for tissue-agnostic immunotherapy. Most patients received empiric systemic therapy rather than site-directed or molecularly targeted regimens. Targetable alterations reported included BRAF, KRAS, HER2, ATM, CHEK2, and PIK3CA. TOO primary site predictions were in concordance with final pathology in only 22 cases (44%). At last follow-up, 29 patients (57%) were deceased. Conclusions: This study characterizes a precision-oncology–relevant subset of CUP in which molecular classifiers fail to assign high-confidence lineage, a population not represented in prior CUP publications. Tumors in this cohort showed broad predicted lineages across multiple organ systems with low rates of high TMB, reflecting substantial molecular heterogeneity. Many displayed nondistinct histology with overlapping immunophenotypes, consistent with diagnostic ambiguity. Concordance between predicted TOO and final pathology was limited, underscoring the challenges of relying solely on molecular classifiers. Actionable targets and markers for immunotherapy were infrequent, and most patients received empiric chemotherapy. Outcomes were poor, underscoring the unmet need for improved integrative molecular approaches and novel precision-oncology strategies in CUP subsets that remain unresolved despite comprehensive molecular assessment.

Impact of prior exposure to erythropoiesis-stimulating agents (ESA) on thromboembolic events in patients with myelodysplastic syndromes receiving luspatercept: A real-world TriNetX-based study.

e18585 Background: Luspatercept demonstrated superior transfusion independence over ESAs in ESA-naive Myelodysplastic Syndromes(MDS). Thromboembolic events were observed with luspatercept in both BELIEVE and COMMANDS trials, particularly among patients with baseline comorbidities. ESAs are also independently associated with higher thromboembolic risk. Whether prior ESA exposure modifies the subsequent thromboembolic risk during luspatercept therapy in older morbid MDS patients remains unknown. We conducted a real-world analysis to determine whether prior ESA exposure compounds the risk of venous thromboembolism and arterial thrombosis after luspatercept initiation in MDS. Methods: We conducted a retrospective cohort study using the TriNetX Global Collaborative Network, including adults with MDS treated with luspatercept and no prior venous or arterial thromboembolism. Patients were stratified into prior ESA-exposed versus ESA-naive cohorts. Propensity score matching (1:1) was performed using demographics, comorbidities/risk factors, anticoagulation/antiplatelet use, and baseline hematologic parameters. Outcomes were assessed at 90 days and 1 year; venous thromboembolism was defined as deep vein thrombosis and/or pulmonary embolism, and arterial thrombosis as acute myocardial infarction or ischemic stroke. Results: A total of 2,149 luspatercept-treated MDS patients were identified, out of which 1218 had prior ESA exposure and 931 were ESA-naïve. After propensity matching, 801 patients remained in each cohort. The overall mean age was 76.6 years, with 55.1% of patients being male. At 90 days interval, venous thromboembolism events occurred in 3.0% of the prior-ESA cohort versus 2.4% of the ESA-naive cohort (RR 1.22, p=0.521). Arterial thrombosis (acute myocardial infarction or ischemic stroke) occurred in 4.3% vs 2.7% (RR 1.60, p=0.090). At 1 year, venous thromboembolism occurred in 4.6% vs 4.7% (RR 0.98, p=0.906), and arterial thrombosis occurred in 7.4% vs 7.2% (RR 1.03, p=0.924). None of the between-group differences were statistically significant at either 90 days or 1 year. Conclusions: These findings suggest that prior ESA therapy does not compound the thrombotic risk associated with luspatercept and support the thrombotic safety profile of luspatercept even in prior ESA-exposed patients and warrant further prospective validation. Thromboembolic outcomes by prior ESA exposure. Time window (post-index) Outcome Prior ESA (%) ESA-naive (%) Effect estimate p-value 90 days Venous thromboembolism (DVT/PE) 3.0 2.4 RR 1.22 0.521 90 days Arterial thrombosis (MI/ischemic stroke) 4.3 2.7 RR 1.60 0.090 1 year Venous thromboembolism (DVT/PE) 4.6 4.7 RR 0.98 0.906 1 year Arterial thrombosis (MI/ischemic stroke) 7.4 7.2 RR 1.03 0.924

Beyond hematologic malignancies: Clonal hematopoiesis of indeterminate potential and the risk of solid tumors in the UK biobank.

10502 Background: Clonal hematopoiesis of indeterminate potential (CHIP) is an age-related condition characterized by somatic mutations in hematopoietic stem cells. While CHIP is associated with hematologic malignancies, its relationship with solid tumors remains incompletely characterized. We conducted a comprehensive analysis of CHIP and cancer risk in a large population-based cohort. Methods: We analyzed 409,150 UK Biobank participants with whole exome sequencing, including 13,709 (3.4%) with CHIP. CHIP was defined as somatic mutations in myeloid driver genes at variant allele frequency (VAF) ≥2%. Incident cancers were ascertained through registry linkage over median 15.5-year follow-up. We performed propensity score-matched Cox regression adjusting for age, sex, ethnicity, smoking, BMI, diabetes, chronic kidney disease, and family history. Gene-specific, VAF-stratified, and multi-hit analyses were conducted. Results: Among 77,695 incident cancers, CHIP was associated with increased risk of any cancer (HR 1.22, 95% CI 1.18-1.27), including both hematologic and solid tumors. Associations were strongest for myeloid neoplasms (HR 10.49, 95% CI 8.64-12.74), including AML (HR 11.09), MDS (HR 9.75), and MPN (HR 8.03). For solid tumors, significant associations were observed for lung cancer (HR 1.50, 95% CI 1.31-1.73), particularly adenocarcinoma (HR 1.79, 95% CI 1.48-2.16), and esophageal cancer (HR 1.52, 95% CI 1.13-2.03). Notably, esophageal squamous cell carcinoma showed particularly elevated risk (HR 2.28, 95% CI 1.37-3.81). VAF-stratified analyses revealed dose-response for esophageal cancer: large clones (defined as VAF ≥20%) conferred HR 2.23 (95% CI 1.32-3.75) and very large clones (defined as VAF ≥30%) HR 2.72 (95% CI 1.39-5.30). Multi-hit CHIP (≥2 mutations) was associated with HR 3.35 (95% CI 1.46-7.71) for esophageal cancer and HR 4.07 (95% CI 1.79-9.21) for esophageal adenocarcinoma. Gene-specific analyses showed TET2 mutations increased esophageal adenocarcinoma risk (HR 2.46, 95% CI 1.21-5.03). Results remained consistent in 5-year landmark analyses. Conclusions: CHIP is associated with significantly increased risk of both hematologic and select solid malignancies. We identify esophageal cancer, particularly squamous cell carcinoma, as a novel CHIP-associated malignancy with clear dose-response by clone burden. These findings warrant external validation and, if confirmed, suggest CHIP features may help refine risk stratification strategies for selected solid tumors.

Efficacy and safety of tremelimumab plus durvalumab (STRIDE) versus durvalumab monotherapy in advanced hepatocellular carcinoma: A systematic review and meta-analysis.

e14577 Background: Durvalumab monotherapy and the STRIDE regimen (tremelimumab plus durvalumab) are approved immunotherapy options for advanced hepatocellular carcinoma (HCC). While both have shown clinical benefit, direct comparisons are limited. Methods: PubMed, Cochrane, Embase, and Scopus were searched systematically for relevant studies from inception to May 2025. Risk ratios (RR) with 95% confidence intervals (CI) were pooled for dichotomous outcomes using Mantel-Haenszel method with random effect model. Heterogeneity was assessed using I² and χ² statistics. The quality of the studies was evaluated using the Cochrane Risk of Bias 2.0 (RoB 2.0) tool. Results: Three studies with 991 patients were included. No significant difference in 18-month survival between STRIDE and durvalumab (RR = 1.18, 95% CI:0.84–1.66; I² = 72%; P = 0.33). The overall response rate favored STRIDE (RR = 1.52, 95% CI:1.03–2.24; I² = 21%; P = 0.03). Partial response and complete response rates were also higher with STRIDE but not statistically significant (PR: RR = 1.44, 95% CI:0.99–2.09; I² = 17%; P = 0.06; CR: RR = 1.95, 95% CI: 0.82–4.68; I² = 0%; P = 0.13). Regarding safety, STRIDE was associated with higher risk of any-grade adverse events and grade 3 adverse events, (RR = 1.78, 95% CI:0.98–3.14; I² = 92%; P = 0.06 and RR = 1.92, 95% CI:1.37–2.68; P = 0.0001 respectively). There was no differences in liver toxicity (ALT: RR = 1.41, 95% CI: 0.46–5.25; I² = 79%; P = 0.56; and AST: RR = 1.25, 95% CI: 0.55–2.88; I² = 64%; P = 0.59) and diarrhea incidence (RR = 1.87, 95% CI: 0.75–4.68, I² = 69%; P = 0.18) between both groups. However, treatment discontinuation due to adverse events (RR = 1.92, 95% CI: 1.07–3.46; I² = 0%; P = 0.03) were higher with STRIDE. Conclusions: STRIDE showed a significantly higher ORR than Durvalumab in advanced HCC, with comparable survival and manageable toxicity. However, further large-scale RCTs are needed to confirm these findings and establish STRIDE as a standard treatment option. Key Words: HCC, STRIDE, Durvalumab.

The oral medicine–oncology nexus: Periodontal disease as a predictor of survival and complications in immunotherapy-treated melanoma.

e21599 Background: Immune checkpoint inhibitors have transformed melanoma treatment, but comorbidities like periodontal disease—pre-existing or immune-related adverse events—may modulate efficacy and toxicity through systemic inflammation and immune dysregulation. This study examined associations between periodontal disease (any-time diagnosis) and clinical outcomes, including mortality and systemic complications, in melanoma patients receiving immunotherapy. Methods: This retrospective cohort study used the TriNetX federated network of deidentified electronic medical records from 88 healthcare organizations. Adults (aged ≥18 years) with melanoma (ICD-10-CM C43-C44) receiving immunotherapy (pembrolizumab or antineoplastic immunotherapy encounter) were stratified by presence (n = 703) or absence (n = 37,419) of periodontal disease (ICD-10-CM K05). Propensity score matching (1:1) balanced race, communicable disease hazards, body mass index, and Eastern Cooperative Oncology Group performance status, yielding 698 patients per group. Analysis was performed January 24, 2026. The primary outcome was all-cause mortality, with secondary outcomes of circulatory (I00-I99), endocrine/metabolic (E00-E89), respiratory (J00-J99), and digestive (K00-K95) diseases; fever of unknown origin (R50); and symptoms involving food/fluid intake (R63), circulatory/respiratory systems (R00-R09), and digestive system/abdomen (R10-R19). Outcomes were assessed from 1 day post-index event (melanoma diagnosis) onward. Analyses included risk differences, ratios, odds ratios, Kaplan-Meier survival (log-rank), Cox hazard ratios, and instance counts excluding prior events. Results: Post-matching, mortality was higher in the periodontal group (41.7% vs 34.8%; risk difference −0.069 [95% CI, −0.120 to −0.018]; P = .008; risk ratio 0.835 [95% CI, 0.730-0.955]; odds ratio 0.746 [95% CI, 0.601-0.927]) with shorter median survival (1271 vs 1626 days; log-rank P = .002; hazard ratio 0.763 [95% CI, 0.643-0.906]; P = .016). Significant associations included fever of unknown origin (hazard ratio 0.682 [95% CI, 0.495-0.939]; log-rank P = .018) and food/fluid intake symptoms (22.3% vs 14.8%; risk difference −0.075 [95% CI, −0.125 to −0.024]; P = .003; hazard ratio 0.603 [95% CI, 0.445-0.817]). No significant differences emerged for circulatory (P = .288), endocrine/metabolic (P = .268), respiratory (P = .999), or digestive symptoms (P = .756 for digestive symptoms; disease analysis precluded due to 0 events in periodontal group). Conclusions: Periodontal disease is linked to poorer survival and select complications in immunotherapy-treated melanoma patients. These findings advocate for routine periodontal evaluation and intervention to potentially enhance oncologic outcomes.

A single-arm, phase II study of ivonescimab in combination with FOLFOX in advanced HER2-negative gastroesophageal adenocarcinomas.

TPS4246 Background: Gastroesophageal adenocarcinoma (GEA) is a major cause of global cancer-related mortality. A common standard of care for frontline advanced GEA is 5-fluorouracil and oxaliplatin (FOLFOX), with the addition of immune checkpoint inhibitors (ICI) in patients with a PD-L1 combined positive score (CPS) of > 1. Adding ICIs improves median overall survival (OS) by 2-4 months, with increased benefit seen with higher CPS scores (Leone et al, ESMO Open , 2024). There remains a significant need to improve ICI outcomes and expand the portion of patients who benefit. In GEA, vascular endothelial growth factors (VEGF) are associated with poor prognosis, influencing immune suppression and T cell exhaustion. There is strong mechanistic rationale for cooperativity between anti-VEGF antibodies and ICI in GEA, as well as evidence of clinical benefit in treatment-refractory gastric cancer (Kim et al, Cancer Immunol Res , 2025). Ivonescimab is a novel bispecific antibody that simultaneously targets PD-1 and VEGF, in a spatially constrained manner, impairing neoangiogenesis and remodeling the suppressive tumor microenvironment. Ivonescimab has evidence of safety and efficacy in colorectal and hepatocellular cancers and is approved for the treatment of PD-L1+ lung cancer in China (Deng et al, ESMO 2024; Li et al, ESMO 2025; Xiong et al, Lancet , 2025). Methods: This single-arm, multi-institutional phase II study seeks to evaluate ivonescimab in combination with FOLFOX as frontline therapy in GEA. Adults with locally advanced unresectable or metastatic HER-2 negative GEA are eligible. Patients must have disease that can be evaluated radiographically and may not have received prior systemic therapy for advanced disease. Positive PD-L1 expression is not required for eligibility. We plan to enroll 40 patients. To allow for collection of early immune modulation data, patients will commence therapy with a single dose of ivonescimab 20 mg/kg at discretion of the treating investigator. Subsequent cycles will continue with the addition of FOLFOX (dosed per institutional standards) to ivonescimab, given on day 1 and day 15 of a 28-day cycle. Imaging will be performed every 6 weeks until week 24, at which point imaging will change to every 9 weeks. Treatment will continue until disease progression, withdrawal of consent, or death. The primary objective is to define efficacy of ivonescimab in combination with FOLFOX as measured by the 6-month progression-free survival (PFS) rate. Secondary objectives include safety and tolerability, objective response and clinical benefit rates, median PFS and OS. Translational correlative studies are planned to explore tumor and immunologic determinants of response. This study is currently open and enrolling (NCT07070466). Clinical trial information: NCT07070466 .

Pan-cancer landscape and real-world management of BRAF non-V600E and MAP2K1 alterations: Analysis from the WAYFIND-R registry.

3080 Background: While targeted therapies have revolutionized the management of BRAF V600E-mutant tumors, non-V600E BRAF and MAP2K1 alterations remain a clinical "gray zone." These variants represent a functionally diverse group of drivers with no established treatment guidelines, often leading to heterogeneous care and suboptimal outcomes. We used the multinational WAYFIND-R registry to map the prevalence and functional landscape of these rare alterations and to evaluate real-world treatment patterns. Methods: We retrospectively analyzed 4,774 patients with solid tumors from the multinational WAYFIND-R registry (NCT04529122). Patients with BRAF and MAP2K1 alterations identified via Next-Generation Sequencing (NGS) were included. Variants were categorized by functional class (Class I-III). Clinical characteristics, co-mutation profiles, and treatment sequences were analyzed to correlate molecular features with management strategies and outcomes. Results: Rare alterations were identified in 38 (0.8%) patients, predominantly in colorectal (29%), lung (13%), and melanoma (13%) cancers. Genomic analysis revealed a complex landscape. Among BRAF-altered patients (n=30), while Class I (V600E/K) accounted for 63.3% (n=19), the remaining 36.7% comprised non-V600E variants distributed between functional Class II (20.0%) (e.g., G469A, L597S) and Class III (13.3%) (e.g., D594N, G466R). MAP2K1 alterations (n=9) were distributed across RAF-dependent Class I (44.4%), RAF-regulated Class II (44.4%), and RAF-independent Class III (11.1%) variants (e.g., E102_I103del). Treatment patterns were highly inconsistent. While first-line therapy typically consisted of standard chemotherapy or anti-PD-1 monotherapy, there was no consensus for subsequent lines. This lack of targeted approach correlated with poor outcomes in lung and colorectal cohorts. However, in contrast to unselected treatments, we observed durable responses with biology-guided MEK inhibitor plus Immunotherapy combinations in two index cases. A patient with MAP2K1 Class III (E102_I103del) NSCLC achieved a near-complete metabolic response with cobimetinib + nivolumab. Similarly, a BRAF Class II (G469A) NSCLC patient achieved a complete metabolic and molecular response (confirmed by ctDNA clearance) with trametinib + pembrolizumab. Conclusions: This analysis confirms that >35% of BRAF-altered patients (nonV600 mutants) and all MAP2K1-altered patients in real-world settings currently lack standardized treatment protocols. The observed clinical heterogeneity highlights a significant unmet need. Our findings suggest that integrating functional subclassification into clinical decision-making and utilizing targeted combinations is necessary and may offer a viable strategy. Clinical trial information: NCT04529122 .

Methylome-wide association study in blood suggests cell type-specific relationships between DNA methylation and lifetime anxiety

Compartment-specific spatial transcriptomics to reveal microenvironmental programs driving omental metastasis in high-grade serous ovarian cancer.

e17601 Background: High-grade serous ovarian cancer (HGSOC) frequently metastasizes to the omentum, yet spatially distinct transcriptional programs within tumor and stromal compartments remain incompletely characterized. Methods: To generate a high-resolution map of these microenvironments, we applied NanoString GeoMx Digital Spatial Profiling to matched primary–omental metastatic HGSOC samples from 15 patients. A nuclear marker and pan-cytokeratin positivity differentiated epithelial tumor regions from surrounding stroma. Digital segmentation enabled compartmentalized whole-transcriptome sequencing, allowing direct comparison of gene expression between matched primary and metastatic disease with specificity for stromal versus tumor compartments. We identified the top 25 differentially expressed mRNA signals from tumor and stroma compartments (p<0.05). These biomarkers were correlated with the Kaplan-Meier Plotter dataset (including GEO, EGA, TCGA, Metabric, Impact, and PubMed) to assess expression–overall survival associations (p<0.05) and generate a candidate gene set driving omental carcinomatosis. Results: Six primary pathways associated with decreased overall survival were identified (Table 1): (1) increased tumor migration and extracellular matrix remodeling; (2) tumor stress hormone upregulation with immune suppression; (3) stromal fibroblast activation; (4) increased stromal GABA transporter; (5) downregulation of hypoxia-driven tumoral pathways; and (6) loss of epigenetic surveillance in stroma. Conclusions: The resulting transcriptional map supports biomarker discovery and identification of actionable pathways in HGSOC. Pathway Genes mRNA Fold Change (Omental Mets vs. Primary) P value (Omental Mets vs. Primary) Survival Difference in Months (High vs. Low Expressors) P Value (High vs. Low Expressors) Tumor migration and remodeling of ECM ENPP2; FN1; TAGLN; CRTAC1 1.44; 2,11; 1.61; 1.55 0.009479; 0.020053; 0.006779; 0.005031 15 vs 17; 14 vs 17; 14 vs 18; 13 vs 16 0.0399; 0.0489; 0.0003; 0.0063 Tumor stress and immune regulation TSC22D3; FKBP5 1.51; 1.53 0.023870; 0.021592 14 vs 18; 13 vs 16 0.0132; 0.0104 Stromal fibroblast activation CD109; ZDHHC7; SMAP2 1.55; 1.61; 1.59 0.001646; 0.000366; 0.000625 13 vs 16; 15 vs 18; 11 vs 15 0.0307; 0.0210; 0.0041 Stromal GABA transporter SLC6A11/GAT3 1.73 0.000256 16 vs 18 0.0029 Downregulation of HIF EGLN3 0.58 0.008467 18 vs 15 0.0170 Loss of replication surveillance MCM6; HELLS; TTC28 0.57; 0.62; 0.71 0.000032; 0.000102; 0.001598 18 vs 16; 17 vs 16; 17 vs 15 0.0116; 0.0075; 0.0198