Citations

Abstract Introduction: COE has become the gold standard in addressing health disparities. COE’s value is emphasized by NIH and leading cancer-related organizations e.g., AACR. The Alliance for COE (The Alliance) mission is to advance the art, science, and impact of COE; promote community health and wellness; and reduce cancer burden and disparities. Methods: This abstract builds on the Alliance’s Science of Community Outreach and Engagement (SoCOE) Conference a NCI R13 funded series. SoCOE Conference is organized by 10 NCI-designated Cancer Centers to provide a platform for COE academic-clinical-policy-community stakeholders to share best practices for solution-focused community responsive cancer prevention, diagnostics, therapeutics and survivorship research and practice. Results: > 500 researchers, clinicians, policymakers, healthcare administrators, with >12% being survivors and/or advocates, participated in our SoCOE Conferences. The delegates’ recommendations are: Decentralize who/entities that hold all the power in science and clinical research (trials). Communities hold valuable, necessary wisdom and resources to inform and guide science Include multisectoral partners early and always Community capacity building and compensation with gratitude Mentoring of junior COE researchers, clinicians, policymakers, healthcare administrators, survivors and advocates with humility is a must Develop virtual communities to share best practices across cancer centers and within regions Develop working groups to identify and explore commonalities across COE work in rural/urban, diverse/homogenous spaces Publish Science of COE papers on various topics related to COE Frameworks and evidenced-solutions Provide community reports on best practice guide that can be widely available to be shared locally and in-person, and digitally Partner and learn from COE teams working with other cultural/ethnic groups Attend to, measure and remedy societal/social drivers of health disparities that mirror barriers to care, and research engagement and participation. Discussion: Informed by our community advisory boards and the findings for our SoCOE Conferences, we present the pillars of the Alliance to tackle population cancer burden and disparities: Training future COE leaders/advocates, Advocating for public health policies/guidelines, Increasing access to cancer prevention and care that will also increase access to studies, Engaging and partnering with multisectoral stakeholders and communities, Building community capacity, Advancing COE science through catchment-relevant research, and Leading bi-directional communication and information and resource sharing between catchment area communities and cancer centers. The expectation and impact of COE can only be achieved if COE teams are valued, supported cancer center members with community advisory boards and community as cancer centers’ priority population. Citation Format: Kimlin Ashing, CHARNITA ZEIGLER-JOHNSON, Hayley S. Thompson, Kim Rhoads, Nadine Barrett, Lisa Carter Bawa, Timiya S. Nolan, Monica Baskin, Vanessa B. Sheppard, Marvella E. Ford, Erica Phillips, Lorna H. McNeill, Folakemi T. Odedina. Advancing the science, practice and impact of COE [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7577.

Abstract Background: Prostate cancer (PCa) is the leading cause of cancer-related deaths globally and disproportionately affects men of African ancestry. African men often present at younger ages and with more aggressive disease, and approximately half of the PCa risk has been attributed to genetic factors. Africa is the most genetically diverse region of the world, and even within the continent, populations show important differences in allele frequencies and patterns of pathogenic variation. Despite this diversity, African men remain underrepresented in genomic studies, creating significant gaps in knowledge that limit precision medicine efforts. Methods: A total of 182 prostate cancer patients were recruited from teaching hospitals in Benin, Côte d’Ivoire, Ghana, Kenya and Nigeria. 50 cancer susceptibility genes using the Aviseq ONCO50 panel and the Illumina MiSeq platform. Variant calling was performed using an in-house Snakemake workflow, followed by additional bioinformatic steps to distinguish true PMS2 variants from PMS2CL-related artifacts. Pathogenicity assessment relied on annotations from ClinVar and LOVD databases. Results: Pathogenic and likely pathogenic variants were detected in 31.3% of the study group identified in 26 genes. The five most frequently altered genes were RAD50, PTCH1, MSH6, MUTYH, and NF1. Nigerian patients accounted for 52.6 percent of all pathogenic variants, with recurrent alterations observed in MSH6, MUTYH, NF1, and PTCH1. In contrast, pathogenic BRCA1 and BRCA2 variants were observed primarily in patients from Kenya, Benin, and Côte d’Ivoire. Several frameshift, nonsense, and splice-site variants identified in this study have not been previously reported in African populations.[SR1] Conclusion: This study provides one of the most detailed assessments of inherited cancer susceptibility gene variants in African PCa patients. The findings reveal substantial regional variation in the spectrum of pathogenic variants and highlight population-specific genomic signatures that may contribute to the unequal burden of PCa in men of African ancestry. These data emphasize the importance of expanding genomic research efforts within Africa to strengthen precision oncology and reduce global PCa disparities. Citation Format: Abimbola F. Onyia, Freeman Okwuchi, Opeyemi C. De Campos, Sylvester Divine, Olutola Olasehinde, Ayinde Yahaya, Ayo Salako, Aminu Zakari, Iya Bassey, Nicholas Titiloye, Bernard Petershie, Isidore GANDAHO, Luc Brun, Tore Sanni, Safiatou Coulibaly, Noel Coulibaly, Michael Fakayode, Coulibaly Issoufou, Valérie Gbonon, Kouame Benjamin, Yao Evrard, Peter Coleman, Ayun Cassell, Victor Ajumbo, Lilac Wattanga, Benson Ochieng Nyambega, Folakemi T. Odedina, Solomon Rotimi. Targeted sequencing of common cancer susceptibility genes reveals germline variants in African prostate cancer patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6289.

Abstract Background: Accurate staging is essential for treatment selection, prognosis assessment, and trial eligibility. While prior studies attempted automating pathological (p) staging, few address clinical (c) staging and no studies have compared LLM-derived staging with clinician staging during clinical visit and retrospective cancer registry staging. Methods: We developed a multi-agent framework using Gemini-2.0-Flash-001 and GPT-5 to (1) identify reports (2) extract key data, and (3) apply AJCC 8th edition staging criteria. LLM staging was compared with clinician documentation and registry staging in breast cancer patients. Clinician-registry agreement served as the reference standard. Two independent breast oncologists reviewed discordant cases. Results: We analyzed 122 randomly selected breast cancer patients across all three Mayo Clinic sites from 2018 - 2023. LLM performance matched or exceeded human inter-rater agreement for pathological staging (95-99.2% vs. 95-98.4%). Clinical staging was more challenging, with LLM concordance of 73-77.9% for cT and 87-89.3% for cN versus 87.8% and 91.0% clinician-registry agreement. Among 89 patients with clinician-registry concordance, LLM concordance was 98.9% for pT/pN, 79.8% for cT, and 91.0% for cN. Experts favored LLM staging in 27.8% (5/18) of cT and 25.0% (2/8) of cN discordances. Error analysis revealed LLM challenges in handling non-mass enhancements (6 cases), selecting radiologic measurements (2 cases), identifying discrete masses (2 cases), and miscellaneous (3 cases). Conclusion: LLMs achieved human-level performance for pathological staging (98%), supporting human-in-the-loop deployment. For clinical staging (79.8% cT, 91.0% cN), future work must enhance multimodal reasoning and integrate physical examination data. Prospective validation is needed to assess real-world impact. With targeted refinements and oversight, automated staging can transform registry workflows while augmenting clinical decision-making. Citation Format: Arshad Mohammed, Umair Ayub, Pooja Advani, Shakeela W. Bahadur, Amye J. Tevaarwerk, Tufia C. Haddad, Elisabeth I. Heath, Brenda J. Ernst, Ben Zhou, Cui Tao, Sara J. Holton, Karthik V. Giridhar, Irbaz B. Riaz. Automating clinical and pathological staging for breast cancer patients [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2744.

Abstract The CoMMpass study (NCT01454297) is a prospective, longitudinal observational study involving 1,141 newly diagnosed multiple myeloma (NDMM) patients. Bone marrow aspirates from the study participants were collected for comprehensive molecular characterization of tumor (using bulk RNA sequencing and whole genome sequencing) and tumor immune microenvironment using 3’ single-cell RNA sequencing (scRNA-seq). To support exploration of this rich multi-omic dataset, we developed CoMMpass Explorer (CE), an intuitive interactive platform that enables real-time analysis of clinical and genomic data from CoMMpass. A central feature of CE is cohort building, where users can filter and stratify patients by clinical, genomic, and survival data elements to create custom cohorts for analysis. CE provides five main views: Overall Summary for clinical feature distribution, Kaplan-Meier curves, and multivariate Cox proportional hazards models; Mutational Profile for visualizing and comparing somatic mutations; Tumor Profile for bulk differential expression and gene set enrichment; Immune Microenvironment for comparing cell type abundance and cell cycle dynamics between cohorts using scRNA-seq data; and Pseudo-Bulk, which aggregates single-cell expression by cell type to enable cohort-level comparisons of transcriptional programs using DEseq2. CE reproduces prior CoMMpass studies, including WEE1 expression being associated with poorer progression-free survival (Simhal, et al.) and different mutation frequencies between African American and European American patients (Manojlovic, et al.). We further demonstrate CE’s utility by interrogating the 2025 Consensus Genomic Staging (CGS) risk classification system (Avet-Loiseau, et al). We compared CGS high-risk (HR, n=249) vs. CGS standard-risk (SR, n=573) patients. HR patients have significantly worse progression-free survival, overall survival, and time to second-line therapy compared with SR patients. Univariate models recapitulated expected differences in the clinical features that define CGS risk. Bulk tumor RNA-seq and ssGSEA highlighted a more proliferative and genomically unstable transcriptional program in HR group, with upregulation of angiogenesis and cell-cycle-related pathways. In the immune microenvironment, monocytes were more abundant in the SR group. Pseudo-bulk RNA-seq of plasma cells showed overexpression of ANXA1, NSD2, IGF1R, MAF and related genes in HR. CE democratizes access to the CoMMpass multi-omic resource, allowing researchers to interrogate clinical and molecular heterogeneity. By enabling hypothesis generation and replication of published results, CE serves as a valuable tool for identifying prognostic biomarkers and meaningful patterns in gene expression, mutation profiles, and the immune microenvironment. Citation Format: Wanxing Zhang, Chaitanya R. Acharya, Steven M. Foltz, David E. Avigan, Samir Parekh, Ravi Vij, Shaji Kunnathu Kumar, Taxiarchis Kourelis, Sagar Lonial, Hearn Cho, Ioannis S. Vlachos, Sacha Gnjatic, Li Ding, Manoj Bhasin, George Mulligan. CoMMpass Explorer: An interactive platform to explore clinical and genomic data from newly diagnosed multiple myeloma patients from the landmark CoMMpass observational study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 5504.

Abstract Background: The increasing volume and diversity of clinical genomic reports pose a significant challenge across healthcare institutions for the interpretation and proper implementation of precision oncology research. Reports from multiple different vendors (e.g., Invitae, Ambry Genetics, Foundation Medicine) are typically PDFs and exhibit substantial heterogeneity in panel design, gene coverage, and reporting standards, which hinders efficient retrospective data mining and patient cohort identification within the electronic medical records. We sought to develop a framework to interrogate all of these reports. Methods: We extracted genomic reports from patients with breast cancer treated with neoadjuvant chemotherapy and developed MolHarmonizer, a novel, scalable framework leveraging Python and Gemini LLMs, designed to process and harmonize genomic data from disparate multi-vendor reports. Gemini LLMs are employed explicitly for robust information extraction, normalization, and structuring of key genomic features, transforming unstructured data into a unified, queryable dataset. Results: Our MolHarmonizer framework successfully processed 1,147 genomic reports from 1703 breast cancer patients (2006-2023) from 23 different companies, demonstrating robust capability to extract and standardize critical actionable biomarkers. Data sources included Invitae (n=554), Ambry Genetics (n=189), Natera (n=95), Mayo Clinic (n=88), Tempus (n=63), Guardant Health (n=47), and Foundation Medicine (n=37), with others contributing less than 20 reports. Of the samples, 827/1147 (72.1%) were germline (blood/saliva). A majority of the patients were tested using the panels due to a personal/family history (n=925). Overall, 413/1147 (36.0%) reports identified at least one mutation. For breast cancer, 75 reports showed BRCA1/2 mutations (37 BRCA1, 37 BRCA2, and one patient with both BRCA1 and 2). Other mutations identified included: PIK3CA (n=40), TP53 (n=96), PTEN (n=21), ESR1 (n=13) and AKT1/2 (n=8). Conclusion: MolHarmonizer, a powerful framework leveraging Gemini LLMs, effectively addresses genomic data heterogeneity by automating biomarker extraction and harmonization. This enables rapid cohort identification and deep retrospective analyses for clinical insights, biomarker discovery, understanding disease history, facilitating novel pattern discovery, e.g., predicting BRCA1 mutations from WSI, and accelerating research within our neoadjuvant BC cohort. Future plans include expanding to include over 20,000 breast cancer patients, developing a user-friendly chatbot, and ensuring inter-institutional adaptability for a variety of complex diseases. Citation Format: Krishna Rani Kalari, Xiaojia Tang, Thanmayee Boyapati, Tanya L. Hoskin, Sumathilatha Myla, Sumedha G. Penheiter, Richard M. Weinshilboum, Liewei Wang, Hamid R. Tizhoosh, Karthik Vikram Giridhar, Matthew P. Goetz, Judy C. Boughey. Development of an LLM framework for analysis of heterogeneous breast cancer patients genomic reports [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2756.

Abstract Introduction: The National Cancer Institute estimated almost 317,000 new breast cancer and over 13,000 new cervical cancer cases in 2025. While the National Breast and Cervical Cancer Early Detection Program has served millions of women over the past two decades, inequities in cancer screening, subsequently delayed diagnosis, and timely and appropriate treatment remain a public health challenge in the United States. Despite the substantial impact of social determinants of health (SDOH) on cancer screening, few studies have accounted for SDOH at the individual level when assessing breast and cervical cancer screening adherence. The primary objective of this study is to examine whether the individual-level SDOH, measured using HOUsing-based SocioEconomic Status (HOUSES) index, is associated with breast and cervical cancer screening non-adherence (not current on respective USPSTF recommendations for age and frequency). Methods: In this retrospective study, patients were identified as those receiving primary care in the Mayo Clinic Health System and Mayo Clinic (Rochester, MN). The Midwest Quality Metrics panel was then used to assess patients due for screening for breast or cervical cancer in July 2023. The HOUSES, using home address and publicly available data on the residence which measures and categorizes housing status into quartiles, was obtained using the electronic health records (EHRs)-linkable patient-level information. Logistic regression analyses were conducted, controlling for Census block group-level adverse social exposome from the Area Deprivation Index, age, and comorbidities. Results are presented as odds ratios (OR) with corresponding 95% confidence intervals (CI). Results: For breast cancer, among 128,462 eligible patients, 20.1% (n= 25,857) were non-adherent. Compared to the highest HOUSES quartile (Q4), lower HOUSES groups showed significantly higher odds of non-adherence: Q3 OR=1.20 (95% CI=1.15, 1.25), Q2 OR=1.45; (95% CI=1.40, 1.53), and Q1 OR=1.90 (95% CI=1.81, 1.99). For cervical cancer, among 175,712 eligible patients, 29.9% (n=52,533) were non-adherent. Compared to the highest HOUSES quartile (Q4), lower HOUSES groups showed significantly higher odds of non-adherence: Q3 OR=1.07 (95% CI=1.03, 1.10), Q2 OR=1.16; (95% CI=1.12, 1.20), and Q1 OR=1.36 (95% CI=1.31, 1.41). Conclusions: A clinically meaningful association between the HOUSES index and cancer screening adherence was observed. A lower HOUSES index was associated with higher odds of non-adherence for both breast and cervical cancer screening, even after accounting for neighborhood-level risk, age, and comorbidities. Incorporating HOUSES into EHRs can be a practical “screening-for-screening” tool that enables close monitoring of higher-risk patients without additional burden of surveys. Citation Format: Eunice Y. Park, Madison Beenken, Dave Watson, Chung Il Wi, Tufia Haddad, Gladys Asiedu, Karla BALLMAN, James R. Cerhan, Carolyn Flock, Brian Lynch, Folakemi T. Odedina, Scott H. Okuno, Young J Juhn, . HOUSES as a screening tool for cancer screening: Patient-level housing-based socioeconomic status and breast and cervical cancer screening adherence [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7599.

Abstract Objectives Microscopic colitis (MC) is an inflammatory disease of the large intestine characterized by chronic diarrhea, normal‐appearing colonic mucosa on endoscopy, and abnormal inflammatory histopathology. While it is a common cause of chronic diarrhea in adults, MC is infrequently reported in children, and data on its presentation, treatment, and outcomes remain limited. This study aimed to analyze single‐center experience and conduct a literature review on pediatric MC. Methods A retrospective review of the Mayo Clinic electronic health records from 2013 to 2024 was performed. Children (≤18 years) diagnosed with MC were included. Demographics, clinical presentation, treatment, and outcomes were collected. A systematic review of pediatric MC literature was completed. Results MC was diagnosed in 42 children (median age 12.5 years; Q1–Q3 7.25–14.75), with 28 (67%) females. Systematic review identified 29 studies with 258 children with MC (mean age 10 years, standard deviation [SD] 4.9). At Mayo Clinic, treatment varied among providers, and budesonide as monotherapy or in combination was most effective. In the literature, the most commonly used treatments included 5‐aminosalicylates (5‐ASA, 36%) and budesonide (31.5%). Combination therapy with budesonide plus a biologic or mesalamine appeared to be the most effective therapy, followed by budesonide monotherapy and corticosteroids. Mucosal remission rate was highest with corticosteroids, followed by budesonide. Mean recurrence rate across all treatments was 69%. Biologics (vedolizumab, infliximab) and bismuth showed the lowest recurrence rates. Conclusion Budesonide, either alone or in combination, appears to be most commonly used and effective in treating pediatric MC, but recurrence rates remain high after therapy discontinuation.

Abstract Background: Chronic inflammation is implicated in ovarian carcinogenesis, but how different inflammation-related exposures individually or jointly affect histotype-specific associations remains unclear. Materials and Methods: We pooled data from 16 case-control studies in the Ovarian Cancer Association Consortium to evaluate associations of eight inflammation-related factors (anti-inflammatory: aspirin use, tubal ligation (TL); pro-inflammatory: endometriosis, obesity, lifetime ovulatory cycles (LOC), smoking, pelvic inflammatory disease (PID), polycystic ovary syndrome (PCOS)) with epithelial ovarian cancer (OvC) by histologic subtype. We examined individual associations and clustering of risk factors across histotypes and computed population attributable risk (PAR) for each factor. We assessed additive and multiplicative interactions for exposure combinations. Results: Associations with OvC risk differed by histotype (e.g., high-grade serous: aspirin: OR=0.90; 95%CI 0.82, 0.99; TL: OR=0.80; 95%CI 0.73, 0.88; overall serous: endometriosis: OR=1.17; 95%CI 1.03, 1.31; high LOC: OR=1.42; 95%CI 1.28, 1.58; obesity (low-grade serous): OR=1.50; 95%CI 1.14, 1.98). Clustering analyses showed highly correlated risk profiles in endometrioid and clear cell (r=0.91). High-grade serous and mucinous profiles were moderately correlated with endometrioid and clear cell (r=0.60) tumors. The profile for low-grade serous (r=0.36) tumors was distinct from other histotypes. PAR estimates suggested modifying aspirin use, TL, and LOCs could substantially reduce burdens of endometrioid, clear cell and mucinous tumors. Out of 28 exposure combinations tested in overall OvC and 189 by histotype, we observed 12 interactions. Not using aspirin regularly showed positive additive interactions with obesity and high LOCs, particularly in endometrioid tumors (obesity relative excess risk due to interaction (RERI)=0.74, 95%CI 0.31, 1.18; Pint=0.001 for; LOCs RERI=0.80, 95%CI 0.03, 1.56; Pint=0.04). Not using aspirin regularly also showed a positive additive interaction with endometriosis in clear cell tumors (RERI=1.77, 95%CI 0.03, 3.52; Pint=0.05). Lack of TL showed positive interactions with obesity in endometrioid (RERI=0.86, 95%CI 0.17, 1.53; Pint=0.01) and mucinous (RERI=1.10, 95%CI 0.23, 1.97; Pint=0.01) tumors, while negative additive interactions were observed for smoking and endometriosis in endometrioid tumors (RERI=-1.12, 95%CI -2.19, -0.05; Pint=0.04). A multiplicative interaction was observed between obesity and endometriosis in mucinous tumors (Pint=0.01). Conclusion: The findings suggest ovarian tumorigenesis is strongly shaped by pro- and anti-inflammatory pathways that act largely independently. Further examining these pathways may clarify the origins of histotype heterogeneity and guide prevention strategies. Citation Format: Maxwell Akonde, Britton Trabert, SHELLEY TWOROGER, Allan Jensen, Kathryn L. Terry, Joshua Sampson, Hoda Anton-Culver, David Bowtell, Elisa V. Bandera, Angela Brooks-Wilson, Andrew Berchuck, Daniel William Cramer, Linda S. Cook, Julie M. Cunningham, Jennifer A. Doherty, Ellen L. Goode, Marc T. Goodman, Holly Ruth Harris, Susanne K. Kjaer, Nhu Le, Alice Wen-Ron Lee, Francesmary Modugno, Kirsten B. Moysich, Celeste Pearce, Malcolm C. Pike, Harvey A. Risch, Mary A. Rossing, Joellen M. Schildkrau, Daniel O. Stram, Rebecca Sutphen, David Van Den Berg, Penelope M. Webb, Anna Wu, Argyrios Ziogas, Nicolas A. Wentzensen. Inflammation-related exposures and histotype- specific ovarian cancer risk in the Ovarian Cancer Association Consortium (OCAC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 6261.

Abstract Large Language Models (LLM) are being widely adopted into the medical field for their impressive ability to analyze and summarize large amounts of text data. These models enable clinicians and researchers to extract meaningful insights from complex datasets and may assist with decision making. Here we present our workflows and application of LLM for the interpretation and summarization of clinical data related to CAR-T cell therapy. Using an LLM (Gemini 2.5 pro) in the Google Cloud Computing (GCP) environment, two applications were developed to analyze CAR-T cell therapy clinical data: 1) extracting and summarizing CRS and ICANS event-related data to streamline the compliance team workflow, and 2) identifying features available at time of CAR-T infusion able to classify patients into high- or low-monitoring needs 14 days post CAR-T. Patient data (vitals, labs, hematology notes, and EKGs) were extracted from the electronic medical record (EMR) using Google BigQuery into SQL tables in GCP. For each application, relevant data fields were retrieved, formatted into JSON objects, and embedded in the LLM prompt for context-aware processing. Both applications have undergone iterative prompt engineering after analyzing the LLM output against ground truth data in the EMR. For the application extracting and summarizing CRS and ICANS events, the LLM was optimized on Mayo Clinic Rochester data. When compared to the IEC compliance database, the LLM achieved 100% accuracy and F1 score for CRS events, and 96% accuracy and 82% F1 score for ICANS events. The match rate for CRS and ICANS grades were 83% and 89% respectively. We applied the same LLM to Mayo Clinic Arizona (MCA) and Mayo Clinic Florida (MCF) cases, who document clinical notes and toxicity flowsheet differently, and achieved an accuracy (MCA: 93%, MCF:93%) and F1 score (MCA: 96%, MCF: 96%) for CRS and accuracy (MCA: 81%, MCF:82%) and F1 score (MCA: 76%, MCF: 75%) for ICANS. LLM was able to capture events missed by manual reviews. For most of the discrepancies where compliance team final adjudication is needed, LLM will be updated to flag discrepancies for review by the compliance team. For the application related to the monitoring needs 14 days post CAR-T, the LLM identified 5 categories predictive of high or low monitoring needs post CAR-T infusion (disease status, inflammatory and tumor burden markers, hematologic status, renal function and performance status). Our model achieved a sensitivity of 85.7%, specificity of 23.8%, and F1 score of 65.5% for our first cohort, compared with data extracted by the LLM from the EMR. For our second cohort, with demographics statistically similar to cohort 1 and using the same 5 categories, the LLM achieved a sensitivity of 83.3%, a specificity of 28.6%, and F1 score of 65.4%. Our workflow and applications of LLM’s provide examples and guidance to others interested in applying LLM’s for clinical and research applications. Citation Format: Emmanuel Contreras Guzman, Matthew Jankowski, Andre De Menezes Silva Corraes, Malvika Gupta, Monica L. Shaw, Madiha Iqbal, Talal Hilal, Saurabh Chhabra, Ricardo Daniel Parrondo, Jody K. Mclean, Kim R. Riester, Kayla Joseph, Melinda Tan, Holly Ross, Cleyonia Barnett, Sylvia Carter, Semy Girmay, Rachel Wolan, Milana Ramsey, Christian Downhour, Kristy Morgan, Shae Sibley, Erica Rushing, Lucy Holmes, Allison Burgstahler, Stephen M. Ansell, Hassan Alkhateeb, Matthew Hathcock, Ramona Bruno, Allison C. Rosenthal, Hemant Murthy, Patrick B. Johnston, Jonas Paludo, Yi Lin. Applications of large language models to CAR-T cell therapy clinical data using Google cloud computing [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2740.

Abstract Background: Prostate cancer remains one of the most common malignancies in men, with significant heterogeneity in clinical outcomes. Early and accurate diagnosis and risk stratification is crucial for effective treatment and improved outcomes. Although Gleason score is an established risk stratification tool, it does not fully explain outcome variability. We present a multimodal deep-learning framework that integrates attention-based whole-slide image (WSI) features with a clinical variable (Age) to predict survival in patients with prostate adenocarcinoma. Design: Clinical data and WSI for 500 patients with prostate adenocarcinoma were obtained from TCGA-PRAD. Clinical data were parsed for recurrence or progression (RoP) events and age. RoP was used as the outcome label, and age was used as a patient feature. Diagnostic WSI tumor regions were annotated and tiled at 224×224 pixels using QuPath (Bankhead Sci Rep 2017). Tile embeddings were extracted with UNI2-h (Mahmood Nat Med 2025). An end-to-end machine learning pipeline was developed to aggregate tile embeddings into slide-level embeddings using Clustering-Constrained Attention Multiple Instance Learning (CLAM) and predict RoP (Lu Nat Biomed 2021). A second pipeline used a Gleason score-trained binary cross entropy (BCE) model to predict RoP in the same patients. Three models were evaluated: WSI-only CLAM model, WSI+age CLAM model, and Gleason-only BCE model. All models were trained and evaluated using 5-fold stratified cross-validation. Each fold was repeated 3 times with independent random initialization. AUROC was averaged across all folds and runs per model. Standard deviation was also computed. Model performance was compared using Friedman’s test, with post hoc pairwise comparisons by Wilcoxon signed-rank test. Results: The Gleason BCE model underperformed (AUROC 0.67 ± 0.07). Both WSI-based models outperformed the Gleason BCE model, achieving AUROC 0.76 ± 0.02 (without age), and AUROC 0.76 ± 0.01 (with age). AUROCs were significantly different by Friedman’s test (p = 0.0224) and showed a trend toward significance in post-hoc comparison between WSI and Gleason-based approaches (p = 0.06). Conclusion: AI models using WSI-derived features, with or without basic clinical context, outperformed traditional Gleason score for recurrence risk prediction in prostate cancer. These findings support the integration of digital pathology and AI into routine prognostic assessment for prostate cancer. Future work will focus on including additional patients, clinical variables, and multi-class prognostic labels to expand prognostic prediction beyond RoP. Citation Format: Justin Johnson, Kingsley Ebare. Whole-slide image and clinical feature integration for superior prostate cancer risk stratification [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1441.