Citations

Abstract Background: Screening mammography is the cornerstone of breast cancer detection and is routinely performed before treatment initiation, providing a unique opportunity for early risk stratification. Beyond tumor detection, screening mammograms contain imaging biomarkers that reflect underlying tumor biology and the surrounding breast microenvironment. However, treatment response and long-term outcomes are driven by complex interactions between imaging phenotypes and patient-specific clinical characteristics that are not fully captured by mammography or conventional clinicopathologic staging alone. Integrating pre-treatment mammograms with longitudinal clinical data enables comprehensive patient representations that can inform therapeutic decision-making before neoadjuvant therapy (NAC) begins. Methods: We developed a multimodal framework to learn comprehensive patient phenotypes by integrating pre-treatment mammograms with longitudinal unstructured multispecialty clinical notes. Mammograms were encoded using MammoFM, a federated breast imaging foundation model pre-trained on large-scale, multi-institutional datasets, with the four standard views (CC and MLO) fused through a multi-view attention mechanism. Longitudinal peri-diagnostic EHR notes (3 months before to 2 months after diagnosis) were processed using a large language model (LLM) to generate semantic clinical embeddings from key patient characteristics. Imaging and clinical embeddings were then integrated through a cross-modal fusion transformer to learn unified multimodal patient phenotypes that capture complementary imaging and clinical information. These transferable phenotypic representations were subsequently leveraged for downstream prediction of pathological complete response (pCR) following NAC and long-term distant recurrence risk. Results: The multimodal phenotyping framework was developed using an internal cohort of 1,064 breast cancer patients from Mayo Clinic (AZ, MN, and FL) and evaluated on an independent held-out test set (n = 35). The learned multimodal patient representations consistently outperformed unimodal approaches across downstream prediction tasks. For long-term recurrence prediction, the multimodal phenotype achieved 85.71% balanced accuracy and an AUROC of 0.760, compared with 71.43% accuracy for mammography-only and 78.57% for clinical-only models. For prediction of pCR, the multimodal phenotype achieved 64.29% accuracy, outperforming mammography-only (35.71%) and clinical-only (46.43%) models. Conclusions: We developed a transferable multimodal patient phenotyping framework that integrates pre-treatment mammography with longitudinal clinical documentation to learn comprehensive representations of breast cancer patients. These multimodal phenotypes consistently outperformed imaging- or clinical-only representations for predicting pCR and long-term recurrence, demonstrating their potential as a generalizable foundation for precision oncology, risk stratification, and personalized treatment planning. Citation Format: Xiaoyang Wei, Madhu Sikha, Zhemin Zhang, Judy W. Gichoya, Imon Banerjee. Learning multimodal breast cancer phenotypes for precision oncology [abstract]. In: Proceedings of the 19th AACR Conference on the Science of Cancer Health Disparities; 2026 Oct 18-21; Atlanta, GA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2026;35(10 Suppl):Abstract nr B040.

Abstract Patients with cancer and their caregivers face challenges accessing clinical trials because trial information, eligibility criteria, search, and referral pathways are complex and fragmented. These barriers can contribute to unequal access across racial, ethnic, and language groups. ALEX (Agents Leveraging Empathy for eXams) is a web-based portal that uses AI-enabled virtual health educators with synthesized voices and lip-synced facial animation to deliver trial education and a large language model to simplify trial information for search and referral. AI-generated content and virtual humans require human and community evaluation for accuracy, authenticity, and trust. This study examined how community feedback informed the iterative design and refinement of ALEX. The dataset included 34 English- and Spanish-language transcripts (about 44 hours) from focus groups and usability interviews using think-aloud procedures to evaluate the portal. The sessions were conducted from 2023 to 2025 at Mayo Clinic, UF-Gainesville, UF-Jacksonville, and the University of Miami. They examined 136 participants’ responses to the virtual health educators and their use of the portal for trial education, search, and referral. Participants reported their race as White (48.5%), Black or African American (47.1%), Native American or Alaska Native (0.7%), or other (3.7%). Separately, 50.7% identified as Hispanic or Latino, and nearly 25% reported Spanish as their primary language. Throughout data collection, feedback was reviewed and translated into design requirements for successive portal updates. We used deductive and inductive qualitative analysis to identify feedback patterns and user needs and reviewed version history to document portal changes. Three of the six themes derived from 1,653 coded excerpts were Trust Before Trial Search (“Who is Alex?”), Layered and Controllable Education (“If it was more in short bullets, maybe better”), and Motivation, Burden, and Research Trust (“There has to be some trust, some transparency or at least some acknowledgement”). Feedback informed portal refinements, including clearer language and institutional branding and content that was easier to scan. The virtual health educators were evaluated on representation, voice authenticity, lip-sync, warmth, and human likeness. Words like “sponsored study,” “diagnosis,” and “symptoms” guided language refinement. Suggested improvements included sample search terms, easy-to-read study cards, the ability to save results, a question guide, and direct coordinator contact. Research trust, privacy, security, and participation burden remained key priorities. To create an AI-enabled clinical trial portal that meets user needs, we used community feedback to identify modifiable barriers to communication, search, and referral. The goal of this formative process is to transparently incorporate that feedback into trial tools that are more credible, usable, acceptable, and equitable. Future work should evaluate whether these revisions improve referral outcomes. Citation Format: Fan Wang, Rashi Ghosh, Christopher You, Trevor Kauer, Andrew H. Maxim, Courtny Franco, Stephen Anton, Elizabeth Flood-Grady, Trudy Gaillard, Thomas J. George, Benjamin Lok, Carolyn Tucker, Melissa Vilaro, Tuo Lin, Fern Webb, Diana Wilkie, Danyell Wilson-Howard, Olveen Carrasquillo, Janice L. Krieger. Can artificial intelligence amplify community voice? Evidence from a multisite formative evaluation of the AI-Powered ALEX Portal to enhance equity in clinical trial access [abstract]. In: Proceedings of the 19th AACR Conference on the Science of Cancer Health Disparities; 2026 Oct 18-21; Atlanta, GA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2026;35(10 Suppl):Abstract nr B034.

Abstract Background: Community Living Lab (CoLLab) learning health system is an actionable model of trust and engagement that fosters ownership and sustainability of interventions in communities with limited access to healthcare resources. The approach has been found to improve quality of life, physical and social health, and cognitive function of the populations studied. The aim of this study is to describe the implementation of CoLLab learning health system as a community-based model to improve prostate health equity among Black men through partnership between the Mayo Clinic Florida and three American Legion Posts in Northeast Florida. Methods: The implementation of CoLLab at the study site was guided by the social-ecological model to ensure that the intervention addresses barriers from individual, interpersonal, organizational, and community levels. Multiple prostate health Resources, Education, Amenities, and Community Health (REACH) services and programs were available to be adopted at each site based on the needs of BM. The CoLLab intervention consists of education, navigation, and support services focused on: (1) social determinants of health; (2) prostate cancer risk reduction, prevention, screening, diagnosis, treatment, survivorship, and advocacy; and (3) cancer clinical trials navigation. The CoLLab sites were activated after an in-person visit to the site to inspect infrastructure, including study devices, materials, tools, and personnel. Sites were opened and provided REACH services to Black men who are members of the American Legion Posts and those residing in the neighborhood around the posts. The Implementation assessments were completed by site leaders, Community Health Workers, research coordinators, and study personnel. Results: The findings demonstrated that CoLLab was highly feasible and generally acceptable across sites. The strongest successes were the establishment of trusted community spaces for prostate risk reduction and prevention, increased prostate cancer awareness, and sustained engagement through peer-led support groups and book clubs. CoLLab intervention evolved beyond a traditional research project into a community-owned health resource. Several intervention components, particularly support groups and book clubs, were embraced by participants as valued community assets and continued independently beyond the study period. However, implementation challenges included participant retention, survey fatigue, technology barriers among older adults, staffing turnover, transportation challenges, and early trust disruptions related to participant incentives and food distribution quality. Conclusion: The CoLLab model successfully transformed trusted American Legion Posts into community-based prostate health learning hubs. The strongest elements were peer-led support groups, book clubs, trusted community relationships, and Community Health Workers engagement. Stakeholders overwhelmingly concluded that the CoLLab model is feasible, acceptable, sustainable, and scalable for reducing prostate cancer disparities among Black men. Citation Format: Opeyemi O. Bolajoko, Quincy A. Wimberly, Shaina E. Flynn, Jade Reeves Ebony, Floyd B. Willis, Toshiko Moultrie, Tracy Gillard, Wayne Ford, Derry Green, Arnold J. Merriweather, Folakemi T. Odedina. Reducing prostate cancer disparities: Lessons from Community Living Lab Learning Health System [abstract]. In: Proceedings of the 19th AACR Conference on the Science of Cancer Health Disparities; 2026 Oct 18-21; Atlanta, GA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2026;35(10 Suppl):Abstract nr A042.

Abstract Background: Prostate Cancer survivorship care requires significant contributions from family members and relatives, especially those who serve as caregivers. This role can be burdensome, but it is vital to treatment outcomes. Given the increased adoption of outpatient models of cancer therapies, there is a substantial need for continuous support for cancer care at home through organized infrastructure to assist family caregivers. The primary objective of this study was to present preliminary findings on caregivers’ unmet survivorship needs and challenges in navigating healthcare and support services and to identify potential approaches to support continuous cancer care at home. Methods: The study used a phenomenological design to obtain data from caregivers. In-depth interviews focused on participants' lived experiences caring for prostate cancer patients and survivors. The interview captured the personal experiences of family caregivers while taking care of patient diagnosed with prostate cancer at home. Questions were asked about the challenges faced in supporting the care of patients at home, healthcare navigation, coping strategies adopted, and survivorship care needs during and after treatment. All interviews were recorded and transcribed verbatim. The transcripts were reviewed, analyzed, and coded, with emerging themes compiled. Results: Twenty-two caregivers participated in this study, and the mean age was 56.4±12.01. Four challenges identified were (1) limited understanding of prostate cancer and care needs; (2) substantial home care burden, including balancing patient care with domestic responsibilities and caregiving for other family members; (3) inadequate knowledge of care navigation, such as identifying specialist providers and scheduling appointments, managing transportation, and completing insurance and treatment-related documentation; (4) financial strain, including maintaining employment, reduced household income, increased out-of-pocket expenses and treatment co-payments. Caregivers identified the need for education and training on cancer care at home, assistance with care navigation, access to social and financial support programs, mental health counseling and emotional support, as well as peer support groups for caregivers. They suggested that effective prostate cancer care should extend beyond the patient to address the informational, emotional, and practical needs of caregivers throughout the entire cancer care continuum. Conclusion: The preliminary findings show that most caregivers assumed the role with little knowledge of prostate cancer and learned through experience, often while coping with emotional distress and navigating complex healthcare systems. Family support, faith, self-education, and maintaining hope helped many caregivers adjust to this new role. Citation Format: Opeyemi O. Bolajoko, Jada K. Melton, Michelle R. Fudge, Mary E. Young, Jocelyn Turner, Toshiko Moultrie, Karen Gainey, Jade Thompson-Reeves, Folakemi T. Odedina. Unmet survivorship care needs of family caregivers of patients with prostate cancer [abstract]. In: Proceedings of the 19th AACR Conference on the Science of Cancer Health Disparities; 2026 Oct 18-21; Atlanta, GA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2026;35(10 Suppl):Abstract nr C005.

ABSTRACT Introduction Changes in peripheral eosinophils follow endoscopic sinus surgery (ESS) in chronic rhinosinusitis with nasal polyps (CRSwNP), with and without aspirin sensitivity. Cytometry by time‐of‐flight (CyTOF) identifies more immune cell subsets than conventional assays or clinical testing. We hypothesized that ESS‐associated changes in peripheral immune cell subsets differ between aspirin‐tolerant (AT)‐CRSwNP and aspirin‐exacerbated respiratory disease (AERD) as measured by CyTOF. Methods Pre‐ and post‐ESS peripheral blood from patients with CRSwNP ( n = 13) or with AERD ( n = 15) was analyzed for immune cell populations using a 37‐marker CyTOF panel. CITRUS analysis identified cell clusters most predictive of patient group or sample timepoint. Results In the combined cohort of 28 patients, circulating CD4 + naïve T cells and CD194 + neutrophils increased after ESS, while eosinophils decreased. When stratified by disease, CD4 + naïve T cells increased in both AERD and CRSwNP post‐ESS; however, eosinophils decreased after ESS only in CRSwNP. Neutrophil subsets showed a mixed response in both patient groups after ESS. Dendritic cells increased in CRSwNP but not in AERD after ESS. Individual timepoint analyses showed more CD141 + monocyte‐derived DCs (moDCs) in CRSwNP than in AERD pre‐ and post‐ESS. Pre‐ESS, both CD185 + and CD194 + neutrophils were higher in AERD compared to CRSwNP. Conclusion CD141 + moDCs were more prevalent in CRSwNP versus AERD, both before and after ESS. Naïve CD4 + T cells increased, and neutrophil subsets changed in all groups after ESS, but eosinophils decreased only in CRSwNP. Our study points to possible systemic immune effects of ESS and candidate immune signatures in CRSwNP and AERD.

Abstract Objective To characterize the prevalence, severity, and clinical correlates of subjective ictal-phase cognitive impairment across the migraine attack cycle in a large, specialty-center cohort. Background Cognitive dysfunction is a prevalent and disabling feature of migraine attacks, yet standardized tools for measuring migraine attack-related cognitive impairment are limited, and the clinical predictors of cognitive burden remain incompletely characterized. Prior work has largely focused on the ictal phase, although cognitive symptoms are reported across the prodromal, ictal, and postdromal phases of the migraine attack. Methods This is a cross-sectional analysis of data from the American Registry for Migraine Research (ARMR), a multicenter registry enrolling patients with migraine diagnosed by headache specialists using ICHD-3 criteria. Participants completed the Migraine-Specific Cognitive Impairment Scale (Mig-SCog; range 0–18, higher scores indicating greater impairment). Associations between cognitive impairment and clinical variables were evaluated using correlation analyses, group comparisons, and multivariable linear regression. Results Among 1,231 patients (87.2% women; mean age 46.0 ± 13.7 years; 71.5% chronic migraine), 1,175/1,231 (95.5%) endorsed cognitive symptoms, with a mean Mig-SCog total score of 8.2 ± 4.8. Executive dysfunction accounted for 71.1%±18.6% of total cognitive burden, but the proportional contribution of language impairment increased with migraine frequency category, from 22.4%±18.9% in low-frequency episodic migraine to 28.8%±19.5% in chronic migraine (p = 0.003). In multivariable regression, migraine frequency classification was not independently associated with cognitive impairment. The strongest independent predictors of cognitive impairment were anxiety symptom score (β = 0.15, p < 0.001) and depression symptom score (β = 0.13, p = 0.002), with standardized coefficients exceeding those of all migraine frequency classification categories. Women reported significantly greater cognitive impairment than men across all domains (8.4 ± 4.8 vs. 6.8 ± 4.7; p < 0.001, small effect). Word-finding difficulty demonstrated the largest ictal symptom-frequency association, with scores rising from 3.3 ± 2.6 among those never reporting word-finding difficulty to 13.6 ± 3.6 among those always reporting it (adjusted p < 0.001, large effect). In phase-based analyses, word-finding difficulty and impaired concentration were associated with 4.6- and 3.7-point higher cognitive impairment scores than in patients who did not report these symptoms during the prodrome and 5.1- and 4.0-point higher scores during the postdrome, respectively (all p < 0.001, large effects). Among patients with chronic migraine and available disease duration data (n = 661), cognitive burden was lower among those within the first 12 months of chronicity compared with those with longer-standing chronic migraine (7.7 ± 4.9 vs. 9.0 ± 4.6; p = 0.010, d = 0.27). Sound hypersensitivity (β = 0.25) and cutaneous allodynia (β = 0.20) were the strongest independent sensory predictors of cognitive impairment (both p < 0.001). Conclusion Migraine-related cognitive dysfunction was nearly universal in this specialty-center cohort and was not adequately captured by migraine frequency classification alone. Anxiety and depression symptom burden, language impairment, full-cycle cognitive symptoms spanning prodrome through postdrome, and sensory sensitization are clinically meaningful markers of cognitive burden that should be assessed alongside more traditional migraine characteristics when evaluating patients who have migraine.

Abstract Background: African American men experience disproportionately high prostate cancer (PCa) mortality and are underrepresented in genomic studies. We evaluated real-world Decipher utilization and oncologic outcomes among African American men treated definitively for localized PCa. Methods: We retrospectively identified African American men with localized prostate cancer treated with radical prostatectomy (RP) or radiation therapy with or without androgen-deprivation therapy (RT±ADT) across 3 Mayo Clinic sites between 2010 and 2024. Baseline variables included age, prostate-specific antigen (PSA ng/mL), clinical T stage, treatment, Gleason score, and radiographic findings. Biochemical progression was defined as PSA ≥0.2 ng/mL after RP or ≥2 ng/mL above nadir after RT; metastases were determined by imaging. All time-to-event outcomes were measured from diagnosis. Kaplan–Meier methods estimated overall survival (OS), biochemical progression-free survival (bPFS), and metastasis-free survival (MFS) by Decipher status and clinical subgroups, with comparisons using log-rank tests. Multivariable Cox proportional-hazards models were fitted, and the proportional-hazards assumption was assessed using Schoenfeld residuals. Two-sided p&lt;0.05 was considered statistically significant. Results: Among 191 identified patients, 69 underwent RP and 122 received RT±ADT; median follow-up was 62 months. Decipher testing was performed in 12 patients (6.3%), and germline testing in 51 (26.7%). Tested patients had lower-grade disease (Gleason &lt;8 in 92% vs 58%; p=0.03). Four deaths occurred, including 1 among Decipher-tested patients and 3 among untested patients. There were 60 bPFS events (31%) and 34 MFS events (18%). In overall population, there was no statistically significant difference in patients who received decipher testing compared to those who did not in terms of OS (p=0.13), bPFS (p=0.14), or MFS (p=0.22). In exploratory multivariable models, PSA ≥20 ng/mL (aHR 11.4, 95% CI 4.8–27.3), Gleason score ≥8 (aHR 2.40, 95% CI 1.01–5.69), and clinical stage T3–T4 (aHR 2.58, 95% CI 1.15–5.81) was associated with worse MFS, whereas the addition of ADT to RT was associated with improved MFS compared to RT alone (aHR 0.15, 95% CI 0.04–0.50). Moreover, RT±ADT improved bPFS compared to RP (aHR 0.21, 95% CI 0.09–0.47) and patients with PSA ≥20 at baseline were associated with worsened bPFS compared to those who had PSA &lt;20 (aHR 2.33, 95% CI 1.12–4.88). OS data was not mature. Conclusions: The utilization of decipher testing was limited in African American men with localized PCa, suggesting a potential gap in access to precision-oncology tools in this patient population. Testing status was not associated with outcomes, although analyses were limited by the small sample size. Exploratory analyses showed association of clinicopathological charatersitics with MFS and bPFS and can be considered for clinical decision making. Prospective, adequately powered studies are needed to evaluate equitable genomic-testing use and prognostic value in African American men. Citation Format: Kaneez Zahra R. Khakwani, Muhammad Ali Khan, Muhammad Umar Afzal, Ahmed A. Khan, Usama Mumtaz Cheema, Syed Arsalan Ahmed Naqvi, Chueh Hsuan Hsu, Ewan Cobran, Irbaz B. Riaz. Low utilization of the Decipher genomic classifier and oncologic outcomes in African American men with localized prostate cancer: A cancer-disparities analysis [abstract]. In: Proceedings of the 19th AACR Conference on the Science of Cancer Health Disparities; 2026 Oct 18-21; Atlanta, GA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2026;35(10 Suppl):Abstract nr A176.

Abstract Black men face a disproportionate burden of prostate cancer, yet screening uptake remains low, underscoring the need for targeted rather than one-size-fits-all screening messages. Grounded in dual-process theories of information processing and the Health Belief Model, this study tested whether narrative framing is more effective than reason-based framing in shaping screening attitudes and intentions among Black men who have never been screened. In a randomized experiment, 268 Black men aged ≥40 years (never screened by PSA or DRE) were assigned to read a reason-based brochure presenting scientific facts, anatomy, and statistics (e.g., 1 in 7 Black men diagnosed; twice the death rate) to urge screening (n = 91), a narrative brochure telling a first-person survivor's story of getting screened (n = 97), or a control brochure containing the same message resource, header, photo, and screening recommendation but no persuasive body text (n = 80). Post-brochure surveys measured affect toward screening and intention to discuss testing with a physician, alongside moderators including attitudes towards screening, perceived behavioral control, cultural orientation, fatalism, knowledge, and demographics; effects were tested using ANCOVA and regression, adjusting for demographics. Narrative framing did not outperform reason-based messaging on affect (F(2,245) = 1.62, p = .201) or intention (F(2,245) = 1.19, p = .305); however, the effect depended on the reader, with a significant condition-by-ethnicity interaction (F(2,245) = 5.22, p = .006) and, most notably, a condition-by-attitude interaction (F(2,242) = 8.11, p &lt; .001). Among men with the least favorable baseline attitudes toward screening, the narrative significantly raised intention over control (+0.61, p &lt; .001), while message style ceased to matter at average or high attitudes toward screening. Attitude was the strongest predictor of intention (b = 0.41, p &lt; .001), followed by perceived behavioral control (b = 0.16, p = .041), cultural orientation (b = 0.15, p = .034), and fatalism (b = −0.16, p = .016), each independently predicting stronger screening intentions (fatalism negatively). The narrative brochure scored highest for reader engagement (F(2,244) = 3.38, p = .036). These findings indicate that there is no universal “best” message; narrative framing is most valuable for reaching the least-convinced men and for outreach engagement, and screening interventions should target the underlying beliefs—attitude, perceived control, and fatalism—that most strongly drive intention. Citation Format: Tingyu Zou, Quincy A. Wimberly, Chelsea C. Reeves, Gui R. Pradieu, Jade Thompson-Reeves, Michelle Fudge, Monica L. Albertie, Folakemi T. Odedina. Motivating Prostate Cancer Screening in Black Men: A Randomized Messaging Experiment [abstract]. In: Proceedings of the 19th AACR Conference on the Science of Cancer Health Disparities; 2026 Oct 18-21; Atlanta, GA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2026;35(10 Suppl):Abstract nr C178.

Abstract Background Obesity has been associated with complications after Achilles tendon surgery, but evidence specific to insertional Achilles tendinopathy is limited and conflicting. We evaluated the association between body mass index and 90-day complications after a standardized insertional Achilles operation. Methods This retrospective procedure-level cohort included all eligible primary operations at a single tertiary referral center from 2013 through 2025. The primary outcome was any adjudicated complication within 90 days. BMI was evaluated with an adjusted three-knot restricted cubic spline and summarized by an odds ratio per 5 kg/m². Logistic regression used patient-clustered robust standard errors and adjusted for age, diabetes, and smoking status at initial surgical evaluation. The BMI ≥ 35 kg/m² comparison was post hoc. Time to documented return to usual activity was analyzed with an adjusted Cox model. Results Among 217 procedures in 194 patients, 32 had a 90-day complication (14.7%). The adjusted spline demonstrated an association between BMI and complications (global p = 0.030) with evidence of nonlinearity (p = 0.023). Each 5-kg/m² increase in BMI was associated with higher average odds of a complication (adjusted OR 1.43, 95% CI 1.11–1.85; p = 0.006). In the post hoc analysis, complications occurred after 21 of 82 procedures with BMI ≥ 35 kg/m² (25.6%) and 11 of 135 with BMI < 35 kg/m² (8.1%; adjusted OR 3.89, 95% CI 1.71–8.85; p = 0.001). Seventeen wound complications occurred (7.8%). BMI was not associated with time to documented return to usual activity (adjusted HR 0.91 per 5 kg/m², 95% CI 0.74–1.11; p = 0.355). Conclusions Higher BMI was associated with greater 90-day complication risk after primary insertional Achilles surgery, with a nonlinear relationship across the observed range. BMI ≥ 35 kg/m² may be a pragmatic counseling reference, but it was selected post hoc and is not a validated threshold or stand-alone contraindication. Compared with prior insertional Achilles series, this study used a standardized 90-day endpoint, nonlinear BMI modeling, blinded complication adjudication, and methods accounting for repeated procedures.