Citations

Traditional home-call managing after-hours pager communications remains a common component of hematology-oncology fellowship and modern practice. Little is known about these patient-related interactions and implications. This project aimed to characterize after-hours pager volumes and evaluate perceived impacts on care, physician wellbeing, and trainee education. We performed a mixed-methods study at a tertiary academic center, analyzing after-hours pager activity from 2020-2024 across Hematology Consult (HCP), Hematology Outpatient (HOP), and Medical Oncology (MOP) pagers. A survey of hematology-oncology fellows evaluated perceptions and attitudes toward the home-call experience using free-text responses and a 5-point Likert scale. After-hours pager burden was substantial and sustained. The MOP received >3,000 pages annually, with medians of 5-6 per weekday and 15-15.5 per weekend day. The HOP volume nearly doubled over the study period (960 to 1,721 annually) with median pager volumes of 1-3 per weekday and 2-7 per weekend day. The HCP volume ranged from 594-768 pages annually. Survey response rate was 58% (19/33). Only 16% of fellows endorsed educational value from patient calls on MOP or HOP, and ≤5% agreed that call tasks required physician-level expertise. Respondents reported that the pager resulted in fatigue, anxiety, and social disruptions. Free text responses revealed themes of low educational value, high non-urgent call volume, and system inefficiencies. Contemporary home-call had substantial pager burdens with limited actionability, adverse effects on wellbeing, and low educational value. Systems engineering, including triage processes and augmented patient education, could help optimize patient care, physician wellbeing, and trainee education.

PURPOSE Incorporation of structured data elements within the electronic health record generates real-time data for many potential uses. However, this requires a change in documentation practices, which may increase documentation effort. The best strategies to support structured documentation interventions have not been thoroughly investigated. Herein, we report the results of extending a data capture tool across multiple disease sites and practice locations. METHODS A note-based data capture tool was developed to collect five data elements using minimum common oncology data elements language. Soft-stop close visit validation (CVV), offloading of data entry tasks to allied health staff, group education, individualized e-mails, and one-on-one assistance were used to support utilization. Baseline and follow-up surveys were collected. For each provider, we compared data capture in the 6 months before and after their CVV go-live. RESULTS Forty-five providers participated (31 physicians, 14 nurse pracationers and physician assistants [NPPAs]), resulting in 23,274 encounters pre-CVV go-live and 23,146 post-CVV go-live visits. Median percentage of all data elements captured increased from 69.5% to 82.6% across all providers. The effect of CVV was immediate and sustained. When surveyed, most respondents agreed with the value propositions of understanding practice population, outcomes, selecting trials to open, and identifying patients for trial enrollment. Most respondents (15/20, 75%) felt the soft stop was a helpful reminder. CONCLUSION Structured data can be successfully collected in provider notes with >80% data capture. Personalized assistance incorporating the structured data elements into users' preexisting templates and soft-stop CVV, along with monitoring the results of implementation and personalized follow-up, was used to support the structured data capture intervention.

Abstract Background Preparing tumor board patient summaries is time intensive. Large-language-model based systems may automate summarization but require real-world evaluation prior to clinical use. We performed an exploratory retrospective evaluation of the Microsoft Healthcare Agent Orchestrator (HAO), deployed in a Mayo Clinic–controlled staged environment, to generate tumor board–style patient summaries from retrospective Electronic Health Record (EHR) notes. Methods HAO generated summaries for breast, hepatobiliary, and neuro-oncology tumor board cases using up to the most recent 1,000 clinical notes. Clinician reviewers evaluated outputs via REDCap surveys across perceived factuality, completeness, clarity/conciseness, temporal cohesion, comparative performance, safety, and clinical utility (0–4 Likert scale). Reviewers were permitted to query the HAO chat interface to address missing details. Automated factuality was assessed using TBFact (bidirectional entailment), reporting precision and recall against available reference summaries. Results Among 57 survey responses from 5 different physicians, mean scores exceeded 2.8 across domains, with medians of 3 for most axes. In an exploratory comparison, oncology fellows required less time to review HAO-generated summaries than to manually generate patient summaries (mean difference 13.57 minutes per patient, p<0.001), although this difference may be influenced by prior familiarity with the same cases; 96% of survey responses indicated that HAO would save time. TBFact evaluations showed higher recall than precision across domains, consistent with broad capture of reference content alongside additional content that was not present in gold-standard summaries. Attribution was viewed favorably but showed issues with primary-source specificity and link reliability. Conclusions In a controlled Mayo environment, HAO demonstrated moderate performance and was associated with reduced review time for tumor board preparation. These findings are promising but preliminary and do not establish clinical safety, noninferiority to manual review, or readiness for routine clinical use. Limitations, including verbosity, specialty-specific content gaps, and inconsistent attribution, highlight the need for iterative refinement and further evaluation. Graphical Abstract Created using ChatGPT

e13661 Background: The process of data abstraction and curation represents a significant proportion of clinical trial costs, requiring up to $30,000 per patient depending on the data complexity. Research variables such as stage, performance status (PS), and cancer disease status are typically present in the electronic health record (EHR) in free text but require abstraction for research use. Structuring these real-world data (RWD) in an exportable format utilizing minimum common oncology data elements (mCODE) could reduce the need for abstraction and the potential for erroneous transcription that has been shown to be 5.51-7.72% in a meta-analysis. We assessed the feasibility, availability, and accuracy of RWD for use across Alliance for Clinical Trials in Oncology sites by comparing RWD to matched electronic case report form data (CRF data) at 2 sites where EHR-integrated tools collect structured RWD for all solid tumor patients. Methods: Stage, PS, and cancer disease status (CDS) were extracted from the EHR as RWD and from eCRFs for 100 patients enrolled in National Clinical Trials Network trials across multiple disease sites from 2022 through 2025. Data from baseline and day 1 of cycles 1-4 were compared (500 encounter timepoints). Results: Stage, PS, and CDS were available from RWD more often than CRFs (85 vs 70 / 100 patients, 436 vs 416 / 500 encounters, and 345 vs 83 / 500 encounters respectively). When stage was present in both (n = 58), there were 5 mismatches (8.6%). PS had 22 mismatches (4%) and CDS had 45 mismatches (11.4%). After correcting for differences in CRF vs RWD definitions, CDS mismatches dropped to 1.2% (n = 4). Discordance between RWD and CRF data was related to differences in the timepoint of data collection (e.g., systemic staging scans completed after the initial consult), CRF data obtained from other EHR notes (e.g., oncology research nurse notes containing PS), inherent discordance between RWD and CRF data choices (e.g., responding vs complete or partial response), and failure to update EHR data capture tool fields. RWD were available 21 days (median) before CRF data. Conclusions: Our results suggest that RWD captured directly within clinical notes match CRF data, are more often present, and are available sooner. RWD could serve as a data source for trials, decreasing the need for manual curation and cost. Although some discordance exists, overall rates are low and close to the published error rate for abstraction.

e13660 Background: We previously showed success with collection of structured data elements within the electronic health record using the minimum common oncology data elements (henceforth, mCODE tool) supported by soft-stop close visit validation (CVV) and staff support interventions in a limited cohort. Since the additional interventions are resource-intensive and difficult to apply across an entire department, we therefore implemented a department-wide roll-out supported by the least resource-intensive interventions while monitoring uptake. Methods: The mCODE tool collects five data elements: stage, disease status, performance status, intent of treatment, and intent to change therapy. The Department of Oncology providers received education on the mCODE tool and were asked to implement use at a faculty meeting; soft-stop CVV was enabled at that time. The percentage of data captured for each data element was collected from the notes of providers who did not participate in prior pilots for the 4 months post-implementation. mCODE tool usage was defined as entry of at least one data element except stage (as stage has long been a structured data element in Epic, can be entered independently of the mCODE tool, and many providers were entering stage prior to mCODE tool introduction). Results: 86 MDs and 20 APPs for a total of 106 providers were included, resulting in 14,973 visits and 8,745 unique patients. 73.6% of providers (n =78) utilized the mCODE tool. Of those who utilized the tool, median overall data capture across the four data elements was 24.0%. 22.6% of providers (n =24) achieved > 80% overall data capture and 32.1% (n =34) had over 50% overall data capture despite minimal intervention. 75.6% of MDs (n = 65) utilized the tool with a median data capture of 16.3% compared to 65% of APPs (n = 13) with a median data capture of 90.3%. Rates of data capture for individual data elements were similar. Conclusions: Use of soft-stop CVV appears to drive utilization of the mCODE tool but is insufficient to promote > 80% data capture for most providers without more intervention. Supportive interventions such as individualized assistance with incorporation into existing templates and directed communication from leadership remain necessary to promote use of structured data elements and will be implemented in sequence to understand how we achieve the target data capture ( > 80%) at scale.

11025 Background: Adjuvant endocrine therapy (AET) improves outcomes in HR+ breast cancer. Nonadherence to AET is common, but manual screening for nonadherence is fraught. High-specificity structured cancer electronic health record (EHR) data, particularly if it includes key variables like stage, could streamline these steps for both clinical care and research purposes. We conducted an analysis of approaches to identifying patients as eligible for a multi-site AET adherence study to understand how variation in structured data informatics-supported screening impacts coordinator time. Methods: Mayo Clinic implemented a two-step identification: (1) EHR structured data (diagnosis date, stage, ER/PR/HER2 positivity, medication script) to prescreen for eligibility; (2) SureScripts fills to flag nonadherence using proportion of days covered < 80% across early fills. Medical College of Wisconsin (MCW) implemented a more manual screening process for Step 1 using only medication scripts. Coordinators at both sites then conducted a medical record review to assess if the resulting list of patients had a plan to continue AET for at least two additional years. Opt-out clinician permission preceded patient contact; pharmacists delivered a patient reported outcome (PRO)-guided intervention and documented encounters. Results: At Mayo, 829 patients were prescreened for study eligibility through informatics workflow; 120 (14.5%) met eligibility criteria. At MCW, 1,763 patients were prescreened; 168 (9.5%) were eligible. EHR filters included non–breast cancer encounters (50% at MCW) and ineligible based on time since cancer (36% at Mayo; 26% at MCW). At Mayo, screening required 5.6 minutes/patient for experienced coordinators. At MCW, experienced coordinators averaged 6.67 minutes/patient and inexperienced coordinators averaged 9.33 minutes/patient. Conclusions: Structured EHR data, pharmacy-fill information, and standardized workflows allowed consistent prescreening, intervention and recruitment processes across sites. Higher-specificity informatics-supported screening reduced the percentage of patients requiring manual review and reduced coordinator screening time. It also supported timely provider-permission steps and patient outreach. These workflows were feasible to implement at both institutions, and study enrollment is ongoing. Clinical trial information: NCT06765707 .

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.

PURPOSE This national study was undertaken to determine screening mammography rates among transgender and gender diverse (TGD) individuals. METHODS TGD individuals and cisgender women who qualified to receive screening mammography based on American College of Radiology guidelines were identified in the longitudinal OptumLabs Data Warehouse. The primary end point was the percentage of patients who had high rates of screening (>75% recommended screenings). Multivariable analyses were conducted in a matched population of transgender men to cisgender women to identify and adjust for factors associated with high adherence. Individuals were matched at a 1:4 ratio (transgender men: cisgender women) on age, race/ethnicity, year of enrollment in their health care plan, and duration of follow-up. RESULTS This study focused on 10,478 TGD individuals (3,778 transgender men; 1,294 transgender women; 5,406 with gender dysphoria not meeting other TGD criteria, NMOT) and 6,218,369 cisgender women. For transgender men, transgender women, individuals with gender dysphoria NMOT, and cisgender women, high mammography rates were observed in 41.1% (95% CI, 39.5 to 42.6), 7.4% (95% CI, 6.0 to 8.8), 11.9% (95% CI, 11.0 to 12.7), and 38.3% (95% CI, 38.3 to 38.4), respectively ( P < .0001). When transgender men were compared with 15,112 matched cisgender women, an age-based differential association was observed between sex and high mammography rates (interaction P = .0003); compared with cisgender women, transgender men appeared more likely to manifest high rates of screening at age 40-49 years (odds ratios [ORs], 1.38 [95% CI, 1.28 to 1.50]; P < .0001) and 50-59 years (OR, 1.57 [95% CI, 1.35 to 1.81]; P < .0001) but not in the older age groups. CONCLUSION The relatively high rates of screening mammography in cisgender women and transgender men suggest clinicians are offering mammography to individuals who are at average or high breast cancer risk, as compared with cisgender women, and not to transgender women at low risk.

405 Background: Clinical trials are critical to advancing prostate cancer treatment, yet patient enrollment remains a major bottleneck. Despite strong interest in AI for trial matching, real-world use remains limited due to reasons such as complex infrastructure requirements. To address this, we developed a lightweight, scalable AI framework leveraging the Google Healthcare Search API—requiring minimal technical expertise—to automate trial eligibility screening. Methods: We used a pre-trained GPT-5 large language model (LLM) to decompose the eligibility criteria of 13 prostate cancer clinical trials into 269 YES/NO questions, such that for an eligible patient of a given trial, the chart review would yield YES for inclusion and NO for exclusion questions. From 348 patients enrolled in one of the trials, we built a dataset linking each eligibility question to each patient, resulting in 4,947 patient–question pairs (2,034 inclusion pairs with YES labels and 2,913 exclusion pairs with NO labels). The task was framed as binary classification. During evaluation, for each patient–question pair, the LLM extracted search phrases from the question to retrieve relevant EHR information via the Google Healthcare Search API, then used the question and retrieved information to predict YES or NO responses. Performance was measured using accuracy, precision, recall, specificity, and F1 score. Results: Across 4,947 question–patient pairs, our framework achieved 0.81 accuracy, 0.85 precision, 0.65 recall, 0.92 specificity and a F1 score of 0.74. Results for different phases of trials is reported in Table 1, which demonstrates the approach’s robustness across trial phases. Predictions for exclusion criteria were highly precise, while inclusion criteria were more challenging. Most errors were observed in criteria involving complex nested logical conditions, interpretable clinical definitions and data not explicitly measurable or recorded in the EHR (e.g., planned treatments or clinician-assessed scores). Conclusions: This work demonstrates a practical AI framework to support clinical trial matching. The system can identify complex disease states (biochemical recurrence vs radiographic metastasis), castration status (sensitive, resistant) and genomics. Future work should focus on developing human-in-the-loop supervision and improving question decomposition to reduce errors in complex eligibility criteria. Screening performance of the proposed framework across different phases of trials. Trial Phase #Patient-Question Pairs #YES/NO labels Accuracy Precision Recall Specificity F1 score I, II, I/II 4,227 Yes: 1,735 No: 2,492 0.80 0.84 0.66 0.92 0.74 III 720 Yes: 299 No: 421 0.82 0.90 0.63 0.95 0.74 All 4,947 Yes: 2,034 No: 2,913 0.81 0.85 0.65 0.92 0.74 #YES/NO labels: The number of patient-question pairs that are labeled YES/NO among total patient-question pairs of the same row.