Citations

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.

AbstractIntroductionStructured data capture requires defined languages such as minimal Common Oncology Data Elements (mCODE). This pilot assessed the feasibility of capturing 5 mCODE categories (stage, disease status, performance status (PS), intent of therapy and intent to change therapy).MethodsA tool (SmartPhrase) using existing and custom structured data elements was Built to capture 4 data categories (disease status, PS, intent of therapy and intent to change therapy) typically documented as free‐text within notes. Existing functionality for stage was supported by the Build. Participant survey data, presence of data (per encounter), and time in chart were collected prior to go‐live and repeat timepoints. The anticipated outcome was capture of >50% sustained over time without undue burden.ResultsPre‐intervention (5‐weeks before go‐live), participants had 1390 encounters (1207 patients). The median percent capture across all participants was 32% for stage; no structured data was available for other categories pre‐intervention. During a 6‐month pilot with 14 participants across three sites, 4995 encounters (3071 patients) occurred. The median percent capture across all participants and all post‐intervention months increased to 64% for stage and 81%–82% for the other data categories post‐intervention. No increase in participant time in chart was noted. Participants reported that data were meaningful to capture.ConclusionsStructured data can be captured (1) in real‐time, (2) sustained over time without (3) undue provider burden using note‐based tools. Our system is expanding the pilot, with integration of these data into clinical decision support, practice dashboards and potential for clinical trial matching.

Abstract Purpose To determine if the COVID‐19 pandemic has further exacerbated racial disparities in late‐stage presentation of breast, colorectal, lung, and prostate cancers. Methods We conducted a registry‐based retrospective study of patients with newly reported diagnoses of breast, colorectal, lung, and prostate cancers between March 2019–June 2019 (pre‐COVID‐19) and March 2020–June 2020 (early‐COVID‐19). We compared the volume of new diagnoses and stage at presentation according to race between both periods. Results During the study period, a total of 3528 patients had newly diagnosed cancer; 3304 of which had known disease stages and were included in the formal analyses. 467 (14.1%) were Blacks, and 2743 were (83%) Whites. 1216 (36.8%) had breast, 415 (12.6%) had colorectal, 827 (25%) had lung, and 846 (25.6%) had prostate cancers, respectively. The pre‐COVID‐19 period included 2120 (64.2%), and the early‐COVID‐19 period included 1184 (35.8%), representing a proportional 44.2% decline in the volume of new cases of breast, colorectal, lung, and prostate cancers, p < 0.0001. Pre‐COVID‐19, 16.8% were diagnosed with metastatic disease, versus 20.4% early‐COVID‐19, representing a proportional increase of 21.4% in the numbers of new cases with metastatic disease, p = 0.01. There was a non‐significant proportional decline of 1.9% in Black patients diagnosed with non‐metastatic breast, colorectal, lung, and prostate cancers early‐COVID‐19 ( p = 0.71) and a non‐significant proportional increase of 7% in Black patients diagnosed with metastatic disease ( p = 0.71). Difference‐in‐difference analyses showed no statistically significant differences in metastatic presentation comparing Black to White patients. Conclusion While we identified substantial reductions in the volume of new cancer diagnoses and increases in metastatic presentations due to the COVID‐19 pandemic, the impact was similar for White and Black patients.

Objective: Reports using large brain tumor and cancer registries suggest that the incidence of vestibular schwannoma is considerably lower in the United States compared with other countries. The current study compares the incidence and disease characteristics of vestibular schwannoma within a large tertiary referral center's brain tumor and cancer registry using nationally mandated reporting protocols to disease incidence in the same population using an externally validated population-based consortium. Study Design: Population-based study spanning 1,945,007 person-years. Setting: Large tertiary referral center. Patients: Adults with sporadic vestibular schwannoma. Main Outcome Measure: Disease incidence rates from 2004 to 2016. Results: From 2004 to 2016, the incidence of vestibular schwannoma in the tumor registry was 1.3 per 100,000 person-years whereas the population-based cohort had an incidence of 4.4 per 100,000 person-years. From 2012 to 2016, the incidence in the tumor registry was 1.4 per 100,000 person-years compared with 5.2 in the population-based cohort. Patients within the population-based cohort were significantly more likely to have smaller tumors at diagnosis (78% intracanalicular versus 45%; p = 0.004) and consequently more likely to undergo management consisting of observation with serial imaging as opposed to treatment with either microsurgery or radiosurgery (71% versus 28%; p = 0.001). Conclusions: The reliance on pathology specimens and cancer-related treatment data for the national registration of new cancer and brain tumor diagnoses may introduce selection bias and underreporting of benign brain tumors that frequently involve observation as a primary treatment modality. This selection bias likely accounts for the discrepant incidence rates of vestibular schwannoma reported between the United States and other countries.

AbstractBackgroundLow molecular weight heparin (LMWH) is the guideline‐endorsed treatment for cancer‐associated venous thromboembolism (cVTE). Study objectives were to compare the efficacy and safety of rivaroxaban and enoxaparin in cVTE.MethodsUsing a cohort study design, consecutive patients with cVTE (3/1/2013‐7/31/2016), enrolled in the Mayo Thrombophilia Clinic Direct Oral Anticoagulants Registry, were compared to contemporary cancer patients receiving enoxaparin. The cumulative incidence of venous thromboembolism (VTE) recurrence, major and clinically relevant non‐major bleeding, and survival were assessed at 3 and 12 months.ResultsNinety‐eight patients received rivaroxaban (51% female, mean age 63 ± 12 years) and 168 enoxaparin (34.5% female, mean age 62 ± 15 years). The most common cancers included gastrointestinal/pancreatic, genitourinary and hematologic cancers. More than half of patients had pulmonary emboli at presentation. More than half had metastases, and two‐thirds were receiving chemotherapy. At 3 months, there were no differences in VTE recurrence (rivaroxaban 1.0% vs enoxaparin 4.2%; P = .15), major bleeding (rivaroxaban 5.1% vs enoxaparin 3.6%; P = .55), or all‐cause mortality (rivaroxaban 4.1% vs enoxaparin 8.9%; P = .14). At 12 months, these outcomes did not differ by treatment strategy.ConclusionThe results of this “real‐world” experience with cVTE suggest that rivaroxaban may offer a safe and effective alternative to LMWH.

284 Background: MM is an uncommon cancer with annual incidence of only 27,000 cases in the US. Our study determined the extent to which the number of MM patients treated annually in a treatment facility affected all-cause mortality. Methods: We used the National Cancer Data Base (NCDB) to identify adult patients with MM diagnosed from 2003-2011. NCDB is sourced from over 1,500 Commission on Cancer-accredited cancer registries representing 70% of cancer cases in the US. We classified treatment facilities by quartiles based on facility volume (mean patients/year): Quartile 1 (Q1: < 3.6), Quartile 2 (Q2: 3.6-6.1), Quartile 3 (Q3: 6.1-10.3) and Quartile 4 (Q4: > 10.3). We used hot deck imputation to account for missing data, Cox regression to analyze the association between facility volume and time-to-death, and random intercepts to adjust for multiple patients per facility. Results: There were 94,722 MM patients cared for at 1,333 facilities. Most patients (73.5%) were diagnosed and treated in the same facility. The median age at diagnosis was 67 years and 54.7% were males. The median annual facility volume was 6.1 patients/year (IQR: 3.6-10.3; range: 0.2-109.9). The distribution of patients according to facility volume was Q1 (5.2%), Q2 (12.6%), Q3 (21.9%) and Q4 (60.3%). The unadjusted median overall survival by facility volume was: Q1: 26.9 months, Q2: 29.1 months, Q3: 31.9 months and Q4: 49.1 months. After multivariable analysis adjusting for demographic (sex, age, race, ethnicity), socioeconomic (income, education, insurance type), geographic (area of residence, treatment facility location), co-morbidity (Charlson-Deyo score), and disease-specific (year of diagnosis) factors, we show that facility volume remains an independent predictor of all-cause mortality. Compared to patients treated at Q4 facilities, patients treated at lower quartile facilities had a higher risk of death (Q3 HR: 1.16 [95% CI, 1.12-1.20]; Q2: 1.21 [1.17-1.26]; Q1: 1.27 [1.22-1.34]). We observed an inverse volume-outcome relationship up to an annual facility volume of approximately 60 patients. Conclusions: MM patients treated at higher volume facilities had lower risk of mortality compared to those treated at lower volume facilities.

Purpose To determine the association between the number of patients with multiple myeloma (MM) treated annually at a treatment facility (volume) and all-cause mortality (outcome). Methods Using the National Cancer Database, we identified patients diagnosed with MM between 2003 and 2011. We classified the facilities by quartiles (Q; mean patients with MM treated per year): Q1: < 3.6; Q2: 3.6 to 6.1, Q3: 6.1 to 10.3, and Q4: > 10.3. We used random intercepts to account for clustering of patients within facilities and Cox regression to determine the volume-outcome relationship, adjusting for demographic (sex, age, race, ethnicity), socioeconomic (income, education, insurance type), geographic (area of residence, treatment facility location, travel distance), and comorbid (Charlson-Deyo score) factors and year of diagnosis. Results There were 94,722 patients with MM treated at 1,333 facilities. The median age at diagnosis was 67 years, and 54.7% were men. The median annual facility volume was 6.1 patients per year (range, 0.2 to 109.9). The distribution of patients according to facility volume was: Q1: 5.2%, Q2: 12.6%, Q3: 21.9%, and Q4: 60.3%. The unadjusted median overall survival by facility volume was: Q1: 26.9 months, Q2: 29.1 months, Q3: 31.9 months, and Q4: 49.1 months ( P < .001). Multivariable analysis showed that facility volume was independently associated with all-cause mortality. Compared with patients treated at Q4 facilities, patients treated at lower-quartile facilities had a higher risk of death (Q3 hazard ratio [HR], 1.12 [95% CI, 1.08 to 1.16]; Q2 HR, 1.17 [95% CI, 1.12 to 1.21]; Q1 HR, 1.22 [95% CI, 1.17 to 1.28]). Conclusion Patients who were treated for MM at higher-volume facilities had a lower risk of mortality compared with those who were treated at lower-volume facilities.

BACKGROUNDPrior studies have shown that a higher hospital volume or physician caseload is associated with better outcomes for complex and uncommon surgical procedures. Similar studies in the medical management of rare diseases such as hematologic cancers are limited. This retrospective, observational study using the US National Cancer Data Base determined the extent to which the number of new non‐Hodgkin lymphoma (NHL) patients treated annually at a treatment facility affected overall survival (OS).METHODSThere were 278,985 patients treated at 1151 facilities from 1998 to 2006. Treatment facilities were classified by quartiles based on the average number of new NHL patients seen annually: quartile 1 (Q1), 2 to 13 patients; quartile 2 (Q2), 14 to 20 patients; quartile 3 (Q3), 21 to 32 patients; and quartile 4 (Q4), 33 or more patients. The outcome of interest was OS according to facility volume.RESULTSThe unadjusted median OS was 61.8 months for Q1, 65.9 months for Q2, 71.4 months for Q3, and 83.6 months for Q4. A multivariate analysis that was adjusted for demographic (sex, age, race, and ethnicity), socioeconomic (income and insurance type), geographic (area of residence), disease‐specific (NHL subtype and stage), and facility‐specific factors (type and location) showed that the facility volume was associated with OS. Compared with patients at Q4 facilities, patients at lower quartile facilities had higher mortality (hazard ratio for Q3, 1.05 [95% confidence interval, 1.04‐1.06]; hazard ratio for Q2, 1.08 [95% confidence interval, 1.07‐1.10]; hazard ratio for Q1, 1.14 [95% confidence interval, 1.11‐1.17]).CONCLUSIONSNHL patients treated at higher volume facilities may survive longer than those treated at lower volume facilities. Further work is needed to understand the mechanisms of these differences and whether volume should be considered in the determination of referrals for NHL patients. Cancer 2016;122:2552–9. © 2016 American Cancer Society.

Background Data collection and review were identified as major contributors to the cost of randomized clinical trials (RCTs). Purpose We proposed and assessed a novel alternative for long-term clinical trial follow-up based on the data captured through an accredited Cancer Registry (CR) that is part of the National Cancer Database (NCDB). Methods Patients from Mayo Clinic, Rochester, enrolled in the North Central Cancer Treatment Group N934653 (COST) trial (98 patients) and the American College of Surgeons Oncology Group Z0030 trial (55 patients) were included in the study. Demographic, treatment, and long-term outcome data were compared between the hospital-based CR and the RCTs’ databases. Concordances were used to estimate the agreement between two databases. Kaplan—Meier curves were plotted to examine the consistency of time-to-event long-term outcomes of the CR and RCT databases. Results High concordances (>95%) were observed for most demographic and treatment variables between the CR data and RCT data. The vital status concordances were 100% and 94.5% between the CR and COST and Z0030 databases, respectively. Three discrepant death dates were observed, one in the COST trial and two in the Z0030 trial. The concordances of disease-free status between the CR and RCT databases were 99.0% and 87.3%, and 15 discrepant disease recurrence cases were identified: 4 for COST and 11 for Z0030. Limitations The analysis has been focused on patients from a single site, Mayo Clinic, Rochester, enrolled in two large RCT evaluating surgical treatments. The findings herein need to be confirmed in a broader setting, such as multi-center, multi-registry including nonsurgical trials. Conclusions CR data were nearly identical to data from two randomized phase III trials in different disease types and conducted by two different cooperative groups. The NCDB Cancer Registries represent a feasible alternative for obtaining long-term follow-up data for large clinical trials. Clinical Trials 2010; 7: 686—695. http:// ctj.sagepub.com