Citations

TPS3650 Background: A total neoadjuvant therapy (TNT) approach improves compliance with chemotherapy and increases rates of tumor response compared to neoadjuvant chemoradiation (CRT) alone in those with locally advanced rectal cancer. Recent data indicate that optimal sequencing of TNT involves consolidation (rather than induction) chemotherapy to improve complete response rates. The use of FOLFIRINOX has shown to improve response and outcomes compared to CRT and surgery alone. Data have also shown that patients with clinical complete response (cCR) after TNT may be managed with a watch and wait approach (WW) instead of preemptive total mesorectal resection (TME). However, the optimal consolidation chemotherapy regimen to improve cCR rates has not been established, and a randomized clinical trial has not robustly evaluated cCR as a primary endpoint. We designed this NCI-sponsored study of chemotherapy intensification to address this and to increase cCR rates, provide opportunity for organ preservation, and survival outcomes. Methods: In this multigroup randomized, seamless phase II/III trial (1:1), up to 760 patients with LARC, T4N0, any T with node positive disease (any T, N+) or T3N0 requiring abdominoperineal resection or coloanal anastomosis and distal margin within 12 cm of anal verge will be enrolled. Stratification factors include tumor stage (T4 vs T1-3), nodal stage (N+ vs N0) and distance from anal verge (0–4; 4–8; 8–12 cm). Patients will be randomized to receive neoadjuvant long-course chemoradiation (LCRT) followed by consolidation doublet (mFOLFOX6 or CAPOX (control arm)) or triplet chemotherapy (FOLFIRINOX (experimental arm)) for 3–4 months. LCRT in both arms involve 4500 cGy in 25 fractions over 5 weeks +900 cGy boost in 5 fractions with a fluoropyrimidine. Patients will undergo assessment 8–12 (±4) weeks post-TNT completion. The primary endpoint for the phase II portion will compare cCR between treatment arms. A total number of 312 patients (156 per arm) will provide statistical power of 90.5% to detect a 17% increase in cCR rate, at a one-sided alpha = 0.048. The primary endpoint for the phase III portion will compare disease-free survival (DFS) between arms. A total of 285 DFS events will provide 85% power to detect an effect size of hazard ratio 0.70 at a one-sided alpha of 0.025, requiring enrollment of 760 patients (380 per arm). Secondary objectives include overall survival, organ preservation time, time to distant metastasis, and adverse event rates. This study has accrued 587 patients as of January 2025, and is investigating exploratory correlatives (e.g., ctDNA). Support: U10CA180821, U10CA180882, U24 CA196171. https://acknowledgments.alliancefound.org. Clinicaltrials.gov ID: NCT05610163. Clinical trial information: NCT05610163 .

3624 Background: Increasing chemotherapy completion and reducing chemotherapy adverse events (AE) are critical to improve survival after colon cancer diagnosis. Sex, as a biological variable, may impact chemotherapy treatment differently, and thus further investigation is needed to examine sex differences in chemotherapy completion and adverse events among patients with colon cancer. Methods: Among an NCI-sponsored trial conducted among patients with stage III colon cancer (CALGB/SWOG 80702), all patients received standard adjuvant chemotherapy FOLFOX (fluorouracil, leucovorin, and oxaliplatin). To signal chemotherapy completion, we utilized relative dose intensity (RDI) calculated as the ratio of the delivered dose intensity to planned dose intensity; and reduced RDI (RDI <85%) was considered as a clinically significant deviation from standard FOLFOX. From clinicians’ records of NCI’s Common Terminology Criteria for Adverse Events (CTCAE), we primarily focused on clinically significant AE such as neutrophils decrease, nausea, platelets decrease, hypertension, peripheral neuropathy, diarrhea, fatigue, gastritis, creatinine increase, gastric ulcer, myocardial ischemia, and cerebral ischemia; and severe AE was defined as the occurrence of any above AE with CTCAE grade ≥3. Using multivariable logistic regression that adjusted for body surface area (BSA) and other clinicopathological confounders, we estimated adjusted odds ratios (OR) for the associations of sex with reduced RDI and severe AE. Results: Of 2201 patients, mean (standard deviation [SD]) age was 60.9 (10.9) years, 1019 (46.3%) were female, 1750 (79.5%) were White, 172 (7.8%) were Hispanic, 964 (43.8%) experienced reduced RDI, and 1156 (52.5%) had severe adverse events. Compared to males, females were at significantly higher risks of experiencing reduced RDI (OR [95% CI]: 1.57 [1.27-1.93], P <0.001) and severe AE (OR [95% CI]: 1.73 [1.41-2.12], P <0.001). Conclusions: Our findings suggested females are more likely than men to experience reduced RDI and severe AE during colon cancer chemotherapy. Clinical Impact: In the era of precision medicine, sex (as a biological variable) should be considered in optimizing colon cancer chemotherapy to improve completion and reduce toxicities. Clinical trial information: NCT01150045 .

e14626 Background: Despite advancements in neoadjuvant therapy and surgical techniques, patients with pancreatic ductal adenocarcinoma (PDAC) often experience high recurrence rates after total neoadjuvant therapy (TNT) and resection. Existing tumor response scoring systems, such as the College of American Pathologists (CAP) guidelines, fail to stratify recurrence risk effectively, emphasizing the need for novel, objective approaches. Methods: This study evaluated spatial characteristics of residual cancer and stroma in a multi-site cohort of 203 patients with resected PDAC following TNT, not having achieved a major pathologic response (CAP 2 or 3). Whole slide images (WSI) of hematoxylin and eosin-stained sections were digitized and analyzed using artificial intelligence to perform tissue segmentation. Metrics quantifying spatial morphology and configuration, inspired by principles of landscape ecology, were computed to explore spatial features of the tumor microenvironment (TME). Statistical models (recursive partitioning and regression tree [RPRT] model and Cox model with backwards eliminations) incorporating these features were developed to stratify recurrence risk. Results: Five metrics were used for risk classification model build. Three models successfully classified patients into high- and low-risk groups based on spatial metrics. Models demonstrated moderate discriminatory power (C-statistics: 0.560–0.566) and showed that increased tumor fragmentation, like stromal patch density and cancer patch shape index, were strongly associated with improved outcomes. Conclusions: Integrating spatial metrics of cancer-stroma configuration enhances risk stratification beyond conventional scoring systems. Given increased tumor fragmentation is associated with improved outcomes, these data indirectly suggest ecological mechanisms of treatment resistance. This approach offers a pathway toward personalized post-operative management strategies to improve outcomes in PDAC. Risk classification models. Model Description Event/n Median DFS (95% CI) HR(95% CI) p-value C-statistics 1(RPRT) High risk: log standard deviation of the stroma shape index > 0.651 87 / 101 8.22(5.72 – 10.1) 1.64(1.20 – 2.25) 0.002 0.562 Low risk: log standard deviation of the stroma shape index ≤ 0.651 71 / 102 11.90(9.24 – 25.1) Ref 2(RPRT) High risk: log standard deviation of the stroma shape index > 0.651 & mean cancer shape index ≤ 1.775 58 / 63 7.23(5.19 – 9.63) 1.82(1.31 – 2.52) 0.0003 0.560 Low risk: (log standard deviation of the stroma shape index > 0.651 & mean cancer shape index > 1.775) or log standard deviation of the stroma shape index ≤ 0.651 100 / 140 11.57(9.24 – 17.92) Ref 3(Backwards elimination) High risk: PI * > median 86 / 101 8.02(6.08 – 9.86) 1.73(1.26 – 2.37) 0.0007 0.566 Low risk:PI * ≤ median 72 / 102 15.72(9.97 – 25.05) Ref *PI: (0.2701 * log mean stroma area) + (0.3905 * log edge density).

e13656 Background: Evidence-based medicine is crucial for clinical decision-making, yet studies suggest that a significant proportion of treatment decisions do not fully incorporate the latest evidence. Large Language Models (LLMs) show promise in bridging this gap, but their reliability for medical recommendations remains uncertain. Methods: We conducted an evaluation study comparing five LLMs’ recommendations across 50 clinical scenarios related to multiple myeloma diagnosis, staging, treatment, and management. The evaluation included three general-purpose LLMs (OpenAI o1-preview, Claude 3.5 Sonnet, Gemini 1.5 Pro), one retrieval-augmented generation (RAG) system (Myelo), and one agentic workflow-based system (HopeAI). General-purpose LLMs generated responses based solely on their internal knowledge, while the RAG system enhanced these capabilities by incorporating external knowledge retrieval. The agentic workflow system extended the RAG approach by implementing multi-step reasoning and coordinating with multiple tools and external systems for complex task execution. Three independent hematologist-oncologists evaluated the LLM-generated responses using standardized scoring criteria developed specifically for this study. Performance assessment encompassed five dimensions: accuracy, relevance, comprehensiveness, hallucination rate, and clinical use readiness. Results: HopeAI demonstrated superior performance across accuracy (82.0%), relevance (85.3%), and comprehensiveness (74.0%), compared to OpenAI o1-preview (64.7%, 57.3%, 36.0%), Claude 3.5 Sonnet (50.0%, 51.3%, 29.3%), Gemini 1.5 Pro (48.0%, 46.0%, 30.0%), and Myelo (58.7%, 56%, 32.7%). Hallucination rates were consistently low across all systems: HopeAI (5.3%), OpenAI o1-preview (3.3%), Claude 3.5 Sonnet (10.0%), Gemini 1.5 Pro (8.0%), and Myelo (5.3%). Clinical use readiness scores were relatively low for all systems: HopeAI (25.3%), OpenAI o1-preview (6.0%), Claude 3.5 Sonnet (2.7%), Gemini 1.5 Pro (4.0%), and Myelo (4.0%). Conclusions: This study demonstrates that while current LLMs show promise in medical decision support, their recommendations require careful clinical supervision to ensure patient safety and optimal care. Further research is needed to improve their clinical use readiness before integration into oncology workflows. These findings provide valuable insights into the capabilities and limitations of LLMs in oncology, guiding future research and development efforts toward integrating AI into clinical workflows.

7547 Background: Minimal residual disease (MRD) has been recently accepted by the Food and Drug Administration (FDA) as an endpoint for accelerated approval in Multiple Myeloma. However, emerging data from recent trials were not included in previous analyses. While literature-based meta-analyses on the correlation between MRD and approved clinical outcomes typically require extensive manual review, leveraging AI with expert-in-the-loop validation can efficiently generate reliable evidence from comprehensive clinical studies with up-to-date outcomes. Methods: An AI-assisted framework was developed to identify relevant studies and extract critical information via two independent objectives. The first objective examined trial-level associations, modeling treatment effects on MRD and clinical endpoints across patient populations using weighted least squares, with association strength measured by coefficients of determination (R²) and 95% confidence intervals (CIs). The second objective analyzed individual-level associations using synthetic individual patient data (SynthIPD) generated from the published Kaplan-Meier plots and summary statistics of patient subgroups. Results: AI-assisted screening identified eligible studies (>50 patients per treatment arm) reporting progression-free survival (PFS), overall survival (OS), and MRD-negative complete response rates (MRD-CR rate) using multi-parameter next-generation flow cytometry or sequencing methods (sensitivity threshold ≥10⁻⁵), expanding previous analyses from 15 to 20 two-arm studies. Trial-level analysis demonstrated an R² of 0.69 (95% CI 0.50–0.89) for PFS log hazard ratio versus MRD-CR rate log odds ratio. Analysis of synthetic individual data from Kaplan-Meier curves using a novel digitization method yielded a global odds ratio of 7.28 (95% CI 5.60–8.95) for individual-level correlation between MRD-CR rates and PFS outcomes. Conclusions: This study validates MRD-CR rate as an endpoint for accelerated approval in MM through rapid AI-assisted literature review and synthetic individual patient data. The findings demonstrate moderate correlation between MRD-CR rate and median PFS at both trial and individual levels, consistent with previous literature but incorporating additional eligible studies. The novel SynthIPD approach presents an efficient alternative to traditional data-sharing methods while maintaining analytical robustness. These results align with current Oncologic Drugs Advisory Committee (ODAC) surrogacy analysis methods and support the utility of MRD assessment in MM clinical trials.

2011 Background: Radiation therapy forms the mainstay of management for patients with brain metastases. Published randomized trials have found improved quality of life with stereotactic radiation (SRS/SRT) over whole brain radiation (WBRT) in patients with ≤ 4 brain metastases; comparative trials in patients with >4 brain metastases are lacking. In addition, prior randomized trials have demonstrated the superiority of hippocampal avoidance WBRT (HA-WBRT) over traditional WBRT, but no study has compared SRS/SRT to HA-WBRT. Accordingly, we conducted a multicenter, phase 3 randomized trial comparing SRS/SRT to HA-WBRT in patients with 5-20 brain metastases. Methods: Eligible patients were age 18-80 with 5-20 brain metastases secondary to a solid primary other than small cell lung cancer, were naïve to prior brain-directed radiation, and lacked leptomeningeal disease. The primary endpoint was the average of patient-reported symptom severity and interference over the first six months post-baseline relative to baseline, using the MD Anderson Symptom Inventory–Brain Tumor (MDASI-BT) module, a validated instrument assessing 22 symptoms and 6 interference measures integral to quality of life, each scored 0-10 with higher scores indicating greater symptomatology/interference in function. The target effect size was a symptom severity of 0.70, corresponding to 50% of the observed difference between patients with a good (90-100) versus poor (≤80) Karnofsky performance status; with 80% power and a two-sided alpha of 0.05, 196 patients were required. Results: Between 4/2017-5/2024, 196 patients enrolled, 98 in each arm. The median number of brain metastases was 14 (IQR 11-18); 25% of patients underwent prior neurosurgical resection. Baseline mean MDASI-BT symptom severity scores were 2.2 (SRS/SRT arm) and 1.9 (HA-WBRT arm), p=0.20; respective interference scores were 3.5 and 3.2 (p=0.40). The average of weighted post-baseline severity and interference scores relative to baseline indicated lower symptomatology/inference in the SRS/SRT arm, meeting the primary endpoint of the study (difference between SRS/SRT and HA-WBRT: -1.06, p<0.001). Averaged post-baseline symptom severity scores minus baseline were -0.03 and 0.59 in the SRS/SRT and HA-WBRT arms, respectively (difference -0.62, with lower symptom severity in the SRS/SRT arm, p<0.001); respective interference estimates were -0.62 and 0.89 (difference -1.50, with lower interference in the SRS/SRT arm, p<0.001). Median survival was 8.3 and 8.5 months in the SRS/SRT and HA-WBRT arms, respectively (p=0.30). Conclusions: This phase 3 randomized trial indicates that patients with 5-20 brain metastases experience fewer symptoms and less interference in function after SRS/SRT as opposed to HA-WBRT, without compromise of survival, supporting SRS/SRT as the standard of care in this population. Clinical trial information: NCT03075072 .

Importance Nontraumatic subarachnoid hemorrhage (SAH) represents the third most common stroke type with unique etiologies, risk factors, diagnostics, and treatments. Nevertheless, epidemiological studies often cluster SAH with other stroke types leaving its distinct burden estimates obscure. Objective To estimate the worldwide burden of SAH. Design, Setting, and Participants Based on the repeated cross-sectional Global Burden of Disease (GBD) 2021 study, the global burden of SAH in 1990 to 2021 was estimated. Moreover, the SAH burden was compared with other diseases, and its associations with 14 individual risk factors were investigated with available data in the GBD 2021 study. The GBD study included the burden estimates of nontraumatic SAH among all ages in 204 countries and territories between 1990 and 2021. Exposures SAH and 14 modifiable risk factors. Main Outcomes and Measures Absolute numbers and age-standardized rates with 95% uncertainty intervals (UIs) of SAH incidence, prevalence, mortality, and disability-adjusted life-years (DALYs) as well as risk factor–specific population attributable fractions (PAFs). Results In 2021, the global age-standardized SAH incidence was 8.3 (95% UI, 7.3-9.5), prevalence was 92.2 (95% UI, 84.1-100.6), mortality was 4.2 (95% UI, 3.7-4.8), and DALY rate was 125.2 (95% UI, 110.5-142.6) per 100 000 people. The highest burden estimates were found in Latin America, the Caribbean, Oceania, and high-income Asia Pacific. Although the absolute number of SAH cases increased, especially in regions with a low sociodemographic index, all age-standardized burden rates decreased between 1990 and 2021: the incidence by 28.8% (95% UI, 25.7%-31.6%), prevalence by 16.1% (95% UI, 14.8%-17.7%), mortality by 56.1% (95% UI, 40.7%-64.3%), and DALY rate by 54.6% (95% UI, 42.8%-61.9%). Of 300 diseases, SAH ranked as the 36th most common cause of death and 59th most common cause of DALY in the world. Of all worldwide SAH-related DALYs, 71.6% (95% UI, 63.8%-78.6%) were associated with the 14 modeled risk factors of which high systolic blood pressure (population attributable fraction [PAF] = 51.6%; 95% UI, 38.0%-62.6%) and smoking (PAF = 14.4%; 95% UI, 12.4%-16.5%) had the highest attribution. Conclusions and Relevance Although the global age-standardized burden rates of SAH more than halved over the last 3 decades, SAH remained one of the most common cardiovascular and neurological causes of death and disabilities in the world, with increasing absolute case numbers. These findings suggest evidence for the potential health benefits of proactive public health planning and resource allocation toward the prevention of SAH.

ABSTRACTBackgroundDeep learning (DL) models for accurate renal tumor characterization may benefit from multi‐center datasets for improved generalizability; however, data‐sharing constraints necessitate privacy‐preserving solutions like federated learning (FL).PurposeTo assess the performance and reliability of FL for renal tumor segmentation and classification in multi‐institutional MRI datasets.Study TypeRetrospective multi‐center study.PopulationA total of 987 patients (403 female) from six hospitals were included for analysis. 73% (723/987) had malignant renal tumors, primarily clear cell carcinoma (n = 509). Patients were split into training (n = 785), validation (n = 104), and test (n = 99) sets, stratified across three simulated institutions.Field Strength/SequenceMRI was performed at 1.5 T and 3 T using T2‐weighted imaging (T2WI) and contrast‐enhanced T1‐weighted imaging (CE‐T1WI) sequences.AssessmentFL and non‐FL approaches used nnU‐Net for tumor segmentation and ResNet for its classification. FL‐trained models across three simulated institutional clients with central weight aggregation, while the non‐FL approach used centralized training on the full dataset.Statistical TestsSegmentation was evaluated using Dice coefficients, and classification between malignant and benign lesions was assessed using accuracy, sensitivity, specificity, and area under the curves (AUCs). FL and non‐FL performance was compared using the Wilcoxon test for segmentation Dice and Delong's test for AUC (p < 0.05).ResultsNo significant difference was observed between FL and non‐FL models in segmentation (Dice: 0.43 vs. 0.45, p = 0.202) or classification (AUC: 0.69 vs. 0.64, p = 0.959) on the test set. For classification, no significant difference was observed between the models in accuracy (p = 0.912), sensitivity (p = 0.862), or specificity (p = 0.847) on the test set.Data ConclusionFL demonstrated comparable performance to non‐FL approaches in renal tumor segmentation and classification, supporting its potential as a privacy‐preserving alternative for multi‐institutional DL models.Evidence Level4.Technical EfficacyStage 2.

Background: Flow cytometric detection of minimal residual disease (MRD) in chronic lymphocytic leukemia (CLL) is complex, time-consuming, and subject to inter-operator variability. Deep neural networks (DNNs) offer potential for standardization and efficiency improvement, but require rigorous validation and monitoring for safe clinical implementation. Methods: We evaluated a DNN-assisted human-in-the-loop approach for CLL MRD detection. Initial validation included method comparison against manual analysis (n = 240), precision studies, and analytical sensitivity verification. Post-implementation monitoring comprised four components: daily electronic quality control, input data drift detection, error analysis, and attribute acceptance sampling. Laboratory efficiency was assessed through a timing study of 161 cases analyzed by five technologists. Results: Method comparison demonstrated 97.5% concordance with manual analysis for qualitative classification (sensitivity 100%, specificity 95%) and excellent correlation for quantitative assessment (r = 0.99, Deming slope = 0.99). Precision studies confirmed high repeatability and within-laboratory precision across multiple operators. Analytical sensitivity was verified at 0.002% MRD. Post-implementation monitoring identified 2.97% of cases (26/874) with input data drift, primarily high-burden CLL and non-CLL neoplasms. Error analysis showed the DNN alone achieved 97% sensitivity compared to human-in-the-loop-reviewed results, with 13 missed cases (1.5%) showing atypical immunophenotypes. Attribute acceptance sampling confirmed 98.8% of reported negative cases were true negatives. The DNN-assisted workflow reduced average analysis time by 60.3% compared to manual analysis (4.2 ± 2.3 vs. 10.5 ± 5.8 min). Conclusions: The implementation of a DNN-assisted approach for CLL MRD detection in a clinical laboratory provides diagnostic performance equivalent to expert manual analysis while substantially reducing analysis time. Comprehensive performance monitoring ensures ongoing safety and effectiveness in routine clinical practice. This approach provides a model for responsible AI integration in clinical laboratories, balancing automation benefits with expert oversight.

ABSTRACT Evidence indicates that the integrity of in utero development influences late life healthy or unhealthy aging; however, specific links between them are unclear. Histone chaperone HIRA is thought to play a role in both life stages, and here, we explore this role using the murine pigmentary system by investigating and comparing the effects of its lineage‐specific knockout, either conditionally during embryogenesis or postnatally. Embryonic knockout of Hira in tyrosinase+ neural crest‐derived lineages, including melanoblasts, led to reduced melanoblast numbers during embryogenesis, with single‐cell RNA sequencing analysis indicating evidence of lineage‐specificity defects. This was supported in an in vitro model using melb‐a melanoblasts in which Hira knockdown affected lineage identity and melanoblast differentiation potential, with ATAC‐seq data indicating a role of HIRA in orchestrating chromatin accessibility. Interestingly, however, newborn Hira knockout mice had wild type numbers of differentiated melanocytes, albeit functionally defective, as demonstrated by very mild hypopigmentation of the first hair coat, increased melanocyte telomere‐associated DNA damage foci, and impaired response to proliferative challenge. Moreover, as they aged, mice with embryonic melanoblast Hira knockout displayed marked defects in melanocyte stem cell maintenance and premature hair graying. Importantly, this phenotype was not observed after postnatal inducible knockout, indicating an essential role for HIRA at embryonic stages that is transmitted to adulthood, rather than a direct postnatal requirement within the pigmentary system. This genetic model shows that HIRA function during early development lays a foundation for maintaining lineage identity and subsequent maintenance of adult tissue‐specific stem cells during aging.