Mount Sinai Medical Center

Although validated prognostic models exist for men with metastatic castration-resistant prostate cancer (mCRPC), current tools do not incorporate genomic biomarkers such as circulating tumor DNA (ctDNA) aneuploidy or pathogenic genetic alterations (PGAs). This study aimed to estimate the prevalence of PGAs in ctDNA, assess their correlation with ctDNA aneuploidy fraction, and evaluate their association with overall survival (OS). Additionally, we developed and validated a clinical-genetic (CG) model to predict OS. We analyzed ctDNA from 776 patients enrolled in the Alliance phase 3 trial (A031201). PGAs were derived using the AR-ctDETECT assay. The net reclassification improvement (NRI) evaluated the added value of the CG model, and the time-dependent area under the receiver operating characteristic curve (tAUC) assessed the accuracy of the OS model. Feature selection using random survival forest identified gains in androgen receptor (AR), AR enhancer, MYC, RSPO2, CCND1, BRAF, and MET; losses or pathogenic variants in ZBTB16, PTEN, MSH6, PPP2R2A, NKX3-1, ZFHX3, TP53, ZNFR3, RB1, FANCA, CHECK1, APC, and CHD1; and a pathogenic AR variant. The CG model significantly outperformed the clinical model, with an average tAUC of 0.77 (95% confidence interval [CI]: 0.73-0.79) for the CG model compared with the tAUC of 0.72 (95% CI: 0.69-0.75, p=0.01) for the clinical model, and NRI of 0.29 between the models. Patients were categorized by the predicted risk scores into poor-, intermediate-, and low-risk groups with median OS of 19.6, 33.6, and 60.8mo, respectively. Incorporation of PGAs from ctDNA into a CG model improved OS prediction by nearly 30% over a clinical model. This model can classify patients into risk groups and is useful for selecting patients in future mCRPC trials.

INTRODUCTION: The kidney is the most commonly transplanted organ and renal transplant is the gold standard treatment for end-stage renal disease. Most kidneys are transplanted into the iliac fossa and usually pre-peritoneal, while the original kidneys are generally left in place. The altered location of the transplanted kidney and ureter as well as the patient’s complex medical history may present many challenges to the gynecologic laparoscopic surgeon. OBJECTIVE: To review preoperative, intraoperative, and postoperative considerations in renal transplant patients and to demonstrate the application of these considerations in a case of a patient with a transplanted kidney. METHODS: Surgical video highlighting perioperative considerations in a 37-year-old patient with a pelvic kidney and symptomatic fibroids undergoing a minimally invasive hysterectomy. RESULTS: This video demonstrates important perioperative considerations for renal transplant patients undergoing laparoscopic hysterectomy. Thorough preoperative planning is essential and includes review of imaging, current medication regimen, transplant history, baseline renal function, and consideration of ureteral stenting prior to surgery. Intraoperatively, we highlight important tips including placement of the uterine manipulator under laparoscopic guidance, careful placement of trocars, identification of the transplanted ureter, cautious intraoperative dissection, and tips to decrease the risk of impaired perfusion to the transplant kidney. Lastly, we highlight unique cystoscopy findings in transplant kidneys, as the ureter is often more anterior and located at the dome of the bladder. CONCLUSIONS: Caring for renal transplant patients can be challenging due to their complex medical history and altered anatomy. This video highlights important considerations to safely perform minimally invasive gynecologic surgeries for renal transplant patients.

Abstract We tested state-of-the-art LLMs under clinical-scale workloads using two designs: a single agent handling all tasks and a multi-agent orchestrator assigning each task to a dedicated worker. Across retrieval, extraction, and dosing tasks, batch sizes ranged from 5–80. Multi-agent accuracy remained high (90.6% at 5 tasks; 65.3% at 80), while single-agent accuracy collapsed (73.1% to 16.6%; p < 0.01). Multi-agent runs used up to 65-fold fewer tokens and limited latency growth. These findings show that lightweight orchestration preserves accuracy and efficiency under mixed-task clinical loads.

Introduction: Computerized intravenous (IV) insulin infusion calculators may offer a safe and efficacious means of achieving glycemic control. Historically, our health system utilized various multiplication factor and fixed-dose, paper-based IV insulin infusion protocols, but transitioned to a computerized insulin infusion calculator protocol based on the insulin sensitivity coefficient (ISC) in 2024. The objective of this study was to compare the safety and efficacy of the computerized ISC-based insulin infusion calculator protocol with the previous protocols. Methods: This was a pre-post intervention analysis (IRB Protocol #2023P002213) of adult patients initiated on a paper-based insulin infusion across 13 institutions from March 2023 to May 2023 for pre-analysis and patients on a computerized, ISC-based insulin infusion from November 2024 to December 2024 for post-analysis. Patients were excluded if the indication was hyperglycemic emergencies, hypertriglyceridemia, hyperkalemia, or hyperglycemia management in obstetric patients. The major endpoint was the percentage of blood glucose levels within target range (80-180 mg/dL). Secondary outcomes included rates of hyperglycemia (>200 mg/dL), severe hyperglycemia (>250 mg/dL), hypoglycemia (< 70 mg/dL), and severe hypoglycemia (< 54 mg/dL). Results: In total, 503 patients were evaluated for inclusion, of which 207 and 237 patients were included in the pre-intervention and post-intervention groups, respectively. The most common reason for exclusion was the use of IV insulin for hyperglycemia emergencies. A total of 14,183 blood glucose values were assessed between the pre- and post-intervention groups. For the major outcome, 4,610 of 6,313 (73%) and 6,163 of 7,870 (78.3%) of blood glucose values were in range in the pre- and post-intervention groups, respectively (p< 0.01). There was no statistical difference in severe hypoglycemia (0.2% vs. 0.1%, p=0.24), but hyperglycemia (16.6% vs. 12.6%, p< 0.01), severe hyperglycemia (5.7% vs. 4.0%, p< 0.01), and hypoglycemia (0.7% vs. 0.2%, p< 0.01) were significantly more common in the pre-intervention group. Conclusions: A computerized ISC-based insulin infusion protocol resulted in more blood glucose values within target range and lower rates of hypoglycemia compared to paper-based IV insulin infusion protocols.

Introduction: Severe reactions to iodinated contrast agents are uncommon, and among those, true anaphylaxis is rare. Most cases involve cutaneous or respiratory symptoms; progression to disseminated intravascular coagulation (DIC) is exceptionally unusual. A literature review by Gelbenegger et al. (2024) noted only a single previously reported case of DIC secondary to anaphylaxis. Here, we present a patient who suffered cardiac arrest shortly after contrast exposure, later found to have developed fulminant DIC. Description: A 56-year-old woman with BRCA2-positive ovarian cancer, status post total laparoscopic hysterectomy and salpingo-oophorectomy, presented for surveillance imaging. She had completed several cycles of carboplatin/paclitaxel and had no known allergies. Immediately following contrast administration for CT imaging, she developed acute cardiovascular collapse and suffered pulseless electrical activity arrest. Return of spontaneous circulation was achieved following ACLS. She was administered corticosteroids and underwent evaluation for ECMO. Laboratory studies revealed leukocytosis, thrombocytopenia, elevated INR, low fibrinogen, and markedly elevated d-dimer—findings consistent with acute consumptive coagulopathy. Serum tryptase was significantly elevated, supporting an anaphylactic etiology. Imaging revealed mediastinal, retroperitoneal, and soft tissue hemorrhage, but no evidence of malignancy recurrence. Despite vasopressors and massive transfusion, she progressed to refractory shock. Goals-of-care discussions led to transition to comfort measures. Discussion: Mechanistically, anaphylaxis can trigger DIC through cytokine release, thrombin generation, and systemic endothelial injury. Prior cases of anaphylaxis-induced DIC have been reported in perioperative and obstetric settings (Kuczkowski, 2003; Yamashita & Matsumoto, 2021), but reports involving contrast-induced reactions are exceedingly rare. This case underscores a rare but catastrophic sequence: contrast-induced anaphylaxis leading to cardiac arrest and DIC. Clinicians should maintain vigilance for coagulopathy in severe anaphylaxis and consider early serum tryptase testing to support diagnosis.

Agentic large language models (LLMs) increasingly rely on retrieved sources and tools, but their ability to reject these tools which undergo adversarial modification is uncertain. We evaluated 21 LLMs on 500 physician-validated emergency department and inpatient vignettes across 12 medical domains. For each vignette, models chose between an authentic guideline excerpt and a sham version with one adversarial modification, presented in random order (10,500 agentic decisions). Models selected the sham in 40.6% of evaluations (59.4% accuracy), with the highest failure rates for safety-critical changes including removed warnings, deleted allergy information, contraindication violations and dosing errors (54.2% to 61.7% failure). Choices were dominated by presentation bias: models favored the first option in 72.7% of decisions, shifting accuracy from 36.7% to 82.3% depending on sham position. Guideline selection in agentic systems is therefore vulnerable to poisoned sources and may require independent verification and ranking safeguards before clinical deployment. This finding is important especially in low-resource environments relying on AI agents as primary public health gatekeepers face disproportionate risks from poisoned tools.

Abstract Background Triple negative breast cancer (TNBC) accounts for approximately 15-20% of incident breast cancers and until recently, was the only BC subtype without targeted therapies. Recent therapeutic advances, including the use of immunotherapy (IO), have highlighted the importance of the spatial tumor microenvironment which includes both the number and location of infiltrating cancer immune cells. We previously reported on a redistribution of recurrence risk in TNBC patients from Mount Sinai using the PreciseBreast (PDxBR) test that includes AI-derived morphologic features of tumor architecture, nuclear size, stroma, mitotic figures and lymphocyte content combined with age, tumor size, anatomic stage and lymph node status. We conducted this study to gain a better understanding of the prognostic performance of PreciseBreast (PDxBR) for TNBC in an independent patient cohort. Methods Patients with H&amp;E stained slides from surgical excisions of TNBC stored in the UCLA biorepository (2008-2017) were sent along with clinical outcomes data to perform the PDxBR test. H&amp;E slides were digitized and imaged on a Philips Ultra-Fast whole slide scanning system. Images were deconstructed to morphologic analytes using an AI-enabled platform to quantify tumor cell and tissue architectural features. Eight quantitative morphologic features (AI-grade model) were combined with 4 clinical factors to determine risk of recurrence. The AI-grade model was also evaluated. We applied an optimized PDxBR test for TNBC with risk scores 0-100 using an adjusted threshold &amp;lt;60 low risk, ≥60 high risk. Recurrence events included local regional (LR), second primary (SP), distant metastasis (DM), or death (D). Prognostic performance was analyzed using hazard ratios (HR), sensitivity (Se), specificity (Sp), negative predictive value (NPV) and positive predictive value (PPV). Results Characteristics of 77 patients with TNBC: median age 58 years (yrs), 66% &amp;gt;50yrs, 86% non-black; 62% tumor size ≤2.5cm and 30% &amp;gt;2.5cm-5cm; 87% Stage 1/2, 68% pN0, 32% pN1-3+, 88% Grade 3; 82% received chemotherapy. There were 14 events (18%) including 6 D, 4 DM, 2 LR, 2 SP with a median follow-up of 7.9 yrs. PDxBR identified 30 (39%) as high risk, 47 (61%) as low risk compared to the AI-grade model with 64 (83%) high risk, 13 (17%) low risk. PDxBR, which includes the AI-grade model + clinical features, was the only model with a significant HR of 3.72 (95% CI, 1.26-10.97, p = 0.0174), with Se 0.57 (95% CI 0.32-0.79), Sp 0.65 (95% CI, 0.53-0.76), PPV 0.27 (95% CI 0.14-0.45) and NPV 0.87 (95% CI, 0.75-0.94). PDxBR classified 8/14 events as high risk including 2 DM, 2LR, 1SP, and 3D. 4 of 14 events (3 D, 1 DM) included tumors that were ≥ 3cm of which 2 had 1-2pN. 6/14 patients with events did not have chemotherapy and 5 of 6 were high risk by PDxBR. The most significant PDxBR morphologic features for event identification included mitotic figure ‘hot spot’ quantification and tumor architecture (i.e. degree of differentiation). Conclusions PDxBR introduces a novel approach towards risk stratification and outcome prediction in a clinically high risk TNBC population. Standardized AI digital pathology-based platforms have the potential to accurately re-assess risk and response to chemotherapy and IO, especially in randomized clinical trials. Citation Format: N. Stanzione, G. Fernandez, S. Vaisman, A. Sainath Madduri, R. Scott, M. Prastawa, X. Zhang, M. Donovan. A novel approach for phenotyping triple negative breast cancer using an Artificial Intelligence digital pathology-based prognostic test to assess recurrence risk and response to therapy [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS4-12-28.