Faculty Bibliography

BACKGROUND:Robotic pancreaticoduodenectomy (RPD) is increasingly performed in the United States. Understanding factors associated with safe adoption of RPD is critical to reducing perioperative mortality during the learning curve. METHODS:Using the Epic Cosmos database, the study identified adult patients (age ≥18 years) who underwent pancreaticoduodenectomy (PD) between 2019 and 2025. Modified Poisson regression was used to assess factors associated with 30-day mortality using adjustment for age, sex, race, ethnicity, insurance, marital status, rural/urban residence, socioeconomic status, and diagnosis. Among surgeons performing two or more RPDs, mortality trends were analyzed across case-number thresholds. Mortality risk was assessed by cumulative RPD and open PD (OPD) experience, with adjustment for age and diagnosis. RESULTS:Among 23,995 patients with a median age of 69 years (interquartile range [IQR], 62-75 years), 1578 (6.6 %) underwent RPD. Use of RPD increased from 4% of PD in 2019 to 10% in 2025. The 30-day mortality was higher for RPD than for OPD (2.7 % vs 2.0 %; adjusted relative risks [aRR], 1.43 (IQR, 1.02-1.95; p = 0.029). In RPD, mortality decreased with increasing surgeon prior experience: 3.9 % (Q1: 0-1 cases), 3.9 % (Q2: 2-4 cases), 2.22 % (Q3: 5-8 cases), 2.67 % (Q4: 9-18 cases), 0.92 % (Q5: 19-71 cases). Increased RPD experience was associated with decreased mortality (per doubling RPD experience: aRR, 0.78 (95 % confidence interval [CI], 0.63-0.96; p = 0.02). The patients who underwent PD between 2023 and 2025 showed no adjusted increase in mortality with robotic technique (aRR, 1.04; 95 % CI, 0.61-1.65; p = 0.85). CONCLUSIONS:Nationwide, adoption of RPD is associated with increased 30-day mortality, which decreases substantially with increasing surgeon RPD experience. These findings suggest that structured, competency-based training pathways are needed to ensure safe dissemination of novel technology, including RPD.

BACKGROUND:Before KAS250 (circles-based allocation), donor service area (DSA) of listing was the largest contributor to deceased donor kidney transplantation (DDKT) rate disparities. Both before and after KAS250, it is unclear to what extent DSA-level disparities are attributable to center-level practice variation. We aimed to disentangle contributions to DDKT rate variation from: (1) center practices, (2) kidney distribution within sharp policy boundaries (DSAs, OPTN Regions), and (3) other geographic variation in kidney scarcity. METHODS:With national transplant registry data, we studied transplant rate variation in the pre-KAS250 era, which prioritized patients based on DSAs and Regions, and under KAS250, which prioritizes patients within 250 nautical mile circles. We modeled candidate DDKT rates with multilevel Poisson regression, adjusting for candidate factors, and calculated median incidence rate ratios (MIRR) to summarize variation attributable to DSAs, OPTN regions, states, census divisions, and to centers within those units. RESULTS:). Adjusted center-level DDKT rates under KAS250 were highly associated with offer acceptance rates (ρ = 0.60, p < 0.001). CONCLUSIONS:Though geographic disparities are driven primarily by center-level practice differences including offer acceptance, KAS250 did reduce DSA-level disparities. Further allocation policy changes are unlikely to substantially reduce geographic variation in DDKT rates.

BACKGROUND AND OBJECTIVES/OBJECTIVE:Traditional medical board examinations present clinical information in static vignettes with multiple-choices (MC), fundamentally different from how physicians gather and integrate data in practice. Recent advances in large language models (LLMs) offer promising approaches to creating more realistic clinical interactive conversations. However, these approaches are limited in neurosurgery, where patient communication capacity varies significantly and diagnosis heavily relies on objective data such as imaging and neurological examinations. We aimed to develop and evaluate a multi-artificial intelligence (AI) agent conversation framework for neurosurgical case assessment that enables realistic clinical interactions through simulated patients and structured access to objective clinical data. METHODS:We developed a framework to convert 608 Self-Assessment in Neurological Surgery first-order diagnosis questions into conversation sessions using 3 specialized AI agents: patient AI for subjective information, system AI for objective data, and clinical AI for diagnostic reasoning. We evaluated generative pretrained transformer 4o's (GPT-4o's) diagnostic accuracy across traditional vignettes, patient-only conversations, and patient + system AI interactions, with human benchmark testing from 10 neurosurgery residents. RESULTS:= .0030) using fewer interactions and reported high educational value of the interactive format. CONCLUSION/CONCLUSIONS:This multi-AI agent framework provides both a more challenging evaluation method for LLMs and an engaging educational tool for neurosurgical training. The significant performance drops in conversational formats suggest that traditional MC testing may overestimate LLMs' clinical reasoning capabilities, while the framework's interactive nature offers promising applications for enhancing medical education.

Since their early development in the 1980s, Simulated Allocation Models (SAMs) have helped policymakers forecast the impact of proposed allocation policy changes on patient outcomes before implementation. In the United States, models like the Kidney-Pancreas Simulated Allocation Model, Liver Simulated Allocation Model, and Thoracic Simulated Allocation Model have been instrumental in shaping organ allocation policies. Analogous models have emerged globally, including the ETKidney and Eurotransplant Liver Allocation System simulators for the Eurotransplant region, to address country and region-specific allocation challenges. This review categorizes and compares SAMs based on their core assumptions, data, and modeling approaches. We highlight challenges in model validation, the use of synthetic data, and model transparency. While simplifying assumptions are often necessary because of limited data, their influence on results should be clearly communicated to ensure policymakers can interpret model predictions accurately. Furthermore, model validation using both retrospective and prospective data is essential to assess performance under evolving policies. Greater transparency through open-source models, detailed reporting of assumptions, and validation efforts can enhance collaboration, reproducibility, and confidence in transplant research. By providing a global perspective on SAMs, this review aims to inform future research and policy development, promoting evidence-based policy development in organ transplantation.

Liver transplant (LT) recipients experience a wide range of comorbidities, leading to frequent healthcare encounters. Until now, national registries, which have limited exposures and outcomes, and laborious small cohort studies have been the main data sources for LT research. Cosmos database offers electronic health record (EHR)-based insights into LT recipients at the national level with granular data. We evaluated if Cosmos data is representative of the entire US LT recipient population. Using Cosmos (N=20,235) and the national Scientific Registry of Transplant Recipients (SRTR) (N=51,281), we identified adult, first-time LT recipients between 7/2016-12/2022. We compared demographics, clinical data, and mortality across datasets, calculating Kaplan-Meier survival estimates and multi-variable Cox regressions. Recipient characteristics were highly comparable (e.g., female: Cosmos=36.5% vs. SRTR=36.4%, Black: 6.8% vs. 7.2%; BMI: 28.5 kg/m2 [24.8-32.9] vs. 28.2 [24.6-32.4]). Lab values were similar across cohorts, including MELD (24 [17-30] vs. 23 [16-30]). Transplant indications, donor characteristics, and 5-year survival (Cosmos 83.1% [82.3-83.8) vs. SRTR 80.9% [80.4-81.3]) were similar. The associations of clinical factors with survival were similar across both groups. Cosmos database demonstrated acceptable generalizability to the general US LT recipient population, which may advance LT research through a better understanding about LT recipients' experiences and outcomes.

Artificial intelligence (AI) is rapidly transforming healthcare, and the field of kidney transplantation (KT) is no exception. While much of the AI-related work has focused on deceased donor KT, there is a growing body of research applying AI tools to living kidney donation (LKD). This review explores AI's current and potential roles in LKD, focusing on predictive and social applications of AI in LKD. Additionally, we discuss the challenges and limitations of implementing AI in clinical settings and highlight emerging research trends. This review consolidates existing research and provides a foundation for both transplant professionals and data scientists seeking to integrate AI responsibly into living donor programs.

PURPOSE OF REVIEW/OBJECTIVE:With the kidney nonuse rate approaching 30%, one-fifth of kidneys placed out of sequence, and unsatisfactory long-term recipient outcomes, U.S. kidney allocation is at a crossroads. This review highlights recent studies and efforts geared toward improving the system. RECENT FINDINGS/RESULTS:The Organ Procurement & Transplantation Network's kidney transplantation committee is tasked with migrating kidney allocation policy, currently based on 250 nautical mile circles, to a fully continuous, points-based system. Challenges in designing a system assured to improve, not worsen, placement efficiency have hampered progress. OPO and transplant center practice patterns have adapted to a rapidly changing donation landscape. Advances in transplant immunology, particularly involving molecular HLA typing methods, are opening doors for more precise donor-recipient matching that appear to hold promise for improved long-term outcomes. SUMMARY/CONCLUSIONS:The largely one-size-fits-all kidney allocation system is in desperate need of an overhaul. The continuous distribution paradigm is flexible enough to accommodate bold, new ideas for addressing major pain points in an equitable way. The OPTN should use policy variances to conduct time-limited, controlled experiments with various continuous distribution policies and build upon what works. Advances in transplant immunology, such as eplet matching, should increasingly be incorporated into kidney offer decision-making and, eventually, the allocation algorithm.

The adoption of large language models (LLMs) in healthcare demands a careful analysis of their potential to spread false medical knowledge. Because LLMs ingest massive volumes of data from the open Internet during training, they are potentially exposed to unverified medical knowledge that may include deliberately planted misinformation. Here, we perform a threat assessment that simulates a data-poisoning attack against The Pile, a popular dataset used for LLM development. We find that replacement of just 0.001% of training tokens with medical misinformation results in harmful models more likely to propagate medical errors. Furthermore, we discover that corrupted models match the performance of their corruption-free counterparts on open-source benchmarks routinely used to evaluate medical LLMs. Using biomedical knowledge graphs to screen medical LLM outputs, we propose a harm mitigation strategy that captures 91.9% of harmful content (F1 = 85.7%). Our algorithm provides a unique method to validate stochastically generated LLM outputs against hard-coded relationships in knowledge graphs. In view of current calls for improved data provenance and transparent LLM development, we hope to raise awareness of emergent risks from LLMs trained indiscriminately on web-scraped data, particularly in healthcare where misinformation can potentially compromise patient safety.