BACKGROUND. Manual extraction of pancreatic cystic lesion (PCL) features from radiology reports is labor-intensive, limiting large-scale studies needed to advance PCL research. OBJECTIVE. The purpose of this study was to evaluate GPT-4o (closed source [ OpenAI]), Llama (Llama-3.1-8B-Instruct, open source), and DeepSeek (DeepSeek-R1-Distill-Llama-8B, open source) large language models (LLMs) for PCL feature extraction, without and with chain-of-thought (CoT) reasoning. METHODS. We curated a dataset of 6469 abdominal MRI or CT reports (2005-2024) that described PCLs from 5615 patients. Llama and DeepSeek were fine-tuned using quantized low-rank adaptation on GPT-4o-generated CoT labels for extracting PCL and main pancreatic duct features. Features were mapped to risk categories per institutional policy. Evaluation was performed on 285 held-out human-annotated reports from 281 patients. Model outputs for 100 cases were independently reviewed by three radiologists. Feature extraction was evaluated using exact-match accuracy, risk categorization with a macro-averaged F1 score, and radiologist-model agreement with Fleiss kappa values. Error analyses were performed to assess how and why models made mistakes. RESULTS. CoT fine-tuned LLMs showed a feature extraction accuracy of 97% (95% CI, 97-98%) for Llama, 98% (95% CI, 97-98%) for DeepSeek, and 97% (95% CI, 97-98%) for GPT-4o. Risk categorization F1 scores were 0.93 (95% CI, 0.89-0.97) for Llama, 0.94 (95% CI, 0.90-0.98) for DeepSeek, and 0.97 (95% CI, 0.93-0.99) for GPT-4o. Radiologist interreader agreement was high (κ = 0.888) and showed no significant difference with the addition of Llama (κ = 0.882; p > .99), DeepSeek (κ = 0.893, p > .99), or GPT-4o (κ = 0.897, p > .99). Across all models, object identification and clinical reasoning were the most frequent error types, accounting for 29.3-37.3% and 18.1-21.1% of total errors, respectively. CONCLUSION. LLMs show feasibility for automatically extracting PCL features from radiology reports. Fine-tuned open-source LLMs achieved performance comparable to that of GPT-4o. CoT reasoning improved accuracy and enabled interpretable error analysis. Model-assigned risk categories showed high agreement with abdominal radiologists. CLINICAL IMPACT. LLMs have the potential to enable creation of large structured registries from existing radiology reports to support population-level research on PCLs.

Background: The existence of sociodemographic disparities in pancreatic cancer has been well-studied but how these disparities have changed over time is unclear. The purpose of this study was to longitudinally assess patient management in the context of sociodemographic factors to identify persisting disparities in pancreatic cancer care. Methods: Using the National Cancer Database, patients diagnosed with pancreatic ductal adenocarcinoma from 2010 to 2017 were identified. The primary outcomes were surgical resection and/or receipt of chemotherapy. Outcome measures included changes in associations between sociodemographic factors (i.e., sex, age, race, comorbidity index, SES, and insurance type) and treatment-related factors (i.e., clinical stage at diagnosis, surgical resection, and receipt of chemotherapy). For each year, associations were assessed via univariate and multivariate analyses. Results: Of 75,801 studied patients, the majority were female (51%), White (83%), and had government insurance (65%). Older age (range of OR 2010-2017 [range-OR]:0.19-0.29), Black race (range-OR: 0.61-0.78), lower SES (range-OR: 0.52-0.94), and uninsured status (range-OR: 0.46-0.71) were associated with lower odds of surgical resection (all p < 0.005), with minimal fluctuations over the study period. Older age (range-OR: 0.11-0.84), lower SES (range-OR: 0.41-0.63), and uninsured status (range-OR: 0.38-0.61) were associated with largely stable lower odds of receiving chemotherapy (all p < 0.005). Conclusions: Throughout the study period, age, SES, and insurance type were associated with stable lower odds for both surgery and chemotherapy. Black patients exhibited stable lower odds of resection underscoring the continued importance of mitigating racial disparities in surgery. Investigation of mechanisms driving sociodemographic disparities are needed to promote equitable care.

BACKGROUND:Prognostic factors in resected pancreatic ductal adenocarcinoma (PDAC) have been determined under the assumption that hazard ratios (HRs) remain static. However, PDAC is a dynamic disease with evolving conditional survival. The aim of this study was to determine if the impact of prognostic factors in PDAC is time-varying. METHODS:This was a multicenter, retrospective cohort study of the prospectively maintained Dutch Pancreatic Cancer Recurrence Database and New York University and Johns Hopkins Hospital Institutional Databases. Patients with complete macroscopic resection of histopathologically proven PDAC between 2014 and 2019 and available follow-up data were included. The time-varying impact of prognostic factors identified by univariable Cox regression was modeled using Aalen's Additive Regression Models (Aalen's models) and visualized as plots of cumulative hazard. RESULTS:In total, 3104 patients were included, of whom 938 (30.2%) received neoadjuvant therapy (NAT), whereas the rest underwent upfront surgery (US). A total of 201 (6.5%) patients achieved observed long-term survival (>5 years). Aalen's models showed that lymphovascular invasion, perineural invasion, and nodal disease were prognostic up to 2 years postoperatively. At varying points thereafter, these variables lost their impact in the NAT but not US patients. Similarly, during the fourth year of follow-up, American Society of Anesthesiology scores became impactful in the NAT but not in the US patients. CONCLUSION:The impact of prognostic factors in resected PDAC across NAT and US patients is time-varying. Our results suggest that aggressive disease drives early mortality but, after NAT, tumor-biological factors lose prognostic importance to frailty and comorbidities over time.

BACKGROUND:IPMN consensus guidelines make implicit judgments on what cancer risk level should prompt surgery. We used decision modeling to estimate this cancer risk threshold (CRT) for BD-IPMN patients. METHODS:We created a decision model to compare quality-adjusted life years (QALYs) following surgery or surveillance for BD-IPMNs. We simulated treatment decisions for hypothetical patients, varying age, comorbidities and lesion location (pancreatic head/tail). The base case was a 60-year-old patient with mild comorbidities and pancreatic head IPMN. Probabilities, life expectancies, and utilities were incorporated from literature/public datasets. CRT was defined as the level of cancer risk at which the expected value of QALYs for surgery first exceeded that of surveillance. RESULTS:In the base case, surgery was preferred over surveillance, yielding 21.90 vs. 21.88 QALYs. The optimal CRT for a BD-IPMN patient depended on age, comorbidities, and location. CRT in the base case was 20 % and 3 % for an IPMN in the head and tail of the pancreas, respectively. Other drivers of preferred treatment were age and likelihood of postoperative mortality. CONCLUSION:For BD-IPMNs, the optimal CRT varies depending on patient age and risk of surgical complications. Personalized risk threshold values could guide treatment decisions and inform future treatment consensus guidelines.

INTRODUCTION:En-bloc celiac axis resection (CAR) was first proposed by Lyon H. Appleby in 1952 for gastric cancer and later modified for pancreatic resections with gastric preservation by Nimura et al. in 1976. CAR remains uncommon, performed in fewer than 0.2 cases annually. Advancements in preoperative imaging and anatomy understanding, ischemic complication management, and centralization of care have improved outcomes. This report presents a robotic distal pancreatectomy (DP) with CAR and superior mesenteric artery (SMA) divestment. CASE REPORT:A 65-year-old woman presented with back pain. Imaging revealed biopsy-proven pancreatic adenocarcinoma in the pancreatic body, encasing the celiac, splenic, and common hepatic arteries with SMA abutment. Following four cycles of neoadjuvant FOLFIRINOX, follow-up imaging demonstrated stable disease without metastasis. The need for hepatic artery reconstruction was assessed intraoperatively, with alternative strategies detailed in the accompanying video. OPERATIVE TECHNIQUE:The patient underwent a distal pancreatectomy and splenectomy with class Ia CAR. Surgery was conducted in a caudal approach, lasted 420 minutes with minimal blood loss (100 ml). Laparoscopic ultrasound (lapUS) and indocyanine green (ICG) perfusion were used to assess resectability, vascular perfusion, and targeted blood vessels. The postoperative course was uneventful, except for a Grade B chyle leak managed conservatively. No liver or gastric ischemia occurred. Adjuvant chemotherapy was initiated two months postoperatively. CONCLUSION:Enhanced visualization, improved dexterity, and adjuncts including lapUS and ICG are potential benefits that are available to surgeons with the robotic platform when performing arterial divestment and CAR via a caudal approach.

BACKGROUND:Prognostic markers for overall survival in resected pancreatic ductal adenocarcinoma are well established but remain unclear following neoadjuvant therapy. This systematic review and meta-analysis aimed to determine factors associated with overall survival following neoadjuvant therapy in resected pancreatic ductal adenocarcinoma. METHODS:The PubMed, Embase, Scopus, Web of Science, and Cochrane CENTRAL databases were systematically searched from January 2010 until May 2024. Studies that reported univariable and multivariable hazard ratios were included if patients underwent neoadjuvant therapy and resection for localized pancreatic ductal adenocarcinoma. Study quality assessment was performed using the Newcastle-Ottawa scale. Meta-analysis was performed using generic inverse-variance random-effects models. RESULTS:Among 2208 unique articles identified by the search, 92 were included in the meta-analysis. Of these, 85 were of "good" and 7 of "poor" quality. The neoadjuvant therapy regimen was described in 84 studies of which 62 included patients treated with FOLFIRINOX. Margin status, nodal disease, American Joint Committee on Cancer (AJCC) T-stage, and normalization of cancer antigen 19-9 (CA19-9) after neoadjuvant therapy were prognostic for overall survival, whereas age, sex, perineural invasion, baseline tumor size, and baseline CA19-9 were not. The test for subgroup differences between ypN substages was not statistically significant in the multivariable model. Neoadjuvant FOLFIRINOX was associated with better survival than other regimens. CONCLUSIONS:This meta-analysis identified margin status, nodal disease, AJCC T-stage, and normalization of CA19-9 after neoadjuvant therapy as prognostic factors for overall survival in patients with resected localized pancreatic ductal adenocarcinoma following neoadjuvant therapy.

Neoadjuvant therapy (NAT) for pancreatic ductal adenocarcinoma (PDAC) is increasingly being used. The aim of this study was to evaluate the association between type, duration, and sequencing of adjuvant therapy (AT) after NAT and overall survival (OS) in patients with resected PDAC. Patients receiving NAT and resection for PDAC (2010–2019) at two high-volume pancreatic surgery centers were included and stratified into groups on the basis of NAT regimen: gemcitabine-based NAT, 5-fluorouracil (5FU)-based NAT, or switched NAT regimen. The maximally selected rank statistic was used to determine the optimal NAT duration. Univariate and multivariable Cox proportional hazards models were used to assess the association between NAT regimen and OS, and between AT and OS. Of 651 patients, 200 (30.7

OBJECTIVE:To investigate the clinicopathologic features and long-term outcomes of cystic and solid pancreatic neuroendocrine tumors (PanNETs). BACKGROUND:PanNETs uncommonly present as cystic lesions. Whether cystic PanNETs represent a distinct clinical entity compared with solid PanNETs is controversial. METHODS:Clinicopathologic data of patients with resected PanNETs were collected from 4 high-volume centers between 2000 and 2019. Clinicopathologic characteristics and outcomes of patients with cystic and solid PanNETs were compared based on a 3 cm tumor size cutoff using the χ 2 test and Mann-Whitney U test. Survival estimates were calculated with the Kaplan-Meier method and log-rank test and multivariable analysis using a Cox proportional hazard model. RESULTS:Of the 1727 patients undergoing pancreatectomy for PanNET, the median age was 58.1 years (IQR: 18.4), and 53.3% were males. Of these, 177 (10.3%) were cystic, and 1550 (89.7%) were solid. Cystic PanNETs were more prevalent in patients with hereditary syndromes, less frequently functional, and more often located in the body/tail of the pancreas. After the exclusion of patients with functional tumors, World Health Organization G3 tumors, hereditary syndromes, neoadjuvant treatment, and metastatic stage, 145 cystic PanNETs were compared with 1059 solid PanNETs, and the median follow-up period of the cohort was 64 months. Cystic PanNETs demonstrated significantly fewer high-risk histopathologic features, lymph node (LN) metastases (5.5% vs 24.0%, P < 0.001), and distant recurrence (4.1% vs 14.4%; P < 0.001). Among tumors, ≤3 cm, cystic PanNETs had a low rate of LN metastases (3.9% vs 17.8%; P < 0.001), recurrence (3.1% vs 8.4%; P = 0.041), and low propensity to recur distantly. Cystic PanNETs had favorable long-term survival regardless of tumor size. CONCLUSIONS:Cystic PanNETs have a more benign course than their solid counterparts, and conservative management can be considered for endoscopic ultrasound-guided fine needle aspiration-proven cystic PanNETs ≤3 cm. Parenchyma and LN-sparing resections are warranted in patients with cystic PanNETs >3 cm. Patients with poor baseline performance status may forego cystic PanNET resection and not affect their overall survival.

OBJECTIVE:The aim of the study was to investigate prognostic factors in context of neoadjuvant therapy (NAT) and develop tools that can allow for accurate and personalized patient prognostication. SUMMARY OF BACKGROUND DATA:NAT might impact the prognostic ability of well-established clinicopathological factors in resected pancreatic ductal adenocarcinoma (PDAC). METHODS:Patients after resection for PDAC were identified from the Dutch Pancreatic Cancer Group Recurrence Database and institutional databases at NYU Langone Health and the Johns Hopkins Hospital (2014-2019). Patients were stratified into NAT and chemo-naïve groups. Overall survival (OS), calculated from the time of resection, was estimated using Kaplan-Meier and compared using log-rank tests. Prognostic factors associated with OS were assessed in both groups using univariable and multivariable Cox-regression analyses and presented using hazard ratios (HR) with corresponding 95% confidence intervals (95%CI). Predictive models were developed and an interactive tool was created to predict survival independently in both groups. RESULTS:Of 2,760 patients with resected PDAC, 778 patients (28%) received NAT. Independent predictors for worse OS in chemo-naïve patients included age ≥65 years, markedly elevated CA19-9 (≥500 U/mL) at diagnosis, higher AJCC-T stage (T3/4 vs T1/2), worsening AJCC N-stage (N2 vs. N1 vs. N0), poor tumor differentiation, perineural invasion, and microscopically positive resection margin (R1 vs. R0). Contrastingly, predictors for worse OS in NAT patients included non-normalization of CA19-9 after NAT (<37 U/mL), presence of nodal disease (N1/2 vs. N0 given no statistical difference between N1 and N2 disease), and grade of treatment response (moderate/poor vs. complete/near complete). CONCLUSION:Prognostic factors for OS in patients with resected PDAC differ between chemo-naïve and NAT patients. Personalized prediction tools for OS in resected PDAC based on these specific factors are available online (www.pancpals.com/tools).

Objective: To investigate whether tangential versus segmental portomesenteric venous resection (PVR) impacts surgical and oncological outcome in patients undergoing pancreatoduodenectomy for pancreatic cancer with portomesenteric vein (PMV) involvement. Summary Background Data: Current comparative studies on tangential versus segmental PVR as part of pancreatoduodenectomy for pancreatic cancer include all degrees of PMV involvement, including cases where tangential PVR may not be a feasible approach, limiting the clinical applicability. Methods: International retrospective study in 10 centers from 5 countries, including all consecutive patients after pancreatoduodenectomy with PVR for pancreatic cancer with ≤180° PMV involvement on cross-sectional imaging at diagnosis (2014-2020). Cox and logistic regression analyses were performed to investigate the association of tangential versus segmental PVR with overall survival (OS) from surgery, recurrence-free survival (RFS), locoregional recurrence, and in-hospital/30-day major morbidity, adjusting for potential confounders. Results: Overall, 357 patients who underwent pancreatoduodenectomy with PVR were included (42% tangential PVR, 58% segmental PVR). The adjusted risk for in-hospital/30-day major morbidity was 23% (95%CI, 17-32) after tangential and 23% (95%CI, 17-30) after segmental PVR (P=0.98). After adjusting for confounders, PVR type was not associated with OS (HR=0.94 [95%CI, 0.69-1.30]), RFS (HR=0.94 [95% CI, 0.69 to 1.28), and locoregional recurrence (OR=0.76 [95%CI, 0.40-1.46]). Conclusions: In patients undergoing pancreatoduodenectomy for pancreatic cancer with ≤180° PMV involvement, the type of PVR (i.e., tangential vs. segmental) was not associated with differences in surgical and oncological outcome. This suggest that if both procedures are technically feasible, surgeons can choose the type of PVR based on their preference.

Antithrombotic therapy (AT) aims to strike a balance between preventing thromboembolic and hemorrhagic complications. However, evidence for AT management after pancreatectomy with vascular reconstruction is lacking. We aimed to provide an overview of the current use of AT for pancreatic surgery with vascular reconstructions. A web-based survey was distributed to 123 surgeons from high-volume pancreas centers (>50 pancreatic resections/year). AT management after different types of vascular reconstruction were investigated. An “aggressive” protocol was defined as the use of any AT protocol other than prophylactic heparin, aspirin, or their combination. The survey was completed by 80 surgeons (59

OBJECTIVE:To establish a novel method for evaluating carbohydrate antigen 19-9 (CA19-9) during neoadjuvant therapy (NAT) and assess its role in predicting overall (OS) and disease-free (DFS) survival in borderline resectable pancreatic adenocarcinoma (BR-PC). SUMMARY BACKGROUND DATA:Static CA19-9 values or percentage changes often fail to capture therapeutic responses in patients with BR-PC undergoing NAT. Improved evaluation methods are essential for guiding the treatment. METHODS:This was a retrospective multicenter study of patients who underwent BR-PC surgery. Two parameters were developed: slope coefficient (SC, change in CA19-9 divided by therapy duration) and mean δf (mδf, calculated as the sum of CA19-9 values over therapy intervals divided by the number of 15-day periods). The main objective was to correlate mδf with OS thresholds derived using a maximally selected log-rank statistic and validated in independent cohorts. RESULTS:Overall, 991 patients (median age 65 [59-71] years; 49% male) were included. The thresholds for mδf were defined as 206 U.mL -1 . month -1 (negative SC) and 115 U.mL -1 . month -1 (positive SC). Patients with mδf below these thresholds had significantly better prognoses, with hazard ratios (HR) for OS (95% CI) of 0.6 (0.4-0.8; P <0.01) and 0.4 (0.2-0.9; P =0.04) for negative and positive SC, respectively. Both thresholds were validated for mOS, with 29 vs 22 months ( P =0.015) and 32 vs 16 months ( P =0.0034) for negative and positive SC, respectively. Similarly, the mDFS was 13 vs 10 months ( P =0.011) and 12 vs 7 months ( P =0.0018), respectively. CONCLUSION:This CA19-9 evaluation approach accurately predicts survival outcomes, offering a valuable tool for optimizing treatment strategies. An mδf calculator is available at https://www.pancpals.com/tools .

BACKGROUND/OBJECTIVES:To evaluate the importance of image processing in a previously validated model for detecting pancreatic neuroendocrine tumors (PanNETs) and to introduce Image2Radiomics, a new framework that ensures reproducibility of the image processing pipeline and facilitates the deployment of radiomics models. METHODS:A previously validated model for identifying PanNETs from CT images served as the reference. Radiomics features were re-extracted using Image2Radiomics and compared to those from the original model using performance metrics. The impact of nine alterations to the image processing pipeline was evaluated. Prediction discrepancies were quantified using the mean ± SD of absolute differences in PanNET probability and the percentage of classification disagreement. RESULTS:The reference model was successfully replicated with Image2Radiomics, achieving a Cohen's kappa coefficient of 1. Alterations to the image processing pipeline led to reductions in model performance, with AUC dropping from 0.87 to 0.71 when image windowing was removed. Prediction disagreements were observed in up to 45% of patients. Even minor changes, such as switching the library used for spatial resampling, resulted in up to 21% disagreement. CONCLUSIONS:Reproducing image processing pipelines remains challenging and limits the clinical deployment of radiomics models. While this study is limited to one model and imaging modality, the findings underscore a common risk in radiomics reproducibility. The Image2Radiomics framework addresses this issue by allowing researchers to define and share complete processing pipelines in a standardized way, improving reproducibility and facilitating model deployment in clinical and multicenter settings.