Using a customized GPT to provide guideline-based recommendations for management of pancreatic cystic lesions

Background and study aims Rising prevalence of pancreatic cysts and inconsistent management guidelines necessitate innovative approaches. New features of large language models (LLMs), namely custom GPT creation, provided by ChatGPT can be utilized to integrate multiple guidelines and settle inconsistencies. Methods A custom GPT was developed to provide guideline-based management advice for pancreatic cysts. Sixty clinical scenarios were evaluated by both the custom GPT and gastroenterology experts. A consensus was reached between experts and review of guidelines and the accuracy of recommendations provided by the custom GPT was evaluated and compared with experts. Results The custom GPT aligned with expert recommendations in 87% of scenarios. Initial expert recommendations were correct in 97% and 87% of cases, respectively. No significant difference was observed between the accuracy of custom GPT and the experts. Agreement analysis using Cohen's and Fleiss' Kappa coefficients indicated consistency among experts and the custom GPT. Conclusions This proof-of-concept study shows the custom GPT's potential to provide accurate, guideline-based recommendations for pancreatic cyst management, comparable to expert opinions. The study highlights the role of advanced features of LLMs in enhancing clinical decision-making in fields with significant practice variability.

Establishing the optimal number of passes during EUS-FNB for diagnosis of pancreatic solid lesions: Prospective multicenter study

Background and study aims The optimal number of needle passes during endoscopic ultrasound-guided fine-needle biopsy (EUS-FNB) is not yet established. We aimed to perform a per-pass analysis of the diagnostic accuracy of EUS-FNB of solid pancreatic lesions using a 22G Franseen needle. Patients and methods Consecutive patients with solid pancreatic lesions referred to 11 Italian centers were prospectively enrolled. Three needle passes were performed; specimens were collected after each pass and processed individually as standard histology following macroscopic on-site evaluation (MOSE) by the endoscopist. The primary endpoint was diagnostic accuracy of each sequential pass. Final diagnosis was established based on surgical pathology or a clinical course of at least 6 months. Secondary endpoints were specimen adequacy, MOSE reliability, factors impacting diagnostic accuracy, and procedure-related adverse events. Results A total of 504 samples from 168 patients were evaluated. Diagnostic accuracy was 90.5% (85.0%-94.1%) after one pass and 97.6% (94.1%-99.3%) after two passes ( P =0.01). Similarly, diagnostic sensitivity and sample adequacy were significantly higher adding the second needle pass (90.2%, 84.6%-94.3% vs 97.5%, 93.8%-99.3%, P =0.009 and 91.1%, 85.7%-94.9% vs 98.2%, 95.8%-99.3%, P =0.009, one pass vs two passes, respectively). Accuracy, sensitivity, and adequacy remained the same after the third pass. The concordance between MOSE and histological evaluation was 89.9%. The number of passes was the only factor associated with accuracy. One case of mild acute pancreatitis (0.6%) was managed conservatively. Conclusions At least two passes should be performed for the diagnosis of solid pancreatic lesions. MOSE is a reliable tool to predict the histological adequacy of specimens.

Comparison of endoscopic ultrasound-guided fine-needle aspiration and fine-needle biopsy to generate pancreatic cancer organoids: Randomized trial

Background and study aims The prognosis for pancreatic cancer remains poor. Molecular diagnostics and customized therapies are becoming increasingly important in clinical routine. Patient-derived, predictive model systems such as organoids have the potential to substantially increase the depth of information from biopsy material by functional and molecular characterization. We compared the extent to which the use of fine-needle aspiration needles (FNA, 22G) or fine-needle biopsy needles (FNB, 22G) influences the generation of pancreatic cancer patient-derived organoids (PDOs) to establish endoscopic standards of organoid technology. Patients and methods Endoscopic ultrasound (EUS)-guided punctures by EUS-FNA and EUS-FNB of pancreatic masses highly suspicious for adenocarcinoma (detected by computed tomography and/or magnetic resonance imaging) were prospectively evaluated. Consecutive patients received EUS-FNA and EUS-FNB in a randomized order without the need to exchange the needle shaft (only the inner needle type (FNA/-B) was exchanged) between the passes. With each needle type, the specimens for histological analysis and for PDOs were obtained separately. Results Fifty patients were enrolled in the study. Histology revealed malignancy in 42 of 50 cases (84%). In total PDOs were generated from 17 patients (34%). Of these, nine were established by FNB only, two by FNA only, and six by both FNA and FNB. Histology revealed malignancy in 13 of 17 PDO cases (76%). In two histologically false-negative cases, PDOs could be established. Conclusions EUS-FNB was superior to EUS-FNA in terms of successful generation of PDOs, although it failed to show statistical significance.

Same-session endoscopic diagnosis and symptom palliation in pancreato-biliary malignancies: Clinical impact of rapid on-site evaluation (ROSE)

Background and study aims Besides increasing adequacy, rapid on-site evaluation (ROSE) during endoscopic ultrasound (EUS) or endoscopic retrograde cholangiopancreatography (ERCP) may impact choices and timing of subsequent therapeutic procedures, yet has been unexplored. Patients and methods This was a retrospective evaluation of a prospectively maintained database of a tertiary, academic centre with availability of ROSE and hybrid EUS-ERCP suites. All consecutive patients referred for pathological confirmation of suspected malignancy and jaundice or gastric outlet obstruction (GOO) between Jan-2020 and Sep-2022 were included. Results Of 541 patients with underlying malignancy, 323 (59.7%) required same-session pathological diagnosis (male: 54.8%; age 70 [interquartile range 63-78]; pancreatic cancer: 76.8%, biliary tract adenocarcinoma 16.1%). ROSE adequacy was 96.6%, higher for EUS versus ERCP. Among 302 patients with jaundice, ERCP-guided stenting was successful in 83.1%, but final drainage was completed in 97.4% thanks to 43 EUS-guided biliary drainage procedures. Twenty-one patients with GOO were treated with 15 EUS-gastroenterostomies and six duodenal stents. All 58 therapeutic EUS procedures occurred after adequate ROSE. With ERCP-guided placement of stents, the use of plastic stents was significantly higher among patients with inadequate ROSE (10/11; 90.9%) versus adequate sampling (14/240; 5.8%) P <0.0001; OR 161; 95%CI 19-1352). Median hospital stay for diagnosis and palliation was 3 days (range, 2-7) and median time to chemotherapy was 33 days (range, 24-47). Conclusions Nearly two-thirds of oncological candidates for endoscopic palliation require contemporary pathological diagnosis. ROSE adequacy allows, since the index procedure, state-of-the-art therapeutics standardly restricted to pathologically confirmed malignancies (e.g. uncovered SEMS or therapeutic EUS), potentially reducing hospitalization and time to oncological treatments.

Comparison of diagnostic outcomes, safety, and cost of Franseen-tip 19G versus 22G needles for endoscopic ultrasound-guided liver biopsies

Background and study aims Favorable outcomes were noted with refinement in newer endoscopic ultrasound-guided liver biopsy (EUS-LB) needle tips. Still, the overall usefulness and benefit are yet to be well explored. Patients and methods This was a retrospective analysis of patients with EUS-LB (Franseen-tip 19G versus 22G FNB needle) over 2 years. EUS-LB was obtained in a one-pass, two-actuation, modified wet suction technique. Diagnostic yield, fragmentation rate, aggregate specimen length (AL), number of complete portal tracts (CPT), length of longest intact core (LIC), adverse events (AEs) (early), and cost of the procedure (1USD = 82 INR) were compared. Results Fifty-four patients (33 [61.1%], female) successfully underwent EUS-LB with a median age of 46 years (interquartile range [IQR] 34-54); the majority 32 (59.2%) underwent 19G biopsies. There was a significantly increased median (IQR) AL in the 19G compared with 22G (20 mm [19-21] vs. 15 [14-15], P < 0.001), respectively. Similarly, significantly lengthier median LIC and CPT were seen, respectively. A nonsignificant diagnostic yield was noted (100% vs. 90.9%, P = 0.082), respectively. The fragmentation rate was higher in 22G FNB needles (36.4% [95% CI 16-56] vs. 12.5% [95% CI 1-24], respectively; P = 0.038). Seven patients (12.9%) had mild AEs with no difference between groups. The average procedure cost with 19G was INR 63000 (768$), and with 22G needle was INR 54500 (664$). Conclusions The Franseen-tip 19G outperforms 22G with a significantly lower fragmentation rate, longer AL, LIC, and a higher number of CPT with a marginal increase in the procedure cost, without any difference in diagnostic yield and safety.

Clinical usefulness of tissue acquisition of pancreatic cystic lesions using an endoscopic ultrasound-guided needle for histological analysis

Background and study aims There are rare data on the usefulness of endosonography-guided tissue acquisition (EUS-TA) in patients with pancreatic cystic lesions (PCLs). This study aimed to determine the accuracy of EUS-TA with ProCore 20G (PC20) for differentiating between mucinous neoplasia (MN) and non-MNs (n-MN) and identifying malignant PCLs, as well as its adverse events (AEs) in patients with PCLs without a classificatory diagnosis by imaging exams. Patients and methods In this observational, retrospective, single-center study, all patients with PCL who underwent EUS-TA due to diagnostic doubts in imaging studies were consecutively recruited from June 2017 to December 2021. The outcomes were to determine the diagnostic accuracy of EUS-TA with PC20 for differentiating between MN and n-MN, identifying malignant PCLs, and the AEs. Results Herein, 145 patients underwent EUS-TA, with 83 women (57.2%) and a mean age of 62.2 years. The mean size was 2.3 cm, with 81 patients (77.9%) having a PCL < 3.0 cm. The final diagnosis was made by EUS-TA (n = 81), surgery (n = 58), and follow-up (n = 6). The sensitivity, specificity, positive and negative predictive values, and accuracy for differentiating between MNs and n-MNs and identifying malignant PCLs were 92.6%, 98.4%, 98.7%, 91.3%, and 95.2% (kappa=0.9), and 92%, 99.2%, 95.8%, 98.3%, and 97.9% (kappa = 0.93), respectively. The AE rate was 2.7%, with no deaths in this cohort. Conclusions EUS-TA with PC20 has high accuracy and technical success with a low AE rate for PCL diagnosis.

Next-generation sequencing of pancreatic cyst wall specimens obtained using micro-forceps for improving diagnostic accuracy

Background and study aims Pancreatic cysts are common incidental findings, with an estimated prevalence of 13% to 15% in imaging done for other reasons. Diagnosis often relies on collection of cyst fluid, but tissue sampling using micro-forceps may allow for a more reliable diagnosis and higher yield of DNA for next-generation sequencing (NGS). The primary aim was to assess the performance of NGS in identifying mucinous cyst. The secondary aims were to assess DNA yield between the cyst fluid and cyst wall tissue, complication rate and performance of conventional investigations. Patients and methods Twenty-four patients referred for endoscopic ultrasound were recruited. Biopsies were taken using micro-forceps and the AmpliSeq Cancer Hotspot panel was used for NGS, a polymerase chain reaction assay targeting several hotspots within 50 genes, including GNAS , KRAS and VHL . Results The concentration of DNA extracted from 24 cyst wall samples was significantly higher than in the nine of 24 available matched cyst fluid samples. The sensitivity, specificity, and diagnostic accuracy of NGS for diagnosing mucinous cyst were 93%, 50% and 84%; for standard of care, they were -66.6%, 50% and 63.1%; and for standard of care with NGS, they were 100%, 50%, and 89.4% respectively. Cyst wall biopsy was able to diagnose 19 of 24 cysts (4 high risk, 7 intraductal papillary mucinous neoplasms, 4 cysts of mucinous origin, and 4 benign). Conclusions NGS data correlate well with histology and may aid in diagnosis and risk stratification of pancreatic cysts. Cyst wall biopsy performs well in diagnosing cysts but was inadequate in five of 24 patients.

Factors Affecting the Learning Curve in the Endoscopic Ultrasound-Guided Sampling of Solid Pancreatic Lesions: A Prospective Study

Background/Aims

Endosonography is associated with a long learning curve. We aimed to assess variables that may influence the diagnostic outcomes in endoscopic ultrasound-guided fine-needle aspiration/biopsy (EUS-FNA/B) of solid pancreatic tumors regarding the level of endoscopists' experience.

Methods

Consecutive patients undergoing EUS-guided puncture of solid pancreatic tumors (eight endosonographers, including six trainees) were prospectively enrolled. An experienced endosonographer was defined as having performed at least 250 EUS examinations, including 75 FNA/Bs. The final diagnosis was determined by cytopathology, histopathology, or clinical follow-up.

Results

In total, 283 EUS-FNA/Bs of solid pancreatic tumors (75.6% malignant) in 239 patients (median age 69 years, 57.6% males) were enrolled. Trainees performed 149/283 (52.7%) of the interventions. Accuracy and sensitivity for detecting malignancy were significantly higher in the expert group than in the trainee group (85.8% vs 73.2%, p=0.01 and 82.5% vs 68.4%, p=0.02). Solid lesions evaluated by an expert using FNB needles showed the best odds for a correct diagnosis (odds ratio, 3.07; 95% confidence interval, 1.15 to 8.23; p=0.02). More experienced endoscopists achieved better accuracy in sampling via the transduodenal approach (86.7% vs 68.5%, p<0.001), in the sampling of malignant lesions (82.5 vs 68.4, p=0.02), and in the sampling of lesions located in the pancreatic head (86.1 vs 69.1, p=0.02). In cases involving these factors, we observed a moderate improvement in the diagnostic accuracy after 40 attempts.

Conclusions

Transduodenal approach, pancreatic head lesions, and malignancy were recognized as the most important clinical factors affecting the learning curve in EUS-FNA/B of solid pancreatic lesions.