• Deeksha Kapoor, Yajushi Desiraju, Vikram A Chaudhari, Afroj Ismail Bagwan, Amit Chopde, ArunKumar Namachivayam, Manish S Bhandare, and Shailesh V Shrikhande.
• Gastrointestinal and Hepato-Pancreato-Biliary Service, Department of Surgical Oncology, Homi Bhabha National Institute, Tata Memorial Hospital, Parel, Mumbai, Maharashtra, 400012, India.
• Ann. Surg. 2024 Aug 13.

ObjectivesTo externally validate the International Study Group of Pancreatic Surgery (ISGPS) classification and test its performance for predicting clinically relevant pancreatic fistula (CRPF) for periampullary tumours (P-amps).BackgroundThe ISGPS is a simple two-factor, four-tier classification of pancreas-related risk for CRPF after a pancreatoduodenectomy (PD). External validation and performance of the classification specific to P-amps are lacking. P-amps have different disease biology, lesser need for neoadjuvant therapy, softer pancreas, and a higher rate of CRPF, underscoring the importance of site-specific prediction.MethodsValidation was performed in a cohort of 1422 patients, with CRPF as the primary outcome. Model performance was tested by plotting the receiver operating curve and calibration plots. After analysing the factors predicting CRPF, the model was optimised for P-amps.ResultsCRPF rate was 22.2% (315/1422), for P-amps being 25.8%. The ISGPS model performed moderately (AUC=0.632, 95% CI 0.598-0.666, P<0.001), with worse performance for P-amps (AUC=0.605, 95% CI 0.566-0.645, P<0.001). On multivariate analysis, soft pancreas (OR 1.689, 95% CI 1.136-2.512, P=0.010), body mass index ≥23 kg/m2 (OR 2.112, 95% CI 1.464-3.046, P<0.001) and pancreatic duct ≤3 mm (OR 2.113 95% CI 1.457-3.064, P<0.001), emerged as independent predictors and the model was optimised. The adjusted ISGPS for P-amps showed improved discrimination (AUC=0.672, P<0.001, 95% CI 0.637-0.707), with adequate performance on internal validation.ConclusionThe adjusted ISPGS performs better than the original ISGPS in predicting CRPF for P-amps. Large-scale multicenter data is needed to generate and validate site-specific predictive models.Copyright © 2024 Wolters Kluwer Health, Inc. All rights reserved.

8 keywords...

hide…