Background: Postoperative pancreatic fistula (POPF) is a major factor for morbidity and mortality after pancreatic resection. Risk stratification for POPF is important for adjustment of treatment, selection of target groups in trials and quality assessment in pancreatic surgery. In this study, we built a risk-prediction model for POPF based on a large number of predictor variables from the German pancreatic surgery registry StuDoQ|Pancreas.
Methods: StuDoQ|Pancreas was searched for patients, who underwent pancreatoduodenectomy from 2014 to 2016. A multivariable logistic regression model with elastic net regularization was built including 66 preoperative und intraoperative parameters. Cross-validation was used to select the optimal model. The model was assessed via area under the ROC curve (AUC) and calibration slope and intercept.
Results: A total of N = 2488 patients were included. In the optimal model the predictors selected were texture of the pancreatic parenchyma (soft versus hard), body mass index, histological diagnosis pancreatic ductal adenocarcinoma and operation time. The AUC was 0.70 (95% CI 0.69-0.70), the calibration slope 1.67 and intercept 1.12. In the validation set the AUC was 0.65 (95% CI 0.64-0.66), calibration slope and intercept were 1.22 and 0.42, respectively.
Conclusion: The model we present is a valid measurement instrument for POPF risk based on four predictor variables. It can be applied in clinical practice as well as for risk-adjustment in research studies and quality assurance in surgery.
Copyright © 2018. Published by Elsevier B.V.