Independent external validation and comparison of existing pancreatic fistula risk scores after laparoscopic pancreaticoduodenectomy with Bing's pancreaticojejunostomy.

BACKGROUND: The fistula risk score (FRS) is the widely acknowledged prediction model for clinically relevant postoperative pancreatic fistula (CR-POPF). In addition, the alternative FRS (a-FRS) and updated alternative FRS (ua-FRS) have been developed. This study performed external validation and comparison of these 3 models in patients who underwent laparoscopic pancreaticoduodenectomy (LPD) with Bing's pancreaticojejunostomy.

METHODS: The FRS total points and predictive probabilities of a-FRS and ua-FRS were retrospectively calculated using patient data from a completed randomized controlled trial. Postoperative pancreatic fistula (POPF) and CR-POPF were defined according to the 2016 International Study Group of Pancreatic Surgery criteria. The correlations of the 4 risk items of the FRS model with CR-POPF and POPF were analyzed and represented using the Cramer V coefficient. The performance of the 3 models was measured using the area under the curve (AUC) and calibration plot and compared using the DeLong test.

RESULTS: This study enrolled 200 patients. Pancreatic texture and pathology had discrimination for CR-POPF (Cramer V coefficient: 0.180 vs 0.167, respectively). Pancreatic duct diameter, pancreatic texture, and pathology had discrimination for POPF (Cramer V coefficient: 0.357 vs 0.322 vs 0.257, respectively). Only the calibration of a-FRS predicting CR-POPF was good. The differences among the AUC values of the FRS, a-FRS, and ua-FRS were not statistically significant (CR-POPF: 0.687 vs 0.701 vs 0.710, respectively; POPF: 0.733 vs 0.741 vs 0.750, respectively). After recalibrating, the ua-FRS got sufficient calibration, and the AUC was 0.713 for predicting CR-POPF.

CONCLUSION: For LPD cases with Bing's pancreaticojejunostomy, the 3 models predicted POPF with better discrimination than predicting CR-POPF. The recalibrated ua-FRS had sufficient discrimination and calibration for predicting CR-POPF.