Predicting pancreatic fistula after central pancreatectomy using current fistula risk scores for pancreaticoduodenectomy and distal pancreatectomy.

ABSTRACT

BACKGROUND: The incidence of clinically relevant postoperative pancreatic fistula (CR-POPF) after central pancreatectomy (CP) is high, yet an effective predictive method is currently lacking. This study aimed to predict CR-POPF after CP by utilizing existing fistula risk scores (FRSs) for pancreaticoduodenectomy (PD) and distal pancreatectomy (DP). METHODS: A retrospective analysis was conducted on patients undergoing CP at our institution between January 2010 and July 2022. The primary outcome was CR-POPF (grade B/C) according to the 2016 International Study Group of Pancreatic Surgery definition. To establish predictive models for CR-POPF after CP, we combined the FRSs for PD and DP using a calculation formula that considers the probability of the union of two events. As a result, we obtained twelve central FRS (C-FRS) models. The performance of each C-FRS was assessed using the area under the curves (AUC) and calibration plots. RESULTS: A total of 115 patients undergoing CP were included. Among them, 38 (33%) were male, with a median age of 53 years. CR-POPF occurred in 35 (30.4%) patients, specifically 33 (28.7%) with grade B and 2 (1.7%) with grade C. Multivariate analysis showed that body mass index (BMI) [odds ratio (OR) 1.260, 95% confidence interval (CI) 1.039-1.528, P = 0.019), pancreatic thickness at the cephalic transection site (OR 1.228, 95% CI 1.074-1.405, P = 0.003), cephalic main pancreatic duct (MPD) size (OR 41.872, 95%CI 7.614-230.265, P < 0.001), and distal MPD size (OR 0.142, 95% CI 0.036-0.561, P = 0.005) were independent predictive factors for CR-POPF. Discrimination was generally acceptable for all C-FRS models, with an AUC ranging from 0.748 (DISPAIR-a-FRS 95% CI, 0.659-0.824) to 0.847 (Intraop-D-a-FRS 95% CI, 0.768-0.907). The models were calibrated with adequate Brier scores ranging from 0.157 to 0.183. The performance in all subgroups was similar as that of the entire cohort. Three preoperative risk groups (low, intermediate, and high) were identified based on the clinical applicability of the Preop-D-Roberts-FRS, with corresponding incidences of CR-POPF as 0% (0/24), 30% (21/70), and 66.7% (14/21), respectively. CONCLUSION: The derived C-FRS models show potential for accurately predicting the development of CR-POPF after CP. However, further validation studies are required to determine the most effective model. In the meantime, the Preop-D-Roberts-FRS is recommended for clinical practice due to its ease of use and preoperative predictability.