This multicenter study sought to evaluate the accuracy of the American College of Surgeons National Surgical Quality Improvement Program's (ACS-NSQIP) surgical risk calculator for predicting outcomes after pancreatoduodenectomy (PD) and to determine whether incorporating other factors improves its predictive capacity.Background:
The ACS-NSQIP surgical risk calculator has been proposed as a decision-support tool to predict complication risk after various operations. Although it considers 21 preoperative factors, it does not include procedure-specific variables, which have demonstrated a strong predictive capacity for the most common and morbid complication after PD – clinically relevant pancreatic fistula (CR-POPF). The validated Fistula Risk Score (FRS) intraoperatively predicts the occurrence of CR-POPF and serious complications after PD.Methods:
This study of 1480 PDs involved 47 surgeons at 17 high-volume institutions. Patient complication risk was calculated using both the universal calculator and a procedure-specific model that incorporated the FRS and surgeon/institutional factors. The performance of each model was compared using the c-statistic and Brier score.Results:
The FRS was significantly associated with 30-day mortality, 90-day mortality, serious complications, and reoperation (all P < 0.0001). The procedure-specific model outperformed the universal calculator for 30-day mortality (c-statistic: 0.79 vs 0.68; Brier score: 0.020 vs 0.021), 90-day mortality, serious complications, and reoperation. Neither surgeon experience nor institutional volume significantly predicted mortality; however, surgeons with a career PD volume >450 were less likely to have serious complications (P < 0.001) or perform reoperations (P < 0.001).Conclusions:
Procedure-specific complication risk influences outcomes after pancreatoduodenectomy; therefore, risk adjustment for performance assessment and comparative research should consider these preoperative and intraoperative factors along with conventional ACS-NSQIP preoperative variables.