RESUMO
BACKGROUND: Early stratification of postoperative pancreatic fistula according to severity and/or need for invasive intervention may improve outcomes after pancreaticoduodenectomy. This study aimed to identify the early postoperative variables that may predict postoperative pancreatic fistula severity. METHODS: All patients diagnosed with biochemical leak and clinically relevant-postoperative pancreatic fistula based on drain fluid amylase >300 U/L on the fifth postoperative day after pancreaticoduodenectomy were identified from a consecutive cohort from Birmingham, UK. Demographics, intraoperative parameters, and postoperative laboratory results on postoperative days 1 through 7 were retrospectively extracted. Independent predictors of clinically relevant-postoperative pancreatic fistula were identified using multivariable binary logistic regression and converted into a risk score, which was applied to an external cohort from Verona, Italy. RESULTS: The Birmingham cohort had 187 patients diagnosed with postoperative pancreatic fistula (biochemical leak: 99, clinically relevant: 88). In clinically relevant-postoperative pancreatic fistula patients, the leak became clinically relevant at a median of 9 days (interquartile range: 6-13) after pancreaticoduodenectomy. Male sex (P = .002), drain fluid amylase-postoperative day 3 (P < .001), c-reactive protein postoperative day 3 (P < .001), and albumin-postoperative day 3 (P = .028) were found to be significant predictors of clinically relevant-postoperative pancreatic fistula on multivariable analysis. The multivariable model was converted into a risk score with an area under the receiver operating characteristic curve of 0.78 (standard error: 0.038). This score significantly predicted the need for invasive intervention (postoperative pancreatic fistula grades B3 and C) in the Verona cohort (n = 121; area under the receiver operating characteristic curve: 0.68; standard error = 0.06; P = .006) but did not predict clinically relevant-postoperative pancreatic fistula when grades B1 and B2 were included (area under the receiver operating characteristic curve 0.52; standard error = 0.07; P = .802). CONCLUSION: We developed a novel risk score based on early postoperative laboratory values that can accurately predict higher grades of clinically relevant-postoperative pancreatic fistula requiring invasive intervention. Early identification of severe postoperative pancreatic fistula may allow earlier intervention.