A Cohort Study for Derivation and Validation of Early Detection of Pancreatic Fistula After Pancreaticoduodenectomy

Background Pancreatic fistula (PF) remains the most important morbidity after pancreaticoduodenectomy (PD). Early drain removal was recently recommended. However, this is not applicable to all cases because the development of severe PF may not be obvious until a later postoperative day (POD). This study aimed to discover ways to detect clinically relevant PF early during the postoperative stage after PD. Methods We studied 120 patients who underwent PD. Grades B/C PF classified according to the International Study Group of Pancreatic Surgery guidelines were defined as clinically relevant PF. Logistic regression was used to identify detection factors for clinically relevant PF. Receiver operating characteristic curves were used to identify the optimal cutoff value for clinically relevant PF, and the k-fold cross-validation model to validate the cutoff value. Results Drain amylase on POD 1 and C-reactive protein (CPR) on POD 2 were independent factors for clinically relevant PF. Drain amylase >1300 IU/l on POD 1 and CRP >12.8 g/dl on POD 2 were the best cutoff values for clinically relevant PF detection and were confirmed by k-fold cross-validation. The sensitivity and specificity values were 79 and 81 %, respectively. Conclusions Values of drain amylase and CRP combined were useful to distinguish clinically relevant PF.