Comparison of preoperative CT-based imaging parameters to predict postoperative pancreatic fistula

Aim To review all reported methods of preoperative computed tomography (CT) in one patient cohort and to identify which were the strongest to predict postoperative pancreatic fistula (POPF) after pancreatoduodenectomy. Materials and methods Consecutive patients undergoing pancreatoduodenectomy were included if they had unenhanced CT images for review. Eighteen variables and two scores were tested. Receiver operator characteristics (ROC) were explored. Results POPF affected 26 of 107 patients (24.3%). Nine variables were significantly related to POPF with pancreatic duct width having the largest area under the ROC curve (AUROC; 0.808, p<0.001). An obese body habitus was associated with POPF with six of nine related variables using data from CT images associated with POPF; of these intra-abdominal wall thickness yielded the largest AUROC (0.713, p=0.001). This corresponded to the finding that body mass index (BMI) was related to POPF (AUROC 0.705, p=0.002). The largest AUROC of all was associated with one of the predictive scores (0.828, p<0.001). Substituting BMI for intra-abdominal wall thickness in this score yielded a non-significant increase to predict POPF (AUROC 0.840, p=0.676). None of the assessments of organ density (in Hounsfield Units) were associated with POPF. Conclusion Data from preoperative CT imaging provides valuable information regarding a patient's risk of POPF. Obesity as assessed by CT images strongly relates to POPF, but the largest single risk factor for POPF is a narrow pancreatic duct.