Preoperative Fistula Risk Prediction Using Computed Tomography Image Before Pancreatoduodenectomy

Abstract

Background: Although many formulas for predicting postoperative pancreatic fistula (POPF) have been introduced, POPF is generally predicted during pancreatic surgery due to pancreatic texture. This study was designed to verify the correlation between Hounsfield units (HU) and pancreatic texture and to suggest a fistula risk score (FRS) that can be used before surgery.

Methods: Data from 545 patients who underwent pancreatoduodenectomy for malignant disease between January 2008 and December 2019 were retrospectively reviewed. The HU level of the pancreas was measured, and odds ratio (OR) of the HU for POPF was analyzed. Additionally, the assessed HU was compared with the pancreatic texture (soft vs. hard) and calculated cutoff level. Finally, the preoperatively chosen pancreatic texture according to HU level was applied to the FRS formula (preoperative-FRS: p-FRS), and the results were compared with a previously reported FRS formula (updated alternative-FRS: ua-FRS).

Results: The Hounsfield unit levels were correlated with clinically relevant POPF (CR-POPF) (odds ratio [OR]: 1.04 (1.01-1.07), p = 0.015). In the receiver operating characteristic curve, the HU showed significant prediction potential for pancreatic texture (area under the curve [AUC]: 0.744, p < 0.001). The p-FRS also showed acceptable results in predicting CR-POPF (AUC = 0.702, p < 0.001). There was no statistically significant difference in the DeLong's test compared with the ua-FRS (p = 0.314). In the Hosmer-Lemeshow test, observed probabilities were correlated with predicted probabilities (p = 0.596).

Conclusions: The HU level on preoperative computed tomography (CT) is a predictive factor for POPF and could represent for pancreatic texture.