Preoperative radiographic measurements can predict postoperative pancreatic fistula formation following pancreatoduodenectomy

Abstract

Background: Postoperative pancreatic fistulas (POPF) are a major contributing factor to pancreatoduodenectomy associated morbidity. Established risk calculators exist to help predict POPF formation, but most rely on subjective or intraoperative assessments. We hypothesized that various objective preoperatively determined computed tomography (CT) measurements could predict POPF as well as validated models and allow for more informed operative consent in high risk patients. Methods: Patients with POPF following elective pancreatoduodenectomy between January 2013 and April 2018 were identified in a prospective database. A propensity matched control cohort was selected for comparison. Clinicopathologic parameters were utilized to generate predictive models for POPF development using receiver operating characteristics (ROC) curves. Pancreatic neck radiodensities (Hounsfield units) were measured by pancreatic protocol CT (venous phase, coronal plane) at 0, 5, and 10mm from the anterior surface of the portal vein. A pancreatic density index (PDI) was created to adjust for differences in contrast timing by dividing the mean of these measurements by the portal vein radiodensity. Total areas of subcutaneous fat and skeletal muscle were calculated at the L3 vertebral level on axial CT. Pancreatic duct (PD) size was determined by CT. Results: Thirty-two patients were identified with POPF and propensity matched to an equal number of patients that did not develop POPF. Age, sex, race, and operative indication were not different between cohorts. Within our dataset a validated predictive model based on BMI, PD size (truncated at 5mm maximum), and gland texture (soft vs. non-soft) performed poorly by ROC curve analysis (AUC 0.590). Additional ROC curves were created using combinations of gland texture, body mass index, skeletal muscle index, sarcopenia, PDI, PD size, and subcutaneous fat area indexed for height (SFI). The model most predictive of POPF included SFI, PDI, and PD size (AUC 0.684). Conclusion: A combination of preoperative objective CT measurements can adequately predict POPF in this small dataset. Validation in a larger dataset would allow for better preoperative stratification of high risk patients and improve informed consent.