A. J. Benjamin1, A. Schneider1, M. M. Buschmann2, B. A. Derstine4, S. C. Wang3,4, W. Dale2, K. Roggin1 1University Of Chicago,Surgery,Chicago, IL, USA 2University Of Chicago,Geriatrics & Palliative Medicine,Chicago, IL, USA 3University Of Michigan,Surgery,Ann Arbor, MI, USA 4University Of Michigan,Morphomic Analysis Group,Ann Arbor, MI, USA
Introduction:
Following pancreaticoduodenectomy (PD), pancreatic fistula (PF) remains a significant cause of morbidity, and current models used to predict PF rely on measures which are only available at the time of operation. Body imaging analysis, such as analytic morphomics (AM), and pre-operative geriatric assessment (GA) have been shown to forecast significant adverse outcomes following PD. We hypothesized preoperative AM and GA can accurately predict PF.
Methods:
An IRB-approved review identified patients (n=63) undergoing PD by experienced pancreatic surgeons who had a non-contrast computed tomography scan (CT), pre-operative geriatric assessments, and prospectively tracked postoperative 90-day outcomes collected between 10/2007 and 3/2016. PF were graded according to the International Study Group for PF (ISGPF) criteria. Pre-operative GA included the Short Physical Performance Battery, self-reported exhaustion on the Center for Epidemiologic Studies Depression Scale (CES-D exhaustion; one of the five criteria of Fried’s frailty), and the Vulnerable Elders Survey (VES-13). CT scans were processed to measure morphomic variables which included measures of psoas muscle area and Hounsfield units (HU), subcutaneous fat measures, visceral fat measures, and total body dimensions. Correlations with the development of a PF were obtained using univariate analysis and multivariate elastic net regression models.
Results:
The median patient age was 67 (37-85) years old and the median BMI was 27.0 (18.9-50.5). In total, 15/63 patients (23%) had a documented PF: 8 patients had a ISGPF grade A PF (53%), 6 had a grade B PF (40%), and one had a grade C PF (7%). On univariate analysis, PF was associated with CES-D exhaustion (p=0.005), VES-13 (p=0.038), subcutaneous fat HU (p=0.009), visceral fat area (p=0.035), visceral fat HU (p=0.001), average psoas HU (p=0.040) and psoas low density muscle area (p=0.049). A predictive model based on demographics, analytic morphomics, and GA had a high AUC for predicting PF (AUC=0.915) when compared to a clinical “base model” including age, BMI, ASA class, and Charlson comorbidity index (AUC=0.685).
Conclusion:
Preoperatively measured AM, in combination with GA, can accurately predict the likelihood of PF following PD. Validation of this model on a larger cohort would provide surgeons with a practical tool to more accurately risk-stratify patients prior to PD.