Back to Annual Meeting Posters

Introduction:
Pancreatoduodenectomy (PD) has become a standard operation with low mortality in high-volume centers, however perioperative morbidity remains substantial, mainly due to postoperative pancreatic fistula (POPF). Development of preoperative protective measures is hampered by a lack of strictly preoperative risk stratification. Predictive power of single parameters can be enhanced by optimally weighed combination of risk factors in an artificial neuronal network (ANN).
Methods:
A panel of clinical and radiological parameters were assessed retrospectively from patients with pancreatoduodenectomy in our institution and risk factors analysis for the endpoint POPF (clinically relevant Grade B/C of ISGPS definition) were identified. Preoperatively available parameters were used for prediction of a high risk pancreas in an ANN. Internal validation of the thereby identified risk group was performed by testing for POPF and other relevant complications.
Results
A total of 471 patients with PD operated from 2001 to 2012 were included. Out of twelve clinical and radiological risk factors for POPF B/C, the most powerful was a soft pancreas. When an ANN was trained to predict a soft high-risk pancreas, correct prediction was achieved in 83% in the test group. Patients predicted to have a high-risk pancreas had a significantly higher rate of POPF and severe complications compared to the low-risk group (POPF B/C (38% vs 8%, p=0.000), intraabdominal abscess (23% vs 10%, p=0.000), severe complications (26% vs 13%, p=0.003), severe postpancreatectomy hemorrhage (18% vs 6%, p=0.012)), as well as a five-fold elevated mortality (5% vs 1%, p=0.034).
Conclusion
Clinical and radiological parameters combined in an ANN model can correctly predict a high-risk pancreas and severe complications already before the operation.