Predicting fatal outcome in the early phase of severe acute pancreatitis by using novel prognostic models

Survival in acute pancreatitis and particularly in severe acute and necrotizing pancreatitis is a combination of therapy-associated and patient-related factors. There are only few relevant methods for predicting fatal outcome in acute pancreatitis. Scores such as Ranson, Imrie, Blamey, and APACHE II are practical in assessing the severity of the disease, but are not sufficiently validated for predicting fatal outcome among patients with severe acute pancreatitis. The aim of this study was to construct a novel prediction model for predicting fatal outcome in the early phase of severe acute pancreatitis (SAP) and to compare this model with previously reported predictive systems.

Hospital records of 253 patients with SAP were retrospectively analyzed. 234 patients with adequate data were included to the test set to construct five logistic regression and three artificial neural network (ANN) models. Two models were tested in an independent prospective validation set of 60 consecutive patients with SAP and compared with previously reported predictive systems.

The prediction model considered optimal was a logistic model with four variables: age, highest serum creatinine value within 60-72 h from primary admission, need for mechanical ventilation, and chronic health status. In the validation set, the predictive accuracy, determined by the area under the receiver operating characteristic curve value, was 0.862 for the chosen model, 0.847 for the ANN model using eight variables, 0.817 for APACHE II, 0.781 for multiple organ dysfunction score, 0.655 for Ranson, and 0.536 for Imrie scores.

Ranson and Imrie scores are inaccurate indicators of the mortality in SAP. A novel predictive model based on four variables can reach at least the same predictive performance as the APACHE II system with 14 variables.