Mortality Risk for Acute Cholangitis (MAC): a risk prediction model for in-hospital mortality in patients with acute cholangitis.

Acute cholangitis is a life-threatening bacterial infection of the biliary tract. Main focus of this study was to create a useful risk prediction model that helps physicians to assign patients with acute cholangitis into different management groups.981 cholangitis episodes from 810 patients were analysed retrospectively at a German tertiary center.Out of eleven investigated statistical models fit to 22 predictors, the Random Forest model achieved the best (cross-)validated performance to predict mortality. The receiver operating characteristics (ROC) curve revealed a mean area under the curve (AUC) of 91.5 %. Dependent on the calculated mortality risk, we propose to stratify patients with acute cholangitis into a high and low risk group. The mean sensitivity, specificity, positive and negative predictive value of the corresponding optimal cutpoint were 82.9 %, 85.1 %, 19.0 % and 99.3 %, respectively. All of these results emerge from nested (cross-)validation and are supposed to reflect the model's performance expected for external data. An implementation of our risk prediction model including the specific treatment recommendations adopted from the Tokyo guidelines is available on http://www2.imse.med.tum.de:3838/ .Our risk prediction model for mortality appears promising to stratify patients with acute cholangitis into different management groups. Additional validation of its performance should be provided by further prospective trails.     «

Acute cholangitis is a life-threatening bacterial infection of the biliary tract. Main focus of this study was to create a useful risk prediction model that helps physicians to assign patients with acute cholangitis into different management groups.981 cholangitis episodes from 810 patients were analysed retrospectively at a German tertiary center.Out of eleven investigated statistical models fit to 22 predictors, the Random Forest model achieved the best (cross-)validated performance to predict...     »