• Rainer Röhrig, Bernd Hartmann, Axel Junger, Joachim Klasen, Dominik Brammen, Florian Brenck, Andreas Jost, and Gunter Hempelmann.
• Department of Anesthesiology, Intensive Care Medicine, and Pain Therapy, Justuts-Liebig University Giessen, Rudolf-Buchheim-Str. 7, D-35392, Giessen, Germany.
• J Clin Monit Comput. 2007 Jun 1; 21 (3): 159-66.

ObjectiveIn anesthesia and intensive care logistic regression analysis are often used to generate predictive models for risk assessment. Strictly seen only independent variables should be represented in such prognostic models. Using anesthesia-information-management-systems a lot of (depending) information is stored in a database during the preoperative ward round. The objective of this study was to evaluate a statistical algorithm to process the different dependent variables without losing the information of each variable on patient's conditions.MethodBased on data about prognostic models in anesthesia an iterative statistical algorithm was initiated to summarize dependent variables to subscores. Seven subscores out of several preoperative variables were calculated corresponding to the proper incidence and the correlation to the occurrence of intraoperative cardiovascular events was evaluated. After that first step logistic regression was used to build a predictive model out of the seven subscores, 10 patient-related, and two surgery-related variables. Performance of the prognostic model was assessed using analysis of discrimination and calibration.ResultFour out of seven subscores together with age, type and urgency of surgery are represented in the prognostic model to predict the occurrence of intraoperative cardiovascular events. The prognostic model demonstrated good discriminative power with an area under the ROC curve (AUC) of 0.734.ConclusionDue to reduced calibration, the clinical use of the prediction model is limited.

20 keywords...

hide…