Abstract 276: A Genetic Algorithm Prognostic Tool Nested Within a 2500-Patient Registry of Post--Cardiac Arrest Care
Objectives: Predicting outcome after cardiac arrest (CA) is challenging. With data collected in the International Cardiac Arrest Registry (INTCAR), a novel prediction tool based on genetic algorithm (GA) was developed.
Methods: 2,564 adult, resuscitated, comatose CA patients were included from 35 sites in Europe and USA. Primary outcome was hospital discharge Cerebral Performance Category (CPC) dichotomized into good (GO=CPC1-2) and poor outcomes (PO=CPC 3-5). The GA-based models were constructed using Experlytics® software including all 180 candidate variables from the registry. A decision tree displaying the relation of the information carrying variables was evolved and pruned to a final model and optimized towards accuracy, sensitivity, specificity and Receiver-Operating Characteristic (ROC) value.
Results: Median age was 64 (IQR 54-73) years, 30% were female, the CA was witnessed in 84%, and initial rhythm was VF/VT in 55%, asystole in 23%, PEA in 17%. Time to ROSC was 19 (10-30) min. 37% had a GO at hospital discharge. 35% had coronary disease, 14% insulin dependent diabetes mellitus (IDDM), 8 % non-IDDM and 14% pulmonary disease. The final tree (figure 1) had an accuracy of 85%, a sensitivity of 91% and a specificity of 72%. The area under the ROC curve was 87% (CI 83-90).
Conclusions: A novel tool based on GA predicted outcome comparable to previous models. The GA could visually display the important information carrying variables and their relation. This model requires external validation, but represents a promising tool to better predict outcome and define the relationship between important risk factors
Figure 1. Relation of variables in a decision tree displaying good and poor outcome. ACD, automatic compression decompression, CPC, cerebral performance category, CPR, cardiopulmonary resuscitation, GCS, Glasgow Coma Scale, IDDM, Insulin dependent diabetes mellitus, MRI , magnetic resonance imaging, ROSC, return of spontaneous circulation.
- © 2012 by American Heart Association, Inc.