Multiparameter prediction model for atrial fibrillation after CABG

The aim of the study was to develop a multiparameter prediction model of atrial fibrillation (AF) after coronary artery bypass grafting (CABG) based on measured P wave parameters. We recorded the standard II lead ECG for at least 48 hours after surgery in 48 patients. In contrast to previous research and in order to enable the analysis of more data we decided to record the ECG continuously. The ECGs were processed offline and a vector of 82 P-wave parameters was calculated for every hour of the record. The segmentation of the ECGs was based on wavelet QRS and P-wave detectors. The calculated P-wave parameters were used for building classification and regression trees. We built several decision trees (models) for discriminating the AF prone patients after CABG. With the best tree model, we were able to achieve specificity (96.55%), sensitivity (54,54%), positive predictivity (85.71%), negative predictivity (84.84 %), accuracy (85,00%).