| Resumo: | Hospital patients recovering from major cardiac surgery are at high risk of postoperative atrial fibrillation (POAF), an arrhythmia which can be life-threatening. With the development of a tool to predict POAF early enough, the development of the arrhythmia could be potentially prevented using prophylactic treatments, thus reducing risks and hospital costs. To date, no reliable method suitable for autonomous clinical integration has been proposed yet. This thesis presents a study on the prediction of POAF using the electrocardiogram. A novel P-wave quality assessment tool to automatically identify high-quality P-waves was designed, and its clinical utility was assessed. Prediction of paroxysmal atrial fibrillation (AF) was performed by implementing and improving a selection of previously proposed methods. This allowed to perform a systematic comparison of those methods, and to test if their combination improved prediction of AF. Finally, prediction of POAF was tested in a clinically relevant scenario. This included studying the 48 hours preceding POAF, and automatically excluding noise-corrupted P-waves using the quality assessment tool. The P-wave quality assessment tool identified high-quality P-waves with high sensitivity (0.93) and good specificity (0.84). In addition, this tool improved the ability to predict AF, since it improved the precision of P-wave measurements. The best predictors of AF and POAF were measurements of the variability in P-wave time- and morphological features. Paroxysmal AF could be predicted with high specificity (0.93) and good sensitivity (0.82) when several predictors were combined. Furthermore, POAF could be predicted 48 hours before its onset with good sensitivity (0.74) and specificity (0.70). This leaves time for prophylactic treatments to be administered and possibly prevent POAF. Despite being promising, further work is required for these techniques to be useful in the clinical setting.
|
|---|