Dynamic risk assessment of the onset of Paroxysmal Atrial Fibrillation

We propose a computational methodology for evaluating the temporal evolution of the risk of onset of Paroxysmal Atrial Fibrillation (PAF) episodes. Firstly, we obtained 75 records of hour long ambulatory electrocardiograms (Holter monitoring) from healthy volunteers. From these we constructed a catalog of normal heart rate behavior patterns. For each record, patterns were defined as the standardized sequences of 5 consecutive values of RR intervals and the catalog was made up of all possible patterns contained in the 75 records. Secondly, 25 records of RR intervals corresponding to one hour long electrocardiograms ending with a PAF episode (group 1) and 25 additional records from healthy volunteers (group 2) were compiled. We then implemented a numerical procedure to compare the patterns belonging to these two groups with those of the catalog: using mobile windows of 6 consecutive minutes, we tested the null hypothesis that the patterns contained in windows of data taken from groups 1 and 2 were statistically indistinguishable from different subsets of patterns within the catalog. Our results demonstrated that the power at which this hypothesis is rejected indicates increased fluctuations in the patterns as we approach arrhythmia. This enabled us to propose an early warning system for the onset of PAF episodes with a specificity of 74% and sensitivity of 88%.