Dec 13, 2012

ECG predicts atrial fibrillation onset

Atrial fibrillation (AF), the most common cardiac arrhythmia, is categorized by different forms. One sub-type is paroxysmal AF (PAF), which refers to episodes of arrhythmia that generally terminate spontaneously after no more than a few days. Although the underlying causes of PAF are still unknown, it's clear that predicting the onset of PAF would be hugely beneficial, not least because it would enable the application of treatments to prevent the loss of sinus rhythm.

Many research groups are tackling the issue of predicting the onset of PAF. Now, however, researchers in Spain have developed a method that assesses the risk of PAF at least one hour before its onset. To date, the approach has not only successfully discriminated healthy individuals and PAF patients, but also distinguished patients far from and close to PAF onset (Physiol. Meas.).

"The ability to assess the risk of arrhythmia at least one hour before its onset is clinically relevant," Arturo Martinez from the University of Castilla-La Mancha told medicalphysicsweb. "Our method assesses the P-wave feature time course from single-lead long-term ECG recordings. Using a single ECG lead reduces the computational burden, paving the way for a real-time system in future."

Analysing sinus rhythm

If the heart is beating normally, the sinus rhythm observed on an ECG will contain certain generic features, such as a P-wave that reflects the atrial depolarization and a large characteristic R peak flanked by two minima representing the depolarization of the heart's right and left ventricles. If an irregular heart beat is suspected, an ECG will be used and typical findings include the absence of a P-wave.

"We hypothesized that different stages of AF could be identified when analysing long-term recordings extracted from patients prone to AF," commented Martinez. "Our method differs to others in that we also use just one single lead to detect small differences in features from the P-wave time course."

P for paroxysmal

Martinez and his collaborators, Raul Alcaraz and Jose Rieta, studied 24-hour Holter ECG recordings from 24 patients in whom PAF had been detected for the first time. For each patient, the longest sinus rhythm interval in the recording was selected, and the two hours preceding the onset of PAF were analysed. These readings were compared with those from 28 healthy individuals. In all cases, only the trace from the V1 ECG lead was considered.

A major challenge for the researchers was to extract the P-wave from the baseline noise. To overcome this, they used an automatic delineator algorithm based on a phasor transform that determines the precise time point relating to the onset, peak and offset of the P-wave. The authors described this algorithm in a previous research paper (Physiol. Meas.).

"All of the recordings in our study were visually supervised by expert cardiologists who corrected the P-wave fiducial points when needed," said Martinez. "Even in the presence of noise, which generated an incredible amount of P-wave distortion, our delineator provided location errors lower than 8 ms."

Chosen parameters

In order to assess which time course features might be useful to predict the onset of PAF, the researchers analysed a number of variables. First, they examined factors representing the duration of the P-wave (P_dur), such as the distance between the P-wave onset and peak (P_ini) and the distance between the P-wave peak and its offset (P_ter). They then studied factors relating P- to R-waves, such as the distance between the two waves' peaks (PR_k) and, finally, beat-to-beat P-wave factors, such as the distance between two consecutive P-wave onset points (PP_on).

"The most remarkable trends were provided by the features measuring P-wave duration," report the authors in their paper. "P_dur identified appropriately 84.21% of all the analysed patients, obtaining a discriminant accuracy of 90.79% and 83.33% between healthy subjects and PAF patients far from PAF and close to PAF, respectively. The metrics related to the PR interval showed the most limited ability to identify patient groups."