Agreement between automatic and manual measurement of atrial and ventricular signal-averaged electrocardiograms in healthy subjects

Background: Although prolonged duration of the signal-averaged (SA) P wave has been proposed as a noninvasive marker of atrial arrhythmias, clinical value of atrial SAECG is limited largely due to the difficulty with detection of the onset and offset of the high gain P wave. The aim of this study was to assess the reliability of automatic measurement of the atrial SAECG. Methods: Fifty-one healthy volunteers (30 men; 32 ± 8 years) underwent a session of 3 atrial and 3 ventricular SAECG recordings. Automatically detected onset and offset of SA QRS complex (QRS(tot)) and SA P wave (P(tot)) were subsequently corrected by two independent observers. For ventricular SAECG, three conventional time-domain parameters were calculated. For atrial SAECG, the following five parameters were measured: P(tot), root mean square voltages of the entire P(tot) (RMS-P) and of the terminal 40, 30, and 20 ms of P(tot). Relative errors of the different pairs of measurements were used to assess the interobserver and observer-computer variability. The Bland-Altman method was applied to express the agreement between measurements. Results: Although the mean interobserver relative errors were low for QRS(tot) and P(tot) (1.1% vs 1.5%), the observer-computer error was significantly higher for P(tot) than for QRS(tot) (1.7% vs 7.1%; P < 0.0001). For the voltage parameters, the lowest interobserver and observer-computer relative errors were found for RMS-P (6.6% vs 7.3%, P = ns). For RMS voltages of the terminal 40-20 ms of P(tot), relative errors exceeded 10%, but the interobserver error was significantly lower than the observer-computer error (P < 0.0001). Conclusion: Automatic detection of the SA P-wave onset and offset is unreliable and the atrial SAECG requires manual correction. Given a good interobserver agreement, such a correction is unlikely to introduce any significant observer-dependent bias.