Influence of filtering techniques on the time-domain analysis of signal-averaged P wave electrocardiogram

The advent of signal-averaged ECG (SAECG) systems for P wave analysis has made it important to determine if the use of different filtering techniques in these systems is diagnostically equivalent.

Three different high-pass filtering techniques and two cutoff frequency values were used: 29- and 40-Hz Butterworth bidirectional filter (BB29, BB40), 29- and 40-Hz Butterworth unidirectional filter (UB29, UB40), and 29- and 40-Hz least mean square filter (LMS29, LMS40). Normal healthy volunteers (n = 36) and patients with documented paroxysmal atrial fibrillation (n = 23) were analyzed. A custom-built SAECG system and standard bipolar orthogonal leads were used. Noise was reduced to < 0.3 microV. P wave total duration, root mean square voltage of the terminal 20, 30, and 40 msec of the filtered vector magnitude, and the area under the curve between the onset and offset of averaged unfiltered and filtered P wave vector magnitude were analyzed. Only the duration of the P wave showed statistically significant differences between groups, being longer in the PAF group for all filters and cutoff frequencies studied. A bias increment of approximately 20 msec was detected in unidirectional and least mean square filters as compared to the bidirectional filter. Sensitivity, specificity, and predictive accuracy were > 70% for all filters; the BB40 filter yielded the best performance.

The normality limits derived from one filter cannot be applied directly to recordings obtained from the other filters. Critical limits must be established individually for different software settings.