The study was designed to investigate the effect of different measuring methodologies on the estimation of P wave duration. The recording length required to ensure reproducibility in unfiltered, signal-averaged P wave analysis was also investigated. An algorithm for automated classification was designed and its reproducibility of manual P wave morphology classification investigated.
Twelve-lead ECG recordings (1 kHz sampling frequency, 0.625 μV resolution) from 131 healthy subjects were used. Orthogonal leads were derived using the inverse Dower transform. Magnification (100 times), baseline filtering (0.5 Hz high-pass and 50 Hz bandstop filters), signal averaging (10 seconds) and bandpass filtering (40–250 Hz) were used to investigate the effect of methodology on the estimated P wave duration. Unfiltered, signal averaged P wave analysis was performed to determine the required recording length (6 minutes to 10 s) and the reproducibility of the P wave morphology classification procedure. Manual classification was carried out by two experts on two separate occasions each. The performance of the automated classification algorithm was evaluated using the joint decision of the two experts (i.e., the consensus of the two experts).
The estimate of the P wave duration increased in each step as a result of magnification, baseline filtering and averaging (100 ± 18 vs. 131 ± 12 ms; P < 0.0001). The estimate of the duration of the bandpass-filtered P wave was dependent on the noise cut-off value: 119 ± 15 ms (0.2 μV), 138 ± 13 ms (0.1 μV) and 143 ± 18 ms (0.05 μV). (P = 0.01 for all comparisons).
The mean errors associated with the P wave morphology parameters were comparable in all segments analysed regardless of recording length (95% limits of agreement within 0 ± 20% (mean ± SD)). The results of the 6-min analyses were comparable to those obtained at the other recording lengths (6 min to 10 s).
The intra-rater classification reproducibility was 96%, while the interrater reproducibility was 94%. The automated classification algorithm agreed with the manual classification in 90% of the cases.
The methodology used has profound effects on the estimation of P wave duration, and the method used must therefore be validated before any inferences can be made about P wave duration. This has implications in the interpretation of multiple studies where P wave duration is assessed, and conclusions with respect to normal values are drawn.
P wave morphology and duration assessed using unfiltered, signal-averaged P wave analysis have high reproducibility, which is unaffected by the length of the recording. In the present study, the performance of the proposed automated classification algorithm, providing total reproducibility, showed excellent agreement with manually defined P wave morphologies.