Un ltered signal-averaged P wave analysis { further validation of the method

Background: The study was designed to explore the impact of di erent methodologies on P wave duration estimates. Furthermore, the required recording length to ensure reproducibility from un ltered signal-averaged P wave analysis was investigated. Finally, the reproducibility of P wave morphology classi cation was investigated and an algorithm for automated classi cation was designed. Methods: 12-lead ECG recordings (1 kHz sampling frequency, 0.625 V resolution) from 131 healthy subjects were used. Orthogonal leads were derived using the inverse Dower transform. When exploring the impact of methodology on P wave duration estimate magni cation (100 times), baseline ltering (0.5 Hz highpass and 50 Hz bandstop lters), signal averaging (10 seconds) and bandpass ltering (40{250 Hz) were used interchangeably. To de ne the required recording length (6 minutes to 10 seconds) and the reproducibility of the P wave morphology classi cation procedure un ltered signal-averaged P wave analysis was performed. The manual classi cation was carried out by two experts on two separate occasions each. The performance of the automated classi cation algorithm was evaluated using the joint decision of the two. Results: The P wave duration estimate increased in each step as a result of magni cation, baseline ltering and averaging (100 18 vs. 131 12 msec; P < 0.0001). The duration estimate of the bandpass ltered P wave was dependent of noise level cut-o value; 119 15 msec (0.2 V), 138 13 msec (0.1 V) and 143 18 msec (0.05 V). (P = 0.01 for all comparisons). The mean errors of the P wave morphology parameters were comparable in all segments analysed regardless of