Diastolic murmur of complete heart block.

tribution. More important, however, the possible existence of correlation among data was not addressed. It is generally known that correlation among data would seriously affect the results of analysis of variance. Therefore, it may be more appropriate to use a time series of analysis technique rather than the analysis of variance method. From the computational formula of the 95% confidence limit for the difference between the control and test periods (which appears to have a typographical error), it is apparent that the authors assumed that variabilities during control and test periods are the same. If the therapeutic intervention, however, reduces VPD frequency during the test period, the variability of these readings also may be reduced dramatically. Therefore, the logarithmic or square root transformation would not be sufficient to create variance homogeneity in comparing the two groups of data. The well-known Behrens-Fisher problem would invalidate this computational formula.1 The authors unfortunately did not address the question of variability of complex ventricular arrhythmias (multifocal VPDs, couplets or ventricular tachycardia) since the control of these arrhythmias may be more significant in the assessment of the efficacy of antiarrhythmic therapy than VPD count alone. It would perhaps be most helpful if the authors would provide the original data for statistical reanalysis. We were pleased to see the interest that Drs. Stein and Lee showed in our manuscript. Statistical analysis of our data required careful consideration because of its complexity, and we are happy to provide additional methodological data as requested by Stein and Lee. The range of variability in our Holter data analysis system was 0.5-18% (average 7.2%). More important, the results of our analysis of variance technique did include the system variability which could not be fully discussed in our lengthy Methods and Results sections. Table 2 listed the sources of variation in ventricular ectopic frequency , and the "between hours" source of variation of 16% included the system variability. The data loss from Holter monitoring over 72 continuous hours in 15 patients was, surprisingly, not high. Only one patient (no. 15) had a total loss of data during 1 day, as noted in table 1. Two of the additional 14 patients had a loss of 10 and 6 hours during one of the 3 days of monitoring, while all other patients had only an occasional hour of lost data during the 72 hours of monitoring. In our initial statistical analysis, we used …