Statistical analysis
Enrollment and sample size were estimated from the results of the PAcing THerapies for Congestive Heart Failure (PATH-CHF) study 17 and 22, based on an expected improvement of 12.5% in the weakest primary end point (Vo2 at the anaerobic threshold). For a 5% significance level and 80% power, a total sample size of 64 full crossover data sets was required. Due to the expected mortality and dropout rates, we planned to enroll 100 patients.
An independent statistician performed all data analyses. End points could be assessed only in patients with no missing data after completion of both crossover periods. Baseline characteristics were compared by the unpaired t test for continuous variables, the Wilcoxon test for walk distance and quality-of-life score, and the Fisher exact test for discrete data. The clinical effects of pacing were tested by calculating for each clinical measure the difference of the second minus the first crossover period value for each patient and then comparing all the patient differences between the two pacing sequences (active first or inactive first) by the unpaired t test for Vo2 measures and by the Wilcoxon rank-sum test for walk distance, quality-of-life score, and NYHA functional classification 26 and 27. Parametric estimation of treatment effects and confidence intervals was used for all variables and calculated from the differences between the active and inactive pacing mean values for the two pacing sequences. The treatment-period interaction (residual or carryover effect) was tested by a t test applied to the individual sums of the first and second period data. When a pacing effect was significant for all patients, the effect was tested by the same statistical methods for the long and short QRS patient groups. A statistical result of p < 0.05 was considered necessary for significance.