Letter by Hamilton and Filardo Regarding Article, “Reduced Ventricular Volumes and Improved Systolic Function With Cardiac Resynchronization Therapy: A Randomized Trial Comparing Simultaneous Biventricular Pacing, Sequential Biventricular Pacing, and Left Ventricular Pacing”
To the Editor:
Rao et al1 report on the effects of left ventricular, simultaneous biventricular, and sequential biventricular pacing on a range of cardiac outcomes. It appears that the authors have attempted to “determine the difference in the sample means” between trial arms for each end point with separate ANOVA models at each time point (baseline, 3 months, and 6 months). No accounting is undertaken for the fact that the echocardiograms recorded at baseline, 3 months, and 6 months are recorded on the same patients (excluding those lost to follow up)—at least no such accounting is noted in the Methods section. Thus, the authors seem to be assuming that measurements recorded on the same patients are completely independent of each other. However, measurements recorded on the same patients can be highly correlated, and inferences based on repeated-measures models would have been more appropriate.
The analysis conducted by Rao et al1 also should account for multiplicity as demonstrated by the 72 P values in Tables 2 through 4 alone (excluding those P values in the baseline demographics table and those additional P values scattered among the text). The probability of a type I error on at least 1 of these 72 tests is <1: (1−0.05)72=97.5% (ie, the overall type I error is controlled at the 97.5% level). Westfall and Young2 have pointed out the deleterious impact that runaway type I error control can have on the results of a clinical trial. However, the authors disregard this critical aspect in their data analysis and conclusions.
Finally, the authors fail to discuss the effect that loss to follow-up may have had on their results. More than 10% of the 306 patients who were randomized did not have 6-month echocardiograms (no information is given on missingness at 3 months). Unless one is willing to assume that patients who are lost to follow-up are missing completely at random (and not, for example, as a result of worsening condition or death), the statistical tests between follow-up and baseline within trial arms could be critically biased.3 An approach such as the one described by McMahon and Harrell4 would have at least accounted for dropouts resulting from patient deaths.
In sum, although this study provides an interesting addition to the literature, the statistical tests provided therein should be interpreted with extreme caution.
Rao RK, Kumar UN, Schafer J, Viloria E, De Lurgio D, Foster E. Reduced ventricular volumes and improved systolic function with cardiac resynchronization therapy: a randomized trial comparing simultaneous biventricular pacing, sequential biventricular pacing, and left ventricular pacing. Circulation. 2007; 115: 2136–2144.
Little R, Rubin D. Statistical Analysis with Missing Data. 2nd ed. Hoboken, NJ: John Wiley & Sons, Inc; 2002.