© 2004 American Thoracic Society
Continuous Positive Airway Pressure in Obstructive Sleep Apnea and Heart FailureTo the Editor:We are writing to point out that the statistical analysis in the article by Mansfield and colleagues (1) was unduly influenced by the presence of an outlier in the control group, thereby impacting the significance of the study in terms of differences in left ventricular ejection fractions (LVEFs). Two subjects, one in the intervention group and one in the control group, developed atrial fibrillation during the study. On the basis of Figure 2 in the study by Mansfield and colleagues (1), if one were to delete these two subjects, the results would lose statistical significance. Figure 2 shows that for the subject in the control group who experienced development of atrial fibrillation, the LVEF dropped from 37 to 22. For the subject in the intervention group in whom atrial fibrillation developed, there was no change in the LVEF. If one were to exclude both subjects, the mean change in LVEF in the control group would increase from 1.5 to 2.3 (eliminating a change of 15 for 1 subject in a sample of 21 would change the mean for the remaining 20 subjects by 0.75), whereas the mean change in LVEF for the continuous positive airway pressure group would be unaltered. Excluding both subjects would result in a nonsignificant p value, on the order of p greater than 0.1, for the differences in LVEFs between the two groups. The standard deviation for the difference in LVEFs for the control group was larger than the standard deviation for the difference in LVEFs for the intervention group (1.4 vs. 1.0). The outlier in the control group probably accounts for much of this difference between the standard deviations. Because the t test assumes that the standard deviations of the two groups are equal, using the t test when there are differences in standard deviations is a violation of the key assumption of the t test. Dropping the two subjects who experienced development of atrial fibrillation would close the gap in the LVEF difference score standard deviations. The essential problem is that the control subject who developed atrial fibrillation is an outlier that pulled down the mean in the control group. Even if the authors left this individual in the study, they should have corrected for this outlier by transforming the data, either by using nonparametric statistics or correcting the t test for unequal variances. Either of these options would have provided a more valid test of the association.
Case Western Reserve University School of Medicine Cleveland, Ohio FOOTNOTES Conflict of Interest Statement: R.P.B. and S.J.Z. do not have a financial relationship with a commercial entity that has an interest in the subject of this letter. REFERENCES
To the Editor: We thank Drs. Blankfield and Zyzanski for their interest in our article (1). In particular, they have highlighted to us a minor error in Figure 2: the asterisk (indicating the baseline to 3-month left ventricular ejection fraction in the patient who experienced development of atrial fibrillation in the group receiving continuous positive airway pressure [CPAP]) should have been beside the data point line 27–25%, and not the 28–28% data point line. We can only assume that this occurred during the printing process and missed our attention on the galley proofs. One patient in each group experienced development of atrial fibrillation during the study period. Left ventricular ejection fraction decreased in the control patient (36–22%) and the CPAP patient (27–25%) despite 7.5 hours per night adherence to 10 cm H2O nasal CPAP. We had no reason to exclude these patients from the final analysis. If these two patients are removed, the statistical significance of the change in left ventricular ejection fraction between groups remains unchanged (+2.3% vs. +5.3%, p = 0.04 using t test for unequal variance and using nonparametric tests). We agree that the variance of the measurements of left ventricular ejection fraction was wide, particularly in the control group, and for this reason t tests for unequal variance were used in the original analysis.
Alfred Hospital Melbourne, Australia FOOTNOTES Conflict of Interest Statement: D.R.M. does not have a financial relationship with a commercial entity that has an interest in the subject of this letter. M.T.N. has served on the Australian Medical Advisory Board for ResMed, a CPAP manufacturer, and received an unconditional grant to conduct a clinical trial; he also receives an honorarium per annum to attend the annual advisory meeting. REFERENCES
This article has been cited by other articles:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
