With a level of significance of 0.05 (type 1 error), a typical power of 0.85 (type 2 error), an inherent variability of 15 percent and a difference of 10 percent between treatments to be regarded as clinically significant—using paried t tests—the sample size required would be 23 subjects. Increasing the power to 0.95 in this case would require a sample size of 32 subjects (SigmaStat, Jan-del Scientific; San Raphael, Calif). Sample size therefore is an essential consideration in evaluating the reliability and power of statistical analysis.
Most studies evaluating CPT treatments in CF patients fail to indicate the statistical power of the study or the size of the change required to be statistically significant and most have sample sizes of less than 20. Hence, the literature becomes even more difficult to interpret in view of sample size and statistical power, particularly those studies where no significant difference was shown between modalities.
Read the rest of this entry »