Chest Physiotherapy and Cystic Fibrosis: Within-Subject Variability and Statistical Power
Subjects with CF vary not only from each other in terms of disease manifestations, but also they vary from day to day in their ability to reproduce pulmonary function tests. In CF subjects, the inherent variability in performing PFTs has been estimated to be as high as 15%; however, the individual’s variability has been shown to be very consistent. Cooper et al (1990) suggest that significant changes in pulmonary function data would be more accurately analyzed by incorporating predetermined variability for each individual rather than using group mean data.
Closely correlated with this variability is the need to consider the statistical power of a study, which is the probability of detecting a difference between therapies or groups, if a true difference exists. For example, a power of 70% indicates that there is a 70% chance of detecting a difference between therapies if a difference truly exists and a 30% chance of missing a difference. In addition, errors can be made with statistical analyses and investigators need to determine the level of acceptable errors or level of significance. natural inhalers for asthma
Two types of errors may occur: type 1 error results when an ineffective therapy is erroneously called effective due to chance alone, and type 2 error occurs when an effective therapy is erroneously called ineffective. Pattishall (1990) investigated the statistical power of 61 negative clinical trials of therapeutic regimens including physiotherapy in CF patients, presenting either a detrimental effect from a form of therapy or no difference between therapies. He found that few studies had the statistical power to detect small or medium differences between therapies. It seems that while investigators usually define the probability for type 1 error, rarely was the probability of type 2 error defined.
The power of the study is related to the levels of significance (probability of type 1 and 2 errors), the size of the difference to be detected, the variability of parameters to be tested, and the statistical tests to be used. These factors need to be considered to calculate the sample size required for statistical reliability. For example, in the case of comparing two forms of CPT treatment, in a group of CF patients, using values from PFTs, the following factors would need to be considered.