Division of Endocrinology (K.L., E.S.H.), Institute of Medicine, Haukeland University Hospital, N-5021 Bergen, Norway; and Department of Clinical Sciences (O.K.), Uppsala University, S-751 85 Uppsala, Sweden
Address correspondence to: Kristian Løvås, M.D., Division of Endocrinology, Institute of Medicine, Haukeland University Hospital, N-5021 Bergen, Norway. E-mail: kristian.lovas{at}med.uib.no.
To the editor:
We recently published the results of a randomized parallel-group clinical trial of dehydroepiandrosterone (DHEA) replacement in adrenal failure, in which we found no benefit for subjective health status and sexuality (1). We agree with Arlt and Allolio (2) that low statistical power infers risk of excluding true benefit (type 2 statistical error). However, statistical power is not only a matter of study size. It also depends on treatment variability and definitions of clinically relevant treatment responses. The wide variation in treatment responses was not known or anticipated at the time of our trial. It seems reasonable that clinically relevant treatment responses should at least exceed the range of measurement variability. The confidence intervals of treatment effects illustrate that there may be true benefit of DHEA replacement, although we did not find any statistically significant difference in effects between DHEA and placebo in our study population (1). Information provided by confidence intervals overcomes the limitations of presenting trial results simply as positive or negative. In fact, our results are consistent with those of Johannsson et al. (3), who did not find significant effects on subjective health status or sexuality (confidence intervals not given). Moreover, if only studies demonstrating significant effects are published, the conclusion of benefit may be severely biased (publication bias). We do not agree with the use of crossover trials (1), because this design is very vulnerable to unblinding, particularly when psychological outcome measures are used (4). Such design is at high risk of reporting statistically significant effects that are not clinically relevant. For instance, we questioned the clinical relevance of negative placebo effects in the two previous crossover trials (5, 6). Taken together, we agree with Arlt and Allolio (2) that larger parallel group studies are required, and we await the conclusions of the larger clinical trial by Gurnell and co-workers (7).
Received April 15, 2003.
References