Amery W K, Hoing M, Debroye M, Dom F
Department of Clinical R & D, Janssen Pharmaceutica NV, Beerse, Belgium.
Neuroepidemiology. 1987;6(4):220-7. doi: 10.1159/000110123.
Statistical aspects need particular attention in reporting on drug studies in migraine. It is important to report on all patient and disease characteristics which have contributed to the diagnosis and on all variables used for the evaluation of the result; the use of median values, accompanied by the extremes, is recommended in this context. Comparability of the study groups should be examined and possibly repeated after exclusion of the defaulters, if their number is substantial. As regards the presentation of the results, it does not suffice to give p values only: the magnitude of the response should be given and the use of techniques estimating the reliability of this magnitude (e.g. confidence intervals) and/or its reproducibility (e.g. power calculations) is recommended. An analysis of three unrelated groups of migraineurs demonstrates that most pertinent variables show a non-Gaussian distribution, and this includes the distribution of treatment results. For this reason, the use of nonparametric tests is advocated.
