Application of a new statistical approach to evaluate a clinical trial with panic disorder patients.

Abstract

In clinical trials in psychiatry, changes in severity are usually measured with ordinal level scales which are applied repeatedly during the trial, showing a constant decline in psychopathology scores as treatment leads to improvement. Previous non-parametric tests for repeated measures in factorial designs did not test the hypothesis that scale scores decrease constantly during the trial. A recently developed "rank test for ordered alternatives in a mixed model" was developed and applied to the data of a clinical trial in panic disorder. Thirty-seven outpatients with panic disorder and agoraphobia (PDA) were treated with imipramine (75-150 mg/day) in an 8-week open, prospective trial. Patients with intercurrent agoraphobia were instructed in practising self-exposure in their agoraphobic situations. The total score on the Panic and Agoraphobia Scale, the Hamilton Anxiety Scale (HAMA) and the Clinical Global Impression Scale (CGI) were used as the main efficacy measures. The new rank test showed significant treatment results in all scales applied. Treatment results were excellent, as was shown by a decrease in the average Panic and Agoraphobia Scale severity scores from 28.9 (range 14-45) to 13.3 (range 0-37; rank statistic Tn = 6.7; p < 0.0001). The largest effect size r(w) of all clinician-rated scales was seen with the observer-rated version of the Panic and Agoraphobia Scale, although closely followed by the CGI and the HAMA. Among the self-rated scales, the Panic and Agoraphobia Scale also showed the largest effect size. All five subscores of the Panic and Agoraphobia Scale showed significant improvements. The highest treatment effect sizes were seen in the "panic attacks" subscore, followed by the "anticipatory anxiety" subscore. The new statistical test applied in this study, which has some advantages in comparison with previously applied tests, is suitable for psychiatric treatment evaluations since it can also be applied in the case of discrete repeated measurements.