Assessing the Signal Detection Properties of 3 Outcome Scales Used in Clinical Trials of Patients with Generalized Anxiety Disorder

Introduction

This analysis assessed correlations between the signal detection properties (drug vs placebo) of 3 outcome scales—the 14-item Hamilton Rating Scale for Anxiety (HAM-A), the Clinician Global Impression of Severity (CGI-S), and the Clinical Global Impression of Improvement (CGI-I)—in generalized anxiety disorder (GAD) clinical trials.

Methods

Data from 5 randomized, double-blind, placebo-controlled venlafaxine XR studies in adult patients with GAD were pooled and examined individually. For all rating scales, Pearson correlation coefficients were calculated for all patients at each visit and by treatment arm. To evaluate signal detection properties, effect sizes and P values based on the pooled and individual study data were examined for the 3 scales.

Results

At pretreatment visits, for the HAM-A and CGI-S, respectively, 1837 and 1831 observations were available, with mean scores of 25.8 and 4.5, and the correlation coefficient between the 2 scales was 0.55 (P < 0.0001). Correlation coefficients at week 1 were 0.69 (HAM-A/CGI-S), 0.66 (HAM-A/CGI-I), and 0.55 (CGI-S/CGI-I). Correlation coefficients increased each week, and at final visit were 0.83 (HAM-A/CGI-S), 0.84 (HAM-A/CGI-I), and 0.82 (CGI-I/CGI-S). All correlations were significant (P < 0.0001) and were of comparable magnitude in the venlafaxine XR and placebo groups. Pooled effect sizes (venlafaxine XR versus placebo) were 0.37, 0.41, and 0.40 for HAM-A, CGI-S, and CGI-I, respectively (week 8 LOCF). Across studies, effect sizes ranged from 0.21 to 0.55, 0.23 to 0.68, and 0.26 to 0.59 for the HAM-A, CGI-S, and CGI-I, respectively. As with the pooled data, however, within studies, they were more consistent across the 3 outcome measures. All 3 outcome measures reached statistical significance (P < 0.05) in 4 of 5 studies.

Conclusions

The 3 scales were consistently correlated in all studies, and the correlations increased during the conduct of the study. Effect sizes based on different scales in the same studies were more similar than effect sizes based on the same scale in different studies. Furthermore, no one scale stood out as having consistently better signal detection properties than the others.