An Analysis of Correlations Among 4 Outcome Scales Employed in Clinical Trials of Patients With Major Depressive Disorder

Abstract

Introduction

The objective of this analysis was to identify and assess any correlations among 4 widely used rating scales—the17-item Hamilton Depression Rating Scale (HAM-D₁₇), the Montgomery-Asberg Depression Scale (MADRS), the Clinical Global Impression-Severity (CGI-S) and -Improvement (CGI-I)—in clinical trials with patients with major depressive disorder (MDD).

Methods

Data from 22 randomized, double-blind, placebo-controlled venlafaxine studies (10 ER studies, 11 IR studies, 1 with both formulations) in adult patients with MDD were pooled and examined from baseline through the first 8 weeks of treatment. For all rating scales, Pearson correlation coefficients were calculated between change scores and by treatment arm for patients at each visit. Correlations between binary outcomes (response defined as CGI-I and CGI-S ≤ 2, 50% decrease in HAM-D₁₇ and MADRS) were determined.

Results

At pretreatment visits, for the HAM-D₁₇, MADRS, and CGI-S, respectively, 5117, 4871, and 5103 observations were available, with mean scores of 23.0, 29.1, and 4.4. Pretreatment correlations ranged from 0.52 (CGI-S and HAM-D₁₇), 0.53 (CGI-S and MADRS), and 0.62 (HAM-D₁₇ and MADRS). Correlations between scales increased at each visit and, at 8 weeks, ranged from 0.87 (CGI-S and CGI-I) to 0.93 (HAM-D₁₇ and MADRS). Correlation coefficients in treatment arm subgroup analyses and between change scores were comparable. Correlation coefficients between binary outcomes were lower, ranging from 0.42 (CGI-I and CGI-S) to 0.61 (HAM-D₁₇ and MADRS) at week 1 and from 0.61 (CGI-I and CGI-S) to 0.81 (HAM-D₁₇ and MADRS) at week 8. All correlation coefficients were significant (P < 0.0001).

Conclusions

Correlations among the four commonly used outcome scales were high; however, correlations among binary outcomes based on the scales were lower. The highest correlations were between the HAM-D₁₇ and the MADRS, which share several items and have similar modes of administration and rating. The modest but consistently lower correlations between the CGI-S and CGI-I scales were unexpected because these scales are sometimes considered interchangeable.