Quality Assurance of Depression Ratings in Psychiatric Clinical Trials.

Background: Extensive experience with antidepressant clinical trials indicates that interrater reliability (IRR) must be maintained to achieve reliable clinical trial results. Contract research organizations have generally accepted 6 points of rating disparity between study site raters and central "master raters" as concordant, in part because of the personnel turnover and variability within many contract research organizations. We developed and tested an "insourced" model using a small, dedicated team of rater program managers (RPMs), to determine whether 3 points of disparity could successfully be demonstrated as a feasible standard for rating concordance.

Methods: Site raters recorded and scored all Montgomery-Åsberg Depression Rating Scale (MADRS) interviews. Audio files were independently reviewed and scored by RPMs within 24 to 48 hours. Concordance was defined as the absolute difference in MADRS total score of 3 points or less. A MADRS total score that differed by 4 or more points triggered a discussion with the site rater and additional training, as needed.

Results: In a sample of 236 ratings (58 patients), IRR between site ratings and blinded independent RPM ratings was 94.49% (223/236). The lowest concordance, 87.93%, occurred at visit 2, which was the baseline visit in the clinical trial. Concordance rates at visits 3, 4, 5, and 6 were 93.75%, 96.08%, 97.30%, and 100.00%, respectively. The absolute mean difference in MADRS rating pairs was 1.77 points (95% confidence interval: 1.58-1.95). The intraclass correlation was 0.984 and an η2 = 0.992 (F = 124.35, P < 0.0001).

Conclusions: Rigorous rater training together with real-time monitoring of site raters by RPMs can achieve a high degree of IRR on the MADRS.