Background: Considering the increasing attention to the study of failed clinical trials, the goal of this study was to identify the sources of unreliability in a failed clinical trial by assessing scores on the Positive and Negative Syndrome Scale (PANSS).
Methods: This study is a substudy from a failed phase 2 double-blind, placebo-controlled trial of schizophrenia. Using the generalizability theory, this substudy assesses reliability on 3 conditions: raters, time points (PANSS evaluations, 1 week apart), subjects for 3 groups (placebo responders, placebo nonresponders, and treatment group).
Results: The placebo response rate was 40.07% (32/71). For all PANSS positive symptom items, the most variability was for raters (range, 33%–72%) for the placebo responders, 31% to 68% for the placebo nonresponders, and 29% to 60% for the treatment group. The variability of the interaction of rater and time point was the second source of unreliability, with an average of 12.28% compared to 12.00% for the placebo nonresponders and 10.00% for the treatment group. All items of the negative symptom subscale showed the most percent variability for raters, for all groups. For general psychopathology items (except preoccupation), raters accounted for the most variability in the scores for placebo responders with an average of 51.00% across items. A similar pattern was observed for the placebo nonresponders and for the treatment group; for the treatment group, the interaction between rater and time point accounted for the most variability for somatic concern and anxiety.
Conclusions: Results confirm the efficacy of applying the generalizability theory to the estimation of reliability to identify a source of unreliability and provide evidence for the relationship between low reliability and failed trials. Findings can be used to guide data monitoring, rater training, and identification of PANSS items, which may require supplementary training.