AMSTERDAM – Novel trial designs can be used to reduce the high placebo response seen in clinical trials of antidepressants and thus increase the efficiency of those trials, according to a drug development scientist who proposed a “filtering” approach at the annual congress of the European College of Neuropsychopharmacology.
“We have found that the response to placebo is the strongest factor in a failed clinical trial,” said Dr. Emilio Merlo-Pich of the Centre of Excellence for External Drug Discovery, GlaxoSmithKline (GSK) in Verona, Italy, who spoke at a session called “The Placebo in Psychiatry.”
The placebo response trumps other reasons proposed to explain the high clinical trial failure rate for novel antidepressants, he said, including trial design features, heterogeneity of the study population, site-subject interactions, and low assay sensitivity related to clinical rating scale deficiencies.
Randomized clinical trial (RCT) outcomes can be improved by learning from past experience and using clinical databases and trial modeling, Dr. Merlo-Pich said. He and his colleagues have proposed an approach to improving RCTs using data to assess the role of recruitment centers and the problem of signal detection.
“Our study supports the implementation of prior-driven data preprocessing into RCT protocols to attenuate their failure rate, leading to a more efficient antidepressant drug discovery,” he said.
A meta-analysis of 52 RCTs indicated that placebo change from baseline to the end of study strongly affects the detection of active treatment superiority (J. Nerv. Ment. Dis. 2003;191:211-8). A statistically significant positive correlation was seen between placebo response magnitude and the advantage of antidepressants over placebo (P less than .0001). Only 21% of antidepressant treatment arms in trials with high placebo responses (more than a 30% mean change from baseline) showed superiority over placebo, compared with 74% in trials with a lower placebo response.
For the present study, a meta-analysis was conducted on nine GSK clinical trials, including 3,953 subjects with major depressive disorder, 1,197 of whom were exposed to placebo and 2,756 who had received antidepressants. The placebo response (reduction in depression score) varied highly according to the center.
A sensitivity analysis indicated that the placebo response by center was relevant for the detection of a treatment effect and for the success of the whole trial, Dr. Merlo-Pich reported. Placebo responses that are “too high” or “too low” generate noise within an RCT, he explained.
“We found that the placebo response can be so strong as to prevent any detection of a signal of pharmacologic effects, even if one is present. Therefore, the performance of each recruitment center is critical for the success of the whole trial,” he said. “In spite of training, recruitment centers manage protocols differently and handle patients differently, and this introduces bias. In fact, we have found the majority of the treatment effect depends on the center’s performance.”
Based on the level of the placebo response, the investigators classified individual centers as “informative” or “noninformative.” This classification was associated with the probability of detecting a signal of a clinically relevant treatment effect. The number of “informative” centers per study is relevant for the clinical trial outcome. “If you have enough informative centers, there is a higher probability of a positive trial,” he said.
In the study, only 60% of centers in the GSK database were classified as informative based on their specific level of placebo response, he reported.
Using this information, Dr. Merlo-Pich and his colleagues then ranked the centers’ performance, varying from 0 (high background nose and no chance to detect a treatment effect) to 100 (low noise and optimal condition for detecting a treatment effect). They then applied a “band-pass filter preprocessing approach” prior to statistical analysis as an enrichment strategy to single out the informative study population, reduce the noise, and improve the outcomes.
A clinical trial simulation was conducted to assess the performance of this “filtering” approach. The result was that the proportion of failed RCTs was reduced from 50% to 10%, he reported.
In implementing the model, the researchers define a-priori per-protocol high and low enrichment criteria to be applied at the end of treatment and before the statistical analysis. This identifies the noninformative centers to be eliminated, leaving the informative centers for the per-protocol efficacy analysis. The noninformative centers are still included in the intention-to-treat efficacy analysis and the safety analysis, he said.
“We believe we can apply this approach to any clinical trial. This will maximize our investment and enhance patient exposure to promising new compounds,” he concluded.
Disclosures: Dr. Merlo-Pich is a full-time employee of GlaxoSmithKline.