Predictors of Placebo Responses in Bipolar Treatment Trials Identified

Caroline Helwick

September 01, 2010

September 1, 2010 (Amsterdam, The Netherlands) — It might be possible to predict which patients in clinical trials of bipolar-mania treatments will respond favorably to placebo, according to a meta-analysis by an international team of investigators, who reported their results here at the 23rd European College of Neuropsychopharmacology Congress.

The study was conducted in response to the high placebo response rates commonly observed in trials of bipolar-mania treatment. "We wanted to ask, what is driving this phenomenon and what can be done to minimize it?" said Aysegül Yildiz, MD, professor of psychiatry at the Dokuz Eylül University in Izmir, Turkey, who presented the findings at a breaking news session.

Harvard Medical School and the International Consortium for Bipolar Disorder Research in Boston, Massachusetts, were also involved in the study.

The identification of factors associated with a response to placebo would have several positive effects, the authors suggested. "This might lead to more cost-effective and feasible trials, with exposure of fewer persons to inactive treatments and facilitation of subject recruitment," Dr. Yildiz explained.

"In response to the many challenges when including placebo controls in treatment trials, we performed a meta-analysis to identify factors that might limit placebo response and enhance drug–placebo contrasts in randomized trials of single treatments for acute bipolar-mania," she said.

The study used unrestricted maximum-likelihood mixed-effects metaregression modeling to assess the influences of study-site count, subject age, sex distribution, diagnostic subgroup, clinical features, and trial completion rates on the corresponding outcome measure, which was the mean difference for change in mania ratings between baseline and end points. Drug–placebo differences in outcomes were also analyzed, as was the rate of response to the drug and placebo.

Researchers identified 38 randomized controlled trials involving 3812 placebo recipients and 6988 drug-treated patients. The studies yielded 56 placebo comparisons (13 with negative results) of 17 active drugs. Most studies (90%) were multicenter industry-sponsored collaborations, with an average site count of 30 (range, 1 to 70).

Several Factors Identified That Predict Placebo Response

Symptomatic improvement over baseline was similar in the placebo groups, regardless of whether or not the active drug proved effective. The mean difference in mania ratings between baseline and end point for the placebo groups was 6.77 for trials showing drug efficacy and 7.61 for trials showing no efficacy. For the drug groups in trials of effective drugs, the pooled effect was a difference of 12.7 points, approximately twice as large as the difference observed with placebo.

Similarly, response rates to placebo were comparable for effective (30.7%) and ineffective drug trials (31.6%). This indicates that most of the contrast in outcomes in trials of apparently effective, compared with ineffective, agents was due to marked differences in drug responses, she noted.

In metaregression modeling, several factors were found to be significantly associated with either a greater or lesser placebo-induced improvement:

  • A larger number of collaborating study sites was associated with a larger placebo effect (P < .0001) and a smaller treatment effect for drug vs placebo (= .00015), but was not associated with a drug effect (P = .43)

  • A larger sample size was associated with a larger placebo effect (P < –.0001) and a smaller treatment effect (P = .008), but not a smaller drug effect

  • Being male was associated with a smaller placebo effect (P = .0024) and a larger drug–placebo contrast (P = .00048), but was not associated with a drug effect

  • Being younger was associated with a smaller placebo effect (P = .0021) and with a larger drug–placebo contrast (P = .00006) and drug effect (P = .0028)

  • Inclusion of more patients with initial psychotic features was associated with a larger drug-associated improvement (P = .00002) and drug–placebo contrast (P = .00009), but was not associated with a placebo effect

  • Inclusion of more patients with mixed-state diagnoses was associated with less drug-associated improvement (P = .001) and less drug–placebo contrast (P =.0035), but was not associated with a placebo effect.

Diagnostic subgroups (mania with or without psychotic features, or manic vs mixed states) had no apparent influence on placebo effect overall; however, the presence of psychotic symptoms increased both drug–placebo contrast and drug-associated benefit. Higher baseline mania ratings predicted greater improvement with the drug.

In addition, a high percentage of trial completion (i.e., a lower dropout rate) in the placebo group was not related to placebo response, although high completion rates in the active groups was associated with both greater drug-associated benefit and drug–placebo contrast.

No association was seen with publication year. However, the exclusion of 2 small academic single-site trials with unusually low placebo effects led to an association of rising placebo effects in more recent years, Dr. Yildiz added.

"The present findings suggest there is a profile of manic patients less likely to improve with placebo, or to show greater drug–placebo contrast — that is, a greater likelihood of a successful trial outcome," she concluded. "The main characteristics include trials with fewer study sites and patients of younger age, male sex, and psychotic features. We propose that this aspect of trial design is readily accomplished and should improve the efficiency of randomized controlled trials, at least for mania."

Guy Goodwin, MD, from Warneford Hospital in Oxford, United Kingdom, moderated a session on the placebo effect in clinical trials, and commented on the current study for Medscape Medical News. "It is the authors' observation that with a fewer number of sites, you are more likely to see a separation of the 2 treatment arms. Whether this reflects [the fact] that with fewer sites companies put more effort into training cannot be determined from the data. But what is important is not the number of sites but the quality of the site. The problem is that you have to accept that some sites perform well and others underperform; the trick is to determine in advance which sites these will be. Limiting the number sounds like a good idea post hoc, but if you don't know in advance which ones these are, then the companies are forced to go with as many sites as they can get. Only later will they discover which are hopeless."

Dr. Goodwin reports financial relationships in the form of grants, honoraria, or material support from Sanofi-Aventis, Servier, AstraZeneca, Bristol-Myers Squibb, Eisai, Lundbeck, P1Vital, Wyeth, and Lilly. The authors have disclosed no relevant financial relationships.

23rd European College of Neuropsychopharmacology (ECNP) Congress: Abstract S.13.02. Presented August 30, 2010.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.