- Depression severity, age, anhedonia, anxious arousal, neuroticism, presence of melancholic features, history of physical abuse and an electroencephalography measure predict response to placebo among patients with major depressive disorder
- A new interactive web-based calculator can help researchers exclude likely placebo responders from clinical trials of antidepressants
- Clinicians may wish to use the calculator to decide which patients need longer-term antidepressant therapy
Several investigational antidepressants have failed phase two or three clinical trials because of high placebo response rates. Most attempts to address this problem in subsequent trials have focused on changing the study design by increasing the sample size or including a placebo lead-in period with limited success.
There's now an alternative. Maurizio Fava, MD, director of the Division of Clinical Research for the Mass General Research Institute, and colleagues have identified eight easy-to-measure patient characteristics that define likely placebo responders with a high degree of accuracy (AUC, area under the curve >.0.73).
Using a web-based calculator to predict the likelihood of placebo response created by Dr. Fava's team, researchers will soon be able to exclude likely placebo responders from clinical trials of antidepressants in order to obtain a purer estimate of any treatment effect.
The findings, reported in Psychotherapy and Psychosomatics, may be of use in practice as well. Clinicians may recommend a brief low-cost, low–side effect intervention for placebo responders, rather than longer-term antidepressant therapy.
The researchers used data from the Establishing Moderators and Biosignatures for Antidepressant Response in Clinical Care (EMBARC) trial to explore close to 300 variables that might predict the likelihood of improvement with placebo. Specifically, they tested:
- 42 clinical and demographic variables
- 14 behavioral and cognitive performance measures
- 15 electroencephalography (EEG) measures
- 212 structural or functional imaging variables
In EMBARC, participants were 18 to 65 years old and had major depressive disorder. Stage one of the trial included an eight-week double-blind, placebo-controlled trial of sertraline that enrolled 309 subjects. Of those, 141 were assigned to placebo and received at least one dose, and they constituted the sample for the new study.
To narrow down the set of variables that predict response to placebo, the researchers used an unusual regression analysis technique called the elastic net. It pulls the regression estimates toward zero while reducing the variance of the estimates in order to increase the overall predictive power.
After the research team applied the elastic net to 100 imputed data sets, eight of the 283 variables were retained in at least 50% of the runs:
- Baseline score on the 17-item Hamilton Rating Scale for Depression
- Anxious arousal as measured by the Mood and Anxiety Symptom Questionnaire–30
- Presence of melancholic features
- History of physical abuse
- Average theta current density in the rostral anterior cingulate cortex on EEG
Using these variables, the researchers conducted a novel Bayesian method to simultaneously predict the degree of symptom change and the probability of response and remission. The EEG measure predicted better outcomes with placebo, while the seven other variables predicted worse outcomes with placebo.
Learn more about the Depression Clinical & Research Center
Learn more about the Division of Clinical Research