Expectancy in Antidepressant Treatment
Bret Rutherford, a psychiatrist at Columbia University and the New York State Psychiatric Institute, and his colleagues have written a series of papers which examine the phenomenon of expectancy, which is a patient’s (or clinical trial participant’s) expectation that the treatment will help. Peter Kramer’s new book, Ordinarily Well: The Case for Antidepressants, comes to somewhat different conclusions about expectancy.
My review of Fabrizio Benedetti’s work identified several neurobiological components of placebo responses--reducing anxiety about symptoms, activating the brain’s reward system, nonconscious conditioned learning, social learning, and reinforcing expectations. He associates these with activity in particular brain regions and circuits.
In a 2013 review, Rutherford and Steven Roose dissect out the contributors to placebo response rates in antidepressant clinical trials. These include measurement issues such as rater bias, response bias, and regression to the mean; the natural history of the illness; patient characteristics; treatment effects, including supportive aspects of the treatment setting; and expectancy effects. They note that both conscious positive expectation of improvement and nonconscious classical conditioning, in which aspects of the treatment or treatment setting become associated with improvement, likely contribute. Most research has studied conscious expectancy; the authors attribute this to the prominent role of verbal information in shaping placebo effects and to the fact that placebo effects occur in treatment-naïve patients. I would add that conscious expectancy may be easier to study than nonconscious conditioning.
Rutherford and Roose cite a number of indirect analyses which imply a large role for patient expectancy in antidepressant drug trials. For example, in one meta-analysis , for each 10% reduction in the probability of receiving placebo, the probability of drug response increased by 1.8% and the probability of placebo response increased by 2.6%. They also discuss patient selection factors—older studies took place in inpatient settings, while today’s subjects are volunteers responding to advertisements. Placebo response rates decrease with the severity of baseline depression scores, which for me raises the question of aspects of the depressed condition, such as imperviousness to hopeful suggestion, may interfere with expectancy. They note that there is little actual evidence for the “assumption of additivity”—that drug effects can be calculated by subtracting placebo response rates from drug response rates--which may be invalid if drug effects overwhelm placebo effects, or if medication interacts with placebo effects in other ways. Nevertheless, clinical trial results are expressed in effect sizes and numbers needed to treat, which are based on this assumption. Untangling this is difficult—deceiving patients about what they may be receiving is considered unethical in depression trials, and newer statistical approaches may not be adequate.
Rutherford and Roose note that clinical trials involve frequent visits to perform assessments and keep patients engaged so they don’t drop out. This support is the “treatment setting” component of the placebo response. It is far more clinical contact than the three quarters of depressed patients who are treated solely by their primary care providers receive, but it is more like care in some psychiatric practices, particularly if the psychiatrist is providing psychotherapy as well as medication.
In another review, Rutherford, Tor Wager, and Roose discuss expectancy in more detail. They note that expectations are likely to vary before, during, and after treatment. Ideas about the prestige, credibility, and sophistication of the treatment affect initial expectations. During treatment, expectations may change in response to the therapeutic alliance, initial treatment effects and side effects, and the severity of illness.
Most research on expectancy has relied on inferences based on re-analyses or meta-analyses of clinical trials. An example is a study by Rutherford’s group which looked at symptom levels in the weeks after patients who had responded to fluoxetine were randomized to placebo or continued fluoxetine. Patients on both the drug and placebo arms had higher depression ratings after they knew they might be switched to placebo, suggesting that the expectation of possibly being switched to placebo generated the increased symptoms. In addition, patients whose symptom levels dropped very early in their original open-label treatment with fluoxetine were more likely to have levels rise when they were later randomized to drug or placebo, which may mean they were a subgroup more responsive to expectancy effects.
The authors reference studies of psychiatric patients in which expectations correlated with therapeutic improvement. Two involved major depression—an NIMH comparison of cognitive-behavioral therapy, interpersonal therapy, imipramine, and placebo with clinical management. For patients treated with psychotherapy or medication, pretreatment expectancy was correlated with lower final depression scores. The same was true in a smaller medication trial.
Rutherford’s group also conducted a pilot study of experimental manipulation of expectancy. They randomized 42 patients to a placebo-controlled track of treatment with escitalopram or placebo, or a comparator track of escitalopram or citalopram. They measured expectancy with the Credibility and Expectancy scale, the most widely used validated scale for this purpose. Those who knew they would receive an active drug had higher expectancy scores than those who knew they had a 50% chance of receiving placebo. There was no significant difference between the placebo-controlled and comparator groups in medication response, but higher initial expectancy scores were significantly correlated with lower final depression scores. This suggestive finding in a small sample is consistent with the hypothesis that expectation influences response in treatment of depression.
Neuroimaging studies in patients with Parkinson’s disease have found dopamine release in the striatum when patients expect improvement after administration of a placebo. Studies of normal individuals’ expectations of pain have found placebo-induced rises in opioid activity in the orbital-frontal and anterior cingulate cortices and amygdala. The same regions were activated in a study in which anxiety was manipulated with a benzodiazepine or a benzodiazepine antagonist. Importantly, in two studies, depressed patients had decreased responses in regions expected to activate in response to emotional stimuli or reward. Two earlier studies found specific differences in brain activity in depressed patients who responded to placebo compared to nonresponders and medication responders, but expectations were not measured in these studies.
In Ordinarily Well, Kramer argues that there is little or no evidence for an expectancy effect with antidepressant medication. He relies primarily on studies comparing placebo with “no treatment,” most involving discontinuation of antidepressants. He also considered data about the timing of responses to placebo and medication—placebo responses are early and transient, and antidepressant responses take longer and persist. His own clinical observations, as well as the profession’s lore, support this pattern. He looks at a small trial which found expectancy effects in the placebo arm but not in the medication arm. He concludes that “the more reliably medicines work, the more they arouse expectations that then inflate placebo responses in drug trials; meanwhile, little or none of that benefit transfers to drugs, who do their work on their own (p158.)”
The situation is murky. Rutherford and colleagues are on the mark when the argue for attempting to measure expectancy directly rather than inferring it from placebo response rates, which are affected by a host of other factors, or from neuro-images showing activation of brain regions believed to be associated with expectancy. Those authors marshal indirect and a little direct evidence that supports the hypothesis of expectancy effects with antidepressants. Kramer looks at different kinds of indirect evidence and a small amount of other direct evidence to support his idea that any expectancy effects are overwhelmed by antidepressant drug effects.
More psychometric research, and including no-treatment arms in clinical trials, would be helpful, but untangling the neurobiology of expectancy and placebo effects with more genetic, imaging, and pharmacological studies of depressed patients will be necessary to assemble a more satisfying picture of what is going on. Important questions include the magnitude of expectancy effects in depressed patients; whether expectancy is negatively correlated with the severity of depression; whether expectancy effects are overwhelmed by antidepressant drug effects; the neural correlates of expectancy in depressed patients; how expectancy changes during treatment; whether expectancy-based symptom reduction can persist; and how expectancy relates to the other placebo effects Benedetti outlined, including anxiety reduction, unconscious conditioning, and social learning.
Thanks to Jonathan Stewart, M.D., for bringing Dr. Rutherford’s work to my attention.
Rutherford BR, Roose SP. A Model of Placebo Responses in Antidepressant Trials. Am J Psychiatry 2013; 170(7):723-733.
Rutherford BR, Wager TD, Roose SP. Expectancy and the Treatment of Depression: A review of Experimental Methodology and Effects on Patient Outcome. Curr Psychiatry Rev 2010; 6(1):1-10.
Kramer, Peter D. Ordinarily Well: The Case for Antidepressants. 2016.