Is there such a thing as a placebo effect? If so, at what point does the drug effect really occur? How do we distinguish between the two?
"Chester" is a 36-year-old engineer who has been depressed for almost a year. I started him on an antidepressant medication three weeks ago, and in his first follow-up visit with me, he says he is doing much better. "I'm almost back to my normal self," he reports.
"That's great," I reply. "When did you first start to feel better?"
"Well, I actually started to feel better about three or four days after starting the medicine," he says almost sheepishly. "That's probably a placebo effect, isn't it?"
The notion that the placebo response occurs early while a true antidepressant response is delayed has become well entrenched in the consciousness of both clinicians and patients alike. It seems to be a given principle of antidepressant pharmacotherapy. But where did this notion come from? In one of the earliest descriptions of antidepressant therapy, Roland Kuhn (1957) suggested that the benefits of imipramine (Tofranil) could be observed within two to three days. Although this observation might be discounted as being anecdotal, it should be recalled that the notion that a pill could ameliorate the course of a depressive illness was somewhat of an anathema at that time, and Kuhn picked imipramine among a plethora of sedating agents as having an antidepressant effect. Nevertheless, the idea that antidepressants take several weeks to work gained increasing acceptance, as evidenced by review articles published over the next two decades by several leading psychopharmacologists--though none of these articles cited original research supporting this belief (Hollister, 1978; Hordern, 1965; Klerman and Cole, 1965).
In 1984, researchers from Columbia University published the first scientific study supporting the delayed antidepressant response hypothesis (Quitkin et al., 1984a). In this study, the investigators pooled the results of three antidepressant trials they had conducted over the years. The study was somewhat novel in that the investigators depicted weekly changes in depression rating scores rather than only end point results. Also, all the trials were six weeks in duration, whereas most previous trials had been only four weeks. The investigators found drug-placebo separation at end point that was comparable to most other research conducted at that time. When they examined the time course of improvement, however, they found that almost all of the drug-placebo separation occurred between week 3 and week 6, and almost no drug-placebo separation occurred during the first three weeks of treatment. This was the first scientific validation of the delayed antidepressant response theory.
The same research group subsequently published a series of studies utilizing an innovative technique called pattern analysis (Quitkin et al., 1987, 1984b). In brief, pattern analysis examines the weekly response rates for each participant over the course of a six-week antidepressant trial. Response status is dichotomized as positive (very much or much improved) or negative (minimally or not improved). A participant who responds during the first week of treatment, for example, and who remains improved through week 6 would have the following pattern of response: positive/positive/positive/positive/ positive/positive. A participant who is unimproved during weeks 1 and 2, but improves at week 3 would have the following pattern: negative/negative/ positive/positive/positive/positive. The researchers hypothesized that the latter pattern of response is more likely to represent a "true drug response," whereas the former pattern is more likely to represent a "placebo response" pattern.
The investigators found that a true drug response pattern (i.e., delayed response) is much more likely to occur for participants receiving an active agent than placebo, whereas placebo response pattern (i.e., early response) occurs with equal frequency among both groups (some responders to active medication may of course be placebo responders). Furthermore, individuals who improve on active medication with a placebo response pattern are more likely to relapse during the maintenance phase of treatment (between six weeks and six months) than individuals who demonstrate a true drug response pattern (McGrath et al., 2000; Quitkin et al., 2003, 1993; Stewart et al., 1998). Presumably, the former cohort had responded to nonspecific factors and, therefore, benefited less from maintenance therapy.
In sum, the Columbia group has produced a body of work that is impressive in its breadth, consistency and rigor, and they have replicated most of their central findings. It is due to this body of research--and largely from this body of research alone--that the delayed antidepressant hypothesis has become so widely accepted.
As impressive as these findings are, several points undermine the delayed antidepressant response hypothesis. First, there is an important flaw in the design of the Columbia group's research. In each of the pattern analysis studies, a single-blind placebo lead-in period was used. As is customary, participants who improved during this lead-in phase were excluded from the active phase of treatment. What this means is that all participants who were purportedly demonstrating an early response to placebo during the double-blind phase of treatment were actually demonstrating a delayed response of at least seven to 10 days (coinciding with the lead-in phase) plus one to two weeks of treatment during the active phase. Thus, the body of research that purportedly depicted the placebo response as occurring early, in reality, is depicting participants who displayed a delayed response of two to three weeks. Second, studies that have examined predictors of response have almost invariably found that improvement occurring during the first two weeks of treatment is the best predictor of response at end point (Nierenberg et al., 2000, 1995). This finding is inconsistent with the delayed antidepressant response theory, since early improvement is only a result of nonspecific (i.e., placebo) factors. Third, in the two decades since pattern analysis was first depicted, it has rarely, if ever, been independently replicated. If pattern analysis can magnify drug-placebo separation as it originally was portrayed as doing, why hasn't the pharmaceutical industry utilized this well-accepted tool to bolster their results?
In an effort to try to reconcile these issues, we conducted a meta-analysis of a large number of antidepressant trials published over the last two decades (Posternak and Zimmerman, 2005). Studies were included in the meta-analysis if they were double-blind, placebo-controlled trials; lasted at least four weeks (most were six weeks); enrolled individuals with major depressive disorder; included an accepted antidepressant agent; utilized the Hamilton Rating Scale for Depression (HAM-D) to measure improvement; and depicted weekly or biweekly changes in HAM-D scores. In total, 47 trials met the inclusion criteria, which included 5,158 participants receiving active medication and 3,418 receiving placebo.
Our goal was to compare the time course of improvement on antidepressant medication and placebo. In particular, we focused on the first two weeks of treatment, since the delayed antidepressant response theory predicts that the benefits of antidepressant therapy should not be apparent during this period. We first asked: How much of the improvement that occurs by the conclusion of a six-week antidepressant trial takes place during the first two weeks of treatment? Based on reductions in HAM-D scores, we found that as much as 60% of the improvement that occurs on antidepressant medication takes place during the first two weeks of treatment. In contrast to what the delayed antidepressant response theory would predict, we found that almost an identical proportion (61.6%) of the improvement that occurs on placebo also takes place during the first two weeks of treatment.
We next compared the time course of improvement on antidepressant medication and placebo. Over the course of six weeks, HAM-D scores decreased by an average of about 13 points for participants receiving active medication and by about nine points for participants receiving placebo--a difference of four points. When does this four-point separation occur? The delayed antidepressant response theory predicts that most or all of the four-point separation should occur between weeks 3 and 6. We found, however, that 57% (2.33 of 4.09 HAM-D points) of the drug-placebo separation actually occurred during the first two weeks of treatment. Rather than a delayed response pattern, our results indicated that drug-placebo separation is most pronounced early in treatment and decreases in a stepwise fashion thereafter. This early effect could not be accounted for by differential dropout rates (Figure) (Due to copyright concerns, this Figure cannot be reproduced online. Please see p34 of the print edition--Ed.).
Examining this question from a different perspective, we asked: Of the patients who ultimately respond to active medication and placebo, when does this response occur? This analysis focused only on treatment responses to active medication and placebo. The delayed antidepressant response theory predicts that placebo responses should be more common during weeks 1 and 2, while active medication responses should be more common from weeks 3 to 6. In fact, we found nearly identical rates of response for each week of a six-week trial, with the placebo response appearing even slightly delayed compared to the antidepressant response.
All our findings suggested that an early antidepressant effect exists, but was there evidence to suggest that this early effect was clinically discernible? To address this issue, we analyzed the results of four of the original 47 trials. These trials published weekly response rates using the Clinical Global Impression (CGI) scale. The CGI scale is a well-accepted instrument that asks clinicians to globally rate whether or not they judge the depression to be much improved. As such, it has clear face validity to address the question of clinical relevance, and it also happens to be the same scale used by the Columbia group in each of their pattern analysis studies.
In this subanalysis, 564 participants were included in the active medication cohort and 521 participants in the placebo cohort. The antidepressants used and sample sizes randomized to active medication for each trial were paroxetine (Paxil) (n=168), imipramine (n=240), venlafaxine (Effexor) (n=79), imipramine (n=38), and zimelidine (n=39). Combining the results of these four trials, at the end of week 1, 100 of 564 (17.7%) participants receiving active medication were judged to be much improved, compared to 60 of 521 (11.5%) participants receiving placebo (χ2=8.32, df=1, p=0.004). After two weeks, 172 of 564 (30.5%) and 104 of 521 (20.0%) were considered much improved on active medication and placebo, respectively (χ2=15.85, df=1, p
Our meta-analysis was undertaken because of patients like the aforementioned Chester who reported feeling better soon after initiating antidepressant therapy. Many such patients exist. An analysis of 50 consecutive patients treated in our own clinical practice, for example, found that over 40% of the patients who responded to an antidepressant trial reported feeling better within the first two weeks of treatment. Traditional teaching would lead us to believe that these patients were responding to nonspecific factors of treatment (e.g., the instillation of hope, the therapeutic effects of receiving care) rather than experiencing a true drug effect. Our meta-analysis, however, suggests otherwise. Not only are the benefits of antidepressant therapy apparent early in treatment, but their effects are actually most pronounced during this period.
How can these findings be reconciled with the results from the Columbia group? After thoroughly considering methodological differences that might account for the disparate findings, we are still unable to explain the contradictory findings. We further believe that the two findings are incompatible. It is not akin to light being simultaneously construed of as both waves and particles; antidepressants can either work quickly or they cannot.
Our meta-analysis was based on a large body of data representative of most of the antidepressant research published over the last two decades. The question of an early antidepressant effect was examined from multiple perspectives, and the findings were unequivocal in showing an early antidepressant effect. Nevertheless, the Columbia group has produced a large and impressive body of research supporting the delayed antidepressant response theory, and this body of research cannot be easily dismissed.
The question of when antidepressants begin to work, therefore, remains an open question. It may seem surprising that 50 years following the introduction of effective pharmacotherapy such a basic question remains unanswered. At a time when millions of research dollars are being spent on imaging techniques, genetic testing and the development of novel treatments, it seems as if research related to the principles of standard antidepressant therapy has been relegated to the pharmaceutical industry. And yet, 50 years of pharmaceutical research has yielded little information other than that a variety of agents are effective in the treatment of depression. It is our hope that there will be renewed interest in evaluating such basic questions as when antidepressants begin to work and how long trials should be continued, so that we may better treat and manage today's generation of patients with depression.
Dr. Posternak is assistant professor of psychiatry and human behavior at Brown University.
Hollister LE (1978), Tricyclic antidepressants (first of two parts). N Engl J Med 299(20):1106-1109.
Hordern A (1965), The antidepressant drugs. N Engl J Med 272:1159-1169.
Klerman GL, Cole JO (1965), Clinical pharmacology of imipramine and related antidepressant compounds. Pharmacol Rev 17:101-141.
Kuhn R (1957), [ber die Behandlung depressiver Zustande mit einem Iminodibenzylderivat.] Schweizerische Medizinische Wochenschrift 87:1135-1140.
McGrath PJ, Stewart JW, Petkova E et al. (2000), Predictors of relapse during fluoxetine continuation or maintenance treatment of major depression. J Clin Psychiatry 61(7):518-524.
Nierenberg AA, Farabaugh AH, Alpert JE et al. (2000), Timing of onset of antidepressant response with fluoxetine treatment. Am J Psychiatry 157(9):1423-1428.
Nierenberg AA, McLean NE, Alpert JE et al. (1995), Early nonresponse to fluoxetine as a predictor of poor 8-week outcome. Am J Psychiatry 152(10):1500-1503 [see comment].
Posternak MA, Zimmerman M (2005), Is there a delay in the antidepressant effect? A meta-analysis. J Clin Psychiatry 66(2):148-158.
Quitkin FM, Petkova E, McGrath PJ et al. (2003), When should a trial of fluoxetine for major depression be declared failed? Am J Psychiatry 160(4):734-740.
Quitkin FM, Rabkin JD, Markowitz JM et al. (1987), Use of pattern analysis to identify true drug response. A replication. Arch Gen Psychiatry 44(3):259-264.
Quitkin FM, Rabkin JG, Ross D, McGrath PJ (1984a), Duration of antidepressant drug treatment. What is an adequate trial? Arch Gen Psychiatry 41(3):238-245.
Quitkin FM, Rabkin JG, Ross D, Stewart JW (1984b), Identification of true drug response to antidepressants. Use of pattern analysis. Arch Gen Psychiatry 41(8):782-786.
Quitkin FM, Stewart JW, McGrath PJ et al. (1993), Further evidence that a placebo response to antidepressants can be identified. Am J Psychiatry 150(4):566-570.
Stewart JW, Quitkin FM, McGrath PJ et al. (1998), Use of pattern analysis to predict differential relapse of remitted patients with major depression during 1 year of treatment with fluoxetine or placebo. Arch Gen Psychiatry 55(4):334-343.