Analysis of Suicidality Data from Adult Antidepressant Trials Litigated by Law Firm Baum Hedlund December 13, 2006

GSK’s May 2006 Analysis of Paxil Clinical Trials Confirms Risk in Adults For its meta-analysis of suicidality data from the adult clinical trials, the FDA asked companies to submit data only from short term depression studies up to 17 weeks, which had at least 30 patients in the study. GSK decided to do its own blinded analysis of its adult clinical trial data. The results of the new analysis showed: “In adults with MDD (all ages), there is a statistically significant increase in the frequency of suicidal behavior in patients treated with paroxetine compared with placebo.” (​VIEW FULL DOCUMENT​) The masking of this risk in previous analyses was the result of a contamination of the data-set from inappropriately included anomalous studies. Once these studies were excluded due to FDA’s criteria, the increased risk of suicidality in adult patients taking Paxil, which had been evident from GSK’s initial submission, reappeared. The odds ratio is 6.7. As a result of GSK’s recent analysis, GSK strengthened Paxil’s label to include this new information and sent a “Dear Doctor” letter to every doctor in the United States including this information. ​Id.

GlaxoSmithKlineThree Franklin Plaza3F0615PO Box 7404Philadelphia PA 19101Tel. 215 751 4000www.gsk.comGlaxoSmithKlineApril 2006IMPORTANT PRESCRIBING INFORMATIONDear Healthcare Professional:GlaxoSmithKline (GSK) would like to advise you of important changes to the ClinicalWorsening and Suicide Risk subsection of the WARNINGS section in the labels forPAXIL (paroxetine HCl) and PAXIL CR (paroxetine HCl Controlled-Release Tablets).These labeling changes relate to your adult patients, particularly those who are youngeradults. Please read the full text of the added WARNINGS following this letter. Fullcopies of the revised package inserts for PAXIL and PAXIL CR are enclosed.Current prescribing information for paroxetine – and for all other antidepressants –contains information in the WARNINGS section (Clinical Worsening and Suicide Risksubsection) stating that “patients with MDD, both adult and pediatric, may experienceworsening of their depression and/or the emergence of suicidal ideation and behavior(suicidality), whether or not they are taking antidepressant medications, and this risk maypersist until significant remission occurs.”GSK has recently conducted a new meta-analysis (an addition to numerous prior analyses)of suicidal behavior and ideation in placebo-controlled clinical trials of paroxetine in adultpatients with psychiatric disorders including Major Depressive Disorder (MDD), otherdepression and non-depression disorders (e.g., dysthymia, panic disorder, generalizedanxiety disorder, obsessive compulsive disorder). These trials included 8958 patientstreated with paroxetine and 5953 with placebo.Results of this analysis showed a higher frequency of suicidal behavior in young adults(prospectively defined as age 18-24) treated with paroxetine compared with placebo(17/776 [2.19%] versus 5/542 [0.92%]). In the older age groups (25-64 years and ≥65years), no such increase was observed. This finding in young adults was not statisticiallysignificant; however, the difference was observed in paroxetine-treated patients with bothdepressive and non-depressive conditions.Further, in the analysis of adults with MDD (all ages), the frequency of suicidal behaviorwas higher in patients treated with paroxetine compared with placebo (11/3455 [0.32%]versus 1/1978 [0.05%]). This difference was statistically significant; however as theabsolute number and incidence of events are small, these data should be interpreted withcaution. All of the reported events of suicidal behavior in the adult patients with MDDwere non-fatal suicide attempts, and the majority of these attempts (8 of 11) were inyounger adults aged 18-30. These MDD data suggest that the higher frequency observed inthe younger adult population across psychiatric disorders may extend beyond the age of 24.The possible increase in risk of suicidal behavior in the MDD studies was observed despitesubstantial evidence for efficacy in the paroxetine-treated patients (compared with placebo)as determined by standardized disease-specific instruments (e.g., Hamilton DepressionMay 2006

Rating Scale and Montgomery-Asberg Depression Rating Scale for depression). Mostpatients had an identified social stressor at the time of the event.It is therefore important that all patients, especially young adults and those who areimproving, receive careful monitoring during paroxetine therapy regardless of thecondition being treated.It is difficult to conclude a causal relationship between paroxetine and suicidality due to thesmall incidence and absolute number of events, the retrospective nature of this meta-analysis, and potential for confounding by the fact that the events of interest are a symptomof the psychiatric illnesses themselves. However, GSK believes it is important to drawyour attention to these findings and is voluntarily amending the paroxetine labeling toreflect this new information and to emphasize the importance of careful monitoring of allpatients during paroxetine therapy. Please read the full text of the added WARNINGSfollowing this letter. Full copies of the revised package inserts for PAXIL and PAXIL CRare enclosed.GSK continues to believe that the overall risk:benefit of paroxetine in the treatment of adultpatients with MDD and other non-depressive psychiatric disorders remains positive.PAXIL is indicated for the treatment of major depressive disorder, obsessive-compulsivedisorder, panic disorder, social anxiety disorder, generalized anxiety disorder, andposttraumatic stress disorder in adults; PAXIL CR is indicated for the treatment of majordepressive disorder, panic disorder, social anxiety disorder, and premenstrual dysphoricdisorder in adults.The medical community can further our understanding of PAXIL and PAXIL CR by reportingadverse events to GlaxoSmithKline at 1-888-825-5249 or to FDA's MedWatch Adverse EventReporting program online (at www.fda.gov/MedWatch/report.htm), by phone (1-800-FDA-1088),or by returning the postage-paid FDA form 3500 (which may be downloaded fromwww.fda.gov/MedWatch/getforms.htm) by mail (to MedWatch, 5600 Fishers Lane, Rockville,MD 20852-9787) or fax (1-800-FDA-0178). GlaxoSmithKline encourages you to familiarize yourself with these revisions to labeling.If you have any questions about the new information, please contact our CustomerResponse Center at 1-888-825-5249. You can find other useful information related to thisissue at gsk.com and to clinical trials involving all other GSK products at our Clinical TrialRegistry website (http://ctr.gsk.co.uk/welcome.asp).Sincerely,John E. Kraus, MD, PhDDirector, Clinical Development Clinical Psychiatry- North America Neurosciences Medicines Development CenterGlaxoSmithKline

The Question of Efficacy Doctors must weigh the benefits of drug treatment versus the risks. In order to do a proper risk benefit analysis, a doctor must be aware of the degree of effectiveness of the drug – not just drug company hype. Is the drug extremely effective or only marginally effective? Doctors know the drug was approved by the FDA, but do they know the FDA’s standards for approving a drug as effective? In an analysis of efficacy data submitted to the FDA between 1987 and 1999 for six of the most popular selective serotonin reuptake inhibitor (SSRI) antidepressants, 75 to 80% of the response to medication was duplicated in placebo groups. (Kirsch and Moore, “The Emperor’s New Drugs: An Analysis of Antidepressant Medication Data Submitted to the U.S. Food and Drug Administration,” ​Prevention & Treatment​, Volume 5, Article 23, July 15, 2002.) These data were the basis on which the medications were approved by the FDA. The researchers explained that the “small difference between the drug response and the placebo response has been a ‘dirty little secret’ known to researchers who conduct clinical trials, FDA reviewers, and a small group of critics who analyzed the published data …” Kirsch, Moore et al., “Antidepressants and Placebos: Secrets, Revelations, and Unanswered Questions,” Prevention & Treatment, Volume 5, Article 33, posted July 15, 2002 (​VIEW FULL DOCUMENT​), Moncrieff and Kirsch,

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html1 of 1312/1/2006 2:00 PMPrevention & Treatment© 2002 by the American Psychological AssociationJuly 15, 2002 Vol. 5, Article 33For personal use only--not for distribution.Antidepressants and Placebos: Secrets, Revelations, and Unanswered QuestionsIrving KirschUniversity of ConnecticutAlan ScoboriaUniversity of ConnecticutThomas J. MooreGeorge Washington University We are very heartened by the thoughtful responses to our article. Unlike some of the responses to aprevious meta-analysis of antidepressant drug effects ( Kirsch & Sapirstein, 1998), there is now unanimous agreement among commentators that the mean difference between response to antidepressant drugs and response to inert placebo is very small. It is so small that, despite sample sizes involving hundreds of participants, 57% of the trials funded by the pharmaceutical industry failed to show a significant difference between drug and placebo. Most of these negative data were not published (see Thase, 2002) and were accessible only by gaining access to U.S. Food and Drug Administration (FDA) documents. The small difference between the drug response and the placebo response has been a \"dirty littlesecret\" ( Hollon, DeRubeis, Shelton, & Weiss, 2002), known to researchers who conduct clinical trials, FDA reviewers, and a small group of critics who analyzed the published data and reached conclusions similar to ours (e.g., Greenberg & Fisher, 1989). It was not known to the general public, depressed patients, or even their physicians. We are pleased that our effort facilitates dissemination of this information.1 The pharmaceutical company data that we have analyzed reveal a mean drug/placebo difference ofless than 2 points on the Hamilton depression scale (HAM-D), a difference that is not clinically significant (see Jacobson, Roberts, Berns, & Mcglinchey, 1999). Another way of describing this difference is in terms of response rates. Thase (2002), for example, notes that 35% - 50% of patients respond to medication compared with 25% - 30% who respond to placebo (see also Brown, 2002 Hollon et al., 2002; ). This is interpreted as indicating that 10% - 20% of depressed clinical trial patients show a true drug effect, in which case it follows that 80% - 90% of these patients do not. Response rate increments of 10% - 20% seem clinically important to some commentators (e.g., Thase, 2002) who have interpreted them as indicating the percentage of patients who show a strong response to antidepressants but would not be helped by placebo. This interpretation is mistaken ( Moncrieff, 2002). To classify patients as responders or nonresponders, one must establish a cutoff point (e.g., a 50% decrease in depressive symptoms; Mulrow et al., 1999). Consider two patients with baseline HAM-D scores of 20. One is randomized to the drug condition and shows a 10-point improvement on the HAM-D. The other is assigned to the placebo group and shows a 9-point improvement. The first would be classified as a responder and the second as a nonresponder, but in fact, the difference in response (1 point) is negligible. What makes this example particularly important is that these two patients are typical, rather thanexceptions. A 50% drug response is the median in published clinical trials ( Mulrow et al., 1999), and a

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html2 of 1312/1/2006 2:00 PM10-point decrease was the mean drug response in the FDA data set. Thus, the 10%-20% difference between drug and placebo response rates could be due entirely to those patients showing a moderately strong response (just above criterion) to drug and those showing almost as strong a response (just below criterion) to placebo. It would be obtained if there were not even one patient with a strong true drug effect. 2Of course, it could also be obtained if there were some patients with a larger true drug effect and others on which the drug had a negative effect or no effect at all. This would be true if there were a hidden moderator variable, a possibility raised by some reviewers and discussed by us below. We have suggested two possible explanations for the small differences between the drug response andthe placebo response: Either drug and placebo effects are not additive (in which case, conventional clinical trials are an inappropriate means of evaluating drug effects because they could lead to the rejection of truly effective medications, the benefits of which are largely masked by placebo), or the drug effect is very small. The commentators have raised three additional possibilities. They suggest that (a) there may be flaws in the ways in which trials sponsored by the pharmaceutical companies are conducted, (b) the small mean differences between drug and placebo obscure strong drug effects produced by some antidepressants in a subset of depressed patients, and (c) the effects of medication may be more stable than those of placebo. In the present reply to the commentaries, we consider each of these possibilities. We also consider the suggestion that the use of antidepressants is justified, no matter how small the effect, as long as it is statistically significant ( Brown, 2002 Moerman, 2002 Salamone, 2002 Thase, 2002; ; ; ).The Adequacy of the Clinical Trials The medications evaluated in the clinical trials that comprise the FDA data set were sponsored by thecompanies that manufacture them and therefore stand to benefit financially from a positive outcome. So, if there were any biases in these trials, one would expect them to favor the investigated drug ( Antonuccio, Burns, & Danton, 2002). Indeed, a quantitative review of the factors affecting response to antidepressants in clinical trials indicated that sponsorship of a trial by a drug company tended to produce effects favoring the sponsor's product ( Freemantle, Anderson, & Young, 2000). Some of the commentators on our analysis of the FDA data set have suggested that the opposite may be the case. Brown (2002) suggests that zealous researchers inflate baseline scores so that mildly depressed potential participants are not screenedout (see also Thase, 2002 Hollon et al. (2002)). maintain that low doses may be used to minimize sideeffects.Brown's (2002) suggestion is particularly troubling; it is tantamount to charging that clinical researchers have been fudging the data. Even more troubling is the news that this charge has been confirmed by a spokesperson for the FDA ( Elias, 2002). Inflating baseline scores also inflates the apparent benefit produced by treatment. Thus, both the drug response and the placebo response might be overestimated. These data were the basis on which the medications were approved by the FDA. If they are suspect, then perhaps the decision to approve the medications should be reconsidered.Hollon et al. (2002) suggest that the manufacturers of fluoxetine may have used low doses in their clinical trials, so as to be able to claim low levels of side effects when marketing their product. As a result, the therapeutic effect might also have been underestimated. This possibility is contradicted by the data. Fluoxetine is prescribed in dosages ranging from 20 to 60 mg daily. The two dose-response studies submitted by Eli Lilly to the FDA evaluated fluoxetine doses of 20, 40, and 60 mg. In one of these trials (conducted on mildly depressed patients), no significant differences were found between doses or between

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html3 of 1312/1/2006 2:00 PMany dose and placebo. In the other study (conducted on moderately to severely depressed patients), the two lower doses were significantly more effective than the high dose, which was not significantly more effective than placebo. According to the package label for fluoxetine, a particularly high dose (80 mg) was used in other clinical trials submitted to the FDA.Moderator Variables It has been suggested that some patients may be more responsive than others to medication or thatsome antidepressants are more effective than others ( Brown, 2002 Hollon et al., 2002 Thase, 2002; ; ). If this is the case, then drug/placebo differences, collapsed across patients and drugs, might underestimate the effect that some medications have on some patients. Our analysis included six different antidepressant medications. Among those medications for whichcomplete data were reported, the range of drug/placebo differences was between 1 and 3 points on the HAM-D. Even for those medications for which data on trials with negative results were withheld, the highest mean difference was 3.21 points. Thus, if there are any mean differences between these medications, they must be very small.Hollon et al. (2002) express the concern that excluding estimates for the three medications for whichscores for unsuccessful trials were not reported might result in the inadvertent omission of \"several of themore powerful current medications\" (¶ 10). They make specific reference to venlafaxine as a medicationthat might \"produce the most robust effects\" (¶ 10) because it works on multiple neurostransmitter systems.This concern is not well founded. First, venlafaxine was not one of the medications that were excluded in estimating effects across medication. Second, a meta-analysis of 105 clinical trials ( Freemantle et al., 2000) indicates that drugs that work on multiple neurotransmitter systems are not more effective than purelyserotonergic medications. Third, the weighted mean drug/placebo difference for the three excluded medications was 2.31 HAM-D points. Because this is an overestimate, in which one third of the trials (those showing particularly poor results for the active drug) were excluded, the real drug/placebo difference for the excluded medications cannot be significantly more than that of the medications for which full data were reported. If the medications do not produce significantly different responses, perhaps there are patient differencesthat mask more potent pharmacological effects. In support of this contention, Hollon et al. (2002) presented data from an unpublished study of paroxetine, in which patients were matched (post hoc) on rank order of change scores (i.e., the patient showing the least improvement in the active drug condition was matched with the patient showing the least improvement in the placebo condition, etc.). Drug/placebo differences between change scores of matched patients varied as a function of change score rank, but even among those pairs of participants showing the most change, the mean difference was only 3.39 points on theHAM-D.Hollon et al. (2002) also report percentages of pairs of patients showing differences of 4 points or greater and differences of 6 points or greater. They interpret the existence of these differences as indicatingthat some patients show a greater drug effect than do others. However, measurement error always produces a distribution around a mean, and Hollon et al.'s data provide no reason to suspect any other explanation. Even if the true drug/placebo differences were always between 2 and 3 points, some of the observed differences would be greater and some would be smaller. Further, there is an exceptionally high

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html4 of 1312/1/2006 2:00 PMdegree of error in the scores assessed by Hollon et al. Change scores are notoriously unreliable, even when pre- and posttest scores are highly reliable ( Campbell & Kenny, 1999). Differences between change scores are doubly unreliable. This unreliability is evident in the study described by Hollon et al., in which the pattern of differences between drug and placebo change scores varied substantially between the two sites at which the study was conducted. One method of coping with unreliability is to examine mean scores from large groups of participantsrather than individual participants or (as in the case of Hollon et al.'s [2002] data) pairs of participants. Even with relatively large samples, differences in sample size produce differences in the reliability of the data.Figure 1 displays drug/placebo differences in the FDA data set as a function of sample size. These differences vary widely in small samples, but they are substantially more reliable in largesamples. In addition, there is a tendency for large samples to show smaller drug/placebo differences. Asseen in Figure 1, a mean difference of about 2 points on the HAM-D characterizes drug/placebo differencesin the larger (and therefore, more reliable) studies. The most frequently proposed hypothetical moderator is baseline severity. The hypothesis is that moreseverely depressed patients show a greater response to medication and a smaller response to placebo. In the FDA data set, there was only one trial conducted on mildly depressed patients. Three trials were conducted on severely depressed, hospitalized inpatients, but the data from two of these trials were not reported because significant drug/placebo differences were not found. The vast bulk of the trials was conducted on patients judged to be moderately to severely depressed, and it is with these patients that the drug/placebo difference was approximately 2 points on the HAM-D. Thus, the drug/placebo difference appears to be negligible for most of the patients to whom antidepressants are prescribed.Figure 2 displays the relation between baseline scores and improvement in the larger ( = 200 or Ngreater), and therefore more reliable, clinical trials described in the FDA data set. More severely depressed patients showed greater improvement, but this is true of patients treated withplacebo as well as those treated with medication. This is due to regression toward the mean ( Campbell & Kenny, 1999), which almost always produces a correlation between baseline scores and change scores. 33 The slope of the regression line appears steeper for drug than for placebo, suggesting a greaterdrug/placebo difference among more severely depressed patients, but a regression analysis indicates that this difference is not statistically significant ( > .20). Nevertheless, because tests of differences in slope phave very low power, a greater drug/placebo difference among more severely depressed patients cannot be ruled out. Therefore, we calculated the drug/placebo differences observed in the six large studies with the lowest baseline depressions scores (range = 17.21 - 24.25) and compared them to the differences observed in the six large studies with the highest baseline depression scores (range = 25.15 - 27.85). Meandrug/placebo differences were 1.46 points in studies with lower baseline scores and 2.56 in studies with higher baseline scores. Further, this is likely to be an overestimate of the drug/placebo difference for the more severely depressed patients because it does not include the unreported data from two trials conducted on severely depressed, hospitalized inpatients, neither of which showed a significant difference between drug and placebo. The bottom line is that even in studies with more severely depressed patients, there is a strong placebo response and a relatively small difference between drug and placebo.

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html5 of 1312/1/2006 2:00 PM Finally, it is possible that different patients respond to different medications ( Hollon et al., 2002). Although there are scant experimental data in support of this hypothesis, clinicians report that patients who do not respond to a particular antidepressant sometimes respond when given a different medication. The results of an early study on the prevention of nausea are pertinent to this issue ( Wolf, Doering, Clark, & Hagans, 1957). Patients were successively given seven different treatments for nausea, following which ipecac was administered?. Response was defined as blocking the emetic effects of ipecac. There was substantial variability in response rates. Some participants responded to all treatments; some to none. Mostresponded to some treatments but not to others. Thus, the pattern of responding was similar to that observed clinically, following the administration of antidepressants: Patients who did not respond to a given medication often responded to another. In the Wolf et al. (1957) study, however, all seven treatments were known to be ineffective for the condition being treated. They were placebos. These data demonstrate that the response to placebo is inconsistent. Thus, finding that patients sometimes respond after switching to a different antidepressant is exactly what one would expect if antidepressants were nothing more than active placebos. The most extensive study of the hypothesis that different patients respond to different medications is theFreemantle et al. (2000) meta-analysis. They reviewed 105 clinical trials in which serotonin uptake inhibitors(SSRIs) were compared with other types of antidepressant and used regression analysis to find predictors of differential outcome. Hypothesized predictors included pharmacological action (noradrenaline reuptake inhibition, serotonin reuptake inhibition, and 5-HT2 receptor antagonism, dual action, and triple action), treatment setting (inpatient vs. outpatient), dose of the comparator drug, method of analysis (LOCF vs. completer data), age of the patient, measurement scale (HAM-D vs. any other), and sponsor of the trial. Only one of these factors bordered on significance: There was \"a trend towards increased efficacy of the sponsor's drug\" ( Freemantle et al., 2000, p. 294).Freemantle et al. (2000) did not use some of the factors that the commentators cited as potential predictors of differential outcome, and one can never rule out the possibility of undetected moderator variables. But if there are hidden moderators, the overall mean difference between drug and placebo (2 points on the HAM-D) constrains the conclusions that can be drawn from them. If the mean drug/placebo difference is greater than 2 points for a subset of medications or patients, then it must be less than 2 points for the others. For example, if the mean difference between drug and placebo is 4 points for half of the patients (which is still a rather small drug effect), then the mean effect of antidepressants on the other patients must be 0, and if it is more than 4 points for half the patients, then the medications must be causingharm to at least some others who would fare better on placebo.Is the Drug Response More Stable Than the Placebo Response? Our analysis was limited to the acute efficacy studies submitted to the FDA. It is possible that short-termdrug/placebo differences are negligible but that drug effects may be more stable than placebo effects, resulting in lower relapse rates ( Brown, 2002 Hollon et al., 2002 Thase, 2002; ; ), just as psychotherapy effects appear be more stable than medication effects ( Hollon, Shelton, & Loosen, 1991). Before looking at the data, however, it is necessary to consider two issues: the way in which the outcome is described and the design of long-term studies. There are two ways in which long-term outcome can be described. One is in terms of relapse followinginitial improvement; the other is in terms of continued response to treatment. Thase (1999), for example,

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html6 of 1312/1/2006 2:00 PMreported 12-month relapse rates of 20% with venlafaxine and 34% with placebo in a pooled analysis of four clinical trials. Thus, there were 1.5 times more relapses with placebo than with medication. Described in terms of continued efficacy, however, these same data indicate that, after 12 months, 80% of the patients receiving medication remained in remission, compared with 66% of patients on placebo. Thus, 83% of the effect of medication was duplicated by placebo. That this effect is similar to that seen in the short-term4trials rules out the possibility that antidepressant drug effects might be substantially greater over a longer period of treatment. A second issue that must be considered in understanding the results of long-term studies is the designof the study. There are two experimental designs that have been used. The more common design is the relapse prevention trial, in which patients who have responded to medication are then randomized to either continue on medication or be switched to placebo. A less common method of assessing long-term responseis with continuation or extension trials, in which patients who have responded to either drug or placebo during an acute efficacy trial are continued on the treatment to which they were originally assigned. The relapse prevention design is particularly biased against finding placebo effects. First, the clinicaltrials from which the patients are drawn typically begin with placebo washout periods, in which all patients who respond to placebo are eliminated from the study. Thus, participation in the relapse prevention trial is limited to patients who have not only responded to the active medication but have also failed to respond to placebo. Second, this design may exacerbate the problem of breaking blind on the basis of perceived side effects. It seems likely that a person who has been on an active medication and is then switched to placebowill be likely to detect the change. Despite this bias, relapse prevention trials often show large placebo responses. For example, one of thetwo 24-week relapse prevention trials reported to the FDA by the manufacturer of citalopram showed 2-yearresponse rates of 69% for placebo and 90% for citalopram. The other showed response rates of 76% for placebo and 86% for citalopram. Thus, in these studies, the long-term placebo response among patients who had been successfully treated with an active antidpressant was between 77% and 88% of the medication response. Similar results were reported by Walach and Maidhof (1999) in a meta-analysis of relapse prevention trials published between 1973 and 1990. In these trials, 71% of the drug response was duplicated by placebo. Walach and Maidhof also examined response to treatment as a function of the duration of the trial. Their data indicate that responses to both drug and placebo decrease over time. Contrary to conventional wisdom, however, the correlation between duration of the trial and response to treatment was higher for active medication ( = -.84) than for placebo ( = -.62), suggesting a steeper r rdecline in effectiveness for active drugs than for placebo. Continuation trials are relatively rare, but the data confirm the relative stability of the placebo responseover time. As Antonuccio et al. (2002) note, the 18-month followup of the NIMH Collaborative DepressionStudy ( Shea et al., 1992) reported equivalent long-term outcomes for both drug and placebo, with a lower relapse rate among remitted patients in the placebo group. A more recent study compared hypericum, sertraline, and placebo ( Hypericum Depression Trial Study Group, 2002) and reported 26-week continuation data for patients who had responded to treatment during the 8-week clinical trial. Of the 79 patients who entered the continuation phase, none of the placebo patients relapsed, none of the setraline patients relapsed, and only one taking hypericum relapsed.What Do We Do Now?

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html7 of 1312/1/2006 2:00 PM As many of the commentators have noted, the basic findings of our analysis are clear and undisputed.The mean difference between response to antidepressant medication and response to placebo is very small. This raises the question of what should be done in clinical practice. The response to both drug and placebo is substantial. How is this therapeutic benefit to be elicited if these medications, which may be little more than active placebos, are abandoned? Some commentators have argued that they should be given todepressed patients, even if their pharmacological effect is negligible. If nothing else, they can be the vehicleby means of which the placebo response is elicited ( Hollon et al., 2002 Moerman, 2002; ). This argument rests on the assumption that a genuine pharmacological effect has been proven. Theemperor's new clothes may not be the elegant suit his subjects were led to expect, but at least they are made with real fabric ( Thase, 2002). We have not denied that antidepressant medication might have genuine pharmacological effects. In fact, we raised the possibility that its effects could be much larger and reliable than the data suggest. The problem is that the effects of antidepressant medication are unknown. They may be very large, vanishingly small, or completely nonexistent. The emperor may be wearing an elegant invisible suit, a fig leaf, or nothing at all. It is true that, on average, there is a significantly greater response to medication than to placebo. Whatis not yet known is the reason for this difference in response. It may be a drug effect, but it may also be an enhanced placebo effect associated with the perception of side effects and the breaking of blind (Greenberg, 2002 Greenberg & Fisher, 1989; ). It is known that antidepressants produce significantly more5side effects than do those observed in inert placebo control groups ( Mulrow et al., 1999), patients assignedto conventional antidepressants are able to break blind to a significant degree ( Rabkin et al., 1986), and a drug effect has not been convincingly demonstrated in studies using active placebos ( Moncrieff, Wessely, & Hardy, 2001), despite the fact that the medications used as active placebos may not prevent patients from breaking blind ( Hollon et al., 2002). These data do not prove that the drug/placebo difference is due tothe breaking of blind, but they do suggest this as a reasonable possibility, one that needs to be ruled out before the claim of even minimal drug effectiveness can be considered to have been substantiated (Greenberg, 2002). In addition, the additivity assumption needs to be tested directly. The indirect data are mixed. As Brown (2002) notes, data showing different brain changes in placebo responders and drug responders ( Leuchter, Cook, Witte, Morgan, & Abrams, 2002) suggest that the effects may not be additive. Conversely, data showing similar brain changes in placebo responders and drug responders ( Mayberg et al., 1999) suggest a common mechanism, which is consistent with the additivity hypothesis. Additivity is also suggested by theparallel increases in response rates for SSRIs and placebos over the years ( Walsh, Seidman, Sysko, & Gould, 2002). Since the composition of the medication has not changed, the change must be due to other factors. We suspect that it is due to the dissemination of information (or misinformation) about the marvelous success of the newer antidepressants. We have raised the possibility of using the balanced placebo design, perhaps with active placebos asan aid in preserving the manipulation, as the most straightforward test of the additivity hypothesis. Hollon et al. (2002) have raised some useful concerns about the success of using active placebos for this purpose.Salamone (2002) has provided suggestions that might address some of those concerns, and Antonuccio et al. (2002) have suggested an alternative (assessing the degree to which blind is broken) that should be implemented routinely in pharmacological research but that does not obviate the need for direct tests of the additivity assumption.

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html8 of 1312/1/2006 2:00 PM There also are other questions that need to be answered. We need to know more about the placeboeffect and its underlying mechanisms. Rehm (2002) has provided many insightful research suggestions aimed at addressing this issue, and we hope that clinical researchers will implement them. We also thinkthat Salamone's (2002) suggestions for new studies of the dose-response relationship are worthwhile. The data sent to the FDA by the pharmaceutical companies included studies in which up to four different doses were evaluated, and except for the finding that lower doses of fluoxetine were more effective than a high dose, they failed to find reliable linear or quadratic differences between doses. It is possible, as Salamone (2002) suggests, that studies involving six or seven different doses would reveal some more complex relationships. In addition, there is a need for additional tests of possible moderator variables (chronicity, cortisol secretion, etc.). In the meantime, what are the alternatives for treating patients? Imagine having a choice between fourtreatments. Treatment A produces a large therapeutic response but also a large number of adverse effects, including diarrhea, nausea, anorexia, sweating, forgetfulness, bleeding, seizures, anxiety, mania, sleep disruption, and sexual dysfunction. Treatments B and C produce therapeutic responses that are almost as great as those produced by treatment A, but without the adverse effects. In fact, the side effects produced by Treatment B are beneficial (e.g., better general physical health). However, the therapeutic effects of Treatments B and C have been evaluated in relatively few studies. Treatment D has been assessed in many comparative studies, in which it has been found to be as effective as Treatment A in the short term and more effective in the long term. It does not produce adverse effects. Given a choice between these alternatives, which would you choose? Of course, these alternatives are not merely hypothetical. Treatment A corresponds to SSRIs, and thelist of side effects is drawn from those that have been shown to be produced by these medications (Antonuccio, Danton, DeNelsky, Greenberg, & Gordon, 1999 Mulrow et al., 1999; ). Treatment B is physical exercise, which has been reported to have lasting therapeutic benefits in the treatment of major depression( Babyak et al., 2000). It may be nothing more than a placebo, but if so, it is one with desirable rather than adverse side effects. Treatment C is bibliotherapy (e.g., Burns, 1999), another low-cost treatment with demonstrated effectiveness ( Jamison & Scogin, 1995 Smith, Floyd, Jamison, & Scogin, 1997; ) and little danger of side effects. Treatment D is psychotherapy. As noted by Antonuccio et al. (2002), \"psychotherapy(particularly cognitive therapy, behavioral activation, and interpersonal therapy) compares favorably with medications in the short term, even when the depression is severe (e.g., DeRubeis, Gelfand, Tang, & Simons, 1999), and appears superior to medications in long-term comparative studies (Antonuccio et al. 1995; Hollon, Shelton, & Loosen, 1991)\" (http://journals.apa.org/prevention/volume5/pre525c.html#p24¶24). Given these data, antidepressant medication might best be considered a last resort, restricted to patients who refuse or fail to respond to other treatments. It may well be that the effects of psychotherapy are due to expectancy, conditioning, and otherpsychological factors that have been hypothesized to be the basis of placebo effect ( Kirsch, 1997). Indeed, changing maladaptive expectations is an essential cornerstone of cognitive therapy. But this does not negate its effectiveness. The contention (e.g., Salamone, 2002 Thase, 2002; ) that the logic of placebo-controlled evaluation should be extended to psychotherapy is mistaken ( Kirsch, 1978). In drug research, placebos are used to distinguish the pharmacologically produced effects of substance administration from its psychologically produced effects. However, there are no pharmacologically producedeffects of psychotherapy. The effects of a psychological intervention can only be due to psychological factors. Control conditions consisting of alternative psychological treatments (including those erroneously

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html9 of 1312/1/2006 2:00 PMcalled placebos) are useful in establishing which psychological factors are important in producing the effects of a particular therapy but not in evaluating the efficacy of that treatment. Like placebos, effective psychotherapies may accomplish their effects by changing expectations ( Kirsch, 1985 1990 1999, , ), but unlike placebos, they do so without deception.ReferencesAntonuccio, D. O., Burns, D. D., & Danton, W. G. (2002). Antidepressants: A triumph of marketing overscience? Prevention & Treatment 5, , Article 25. Available on the World Wide Web:http://www.journals.apa.org/prevention/volume5/pre525c.html.Antonuccio, D. O., Danton, W. G., DeNelsky, G. Y., Greenberg, R. P., & Gordon, J. S. (1999). Raisingquestions about antidepressants. Psychotherapy and Psychosomatics 68, , 3-14. Babyak, M., Blumenthal, J. A., Herman, S., Khatri, P., Doraiswamy, M., Moore, K., Craighead, E.,Baldewicz, T., & Krishnan, K. R. (2000). Exercise treatment for major depression: Maintenance oftherapeutic benefit at 10 months. Psychosomatic Medicine 62, , 633-638. Brown, W. A. (2002). Are antidepressants as ineffective as they look? Prevention & Treatment, 5http://www.journals.apa.org/prevention/volume5/pre526c.html .Burns, D. D. (1999). Feeling good: The new mood therapy (Rev. Ed.). New York: Avon.Campbell, D. T., & Kenny, D. A. (1999). Primer on regression artifacts. New York: Guilford Press.DeRubeis, R. J., Gelfand, L. A., Tang, T. Z., & Simons, A. D. (1999). Medications versus cognitive behaviortherapy for severely depressed outpatients: Meta-analysis of four randomized comparisons. American Journal of Psychiatry 156, , 1007-1013. Elias, M. (2002, July 7). Study: Antidepressant barely better than placebo. USA Today, Retrieved July 8,2002 from http://usatoday.com/news/healthscience/health/drugs/2002-07-08-antidepressants.htmFreemantle, N., Anderson, I. M., & Young, P. (2000). Predictive value of pharmacological activity for therelative efficacy of antidepressant drugs: Meta-regression analysis. British Journal of Psychiatry 177, , 292-302. Greenberg, R. P. (2002). Reflections on the emperor's new drugs. Prevention & Treatment 5, , Article 27.AvailableontheWorldWideWeb:http://www.journals.apa.org/prevention/volume5/pre527c.html.Greenberg, R. P., & Fisher, S. (1989). Examining antidepressant effectiveness: Findings, ambiguities, andsome vexing puzzles. In S. Fisher & R. P. Greenberg (Eds.), The limits of biological treatments for psychological distress: Comparisons with psychotherapy and placebo(pp. 1 37). Hillsdale, NJ: Erlbaum.Hollon, S. D., DeRubeis, R. J., Shelton, R. C., & Weiss, B. (2002). The emperor's new drugs: Effect sizeand moderation effects. Prevention & Treatment 5, , Article 28. Available on the World Wide Web:http://www.journals.apa.org/prevention/volume5/pre528c.html.Hollon, S. D., Shelton, R. C., & Loosen, P. T. (1991). Cognitive therapy and pharmacotherapy fordepression. Journal of Consulting & Clinical Psychology 59, , 88-99. Hypericum Depression Trial Study Group. (2002). Effect of Hypericum Perforatum (St John's wort) in major depressive disorder: A randomized controlled trial. Journal of the American Medical Association 287, , 1807-1814. Jacobson, N. S., Roberts, L. J., Berns, S. B., & McGlinchey, J. B. (1999). Methods defining and determiningthe clinical significance of treatment effects: Description, application, and alternatives. Journal of Consulting & Clinical Psychology 67, , 300-307.

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html10 of 1312/1/2006 2:00 PMJamison, C., & Scogin, F. (1995). Outcome of cognitive bibliotherapy with depressed adults. Journal of Consulting & Clinical Psychology 63, , 644-650. Khan, A., Warner, H. A., & Brown, W. A. (2000). Symptom reduction and suicide risk in patients treated withplacebo in antidepressant clinical trials: An analysis of the Food and Drug Administration database. Archives of General Psychiatry 57, , 311-317. Kirsch, I. (1978). The placebo effect and the cognitive-behavioral revolution. Cognitive Therapy and Research 2, , 255-264. Kirsch, I. (1985). Response expectancy as a determinant of experience and behavior. American Psychologist 40, , 1189-1202. Kirsch, I. (1990). Changing expectations: A key to effective psychotherapy. Pacific Grove, CA: Brooks/Cole.Kirsch, I. (1997). Specifying nonspecifics: Psychological mechanisms of placebo effects. In A. Harrington (Ed.), The placebo effect: An interdisciplinary exploration(pp. 166-186). Cambridge, MA: Harvard University Press.Kirsch, I. (1999). How expectancies shape experience. Washington, DC: American Psychological Association.Kirsch, I., & Sapirstein, G. (1998). Listening to Prozac but hearing placebo: A meta analysis ofantidepressant medication. Prevention & Treatment 1, , Article 0002a. Available on the World Wide Web:http://www.journals.apa.org/prevention/volume1/pre0010002a.htmlLeber, P. (1998, May 4). Approvable action on Forrest Laboratories, Inc. NDA 20-822 Celexa (citalopram HBr) for the management of depression. Memoradum to the Department of Health and Human Services, Public Health Service, Food and Drug Administration, Center for Drug Evaluation and Research. Washington, DC.Leuchter, A. F., Cook, I. A., Witte, E. A., Morgan, M., & Abrams, M. (2002). Changes in brain function ofdepressed subjects during treatment with placebo. American Journal of Psychiatry 159, , 122-129. Mayberg, H. S., Liotti, M., Brannan, S. K., McGinnis, S., Mahurin, R. K., Jerabek, P. A., Silva, A., Tekell, J.L., Martin, C. C., Lancaster, J. L., & Fox, P. T. (1999). Reciprocal limbic-cortical function and negativemood: Converging PET findings in depression and normal sadness. American Journal of Psychiatry, 156, 675-682. Moerman, D. E. (2002). \"The Loaves and the Fishes\": A Comment on \"The Emperor's New Drugs: An Analysis of Antidepressant Medication Data Submitted to the U.S. Food and Drug Administration.\" Prevention & Treatment 5, , Article 29. Available on the World Wide Web:http://www.journals.apa.org/prevention/volume5/pre529c.html.Moncrieff, J. (2002). The antidepressant debate. British Journal of Psychiatry 180, , 193-194. Moncrieff, J., Wessely, S., & Hardy, R. (2001). Antidepressants using active placebos [Cochrane Review].Cochrane Database Systematic Review, 2 CD003012.Mulrow, C. D., Williams, J. W. , Jr., Trivedi, M., Chiquette, E., Aguilar, C., Cornell, J. E., Badgett, R., Noel,P. H., Lawrence, V., Lee, S., Luther, M., Ramirez, G., Richardson, W. S., & Stamm, K. (1999). Treatmentof depression: Newer pharmacotherapies. Evidence Report/Technology Assessment No. 7 (AHCPRPublication No. 99-E014; prepared by the San Antonio Evidence-based Practice Center based at the University of Texas Health Science Center at San Antonio under Contract 290-97-0012). Rockville, MD: Agency for Health Care Policy and Research.Muñoz, R. (2002). Comment on Kirsch,Moore,Scoboria,and Nicholls (2002). Prevention & Treatment 5, ,

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html11 of 1312/1/2006 2:00 PMArticle 30. Available on the World Wide Web:http://www.journals.apa.org/prevention/volume5/pre530c.html.Rabkin, J. G., Markowitz, J. S., Stewart, J. W., McGrath, P. J., Harrison, W., Quitkin, F. M., & Klein, D. F.(1986). How blind is blind? Assessment of patient and doctor medication guesses in a placebo-controlledtrial of imipramine and phenelzine. Psychiatry Research 19, , 75-86. Rehm, L. P. (2002). How can we better disentangle placebo and drug effects? Prevention & Treatment 5, , Article 31. Available on the World Wide Web:http://www.journals.apa.org/prevention/volume5/pre531c.htmlSalamone, J. D. (2002). Antidepressants and placebos: Conceptual problems and research strategies. Prevention & Treatment 5, , Article 24. Available on the World Wide Web:http://www.journals.apa.org/prevention/volume5/pre524c.htmlShea, M. T., Elkin, I., Imber, S. D., Sotsky, S. M., Watkins, J. T., Collins, J. F., Pilkonis, P. A., Beckham, E.,Glass, D. R., Dolan, R. T., & Parloff, M. B. (1992). Course of depressive symptoms over follow-up:Findings from the National Institute of Mental Health Treatment of Depression Collaborative Research Program. Archives of General Psychiatry 49, , 782-787. Smith, N. M., Floyd, M. R., Jamison, C., & Scogin, F. (1997). Three-year follow-up of bibliotherapy fordepression. Journal of Consulting & Clinical Psychology 65, , 324-327. Thase, M. E. (1999). How should efficacy be evaluated in randomized clinical trials of treatments for depression? Journal of Clinical Psychiatry 60, (Suppl. 4). , 23-31. Thase, M. E. (2002). Antidepressant effects: The suit may be small, but the fabric is real. Prevention & Treatment 5, , Article 32. Available on the World Wide Web:http://www.journals.apa.org/prevention/volume5/pre532c.html.Walach, H., & Maidhof, C. (1999). Is the placebo effect dependent on time? A meta-analysis. In I. Kirsch(Ed.), How expectancies shape experience (pp. 321-332). Washington, DC: American Psychological Association.Walsh, B. T., Seidman, S. N., Sysko, R., & Gould,, M. (2002). Placebo response in studies of majordepression. Variable, substantial, and growing. Journal of the American Medical Association 287, , 1840-1847. Wolf, S., Doering, C. R., Clark, M. L., & Hagans, J. A. (1957). Chance distribution and the placebo \"reactor.\"Journal of Laboratory and Clinical Medicine 49, , 837-841. 1An internal memorandum by the Director of the Division of Neuropharmacological Drug Products indicates FDA awareness of this situation:The Clinical Efficacy Trials subsection within the Clinical Pharmacology section not only describes the clinicaltrials providing evidence of citalopram's antidepressant effects, but make mention of adequate and well controlled clinical studies that failed to do so. I am mindful, based on prior discussions of the issue, that the Office Director is inclined toward the view that the provision of such information is of no practical value to either the patient or prescriber. I disagree. I believe it is useful for the prescriber, patient, and 3rd-party payer to know, without having to gain access to official FDA review documents, that citalopram's antidepressants (sic) effects were not detected in every controlled clinical trial intended to demonstrate those effects. I am aware that clinical studies often fail to document the efficacy of effective drugs, but I doubt that the public, or even the majority of the medical community, is aware of this fact. I am persuaded that they not only have a right to know but that they should know. Moreover, I believe that labeling that selectively describes positive studies and excludes mention of negative ones can be viewed as potentially \"false and misleading\" (Leber, 1998, p. 11).

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html12 of 1312/1/2006 2:00 PMWe agree that the public and the medical community should be informed of these data.2The drug effect is conventionally interpreted as the difference between the response to the drug and the response to placebo. It is that part of the drug response that is due to the pharmacological action of the drug. In contrast, the drug response includes the effect of the drug, the placebo effect, spontaneous remission, regression to the mean, and any other factors that might contribute to changes observed following the administration of medication. Similarly, the placebo effect is that portion of the placebo response that is actually due to the administration of a placebo.3Regression artifacts also affect the correlation between drug response and placebo response. Moerman reviewed priormeta-analyses showing that these correlations range between .43 and .90. Similarly, if the analysis is limited to the morereliable clinical trials in the FDA data set?those with samples of 200 or greater?the correlation between drug responseand placebo response in the FDA data set is .72. However, because baseline scores are necessarily correlated withchange scores (Campbell & Kenny, 1999), when the baseline scores of two groups are correlated (as they are inmeta-analyses of drug and placebo responses), the change scores will also be correlated. For example, when normallydistributed baseline scores in two samples are perfectly correlated with each other, and normally distributed posttestscores in these samples are uncorrelated either with each other or with baseline scores, the correlation between the twosets of change scores will be .50.4The way in which this method of describing the data can be misleading can be illustrated by applying it to the Khan, Warner, and Brown (2000) analysis of the data submitted to the FDA. These data indicate that there were twice as many suicide attempts among patients given active medication than among those given placebo. This might be interpreted as indicating that medication doubled the risk of suicide attempts, but this method of describing the data is misleading. The actual rates were 0.8% with antidepressant medication and 0.4% with placebo. Thus, suicide attempts were absent in 99.2% of patients treated with drugs and 99.6% of patients treated with placebo, and as noted by Khan et al. (2000), the difference was not statistically significant.5Contrary to the implication in Muñoz's (2002) commentary, this hypothesis does not suggest duplicity on the part of patients or researchers.Correspondence concerning this article should be addressed to Irving Kirsch, Department of Psychology, University of Connecticut 406 Babbidge Road, U-20, Storrs, CT 06269-1020E-mail: [email protected] Figure 1. Drug/placebo differences as a function of sample size.

PsycARTICLES - Antidepressants and Placebos: Secrets, Revelations, and Una...http://content.apa.org/journals/pre/5/1/33r.html13 of 1312/1/2006 2:00 PM Figure 2. Regression toward the mean in response to drug and placebo.

“Efficacy of antidepressants in adults”BMJJuly 2005(VIEW FULL DOCUMENT).

Education and debateEfficacy of antidepressants in adultsJoanna Moncrieff, Irving KirschMost people with depression are initially treated with antidepressants. But how well do the datasupport their use, and should we reconsider our strategy?The National Institute for Health and ClinicalExcellence (NICE) recently recommended that anti-depressants, in particular selective serotonin reuptakeinhibitors, should be first line treatment for moderateor severe depression. This conclusion has broadly1been accepted as valid. The message is essentially the2same as that of the Defeat Depression Campaign in theearly 1990s, which probably contributed to the 253%rise in antidepressant prescribing in 10 years. From1our involvement in commenting on the evidence basefor the guideline we believe these recommendationsignore NICE data. The continuing concern that selec-tive serotonin reuptake inhibitors may increase the riskof suicidal behaviourw1w2means there needs to befurther consideration of evidence for the efficacy ofantidepressants in adults as there has been in children.EfficacyAlthough the NICE meta-analysis of placebo control-led trials of selective serotonin reuptake inhibitorsfound significant differences in levels of symptoms,these were so small that the effects were deemedunlikely to be clinically important. The conclusion that1the drugs had clinically important benefits was basedon analysis of response and remission rates. However,in our comments on the draft guidelines, we pointedout that these categorical outcomes were derived fromthe same continuous data for symptoms scores thatwere found to show no clinically relevant effects. AsNICE notes, “dichotomising scores into remission andnon-remission creates an artificial boundary, withpatients just over the cut-off score often being clinicallyindistinguishable from those just under the cut-off.”1The hypothetical data in the figure show how smalldifferences may be magnified by transformation ofcontinuous data into categorical data.3In thisexample, response was defined as a minimum 12point improvement on the Hamilton rating scale fordepression. Difference in mean change of scoresbetween drug and placebo groups was 1 point. Thisscenario yields response rates of 50% in the drugcondition and 32% in the placebo condition. Thus, ifimprovement is normally distributed and the criterionfor response is close to the mean improvement rate(which it generally is), a very small difference insymptom score can push a large proportion ofpatients into different categories.The small effects found on continuous measuresare consistent with results of other recent meta-analyses of symptom scores. Khan et al found a 10%difference in levels of symptoms in two meta-analyses,45and Kirsch et al included unpublished stud-ies in their latest analysis and found an overall meandifference of 1.7 points on the Hamilton scale. No6research evidence or consensus is available about whatconstitutes a clinically meaningful difference inHamilton scores, but it seems unlikely that a differenceof less than 2 points could be considered meaningful.NICE required a difference of at least 3 points as thecriterion for clinical importance but gave no justifica-tion for this figure. The most commonly used 17 item1version of the Hamilton scale has a maximum score of52 and contains seven items concerning sleep andanxiety, with each item on sleep scoring up to 6 points.Hence any drug with some sedative properties, includ-ing many antidepressants, could produce a differenceof 2 points or more without exerting any specific anti-depressant effect. Other recent meta-analyses thatpresent categorical outcomes also find modestdifferences of between 14% and 18% in improvementor response rates.w3-w5References w1-w20 are on bmj.comChange in Hamilton scoreFrequency (%)05101501015205DrugPlaceboNormal distribution of scores for Hamilton rating scale fordepression with mean score of 11.5 for antidepressant and 10.5 forplaceboDepartment ofMental HealthSciences, UniversityCollege London,London W1N 8AAJoanna Moncrieffsenior lecturer insocial and communitypsychiatrySchool of Healthand Social work,University ofPlymouth,PlymouthIrving Kirschprofessor of psychologyCorrespondence to:J [email protected];331:155–9155BMJVOLUME 33116 JULY 2005bmj.com on 15 July 2005 bmj.comDownloaded from

Severity of depressionA key claim in the NICE guideline is that the superior-ity of antidepressants over placebo correlates posi-tively with the severity of depression being treated.This belief is an old one. In 1958 Kuhn suggested thatendogenous depression was more responsive toantidepressants than neurotic or reactive depression,which was generally regarded as less severe.7Regression to the mean may account for this impres-sion since it entails that people with more severedepression at baseline will show greatest overalllevels of improvement. But it does not explaindrug-placebo differences, because greater improve-ment among patients with more severe depressionoccurs regardless of whether they are treated with adrug or placebo.An early review of controlled trials found thatevidenceaboutwhetherendogenoussymptomspredicted response was inconsistent. Recent evidence8comes from post-hoc analysis in trials with otherwisenegative resultsw6 w7and from meta-analyses. The meta-analysis by Angst et al is often cited in support of theseverity hypothesis, but severity effects were weak andmostly non-significant.9Effects in another meta-analysis were more impressive, but data were providedonly for investigational antidepressants and not estab-lished ones, where the evidence seemed to be weaker.10In contrast, another recent meta-analysis found norelation between severity and antidepressant effect,11and a meta-analysis of older studies showed that differ-ences between antidepressants and placebo weresmaller and non-significant in inpatient trials com-pared with outpatient trials. The NICE meta-analysis12failed to find a consistent gradient of effect from“moderate” (Hamilton score 14-18) through “severe”(19-22) to “very severe” depression ( 23). In fact, the≥1middle group, which would generally be referred to asmoderately depressed, tended to show larger effectsthan either of the other two, but numbers of studieswere small.Thus there seems to be little support for thesuggestion that recent failure to find markeddifferences between antidepressants and placebo isdue to recruitment of patients with mild depressionthat is less responsive to antidepressants. Indeed, in1the meta-analysis by Kirsch et al, all but one of the trialswere conducted in patients with severe to very severedepression according to NICE criteria. The possibility6that patients in the mid-range of severity show agreater antidepressant response, as suggested by theNICE data and by Joyce and Paykel, would not be8expected from a simple biological effect. It mayindicate that this group is more susceptible to somemethodological artefact such as infringement of thedouble blind (see below).Methodological issues in antidepressanttrialsSeveral commentators have suggested that the smalleffects of antidepressants compared with placebos maybe attributable to methodological factors or selectivepresentation of data from antidepressant trials.w8-w10These include concerns that trials of antidepressantsmay not be truly double blind. This is becauseparticipants may be able to detect differences betweenplacebos and drugs because the drugs cause noticeablephysiological effects including, but not limited to,recognised side effects. Other concerns include thevalidity of outcome measures, that discontinuationeffects may confound continuation trials, and thatresults may be inflated by exclusion of people whowithdraw early from the analysis. Evidence also showsthat trials of antidepressants with negative results areless likely to be published than those with positiveresults and that, within published trials, negativeoutcomes may not be presented.13A neglected aspect of antidepressant trials is thesubstantial heterogeneity of their findings. Although12many trials do find antidepressants are superior toplacebo, many do not, including some of the largestand most well known landmark trials such as theMedical Research Council trial and the early NationalInstitute for Mental Health trial.w11w12In addition,many trials find that substances as diverse asmethylphenidate, benzodiazepines, and antipsychoticsCampaigns raising awareness of depression have contributed toincreased prescribing of antidepressantsEducation and debate156BMJVOLUME 331 16 JULY 2005bmj.com on 15 July 2005 bmj.comDownloaded from

can have antidepressant effects, suggesting that theseeffects may be attributable to non-specific pharmaco-logical or psychological mechanisms.w10Effect of antidepressantsLongitudinal follow-up studies show very pooroutcomes for people treated for depression both inhospital14and in the community,15and the overallprevalence of depression is rising despite increased useof antidepressants.16Two studies that prospectivelyassessed outcome in depressed patients treatednaturalistically by general practitioners and psychia-trists found that people prescribed antidepressants hada slightly worse outcome than those not prescribedthem, even after baseline severity had been taken intoaccount.17 18No comparable studies could be foundthat showed a better outcome in people prescribedantidepressants.Some authors have suggested a causal associationbetween increased antidepressant prescribing since1990 and reduction of overall suicide rates observed insome countries.w13 w14However, others have pointed outthat falls in overall suicide rates started long before thisperiod,w15-w17and suicide rates have increased in someage groupsw15and some countriesw18despite increasedantidepressant prescribing. Meta-analyses of data fromcontrolled trials have not found reduced rates ofsuicide or suicidal behaviour in drug arms comparedwith placebo arms.4 5 w19 w20ConclusionsThe NICE review data suggest that selective serotoninreuptake inhibitors do not have a clinically meaning-ful advantage over placebo, which is consistent withother recent meta-analyses. In addition, methodologi-cal artefacts may account for the small effect seen. Evi-dence that antidepressants are more effective in moresevere conditions is not strong, and data on long termoutcome of depression and suicide do not provideconvincing evidence of benefit. In children, thebalance of benefits to risks is now recognised asunfavourable. We suggest this may also be the case foradults, given the continuing uncertainty about thepossible risk of increased suicidality as well as otherknown adverse effects. This conclusion implies theneed for a thorough re-evaluation of currentapproaches to depression and further development ofalternatives to drug treatment. Since antidepressantshave become society’s main response to distress,expectations raised by decades of their use will alsoneed to be addressed.We thank other members of the Critical Psychiatry Networkwho contributed to the response to the NICE depression reviewand especially Duncan Double, who coordinated the response.Contributors and sources: Both authors have conductedseparate meta-analyses of antidepressant trials and reviews ofantidepressant literature. JM has recently obtained an MD inantidepressant research methodology. The article draws onthese sources, as well as the data contained in the NICE review.JM and IK contributed to the response to the NICE review. JMhad the idea to write the paper. JM and IK drafted and revisedthe current manuscript. JM will act as guarantor.Competing interests: IK has received consulting fees fromSquibb and Pfizer. JM is co-chair of the Critical PsychiatryNetwork.1National Institute for Clinical Excellence.Depression:management of depres-sion in primary and secondary care.Clinical practice guideline No 23. London:NICE, 2004. www.nice.org.uk/page.aspx?o = 235213 (accessed 24 May2005).2Middleton H, Shaw I, Feder G. NICE guidelines for the management ofdepression.BMJ2005;330:267-8.3Kirsch I. St John’s wort, conventional medication and placebo: anegregious double standard.Complement Ther Med2003;11:193-5.4Khan A, Warner HA, Brown WA. Symptom reduction and suicide risk inpatients treated with placebo in antidepressant clinical trials.Arch GenPsychiatry2000;57:311-24.5Khan A, Khan SR, Leventhal RM, Brown WA. Symptom reduction andsuicide risk in patients treated with placebo antidepressant clinical trials:a replication analysis of the Food and Administration Database.Int JNeuropsychopharmacol2001;2:113-8.6Kirsch I, Moore TJ, Scoboria A, Nicholls SS. The emperor’s new drugs:an analysis of antidepressant medication data submitted to the USFood and Drug Administration.Prev Treat2002;5.www.journals.apa.org/prevention/volume5/pre0050023a.html (accessed 20 May 2005).7Kuhn R. The treatment of depressive states with G22355 (imipraminehydrochloride).Am J Psychiatry1958;115:459-64.8Joyce PR, Paykel ES. Predictors of drug response in depression.Arch GenPsychiatry1989;46:89-99.9Angst J, Scheidegger P, Stabl M. Efficacy of moclobemide in differentpatient groups: results of new subscales of the Hamilton Rating Scale.Clin Neuropharmacol1993;16 (suppl 2):S55-62.10 Khan A, Leventhal RM, Khan SR, Brown WA. Severity of depression andresponse to antidepressants and placebo: An analysis of the Food andDrug Administration database.J Clin Psychopharmacol2002;22:40-5.11 Kirsch I, Scoboria A, Moore TJ. Antidepressants and placebos: secrets,revelations, and unanswered questions.Prev Treat2002;5. www.journals.apa.org/prevention/volume5/pre0050033r.html(accessed20May2005).12 Moncrieff J. A comparison of antidepressant trials using active and inertplacebos.Int J Methods Psychiatric Res2003;12:117-27.13 Melander H, Ahlqvist-Rastad J, Meijer G, Beermann B. Evidence b(i)asedmedicine selective reporting from studies sponsored by pharmaceutical—industry: review of studies in new drug applications.BMJ2003;326:1171-3.14 Tuma TA. Outcome of hospital treated depression at 4.5 years. An elderlyand a younger cohort compared. Br J Psychiatry2000;176:224-8.15 Goldberg D, Privett M, Ustun B, Simon G, Linden M. The effects of detec-tion and treatment on the outcome of major depression in primary care:a naturalistic study in 15 cities.Br J Gen Pract1998;48:1840-4.16 Fombonne E. Increased rates of depression: update of epidemiologicalfindings and analytical problems.Acta Psychiatr Scand1994;90:145-56.17 Brugha TS, Bebbington P, MacCarthy B, Stuart E, Wykes T.Antidepressants may not assist recovery in practice: a naturalisticprospective survey.Acta Psychiatr Scand1992;86:5-11.18 Ronalds C, Creed F, Stone K, Webb S, Tomenson B. The outcome ofanxiety and depressive disorders in general practice.Br J Psychiatry1997;171:427-33.(Accepted 11 May 2005)Summary pointsRecent meta-analyses show selective serotoninreuptake inhibitors have no clinically meaningfuladvantage over placeboClaims that antidepressants are more effective inmore severe conditions have little evidence tosupport themMethodological artefacts may account for thesmall degree of superiority shown over placeboAntidepressants have not been convincinglyshown to affect the long term outcome ofdepression or suicide ratesGiven doubt about their benefits and concernabout their risks, current recommendations forprescribing antidepressants should bereconsideredEducation and debate157BMJVOLUME 33116 JULY 2005bmj.com on 15 July 2005 bmj.comDownloaded from

FDA approval of these drugs implies that the data were strong enough and reliable enough to warrant approval, however, as one FDA memorandum written by Dr. Paul Leber illustrates, the FDA’s standards for approving antidepressants as effective are not robust “Approval [of the antidepressant] may … come under attack by constituencies that do not believe the agency is as demanding as it ought to be in regard to its standards for establishing the efficacy on antidepressant drug products.” (​VIEW FULL DOCUMENT​)

MemorandumDepartmentofHealthandHumanServicesPublicHealthServiceFoodandDrugAdministrationCenterforDrugEvaluationandResearchDATE:December24,1991FROM:PAULLEBER,M.D.DIRECTOR,DIVISIONOFNEUROPHARI4ACOLOQICALDRUGPRODUCTSSUBJECT:RecommendationtoApproveNDA19439(Zoloft;tertrallne)TO:RobertTemple,U.D...-.;.-•:-..••..-.-Director,.OO.B.I....-.....•>.•.-./.••.:.-..:{.-,. , . / . v .:>\".••..•.A . v \"FileNOA19-*39''':ThismemorandumconveysmyformalrecommendationthatPilzer'sNDA19-839forZOLOFT(sertraJine)beapproved.''TheadministrativefiledocumentsthatNDAreviewteamunderthedirectionofDr.ThomasLaughrenhascompleteditsreviewofFiber'sNDAforsertraline,Including,inparticular,thefirm'sresponsetotheagency's10/30/91approvableactionletter.Thedivision'sreviewhasfailedtofindthatanyofthegroundsenumeratedinSection505(d)of theFederal Food,Drug,andCosmeticActforthedisapprovalofanNDAapply,andconsequently,werecommendthattheNDAboapprovedasprovidedbySection505(c)(1)(A)oftheActItisnoteworthythatseveralforeignnationaldrugregulatoryauthorities(seeDr.Laugnren's12/10/91memorandum),presumablyprovidedwiththesamebodyofInformationthatIscontainedIn theNDAsubmittedtous,havenotyetbeenwillingtoallowsertraJtne'smarketingintheirrespectivecountries.Itisourunderstandingthattheseregulatoryauthoritiesarenotconcernedaboutthe risksofsartrallnoforuse,butbywhatmaybeconsideredtheMackofrobustness*ofthecfinlca!evidencesupportingItsefficacyinthetreatmentofdepression.ThisturnofeventsmayseemsomewhatsurprisingInviewofthefactthattheagencyistraditionallymoreconservativethanitsEuropeancounterparts.Obviously,changesareunderwaythroughoutthewesternEurope,perhapsinresponsetotheEEC'sharmonizationInitiatives.Inanycase,withtheMotus/PfizerDocs046941

LebenZOLOFTApproval Action [12/24/91]page2of3ImportantexceptionoftheUK'sCSM/MCA,standardsforantfdepressantdrugproductapprovalseemtobebecomingmoredemandinginregardto1)thedurationofcontrolledtrialsservingassourcesofevidenceofefficacy,2)theneedtodocumentefficacyInhospitalizeddepressedpatients(becausethesearepresumed,arguably,tobemoreseverelydepressed),3)theneedtoshowefficacyinmaintainingremission,4)theneedtoshowefficacyinpreventingrelapseofeuthymicpatientswithahistoryofrecurrentepisodesofaffectiveillness,and5)aneedtoestablishequivalencyand/orsuperiorityofanewantidepressanttoalreadymarketeddrugproducts.ManyoftheseforeignregulatoryInitiativeshavepotentialmerit,but,given..the-; perceivedairgericy.we. expressas. an .-.Institutionfo<..exped!tlng.,ihe...-.....-:public's'accesslDnew,potemialfy promising;-drags,idonotbelievewe.can'successfullyintroducesimilar,moredemanding,requirementsdomestically,atleastuntil rttmreisasignificant'seachange'Inoursociety'scollectiveattitudetowanfc Federalregulationofnewdrugapprovals.Incidentally,ifyoudisagreewithmyassessment,Iwoulddketoknowbecausethedivisionwouldcertainlybewillingtoproposeadditionalrequirements,provided,ofcourse,thatwecouldbeassuredofsupportfortheinitiativefromboththe, OfficeandtheCenter.inanycase,baseduponourcurrentinterpretationoftheAct s1requirements,Pfizer'sNOAforsertrailnemustbeapproved.Sertraiineissafeforuse,effectiveinuse,andadequatelylabeled,aviewconfirmed,albeitnotunanimously,bythevoteofourpublicAdvisoryCommittee(i.e.,PQAC).Furthermore,althoughsertrailnemaynotbethemostrobustlypowerfulantidepressantdrugproducteverintroduced(apointmadeforcefullybysomeofourjBdvisors),ithassomepotentialadvantages.Asa'pure1serotonergicauptakeinhibitoritshares.withProzac(fluoxetine),theonlycurrentlymaketeddrugofthis, type,freedomfromthetroublingsideeffectsassociatedwiththeuseofthedassictricycfleantidepressants(e.g.,imipramine,desfpramine,amitriptyllne,etc.)andthedietaryrestrictionsnecessarily-associatedwiththeuseofmonamlneoxidaseinhibitors.Sertrallneand fluoxHine havenotundergoneheadtoheadcomparativeclinicaltesting,butsertrallne,and .Itsmajormetabolite,nor-sertrailne,havesomewhatshorterelimination\" half-livesrespectivelythanfluoxltineandnor-fiuoxitine.apotentialadvantage..Insum,theapprovalofSertralineisreadilyjustifiedunderexistingrulesMotus/PfizerDocs046 942

Leben20L0FTApproval jetton[12/24*1]page of 33•andregulations.Approvalmay,however,forthereasonsenumeratedabove,comeunderattackbyconstituenciesthatdonotbelievetheagencyIsasdemandingasit oughttobeinregardtoitsstandardsforestablishingtheefficacyofantldepressantdrugproducts.Paul Leber,M.O.O A , ..;.HFD-10b:Temple'••HFD-120:Katz,Laughren,Lee,Knudsen,DavidMotus/Pfizer046943

No Scientifically Reliable Evidence that Declining Suicide Rates are the Result of Increased Prescriptions of Antidepressants On the issue of national suicide rates going down and, in particular, on the possible impact of antidepressants on these rates, as one renowned expert has noted: “This argument is like saying that, because there has been an increase in storks seen recently and a coincidental increase in births, babies are therefore brought by storks.” (Declaration of Dr. David Healy, ​Tucker v. GSK​.) In fact, according to Gunnel et al. , “Antidepressants and suicide: ​1​what is the balance of benefit and harm,” British Medical Journal (BMJ), 2004; 329:34-38 (3 July): Surprisingly, direct evidence that antidepressants prevent suicide is hard to find. … In the most comprehensive synthesis of data from randomised trials, Khan and colleagues found no evidence of a beneficial effect of antidepressants on suicide. Gunnell, citing Khan A, Khan S, Kolts R, Brown WA. “Suicide rates in clinical trials of SSRIs, other antidepressants, and placebo: analysis of FDA reports,” Am J Psychiatry 2003;160: 790-2 The authors also pointed out that “Suicide is rare, even among people with depression. [Cite omitted.] Thus, most clinical trials have insufficient power to provide clear evidence on the effect of antidepressants on suicide.” Gunnell ​citing ​Jick SS, Dean AD, Jick H. Antidepressants and suicide. BMJ 1995;310: 215-8. According to a study by Herman Van Praag published recently in World Journal of Biological Psychiatry ​titled “A Stubborn Behaviour: the Failure of Antidepressants to Reduce Suicide Rates,” despite the increased use of antidepressants “completed suicide has remained quite stable” and “suicide attempts even seem[] to have increased.” 1 ​https://www.baumhedlundlaw.com/our-successes/advocacy-campaigns/antidepressant-trials-adult-suicidality-data/#ftn_1