Symptom Reduction and Suicide Risk in Patients Treated With Placebo in Antidepressant Clinical Trials: Title and subTitle BreakAn Analysis of the Food and Drug Administration Database

During the past half century, the double-blind, randomized, placebo-controlled clinical trial has become the gold standard for treatment evaluation. In the United States, placebo controls are widely regarded by many scientific investigators and the Food and Drug Administration (FDA) as necessary to establish the efficacy of a putative treatment. Nonetheless, in recent years, the scientific and ethical grounds for placebo controls are being increasingly questioned by the scientific community, by the federal government, and most vociferously by the popular press. The evaluation of new antidepressants brings these issues into focus. Depressive illness is common and, in part for that reason, the past 30 years have seen an unabated flow of new antidepressants.

In a recent report, Rothman and Michels¹ argue that it is unethical for clinical trials to include a placebo control when, as in the case of depression, an effective treatment is available. Others² question the scientific necessity and value of placebo controls.

During the past 30 years, placebo-controlled antidepressant clinical trials have consistently shown placebo response rates of 30% to 50%, drug response rates of 45% to 70%, and drug-placebo differences of 18% to 25%.^{3 - 4} Nonetheless, the extent to which depressed patients who are deprived of an available or an experimental antidepressant are exposed to excess morbidity is a matter of controversy. This controversy is not merely academic. Institutional review boards in the United States regularly struggle with the ethics of assigning depressed patients to placebo, and ethics committees in most European countries simply do not allow placebo controls in antidepressant trials.

Much of the debate over the potential harmfulness of assigning depressed patients in clinical trials to placebo treatment is fueled by misconceptions about the dangers of mental illness and generalizations about severely depressed patients seen in clinical settings being applied to subjects of clinical trials. The debate does not seem informed by what actually happens to depressed patients in clinical trials who are assigned to placebo.

To address this information gap, we examined the FDA database for antidepressants approved from January 1, 1987, through December 31, 1997. We hypothesized that rates of suicide and suicide attempts among patients assigned to placebo treatment would not differ from those among patients assigned to antidepressant treatment. We also hypothesized that drug-placebo differences derived from this database would be smaller than those derived from published studies.

Little is known about the rates of suicide and suicide attempts in patients participating in antidepressant clinical trials. The rate of these events is low, so single clinical trials that typically involve several hundred patients or fewer do not have a sufficient number of subjects to estimate the rates. Meta-analyses of clinical trials, which are based on published studies, do not include data on these events.

Several meta-analyses^{5 - 8} of clinical trials have focused on comparative symptomatic outcomes for depressed patients who are assigned to antidepressant and placebo treatment. Although these meta-analyses generally conclude that antidepressants offer an advantage over placebo, their conclusions differ regarding the magnitude of the drug-placebo difference and its clinical implications. Some of the inconsistencies among these meta-analyses can be accounted for by differences in the studies selected for inclusion and by methods of analysis. However, all of these meta-analyses are limited by the bias inherent in reliance on published studies, ie, positive findings are more likely to be reported than negative ones. Therefore, meta-analyses based on published studies have to contend with the "file-drawer effect," ie, the negative findings that are never reported.

The FDA database on investigational medications is uniquely suited for providing relatively unbiased data on the fate of depressed patients who are assigned to placebo. First, the FDA requires full disclosure of all data from all studies conducted worldwide for a drug under development. Second, the FDA determines whether an investigational drug is efficacious based on studies deemed well designed and well controlled (ie, pivotal) under Title 21 of the Code of Federal Regulations.⁹ For antidepressants, these pivotal studies by convention and practice are randomized, double blind, and placebo controlled, with established criteria for the sample under study and defined criteria for response. (The inclusion and exclusion criteria for the samples select a group of depressed patients not representative of those seen in clinical practice.) The data from these studies must be sufficient to assess effects according to specific methods of analysis. Typically, these are phase 2 and 3 studies with several hundred patients. Third, studies are designated as pivotal without regard to outcome; many pivotal studies do not demonstrate statistical superiority of antidepressants over placebo.

We used this entire database to determine the incidence of suicide and suicide attempts in patients participating in antidepressant clinical trials, and we analyzed the data from pivotal studies to assess comparative symptom reduction among patients who are assigned to antidepressant and placebo treatments.

We obtained FDA clinical trial data for all antidepressants approved in the United States during the 11-year study period. Specifically, under the Freedom of Information Act,¹⁰ we obtained public domain data on FDA-reviewed studies for fluoxetine hydrochloride (Prozac), sertraline hydrochloride (Zoloft), paroxetine hydrochloride (Paxil), venlafaxine hydrochloride (Effexor), nefazodone hydrochloride (Serzone), mirtazapine (Remeron), and bupropion hydrochloride (Wellbutrin SR), by a specific written request to the FDA offices, Rockville, Md. The data were sent by mail for a small fee on microfiche and consisted of more than 5200 pages of paper.

To assess safety, we reviewed all available clinical trial data on incidence of suicide attempts and suicides. These data were available for 5 of the 7 antidepressants. (Suicide and suicide attempt data for fluoxetine studies were not included in the FDA database, and venlafaxine studies combined the incidence of suicide and suicide attempts.) These data encompassed all worldwide clinical trial data obtained during the drugs' development (Table 1). The data were not limited to pivotal studies, but included nonpivotal studies (some examples were pharmacokinetic, dose tolerance, bioavailability, and open-label studies and preliminary studies in obesity, obsessive-compulsive disorder, and panic disorder). In addition to the frequency data, we were able to estimate the incidence of suicide and suicide attempts based on patient exposure years (PEY; ie, cumulative time that subjects were exposed to investigational antidepressant, active comparator, or placebo while in a research program). Testing with χ² analysis was applied to assess the statistical significance for frequency of suicide and suicide attempts in the placebo-treated compared with drug-treated subjects (including active comparator and investigational antidepressant).

To assess efficacy, we used the data from clinical trials reviewed by the FDA in support of a drug's indication and deemed pivotal studies. From the 7 antidepressants' research programs, the FDA considered 48 studies to be pivotal; of these, 3 were excluded from our analysis. Two studies were excluded because of insufficient data, such as mean total Hamilton Depression Rating¹¹ (HAM-D) scores; and 1 study, because it focused on relapse prevention rather than acute treatment response.

Of the 45 studies analyzed (Table 2), 23 were 2-armed: study-drug vs placebo studies; 22 were 3-armed: study-drug vs active comparator (imipramine, amitriptyline, or trazodone) vs placebo studies. Study duration was 4 weeks for 6 studies (study design for 5 paroxetine studies was indicated as 6 weeks, but end-point analysis was based on week 4), 5 weeks for 3 studies, 6 weeks for 29 studies, and 8 weeks for 7 studies.

Our original aims were to establish rates of therapeutic response, to evaluate the effect size between placebo and antidepressants, and to establish confidence intervals for effects of placebo. However, these public domain data do not lend themselves to such analyses; the only available symptom intensity scores were the mean total HAM-D scores at double-blind randomization (baseline) and the mean total HAM-D scores at the last observation carried forward (LOCF). Mean HAM-D scores were not available on a week-by-week basis, and HAM-D scores were not available for individual patients. Also, we were unable to analyze HAM-D scores exclusively for patients who completed the studies (observed cases). Accordingly, we used the LOCF technique to assess symptom reduction. In this technique, patients prematurely terminating from the trial are assumed to experience no further improvement; thus, the last measured HAM-D scores are considered the final (or end-of-trial) scores. We used the mean total scores on the intent-to-treat population, defined as those having at least 1 dose of double-blind medication and at least 1 efficacy rating during the double-blind phase.

A subgroup of studies (21/45) included the SEM or the SD for mean change in total HAM-D from baseline to LOCF, allowing us to determine effect size. For studies reporting SEM, we calculated the SD of change in LOCF (C-LOCF) scores for placebo- and antidepressant-treated patients. We calculated effect size by dividing the mean change difference in C-LOCF between drug- and placebo-treated patients by the SD of the applicable drug cell (investigational antidepressant or active comparator).

Each of the studies used the following standard subject inclusion and exclusion criteria: (1) 18 years of age or older; (2) diagnosis of major depression of moderate to severe degree, according to DSM-III,¹² DSM-III-R,¹³ or research diagnostic criteria¹⁴ ; (3) signed informed consent form (witnessed); (4) minimum score at screen evaluation of 18 or 20 on total HAM-D (HAM-17 and HAM-21, respectively); (5) 4 to 10 days of placebo run-in; (6) less than a 20% decrease in HAM-D score between screen and baseline; (7) not actively suicidal or posing a serious suicidal risk; and (9) no significant concurrent or previous other psychiatric illnesses.

Each study had the following design features: (1) primary efficacy measures included HAM-D; (2) mean total raw scores and change in scores on the HAM-D were available for calculations and comparisons; (3) depressed patients underwent evaluation at weekly intervals during the first month and at least biweekly after 4 weeks; and (4) the dose of the antidepressants was adequate based on standard practice.

Table 1 describes the incidence of suicides and suicide attempts for the 19,639 patients participating in clinical trials evaluating 5 of the 7 antidepressants. Thirty-four patients committed suicide, 27 while receiving the investigational antidepressant, 5 while receiving the active comparator, and 2 while receiving placebo. Using PEY, overall incidence of suicide was 757/100 000 per year (34/4491) among the 19 639 patients. Among patients receiving the active comparator, incidence of suicide was 686/100 000 per year (5/729); among patients receiving investigational antidepressants, 842/100 000 per year (27/3206); and among patients receiving placebo, 360/100 000 per year (2/556). The differences in suicide rates did not reach statistical significance among the 3 treatment groups (χ²₂=1.53; P=.46).

A total of 130 patients attempted suicide, suggesting an overall risk of 2895/100 000 per year. Among patients receiving the active comparator, incidence of suicide attempts was 3429/100 000 per year (25/729); among patients receiving investigational antidepressants, 2807/100 000 per year (90/3206); and among patients receiving placebo, 2698/100 000 per year (15/556). The differences in rates of suicide attempts did not reach statistical significance among the 3 treatment groups (χ²₂=0.90; P=.64). (In venlafaxine studies, the combined incidence of suicide and suicide attempts was 3 of the 2181 patients randomized to venlafaxine, 1 of the 591 patients randomized to active comparator, and 1 of the 451 patients randomized to placebo.)

Table 2 describes the 45 pivotal studies of the 7 antidepressants. A total of 8731 depressed patients participated in these studies; 4510 (51.7%) received the investigational antidepressant; 1416 (16.2%) received active comparator; and 2805 (32.1%) received placebo. Table 3, Table 4, and Table 5 delineate the mean baseline total HAM-D scores and mean change in total HAM-D at LOCF. Among the 4510 depressed patients receiving the investigational antidepressant, the mean decrease in total HAM-D score was 40.7%; among the 1416 depressed patients receiving active comparator, 41.7%; and among the 2805 depressed patients receiving placebo, 30.9%.

Regardless of treatment condition, total HAM-D scores showed greater decreases as study duration increased. Among depressed patients receiving the investigational antidepressant, the decreases in total HAM-D scores were 40.0% in 4-week trials, 40.5% in 5-week trials, 40.6% in 6-week trials, and 43.9% in 8-week trials. Similarly, among the depressed patients receiving active comparator, these decreases were 28.2% in 4-week trials, 44.0% in 6-week trials, and 48.1% in 8-week trials. Among the placebo-treated patients, these decreases were 24.7% in 4-week trials, 31.5% in 5-week trials, 30.5% in 6-week trials, and 36.1% in 8-week trials.

For 21 of the 45 studies, we could estimate effect size (Table 3, Table 4, and Table 5). Nearly half the studies had less than medium effect size (<0.5), ie, 9 (43%) of those comparing placebo and investigational antidepressants and 5 (45%) of those comparing placebo and active comparator.

The patient study completion rates (Table 3, Table 4, and Table 5) favored investigational antidepressants, although not by a large margin. In 42 studies with patient completion rates, 18 favored investigational antidepressant, 11 favored active comparator, and 15 favored placebo.

Our aim was to assess the fate of depressed patients who are assigned to placebo treatment in antidepressant clinical trials. Specifically, we assessed risks for suicide and suicide attempt and the magnitude of change in HAM-D scores for patients who are randomly assigned to placebo or antidepressant treatment.

As expected, incidence of suicide and suicide attempts was low for patients participating in antidepressant clinical trials. Suicide rate was approximately 757/100 000 per year. There are no other clinical trial data with which to compare our suicide risk findings. The annual US suicide rate in the general population is 11/100 000.¹⁵ Risk for suicide among depressed patients is estimated to vary from 275/100 000¹⁶ to 1352/100 000 per year.¹⁷ Although rates of suicide and attempted suicide were numerically lower in the placebo compared with active comparator– or investigational antidepressant–treated group (Table 1), the differences did not reach statistical significance.

Regarding efficacy, patients who are assigned to placebo treatment experienced substantial symptom reduction, although not of the magnitude experienced by those assigned to antidepressants. There was a positive relationship between duration of clinical trial and reduction in depression; ie, the longer the trial, the greater the decrease in depressive symptoms, regardless of treatment. Effect sizes between placebo and antidepressant were notably variable (Table 3, Table 4, and Table 5). Nearly half the studies showed only a modest effect for antidepressants (effect size, <0.5), whether they were investigational antidepressants or active comparators. Similar trends are reported by earlier authors.^{5 - 8} The less-than-impressive difference between drug and placebo in this and other studies of clinical trials does not speak directly to the effectiveness of antidepressants in clinical practice. Participants in antidepressant clinical trials are a highly selected group, not representative of the general population of depressed patients. They are not actively suicidal; they are almost always moderately (not severely or mildly) depressed outpatients; and they are free of comorbid physical or psychiatric illness. They are likely to have a higher placebo response rate than depressed patients who are more severely ill.

Further, clinical trials are not primarily designed to identify the optimal effect of antidepressants, but rather to rapidly assess their efficacy. Therefore, dose, duration, and diagnosis in clinical trials are not necessarily ideally suited to identify the optimal effects of antidepressants. Accordingly, clinical trials may identify the lower bound of the effect size.

The less-than-impressive difference in this and other studies between drug and placebo also recalls that patients who are assigned to placebo treatment in clinical trials are not untreated. The capsule they receive is pharmacologically inert, but hardly inert with respect to its symbolic value and its power as a conditioned stimulus. In addition, placebo-treated patients receive all the components of the treatment situation common to any treatment, ie, a thorough evaluation; an explanation for distress; an expert healer; a plausible treatment; expectation of improvement; a healer's commitment, enthusiasm, and positive regard; and an opportunity to verbalize their distress. Frank and Frank¹⁸ make a compelling case that these ingredients of the treatment situation are the active ingredients of all the psychotherapies.

Although our analysis supports our hypotheses, there are several deficiencies in the FDA database that limit our conclusions. We were unable to calculate specific response rates to drug vs placebo, to establish acceptable confidence intervals between drug and placebo, or to establish the effect on our results of patients who terminated clinical trials prematurely. However, degree of symptom reduction is a valid method to assess outcome in acute treatment clinical trials and is routinely used in reporting of results in antidepressant clinical trials.

Our data indicate that depressed patients in clinical trials who are assigned to placebo treatment are not at greater risk for suicide or suicide attempts than those assigned to active treatment. Depressed patients who are assigned to placebo experience substantial symptom reduction. We hope that these and other pertinent data can usefully inform deliberations about the risks, benefits, and ethics of placebo-controlled antidepressant studies.

Accepted for publication September 13, 1999.

Reprints: Arif Khan, MD, 1900,116th Ave NE, No. 112, Bellevue, WA 98004 (e-mail: arif@accessone.com).