Clinical Response and Risk for Reported Suicidal Ideation and Suicide Attempts in Pediatric Antidepressant Treatment:  A Meta-analysis of Randomized Controlled Trials FREE

The US Food and Drug Administration (FDA), in its review and meta-analysis of 24 placebo-controlled trials assessing use of antidepressant medications among more than 4400 children and adolescents, concluded that these medications pose a 2-fold (4% vs 2%) increased risk for “suicidal behavior or suicidal ideation,” although no suicides were reported.^1- 2 In the FDA analysis, suicidal behavior comprised actual suicide attempts and preparatory actions toward imminent suicidal behavior (Box). Subsequently, the FDA mandated that a boxed warning be put on the labels of all antidepressants indicating an increased risk of suicidal thoughts and behavior in youth taking these medications but stopped short of prohibiting their use.

^aSuicidal ideation and suicidal behaviors that occurred during the double-blind acute-treatment period or within 1 day of the end of this period.¹

The concern about antidepressants must be considered in the context of possible benefit. The extant evidence supports the efficacy of antidepressant treatment for pediatric major depressive disorder (MDD), obsessive-compulsive disorder (OCD), and non-OCD anxiety disorders, although the strength of the treatment effect varies by indication and by the specific antidepressants.^4- 19 A prior meta-analysis²⁰ examined the risk-benefit profile of selective serotonin reuptake inhibitors (SSRIs) in pediatric depression, but not other indications, and predated the publication of 2 large clinical trials of pediatric depression among more than 700 children and adolescents.^5,18 We therefore conducted a more current review and meta-analysis of pediatric trials assessing use of antidepressants across the indications of MDD, OCD, and non-OCD anxiety disorders and also examined patient- and trial-level moderators of benefit and risk.

A PubMed literature search was conducted (from January 1988 through July 2006) using the MeSH keywords SSRI; serotonin reuptake inhibitors; antidepressive agents: second-generation; child; adolescent; randomized controlled trial; specific names of antidepressant medications; and names of researchers known to the investigators for having conducted at least 1 clinical trial of pediatric antidepressants. Tricyclic antidepressants and monoamine oxidase inhibitors were not included in our search, because the FDA report specifically targeted the newer classes of antidepressants.^1- 2 Relevant studies were also identified through (1) references of originally identified articles; (2) FDA and British Medicines and Healthcare Products Regulatory Agency reports^{1- 2,21- 22}; (3) scientific proceedings of the American Psychiatric Association (2000-2005), the American Academy of Child and Adolescent Psychiatry (1998-2005), and the New Clinical Drug Evaluation Unit (2001-2006); (4) clinical trial registries (primarily www.clinicaltrials.gov); and (5) contact with individual investigators.

We included randomized, placebo-controlled, parallel-group trials that studied SSRIs as well as other second-generation agents in children and adolescents (<19 years) with MDD, OCD, or non-OCD anxiety disorders (Table 1). Trials were included if efficacy or treatment-emergent data on suicidal ideation/suicide attempt were available for both antidepressant-treated and placebo-treated participants. For the majority of studies, data on suicidal ideation/suicide attempt were obtained from the FDA report.^1- 2 For studies not included in the FDA review, data on suicidal ideation/suicide attempt were obtained either from the principal investigators of each study^{11- 12,30- 31,33} or from study reports.^10,18

We excluded trials that used a crossover design³⁵; had only an active comparator group³⁶; only randomized treatment responders³⁷; provided adjunctive cognitive behavioral therapy to all patients³⁸; were currently recruiting patients (www.clinicaltrials.gov NCT00107120, NCT00352768, NCT00150306, and NCT00052078); studied anxiety-related indications other than separation, social, or generalized anxiety disorder^39- 40; or for which the primary author could not provide data.⁴¹

The quality of published trial reports for which data were available was independently rated by 2 coauthors (J.A.B., C.B.S.) according to the criteria of Detsky et al,⁴² with final quality ratings based on consensus (intraclass correlation coefficient between raters, 0.94; 95% confidence interval [CI], 0.92 to 0.95) (Table 1).

The primary study-defined measure of treatment response and the prospectively identified scalar variable assessing change in symptoms from baseline to the end of treatment were the efficacy outcomes used for this meta-analysis. For studies that were included in the FDA analysis,^1- 2 the primary risk outcome was suicidal ideation/suicide attempt (Box), as rated from suicidal adverse event narratives by an independent panel of suicide experts assembled by Columbia University on behalf of the FDA³; a breakdown of suicidal ideation/suicide attempt by age and type (suicidal ideation, suicide attempt/preparatory actions toward imminent suicidal behavior) was provided by the study author (T.A. Hammad, MD, PhD, MSc, MS, US Food and Drug Administration, written communication, October 3 and December 19, 2006). For studies not included in the FDA analysis,^{10- 12,18,30- 31,33} we rated suicidal ideation/suicide attempt according to the Columbia coding scheme.³

Trial characteristics were compared among treatment indications by using standard parametric and nonparametric statistics. The primary dichotomous outcomes (eg, symptomatic response) were expressed as Mantel-Haenszel risk differences, indicating the drug-placebo difference in the proportion of participants with each specific outcome. Use of risk differences permits all trials, including those with no events, to be included in analyses.

We used the DerSimonian and Laird random-effects model⁴³ to obtain a pooled estimate of the risk difference and corresponding 95% CIs. For the primary scalar measure of efficacy, we calculated the pooled random-effects mean difference and 95% CI between the treated and placebo groups, divided by the pooled standard deviation (Hedges g).^43- 44 The random-effects approach to meta-analysis assumes that there is heterogeneity across studies that is not fully accounted for by observed covariates, such as age, and provides more conservative estimates of effect by incorporating within- and between-study heterogeneity, whereas the fixed-effects approach assumes that the effect size is constant across studies. To test the validity of the DerSimonian and Laird model, we performed a sensitivity analysis comparing inferences of the random-effects model with those of a fixed-effects model. Estimates, CIs, and P values based on the 2 models were very close when the test for homogeneity did not reject; when it did, the random-effects approach gave wider CIs and larger P values, as expected (data available on request). When outcomes were significant, we calculated the number needed to treat to benefit (NNT) and number needed to treat to harm (NNH) according to the methods of Sackett et al,⁴⁵ with 95% CIs computed using the method of Altman.⁴⁶

To provide a common effect-size comparison between the primary categorical and scalar efficacy outcomes, we computed g for the dichotomous response variable of each study by means of transformation from the original log odds ratio scale to the standardized mean difference and then from this difference to g.⁴⁷ Heterogeneity between trials was assessed by the Cochran Q statistic (P≤.10) and the I² statistic, a transformation of Q that indicates the percentage of variation in the effect-size estimate attributable to heterogeneity rather than sampling error.⁴⁸ Sensitivity analyses were conducted by iteratively deleting each study and calculating the resulting effect sizes. As a comparison with the FDA's analyses,^1- 2 we also used fixed- and random-effects models to compute the pooled risk differences and risk ratios (RRs) of suicidal ideation/suicide attempt, where an RR greater than 1 indicates a higher relative risk for patients treated with antidepressants than for those treated with placebo. In contrast to the FDA report, we estimated pooled risk differences for each indication and combined 6 trials into 3, because these trials were pooled for publication.^15,29,32 In trials in which no suicidal ideation/suicide attempt occurred in 1 of the trial groups, 0.5 was added to each of the 4 cells to reduce bias and variance of the RR estimates, but trials without any events could not be included in the pooled estimate. We tested for a treatment × indication × suicidal ideation/suicide attempt interaction using an exact homogeneity test.

Secondary analyses included age-grouped (<12 or 13 years [depending on study definition] vs ≥12 or 13 years) comparisons of efficacy and suicidal ideation/suicide attempt, and tests of continuous and categorical moderators. For ease of interpretation, we refer to the younger age group as younger than 12 years and to the older age group as 12 years or older for all comparisons. For continuous moderators, regression was used to explore whether methodological quality, number of treatment sites, proportion of female participants, and duration of illness at baseline influenced outcome. Continuous moderators that showed an association with outcome were dichotomized by median split and reanalyzed as categorical variables. Trial-level categorical moderators of outcome, namely publication status, primary funding source, study location, drug class, and use of placebo run-in period were also examined. Publication bias was assessed visually using a funnel plot and quantitatively using an adjusted rank correlation test⁴⁹ and a regression procedure to measure funnel plot asymmetry.⁵⁰ To reduce heterogeneity in anxiety disorders trials, we excluded 1 small study³¹ with very large treatment effects from all regression and moderator analyses of efficacy outcomes. Statistical analyses were performed with Comprehensive Meta-analysis version 2.1 (Biostat, Englewood, NJ), SPSS version 13.1 (SPSS Inc, Chicago, Ill), and StatXact-7 (Cytel Software Corp, Cambridge, Mass). We present P values and CIs unadjusted for the multiple primary and secondary hypotheses tested. P<.05 was used to determine statistical significance. The reader can make adjustments with such methods as Bonferroni.

Our search identified 27 prospective, randomized, placebo-controlled clinical trials that compared the relative efficacy of antidepressants against placebo in patients younger than 19 years for the treatment of MDD (15 trials, n = 3430), OCD (6 trials, n = 718), and non-OCD anxiety disorders (6 trials, n = 1162) (Figure). Response rates were available for all but 2 MDD trials,^23,51 and a primary scalar measure of efficacy was available for all trials. Complete data with regard to suicidal ideation/suicide attempt were available except in 1 trial, in which data on suicidal ideation/suicide attempt were obtained from only 1 of 2 treatment sites.³⁰

Characteristics of the 27 included trials are shown in Table 1. The median number of treatment weeks for MDD, OCD, and non-OCD anxiety disorders was 8, 11, and 11, respectively. Almost half (7/15 [47%]) of the MDD trials used a placebo run-period, compared with 2 (33%) of 6 OCD trials, and 1 of 6 (17%) non-OCD anxiety disorders trials (P = .51). All but 4 of the studies used a flexible dosing schedule. The majority (23/27 [85%]) of studies were multisite (median No. of sites, 21; interquartile range, 5-36), with no differences in number of study sites by treatment indication (P = .63). Most studies were conducted solely in the United States (19/27 [70%]). All OCD and non-OCD anxiety disorders trials were published or in press, compared with 12 (80%) of 15 MDD trials (P = .38). The median overall Detsky score for the 23 trials in which quality was assessed was 0.88 (interquartile range, 0.80-0.95), indicating generally good methodological quality. Failure to analyze data on a strictly intent-to-treat basis and an inadequate description of the randomization and exclusionary procedures were the most common methodological flaws. There were no differences in methodological quality by treatment indication (P = .91).

In MDD trials, the Clinical Global Inventory–Improvement scale⁵² was used as a primary measure of response in 8 (53%) of 15 trials. The Children's Depression Rating Scale–Revised⁵³ was used as the primary scalar measure of efficacy in 11 (73%) of 15 trials. In OCD trials, a reduction of 25% or more on scores from the Children's Yale-Brown Obsessive Compulsive Scale^54- 55 was used as the primary measure of response in 3 of 6 trials. All OCD trials used change in the score from this latter scale, from baseline to end of treatment, as the primary scalar measure of efficacy. In non-OCD anxiety disorders trials, the Clinical Global Inventory–Improvement scale was used as the primary measure of response in all 6 trials, and the Pediatric Anxiety Rating Scale⁵⁶ was used as the primary scalar measure of efficacy in 2 (33%) of 6 studies.^11- 12

Efficacy outcome data are summarized in eTable 1,^57- 58eTable 2, and Table 2. Treatment-emergent suicidal ideation/suicide attempt data are summarized in Table 3 and Table 4, where separate analyses of suicidal ideation and suicide attempt/preparatory actions are presented for comparison with the overall rates of suicidal ideation/suicide attempt.

Major Depressive Disorder. Based on data from 13 trials and 2910 participants, pooled absolute rates of response were 61% (95% CI, 58% to 63%) in participants treated with antidepressants and 50% (95% CI, 47% to 53%) in those treated with placebo, yielding a pooled risk difference of 11% (95% CI, 7% to 15%), an NNT of 10 (95% CI, 7 to 15), and a g of 0.25 (95% CI, 0.16 to 0.34). A similar effect was found in pooled analyses of continuous measures of mean improvement in depression symptomatology (g = 0.20; 95% CI, 0.12 to 0.29). Pooled absolute rates of suicidal ideation/suicide attempt were 3% (95% CI, 2% to 4%) in antidepressant-treated participants and 2% (95% CI, 1% to 2%) in those receiving placebo. The pooled risk difference was 1% (95% CI, −0.1% to 2%; P = .08), so the NNH was 112.

Obsessive-Compulsive Disorder. In 6 OCD trials of 705 participants, pooled rates of response were 52% (95% CI, 46% to 57%) in SSRI-treated participants and 32% (95% CI, 27% to 37%) in those receiving placebo, and the pooled risk difference was 20% (95% CI, 13% to 27%), yielding an NNT of 6 (95% CI, 4 to 8), and a g of 0.48 (95% CI, 0.30 to 0.65). The pooled estimate of g for the scalar measure of efficacy was 0.41 (95% CI, 0.26 to 0.56). Pooled absolute rates of suicidal ideation/suicide attempt were 1% (95% CI, 0% to 2%) in SSRI-treated participants and 0.3% (95% CI, −0.3% to 1%) in those receiving placebo, and the pooled risk difference was 0.5% (95% CI, −1% to 2%; P = .57), for an NNH of 200.

Non-OCD Anxiety Disorders. In 6 non-OCD anxiety disorders trials of 1136 participants, pooled rates of response were 69% (95% CI, 65% to 73%) in antidepressant-treated participants and 39% (95% CI, 35% to 43%) in those receiving placebo, and the pooled risk difference was 37% (95% CI, 23% to 52%), yielding an NNT of 3 (95% CI, 2 to 5). While all studies favored antidepressant treatment, there was large variation in the degree of benefit across trials (risk difference range, 19%-82%; Q₅ = 32.88; P<.001). One small pilot study³¹ had the highest point estimate. Exclusion of this trial diminished the prior result to a risk difference of 30% (95% CI, 19% to 41% [NNT, 4; 95% CI, 3 to 6]); however, significant heterogeneity remained (Q₄ = 14.71, P = .01). For the primary scalar measure of efficacy, we calculated the pooled estimate of g after excluding the pilot study (0.69; 95% CI, 0.47 to 0.91; Q₄ = 12.19; P = .02), which was similar to the primary risk difference of response expressed as g (0.65; 95% CI, 0.52 to 0.79). Pooled absolute rates of suicidal ideation/suicide attempt were 1% (95% CI, 0.2% to 2%) in antidepressant-treated participants and 0.2% (95% CI, −0.2% to 0.5%) in those receiving placebo, and the pooled risk difference was 0.7% (95% CI, −0.4% to 2%; P = .21), yielding an NNH of 143.

Among participants treated with placebo, the risk of suicidal ideation/suicide attempt was greater in MDD trials compared with non-OCD anxiety disorders trials (odds ratio, 9.9; 95% CI, 1.6 to 406.3) and OCD trials (odds ratio, 5.8; 95% CI, 0.9 to 237.3); these effects were limited to adolescent participants. There was no evidence of a treatment × indication × suicidal ideation/suicide attempt interaction (P = .53).

As shown in Table 5, our estimates of the risk difference for suicidal ideation/suicide attempt are lower than those reported by the FDA because we used a random-effects model to combine data, whereas the FDA used a fixed-effects model, and because we included 7 additional studies,^{10- 12,18,30- 31,33} 3 of which were not available at the time of the FDA report.^10,18,33 The FDA used primarily fixed-effects models to combine data from the randomized trials but also conducted random-effects models as part of a sensitivity analysis. The estimates derived from the latter analyses were more similar to our results, since random-effects models tended to yield smaller RRs and risk differences.

In MDD trials, the number of trial sites was inversely associated with risk differences in response (β = −.002, Q₁ = 3.8, P = .05) and the scalar measure of efficacy (β = −.01, Q₁ = 3.7, P = .05), suggesting a decrease in the magnitude of antidepressant treatment effects as the number of study sites increased. In 9 of 15 MDD trials for which data were available, duration of the depressive episode at intake was inversely associated with risk differences in response (β = −.01, Q₁ = 5.4, P = .02) and the scalar measure of efficacy (β = −.02, Q₁ = 9.4, P = .002), indicating less efficacy with longer duration of illness. Study quality and proportion of female participants were not associated with outcome.

In OCD and non-OCD anxiety disorders trials, no patient characteristics or elements of trial design were associated with outcome (P ≥.20 for all).

In the 10 MDD trials for which age-grouped data were available,^{4- 5,14- 17,24- 26,29} the pooled risk difference in response was lower in children than in adolescents and not significantly different than zero, in part related to a higher placebo response in children (χ²₁ = 5.5, P = .02) (Table 2). Limiting trials to only fluoxetine studies resulted in statistically significant pooled risk differences in response that were similar for children (21%; 95% CI, 4% to 37%; P = .01) as well as adolescents (20%; 95% CI, 7% to 33%; P = .002). The trial-level variables of drug class, publication status, and placebo run-in period were not significant moderators (eTable 3). The potential role of funding source was not examined, because trials funded by the National Institute of Mental Health all investigated the efficacy of fluoxetine.

In OCD trials, the pooled measures of efficacy outcome were comparable in children and adolescents, and the use of a placebo run-in period was not a significant moderator of outcome.

In the 3 non-OCD anxiety disorders studies for which age stratification was possible,^11- 13 there was a trend for the risk difference to be larger among adolescents (46% vs 29%, Q₁ = 3.71, P = .05) (Table 2). Trials of SSRIs were associated with larger effect sizes of efficacy favoring antidepressants compared with non-SSRI trials and with eliminating intertrial heterogeneity (eTable 3).

With respect to suicidal ideation/suicide attempt, none of the patient characteristics or trial features was an important moderator in MDD, OCD, and non-OCD anxiety disorders trials.

Separate analyses of publication bias were conducted for MDD, OCD, and non-OCD anxiety disorders trials. For each treatment indication, there was no evidence of publication bias based on visual analysis of the funnel plots, the adjusted rank correlation tests, and the regression intercept approach (data available on request).

This meta-analysis of all available randomized clinical trials of antidepressant treatment of pediatric MDD, OCD, and non-OCD anxiety disorders shows evidence of efficacy for all 3 indications, although the effects were strongest for non-OCD anxiety disorders, intermediate for OCD, and more modest in MDD. We found some evidence that age moderated outcome both for the treatment of depression and anxiety, and duration of depression was also a moderator. One trial-level moderator was found in studies of MDD, in which efficacy was inversely proportional to the number of trial sites. Consistent with the analyses of the FDA,¹ we found evidence of an overall small but increased risk of treatment-emergent suicidal ideation/suicide attempt. However, the pooled random-effects risk differences of suicidal ideation/suicide attempt for each indication were all less than 1%. There were no completed suicides in these trials. Before we discuss each of these findings, we first place these results within the limitations of this study.

The main limitations are those imposed on the meta-analytic venture, which is the quality and quantity of data available for meta-analysis. With relatively few trials in any of the 3 diagnostic categories, our ability to draw inferences about design and diagnostic moderators of treatment response and of suicidal ideation/suicide attempt are limited to major effects. Moreover, data on factors that might explain differences in outcome between trials relative to baseline differences in participants, dosing, or adherence to treatment are not consistently accrued in clinical trials and therefore were not routinely available for this meta-analysis. We did not adjust the P values for multiple comparisons, because the contrasts for the primary outcomes were planned and the secondary analyses were exploratory and should be regarded with caution. No study was designed to examine suicidal ideation/suicide attempt as a study outcome, and in fact most trials were conducted in patients who had been carefully screened to exclude youths at risk. Thus, unlike efficacy data for which the outcome is defined a priori, most studies only collected data on suicidal ideation/suicide attempt as part of adverse events reporting, which is generally retrospective and not systematically assessed. Finally, our risk analysis was limited to only suicidal ideation/suicide attempt. We did not consider other common adverse effects, such as activation, insomnia, and gastrointestinal tract symptoms, that may be more common in children than in adolescents,⁵⁹ or other adverse effects of the medication, such as disinhibition, mania, irritability, and agitation, any of which may precipitate suicidal ideation/suicide attempt.⁶⁰

Adolescents appeared to respond better than children to antidepressants in trials of both depression and anxiety. While there was evidence of efficacy for antidepressant treatment for depressed adolescents across several antidepressant types, only 1 agent, fluoxetine, outperformed placebo in depressed children younger than 12 years. The reason why depressed children seemed to respond better to fluoxetine compared with other agents is unclear but could be due to study quality, location, or properties of the medication itself, such as its long half-life. These results are consistent with recent reports examining the age-stratified performance of fluoxetine and other antidepressants in pediatric major depressive disorder^29,61- 64 and suggest that the lack of a significant treatment effect may be due in part to a higher placebo response in children. The Treatment of Adolescent Depression Study (TADS)⁵ found that fluoxetine showed a larger effect relative to placebo in more severely depressed individuals (effect size, 0.69 vs 0.39), so perhaps greater separation between drug and placebo would be expected in more severely ill children. Adolescents also showed a vigorous response to antidepressants for the treatment of anxiety compared with children, although children still showed a significant and positive effect. This may be because separation anxiety is more common in children than in adolescents and, at least in 1 study, did not appear to respond as well to antidepressant treatment as did the other anxiety disorders.¹² Despite some age differences in response, antidepressants appear to be more efficacious in the treatment of pediatric anxiety in both children and adolescents, compared with the effects found for depression. This may be due to chronicity and stability of anxiety disorders relative to depression⁶⁵ and greater homogeneity of children with anxiety disorders, compared with depressed children, some of whom do not go on to have recurrent depressive disorders.⁶⁶

In depression trials, efficacy was inversely proportional to the duration of the depressive disorder. While these results should be interpreted with caution because there was a large amount of missing data (6/15 trials [40%]) for this variable, these findings were also reported in 1 large individual trial,⁶¹ as well as in the Sequenced Treatment Alternatives to Relieve Depression (STAR-D) study in adults.⁶⁷ This suggests that efforts to identify depressed youth earlier in their depressive course may improve outcome. Given that anxiety is a frequent precursor of depression and that the efficacy of antidepressants for non-OCD anxiety disorders is larger than those for depression, identification and treatment of anxiety disorders may be another strategy to reduce the public health burden of child and adolescent depression.⁶⁵

Effect sizes for efficacy outcomes in depression trials were inversely proportional to the number of sites. This suggests the need to either limit studies to fewer, larger sites, or to improve quality control if a study requires a larger number of sites, perhaps by centralizing the determination of eligibility and the assessment and monitoring of treatment quality to a central site.⁶⁸

Similar to the FDA's analysis, we found an overall increased risk of suicidal ideation/suicide attempt associated with antidepressant treatment. Our pooled estimates of risk differences are smaller than those from the FDA report, because we used random-effects rather than fixed-effects models for combining studies and also because we included 7 additional studies not covered in the FDA report.^{10- 12,18,30- 31,33} When we reanalyzed the data using fixed-effects models, our results were more similar to those of Hammad et al.¹ While either approach to meta-analysis is defensible, we believe that a random-effects approach is more appropriate for pooling data because it does not assume that there is a common effect across all studies. In light of quantitative and qualitative evidence of heterogeneity among trials, this is a more conservative assumption and provides estimates of risk differences that are less vulnerable to heterogeneity among studies. Moreover, we deemphasized the use of the metric of relative risk because it excludes studies in which there were no events in either cell, which then is not an accurate reflection of the risk found in the population of all studies. Our findings are also consistent with those of a recent reanalysis⁶⁹ of the data from the study by Hammad et al¹ that used Bayesian meta-analysis to include all available trials and found an increased risk of suicidal ideation/suicide attempt in the subset of MDD trials only. Using the metric of relative risk, the risk of suicidal ideation/suicide attempt in our analysis is increased at least as much in the OCD and non-OCD anxiety disorders groups as in the depressed group. However, a 2-fold increase in suicidal ideation/suicide attempt is of potentially greater clinical significance in depression, because the base rate of incident suicidal ideation/suicide attempt is much higher in the placebo-treated depressed group compared with the OCD or non-OCD anxiety disorders placebo groups.

This meta-analysis cannot shed light on exactly what characteristics of patient, medication, or trial might affect incident suicidal ideation/suicide attempt, but at a study level we did not find any effect of sex or use of a placebo run-in period. While there has been some speculation that emergent suicidal ideation/suicide attempt might be induced by withdrawal effects found particularly in those drugs with shorter half-lives, we found no such association when comparing the risk of suicidal ideation/suicide attempt in fluoxetine, a drug with a longer half-life (5 days) than those of the other drugs (all substantially shorter than 24 hours) (P = .58). Furthermore, the risk of incident suicidal ideation/suicide attempt does not appear to be a consequence of lack of efficacy, insofar as the magnitude of increased risk is as similar for anxiety as it is for depression in adolescents, even though antidepressants are more efficacious for anxiety than for depression.

A prior meta-analysis of SSRIs in pediatric depression²⁰ showed that published trials produced a more favorable risk-to-benefit profile than unpublished reports. In contrast, we found no evidence of a publication bias for studies of depression, because many of the negative trials that were unpublished at the time of the first report were either published or in press by the time of the current meta-analysis. Nevertheless, there is a need for mandatory clinical trial registration at a study's inception, because trial registries provide a step toward greater transparency.⁷⁰ To promote full disclosure, performance and outcome data from clinical trials should also be made publicly available and published in a timely manner, regardless of results.

There are several recommendations for future studies that emerge from what could not be gleaned from this meta-analysis. Baseline characteristics that are likely to be related to treatment response, as well as emergent suicidal ideation/suicide attempt, should be assessed and publicly available so that future analyses can elucidate the impact of these variables on outcome. For example, self-reported suicidal ideation and irritability at intake have been found to predict future suicidal ideation/suicide attempt.^71- 72 Future research should focus on identifying the most efficient methods for systematically monitoring children and adolescents taking antidepressant medications for both clinical response and for adverse events. An effective monitoring system could increase the confidence of clinicians and parents in the use of these medications. Finally, although it would increase the cost of drug trials, assays of drug and metabolites levels could identify those who are nonadherent or who have either very slow or very rapid metabolism and thus are not responding to the treatment for reasons other than intrinsic properties of the medication. Pharmacogenetic predictors of response and adverse events may also hold promise.^73- 74

In summary, depending on treatment indication, the NNT ranges from 3 to 10, while the NNH via emergence of suicidal ideation/suicide attempt ranges from 112 to 200, indicating a favorable overall risk-to-benefit profile for antidepressants in the treatment of pediatric MDD, OCD, and non-OCD anxiety disorders. Although our meta-analysis of antidepressant treatment found an overall increased risk for suicidal ideation/suicide attempt compared with placebo, the pooled risk differences for each specific treatment indication, while in the direction of increased risk, were not statistically significant. Some may argue that any risk of suicidal ideation/suicide attempt cannot possibly justify treatment with antidepressants for children and adolescents. Instead, we believe that the strength of evidence presented here supports the cautious and well-monitored use of antidepressant medications as one of the first-line treatment options, with the recognition that efficacy appears greatest for non-OCD anxiety disorders, intermediate for OCD, and more modest for MDD. Since the choice of treatment should be the result of a collaborative discussion between clinician, family, and patient, the information presented in this report should allow for an informed evaluation of the potential benefits and risks of these medications vs no treatment and provide a framework for their comparison with nondrug treatments as well.^75- 76