Concerning the harm-benefit ratio of escitalopram for pediatric generalized anxiety disorder. A critical viewpoint on the evidence and approval process

Abstract

In this commentary we review the recent approval of escitalopram for generalized anxiety disorder (GAD) in children and adolescents. We critically discuss the FDA approval document and the approval trial. In the approval trial, efficacy was not clinically meaningful, statistical significance uncertain, and there were significantly more adverse events with escitalopram than with placebo. Relative to placebo, children and adolescents exposed to escitalopram were more likely to become suicidal than to experience a clinically relevant improvement in anxiety. Overall, the harm-benefit ratio seems problematic for escitalopram for pediatric GAD.

In this opinion piece, we review the recent Food and Drug Administration (FDA) approval of escitalopram for generalized anxiety disorder (GAD) in children and adolescents which seemed to violate the precautionary principle since benefits demonstrated were weak and harms were considerable. In 2019, 8.4% of children in the US aged 5 to 17 years had taken medication for their mental health in the preceding 12 months according to the CDC. ¹ Escitalopram is the second antidepressant to be approved for use in GAD in children, with duloxetine approved in 2014 by the FDA. ² Widespread use of escitalopram for GAD in clinical practice as a result of this approval may cause more harm than good for the majority of patients, which is particularly concerning given the rising prescription rates of antidepressants in children and adolescents in the US. ³

Efficacy

The primary outcome of the approval trial was the Pediatric Anxiety Rating Scale (PARS) after 8-week exposure to escitalopram or placebo, with a score-range of 0 to 25 points. The drug-placebo difference had a mean (M) of −1.4 (95% confidence interval CI: −2.7 to −0.2, p = 0.03). ⁴ We calculated a standardized mean difference (SMD) of −0.3 (CI −0.5 to −0.0). ⁵ Strawn et al. ⁶ and the FDA ^{7 (p. 13, 55)} considered this both “clinically meaningful” and “numerically small,” yet, the minimal important difference (MID) for the PARS has been estimated to be ≥4 points or ≥20% improvement. ⁸ This is consistent with findings from other fields, where the MID ranged between 12 and 18% of clinician rating scales, ⁹ corresponding with 3 to 4.5 points on the PARS. Furthermore, the MID in antidepressant research converges at a standardized mean difference (SMD) of 0.5. ¹⁰ Symptom reduction with placebo was 83% of that with escitalopram, attributable to variation, regression to the mean, and placebo effects. Thus, efficacy found in the approval trial is likely not clinically significant.

In a letter, Strawn and Mills defended the efficacy as clinically meaningful and consistent with findings from their previous study. ⁶ Still, they also argued that the efficacy of the approval trial was smaller due to anxiety-relieving effects of pandemic-related school closures, or the high placebo response rate possibly caused by the numerous study sites. However, efficacy in the previous 2020 study ¹¹ corresponded with an SMD of −4.30 (CI −5.29 to −3.29), which is not consistent with the approval study and suspect in its magnitude. Moreover, the 33% placebo response rate in the approval trial is very typical according to a review coauthored by Strawn, with a median of 32% for both the 14 antidepressant studies and for the 7 studies conducted in one or two sites only. 12 The claim about the pandemic is hypothetical and may apply more to social anxiety than GAD. Furthermore, anxiety symptoms increased in adolescents during this period, ¹¹ and baseline symptom severity was not lower than in previous studies. ¹²

Importantly, in the approval trial, statistical significance was lost in the more informative sensitivity analyses with imputed missing data (Table 7). ⁷ Here, the FDA commented: “the applicant’s post hoc sensitivity analysis does not ensure that the relationships between efficacy outcome and other factors are treated in a consistent way between the imputation model and the analysis model.” This post-hoc change was not explained in the publication. ⁴

Furthermore, all 6 secondary efficacy outcomes (remission/response on different scales) were non-significant and nearly zero (e.g., odds ratio OR = 1.2 (CI 0.6 to 2.4) for response on the PARS).

The FDA approval also referred to evidence for escitalopram for pediatric depression and adult GAD where efficacy for pediatric depression was only M = −2.6 (CI −5.3 to 0.0) points (SMD = 0.2) on the Childhood Depression Rating Scale with a MID of 5 points (SMD = 0.3). ¹³ The FDA approval also referred to the evidence from studies in adult GAD, where efficacy was −2.5 (CI −3.3 to −1.6) points on the Hamilton Anxiety Rating Scale. ¹⁴ Estimations of the MID are unavailable; a 2.5 point change corresponds to only 4% of the scale range.

Clinician rating scales can lead to unblinding and bias compared to patient-reported outcomes. No significant effects were found for self-report instruments in pediatric depression trials, ¹³ though this was not included in this trial.

Harms

Strawn et al. and the FDA considered treatment emergent suicidality as being rare. However, 9.5% patients on escitalopram experienced suicide ideation, compared to 1.5% with placebo, a significant difference (OR = 6.7, CI 1.8 to 47.2). This was not discussed by Strawn et al. ⁴ in the original publication which later referred to baseline imbalances in lifetime suicidality (16% escitalopram, 12% placebo). ⁶ Still, the difference was also significant for first emergence of suicidality during the trial (5.1% vs. 0.7%, OR = 6.6, CI 1.1 to 168.7). The FDA reported a number needed to harm (NNH) of 14. A previous study from Strawn et al. reported elevated levels of suicidality with escitalopram compared to placebo, too. ¹⁵ An increased risk for suicide ideation/behavior with escitalopram and SSRIs in general was also reported in trials for major depression in children, adolescents, and young adults.^13,16

Adverse events other than suicidality (e.g., insomnia, nausea, diarrhea) were more common for patients on escitalopram versus placebo (55.5% vs. 37.5%), OR = 2.1, CI 1.3 to 3.4, p < 0.01, NNH = 6.

Spontaneous reporting of individual adverse events is known to underestimate harm. This is well known for suicidality and was evident in the study by Strawn et al., where systematic assessment (Columbia Suicide Rating Scale) but not spontaneous reports showed significant differences. Similar underreporting is found for sexual dysfunction, ¹⁷ a frequent adverse event for escitalopram, yet not assessed in this study.

The FDA reported a major protocol deviation, not mentioned in the publication, for the downtaper period, where participants erroneously continued the 10 mg dose, impacting the safety evaluation. Additionally, safety assessment of tapering in the trial underestimates problems occurring in clinical practice with usually longer treatments, as withdrawal problems increase with length of treatment. ¹⁸

Conclusion

The FDA concluded that “the benefits of the product outweigh the risks.” ^{5 (p. 14)} However, as we have outlined, efficacy was likely not clinically meaningful, statistical significance was uncertain (as shown in the sensitivity analyses), and there were significantly more adverse events with escitalopram than with placebo. In particular, children exposed to escitalopram were six to seven times more likely to become suicidal than those exposed to placebo. Indeed, relative to placebo, children and adolescents exposed to escitalopram were more likely to become suicidal than to experience a clinically relevant improvement in anxiety. A limitation is that our arguments are based on secondary analyses. Independent re-analyses of the raw-data of pediatric antidepressant trials have shown problematic research practices exaggerating benefit and downplaying harms.^19–21 For example, adverse events, including suicidality, were sometimes recategorized into less severe categories, were only reported when occurring above a certain frequency, or were distributed across categories to remain below this threshold. ¹⁷ Without access to the raw data, we do not know whether such biases were pertinent in the escitalopram trial discussed here.

Overall, the harm-benefit ratio seems problematic for escitalopram for pediatric GAD. The findings for suicidality support the FDA’s black box warning and are consistent with evidence from other randomized controlled trials. This is relevant in light of ongoing debates about the warning, which have been based on relatively weak evidence.^22,23 A common response to an overall negative harm-benefit ratio for the entire study population is arguing that a subset of patients may experience significant net benefit from the medication. As we have discussed elsewhere, ²⁴ for the treatment of depression with antidepressants, there is no robust evidence for this claim, nor for identifying predictors of a good response (relative to placebo). However, it is possible that future research might reveal which anxious children may benefit especially well from the psychotropic effects of antidepressants. Nonetheless, this would not change the current conclusion that the majority of patients experience net harm.

We suggest that the FDA should reconsider the approval decision, given the evidence suggests escitalopram is more likely to make this vulnerable population of anxious children and adolescents suicidal (OR = 6.7, CI 1.8 to 47.2 in the study by Strawn et al.) than to achieve response in anxiety symptoms (OR = 1.22, CI 0.6 to 2.4). While the FDA’s approval criteria were met, our re-analysis reveals potential flaws in current approval procedures, which might need to be re-examined. The inclusion of a “boxed warning” about suicidality may be insufficient to substantially modify the risk of suicidality. If approval is maintained guideline committees and clinicians need to provide patients and families with honest information about the lack of evidence for clinically meaningful benefit, the evidence of harms, alternative approaches, and, if medication is used, close monitoring (often lacking in clinical practice). ²⁵