The problem is their argument presents a false narrative that is factually inaccurate in ways that are directly verifiable from the evidence their article, itself, presents. We write as authors of peer-reviewed publications that bear directly on the claims the article makes, and as scientists who reviewed and critiqued a draft of their manuscript before its submission to Regulation. Our critique identified specific errors and directed the authors to published work contradicting their conclusions. Given the importance of this issue, we were concerned to see the article was published without correcting any identified error and without citing any of the identified literature. We also note that the article quotes a passage from our written review without our permission and instead falsely attributes it to “the FDA,” even though none of us is currently affiliated with that agency. Our positions reflect analysis of the evidence rather than defense of a regulatory outcome: Each major empirical claim cited here is independently reviewable in publicly available data or peer-reviewed publications. We believe Regulation’s readers deserve a fuller account of what the evidence shows.

A Real Signal, Rigorously Confirmed

Khan et al. characterize the data considered at the 2004 and 2006 FDA advisory committee meetings as “mixed and inconclusive” and as having “failed to showcase statistical significance.” This is incorrect. The 2004 pediatric analysis found a relative risk for suicidal thinking or behavior of 1.95 with a 95 percent confidence interval of 1.28 to 2.98 and p‑value < 0.002—a result that is, by any standard definition, statistically significant (Hammad et al. 2006). In other words, the analysis found that suicidal thinking or behavior was twice as frequent among the children who received antidepressant drugs in the trials compared to the children who received placebo.

Following that finding, the FDA requested that manufacturers submit adult trial data so the same question could be examined across the full age spectrum. This was presented at the 2006 meeting. The result was the most rigorous analysis of this question ever conducted: a meta-analysis of 372 double-blind, randomized, placebo-controlled trials involving 99,231 adults, using primary patient-level data with validated, blinded adjudication of every event (Stone et al. 2009). The findings were unambiguous. For adults under 25, the odds for a suicide attempt were 2.3 times higher for patients who took antidepressants compared to those given placebo, approaching the risk seen in children and adolescents. Risk declined continuously with age, showing a net neutral effect in adults aged 25–64 and a protective effect in those 65 and older. This age-dependent gradient was statistically significant, robust across multiple statistical methods, and consistent across drug classes and diagnostic categories. Two of the three authors of this article are authors of that study.

In other words, the boxed warning (the correct term, “black box” is pejorative) reflects a confirmed, biologically plausible signal with a clear mechanistic rationale: known age-related differences in impulse control and neurobiological sensitivity. Removing the boxed warning would not correct a regulatory error; it would eliminate clinically valuable information about a real adverse drug effect from the prescribing encounter.

The Causal Claim Fails on Its Own Evidence

The Khan et al. article does not merely suggest an association; it attributes specific excess deaths to the warning and calls for its removal on that basis. Its argument rests on a specific causal chain: The boxed warning suppressed antidepressant prescribing; suppressed prescribing meant fewer young people received treatment; and fewer treated patients meant more suicides. This chain breaks down at every link.

Wrong timing / Their article’s own Figure 2 shows that suicide rates among 10- to 14-year-olds decreased for several years after the 2004–2005 warning and did not begin rising until 2007–2009, three to five years later. A cause cannot consistently precede its alleged effect by that margin without explanation (Stone 2014). Khan et al. offer none nor do they consider other explanations for that increase in suicides.

Misrepresented comparison group / The article presents adults aged 55 and older as a control population whose stable suicide rates confirm that the warning uniquely harmed younger people. In fact, published CDC data show that suicide rates for men aged 55–64 and women aged 55–64 and 65–74 were 25–44 percent higher in 2021 than in 2001 (Garnett et al. 2023)—rates of increase comparable to those the article attributes to the boxed warning in adolescents. The pattern the article presents as specific to warned populations is not specific at all.

Any prescribing suppression ended long ago / The causal chain requires that the warning had sustainably suppressed antidepressant prescribing to young people. It did not. Prescribing returned to pre-warning levels by 2010–2012. Between 2016 and 2022, antidepressant prescriptions for adolescents aged 12–17 increased by 43 percent, and among those aged 18–25 by more than 47 percent (Chua et al. 2024). Antidepressant use among young people is now substantially higher than before the warning was issued, yet suicide rates have continued to rise. Their article offers no account of this.

Our Figures 1 and 2 are from CDC data for suicide rates by age and sex from 1999 through 2024. They compare rates for ages 10–14 and 15–19 with those for ages 55–59 and 60–64 (data for all age ranges between 10 and 84 are provided in Tables 1 and 2) and illustrate the misleading way this was presented.

Between the issuance of the boxed warning in early 2005 and the end of 2007, a period during which the rate of youth antidepressant prescribing declined by 20 percent, suicide rates for girls aged 10–14 and 15–19 decreased by 46 percent and 30 percent, respectively (Stone 2018). For boys, the declines were 29 percent and 13 percent, respectively. In contrast, suicide rates for women aged 55–64 increased by about 20 percent, with about a 10 percent increase for men of the same ages.

Youth prescribing began to increase in 2008, returning to 2004 levels in 2011. For both girls and boys, suicide rates increased relative to their 2007 nadir but remained near or below 2004 levels. For adults ages 55–64, suicide rates continued to increase.

Since 2011, youth prescribing rates have steadily increased. As our figures show, suicide rates for girls began a dramatic increase beginning 2013–2014, more than eight years after the boxed warning began. The rate has plateaued, beginning in 2017, for girls aged 15–19, but the rate has continued to increase for girls aged 10–14. Boys showed comparable rates of increase through 2017–2018, but rates then began to sharply decrease and were about 30 percent lower in 2024 than at their peak.

This sex-specific divergence is itself strongly inconsistent with the proposed mechanism of the Khan et al. thesis: Trends in prescribing rates followed similar trajectories in boys and girls (Chua et al. 2024), so a warning-driven suppression of prescribing does not plausibly explain why suicide rates rose in girls while simultaneously declining sharply in boys over the same period. For women aged 55–64, suicide rates stopped increasing around 2015–2016 and have since shown a decline. For men aged 55–64, rates peaked in 2018 and have stayed near that level or slightly decreased.

In general, trends in suicide rates for youth aged 15–19 have resembled those seen in adults. Rates for those aged 10–14 have been more volatile in both increases and declines; given that suicide is about eight times more frequent among 15- to 19-year-olds and adults than among 10- to 14-year-olds, year-to-year fluctuations in this youngest group are statistically unreliable as indicators of underlying trend. What is meaningful is the multi-year pattern, and that pattern is shared across age groups, making it implausible that it reflects a regulatory action specifically targeting pediatric prescribing.

European evidence inverts prediction / If the boxed warning in the United States caused increased youth suicide by discouraging antidepressant use, countries with stronger restrictions should show worse outcomes. Cross-national comparisons cannot isolate the effect of labeling policy, but they do test the directional prediction implied by the Khan et al. hypothesis: that more restrictive warnings should be associated with worse youth suicide outcomes. European product labeling states that antidepressants should not be used to treat depression in children and adolescents under 18, a categorically more restrictive position than that of the FDA. Yet Eurostat data on intentional self-harm mortality among 15–19-year-olds show declining rates across EU member states from 2011 through 2022 (Eurostat 2025). Khan et al. do not mention this.

Suicide and antidepressants / The causal chain assumes that changes in physician prescribing would meaningfully affect population suicide rates. This requires a substantial proportion of suicide decedents to have been under the active care of physicians who could have prescribed antidepressants but were deterred by the warning. The evidence does not support this. Only about 30 percent of people who die by suicide had any contact with mental health services in the year before their death, and more than one third had no healthcare contact of any kind in the preceding year (Walby et al. 2018; Porter et al. 2026). Among those who do have healthcare contact, many have conditions that are not primarily treated with antidepressants: schizophrenia, bipolar disorder, substance use disorders, and personality disorders. The population whose suicide risk could be reduced by changes in antidepressant prescribing is a small fraction of total suicide decedents.

Prescribing decline and the boxed warning / Even when prescribing did decline, Khan et al. never show that the FDA’s labeling decision—rather than the broader public controversy surrounding it—was responsible. The BBC Panorama broadcasts, news media coverage, and lawsuits all preceded the formal warning and were independently capable of influencing physician behavior. Nor is it clear that the controversy was responsible for a decline in prescribing. Antidepressant prescribing declined in 2005 across every age group through 54, not just adolescents (Stone 2014)—precisely what one would expect from reduced pharmaceutical marketing rather than from a pediatric-specific warning. Popular brand-name antidepressants went off patent in this period, generic market share rose sharply (Ventimiglia & Kalali 2010), and annual promotional expenditures by manufacturers fell by approximately 35 percent, or $800 million, between 2004 and 2006 (Pamer et al. 2010). The refocusing of remaining promotional spending toward patients soon to be eligible for the new Medicare Part D benefit further explains why prescribing increased among adults aged 55 and older during the same period that it declined in younger age groups, a pattern Khan et al. attribute entirely to the warning. These are the mundane economics of what the pharmaceutical industry views as a maturing drug market, not the fingerprints of a chilling regulatory action.

Prescribing rates and suicide rates / The Khan et al. article’s entire architecture assumes that more antidepressant prescribing produces less suicide. This assumption is not supported by evidence. When antidepressant use is low, it is concentrated among the highest-risk patients; as use expands, it reaches progressively lower-risk patients and the marginal protective benefit declines (Stone 2014).

Moreover, such a relationship incorrectly presumes that antidepressants are highly effective. Practitioners tend to overestimate the effectiveness of antidepressants, in part because large placebo effects create a misleading impression of drug response. The central empirical finding from an individual participant data analysis of 232 trials for treatment of major depression submitted to FDA (Stone et al. 2022) is that benefit is concentrated in a minority of responders; approximately 85 percent of participants randomized to antidepressants demonstrate no treatment-specific improvement beyond what is observed under placebo conditions. Among children and adolescents, the effect is even smaller, and only two antidepressants have gained FDA approval for pediatric depression.

This distributional finding sharply constrains the population-level implications of prescribing changes. Expanding antidepressant use primarily increases exposure among individuals unlikely to derive antidepressant-specific benefit, while still subjecting them to drug-related risks, including treatment-emergent suicidal thinking or behavior. A causal account in which reduced prescribing yields large increases in suicide therefore requires—without evidence—that those deprived of treatment were disproportionately drawn from the small subset of true responders and that these responders account for a substantial fraction of suicide mortality.

Finally, because few antidepressant prescriptions have significant beneficial effect on the risk of suicide, change in antidepressant use is more plausibly understood as a response to changes in the underlying prevalence of depression and suicide risk in the population, not a cause. Across 20 age and sex subgroups examined over more than a decade of national data, suicide rates and antidepressant prescribing prevalence showed a strong positive correlation in 13 subgroups—including all subgroups under age 45—and a strong negative correlation in only five (Stone 2018).

The Cited Evidence Is Methodologically Invalid for This Purpose

Throughout this analysis, we rely on population-level (ecological) data not to establish causation, but to evaluate whether the specific causal mechanism proposed by Khan et al. is consistent with observable patterns; they can rule out explanations that require tight temporal coupling, population specificity, or sustained prescribing suppression. We referred to this evidence to test internal coherence, not to infer causality from correlation.

The empirical literature the article marshals relies almost entirely on interrupted time series analyses and similar before-and-after comparisons of ecological data. These methods have intuitive appeal—they seem to show what happened before and after a policy change—but they rest on assumptions that fail in this context. Ecological data, moreover, cannot prove causal effects. Even when such before-and-after comparisons appear visually persuasive, they can generate compelling but spurious causal narratives precisely because they absorb pre-existing trends and unrelated temporal changes into the apparent effect of a policy intervention. Peer-reviewed analysis has also documented in detail that these studies involve unvalidated proxy outcome measures and cherry-picked time frames (Stone 2014; Stone 2018). A widely cited study (Gibbons 2007), for example, presented two graphs side by side—one showing a downtick in adolescent antidepressant prescriptions, the other an uptick in adolescent suicide rates—as evidence of a causal link. But the graphs were not aligned to the same period: The suicide uptick occurred in 2004, before issuance of the boxed warning, but the prescription downtick did not occur until 2005. Despite that error, which was pointed out immediately in correspondence to the journal (Jureidini 2007), the paper was never retracted and continues to be cited, including by Khan et al. (Stone 2014). The methodological critiques cited above were provided to the authors before publication. Neither appears in the Khan et al. article’s reference list, nor did they alter their argument in consideration of these criticisms.

What Removing the Warning Would Actually Accomplish

Khan et al. propose relocating the risk language from a boxed warning to the Warnings and Precautions section of the label. Even setting aside the failures of their causal argument, their proposed remedy would not follow from its own diagnosis. The considerations that follow concern regulatory design and clinical risk management rather than population-level causal claims about suicide trends.

The risk information would remain in the label. Clinicians and plaintiffs’ attorneys would remain aware of it. There is no published evidence that a warning carries less clinical or legal weight in Section 5 than in a box. A physician deterred from prescribing by the documented risk of drug-induced suicidal thoughts or behavior in young patients—or by concern about litigation if it occurs—would face the same considerations regardless of typographic placement.

As we explained in our pre-submission review of the Khan et al. manuscript, the purpose of the boxed warning is not to stigmatize antidepressants or to imply that they are categorically harmful. Rather, it serves to alert clinicians to a counterintuitive and clinically dangerous possibility: that new or worsening suicidal thoughts or behavior shortly after treatment initiation may be an adverse drug reaction rather than an inadequate treatment response. Without this awareness, clinicians may respond by increasing the dose or persisting with treatment under the assumption that symptoms will improve with time, thereby potentially exacerbating harm. This risk of mismanagement justifies the prominence of a boxed warning as opposed to relegation to a routine warning section.

Consider who Khan et al. claim the warning is harming: young patients with depression whose physicians are being deterred from prescribing. These are, by definition, patients who are already in the healthcare system, have an established diagnosis, and are under the care of a clinician who has evaluated them. This is not the population that accounts for most suicide deaths. Removing the warning would reduce awareness of a clinically important adverse reaction among the patients who are being actively managed—the ones who, by the logic of the article’s own argument, are most likely to benefit from treatment. This could lead to an increase in suicidal thoughts and behavior. Efficacy data are lacking in younger patients for most antidepressants. To the extent that it exists, it points to lesser efficacy than seen in adults. Any positive effect on suicide, resulting from a higher proportion of young, severely depressed patients receiving antidepressants, is apt to be quite small. The label provides quantitative estimates of the risk of suicidal thinking and behavior due to treatment. For patients with severe depression who are at higher risk for suicide, clinicians can weigh the potential for benefit from treatment against the potential for harm. For patients whose depression is not severe or who have non-depression indications, such as generalized anxiety disorder, the risk of suicidal thinking or behavior from their disorder is less than the risk of suicidal thinking or behavior being caused by the drug. If removing the boxed warning has any effect on encouraging prescribing in this population, the result may be symptomatic relief but also an increase in suicidal thinking or behavior and, possibly, death from suicide.

Conclusion

The concern that the FDA’s antidepressant boxed warning has caused the deaths of thousands of American children is serious enough to demand rigorous examination. On examination, the causal account advanced by Khan et al. is inconsistent with trial evidence, prescribing data, timing patterns, and healthcare-contact realities, and cannot account for observed suicide trends. The methodology underlying their cited evidence has been shown in peer-reviewed publications to be invalid for this purpose—publications Khan et al. possessed but did not cite.

We share the stated goal of improving mental health care and outcomes for children and adolescents. Achieving that goal requires evaluations of antidepressant therapy that are neutral, methodologically rigorous, and grounded in the totality of the evidence. The boxed warning conveys correct and clinically essential information about a real drug effect. The appropriate policy response to a confirmed safety signal is not to suppress the warning to encourage use of the drug; it is to ensure that clinicians understand it correctly, that the benefits of treatment are communicated to patients, parents, and caregivers alongside the drug’s risks, and that the healthcare system reaches more of the people at risk who are currently invisible to it. Those are the right policy questions. Khan et al. do not ask them. A more balanced and scientifically grounded assessment would better serve clinicians, patients, and regulators alike.

Readings

• Chua, K.P., A. Volerman, J. Zhang, et al., 2024, “Antidepressant Dispensing to US Adolescents and Young Adults: 2016–2022,” Pediatrics 154(3): e2024066904.
• Eurostat, 2025, “Young People—Health,” Statistics Explained, European Commission.
• Garnett, M.F., M.R. Spencer, & J.D. Weeks, 2023, “Suicide among Adults Age 55 and Older, 2021,” Data Brief No. 483, National Center for Health Statistics, US Department of Health and Human Services.
• Gibbons, R.D., C.H. Brown, K. Hur, et al., 2007, “Early Evidence on the Effects of Regulators’ Suicidality Warnings on SSRI Prescriptions and Suicide in Children and Adolescents,” American Journal of Psychiatry 164: 1356–1363.
• Hammad, T.A., T. Laughren, & J. Racoosin, 2006, “Suicidality in Pediatric Patients Treated with Antidepressant Drugs,” Archives of General Psychiatry 63(3): 332–339.
• Jureidini, J., 2007, “The Black Box Warning: Decreased Prescriptions and Increased Youth Suicide?” American Journal of Psychiatry 164(12): 1907.
• Khan, A., A. Arora, & A. Prasad, 2026, “A Tragic Unintended Consequence,” Regulation 49(1): 26–29.
• Pamer, C.A., T.A. Hammad, Y. Wu, et al., 2010, “Changes in US Antidepressant and Antipsychotic Prescription Patterns during a Period of FDA Actions,” Pharmacoepidemiology and Drug Safety 19: 158–174.
• Porter, A., K. Allison, S.B. Gokarakonda, et al., 2026, “Healthcare Utilization in the Year Before Death by Suicide: A Multinomial Approach,” Journal of Behavioral Health Services & Research 53(2): 291–296.
• Stone, M.B., 2014, “The FDA Warning on Antidepressants and Suicidality—Why the Controversy?” New England Journal of Medicine 371: 1668–1671.
• Stone, M.B., 2018, “In Search of a Pony: Sources, Methods, Outcomes, and Motivated Reasoning,” Medical Care 56: 375–381.
• Stone, M., T. Laughren, M.L. Jones, et al., 2009, “Risk of Suicidality in Clinical Trials of Antidepressants in Adults: Analysis of Proprietary Data Submitted to US Food and Drug Administration,” BMJ 339: b2880.
• Stone, M.B., Z.S. Yaseen, B.J. Miller, et al., 2022, “Response to Acute Monotherapy for Major Depressive Disorder in Randomized, Placebo Controlled Trials Submitted to the US Food and Drug Administration: Individual Participant Data Analysis,” BMJ 378: e067606.
• Ventimiglia, J., & A.H. Kalali, 2010, “Generic Penetration in the Retail Antidepressant Market,” Psychiatry (Edgmont) 7(6): 9–11.
• Walby, F.A., M.Ø. Myhre, & A.T. Kildahl, 2018, “Contact with Mental Health Services Prior to Suicide: A Systematic Review and Meta-Analysis,” Psychiatric Services 69(7): 751–759.