The Denominator They Buried

SSRIs, Pregnancy, and Teen Depression in a Culture That Mistakes Warnings for Knowledge

One of the easiest ways to lie with medicine is to hide the denominator.

You do not have to fabricate data. You do not have to invent a mechanism. You do not have to forge a clinical trial. You only have to say “SSRIs increase risk” and then refuse to say compared with what, in whom, during which exposure window, at what dose, for which drug, and against which untreated condition.

That is how public medicine becomes theater. The molecule becomes a villain. The patient becomes a symbol. The fetus becomes a prop. The teenager becomes a warning label with a pulse.

The SSRI debate has now reached that stage.

On July 21, 2025, FDA convened an expert panel on selective serotonin reuptake inhibitors and pregnancy. The panel included several critics of psychiatric medication, developmental-risk researchers, and one obstetrician-psychiatry voice speaking directly to the clinical role of SSRIs in perinatal care. ACOG responded the same day, calling the panel “alarmingly unbalanced” and warning that it failed to account for the harms of untreated perinatal mood disorders. FDA framed the event as an expert review of SSRI safety in pregnancy; ACOG framed the backlash as a threat to evidence-based access for pregnant patients with depression and anxiety.

Both the public and the press wanted the primitive question: Are SSRIs safe?

That question should be rejected.

The rational question is harder: Safe compared with what?

Compared with a healthy pregnant woman who never had major depression? Compared with a woman with recurrent severe depression who relapsed after previous discontinuation? Compared with psychotherapy that is available next week? Compared with psychotherapy that exists only on a referral sheet? Compared with late-pregnancy exposure? Compared with first-trimester exposure? Compared with a teen prescribed fluoxetine plus cognitive behavioral therapy and weekly monitoring? Compared with a teen handed medication and told to come back in three months?

There is no single “SSRI risk.” There are only risk structures.

The first duty of rationalism is not to pick a tribe. It is to stop accepting corrupted reference classes.

The SSRI Is Not the Unit of Analysis

The word “SSRI” pretends to simplify. It actually conceals.

Sertraline is not fluoxetine. Fluoxetine is not paroxetine. Escitalopram is not citalopram. A first-trimester exposure is not a third-trimester exposure. A prescription fill is not ingestion. A neonatal respiratory-support endpoint is not autism. A teenager with mild depressive symptoms is not a teenager with major depression, self-harm, insomnia, school collapse, and bipolar family history.

The public debate compresses all of these into one object because one object is easier to weaponize.

Science cannot do that.

A defensible claim about SSRIs must state the patient population, indication, drug, dose, gestational or developmental timing, comparator, endpoint, absolute risk, and monitoring environment. If one of those terms is missing, the claim is not yet a claim. It is a mood.

The collapse of this precision explains why the argument has become so stupid. One side says SSRIs are safe in pregnancy. The other says SSRIs harm babies. Both sentences fail unless they name the outcome.

For congenital heart defects, the best-adjusted evidence does not support a large class-wide teratogenic signal. For late-pregnancy neonatal adaptation, the signal looks more real. For persistent pulmonary hypertension of the newborn, the outcome is serious but rare, and absolute risk matters. For autism and ADHD, crude association is not causation, and better comparator designs weaken the causal story. For teen depression, medication can show benefit in some trial structures, but the benefit is modest, diagnosis-specific, and inseparable from monitoring.

That is not equivocation. That is the map.

The slogan is not the map.

Pregnancy Is a Two-Patient Problem

Pregnancy does not make drug safety simpler. It makes false simplicity more dangerous.

Every decision involves at least two patients: the pregnant woman and the fetus or newborn. The moral error in most public SSRI debate is that one of them disappears.

The anti-SSRI version erases the mother. It treats untreated depression as a clean baseline, as if severe psychiatric illness were merely the absence of medication. That is false. Antenatal depression has been associated with preterm birth, low birth weight, and intrauterine growth restriction; one 2021 updated meta-analysis reported elevated risks for preterm birth, low birth weight, and IUGR, while also noting that adjustment for confounders changed some estimates.

The pro-SSRI version can erase the newborn. It treats reassurance as if it were consent. That is also false. Late-pregnancy SSRI exposure has measurable neonatal signals, and patients deserve the numbers, not a vibe.

The only honest position begins here: untreated illness is an exposure, and medication is an exposure.

The decision is not between risk and no risk. It is between risk portfolios.

ACOG’s perinatal mental health materials make that point clinically: clinicians should counsel on treated versus untreated illness, may initiate pharmacotherapy or refer for behavioral health care when indicated, should use validated tools to monitor response, and should not withhold or discontinue medication solely because a patient is pregnant or lactating.

That guidance does not mean every pregnant woman with sadness needs sertraline.

It means pregnancy status alone is not an argument.

Birth Defects: The Comparator Changes the Story

The most important number in an SSRI pregnancy study is often not the risk ratio. It is the comparator group.

If a study compares SSRI-exposed pregnant women to the general unexposed population, it risks laundering depression severity, anxiety, smoking, substance use, obesity, diabetes, socioeconomic stress, health-care access, surveillance intensity, and comorbidity into the drug effect.

That is confounding by indication. It is not a footnote. It is the battlefield.

Huybrechts and colleagues studied 949,504 pregnant women in Medicaid and found that apparent associations between antidepressant use and cardiac defects attenuated with better adjustment. The unadjusted relative risk for any cardiac defect with SSRI use was 1.25. Restricted to women with depression, it fell to 1.12. In the fully adjusted depression-restricted analysis, it fell to 1.06 with a 95% confidence interval crossing null. The study also did not find significant associations for paroxetine with right ventricular outflow tract obstruction or sertraline with ventricular septal defects after analysis. PMID: 24941178; DOI: 10.1056/NEJMoa1312828.

A 2018 BMC Medicine meta-analysis of 29 cohort studies including more than 9 million births found small increased relative risks for major congenital anomalies and congenital heart defects in broad analyses. But when the analysis restricted comparison to women with psychiatric diagnoses, the major congenital anomaly estimate was 1.04 and the congenital heart defect estimate was 1.06, both statistically nonsignificant. PMID: 30415641; DOI: 10.1186/s12916-018-1193-5.

That is the denominator problem in plain sight.

The headline wants “SSRIs increase birth defects.”

The better reading is: broad comparisons show small signals; psychiatric-comparator analyses weaken the claim of a substantial class-wide teratogenic effect.

That does not prove no molecule-specific risk. Paroxetine has carried historical concern. Some individual-drug and subtype estimates remain unstable. Rare outcomes can be underpowered even in large datasets when subdivided by drug, dose, defect type, and timing. But the evidence does not justify the popular move from “small observational signals exist” to “SSRIs are proven teratogens.”

A rationalist does not need institutional reassurance.

A rationalist needs the right control group. (Article continues below…)

Late Pregnancy: The Neonatal Signal Is Harder to Dismiss

The strongest short-term pregnancy concern does not come from the congenital-malformation storyline. It comes from late gestation.

A 2024 population-based cohort study of 280,090 term infants found delayed neonatal adaptation in 11.2% of infants exposed to SSRIs after 20 weeks versus 4.4% of unexposed infants. After adjustment, SSRI exposure remained associated with delayed neonatal adaptation, with adjusted odds ratio 2.14. The study reported dose dependence and found higher associated risks for escitalopram and fluoxetine than for some other SSRIs. PMID: 38071585; DOI: 10.1136/archdischild-2023-326049.

That is a real signal.

It is not proof of permanent injury. It is not a congenital-malformation result. It is not autism. It is not a reason to terrorize pregnant women. But it is also not nothing.

The endpoint matters. “Delayed neonatal adaptation” in that study included low five-minute Apgar score, resuscitation at birth, or NICU admission for respiratory support. Those outcomes belong in the same clinical neighborhood, but they do not carry identical severity or long-term meaning. Composite endpoints can reveal a pattern while still blurring the moral weight of their parts.

This is where public medicine often fails. It either inflates a short-term neonatal signal into irreversible developmental catastrophe, or it dilutes it into “manageable side effects.”

Neither move respects the patient.

The right sentence is: late-pregnancy SSRI exposure appears associated with higher risk of neonatal adaptation problems; that risk should be discussed, anticipated, and weighed against maternal relapse risk and available alternatives.

A risk that can be prepared for is not a risk that can be ignored.

PPHN: Serious, Rare, and Easy to Misuse

Persistent pulmonary hypertension of the newborn is the kind of outcome that makes public communication treacherous. It is serious. It can be life-threatening. It is also rare.

A 2014 BMJ systematic review and meta-analysis found that late-pregnancy SSRI exposure was associated with persistent pulmonary hypertension of the newborn, with odds ratio 2.50. But the same analysis estimated the absolute risk difference at roughly 2.9 to 3.5 additional cases per 1000 infants, with an estimated 286 to 351 women needing late-pregnancy SSRI exposure for one additional case. DOI: 10.1136/bmj.f6932.

A later JAMA Medicaid cohort study found PPHN diagnoses in 20.8 per 10,000 unexposed infants and 31.5 per 10,000 SSRI-exposed infants before deeper adjustment. After restricting to women with depression and applying high-dimensional propensity-score adjustment, the SSRI odds ratio attenuated to 1.10, with a confidence interval crossing null; when the outcome was restricted to primary PPHN, the adjusted SSRI odds ratio was 1.28. DOI: 10.1001/jama.2015.5605; PMID: 26034955.

This is how rational risk works.

A relative risk can double and still describe a rare event. A rare event can still matter. Both sentences are necessary. If either sentence disappears, someone is manipulating you.

Postpartum Hemorrhage: Modest Signal, Practical Consequence

SSRIs may also matter at delivery because serotonin participates in platelet function. The evidence here is not a cartoon, but it is clinically relevant.

A 2016 systematic review and meta-analysis found increased postpartum hemorrhage risk among antidepressant users, including SSRI users, with pooled SSRI relative risk 1.20. The increased risk was more evident for current and recent late-gestation exposure than for past exposure. PMID: 27637098; DOI: 10.1016/j.jpsychires.2016.09.001.

That does not mean “stop SSRIs before birth” as a universal rule. Discontinuation can destabilize psychiatric illness. Late pregnancy and postpartum already carry relapse risk. But obstetric teams can plan for bleeding risk. Medication history should not be a secret from the delivery team.

The practical conclusion is boring and therefore valuable: document the exposure, prepare the team, and stop pretending that risk counseling equals fearmongering.

Autism and ADHD: The Dirtiest Causal Terrain

Autism and ADHD claims are where the SSRI debate most reliably degrades into association laundering.

A crude observational association is easy to obtain: compare children exposed to antidepressants in utero with all unexposed children, then report a higher rate of neurodevelopmental diagnosis. That design cannot answer the causal question. It is too vulnerable to maternal psychiatric illness, paternal psychiatric history, genetics, environmental stress, surveillance intensity, health-care contact, and diagnostic ascertainment.

A 2018 BMC Medicine systematic review examined antidepressant use during pregnancy and risk of autism spectrum disorder and ADHD, then showed how design choice changes interpretation. The broad adjusted estimate for autism remained elevated in classical exposed-versus-unexposed comparisons, but alternative approaches—pre-pregnancy maternal exposure, paternal exposure, maternal affective-disorder history, and sibling comparisons—undermined the simple causal story. The sibling-comparison estimates were not consistent with a strong direct causal effect. DOI: 10.1186/s12916-017-0993-3.

This is not a minor technical adjustment. It is the core of the question.

If prenatal SSRI exposure truly causes autism independent of inherited and familial risk, the signal should survive stronger causal tests. When it weakens under sibling and negative-control designs, the responsible conclusion is not “SSRIs cause autism.” The responsible conclusion is: the crude association is contaminated.

That does not prove zero developmental effect. Serotonin matters in development. Animal models matter for mechanism generation. Cohort studies matter for surveillance. But mechanism plus association does not equal causation.

The weaker the comparator, the louder the headline.

Discontinuation Is an Intervention

The most dishonest phrase in this debate is “just stop.”

Stopping a medication is not the removal of an intervention. It is a new intervention, with its own exposure profile.

Cohen and colleagues’ 2006 JAMA study remains central because it examined relapse during pregnancy among women who maintained versus discontinued antidepressant treatment. The study reported higher relapse risk after discontinuation, but it also carries a serious disclosure problem: JAMA later published a correction for omitted financial disclosures among authors. PMID: 16449615; DOI: 10.1001/jama.295.5.499.

That is how it should be handled.

A conflicted study is not automatically false. An undisclosed conflict is automatically relevant.

The relapse question cannot be dismissed merely because industry-adjacent investigators studied it. Women with recurrent major depression do relapse. Pregnancy does not immunize the brain against psychiatric illness. But the evidence must carry its disclosure history with it.

The right clinical frame is individualized relapse risk: number of prior episodes, severity, suicidality, hospitalization, prior taper failure, current stability, psychotherapy access, social support, postpartum vulnerability, and patient values.

The wrong frame is “pregnant women should stop psychiatric drugs.”

That is not caution. It is negligence wearing the costume of purity.

Teens: The Warning Label Is Not the Whole Map

The adolescent SSRI debate has its own denominator problem.

The FDA pediatric antidepressant warning was based on short-term placebo-controlled trials involving more than 4,400 pediatric patients. FDA reported suicidal thinking or behavior in 4% of antidepressant-treated patients versus 2% on placebo, with no suicides in those trials. The agency directed close observation for worsening, suicidality, agitation, irritability, and unusual behavioral changes, especially during the first months of treatment and around dose changes.

That warning is real.

It is also not a complete causal map.

Suicidal ideation, suicide attempt, completed suicide, akathisia, agitation, activation, bipolar conversion, depressive worsening, impulsivity, and family crisis are not interchangeable endpoints. A warning label bundles because regulators must communicate across drugs and populations. Scientists and clinicians must unbundle.

The black-box warning should not be dismissed as anti-drug panic. It should not be weaponized as proof that antidepressants make every teen suicidal.

It is a demand for disciplined care.

The failure is not that SSRIs require monitoring. The failure is prescribing them into systems that cannot monitor.

Benefit Is Diagnosis-Specific

Teen SSRI evidence depends heavily on indication.

Bridge and colleagues’ 2007 JAMA meta-analysis of pediatric antidepressant trials found that response benefits differed across major depressive disorder, obsessive-compulsive disorder, and non-OCD anxiety disorders. The number needed to treat was 10 for major depressive disorder, 6 for OCD, and 3 for non-OCD anxiety disorders. The pooled risk difference for suicidal ideation or suicide attempt was 0.7%, corresponding to number needed to harm 143, with no completed suicides. DOI: 10.1001/jama.297.15.1683.

That means the same warning sits on top of unequal benefit.

An SSRI for pediatric OCD is not the same decision as an SSRI for adolescent major depression. A panic-disorder case is not an anhedonic, suicidal depression case. A teen with mild symptoms and intact function is not a teen who cannot sleep, cannot attend school, cannot eat, and is cutting.

The public wants one rule.

Medicine needs triage.

TADS: Medication Alone Was Never the Best Story

The Treatment for Adolescents With Depression Study remains important because it did not merely ask whether fluoxetine works. It compared fluoxetine, CBT, combination treatment, and placebo.

At 12 weeks, response rates were 71.0% for fluoxetine plus CBT, 60.6% for fluoxetine alone, 43.2% for CBT alone, and 34.8% for placebo. The authors concluded that combination treatment offered the most favorable benefit-risk tradeoff for adolescents with major depressive disorder. DOI: 10.1001/jama.292.7.807.

That result does not vindicate lazy prescribing. It indicts it.

The strongest practical lesson from TADS is not “give teens Prozac.” It is: the treatment container matters.

Medication plus therapy is not the same intervention as medication alone. Medication plus weekly monitoring is not the same intervention as medication plus silence. Medication after bipolar screening is not the same intervention as medication without asking why the teen sometimes sleeps three hours and feels invincible.

Guidelines reflect this. GLAD-PC recommends active monitoring for mild adolescent depression, evidence-based medication and psychotherapy for moderate or severe depression, close side-effect monitoring, consultation or comanagement with mental health specialists, and ongoing tracking of outcomes. PMID: 29483201; DOI: 10.1542/peds.2017-4082.

A prescription is not a treatment plan.

A treatment plan has measurement, follow-up, family instruction, adverse-event surveillance, crisis response, dose discipline, and an exit strategy.

The Fluoxetine Problem Has Become Less Settled

For years, fluoxetine has occupied the privileged position in adolescent depression. It had the best-known trial evidence and guideline support. That position now deserves renewed pressure.

Cochrane’s 2021 network meta-analysis concluded that newer-generation antidepressants may reduce symptoms compared with placebo, but the reductions were small and may not be experienced as important by youth, parents, caregivers, or clinicians. The review stressed that suicidal youth were often excluded from trials and that close monitoring of treatment effects and suicide-related outcomes remains critical. DOI: 10.1002/14651858.CD013674.pub2.

A 2026 Journal of Clinical Epidemiology commentary by Plöderl, Lyus, Horowitz, and Moncrieff went further, arguing that fluoxetine’s estimated efficacy in pediatric depression has declined over time into clinical equivalence with placebo, likely reflecting novelty bias or expectancy effects, and that guideline recommendations have not adequately acknowledged this. PMID: 41548966; DOI: 10.1016/j.jclinepi.2025.112016.

That does not instantly erase TADS. It does not mean no adolescent ever benefits from fluoxetine. It does mean the old confidence needs revision.

A 2025 BMC Psychiatry network meta-analysis, searching trials through October 2024, reported statistically significant improvement on some scales for fluoxetine, sertraline, and other medications, while also warning that limited study numbers, indirect comparisons, and type I and type II error risks require cautious interpretation. DOI: 10.1186/s12888-025-06941-x.

So the live evidence is no longer cleanly summarized by “fluoxetine works best.”

A better summary is: fluoxetine has historically had the strongest adolescent depression evidence, but newer syntheses challenge the clinical magnitude of that benefit; any use must be severity-justified, monitored, and preferably embedded in psychotherapy and measurement-based care.

That is not anti-medication.

It is anti-ritual.

The Real Scandal Is System Design

The SSRI debate keeps pretending that the central object is the drug.

Often, the central object is the system around the drug.

A monitored SSRI trial in a severely depressed teen with family involvement, CBT, dose discipline, bipolar screening, suicidality tracking, and rapid access to care is one intervention.

An SSRI prescription written during a rushed visit, with no therapy access, no family education, no early follow-up, no side-effect tracking, and no suicide plan is another intervention.

They share a molecule. They do not share a safety profile.

The same applies in pregnancy.

A pregnant woman with recurrent severe depression, prior hospitalization, prior relapse after discontinuation, current stability on sertraline, an obstetric team aware of the exposure, neonatal planning for adaptation symptoms, and postpartum psychiatric follow-up is in one decision class.

A pregnant woman with mild first-episode symptoms, no prior severe course, immediate access to psychotherapy, no suicidality, and strong support is in another.

A public debate that gives both women the same answer has abandoned medicine.

How Bad Evidence Wins

Bad evidence wins by exploiting urgency.

It says: babies are at risk. Teens are at risk. Mothers are at risk. Act now.

But urgency does not repeal inference.

Bad evidence wins through confounding by indication: sicker patients get medication, then their outcomes get attributed to the medication.

It wins through comparator corruption: medicated depressed women get compared with the general unmedicated population, not with comparable women with psychiatric illness.

It wins through window bias: first-trimester exposure gets used to imply late-neonatal outcomes, or late-pregnancy exposure gets used to imply organogenesis.

It wins through endpoint bundling: low Apgar, NICU respiratory support, congenital malformation, autism, ADHD, suicide attempt, and suicidal ideation get melted into a single blob called “harm.”

It wins through surveillance bias: children of medicated mothers and teens in treatment get watched more closely, and what is watched gets diagnosed.

It wins through prescription-fill misclassification: a dispensed prescription becomes ingestion, and absence of a recorded prescription becomes no exposure.

It wins through absolute-risk suppression: rare outcomes are discussed only as relative risks because relative risks frighten better.

It wins through undisclosed conflicts: sponsor bias, academic prestige, institutional defensiveness, and anti-psychiatry incentives all distort interpretation.

And it wins through moral simplification: one side owns babies, the other owns mothers; one side owns teen safety, the other owns teen treatment.

That is not science.

That is custody battle epistemology.

The Rational Decision Grid

No article should tell a specific pregnant woman or teenager to start, stop, continue, taper, or switch medication. That belongs in clinical care with the patient, family where appropriate, prescriber, obstetric or pediatric team, and mental health support.

But public reasoning can still define the decision grid.

For pregnancy, the relevant questions are not “SSRI or no SSRI?” They are: What is the diagnosis? How severe is it? How many prior episodes occurred? Did prior discontinuation cause relapse? Is there suicidality? Which drug? What dose? Which trimester? Is this first-trimester malformation risk, late-pregnancy neonatal adaptation risk, postpartum hemorrhage risk, or relapse risk? Is psychotherapy available in reality or only in principle? What monitoring exists postpartum? What does the patient value most after seeing absolute risks?

For teens, the relevant questions are: Is this mild depression, moderate depression, severe depression, OCD, anxiety, or something else? Has bipolar disorder been screened? Is there self-harm? What is baseline suicidality? Is therapy available? Will caregivers monitor daily during initiation and dose changes? Is there follow-up within weeks, not months? What scale will track response? What adverse effects trigger action? What is the plan if there is no improvement after six to eight weeks? What is the taper plan if stopping becomes appropriate?

A treatment system that cannot answer those questions has no business acting confident.

The Moral Error

The SSRI fight keeps using vulnerable people as shields.

Pregnant women become symbols of fetal risk. Teens become symbols of pharmaceutical overreach. Mothers with severe depression become symbols of access. Children with neurodevelopmental diagnoses become symbols of regulatory failure.

The actual people vanish.

A pregnant woman is not a vessel for a teratology argument. A fetus is not a rhetorical inconvenience. A depressed teenager is not a black-box warning. A newborn with respiratory distress is not an acceptable externality. A mother with recurrent major depression is not a lifestyle variable. A family trying to keep a suicidal adolescent alive is not a talking point in a culture war about psychiatry.

The rational position is not to defend SSRIs.

The rational position is not to fear SSRIs.

The rational position is to stop hiding the denominator.

SSRIs are tools used under uncertainty. In pregnancy, they require two-patient reasoning. In teenagers, they require developmental reasoning. In both, they require absolute risks, honest comparators, molecule-specific claims, exposure-window discipline, endpoint separation, conflict disclosure, and active monitoring.

Anything less is not informed consent.

It is narrative capture with a prescription pad.