were kids on ssris nearly twice as likely to respond in the 2025 pediatric ocd meta-analysis?
“kids taking SSRIs turned out to be nearly twice as likely to achieve a treatment response, which they define as at least a 35% reduction in symptoms.”
a host, a psychiatrist said on The Carlat Psychiatry Podcast, 2026-10-12, at 04:04. hear it in context → the speaker cited: a 2025 study by Cohen and colleagues.
the episode: How Much Do SSRIs Really Help Kids with OCD? (at 04:04). we link the show's own page; we do not re-host audio.
first, the part that holds
the study is real and was described correctly: a 2025 meta-analysis of individual participant data in the journal of the american academy of child and adolescent psychiatry defined response as a reduction of at least 35% on the children's yale-brown obsessive-compulsive scale (cy-bocs), and found a response odds ratio of 1.89 in favor of ssris over placebo. "nearly twice" is a fair reading of 1.89.
what we found when we went and looked
we read the paper's abstract on pubmed's record service and the full version of record. across 4 placebo-controlled trials with 614 children and adolescents aged 6 to 18, ssris produced 3.0 more points of cy-bocs improvement than placebo (95% ci 2.5 to 3.5, hedges' g 0.38). analysis of response (at least a 35% cy-bocs reduction) gave an odds ratio of 1.89 (95% ci 1.45 to 2.45). the number is correct. the trouble is what it measures: 1.89 is a ratio of odds, and "nearly twice as likely" reads it as a ratio of probabilities. whenever some placebo patients respond, the ratio of probabilities comes out smaller than the odds ratio, so the drug's advantage in "likelihood" terms is less than double.
assertion by assertion
| assertion | verdict | what the record says |
|---|---|---|
| the speaker described a 2025 meta-analysis by a first author and colleagues comparing ssris with placebo in children | confirmed | a systematic review and individual-participant-data meta-analysis of randomized, placebo-controlled ssri trials for pediatric ocd, published in the journal of the american academy of child and adolescent psychiatry in july 2025 (64(7):775-785, doi 10.1016/j.jaac.2025.01.001). it covered 4 trials with 614 patients (319 on an ssri, 295 on placebo). the authors say this is 86% of everyone ever enrolled in double-blind placebo-controlled ssri trials for pediatric ocd. contracts with the drug makers meant the paper could not name which ssri was used in which trial. [1][2] |
| response was defined as at least a 35% reduction in cy-bocs symptoms | confirmed | the abstract says: "odds ratio (or) with ≥35% cy-bocs reduction was used as the responder outcome measure." the full text measures the change from baseline to week 12, or to the visit closest to week 12. [1] |
| children on ssris were nearly twice as likely as those on placebo to achieve response | overstated | the abstract says: "analysis of response showed an odds ratio of 1.89 (95% ci 1.45-2.45)." the one-stage model gave 1.91 (1.35 to 2.72). but these are odds ratios, not risk ratios. "twice as likely" in plain speech means twice the probability, and when the placebo response rate is above zero, the probability ratio is always smaller than the odds ratio. the main text we read does not report response rates for each arm, so we cannot give the exact probability ratio. for illustration only: if 20% to 30% of the placebo group had responded, an odds ratio of 1.89 would mean roughly 1.5 to 1.6 times as likely, not about 2. the direction is right and statistically significant. the size, in the words used, is inflated. [1] |
| context: the average benefit over placebo was small, and it shrank as baseline severity rose | confirmed | the abstract describes 3.0 cy-bocs points as "a small effect size (0.38 hedges' g)" and reports that baseline severity "was correlated negatively with odds ratio for response (β = -0.92, p = .0074)." the full text puts the minimal important difference for the cy-bocs at 5.1 points, above the 3.0-point average gap. in a post hoc split by severity, the response odds ratio was significant only for mild ocd (2.50, 95% ci 1.16 to 5.37). it was not significant for moderate (1.68, 0.67 to 4.17) or severe ocd (1.61, 0.90 to 2.89). [1] |
both directions
for the speaker: every factual anchor in the claim checks out: the year, the study design, the comparison with placebo, the 35% response threshold, and the 1.89 figure that "nearly twice" rounds. the confidence interval (1.45 to 2.45) is entirely above 1, so the advantage over placebo holds up statistically. it held in the one-stage model (1.91) and in sensitivity analyses using completers only and multiple imputation. the authors rated the certainty of evidence for the response outcome as high. a response gap at the 35% threshold is also the kind of result a parent or patient can picture more easily than a mean point difference.
against: "nearly twice as likely" turns an odds ratio into a probability ratio, and a probability ratio here has to be smaller than 1.89, probably by a lot. the paper's own framing is more cautious than the soundbite. the authors call the effect "small" (hedges' g 0.38) and "clinically modest." the average 3.0-point gap over placebo is below the 5.1-point minimal important difference they cite for children. the response advantage was weaker in sicker children: it was significant only in the mild-severity subgroup, and the moderate and severe subgroups did not reach significance (post hoc analysis). a listener could take "nearly twice as likely" to mean the drug doubles a child's chances of getting better. the study does not show that.
what this cannot say
the main text we read does not report response percentages for each arm, so we could not compute the exact probability ratio. our 1.5-to-1.6 figure is an illustration based on assumed placebo response rates, not a number from the paper. the supplement may have the arm rates. we did not read it. the trials ran 10 to 13 weeks, began between 1991 and 2000, took place almost entirely in the united states, and enrolled about 88% white participants. two small trials (combined n = 99) were left out because their data could not be obtained. the paper could not say which ssri was tested in which trial, so nothing here applies to any single drug. we read the full text in the authors' institutional repository copy of the version of record, which our sourcing rules do not let us link. every number in the assertion rows marked with source 1 also appears in the pubmed abstract except the one-stage odds ratio, the subgroup figures and the 5.1-point minimal important difference, which come from the full text. we heard the claim as transcribed and did not listen to the rest of the episode, which may have given this context.
sources, each one fetched and read
- systematic review and meta-analysis of individual participant data: randomized, placebo-controlled trials of selective serotonin reuptake inhibitors for pediatric obsessive-compulsive disorder. j am acad child adolesc psychiatry, 2025 jul;64(7):775-785. doi 10.1016/j.jaac.2025.01.001. pmid 39799995 (pubmed record via ncbi e-utilities) What we read there: "odds ratio (or) with ≥35% cy-bocs reduction was used as the responder outcome measure." "data were obtained from 4 studies comprising 614 patients." "meta-analysis showed reduction of 3.0 cy-bocs points compared with placebo (95% ci 2.5-3.5), corresponding to a small effect size (0.38 hedges' g). analysis of response showed an odds ratio of 1.89 (95% ci 1.45-2.45). of all possible modifiers, severity was correlated negatively with odds ratio for response (β = -0.92, p = .0074)." Fetched 2026-10-15 (HTTP 203). https://pubmed.ncbi.nlm.nih.gov/39799995/
- editorial: meta-analysis of individual participant data from trials of selective serotonin reuptake inhibitors for pediatric obsessive-compulsive disorder: a gold standard approach. j am acad child adolesc psychiatry, 2025 jul;64(7):768-769. doi 10.1016/j.jaac.2025.03.016. pmid 40139344 What we read there: "pediatric obsessive-compulsive disorder (ocd) is a common (2% prevalence) and potentially debilitating condition." the editorial introduces the 2025 individual-participant-data meta-analysis of ssris against placebo and its look at baseline variables that moderate outcome. Fetched 2026-10-15 (HTTP 203). https://pubmed.ncbi.nlm.nih.gov/40139344/
speakers are described by role, not named. one sentence is quoted under fair use, with attribution and a link to the original at the timestamp; we do not re-host audio or video. nothing on this page is a reason to start, stop or change a medication. if you have the file behind anything marked unverifiable, the research board is where people can answer. adam