do antidepressants beat placebo by only one or two points on depression scales?
“if you improve on one of the major scales that's used by one or two points on a 60-point scale, it may be enough to show statistical significance”
a guest, a historian of psychiatry (Andrew Scull) said on Lex Fridman Podcast, 2026-09-21, at 09:24. hear it in context →
first, the part that holds
the speaker has the size right. on the scale most antidepressant trials used, the average gap between drug and placebo is about 1.8 to 2.1 points. that number is statistically solid, and it falls below the clinical-benefit cutoffs those same papers cite.
what we found when we went and looked
we read three analyses built on trial data. the first used the trials drug companies submitted to the FDA; it found a 1.80-point gap on the Hamilton scale (HRSD), against a 3-point benefit threshold set by NICE, England's health guidance body. a 2022 analysis of individual patients in 232 FDA-submitted trials (73,388 participants) found a 1.75-point gap (95% CI 1.63 to 1.86). a 2020 meta-analysis found 2.07 points on the 17-item Hamilton scale and 2.99 points on the MADRS (Montgomery-Åsberg scale). one detail is off: the Hamilton scale runs 0-52, not 0-60. the scale that does run 0-60, the MADRS, shows a gap of about 3 points. both scales give the same standardized effect of about 0.3, and the 2022 paper found the average hides a group, about 15% of participants, who get a large benefit beyond placebo.
assertion by assertion
| assertion | verdict | what the record says |
|---|---|---|
| on the main depression rating scale, antidepressants beat placebo on average by only about one to two points | confirmed | FDA-submitted trials: 9.60 points of improvement on drug vs 7.80 on placebo, a 1.80-point HRSD gap. individual-patient analysis of 232 FDA trials: 1.75 points (95% CI 1.63-1.86), in Hamilton-equivalent units. 2020 meta-analysis: 2.07 points (1.76-2.37) on the 17-item Hamilton scale. limit: these are adult acute-phase trials of 6-12 weeks, mostly drawn from regulatory files. [1][2][4] |
| the scale in question is a 60-point scale | misattributed | the Hamilton scale (17-item), which produced the 1.75-2.07 figures, has a possible range of 0-52. the MADRS has a 0-60 range, and on it the average gap was 2.99 points (2.24-3.74) across 28 trials. the 1-2 point figure belongs to the 52-point scale. on the 60-point scale it is closer to 3. as a share of the scale it is the same: standardized effect 0.27 vs 0.30. [4] |
| a gap this small can still reach statistical significance | confirmed | every pooled estimate we read has a confidence interval that excludes zero. a network meta-analysis of 522 trials (116,477 participants) found all 21 antidepressants more effective than placebo on response, with odds ratios from 1.37 to 2.13. [2][3][4] |
| that average gap falls short of clinical significance | confirmed | the 1.80-point gap is below the 3-point clinical-benefit threshold set by NICE. the 2020 meta-analysis cites published estimates that a clinician needs about 7 Hamilton points or about 8 MADRS points to see minimal improvement, and calls the effects 'marginally small and with questionable importance for the average patient.' limit: where to set the threshold is disputed, and an average says nothing about any one patient. [1][4] |
both directions
for the speaker: three independent analyses of large trial sets agree, each using a different method. trial-level FDA data gives 1.80 points, individual-patient FDA data gives 1.75, and published trials give 2.07 on the 17-item Hamilton scale. the 2008 analysis found the gap only reached the NICE 3-point threshold for patients at the very top of the very-severe range. it also found that the growth with severity came from a weaker placebo response, not a stronger drug response. the 2020 authors conclude the effects are of questionable importance for the average patient.
against: the 1-2 point figure is a Hamilton-scale (0-52) number. on the 60-point MADRS the gap is about 3 points, which meets or exceeds some of the smaller clinical cutoffs. the average is not a uniform small effect. the 2022 individual-patient analysis found three response groups, and 24.5% of people on drug had a large response vs 9.6% on placebo. the authors read that as about 15% of participants getting a substantial benefit beyond placebo. so 'about 2 points on average' does not mean 'about 2 points for everyone'. the 522-trial network meta-analysis also found every drug studied beat placebo on response rates.
what this cannot say
'clinically significant' has no single agreed cutoff. the 3-point NICE figure and the 7-8 point minimal-improvement figures come from different methods, and other published estimates are lower, so the verdict on clinical significance depends on which cutoff you use. the data are mostly short adult acute-treatment trials. they say little about long-term use, relapse prevention, children, or any one person's response. we read abstracts, plus specific passages pulled from the full text of the 2020 paper, not every table. publisher pages for the BMJ and Lancet papers blocked us, so we read those abstracts through the PubMed record. we did not check the episode's wider context beyond the quoted line.
sources, each one fetched and read
- Initial severity and antidepressant benefits: a meta-analysis of data submitted to the Food and Drug Administration. PLoS Medicine, 2008. doi:10.1371/journal.pmed.0050045 What we read there: "weighted mean improvement was 9.60 points on the HRSD in the drug groups and 7.80 in the placebo groups, yielding a mean drug–placebo difference of 1.80"; NICE defined clinical benefit as a 3-point HRSD difference, met only at the upper end of very severe depression. Fetched 2026-10-06 (HTTP 200). https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.0050045
- 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 2022;378:e067606. doi:10.1136/bmj-2021-067606 (PubMed record 35918097) What we read there: 232 trials, 73,388 participants; random effects mean difference favoring drug 1.75 points (95% CI 1.63 to 1.86) in HAMD17-equivalent units; large responses 24.5% on drug vs 9.6% on placebo; 'about 15% of participants have a substantial antidepressant effect beyond a placebo effect.' Fetched 2026-10-06 (HTTP 203). https://pubmed.ncbi.nlm.nih.gov/35918097/
- Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis. Lancet 2018;391(10128):1357-1366 (PubMed record 29477251) What we read there: 522 trials, 116,477 participants; all 21 antidepressants more efficacious than placebo, odds ratios for response ranging from 1.37 to 2.13. Fetched 2026-10-06 (HTTP 203). https://pubmed.ncbi.nlm.nih.gov/29477251/
- Efficacy of new-generation antidepressants assessed with the Montgomery-Asberg Depression Rating Scale, the gold standard clinician rating scale: a meta-analysis of randomised placebo-controlled trials. PLoS ONE, 2020. doi:10.1371/journal.pone.0229381 (PMC7043778) What we read there: drug-placebo raw difference 2.07 (1.76-2.37) points on the HDRS-17 (range 0-52) and 2.99 (2.24-3.74) on the MADRS (range 0-60); standardized effects 0.27 and 0.30; minimal clinician-detectable improvement about 7 HDRS-17 or about 8 MADRS points; effects 'marginally small and with questionable importance for the average patient.' Fetched 2026-10-06 (HTTP 200). https://pmc.ncbi.nlm.nih.gov/articles/PMC7043778/
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