is the evidence that cbt works for schizophrenia poor?
“If I look at the literature on CBT, for example, and its uses in cases of schizophrenia, the evidence is pretty poor that it works.”
a guest, a historian of psychiatry said on Lex Fridman Podcast, 2026-09-21, at 08:54. hear it in context → the speaker cited: the literature on CBT.
first, the part that holds
the speaker's description holds up. the literature's own quality grades rate the evidence for CBT in schizophrenia as weak to low certainty, and the effects it finds are small. they get smaller still when the people scoring symptoms don't know who got CBT.
what we found when we went and looked
we read four primary syntheses. a 2014 meta-analysis of 34 trials in the british journal of psychiatry found small end-of-treatment effects: SMD −0.33 for overall symptoms, −0.25 for positive symptoms and −0.13 for negative symptoms. in trials where outcome raters were blinded, the effect on positive symptoms fell to −0.08, against −0.57 where raters weren't blinded. a 2024 umbrella review of 26 meta-analyses graded none of 70 comparisons as convincing or highly suggestive. 57.1% were non-significant, and the authors called on guidelines to reconsider their recommendations. a cochrane review of 36 trials (3,542 people) found no clear advantage for CBT over other psychosocial therapies, mostly on low or very low quality evidence. one caveat: none of this shows that CBT does nothing. a 2018 network meta-analysis found CBT 'seems to be effective' on positive symptoms, with lower-to-medium effect sizes, at moderate to very low confidence.
assertion by assertion
| assertion | verdict | what the record says |
|---|---|---|
| the evidence that CBT is effective for schizophrenia is poor | confirmed | the 2024 umbrella review (26 meta-analyses, 70 comparisons) graded 0% of comparisons class I (convincing) or class II (highly suggestive). 7.1% were class III (suggestive), 35.7% class IV (weak) and 57.1% non-significant. the cochrane review rated its evidence mostly low or very low quality, and the 2018 network meta-analysis rated its confidence moderate to very low. limitation: 'poor' describes certainty and size. it doesn't mean the pooled effects are zero, and at end of treatment they are mostly statistically significant. [2][3] |
| the measured benefit of CBT on schizophrenia symptoms is small, and smaller under blinded assessment | confirmed | pooled end-of-treatment SMDs were −0.33 overall (34 studies), −0.25 positive (33) and −0.13 negative (34). with blinded raters, overall symptoms went from −0.62 to −0.15 (P = 0.001) and positive symptoms from −0.57 to −0.08 (P<0.001). limitation: this is subgroup moderator analysis across trials, not a head-to-head test. [1] |
| CBT has no clear advantage over other, simpler psychosocial therapies for schizophrenia | confirmed | cochrane, 36 RCTs, 3,542 participants: long-term relapse RR 1.05 (95% CI 0.85 to 1.29), long-term rehospitalisation RR 0.96 (0.82 to 1.14), no clear difference in long-term mental state, social functioning or quality of life. conclusion: 'no clear and convincing advantage.' evidence quality was mostly low or very low. [3] |
| popular framing: CBT does not work at all for schizophrenia | overstated | the speaker did not say this. if it's repeated in this form, the record doesn't support it. the 2018 network meta-analysis (53 RCTs, 4,068 people, all also on antipsychotics) found CBT reduced positive symptoms compared with inactive control (SMD −0.29), treatment as usual (−0.30) and supportive therapy (−0.47). the umbrella review describes a small-to-medium effect at end of treatment that becomes inconsistent at follow-up. [2] |
both directions
for the speaker: every synthesis we read puts the effect in the small range or rates the evidence low. none of 70 umbrella-review comparisons reached convincing or highly suggestive. when raters were blinded, the effect on positive symptoms fell to SMD −0.08. cochrane found no clear long-term advantage over other psychosocial therapies on relapse, rehospitalisation, mental state, functioning or quality of life. the umbrella authors say guidelines should reconsider their recommendations.
against: 'poor' could be read as 'no effect', and the record doesn't support that reading. across meta-analyses the end-of-treatment effects on overall and positive symptoms are statistically significant, roughly SMD 0.25 to 0.36. the 2018 network meta-analysis concluded CBT 'seems to be effective' on positive symptoms in moderately ill patients, and some of its estimates had moderate confidence. the umbrella review also found suggestive evidence for functioning at end of treatment. CBT for psychosis is still a guideline-recommended treatment. we tried to read the UK national guideline page directly, but the fetch was blocked, so we don't cite it.
what this cannot say
we read abstracts and the publisher or PMC summaries, not every included trial. the meta-analyses overlap heavily in the trials they pool, so they're not independent confirmations. 'CBT' covers varied protocols, therapist skill levels and patient groups. nearly all trials add CBT on top of antipsychotics, so they can't speak to CBT on its own. most results are end-of-treatment, and follow-up data are thinner and inconsistent. effect sizes are averages: a small pooled effect can hide some people who benefit a lot and some who don't benefit at all. 'poor' is a judgment word. we matched it to the reviews' own certainty grades, and a reader who weights statistical significance over certainty could fairly call the claim overstated. one fetch gave a dropout figure that looked garbled, so we left it out.
sources, each one fetched and read
- cognitive–behavioural therapy for the symptoms of schizophrenia: systematic review and meta-analysis with examination of potential bias. british journal of psychiatry, 2014;204(1):20-29 What we read there: pooled effect sizes of −0.33 overall (34 studies), −0.25 positive and −0.13 negative symptoms. with blinded raters, overall fell from −0.62 to −0.15 and positive from −0.57 to −0.08. the authors describe a therapeutic effect 'in the small range' that 'reduces further when sources of bias, particularly masking, are controlled for.' Fetched 2026-10-27 (HTTP 200). https://www.cambridge.org/core/journals/the-british-journal-of-psychiatry/article/cognitivebehavioural-therapy-for-the-symptoms-of-schizophrenia-systematic-review-and-metaanalysis-with-examination-of-potential-bias/10B5A4BB22C8A82FE55E578A15975753
- cognitive behavioural therapy for the treatment of schizophrenia spectrum disorders: an umbrella review of meta-analyses of randomised controlled trials. eClinicalMedicine, 2024 What we read there: 26 meta-analyses, 70 comparisons. none were class I or II, 7.1% class III, 35.7% class IV, 57.1% non-significant. general psychopathology g = −0.33, hallucinations 0.32, delusions 0.36, negative symptoms −0.13. 'small to medium effect … at the end of treatment, while evidence became inconsistent at follow-up'. guidelines should 'reconsider their recommendations'. Fetched 2026-10-27 (HTTP 200). https://pmc.ncbi.nlm.nih.gov/articles/PMC10809079/
- cochrane schizophrenia group. cognitive behavioural therapy plus standard care versus standard care plus other psychosocial treatments for people with schizophrenia. cochrane database of systematic reviews, 2018, CD008712.pub3, PMID 30480760 (pubmed abstract) What we read there: 36 trials, 3,542 participants. long-term relapse RR 1.05 (0.85 to 1.29), rehospitalisation RR 0.96 (0.82 to 1.14), no clear difference in mental state, social functioning or quality of life. 'no clear and convincing advantage for cognitive behavioural therapy over other … psychosocial therapies'. evidence mostly low or very low quality. Fetched 2026-10-27 (HTTP 203). https://pubmed.ncbi.nlm.nih.gov/30480760/
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