Response Rates in Psychiatric Drug Trials Are Statistical Nonsense

Response Rates in Psychiatric Drug Trials Are Statistical Nonsense
by Peter C. Gøtzsche at Brownstone Institute

The outcomes used in psychiatric drug trials are not meaningful, and psychiatric diagnoses and names of drug classes are also problematic. According to DSM-5, major depression “causes clinically significant distress or impairment in social, occupational, or other important areas of functioning.”1 It is the other way around. People become depressed because they have difficulties in their lives, not because they are attacked by some depression monster that can be killed with so-called antidepressants, like antibiotics that can kill bacteria. 

The patients want to have normal levels of functioning and to enjoy a meaningful life.² Yet, I have not seen a single placebo-controlled trial of depression drugs that reported on such outcomes except one, which was unethical because the drugs were abruptly withdrawn in half the patients, which harmed them substantially, as they developed abstinence symptoms.³ Patients on paroxetine reported statistically significant deterioration in functioning at work, relationships, social activities, and overall functioning. The study was sponsored by Eli Lilly, the manufacturer of fluoxetine, which has an active metabolite with a half-life of one to two weeks. Thus, little harm will be done to patients on fluoxetine during a five-day period where the drug was changed to placebo unbeknownst to the patients. 

The outcomes in psychiatric drug trials are measured on rating scales even though the results cannot tell us whether the patients have improved in any way that is important for them. But we can exclude this possibility because the effects obtained with such scales are considerably lower than the least clinically relevant difference to placebo, both for depression drugs and psychosis drugs.⁴ Thus, the drugs don’t work, not even for very severe degrees of depression.⁴ This is not what the patients are being told. 

Statistical Hocus Pocus

We constantly hear about large effects of psychiatric drugs. This is usually because data on a ranking scale are dichotomised into the number of patients improved in an act of statistical hocus pocus.

A recent anonymous editorial in the Lancet illustrates this.⁵ It quoted a 2018 network meta-analysis by Cipriani et al.,⁶ noting that “all antidepressants are more efficacious than placebo in adults with a diagnosis of major depressive disorder, with odds ratios ranging between 2.23 and 1.37” (there was no average for all the drugs in the meta-analysis but it would have been around 1.7).

An almost doubling of the response rate looks very impressive but it wasn’t.⁷ Cipriani et al. also reported that the standardised mean difference was only 0.30, similar to other meta-analyses.^8,9 The difference to placebo is only about 2 on the Hamilton depression scale,^6,8-10 far less than what is clinically relevant. The smallest effect that can be perceived on this scale is 5-6,¹¹ and the minimal clinically relevant effect is of course larger than the bare minimum that can be perceived.

It is highly misleading to dichotomise data on a ranking scale and report on patients who have improved by a certain amount. This statistical hocus pocus spins straw into gold transforming ineffectiveness into the much-trumpeted idea that antidepressants work,¹² as expressed in a headline in the Guardian when the Cipriani meta-analysis was published.¹³ By categorising people into responders and non-responders, Cipriani et al. transformed a tiny 2-point difference in depression symptoms scores¹⁰ into the illusion that you are twice as likely to respond if you take a depression drug compared to placebo. 

The “response” that is reported in trials is an artificial figure constructed by categorising the data using an arbitrary cut-off point. There is no natural distinction between showing a response and not showing a response.¹² People improve to different degrees. 

Unsurprisingly, statisticians have advised not to categorise data from scales in this way.^14,15 Response rates derived from continuous measures do not add information, and they can create an unwarranted illusion of clinical effectiveness. Psychologist Irving Kirsch and psychiatrist Joanna Moncrieff have shown how absurd this is.¹⁶ Relatively small differences in improvement scores can produce relatively large differences in response rates. 

The most commonly used definition of response, which was also used in the Cipriani paper, is a 50% decrease in depressive symptoms.¹⁶ As the mean baseline Hamilton score in clinical trials is about 24, the criterion for response for an average patient would be 12. Thus, a patient with an improvement of 11 would be classified as a non-responder, even though the improvement is over five times larger than the drug-placebo difference of 2. 

Number Needed to Treat Is Also Hocus Pocus

What I have just discussed applies to all psychiatric drugs. In other areas of medicine, we would not accept such manipulations. 

The number needed to treat (NNT) to benefit one patient is also hocus pocus. It is not the number of patients one must treat for one extra person to get better; it is the number of patients one must treat to push one extra person over the arbitrary and meaningless response criterion.¹⁶

An article, which in its headline claimed that the NNT is an underused measure of treatment effect in psychopharmacology, reported that the NNT for drugs used for depression, mania, bipolar disorder, schizophrenia, panic disorder, social phobia, and obsessive-compulsive disorder were in the range of 3 to 6.¹⁷

For ADHD, a poor-quality meta-analysis, which did not assess the risk of bias in the individual studies, reported huge effects of stimulants, which the authors translated into an NNT of only about 2-3.¹⁸ Two Cochrane reviews performed by my employees found that every single trial ever performed of methylphenidate for ADHD was at high risk of bias,^19,20 and a third Cochrane review that did not pay due attention to this was withdrawn after we had protested.²¹ 

In 2014, leading British psychiatrists claimed that antidepressants are among the most effective drugs we have in the whole of medicine and that they have an impressive ability to prevent recurrence of depression, with an NNT of around three.²² The problem with this is that, in the trials that have shown these effects, half of the patients continued with their depression drug after they had recovered, while the other half were switched to placebo and developed withdrawal symptoms that were misinterpreted as relapse.^4,23 As only two patients are needed to get one with withdrawal symptoms when a drug is stopped,²⁴ there cannot exist an NNT to prevent recurrence, only a number needed to harm (NNH), which is two.

The most important reason why the NNT for a psychiatric drug is an illusion²⁵ is that more patients are harmed than those who benefit. Harms and benefits are rarely measured on the same scale, but when patients in a placebo-controlled trial decide whether it is worthwhile to continue in the trial, they make a judgment about if the benefits they perceive exceed the harms. 

My research group did such an analysis based on clinical study reports we had obtained from drug regulators, and we found that 12% more patients dropped out on a depression pill than on placebo (P < 0.00001).²⁶ This means that there cannot be an NNT for depression pills, only an NNH. Our meta-analysis showed that this number is about 25.

The psychiatric narrative, which speaks about effective and safe drugs,⁷ is misleading. If we break a leg, we would not be satisfied with a treatment that reduces the pain so little that we cannot feel the difference from placebo, while the leg is still broken. And whether we have a psychiatric issue or a physical one, we want to be healed, which no psychiatric drug can accomplish.⁴ 

References

1 American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 5th ed. Washington: American Psychiatric Publishing Group; 2013.

2 Gøtzsche PC. A new paradigm for testing psychiatric drugs is needed. Mad in America 2023; Feb 25.

3 Michelson D, Fava M, Amsterdam J, et al. Interruption of selective serotonin reuptake inhibitor treatment. Double-blind, placebo-controlled trial. Br J Psychiatry 2000;176:363-8.

4 Gøtzsche PC. Critical psychiatry textbook. Copenhagen: Institute for Scientific Freedom; 2022, pages 45 and 72 (freely available).

5 50 years of SSRIs: weighing benefits and harms. Lancet 2025;405:1641.

6 Cipriani A, Furukawa TA, Salanti G, et al. 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;391:1357-66.

7 Gøtzsche PC. Protecting the false narrative about antidepressants. Mad in America 2025; July 7.

8 Jakobsen JC, Katakam KK, Schou A, et al. Selective serotonin reuptake inhibitors versus placebo in patients with major depressive disorder: a systematic review with meta-analysis and trial sequential analysis. BMC Psychiatry 2017;17:58.

9 Stone MB, Yaseen ZS, Miller BJ, et al. 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.

10 Munkholm K, Paludan-Müller AS, Boesen K. Considering the methodological limitations in the evidence base of antidepressants for depression: a reanalysis of a network meta-analysis. BMJ Open 2019;9:e024886.

11 Leucht S, Fennema H, Engel R, et al. What does the HAMD mean? J Affect Disord 2013;148:243-8.

12 Moncrieff J. Chemically imbalanced: The making and unmaking of the serotonin myth. Padstow: Flint; 2025.

13 Boseley S. The drugs do work: antidepressants are effective, study shows. the Guardian 2018; Feb 22. 

14 Royston P, Altman DG, Sauerbrei W. Dichotomizing continuous predictors in multiple regression: a bad idea. Stat Med 2006;25:127-41.

15 Altman DG, Royston P. The cost of dichotomising continuous variables. BMJ 2006;332:1080.

16 Kirsch I, Moncrieff J. Clinical trials and the response rate illusion. Contemporary Clinical Trials 2007;28:348-51.

17 Pinson L, Gray GE. Psychopharmacology: number needed to treat: an underused measure of treatment effect. Psychiatr Serv 2003;54:145-6.

18 Faraone SV, Glatt SJ. A comparison of the efficacy of medications for adult attention-deficit/hyperactivity disorder using meta-analysis of effect sizes. J Clin Psychiatry 2010;71:754-63.

19 Storebø OJ, Ramstad E, Krogh HB, et al. Methylphenidate for children and adolescents with attention deficit hyperactivity disorder (ADHD). Cochrane Database Syst Rev 2015;11:CD009885.

20 Boesen K, Paludan-Müller AS, Gøtzsche PC, et al. Extended-release methylphenidate for attention deficit hyperactivity disorder (ADHD) in adults. Cochrane Database Syst Rev 2022;2:CD012857.

21 Boesen K, Saiz LC, Erviti J, et al. The Cochrane Collaboration withdraws a review on methylphenidate for adults with attention deficit hyperactivity disorder. Evid Based Med 2017;22:143-7.

22 Nutt DJ, Goodwin GM, Bhugra D, et al. Attacks on antidepressants: signs of deep-seated stigma? Lancet Psychiatry 2014;1:103-4.

23 Gøtzsche PC, Demasi M. Interventions to help patients withdraw from depression drugs: A systematic review. Int J Risk Saf Med 2024;35:103-16.

24 Davies J, Read J. A systematic review into the incidence, severity and duration of antidepressant withdrawal effects: Are guidelines evidence-based? Addict Behav 2019;97:111-21.

25 Gøtzsche PC. Number needed to treat with a psychiatric drug to benefit one patient is an illusion. Mad in America 2022; Dec 13.

26 Sharma T, Guski LS, Freund N, et al. Drop-out rates in placebo-controlled trials of antidepressant drugs: A systematic review and meta-analysis based on clinical study reports. Int J Risk Saf Med 2019;30:217-32.

Response Rates in Psychiatric Drug Trials Are Statistical Nonsense
by Peter C. Gøtzsche at Brownstone Institute – Daily Economics, Policy, Public Health, Society