Individual Differences in Response to Antidepressants: A Meta-analysis of Placebo-Controlled Randomized Clinical Trials | Depressive Disorders | JAMA Psychiatry | JAMA Network
[Skip to Navigation]
Sign In
Figure 1.  Study Selection Process
Study Selection Process

AD indicates antidepressant; RCT, randomized clinical trial.

Figure 2.  Comparisons of Variability Between Antidepressants (ADs) and Placebo, Separated by Class
Comparisons of Variability Between Antidepressants (ADs) and Placebo, Separated by Class

Comparisons of placebo (the reference group) and AD (or AD class) indicate the difference between the 2 groups in ln σ̂. These estimates have been exponentiated, with a bln σ̂ less than 1 representing less variability than placebo. NA indicates norepinephrinergic agent; NDRI, norepinephrine-dopamine reuptake inhibitor; SSRI, selective serotonin reuptake inhibitor; SNRI, serotonin and norepinephrine reuptake inhibitor.

Figure 3.  Comparisons of Variability Between Selective Serotonin Reuptake Inhibitors (SSRIs) and Antidepressant (AD) Classes
Comparisons of Variability Between Selective Serotonin Reuptake Inhibitors (SSRIs) and Antidepressant (AD) Classes

Comparisons of SSRI (the reference group) and AD classes indicate the difference between the two groups in ln σ̂. These estimates have been exponentiated, with a bln σ̂ less than 1 representing less variability than SSRI. NDRI indicates norepinephrine-dopamine reuptake inhibitor; NA, norepinephrinergic agent; SNRI, serotonin and norepinephrine reuptake inhibitor.

Table.  Results of Secondary Analyses Examining Baseline Depression Severity, and Randomized Clinical Trial Publication Year
Results of Secondary Analyses Examining Baseline Depression Severity, and Randomized Clinical Trial Publication Year
Supplement.

eMethods. Justification for not using variation and coefficient of variation ratios (VRs and CVRs) and detailed descriptions of random-slope mixed-effects models (RSMMs)

eFigure 1. Association between endpoint means and standard deviations of total depression scores

eFigure 2. Association between endpoint means and coefficients of variation of total depression scores

eResults. Treatment effect heterogeneity: upper bounds and simulation

eTable 1. Upper bounds on SDte based on correlations (ρ) between the placebo response and treatment effect

eTable 2. Testing a quadratic effect: Primary analysis

eTable 3. Controlling for scale: Primary analysis

eTable 4. Controlling for scale: Secondary analysis of baseline depression severity

eTable 5. Controlling for Scale: Secondary analysis of AD class (as compared to placebo)

eTable 6. Controlling for Scale: Secondary analysis of AD class (as compared to SSRI)

eTable 7. Controlling for Scale: Secondary analysis of RCT publication year

eTable 8. Controlling for treatment duration: Primary analysis

eTable 9. Controlling for treatment duration: Secondary analysis of baseline depression severity

eTable 10. Controlling for treatment duration: Secondary analysis of AD class (as compared to placebo)

eTable 11. Controlling for treatment duration: Secondary analysis of AD class (as compared to SSRI)

eTable 12. Controlling for treatment duration: Secondary analysis of RCT publication year

eTable 13. Expanded priors: Primary analysis

eTable 14. Expanded priors: Secondary analysis of baseline depression severity

eTable 15. Expanded priors: Secondary analysis of AD class (as compared to placebo)

eTable 16. Expanded priors: Secondary analysis of AD class (as compared to SSRI)

eTable 17. Expanded priors: Secondary analysis of RCT publication year

eFigure 3. SDte upper bounds against a range of ρ for bln σ ̂ = 0.99 and bln σ ̂ = 1.05

eFigure 4. A plot of simulated SDte and VRs

eFigure 5a. Density plot of simulated variability ratios

eFigure 5b. Density plot of simulated treatment effect standard deviations (SDte)

eFigure 5c. Density plot of simulated outcomes under placebo and antidepressant (AD) conditions

eReferences.

1.
American Psychiatric Association.  Diagnostic and Statistical Manual of Mental Disorders. 5th ed. Washington, DC: American Psychiatric Association; 2013.
2.
National Institute for Health and Care Excellence. Depression in adults: recognition and management: clinical guideline. Published October 28, 2009. Accessed January 15, 2021. https://www.nice.org.uk/guidance/cg90/resources/depression-in-adults-recognition-and-management-pdf-975742636741
3.
Thomas  L, Kessler  D, Campbell  J,  et al.  Prevalence of treatment-resistant depression in primary care: cross-sectional data.   Br J Gen Pract. 2013;63(617):e852-e858. doi:10.3399/bjgp13X675430PubMedGoogle ScholarCrossref
4.
Noma  H, Furukawa  TA, Maruo  K,  et al.  Exploratory analyses of effect modifiers in the antidepressant treatment of major depression: individual-participant data meta-analysis of 2803 participants in seven placebo-controlled randomized trials.   J Affect Disord. 2019;250:419-424. doi:10.1016/j.jad.2019.03.031PubMedGoogle ScholarCrossref
5.
Uher  R.  Genes, environment, and individual differences in responding to treatment for depression.   Harv Rev Psychiatry. 2011;19(3):109-124. doi:10.3109/10673229.2011.586551PubMedGoogle ScholarCrossref
6.
Winkelbeiner  S, Leucht  S, Kane  JM, Homan  P.  Evaluation of differences in individual treatment response in schizophrenia spectrum disorders: a meta-analysis  [published online June 3, 2019].  JAMA Psychiatry. 2019;76(10):1063-1073. doi:10.1001/jamapsychiatry.2019.1530PubMedGoogle ScholarCrossref
7.
Volkmann  A. On the relationship between treatment effect heterogeneity and the variability ratio effect size statistic. arXiv. Accessed January 15, 2021. https://arxiv.org/pdf/2006.11848.pdf
8.
Senn  S.  Mastering variation: variance components and personalised medicine.   Stat Med. 2016;35(7):966-977. doi:10.1002/sim.6739PubMedGoogle ScholarCrossref
9.
Cuijpers  P, Christensen  H.  Are personalised treatments of adult depression finally within reach?   Epidemiol Psychiatr Sci. 2017;26(1):40-42. doi:10.1017/S204579601600007XPubMedGoogle ScholarCrossref
10.
Kessler  RC.  The potential of predictive analytics to provide clinical decision support in depression treatment planning.   Curr Opin Psychiatry. 2018;31(1):32-39. doi:10.1097/YCO.0000000000000377PubMedGoogle ScholarCrossref
11.
Ozomaro  U, Wahlestedt  C, Nemeroff  CB.  Personalized medicine in psychiatry: problems and promises.   BMC Med. 2013;11:132. doi:10.1186/1741-7015-11-132PubMedGoogle ScholarCrossref
12.
Simon  GE, Perlis  RH.  Personalized medicine for depression: can we match patients with treatments?   Am J Psychiatry. 2010;167(12):1445-1455. doi:10.1176/appi.ajp.2010.09111680PubMedGoogle ScholarCrossref
13.
Volkmann  C, Volkmann  A, Müller  CA.  On the treatment effect heterogeneity of antidepressants in major depression: a Bayesian meta-analysis and simulation study.   PLoS One doi:10.1371/journal.pone.0241497 Google Scholar
14.
Mills  HL, Higgins  JP, Morris  RW,  et al.  Detecting heterogeneity of intervention effects using analysis and meta-analysis of differences in variance between arms of a trial.   medRxiv. Accessed May 12, 2020. doi:10.1101/2020.03.07.20032516Google Scholar
15.
Winkelbeiner  S, Homan  P.  Is variance ratio a valid indicator of heterogeneous treatment effect? reply.   JAMA Psychiatry. 2020;77(2):217-218. doi:10.1001/jamapsychiatry.2019.3382PubMedGoogle ScholarCrossref
16.
Bae  S.  Is variance ratio a valid indicator of heterogeneous treatment effect?   JAMA Psychiatry. 2020;77(2):216-217. doi:10.1001/jamapsychiatry.2019.3372PubMedGoogle ScholarCrossref
17.
Guo  X, McCutcheon  RA, Pillinger  T,  et al.  The magnitude and heterogeneity of antidepressant response in depression: A meta-analysis of over 45,000 patients.   J Affect Disord. 2020;276:991-1000. doi:10.1016/j.jad.2020.07.102PubMedGoogle ScholarCrossref
18.
Nakagawa  S, Poulin  R, Mengersen  K,  et al  Meta-analysis of variation: ecological and evolutionary applications and beyond.   Methods Ecol Evol. 2015;6(2):143-152. doi:10.1111/2041-210X.12309Google ScholarCrossref
19.
Plöderl  M, Hengartner  MP.  What are the chances for personalised treatment with antidepressants? detection of patient-by-treatment interaction with a variance ratio meta-analysis.   BMJ Open. 2019;9(12):e034816. doi:10.1136/bmjopen-2019-034816PubMedGoogle Scholar
20.
Munkholm  K, Winkelbeiner  S, Homan  P. Individual response to antidepressants for depression in adults–a simulation study and meta-analysis. PloS One. 2020;15(8):e0237950.
21.
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(10128):1357-1366. doi:10.1016/S0140-6736(17)32802-7PubMedGoogle ScholarCrossref
22.
Fournier  JC, DeRubeis  RJ, Hollon  SD,  et al.  Antidepressant drug effects and depression severity: a patient-level meta-analysis.   JAMA. 2010;303(1):47-53. doi:10.1001/jama.2009.1943PubMedGoogle ScholarCrossref
23.
Stone  M, Kalaria  S, Richardville  K,  et al Components and trends in treatment effects in randomized placebo-controlled trials in major depressive disorder from 1979-2016. Presented at: American Society of Clinical Psychopharmacology Annual Conference; May 30, 2018; Miami, FL.
24.
Furukawa  TA, Salanti  G, Atkinson  LZ,  et al.  Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis.   BMJ Open. 2016;6(7):e010919. doi:10.1136/bmjopen-2015-010919PubMedGoogle Scholar
25.
Hamilton  M.  A rating scale for depression.   J Neurol Neurosurg Psychiatry. 1960;23(1):56-62. doi:10.1136/jnnp.23.1.56PubMedGoogle ScholarCrossref
26.
Bech  P.  Clinical Psychometrics. Oxford, England: Wiley Blackwell; 2012. doi:10.1002/9781118511800
27.
Williams  JBW, Link  MJ, Rosenthal  NE, Terman  M.  Structured Interview Guide for the Hamilton Depression Rating Scale, Seasonal Affective Disorders Version (SIGH-SAD). New York, New York: State Psychiatric Institute, Biometrics Research; 1988.
28.
Roberts  JN, Lei  HH, Krishen  A,  et al Identifying sub-scales of the 31-item Hamilton Depression Scale. Poster presented at: New Clinical Drug Evaluation Unit (NCDEU) meeting; May 28-31, 2001; Phoenix, AZ.
29.
Montgomery  SA, Asberg  M.  A new depression scale designed to be sensitive to change.   Br J Psychiatry. 1979;134:382-389. doi:10.1192/bjp.134.4.382PubMedGoogle ScholarCrossref
30.
Rush  AJ, Giles  DE, Schlesser  MA, Fulton  CL, Weissenburger  J, Burns  C.  The Inventory for Depressive Symptomatology (IDS): preliminary findings.   Psychiatry Res. 1986;18(1):65-87. doi:10.1016/0165-1781(86)90060-0PubMedGoogle ScholarCrossref
31.
Raudenbush  SW, Bryk  AS.  Examining correlates of diversity.   J Educ Stat. 1987; 12(3):241-69. doi:10.3102/10769986012003241Google ScholarCrossref
32.
Ghaemi  SN.  A new nomenclature for psychotropic drugs.   J Clin Psychopharmacol. 2015;35(4):428-433. doi:10.1097/JCP.0000000000000341PubMedGoogle ScholarCrossref
33.
Owens  MJ, Morgan  WN, Plott  SJ, Nemeroff  CB.  Neurotransmitter receptor and transporter binding profile of antidepressants and their metabolites.   J Pharmacol Exp Ther. 1997;283(3):1305-1322.PubMedGoogle Scholar
34.
R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Published 2019. Accessed January 12, 2020. https://www.R-project.org/
35.
Hadfield  J.  MCMC methods for multi-response generalized linear mixed models: the MCMCglmm R package.   J Stat Softw. 2010;33(2):1-22. doi:10.18637/jss.v033.i02Google ScholarCrossref
36.
Open Science Foundation. Individual differences in response to antidepressants: R Code. Accessed November 13, 2020. https://mfr.ca-1.osf.io/render?url=https://osf.io/m6u89/?direct%26mode=render%26action=download%26mode=render
37.
Chekroud  AM, Gueorguieva  R, Krumholz  HM, Trivedi  MH, Krystal  JH, McCarthy  G.  Reevaluating the efficacy and predictability of antidepressant treatments: a symptom clustering approach.   JAMA Psychiatry. 2017;74(4):370-378. doi:10.1001/jamapsychiatry.2017.0025PubMedGoogle ScholarCrossref
38.
Furukawa  TA, Maruo  K, Noma  H,  et al.  Initial severity of major depression and efficacy of new generation antidepressants: individual participant data meta-analysis.   Acta Psychiatr Scand. 2018;137(6):450-458. doi:10.1111/acps.12886PubMedGoogle ScholarCrossref
39.
Rabinowitz  J, Werbeloff  N, Mandel  FS, Menard  F, Marangell  L, Kapur  S.  Initial depression severity and response to antidepressants v. placebo: patient-level data analysis from 34 randomised controlled trials.   Br J Psychiatry. 2016;209(5):427-428. doi:10.1192/bjp.bp.115.173906PubMedGoogle ScholarCrossref
40.
Gibbons  RD, Hur  K, Brown  CH, Davis  JM, Mann  JJ.  Benefits from antidepressants: synthesis of 6-week patient-level outcomes from double-blind placebo-controlled randomized trials of fluoxetine and venlafaxine.   Arch Gen Psychiatry. 2012;69(6):572-579. doi:10.1001/archgenpsychiatry.2011.2044PubMedGoogle ScholarCrossref
41.
Fried  EI, Nesse  RM.  Depression is not a consistent syndrome: An investigation of unique symptom patterns in the STAR*D study.   J Affect Disord. 2015;172:96-102. doi:10.1016/j.jad.2014.10.010PubMedGoogle ScholarCrossref
42.
Fried  EI, Nesse  RM.  Depression sum-scores don’t add up: why analyzing specific depression symptoms is essential.   BMC Med. 2015;13(1):72. doi:10.1186/s12916-015-0325-4 PubMedGoogle ScholarCrossref
43.
Niculescu  AB, Le-Niculescu  H, Levey  DF,  et al.  Precision medicine for suicidality: from universality to subtypes and personalization.   Mol Psychiatry. 2017;22(9):1250-1273. doi:10.1038/mp.2017.128PubMedGoogle ScholarCrossref
44.
Green  E, Goldstein-Piekarski  AN, Schatzberg  AF, Rush  AJ, Ma  J, Williams  L.  Personalizing antidepressant choice by sex, body mass index, and symptom profile: An iSPOT-D report.   Personalized Medicine in Psychiatry. 2017;1:65-73. doi:10.1016/j.pmip.2016.12.001Google ScholarCrossref
45.
Picardi  A, Viroli  C, Tarsitani  L,  et al.  Heterogeneity and symptom structure of schizophrenia.   Psychiatry Res. 2012;198(3):386-394. doi:10.1016/j.psychres.2011.12.051PubMedGoogle ScholarCrossref
46.
Tandon  R, Gaebel  W, Barch  DM,  et al.  Definition and description of schizophrenia in the DSM-5.   Schizophr Res. 2013;150(1):3-10. doi:10.1016/j.schres.2013.05.028PubMedGoogle ScholarCrossref
47.
Durisko  Z, Mulsant  BH, Andrews  PW.  An adaptationist perspective on the etiology of depression.   J Affect Disord. 2015;172:315-323. doi:10.1016/j.jad.2014.09.032PubMedGoogle ScholarCrossref
48.
Thase  ME, Entsuah  AR, Rudolph  RL.  Remission rates during treatment with venlafaxine or selective serotonin reuptake inhibitors.   Br J Psychiatry. 2001;178(3):234-241. doi:10.1192/bjp.178.3.234PubMedGoogle ScholarCrossref
49.
Radua  J, Davies  C, Fusar-Poli  P.  Evaluation of variability in individual response to treatments in the clinical high-risk state for psychosis: a meta-analysis.   Schizophr Res.S0920-9964(20)30262-0. doi:10.1016/j.schres.2020.05.010PubMedGoogle Scholar
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    2 Comments for this article
    EXPAND ALL
    Multiple testing correction ?
    Roland Dardennes, MD | Université de Paris
    Dear Sir,
    We read with Great interest this new article from the Meta-analytic Oxford Team. We have one question: did. the authors adjust their résults from multiple-testing fallacy? As an example, a p-value of .02 between the variability of response between noradrenergic and sérotoninergic antidepressant would not be of any significance if no multiple-testing adjustment was performed.
    Best wishes
    Pr. Roland Dardennes and Mrs Nebal Alanbar, PhD
    CONFLICT OF INTEREST: None Reported
    Noradrenergic antidepressants and larger response variability
    Takeshi Terao, MD, PhD | Department of Psychiatry, , Oita University Faculty of Medicine
    To the Editor: I read with great interest the article by Maslej et al1 in which they showed that the responses to noradrenergic antidepressants (NAs) were significantly more variable than those to selective serotonin reuptake inhibitors (SSRIs). Although the authors mentioned that antidepressants primarily affecting noradrenaline may have a greater effect than those exclusively affecting synaptic serotonin, they do not address the potential causes underlying the greater response variation of NAs compared with SSRIs. Moreover, the authors mentioned that functional unblinding during randomized clinical trials (RCTs) involving NAs might explain this finding because dropout rates due to adverse effects of NAs are generally higher than those for SSRIs, suggesting that observers are not blinded to treatment allocation in some RCTs. Furthermore, there is insufficient justification provided for the final statement of the article in which the authors suggest that NAs may be the most suitable therapy for personalized treatment. I would like to elaborate on this topic.

    It is well-known that noradrenaline triggers a manic switch in patients with depression.2 Furthermore, imipramine and serotonin–norepinephrine reuptake inhibitors (SNRIs) have induced a manic switch significantly more often than placebo in 40 observational studies, unlike SSRIs.3 According to the Diagnostic and Statistical Manual of Mental Disorders (5th Edition),4 full-manic or full-hypomanic episodes occurring during antidepressant treatment, which persist at a fully syndromal level beyond the physiological effect of the treatment, are strongly suggestive of a manic or hypomanic episode, respectively. Therefore, patients displaying a manic or hypomanic switch during NA treatment would have been diagnosed with bipolar I and bipolar II disorder, respectively. The former may have dropped out from the RCT with low scores of intention-to-treat analyses, while the latter completed the RCT with high scores. The patients with true depression responded to NAs normally. Therefore, responses to NAs may have been highly variable because the cohort included patients with true depression, bipolar I disorder, and bipolar II disorder. It is noteworthy to mention that a substantial portion of patients with unipolar depression present manic and/or hypomanic episodes, resulting in a change of diagnosis to bipolar disorder.5 Considering the diagnostic heterogeneity underlying the variable responses to NAs, clinicians should change pharmacotherapeutic NAs to mood stabilizers when appropriate, and thus adopt a personalized treatment approach.

    REFERENCES
    1. Maslej MM, Furukawa TA, Cipriani A, Andrews PW, Sanches M, Tomlinson A, et al. Individual differences in response to antidepressants: a meta-analysis of placebo-controlled randomized clinical trials. Published online [February 17, 2021]. JAMA Psychiatry. doi:10.1001/jamapsychiatry.2020.4564
    2. Bunney WE Jr, Murphy DL, Goodwin FK, Borge GF. The switch process from depression to mania: relationship to drugs which alter brain amines. Lancet. 1970;1(7655):1022-1027.
    3. Allain N, Leven C, Falissard B, Allain JS, Batail JM, Polard E, et al. Manic switches induced by antidepressants: an umbrella review comparing randomized controlled trials and observational studies. Acta Psychiatr Scand. 2017;135(2):106-116.
    4. American Psychiatric Association, Diagnostic and Statistical Manual of Mental Disorders. 5th ed. Washington, DC; 2013.
    5. Terao T, Ishii N, Hirakawa H. A specific group of patients with diagnostic conversion from depression to bipolar disorder
    CONFLICT OF INTEREST: None Reported
    READ MORE
    Original Investigation
    February 17, 2021

    Individual Differences in Response to Antidepressants: A Meta-analysis of Placebo-Controlled Randomized Clinical Trials

    Author Affiliations
    • 1Centre for Addiction and Mental Health, Toronto, Ontario, Canada
    • 2Department of Psychiatry, University of Toronto, Toronto, Ontario, Canada
    • 3Department of Health Promotion and Human Behavior, Kyoto University Graduate School of Medicine, School of Public Health, Yoshida-Konoe, Sakyo, Kyoto, Japan
    • 4Department of Clinical Epidemiology, Kyoto University Graduate School of Medicine, Kyoto University School of Public Health, Yoshida-Konoe, Sakyo, Kyoto, Japan
    • 5Department of Psychiatry, University of Oxford, Oxford, England
    • 6Oxford Health NHS Foundation Trust, Warneford Hospital, Oxford, England
    • 7Department of Psychology, Neuroscience & Behaviour, McMaster University, Hamilton, Ontario, Canada
    • 8Department of Psychiatry and Psychotherapy, Charité Universitätsmedizin Berlin, Berlin, Germany
    • 9Institute of Psychiatry, Psychology and Neuroscience, Department of Psychosis Studies, King’s College of London, London, England
    • 10Department of Psychiatry, Renmin Hospital of Wuhan University, Wuhan, China
    JAMA Psychiatry. 2021;78(5):490-497. doi:10.1001/jamapsychiatry.2020.4564
    Key Points

    Question  Is there evidence that response to antidepressants varies based on individual differences?

    Findings  In this meta-analysis of 91 randomized clinical trials (18 965 participants) on the use of antidepressants in major depression, no evidence was found of more variability in response to antidepressants than to placebo. Variability was not associated with baseline depression severity or study year, but variability in response to noradrenergic agents was higher than that of selective serotonin reuptake inhibitors.

    Meaning  Individual differences may not underlie variability in the association between total depression scores and antidepressant treatment; future efforts toward personalization should focus on individual symptoms or biomarkers.

    Abstract

    Importance  Antidepressants are commonly used to treat major depressive disorder (MDD). Antidepressant outcomes can vary based on individual differences; however, it is unclear whether specific factors determine this variability or whether it is at random.

    Objective  To investigate the assumption of systematic variability in symptomatic response to antidepressants and to assess whether variability is associated with MDD severity, antidepressant class, or study publication year.

    Data Sources  Data used were updated from a network meta-analysis of treatment with licensed antidepressants in adults with MDD. The Cochrane Central Register of Controlled Trials, CINAHL, Embase, LILACS database, MEDLINE, MEDLINE In-Process, and PsycInfo were searched from inception to March 21, 2019. Additional sources were international trial registries and sponsors, drug companies and regulatory agencies’ websites, and reference lists of published articles. Data were analyzed between June 8, 2020, and June 13, 2020.

    Study Selection  Analysis was restricted to double-blind, randomized placebo-controlled trials with depression scores available at the study’s end point.

    Data Extraction and Synthesis  Baseline means, number of participants, end point means and SDs of total depression scores, antidepressant type, and publication year were extracted.

    Main Outcomes and Measures  Log SDs (bln σ̂) were derived for treatment groups (ie, antidepressant and placebo). A random-slope mixed-effects model was conducted to estimate the difference in bln σ̂ between treatment groups while controlling for end point mean. Secondary models determined whether differences in variability between groups were associated with baseline MDD severity; antidepressant class (selective serotonin reuptake inhibitors and other related drugs; serotonin and norepinephrine reuptake inhibitors; norepinephrine-dopamine reuptake inhibitors; noradrenergic agents; or other antidepressants); and publication year.

    Results  In the 91 eligible trials (18 965 participants), variability in response did not differ significantly between antidepressants and placebo (bln σ̂, 1.02; 95% CI, 0.99-1.05; P = .19). This finding is consistent with a range of treatment effect SDs (up to 16.10), depending on the association between the antidepressant and placebo effects. Variability was not associated with baseline MDD severity or publication year. Responses to noradrenergic agents were 11% more variable than responses to selective serotonin reuptake inhibitors (bln σ̂, 1.11; 95% CI, 1.01-1.21; P = .02).

    Conclusions and Relevance  Although this study cannot rule out the possibility of treatment effect heterogeneity, it does not provide empirical support for personalizing antidepressant treatment based solely on total depression scores. Future studies should explore whether individual symptom scores or biomarkers are associated with variability in response to antidepressants.

    Introduction

    Major depressive disorder (MDD) is a common and heterogeneous mental condition characterized by emotional, cognitive, somatic, and behavioral symptoms.1 Antidepressants are one of the first-line interventions for the treatment of depression,2 but their efficacy appears to be variable. While a significant proportion of individuals experience remission of depression after 8 weeks of treatment, more than 50% of patients improve very little or their depression worsens.3 This observed variation has prompted efforts to identify moderators of antidepressant efficacy and to personalize treatments by matching specific antidepressants with the unique characteristics of individual patients.4-6 The variability in the efficacy of psychiatric medications is typically deduced from aggregate data of randomized clinical trials (RCTs), which estimate average treatment effects.5,6 To some degree, treatment effects in RCTs vary between individuals owing to random factors or other factors, such as placebo effects, regression to the mean, or measurement error.6,7 The ability to personalize treatments rests on the assumption that individual differences contribute to this variability, but detecting treatment by individual interactions requires more complex study designs.8

    Despite the paucity of studies designed to detect treatment by individual interactions, there is a widely held assumption that individual differences moderate the effect of antidepressants on depressive symptoms (ie, response).9-12 Data from RCTs show that individuals with symptoms of depression assigned to receive the same antidepressant at the same dose and for the same period can experience very different outcomes.5 The source of this variability is believed to result from individual differences in clinical or biological factors.10 If indeed some antidepressants (or classes of antidepressants) are more effective at treating patients whose MDD is characterized by specific clinical or biological factors, such differences would be consistent with the assumption that individuals vary systematically in their response to antidepressants. However, if this variability is driven by other factors (eg, placebo response and measurement error), it may not be possible to personalize antidepressant treatment. Given that the potential for personalization rests on the validity of an assumption of systematic variability in observed response to antidepressants, it is important to evaluate it rigorously. Thus, we compared variability in observed outcomes of patients with MDD assigned to receive antidepressants or placebo to assess whether the observed variability in response to antidepressants is owing to systematic, nonrandom factors. We hypothesized that if variability in observed response to antidepressants includes an individual by treatment interaction, it would differ from the variability in observed response to placebo.

    A vigorous methodologic debate has been taking place on the best way to assess variability in response to psychiatric treatment.13-17 This debate involves how to deal with the problem that means and their variability (ie, SDs) are often not independent.13,17,18 Several previously published articles19,20 have relied on methods that make assumptions about the nature of the mean-SD relationship, which could lead to biased estimates if these assumptions are not met. To address this issue, we use a random-slope mixed-effects model (RSMM) that accounted for the association between end point depression scores and variability by modeling this association directly from the data.13,18 Furthermore, we updated the open data set of RCTs evaluating outcomes of antidepressants in patients with MDD from Cipriani et al21 that has been used in prior studies of variability in response to antidepressants.13,19,20

    We also examined whether baseline severity of depression, antidepressant class, or the year in which studies were published is associated with variability in response to antidepressants. Because on average, the effects of antidepressants are modest,21 we expected variability to increase for moderators associated with increased response. Thus, we examined whether responses in groups of participants whose symptoms were initially more severe would be more variable because the effects of antidepressants may be more pronounced in individuals with severe depression.22,23 We also expected that variability might differ based on the way that different antidepressant classes interact with different neurotransmitter systems. Specifically, antidepressants affecting multiple neurotransmitter systems might produce more variable outcomes than would antidepressants with more selective effects.

    Methods

    We started with publicly available data from a published network meta-analysis of 522 RCTs evaluating the effects of antidepressants on MDD.21 The methods and descriptive statistics for this meta-analysis are published elsewhere.21,24 Briefly, selected databases (the Cochrane Central Register of Controlled Trials, CINAHL, Embase, LILACS database, MEDLINE, MEDLINE In-Process, and PsycInfo) were searched from their inception to January 8, 2016, using terms that included references to depression in combination with a list of antidepressants. Additional sources were international trial registries, drug approval agency websites, and key scientific journals. We updated the search to include 15 additional RCTs published between January 9, 2016, and March 21, 2019, resulting in a total of 537 RCTs (Figure 1). Of these 537 RCTs, 256 (48%) were conducted in North America, 142 (26%) in Europe, and 42 (8%) in Asia, with the remaining studies being cross-continental or from other regions. There was a total of 89 838 participants allocated to an antidepressant and 30 251 allocated to a placebo. Participants’ mean (SD) age was 44.18 (11.21) years, and 62% were women. The included studies assessed depressive symptoms using one of several versions of the Hamilton Rating Scale for Depression (HAMD-17,25 HAMD-21,25 HAMD-24,26 HAMD-29,27 and HAMD-3128), the Montgomery Åsberg Depression Rating Scale,29 or the Inventory of Depressive Symptomatology.30 End point scores were extracted as close to 8 weeks after the start of antidepressant treatment or placebo as possible, with the median duration of extracted scores being 8 weeks (interquartile range, 6-8 weeks). The data were analyzed between June 8, 2020, and June 13, 2020.

    Eligibility Criteria

    Our analysis included observations from placebo-controlled RCTs with available data at end point (means and SDs of total depression scores and number of participants assessed in each group). Figure 1 depicts our selection process and the resulting number of included RCTs. From the publicly available data21 corresponding to the eligible RCTs, we extracted baseline and end point means, SDs, number of participants in each group, antidepressant drug, and, when available, the year of RCT publication.

    Statistical Analysis
    Primary Analysis

    Based on investigations into the appropriate approach to quantify variability in our included RCTs (described in the eMethods in the Supplement), we used an RSMM to estimate differences in variability between treatment groups, while controlling for the end point mean score.13,18 Following Nakagawa et al,18 we used an unbiased estimator of the natural logarithm of the population SD and its sampling variance.31 For each observation corresponding to an antidepressant or placebo group, we calculated the log SD:

    Image description not available.

    where s refers to the SD and n refers to the number of participants for that group.18

    We used the following formula to derive sampling variances:

    Image description not available.

    where n refers to the number of participants for that group.18 In our primary RSMM, we specified each ln σ̂ (fitted with its associated sampling variance) as the response variable. We specified treatment group as a categorical predictor indicating whether the ln σ̂ corresponded to antidepressant (coded as 1) or placebo (coded as 0). To account for the mean-SD association, we added the log end point mean score (z-transformed across the entire data set) as a predictor. Because the effect of treatment group, estimated by bln σ̂, represented the difference in ln σ̂ between antidepressant and placebo,18 we report the exponentiated bln σ̂. A group effect larger than 1 indicated higher variability in antidepressant groups than placebo groups; conversely, a group effect lower than 1 indicated less variability in the antidepressant groups compared with placebo groups.18 We used methods from previous work7,13 to explore possible values for treatment effect SD, based on findings of this primary analysis and a range of possible associations between the treatment and placebo effects.

    Secondary Analyses

    We repeated our primary analyses adding a treatment group by baseline depression interaction. We used all observations with a baseline depression severity score available (Figure 1) and z-transformed log baseline means across the entire data set. A treatment group by baseline depression interaction effect larger than 1 was consistent with our hypothesis, indicating more variability in response to antidepressant as compared with placebo with increasing baseline depression severity.

    For each available observation from the eligible RCTs, we categorized the antidepressants into 1 of the following 5 classes based on their main putative mechanisms of action32,33: selective serotonin reuptake inhibitors (SSRIs) and other related drugs (citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, sertraline, and vilazodone); serotonin and norepinephrine reuptake inhibitors (SNRIs) (desvenlafaxine and venlafaxine); norepinephrine-dopamine reuptake inhibitors (NDRIs) (bupropion); noradrenergic agents (NAs) (amitriptyline and reboxetine); and other antidepressants (agomelatine, mirtazapine, and trazodone). Randomized clinical trials reporting observations for 4 antidepressants (ie, duloxetine, levomilnacipran, nefazadone, and vortioxetine) did not report end point means and SDs and were not included in our analysis. We repeated our primary analyses replacing the categorical treatment group predictor with antidepressant class. We specified placebo as the reference group, so that an effect of antidepressant class (exponentiated) larger than 1 indicated greater variability in symptomatic response for that class than placebo. To compare variability in response among antidepressant classes, we restricted our analyses to antidepressant groups only, specifying SSRIs as the reference group. Any antidepressant class effects larger than 1 were consistent with our hypothesis, indicating greater symptomatic variability for classes with less selective effects than SSRIs.

    Finally, we investigated whether the year in which RCTs were published was associated with differences in variability. For RCTs with a publication year available (Figure 1), we rescaled years by subtracting the median year (ie, 2001). We conducted an RSMM including a year by treatment group interaction, with an interaction effect greater than 1 indicating more variability in response to antidepressant than placebo in RCTs conducted more recently.

    Sensitivity Analyses

    In the included RCTs, baseline and end point means were measured using the HAMD-17, HAMD-21, HAMD-24, and MADRS, introducing the possibility of spurious effects associated with the use of different scales.13 The duration of treatment also differed between RCTs (ranging from 4-12 weeks). To examine whether different scales and treatment durations affected our results, we repeated our primary and secondary analyses adding scale as a categorical predictor and duration of treatment in weeks as a continuous predictor.

    All analyses were completed in R, version 3.5.3 (the R Foundation),34 using the MCMCglmm package.35 We used functions provided by Nakagawa et al18 to calculate ln σ̂ and sampling variances and the MCMCglmm function for generating RSMMs.35 Detailed descriptions of model specifications and priors are provided in the eMethods in the Supplement. To ensure that choice of priors did not affect our results, we repeated our primary and secondary analyses with expanded priors.18 The significance threshold was .05, and significance testing was 2-sided. To ensure reproducibility, our code36 is freely available online. The updated data set is available on request from the authors.

    Results
    Primary Analysis

    A total of 91 RCTs comprising 18 965 unique participants met our inclusion criteria. To measure outcomes at end point, 44 RCTs (47%) used the HAMD-17 and 33 (37%) used the HAMD-21. Because some RCTs compared placebo with more than 1 antidepressant (ie, multiarm trials), 207 observations were available (Figure 1). End point means were associated with variability (bln σ̂ = 1.07; 95% CI, 1.04-1.10; P < .001). We tested for nonlinearity in this association by adding a quadratic term for end point means, and there was no evidence that adding this term improved the model (eTable 1 in the Supplement). After controlling for the linear association, there was 2% more variability in responses to antidepressants compared with placebo, but this difference was not statistically significant (bln σ̂ = 1.02; 95% CI, 0.99-1.05; P = .19) (Figure 2).

    We calculated the upper bounds on the treatment effect SD, based on this finding and a range of possible associations between the treatment and placebo effects (details of this calculation and simulation are provided in the eResults and eFigures 3-5 in the Supplement). These analyses suggest that bln σ̂ = 1.05 (the upper limit of our estimated bln σ̂) is compatible with a range of treatment effect SDs (eFigure 3 in the Supplement) up to 16.10 (eTable 1 in the Supplement).

    Secondary Analyses
    Depression Severity

    There were 89 placebo-controlled RCTs with both baseline and end point means available, corresponding to 201 observations (Figure 1). We found no evidence that the association between treatment group and variability was associated with baseline depression severity (bln σ̂ = 0.99; 95% CI, 0.96, 1.02; P = .58) (results for model predictors are provided in the Table).

    Antidepressant Class

    Figure 2 provides the number of RCTs reporting outcomes for each antidepressant class and the number of available observations. Responses to all antidepressant classes were not more variable than responses to placebo (depicted in Figure 2). Responses to NDRIs, SNRIs, or other antidepressants were not more variable than responses to SSRIs (ie, the reference group); however, responses to NAs were 11% more variable than responses to SSRIs (bln σ̂, 1.11; 95% CI, 1.01-1.21; P = .02) (Figure 3).

    RCT Publication Year

    The 71 RCTs with a publication year available, corresponding to 158 observations (Figure 1), were published between 1979 and 2018. We found no evidence that the association between treatment group and variability association with the year that an RCT was published (bln σ̂ = 0.999; 95% CI, 0.996-1.002; P = .42) (see the Table for model predictor results).

    Sensitivity Analyses

    We examined whether results from our primary and secondary analyses were affected by depression scales used to measure end point means and SDs across RCTs. Results were not qualitatively different when controlling for scale (full results are provided in eTables 2-7 in the Supplement). One difference emerged when controlling for treatment duration: the difference in variability between SSRIs and NAs was qualitatively similar to our secondary analysis, but it was no longer statistically significant (bln σ̂, 1.108; 95% CI, 0.997-1.227; P = .05) (eTables 8-12 in the Supplement). Results from our primary and secondary analyses were not qualitatively different in RSMMs using expanded priors (eTables 13-17 in the Supplement).

    Discussion

    We examined whether there is evidence of systematic variability in the overall symptomatic response to antidepressants by comparing variability in outcomes in RCTs between participants with MDD assigned to receive antidepressants or placebo. However, contrary to our hypothesis, we found no evidence that variability in observed response (as measured by total depression scores) among participants receiving antidepressants was greater than among those receiving a placebo. Our results do not identify variability based on total depression scores. In other words, our findings do not provide empirical support for efforts to personalize antidepressant treatment based on total depression scores. Nevertheless, these findings do not rule out the possibility of treatment effect heterogeneity; they are consistent with a range of treatment effect SDs that are greater than zero (up to 16.10), depending on negative associations between the treatment and placebo effects (eTable 1 in the Supplement). These findings also do not address the potential for personalization based on individual symptoms or biomarkers, where it may be possible to select specific antidepressants according to some specific clinical or biological characteristics of individual patients.37

    Our findings are consistent with other work suggesting that random or nonspecific factors may account for the variability in observed response to antidepressants. In these studies, variability ratios (VRs) were calculated to quantify differences in SDs of pre/post differences in depression between antidepressant and placebo groups in an earlier open data set.21 Consistent with our results, meta-analyses of VRs13,19,20 found no evidence of larger variability in antidepressant groups as compared with placebo. Our findings are also consistent with previous work in another field of psychiatry,6 which showed that overall responses to antipsychotics are not more variable than responses to placebo in patients with schizophrenia.

    Findings from our secondary analyses suggest that traditional MDD subtypes based on symptom severity are not associated with variability in observed response. This result is consistent with other studies showing that the efficacy of antidepressants is comparable across the continuum of symptom severity of MDD.38-40 Coupled with our primary finding, these results do not support the assumption that there are moderators of observed responses to antidepressants associated with overall severity of depression. At the same time, if MDD is a heterogeneous condition with diverse symptom profiles or subtypes, total depression scores may not adequately capture its typology or even its severity. Some have argued that examining responses to antidepressants based on total scores conceals their effects,41,42 which may also apply to the variability in these effects. Although our results do not provide evidence in support of personalizing MDD treatments based on total depression scores, they do not rule out the possibility of selecting antidepressant treatments based on scores corresponding to individual symptoms or biomarkers. This is consistent with the suggestion that a focus on total depression scores may thwart progress in personalizing treatments for MDD, since there may not be individual differences or biomarkers that capture the shared variance of all its symptoms.42 The next step toward optimizing treatments should involve examining the evidence for variability in responses to antidepressants based on individual symptoms (eg, suicidality43), symptom profiles,44 sets of biomarkers, or a combination of them.

    In our secondary analysis, we did not find any differences in the variability of responses to antidepressant classes and placebo. However, a difference emerged among antidepressant classes, with responses to SSRIs being less variable than responses to NAs. In our sensitivity analysis, this association may be owing to longer treatment duration with NAs. Nevertheless, responses to NAs may be more variable than responses to SSRIs because antidepressants affecting primarily norepinephrine (eg, amitriptyline) may have a greater effect on depressive symptoms than antidepressants that affect only synaptic serotonin.45-48 Alternatively, functional unblinding in RCTs of NAs may account for this finding. Cipriani et al21 reported that, compared with SSRIs, dropout rates owing to adverse effects were generally higher for antidepressants with noradrenergic effects, introducing the possibility that raters in some RCTs were unblinded to treatment allocation. Nevertheless, our finding does not provide insight into the association of individual differences in response to NAs compared with placebo, but it suggests that, among the antidepressant classes we examined, NAs may be most amenable to personalization.

    Limitations

    The results of our study should be considered in the context of some methodologic challenges and limitations. The analysis of variability in response to psychotropic medications has attracted a renewed interest with the availability of novel methods to summarize this variability.18 To our knowledge, the first study to use one of these methods in relation to psychotropic medications was published in 20196; since then, more than 5 published studies13,14,17,19,20 (including this analysis) and other documents15,16 have described the merits and shortcomings of these approaches. When comparing variability between antidepressant and placebo groups, coefficient of variation ratios (CVRs) have been used to account for associations between end point scores and SDs. However, CVRs can yield biased results depending on the nature of these associations, specifically by inflating variability in the group with a lower end point score.13,17,18 Previous findings of more variability in response to placebo as compared with antidepressant derived with the CVR might therefore be associated with a focus in these studies on pre/post difference scores,19,20 which tend to be lower in placebo groups. Owing to the nature of the mean-SD association in our data, the use of VRs and CVRs could produce biased results (as described in the eMethods and depicted in eFigures 1 and 2 in the Supplement). Given that similar investigations of variability in response to other psychiatric treatments and interventions are emerging,49 it is crucial to consider when the VR and CVR are not appropriate and when mixed models that control for the effect of the mean on variability, such as the RSMM, are optimal.13 Also, our findings do not rule out heterogeneity in treatment effects (ie, the difference in outcomes if the same individuals received antidepressant and placebo). According to a 2020 analysis,7 a finding of increased variability in response to treatment suggests treatment effect heterogeneity. Adapting this analysis7 in the context of our study, we show that our finding of no increased variability in response to antidepressant is consistent with a range of potential treatment effect heterogeneities, up to twice the size of the SD in those treated with placebo (described in the eResults in the Supplement).

    The validity of our results rests on the quality of RCTs that were included in our data set, which in some cases was low.21 Because participants in RCTs are selected to be relatively homogenous, our results may also not generalize to patients with MDD seen in clinical practice; it is possible there may be more variability in overall response to antidepressants in patients with greater variability in clinical characteristics. Furthermore, our analysis was limited to RCTs with available data at end point, as well as the 15 drugs from the eligible RCTs. It is possible that some antidepressants that we did not include would produce different results because of their specific mechanisms of action. We grouped antidepressants into classes; therefore, it is likely that our results apply to other antidepressants belonging to the same class. However, few comparisons were available for SNRIs and NDRIs, and our analyses involving these antidepressant classes may have been underpowered.

    Given that we relied on published data, we were limited in the moderators of response variability that we could address. Without access to patient-level or item-level data, we could not examine how various subgroups of patients or symptom profiles affected response variability. This remains an important direction for future work. As discussed in previous work,6,15,16 our analysis does not rule out the possibility of treatment by subgroup interactions or address the potential to personalize treatments based on subsets of symptoms. Finally, our findings also do not address variability in response to nonpharmacological treatments, such as psychotherapies, convulsive therapies, and neuromodulatory interventions, which might possibly be personalized based on their effects.

    Conclusions

    In conclusion, our results do not support the widely held assumption that individual differences underlie the variability in the association between total depression scores and antidepressant treatment. Future efforts toward personalizing treatments for MDD should focus on whether individual symptoms or biomarkers are associated with variability in response to antidepressants or whether there is evidence of variability in response to other treatment types.

    Back to top
    Article Information

    Corresponding Author: Benoit H. Mulsant, MD, MS, Centre for Addiction and Mental Health, University of Toronto, 250 College St, Ste 835, Toronto, ON M5T 1R8, Canada (benoit.mulsant@utoronto.ca).

    Accepted for Publication: November 28, 2020.

    Published Online: February 17, 2021. doi:10.1001/jamapsychiatry.2020.4564

    Author Contributions: Dr Maslej had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

    Concept and design: Maslej, Cipriani, Howes, Mulsant.

    Acquisition, analysis, or interpretation of data: All authors.

    Drafting of the manuscript: Maslej, Andrews, Tomlinson, McCutcheon, Mulsant.

    Critical revision of the manuscript for important intellectual content: Furukawa, Cipriani, Andrews, Sanches, Tomlinson, Volkmann, McCutcheon, Howes, Guo, Mulsant.

    Statistical analysis: Maslej, Sanches, Volkmann, McCutcheon, Guo.

    Obtained funding: Maslej, Cipriani.

    Administrative, technical, or material support: Tomlinson.

    Supervision: Cipriani, Sanches, Howes, Mulsant.

    Conflict of Interest Disclosures: Dr Furukawa reported receiving honoraria for lectures from MSD and consulting fees from Mitsubishi-Tanabe and Shionogi; receiving a grant from Mitsubishi-Tanabe outside the submitted work; and having patent 2018-177688 pending and a copyright concerning Kokoro-app issued. Dr Cipriani has received research and consultancy fees from INCiPiT (Italian Network for Paediatric Trials), CARIPLO Foundation, and Angelini Pharma. Dr Tomlinson has received research and consultancy fees from INCiPiT (Italian Network for Paediatric Trials) and Angelini Pharma. Dr Howes has received investigator-initiated research funding from and/or participated in advisory/speaker meetings organised by Angelini, AstraZeneca, Autifony, Biogen, Boehringer Ingelheim, Eli Lilly, Heptares, Invicro, Janssen, Lundbeck, Lyden-Delta, Mylan, Neurocrine, Otsuka, Sunovion, Rand, Recordati, and Roche. Neither Dr Howes or his family have been employed by or have holdings/a financial stake in any pharmaceutical company. Dr Mulsant reported currently receiving research support from Brain Canada, the Canadian Institutes of Health Research, the CAMH Foundation, the Patient-Centered Outcomes Research Institute, the National Institutes of Health (NIH), Capital Solution Design LLC (software used in a study founded by the CAMH Foundation), and HAPPYneuron (software used in a study founded by Brain Canada); owning stock in General Electric (less than $5000); and having received research support from Eli Lilly (medications for an NIH-funded clinical trial) and Pfizer (medications for a NIH-funded clinical trial). Dr Mulsant holds and receives support from the Labatt Family Chair in Biology of Depression in Late-Life Adults at the University of Toronto. No other disclosures were reported.

    Funding/Support: Dr Maslej is funded by the CIHR Health Systems Impact Postdoctoral Fellowship. Dr Cipriani is supported by the National Institute for Health Research (NIHR) Oxford Cognitive Health Clinical Research Facility, by an NIHR Research Professorship (grant RP-2017-08-ST2-006), by the NIHR Applied Research Collaboration Oxford and Thames Valley at Oxford Health NHS Foundation Trust, and by the NIHR Oxford Health Biomedical Research Centre (grant BRC-1215-20005). Dr McCutcheon is funded by the NIHR. Dr Howes is funded by Medical Research Council-UK (MC U120097115), Maudsley Charity (666), Brain and Behavior Research Foundation, and Wellcome Trust (094849/Z/10/Z) grants to Dr Howes, and the National Institute for Health Research (NIHR) Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. Dr Guo is funded by the National Key R&D Program of China (2018YFC1314600).

    Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

    Disclaimer: The views expressed are those of the authors and not necessarily those of the UK National Health Service, National Institute for Health Research, or UK Department of Health and Social Care.

    Additional Contributions: We thank Cipriani et al21 for making data from their network meta-analysis available online and Nakagawa et al18 for making their code available online. We also thank an anonymous, uncompensated reviewer for suggesting the secondary analysis involving RCT publication year. We thank Benjamin M. Bolker, PhD, Departments of Biology and Mathematics & Statistics, McMaster University, Hamilton, Ontario, Canada, for reviewing our primary analysis and providing his edits and suggestions regarding our code.

    References
    1.
    American Psychiatric Association.  Diagnostic and Statistical Manual of Mental Disorders. 5th ed. Washington, DC: American Psychiatric Association; 2013.
    2.
    National Institute for Health and Care Excellence. Depression in adults: recognition and management: clinical guideline. Published October 28, 2009. Accessed January 15, 2021. https://www.nice.org.uk/guidance/cg90/resources/depression-in-adults-recognition-and-management-pdf-975742636741
    3.
    Thomas  L, Kessler  D, Campbell  J,  et al.  Prevalence of treatment-resistant depression in primary care: cross-sectional data.   Br J Gen Pract. 2013;63(617):e852-e858. doi:10.3399/bjgp13X675430PubMedGoogle ScholarCrossref
    4.
    Noma  H, Furukawa  TA, Maruo  K,  et al.  Exploratory analyses of effect modifiers in the antidepressant treatment of major depression: individual-participant data meta-analysis of 2803 participants in seven placebo-controlled randomized trials.   J Affect Disord. 2019;250:419-424. doi:10.1016/j.jad.2019.03.031PubMedGoogle ScholarCrossref
    5.
    Uher  R.  Genes, environment, and individual differences in responding to treatment for depression.   Harv Rev Psychiatry. 2011;19(3):109-124. doi:10.3109/10673229.2011.586551PubMedGoogle ScholarCrossref
    6.
    Winkelbeiner  S, Leucht  S, Kane  JM, Homan  P.  Evaluation of differences in individual treatment response in schizophrenia spectrum disorders: a meta-analysis  [published online June 3, 2019].  JAMA Psychiatry. 2019;76(10):1063-1073. doi:10.1001/jamapsychiatry.2019.1530PubMedGoogle ScholarCrossref
    7.
    Volkmann  A. On the relationship between treatment effect heterogeneity and the variability ratio effect size statistic. arXiv. Accessed January 15, 2021. https://arxiv.org/pdf/2006.11848.pdf
    8.
    Senn  S.  Mastering variation: variance components and personalised medicine.   Stat Med. 2016;35(7):966-977. doi:10.1002/sim.6739PubMedGoogle ScholarCrossref
    9.
    Cuijpers  P, Christensen  H.  Are personalised treatments of adult depression finally within reach?   Epidemiol Psychiatr Sci. 2017;26(1):40-42. doi:10.1017/S204579601600007XPubMedGoogle ScholarCrossref
    10.
    Kessler  RC.  The potential of predictive analytics to provide clinical decision support in depression treatment planning.   Curr Opin Psychiatry. 2018;31(1):32-39. doi:10.1097/YCO.0000000000000377PubMedGoogle ScholarCrossref
    11.
    Ozomaro  U, Wahlestedt  C, Nemeroff  CB.  Personalized medicine in psychiatry: problems and promises.   BMC Med. 2013;11:132. doi:10.1186/1741-7015-11-132PubMedGoogle ScholarCrossref
    12.
    Simon  GE, Perlis  RH.  Personalized medicine for depression: can we match patients with treatments?   Am J Psychiatry. 2010;167(12):1445-1455. doi:10.1176/appi.ajp.2010.09111680PubMedGoogle ScholarCrossref
    13.
    Volkmann  C, Volkmann  A, Müller  CA.  On the treatment effect heterogeneity of antidepressants in major depression: a Bayesian meta-analysis and simulation study.   PLoS One doi:10.1371/journal.pone.0241497 Google Scholar
    14.
    Mills  HL, Higgins  JP, Morris  RW,  et al.  Detecting heterogeneity of intervention effects using analysis and meta-analysis of differences in variance between arms of a trial.   medRxiv. Accessed May 12, 2020. doi:10.1101/2020.03.07.20032516Google Scholar
    15.
    Winkelbeiner  S, Homan  P.  Is variance ratio a valid indicator of heterogeneous treatment effect? reply.   JAMA Psychiatry. 2020;77(2):217-218. doi:10.1001/jamapsychiatry.2019.3382PubMedGoogle ScholarCrossref
    16.
    Bae  S.  Is variance ratio a valid indicator of heterogeneous treatment effect?   JAMA Psychiatry. 2020;77(2):216-217. doi:10.1001/jamapsychiatry.2019.3372PubMedGoogle ScholarCrossref
    17.
    Guo  X, McCutcheon  RA, Pillinger  T,  et al.  The magnitude and heterogeneity of antidepressant response in depression: A meta-analysis of over 45,000 patients.   J Affect Disord. 2020;276:991-1000. doi:10.1016/j.jad.2020.07.102PubMedGoogle ScholarCrossref
    18.
    Nakagawa  S, Poulin  R, Mengersen  K,  et al  Meta-analysis of variation: ecological and evolutionary applications and beyond.   Methods Ecol Evol. 2015;6(2):143-152. doi:10.1111/2041-210X.12309Google ScholarCrossref
    19.
    Plöderl  M, Hengartner  MP.  What are the chances for personalised treatment with antidepressants? detection of patient-by-treatment interaction with a variance ratio meta-analysis.   BMJ Open. 2019;9(12):e034816. doi:10.1136/bmjopen-2019-034816PubMedGoogle Scholar
    20.
    Munkholm  K, Winkelbeiner  S, Homan  P. Individual response to antidepressants for depression in adults–a simulation study and meta-analysis. PloS One. 2020;15(8):e0237950.
    21.
    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(10128):1357-1366. doi:10.1016/S0140-6736(17)32802-7PubMedGoogle ScholarCrossref
    22.
    Fournier  JC, DeRubeis  RJ, Hollon  SD,  et al.  Antidepressant drug effects and depression severity: a patient-level meta-analysis.   JAMA. 2010;303(1):47-53. doi:10.1001/jama.2009.1943PubMedGoogle ScholarCrossref
    23.
    Stone  M, Kalaria  S, Richardville  K,  et al Components and trends in treatment effects in randomized placebo-controlled trials in major depressive disorder from 1979-2016. Presented at: American Society of Clinical Psychopharmacology Annual Conference; May 30, 2018; Miami, FL.
    24.
    Furukawa  TA, Salanti  G, Atkinson  LZ,  et al.  Comparative efficacy and acceptability of first-generation and second-generation antidepressants in the acute treatment of major depression: protocol for a network meta-analysis.   BMJ Open. 2016;6(7):e010919. doi:10.1136/bmjopen-2015-010919PubMedGoogle Scholar
    25.
    Hamilton  M.  A rating scale for depression.   J Neurol Neurosurg Psychiatry. 1960;23(1):56-62. doi:10.1136/jnnp.23.1.56PubMedGoogle ScholarCrossref
    26.
    Bech  P.  Clinical Psychometrics. Oxford, England: Wiley Blackwell; 2012. doi:10.1002/9781118511800
    27.
    Williams  JBW, Link  MJ, Rosenthal  NE, Terman  M.  Structured Interview Guide for the Hamilton Depression Rating Scale, Seasonal Affective Disorders Version (SIGH-SAD). New York, New York: State Psychiatric Institute, Biometrics Research; 1988.
    28.
    Roberts  JN, Lei  HH, Krishen  A,  et al Identifying sub-scales of the 31-item Hamilton Depression Scale. Poster presented at: New Clinical Drug Evaluation Unit (NCDEU) meeting; May 28-31, 2001; Phoenix, AZ.
    29.
    Montgomery  SA, Asberg  M.  A new depression scale designed to be sensitive to change.   Br J Psychiatry. 1979;134:382-389. doi:10.1192/bjp.134.4.382PubMedGoogle ScholarCrossref
    30.
    Rush  AJ, Giles  DE, Schlesser  MA, Fulton  CL, Weissenburger  J, Burns  C.  The Inventory for Depressive Symptomatology (IDS): preliminary findings.   Psychiatry Res. 1986;18(1):65-87. doi:10.1016/0165-1781(86)90060-0PubMedGoogle ScholarCrossref
    31.
    Raudenbush  SW, Bryk  AS.  Examining correlates of diversity.   J Educ Stat. 1987; 12(3):241-69. doi:10.3102/10769986012003241Google ScholarCrossref
    32.
    Ghaemi  SN.  A new nomenclature for psychotropic drugs.   J Clin Psychopharmacol. 2015;35(4):428-433. doi:10.1097/JCP.0000000000000341PubMedGoogle ScholarCrossref
    33.
    Owens  MJ, Morgan  WN, Plott  SJ, Nemeroff  CB.  Neurotransmitter receptor and transporter binding profile of antidepressants and their metabolites.   J Pharmacol Exp Ther. 1997;283(3):1305-1322.PubMedGoogle Scholar
    34.
    R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Published 2019. Accessed January 12, 2020. https://www.R-project.org/
    35.
    Hadfield  J.  MCMC methods for multi-response generalized linear mixed models: the MCMCglmm R package.   J Stat Softw. 2010;33(2):1-22. doi:10.18637/jss.v033.i02Google ScholarCrossref
    36.
    Open Science Foundation. Individual differences in response to antidepressants: R Code. Accessed November 13, 2020. https://mfr.ca-1.osf.io/render?url=https://osf.io/m6u89/?direct%26mode=render%26action=download%26mode=render
    37.
    Chekroud  AM, Gueorguieva  R, Krumholz  HM, Trivedi  MH, Krystal  JH, McCarthy  G.  Reevaluating the efficacy and predictability of antidepressant treatments: a symptom clustering approach.   JAMA Psychiatry. 2017;74(4):370-378. doi:10.1001/jamapsychiatry.2017.0025PubMedGoogle ScholarCrossref
    38.
    Furukawa  TA, Maruo  K, Noma  H,  et al.  Initial severity of major depression and efficacy of new generation antidepressants: individual participant data meta-analysis.   Acta Psychiatr Scand. 2018;137(6):450-458. doi:10.1111/acps.12886PubMedGoogle ScholarCrossref
    39.
    Rabinowitz  J, Werbeloff  N, Mandel  FS, Menard  F, Marangell  L, Kapur  S.  Initial depression severity and response to antidepressants v. placebo: patient-level data analysis from 34 randomised controlled trials.   Br J Psychiatry. 2016;209(5):427-428. doi:10.1192/bjp.bp.115.173906PubMedGoogle ScholarCrossref
    40.
    Gibbons  RD, Hur  K, Brown  CH, Davis  JM, Mann  JJ.  Benefits from antidepressants: synthesis of 6-week patient-level outcomes from double-blind placebo-controlled randomized trials of fluoxetine and venlafaxine.   Arch Gen Psychiatry. 2012;69(6):572-579. doi:10.1001/archgenpsychiatry.2011.2044PubMedGoogle ScholarCrossref
    41.
    Fried  EI, Nesse  RM.  Depression is not a consistent syndrome: An investigation of unique symptom patterns in the STAR*D study.   J Affect Disord. 2015;172:96-102. doi:10.1016/j.jad.2014.10.010PubMedGoogle ScholarCrossref
    42.
    Fried  EI, Nesse  RM.  Depression sum-scores don’t add up: why analyzing specific depression symptoms is essential.   BMC Med. 2015;13(1):72. doi:10.1186/s12916-015-0325-4 PubMedGoogle ScholarCrossref
    43.
    Niculescu  AB, Le-Niculescu  H, Levey  DF,  et al.  Precision medicine for suicidality: from universality to subtypes and personalization.   Mol Psychiatry. 2017;22(9):1250-1273. doi:10.1038/mp.2017.128PubMedGoogle ScholarCrossref
    44.
    Green  E, Goldstein-Piekarski  AN, Schatzberg  AF, Rush  AJ, Ma  J, Williams  L.  Personalizing antidepressant choice by sex, body mass index, and symptom profile: An iSPOT-D report.   Personalized Medicine in Psychiatry. 2017;1:65-73. doi:10.1016/j.pmip.2016.12.001Google ScholarCrossref
    45.
    Picardi  A, Viroli  C, Tarsitani  L,  et al.  Heterogeneity and symptom structure of schizophrenia.   Psychiatry Res. 2012;198(3):386-394. doi:10.1016/j.psychres.2011.12.051PubMedGoogle ScholarCrossref
    46.
    Tandon  R, Gaebel  W, Barch  DM,  et al.  Definition and description of schizophrenia in the DSM-5.   Schizophr Res. 2013;150(1):3-10. doi:10.1016/j.schres.2013.05.028PubMedGoogle ScholarCrossref
    47.
    Durisko  Z, Mulsant  BH, Andrews  PW.  An adaptationist perspective on the etiology of depression.   J Affect Disord. 2015;172:315-323. doi:10.1016/j.jad.2014.09.032PubMedGoogle ScholarCrossref
    48.
    Thase  ME, Entsuah  AR, Rudolph  RL.  Remission rates during treatment with venlafaxine or selective serotonin reuptake inhibitors.   Br J Psychiatry. 2001;178(3):234-241. doi:10.1192/bjp.178.3.234PubMedGoogle ScholarCrossref
    49.
    Radua  J, Davies  C, Fusar-Poli  P.  Evaluation of variability in individual response to treatments in the clinical high-risk state for psychosis: a meta-analysis.   Schizophr Res.S0920-9964(20)30262-0. doi:10.1016/j.schres.2020.05.010PubMedGoogle Scholar
    ×