Advancing the Treatment of Depression With Personalized Medicine

Andrew A. Nierenberg, MD

Department of Psychiatry and the Depression Clinical Research Program, Harvard Medical School and Massachusetts General Hospital, Boston

The goal of personalized medicine is to enable clinicians to match specific treatments to individual patients to increase the probability of an optimal outcome. However, to implement personalized medicine, clinicians need a clinically useful taxonomy to characterize groups of patients by variables. To create a useful classification system of personalizing variables, many different groups of patients, the relationship between these groups and their treatment outcomes, and their differential responses to many treatments, need to be studied. Variables from any domain (eg, demographic, clinical, physiologic, and genomic) can provide information but, to be useful, must not provide redundant information.

Types of Variables

A predictor of response is a variable that suggests that a person will respond to a given treatment. A negative predictor is one that decreases the probability of response to any treatment. Other types of variables that influence response are moderators and mediators.1 Moderators are baseline variables that predict a difference in outcomes between one medication and another. Moderators, therefore, indicate in which person, or under what circumstance, a treatment would have the most clinically significant effect. Mediators are variables that change early in treatment and predict the difference in outcomes between one treatment and another. For example, if an ultimate outcome of nonresponse can be predicted by week 2 because of a certain mediator, then that knowledge can inform the clinician to change treatment earlier rather than later.

Promises and Pitfalls of Personalized Medicine

Personalized medicine holds promise, but its pitfalls should not be overlooked. Tailoring treatments to individual patients promises improved outcomes (such as sustained remission) beyond that which a reasonably informed clinician could achieve through a trial and error process. The precision of predicting response to treatments will increase, and the probability of intolerable side effects or other adverse events occurring will decrease.

This premise may sound simple, but, currently, data for many variables are insufficient to match treatments to patients. Very large trials may be needed because raising the number of variables in a study increases the number of strata, which can mean that too few patients are in each stratum to study the variables efficiently.2 Increasing the sample size of studies, however, increases costs. With too few subjects, determining the statistical versus clinical significance of a particular variable is difficult. The effect size of the variable may be too small to predict which treatment should be given to a patient even though a statistically significant difference occurred. Another problem is that comparative head-to-head effectiveness studies in a generalized population do not exist for many antidepressants. So, much more research is required before reliable personalized medicine is available.

Studies of Personalizing Variables

Demographic and clinical variables. The personalizing variables that are typically employed in antidepressant studies are demographic (eg, age and sex) and clinical (eg, functional status, general medical burden, the presence or absence of psychosis, and subtype of depression). These variables are not mutually exclusive.

Older age is a negative predictor of antidepressant response; in patients above age 65 or 70 years, treatment is expected to be less effective than in younger patients, and no one treatment has been shown to be more effective than another.3 Similarly, worse functioning and greater medical burden are negative predictors of response,4 and, in patients with these features, no treatment type is preferred over another.

Sex and hormonal status, however, may be moderators of antidepressant response. Some treatments may have differential efficacy in women versus men. Kornstein and colleagues5 found that premenopausal women responded better to SSRIs than to TCAs but men responded better to TCAs than to SSRIs. Postmenopausal women responded similarly to both medications. However, clear evidence that sex can be used as a personalizing variable is not yet available. Parker et al6 failed to find a difference between men and women in SSRI and TCA response, but the sample sizes were small. Quitkin and colleagues7 studied response to TCAs, MAOIs, and an SSRI using the variables sex, age, and menopausal status but, again, small sample sizes yielded inconclusive results. Thase et al8 reported that remission rates with SSRIs were equivalent (about 35%) among women under 50 years, women taking HRT, and men of any age, but were lower (about 27%) among women 50 years or older and those not taking HRT. Replication of findings is needed to confirm clinically useful moderators.

In practice, clinicians may be faced with a conundrum when considering multiple variables. Although a variable like female sex may be a moderator of response to a particular treatment, a female patient may present with additional factors like older age, poor functioning, and several comorbid medical conditions, all of which are negative predictors of treatment response. The predictors will not guide treatment, but moderators will.


Melancholic, atypical, anxious, and psychotic subtypes of depression have also been evaluated as personalizing variables. Patients with melancholia may have a preferential response to TCAs versus SSRIs, but this long-standing debate is unresolved.9 Atypical depression is predictive of a better response to MAOIs than TCAs.10 Anxiety is a negative predictor of remission,11 and no difference in response between SSRIs and bupropion was found in those with anxious depression.12 Psychosis appears to be a moderating variable of response,13 but this result needs to be replicated. In 2 placebo-controlled trials, Rothschild and colleagues13 examined antipsychotic monotherapy and an antipsychotic plus an antidepressant in patients with psychotic depression. Subjects showed significantly better response to the combination therapy than to monotherapy in 1 trial (P = .027) but not in the other trial.

Neurophysiologic variables. Some physiologic features are personalizing variables and others are not. Structural MRI, for example, can show hippocampal atrophy, possibly indicating chronic depression, which is associated with decreased antidepressant response3—but chronic depression is the personalizing variable, not hippocampal atrophy. However, quantitative EEG can detect a biomarker change, which, at baseline and after 1 week of SSRI treatment, predicted response and remission 6 weeks later with 74% accuracy.14 This biomarker is a mediating variable in that it can predict outcomes early in treatment and help personalize treatment over time. Other physiologic features may be examined by functional neuroimaging studies, peripheral measures of allostatic load, and other tests, but differential responses to treatment are not yet predictable by these methods.

Genomic variables. Applying genomics to psychotropic medication selection is more complicated than originally thought, and personalizing variables cannot yet be clearly distinguished. Polymorphisms in serotonin transporter genes, BDNF genes, and many other genes may be related to antidepressant response.15 Studies of genes associated with antidepressant pharmacodynamics and pharmacokinetics have had heterogeneous findings.15 Combining genetics with neuroimaging may clarify how circuits are coded structurally and functionally.16 Larger sample sizes are needed for genomic studies, however. Even large studies such as STAR*D have not been big enough to achieve genome-wide significance.16 In addition to genomics, transcriptomics and proteomics—fields that provide information about mRNA and protein products made by genes, respectively—may need to be involved in predicting individual antidepressant response.16

Individualized Comparative Effectiveness Studies

AV 1. Individualized Comparative Effectiveness Studies: Matching Treatment to Patients (00:43)

Individualized comparative effectiveness studies in a generalized population are needed to decide which individuals should receive particular treatments. While comparative effectiveness studies can find which treatment is better overall, an individualized comparative effectiveness study can determine who should get treatment A and not B, or B and not A, and so forth (AV 1).

Several individualized comparative effectiveness studies indicate variables that could be used in practice. For example, Trivedi and colleagues17 examined whether adjunctive light or heavy aerobic exercise would improve outcomes in patients with moderate MDD who had a partial response to an SSRI. Sex and family history of mental illness, the treatment response moderators, only showed significant differences between groups when analyzed separately (AV 2).

AV 2. Sex and Family History (FH) of Mental Illness as Moderators of Exercise Augmentation for Depression (00:40)

Data from Trivedi et al17
KKW = kcal per kg per week using treadmills and/or cycle ergometers

In another study,18 responses to either an antidepressant or CBASP were compared in patients with chronic MDD who had or had not experienced childhood trauma. For achieving remission, the antidepressant was more effective in those without childhood trauma, while psychotherapy was more effective for those with childhood trauma.

A large individualized comparative effectiveness study, EMBARC, is underway to identify clinical, neuroimaging, neurophysiologic, and behavioral moderators and mediators of response to 2 antidepressants versus placebo. If replicable, EMBARC may provide generalizable results to personalize treatment. Other research to identify genetic, neuroimaging, and cognitive markers of antidepressant response is also ongoing.


Identifying personalized variables that can predict antidepressant response offers the opportunity to match treatments to patients, which can improve efficiency, reduce adverse events, and have a major public health impact. Challenges to making personalized medicine a reality include limited funding for studies, the need for carefully structured and replicable studies that provide generalizable results, and the complex implementation process for clinicians to establish personalized treatment in practice.

For Clinical Use


  • Multiple variables may affect treatment outcomes in a patient
  • Treatment outcomes may be influenced by sex, hormonal status, type of depression, childhood trauma, family history of mental illness, and some biomarkers and genetic polymorphisms


BDNF = brain-derived neurotrophic factor, CBASP = cognitive-behavioral analysis system of psychotherapy, EEG = electroencephalography, EMBARC = Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care for Depression, FH = family history, HRT = hormone replacement therapy, KKW = kcal per kg per week, MAOI = monoamine oxidase inhibitor, MDD = major depressive disorder, MRI = magnetic resonance imaging, mRNA = messenger ribonucleic acid, SSRI = selective serotonin reuptake inhibitor, STAR*D = Sequenced Treatment Alternatives to Relieve Depression, TCA = tricyclic antidepressant

Take the online posttest.


  1. Kraemer HC, Wilson GT, Fairburn CG, et al. Mediators and moderators of treatment effects in randomized clinical trials. Arch Gen Psychiatry. 2002;59(10):877–883. PubMed
  2. Leon AC. Two clinical trial designs to examine personalized treatments for psychiatric disorders. J Clin Psychiatry. 2011;72(5):593–597. Abstract
  3. Fournier JC, DeRubeis RJ, Shelton RC, et al. Prediction of response to medication and cognitive therapy in the treatment of moderate to severe depression. J Consult Clin Psychol. 2009;77(4):775–787. PubMed
  4. Sinyor M, Schaffer A, Levitt A. The sequenced treatment alternatives to relieve depression (STAR*D) trial: a review. Can J Psychiatry. 2010;55(3):126–135. PubMed
  5. Kornstein SG, Schatzberg AF, Thase ME, et al. Gender differences in treatment response to sertraline versus imipramine in chronic depression. Am J Psychiatry. 2000;157(9):1445–1452. PubMed
  6. Parker G, Parker K, Austin MP, et al. Gender differences in response to differing antidepressant drug classes: two negative studies. Psychol Med. 2003;33(8):1473–1477. PubMed
  7. Quitkin FM, Stewart JW, McGrath PJ, et al. Are there differences between women's and men's antidepressant responses? Am J Psychiatry. 2002;159(11):1848–1854. PubMed
  8. Thase ME, Entsuah R, Cantillon M, et al. Relative antidepressant efficacy of venlafaxine and SSRIs: sex-age interactions. J Women's Health (Larchmt). 2005;14(7):609–616. PubMed
  9. Uher R, Dernovsek MZ, Mors O, et al. Melancholic, atypical and anxious depression subtypes and outcome of treatment with escitalopram and nortriptyline. J Affect Disord. 2011;132(1–2):112–120. PubMed
  10. Quitkin FM. Depression with atypical features: diagnostic validity, prevalence, and treatment. Prim Care Companion J Clin Psychiatry. 2002;4(3):94–99. Abstract
  11. Fava M, Rush AJ, Alpert JE, et al. Difference in treatment outcome in outpatients with anxious versus nonanxious depression: a STAR*D report. Am J Psychiatry. 2008;165(3):342–351. PubMed
  12. Rush AJ, Batey SR, Donahue RM, et al. Does pretreatment anxiety predict response to either bupropion or sertraline? J Affect Disord. 2001;64(1):81–87. PubMed
  13. Rothschild AJ, Williamson DJ, Tohen MF, et al. A double-blind, randomized study of olanzapine and olanzapine/fluoxetine combination for major depression with psychotic features. J Clin Psychopharmacol. 2004;24(4):365–373. PubMed
  14. Leuchter AF, Cook IA, Marangell LB, et al. Comparative effectiveness of biomarkers and clinical indicators for predicting outcomes of SSRI treatment in major depressive disorder: results of the BRITE-MD study. Psychiatry Res. 2009;169(2):124–131. PubMed
  15. Porcelli S, Drago A, Fabbri C, et al. Pharmacogenetics of antidepressant response. J Psychiatry Neurosci. 2011;36:(2):87–113. PubMed
  16. Laje G, McMahon FJ. Genome-wide association studies of antidepressant outcome: a brief review. Prog Neuropsychopharmacol Biol Psychiatry. 2011;35:(7):1553–1557. PubMed
  17. Trivedi MH, Greer TL, Church TS, et al. Exercise as an augmentation treatment for nonremitted major depressive disorder: a randomized, parallel dose comparison. J Clin Psychiatry. 2011;72(5):677–684. Abstract
  18. Nemeroff C, Heim C, Thase M, et al. Differential responses to psychotherapy versus pharmacotherapy in patients with chronic forms of major depression and childhood trauma. Proc Natl Acad Sci USA. 2003;100(24):14293–14296. PubMed