Clinical Predictors in Major Depressive Disorder

Madhukar H. Trivedi, MD
Benji T. Kurian, MD
Bruce D. Grannemann, MA

Dr. Trivedi is the Lydia Bryant Test Professor in Psychiatric Research, professor of psychiatry, and director of the Mood Disorders Research Program, Dr. Kurian is a post-doctoral research fellow; and Mr. Grannemann is a faculty member in the Department of Psychiatry at the University of Texas Southwestern in Dallas.

Abstract

Remission is the goal for current depression treatment. However, the Sequenced Treatment Alternatives to Relieve Depression study has shown that the majority of patients will fail to achieve remission with a first-line antidepressant agent. Previous research has attempted to identify which depression treatments are preferred for whom by assessing baseline predictors. Of predictors, sex, age, severity of illness, depressive subtype, and comorbidity have predicted treatment response/nonresponse. However, these predictors have not always provided meaningful clinical correlation. Furthermore, based on the results of recent research, it is clear that clinicians need predictive variables to identify “next-best” preferred depression treatments for patients. This article defines these predictive variables as process predictors—that is they include clinical features that appear during the treatment process and are associated with outcomes.

INTRODUCTION

Depressive disorders are one of the leading causes of disability-adjusted life years, worldwide.¹ In the United States, depressive disorders account for >31 billion dollars a year in lost employee productivity.² This debilitating illness will affect up to 16.2% of Americans by the end of their lifetime.³

Antidepressant medications are the most common form of treatment for adults suffering from major depressive disorder (MDD).⁴ Among antidepressants, selective serotonin reuptake inhibitors (SSRIs) are the most commonly prescribed agents in the US.⁴ However, based on recent clinical trials evaluated in “real-world” settings, first-step treatment with an SSRI is only 33% effective in achieving remission of depressive symptoms.⁵ Furthermore, about one-third of patients with MDD will become treatment resistant (ie, remain symptomatic despite multiple trials of antidepressant medications).⁶

Since a large number of patients fail to respond to antidepressants, an emphasis has been placed on identifying predictors of treatment success. To date, most research on predicting treatment response has focused on baseline predictors. This article includes not only a review of new research on baseline predictors but will also examine a new type of predictor—process predictors—that include changes that occur early in treatment, which may relate to final outcomes and can be used to tailor treatment to individual patients once treatment has been initiated.

CURRENT FINDINGS: Baseline Predictors

Baseline predictors are based on information that is available to clinicians at the initiation of treatment. Baseline predictors are important because patient features at the point of treatment initiation could inform clinicians as to what treatments may be best for a specific patient or which treatments may be more effective than other treatments. Although this type of predictor would be of the greatest use to clinicians and the majority of research has pursued the identification of baseline predictors, the majority of findings examining baseline predictors seldom prove useful in the clinical setting because of their modest to small effect size.⁷ Baseline predictors are divided here into five different categories: sociodemographic factors, depressive illness features, symptom factors, comorbidity factors, and a review of the burgeoning field of biological predictors.

Sociodemographic Predictors

Most research to date has focused on assessing whether baseline factors (ie, sociodemographic, illness features, symptom presentation, and comorbid conditions) predict future antidepressant response (Table).^5,8-16 Few markers have been proven to reliably predict a positive antidepressant response.^11,12,17-19 Some replicated studies have forecasted that less education, single living status, and low baseline quality of life predict treatment non-response to antidepressants.^8,9,13,19,20 In fact, Trivedi and colleagues^8,9 found that living status (married or cohabiting) predicted better treatment response than living alone (single, engaged, divorced, widowed). Furthermore, studies have reliably replicated the effect of gender and age; specifically, younger women are more likely to respond to SSRIs and less likely to tolerate or respond to tricyclic antidepressants (TCAs).^10,16,21 Additionally, Kornstein and colleagues¹⁶ hypothesize that female sex hormones may account for this effect (ie, premenopausal status may predict a better treatment response to SSRIs versus postmenopausal status).

Illness Features

Several illness characteristics can be used as negative predictors. For example, patients with increased baseline depression severity are less likely to respond to antidepressants.^5,15 Another factor associated with MDD, the illness length both of the current episode⁵ and of prior episodes,^15,22 has shown to negatively predict treatment response. However, age of onset of first depressive episode has variably predicted symptom response. Some studies find younger age of onset to have less favorable treatment response,^15,23 while others found no association.^14,24 Thus, a lower rate of response with more severe illness, as indicated by severity at baseline and longer duration of episodes.

Symptom Features

For years, clinical symptom presentations have been used to describe depressive subtypes. One of the oldest subtypes is melancholic depression (ie, “endogenous depression”) which is described in the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR) as having an essential feature of “loss of interest or pleasure in all, or almost all, activities or a lack of reactivity to usually pleasurable stimuli.”²⁵ As with other predictors of MDD, no clear factors associated with melancholia have reliably predicted treatment response to antidepressant medications.^26,27 Past studies have shown that £50% of patients suffering from melancholic depression have abnormal baseline dexamethasone suppression tests; however, no correlation to pharmacologic response has been associated.^28,29

Another well-studied depressive subtype is atypical depression, which is primarily characterized by mood reactivity or an ability to be “cheered up” by positive events.^25,30 A seminal finding by Quitkin and colleagues³¹ is that those suffering from atypical depression are more likely to respond to monoamine oxidase inhibitors (MAOIs) than TCAs.^31-33 However, when assessing SSRI versus MAOI treatment response in depressed patients with atypical features, results have been mixed and unclear.^10,34 With regard to clinical prediction and further delineation of the atypical diagnoses, Stewart and colleagues³⁵ found that patients with atypical depression who had early onset of depressive illness and disease chronicity responded poorly to TCAs.

Despite the fact that the DSM-IV-TR does not currently hold a specifier for depression with anxious features, a number of studies have described this correlate in the literature.^36-40 In fact, based on research from the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) trial, 46% of patients met baseline criteria for anxious depression.³⁶ Literature has shown that patients with MDD and comorbid anxiety are more likely to suffer increased disease burden and longer duration of illness.^39,41 Furthermore, some studies have shown that patients suffering from anxious depression are less likely to respond to antidepressants,^41-43 while other studies report similar response rates for anxious and non-anxious depression, but a longer time to response for the former group.^44,45

Irritability, although not a distinct depressive subtype, does correlate with increased depression severity and further studies to assess the prediction of treatment response are warranted.⁴⁶ Additionally, physical pain is a common symptom accompanying MDD that appears to be associated with increased severity of illness and may have a slightly better treatment response to dual-action antidepressants.^47,48

Comorbidity Predictors

MDD that is comorbid with medical illnesses has also shown mixed results with regard to prediction, with some large studies predicting a lower response rate to antidepressants, while other smaller studies report a potentially insignificant difference between medical illness and no medical illness.^49,50 Meanwhile, depressive illnesses comorbid with personality disorders have been associated with both delayed response to treatment⁵¹ and a negative predictive value.^52,53

Neuroticism is a personality trait that is a reliable predictor of treatment non-response.^13,18,54,55 Katon and colleagues¹³ define patients with neuroticism as having low tolerance for stress, feeling alienated, having low-self-esteem, and being anxious worriers, and feeling easily victimized and resentful.

Biological Predictors

This article seeks to provide clinicians with a practical guide to predictors of treatment response and, although biological predictors (eg, neuroimaging, pharmacogenetics, quantitative electroenchalography [QEEG]) are yet to be clinically applicable, it is important to present a brief introduction into new directions in the field. A recent study reported that patients with hypofolatemia and/or associated brain white matter hyperintensities, measured via magnetic resonance imaging (MRI), had a lower probability to respond to antidepressants.⁵⁶

More recently, researchers have focused on pharmacogenetic predictors of antidepressant response, and results have been mixed.^57-59 The interplay of pharmacogenetic markers and pharmacologic response is complex and potentially variable among ethnic populations.^58,59 Researchers have also studied neurophysiologic factors as potential predictors of treatment response, of which prefrontal QEEG cordance appears to hold the most promise.^60-63 Perlis⁶⁴ has extensively described the role of genetic predictors.

Although the development of a clinical factor that reliably predicts pharmacologic response is of great importance, a vast majority of patients will not achieve symptom remission with a first-line agent. Thus, clinicians should devise a methodology to predict response for the “next-best treatment.” The STAR*D study was designed and executed with the knowledge that depression is a chronic illness with the goal of attaining remission in “real-world” settings.⁶⁵ Therefore, it is necessary to identify process predictors that will allow clinicians to identify next-best treatment options for patients who have failed prior approaches.

A critical factor in evaluating baseline predictor research is that the majority of research examines one or two active treatment medications. Thus, identifying variables that could differentiate between treatments becomes very difficult.

NEW DIRECTIONS: Process Predictors

Process predictors are based on information that becomes available to the clinician during the process of treatment. Examples of process predictors include timing and nature of change in symptom severity in the first few weeks of treatment, side-effect burden, and patient adherence. The concept of process predictors has become important because of the new direction that depression treatment has taken over the last 10 years. Since the goal of treatment has become remission as opposed to response, results of large clinical trials show that for the majority of patients, the first treatment is unlikely to achieve this goal.⁵ Thus, the ability to predict early in the course of treatment which particular patients will achieve remission and which patients will not achieve remission is vitally important.

A clinical example of a process predictor is using measures of depression severity during the early weeks of treatment to predict antidepressant success/failure.⁶⁶ That is, the changes from baseline scores on standard symptom severity measures in the first few weeks of treatment could be used to predict the patient’s final status. Other measures of this type may include treatment adherence, side-effect burden, or changes in specific symptom clusters within the MDD domain, such as anxiety, cognition, and core emotions. Nierenberg and colleagues⁶⁷ described this approach in 1995, as they used changes from baseline in the first 4 weeks of treatment to predict final status at the end of an 8-week trial. In this study, the authors found, for patients with a reduction in Hamilton Rating Scale for Depression (HAM-D)⁶⁸ scores of <20%, there was a <30% chance of showing a full response (response defined as a 50% HAM-D improvement from baseline) on week 8.

Another study conducted by Nierenberg and colleagues⁶⁹ was designed to examine the timing of antidepressant response with fluoxetine. While there was no evidence for a specific time for response, this study did show that patients who had not shown a response in the first 4 weeks of treatment were unlikely to respond. In fact among patients who had not shown a response in the first 2 weeks of treatment, <75% showed response by the end of 8 weeks. This study suggests that the initial response to treatment may be a critical predictor of whether or not a patient will eventually respond to a treatment.

In another study, Perlis and colleagues¹⁹ examined three different augmentation strategies following a 4-week trial of fluoxetine 20 mg/day to assess predictors of treatment augmentation. The researchers compared increasing the dose of fluoxetine, addition of desipramine, or addition of lithium. While the study failed to find any predictors of specific treatment, the study reported three elements that predicted whether or not patients would achieve a response. Two of these elements were baseline predictors marital status and age of depression onset. In addition, the study also reported a process type predictor: the severity of patients’ HAM-D score after 4 weeks of treatment predicted likelihood of response with the augmentation.

Trivedi and colleagues⁷⁰ have also used changes in symptom status as a predictor of response. In this study, patients who were enrolled in an open-label 12-week trial and had not shown a response by week 4 were assessed. The changes in symptom clusters during the first 4 weeks of treatment were used to predict whether or not these patients would achieve a full response by the end of the 12-week trial. Findings suggest that changes in the symptom cluster scores between weeks 3 and 4 were best at determining which patients would eventually achieve a full response. A discriminant analysis correctly identified 70% of the late responders and 64% of the non-responders.

In general, predictor research has focused on identifying variables that inform clinicians of the best treatment for a given patient. However, emphasis on process predictor research is designed to identify which patients should be continued on current treatment and which patients are unlikely to benefit from the current treatment during various time points in the acute phase. This may be one of the most beneficial types of prediction to reduce unnecessarily long and ineffective treatment trials, further enhancing the chances of identifying the most effective treatment at the earliest time and potentially reducing attrition from treatment. This benefit is particularly true if the treatment goal is to achieve full remission. Based on the STAR*D study, patients achieve only modest rates of remission with the first treatment.⁵ Therefore, if researchers can shorten unsuccessful treatment time, clinicians should be able to more rapidly switch patients to a more effective treatment.

DISCUSSION

Treatment of MDD is complicated because of available antidepressants’ slow onset and low remission rates; uncertainty about the optimal duration of an adequate acute treatment trial; high rates of residual symptomatology; and need for sequential treatment steps and associated response heterogeneity. Furthermore, the improvement trajectory after an antidepressant is initiated as a determinant of eventual outcome has proved to be of limited utility. These difficulties have led to a treatment process in clinical care that is based on trial and error until an effective treatment is identified. The recently completed STAR*D trial found that up to 67% of patients can achieve remission with four treatment steps used sequentially,^5,71-76 and that well-defined measurement-based care can be used to tailor treatment during each of the acute phase steps to maximize outcomes.^5,77,78 (Measurement-based care involves use of response as determined by rating scales to guide dose adjustment). One conclusion that can be drawn from the sequential treatment trials is that there is a clear need for methods to predict which treatment will work for which patient. Well-defined predictors are likely to assist in identification of the most appropriate treatment; determination of the duration of an adequate treatment trial for a given patient; identification of the next best step treatment, and determination of the nature and timing of intervention for residual or breakthrough symptoms during the long-term management of patients with MDD. There are at least two types of clinical predictors that could be of use to practicing clinicians. First, information used at the initiation of treatment can aid in the treatment choice. Second, information accrued during the treatment course can determine whether continuation of the current treatment will produce full and sustained remission of depressive symptoms.

Research on predictors has predominantly focused on identification of a number of variables that could be used to predict outcomes for groups of patients as a whole. Thus, researchers focus on variables that help them understand the underlying etiology or pathophysiology of the disorder. Since there are no direct or absolute measures of depression (ie, the measurement of the underlying depression is typically based on patient reports or clinician observations of patient behavior) there remains a fair amount of uncertainty about the contribution of each of the variables in predicting outcomes. In contrast, clinicians have to apply the results of studies based on group data to individual patients. Specifically, there is a need for predictors that can be used to guide treatment in terms of initiating the “right” antidepressant treatment; determining the timing of a treatment change, and assessing the need for adjunctive treatments for associated symptoms, side effects, or residual symptoms. Clinicians are also required to assist patients by educating them about what to expect in terms of symptom change as well as disease self-management in order to personalize treatment and engage the patient in the treatment process. Achieving these goals would not be a problem if the clinical and or biological variables identified as significant predictors were more robust. However, in a typical depression study, any given predictor is likely to explain only a small percentage of the variance. This results in a situation in which characteristics that have been identified as predictors in research studies may not be of practical use in predicting the outcome for a given patient. This situation has been described in several previous studies of depression predictors.^7,18,79 Another reason why clinicians are often unable to assess the usefulness of a predictor from research studies is the fact that effect size estimates are not provided. One attempt to overcome this shortcoming has been the inclusion of odds ratios or “number needed to treat” as part of the measurement of the outcome effect.

There have been attempts at evaluating predictors in terms of their usefulness to a practicing clinician. Many of these attempts have also focused on changes occurring during the course of treatment (ie, process predictors). Effective baseline predictors would be most useful for practicing clinicians since beginning patients on a treatment that will work immediately is ideal. However, process predictors appear to have larger predictive power, thereby providing clinicians with more relevant clinical utility.

CONCLUSION

Studies should report effect sizes in order to assist clinicians in evaluating the strength of individual predictors. Furthermore, in order to evaluate the effectiveness of the predictor, sensitivity as well as specificity of predictors should be reported so that both the positive and negative prediction for a given variable can be quantified. Most studies of baseline predictors report the clinical significance of their effects on treatment response, however this effect is modest. Thus, process predictors provide the most clinically useful guide during treatment decision making. The utility of process predictors depends on the routine measurement of variables using standardized rating instruments and has led to the development of measurement-based care as used in the STAR*D study.

Disclosures: Dr. Trivedi is a consultant to AstraZeneca, Bristol-Myers Squibb, Cephalon, Cyberonics, Eli Lilly, Forest, GlaxoSmithKline, Neuronetics, Novartis, VantagePoint, and Wyeth; is on the speaker’s bureaus of Bristol-Myers Squibb, Cephalon, Cyberonics, Eli Lilly, Forest, GlaxoSmithKline, and Wyeth; and receives research support from Bristol-Myers Squibb, Cephalon, Corcept, Cyberonics, Forest, the National Alliance for Research in Schizophrenia and Depression, the National Institute of Mental Health, Novartis, Predix, and Wyeth.

Please direct all correspondence to: Madhukar H. Trivedi, MD, Mood Disorders Program and Clinic, Department of Psychiatry, University of Texas Southwestern, 6363 Forest Park Rd, Suite 13.354, Dallas, TX 75390-9119; Tel: 214-648-0188; Fax: 214-648-0167; E-mail: [email protected].

References

1. Ustun TB. The global burden of mental disorders. Am J Public Health. 1999;89(9):1315-1318.

2. Stewart WF, Ricci JA, Chee E, Hahn SR, Morganstein D. Cost of lost productive work time among US workers with depression. JAMA. 2003;289(23):3135-3144.

3. Kessler RC, Berglund P, Demler O, et al, and the National Comorbidity Survey Replication. The epidemiology of major depressive disorder: results from the National Comorbidity Survey Replication (NCS-R). JAMA. 2003;289(23):3095-3105.

4. Olfson M, Marcus SC, Druss B, Elinson L, Tanielian T, Pincus HA. National trends in the outpatient treatment of depression. JAMA. 2002;287(2):203-209.

5. Trivedi MH, Rush AJ, Wisniewski SR, et al, and the STAR*D Study Team. Evaluation of outcomes with citalopram for depression using measurement-based care in STAR*D: implications for clinical practice. Am J Psychiatry. 2006;163(1):28-40.

6. Fava M, Davidson KG. Definition and epidemiology of treatment-resistant depression. Psychiatr Clin North Am. 1996;19(2):179-200.

7. Rush AJ, Prien RF. From scientific knowledge to the clinical practice of psychopharmacology: can the gap be bridged? Psychopharmacol Bull. 1995;31(1):7-20.

8. Trivedi MH, Morris DW, Pan JY, Grannemann BD, Rush A. What moderator characteristics are associated with better prognosis for depression? Neuropsychiatr Dis Treat. 2005;1(1):51-57.

9. Trivedi MH, Rush AJ, Pan JY, Carmody TJ. Which depressed patients respond to nefazodone and when? J Clin Psychiatry. 2001;62(3):158-163.

10. Joyce PR, Mulder RT, Luty SE, et al. Patterns and predictors of remission, response and recovery in major depression treated with fluoxetine or nortriptyline. Aust N Z J Psychiatry. 2002;36(3):384-391.

11. Hirschfeld RM, Russell JM, Delgado PL, et al. Predictors of response to acute treatment of chronic and double depression with sertraline or imipramine. J Clin Psychiatry. 1998;59(12):669-675.

12. Morishita S, Arita S. Possible predictors of response to fluvoxamine for depression. Hum Psychopharmacol. 2003;18(3):197-200.

13. Katon W, Russo J, Frank E, et al. Predictors of nonresponse to treatment in primary care patients with dysthymia. Gen Hosp Psychiatry. 2002;24(1):20-27.

14. Joyce PR, Mulder RT, Cloninger CR. Temperament predicts clomipramine and desipramine response in major depression. J Affect Disord. 1994;30(1):35-46.

15. O’Leary D, Costello F, Gormley N, Webb M. Remission onset and relapse in depression. An 18-month prospective study of course for 100 first admission patients. J Affect Disord. 2000;57(1-3):159-171.

16. 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.

17. Bagby RM, Ryder AG, Cristi C. Psychosocial and clinical predictors of response to pharmacotherapy for depression. J Psychiatry Neurosci. 2002;27(4):250-257.

18. Nierenberg AA. Predictors of response to antidepressants general principles and clinical implications. Psychiatr Clin North Am. 2003;26(2):345-352.

19. Perlis RH, Alpert J, Nierenberg AA, et al. Clinical and sociodemographic predictors of response to augmentation, or dose increase among depressed outpatients resistant to fluoxetine 20 mg/day. Acta Psychiatr Scand. 2003;108(6):432-438.

20. Kocsis JH, Mason BJ, Frances AJ, Sweeney J, Mann JJ, Marin D. Prediction of response of chronic depression to imipramine. J Affect Disord. 1989;17(3):255-260.

21. Joyce PR, Mulder RT, Luty SE, McKenzie JM, Rae AM. A differential response to nortriptyline and fluoxetine in melancholic depression: the importance of age and gender. Acta Psychiatr Scand. 2003;108(1):20-23.

22. Spijker J, de Graaf R, Bijl RV, Beekman AT, Ormel J, Nolen WA. Determinants of persistence of major depressive episodes in the general population. Results from the Netherlands Mental Health Survey and Incidence Study (NEMESIS). J Affect Disord. 2004;81(3):231-240.

23. Moses T, Leuchter AF, Cook I, Abrams M. Does the clinical course of depression determine improvement in symptoms and quality of life? J Nerv Ment Dis. 2006;194(4):241-248.

24. Klein DN, Schatzberg AF, McCullough JP, et al. Age of onset in chronic major depression: relation to demographic and clinical variables, family history, and treatment response. J Affect Disord. 1999;55(2-3):149-157.

25. Diagnostic and Statistical Manual of Mental Disorders. 4th ed, text rev. Washington, DC: American Psychiatric Association; 2000.

26. Brown WA. Treatment response in melancholia. Acta Psychiatr Scand Suppl. 2007(433):125-129.

27. Peselow ED, Sanfilipo MP, Difiglia C, Fieve RR. Melancholic/endogenous depression and response to somatic treatment and placebo. Am J Psychiatry. 1992;149(10):1324-1334.

28. Carroll BJ, Curtis GC, Mendels J. Neuroendocrine regulation in depression. II. Discrimination of depressed from nondepressed patients. Arch Gen Psychiatry. 1976;33(9):1051-1058.

29. Ribeiro SC, Tandon R, Grunhaus L, Greden JF. The DST as a predictor of outcome in depression: a meta-analysis. Am J Psychiatry. 1993;150(11):1618-1629.

30. Parker GB. Atypical depression: a valid subtype? J Clin Psychiatry. 2007;68(suppl 3):18-22.

31. Quitkin FM, Harrison W, Stewart JW, et al. Response to phenelzine and imipramine in placebo nonresponders with atypical depression. A new application of the crossover design. Arch Gen Psychiatry. 1991;48(4):319-323.

32. Liebowitz MR, Quitkin FM, Stewart JW, et al. Antidepressant specificity in atypical depression. Arch Gen Psychiatry. 1988;45(2):129-137.

33. Quitkin FM, McGrath PJ, Stewart JW, et al. Atypical depression, panic attacks, and response to imipramine and phenelzine. A replication. Arch Gen Psychiatry. 1990;47(10):935-941.

34. McGrath PJ, Stewart JW, Janal MN, Petkova E, Quitkin FM, Klein DF. A placebo-controlled study of fluoxetine versus imipramine in the acute treatment of atypical depression. Am J Psychiatry. 2000;157(3):344-350.

35. Stewart JW, McGrath PJ, Quitkin FM. Do age of onset and course of illness predict different treatment outcome among DSM IV depressive disorders with atypical features? Neuropsychopharmacology. 2002;26(2):237-245.

36. Fava M, Alpert JE, Carmin CN, et al. Clinical correlates and symptom patterns of anxious depression among patients with major depressive disorder in STAR*D. Psychol Med. 2004;34(7):1299-1308.

37. Fava M, Rankin MA, Wright EC, et al. Anxiety disorders in major depression. Compr Psychiatry. 2000;41(2):97-102.

38. Fava M, Rush AJ, Alpert JE, et al. What clinical and symptom features and comorbid disorders characterize outpatients with anxious major depressive disorder: a replication and extension. Can J Psychiatry. 2006;51(13):823-835.

39. Joffe RT, Bagby RM, Levitt A. Anxious and nonanxious depression. Am J Psychiatry. 1993;150(8):1257-1258.

40. Rush AJ, Zimmerman M, Wisniewski SR, et al. Comorbid psychiatric disorders in depressed outpatients: demographic and clinical features. J Affect Disord. 2005;87(1):43-55.

41. VanValkenburg C, Akiskal HS, Puzantian V, Rosenthal T. Anxious depressions. Clinical, family history, and naturalistic outcome--comparisons with panic and major depressive disorders. J Affect Disord. 1984;6(1):67-82.

42. Cassano P, Soares CN, Cohen LS, Lyster AK, Fava M. Sex- and age-related differences in major depressive disorder with comorbid anxiety treated with fluoxetine. Arch Womens Ment Health. 2004;7(3):167-171.

43. Fava M, Uebelacker LA, Alpert JE, Nierenberg AA, Pava JA, Rosenbaum JF. Major depressive subtypes and treatment response. Biol Psychiatry. 1997;42(7):568-576.

44. Brown C, Schulberg HC, Madonia MJ, Shear MK, Houck PR. Treatment outcomes for primary care patients with major depression and lifetime anxiety disorders. Am J Psychiatry. 1996;153(10):1293-1300.

45. Russell JM, Koran LM, Rush J, et al. Effect of concurrent anxiety on response to sertraline and imipramine in patients with chronic depression. Depress Anxiety. 2001;13(1):18-27.

46. Perlis RH, Fraguas R, Fava M, et al. Prevalence and clinical correlates of irritability in major depressive disorder: a preliminary report from the Sequenced Treatment Alternatives to Relieve Depression study. J Clin Psychiatry. 2005;66(2):159-166.

47. Trivedi MH. The link between depression and physical symptoms. Prim Care Companion J Clin Psychiatry. 2004;6(suppl 1):12-16.

48. Trivedi MH, Clayton AH, Frank E. Treating depression complicated by comorbid medical illness or anxiety. J Clin Psychiatry. 2007;68(1):e01.

49. Iosifescu DV. Treating depression in the medically ill. Psychiatr Clin North Am. 2007;30(1):77-90.

50. Perlis RH, Iosifescu DV, Alpert J, Nierenberg AA, Rosenbaum JF, Fava M. Effect of medical comorbidity on response to fluoxetine augmentation or dose increase in outpatients with treatment-resistant depression. Psychosomatics. 2004;45(3):224-229.

51. Patience DA, McGuire RJ, Scott AI, Freeman CP. The Edinburgh Primary Care Depression Study: personality disorder and outcome. Br J Psychiatry. 1995;167(3):324-330.

52. Ezquiaga E, Garcia A, Bravo F, Pallares T. Factors associated with outcome in major depression: a 6-month prospective study. Soc Psychiatry Psychiatr Epidemiol. 1998;33(11):552-557.

53. Ezquiaga E, Garcia-Lopez A, de Dios C, Leiva A, Bravo M, Montejo J. Clinical and psychosocial factors associated with the outcome of unipolar major depression: a one year prospective study. J Affect Disord. 2004;79(1-3):63-70.

54. Joyce PR, Paykel ES. Predictors of drug response in depression. Arch Gen Psychiatry. 1989;46(1):89-99.

55. Katon W, Lin E, von Korff M, et al. The predictors of persistence of depression in primary care. J Affect Disord. 1994;31(2):81-90.

56. Papakostas GI, Iosifescu DV, Renshaw PF, et al. Brain MRI white matter hyperintensities and one-carbon cycle metabolism in non-geriatric outpatients with major depressive disorder (Part II). Psychiatry Res. 2005;140(3):301-307.

57. Kraft JB, Peters EJ, Slager SL, et al. Analysis of association between the serotonin transporter and antidepressant response in a large clinical sample. Biol Psychiatry. 2007;61(6):734-742.

58. Malhotra AK, Murphy GM Jr, Kennedy JL. Pharmacogenetics of psychotropic drug response. Am J Psychiatry. 2004;161(5):780-796.

59. Serretti A, Artioli P, Quartesan R. Pharmacogenetics in the treatment of depression: pharmacodynamic studies. Pharmacogenet Genomics. 2005;15(2):61-67.

60. Bares M, Brunovsky M, Kopecek M, et al. Changes in QEEG prefrontal cordance as a predictor of response to antidepressants in patients with treatment resistant depressive disorder: a pilot study. J Psychiatr Res. 2007;41(3-4):319-325.

61. Cook IA, Leuchter AF, Morgan M, et al. Early changes in prefrontal activity characterize clinical responders to antidepressants. Neuropsychopharmacology. 2002;27(1):120-131.

62. Cook IA, Leuchter AF, Witte E, et al. Neurophysiologic predictors of treatment response to fluoxetine in major depression. Psychiatry Res. 1999;85(3):263-273.

63. Leuchter AF, Cook IA, Witte EA, Morgan M, Abrams M. Changes in brain function of depressed subjects during treatment with placebo. Am J Psychiatry. 2002;159(1):122-129.

64. Perlis RH. Genetic predictors of antidepressant treatment response: progress towards clinical pharmacogenetics. Primary Psychiatry. 2007;14(6):53-58.

65. Rush AJ, Trivedi M, Fava M. Depression, IV: STAR*D treatment trial for depression. Am J Psychiatry. 2003;160(2):237.

66. Trivedi MH, Baker SM. Clinical significance of monitoring early symptom change to predict outcome. J Clin Psychiatry. 2001;62(suppl 4):27-33.

67. Nierenberg AA, McLean NE, Alpert JE, Worthington JJ, Rosenbaum JF, Fava M. Early nonresponse to fluoxetine as a predictor of poor 8-week outcome. Am J Psychiatry. 1995;152(10):1500-1503.

68. Hamilton M. Development of a rating scale for primary depressive illness. Br J Soc Clin Psychol. 1967;6(4):278-296.

69. Nierenberg AA, Farabaugh AH, Alpert JE, et al. Timing of onset of antidepressant response with fluoxetine treatment. Am J Psychiatry. 2000;157(9):1423-1428.

70. Trivedi MH, Morris DW, Grannemann BD, Mahadi S. Symptom clusters as predictors of late response to antidepressant treatment. J Clin Psychiatry. 2005;66(8):1064-1070.

71. Fava M, Rush AJ, Wisniewski SR, et al. A comparison of mirtazapine and nortriptyline following two consecutive failed medication treatments for depressed outpatients: a STAR*D report. Am J Psychiatry. 2006;163(7):1161-1172.

72. McGrath PJ, Stewart JW, Fava M, et al. Tranylcypromine versus venlafaxine plus mirtazapine following three failed antidepressant medication trials for depression: a STAR*D report. Am J Psychiatry. 2006;163(9):1531-1541.

73. Nierenberg AA, Fava M, Trivedi MH, et al. A comparison of lithium and T(3) augmentation following two failed medication treatments for depression: a STAR*D report. Am J Psychiatry. 2006;163(9):1519-1530.

74. Rush AJ, Trivedi MH, Wisniewski SR, et al. Acute and longer-term outcomes in depressed outpatients requiring one or several treatment steps: a STAR*D report. Am J Psychiatry. 2006;163(11):1905-1917.

75. Rush AJ, Trivedi MH, Wisniewski SR, et al. Bupropion-SR, sertraline, or venlafaxine-XR after failure of SSRIs for depression. N Engl J Med. 2006;354(12):1231-1242.

76. Trivedi MH, Fava M, Wisniewski SR, et al, and the STAR*D Study Team. Medication augmentation after the failure of SSRIs for depression. N Engl J Med. 2006;354(12):1243-1252.

77. Trivedi MH, Daly EJ. Measurement-based care for refractory depression: A clinical decision support model for clinical research and practice. Drug Alcohol Depend. 2007;88(suppl 2):S61-S71.

78. Trivedi MH, Rush AJ, Gaynes BN, et al. Maximizing the adequacy of medication treatment in controlled trials and clinical practice: STAR(*)D measurement-based care. Neuropsychopharmacology. 2007. In press.

79. Kraemer HC, Stice E, Kazdin A, Offord D, Kupfer D. How do risk factors work together? Mediators, moderators, and independent, overlapping, and proxy risk factors. Am J Psychiatry. 2001;158(6):848-856.