Objective: The primary objective of this study was to assess the rate of bipolar disorder (BPD) risk among patients unsuccessfully treated for major depression and to identify predictors of bipolarity.
Patients, Design, and Setting: Psychiatrists from community and private practice settings sequentially selected patients with unipolar depression who exhibited nonresponse to at least 1 antidepressant (AD) trial.
Main Outcome Measures: Patients self-reported their demographics, family history, comorbid health status, and legal problems. Current depression symptoms were assessed via the Centers for Epidemiologic Studies - Depression (CES-D) scale. Bipolar screening was performed using the Mood Disorder Questionnaire (MDQ). A psychiatrist recorded patient history and current and prior AD medication use.
Results: Of 602 patients enrolled, 18.6% screened positive on the MDQ. This rate was not affected by the number of prior AD failures or patient demographics. A prior history of BPD was reported by 12.3% of patients of which the psychiatrist was not aware or did not report, and 41.2% of these patients were MDQ+. Stepwise logistic regression identified 5 variables associated with bipolar disorder risk: the CES-D item "people were unfriendly," comorbid anxiety, initial depression diagnosis within 5 years, family history of BPD, and legal problems. In the subset of patients with complete data for the 5 variables (n = 483), 41.3% of the patients endorsing any 3 or more risk factors (n = 109) were MDQ+.
Conclusions: Almost 20% of patients with AD nonresponsive unipolar depression screened positive for BPD and the number of past medication trials had no effect on bipolarity. This suggests that clinicians should carefully screen for BPD in patients who have failed at least 1 antidepressant. Comorbid anxiety, feelings of people being unfriendly, recent depression diagnosis, BPD family history, and legal problems may prove useful indicators of BPD risk among patients who have failed at least 1 antidepressant. Future studies are needed to confirm that these risk factors are useful clinical variables to screen for bipolarity.


The past several decades have seen progress in the recognition, diagnosis, and therapeutic management of major depression, especially with the introduction of new classes of antidepressants and novel approaches to management. However, years of clinical experience show that many patients do not respond to antidepressant therapy as well as could be anticipated. At least 30% of patients with depression fail to respond fully to adequate antidepressant therapy,[1] and remission rates in controlled clinical trials remain below 50%.[2,3] A meta-analysis of 36 clinical trials of antidepressants in unipolar depression showed that between 29% and 46% of patients failed to respond fully to antidepressant treatment of adequate dose and duration. Partial response was achieved in 12% to 15% of the depressed patients studied, and nonresponse was observed in 19% to 34%.[4] When depressed patients were followed longitudinally over a period of 8-11 years,[5] two thirds of the patients who responded to their initial treatment subsequently suffered a recurrence. The long-term outcome of depression shows high recurrence rates and does not appear to have changed in the last 20 years.[6] Although most patients with chronic depression improve substantially with vigorous somatic therapy, few reach premorbid levels.[5]

A number of factors are associated with treatment failure with antidepressants, among them nonadherence to treatment and undertreatment (lack of adequate dose or duration of treatment), as well as comorbid medical illnesses and comorbid substance abuse or dependency.[7] More recent research suggests that a subset of patients diagnosed with unipolar major depression in fact suffer from bipolar disorder.[8] Studies show that in as many as one third of patients with bipolar disorder required at least 10 years from illness onset to correct bipolar disorder diagnosis.[9-11] The most common incorrect diagnosis (for 60% of respondents) is unipolar depression.[11] Misdiagnosed patients typically experience depression as the first symptom of their illness.[9] Most patients with bipolar disorder seek treatment for depression, and not for mania or hypomania,[12] adding to the frequency of misdiagnosis in both psychiatric and primary care settings.

Recent studies have shown that between 21% and 26% of unipolar depressed patients in primary care settings had evidence of bipolarity after careful screening.[13,14] For example, of 649 patients taking antidepressants for depression diagnosed by a primary care practitioner, 21% screened positive for bipolar disorder on the MDQ. Approximately half of these people had never been diagnosed with bipolar disorder.[13] This lack of correct diagnosis of bipolar disorder in the primary care setting has been underscored in a recent study [15] in which nearly 90% of the patients who screened positive for bipolar disorder had been previously diagnosed with a mental disorder by a healthcare professional, but only 8% of these reported a previous diagnosis of bipolar disorder. These patients were most likely to receive a diagnosis of depression (80% of the cases). Overall in this study, of 1146 participants who presented to the primary care practice for a variety of complaints, 10% had a positive screen for a lifetime history of bipolar disorder.[15]

The prevalence of bipolar disorder in psychiatric settings is substantially higher than in community and primary care settings. In Europe the prevalence of bipolar disorder in psychiatric settings is reported to be between 40% and 50% among patients in treatment for major depression.[16,17] When the broad definition of bipolar spectrum disorder (BSD) [18.19] is used, the prevalence of bipolar disorders in major depressive populations may be as high as 60%. For example, in a study conducted in Poland, the rate of bipolar I, II, and BSD was reported to be 61% in psychiatric outpatients who had had at least 1 major depressive episode.[20] In their study, bipolar I and II disorders were diagnosed on the basis of the DSM-IV criteria, and BSD was diagnosed using the MDQ and the criteria used by Ghaemi and colleagues,[21] who proposed a heuristic definition of "bipolar spectrum disorder" that included all forms of bipolar illness and went beyond the type I or II definition. It is also hypothesized that a large number of treatment-refractory major depressive patients (up to 60%), who show poor response to antidepressant therapy in spite of continued use of antidepressants, may have a bipolar diathesis or belong to a "soft" bipolar spectrum group.[21.22]

Our study was designed to measure the relationship between BPD risk and antidepressant failures among patients with unipolar depression who are not responding to antidepressant medication. The study hypothesized that a proportion of patients currently receiving antidepressant treatment actually suffer from BPD, rather than unipolar depression, and are thus incorrectly diagnosed and treated. A secondary objective was to identify variables possibly related to bipolar disorder risk among patients currently being treated with antidepressants for major depression and to assess the patterns of antidepressant use.


Participants and Site Recruitment

Selection of Psychiatrists. An invitation letter to participate in "an observational study of patients with major depression who have not responded to antidepressant medication" was sent to a national sample of approximately 4500 general psychiatrists selected at random from the American Medical Association listing of US psychiatrists. Responding sites were then screened for eligibility. To qualify for participation, study psychiatrists had to be board certified/eligible in psychiatry, and treating a minimum of 20 patients with major depression per month. With the exception of 2 research sites identified from a clinical trials database, all participating study practices have been involved in no more than 1 clinical study in the past 2 years. All sites possessed sufficient resources to identify, recruit, and assist patients throughout the study, and to complete the medical record abstractions. Prior to study initiation, all study psychiatrists and appropriate staff members participated in a telephone briefing regarding study protocol.

Selection of Subjects. Patients were eligible for this study if they were 18 years of age and older and if (1) they had received treatment for major depression for at least 3 months and had 1 or more medication changes during their current episode or (2) if treated less than 3 months, they had changed medications at least 3 times. Patients with a prior diagnosis of bipolar disorder, obsessive-compulsive disorder, schizophrenia, or schizoaffective disorder were excluded. Study psychiatrists were asked to sequentially take their next 10 eligible patients in order of presentation to their facility, regardless of their other characteristics.

Instruments. The study included 2 components: a self-report patient survey and a medical records abstraction form, completed by the psychiatrists or their designated office staff members.

Patient Survey. The patient survey included the Patient Information and Consent Form and the following inventories: demographic and family history form (including general demographic items like sex, race, education, marital status, the amount of schooling received, and annual income); the CES-D [23]; a healthcare resource use form; the MDQ[8,24-25]; and a comorbid health problems form. Legal problems were assessed with the legal status section of the Addiction Severity Index (ASI).[26,27]

The primary measure in the study was the MDQ.[8,24-25] The MDQ is a screening instrument for bipolar I and II disorders that contains 13 yes-or-no questions asking about mood and behaviors that are typically associated with mania, and 2 additional questions asking about the co-occurrence of symptoms and about the severity of functional impairment caused by the symptoms. A positive MDQ screen is defined as endorsement of at least 7 of the 13 yes or no questions, co-occurrence of 2 or more symptoms, and moderate to severe functional impairment. The MDQ has been validated in both a psychiatric setting and the general population. The MDQ has a sensitivity of 0.28 and a specificity of 0.97 in the general population,[25] and a sensitivity of 0.73 and a specificity of 0.90 in an outpatient psychiatric sample.[24]

An alternative MDQ scoring was also used that does not require endorsement of the co-occurrence or impairment items of the MDQ. A positive screen is simply defined as endorsement of any 7 of the total of 15 items. This scoring approach lowers the symptom threshold (potentially to 5) and increases the sensitivity of the instrument, but also lowers its specificity. In the clinical sample analyzed earlier by Hirschfeld and coworkers[24] this approach increased the sensitivity of the MDQ to 87%, but lowered the specificity to 61% (re-analysis of clinic validation study data published by Hirschfeld and associates, 2000).[24] In a general population sample this approach yielded a sensitivity of 57% and specificity of 82% (re-analysis of the general population sample data published by Hirschfeld and colleagues, 2003).[25]

The CES-D is a 20-item self-administered instrument devised for measuring depressive symptomatology. Respondents are asked to indicate the frequency of symptoms on a 4-point scale from "rarely or none of the time (less than 1 day)" to "most of the time (5-7 days)" that they experienced during the past week. A score of 16 is the recommended threshold level indicating the likely presence of clinical depression.[23] The CES-D was used in the present study to assess current depression symptoms.

The patient survey also asked about the patterns of depression, including the age of depression onset, age of first medical consultation for depression, and the pattern of visits to medical professionals. A comorbid health problems checklist asked participants to indicate if they or one or more members of their immediate family were ever diagnosed with bipolar disorder.

Medical Records Abstraction

The medical records abstraction form was completed by the study psychiatrists or their support staff and included the following information drawn from the medical records: patient and family health history, lifetime history of major depression, current episode of major depression, prescription drug treatment history, number of prior antidepressant medication failures, healthcare resource use, and outcomes of treatment (including no response, partial response, and remission/sustained remission).


Following instrument pretesting and IRB approval, the study was conducted between April and August 2005. Designated staff at each site identified qualified patients and administered the survey to patient volunteers before or after a regularly scheduled office visit. Patients were asked to complete the patient survey. Designated staff at each site abstracted specified information from each participant's medical records and returned this form with the patient survey. Patients were not identified in any way on either the Medical Record Abstraction form or the Patient Survey

Statistical Analysis

The data from both study components were analyzed using SPSS version 12.0. The study hypotheses were evaluated using cross-tabulation analysis, Chi-square, and ANOVA, where appropriate. The overall MDQ positive rate was calculated using the original scoring algorithm as well as the modified algorithm. The relationships specified in the study hypotheses were considered statistically significant if the results of statistical testing were significant at the P < .05 level. Stepwise logistic regression was used to identify variables that optimized the detection of MDQ+ versus MDQ- patients


Sixty-three sites participated in the study, each recruiting 1-20 patients with an average of 9.6 patients per site. In total, 602 patients were recruited. Most patients were recruited by psychiatrists in private practice (n = 409, 68.7%), followed by community mental health clinics (n = 116, 19.2%), hospitals (n = 51, 8.5%), and research centers (n = 22, 3.6%). The mean age of participants was 47.9 (SD = 12.6 years). The majority of subjects enrolled in this study were married, white females with some college education ( Table 1 ).

Patterns of Current and Lifetime Clinical Depression and Antidepressant Use

Of the 602 participants enrolled in the study, 601 completed the CES-D (1 CES-D questionnaire could not be used because of missing values). CES-D based severity measures indicated that 61% were severely depressed, 11.4% were mildly to moderately depressed, and 17.4% were not depressed. Study participants reported a mean age of onset of clinical depression of 27.5 years, a mean age at first physician consultation of 33.2 years, and a mean duration of depressive symptoms of 20.4 years. The participants also reported a median of 5 lifetime depressive episodes and a median of 4 lifetime antidepressants taken. Among those reporting that they are currently depressed, the mean duration of the current episode was reported to be 2.9 years. Only 5.8% of participants reported prior treatment with electroconvulsive therapy.

The study psychiatrists reported on the participant's antidepressant medication use in the current episode of depression. Two (39.4% of participants) or 3 (31.4%) antidepressant medication failures were reported during the current episode ( Table 2 ). On average, patients failed a mean of 2.9 antidepressant medications in their current episode of depression. The number of antidepressant regimens, or combinations of antidepressant medications, was also determined for the current episode of depression. Most patients (54.9%) were treated with a regimen of 2 antidepressants in their current episode, and 25.3% of patients were on a regimen of 3 antidepressants. The mean number of unique antidepressant regimens was 2.6 for the current episode of depression ( Table 2 ).

Patterns of Bipolar Disorder

Of the 602 participants who completed the MDQ, 112 (18.6%) screened positive for BPD. Increasing the number of prior antidepressant failures above 1 (all patients were required to have at least 1 antidepressant failure) did not affect the rate of positive MDQ screens. The rate of MDQ positive screens was not affected by patient demographics ( Table 1 ). The mean age of onset of bipolar symptoms was 28.6 years, and the mean duration of bipolar symptoms was 18.1 years. The rate of positive MDQ screens was similar among female (17.8%) and male (17.9%) participants. Younger participants were somewhat more likely to screen positive for BPD on the MDQ than older participants ( Table 1 ). However, these differences were not statistically significant (on the basis of a Chi-square analysis). There were 74 patients (12.3%) who reported a prior history of BPD where the psychiatrist was not aware of this diagnosis or did not report it. The positive MDQ rate in this subgroup was 41.9%.

When the rate of positive bipolar screens was analyzed using the alternative scoring approach (endorsement of any 7 or more of the total 15 items), the overall MDQ positive rate increased to 36.15% (N = 217). However, a relationship with antidepressant failure was still not seen.

Predictors of Bipolar Disorder Risk

Potential predictors of MDQ status were first assessed via univariate analyses, which yielded the following statistically significant predictors: comorbid anxiety (including panic attacks), family history of BPD, family history of mania or manic depression, family history of allergies, legal problems, recent depression diagnosis (within 5 years), and the following 5 CES-D items, treated as dichotomous variables: "I thought my life had been a failure," "My sleep was restless," "People were unfriendly," "I had crying spells," and "I felt that people disliked me." These variables were entered into a stepwise logistic regression that identified 5 variables associated with BPD risk (MDQ+): the CES-D item "people were unfriendly" (OR = 2.59, P < .001), comorbid anxiety (OR = 2.99, P < .002), recent depression diagnosis (within 5 years) (OR = 2.48, P <. 001), family history of BPD (OR = 2.02, P < .010), and legal problems (OR = 1.74, P < .026) ( Table 3 ).

The 5 risk factors were scored as present or absent and summed for each patient. Patients with more risk factors had a significantly higher probability of being MDQ+. For instance, of those patients with 2 or more risk factors, 24.8% were MDQ+ compared with only 8.4% with 0 or 1 risk factors (chi = 20.9, P < .001). Of the patients with no risk factors (n = 41), 2.4% were MDQ+. The probability of being MDQ+ increased to 10% in patients with only 1 risk factor, to 15% in patients with only 2 risk factors, and to 43% with only 3 risk factors (Figure 1). The percentage of patients with only 4 risk factors (n = 25) or all 5 risk factors (n = 3) who were MDQ+ was 25% and 100%, respectively. Of the patients with any 3 or more risk factors (n = 109), 41.3% were MDQ+.

Figure 1.

Association between the number of risk factors (in any combination) and being positive on the MDQ (MDQ+).

Among patients endorsing "people were unfriendly" (n = 103), 31.1% were MDQ+. Among patients endorsing "people were unfriendly" and comorbid anxiety (n = 82), MDQ+ rate increased to 35.4%. Adding recent depression onset (n = 17) increased the MDQ+ rate to 41.2%, and adding family history (n = 4) increased the MDQ+ rate to 75%. One-hundred percent of those endorsing all 5 factors (n = 3) were MDQ+ (Figure 2).

Figure 2.

Percentage of MDQ-positive patients (MDQ+) endorsing the following predictor variables: people were unfriendly (single predictor variable), plus comorbid anxiety (2 predictor variables), recent depression (3 predictor variables), family history of bipolar disorder (4 predictor variables), and legal problems (5 predictor variables).


In the present study, we first analyzed the pattern of antidepressant use and behavior of the patients currently treated with antidepressants for major depression by a psychiatrist. We found that more than 60% of the participants in the present study were severely depressed (on the basis of CES-D findings), and that they were taking an average of 3 antidepressant medications for their current episode of depression. Of these patients, nearly 20% had a positive screen for bipolar disorder. The rate of positive MDQ screens was not affected by either patient demographics or the number of prior antidepressant failures. In the 12% of patients with a history of bipolar disorder of which the psychiatrist was not previously aware or did not report, the rate of positive MDQ screens jumped to 42%.

In our study nearly 1 in 5 patients treated with antidepressants for unipolar depression by a psychiatrist screened positive for bipolar disorder on the MDQ, and the majority of these patients had never been diagnosed with bipolar disorder. Therefore, the likelihood of receiving mood stabilizers and other appropriate medications for bipolar disorder was low. Although the MDQ is not a diagnostic tool, its sensitivity (73%) and specificity (90%) in a psychiatric-based setting suggest that a patient with a positive screen has a reasonable probability of having bipolar disorder.[24]

In our sample 18.6% of depressed patients screened positive on the MDQ. This rate is significantly lower than the reported prevalence of positive MDQ screens and ultimate bipolar disorder in psychiatric samples (between 40% and 60%), and more comparable to the prevalence of bipolar disorder in primary care settings. Part of the explanation may lie in the fact that most recent studies of psychiatric patients were aimed at detecting "soft" bipolar disorder, and the MDQ used in the present analysis is best at screening for bipolar I disorder, and not as sensitive to bipolar II and not otherwise specified (NOS) disorder.[28] The sensitivity of the MDQ is 70% in bipolar I disorder, but only 30% in bipolar II/NOS disorder.[28]

An alternative scoring algorithm was used in the present study in an attempt to better estimate the risk of bipolar II disorder. This approach did not require endorsement of the co-occurrence or impairment items of the MDQ, and a positive screen was defined as endorsement of any 7 or more of the total of 15 items. This approach was similar to that used previously by others in an attempt at better detecting bipolar II cases.[20,29] This scoring approach lowers the symptom threshold (potentially to 5) and increases the sensitivity (87%) of the instrument but also lowers its specificity (61%, on the basis of re-analysis of clinic validation study data published by Hirschfeld and coworkers). Using this approach yielded an overall MDQ positive rate of 36.1% (vs 18.6% positive MDQ rate using the standard scoring criterion).

A secondary goal of the present study was to identify predictors that would alert physicians to the possibility of bipolar depression in their patients currently being treated for major depression. Our study identified 5 significant discriminators of bipolar disorder risk (MDQ+). Over one third of patients who endorsed the item "people were unfriendly" on the CES-D scale and reported comorbid anxiety screened positive for bipolar disorder. These 2 clinical features plus recent depression onset, family history of bipolar disorder, and legal problems may prove to be useful indicators of bipolar disorder risk among patients currently treated for major depression. In our study, 100% of the patients (n = 3) who endorsed all 5 predictor variables had positive MDQ screens for bipolar disorder. These findings are consistent with previous research. Family history of BPD may be one the most useful predictors of the bipolar disorder diagnosis.[30,31] The finding that comorbid anxiety predicts BPD risk is also consistent with recent findings suggesting that more than 50% of patients with bipolar disorder experience at least 1 comorbid anxiety disorder.[32,33] The association between bipolar disorder and legal problems also has been established previously.[34] More subjects who screen positive for bipolar disorder are jailed, arrested, or convicted of crimes other than drunk-driving than those who screen negative for bipolar disorder.[34]

On the basis of the results of this study, these questions may prove beneficial in identifying increased bipolar disorder risk among patients who remain depressed after taking at least 1 antidepressant: (1) Have you ever been told by your healthcare provider that you have anxiety? (2) Do you sometimes feel that people are acting unfriendly toward you? (3) Have you been diagnosed with depression within the past 5 years? (4) Has anyone in your family ever been diagnosed with bipolar disorder? (5) Have you ever had any legal problems? Further research is needed to test the validity and reliability of the questions as well as the usefulness and acceptance of the questions by patients and their health care providers. Although these questions may be helpful in screening for bipolarity they should not be a substitute for a thorough clinical evaluation.

There are several limitations to this study. The MDQ was used as the sole screening instrument for bipolar disorder and the positive MDQ cases were not validated by the SCID diagnostic interview. As has been previously mentioned, although the MDQ is the most effective screening instrument currently available, it is less sensitive in detecting bipolar II vs bipolar I disorder, and there may be other screens better suited for detecting bipolar spectrum disorders, including hypomania and cyclothymia,[35,36] such as the SCID interview used in other studies. However, using the SCID diagnostic interview was not feasible in the present study, and we know from previous research that the MDQ is a sensitive instrument for screening for bipolar disorder risk. In future studies, it may be useful to conduct interviews with the psychiatrists who did not make the diagnosis of bipolar disorder in patients screening positive on the MDQ to clarify what factors lead clinicians to diagnose unipolar depression in patients who may suffer from bipolar depression. Seventy-four patients (12.3%) reported receiving a prior diagnosis of bipolar disorder even though their current physician diagnosed them with major depressive disorder. It is unknown whether a prior diagnosis of bipolar disorder leads patients to answer the MDQ questions differently from if they had not received a bipolar disorder diagnosis. However, these patients were included in the current study because they were currently being treated for unipolar depression.

Nevertheless, this study suggests that a proportion of patients being treated with antidepressants for major depression in general psychiatry settings may be at risk for unrecognized bipolar disorder. The consequences of incorrectly diagnosing bipolar depression in patients with unipolar depression and treating them with antidepressants can be profound. Antidepressant treatment in bipolar disorder may exacerbate the problem by inducing mania or cycle acceleration, and the long-term benefits of antidepressants in bipolar depression have not been adequately established.[37-39] Practice guidelines, including those of the American Psychiatric Association, recommend that antidepressants be used with caution in bipolar depression, seldom without the use of a concomitant antimanic agent, and rarely for long periods of time.[40] Possible predictors of bipolar risk in patients currently being treated for unipolar depression who have failed at least 1 antidepressant have been outlined, and clinicians are advised to actively screen for signs of bipolarity in their depressed patients.

Table 1. Patient Demographic Characteristics Stratified by MDQ Status

Study Patients





















Age (Mean = 47.9, Median = 48.0)





















Ethnic Background

African American










Asian/Pacific Islander





Native American










Mixed / Multicultural





Of Spanish or Hispanic heritage





Marital Status