Predominant recurrence polarity among 928 adult international bipolar I disorder patients


  • R. J. Baldessarini,

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Department of Psychiatry, Harvard Medical School, Boston, MA, USA
    Search for more papers by this author
  • J. Undurraga,

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Department of Psychiatry, Harvard Medical School, Boston, MA, USA
    3. Bipolar Disorders Program, Hospital Clinic, University of Barcelona, IDIBAPS, CIBERSAM, Barcelona, Spain
    Search for more papers by this author
  • G. H. Vázquez,

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Department of Clinical Neuroscience, Palermo University, Buenos Aires, Argentina
    Search for more papers by this author
  • L. Tondo,

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Department of Psychiatry, Harvard Medical School, Boston, MA, USA
    3. Lucio Bini Mood Disorder Center, Cagliari, Sardinia
    Search for more papers by this author
  • P. Salvatore,

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Department of Psychiatry, Harvard Medical School, Boston, MA, USA
    3. Section of Psychiatry, Department of Neuroscience, University of Parma, Parma, Italy
    Search for more papers by this author
  • K. Ha,

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Department of Psychiatry, Harvard Medical School, Boston, MA, USA
    3. Department of Psychiatry, Seoul National University Bundang Hospital, Bundang, Gyeonggi, Korea
    Search for more papers by this author
  • H.-M. K. Khalsa,

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Department of Psychiatry, Harvard Medical School, Boston, MA, USA
    Search for more papers by this author
  • B. Lepri,

    1. Lucio Bini Mood Disorder Center, Cagliari, Sardinia
    Search for more papers by this author
  • T. H. Ha,

    1. Department of Psychiatry, Seoul National University Bundang Hospital, Bundang, Gyeonggi, Korea
    Search for more papers by this author
  • J. S. Chang,

    1. Department of Psychiatry, Seoul National University Bundang Hospital, Bundang, Gyeonggi, Korea
    Search for more papers by this author
  • M. Tohen,

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Department of Psychiatry, Harvard Medical School, Boston, MA, USA
    3. Department of Psychiatry, University of Texas Health Science Center, San Antonio, TX, USA
    Search for more papers by this author
  • E. Vieta

    1. International Consortium for Bipolar Disorder Research, McLean Division of Massachusetts General Hospital, Belmont, MA
    2. Bipolar Disorders Program, Hospital Clinic, University of Barcelona, IDIBAPS, CIBERSAM, Barcelona, Spain
    Search for more papers by this author

Ross J. Baldessarini, Mailman Research Center 312, McLean Hospital, 115 Mill Street, Belmont, MA 02478-9106 USA.


Baldessarini RJ, Undurraga J, Vázquez GH, Tondo L, Salvatore P, Ha K, Khalsa H-MK, Lepri B, Ha TH, Chang JS, Tohen M, Vieta E. Predominant recurrence polarity among 928 adult international bipolar I disorder patients.

Objective:  To test the hypothesis that patients with bipolar disorder (BPD) differ demographically and clinically within subgroups based on the predominant-polarity of major recurrences.

Method:  We tested factors for association with predominantly (≥2 : 1) depressive vs. mania-like episodes with 928 DSM-IV type-I BPD subjects from five international sites.

Results:  Factors preliminarily associated with predominant-depression included: electroconvulsive treatment, longer latency-to-BPD diagnosis, first episode depressive or mixed, more suicide attempts, more Axis-II comorbidity, ever having mixed-states, ever married, and female sex. Predominant-mania was associated with: initial manic or psychotic episodes, more drug abuse, more education, and more family psychiatric history. Of the 47.3% of subjects without polarity-predominance, risks for all factors considered were intermediate. Expanding the definition of polarity-predominance to ≥51% added little, but shifting mixed-states to ‘predominant-depression’ increased risk of suicidal acts from 2.4- to 4.5-fold excess over predominant-mania–hypomania, and suicidal risk was associated continuously with increasing proportions of depressive or mixed episodes.

Conclusion:  Subtyping by predominant-polarity yielded predictive associations, including the polarity of first episodes and risk of suicide attempts. Such subtyping may contribute to improve planning of clinical care and to biological studies of BPD.

Significant outcomes

  •  The findings support subtyping by predominant-polarity in bipolar disorders, indicate that long-term morbidity is predicted by initial presentations, and suggest including mixed-states with predominant-depression.
  •  Predominant-depression was associated, notably, with depressive or mixed onset, more mixed-states, and higher suicidal risk; predominant-mania was associated with initial mania or psychosis and more family history.
  •  Including mixed-states with predominant depressions markedly increased association with suicidal risk.


  •  Many patients with bipolar I disorder (47%) could not be characterized by predominant-polarity.
  •  Rates of predominant-polarity varied among sites, supporting pooling of data across multiple international sites to enhance representativeness of findings.
  •  Some factors associated with predominant-depression or mania were expected but support the face validity of subtyping by predominant polarity.


The concept of predominant polarity in bipolar disorders (BPD) was proposed by Colom et al. (1) to define discrete subgroups of patients who mainly experience recurrences of depression or of mania-like episodes (manic, hypomanic, or mixed). This categorization, based on at least a two-fold excess of one polarity, has been applied to test for associations with various clinical characteristics in several published reports appearing since 2006 (1–5). Approximately half of patients with BPD could be characterized: of these, 54.1% had mainly depressions, and 45.9% experienced predominantly manic–hypomanic or mixed recurrent major episodes of BPD. However, these proportions are influenced by the inclusion of mainly depressed, type-II BPD patients at 48.9% (1) and 76.6% (2) in earlier reports on this topic. Initially, we summarized characteristics of the 847 patients with predominant-mania-like or depressive recurrences in these reports to generate hypotheses for this study (Table 1). Six factors showed ≥25% higher prevalence among subjects with predominant-depressions: (i) ever psychotic, (ii) depressive first lifetime episodes, (iii) suicidal acts, (iv) ever rapid cycling (≥4 recurrences/12 months), (v) ever married, and (vi) female sex. Only substance abuse was at least 25% more prevalent among ‘predominantly manic’ patients (Table 1).

Table 1.   Summary of literature reports of prevalence rates vs. predominant polarity in patients with bipolar disorder
FactorsStudiesPredominant polarity (95%CI)Relative risk
  1. Relative risk is for predominant depression vs. mania–hypomania-mixed states (excluding unclassifiable patients). Data are averages weighted by the numbers of subjects per study arm with their 95% confidence intervals (CI, when there were at least three reports) and are ranked by relative risk. Total: 847 subjects.

Ever psychotic (%)(4)
Depressive first episode (%)(1, 3)80.335.32.27
Suicidal acts (%)(1–3, 5)31.8 (9.48–54.0)15.6 (3.55–27.6)2.04
Ever rapid-cycling (%)(1, 2, 4)47.0 (15.4–34.0)23.7 (–2.14 to 49.5)1.98
Married (%)(5)60.834.61.76
Women (%)(1–5)62.2 (40.5–86.9)47.3 (31.4–63.2)1.32
Any substance abuse (%)(1, 2, 5)42.1 (30.3–53.9)52.8 (45.0–60.6)1/1.25
Family history (%)(1–3, 5)71.8 (63.2–80.4)58.7 (35.6–81.8)1.22
Employed (%)(1, 2)59.370.21/1.19
Ever hospitalized (%)(5)20.717.71.17
Current age(1–5)43.4 (40.9–45.9)40.2 (34.3–46.1)1.08
Years of illness(1–5)15.4 (11.0–19.8)14.7 (10.5–18.1)1.05
Onset age(1–5)26.5 (23.1–29.9)25.4 (22.4–28.4)1.04
Psychiatric comorbidity (%)(1, 3)49.748.51.02
Episodes/year(1–3, 5)0.811 (0.070–1.55)0.804 (0.347–1.26)1.01

These findings guided analyses in the present international collaborative study aimed at testing replicability of the published findings and their possible extension, and that secondarily considered alternative definitions of polarity-predominance. This study involved 928 patients with DSM-IV type-I BPD providing data relevant to predominant-polarity, from five collaborating mood disorder research sites in Argentina, Spain, South Korea, Italy, and the United States.

Aims of the study

To compare and extend reported associations with predominant-polarity with clinical factors in a large international sample of patients with bipolar I disorder.

Furthermore, to evaluate effects of broader definitions of predominant-polarity, based on broadened predominance criteria (from ≥67% to ≥51%).

Finally, to evaluate effects of including mixed-states with predominant-depressive rather than predominant-mania-like recurrences, including associations with initial episode-type and suicidal risk.

Material and methods

Patient–subjects were evaluated by DSM-IV diagnostic categorization of BPD diagnoses and polarities of major recurrences, based on clinical historical assessments, backed by Structured Clinical Interviews for DSM-IV Axes I and II [SCID (6)], as well as by regular, prospective clinical re-assessments and treatments at the collaborating mood disorder centers. Collaborating sites were located in: Barcelona, Catalonia (Spain); Belmont, Massachusetts (USA); Buenos Aires, (Argentina); Cagliari, Sardinia (Italy); and Bundang, Gyeonggi (Korea), with respective contributors as defined in the authorship above. All study participants were adults (aged ≥18 years), who provided written, informed consent for study participation and for aggregate and anonymous reporting of their clinical data, with the approval by local institutional ethical review committees. All subjects were treated individually by clinical requirements not guided by research protocols. None had participated in similar studies previously, except for 24% of Barcelona subjects.

We compared salient demographic and clinical characteristics, initially between patients with mainly depressive vs. mainly ‘mania-like’ (manic, hypomanic, or mixed) episodes as their predominant (≥2 : 1) polarity, with ≥66.7% of episodes/time-at-risk of either polarity (1). In secondary analyses, we also compared predominantly manic or depressive patients with those with <2 : 1 excess of a single polarity, including subgroups with only ≥51% excess of one polarity as well as those with intermediate levels of polarity excess (51.0–66.7%). In addition, when possible (Barcelona and Belmont subjects), mixed-states were shifted to ‘predominant-depression’ (depression-mixed) to compare with subjects with predominant-mania or hypomania without mixed-states.

We compared the prevalence of selected factors between predominant-polarity categories (as well as among subjects who could not be categorized), using contingency tables (χ2) for categorical measures and anova methods (F) for continuous variables. Factors that appeared to differ between polarity subgroups ( 0.05) were further evaluated with multivariate logistic regression modeling for independent association with a predominant-polarity. Suicidal risk was of particular interest and was considered by previously reported studies as well as at each collaborating study site. We utilized random effects meta-analysis to test for a predicted association of suicidal acts with predominant-depression or with depression plus mixed-states. We also considered suicidal risk as a function of percentage of recurrences in depression or mixed-states as a continuous measure. Data are reported as means ± standard deviations (SD), means weighted by subject number, medians with interquartile ranges (IQR), or pooled risk ratios with 95% confidence intervals (CI). Computations of predictive values were based on standard Bayesian methods (7). Statistical analyses employed commercial programs (Statview-5®; SAS Institute, Cary, NC, USA; or Stata-8®; StataCorp, College Station, TX, USA).


Patients with DSM-IV-TR type-I BPD (N = 928) from the five collaborating sites were classified by predominant-polarity using criteria initially proposed by Colom et al. (1), with excesses of depressive or mania-like (mania, hypomania, and mixed-states) DSM-IV recurrent major episodes of BPD illness as ≥ 66.7% of all recurrences per years at risk. Subjects/site ranked: Italy (n = 232; 24.7%), Korea (n = 218; 23.2%), Spain (n = 203; 21.6%), USA (n = 187; 19.9%), Argentina (n = 88; 9.4%). Age averaged 36.5 ± 14.1 years [median (IQR): 34.0 (20.8) years], with a slight excess (51.6%) of women. Subjects were ill for 13.2 ± 11.4 years [median (IQR): 9.71 (14.8); 15.2 ± 11.6 [median: 11.0 (16.9)] for predominantly depressed, 10.6 ± 9.6 [median: 7.27 (10.3)] years for predominantly manic cases]. There were major differences in morbidity patterns among the study sites, encouraging data pooling to enhance generalization. For example, years of illness from onset ranked: Argentina (19.7 ± 12.4), Spain (18.0 ± 10.3), Sardinia (14.4 ± 12.3), Korea (10.1 ± 9.2), and United States (4.8 ± 2.4 years; = 55.4, < 0.0001). Also, the ratio of proportions (%) of predominantly depressed/predominantly mania-like polarities ranked: Argentina (30.7/20.5), Sardinia (23.7/23.7), Korea (24.8/31.7), Spain (15.8/22.7), and United States (16.6/54.5; χ2 = 79.3, < 0.0001).

Polarity categorizations, overall, ranked: predominant-depression (n = 199; 21.4%) < predominant-mania, hypomania, or mixed-states (n = 290; 31.3%) < neither (n = 439; 47.3%); that is, 52.7% of subjects could be classified as having at least a two-fold excess of one major polarity. Age-at-onset of BPD did not differ among subgroups (for means: = 1.00, = 0.37): median (IQR) onset age was 22.0 (12.2) years overall, 22.2 (13.7) with predominant-depression, 22.0 (12.0) for unclassifiable cases, and 21.9 (10.2) years with predominant-mania. Duration of illness (years from onset) averaged 13.2 ± 11.4 [median: 11.4 (IQR: 14.8)] years, overall, ranking: predominant-depression [15.2 ± 11.6; 12.0 (16.2)] ≥neither [14.9 ± 11.9; 11.0 (16.9)] > predominant-mania-like [10.6 ± 9.6; 7.27 (10.3)] years; = 15.5, < 0.0001).

Categorization by predominant-polarity appeared to be quite stable over time in that the proportion of those considered predominantly depressed did not increase significantly over quintiles by years of illness, ranging from 6.7 to 36 years, including an initial 5 years for the average latency to diagnosis of BPD (r = +0.37, = 0.54). Moreover, the proportion of all subjects considered predominantly depressed was nearly identical at the shortest and longest quintiles of exposure time (22.7% vs. 23.3%).

Morbidity (total episodes/year) ranked: predominantly mania-like (0.543 ± 0.431) < predominantly depression (0.563 ± 0.365) < unclassifiable (0.915 ± 0.597). However, as expected, depressions/year ranked: predominant-mania (0.028 ± 0.088) < unclassifiable (0.408 ± 0.252) < predominant-depression (0.563 ± 0.365); mania-like episodes/year ranked: predominantly depressed (0.106 ± 0.111) < neither (0.360 ± 0.317) < predominantly manic (0.411 ± 0.687; all  10.2, all < 0.0001). Data from the Belmont site allowed estimates of the proportion of weeks ill during 2 years of prospective follow-up following first episodes. Predominant-depression was only weakly associated with more time ill than predominant-mania (24.8 ± 25.8% vs. 16.9 ± 21.5%; = 3.02, = 0.08).

Factors associated with predominant-polarity

We considered characteristics of patients with predominant-depression vs. predominant-mania-like recurrences in preliminary bivariate comparisons. Several contrasts were statistically significant, even with Bonferroni adjustment of significance criteria for multiple comparisons (Table 2). Of factors with previously reported, substantial (≥25%) differences in prevalence by predominant-polarity (Table 1), five of seven were sustained among study subjects (Table 2). With predominant-depression, these included excesses of: (i) depressive (or mixed) first episodes, (ii) suicidal acts, (iii) having married, and (iv) female sex (Table 2). Substance abuse was previously reported to be associated with predominant-mania-like episodes (Table 1) and replicated (Table 2). Minor differences were reported previously for onset age, current age, years of illness, and recurrences/year (Table 1). Four new factors also differed significantly by predominant-polarity; four with predominant-depression: electroconvulsive treatment (ECT), latency to BPD diagnosis, any mixed-state episodes, and one with predominant-mania-like episodes: more education (Table 2).

Table 2.   Comparison of clinical features associated with depressive vs. manic predominant recurrence polarity, and without predominance, among 928 patients with bipolar I disorder from five international sites
FactorsPredominant polarityRisk ratioF or χ2P-valueNo predominance (n = 439)Risk rank
Depression (D) (n = 199)Mania (M) (n = 290)
  1. Based on comparing cases with ‘predominant mania’ (mania + hypomania + mixed-states) or predominant depression (≥2 : 1). Of the 489 classifiable cases (52.7%), 290 (59.3)% were predominantly manic (31.3% of all cases), 199 (40.7%) predominantly depressed (21.4% of all cases), and 439 (47.3% of all patients) fit neither category. Factors are ranked by risk ratio within predominance categories; inverse risk ratios indicate association with predominant-mania > depression. Risk rank indicates the order of risk for each factor for predominant depression (D), mania-like states (M), or neither (N). Years to diagnosis: time from initial syndromal illness (onset) to clinical recognition and treatment as bipolar disorder; ECT = electroconvulsive treatment. For the 12 factors associated with a predominant polarity, Bonferroni correction for multiple comparisons would lead to a requirement of  0.004 for significance, as met by eight of the 12 factors (not for ECT, Axis-II comorbidity, sex, or drug abuse). Note that cases that could not be classified (N) were consistently intermediate in risk for all 12 factors between predominant depression (D) and mania (M). The average contrast in risks between D and M subgroups to D or M vs. N decreased from 2.48 ± 0.95- to 1.46 ± 0.47-fold, indicating much more moderate separation of the extremes of predominance polarity from the unclassifiable cases than from each other.

Predominant depression > mania
 Ever received ECT (%)34.48.703.958.010.00512.8D > N > M
 Years to diagnosis5.33 ± 7.611.58 ± 3.403.37148<0.00014.66 ± 7.27D > N > M
 First episode depressive or mixed (%)80.426.63.02148<0.000159.3D > N > M
 Suicide attempted (%)28.410.72.6518.0<0.000122.8D > N > M
 Axis-II comorbidity (%)31.313.02.413.840.0523.2D > N > M
 Ever in a mixed-state (%)65.829.92.2014.10.000251.6D > N > M
 Ever married (%)57.638.71.4912.50.000450.7D > N > M
 Women (%) > N > M
Predominant-mania > depression
 First episode manic or psychotic (%)19.375.11/3.89148<0.000140.7M > N > D
 Drug abuse (%)18.850.01/2.667.890.00530.4M > N > D
 Educated ≥12 years (%)43.967.21/1.5316.3<0.000150.7M > N > D
 Family history of affective illness (%)54.574.51/1.378.820.00371.4M > N > D

Of the total of 12 factors that differed significantly by predominant-polarity (Table 2), eight were associated with predominant-depression and ranked by prevalence difference (rate ratio): (i) ever given ECT (4.0-fold excess), (ii) longer latency from onset to diagnosis of BPD (3.4-fold), (iii) first lifetime major episode depressive or mixed (3.2), (iv) suicide attempted (2.6), (v) DSM-IV Axis-II comorbidity (2.4), (vi) ever being in a mixed-state (2.2), (vii) ever married (1.5) (all < 0.005), and (viii) female sex (1.2-fold; = 0.05; Table 2). Four factors were associated with predominant-mania-like recurrences: (ix) first lifetime episode manic, hypomanic, or psychotic (by 3.9-fold), (x) drug abuse (2.7-fold), (xi) more education (1.5), and (xii) more family history of affective illness (1.3) (all  0.005; Table 2). Among subjects who were unclassifiable by predominant-polarity, rates for all 12 factors considered in Table 2 were consistently intermediate between subjects with predominant-depression or mania.

Particularly, notable were strong associations of predominant-depression or mania with the polarity of initial episodes. Overall, first episode frequency ranked: mania or hypomania (45.1%) > depression (38.1%) > mixed-states (10.6%) > psychosis (3.69%) > other (2.50%, mainly anxiety states). In addition, initial and later mixed-states were strongly associated with predominant-depression vs. predominant-mania (Table 2).

Another 17 factors were not significantly associated with a predominant-mania or depression (all relative rates 1.02–1.84, = 0.08–0.85). In descending level of association (relative rates), they were the following: (i) ever rapid cycling (≥4 episodes in a year), (ii) episodes/year, (iii) comorbid DSM-IV anxiety disorder, (iv) having children, (v) ever psychotic, (vi) alcohol abuse history, (vii) years of total illness, (viii) lower functional status (based on a pooled assessment of employment, having married, and living independently), (ix) mood-switch (hypomanic, manic, or mixed) during treatment with an antidepressant or stimulant, (x) current age, (xi) any DSM-IV Axis-I comorbidity, (xii) onset age, (xiii) ever hospitalized, (xiv) clinically apparently responsive to antidepressant treatment for depression, (xv) living independently, (xvi) treated with a mood-stabilizer (lithium or an anticonvulsant), and (xvii) being employed.

In sum, these associations verified 5/7 factors previously associated with a predominant-polarity among patients with different forms of BPD and added seven more for patients with type-I BPD for a total of 12 associated factors, eight of which were statistically significant even with the adjustment for multiple comparisons (Table 2). Particularly, noteworthy were strong associations with initial episode types and long-term predominant polarity, as well as a strong association of suicidal risk with predominant-depression (Table 2). Among the 47.3% of patients who did not meet criteria for polarity-predominance (unclassifiable cases), risks of all 12 factors in Table 2 were intermediate between rates among subjects with predominant-polarities. As expected, some factors associated with predominant-depression or mania were not surprising. For example, predominant-depression was associated with: more use of ECT, longer latency-to-BPD diagnosis, suicidal risk, female sex, and being married (Table 2). These expected associations, as well as the consistently intermediate rates among unclassifiable subjects can be considered as supporting subtyping based on predominant-polarity.

Predictive value of predominant-polarity

To assess the potential clinical predictive value of subtyping by predominant-polarity, we computed Bayesian characteristics of observed associations of selected factors with predominant-depression vs. predominant-mania-like episodes. Positive predictive value (PPV: true-positive outcomes/all positive outcomes) was of particular interest. For the association of depressive or mixed first lifetime episodes with later predominant-depression, PPV = 80.4%; for the association of manic, hypomanic, or psychotic first episodes with later predominant-mania-like episodes, PPV = 75.2%; and for family history of affective illness with predominant-mania, PPV = 74.5%; however, for suicidal behaviour with predominant-depression, PPV = 26.6%. The low PPV for suicide attempts in association with predominant-depression probably reflected the low prevalence (8) of suicidal acts (18.0% overall). However, the negative predictive value (NPV) of predominant mania with lack of suicidal acts/all cases of predominant-mania) was much higher (NPV = 86.6%).

Comparisons of predominance criteria

We compared all of the 29 factors considered previously, using two definitions of predominance: the previously employed ≥66.7% excess (≥2 : 1) of one polarity vs. the broader criterion of ≥51%. As expected, the broader criterion allowed more cases [28/3%: 752/928 (81.0%) vs. 489/928 (52.7%)] to be characterized as having either predominantly depressive or predominantly mania-like recurrences. However, differences in information gained were minor. This conclusion is supported by consideration of the relative rates of association of specific factors with predominant-depression or predominant-mania-like episodes with each predominance criterion. Regression of 29 relative rates obtained with the criterion of ≥66.7% predominance vs. ≥51.0% predominance yielded a nearly 1 : 1 correlation (r = +0.939), with a slope function that did not differ significantly from unity (slope = 1.06; 95% CI: 0.90–1.22).

However, as the broader definition (≥51%) included the narrower definition (≥66.7% excess of depressive or mania-like episodes), we considered the contributions of subjects with predominance of depression or mania between 51.0% and 66.7% of episodes compared to those with predominance of ≥66.7% and tested for all factors tested in Table 2. Only one – the presence of any form of psychiatric or substance use comorbidity – showed a marginally significant excess association with predominant-depression [79.8%/65.6% = 1.22, χ2 (df = 1) = 3.86, = 0.05] not found with the ≥2 : 1 definition of predominance. These observations indicate that the broader criterion for predominant-polarity of recurrences can include more patients, but without apparent gain in informational value.

Multivariate modeling

Multivariate logistic regression modeling of factors found to be at least tentatively associated with predominant-mania or depression in Table 2 (at  0.05) indicated that only two remained significantly and independently associated with predominant-depression. These were the following: longer diagnostic latency (more years from initial illness to diagnosis as BPD; χ2 = 10.9, = 0.001) and suicidal acts (χ2 = 7.80, = 0.005).

Effect of categorization of mixed-states

We also considered effects of shifting mixed-state episodes (meeting DSM-IV criteria for both depression and mania) from the ‘mania-like’ to the ‘depression-like’ category in defining predominant recurrence polarity, again employing the ≥2 : 1 predominance criterion (≥66.7% of episodes of one polarity). We compared this definition with the previously recommended combining of mixed-states with mania and hypomania (1). Only two study sites (Barcelona and Belmont) provided information on mixed-states, separately, to support this secondary analysis, and 12 factors differed significantly between ‘depression-like’ and ‘mania-like’ episode predominance by at least one of the alternative definitions (not shown). Of these, seven were not made stronger (larger relative rate or lower P-value) by shifting mixed-states to the depression-like category. The five remaining factors with greater contrasts included the circular and trivial factor ‘ever being in a mixed state’ and four others. Two of the four were more strongly associated with predominant ‘depression-mixed’ than with ‘mania–hypomania’: suicide attempts (from 2.16- to 4.37-fold) and Axis-II comorbidity (from 2.41- to 3.86-fold, a 60.2% increase). In addition, drug abuse was somewhat more strongly associated with mania-like recurrences when mixed-states were removed (1.28- to 1.39-fold, or 8.59% greater), whereas family history of affective disorders became less strongly associated with mania-like episodes (1.70- from 2.19-fold, or a decrease by 28.8%). Of these findings, the strongest and most interesting observation was a marked increase in suicidal risk (by more than two-fold) when depressive and mixed episodes were combined.

Another noteworthy finding was that the type of first lifetime DSM-IV affective episode anticipated later predominant polarity. Predominant depressions + mixed-states were strongly associated with initial depressions + mixed-states + anxiety episodes, whereas later predominant-mania-like (mania, hypomania) or psychotic episodes were strongly associated with initial manias + hypomanias + psychoses (Fig. 1).

Figure 1.

 Relationship of type of first lifetime episode of DSM-IV illness vs. later predominant polarity: Pred-D, predominant depression or mixed-states; Pred-M, predominant-mania, hypomania, or psychosis. Initial episodes are major depression or mainly anxiety disorders, mixed manic-depressive states, or mania or psychosis. Ratios (Pred-D/Pred-M) for the prevalence of first episode types are as follows: mixed-states (46.7) depression–anxiety (7.19), mania–psychosis (1/5.97). These differences are highly significant (χ2 = 166, < 0.0001).

Association of suicidal acts with predominant-polarity

Given the strong association of suicide attempts with depressive or mixed episodes, we carried out a random effects meta-analysis of findings provided in four published reports (1–3, 5) and from the sites of the present study, each considered separately. This analysis indicated a very strong association of suicidal acts with predominant-depression, based on the 2 : 1 predominance criterion compared with mania-like episodes that included mania, hypomania, and mixed-states, with a pooled risk ratio of 1.98 (95% CI: 1.51–2.60; = 4.95, < 0.0001; Fig. 2).

Figure 2.

 Random effects meta-analysis of suicide attempt risk with predominant depression versus mania. Based on nine studies [four previously reported (1, 2, 3, 5), five new from components of the present report: Ha et al. (Ha K, Ha TH, Chang JS. Unpublished data) in Korea; Salvatore et al. (28) in the United States; Tondo et al. (25) in Italy; Undurraga et al. (12) in Spain; and Vázquez et al. (Vázquez G, Lolich M, Leiderman E., Unpublished data) and in Argentina] for adult DSM-IV bipolar I disorder patients with predominantly (≥66.7%) depressive or manic recurrent episodes (mixed-states classified with mania-like episodes). Risk was greater than the null (1.0; vertical line) in 7/9 of the studies, and the pooled relative risk (RR; filled diamond and vertical dashed line) was 1.98 (95% CI: 1.51–2.60; = 4.95, < 0.0001).

We also carried out meta-analyses with data from collaborating sites with specific information concerning mixed-states to compare the association of suicidal acts with depression alone vs. with depression-plus-mixed-states at the same sites. By shifting mixed-states from mania-like to depression-like recurrences, the association of suicidal acts with predominant-depression-plus-mixed states nearly doubled, from relative risk = 2.48 (CI: 0.41–14.8; = 0.99, = 0.32) to 4.62 (CI: 2.16–9.89; = 3.95, < 0.0001). These findings support an expected, strong association of suicidal behaviour with both depression and mixed-states.

Finally, we tested for a continuous association of the proportion of depressive or mixed-state episodes from sites providing such information (Barcelona and Belmont) with rates of suicide attempts in approximate quintiles defined by graded levels of depressive or mixed episodes in 390 patients (Fig. 3). This analysis indicated a strong, continuous relationship of increasing proportions of depressive or mixed episodes among all recurrences per year with higher risk of suicidal acts. The association was more nearly continuous and much more significant when both depressive and mixed recurrences were considered together than with depressions alone (Fig. 3).

Figure 3.

 Risk of suicidal acts (% of patients) vs. quintiles of proportions of depressive (shaded bars) or depressive-plus-mixed episodes (striped bars) of all recurrences per year in 390 patients with type-I bipolar disorder from Barcelona and Belmont. For depressive recurrences, χ2 [df = 4] = 11.4, = 0.02; for depressive + mixed recurrences, χ2 [df = 4] = 24.6, < 0.0001.


The present findings from a large, international sample of 928 patients with DSM-IV type-I BPD support and extend the original proposal by Colom and his colleagues (1) of subtyping by predominant polarity in BPD. Separating patients with type-I BPD by the predominant-polarity of episode recurrences was feasible in approximately half of the patients (53%). Broadening the criterion of predominant-polarity from a ≥2 : 1 excess to only ≥51% excess, as expected, increased the proportion of categorizable patients, as expected, but added very little to detecting clinical differences between predominant-polarity subgroups, and is likely to reduce the specificity of the predominance contrasts. In addition, shifting mixed-states from the mania-like to the depression-like category added little in detecting differences between predominant-polarity subgroups for most factors considered, with the notable exception that the risk of suicidal behaviour nearly doubled, to a 4.5-fold excess when mixed-states were combined with depressions. There was also a stronger continuous relationship of suicidal risk with the proportion of total recurrences that were depressive-or-mixed vs. depressive (Fig. 3). These findings and a meta-analysis of previous reports and data from the sites of the current study (Fig. 2) support other recent indications that mixed-states and well as depressions are strongly associated with suicidal behaviour in patients with BPD (8–12).

Specific associations with predominant-polarity replicated 5/7 characteristics found to be in ≥25% excess in previous, comparable studies (1–5; Table 1) and added eight new associations: eight factors were preliminarily associated with predominant-depression, and four with predominant-mania-like recurrences (Table 2). Also, the percentage of weeks ill was somewhat higher with predominant-depression. Multivariate modeling sustained significant and independent association of predominant-depression with longer delay of BPD diagnosis and more suicidal behaviour.

Several of these associations were to be expected, particularly with predominant-depression, and can be considered as supporting subtyping by predominant-polarity. Others are of greater potential research interest. Particularly, important were strong associations between the polarity of first lifetime episodes with later predominant-polarity (Fig. 1; Table 2), consistent with previous associations between the polarity of index and subsequent episodes, or of first and later episodes (10, 13–18). The association of first episode type and later predominant-polarity showed substantial positive predictive value (PPV, 75–80%), suggesting that early illness types can support prediction of future morbidity. Such predictions may have clinical value in planning for long-term care of patients with BPD and advising about likely prognosis and responses to particular treatments, with the important limitation that only half of subjects could be classified by predominant polarity. Nevertheless, predominant-polarity often can be established within several years of onset. There also is emerging evidence that specific mood-stabilizing agents may have particular benefits against recurrences of bipolar depression (e.g., lamotrigine, lithium, and quetiapine) – the far more difficult to treat component of BPD (19) – whereas many mood-stabilizers and antipsychotic drugs are more effective in mania and mania-like recurrences (20, 21). It may be feasible to recommend particular treatment choices of short- and long-term treatments relatively early in the course of BPD, and such decisions may become more important as additional mood-stabilizing agents with selective beneficial effects are developed (20–24).

There were strong associations of predominant-depression, especially with mixed-episodes included, with suicidal behaviour (Table 2, Figs 2 and 3). However, this association provided limited PPV (only 27%), probably reflecting the relatively low prevalence (7) of suicide attempts (18%) even in a high-risk group like patients with BPD (25). On the other hand, the NPV of the association of suicidal behaviour with predominant-depression was quite high (NPV = 89%). That is, predominant-mania was strongly associated with the absence of suicide attempts (PPV = 89%).

It is noteworthy that patients eventually diagnosed with BPD have had an extended history of early symptomatic expression of abnormalities of mood and behaviour including depression (18, 26–28), with typical latencies to diagnosis of BPD of 4–10 years, and averaging 4.05 ± 6.69 years in this study. This latency was 3.4-times longer with predominant-depression than predominant-mania, as expected, since early depressive recurrences were sometimes followed later by mania (Table 2). Nevertheless, it is evidently often possible to characterize predominant-polarity in patients with type-I BPD relatively early in the illness course (within the first 4–5 years), or by the time the diagnosis of BPD is established, adding to the potential prognostic value of such categorization.

As subgrouping patients with BPD by predominant-polarity has been proposed only recently, its full significance and potential clinical and research value require further study. It is notable that, as in many efforts at categorical nosology in psychiatric disorders, there may be more or less archtypical groups at the extremes of predominant-polarities, with a large intermediate group (47.3% of the present cases) without clear polarity-predominance. Such subtyping implies that type-I BPD almost certainly is not a homogenous disorder, despite its relatively stereotyped adult presentations and remarkable long-term diagnostic stability (28). Subtyping by predominant-polarity may contribute to more focused genetic or other psychobiological efforts to characterize more homogeneous subgroups, particularly in view of the observed association of predominant-mania with familial psychiatric illness (Table 2).

Remaining uncertainties about subgrouping patients with BPD by predominant-polarity include the need to consider the proportion of time ill in particular morbid states during follow-up, and not only counts or rates recurrences of particular of major episode types (18). It also may be important to consider the effects of high levels of inter-episode ‘subsyndromal’ morbidity, particularly of dysthymia and dysphoria, which are likely to contribute to adverse outcomes (11, 13, 18). Also needed are comparisons of subtypes of patients with BPD defined by predominant-polarity vs. other criteria [such as types I, II, and other mainly depressed cases (29, 30)]. Finally, the 47% of patients with type-I BPD who could not be categorized by predominant-polarity need further study.

Limitations of this study, notably, include substantial site variance in the ascertainment of predominant-polarities. Such variance may limit generalizations, despite use of a large, international sample to limit effects of local variation (1). Basing findings on geographically and culturally different sites tends to support the robustness of subdividing patients with type-I BPD by predominant-polarity of major episodes. In addition, some distinctions of cases with dissimilar predominant-polarities from each other or from unclassifiable patients with BPD were limited, and some factors were expected. The low prevalence of suicide attempts limits the potential value of predicting suicidal risk by predominant-polarity. Nevertheless, other factors, such as prediction of later course from initial presentations, and the association of family history with predominant-mania appeared to have more robust predictive value, with the major caveat that nearly half of patients could not be classified by predominant polarity even after several years of illness.

In conclusion, the present findings support and extend the proposal that patients diagnosed by DSM-IV criteria with type-I BPD can usefully be subdivided by predominant-polarity of recurrent major episodes, based on at least a two-fold excess of one polarity type. Several factors found to be associated with predominant-depression or mania were expected and well-known clinically. Nevertheless, future, long-term morbidity patterns could be distinguished from very early stages of BPD illness, and support was found for considering mixed-states with depression rather than mania. Subtyping by predominant polarity may have value for future research as well as prognostic and potential therapeutic clinical implications.


Supported in part by a grant from the Bruce J. Anderson Foundation and the McLean Private Donors Psychopharmacology Research Fund (to RJB), by a Traveling Research Fellowship from the University of Barcelona and the Fundación Española de Psiquiatría y Salud Mental (to JU), by the Lucio Bini Private Donors Mood Disorders Research Fund (to LT), and by research grant A-101905 from the Korean Ministry of Health (to KH).

Declaration of interests

Doctor K. Ha has received research support or honoraria from Pfizer, Otsuka, AstraZeneca, and Eli Lilly Corporations. Doctor Tohen is a former employee of Eli Lilly (to 2008) and has received honoraria or consulted for AstraZeneca, Bristol Myers Squibb, Glaxo Smith Kline, Eli Lilly, Johnson & Johnson, Sepracor, Otsuka, Merck, Sunovion, Forest, Lundbeck and Wyeth Corporations; his spouse is a current employee and minor stockholder at Eli Lilly. Doctor Vieta has been a consultant or grant recipient with the following companies: Almirall, Astra-Zeneca, Bristol-Myers-Squibb, Eli Lilly, Forest Research Institute, Geodon Richter, Glaxo-Smith-Kline, Janssen-Cilag, Jazz, Lundbeck, Merck, Novartis, Otsuka, Pfizer, Sanofi, Servier, Schering-Plough, Takeda, and United Biosource Corporations. No other author or close family member has current financial relations that might represent a conflict of interest in the work presented.