In contrast to lifetime prevalence data reported for major depressive disorder, in which the ratio of women to men is 2:1, epidemiologic studies demonstrate that the gender distribution for patients with bipolar disorder is 1:1 (1—4); men and women are equally likely to develop bipolar disorder over their lifetime. Gender differences have consistently been identified, however, in certain aspects of the disorder, including risk for rapid cycling, mixed states, and age at onset of illness (5—9). Whether gender differences exist in either the proportion of time spent ill or the proportion of overall time spent specifically in the depressed phase of the disorder remains to be further clarified. There is ongoing debate on whether women are more likely than men to spend time in depression over the course of their bipolar illness.
Few reports have systematically and prospectively followed patients with bipolar disorder to evaluate time spent in depression, mania, and euthymia as a function of gender. In the earliest long-term (16 years) prospective study, Angst (10) observed that depressive episodes were more likely to predominate in women than in men with bipolar disorder. Since then, some studies (11—13), but not all (14—22), have observed this gender difference, and only four of these studies (10, 13—15) have been prospective.
Because time in depression is inversely correlated with quality of life measures (21—26), a greater incidence of depressive symptoms in women with bipolar disorder would place women at an elevated risk for impaired overall functioning (10, 23). In an effort to clarify previous observations in the literature, we evaluated a large-scale prospective database combining patients from seven psychiatric sites in the United States (four sites) and Europe (three sites). The Stanley Foundation Bipolar Treatment Outcome Network prospectively tracked patients for up to 7 years in a naturalistic follow-up study. We used data from 711 patients with bipolar I or bipolar II disorder to evaluate our primary research hypothesis, which was that in patients with bipolar disorder, women would spend a greater proportion of their time over follow-up in the depressive pole of the illness than would men.
The mission of the Stanley Foundation Bipolar Treatment Outcome Network (1995—2002) has been described in detail elsewhere (27, 28). Briefly, patients with bipolar disorder were recruited from private, academic, and community outpatient settings by referral and advertisements and were enrolled if they were at least 18 years old; had a DSM-IV diagnosis of bipolar I or bipolar II disorder; were willing and able to perform prospective daily mood charting and to attend monthly evaluation appointments; were willing to be in some form of ongoing treatment with a psychiatrist; and were capable of providing written informed consent after the study procedures had been fully explained. There was no payment for participation.
Diagnoses were made using the Structured Clinical Interview for DSM-IV (29) at entry to the Stanley network. Patients were followed naturalistically and prospectively on a monthly basis, and ongoing medication changes were made on the basis of need. At each visit, the Inventory of Depressive Symptomatology—Clinician-Rated version (IDS; 30), the Young Mania Rating Scale (YMRS; 31), and the Clinical Global Impressions scale adapted for bipolar disorder (32) were administered. Across all sites, high interrater reliability was maintained (kappa values were 0.70 for the YMRS and 0.85 for the IDS). While in the network, patients were enrolled in the naturalistic follow-up study and could additionally enroll separately in defined open or double-blind clinical trials.
In this study we analyzed data from 711 patients with bipolar I and II disorders collected while patients were in naturalistic follow-up. Data from patients with bipolar disorder not otherwise specified and schizoaffective disorder were not included in this analysis because of small sample sizes. Also, data from periods when patients were in open or double-blind clinical trials were not included in this analysis because their patterns of mood states were likely to differ from those in the naturalistic follow-up study. The clinical trials attempted to directly influence mood state by treating a depression or mania, and this correspondingly required visits on a biweekly rather than a monthly basis. Thus the trial visits, which accounted for a minority of visits in the network, were not comparable either qualitatively or in number to the naturalistic clinical visits.
Observations were included in the analysis if both the YMRS and IDS were completed on the same day as the visit and the data were available. To identify visits on which patients had symptoms of depression, we used the IDS and the YMRS scales. The definitions for the presence of manic and depressive symptoms were established a priori by the Stanley Foundation Bipolar Network consortium consensus guidelines (referred to as the "Stanley criteria"; see www.npistat.com/stanley/definitions.asp). The presence of depressive symptoms was operationalized as an IDS score ≥14 in addition to a YMRS score <8. (Given that we have previously reported on high rates of depression in women who have hypomania , for this analysis we focused not on mixed states but only on depressive symptoms that occurred in the absence of hypomania or mania in bipolar illness.) Additionally, we defined three subcategories of depression according to the Stanley criteria: mild depression, with an IDS score of 14—21 (subsyndromal depressive symptoms); moderate depression, with a score of 22—30; and severe depression, with a score ≥31 (34). The presence of hypomania or mania was defined as a YMRS score ≥8. We also divided mania into two subcategories, with mild mania defined as a YMRS score of 8—20 and moderate-severe mania defined as a score ≥21. The presence of euthymia was defined as an IDS score <14 and a YMRS score <8.
The primary analysis consisted of a sequence of repeated-measures mixed-effects regressions using a logistic link function to model, in turn, the probability that an individual was depressed, manic, or euthymic at a given visit as a function of gender and other covariates. This structure allowed us to account correctly for the multiple observations per subject. Our primary outcome measure was whether a patient had depressive symptoms at a given visit. We first fitted a model with gender as the only independent variable to determine whether there was a significant difference between men and women. This was followed by a second model additionally controlling for diagnosis, age at study entry, rapid cycling, history of an anxiety disorder, history of a substance use disorder, and time in the Stanley network to determine whether these factors helped account for any observed gender differences. Similar pairs of models were fitted with indicators for hypomania or mania and euthymia as the dependent variables. The models were fitted using the generalized linear mixed model as implemented in the lme4 package of the R statistical programming language (35). Analyses were conducted using gender, diagnosis (bipolar I or II disorder), age, and binary indicators for rapid cycling and history of anxiety or substance use disorders (all as determined at study entry) and all two- and three-way interactions of these predictors as fixed effects. Time in the Stanley network was treated as a time-varying covariate (changing with each visit), and a random intercept was included for each participant to account for correlations among multiple visits. The inclusion of patient-level random effects helps to ensure that any observed gender differences are not due simply to a small number of patients in one group having a very large number of visits in a particular mood state. Because we were interested in understanding whether each of these variables was a significant predictor of the likelihood of depressive symptoms after adjusting for the other covariates (i.e., had independent explanatory power), we report p values for the z tests of the individual parameter estimates in the joint model rather than follow-up contrasts.
The database contained 711 patients (406 women and 305 men) with bipolar I disorder (N=572) or bipolar II disorder (N=139) who had visits during which they were not actively participating in a clinical trial. These patients had a total of 13,191 visits during naturalistic follow-up on which both the YMRS and the IDS were completed. Table 1 gives an overview of the demographic and clinical characteristics of the patient population. The mean age of the sample was 42 years (SD=11.6). On average, men were slightly but significantly older than women and had a significantly later onset of illness, but the mean duration of illness did not differ between genders. Despite the similar duration of illness, women had a significantly greater number of prior depressive episodes and hospitalizations for depression. While men and women did not significantly differ in the number of lifetime episodes of mania, there was a trend suggesting that men had a greater number of lifetime hospitalizations for mania. Men were more likely than women to have a lifetime history of substance use disorders, and women were more likely than men to be rapid cyclers and to have a lifetime history of an anxiety disorder. Rapid cycling status and history of anxiety disorders were also significantly associated (χ2=9.73, df=1, p=0.003), with patients diagnosed with a lifetime anxiety disorder being almost twice as likely to be rapid cyclers (odds ratio=1.6).
Table 1. Demographic and Clinical Characteristics of 711 Patients With Bipolar Disorder
Gender differences in patterns of study visits were also examined to verify that these were not the basis of observed differences in mood state (Table 2). Most of the variables were skewed. Follow-up ranged from 0 to 2,437 days, with a mean of 765 days and a median of 602 days. Patients had an average of 18.5 visits (median=14) and an average time between visits of 44.8 days (median=33), which was close to our target for monthly sessions. There was no evidence of significant differences in visit patterns by gender in terms of number of visits, follow-up time, or spacing of visits, which suggests that any gender differences were not attributable to a small abnormal subgroup in one of the samples (an issue further addressed by the inclusion of patient-level random effects in our models).
Table 2. Visit Characteristics of 711 Patients With Bipolar Disordera
Proportion of Visits in Depression
Figures 1—3 present the whole-sample empirical values of proportion of visits in different mood states for all patients with bipolar disorder (Figure 1) and then for patients with only bipolar I disorder (Figure 2) or bipolar II disorder (Figure 3). For ease of interpretation, we present the raw proportions rather than the model estimates adjusted for the repeated-measures structure of the data. However, all reported significance values are adjusted to ensure that they properly account for the different numbers of and correlations between the observations for each subject.
Figure 1. Proportion of Time Spent Ill During Clinical Visits for Women and Men With Bipolar I or II Disorder (N=711)a
aWomen and men, 35.6% compared with 28.7% of visits depressed; 50.4% compared with 56.9% of visits euthymic; 14.1% compared with 14.4% of visits hypomanic or manic.
Figure 2. Proportion of Time Spent Ill During Clinical Visits for Women and Men With Bipolar I Disorder (N=572)a
aWomen and men, 35.4% compared with 29.3% of visits depressed; 49.3% compared with 56.1% of visits euthymic; 15.1% compared with 14.6% of visits hypomanic or manic.
Figure 3. Proportion of Time Spent Ill During Clinical Visits for Women and Men With Bipolar II Disorder (N=139)a
aWomen and men, 36.1% compared with 26.2% of visits depressed; 55.7% compared with 60.2% of visits euthymic; 8.1% compared with 13.7% of visits hypomanic or manic.
Women had depressive symptoms at a significantly higher proportion of their visits than men in the model without covariates (35.6% compared with 28.7%; z=3.33, p<0.001, odds ratio=1.41). This difference was accounted for by the higher rates in women of both rapid cycling and anxiety disorders, each of which was associated with a greater likelihood of depression. None of the other covariates or interactions were significant. In the follow-up models adjusted for rapid cycling, bipolar diagnosis, age, time in study, anxiety and substance abuse comorbid disorders, and interactions, gender was no longer a significant predictor, but both rapid cycling (z=2.47, p=0.014, odds ratio=2.01) and comorbid anxiety disorders (z=4.22, p<0.001, odds ratio=2.26) were associated with a higher probability of depressive symptoms. Additionally, the longer a patient was part of the Stanley network, the less likely he or she was to have depression at a given visit (z=7.92, p<0.001).
A follow-up analysis of the pattern of severity of depression, again using mixed regression models, revealed no significant differences by gender. For men, 45.1% of the visits with depressive symptoms were for mild depression, 29.3% for moderate depression, and 25.6% for severe depression. For women, 39.9% of the visits were for mild depression, followed by 30.4% for moderate depression and 29.7% for severe depression. A history of an anxiety disorder was associated with more severe depressive symptoms (z=3.89, p<0.001), but none of the other covariates or interactions were significant.
Proportion of Visits in Euthymia and Hypomania or Mania
Parallel analyses were conducted with euthymia and hypomania or mania as the outcome variables. There was no evidence of a gender difference for hypomania or mania, even without adjusting for covariates (women, 14.1% of visits; men, 14.4% of visits; z=0.59, p=0.07, odds ratio=1.07 in the gender-only model). However, in the full model there were main effects of both time in study (z=7.91, p<0.001) and diagnosis (bipolar I or bipolar II disorder, z=3.41, p<0.001). As with depression, time in the network was inversely associated with likelihood of hypomanic or manic symptoms and, as expected, patients with bipolar I disorder spent significantly more time in hypomania or mania (z=3.41, p<0.001) than those with bipolar II disorder spent in hypomania. Additionally, although there was no main effect of gender in this model, there was a significant interaction between gender and bipolar diagnosis (z=2.55, p=0.01). Post hoc tests revealed that while there was no difference in the proportion of time spent hypomanic or manic in men and women with bipolar I disorder (14.6% compared with 15.4% of visits; z=0.05, p=0.96, odds ratio=1.01) (Figure 2), among patients with bipolar II disorder, men spent significantly more visits hypomanic than did women (13.7% compared with 8.1% of visits; z=2.94, p=0.003, odds ratio=2.79) (Figure 3).
Men spent a greater proportion of time in euthymia than women (56.9% compared with 50.4% of visits; z=2.44, p=0.015, odds ratio=0.66 for women compared with men in the model without covariates). However, as with depression, this difference was accounted for by the higher rates in women of rapid cycling and lifetime comorbid anxiety disorders, both of which were associated with a lower likelihood of euthymia (z=2.72, p<0.001, odds ratio=0.47, and z=4.14, p<0.001, odds ratio=0.45, respectively). While time in the study was associated with time in euthymia (z=10.34, p<0.001), neither gender nor any of the other covariates and interaction effects were significant in the full model.
In this analysis of a prospectively observed cohort with bipolar disorder, there were four main findings. First, the proportion of visits spent in a depressed mood state over 7 years was significantly greater for women than for men. Since there were no differences between men and women in duration of study participation, our results suggest that the observed gender difference in depression visits can be accounted for by the fact that a higher fraction of women in the study were rapid cyclers or had lifetime comorbid anxiety disorders (both of which were significant predictors of proportion of visits with depressive symptoms). With the data we had available to us, however, it was difficult to tease apart the causal mechanisms and their relative contributions to the gender effect. Second, while no gender difference was observed in time spent in hypomania or mania for men compared with women in the overall cohort, among patients with bipolar II disorder, men were significantly more likely than women to be hypomanic at a given visit. Third, women were more likely than men to be ill at a visit (depressed or hypomanic or manic), but this difference was also accounted for by their higher rates of rapid cycling and lifetime comorbid anxiety disorders. Finally, on average, the longer patients participated in the network, the less likely they were to be ill at a visit.
To our knowledge, only four prospective studies tracking mood as a function of gender have been reported in the literature. Angst (10) followed 95 patients over a 16-year period, recording number and type of mood episodes following an initial inpatient admission. While a preponderance of the data were prospective, some retrospective data were included. Women had nearly twice the proportion of depressive episodes that men had (60% compared with 35.5%), and their proportion of manic episodes was nearly one-third that of men (13.6% compared with 35.5%.) Thus, women not only spent a significantly greater proportion of time than men in depression, but they also spent a greater proportion of their ill time in the depressive (60%) compared with the manic (13.6%) phase than men, who demonstrated no difference in ill time spent in the depressed (35.5%) compared with the manic (35.5%) phase. In another study, Christensen et al. (13) followed 56 bipolar patients prospectively. Patients were interviewed every 3 months for up to 3 years with structured mood rating scales (e.g., the Hamilton Depression Rating Scale, the Newcastle scale, and the Bech-Rafaelsen Mania Scale) to evaluate mood state changes. As in our study, the proportion of examinations in a given mood state were analyzed. The investigators found that women spent a significantly greater proportion of time in depression than men (10% compared with 5% of clinical visits) and men a significantly greater proportion of time in a manic phase than women (13% compared with 5% of clinical visits.) These two studies, like ours, suggest higher proportions of time spent in depression for women compared with men. Two other prospective studies, however, did not observe these gender differences. Benedetti et al. (15), in a 48-week prospective longitudinal study of 27 men and 45 women bipolar I patients, found no difference between men and women in median percentages of time meeting DSM-IV threshold criteria for depression or mania. Patients were interviewed every 16 weeks after treatment in a partial hospital program. Additionally, no differences were observed in time spent in subclinical depression or mania, based on the Longitudinal Interval Follow-Up Evaluation (36). Men and women in this approximately 1-year follow-up period exhibited no differences in rates of subthreshold or threshold mania or depressive symptoms and similar rates of time in remission. The length of follow-up was certainly shorter than in our study and the other two reported above, and the interval between follow-up visits was the longest (4 months). In the fourth and last of the prospective studies, Winokur et al. (14) followed the naturalistic course of 131 bipolar I patients over 10 years with interviews initially every 6 months and then yearly during the last 5 years of follow-up. Patients were assessed using the Longitudinal Interval Follow-Up Evaluation. As with the study conducted by Benedetti et al., no significant differences were found between men and women in the number of threshold depressions or manias as defined by the Research Diagnostic Criteria. The mean number of depressive episodes was 2.02 for men and 2.54 for women. It is interesting that the two prospective negative studies that report no gender differences both used the Longitudinal Interval Follow-Up Evaluation, with which patients are usually seen at 4- to 6-month intervals and asked to report retrospectively on their mood over the previous 6 months. Thus, while these two studies were prospectively tracking patients, they actually relied on retrospective recall over a long period for depressive and manic symptoms. This may have led to an underdiagnosing of overall symptoms because of prior recall or biased recollection of mood states.
Eleven other retrospective or cross-sectional studies have provided data on gender differences in bipolar depression, although only some of these studies focused on gender. Of the 11, only six (11, 12, 16, 18—20) were designed to examine gender differences within the course of bipolar disorder. Of these six, two (11, 12) presented data consistent with the findings we present here and four did not.
Our data, consisting of a large cohort of patients who were prospectively assessed monthly for syndromal and subsyndromal manic and depressive symptoms, add to a small literature assessing gender differences in bipolar disorder. Our observational longitudinal data set, involving prospective monthly clinical assessments, suggests that women have a significantly greater proportion of clinical visits with depressive symptoms than do men. Our findings additionally reveal that this is accounted for by women's greater rate of both rapid cycling and comorbid anxiety disorders. However, it is unclear whether it is one of these factors individually or a combination of both that drives the difference. For example, perhaps rapid cycling leads to higher levels of anxiety, which in turn lead to increased depression. Alternatively, a history of an anxiety disorder may increase vulnerability to depression, and depression in turn can increase anxiety. In our cohort, rapid cycling was associated with a history of an anxiety disorder, and both are associated with gender, making it difficult to disentangle their relative importance in explaining the gender difference in intermediate covariate models. Yet there is some suggestion in our data that anxiety may make the more important contribution. High rates of comorbid anxiety have been reported in both men and women with bipolar disorder (37—41), and comorbid anxiety has a negative impact on quality of life (1, 42—45). The data in the present study show gender differences in rates of specific phobia, obsessive-compulsive disorder, and posttraumatic stress disorder. Gender differences have been reported in rates of anxiety disorders (46—48) as well as comorbid anxiety disorders in the bipolar population (41) and may be related to the higher rates of depressive symptoms in women in ways that remain to be explored.
In the present study, we only evaluated depression without concurrent mania. In a prior study by our group (33), we examined rates of mixed hypomania or mania in women compared with men. The probability of having depressive symptoms during hypomania or mania was 72% for women, compared with 42% for men, which suggests a depressive diathesis for women even during hypomania or mania. This finding, combined with our current findings, suggests that while bipolar disorder affects men and women equally, there is a considerable depressive morbidity for women with the disorder in both the depressed and the manic poles. This is consistent with previous reports suggesting that there may be more mania-prone and depression-prone presentations of bipolar disorder, with women having a susceptibility to the depression-prone presentation (49).
Our study had several limitations. First, the total amounts of time spent depressed, euthymic, and hypomanic or manic (rather than the proportion of visits in which patients exhibited particular symptoms) were not directly available in this study cohort, nor were the number or timing of mood changes between visits. Thus, we chose to use status at visit as our outcome measure. Because patients adhered well to the regular visit pattern prescribed in the study design, we believe that this provided a good proxy for time spent in the different mood states. However, this point merits further study. Second, while our analyses suggested that the observed gender differences in likelihood of depressive or hypomanic or manic symptoms could be accounted for by the greater rates of rapid cycling and comorbid anxiety disorders in women, the data on rapid cycling status and anxiety disorders were only obtained at the initial visit. To tease out the mechanism underlying the gender differences, it would be helpful to have repeated measures on these covariates as well as on mood state. Third, studies in the literature (including our own) have varied on the lower IDS cutoff score consistent with a subsyndromal depression. Subsyndromal depressive symptom criteria have been established to have a low cutoff score of 12—15 on the IDS to denote mild subsyndromal symptoms (see http://www.ids-qids.org/index2.html#table4; adapted from Rush et al.  and Trivedi et al. ). In the present study, we used the cutoff score developed by the Stanley consortium of a conservative lower boundary of 14 for the presence of mild depressive symptoms (see the Stanley criteria at www.npistat.com/stanley/definitions.asp). The field could benefit from a consensus regarding the definition of subsyndromal depression so that future studies can be easily compared.
Our results suggest that women with bipolar disorder spend a greater proportion of their visits in depression than do men. Prospective, naturalistic follow-up studies of patients with bipolar disorder that assess for symptoms of depression and mania are rare. In this study, we assessed for the likelihood of the presence of manic or depressive symptoms at patients' visits. Our findings, combined with data reporting higher rates of mixed (depressive) mania in women compared with men, provide support for the notion that women spend a greater proportion of their time ill struggling with various degrees of depressive symptoms whether they are in the manic or the depressive pole of the disorder. The reasons for this, which may be biological (hormonal), psychosocial, or cultural, remain to be determined. Our study additionally suggests that future research should evaluate the association of rapid cycling and comorbid anxiety disorders in relation to depressive symptoms in bipolar disorder.