Risk of Recurrence in Women With Bipolar Disorder During Pregnancy: Prospective Study of Mood Stabilizer Discontinuation

Pregnant women (N=89) diagnosed with DSM-IV type I (N=61) or II (N=28) bipolar disorder were enrolled in this prospective, observational study of pregnancy between March 1, 1999, and Aug. 31, 2004, at the Perinatal and Reproductive Psychiatry Clinical Research Program at Massachusetts General Hospital, Boston. The subjects were recruited among women planning pregnancy and seeking specialized psychiatric consultation, as recommended by their obstetricians or by self-referral.

Study subjects were eligible if they 1) had a history of bipolar disorder before pregnancy, 2) were euthymic for at least 4 weeks before their last menstrual period, 3) were receiving mood stabilizer therapy or 4) discontinued pharmacotherapy within 6 months before conception or 12 weeks after conception, and 5) were enrolled before 24 weeks of gestation. Patients were excluded if they 1) were actively suicidal, 2) had discontinued all mood stabilizer therapy >6 months before conception, or 3) met DSM-IV criteria for a primary psychotic, schizoaffective, or organic mental disorder or mental retardation. All subjects provided written informed consent to participate after approval of the study protocol by the Massachusetts General Hospital Institutional Review Board.

The subjects were followed through the end of pregnancy and for 12 months postpartum, regardless of their decisions concerning continued use of psychotropic medication. All subjects initially received individual evaluations by the first author (A.C.V.), including an extensive clinical assessment that included the Structured Clinical Interview for DSM-IV (SCID) (29), and comprehensive review of potential risks and benefits of continuing or stopping treatment, as detailed elsewhere (17) . This review covered current knowledge of the potential teratogenicity of mood stabilizing, antipsychotic, sedative-hypnotic, and antidepressant drugs, their potential adverse effects on neonates, and maternal and potential fetal risks associated with discontinuing treatment.

Initial SCID assessment confirmed a DSM-IV lifetime diagnosis of bipolar disorder and evaluated the presence of comorbid psychiatric illnesses. Demographic and clinical characteristics of interest were recorded at baseline, including family history, estimated age at onset of bipolar disorder, approximate number of prior episodes and hospitalizations, and occurrence of suicide attempts, as well as the type, severity, and time since the approximate onset and end of the last major affective episode and treatment history (medicines, doses, and estimated exposure times). The subjects were followed up prospectively by A.C.V. at a study visit each trimester and at 6, 12, 24, and 52 weeks postpartum to ascertain the presence of major symptoms and their clinical severity (mild, moderate, or severe) and current treatments (drugs, doses, apparent benefits, and adverse effects), noting any changes. In addition, the primary outcome variable of recurrence of a new DSM-IV illness episode was determined by an independent rater (trained and experienced research assistant) who was blind to treatment status with the SCID mood module, as reported previously in a prospective, longitudinal study of risk recurrence among pregnant women with recurrent major depressive disorder (30) . The blinded assessment covered the interval from the last to the current study visit, with a best estimate of the gestational week of illness onset, all verified with A.C.V. when a recurrence was detected. All subjects were managed clinically by their treating psychiatrist without blinding to treatment status.

The primary outcome variables were recurrence and weeks to the start of an illness recurrence fulfilling DSM-IV criteria based on the SCID mood modules for mania, hypomania (lasting ≥1 week), major depression, or a mixed state. Neonatal and postpartum outcomes will be reported separately. Given the complexity and multiple changes in pharmacologic regimens among patients with bipolar disorder, we stratified the study group into two treatment groups based on mood stabilizer treatment status only: 1) use of at least one mood stabilizer at conception and continued at least through the first 12 weeks of pregnancy or 2) discontinuation of all mood stabilizer treatment during the period ranging from 6 months before conception to 12 weeks of gestation.

We first compared the distribution of demographic and clinical characteristics between the treatment groups to identify potential confounding factors. Associations of treatment status with risk and latency of recurrence were then assessed. Kaplan-Meier survival analyses determined median weeks to start of a first recurrence with 95% confidence intervals (CIs), with unadjusted univariate Mantel-Cox log-rank tests (χ ² ) to compare survival times between treatment groups, and censoring at birth or miscarriage (31) . Cox multivariate proportional hazards modeling estimated the hazard ratio and 95% CI for median time to recurrence between treatment groups, with adjustment for predictors of recurrence suggested by preliminary univariate analyses. This modeling employed forward variable selection to adjust for the effects of potential confounders and to identify potential risk factors for illness recurrence. We also used survival analysis to compare weeks to recurrence dated from the point of discontinuation of mood stabilizer treatment to compare subgroups that discontinued abruptly or rapidly (1–14 days) versus gradually (≥15 days of dose reduction). Finally, we used logistic regression modeling to evaluate risk factors for significant and independent association with recurrence.

Summary data are reported as means with standard deviations for continuous variables, survival-computed median weeks to events and hazard ratios with 95% CIs, and proportions (percentages) for categorical data, which were compared with contingency tables (χ ² ) or Fisher’s exact test (p). Statistical tests of hypotheses were two-sided, at α=p≤0.05. Analyses used commercial statistical programs (STATA version 8.2, College Station, Tex.).

Among 89 women with bipolar disorder (69% bipolar I disorder, 31% bipolar II disorder) in the study group, 85 had a live birth, two subjects had stillbirths at term, one woman miscarried after a recurrence, and one subject dropped out of the study after a recurrence. Table 1 illustrates demographic, clinical, and treatment characteristics of all participants as well as those who discontinued or maintained treatment with a mood stabilizer. Overall, the mean age of the subjects was 32.7 years (SD=5.4). The majority of subjects were Caucasian (96.6%), had ≥12 years of education (98.9%), and were married (82%), employed outside the home (76.0%), and multiparous (64%). Most subjects (>70%) were taking more than one psychotropic, which included a mood stabilizer in combination with an antidepressant and/or antipsychotic ( Table 1 ). Across the study group, primary mood stabilizer type ranked as follows: lithium (55 of 89=61.8%) > anticonvulsants (32 of 89=36.0%; valproic acid, N=15; lamotrigine, N=8; carbamazepine, N=6; gabapentin, N=3) > atypical antipsychotics (two of 89=2.3%; olanzapine, N=1; and quetiapine, N=1). Over half the study group were exposed to an antidepressant (51.7%, 46 of 89; bupropion, N=15; sertraline, N=7; fluoxetine, N=8; fluvoxamine, N=1; paroxetine, N=5; venlafaxine, N=2; citalopram, N=4; selegiline, N=1; tricyclic, N=1; tranylcypromine, N=2) in addition to a mood stabilizer. Among subjects who continued taking a mood stabilizer, 18.5% used adjunctive antidepressants versus 66.1% (41 of 62) of those who discontinued mood stabilizer. In addition, 27% (24 of 89) of the cohort was exposed to an antipsychotic (olanzapine, N=6; quetiapine, N=2; perphenazine, N=4; risperidone, N=4; thiothixine, N=2; haloperidol, N=3; or ziprasidone, N=3), and the proportion of adjunctive antipsychotic use was higher among subjects who maintained versus discontinued a mood stabilizer ( Table 1 ).

At study entry, the two treatment groups did not differ significantly with respect to demographic features (i.e., age, race, marital status, years of education, employment, and previous pregnancy) and several characteristics, including unplanned pregnancy, average episodes/year, current comorbidity, prior hospitalizations, prior suicide attempts, time since last episode, and prior perinatal and postpartum episodes ( Table 1 ), as well as family history of mood disorders, age at diagnosis and at first mood stabilizer treatment, prior mixed episodes, and recent sedative use (not shown). However, the two groups did differ along several important measures of illness severity ( Table 1 ). Those who continued maintenance mood stabilizers during pregnancy were more likely to 1) have bipolar I disorder, 2) be maintained with lithium as the primary mood stabilizer, 3) have a history of psychotic features, and 4) be treated with adjunctive antipsychotics. The women who discontinued mood stabilizer treatment were more likely to 1) have a bipolar II disorder diagnosis, 2) have discontinued an anticonvulsant, 3) have experienced onset of bipolar disorder at a younger age and with a depressive first episode, 4) be treated with ≥two psychotropics, 5) be ill more total years, 6) experience a history of rapid cycling (≥4 episodes in any year), and 7) be treated with an antidepressant ( Table 1 ).

During pregnancy, a total of 70.8% (63 of 89) of the subjects experienced at least one episode of illness meeting DSM-IV SCID criteria. Overall, there were a total of 81 episodes, including shifts in polarity without intervening recovery, yielding an average of 1.3 episodes per affected subject (81 in 63) during the study period ( Table 2 ). Recurrence risk was 2.3 times greater after discontinuation of mood stabilizer treatment (53 of 62, 85.5%) than with continued treatment (10 of 27, 37.0%; Table 2 ). In addition, the proportion of time spent ill (i.e., with a mood episode) during pregnancy was nearly a third (33%) of the pregnancy across the entire cohort. The subjects who discontinued the mood stabilizer spent over 40% of pregnancy in an illness episode, versus only 8.8% of pregnancy among subjects who maintained the mood stabilizer ( Table 2 ).

Based on Kaplan-Meier survival analyses, the median time to first recurrence was 9.0 (95% CI=8.0–13.0) weeks after discontinuing treatment and >40 weeks (95% CI indeterminate) with continued treatment ( Figure 1 ). With respect to the timing of recurrences, the majority of new episodes emerged early in pregnancy: 47.2% risk in the first trimester, 31.9% in the second, and 18.8% in the third ( Table 2 ).

Women who discontinued abruptly or rapidly (1–14 days; N=35) experienced a 50% risk of recurrence within 2.0 (95% CI=1.0–6.0) weeks, and those who discontinued gradually (≥15 days, N=27) required 22.0 (95% CI=16.0–38.0) weeks to reach the same level of recurrence risk (χ ² =25.9, df=1, p<0.0001; not shown). It is noteworthy, and perhaps not surprising, that unplanned pregnancy covaried with greater likelihood of rapid discontinuation of mood stabilizer treatment (23 of 24, 95.8%, versus 12 of 59, 20.3%, in planned pregnancies; p<0.0001, Fisher’s exact test).

The distribution of polarities across all first recurrences ranked as follows: major depression (41.3%, 26 of 63) > mixed states (38.1%, 24 of 63) > hypomania (11.1%, seven of 63) > mania (9.5%, six of 63), indicating a 3.8-fold excess of depressive or dysphoric (mixed) versus manic or hypomanic recurrences (79.4%, 50 of 63, versus 20.6%, 13 of 63). The excess of depressive-dysphoric versus manic-hypomanic episodes was even greater after discontinuation of mood stabilizer treatment (55 of 62 recurrences, 88.7%, versus 12 of 62, 19.3%, or 4.6-fold) compared to continued treatment (five of 27, 18.5%, versus nine of 27, 33.3%, or only a 1.8-fold difference). The excess of new depressive-dysphoric versus manic-hypomanic illness was found among both bipolar I disorder cases (25 of 61, 40.9%, versus 12 of 61, 19.6%, or 2.1-fold) and bipolar II disorder subjects who demonstrated a 24-fold excess of depression over hypomania (25 of 28, 89.3%, versus one of 28, 3.7%).

We examined whether certain demographic or clinical variables, other than discontinuation of mood stabilizer, were associated with recurrence during pregnancy. No statistically significant association was noted between recurrence risk and race, educational status, or marital status. However, several other clinical factors, including illness history, pregnancy, and treatment-related factors were associated with illness recurrence during pregnancy. Significant illness-history-related predictors included the following in rank order by risk ratio: 1) ≥5 years of illness (risk ratio=1.7, p<0.001), 2) younger age at onset (risk ratio=1.6, p<0.001), 3) bipolar II disorder diagnosis (risk ratio=1.5, p<0.002), 4) history of rapid cycling (risk ratio=1.5, p<0.002), 5) shorter clinical stability since the last episode before conception (risk ratio=1.5, p<0.004), 6) one or more prior episodes per year (risk ratio=1.5, p=0.004), 7) previous mixed-state episodes (risk ratio=1.5, p=0.004), 8) prior suicide attempts (risk ratio=1.4, p=0.01), and 9) current psychiatric comorbidity (risk ratio=1.4, p=0.02). Pregnancy-related risk factors associated with recurrence only included 10) unplanned index pregnancy (risk ratio=1.5, p=0.006), but not previous live birth or prior history of a mood episode during pregnancy or the postpartum period. Treatment-related risk factors, besides discontinuation of mood stabilizer, included 11) polytherapy with two or more psychotropic agents (risk ratio=2.3, p<0.001), 12) use of antidepressants (risk ratio=2.0, p<0.001), 13) primary mood stabilizer other than lithium (risk ratio=1.6, p<0.001), 14) previous switch from depression to mania-hypomania during treatment with an antidepressant (risk ratio=1.5, p<0.009), and 15) abrupt discontinuation of mood stabilizer (risk ratio=1.4, p=0.008). Factors not associated with recurrence included race, age, education, marital status, employment, previous pregnancy, prior illness in pregnancy or during the postpartum period, first episode depressive, any family history of mood disorder, and current use of more than one mood stabilizer, among others (not shown).

The preceding factors preliminarily associated with recurrence of bipolar disorder illness during pregnancy were subjected to multivariate analysis with Cox proportional hazards modeling to test for factors remaining independently associated with recurrence latency (not shown). Without adjustment for covariates, time to recurrence was much shorter after treatment discontinuation (hazard ratio=3.80, 95% CI=1.90–7.60) and remained so (hazard ratio=2.50, 95% CI=1.20–5.20) even after adjustment for indices of illness severity (lifetime years ill and number of prior episodes), bipolar disorder diagnostic subtype (I or II), and antidepressant use. Furthermore, in this multivariate model, indices of illness severity and bipolar type were no longer associated with increased hazard, but antidepressant use remained a robust predictor of recurrence risk along with treatment discontinuation (hazard ratio=2.2, 95% CI=1.2–4.2, p=0.02).