Early Intervention for Symptomatic Youth at Risk for Bipolar Disorder: A Randomized Trial of Family-Focused Therapy

Children between ages 9 years, 0 months and 17 years, 11 months who had a first-degree relative with BD I or II were recruited at the University of Colorado or the Stanford University School of Medicine between June 2008 and August 2010. Referrals originated from community practitioners, parent support groups, e-mail advertisements, and inpatient settings. After a full explanation of the procedures, children and parents read and signed University Institutional Review Board–approved assent and consent forms.

Eligibility criteria included the following: English speaking; at least one first-degree relative who met DSM-IV-TR criteria for BD I or II, based on the Mini-International Neuropsychiatric Interview (MINI)¹⁹; child met DSM-IV-TR criteria for a lifetime diagnosis of BD-NOS, MDD, or cyclothymic disorder, and had significant current affective symptoms (1-week Young Mania Rating Scale [YMRS] score >11 or 2-week Children’s Depression Rating Scale, Revised [CDRS-R] score >29).^20,21 The diagnostic criteria for BD-NOS were the same as those used in the Course and Outcome in Bipolar Youth (COBY) study and included the following: a distinct period of abnormally elevated, expansive, or irritable mood plus two (three, if irritable only) DSM-IV-TR symptoms of mania that caused a change in functioning, lasted ≥ 4 hours in a day, and occurred for a total of 4 or more days across the child’s lifetime.² If the main diagnosis was MDD, the youth must have had a full DSM-IV major depressive episode (MDE) within the previous 2 years. Cyclothymic disorder required that, in the preceding year, the child had multiple brief (i.e., 1- 2-day) periods of hypomanic, depressive or subsyndromal mixed symptoms and no more than a 2-month period without mood symptoms, without meeting the DSM-IV criteria for MDD or the COBY criteria for BD-NOS.

Trained MA/MD/PhD level diagnosticians with at least 2 years of experience administered the Washington University Schedule for Affective Disorders and Schizophrenia for Children (K-SADS).²² The K-SADS interviews were conducted separately with the youth and at least one parent; when discrepancies between reports occurred, they were interviewed conjointly. Current and lifetime K-SADS ratings were based on the most severe 1 to 2 weeks in the prior 4 months and the most severe 1 to 2 weeks in the subject’s lifetime, respectively. Board-certified psychiatrists conducted separate evaluations with the youth and at least one parent. Final diagnoses were based on information obtained from all sources.

Interrater reliability between three expert diagnosticians at Stanford was 0.90 (intraclass r) on K-SADS mood items and 0.75 (kappa) on the presence/absence of lifetime comorbid disorders. Interrater reliability (intraclass r) among 11 Colorado raters for K-SADS depression items was 0.89 and for mania items was 0.97; and 0.70 (kappa) for lifetime comorbid disorders. A cross-site reliability evaluation (seven raters at Colorado and six at Stanford; 54 ratings) yielded the following intraclass r values between raters for K-SADS items: depressed mood, 0.77; elevated mood, 0.90; irritability, 0.85; and grandiosity, 0.81 (p < .0001 for all).

At the initial or a second evaluation visit, we administered the MINI diagnostic interview to all available first-degree relatives. Interrater reliabilities at the two sites for diagnoses of relatives ranged from 0.76 to 0.87 (kappas) for presence/absence of mood disorders.

Five-minute speech samples (FMSSs) were obtained from parents (or, in one case, a grandparent) who were primary caregivers (i.e., lived with and had a minimum of 4 hours per week of contact with the child). Research staff members instructed the parent to “talk for 5 minutes about what kind of person (your child) is and how the two of you get along together.” In cases where two parents were available, FMSSs were obtained individually from each parent. FMSS audiotapes were coded by Ana Magana-Amato, M.A., who developed the FMSS-EE coding system.²³ Following convention, families were classified as high in EE if, during the FMSSs, one or both parents made at least one critical comment (e.g., “I resent his oppositional attitude”) or showed evidence of hostility or emotional overinvolvement. Families in which neither parent met the high-EE criteria were classified as low-EE. Single-parent families were assigned the EE rating of the individual parent. Reliability between Ms. Magana-Amato and 12 trained raters, each of whom completed 10 independent ratings, averaged 0.95 for agreement on classifications of parents as high or low EE, and 0.84 on the EE components (e.g., critical comments).

A data analyst who was not otherwise involved in the study randomly assigned participants to a treatment condition in a 50:50 split (FFT-HR or EC) using a modification of Efron’s biased coin toss.²⁴ The assignments were balanced within sites on age (< 13 years or ≥ 13 years), initial diagnosis (BD-NOS/cyclothymic versus MDD), and medications (mood stabilizer/atypical antipsychotic or not). Youth were not required to take medications to participate; 40% were medication free at entry (Table 1).

When families and youth opted for medication management through the study, psychiatrists saw them weekly for the first month, and then a minimum of once per month for the study year. Physicians followed a pharmacotherapy algorithm that specified first-, second-, and third-line agents, dosage adjustments, and rescue medications. Treatments for depression included citalopram, sertraline, and buproprion as first-, second-, or third-line treatments or lamotrigine if the child had a history of antidepressant-induced mania. Treatment of mania symptoms included aripiprazole, quetiapine, risperidone or lithium as first-, second-, third-, or fourth-line treatments. Treatment of comorbid attention-deficit/hyperactivity disorder (ADHD) included stimulants as first-line agents and atomoxetine as second-line, whereas treatment of comorbid anxiety disorders included citalopram, sertraline, and clonazepam as first-, second-, and third-line treatments, respectively. Study psychiatrists carefully monitored patients on antidepressants for signs of mood switching or increased cycling, and discontinued antidepressants when warranted.

FFT-HR was administered to the youth, parent(s), and available siblings in 12 hour-long sessions over 4 months (eight weekly, four biweekly) by trained therapists who followed a session-by-session manual (available at www.semel.ucla.edu/champ/resources). The goals of FFT-HR are to assist youth and family members to recognize and intervene early with symptoms of mood episodes and enhance intrafamilial communication and problem-solving. In the first four sessions (psychoeducation), the child and family develop an individualized mood management plan involving a summary of stress triggers, signs of worsening mood (as revealed through daily monitoring), and regulation strategies (e.g., regulating sleep cycles, removing oneself from volatile situations, or arranging for medication changes). In the second module (sessions 5–8; communication enhancement training), participants learn to down-regulate impulsive expressions of negative emotions through practicing expressing positive feelings, active listening, and diplomatically requesting changes in each other’s behavior. In the problem-solving module (sessions 9–12), participants learn to break down large problems (e.g., “We can’t get along”) into smaller ones (e.g., “We need to use lower tones of voice.”), generate and evaluate the pros and cons of various solutions, and select solutions to implement (e.g., alerting each other to aggressive voice tones).

EC consisted of 1 to 2 family sessions in which clinicians summarized the results of the diagnostic assessment, introduced daily mood monitoring, and offered recommendations and instructional handouts on managing mood swings. Families in both treatments could request additional crisis sessions from study clinicians during the post-therapy interval, or referrals for additional psychosocial services if needed. Children continued to see their study psychiatrists for the full year.

After an initial training workshop, two of the authors (D.J.M. and E.L.G.) provided biweekly or monthly supervision for clinicians, and made quality control ratings of audiotapes of FFT-HR and EC sessions using the Therapist Competence and Adherence Scales, Revised (TCAS-R).^14,25 The TCAS-R includes 13 clinician adherence/skill items rated 1 (poor) to 7 (excellent). Interrater reliabilities for TCAS-R items ranged from 0.61 to 0.86 (intraclass r) among four independent raters. Clinicians had to maintain overall TCAS ratings of ≥ 5.0 (very good) on >75% of sessions to work in each modality. Colorado and Stanford therapists had nearly identical fidelity ratings across the treatment conditions (Colorado: mean = 5.9 ± 1.3; Stanford: mean = 5.8 ± 1.4; 49 ratings, p = .65).

A broad age range (9–17 years) was chosen, because the cases referred to our clinics often included younger children, and because FFT-HR had been adapted to school-aged children during the open trial.¹⁸ FFT-HR included simplified handouts that emphasized visual stimuli (e.g., mood charts with facial images portraying emotional states), flipcharts to explain key concepts, and flexible session structures. Younger children responded better to training in communication and problem solving when the tasks resembled games (e.g., choosing skills to practice out of a hat). Parents played more active roles in directing younger children during in-session skill rehearsals, whereas adolescents were given more autonomy in these tasks.

Independent evaluators (IEs) interviewed the youth and at least one parent at baseline (covering the prior 4 months) and again at 4, 8, and 12 months post-randomization. The IEs remained blinded to treatment assignments by instructing patients and family members not to disclose their treatment condition during follow-up interviews, communicating clinical concerns to the principal investigator instead of the therapists, and scheduling follow-up interviews at different times than therapy sessions.

IEs administered the YMRS and CDRS-R at each evaluation and rated the prior 1 or 2 weeks based on a consensus of parent and child reports. Next, the IE rated each week of the 4-month interval using the Adolescent Longitudinal Interval Follow-up Evaluation (A-LIFE) Psychiatric Status Ratings (PSRs), which cover the severity of depression, mania, and hypomania symptoms on point scales (1–6) of severity.^2,26 PSRs are numeric values linked to DSM-IV criteria: ratings of 1 or 2 indicate asymptomatic or mildly symptomatic status with no or minimal impairment; 3 or 4, subsyndromal symptoms with mild-to-moderate impairment; and 5 or 6, fully syndromal states with severe impairment. Weekly PSRs, based on inquiries regarding symptom intensity and change points, were used to clarify the timing of recovery, onsets of new episodes, and weeks in remission (all PSR mood scales ≤ 2). Cross-site interrater reliability for 12 PSR ratings was r = 0.74 (intraclass correlation coefficient, p < .01).

Remission status at baseline was assigned if the youth’s depression or hypomania PSRs were rated ≤ 2 for each of the 4 weeks before randomization. Depressive or hypomanic status at baseline was assigned if the youth had ≥ 2 weeks with PSR depression scores ≥ 5 or ≥ 1 week with hypomania scores ≥ 5. Baseline status was classified as subthreshold if the youth was neither in remission nor in episode in the prior 4 weeks, but had a PSR mood score between 3 and 4 for ≥ 1 week. For the 1-year prospective period, we identified the following: the first 8-week period with PSR depression, mania, and hypomania scores ≤ 2 (recovery); the first occurrence of a ≥ 2-week depressive or ≥ 1-week (hypo)manic episode (PSRs ≥ 5); or the first occurrence of a subthreshold hypomanic episode of at least moderate intensity (PSR score of 4 lasting ≥ 1 week) after a period of recovery.²

All analyses were by intent-to-treat. Time to recovery was calculated as the number of weeks from the date of randomization to the beginning of the first 8-week recovery period. We used Kaplan–Meier models to compare the group survival curves on weeks to full mood recovery (all PSR scales ≤ 2), with separate calculations for time to recovery from baseline depressive and (hypo)manic symptoms. We used Cox Proportional Hazards models to estimate the independent effects of treatment group on time to recovery with each of the following covariates: family EE status (low, high), site, sex, age, baseline clinical status, diagnosis, and baseline medications. Patients who had been remitted for the 4 weeks before randomization were not included in the analyses of time to recovery. Those who did not recover or left the study prematurely were “censored” at the point of their final assessment.

The secondary hypothesis was that youth in FFT-HR would spend proportionately more weeks in remission from all mood symptoms than youth in EC during the 1-year study. This hypothesis was tested in a Poisson regression model that adjusted for total number of weeks in the study, site, age, sex, family EE, and the percentage of weeks with subthreshold or threshold symptoms (PSR depression or hypomania scores ≥ 3) in the 4 months before randomization. Finally, mixed effects regression (growth curve) models with random subject effects were used to explore the effects of treatments on CDRS-R and YMRS scores from baseline to 4, 8, and 12 month follow-ups.

Secondary analyses explored, first, whether comorbid disorders explained the effects of treatment condition on the primary outcome variables once concurrent mood states were covaried, and second, whether the effects of FFT-HR remained robust when covarying medication regimens. With 40 subjects, a one-sided log rank test of survival curves had 78% power (p < .05) to detect a 25% difference in recovery rates over 1 year, assuming 20% attrition. With three follow-up points, mixed effects regression models had 80% power to detect a 0.92 standard deviation (SD) group difference in means at follow-up (adjusted for baseline scores).

The mean age ± SD of the 40 participants was 12.3 ± 2.8 years (Table 1); 17 (42.5%) were female. Participants were most frequently diagnosed with BD-NOS (n = 20, 50%) or MDD (n = 17, 42.5%); three (7.5%) had cyclothymic disorder. Although there was variability in clinical status in the month before randomization, no participant met criteria for a DSM-IV-TR manic or mixed episode. One patient did not have PSR ratings and was dropped from further analyses. Patients assigned to FFT-HR and EC did not differ on any of the demographic, clinical, or illness history variables in Table 1, nor were there any treatment × site interactions on these variables (p > .10 for all).

MINI diagnostic interviews were conducted with 82 first-degree relatives. Lifetime diagnoses of BD were observed in 15 fathers (seven with BD I, eight with BD II), 23 biological mothers (14 with BD I, nine with BD II) (Table 1), and four full siblings of the index child; 40 relatives had no lifetime history of mood disorder. Thirty-five of 40 (87.5%) relatives with BDI or II were living with the high-risk child, as were 34 of 37 (91.9%) relatives with no mood disorder (data on living situation were missing on five relatives).

A total of 42 relatives (12 fathers, 29 mothers, 1 grandmother) of 35 youths met the caregiver inclusion criteria for the FMSS-EE assessment. In five families, EE ratings were unavailable because of inaudible tapes. Of the 42 caregivers, 17 (40.48%) were rated high in EE (three of 12 fathers [25.0%], 13 of 29 mothers [44.8%], and one grandmother) and 25 (59.5%) were rated low in EE. The EE classifications did not differ between relatives with and without BD (χ²[2] = 2.16, p = .34). Using the conventional EE classification criteria (mentioned above), 17 families (48.6%) were rated high-EE, a rating almost identical to that previously observed in families of adolescents with BD I or II (46.2%).¹⁶ Family EE status was unrelated to whether the family had two parents (n = 23) or a single parent (n = 17) (χ²[1] =.27, p = .60), or whether there were two biological parents (n = 21) or a single biological parent living in the home (n = 19; χ²[1] =.03, p = .86).

Treatment completion was equivalent in the two conditions (Figure 1). The mean number of FFT-HR sessions was 12.43 ± 5.75 (range, 0–18), whereas the mean number of EC sessions was 2.26 ± 1.19 (range, 0–5). The number of sessions exceeded protocol specifications when families requested crisis sessions during the year. The groups did not differ on the number of extra sessions (F_1,38 = 0.48, p = .49), or on use of nonprotocol outpatient or school-based services at follow-up (for all, p > .10). Participants in FFT-HR completed an average of 45.0 ± 12.44 weeks of research follow-up, whereas participants in EC completed 44.88 ± 14.7 weeks (t[32] = .03, p = .98; Figure 1).

Of the 32 patients who were not remitted at baseline, 26 (81.25%) recovered (all PSR mood scales ≤ 2 for ≥ 8 weeks) during the 1-year study. Of the six patients who did not recover, five (83.3%) were in EC and one (16.7%) was in FFT-HR. Patients in FFT-HR recovered in an average of 13.0 weeks (SE = 2.93; 25% quartile = 1 week, 95% confidence interval [CI] = 0–8) and in the EC group, 21.25 weeks (SE = 4.22; 25% quartile = 7.5 weeks, 95% CI = 0–14), a significant difference (χ²[1] = 5.24, p = .022; Kaplan–Meier model, Figure 2). The effects of FFT-HR on time to recovery were robust (χ²[1] = 3.96, p = .047; HR = 2.69) when baseline hypomania status (χ²[1] = 16.22, p < .0001; HR = 5.88) and baseline depression status (χ²[1] = 7.52, p = .006; HR = 3.0) were included in a Cox proportional hazards model. The treatment effect was marginally significant (χ²[1] = 2.71, p = .099; HR = 2.02) when the child’s living status (with both biological parents or not) was covaried, although there was no independent effect of living status on time to recovery (χ²[1] = .31, p = .58; HR = 0.79).

In the subgroup of patients with subthreshold (n = 20) or syndromal (n = 9) depressive symptoms at baseline, FFT-HR was associated with faster recovery from depressed mood (mean = 9.2 weeks, SE = 2.98; 25% quartile = 1 week, 95% CI = 0–8) than EC (mean = 21.38 weeks, SE = 4.05; 25% quartile = 11 weeks, 95% CI = 6–18) (χ²[1] = 7.69, p = .006; HR = 2.63). A similar analysis could not be conducted for recovery from hypomania because only 10 participants entered with at least subthreshold hypomania symptoms (PSR scores ≥ 3).

Regardless of treatment assignment, patients in high-EE households took significantly longer to recover from their initial mood symptoms (mean = 21.2 weeks, SE = 3.32) than patients in low-EE households (mean = 11.38 weeks, SE = 3.29) and were more likely to remain symptomatic for the full follow-up (five of 14 [35.7%] versus one of 16 [6.25%]) (χ² = 9.62, p = .002). In a Cox proportional hazards model, separate main effects were observed for treatment group (χ² = 4.14, p = .042; HR = 2.47) and EE (χ² = 7.37, p = .007, HR = 0.29). Although Cox models were underpowered to detect an interaction between treatment group and EE, it is notable that the hazard ratio for time to recovery (favoring FFT-HR) was numerically larger in the high-EE subgroup (HR = 4.59, p = .014; n = 14) than in the low-EE subgroup (HR = 1.46; p = .11; n = 16).

Over the 1-year period, participants spent an average of 25.15 ± 14.75 weeks in remission from all mood symptoms (all PSR scores ≤ 2), 17.15 weeks ± 12.75 in subthreshold states (PSR scores, 3 or 4) and 2.65 ± 4.32 weeks in threshold states (at least one PSR scale ≥ 5). In a Poisson regression model that included site, sex, age, treatment, weeks followed, family EE, and percentage of weeks with subthreshold or threshold symptoms in the 4 months before randomization, youth in FFT-HR had significantly more weeks in full remission (mean = 26.80 weeks, 95% CI = 24.03–29.89) than youth in EC (mean = 19.50 weeks, 95% CI = 17.19– 22.11) (χ²[1] = 15.80, p < .0001). The treatment group × EE interaction on weeks in remission also was significant (χ²[1] = 7.62, p = .006), indicating that participants in high-EE families who received EC had fewer remitted weeks at follow-up (mean = 11.95 weeks, 95% CI = 9.53–14.99) than participants in high-EE families who received FFT-HR (mean = 26.58 weeks, 95% CI 22.95 to 30.78), or youth in low-EE families who received FFT-HR (mean = 27.20 weeks, 95% CI = 23.14–31.96) or EC (mean = 26.18 weeks, 95% CI = 22.67–30.24).

A separate Poisson model with the same predictors and baseline covariates examined treatment in relation to weeks in subthreshold states. Consistent with the remission findings, there was a main effect of treatment group (χ²[1] = 4.49, p = .034) and a treatment × EE interaction (χ²[1] = 4.98, p = .026), indicating more subthreshold weeks among youth in high-EE families who received EC (mean = 26.29 weeks ± 15.82) than among youth in high-EE families who received FFT-HR (mean = 14.88 ± 13.33).

Over the 1-year follow-up, new MDEs after a period of mood recovery were observed in eight of 32 patients (25.0%), of whom five of eight (62.5%) were in EC and three of eight (37.5%) were in FFT-HR (χ²[1] = 0.64, p =.42). Following mood recovery, seven participants developed subsyndromal hypomanic episodes of moderate severity (PSR = 4) that lasted at least 1 week. Of these seven participants, three (42.8%) were in EC and four (57.1%) were in FFT-HR (χ² [1] = 0.06, p =.80). One FFT-HR participant with BD-NOS entered the study in a MDE and, at 5 weeks post-randomization, developed a 5-week fully syndromal hypomanic episode (PSR = 5). This participant was judged by two independent raters to have “converted” to BD II. The remaining participants retained their initial diagnoses at follow-up.

The treatment by time (study visit) interaction on follow-up CDRS-R scores was nonsignificant (F_1,84.1 = 1.52, p = .22). There was, however, a treatment × time interaction on YMRS scores (Figure 3; F_1,98 = 4.55, p = .035), indicating a greater improvement slope for YMRS scores from baseline to 12 months among youth in FFT-HR (mean = −8.73 + 10.92) as compared with youth in EC (mean = −4.38 + 6.10) (Cohen’s d = 0.49). The effects of treatment group on YMRS scores remained significant when baseline YMRS score, pretreatment hypomania status (PSR scale), diagnosis (BD-NOS/cyclothymic or MDD), sex, or age were included in mixed regression models. There was no effect of EE or an interaction between treatment group and EE on CDRS-R or YMRS scores (p > .10 for all).

Comorbid disorders were common in this sample (Table 1). Individual comorbid disorders were not more common among youth with BD-NOS/cyclothymic disorder than among youth with MDD (χ²[4] = 1.78, p = .78). To determine the relation of comorbidity to the primary and secondary outcomes, we created an ordinal variable based on number of comorbid disorders: none (n = 10; 26.3%), one (n = 10, 26.3%), two (n = 6, 15.8%), or three (n = 12, 31.6%) (comorbidity data were missing for two cases). In a Cox proportional hazards model, comorbid disorders were not related to time to full mood recovery (χ²[1] = 0.14, p = .70; HR = 0.93) when baseline hypomania (χ²[1] = 14.98, p < .0001; HR = 5.88) and depression status (χ²[1] = 5.51, p = .019; HR = 2.76) were covaried. FFT-HR retained its association (HR = 2.85) with time to recovery in this model, although at a lower level of significance (χ²[1] = 3.21, p = .07). In a mixed effects regression model, number of comorbid disorders were not associated with YMRS scores over time (F_1,32 = 2.19, p = .15), but the treatment group by time interaction remained significant (F_1,93.8 = 4.40, p = .04), indicating more improvement in YMRS scores in the FFT-HR group. These results were virtually identical when comorbidity was operationalized in terms of subtype (externalizing [attention-deficit disorder (ADD), oppositional defiant disorder (ODD), or conduct disorder, n = 6], internalizing [anxiety only, n = 2], or both internalizing and externalizing [n = 20]).

At study entry, 60% of the participants were taking at least one psychiatric medication (Table 1). Of the 40 participants, 30 (75.0%) received ongoing medication sessions from study psychiatrists (mean 6.27 ± 3.34 visits, range 2–14 visits). Patients in FFT-HR and EC did not differ at randomization on use of mood stabilizers, atypical antipsychotics, psychostimulants, or antidepressants (p > .10 for all) and were equally likely to receive follow-up medication management from a study psychiatrist (χ²[1] = 0.03, p = .86).

In most cases, study psychiatrists did not significantly alter medication regimens over the study year. Of 10 patients in FFT who began on a psychiatric medication, eight remained treated at 1 year (80%); of 14 in EC, 10 remained treated (71.4%). Log-linear models indicated that changes in prescriptions for mood stabilizers, atypical antipsychotics, psychostimulants, or antidepressants over the year were unrelated to treatment group (p > .10 for all). When the number of mood stabilizers and/or atypical antipsychotics at baseline were covaried, youth in FFT-HR still spent more weeks in full remission over the year than youth in EC (χ² [1] = 24.30, p < .0001). An independent effect of number of mood stabilizers/atypicals (χ² [1] = 26.81, p < .0001) indicated fewer weeks in remission among those youth who entered the study with more complex medication regimens.