- Email: [email protected]
The predictive performance of the bipolarity index in a Dutch epidemiological sample manuscript
Journal of Affective Disorders xxx (xxxx) xxx–xxx
Contents lists available at ScienceDirect
Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad
Research paper
The predictive performance of the bipolarity index in a Dutch epidemiological sample manuscript Wendela G. ter Meulena,b, , Stasja Draismaa,b, Aartjan T.F. Beekmana,b, Brenda W.J.H. Penninxa, Ralph W. Kupkaa,b ⁎
a b
Amsterdam UMC, Vrije Universiteit, Psychiatry, Amsterdam Public Health research institute, The Netherlands GGZ inGeest Specialized Mental Health Care, Amsterdam, The Netherlands
ABSTRACT
Background: No instrumnt exists that can predict the incidence of bipolar disorders (BD). The Bipolarity index (BI), originally developed to improve diagnostic confidence for a lifetime diagnosis of BD, may predict incident BD. Aim: To assess the predictive performance of the BI for incident BD in persons with a lifetime depression. Methods: The BI score was composed from different questionnaires and interviews in n = 1857 subjects without BD and with a lifetime unipolar depressive disorder from the Netherlands Study of Depression and Anxiety, a longitudinal cohortstudy. The incidence of DSM-IV defined BD I or II as a criterion diagnosis was established with the Composite International Diagnostic Interview after 2, 4, 6 and 9 years of follow-up. Cox regression analyses calculated whether the BI predicts incident BD during 9-years of follow-up. The area Under the Curve (AUC) was determined. At the optimal cut-off score, sensitivity, specificity, positive, and negative predictive values (PPV and NPV) were calculated. Results: Over the course of 9 years, bipolar conversion occurred in n = 46 subjects (2.5%). Each point increase in BI score significantly predicted incident BD (HR [95%CI]= 1.047[1.018–1.076], p = 0.001). The AUC was 0.61 (95%CI: 0.54–0.68). At the optimal cut-off of 30, sensitivity was 67%, specificity 52%, PPV 3% and NPV 98%. Limitations: Not all items of the BI were fully covered; mean age of the sample of 42. Conclusion: The BI score predicts bipolar conversion over 9 years in those with a lifetime depression. However, given the modest performance metrics, the BI cannot guide clinical decision making yet.
1. Introduction Bipolar disorders (BD) often debut as unipolar depressive disorders (UD), whereas frank hypomanic or manic episodes often appear later in life (Angst, 2006). The yearly rate of such bipolar conversion from an initial diagnosis of UD to BD declines from 3.9% in the first year after study entry with a diagnosis of UD to 0.8% in years 5–10 (Kessing et al., 2017). Conversion rates also depend on treatment setting, and increase to 1 out of 3 patients hospitalized for UD (Angst et al., 2005). Timely identification of those persons with UD who are at risk for bipolar conversion is relevant, as prognosis and outcome are poorer for persons with BD compared to UD (Kessing et al., 2017), and because BD requires different treatment (Kupka et al., 2015) than UD. Several predictors for bipolar conversion have been identified (Kessing et al., 2017), such as family history of BD and age at onset of depression. However, prodromal symptoms and risk factors for bipolar conversion are inconsistent across studies
due to variations in methodology (Kessing et al., 2017; Skjelstad et al., 2010), and their specificity for the prediction of incident BD is low (Skjelstad et al., 2010). Persons presenting with a UD frequently show subthreshold manic symptoms or other features of “bipolarity” to some extent (Angst et al., 2011). Inspired by this clinical experience and driven by an intent to assign a “bipolar profile” to the individual treatment-seeking patient with a probable mood disorder, experienced investigators in the field of BD created the bipolarity index (BI) in 2004 (Sachs, 2004). The BI is a 0–100 continuous scale that covers five illness dimensions with a maximum of 20 points per domain: I. signs and symptoms; II. age of onset; III. course of illness; IV. response to treatment, and V. family history. “Classic” BD according to the authors of the BI would be characterized by: I. at least one euphoric manic episode; II. early age of onset; III. recurrent and fully remitting illness course; IV. positive response to a mood stabilizer; and V. having a first-degree family member with BD.
Corresponding author at: W.G. Wendela ter Meulen, GGZ inGeest Specialized Mental Health Care, Research and Innovation, Oldenaller 1, 1081 HJ Amsterdam, The Netherlands. E-mail addresses: [email protected] (W.G. ter Meulen), [email protected] (S. Draisma), [email protected] (A.T.F. Beekman), [email protected] (B.W.J.H. Penninx), [email protected] (R.W. Kupka). ⁎
https://doi.org/10.1016/j.jad.2019.10.055 Received 27 May 2019; Received in revised form 10 September 2019; Accepted 28 October 2019 Available online 30 October 2019 0165-0327/ © 2019 Elsevier B.V. All rights reserved.
Please cite this article as: Wendela G. ter Meulen, et al., Journal of Affective Disorders, https://doi.org/10.1016/j.jad.2019.10.055
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
The five dimensions of the BI highlight that the conceptualization of “bipolarity” that underlies the BI goes beyond the mere symptomatic assessment of lifetime affective symptoms, as the DSM-IV categorization requires. Instead, the authors of the BI added additional state and trait variables, based on their clinical experience and on the earlier theories of Kraepelin. By doing so, the BI represents a broader view of mood disorders that is now termed “bipolarity”, that represents a more conservative view of the classic conceptualization of mood disorders, which Kraepelin considered to include risk for shifts to elevated mood states. Hence, it is well possible that the BI estimates a latent trait of bipolarity that may become apparent as bipolar conversion at a later stage in those with a lifetime UD. Nevertheless, no studies to date have investigated the predictive performance of the BI for incident BD in persons with a lifetime UD, an event also known as bipolar conversion. Studies into the concurrent validity of the BI against a lifetime DSM-IV classification of BD, although potentially flawed by observer bias (Holman et al., 2015) as the diagnosis of BD and BI answers are often provided by the same clinician, found good to excellent metrics (Aiken et al., 2015; Ma et al., 2016; Mosolov et al., 2014). Although the BI was not designed for the prediction of BD, investigation into the predictive performance of the BI for bipolar conversion could be of particular interest for clinical practice in those with a lifetime UD who present with risk factors of BD conversion and/or subthreshold bipolar symptoms (Angst et al., 2011). Such research may also substantiate the validity of the BI as a measure of “bipolarity”. Therefore, in this paper, we will report on the predictive performance of the BI for incident BD by using data from the Netherlands Study of Depression and Anxiety disorders (NESDA) sample. The longitudinal design of NESDA with 4 measurements in 9 years of follow-up allows prospective investigation into incident BD in those with a lifetime UD. NESDA has a high prevalence of affective and anxiety disorders and comorbid conditions such as substance use disorders, which means a high density of bipolar risk factors (Angst et al., 2005; Coryell et al., 1995; Østergaard et al., 2014) that are included in the BI. The predictive validity of the BI will be estimated for the incidence of BD type I or II after 2, 4, 6 and 9 years of follow-up as a criterion diagnosis in those with a lifetime UD at baseline who were recruited from the community, primary care, and outpatient mental health care services. Analyses will be repeated for the subsample of those aged < 30 years at baseline, as most bipolar conversion occurs before this age (Angst et al., 2005). BD, lifetime UD and comorbid conditions such as anxiety disorders will be assessed according to DSMIV criteria with the Composite International Diagnostic Interview (CIDI) (World Health Organization, 1997). We hypothesize that a higher BI score at baseline predicts an increased risk of bipolar conversion at follow-up assessments. Finally, the individual contribution of each of the five dimensions of the BI to the incidence of BD in this sample will be explored.
elsewhere (Penninx et al., 2008). In short, the baseline assessment (T0) consisted of an extensive face-to-face interview by trained interviewers, including a diagnostic psychiatric interview with the CIDI (World Health Organization, 1997), and paper-and-pencil questionnaires. Follow-up assessments took place after 2 (T2), 4 (T4), 6 (T6) and 9 (T9) years. Since NESDA was originally not aimed at BD, this diagnosis was an exclusion criterion at sample selection, well before baseline assessments took place. In the first follow-up assessments that were conducted after two years, the first CIDI assessment of lifetime BD took place. At that T2 assessment, lifetime BD I or II had been retrospectively reclassified with the CIDI in n = 116 subjects with UD at T0. For the present analyses, we selected all n = 1857 (1973–116) respondents with lifetime UD and without lifetime BD at baseline. NESDA was approved centrally by the Ethical Review Board of the VU University Medical Centre and subsequently by the local review boards of each participating center, and all participants signed written informed consent. 2.2. Bipolarity Index (BI) scoring The BI is a clinician-rated instrument that covers five clinical domains based on the original work of Robins and Guze to validate psychiatric disorders (Robins, E., Guze, 1970): I. signs and symptoms; II. age of onset; III. course of illness; IV. response to treatment, and V. family history. Per domain, several items can be assessed as either present or absent. These items are considered common clinical features with an ordinal score ranging from 0–20 (with possible values of 0, 2, 5, 10 15, and 20). A maximum score of 20 per domain indicates the presence of an item within that domain most characteristic of ‘classic’ BD type I. The total BI score is the sum of the highest rated items on each of the five domains, thus ranging from 0–100. For the current study, the BI score at T0 was constructed from data that had been retrieved with several questionnaires and interviews during the NESDA assessments. Before assessing the items of the BI, scoring decisions were discussed by the authors [RK, WtM]. Table 1 shows the complete BI interview as originally described by Sachs et al. (Sachs, 2004), the NESDA research instruments per item that were chosen for scoring, relevant scoring decisions and their pertaining references, and descriptives. Among these items, the presence of (hypo) manic episodes at T0 in domain I and III of the BI had been retrospectively assessed with the CIDI-BD interview at T2. Furthermore, no validated epidemiological instrument was available to assess the precise effects of medication use (BI domain IV). In NESDA, medication use had been assessed through a medication interview that covered current (T0) and retrospective medication use over a period of 5 years before T0 (Penninx et al., 2008), documented according to the Anatomical Therapeutic Chemical (ATC) classification system of the World Health Organizations (WHO). Current (T0) and previous (covering the 5 years before T0) use of antidepressants (ATC code N06A), antipsychotics (ATC code N05A, including lithium N05AN) and mood stabilizers (ATC code N03A) was selected. Finally, the BI total score at T0 was calculated by summing the subscores of the five domains of the BI. To establish the subscore on each domain, we assessed the highest scored item on that domain in accordance with the original BI procedure. For example, when a subject rated positive in domain III on the items “comorbid personality disorder” (rated 5 points) and “comorbid substance use disorder” (rated 10 points), the score of 10 prevailed. We chose a conservative approach for assigning a BI-score at baseline, as this most closely resembles the routine clinical practice of the BI interview. Conservative means that if no or ambiguous information was available on a certain item, a score of 0 was assigned. The interviewers who had performed the psychiatric interviews at T0 were unaware that the data would be used for BI rating later on.
2. Method 2.1. Study sample Data were retrieved from the Netherlands Study of Depression and Anxiety (NESDA) (Penninx et al., 2008), a multisite naturalistic cohort study designed to investigate the course and consequences of the most common depressive and anxiety disorders in adults (18–65 years). To represent various settings and stages of psychopathology, persons were recruited from the community (19%), primary care (54%) and outpatient mental health care services (27%). Subjects in NESDA at baseline (n = 2981) include n = 1973 (66%) subjects with a lifetime DSMIV unipolar depressive disorder, defined as major depressive disorder and/or dysthymia as assessed with the Composite International Diagnostic Interview (CIDI) version 2.1 (World Health Organization, 1997). Aims, design and method of recruitment were extensively described 2
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
Table 1 Original BI interview; scoring instruments to rate the BI in this sample; relevant scoring decisions; and descriptives: see below. Bipolarity Index(Sachs, 2004) For each of the items below circle the item characteristic of the patient: Most convincing characteristic: 20 Convincing characteristic of Bipolar Disorder: 15 Known associated feature of Bipolar Disorder: 10 Non-specific feature suggestive of Bipolar Disorder: 5 Feature with possible relationship to Bipolar Disorder: 2 No evidence of Bipolar Disorder: 0 Score & BI domain Instrument Domain I: Episode characteristics 20 1. Documented acute mania or mixed episode with prominent euphoria, grandiosity or expansiveness and no significant general medical or known secondary etiology 15 Clear-cut acute mixed episode or dysphoric, or irritable mania with no significant general medical or known secondary etiology 10 2. Clear-cut hypomania with no significant general medical or known secondary etiology Clear-cut cyclothymia with no significant general medical or known secondary etiology Clear cut mania secondary to antidepressant use 5
Missing% 0%
Mania at/before T0 according to DSM-IV criteria as established with the CIDI 2.1
n = 0 (0.0%)
CIDI 2.1 BD (World Health Organization, 1997) at T2
0%
n = 1 (0.1%)
CIDI 2.1 BD (World Health Organization, 1997) at T2 *
0%
Mixed episode at/before T0 according to DSM-IV criteria as established with the CIDI 2.1 Hypomania at/before T0 according to DSMIV criteria as established with the CIDI 2.1 Cyclothymia not assessed with the CIDI 2.1
*
CIDI 2.1 BD (World Health Organization, 1997) at T2
0%
Clear cut hypomania secondary to antidepressant use Episodes with characteristics signs of hypomania but symptoms, duration or intensity are subthreshold for hypomania or cyclothymia
CIDI 2.1 BD (World Health Organization, 1997) at T2 Mood Disorder Questionnaire (MDQ) (Hirschfeld et al., 2000) at T0
0%
A single major depressive episode with psychotic or atypical features (Atypical 2 of 3: hypersomnia, hyperphagia, leaden paralysis of limbs Any postpartum depression (PPD)
CIDI 2.1 MDD (World Health Organization, 1997) at T0
0%
Edinburgh Postnatal Depression Scale (Cox et al., 1987) at T4
10.6%b
Any recurrent typical unipolar major CIDI 2.1 MDD (World Health depressive disorder (MDD) Organization, 1997) at T0 History of any kind of psychotic episode(i.e. * presence of delusions, hallucinations, ideas of reference, or magical thinking) 0 No history of significant mood elevation, recurrent depression, or psychosis Domain II: Age of onset (1st affective episode ⁄ syndrome) 20 First episode age 15–19 years CIDI 2.1 MDD (World Health 15 First episode age < 15 or 20–30 Organization, 1997) at T0 10 First episode age 30–45 CIDI 2.1 BD (World Health 5 First episode > 45 Organization, 1997) at T2 0 No history of affective illness (no episodes, cyclothymia, dysthymia or BP NOS) Domain III: Course of illness⁄associated features 20 Recurrent distinct manic episodes separated by Retrospective lifechart interview periods of full recovery (Lyketsos et al., 1994) that covered the last 3 years before T0 Recurrent distinct manic episodes with incomplete interepisode recovery Recurrent distinct hypomanic episodes with full interepisode recovery 10
Comorbid substance abuse
Psychotic features only during acute mood episodes Incarceration or repeated legal offences related to manic behaviour (e.g. shoplifting, reckless driving, bankruptcy)
Descriptives
CIDI 2.1 BD (World Health Organization, 1997) at T2
2
15
Relevant scoring decisions
Retrospective lifechart interview (Lyketsos et al., 1994) that covered the last 3 years before T0 Retrospective lifechart interview (Lyketsos et al., 1994) that covered the last 3 years before T0 Fagestrom Test of Nicotine Dependence (FTND) (Fagerström, 1978) CIDI 2.1 alcohol use disorders (AUD) (World Health Organization, 1997) CIDI 2.1 BD (World Health Organization, 1997) at T2 Retrospective lifechart interview Retrospective lifechart interview (Lyketsos et al., 1994)
3.0%
0% *
0 (0)%
Positive if subject reports any mania or mixed episode on CIDI secondary to antidepressant use at/before T0. Positive if subject reports any hypomania secondary to antidepressant use at/before T0. Positive if in A-section either first question (good, hyper) and ≥3 other questions are answered “yes”, or second question (irritable) and ≥4 other questions are answered “yes”; and B-section “yes”; and C-section shows minor problems Positive if any MDD and any episode characterized by at least 2 atypical features
12.9% 12.9% 0%
0% 0%
* n = 3 (0.1%) n = 0 (0.0%) n = 246 (13.2%)
n = 278 (15.0%)
Positive if score >12 on EPDS (Meltzer-Brody et al., 2013) and current or lifetime MDD and/or dysthymia before T0 Recurrent MDD according to DSM-IV criteria as established with the CIDI 2.1 Primary psychotic disorder was exclusion criterion in NESDA at baseline.
n = 934 (50.3%)
Those without score 2–20 on domain I
n = 1550 (54.2%)
Question in CIDI on age of onset first affective episode (episode of mania, hypomania, MDD, or dysthymia)
n = 347 n = 721 n = 468 n = 208 n = 113
Those without score 5–20 on domain II 12.9%
n = 17 (0.9%)
Any (either medication-induced or not) manic or mixed episode (CIDI) and ≥15% of time asymptomatic (Verduijn et al., 2015)a Any manic or mixed episode (CIDI) and <15% of time asymptomatic (Verduijn et al., 2015)a Any hypomanic episode (CIDI) and ≥15% of time asymptomatic (Verduijn et al., 2015)a Positive if score ≥5 on FTND, corresponding with high nicotine dependence (Fagerström, 1978), and/or if any AUD in the past 6 months Positive if any psychotic symptoms were present in CIDI-BD Screening on the following words: legal, offences, shoplifting, reckless driving, bankruptcy
n = 250 (13.5%)
(18.7%) (38.8%) (25.2%) (11.2%) (6.1%)
n = 2 (0.1%)
n = 0 (0.0%) n = 14 (0.8%) FTND: n = 138 (7.4%) AUD: n = 130 (7.0%) n = 5 (0.3%) n = 0 (0%)
(continued on next page)
3
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
Table 1 (continued) 5
Recurrent unipolar MDD with three or more major depressive episodes Recurrent distinct hypomanic episodes without full interepisode recovery
Comorbid borderline personality
Personality Diagnostic Questionnaire-4 (PDQ-4)
0%
Comorbid anxiety or eating disorders (e.g. OCD, panic disorder, bulimia) History of ADHD in childhood and periods of above average scholastic or social function
CIDI 2.1 anxiety disorders (AD) at T0 and T2 Conners Adult ADHD Rating Scale (CAARS) (Conners et al., 1999) at T0
0%
Gambling, risky investment, overspending, or sexual indiscretions have (or would if not concealed) pose a problem for patient, friends, or family Behavioral evidence of perimenstrual exacerbation of mood symptoms Baseline hyperthymic personality (when not manic or depressed)
*
*
Positive if MDD and three or more major depressive episodes Any hypomanic episode (CIDI) and <15% of time asymptomatic (Verduijn et al., 2015)a If ≥2 antidepressant medications (ATC code N06A) or mood stabilizers (ATC code N03A) were used <50% of the time that was prescribed in the past 5 years (consensus RK / WtM). Positive if ≥4 items of borderline personality disorder present. The PDQ-4 has high sensitivity and moderate specificity for borderline personality disorder. A score of >4 is highly suggestive of BPD (Sansone et al., 2008) Positive if any recent (<6 month recency) anxiety disorder Positive if current ADHD symptoms on the CAARS and a positive score on childhood or early-adolescent ADHD indicators, as described elsewhere (Bron et al., 2016) *
*
*
*
NEO-Five Factor Inventory (NFFI) (Hoekstra et al., 1996)
1.4%
Married three or more times (including remarriage to same individual) In two or more years has started a new job and changed jobs after less than year Has more than two advanced degrees 0 None of the above Domain IV: Response to treatment 20 Full recovery within 4 weeks of therapeutic treatment with mood stabilizing medication 15 Full recovery within 12 weeks of therapeutic treatment with mood stabilizing medication or relapse within 12 weeks of discontinuing treatment
*
*
Positive if extraversion score was >1.5 SD above the mean extraversion score (consensus RK / WtM). *
*
*
*
*
*
* Those without score 2–20 on domain III
*
*
Retrospective medication interview T0
0%
CIDI 2.1 BD (World Health Organization, 1997) at T2
0%
Retrospective medication interview T0 CIDI 2.1 BD (World Health Organization, 1997) at T2
0%
Time of response to mood stabilizers was not explicitly asked. Positive if previous (covering the 2 years before T2) use of antipsychotics (ATC code N05A, including lithium N05AN) and/or mood stabilizers (ATC code N03A); and subject reports response as reason why medication was ended (consensus RK / WtM). Positive if subject reports any mania or mixed episode on CIDI secondary to antidepressant use. Positive if previous (covering the 2 years before T0) use of antidepressants (ATC code N06A) was ended and subject reports manic or mixed symptoms (free text) as reason why medication was ended (consensus RK / WtM). Positive if subject reports any hypomanic episode at/before T0 on CIDI secondary to antidepressant use.
Recurrent medication non-compliance
2
10
5
2 0
Affective switch to mania (pure or mixed) within 12 weeks of starting a new antidepressant or increasing dose Worsening dysphoria or mixed symptoms during antidepressant treatment subthreshold for mania
CIDI 2.1 MDD (World Health Organization, 1997) at T0 Retrospective lifechart interview (Lyketsos et al., 1994) that covered the last 3 years before T0 Current and retrospective medication interview that covered the previous 5 years before T0
0% 0% 0%
0%
n = 638 (34.4%) n = 3 (0.2%) n = 29 (1.6%)
n = 319 (17.2%)
n = 982 (52.9%) n = 149 (8.0%)
n = 53 (2.9%)
Partial response to one or two mood stabilizers within 12 weeks of therapeutic treatment Antidepressant-induced new or worsening rapid-cycling course
*
*
Retrospective medication interview T0
0%
Treatment resistance: lack of response to complete trials of three or more antidepressants Affective switch to mania or hypomania with antidepressant withdrawal Immediate near complete response to antidepressants (in 1 week or less) None of the above or no treatment
Current and retrospective medication interview at T0
0%
*
*
Positive if previous (covering the 2 years before T2) use of antidepressants (ATC code N06A) was ended and subject reports rapid cycling (free text) as reason why medication was ended (consensus RK / WtM). Positive if ≥3 antidepressant medications (ATC code N06A) were used on a daily base in the past 5 years *
*
*
* Those without score 2–20 on domain IV
N = 428 (23.6%)
n = 0 (0.0%)
n = 2 (0.1%) n = 1 (0.1%)
n = 0 (0.0%)
n = 0 (0%)
n = 0 (0.0%)
n = 1854 (99.8%)
(continued on next page)
4
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
Table 1 (continued) Domain V: Family history 20 At least one first-degree relative with documented bipolar illness 10 First-degree relative with documented recurrent unipolar MDD or schizoaffective disorder 10 Any relative with documented bipolar illness
5 2
0
Family tree interview that covers 1st and 2nd degree family members
0% 0% 0%
Any relative with documented recurrent unipolar MDD and behavioural evidence suggesting bipolar illness First-degree relative with documented substance abuse, or any relative with possible bipolar illness First degree relative with possible recurrent unipolar MDD First degree relative with diagnosed related illness: anxiety disorders, eating disorders, ADD None of the above or no family psychiatric illness
0% 0% 0% 0%
Positive if subject reports any BD in first degree family member Positive if subject reports any recurrent unipolar MDD or schizoaffective disorder in first degree family member Positive if subject reports any BD in second degree family member Positive if subject reports any recurrent unipolar MDD or schizoaffective disorder in second degree family member Positive if subject reports any substance use disorder in first degree family member Possible diagnosis was not questioned as such; therefore, no scoring possible Positive if subject reports any anxiety disorder or ADD in first degree family member Those without score 2–20 on domain V
n = 56 (3.0%) n = 45 (2.4%)
n = 1469 (79.1%) n = 832 (44.8%)
n = 1363 (73.4%) n = 152 (8.2%)
* = no scoring possible as no instrument was available to score this variable. a. = The level of recovery in domain III was based on the percentage of asymptomatic time as established by a retrospective life chart interview (Lyketsos et al., 1994), and is in line with previous reports from our group (Verduijn et al., 2015). b. Missings due to attrition.
2.3. Assessment of unipolar and bipolar disorder (BD)
3. Results
At baseline, the presence of a unipolar disorder (UD) was defined as either a lifetime major depressive disorder and/or lifetime dysthymia, as assessed with CIDI 2.1. Incident BD I or II after 2, 4, 6 and 9 years as established with the CIDI 2.1 (World Health Organization, 1997) was the criterion diagnosis against which the predictive performance of the BI was tested. The CIDI 2.1 is a structured lay-administered interview that is based on DSM-IV criteria (First et al., 1997). A new BD diagnosis was considered present if a subject fulfilled DSM-IV criteria (First et al., 1997) for BD I or II for the first time.
3.1. Sample Table 2 shows the sample of n = 1857 subjects with a lifetime UD at baseline. As would be expected, the majority of UD subjects had a comorbid anxiety disorder (Hirschefeld, 2001) and almost a third had experienced both lifetime MDD and lifetime dysthymia. Other sample characteristics are represented in Table 1, which includes a quantitative description of all the items of the BI. 3.2. Predictive validity
2.4. Statistical analyses
In the total of 9 years of follow-up in the whole sample (n = 1857), incident BD occurred in n = 49 subjects (2.6%): 4 at T2 (0.2%), 20 at T4 (1.1%), 16 at T6 (0.9%), and 9 at T9 (0.5%). Cox regression analyses revealed that each point increase in BI score at baseline was significantly associated with a higher hazard of incident BD over 9 years (HR[95%CI] = 1.05[1.02–1.08], p = 0.001). The AUC was 0.60 (95%CI: 0.52–0.81), and the cut-off of the BI score with the highest
Cox regression analysis was performed to determine the association between the BI score at baseline and the incidence of BD over 9-years of follow-up, with measurements at 4 different time points: after 2, 4, 6 and 9 years. Time at risk was measured from T0 until the moment either the person had incident BD, or was censored at the last recorded followup. The BI cut-off score that coincided with a maximum sum of sensitivity and specificity for incident BD during 9 years of follow-up was calculated by the Youden Index (Perkins and Schisterman, 2006) in a logistic regression model. Based on this BI cut-off score, sensitivity and specificity, positive and negative predictive values (PPV and NPV), and their corresponding 95% CI's were calculated by cross-tabulation. In order to estimate the contribution of each separate BI domain to the incidence of BD, a multivariate logistic regression model was created with stepwise regression analyses. Multicollinearity between the domainscores was checked by Variance Inflation Factors (VIF) (Menard, 2002). For the multivariate stepwise logistic regression model, the incidence of BD was defined as the incidence of BD at any time point between baseline and 9 years. The goodness of fit of the model for each added dimension was expressed as the change to the Nagelkerke correlation coefficient (R2). All analyses were conducted using IBM SPSS Statistics version 22.0 (SPSS Inc., Chicago, USA) (IBM Corp. Released, 2011).
Table 2 Baseline sociodemographic and clinical characteristics of the sample of n = 1857 subjects with a lifetime UD and without BD at baseline. Characteristics Sociodemographic Gender (% female) Age at intake (mean in years, SD) University education completed (%) Unemployed (%) Clinical Mean BI score (SD) Lifetime anxiety disorders (%) Lifetime MDD (%) Lifetime dysthymia (%) Lifetime MDD and dysthymia (%) Current (6 month recency) MDD (%) Current (6 month recency) anxiety disorder (%)
Lifetime UD (n = 1857) 68.8 42.1 (12.4) 9.7 17.6 28.5 (9.1) 71.2 97.4 32.9 30.3 46.5 55.5
Abbreviations: MDD (major depressive disorder); SD (standard deviation); UD (unipolar depressive disorder). 5
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
unipolar depressive disorder (UD), an event also known as bipolar conversion. The BI significantly and positively predicted bipolar conversion after 2, 4, 6, and 9 years of follow-up in those with lifetime UD who were recruited from the community, primary care and outpatient mental health services. This positive association was found in the whole sample as well as in the subsample of those aged <30 years at baseline. In this subsample, the HR per point increase of the BI was somewhat higher (HR 1.10 (95%CI 1.02–1.18)) compared to the HR in the whole sample (HR 1.05 (95%CI 1.02–1.08)). In the whole sample, the variance in two domains of the BI significantly contributed to the incidence of BD (domain III and IV), although their relative contribution to the incidence of BD was low: they accounted for only 6% of the prediction of bipolar conversion. A closer look into the performance metrics of the BI score reveals that the contribution of the BI to the prediction of bipolar conversion is modest: Taking into account that an AUC of 1.00 represents a perfect scale, and 0.50 a useless scale (Hanley and McNeil, 1982), the AUC of 0.60 in the whole sample and of 0.57 in the subsample of those aged <30 years at baseline indicates that the BI should be interpreted as a relatively poor test to separate those with from those without future incident BD. At the optimal cut-off of 30 in the whole sample, sensitivity of the BI was fair at 74%. This means, that for those with a score below 30, one can fairly – but not fully – rule out the incidence of BD over the next 9 years. At the optimal cut-off of 30, specificity was deficient with 45%, which means that at this cut-off the number of false positives and true negatives is almost equal. The high number of false positives also translates into a poor PPV of 4%, whereas we found a NPV of 98%. This high NPV can be explained by the low incidence of BD, and hence the large a priori likelihood that there would not be a bipolar conversion. The poor PPV means that the percentage of subjects with a BI of ≥30 who actually convert to BD is very low, and that the likelihood of a type-I error is very high. Our study also compared the psychometric performance of the whole sample (mean age of 42 years) with those aged < 30 years at baseline, as most bipolar conversion occurs in this younger age group (Angst et al., 2005). This comparison showed that at the same BI cut-off of for example 30, sensitivity is higher in those aged < 30 years than in the whole sample, at the expense of specificity (see Table 3). The higher the sensitivity, the better this advantage may weight against the effort and investment of an elaborate instrument such as the BI. However, no firm conclusions can be drawn yet given the low number of incident cases (n = 12) in this subsample. The poor discriminatory ability of the BI to separate those with from those without incident BD that we found in the whole sample as well in those aged <30 years at baseline, is related to the high density of prodromal symptoms and risk factors of BD in our sample, which challenges the instrument to separate those at risk for bipolar conversion to those who are not at risk. Still, a sample high in prodromal symptoms and risk factors is a valid sample to investigate the BI, as the BI is an elaborate and time-consuming instrument that would only be applied in such samples “at risk”, where the basic idea of performing the BI as a predictive test would be to yield a high sensitivity. High-risk subjects with a lifetime UD and unresponsive to conventional antidepressant treatment who have a high BI score, could potentially benefit from mood stabilizers for treating depression. In the true positives, such treatments aimed at bipolar spectrum disorders might hypothetically even delay bipolar conversion. By contrast, in samples with a lower a priori likelihood of bipolar conversion, the high proportion of false positives would be incompatible with clinical decision making. Nevertheless, based on our study the BI cannot be recommended yet as tool to predict bipolar conversion in clinical practice, or to guide treatment decisions. The modest performance of the BI for the prediction of bipolar conversion in our study should be interpreted with some caution. First, the incidence of BD in the NESDA sample is relatively low, compared to the incidence in other samples at risk, such as hospitalized subjects or
Table 3 Psychometric performance of the BI at different cut-offs. Cut-offs are displayed with intervals of 3 points around the optimal cut-off of 30 (whole sample, n = 1857). In the subsample of those aged <30 years at baseline (n = 377), psychometric values are only presented at the optimal cut-off of 39 and again at the cut-off of 30. Cut-off
Whole sample Sens
Age <30 years at baseline Spec PPV NPV Sens
18 21 24 27 30 33 36 39 42
91.8 89.8 78.8 75.5 73.5* 40.8 24.5 18.4 10.2
13.1 20.2 24.8 37.1 44.6* 69.2 82.4 88.0 95.1
2.8 3.0 3.1 3.1 3.5* 3.8 3.6 4.0 5.4
98.3 98.7 98.7 98.2 98.4* 97.7 97.6 97.5 97.5
Spec
PPV
NPV
83.3
22.5
4.1
97.6
41.7*
80.0*
6.4*
97.2*
Abbreviations: Sens = sensitivity (%); spec = specificity (%); PPV = positive predictive value (%); NPV = negative predictive value (%). ⁎ Values around the optimal cut-off, i.e. with the highest Youden Index, are displayed with an asterix.
Youden Index (1,18) was 30. This cut-off of 30 corresponded with a sensitivity of 73% (36/49), specificity of 45% (806/1808), positive predictive value (PPV) of 3% (36/1038) and negative predictive value of 98% (806/819). In the subsample of those aged < 30 years at baseline (n = 377), incident BD occurred in n = 12 subjects (3.1%). Cox regression analyses showed that each point increase in BI score at baseline was significantly associated with a higher hazard of incident BD over 9 years (HR[95%CI] = 1.10[1.02–1.18], p = 0.02). The AUC was 0.57 (95%CI: 0.39–0.76), and the cut-off of the BI score with the highest Youden Index (1,22) was 39. The corresponding performance metrics are displayed in Table 3. 3.3. Performance of the BI at different cut-offs Table 3 shows the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) at different cut-offs of the BI in the whole sample. The cut-offs are randomly chosen with intervals of 3 points around the optimal cut-off of 30. In the subsample of those aged <30 years at baseline, psychometric values in Table 3 are only presented at the optimal cut-off of 39 and again at the cut-off of 30. 3.4. Contribution of the five dimensions of the BI VIF values indicated that multicollinearity between the five dimensions was absent. In the multivariate logistic regression model, the BI dimensions were entered as independent variables and incident BD at any timepoint as dependent variable. Those BI domainscores that were skewed, were first log-transformed before they were entered as independent variables in the model. These multivariate analyses showed that in the whole sample domain III (β = =3.00, p = 0.034) and domain IV (β = =1.52, p < 0.001) were significantly associated with the outcome variable. Next, stepwise logistic regression with these two variables in the model revealed that the goodness of fit of the model expressed as the Nagelkerke R2with only domain III was 0.029; and with both domain III and IV the R2 was 0.061 (difference: 0.032). We considered the number of incident cases (n = 12) in the subsample of those aged < 30 years at baseline too small to calculate the contribution of the five dimensions. 4. Discussion The main goal of our study was to report on the predictive performance of the bipolarity index (BI) for the incidence of DSM-IV defined bipolar disorders (BD) as a criterion diagnosis in those with a lifetime 6
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
those in outpatient mental health care services specialized in mood disorders. This low incidence is probably due to the high mean age at baseline (42 ± 12 years) and to the fact that a BD diagnosis was an exclusion criterion before baseline assessments. Any clinical diagnosis with a low incidence is generally very difficult to predict by an instrument. Still, in our study we did find a significant effect, which is relevant given the high impact and clinical implications of a diagnosis of BD. Therefore, the BI appears a suitable instrument to evaluate risk of bipolar conversion, when incorporated in a broader risk assessment including other known risk factors for bipolarity such as environmental factors (McGuffin et al., 2003). For example, the impact of stressful life events such as childhood physical abuse on the risk of developing BD is well investigated (Sugaya et al., 2012) but is not included in the BI. Another explanation for the modest performance of the BI and in particular the poor PPV, is that a DSM classification of BD - a pure description of symptomatic course - may not be the correct gold standard for the diagnosis of BD. In other words, the construct of bipolarity may be better represented through the BI itself than by DSM classification criteria. This tentative explanation raises the question what “bipolarity” is as captured by the BI. Although the pathogenesis of BD is not known, family and twin studies show that inherited factors are considerably involved (Barnett and Smoller, 2009; McGuffin et al., 2003), which in the BI is represented by domain V (family history). Various items of domain III and the lower scored items in domain V reflect that inherited factors are pleiotropic (Lichtenstein et al., 2009), and that innate susceptibility confers a risk for other types of psychopathology too (Serretti and Fabbri, 2013) such as UD, anxiety and ADHD. The model of BD as a dopamine dysregulation syndrome describes an increased dopaminergic drive in mania and the opposed in depression (Berk et al., 2007). This model would also suggests a mechanism for antidepressant-induced manic symptoms, and for the particular response of BD to mood stabilizers that regulate dopaminergic transmission. In the BI, adverse reactions to regular antidepressant therapies and response to mood stabilizers are captured in domain I and IV. A few limitations should be considered. First, since the BI was not part of the original NESDA set of questionnaires, not all items of the BI could be completely addressed with the NESDA data, which resulted in a conservative scoring. A better psychometric performance could be expected when all items of the BI had been fully covered. Another limitation is the inevitable rate of attrition in longitudinal studies including NESDA (Lamers et al., 2012). A post-hoc comparison showed no significant difference between the mean BI score at baseline of those who during any of the follow-up assessments had attrited (28.3) versus those who did not (28.5), which suggests that those at a higher risk of bipolar conversion did not have a higher rate of attrition though. Finally, the high mean age at baseline (42 ± 12 years) means that all subjects but those aged <30 years (the youngest quartile) would be considered beyond the period of substantial risk for conversion (Angst et al., 2005). The higher mean age at baseline as well as the exclusion of lifetime BD or psychosis at baseline limits the representativeness of the sample, and means that the whole sample may be enriched for low risk subjects. In a sample with more subjects at risk, the state-related items (I and III) may be related to incidence of BD over time. Exclusion of subjects with BD however may have flattened the BI to essentially reflect predominantly the trait-related domains (II, IV and V). Another limitation is that the psychiatric diagnoses were assessed by the CIDI, which is a fully structured diagnostic interview administered by trained interviewers without clinical experience. The CIDI however might not assess bipolar spectrum disorders as accurately as a semistructured interview administered by clinicians – such as the Structured Clinical Interview for DSM (SCID) - does (Kessler et al., 2006; Regeer et al., 2004). This disadvantage should be weighed against the major advantage of the CIDI for large cohort studies that interviewers can relatively easily be trained to employ the instrument. Advantage of this study was that the BI rating was based on data from elaborate questionnaires and interviews from a large
epidemiological sample from the Netherlands. This allowed to minimize the risk of observer bias, and to provide transparent and reproducible data. Furthermore, NESDA is unique as a large sample with affective and anxiety disorders – a high risk population for bipolar conversion. The high density of risk factors for bipolar conversion considerably challenges the discriminatory ability of the BI. Altogether, some recommendations for future studies can be made. First, because the incidence of BD on average occurs at early adulthood, we suggest to perform future studies into the predictive performance of the BI in younger samples than in our study. Second, adaptations to the ranking or rating of the BI might increase the psychometric performance of the BI to predict incident BD. Third, to assess the gold standard diagnosis of BD, we suggest to employ diagnostic instruments with high concordance rates for BD, for example the clinician-rated SCID. 5. Conclusion The BI significantly and positively predicts the incidence of BD over 9 years of follow-up in a sample of subjects with a lifetime depressive disorder, recruited from the community, primary care and outpatient mental health care services. This is relevant, as to date no instrument exists that predicts BD, whereas that could be of clinical benefit for those persons at risk for conversion from lifetime UD to BD. Given the modest performance metrics of the BI in this study, the BI cannot be recommended yet as an instrument to predict bipolar conversion in routine clinical practice, or to guide clinical decision making. However, the BI may be a helpful tool to evaluate the risk of bipolar conversion, when integrated with other means of risk assessment. To further optimize the BI for the prediction of BD, we recommend research in younger samples at risk, and further development of the items of the BI. Author statement Wendela G. ter Meulen, MD1, 2 [email protected] chaired and executed the concept, design, analyses, drafting, and writing of the manuscript. Stasja Draisma, PhD1 [email protected] participated in the concept, design, analyses, interpretation and writing of the manuscript; and thoroughly participated in drafting and revision of the manuscript. Aartjan T.F. Beekman, MD, PhD1 [email protected] participated in the concept, design (including statistical design), interpretation and writing of the manuscript; and thoroughly participated in drafting and revision of the manuscript. Brenda W.J.H. Penninx, MD, PhD1 [email protected] intensively participated in the data acquisition; and had a major role in drafting and revision of the manuscript. Ralph W. Kupka, MD, PhD1 [email protected] participated in the concept, design (including methodological design), interpretation and writing of the manuscript; and thoroughly participated in drafting and revision of the manuscript. Declaration of Competing Interest None. Acknowledgements The infrastructure for the NESDA study www.nesda.nl is funded through the Geestkracht program of the Netherlands Organisation for Health Research and Development (ZonMw, grant number 10-0001002) and financial contributions by participating universities and mental health care organizations (VU University Medical Center, GGZ inGeest, Leiden University Medical Center, Leiden University, GGZ Rivierduinen, University Medical Center Groningen, University of Groningen, Lentis, GGZ Friesland, GGZ Drenthe, Rob Giel Onderzoekscentrum). 7
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
Supplementary materials
https://www.nvvp.net/stream/richtlijn-bipolaire-stoornissen-2015. Lamers, F., Hoogendoorn, A.W., Smit, J.H., Van Dyck, R., Zitman, F.G., Nolen, W.A., Penninx, B.W., 2012. Sociodemographic and psychiatric determinants of attrition in the Netherlands study of depression and anxiety (NESDA). Compr. Psychiatry 53, 63–70. https://doi.org/10.1016/j.comppsych.2011.01.011. Lichtenstein, P., Yip, B.H., Björk, C., Pawitan, Y., Cannon, T.D., Sullivan, P.F., Hultman, C.M., 2009. Common genetic determinants of schizophrenia and bipolar disorder in Swedish families: a population-based study. Lancet. https://doi.org/10.1016/S01406736(09)60072-6. Ma, Y., Gao, H., Yu, X., Si, T., Wang, G., Fang, Y., Liu, Z., Sun, J., Yang, H., Wang, X., Li, J., Zhang, Y., Sachs, G., 2016. Bipolar diagnosis in China: evaluating diagnostic confidence using the bipolarity index. J. Affect. Disord. 202, 247–253. https://doi. org/10.1016/j.jad.2016.05.039. McGuffin, P., Rijsdijk, F., Andrew, M., Sham, P., Katz, R., Cardno, A., 2003. The heritability of bipolar affective disorder and the genetic relationship to unipolar depression. Arch. Gen. Psychiatry. https://doi.org/10.1001/archpsyc.60.5.497. Menard, S., 2002. Applied Logistic Regression Analysis. Sage. https://doi.org/10.4135/ 9781412983433. Mosolov, S., Ushkalova, A., Kostukova, E., Shafarenko, A., Alfimov, P., Kostyukova, A., Angst, J., 2014. Bipolar II disorder in patients with a current diagnosis of recurrent depression. Bipolar Disord. 16, 389–399. https://doi.org/10.1111/bdi.12192. Østergaard, S.D., Straszek, S., Petrides, G., Skadhede, S., Jensen, S.O.W., MunkJørgensen, P., Nielsen, J., 2014. Risk factors for conversion from unipolar psychotic depression to bipolar disorder. Bipolar Disord. 16, 180–189. https://doi.org/10. 1111/bdi.12152. Penninx, B.W.J.H., Beekman, A.T.F., Smit, J.H., Zitman, F.G., Nolen, W.A., Spinhoven, P., Cuijpers, P., De Jong, P.J., Van Marwijk, H.W.J., Assendelft, W.J.J., van der Meer, K., Verhaak, P., Wensing, M., de Graaf, R., Hoogendijk, W.J., Ormel, J., van Dyck, R., 2008. The Netherlands study of depression and anxiety (NESDA): rationale, objectives and methods. Int. J. Methods Psychiatr. Res. 17, 121–140. https://doi.org/10. 1002/mpr.256. Perkins, N.J., Schisterman, E.F., 2006. The inconsistency of “optimal” cutpoints obtained using two criteria based on the receiver operating characteristic curve. Am. J. Epidemiol. 163, 670–675. https://doi.org/10.1093/aje/kwj063. Regeer, E.J., Ten Have, M., Rosso, M.L., Hakkaart-Van Roijen, L., Vollebergh, W., Nolen, W.A., 2004. Prevalence of bipolar disorder in the general population: a reappraisal study of the Netherlands mental health survey and incidence study. Acta Psychiatr. Scand. 110, 374–382. https://doi.org/10.1111/j.1600-0447.2004.00363.x. Robins, E., Guze, S.B., 1970. Establishment of diagnostic validity in psychiatric illness: its application to schizophrenia. Am. J. Psychiatry 126, 983–987. Sachs, G.S., 2004. Strategies for improving treatment of bipolar disorder: integration of measurement and management. Acta Psychiatr. Scand. 110, 7–17. https://doi.org/ 10.1111/j.1600-0447.2004.00409.x. Serretti, A., Fabbri, C., 2013. Shared genetics among major psychiatric disorders. Lancet. https://doi.org/10.1016/S0140-6736(13)60223-8. Skjelstad, D.V., Malt, U.F., Holte, A., 2010. Symptoms and signs of the initial prodrome of bipolar disorder: a systematic review. J. Affect. Disord. https://doi.org/10.1016/j. jad.2009.10.003. Sugaya, L., Hasin, D.S., Olfson, M., Lin, K.H., Grant, B.F., Blanco, C., 2012. Child physical abuse and adult mental health: a national study. J. Trauma. Stress. https://doi.org/ 10.1002/jts.21719. World Health Organization, 1997. Composite International Diagnostic Interview, Core version 2.1: Interviewer's Manual. Syndey, Aust.
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.jad.2019.10.055. References Aiken, C.B., Weisler, R.H., Sachs, G.S., 2015. The bipolarity index: a clinician-rated measure of diagnostic confidence. J. Affect. Disord. 177, 59–64. https://doi.org/10. 1016/j.jad.2015.02.004. Angst, J., 2006. Do many patients with depression suffer from bipolar disorder? Can. J. Psychiatry. Angst, J., Azorin, J.M., Bowden, C.L., Perugi, G., Vieta, E., Gamma, A., Young, A.H., 2011. Prevalence and characteristics of undiagnosed bipolar disorders in patients with a major depressive episode: the bridge study. Arch. Gen. Psychiatry. https://doi.org/ 10.1001/archgenpsychiatry.2011.87. Angst, J., Sellaro, R., Stassen, H.H., Gamma, A., 2005. Diagnostic conversion from depression to bipolar disorders: results of a long-term prospective study of hospital admissions. J. Affect. Disord. 84, 149–157. https://doi.org/10.1016/S0165-0327(03) 00195-2. Barnett, J.H., Smoller, J.W., 2009. The genetics of bipolar disorder. Neuroscience 164, 331–343. https://doi.org/10.1016/j.neuroscience.2009.03.080. Berk, M., Dodd, S., Kauer-Sant’Anna, M., Malhi, G.S., Bourin, M., Kapczinski, F., Norman, T., 2007. Dopamine dysregulation syndrome: implications for a dopamine hypothesis of bipolar disorder. Acta Psychiatr. Scand. 116, 41–49. https://doi.org/10.1111/j. 1600-0447.2007.01058.x. Coryell, W., Endicott, J., Maser, J.D., Keller, M.B., Leon, A.C., Akiskal, H.S., 1995. Longterm stability of polarity distinctions in the affective disorders. Am. J. Psychiatry 152, 385–390. https://doi.org/10.1176/ajp.152.3.385. First, M.B., Spitzer, R.L., Gibbon, M., Williams, J.B.W., 1997. In: Structured Clinical Interview for DSM-IV Axis I Disorders. Clinician Version (SCID-CV), for DSMIV. Hanley, J.A., McNeil, B.J., 1982. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143, 29–36. https://doi.org/10.1148/ radiology.143.1.7063747. Hirschefeld, R.M.A., 2001. The comorbidity of major depression and anxiety disorders: recognition and management in primary care. Prim Care Companion J. Clin. Psychiatry. https://doi.org/10.3109/00048679709062683. Holman, L., Head, M.L., Lanfear, R., Jennions, M.D., 2015. Evidence of experimental bias in the life sciences: why we need blind data recording. PLoS Biol. 13. https://doi. org/10.1371/journal.pbio.1002190. IBM Corp. Released, 2011. IBM SPSS Statistics for Windows, Version 22.0. 2011. Kessing, L.V., Willer, I., Andersen, P.K., Bukh, J.D., 2017. Rate and predictors of conversion from unipolar to bipolar disorder: a systematic review and meta-analysis. Bipolar Disord. https://doi.org/10.1111/bdi.12513. Kessler, R.C., Akiskal, H.S., Angst, J., Guyer, M., Hirschfeld, R.M.A., Merikangas, K.R., Stang, P.E., 2006. Validity of the assessment of bipolar spectrum disorders in the WHO CIDI 3.0. J. Affect. Disord. 96, 259–269. https://doi.org/10.1016/j.jad.2006. 08.018. Kupka, R.W., Goossens, P., van Bendegem, M., Daemen, P., Daggenvoorde, D.M., Dols, A., van Duin, D., Hillegers A., H., ter Kulve, E., Peetoom, T., Schulte, R., Stevens, A., 2015. Multidisciplinaire Richtlijn Bipolaire Stoornissen. Utrecht, De Tijdstroom.
8
Contents lists available at ScienceDirect
Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad
Research paper
The predictive performance of the bipolarity index in a Dutch epidemiological sample manuscript Wendela G. ter Meulena,b, , Stasja Draismaa,b, Aartjan T.F. Beekmana,b, Brenda W.J.H. Penninxa, Ralph W. Kupkaa,b ⁎
a b
Amsterdam UMC, Vrije Universiteit, Psychiatry, Amsterdam Public Health research institute, The Netherlands GGZ inGeest Specialized Mental Health Care, Amsterdam, The Netherlands
ABSTRACT
Background: No instrumnt exists that can predict the incidence of bipolar disorders (BD). The Bipolarity index (BI), originally developed to improve diagnostic confidence for a lifetime diagnosis of BD, may predict incident BD. Aim: To assess the predictive performance of the BI for incident BD in persons with a lifetime depression. Methods: The BI score was composed from different questionnaires and interviews in n = 1857 subjects without BD and with a lifetime unipolar depressive disorder from the Netherlands Study of Depression and Anxiety, a longitudinal cohortstudy. The incidence of DSM-IV defined BD I or II as a criterion diagnosis was established with the Composite International Diagnostic Interview after 2, 4, 6 and 9 years of follow-up. Cox regression analyses calculated whether the BI predicts incident BD during 9-years of follow-up. The area Under the Curve (AUC) was determined. At the optimal cut-off score, sensitivity, specificity, positive, and negative predictive values (PPV and NPV) were calculated. Results: Over the course of 9 years, bipolar conversion occurred in n = 46 subjects (2.5%). Each point increase in BI score significantly predicted incident BD (HR [95%CI]= 1.047[1.018–1.076], p = 0.001). The AUC was 0.61 (95%CI: 0.54–0.68). At the optimal cut-off of 30, sensitivity was 67%, specificity 52%, PPV 3% and NPV 98%. Limitations: Not all items of the BI were fully covered; mean age of the sample of 42. Conclusion: The BI score predicts bipolar conversion over 9 years in those with a lifetime depression. However, given the modest performance metrics, the BI cannot guide clinical decision making yet.
1. Introduction Bipolar disorders (BD) often debut as unipolar depressive disorders (UD), whereas frank hypomanic or manic episodes often appear later in life (Angst, 2006). The yearly rate of such bipolar conversion from an initial diagnosis of UD to BD declines from 3.9% in the first year after study entry with a diagnosis of UD to 0.8% in years 5–10 (Kessing et al., 2017). Conversion rates also depend on treatment setting, and increase to 1 out of 3 patients hospitalized for UD (Angst et al., 2005). Timely identification of those persons with UD who are at risk for bipolar conversion is relevant, as prognosis and outcome are poorer for persons with BD compared to UD (Kessing et al., 2017), and because BD requires different treatment (Kupka et al., 2015) than UD. Several predictors for bipolar conversion have been identified (Kessing et al., 2017), such as family history of BD and age at onset of depression. However, prodromal symptoms and risk factors for bipolar conversion are inconsistent across studies
due to variations in methodology (Kessing et al., 2017; Skjelstad et al., 2010), and their specificity for the prediction of incident BD is low (Skjelstad et al., 2010). Persons presenting with a UD frequently show subthreshold manic symptoms or other features of “bipolarity” to some extent (Angst et al., 2011). Inspired by this clinical experience and driven by an intent to assign a “bipolar profile” to the individual treatment-seeking patient with a probable mood disorder, experienced investigators in the field of BD created the bipolarity index (BI) in 2004 (Sachs, 2004). The BI is a 0–100 continuous scale that covers five illness dimensions with a maximum of 20 points per domain: I. signs and symptoms; II. age of onset; III. course of illness; IV. response to treatment, and V. family history. “Classic” BD according to the authors of the BI would be characterized by: I. at least one euphoric manic episode; II. early age of onset; III. recurrent and fully remitting illness course; IV. positive response to a mood stabilizer; and V. having a first-degree family member with BD.
Corresponding author at: W.G. Wendela ter Meulen, GGZ inGeest Specialized Mental Health Care, Research and Innovation, Oldenaller 1, 1081 HJ Amsterdam, The Netherlands. E-mail addresses: [email protected] (W.G. ter Meulen), [email protected] (S. Draisma), [email protected] (A.T.F. Beekman), [email protected] (B.W.J.H. Penninx), [email protected] (R.W. Kupka). ⁎
https://doi.org/10.1016/j.jad.2019.10.055 Received 27 May 2019; Received in revised form 10 September 2019; Accepted 28 October 2019 Available online 30 October 2019 0165-0327/ © 2019 Elsevier B.V. All rights reserved.
Please cite this article as: Wendela G. ter Meulen, et al., Journal of Affective Disorders, https://doi.org/10.1016/j.jad.2019.10.055
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
The five dimensions of the BI highlight that the conceptualization of “bipolarity” that underlies the BI goes beyond the mere symptomatic assessment of lifetime affective symptoms, as the DSM-IV categorization requires. Instead, the authors of the BI added additional state and trait variables, based on their clinical experience and on the earlier theories of Kraepelin. By doing so, the BI represents a broader view of mood disorders that is now termed “bipolarity”, that represents a more conservative view of the classic conceptualization of mood disorders, which Kraepelin considered to include risk for shifts to elevated mood states. Hence, it is well possible that the BI estimates a latent trait of bipolarity that may become apparent as bipolar conversion at a later stage in those with a lifetime UD. Nevertheless, no studies to date have investigated the predictive performance of the BI for incident BD in persons with a lifetime UD, an event also known as bipolar conversion. Studies into the concurrent validity of the BI against a lifetime DSM-IV classification of BD, although potentially flawed by observer bias (Holman et al., 2015) as the diagnosis of BD and BI answers are often provided by the same clinician, found good to excellent metrics (Aiken et al., 2015; Ma et al., 2016; Mosolov et al., 2014). Although the BI was not designed for the prediction of BD, investigation into the predictive performance of the BI for bipolar conversion could be of particular interest for clinical practice in those with a lifetime UD who present with risk factors of BD conversion and/or subthreshold bipolar symptoms (Angst et al., 2011). Such research may also substantiate the validity of the BI as a measure of “bipolarity”. Therefore, in this paper, we will report on the predictive performance of the BI for incident BD by using data from the Netherlands Study of Depression and Anxiety disorders (NESDA) sample. The longitudinal design of NESDA with 4 measurements in 9 years of follow-up allows prospective investigation into incident BD in those with a lifetime UD. NESDA has a high prevalence of affective and anxiety disorders and comorbid conditions such as substance use disorders, which means a high density of bipolar risk factors (Angst et al., 2005; Coryell et al., 1995; Østergaard et al., 2014) that are included in the BI. The predictive validity of the BI will be estimated for the incidence of BD type I or II after 2, 4, 6 and 9 years of follow-up as a criterion diagnosis in those with a lifetime UD at baseline who were recruited from the community, primary care, and outpatient mental health care services. Analyses will be repeated for the subsample of those aged < 30 years at baseline, as most bipolar conversion occurs before this age (Angst et al., 2005). BD, lifetime UD and comorbid conditions such as anxiety disorders will be assessed according to DSMIV criteria with the Composite International Diagnostic Interview (CIDI) (World Health Organization, 1997). We hypothesize that a higher BI score at baseline predicts an increased risk of bipolar conversion at follow-up assessments. Finally, the individual contribution of each of the five dimensions of the BI to the incidence of BD in this sample will be explored.
elsewhere (Penninx et al., 2008). In short, the baseline assessment (T0) consisted of an extensive face-to-face interview by trained interviewers, including a diagnostic psychiatric interview with the CIDI (World Health Organization, 1997), and paper-and-pencil questionnaires. Follow-up assessments took place after 2 (T2), 4 (T4), 6 (T6) and 9 (T9) years. Since NESDA was originally not aimed at BD, this diagnosis was an exclusion criterion at sample selection, well before baseline assessments took place. In the first follow-up assessments that were conducted after two years, the first CIDI assessment of lifetime BD took place. At that T2 assessment, lifetime BD I or II had been retrospectively reclassified with the CIDI in n = 116 subjects with UD at T0. For the present analyses, we selected all n = 1857 (1973–116) respondents with lifetime UD and without lifetime BD at baseline. NESDA was approved centrally by the Ethical Review Board of the VU University Medical Centre and subsequently by the local review boards of each participating center, and all participants signed written informed consent. 2.2. Bipolarity Index (BI) scoring The BI is a clinician-rated instrument that covers five clinical domains based on the original work of Robins and Guze to validate psychiatric disorders (Robins, E., Guze, 1970): I. signs and symptoms; II. age of onset; III. course of illness; IV. response to treatment, and V. family history. Per domain, several items can be assessed as either present or absent. These items are considered common clinical features with an ordinal score ranging from 0–20 (with possible values of 0, 2, 5, 10 15, and 20). A maximum score of 20 per domain indicates the presence of an item within that domain most characteristic of ‘classic’ BD type I. The total BI score is the sum of the highest rated items on each of the five domains, thus ranging from 0–100. For the current study, the BI score at T0 was constructed from data that had been retrieved with several questionnaires and interviews during the NESDA assessments. Before assessing the items of the BI, scoring decisions were discussed by the authors [RK, WtM]. Table 1 shows the complete BI interview as originally described by Sachs et al. (Sachs, 2004), the NESDA research instruments per item that were chosen for scoring, relevant scoring decisions and their pertaining references, and descriptives. Among these items, the presence of (hypo) manic episodes at T0 in domain I and III of the BI had been retrospectively assessed with the CIDI-BD interview at T2. Furthermore, no validated epidemiological instrument was available to assess the precise effects of medication use (BI domain IV). In NESDA, medication use had been assessed through a medication interview that covered current (T0) and retrospective medication use over a period of 5 years before T0 (Penninx et al., 2008), documented according to the Anatomical Therapeutic Chemical (ATC) classification system of the World Health Organizations (WHO). Current (T0) and previous (covering the 5 years before T0) use of antidepressants (ATC code N06A), antipsychotics (ATC code N05A, including lithium N05AN) and mood stabilizers (ATC code N03A) was selected. Finally, the BI total score at T0 was calculated by summing the subscores of the five domains of the BI. To establish the subscore on each domain, we assessed the highest scored item on that domain in accordance with the original BI procedure. For example, when a subject rated positive in domain III on the items “comorbid personality disorder” (rated 5 points) and “comorbid substance use disorder” (rated 10 points), the score of 10 prevailed. We chose a conservative approach for assigning a BI-score at baseline, as this most closely resembles the routine clinical practice of the BI interview. Conservative means that if no or ambiguous information was available on a certain item, a score of 0 was assigned. The interviewers who had performed the psychiatric interviews at T0 were unaware that the data would be used for BI rating later on.
2. Method 2.1. Study sample Data were retrieved from the Netherlands Study of Depression and Anxiety (NESDA) (Penninx et al., 2008), a multisite naturalistic cohort study designed to investigate the course and consequences of the most common depressive and anxiety disorders in adults (18–65 years). To represent various settings and stages of psychopathology, persons were recruited from the community (19%), primary care (54%) and outpatient mental health care services (27%). Subjects in NESDA at baseline (n = 2981) include n = 1973 (66%) subjects with a lifetime DSMIV unipolar depressive disorder, defined as major depressive disorder and/or dysthymia as assessed with the Composite International Diagnostic Interview (CIDI) version 2.1 (World Health Organization, 1997). Aims, design and method of recruitment were extensively described 2
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
Table 1 Original BI interview; scoring instruments to rate the BI in this sample; relevant scoring decisions; and descriptives: see below. Bipolarity Index(Sachs, 2004) For each of the items below circle the item characteristic of the patient: Most convincing characteristic: 20 Convincing characteristic of Bipolar Disorder: 15 Known associated feature of Bipolar Disorder: 10 Non-specific feature suggestive of Bipolar Disorder: 5 Feature with possible relationship to Bipolar Disorder: 2 No evidence of Bipolar Disorder: 0 Score & BI domain Instrument Domain I: Episode characteristics 20 1. Documented acute mania or mixed episode with prominent euphoria, grandiosity or expansiveness and no significant general medical or known secondary etiology 15 Clear-cut acute mixed episode or dysphoric, or irritable mania with no significant general medical or known secondary etiology 10 2. Clear-cut hypomania with no significant general medical or known secondary etiology Clear-cut cyclothymia with no significant general medical or known secondary etiology Clear cut mania secondary to antidepressant use 5
Missing% 0%
Mania at/before T0 according to DSM-IV criteria as established with the CIDI 2.1
n = 0 (0.0%)
CIDI 2.1 BD (World Health Organization, 1997) at T2
0%
n = 1 (0.1%)
CIDI 2.1 BD (World Health Organization, 1997) at T2 *
0%
Mixed episode at/before T0 according to DSM-IV criteria as established with the CIDI 2.1 Hypomania at/before T0 according to DSMIV criteria as established with the CIDI 2.1 Cyclothymia not assessed with the CIDI 2.1
*
CIDI 2.1 BD (World Health Organization, 1997) at T2
0%
Clear cut hypomania secondary to antidepressant use Episodes with characteristics signs of hypomania but symptoms, duration or intensity are subthreshold for hypomania or cyclothymia
CIDI 2.1 BD (World Health Organization, 1997) at T2 Mood Disorder Questionnaire (MDQ) (Hirschfeld et al., 2000) at T0
0%
A single major depressive episode with psychotic or atypical features (Atypical 2 of 3: hypersomnia, hyperphagia, leaden paralysis of limbs Any postpartum depression (PPD)
CIDI 2.1 MDD (World Health Organization, 1997) at T0
0%
Edinburgh Postnatal Depression Scale (Cox et al., 1987) at T4
10.6%b
Any recurrent typical unipolar major CIDI 2.1 MDD (World Health depressive disorder (MDD) Organization, 1997) at T0 History of any kind of psychotic episode(i.e. * presence of delusions, hallucinations, ideas of reference, or magical thinking) 0 No history of significant mood elevation, recurrent depression, or psychosis Domain II: Age of onset (1st affective episode ⁄ syndrome) 20 First episode age 15–19 years CIDI 2.1 MDD (World Health 15 First episode age < 15 or 20–30 Organization, 1997) at T0 10 First episode age 30–45 CIDI 2.1 BD (World Health 5 First episode > 45 Organization, 1997) at T2 0 No history of affective illness (no episodes, cyclothymia, dysthymia or BP NOS) Domain III: Course of illness⁄associated features 20 Recurrent distinct manic episodes separated by Retrospective lifechart interview periods of full recovery (Lyketsos et al., 1994) that covered the last 3 years before T0 Recurrent distinct manic episodes with incomplete interepisode recovery Recurrent distinct hypomanic episodes with full interepisode recovery 10
Comorbid substance abuse
Psychotic features only during acute mood episodes Incarceration or repeated legal offences related to manic behaviour (e.g. shoplifting, reckless driving, bankruptcy)
Descriptives
CIDI 2.1 BD (World Health Organization, 1997) at T2
2
15
Relevant scoring decisions
Retrospective lifechart interview (Lyketsos et al., 1994) that covered the last 3 years before T0 Retrospective lifechart interview (Lyketsos et al., 1994) that covered the last 3 years before T0 Fagestrom Test of Nicotine Dependence (FTND) (Fagerström, 1978) CIDI 2.1 alcohol use disorders (AUD) (World Health Organization, 1997) CIDI 2.1 BD (World Health Organization, 1997) at T2 Retrospective lifechart interview Retrospective lifechart interview (Lyketsos et al., 1994)
3.0%
0% *
0 (0)%
Positive if subject reports any mania or mixed episode on CIDI secondary to antidepressant use at/before T0. Positive if subject reports any hypomania secondary to antidepressant use at/before T0. Positive if in A-section either first question (good, hyper) and ≥3 other questions are answered “yes”, or second question (irritable) and ≥4 other questions are answered “yes”; and B-section “yes”; and C-section shows minor problems Positive if any MDD and any episode characterized by at least 2 atypical features
12.9% 12.9% 0%
0% 0%
* n = 3 (0.1%) n = 0 (0.0%) n = 246 (13.2%)
n = 278 (15.0%)
Positive if score >12 on EPDS (Meltzer-Brody et al., 2013) and current or lifetime MDD and/or dysthymia before T0 Recurrent MDD according to DSM-IV criteria as established with the CIDI 2.1 Primary psychotic disorder was exclusion criterion in NESDA at baseline.
n = 934 (50.3%)
Those without score 2–20 on domain I
n = 1550 (54.2%)
Question in CIDI on age of onset first affective episode (episode of mania, hypomania, MDD, or dysthymia)
n = 347 n = 721 n = 468 n = 208 n = 113
Those without score 5–20 on domain II 12.9%
n = 17 (0.9%)
Any (either medication-induced or not) manic or mixed episode (CIDI) and ≥15% of time asymptomatic (Verduijn et al., 2015)a Any manic or mixed episode (CIDI) and <15% of time asymptomatic (Verduijn et al., 2015)a Any hypomanic episode (CIDI) and ≥15% of time asymptomatic (Verduijn et al., 2015)a Positive if score ≥5 on FTND, corresponding with high nicotine dependence (Fagerström, 1978), and/or if any AUD in the past 6 months Positive if any psychotic symptoms were present in CIDI-BD Screening on the following words: legal, offences, shoplifting, reckless driving, bankruptcy
n = 250 (13.5%)
(18.7%) (38.8%) (25.2%) (11.2%) (6.1%)
n = 2 (0.1%)
n = 0 (0.0%) n = 14 (0.8%) FTND: n = 138 (7.4%) AUD: n = 130 (7.0%) n = 5 (0.3%) n = 0 (0%)
(continued on next page)
3
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
Table 1 (continued) 5
Recurrent unipolar MDD with three or more major depressive episodes Recurrent distinct hypomanic episodes without full interepisode recovery
Comorbid borderline personality
Personality Diagnostic Questionnaire-4 (PDQ-4)
0%
Comorbid anxiety or eating disorders (e.g. OCD, panic disorder, bulimia) History of ADHD in childhood and periods of above average scholastic or social function
CIDI 2.1 anxiety disorders (AD) at T0 and T2 Conners Adult ADHD Rating Scale (CAARS) (Conners et al., 1999) at T0
0%
Gambling, risky investment, overspending, or sexual indiscretions have (or would if not concealed) pose a problem for patient, friends, or family Behavioral evidence of perimenstrual exacerbation of mood symptoms Baseline hyperthymic personality (when not manic or depressed)
*
*
Positive if MDD and three or more major depressive episodes Any hypomanic episode (CIDI) and <15% of time asymptomatic (Verduijn et al., 2015)a If ≥2 antidepressant medications (ATC code N06A) or mood stabilizers (ATC code N03A) were used <50% of the time that was prescribed in the past 5 years (consensus RK / WtM). Positive if ≥4 items of borderline personality disorder present. The PDQ-4 has high sensitivity and moderate specificity for borderline personality disorder. A score of >4 is highly suggestive of BPD (Sansone et al., 2008) Positive if any recent (<6 month recency) anxiety disorder Positive if current ADHD symptoms on the CAARS and a positive score on childhood or early-adolescent ADHD indicators, as described elsewhere (Bron et al., 2016) *
*
*
*
NEO-Five Factor Inventory (NFFI) (Hoekstra et al., 1996)
1.4%
Married three or more times (including remarriage to same individual) In two or more years has started a new job and changed jobs after less than year Has more than two advanced degrees 0 None of the above Domain IV: Response to treatment 20 Full recovery within 4 weeks of therapeutic treatment with mood stabilizing medication 15 Full recovery within 12 weeks of therapeutic treatment with mood stabilizing medication or relapse within 12 weeks of discontinuing treatment
*
*
Positive if extraversion score was >1.5 SD above the mean extraversion score (consensus RK / WtM). *
*
*
*
*
*
* Those without score 2–20 on domain III
*
*
Retrospective medication interview T0
0%
CIDI 2.1 BD (World Health Organization, 1997) at T2
0%
Retrospective medication interview T0 CIDI 2.1 BD (World Health Organization, 1997) at T2
0%
Time of response to mood stabilizers was not explicitly asked. Positive if previous (covering the 2 years before T2) use of antipsychotics (ATC code N05A, including lithium N05AN) and/or mood stabilizers (ATC code N03A); and subject reports response as reason why medication was ended (consensus RK / WtM). Positive if subject reports any mania or mixed episode on CIDI secondary to antidepressant use. Positive if previous (covering the 2 years before T0) use of antidepressants (ATC code N06A) was ended and subject reports manic or mixed symptoms (free text) as reason why medication was ended (consensus RK / WtM). Positive if subject reports any hypomanic episode at/before T0 on CIDI secondary to antidepressant use.
Recurrent medication non-compliance
2
10
5
2 0
Affective switch to mania (pure or mixed) within 12 weeks of starting a new antidepressant or increasing dose Worsening dysphoria or mixed symptoms during antidepressant treatment subthreshold for mania
CIDI 2.1 MDD (World Health Organization, 1997) at T0 Retrospective lifechart interview (Lyketsos et al., 1994) that covered the last 3 years before T0 Current and retrospective medication interview that covered the previous 5 years before T0
0% 0% 0%
0%
n = 638 (34.4%) n = 3 (0.2%) n = 29 (1.6%)
n = 319 (17.2%)
n = 982 (52.9%) n = 149 (8.0%)
n = 53 (2.9%)
Partial response to one or two mood stabilizers within 12 weeks of therapeutic treatment Antidepressant-induced new or worsening rapid-cycling course
*
*
Retrospective medication interview T0
0%
Treatment resistance: lack of response to complete trials of three or more antidepressants Affective switch to mania or hypomania with antidepressant withdrawal Immediate near complete response to antidepressants (in 1 week or less) None of the above or no treatment
Current and retrospective medication interview at T0
0%
*
*
Positive if previous (covering the 2 years before T2) use of antidepressants (ATC code N06A) was ended and subject reports rapid cycling (free text) as reason why medication was ended (consensus RK / WtM). Positive if ≥3 antidepressant medications (ATC code N06A) were used on a daily base in the past 5 years *
*
*
* Those without score 2–20 on domain IV
N = 428 (23.6%)
n = 0 (0.0%)
n = 2 (0.1%) n = 1 (0.1%)
n = 0 (0.0%)
n = 0 (0%)
n = 0 (0.0%)
n = 1854 (99.8%)
(continued on next page)
4
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
Table 1 (continued) Domain V: Family history 20 At least one first-degree relative with documented bipolar illness 10 First-degree relative with documented recurrent unipolar MDD or schizoaffective disorder 10 Any relative with documented bipolar illness
5 2
0
Family tree interview that covers 1st and 2nd degree family members
0% 0% 0%
Any relative with documented recurrent unipolar MDD and behavioural evidence suggesting bipolar illness First-degree relative with documented substance abuse, or any relative with possible bipolar illness First degree relative with possible recurrent unipolar MDD First degree relative with diagnosed related illness: anxiety disorders, eating disorders, ADD None of the above or no family psychiatric illness
0% 0% 0% 0%
Positive if subject reports any BD in first degree family member Positive if subject reports any recurrent unipolar MDD or schizoaffective disorder in first degree family member Positive if subject reports any BD in second degree family member Positive if subject reports any recurrent unipolar MDD or schizoaffective disorder in second degree family member Positive if subject reports any substance use disorder in first degree family member Possible diagnosis was not questioned as such; therefore, no scoring possible Positive if subject reports any anxiety disorder or ADD in first degree family member Those without score 2–20 on domain V
n = 56 (3.0%) n = 45 (2.4%)
n = 1469 (79.1%) n = 832 (44.8%)
n = 1363 (73.4%) n = 152 (8.2%)
* = no scoring possible as no instrument was available to score this variable. a. = The level of recovery in domain III was based on the percentage of asymptomatic time as established by a retrospective life chart interview (Lyketsos et al., 1994), and is in line with previous reports from our group (Verduijn et al., 2015). b. Missings due to attrition.
2.3. Assessment of unipolar and bipolar disorder (BD)
3. Results
At baseline, the presence of a unipolar disorder (UD) was defined as either a lifetime major depressive disorder and/or lifetime dysthymia, as assessed with CIDI 2.1. Incident BD I or II after 2, 4, 6 and 9 years as established with the CIDI 2.1 (World Health Organization, 1997) was the criterion diagnosis against which the predictive performance of the BI was tested. The CIDI 2.1 is a structured lay-administered interview that is based on DSM-IV criteria (First et al., 1997). A new BD diagnosis was considered present if a subject fulfilled DSM-IV criteria (First et al., 1997) for BD I or II for the first time.
3.1. Sample Table 2 shows the sample of n = 1857 subjects with a lifetime UD at baseline. As would be expected, the majority of UD subjects had a comorbid anxiety disorder (Hirschefeld, 2001) and almost a third had experienced both lifetime MDD and lifetime dysthymia. Other sample characteristics are represented in Table 1, which includes a quantitative description of all the items of the BI. 3.2. Predictive validity
2.4. Statistical analyses
In the total of 9 years of follow-up in the whole sample (n = 1857), incident BD occurred in n = 49 subjects (2.6%): 4 at T2 (0.2%), 20 at T4 (1.1%), 16 at T6 (0.9%), and 9 at T9 (0.5%). Cox regression analyses revealed that each point increase in BI score at baseline was significantly associated with a higher hazard of incident BD over 9 years (HR[95%CI] = 1.05[1.02–1.08], p = 0.001). The AUC was 0.60 (95%CI: 0.52–0.81), and the cut-off of the BI score with the highest
Cox regression analysis was performed to determine the association between the BI score at baseline and the incidence of BD over 9-years of follow-up, with measurements at 4 different time points: after 2, 4, 6 and 9 years. Time at risk was measured from T0 until the moment either the person had incident BD, or was censored at the last recorded followup. The BI cut-off score that coincided with a maximum sum of sensitivity and specificity for incident BD during 9 years of follow-up was calculated by the Youden Index (Perkins and Schisterman, 2006) in a logistic regression model. Based on this BI cut-off score, sensitivity and specificity, positive and negative predictive values (PPV and NPV), and their corresponding 95% CI's were calculated by cross-tabulation. In order to estimate the contribution of each separate BI domain to the incidence of BD, a multivariate logistic regression model was created with stepwise regression analyses. Multicollinearity between the domainscores was checked by Variance Inflation Factors (VIF) (Menard, 2002). For the multivariate stepwise logistic regression model, the incidence of BD was defined as the incidence of BD at any time point between baseline and 9 years. The goodness of fit of the model for each added dimension was expressed as the change to the Nagelkerke correlation coefficient (R2). All analyses were conducted using IBM SPSS Statistics version 22.0 (SPSS Inc., Chicago, USA) (IBM Corp. Released, 2011).
Table 2 Baseline sociodemographic and clinical characteristics of the sample of n = 1857 subjects with a lifetime UD and without BD at baseline. Characteristics Sociodemographic Gender (% female) Age at intake (mean in years, SD) University education completed (%) Unemployed (%) Clinical Mean BI score (SD) Lifetime anxiety disorders (%) Lifetime MDD (%) Lifetime dysthymia (%) Lifetime MDD and dysthymia (%) Current (6 month recency) MDD (%) Current (6 month recency) anxiety disorder (%)
Lifetime UD (n = 1857) 68.8 42.1 (12.4) 9.7 17.6 28.5 (9.1) 71.2 97.4 32.9 30.3 46.5 55.5
Abbreviations: MDD (major depressive disorder); SD (standard deviation); UD (unipolar depressive disorder). 5
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
unipolar depressive disorder (UD), an event also known as bipolar conversion. The BI significantly and positively predicted bipolar conversion after 2, 4, 6, and 9 years of follow-up in those with lifetime UD who were recruited from the community, primary care and outpatient mental health services. This positive association was found in the whole sample as well as in the subsample of those aged <30 years at baseline. In this subsample, the HR per point increase of the BI was somewhat higher (HR 1.10 (95%CI 1.02–1.18)) compared to the HR in the whole sample (HR 1.05 (95%CI 1.02–1.08)). In the whole sample, the variance in two domains of the BI significantly contributed to the incidence of BD (domain III and IV), although their relative contribution to the incidence of BD was low: they accounted for only 6% of the prediction of bipolar conversion. A closer look into the performance metrics of the BI score reveals that the contribution of the BI to the prediction of bipolar conversion is modest: Taking into account that an AUC of 1.00 represents a perfect scale, and 0.50 a useless scale (Hanley and McNeil, 1982), the AUC of 0.60 in the whole sample and of 0.57 in the subsample of those aged <30 years at baseline indicates that the BI should be interpreted as a relatively poor test to separate those with from those without future incident BD. At the optimal cut-off of 30 in the whole sample, sensitivity of the BI was fair at 74%. This means, that for those with a score below 30, one can fairly – but not fully – rule out the incidence of BD over the next 9 years. At the optimal cut-off of 30, specificity was deficient with 45%, which means that at this cut-off the number of false positives and true negatives is almost equal. The high number of false positives also translates into a poor PPV of 4%, whereas we found a NPV of 98%. This high NPV can be explained by the low incidence of BD, and hence the large a priori likelihood that there would not be a bipolar conversion. The poor PPV means that the percentage of subjects with a BI of ≥30 who actually convert to BD is very low, and that the likelihood of a type-I error is very high. Our study also compared the psychometric performance of the whole sample (mean age of 42 years) with those aged < 30 years at baseline, as most bipolar conversion occurs in this younger age group (Angst et al., 2005). This comparison showed that at the same BI cut-off of for example 30, sensitivity is higher in those aged < 30 years than in the whole sample, at the expense of specificity (see Table 3). The higher the sensitivity, the better this advantage may weight against the effort and investment of an elaborate instrument such as the BI. However, no firm conclusions can be drawn yet given the low number of incident cases (n = 12) in this subsample. The poor discriminatory ability of the BI to separate those with from those without incident BD that we found in the whole sample as well in those aged <30 years at baseline, is related to the high density of prodromal symptoms and risk factors of BD in our sample, which challenges the instrument to separate those at risk for bipolar conversion to those who are not at risk. Still, a sample high in prodromal symptoms and risk factors is a valid sample to investigate the BI, as the BI is an elaborate and time-consuming instrument that would only be applied in such samples “at risk”, where the basic idea of performing the BI as a predictive test would be to yield a high sensitivity. High-risk subjects with a lifetime UD and unresponsive to conventional antidepressant treatment who have a high BI score, could potentially benefit from mood stabilizers for treating depression. In the true positives, such treatments aimed at bipolar spectrum disorders might hypothetically even delay bipolar conversion. By contrast, in samples with a lower a priori likelihood of bipolar conversion, the high proportion of false positives would be incompatible with clinical decision making. Nevertheless, based on our study the BI cannot be recommended yet as tool to predict bipolar conversion in clinical practice, or to guide treatment decisions. The modest performance of the BI for the prediction of bipolar conversion in our study should be interpreted with some caution. First, the incidence of BD in the NESDA sample is relatively low, compared to the incidence in other samples at risk, such as hospitalized subjects or
Table 3 Psychometric performance of the BI at different cut-offs. Cut-offs are displayed with intervals of 3 points around the optimal cut-off of 30 (whole sample, n = 1857). In the subsample of those aged <30 years at baseline (n = 377), psychometric values are only presented at the optimal cut-off of 39 and again at the cut-off of 30. Cut-off
Whole sample Sens
Age <30 years at baseline Spec PPV NPV Sens
18 21 24 27 30 33 36 39 42
91.8 89.8 78.8 75.5 73.5* 40.8 24.5 18.4 10.2
13.1 20.2 24.8 37.1 44.6* 69.2 82.4 88.0 95.1
2.8 3.0 3.1 3.1 3.5* 3.8 3.6 4.0 5.4
98.3 98.7 98.7 98.2 98.4* 97.7 97.6 97.5 97.5
Spec
PPV
NPV
83.3
22.5
4.1
97.6
41.7*
80.0*
6.4*
97.2*
Abbreviations: Sens = sensitivity (%); spec = specificity (%); PPV = positive predictive value (%); NPV = negative predictive value (%). ⁎ Values around the optimal cut-off, i.e. with the highest Youden Index, are displayed with an asterix.
Youden Index (1,18) was 30. This cut-off of 30 corresponded with a sensitivity of 73% (36/49), specificity of 45% (806/1808), positive predictive value (PPV) of 3% (36/1038) and negative predictive value of 98% (806/819). In the subsample of those aged < 30 years at baseline (n = 377), incident BD occurred in n = 12 subjects (3.1%). Cox regression analyses showed that each point increase in BI score at baseline was significantly associated with a higher hazard of incident BD over 9 years (HR[95%CI] = 1.10[1.02–1.18], p = 0.02). The AUC was 0.57 (95%CI: 0.39–0.76), and the cut-off of the BI score with the highest Youden Index (1,22) was 39. The corresponding performance metrics are displayed in Table 3. 3.3. Performance of the BI at different cut-offs Table 3 shows the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) at different cut-offs of the BI in the whole sample. The cut-offs are randomly chosen with intervals of 3 points around the optimal cut-off of 30. In the subsample of those aged <30 years at baseline, psychometric values in Table 3 are only presented at the optimal cut-off of 39 and again at the cut-off of 30. 3.4. Contribution of the five dimensions of the BI VIF values indicated that multicollinearity between the five dimensions was absent. In the multivariate logistic regression model, the BI dimensions were entered as independent variables and incident BD at any timepoint as dependent variable. Those BI domainscores that were skewed, were first log-transformed before they were entered as independent variables in the model. These multivariate analyses showed that in the whole sample domain III (β = =3.00, p = 0.034) and domain IV (β = =1.52, p < 0.001) were significantly associated with the outcome variable. Next, stepwise logistic regression with these two variables in the model revealed that the goodness of fit of the model expressed as the Nagelkerke R2with only domain III was 0.029; and with both domain III and IV the R2 was 0.061 (difference: 0.032). We considered the number of incident cases (n = 12) in the subsample of those aged < 30 years at baseline too small to calculate the contribution of the five dimensions. 4. Discussion The main goal of our study was to report on the predictive performance of the bipolarity index (BI) for the incidence of DSM-IV defined bipolar disorders (BD) as a criterion diagnosis in those with a lifetime 6
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
those in outpatient mental health care services specialized in mood disorders. This low incidence is probably due to the high mean age at baseline (42 ± 12 years) and to the fact that a BD diagnosis was an exclusion criterion before baseline assessments. Any clinical diagnosis with a low incidence is generally very difficult to predict by an instrument. Still, in our study we did find a significant effect, which is relevant given the high impact and clinical implications of a diagnosis of BD. Therefore, the BI appears a suitable instrument to evaluate risk of bipolar conversion, when incorporated in a broader risk assessment including other known risk factors for bipolarity such as environmental factors (McGuffin et al., 2003). For example, the impact of stressful life events such as childhood physical abuse on the risk of developing BD is well investigated (Sugaya et al., 2012) but is not included in the BI. Another explanation for the modest performance of the BI and in particular the poor PPV, is that a DSM classification of BD - a pure description of symptomatic course - may not be the correct gold standard for the diagnosis of BD. In other words, the construct of bipolarity may be better represented through the BI itself than by DSM classification criteria. This tentative explanation raises the question what “bipolarity” is as captured by the BI. Although the pathogenesis of BD is not known, family and twin studies show that inherited factors are considerably involved (Barnett and Smoller, 2009; McGuffin et al., 2003), which in the BI is represented by domain V (family history). Various items of domain III and the lower scored items in domain V reflect that inherited factors are pleiotropic (Lichtenstein et al., 2009), and that innate susceptibility confers a risk for other types of psychopathology too (Serretti and Fabbri, 2013) such as UD, anxiety and ADHD. The model of BD as a dopamine dysregulation syndrome describes an increased dopaminergic drive in mania and the opposed in depression (Berk et al., 2007). This model would also suggests a mechanism for antidepressant-induced manic symptoms, and for the particular response of BD to mood stabilizers that regulate dopaminergic transmission. In the BI, adverse reactions to regular antidepressant therapies and response to mood stabilizers are captured in domain I and IV. A few limitations should be considered. First, since the BI was not part of the original NESDA set of questionnaires, not all items of the BI could be completely addressed with the NESDA data, which resulted in a conservative scoring. A better psychometric performance could be expected when all items of the BI had been fully covered. Another limitation is the inevitable rate of attrition in longitudinal studies including NESDA (Lamers et al., 2012). A post-hoc comparison showed no significant difference between the mean BI score at baseline of those who during any of the follow-up assessments had attrited (28.3) versus those who did not (28.5), which suggests that those at a higher risk of bipolar conversion did not have a higher rate of attrition though. Finally, the high mean age at baseline (42 ± 12 years) means that all subjects but those aged <30 years (the youngest quartile) would be considered beyond the period of substantial risk for conversion (Angst et al., 2005). The higher mean age at baseline as well as the exclusion of lifetime BD or psychosis at baseline limits the representativeness of the sample, and means that the whole sample may be enriched for low risk subjects. In a sample with more subjects at risk, the state-related items (I and III) may be related to incidence of BD over time. Exclusion of subjects with BD however may have flattened the BI to essentially reflect predominantly the trait-related domains (II, IV and V). Another limitation is that the psychiatric diagnoses were assessed by the CIDI, which is a fully structured diagnostic interview administered by trained interviewers without clinical experience. The CIDI however might not assess bipolar spectrum disorders as accurately as a semistructured interview administered by clinicians – such as the Structured Clinical Interview for DSM (SCID) - does (Kessler et al., 2006; Regeer et al., 2004). This disadvantage should be weighed against the major advantage of the CIDI for large cohort studies that interviewers can relatively easily be trained to employ the instrument. Advantage of this study was that the BI rating was based on data from elaborate questionnaires and interviews from a large
epidemiological sample from the Netherlands. This allowed to minimize the risk of observer bias, and to provide transparent and reproducible data. Furthermore, NESDA is unique as a large sample with affective and anxiety disorders – a high risk population for bipolar conversion. The high density of risk factors for bipolar conversion considerably challenges the discriminatory ability of the BI. Altogether, some recommendations for future studies can be made. First, because the incidence of BD on average occurs at early adulthood, we suggest to perform future studies into the predictive performance of the BI in younger samples than in our study. Second, adaptations to the ranking or rating of the BI might increase the psychometric performance of the BI to predict incident BD. Third, to assess the gold standard diagnosis of BD, we suggest to employ diagnostic instruments with high concordance rates for BD, for example the clinician-rated SCID. 5. Conclusion The BI significantly and positively predicts the incidence of BD over 9 years of follow-up in a sample of subjects with a lifetime depressive disorder, recruited from the community, primary care and outpatient mental health care services. This is relevant, as to date no instrument exists that predicts BD, whereas that could be of clinical benefit for those persons at risk for conversion from lifetime UD to BD. Given the modest performance metrics of the BI in this study, the BI cannot be recommended yet as an instrument to predict bipolar conversion in routine clinical practice, or to guide clinical decision making. However, the BI may be a helpful tool to evaluate the risk of bipolar conversion, when integrated with other means of risk assessment. To further optimize the BI for the prediction of BD, we recommend research in younger samples at risk, and further development of the items of the BI. Author statement Wendela G. ter Meulen, MD1, 2 [email protected] chaired and executed the concept, design, analyses, drafting, and writing of the manuscript. Stasja Draisma, PhD1 [email protected] participated in the concept, design, analyses, interpretation and writing of the manuscript; and thoroughly participated in drafting and revision of the manuscript. Aartjan T.F. Beekman, MD, PhD1 [email protected] participated in the concept, design (including statistical design), interpretation and writing of the manuscript; and thoroughly participated in drafting and revision of the manuscript. Brenda W.J.H. Penninx, MD, PhD1 [email protected] intensively participated in the data acquisition; and had a major role in drafting and revision of the manuscript. Ralph W. Kupka, MD, PhD1 [email protected] participated in the concept, design (including methodological design), interpretation and writing of the manuscript; and thoroughly participated in drafting and revision of the manuscript. Declaration of Competing Interest None. Acknowledgements The infrastructure for the NESDA study www.nesda.nl is funded through the Geestkracht program of the Netherlands Organisation for Health Research and Development (ZonMw, grant number 10-0001002) and financial contributions by participating universities and mental health care organizations (VU University Medical Center, GGZ inGeest, Leiden University Medical Center, Leiden University, GGZ Rivierduinen, University Medical Center Groningen, University of Groningen, Lentis, GGZ Friesland, GGZ Drenthe, Rob Giel Onderzoekscentrum). 7
Journal of Affective Disorders xxx (xxxx) xxx–xxx
W.G. ter Meulen, et al.
Supplementary materials
https://www.nvvp.net/stream/richtlijn-bipolaire-stoornissen-2015. Lamers, F., Hoogendoorn, A.W., Smit, J.H., Van Dyck, R., Zitman, F.G., Nolen, W.A., Penninx, B.W., 2012. Sociodemographic and psychiatric determinants of attrition in the Netherlands study of depression and anxiety (NESDA). Compr. Psychiatry 53, 63–70. https://doi.org/10.1016/j.comppsych.2011.01.011. Lichtenstein, P., Yip, B.H., Björk, C., Pawitan, Y., Cannon, T.D., Sullivan, P.F., Hultman, C.M., 2009. Common genetic determinants of schizophrenia and bipolar disorder in Swedish families: a population-based study. Lancet. https://doi.org/10.1016/S01406736(09)60072-6. Ma, Y., Gao, H., Yu, X., Si, T., Wang, G., Fang, Y., Liu, Z., Sun, J., Yang, H., Wang, X., Li, J., Zhang, Y., Sachs, G., 2016. Bipolar diagnosis in China: evaluating diagnostic confidence using the bipolarity index. J. Affect. Disord. 202, 247–253. https://doi. org/10.1016/j.jad.2016.05.039. McGuffin, P., Rijsdijk, F., Andrew, M., Sham, P., Katz, R., Cardno, A., 2003. The heritability of bipolar affective disorder and the genetic relationship to unipolar depression. Arch. Gen. Psychiatry. https://doi.org/10.1001/archpsyc.60.5.497. Menard, S., 2002. Applied Logistic Regression Analysis. Sage. https://doi.org/10.4135/ 9781412983433. Mosolov, S., Ushkalova, A., Kostukova, E., Shafarenko, A., Alfimov, P., Kostyukova, A., Angst, J., 2014. Bipolar II disorder in patients with a current diagnosis of recurrent depression. Bipolar Disord. 16, 389–399. https://doi.org/10.1111/bdi.12192. Østergaard, S.D., Straszek, S., Petrides, G., Skadhede, S., Jensen, S.O.W., MunkJørgensen, P., Nielsen, J., 2014. Risk factors for conversion from unipolar psychotic depression to bipolar disorder. Bipolar Disord. 16, 180–189. https://doi.org/10. 1111/bdi.12152. Penninx, B.W.J.H., Beekman, A.T.F., Smit, J.H., Zitman, F.G., Nolen, W.A., Spinhoven, P., Cuijpers, P., De Jong, P.J., Van Marwijk, H.W.J., Assendelft, W.J.J., van der Meer, K., Verhaak, P., Wensing, M., de Graaf, R., Hoogendijk, W.J., Ormel, J., van Dyck, R., 2008. The Netherlands study of depression and anxiety (NESDA): rationale, objectives and methods. Int. J. Methods Psychiatr. Res. 17, 121–140. https://doi.org/10. 1002/mpr.256. Perkins, N.J., Schisterman, E.F., 2006. The inconsistency of “optimal” cutpoints obtained using two criteria based on the receiver operating characteristic curve. Am. J. Epidemiol. 163, 670–675. https://doi.org/10.1093/aje/kwj063. Regeer, E.J., Ten Have, M., Rosso, M.L., Hakkaart-Van Roijen, L., Vollebergh, W., Nolen, W.A., 2004. Prevalence of bipolar disorder in the general population: a reappraisal study of the Netherlands mental health survey and incidence study. Acta Psychiatr. Scand. 110, 374–382. https://doi.org/10.1111/j.1600-0447.2004.00363.x. Robins, E., Guze, S.B., 1970. Establishment of diagnostic validity in psychiatric illness: its application to schizophrenia. Am. J. Psychiatry 126, 983–987. Sachs, G.S., 2004. Strategies for improving treatment of bipolar disorder: integration of measurement and management. Acta Psychiatr. Scand. 110, 7–17. https://doi.org/ 10.1111/j.1600-0447.2004.00409.x. Serretti, A., Fabbri, C., 2013. Shared genetics among major psychiatric disorders. Lancet. https://doi.org/10.1016/S0140-6736(13)60223-8. Skjelstad, D.V., Malt, U.F., Holte, A., 2010. Symptoms and signs of the initial prodrome of bipolar disorder: a systematic review. J. Affect. Disord. https://doi.org/10.1016/j. jad.2009.10.003. Sugaya, L., Hasin, D.S., Olfson, M., Lin, K.H., Grant, B.F., Blanco, C., 2012. Child physical abuse and adult mental health: a national study. J. Trauma. Stress. https://doi.org/ 10.1002/jts.21719. World Health Organization, 1997. Composite International Diagnostic Interview, Core version 2.1: Interviewer's Manual. Syndey, Aust.
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.jad.2019.10.055. References Aiken, C.B., Weisler, R.H., Sachs, G.S., 2015. The bipolarity index: a clinician-rated measure of diagnostic confidence. J. Affect. Disord. 177, 59–64. https://doi.org/10. 1016/j.jad.2015.02.004. Angst, J., 2006. Do many patients with depression suffer from bipolar disorder? Can. J. Psychiatry. Angst, J., Azorin, J.M., Bowden, C.L., Perugi, G., Vieta, E., Gamma, A., Young, A.H., 2011. Prevalence and characteristics of undiagnosed bipolar disorders in patients with a major depressive episode: the bridge study. Arch. Gen. Psychiatry. https://doi.org/ 10.1001/archgenpsychiatry.2011.87. Angst, J., Sellaro, R., Stassen, H.H., Gamma, A., 2005. Diagnostic conversion from depression to bipolar disorders: results of a long-term prospective study of hospital admissions. J. Affect. Disord. 84, 149–157. https://doi.org/10.1016/S0165-0327(03) 00195-2. Barnett, J.H., Smoller, J.W., 2009. The genetics of bipolar disorder. Neuroscience 164, 331–343. https://doi.org/10.1016/j.neuroscience.2009.03.080. Berk, M., Dodd, S., Kauer-Sant’Anna, M., Malhi, G.S., Bourin, M., Kapczinski, F., Norman, T., 2007. Dopamine dysregulation syndrome: implications for a dopamine hypothesis of bipolar disorder. Acta Psychiatr. Scand. 116, 41–49. https://doi.org/10.1111/j. 1600-0447.2007.01058.x. Coryell, W., Endicott, J., Maser, J.D., Keller, M.B., Leon, A.C., Akiskal, H.S., 1995. Longterm stability of polarity distinctions in the affective disorders. Am. J. Psychiatry 152, 385–390. https://doi.org/10.1176/ajp.152.3.385. First, M.B., Spitzer, R.L., Gibbon, M., Williams, J.B.W., 1997. In: Structured Clinical Interview for DSM-IV Axis I Disorders. Clinician Version (SCID-CV), for DSMIV. Hanley, J.A., McNeil, B.J., 1982. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143, 29–36. https://doi.org/10.1148/ radiology.143.1.7063747. Hirschefeld, R.M.A., 2001. The comorbidity of major depression and anxiety disorders: recognition and management in primary care. Prim Care Companion J. Clin. Psychiatry. https://doi.org/10.3109/00048679709062683. Holman, L., Head, M.L., Lanfear, R., Jennions, M.D., 2015. Evidence of experimental bias in the life sciences: why we need blind data recording. PLoS Biol. 13. https://doi. org/10.1371/journal.pbio.1002190. IBM Corp. Released, 2011. IBM SPSS Statistics for Windows, Version 22.0. 2011. Kessing, L.V., Willer, I., Andersen, P.K., Bukh, J.D., 2017. Rate and predictors of conversion from unipolar to bipolar disorder: a systematic review and meta-analysis. Bipolar Disord. https://doi.org/10.1111/bdi.12513. Kessler, R.C., Akiskal, H.S., Angst, J., Guyer, M., Hirschfeld, R.M.A., Merikangas, K.R., Stang, P.E., 2006. Validity of the assessment of bipolar spectrum disorders in the WHO CIDI 3.0. J. Affect. Disord. 96, 259–269. https://doi.org/10.1016/j.jad.2006. 08.018. Kupka, R.W., Goossens, P., van Bendegem, M., Daemen, P., Daggenvoorde, D.M., Dols, A., van Duin, D., Hillegers A., H., ter Kulve, E., Peetoom, T., Schulte, R., Stevens, A., 2015. Multidisciplinaire Richtlijn Bipolaire Stoornissen. Utrecht, De Tijdstroom.
8