Está en la página 1de 78

Accepted Manuscript

Predictors of switch from depression to mania in bipolar disorder

Tomihisa Niitsu, Chiara Fabbri, Alessandro Serretti, M.D., Ph.D.

PII: S0022-3956(15)00117-X
DOI: 10.1016/j.jpsychires.2015.04.014
Reference: PIAT 2614

To appear in: Journal of Psychiatric Research

Received Date: 6 November 2014


Revised Date: 14 April 2015
Accepted Date: 14 April 2015

Please cite this article as: Niitsu T, Fabbri C, Serretti A, Predictors of switch from depression to mania in
bipolar disorder, Journal of Psychiatric Research (2015), doi: 10.1016/j.jpsychires.2015.04.014.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT

Predictors of switch from depression to mania in bipolar disorder

Tomihisa Niitsu 1,2, Chiara Fabbri 1, Alessandro Serretti 1 *

1
Department of Biomedical and NeuroMotor Sciences, University of Bologna, Bologna,

PT
Italy

RI
2
Research Center for Child Mental Development, Chiba University Graduate School of

SC
Medicine, Chiba, Japan

U
AN
Running title: Predictors of manic switch in bipolar depression
M
D
TE

* Corresponding author: Alessandro Serretti, M.D., Ph.D.


EP

Department of Biomedical and NeuroMotor Sciences, University of Bologna,

Viale Carlo Pepoli 5, 40123 Bologna, Italy


C

Tel +39 051 6584233; Fax +39 051 521030


AC

Email: alessandro.serretti@unibo.it

1
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

Abstract

Manic switch is a relevant issue when treating bipolar depression. Some risk factors

have been suggested, but unequivocal findings are lacking. We therefore investigated

predictors of switch from depression to mania in the Systematic Treatment

PT
Enhancement Program for Bipolar Disorder (STEP-BD) sample. Manic switch was

RI
defined as a depressive episode followed by a (hypo)manic or mixed episode within the

following 12 weeks. We assessed possible predictors of switch using generalized linear

SC
mixed models (GLMM).

8403 episodes without switch and 512 episodes with switch (1720 subjects) were

U
included in the analysis. Several baseline variables were associated with a higher risk of
AN
switch. They were younger age, previous history of: rapid cycling, severe manic
M

symptoms, suicide attempts, amphetamine use and some pharmacological and

psychotherapeutic treatments. During the current depressive episode, the identified risk
D

factors were: any possible mood elevation, multiple mania-associated symptoms with at
TE

least moderate severity, and comorbid panic attacks.

In conclusion, our study suggests that both characteristics of the disease history and
EP

clinical features of the current depressive episode may be risk factors for manic switch.
C
AC

Keywords: bipolar depression; manic switch; manic symptom; panic attack;

venlafaxine; suicide; rapid-cycling; amphetamine use; onset

2
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

Highlights

− Manic switch is a primary issue when treating bipolar depression


− Concomitant manic symptoms and panic attacks are the strongest risk factors
− Antidepressant treatment may not represent a primary risk factor

PT
In low risk cases, younger age, past treatments or suicide attempts may be
relevant

RI
U SC
AN
M
D
TE
C EP
AC

3
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

1. Introduction

Bipolar disorder (BD) is a chronic disease characterized by the alternation of

periods of (hypo)mania and depression, resulting in high personal and socio-economic

PT
burden in terms of poor quality of life, increased rates of suicide, direct and indirect

RI
costs (Ferrari et al. 2012; Conus et al. 2013; Fagiolini et al. 2013; Kleine-Budde et al.

2013; Whiteford et al. 2013). The course of BD is highly variable, thus the identification

SC
of individuals at higher risk of experiencing a more severe course has relevant

U
implications for both patients and public health care.
AN
Switch from depression to (hypo)mania or mixed status is a critical issue since
M

its high impact on disease severity. Indeed, patients with one or more episodes of switch
D

have longer periods of illness lifetime and higher risk of suicide; further, they
TE

experience more comorbidities (e.g. substance abuse) (Maj et al. 2002; MacKinnon et al.
EP

2003; MacKinnon et al. 2003; MacKinnon et al. 2005). Thus, the identification of

switch risk factors represents a pivotal issue in order to prevent this detrimental event
C
AC

and improve BD prognosis. Previous studies investigated antidepressant (AD)-related

switch in patients with BD, and they suggested a number of risk factors (see

Supplementary Table S1 for a summary). On the basis of the results observed in the

largest samples, the most replicated risk factors were previous history of switch or rapid

cycling, younger onset age, and BD I diagnosis (Serretti et al. 2003; Perlis et al. 2010;

4
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

Valenti et al. 2012). One of these studies was performed on the systematic Treatment

Enhancement Program for Bipolar Disorder (STEP-BD) sample and it reported several

associations between switch and baseline clinical-demographic features (greater number

PT
of past depressive episodes, recent or lifetime rapid cycling, alcohol use disorder,

RI
previous suicide attempt, past history of AD-emergent switch, and greater manic

SC
symptom severity) (Perlis et al. 2010). Some of these risk factors were dependent from

U
AD exposure (younger onset age in patients not treated with ADs; previous suicide
AN
attempt and BD type I diagnosis in AD-exposed patients).

Previous studies were mainly focused on the risk of switch during AD


M

treatment (Supplementary Table S1), but some patients with BD experience switch even
D

in the absence of AD treatment. Indeed, some randomized controlled trials (RCTs)


TE

suggested that a mood stabilizer plus adjunctive AD therapy, compared with a mood
EP

stabilizer plus a placebo or mood stabilizer monotherapy, showed no association with


C

higher risk of AD-emergent affective switch (Keck et al. 2005; Sachs et al. 2007). The
AC

first of these RTCs (Sachs et al. 2007) was carried out on a subsample of the STEP-BD

study and its results could have been biased due to the intervention of clinicians in the

process of randomization (tendency to avoid the randomization of patients at higher risk

of switch). Despite the risk of inclusion bias deriving from this type of study design,

5
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

several neurobiological mechanisms support the risk of spontaneous switch (Salvadore

et al. 2010), thus risk factors of switch should not be investigated only in relation to

concomitant AD treatment.

PT
Previous studies focused on the risk of AD-induced mood switch suggested that

RI
tricyclic ADs (Boerlin et al. 1998; Bottlender et al. 2001; Nemeroff et al. 2001;

SC
Bottlender et al. 2004), monoamine oxidase inhibitors (MAOIs) (Boerlin et al. 1998)

U
and venlafaxine (Leverich et al. 2006; Post et al. 2006) may be associated with higher
AN
risk than SSRIs or bupropion.

All the above-mentioned studies (with the exception of a small naturalistic


M

study (Boerlin et al. 1998)) investigated a patient-based phenotype of switch from


D

depression to (hypo)mania, i.e. they compared patients with and without switch, but
TE

they did not consider an episode-based phenotype, i.e. the comparison between
EP

depressive episodes followed or not by switch. Despite the fact that the disease course
C

usually keeps a similar pattern in the same patient during time (van Rossum et al. 2008;
AC

Uher et al. 2013), the recurrence of acute episodes increases the level of functional

impairment (Bonnin et al. 2013) and it possibly affects the clinical features and

frequency of the following episodes. Thus, the specific features of each episode should

be considered in addition to the general characteristics of BD in each patient in order to

6
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

provide an accurate prognosis and suitable treatment interventions.

Given the hypothesized clinical relevance of each episode in every subject, the

present study investigated the clinical predictors of switch from major depressive

PT
episodes to (hypo)manic/mixed episodes including into the analysis all the depressive

RI
episodes occurring for each patient during follow-up. Previous evidence suggested that

SC
the risk of switch is not influenced only by concomitant AD treatment, thus supporting

U
our choice. Thus, baseline clinical-demographic characteristics, past clinical history, and
AN
current clinical characteristics (including treatment) at each depressive episode were

investigated for a possible correlation with the risk of switch from depression to
M

(hypo)manic/mixed state.
D
TE

2. Methods
EP

2.1. Study Overview


C

STEP-BD was a large, multi-center and naturalistic cohort study, conducted in


AC

the USA between 1999 and 2005. The aim of STEP-BD was to evaluate prospective

outcomes among patients with bipolar disorder treated in a naturalist setting. STEP-BD

clinicians were encouraged to treat patients in accordance with contemporary practice

guidelines. The study design was approved by the Human Subjects Panel from each site.

7
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

More detailed methods of the STEP-BD are described elsewhere (Sachs et al. 2003;

Perlis et al. 2006).

2.2. Participants

PT
Entry criteria involved meeting DSM-IV criteria for BD I, II or not otherwise

RI
specified, or schizoaffective disorder bipolar type, and ability to provide informed

SC
consent. All patients were outpatients; inpatients were eligible to enter the study

U
following discharge (Sachs et al. 2003; Perlis et al. 2006). Overall, 4360 participants
AN
were included. Participants received pharmacological interventions as clinically

indicated and clinicians were encouraged to follow the principles of evidence-based


M

treatment and published guidelines, updated annually in the STEP-BD Clinicians


D

Handbook. The protocol promoted the application of evidence-based treatments at


TE

regular clinical assessments during follow-up, in accordance with the needs of the
EP

participant and did not follow compliance to a specific treatment algorithm. This
C

naturalistic study employed ongoing assessments of treatment and outcome information.


AC

For the present study, further inclusion criteria were: (i) 6 months (180 days) or

more follow-up duration; (ii) at least one depressive clinical episode during follow-up,

according to the Clinical Monitoring Form (CMF) (Sachs et al. 2002; Sachs et al. 2003)

of the STEP-BD Standard Care Pathway (SCP).

8
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

2.3. Assessments

The diagnosis of BD was assessed with the Mini International

Neuropsychiatric Interview (MINI) (Sheehan et al. 1998) and Affective Disorder

PT
Evaluation (ADE) form (Sachs et al. 2003; Sachs 2004) upon study entry together with

RI
a socio-demographic form (Demographic Form (DF)) (Sachs et al. 2003; Sachs 2004).

SC
The CMF was used as source document for key prospective outcome measures and

U
provided the progress note in the patient’s medical record (Sachs et al. 2002).
AN
Regarding symptoms of depression and mania, sums of classified scores for

depression (SUM-D) and mania (SUM-M) were calculated using the Scoring
M

Conventions and Rules for Imputation (Sachs et al. 2002).


D

2.4. Definition of “depressive episode with switch” and “patient with switch”
TE

We used the clinical status variable in the CMF form to define the occurrence
EP

of switch. A “depressive episode with switch” was defined as a depressive episode that
C

was followed by a manic, hypomanic or mixed episode within 12 weeks (84 days). The
AC

12 weeks cut-off was chosen according to previous literature (Vieta et al. 2009). The

other depressive episodes were defined as “depressive episode without switch”. Patients

with at least one “depressive episode with switch” during follow-up were defined as

“patients with switch”, otherwise they were considered “patients without switch”.

9
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

2.5 Hypothesis under analysis

The risk of switch from depression to (hypo)manic/mixed state was tested for

association with: 1) past clinical history, including previous treatments and disease

PT
course; 2) clinical-demographic characteristics at study inclusion; and 3) the clinical

RI
features of the current depressive episode, including treatment(s).

SC
2.6. Statistical analysis

U
We employed a generalized linear mixed model (GLMM) (Stroup 2012) for
AN
investigating predictors of switch, because this model allowed the inclusion of data

from all depressive episodes for each patient. A dichotomous variable coding for the
M

affective switch was used as dependent variable, and a binomial distribution was
D

applied. Identification ID Number of each patient was considered as a random effect,


TE

while the other clinical variables were considered as fixed effects.


EP

To identify possible confounders mediating the association between switch and


C

clinical variables, several regression models were constructed in a stepwise manner.


AC

Schemas of data structure and of each model are shown in Supplementary Figure S1.

Firstly, we assessed crude associations between switch and each clinical variable

(Model 1). Secondly, on the base of clinical considerations and STEP-BD design, we

included baseline demographic and socio-economic status (SES) variables and

10
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

STEP-BD time-related variables as covariates (age, gender, ethnicity, education,

employment, annual personal income, marital status, total follow-up duration, and time

from baseline to a depressive episode (Model 2)), in accordance to the criteria reported

PT
in Supplementary Figure S1. Then a combination of variables describing past clinical

RI
history was selected and included in Model 3 together with significant (see below for

SC
further details) covariates of Model 2. Similarly, a combination of variables describing

U
baseline clinical features was selected and included in Model 4 together with significant
AN
covariates of Model 3. Finally, a combination of variables describing current (at

depressive episode) clinical features was selected and included in Model 5 together with
M

significant covariates of Model 4. The variables included in each model were selected
D

according to three criteria that were applied in the following order. 1) To avoid the
TE

inclusion of ineffective variables and loss of sample size, the following selection criteria
EP

were applied: (i) p < 0.05 with a moderate effect (F-value > 5.00); (ii) available data
C

from at least 6000 depressive episodes (about 1200 patients). 2) If two or more
AC

candidate variables were similar in terms of clinical implications, the most inclusive and

clinically meaningful one was selected, e.g. history of AD treatment was selected rather

than history of sertraline treatment. 3) Stepwise combinations of the candidate variables

were added to each model (Models 2-4), and the combination minimizing both the

11
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

corrected Akaike’s Information Criterion (AICc) and the Bayesian Information Criterion

(BIC) was selected and included into the following model. The AICc and BIC are useful

criteria to choose the models that best fit the data (Stroup 2012; Vrieze 2012). Possible

PT
multicollinearity among the selected covariates was investigated. A correlation matrix of

RI
all the candidate variables was scanned for the presence of very high correlation (|r| >

SC
0.90), and the variance inflation factor and tolerance were calculated (see

U
Supplementary Figure S1). No evidence of multicollinearity among the selected
AN
covariates was found.

Regarding dichotomous variables that were found associated with the risk of
M

switch, odds ratio (OR) were also reported. Nevertheless, we underline that these OR
D

were calculated on episode-based data (not subject-based data) and did not take into
TE

consideration the effect of covariates (among which the random covariate that
EP

represented the subject).


C

For comparisons between patients with switch and patients without switch,
AC

t-test, Mann-Whitney U test and chi-square test with two-tailed p-values were used as

appropriate. All statistical analyses were performed using SPSS version 21.0 (IBM,

Tokyo, Japan) with alpha value of 0.05. A conservative alpha value was chosen because

the present analyses were based on hypotheses with previous evidence in literature.

12
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

3. Results

3.1. Participant flow

CMF data were available for 3721 patients. Of these, 946 patients were

PT
excluded because of short follow-up duration (< 6 months). Among the remaining

RI
patients, 974 patients were excluded because they did not show any depressive episode

SC
during follow-up. Further 78 patients were excluded because their baseline data from

U
ADE and/or DF forms were unavailable. Finally, three patients reported transsexual
AN
gender and were excluded because of the small size of the group. The final sample

included 1720 patients (Supplementary Figure S2).


M

Among the included patients, 1368 patients showed no switch during follow-up,
D

while 352 patients showed at least one switch (Table 1). A total of 8915 depressive
TE

episodes were available, among which 8403 (5898 from patients without switch and
EP

2505 from patients with switch) were labeled as “episodes without switch”, while 512
C

were labeled as “episodes with switch”.


AC

3.2. Baseline clinical characteristics in patients with and without switch

In patients with switch, the total and mean annual number of depressive

episodes during follow-up was higher than those of patients without switch (both p <

0.001, Table 1). Lifetime diagnosis did not differ between the groups with and without

13
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

switch (Table 1). In patients with switch a higher prevalence of females and patients

with low annual income was observed (p = 0.022 and p = 0.013, respectively). Age at

the first (hypo)manic episode in patients with switch was younger (p = 0.043).

PT
Regarding other demographic and SES characteristics at baseline, there was no

RI
significant difference between the groups.

SC
3.3. Predictors of switch

U
3.3.1. Demographic variables at baseline
AN
In Model 1, a younger age at baseline and a shorter time from baseline to the

first depressive episode were associated with the risk of switch (F = 9.891, p = 0.002, β
M

= -0.011; F = 4.151, p = 0.042, β = <-0.001, respectively; Supplementary Table S2). In


D

Model 2, after adjustment for demographic, SES, and STEP-BD time-related variables,
TE

these associations continued to be significant (F = 7.788, p = 0.005, β = -0.012; F =


EP

4.198, p = 0.041, β = <-0.001; respectively).


C

3.3.2. Past history of disease


AC

In Model 1, a younger age at onset of the first (hypo)manic episode was

associated with switch (F = 5.320, p = 0.021, β = -0.010; Supplementary Table S2),

although this association was not observed in Model 2. A higher number of previous

(hypo)manic, depressive and total episodes, and history of last year rapid cycling were

14
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

associated with the risk of switch. In Model 2, history of suicide attempt, and history of

amphetamine or opiates use showed association with switch. History of ethanol abuse

treatment was associated with switch in Model 2, although the number of observations

PT
available for this variable was relatively small. Regarding past treatment history in

RI
Model 2, lower risk of switch was associated with each type of AD treatment (all ADs,

SC
SSRIs and non-SSRIs), some specific AD treatments (bupropion, sertraline, MAOI and

U
heterocyclic ADs) and a higher number of ADs used (all ADs and non-SSRIs). Further,
AN
a higher risk of switch was associated with past history of switch (based on patient’s

attributions) caused by SSRIs and some specific drugs (fluoxetine, sertraline,


M

paroxetine). Past treatment with quetiapine or clozapine was associated with the risk of
D

switch, although the prevalence of past clozapine treatment was relatively small. Past
TE

history of psychotherapy was associated with a lower risk of switch. Among the above
EP

variables, a combination was selected to be included in Model 3 according to the criteria


C

stated above (2.5. Statistical analysis), as reported in Table 2.


AC

3.3.3. Clinical status at baseline

In Model 3, a higher number of days with elevated mood over the last 2 weeks

and a higher score of Clinical Global Impression - Bipolar disorder (CGI-BP) -

Elevation subscale in the last month were associated with switch (Supplementary Table

15
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

S3). Consistently, higher scores of SUM-M and higher scores of mania-associated

symptoms, as well as ideas of reference, were risk factors for switch. On the other hand,

dysthymia showed association with a lower risk of switch. There was no association

PT
between switch and treatments at baseline in Model 3. From the above variables, a

RI
combination was selected to be included in Model 4 according to the criteria stated

SC
above (2.5. Statistical analysis), as reported in Table 2.

U
3.3.4. Clinical variables at current depressive episode
AN
In Model 4, mood elevation and irritability over a last 4-7 day period, and a

higher percent of days with elevated mood over the past 10 days, were associated with
M

the risk of switch (Table 3). Regarding current associated symptoms at depressive
D

episodes, a higher score of SUM-M and two or more mania-associated symptoms with
TE

at least moderate severity showed association with switch. Comorbidity with panic
EP

attacks was associated with switch. Although there was no association between
C

venlafaxine treatment during the current depressive episode and switch in Model 1 (F =
AC

0.633, p = 0.426, β = 0.108, Table 3), this association was found after adjustment for

covariates in Model 4 (F = 4.310, p = 0.038, β = 0.360). Among the above variables, a

combination was selected to be included in Model 5 according to the criteria stated

above (2.5. Statistical analysis), as reported in Table 2.

16
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

3.3.5. Full-adjusted model for prediction of switch

A summary of switch predictors found by Model 5 is shown in Table 4. Mood

elevation over the last 4-7 day period before the depressive episode was associated with

PT
the risk of switch (F = 6.491, p = 0.002; for probable mood elevation, β = 0.547, p =

RI
0.012; for definite mood elevation, β = 0.738, p = 0.005). Two or more mania-associated

SC
symptoms with at least moderate severity during the depressive episode were associated

U
with switch (F = 6.687, p = 0.010, β = 0.340). Comorbidity with panic attacks during
AN
the depressive episode was associated with higher risk of switch (F = 7.376, p = 0.007,

β = 0.376). Even when the effect of past history of panic disorder (PD) was included as
M

covariate in addition to the other covariates of Model 5, this association continued to be


D

significant (F = 7.345, p = 0.007, β = 0.387).


TE

The presence of at least one of these three identified risk factors in depressive episodes
EP

followed or not by switch was reported in Table 5. When all the three risk factors were
C

present the OR of switch is 7.28 (95% CI 4.15-12.78, p<0.001). The three risk factors
AC

identified by Model 5 provided a positive predictive value for predicting switch of 0.25

and a negative predictive value of 0.96.

17
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

4. Discussion

This study investigated the clinical-demographic predictors of switch from a

depressive phase to a (hypo)manic or mixed phase in patients with BD. The large,

PT
prospective and long term STEP-BD sample (Sachs et al. 2003) was analyzed for this

RI
purpose. After adjustment for covariates, the variables associated with the risk of switch

SC
were any possible mood elevation, multiple mania-associated symptoms with at least

U
moderate severity, comorbid panic attacks, and venlafaxine treatment during the index
AN
depressive episode.

To our knowledge, this is the first large study that investigated the predictors of
M

manic switch in a prospective design and using an episode-based approach, i.e.


D

including all the ascertained depressive episodes in each patient during follow-up.
TE

Indeed, previous studies used a patient-based approach, i.e. they compared patients with
EP

switch versus patients without switch (see Supplementary Table S1), with the exception
C

of a study on a very small sample (Boerlin et al. 1998). In addition to this


AC

methodological difference, the present study used a multivariate regression approach

that allowed to estimate the effect size and independence of each variable of interest.

Crude associations (i.e. associations found using models without covariates) provided a

rough evidence considering the complexity of the switch phenomenon. Thus, variables

18
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

showing crude associations were included as covariates in multivariate models in a

stepwise fashion. This approach represents a further methodological difference

compared to a previous study performed on the same sample (Perlis et al. 2010).

PT
The association between manic symptoms during the index depressive episode

RI
and switch could represent an expected finding, since manic symptoms often

SC
accompany bipolar depressive episodes (Goldberg et al. 2009). However, manic

U
symptoms may be easily overlooked when they appear less prominent compared to
AN
depressive symptoms (Goldberg et al. 2009). Therefore, our results confirmed the

clinical importance of carefully monitoring both depressive and manic symptoms during
M

bipolar depressive episodes. Further, higher manic symptom severity at inclusion in the
D

study was also associated with switch, consistently with previous findings (Houston et
TE

al. 2007; Perlis et al. 2010).


EP

The comorbidity with panic attacks was demonstrated as a promising predictor


C

of manic switch in depressive patients with BD and it was not investigated in the
AC

previous study on the STEP-BD sample (Perlis et al. 2010). Anxiety disorders,

including PD, were reported as the most common psychiatric comorbidities among

STEP-BD patients (51.2% lifetime; 30.5% current) (Simon et al. 2004; Bowden et al.

2012). In our sample, 36.8% of patients reported a probable or definite past history of

19
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

PD at baseline. Past history of PD showed association with switch only before

adjustment for covariates (Model 1), suggesting a weaker effect size than comorbid

panic attacks during the current depressive episode (but possibly this is due to the lower

PT
accuracy of past history recollection compared to current evaluations). Anyway, the

RI
observation of more frequent previous history of PD in patients with switch supports the

SC
hypothesis that PD could share some pathogenic mechanisms with BD at high switch

U
risk rather than panic symptoms could simply be an epiphenomenon of the imminent
AN
switch. Indeed, comorbidity between PD and BD with clinical markers of severity

demonstrated familiar aggregation (Goes et al. 2012). Further, the diagnosis of PD was
M

demonstrated to increase the risk of rapid mood switching in BD (MacKinnon et al.


D

2003). Broadly speaking, higher comorbidity with anxiety disorders were previously
TE

identified as a risk factor for rapid switching in BD (Nwulia et al. 2008) and for higher
EP

mental health service use (Sala et al. 2012). Thus, current anxiety symptoms and
C

particularly comorbid panic attacks should be carefully investigated during bipolar


AC

depressive phases.

Regarding AD treatment during bipolar depressive episodes,

venlafaxine treatment showed an association with switch only after adjustment for

potential baseline risk factors of switch. Anyway, in bivariate models venlafaxine

20
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

treatment did not increase the risk of switch neither in low risk depressive episodes

(showing none of the other three risk factors identified by Model 5: OR=1.04, 95%

CI=0.83-1.83, p=0.32) nor in high risk depressive episodes (with at least one of the

PT
other three risk factors identified by Model 5: OR=0.86, 95%CI=0.56-1.33, p=0.53).

RI
Thus venlafaxine treatment may show a marginal effect on the risk of switch that is

SC
probably highly influenced by the interaction with other clinical variables. Previous

U
studies investigating mood switch during AD treatment found heterogeneous results.
AN
Previous meta-analyses suggested that the rate of treatment-emergent switch occurred

more often with tri- and tetra-cyclics (Peet 1994) and venlafaxine (Gijsman et al. 2004;
M

Sidor and Macqueen 2011) than with SSRIs and bupropion. One study (Post et al. 2006)
D

reported that venlafaxine as adjunctive therapy to mood stabilizers increased the risk of
TE

mood switch compared to bupropion or sertraline, especially in patients with a history


EP

of rapid cycling. In contrast, a retrospective study reported no difference in the


C

incidence of switch between venlafaxine, bupropion and SSRIs monotherapy in rapid


AC

cycling patients, while monotherapy with second-generation ADs as a group showed

association with treatment-induced mania in rapid cyclers (Gao et al. 2008). Given that

in the STEP-BD sample the prescription of TCAs and MAOIs was very infrequent, we

cannot exclude that this study did not find any effect of these ADs on switch risk

21
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

because of the lack of statistical power.

Several clinical-demographic variables showed evidence of association in

non-adjusted models (Models 2-5, Supplementary Figure S1). In detail, history of

PT
previous AD treatment and number of AD(s) prescribed, history of suicide attempt,

RI
history of amphetamine or opiate use, higher number of previous (hypo)manic and

SC
depressive episodes, last year rapid cycling, younger age at onset and at inclusion in the

U
study represent potential risk factors with probable low effect size that did not survive in
AN
multi-regression models. An impact of these variables should not be excluded especially

in patients with absent or mild manic symptoms and no comorbid panic attacks.
M

Younger age was associated with switch even after adjustment for demographic
D

and SES covariates including STEP-BD time-related variables. This finding is not
TE

supported by a previous quite large study investigating this variable (Serretti et al. 2003)
EP

and it is difficult to conclude whether age at baseline could be a risk factor of switch or
C

a confounder, i.e. age may impact on medication choice by clinicians. Thus, the
AC

meaning of demographic and SES variables as risk factors of switch should be

considered with caution.

According to previous evidence (Perlis et al. 2010; Valenti et al. 2012), the

present study confirmed that age at onset in patients with switch is younger than in

22
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

patients without switch. However, an earlier age at onset of the first (hypo)manic

episode showed no association with switch in Model 2. Other disease history

characteristics that were associated with mood switch were a higher number of previous

PT
depressive episodes, last year rapid cycling, and history of suicide attempt, consistently

RI
with previous results (Perlis et al. 2010).

SC
A higher number of (hypo)manic episodes during lifetime was proposed as a

U
risk factor for switch. On the other hand, a previous retrospective study suggested that a
AN
lower number of previous manic episodes may be associated with switch (Serretti et al.

2003). The difficulty derived from the retrospective collection of the number of
M

episodes may have impacted on this result explaining the inconsistency with our present
D

findings.
TE

Previous history of substance-related disorders and some previous treatments


EP

were also identified as possible risk factors for mood switch. Indeed, history of
C

amphetamine use was associated with a higher risk of switch. The effect of
AC

amphetamine use (past two months) on the risk of switch during a depressive phase

showed an OR=1.92 in the present study. Abnormalities in catecholamine levels were

associated with both spontaneous and treatment-emergent switch (Salvadore et al. 2010).

Given their promoting effect on dopamine release and the inhibition of dopamine

23
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

reuptake, amphetamines were demonstrated to precipitate hypomania in patients with

BD and induce a hypomanic-like status in healthy subjects (Jacobs and Silverstone

1986; Asghar et al. 2003). The impact of amphetamines on sleep may also mediate this

PT
effect. Regarding previous treatments, history of psychotherapy may predict lower risk

RI
of switch (OR=0.68). A recent study on the STEP-BD sample suggested that patients

SC
with excessively pessimistic or optimistic attributions may have a higher risk of manic

U
switch, despite psychotherapy per se was not investigated as a factor involved in switch
AN
risk. Thus, psychotherapy-induced cognitive flexibility may contribute to the reduction

of switch risk (Stange et al. 2013). In addition, evidence obtained thanks to STEP-BD
M

randomized interventions suggested that adjunctive psychosocial interventions but not


D

adjunctive ADs produced better outcomes than those achieved with mood stabilizers
TE

alone (Miklowitz et al. 2006; Miklowitz et al. 2007; Miklowitz et al. 2007; Bowden et
EP

al. 2012). Nevertheless, the lower frequency of psychotherapy in patients with switch
C

may be due to the tendency of avoiding this type of treatment in patients with more
AC

severe BD, resulting in a selection bias. Among pharmacological interventions, history

of previous quetiapine treatment and no history of previous AD treatment were

associated with switch. These variables most probably represent confounding factors

rather than predictors of switch. Considering the clinical guidelines for BD treatment at

24
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

STEP-BD recruitment time (Sachs et al. 2000), it is likely that clinicians would

prescribe quetiapine to depressive patients with history of previous mood switch, while

they would try an AD in depressive patients without such history.

PT
Limitations of the present study should be mentioned. Firstly, switch from

RI
depression to mania was not directly reported in the CMF form thus it was derived from

SC
clinical status changes during follow-up. The CMF clinical status was based on several

U
clinical evaluations of depressive and manic symptoms that were tested for being switch
AN
risk factors. In other words, no independent instruments were used to assess the

diagnosis of depression (that was used to build the dependent variable) and some of the
M

clinical characteristics that were used as independent variables. Secondly, some possible
D

stratification factors could have biased our results. In particular, the effect of drug doses
TE

and drug combinations was not included, as well as different bipolar spectrum
EP

diagnoses. Anyway, other studies focusing on the phenomenon of switch considered the
C

diagnosis of manic/mixed episode as the inclusion criteria rather than a definite type of
AC

bipolar spectrum disorder (Vieta et al. 2009). Finally, the three risk factors identified by

Model 5 provide a negative predictive value of 0.96 but a positive predictive value of

0.25 only, thus they are useful only to exclude the risk of switch.

25
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

5. Conclusion

Our study found that a number of past and baseline clinical-demographic

features may be risk factors of mood switch. In detail, younger age, higher number of

PT
previous (hypo)manic and depressive episodes, history of: rapid cycling, suicide attempt,

RI
amphetamine use and psychotherapy were past history characteristics that were

SC
associated with switch. More severe baseline manic symptoms and shorter time from

U
baseline to the first depressive episode were also associated with switch. After
AN
adjustment for covariates, our results suggested that three variables may be independent

predictors of manic switch, i.e. any possible mood elevation, multiple mania-associated
M

symptoms with at least moderate severity, and comorbid panic attacks during the index
D

depressive episode. Venlafaxine treatment may be a marginal risk factor for manic
TE

switch and it is probably influenced by other clinical variables. These findings may
EP

represent useful clinical information for monitoring the risk of manic switch when
C

treating depressive bipolar patients.


AC

26
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

Supplementary online materials

Supplementary Figure S1: Schemas of data structure and models, criteria for covariate

selection and multicollinearity assessment.

PT
Supplementary Figure S2: Sample flowchart.

RI
Supplementary Table S1: summary of mood switch rates and risk factors of mood

SC
switch from depression to (hypo)mania in bipolar disorder (BD) according to previous

U
literature. AN
Supplementary Table S2: Associations between manic switch and demographic,

socio-economic, STEP-BD time-related variables and clinical past history.


M

Supplementary Table S3: Associations between switch and variables describing


D

clinical status at baseline.


TE
C EP
AC

27
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

References

Asghar SJ, Tanay VA, Baker GB, Greenshaw A and Silverstone PH.
Relationship of plasma amphetamine levels to physiological,
subjective, cognitive and biochemical measures in healthy volunteers.
Hum Psychopharmacol 2003; 18: 291-299.

PT
Boerlin HL, Gitlin MJ, Zoellner LA and Hammen CL. Bipolar depression and
antidepressant-induced mania: a naturalistic study. J Clin Psychiatry

RI
1998; 59: 374-379.
Bonnin CM, Torrent C, Goikolea JM, Reinares M, Sole B, Valenti M, et al.

SC
The impact of repeated manic episodes and executive dysfunction on
work adjustment in bipolar disorder. Eur Arch Psychiatry Clin
Neurosci 2013.

U
Bottlender R, Rudolf D, Strauss A and Moller HJ. Mood-stabilisers reduce
the risk of developing antidepressant-induced maniform states in
AN
acute treatment of bipolar I depressed patients. J Affect Disord 2001;
63: 79-83.
M

Bottlender R, Sato T, Kleindienst N, Strauss A and Moller HJ. Mixed


depressive features predict maniform switch during treatment of
depression in bipolar I disorder. J Affect Disord 2004; 78: 149-152.
D

Bowden CL, Perlis RH, Thase ME, Ketter TA, Ostacher MM, Calabrese JR,
TE

et al. Aims and results of the NIMH systematic treatment


enhancement program for bipolar disorder (STEP-BD). CNS Neurosci
Ther 2012; 18: 243-249.
EP

Conus P, Macneil C and McGorry PD. Public health significance of bipolar


disorder: implications for early intervention and prevention. Bipolar
Disord 2013.
C

Fagiolini A, Forgione R, Maccari M, Cuomo A, Morana B, Dell'Osso MC, et al.


AC

Prevalence, chronicity, burden and borders of bipolar disorder. J Affect


Disord 2013; 148: 161-169.
Ferrari AJ, Saha S, McGrath JJ, Norman R, Baxter AJ, Vos T, et al. Health
states for schizophrenia and bipolar disorder within the Global
Burden of Disease 2010 Study. Popul Health Metr 2012; 10: 16.
Gao K, Kemp DE, Ganocy SJ, Muzina DJ, Xia G, Findling RL, et al.
Treatment-emergent mania/hypomania during antidepressant
monotherapy in patients with rapid cycling bipolar disorder. Bipolar

28
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

Disord 2008; 10: 907-915.


Gijsman HJ, Geddes JR, Rendell JM, Nolen WA and Goodwin GM.
Antidepressants for bipolar depression: a systematic review of
randomized, controlled trials. Am J Psychiatry 2004; 161: 1537-1547.
Goes FS, McCusker MG, Bienvenu OJ, Mackinnon DF, Mondimore FM,
Schweizer B, et al. Co-morbid anxiety disorders in bipolar disorder

PT
and major depression: familial aggregation and clinical characteristics
of co-morbid panic disorder, social phobia, specific phobia and

RI
obsessive-compulsive disorder. Psychol Med 2012; 42: 1449-1459.
Goldberg JF, Perlis RH, Bowden CL, Thase ME, Miklowitz DJ, Marangell LB,

SC
et al. Manic symptoms during depressive episodes in 1,380 patients
with bipolar disorder: findings from the STEP-BD. Am J Psychiatry
2009; 166: 173-181.

U
Houston JP, Lipkovich IA, Ahl J, Rotelli MD, Baker RW and Bowden CL.
Initial symptoms of manic relapse in manic or mixed-manic bipolar
AN
disorder: post hoc analysis of patients treated with olanzapine or
lithium. J Psychiatr Res 2007; 41: 616-621.
M

Jacobs D and Silverstone T. Dextroamphetamine-induced arousal in human


subjects as a model for mania. Psychol Med 1986; 16: 323-329.
Keck PE, Jr., Corya SA, Altshuler LL, Ketter TA, McElroy SL, Case M, et al.
D

Analyses of treatment-emergent mania with olanzapine/fluoxetine


TE

combination in the treatment of bipolar depression. J Clin Psychiatry


2005; 66: 611-616.
Kleine-Budde K, Touil E, Moock J, Bramesfeld A, Kawohl W and Rossler W.
EP

Cost of illness for bipolar disorder: a systematic review of the economic


burden. Bipolar Disord 2013.
Leverich GS, Altshuler LL, Frye MA, Suppes T, McElroy SL, Keck PE, Jr., et
C

al. Risk of switch in mood polarity to hypomania or mania in patients


AC

with bipolar depression during acute and continuation trials of


venlafaxine, sertraline, and bupropion as adjuncts to mood stabilizers.
Am J Psychiatry 2006; 163: 232-239.
MacKinnon DF, Potash JB, McMahon FJ, Simpson SG, Depaulo JR, Jr.,
Zandi PP, et al. Rapid mood switching and suicidality in familial
bipolar disorder. Bipolar Disord 2005; 7: 441-448.
MacKinnon DF, Zandi PP, Gershon E, Nurnberger JI, Jr., Reich T and
DePaulo JR. Rapid switching of mood in families with multiple cases

29
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

of bipolar disorder. Arch Gen Psychiatry 2003; 60: 921-928.


MacKinnon DF, Zandi PP, Gershon ES, Nurnberger JI, Jr. and DePaulo JR,
Jr. Association of rapid mood switching with panic disorder and
familial panic risk in familial bipolar disorder. Am J Psychiatry 2003;
160: 1696-1698.
Maj M, Pirozzi R, Magliano L and Bartoli L. The prognostic significance of

PT
"switching" in patients with bipolar disorder: a 10-year prospective
follow-up study. Am J Psychiatry 2002; 159: 1711-1717.

RI
Miklowitz DJ, Otto MW, Frank E, Reilly-Harrington NA, Kogan JN, Sachs
GS, et al. Intensive psychosocial intervention enhances functioning in

SC
patients with bipolar depression: results from a 9-month randomized
controlled trial. Am J Psychiatry 2007; 164: 1340-1347.
Miklowitz DJ, Otto MW, Frank E, Reilly-Harrington NA, Wisniewski SR,

U
Kogan JN, et al. Psychosocial treatments for bipolar depression: a
1-year randomized trial from the Systematic Treatment Enhancement
AN
Program. Arch Gen Psychiatry 2007; 64: 419-426.
Miklowitz DJ, Otto MW, Wisniewski SR, Araga M, Frank E,
M

Reilly-Harrington NA, et al. Psychotherapy, symptom outcomes, and


role functioning over one year among patients with bipolar disorder.
Psychiatr Serv 2006; 57: 959-965.
D

Nemeroff CB, Evans DL, Gyulai L, Sachs GS, Bowden CL, Gergel IP, et al.
TE

Double-blind, placebo-controlled comparison of imipramine and


paroxetine in the treatment of bipolar depression. Am J Psychiatry
2001; 158: 906-912.
EP

Nwulia EA, Zandi PP, McInnis MG, DePaulo JR, Jr. and MacKinnon DF.
Rapid switching of mood in families with familial bipolar disorder.
Bipolar Disord 2008; 10: 597-606.
C

Peet M. Induction of mania with selective serotonin re-uptake inhibitors and


AC

tricyclic antidepressants. Br J Psychiatry 1994; 164: 549-550.


Perlis RH, Ostacher MJ, Goldberg JF, Miklowitz DJ, Friedman E, Calabrese
J, et al. Transition to mania during treatment of bipolar depression.
Neuropsychopharmacology 2010; 35: 2545-2552.
Perlis RH, Ostacher MJ, Patel JK, Marangell LB, Zhang H, Wisniewski SR,
et al. Predictors of recurrence in bipolar disorder: primary outcomes
from the Systematic Treatment Enhancement Program for Bipolar
Disorder (STEP-BD). Am J Psychiatry 2006; 163: 217-224.

30
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

Post RM, Altshuler LL, Leverich GS, Frye MA, Nolen WA, Kupka RW, et al.
Mood switch in bipolar depression: comparison of adjunctive
venlafaxine, bupropion and sertraline. Br J Psychiatry 2006; 189:
124-131.
Sachs GS. Strategies for improving treatment of bipolar disorder: integration
of measurement and management. Acta Psychiatr Scand Suppl

PT
2004;7-17.
Sachs GS, Guille C and McMurrich SL. A clinical monitoring form for mood

RI
disorders. Bipolar Disord 2002; 4: 323-327.
Sachs GS, Nierenberg AA, Calabrese JR, Marangell LB, Wisniewski SR,

SC
Gyulai L, et al. Effectiveness of adjunctive antidepressant treatment
for bipolar depression. N Engl J Med 2007; 356: 1711-1722.
Sachs GS, Printz DJ, Kahn DA, Carpenter D and Docherty JP. The Expert

U
Consensus Guideline Series: Medication Treatment of Bipolar
Disorder 2000. Postgrad Med 2000; Spec No: 1-104.
AN
Sachs GS, Thase ME, Otto MW, Bauer M, Miklowitz D, Wisniewski SR, et al.
Rationale, design, and methods of the systematic treatment
M

enhancement program for bipolar disorder (STEP-BD). Biol


Psychiatry 2003; 53: 1028-1042.
Sala R, Goldstein BI, Morcillo C, Liu SM, Castellanos M and Blanco C.
D

Course of comorbid anxiety disorders among adults with bipolar


TE

disorder in the U.S. population. J Psychiatr Res 2012; 46: 865-872.


Salvadore G, Quiroz JA, Machado-Vieira R, Henter ID, Manji HK and Zarate
CA, Jr. The neurobiology of the switch process in bipolar disorder: a
EP

review. J Clin Psychiatry 2010; 71: 1488-1501.


Serretti A, Artioli P, Zanardi R and Rossini D. Clinical features of
antidepressant associated manic and hypomanic switches in bipolar
C

disorder. Prog Neuropsychopharmacol Biol Psychiatry 2003; 27:


AC

751-757.
Sheehan DV, Lecrubier Y, Sheehan KH, Amorim P, Janavs J, Weiller E, et al.
The Mini-International Neuropsychiatric Interview (M.I.N.I.): the
development and validation of a structured diagnostic psychiatric
interview for DSM-IV and ICD-10. J Clin Psychiatry 1998; 59 Suppl
20: 22-33;quiz 34-57.
Sidor MM and Macqueen GM. Antidepressants for the acute treatment of
bipolar depression: a systematic review and meta-analysis. J Clin

31
Niitsu et al. Predictors of manic switch in bipolar depression
ACCEPTED MANUSCRIPT

Psychiatry 2011; 72: 156-167.


Simon NM, Otto MW, Wisniewski SR, Fossey M, Sagduyu K, Frank E, et al.
Anxiety disorder comorbidity in bipolar disorder patients: data from
the first 500 participants in the Systematic Treatment Enhancement
Program for Bipolar Disorder (STEP-BD). Am J Psychiatry 2004; 161:
2222-2229.

PT
Stange JP, Sylvia LG, Magalhaes PV, Frank E, Otto MW, Miklowitz DJ, et al.
Extreme attributions predict transition from depression to mania or

RI
hypomania in bipolar disorder. J Psychiatr Res 2013; 47: 1329-1336.
Stroup WW. Generalized Linear Mixed Models, Modern Concepts, Methods

SC
and Applications. USA, CRC Press 2012.
Uher R, Mantere O, Suominen K and Isometsa E. Typology of clinical course
in bipolar disorder based on 18-month naturalistic follow-up. Psychol

U
Med 2013; 43: 789-799.
Valenti M, Pacchiarotti I, Bonnin CM, Rosa AR, Popovic D, Nivoli AM, et al.
AN
Risk factors for antidepressant-related switch to mania. J Clin
Psychiatry 2012; 73: e271-276.
M

van Rossum I, Haro JM, Tenback D, Boomsma M, Goetz I, Vieta E, et al.


Stability and treatment outcome of distinct classes of mania. Eur
Psychiatry 2008; 23: 360-367.
D

Vieta E, Angst J, Reed C, Bertsch J, Haro JM and board Ea. Predictors of


TE

switching from mania to depression in a large observational study


across Europe (EMBLEM). J Affect Disord 2009; 118: 118-123.
Vrieze SI. Model selection and psychological theory: a discussion of the
EP

differences between the Akaike information criterion (AIC) and the


Bayesian information criterion (BIC). Psychol Methods 2012; 17:
228-243.
C

Whiteford HA, Degenhardt L, Rehm J, Baxter AJ, Ferrari AJ, Erskine HE, et
AC

al. Global burden of disease attributable to mental and substance use


disorders: findings from the Global Burden of Disease Study 2010.
Lancet 2013; 382: 1575-1586.

32
ACCEPTED MANUSCRIPT

Table 1. Demographic data for patients with and without switch.


Total sample Patients without switch Patients with switch Statistics df P
(N = 1720) (N = 1368) (N = 352)

PT
Total number of analyzed depressive 8915 5898 3017
episodes, episode

RI
Without switch, episode 8403 5898 2505
With switch, episode 512 - 512

SC
Total number of depressive episodes 4.92±6.57 (1720) 4.31±5.30 (1368) 8.57±8.69 (352) Z = 12.52 - <0.001***
during follow-up period, episode

U
Total follow-up duration, day 762.3±422.5 (1720) 743.1±417.4 (1368) 836.9±434.6 (352) t = -3.73 1718 <0.001***

AN
Mean annual number of depressive 2.92±3.15 (1720) 2.63±2.97 (1368) 4.07±3.53 (352) Z = 10.08 - <0.001***
episodes during follow-up period,

M
episode/year
Lifetime diagnosis

D
Bipolar I 64.5 (1109) 77.8 (863) 22.2 (246) χ2 = 5.67 3 0.129

TE
Bipolar II 29.0 (499) 82.6 (412) 17.4 (87)
Bipolar NOS 5.1 (88) 83.0 (73) 17.0 (15)
EP
Schizoaffective Bipolar type 1.4 (24) 83.3 (20) 16.7 (4)
Gender χ2 = 5.26 1 0.022*
C

Male 40.0 (688) 82.3 (566) 17.7 (122)


AC

Female 60.0 (1032) 77.7 (802) 22.3 (230)


Age, year 40.8±12.3 (1720) 41.1±12.5 (1368) 39.8±11.5 (352) t = 1.97 583.8 0.050
Age at onset of mood disorder, year 16.9±8.5 (1691) 17.1±8.7 (1345) 16.3±7.8 (346) t = 1.81 585.0 0.071
2
Polarity at onset of mood disorder χ = 1.13 2 0.568
ACCEPTED MANUSCRIPT

Depression 58.0 (980) 80.1 (785) 19.9 (195)


a
Mixed 19.1 (323) 77.4 (250) 22.6 (73)
(Hypo)mania 22.9 (388) 79.9 (310) 20.1 (78)

PT
Duration of illness as mood disorder, year 23.9±12.6 (1691) 24.0±12.9 (1345) 23.4±11.4 (346) t = 0.88 593.9 0.377
Age at onset of (hypo)manic episode, year 21.7±10.2 (1623) 22.0±10.4 (1288) 20.7±9.6 (335) t = 2.03 1621 0.043*

RI
b
Duration of illness as bipolar disorder , 19.0±12.4 (1623) 19.1±12.6 (1288) 18.7±11.6 (335) t = 0.61 556.7 0.541
year

SC
Marital status χ2 = 0.59 2 0.744
Never married 34.4 (584) 79.8 (466) 20.2 (118)

U
Married/living as married 38.8 (659) 78.8 (519) 21.2 (140)

AN
Separated/divorced/widowed 26.8 (454) 80.6 (366) 19.4 (88)
Race/ethnicity χ2 = 0.18 1 0.674

M
White/Caucasian 91.7 (1578) 79.7 (1257) 20.3 (321)
Others 8.3 (142) 78.2 (111) 21.8 (31)

D
Education χ2 = 2.34 1 0.127

TE
Technical School or less 54.0 (913) 78.2 (714) 21.8 (199)
College Diploma or more 46.0 (777) 81.2 (631) 18.8 (146)
EP
Employment status χ2 = 0.92 1 0.337
Others 69.7 (1174) 79.1 (929) 20.9 (245)
C

Full-time 30.3 (510) 81.2 (414) 18.8 (96)


AC

Annual personal income χ2 = 6.17 1 0.013*


Less than $20,000 62.9 (1038) 77.7 (807) 22.3 (231)
$20,000 or more 37.1 (612) 82.8 (507) 17.2 (105)
Medical insurance χ2 = 0.04 1 0.838
ACCEPTED MANUSCRIPT

No 15.9 (268) 80.2 (215) 19.8 (53)


Yes 84.1 (1417) 79.7 (1129) 20.3 (288)
Family history of bipolar illness χ2 = 0.02 1 0.888

PT
No 9.0 (151) 80.1 (121) 19.9 (30)
Yes 91.0 (1533) 79.6 (1221) 20.4 (312)

RI
Values represent mean ± SD (N) for continuous variables, and % (N) for categorical variables (vertical % values are reported in the column total sample and row % are
used to describe each categorical variable in the switch and non-switch groups). N represents number of patients. p < 0.05*, p < 0.01**, p < 0.001***

SC
Patients with at least one depressive episode with following manic switch during follow-up period were defined as “patients with switch, while the other patients were as
“patients without switch”.

U
a, Both depressive and manic episodes were diagnosed within the year at onset of mood disorder. b, Duration since the first (hypo)manic episode was diagnosed.

AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT
Table 2. Summary of the covariates selected from each Model.
Fixed Effect Fixed Coefficient
F P β (SE) t 95% CI P
Past history variables selected from Model 2 and included in Model 3
Number of (hypo)manic episodes in 6.267 0.012 0.249 (0.100) 2.503 0.054 – 0.445 0.012
lifetime (0-9 times/ 10 or more)
Number of depressive episodes in 9.176 0.002 0.336 (0.111) 3.029 0.118 – 0.553 0.002

PT
lifetime (0-9 times/ 10 or more)
History of the last-year rapid 6.982 0.008 0.276 (0.105) 2.642 0.071 – 0.482 0.008

RI
cycling (No/Yes)
Suicide attempt (No/Yes) 5.728 0.017 0.218 (0.091) 2.393 0.039 – 0.397 0.017
Amphetamine use (No/Yes) 5.710 0.017 0.515 (0.216) 2.390 0.093 – 0.937 0.017

SC
Antidepressants treatment (No/Yes) 6.407 0.011 -0.359 (0.142) -2.531 -0.637 – -0.081 0.011
Quetiapine treatment (No/Yes) 7.452 0.006 0.300 (0.110) 2.730 0.085 – 0.515 0.006

U
Psychotherapy (No/Yes) 5.639 0.018 -0.220 (0.093) -2.375 -0.401 – -0.038 0.018
Clinical status variables at baseline selected from Model 3 and included in Model 4
AN
CGI-BP-Elevation subscale in the 14.036 <0.001 0.154 (0.154) 3.746 0.073 – 0.234 <0.001
last month
M

SUM-M at baseline 18.184 <0.001 0.109 (0.026) 4.264 0.059 – 0.159 <0.001
Clinical status variables at depressive episodes selected from Model 4 and included in Model 5
Mood elevation over a last 4-7 day 9.606 <0.001
D

period
TE

No -
Probable 0.682 (0.210) 3.248 0.271 – 1.094 0.001
Definite 0.838 (0.263) 3.186 0.323 – 1.354 0.001
EP

Multiple mania-associated 12.566 <0.001 0.449 (0.127) 3.545 0.201 – 0.698 <0.001
symptoms with at least moderate
severity (No/Yes)
C

Panic attacks (No/Yes) 10.796 0.001 0.448 (0.136) 3.286 0.181 – 0.715 0.001
AC

Abbreviations: CGI-BP, Clinical Global Impression scale-Bipolar disorder; SUM-M, Summary of classified scores
for mania.
The covariates in Model 2 consist of demographic, socio-economic status and the STEP-BD time-related variables;
age, gender, ethnicity, education, employment, annual personal income, marital status, total follow-up duration, and
time from baseline to a depressive episode. The combination of variables in each model was selected and included
together with the covariates of each model into the next step model. For each variable, F values indicate the effect size
and beta coefficients show whether the relationship between the variable and switch risk is positive or negative.
ACCEPTED MANUSCRIPT

Table 3. Associations between manic switch and variables of current clinical status at depressive episodes.
Total No. of Episodes Episodes with Model 1 (Crude) Model 4a (Baseline adjusted)
Patients / without switch switch

PT
Episodes
Fixed Effect Fixed Coefficient Fixed Effect Fixed Coefficient

RI
(N = 8403) (N = 512) F P β P F P β P
Current clinical symptoms at depressive episode

SC
% of days with elevated 1720/8908 4.0±13.6 (8397) 11.2±21.7 (511) 81.836 <0.001* 0.018 <0.001* 32.789 <0.001 0.016 <0.001**

U
mood over the past 10 days ** ** *** *

AN
Mood elevation over a last 1718/8890 23.655 <0.001* 9.606 <0.001
4-7 day period ** ***

M
No 94.8 (7917) 5.2 (435) - -
Probable 87.6 (319) 12.4 (45) 0.769 <0.001* 0.682 0.001**

D
**

TE
Definite 83.3 (145) 16.7 (29) 1.112 <0.001* 0.838 0.001**
**
EP
% of days with irritability 1720/8906 36.3±39.2 (8396) 46.6±39.9 (510) 17.798 <0.001* 0.004 <0.001* 3.588 0.058 0.003 0.058
over the past 10 days ** **
C

Irritability over a last 4-7 1718/8859 7.701 <0.001* 4.536 0.011*


AC

day period **
No 95.1 (5419) 4.9 (279) - -
Probable 94.2 (1123) 5.8 (69) 0.116 0.372 -0.021 0.902
Definite 91.9 (1810) 8.1 (159) 0.386 <0.001* 0.364 0.004**
**
ACCEPTED MANUSCRIPT

Current associated symptoms at depressive episode

SUM_D# in CMF 1707/8664 7.72±2.11 (8170) 8.09±2.30 (494) 7.759 0.005** 0.054 0.005** 1.498 0.221 0.032 0.221

PT
Psychomotor agitation as a 1717/8856 0.31±0.43 (8350) 0.42±0.48 (506) 17.025 <0.001* 0.378 <0.001* 7.349 0.007** 0.328 0.007**
depressive symptom ** **

RI
Suicidal Ideation 1720/8895 0.37±0.43 (8384) 0.43±0.46 (511) 5.380 0.020* 0.221 0.020* 0.090 0.765 0.040 0.765
Sleep 1720/8901 0.88±0.52 (8390) 0.97±0.54 (511) 6.941 0.008** 0.215 0.008** 2.829 0.093 0.180 0.093

SC
Appetite 1720/8905 0.61±0.51 (8393) 0.66±0.52 (512) 2.767 0.096 0.139 0.096 0.425 0.515 -0.071 0.515
#
SUM_M in CMF 1714/8782 1.14±1.14 (8280) 1.67±1.46 (502) 57.770 <0.001* 0.246 <0.001* 15.665 <0.001 0.175 <0.001**

U
** ** *** *

AN
Need for sleep 1720/8904 0.04±0.19 (8392) 0.07±0.26 (512) 8.100 0.004** 0.492 0.004** 0.422 0.516 0.168 0.516
Talking 1720/8908 0.06±0.19 (8396) 0.12±0.28 (512) 38.321 <0.001* 1.009 <0.001* 4.859 0.028* 0.505 0.028*

M
** **
Flight of Ideas 1719/8905 0.19±0.37 (8393) 0.35±0.48 (512) 51.573 <0.001* 0.671 <0.001* 16.608 <0.001 0.525 <0.001**

D
** ** *** *

TE
Goal Directed Activity 1718/8858 0.51±0.52 (8391) 0.56±0.54 (511) 25.317 <0.001* 0.948 <0.001* 5.688 0.017* 0.628 0.017*
** **
EP
Psychomotor agitation as a 1716/8851 0.26±0.41 (8343) 0.38±0.47 (508) 23.607 <0.001* 0.448 <0.001* 9.821 0.002** 0.386 0.002**
manic symptom ** **
C

High Risk Behavior 1719/8885 0.04±0.17 (8375) 0.09±0.26 (510) 28.670 <0.001* 0.933 <0.001* 7.697 0.006** 0.640 0.006**
AC

** **
Number of mania-associated 1720/8915 0.76±0.93 (8403) 1.11±1.17 (512)
symptoms with at least
##
moderate severity
1 or more symptoms 6.961 0.008** 0.956 0.328
ACCEPTED MANUSCRIPT

No 95.2 (4141) 4.8 (210) - -


Yes (1 or more) 93.4 (4262) 6.6 (302) 0.227 0.008** 0.109 0.328
Multiple symptoms 39.753 <0.001* 12.566 <0.001

PT
** ***
No (no or 1) 95.3 (6804) 4.7 (338) - -

RI
Yes (2 or more) 90.2 (1599) 9.8 (174) 0.592 <0.001* 0.449 <0.001**
** *

SC
Current psychoactive substance use and associated features at depressive episode

U
Caffeine use, cups/day 1676/8288 1.41±2.01 (7811) 1.67±2.17 (477) 4.596 0.032* 0.040 0.032* 1.459 0.227 0.029 0.227

AN
Nicotine use, packs/day 1677/8316 0.30±0.57 (7840) 0.36±0.66 (476) 2.727 0.099 0.119 0.099 2.847 0.092 0.158 0.092
Panic attacks 1712/8768 10.967 0.001** 10.796 0.001**

M
No 94.8 (6899) 5.2 (382) - -
Yes 91.9 (1366) 8.1 (121) 0.347 0.001** 0.448 0.001**

D
Paranoid Ideation severity 1717/8846 0.13±0.46 (8335) 0.18±0.57 (511) 2.714 0.100 0.134 0.100 1.834 0.176 0.139 0.176

TE
Hallucinations severity 1717/8838 0.06±0.32 (8328) 0.10±0.44 (510) 5.358 0.021* 0.241 0.021* 1.967 0.161 0.185 0.161
Delusions severity 1715/8832 0.03±0.26 (8323) 0.06±0.33 (509) 2.930 0.087 0.218 0.087 0.710 0.399 0.125 0.399
EP
Current treatment at depressive episode

Valproate 1720/8915 3.421 0.064 0.013 0.908


C

No 94.6 (6168) 5.4 (350) - -


AC

Yes 93.2 (2235) 6.8 (162) 0.172 0.064 0.014 0.908


Number of Current Mood 1720/8915 1.40±0.84 (8403) 1.53±0.93 (512) 6.684 0.010* 0.128 0.010* 3.078 0.079 0.113 0.079
Stabilizers (range: 0-6)
Venlafaxine 1720/8915 0.633 0.426 4.310 0.038*
ACCEPTED MANUSCRIPT

No 94.3 (7543) 5.7 (452) - -


Yes 93.5 (860) 6.5 (60) 0.108 0.426 0.360 0.038*
Values represent mean ± SD (N) for continuous variables, and % (N) for categorical variables. N represents number of depressive episodes.

PT
Variables with statistical significance (p < 0.05*, p < 0.01**, p < 0.001***) or trend (p < 0.10) in Model 1 or 4 are reported.
Abbreviations: CMF, Clinical Monitoring Form; SUM-D, Summary of classified scores for depression; SUM-M, Summary of classified scores for mania.

RI
a Sample size in Model 4: number of patients = 1052; number of depressive episodes = 5215.
#, SUM-D and SUM-M were calculated from the data “Associated Symptoms for the Past Week” using the imputation methods by Sachs et al. (2002).

SC
##, Number of mania-associated symptoms with at least moderate severity was counted from the data of mood elevation items in “Associated Symptoms for the Past Week”
using the imputation methods by Sachs et al. (2002). For each variable, F values indicate the effect size and beta coefficients show whether the relationship between the variable

U
and switch risk is positive or negative.

AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT
Table 4. Summary of predictors for manic switch at depressive episodes in patients with bipolar
disorder (Model 5).
Fixed Effect Fixed Coefficient
F P β (SE) t 95% CI P
Mood elevation over a last 4-7 day 6.491 0.002
period
No -

PT
Probable 0.547 (0.217) 2.519 0.121 – 0.972 0.012
Definite 0.738 (0.265) 2.784 0.218 – 1.258 0.005

RI
Multiple mania-associated symptoms 6.687 0.010 0.340 (0.131) 2.586 0.082 – 0.597 0.010
with at least moderate severity
(No/Yes)

SC
Panic attacks (No/Yes) 7.376 0.007 0.376 (0.138) 2.716 0.105 – 0.647 0.007
Sample size in the final Model 5: number of patients = 1052; number of depressive episodes = 5116.

U
The covariates (fixed effects) in the final Model 5 consist of the covariates shown in Table 2 and its annotation.
Identification ID Number of each patient was considered as a random effect. For each variable, F values indicate the
AN
effect size and beta coefficients show whether the relationship between the variable and switch risk is positive or
negative.
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT
Table 5. Distribution of the three predictors of switch identified by Model 5 split by
switch group.
No switch Switch OR 95% CI P

Depressive episodes with 2.080 1.735-2.492 < 0.001***


None of the three predictors 95.7 (5546) 4.3 (249)
At least one of the three predictors 91.5 (2699) 8.5 (252)
Depressive episodes with 7.282 4.151-12.775 < 0.001***

PT
None of the three predictors 95.7 (5546) 4.3 (249)
All of the three predictors 75.4 (52) 24.6 (17)

RI
Values represent % (N) for categorical variables. N represents the number of depressive episodes.
The three identified predictors of switch at depressive episodes are any possible mood elevation, multiple
mania-associated symptoms with at least moderate severity, and panic attacks.

U SC
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

Acknowledgements

Data and biomaterials were collected for the Systematic Treatment

Enhancement Program for Bipolar Disorder (STEP-BD), a multi-center, longitudinal

PT
(5-8 year) project selected from responses to RFP #NIMH-98-DS-0001, “Treatment for

RI
Bipolar Disorder.” The project was led by Gary Sachs, M.D., and coordinated by

SC
Massachusetts General Hospital in Boston, MA. The NIMH grant number was

U
2N01MH080001-001. AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

Conflict of Interest

Dr. Serretti is or has been a consultant / speaker for: Abbott, Astra Zeneca,

Clinical Data, Boheringer, Bristol Myers Squibb, Eli Lilly, GlaxoSmithKline, Janssenn,

PT
Lundbeck, Pfizer, Sanofi, and Servier. The other authors have no conflict of interest to

RI
declare.

U SC
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

Contributors

Dr. Niitsu prepared the dataset, performed statistical analyses, and wrote the

first draft of the manuscript. Dr. Fabbri and Dr. Serretti assisted in the preparation of the

PT
dataset and statistical analyses and revised the manuscript. All authors have approved

RI
the final manuscript.

U SC
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

Role of Funding Source

The work of Dr. Niitsu was supported by a grant for research abroad from

SENSHIN Medical Research Foundation, Japan. The funding body had no role in the

PT
study design, the collection of data, the writing of this report and decision to submit it

RI
for publication.

U SC
AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

Supplementary Table S1: summary of mood switch rates and risk factors of mood switch from depression to (hypo)mania in bipolar disorder (BD) according to
previous literature. AD=antidepressant medication. PL=placebo. * The number and percent of patients with mood switch were reported. § These studies were
specifically focused on the study of switch risk factors after the introduction of an antidepressant treatment. # These studies were specifically focused on the
study of switch risk associated with antidepressant medications.

PT
Study Design Diagnosis Sample size Switch* Risk factors
(Serretti et al. Observational BD I and II 416 169 (40.6%) Older age, absence of delusions, depressive
2003) retrospective polarity at onset, less previous manic episodes,

RI
previous history of switch, BD I, no maintenance
with mood stabilizers

SC
(Valenti et al. Experimental BD I and II 221 54 (24.4%) Earlier age at onset, higher rate of previous
2012) § prospective switches, lower rate of AD response
(Perlis et al. Observational BD I and II 2166 - 461 (21.3%) in the whole group Younger onset age in patients not treated with

U
2010) retrospective - 289 (19.6%) in the group treated with ADs, greater number of past depressive episodes,
an AD greater manic symptom severity, recent or

AN
lifetime rapid cycling, alcohol use disorder,
previous suicide attempt in AD-exposed patients,
history of switch while treated with ADs, BD I in

M
AD-exposed patients
(Altshuler et al. Experimental BD I and II 184 17 (9.2%) BD I

D
2006) § prospective
(Leverich et al. Experimental BD I, II and 159 - 31 (19.3%) during acute phase (10 BD I, mood stabilizer + venlafaxine compared to

TE
2006) § prospective BD not weeks) mood stabilizer + sertraline or bupropion
otherwise - 58 (36.8%) during continuation phase
specified
EP
(Bottlender et al. Observational BD I 158 39 (24.7%) Number of mixed depressive symptoms at
2004) and retrospective baseline, treatment with tricyclic ADs, no
(Bottlender et al. treatment with a mood stabilizer
C

2001)
AC

(Boerlin et al. Observational BD I 29 with a 22 episodes (28%) Higher number of previous manic episodes,
1998) retrospective total of 79 tricyclic ADs and monoamine oxidase inhibitors
depressive compared to fluoxetine
episodes
(Henry et al. Observational BD I and II 44 12 (27.3%) High hyperthymia score (Semistructured
2001) prospective Affective Temperament Interview) and treatment
with a mood stabilizer different from lithium
(Post et al. 2006)# RCT BD I, II and 174 - 2 (4%) for bupropion + mood stabilizer Mood stabilizer + venlafaxine compared to mood
ACCEPTED MANUSCRIPT

BD not - 4 (7%) for sertraline + mood stabilizer stabilizer + sertraline or bupropion


otherwise - 10 (15%) for venlafaxine + mood
specified stabilizer
(Sachs et al. RCT BD I and II 366 - 18 (10.1%) for mood stabilizer + AD Adjunctive AD does not increase switch risk
2007)# - 20 (10.7%) for mood stabilizer + PL

PT
(Nemeroff et al. RCT BD I and II 117 - 3 (7.7%) for mood stabilizer + Imipramine compared with paroxetine
2001)# imipramine
- 1 (2.3%) for mood stabilizer + PL

RI
- 0 for mood stabilizer + paroxetine
(Keck et al. RCT BD I 833 - 21 (5.7%) for olanzapine /
2005)# - 25 (6.7%) for PL

SC
- 5 (6.4%) for olanzapine + fluoxetine
(Carlson et al. Observational BD I with 123 17 (22.4%) /

U
2007) retrospective psychotic
features

AN
M
D
TE
C EP
AC
ACCEPTED MANUSCRIPT

Supplementary Table S2. Associations between manic switch and demographic, socio-economic, STEP-BD time-related variables and clinical
past history.
Total Total Episodes Episodes with Model 1 (Crude) Model 2a (Demographic adjusted)

PT
No. of No. of without switch switch
Patien Episo

RI
ts des
Fixed Effect Fixed Coefficient Fixed Effect Fixed Coefficient

SC
(N = 8403) (N = 512) F P Estim P F P Estim P
ate (β) ate (β)

U
Demographic data and socio-economic status at baseline

AN
Age, year 1720 8915 42.3 ± 12.0 40.0 ± 11.3 9.891 0.002** -0.011 0.002** 7.788 0.005** -0.012 0.005**

M
(8403) (512)
Gender 1720 8915 0.942 0.332 0.081 0.776

D
Male 94.7 (3302) 5.3 (186) - -

TE
Female 94.0 (5101) 6.0 (326) 0.086 0.332 0.027 0.776
Ethnicity 1720 8915 0.075 0.784 0.335 0.563
EP
White 94.3 (7720) 5.7 (468) -
Others 93.9 (683) 6.1 (44) 0.042 0.784 -0.097 0563
C

Education 1690 8762 0.489 0.485 0.045 0.832


AC

Technical School or less 94.1 (4501) 5.9 (284) - -


College Diploma or more 94.5 (3760) 5.5 (217) -0.061 0.485 -0.020 0.832
Employment status 1684 8739 0.007 0.936 0.188 0.665
Others 94.3 (6134) 5.7 (371) - -
ACCEPTED MANUSCRIPT

Full-time 94.4 (2108) 5.6 (126) -0.008 0.936 0.055 0.665


Annual Personal Income 1650 8536 0.757 0.384 0.629 0.428
Less than $20,000 94.1 (5342) 5.9 (334) - -

PT
$20,000 or more 94.8 (2710) 5.2 (150) -0.082 0.384 -0.097 0.428
Marital status 1697 8782 0.076 0.927 0.747 0.474

RI
Never Married 94.2 (2866) 5.8 (177) - -
Married 94.2 (3233) 5.8 (199) -0.002 0.982 0.140 0.224

SC
-0.040 0.724 0.103 0.421
94.5 (2180) 5.5 (127)
Separated/divorced/widowed

U
Total follow-up duration, day 1720 8915 894.3 ± 434.1 877.9 ± 435.5 0.367 0.545 -0.000 0.545 1.016 0.313 0.000 0.313

AN
(8403) (512)
Time from baseline to a 1720 8915 395.4 ± 355.8 349.1 ± 346.4 4.151 0.042* -0.000 0.042* 4.198 0.041* -0.000 0.041*

M
depressive episode (CMF (8403) (512)
assessment), day

D
Age at onset and duration of illness

Age at onset of mood 1691 8793 17.6 ±


TE
8.8 16.6 ± 8.3 2.778 0.096 -0.009 0.096 0.307 0.580 -0.003 0.580
EP
disorder, year (8294) (504)
Duration of illness as mood 1691 8793 24.8 ± 12.2 23.3 ± 11.3 3.479 0.062 -0.007 0.062 0.307 0.580 0.003 0.580
C

disorder, year (8289) (504)


AC

Age at onset of (hypo)manic 1623 8452 22.4 ± 10.4 20.8 ± 9.7 5.320 0.021* -0.010 0.021* 0.706 0.401 -0.004 0.401
episode, year (7964) (488)
Number of episodes before baseline
ACCEPTED MANUSCRIPT

No. of (hypo)manic episodes 1523 7640 5.592 0.018* 6.267 0.012*


in lifetime
0-9 times 95.1 (2821) 4.9 (144) - -

PT
10 or more 93.3 (4364) 6.7 (311) 0.227 0.018* 0.249 0.012*
No. of (hypo)mania in the 1565 8150 2.97 ± 7.02 3.76 ± 6.56 3.499 0.061 0.010 0.061 1.657 0.198 0.007 0.198

RI
last year (7690) (460)
No. of depressive episodes in 1534 7670 10.866 0.001** 9.176 0.002**

SC
lifetime
0-9 times 95.6 (2127) 4.4 (97) - -

U
10 or more 93.3 (5080) 6.7 (366) 0.461 0.001** 0.336 0.002**

AN
No. of phases in lifetime 1127 5733 8.007 0.005** 6.586 0.010*
0-52 times 95.0 (3501) 5.0 (185) - -

M
53 or more 92.3 (1895) 7.4 (152) 0.409 0.005** 0.384 0.010*
History of the last-year rapid 1353 6891 9.332 0.002** 6.982 0.008**

D
cycling (No. of phases in the

TE
last year)
No (0-3 times) 95.0 (4438) 5.0 (234) - -
EP
Yes (>= 4 times) 92.5 (2053) 7.5 (166) 0.304 0.002** 0.276 0.008**
Estimated rapid cycling in 1099 5521 4.861 0.028* 3.155 0.076
C

every year (No. of phases in


AC

every year)
No (0-3 times) 95.0 (3501) 5.0 (185) - -
Yes (>= 4 times) 92.6 (1895) 7.4 (152) 0.245 0.028* 0.203 0.076
ACCEPTED MANUSCRIPT

Psychiatric history before baseline

Panic 1707 8854 2.697 0.067 2.576 0.076

PT
No 94.7 (4937) 5.3 (276) - -
Probable 94.9.1 (930) 5.1 (50) -0.025 0.863 -0.101 0.510

RI
Definite 93.0 (2476) 7.0 (185) 0.205 0.027* 0.190 0.052

SC
GAD 1696 8742 2.995 0.050 2.310 0.099
No 94.6 (5597) 5.4 (321) - -
Probable 95.1 (807) 4.9 (42) -0.062 0.686 -0.045 0.777

U
Definite 92.7 (1831) 7.3 (144) 0.228 0.022* 0.210 0.041*

AN
Suicide attempt 1682 8735 9.042 0.003** 5.728 0.017*
No 95.1 (4710) 4.9 (242) - -

M
Yes 93.0 (3520) 7.0 (263) 0.259 0.003** 0.218 0.017*
Psychosis 1676 8667 3.432 0.064 3.241 0.072

D
No 94.8 (5332) 5.2 (294) - -

TE
Yes 93.5 (2842) 6.5 (199) 0.165 0.064 0.169 0.072
Psychoactive substance use history before baseline
EP
Ethanol abuse treatment 693 3161 3.558 0.059 5.115 0.024*
C

No 94.7 (2297) 5.3 (128) - -


AC

Yes 92.3 (679) 7.7 (57) 0.299 0.059 0.369 0.024*


Caffeine use for the past 2 1650 8550 1.92 ± 2.64 2.25 ± 2.99 3.992 0.046* 0.029 0.046* 2.862 0.091 0.028 0.091
months, c/d (8064) (486)
Nicotine use for the past 2 1651 8601 0.36 ± 1.00 0.47 ± 1.12 3.692 0.055 0.059 0.055 2.797 0.095 0.053 0.095
ACCEPTED MANUSCRIPT

months, p/d (8114) (487)


Marijuana use for the past 2 1685 8661 2.797 0.094 0.711 0.399
months

PT
No 94.5 (7218) 5.5 (422) - -
Yes 92.8 (947) 7.2 (74) 0.210 0.094 0.114 0.399

RI
Onset of Marijuana abuse, 512 2382 18.7 ± 7.2 17.4 ± 6.3 4.890 0.027* -0.037 0.027* 3.049 0.081 -0.035 0.081
year (2230) (152)

SC
Amphetamine use for the 1679 8633 5.860 0.016* 5.710 0.017*
past 2 months

U
No 94.4 (7919) 5.6 (469) - -

AN
Yes 89.8 (220) 10.2 (25) 0.518 0.016* 0.515 0.017*
Amphetamine abuse 227 1042 5.463 0.020* 2.271 0.132

M
treatment
No 93.8 (800) 6.2 (53) - -

D
Yes 87.3 (165) 12.7 (24) 0.832 0.020* 0.592 0.132

TE
Cocaine abuse treatment 281 1339 3.831 0.051 3.097 0.079
No 93.3 (952) 6.7 (68) - -
EP
Yes 89.3 (285) 10.7 (34) 0.542 0.051 0.541 0.079
LSD abuse treatment 175 794 3.199 0.074 1.780 0.183
C

No 93.2 (683) 6.8 (50) - -


AC

Yes 86.9 (53) 13.1 (8) 0.876 0.074 0.752 0.183


Opiates use for the past 2 1678 8623 4.581 0.032* 5.286 0.022*
months
No 94.4 (7931) 5.6 (469) - -
ACCEPTED MANUSCRIPT

Yes 90.1 (201) 9.9 (22) 0.485 0.032* 0.541 0.022*


Past treatment history before baseline

PT
Buproprion 1681 8706 4.321 0.038* 3.895 0.048*
No 93.5 (3714) 6.5 (257) - -

RI
Yes 95.0 (4496) 5.0 (239) -0.180 0.038* -0.177 0.048*

SC
Affective switch caused by 703 4078 4.169 0.041* 2.509 0.113
Buproprion

U
No 95.4 (3152) 4.6 (152) - -
Yes 92.9 (719) 7.1 (55) 0.315 0.041* 0.261 0.113

AN
MAOI 1682 8706 7.068 0.008** 4.720 0.030*
No 93.9 (7259) 6.1 (470) - -

M
Yes 97.0 (948) 3.0 (29) -0.424 0.008** -0.362 0.030*
Affective switch caused by 367 2241 2.817 0.093 2.708 0.100

D
Citalopram

TE
No 95.1 (1776) 4.9 (91) - -
Yes 91.2 (341) 8.8 (33) 0.497 0.093 0.508 0.100
EP
Affective switch caused by 731 4410 7.571 0.006** 7.869 0.005**
Fluoxetine
C

No 95.6 (3105) 4.4 (143) - -


AC

Yes 92.7 (1077) 7.3 (85) 0.367 0.006** 0.383 0.005**


Sertraline 1680 8655 5.430 0.020* 6.249 0.012*
No 93.5 (4011) 6.5 (279) - -
Yes 95.1 (4151) 4.9 (214) -0.203 0.020* -0.226 0.012*
ACCEPTED MANUSCRIPT

Affective switch caused by 609 3688 4.378 0.036* 4.300 0.038*


Sertraline
No 95.8 (2898) 4.2 (127) - -

PT
Yes 93.1 (617) 6.9 (46) 0.351 0.036* 0.360 0.038*
Affective switch caused by 572 3447 7.660 0.006** 7.617 0.006**

RI
Paroxetine
No 95.2 (2689) 4.8 (135) - -

SC
Yes 91.5 (570) 8.5 (53) 0.449 0.006** 0.473 0.006**
Affective switch caused by 110 742 3.531 0.061 2.462 0.117

U
Fluvoxamine

AN
No 94.3 (649) 5.7 (39) - -
Yes 79.6 (43) 20.4 (11) 1.138 0.061 1.121 0.117

M
Venlafaxine 1681 8674 3.905 0.048* 2.977 0.084
No 93.7 (4869) 6.3 (328) - -

D
Yes 95.1 (3306) 4.9 (171) -0.178 0.048* -0.162 0.084

TE
Heterocyclic (TCA) 1682 8697 3.858 0.050* 4.117 0.042*
No 93.9 (6493) 6.1 (421) - -
EP
Yes 95.6 (1705) 4.4 (78) -0.224 0.050* -0.246 0.042*
Affective switch caused by 234 1487 3.163 0.076 0.387 0.534
C

Heterocyclic (TCA)
AC

No 96.0 (1237) 4.0 (52) - -


Yes 91.9 (182) 8.1 (16) 0.504 0.076 0.212 0.534
Number of Antidepressants 1720 8915 3.51 ± 2.37 3.11 ± 2.27 6.935 0.008** -0.049 0.008** 5.541 0.019* -0.046 0.019*
used before the baseline (8403) (512)
ACCEPTED MANUSCRIPT

(0-11: CIT, FLUO, SER,


PAR, FLUV, BUP, MIRT,
MAOI, VEN, NEF, TCA)

PT
Antidepressants used before 1720 8915 7.568 0.006** 6.407 0.011*
the baseline (CIT, FLUO,

RI
SER, PAR, FLUV, BUP,
MIRT, MAOI, VEN, NEF,

SC
TCA)
No 91.4 (687) 8.6 (65) - -

U
Yes 94.5 (7716) 5.5 (447) -0.372 0.006** -0.359 0.011*

AN
Number of Antidepressants 1720 8915 0.51 ± 0.89 0.68 ± 1.13 10.682 0.001** 0.133 0.001** 10.181 0.001** 0.140 0.001**
causing Affective Switch (8403) (512)

M
(0-11: CIT, FLUO, SER,
PAR, FLUV, BUP, MIRT,

D
MAOI, VEN, NEF, TCA)

TE
Affective Switch caused by 1720 8915 3.069 0.080 2.259 0.133
Antidepressants (CIT,
EP
FLUO, SER, PAR, FLUV,
BUP, MIRT, MAOI, VEN,
C

NEF, TCA)
AC

No 94.7 (5521) 5.3 (310) - -


Yes 93.5 (2882) 6.5 (202) 0.154 0.080 0.137 0.133
Number of SSRIs used 1720 8915 1.90 ± 1.33 1.76 ± 1.32 2.970 0.085 -0.056 0.085 3.081 0.079 -0.060 0.079
before the baseline (0-5: CIT, (8403) (512)
ACCEPTED MANUSCRIPT

FLUO, SER, PAR, FLUV)


SSRIs used before the 1720 8915 2.955 0.086 1.801 0.180
baseline (CIT, FLUO, SER,

PT
PAR, FLUV)
No 93.0 (1379) 7.0 (104) - -

RI
Yes 94.5 (7024) 5.5 (408) -0.187 0.086 -0.153 0.180
Number of SSRIs causing 1720 8915 0.32 ± 0.63 0.45 ± 0.85 11.705 0.001** 0.196 0.001** 13.361 < 0.217 <

SC
Affective Switch (0-5: CIT, (8403) (512) 0.001** 0.001**
FLUO, SER, PAR, FLUV) * *

U
Affective Switch caused by 1720 8915 7.341 0.007** 8.220 0.004**

AN
SSRIs (CIT, FLUO, SER,
PAR, FLUV)

M
No 94.8 (6360) 5.2 (351) - -
Yes 92.7 (2043) 7.3 (161) 0.255 0.007** 0.276 0.004**

D
Number of other 1720 8915 1.60 ± 1.40 1.35 ± 1.32 8.022 0.005** -0.091 0.005** 5.493 0.019* -0.079 0.019*

TE
Antidepressants used (8403) (512)
before the baseline (0-6:
EP
BUP, MIRT, MAOI, VEN,
NEF, TCA)
C

Other Antidepressants used 1720 8915 4.503 0.034* 3.547 0.060


AC

before the baseline (BUP,


MIRT, MAOI, VEN, NEF,
TCA)
No 93.1 (2147) 6.9 (160) - -
ACCEPTED MANUSCRIPT

Yes 94.7 (6256) 5.3 (352) -0.199 0.034* -0.183 0.060


Clozapine 1681 8680 8.163 0.004** 6.258 0.012*
No 94.4 (7930) 5.6 (468) - -

PT
Yes 89.4 (252) 10.6 (30) 0.563 0.004** 0.532 0.012*
Quetiapine 1682 8689 9.089 0.003** 7.452 0.006**

RI
No 94.7 (6803) 5.3 (378) - -
Yes 92.1 (1389) 7.9 (119) 0.317 0.003** 0.300 0.006**

SC
Number of Atypical 1720 8915 0.89 ± 1.05 1.04 ± 1.11 5.683 0.017* 0.092 0.017* 3.476 0.062 0.077 0.062
Antipsychotics used before (8403) (512)

U
the baseline (0-6: RIS,

AN
CLOZ, OLAN, QUET, ZIP,
ARI)

M
Atypical Antipsychotics used 1720 8915 4.216 0.040* 1.715 0.190
before the baseline (RIS,

D
CLOZ, OLAN, QUET, ZIP,

TE
ARI)
No 95.0 (3821) 5.0 (200) - -
EP
Yes 93.6 (4582) 6.4 (312) 0.178 0.040* 0.119 0.190
Number of Antipsychotics 1720 8915 0.95 ± 1.14 1.10 ± 1.18 5.100 0.024* 0.080 0.024* 3.058 0.080 0.067 0.080
C

used before the baseline (0-7: (8403) (512)


AC

HAL, RIS, CLOZ, OLAN,


QUET, ZIP, ARI)
Antipsychotics used before 1720 8915 4.415 0.036* 1.881 0.170
the baseline (HAL, RIS,
ACCEPTED MANUSCRIPT

CLOZ, OLAN, QUET, ZIP,


ARI)
No 95.1 (3751) 4.9 (195) - -

PT
Yes 93.6 (4652) 6.4 (317) 0.183 0.036* 0.125 0.170
Psychotherapy (Verbal 1682 8717 8.088 0.004** 5.639 0.018*

RI
therapy)
No 93.4 (4695) 6.6 (330) - -

SC
Yes 95.4 (3523) 4.6 (169) -0.254 0.004** -0.220 0.018*
Affective switch caused by 516 2995 5.306 0.021* 5.926 0.015*

U
psychotherapy (Verbal

AN
therapy)
No 95.6 (2667) 4.4 (124) - -

M
Yes 90.7 (185) 9.3 (19) 0.580 0.021* 0.639 0.015*
Values represent mean ± SD (N) for continuous variables, and % (N) for categorical variables. (N represents number of depressive episodes.)

D
Variables with statistical significance (p < 0.05*, p < 0.01**, p < 0.001***) or trend (p < 0.10) in Model 1 are reported.

TE
The covariates in Model 2 consist of demographic (DEM), socioeconomic status (SES) and the STEP-BD time-related variables: age, gender, ethnicity, education, employment,
annual personal income, marital status, total follow-up duration, and time from baseline to a depressive episode.
EP
The combination of yellow-colored variables describing past history features were selected and included together with the covariates of Model 2 into Model 3.
a Sample size in Model 2: number of patients = 1640; number of depressive episodes = 8506. For each variable, F values indicate the effect size and beta coefficients show
C

whether the relationship between the variable and switch risk is positive or negative.
AC
ACCEPTED MANUSCRIPT

Supplementary Table S3. Associations between switch and variables describing clinical status at baseline.
Total Total Episodes Episodes with Model 1 (Crude) Model 3a (+ Past history adjusted)
No. of No. of without switch switch

PT
Patien Episod
ts es

RI
Fixed Effect Fixed Coefficient Fixed Effect Fixed Coefficient
(N = 8403) (N = 512) F P Estima P F P Estima P

SC
te (β) te (β)
Clinical symptoms at baseline

U
AN
No. of days with elevated 1719 8908 1.58 ± 3.20 2.64 ± 4.06 31.903 < 0.061 < 9.858 0.002** 0.047 0.002**
mood over the last 2 weeks (8396) (512) 0.001** 0.001**

M
* *
Severity of elevated mood 1154 6108 1.03 ± 1.18 1.42 ± 1.29 20.380 < 0.238 < 2.715 0.100 0.093 0.100

D
(5718) (390) 0.001** 0.001**

TE
* *
% of days with elevated 1668 8690 17.5 ± 19.4 21.7 ± 20.9 12.633 < 0.007 < 2.344 0.126 0.004 0.126
EP
mood in the last year (8190) (500) 0.001** 0.001**
* *
C

No. of days with irritability 1719 8908 5.16 ± 5.43 5.92 ± 5.53 5.141 0.023* 0.017 0.023* 0.736 0.391 0.009 0.391
AC

over the last 2 weeks (8396) (512)


Severity of irritability 1423 7629 1.74 ± 1.18 1.92 ± 1.21 6.046 0.014* 0.127 0.014* 0.883 0.347 0.049 0.347
(7167) (462)
Total bipolar index 1687 8714 69.8 ± 15.9 72.3 ± 15.0 6.142 0.013* 0.007 0.013* 0.152 0.697 0.001 0.697
(8214) (500)
ACCEPTED MANUSCRIPT

CGI-BP-Elevation subscale 1702 8844 1.86 ± 1.17 2.27 ± 1.47 33.768 < 0.185 < 14.036 < 0.154 <
in the last month (8337) (507) 0.001** 0.001** 0.001** 0.001**
* * * *

PT
CGI-BP-Overall in the last 1709 8865 3.59 ± 1.17 3.72 ± 1.14 2.948 0.086 0.064 0.086 0.695 0.405 0.040 0.405
month (8355) (510)

RI
Associated symptoms at baseline

SC
SUM-D# at baseline 1694 8728 6.38 ± 3.63 6.83 ± 3.74 3.765 0.052 0.023 0.052 0.995 0.319 0.017 0.319
(8232) (496)

U
Concentration at baseline 1716 8875 0.72 ± 0.56 0.78 ± 0.56 2.934 0.087 0.131 0.087 1.548 0.213 0.134 0.213

AN
(absolute value of the score of (8368) (507)
DEPCONCN)

M
Guilt at baseline (the score of 1710 8828 0.48 ± 0.54 0.56 ± 0.60 5.064 0.024* 0.172 0.024* 3.270 0.071 0.185 0.071
DEPGUILT) (8319) (509)

D
Distractibility as a depressive 1712 8876 0.54 ± 0.53 0.62 ± 0.54 5.597 0.018* 0.185 0.018* 2.502 0.114 0.172 0.114

TE
symptom at baseline (the score (8367) (509)
of DEPDIST)
EP
Psychomotor agitation as a 1714 8885 0.34 ± 0.47 0.44 ± 0.48 12.589 < 0.301 < 1.393 0.238 0.144 0.238
depressive symptom at (8378) (507) 0.001** 0.001**
C

baseline (the score of * *


DEPPMA)
AC

Sleep at baseline (absolute 1719 8908 0.79 ± 0.59 0.88 ± 0.64 7.293 0.007** 0.192 0.007** 1.899 0.168 0.129 0.168
value of the score of (8396) (512)
DEPSLEEP)
SUM-M# at baseline 1706 8829 1.80 ± 2.01 2.57 ± 2.49 40.921 < 0.115 < 18.184 < 0.109 <
ACCEPTED MANUSCRIPT

(8322) (507) 0.001** 0.001** 0.001** 0.001**


* * * *
Need for sleep at baseline 1719 8908 0.13 ± 0.38 0.24 ± 0.51 22.396 < 0.423 < 8.526 0.004** 0.360 0.004**

PT
(absolute value of the score (8396) (512) 0.001** 0.001**
of ELVSLEEP) * *

RI
Elevated Self-Esteem at 1719 8908 0.06 ± 0.22 0.11 ± 0.29 12.524 < 0.545 < 6.432 0.011* 0.544 0.011*
baseline (the score of (8396) (512) 0.001** 0.001**

SC
ELVSELFE) * *
Talking (the score of 1719 8908 0.21 ± 0.40 0.32 ± 0.49 21.582 < 0.423 < 7.642 0.006** 0.353 0.006**

U
ELVTALK) (8396) (512) 0.001** 0.001**

AN
* *
Flight of Ideas at baseline 1718 8895 0.32 ± 0.47 0.45 ± 0.51 19.316 < 0.363 < 10.030 0.002** 0.382 0.002**

M
(the score of ELVFOI) (8386) (509) 0.001** 0.001**
* *

D
Distractibility as a manic 1717 8901 0.52 ± 0.55 0.63 ± 0.57 10.576 0.001** 0.246 0.001** 8.913 0.003** 0.314 0.003**

TE
symptom at baseline (the (8390) (511)
score of ELVDISTR)
EP
Goal Directed Activity at 1710 8850 0.14 ± 0.33 0.23 ± 0.44 19.837 < 0.470 < 6.267 0.012* 0.347 0.012*
baseline (the score of (8339) (511) 0.001** 0.001**
C

ELVGDACT) * *
AC

Psychomotor agitation as a 1716 8896 0.32 ± 0.46 0.44 ± 0.49 16.963 < 0.347 < 3.824 0.051 0.231 0.051
manic symptom at baseline (8384) (512) 0.001** 0.001**
(the score of ELVPMA) * *
High Risk Behavior at 1718 8899 0.09 ± 0.27 0.15 ± 0.33 14.479 < 0.500 < 6.244 0.012* 0.447 0.012*
ACCEPTED MANUSCRIPT

baseline (the score of (8387) (512) 0.001** 0.001**


ELVHRB) * *
Associated features at baseline

PT
Paranoid ideation 1707 8817 11.118 0.001** 2.420 0.120

RI
No 94.7 (7104) 5.3 (400) - -
Yes 91.6 (1203) 8.4 (110) 0.362 0.001** 0.223 0.120

SC
Ideas of reference 1706 8815 6.824 0.009** 3.950 0.047*
No 94.4 (7724) 5.6 (454) - -

U
Yes 91.2 (581) 8.8 (56) 0.378 0.009** 0.397 0.047*

AN
Hallucinations 1706 8815 6.776 0.009** 1.044 0.307
No 94.4 (7886) 5.6 (467) - -

M
Yes 90.7 (419) 9.3 (43) 0.427 0.009** 0.209 0.307
Delusions 1706 8815 3.954 0.047* 0.432 0.511

D
No 94.4 (7925) 5.6 (474) - -

TE
Yes 91.3 (380) 8.7 (36) 0.352 0.047* 0.167 0.511
Dysthymia 1681 8761 5.435 0.020* 3.871 0.049*
EP
No 94.0 (7246) 6.0 (465) - -
Yes 96.6 (1014) 3.4 (36) -0.350 0.020* -0.422 0.049*
C

Treatment at baseline
AC

Bupropion at baseline 1720 8915 2.909 0.088 2.371 0.124


assessment
No 94.0 (7084) 6.0 (452) - -
Yes 95.6 (1319) 4.4 (60) -0.215 0.088 -0.255 0.124
ACCEPTED MANUSCRIPT

Fluvoxamine at baseline 1720 8915 2.716 0.099 0.059 0.808


assessment
No 94.3 (8297) 5.7 (500) - -

PT
Yes 89.8 (106) 10.2 (12) 0.503 0.099 -0.112 0.808
Number of SSRIs at baseline 1720 8915 0.29 ± 0.47 0.24 ± 0.45 2.823 0.093 -0.161 0.093 0.020 0.887 -0.018 0.887

RI
(0-5: CIT, FLUO, SER, PAR, (8403) (512)
FLUV)

SC
SSRIs at baseline (CIT, 1720 8915 3.592 0.058 0.079 0.779
FLUO, SER, PAR, FLUV)

U
No 93.8 (6011) 6.2 (394) - -

AN
Yes 95.3 (2392) 4.7 (118) -0.187 0.058 -0.037 0.779
Number of Antidepressants 1720 8915 0.62 ± 0.69 0.52 ± 0.64 5.418 0.020* -0.150 0.020* 0.882 0.348 -0.080 0.348

M
at baseline (0-10: CIT, (8403) (512)
FLUO, SER, PAR, FLUV,

D
BUP, MIRT, VEN, NEF,

TE
TCA)
Antidepressants at baseline 1720 8915 4.006 0.045* 0.180 0.672
EP
(CIT, FLUO, SER, PAR,
FLUV, BUP, MIRT, VEN,
C

NEF, TCA)
AC

No 93.6 (4201) 6.4 (288) - -


Yes 94.9 (4202) 5.1 (224) -0.171 0.045* -0.049 0.672
Quetiapine at baseline 1720 8915 6.469 0.011* 0.579 0.447
assessment
ACCEPTED MANUSCRIPT

No 94.5 (7573) 5.5 (438) - -


Yes 91.8 (830) 8.2 (74) 0.324 0.011* -0.152 0.447
Number of antipsychotics at 1720 8915 0.36 ± 0.52 0.42 ± 0.54 3.077 0.079 0.141 0.079 0.059 0.808 -0.027 0.808

PT
baseline (0-7: HAL, RIS, (8403) (512)
CLOZ, OLAN, QUET, ZIP,

RI
ARI)
Antipsychotics at baseline 1720 8915 2.938 0.087 0.048 0.827

SC
(HAL, RIS, CLOZ, OLAN,
QUET, ZIP, ARI)

U
No 94.7 (5512) 5.3 (310) - -

AN
Yes 93.5 (2891) 6.5 (202) 0.151 0.087 -0.026 0.827
Values represent mean ± SD (N) for continuous variables, and % (N) for categorical variables. (N represents number of depressive episodes.)

M
Variables with statistical significance (p < 0.05*, p < 0.01**, p < 0.001***) or trend (p < 0.10) in Model 1 are reported.
The combination of variables describing past history features was selected and included together with the covariates of Model 2 into Model 3.

D
The combination of yellow-colored variables describing baseline clinical features were selected and included together with the covariates of Model 3 into Model 4.

TE
Abbreviations: CGI-BP, Clinical Global Impression scale-Bipolar disorder; SUM-D, Summary of classified scores for depression; SUM-M, Summary of classified scores for
mania.
EP
a Sample size in Model 3: number of patients = 1067; number of depressive episodes = 5297.
# SUM-D and SUM-M were calculated from the data “Associated Symptoms for the Past Week” using the imputation methods by Sachs et al. (2002).
C

For each variable, F values indicate the effect size and beta coefficients show whether the relationship between the variable and switch risk is positive or negative.
AC
ACCEPTED MANUSCRIPT

PT
Predictors of switch from depression to mania in bipolar

RI
disorder

U SC
Supplementary Figure S1.

AN
M
D
Tomihisa Niitsu1,2, Chiara Fabbri1, Alessandro Serretti1
TE
EP
1. Department of Biomedical and NeuroMotor Sciences, University of Bologna,
Bologna, Italy
C

2. Research Center for Child Mental Development, Chiba University Graduate


AC

School of Medicine, Chiba, Japan

Feb/11/2014 1
ACCEPTED MANUSCRIPT

Index

PT
RI
• The STEP-BD time course and analyzed datasets P3

SC
• Definitions of phenotypes (switch) P4

U
• Data structures and each model (Models 1 to 5) P5

AN
• How to select the combination of variables to include

M
in each model P10

D
• Multicollinearity diagnostics P11
TE
C EP
AC

2
ACCEPTED MANUSCRIPT

The STEP-BD time course and analyzed datasets


Baseline Clinical assessment Clinical assessment

PT
Time

RI
SC
ADE & DF CMF dataset

U
datasets

AN
Current data Current data
Demographic Dep. episode Dep. episode

M
SES

D
TE
Past history EP
Baseline data
C
AC

• Questionnaires
– ADE: Affective Disorders Evaluation
– DF: Demographic Form
– CMF: Clinical Monitoring Form
3
ACCEPTED MANUSCRIPT

Definitions of “Depressive episode with switch” and


“Patient with switch” using the CMF dataset
• Definition of “Depressive episode with switch”

PT
– Depressive episode followed by manic/hypomanic/mixed episode within 12

RI
weeks (84 days)

SC
Mania =< 12 weeks (84 days)

U
AN
M
Mood

Depressive Manic/hypo/

D
Time
episode mixed episode
TE
C EP
AC

Depression

• Definition of “Patient with Switch”


– Patient with at least one depressive episode with switch during total
follow-up period 4
ACCEPTED MANUSCRIPT

Associations of variables & Model 1: Crude (not adjusted)


Manic/hypomanic/
Baseline Depressive episode mixed episode

PT
(ADE & DF dataset) (CMF dataset) (CMF dataset)
Duration Time

RI
SC
DEM SES

U
AN
Past history

M
D
TE
Baseline data
Switch
C EP

Current data
AC

5
ACCEPTED MANUSCRIPT

Model 2: Adjusted for DEM, SES and Durations


Manic/hypomanic/
Baseline Depressive episode mixed episode

PT
(ADE & DF dataset) (CMF dataset) (CMF dataset)
Duration Time

RI
SC
DEM SES

U
AN
Past history

M
D
TE
Baseline data
Switch
C EP

Current data
AC

• Variables of past history


were associated with
switch via blue arrows.
6
ACCEPTED MANUSCRIPT

Model 3: Adjusted for the covariates in Model 2 + variables selected from the
significant variables of past history in Model 2
Manic/hypomanic/
Baseline Depressive episode mixed episode

PT
(ADE & DF dataset) (CMF dataset) (CMF dataset)
Duration Time

RI
SC
DEM SES

U
AN
Past history

M
D
TE
Baseline data
Switch
CEP

Current data
AC

• Variables of baseline
clinical data were
associated with switch via
green arrows. 7
ACCEPTED MANUSCRIPT

Model 4: Adjusted for the covariates in Model 3 + variables selected from the
significant variables baseline clinical data in Model 3
Manic/hypomanic/
Baseline Depressive episode mixed episode

PT
(ADE & DF dataset) (CMF dataset) (CMF dataset)
Duration Time

RI
SC
DEM SES

U
AN
Past history

M
D
TE
Baseline data
Switch
CEP

Current data
AC

• Variables of current
clinical data were
associated with switch via
an orange arrow. 8
ACCEPTED MANUSCRIPT

Model 5 (Full adjusted): Adjusted for the covariates in Model 4 + variables


selected from the significant variables current clinical data in Model 4
Manic/hypomanic/
Baseline Depressive episode mixed episode

PT
(ADE & DF dataset) (CMF dataset) (CMF dataset)
Duration Time

RI
SC
DEM SES

U
AN
Past history

M
D
TE
Baseline data
Switch
CEP

Current data
AC

A
• Each variables of current
clinical data was
associated with switch via
an orange arrow. B
9
ACCEPTED MANUSCRIPT

How to select the combination of variables to include in


each model (Model 3 – 5)

PT
1. Firstly, to avoid including practically ineffective variables as covariates
and losing sample size, the following selection criteria of the candidate

RI
variables were set;

SC
1. statistically significant with a moderate fixed effect (F > 5.00)
2. with available data from at least 6,000 episodes (about 1,200 patients)

U
AN
2. Secondly, if two or more candidate variables were similar in clinical
implications, the most inclusive and clinically meaningful one was

M
selected.

D
– e.g. history of ADs treatment was selected rather than history of sertraline

TE
treatment.
3. Finally, stepwise combinations of the candidate variables were added to
EP
each model (Model 2 – 4), and the combination minimizing both the
C

corrected Akaike’s Information Criterion (AICc) and the Bayesian


AC

Information Criterion (BIC) was selected and included into the next step
model.
– The AICc and BIC are useful tools to choose a better forecasting model (Vrieze,
2012).
10
ACCEPTED MANUSCRIPT

Multicollinearity diagnostics
• A correlation matrix of all the candidate variables was scanned for

PT
presence of very high correlation (|r| > 0.90). (Spearman’
correlation coefficient are used)

RI
• Variance inflation factor (VIF) > 10: The VIF indicates whether a

SC
predictor has a strong linear relationship with the other predictor(s).

U
(Myers, 1990)

AN
• Tolerance (= 1 / VIF) < 0.2: Values below 0.2 are worthy of concern,

M
below 0.1 indicate serious problems. (Menard, 1995)

D
– In SPSS, the VIF and tolerance could be provided using its linear regression

TE
model. (Fields, 2009)
C EP
AC

11
ACCEPTED MANUSCRIPT
Supplementary Figure S2. Flowchart of the sample selection from the STEP-BD study.

4,107 patients with clinical data of the Standard


Care Pathway (SCP).

Excluded: 386 (9.4%) patients had CMF


unavailable clinical data.

PT
RI
3,721 (90.6%) patients with clinical data of the
Clinical Monitoring Form (CMF).

SC
Excluded: 946 (25.4%) patients had short
follow-up duration (< 6 months).

U
AN
2,775 (74.6%) patients with at least 6 months
follow-up duration.
M

Excluded: 974 (35.1%) patients had no depressive


D

episode during follow-up.


TE

1,801 (64.9%) patients with at least one depressive


episode during follow-up.
EP

Excluded: 78 (4.3%) patients had unavailable


C

data for the Affective Disorder Evaluation (ADE)


form and /or Demographic form (DF).
AC

3 (0.2%) patients reported transsexual gender.

1,720 (95.5%) patients (8,915 depressive


episodes) were included.