Clinical correlates of age at onset distribution in bipolar disorder: a comparison between diagnostic subgroups
Manchia et al. Int J Bipolar Disord
Clinical correlates of age at onset distribution in bipolar disorder: a comparison between diagnostic subgroups
Mirko Manchia 0 2
Giuseppe Maina 1
Bernardo Carpiniello 0
Federica Pinna 0
Luca Steardo 5
Virginia D'Ambrosio 1
Virginio Salvi 1
Martin Alda 4
Alfonso Tortorella 5
Umberto Albert 3
0 Section of Psychiatry, Department of Medical Science and Public Health, University of Cagliari , Via Liguria, 13, 09127 Cagliari , Italy
1 Department of Mental Health, “San Luigi-Gonzaga” Hospital, University of Turin , Orbassano , Italy
2 Department of Pharmacology, Dalhousie University , Halifax, NS , Canada
3 Rita Levi Montalcini Department of Neuroscience, Anxiety and Mood Disorders Unit, University of Turin , Turin , Italy
4 Department of Psychiatry, Dalhousie University , Halifax, NS , Canada
5 Department of Psychiatry, University of Naples SUN , Naples , Italy
Background: Admixture analysis of age at onset (AAO) has helped delineating the clinical profile of early onset (EO) bipolar disorder (BD). However, there is scarce evidence comparing the distributional properties of AAO as well as the clinical features of EO BD type 1 (BD1) with EO BD type 2 (BD2). To this end, we studied 515 BD patients (224 BD1, 279 BD2, and 12 BD not otherwise specified [NOS]) diagnosed according to DSM-IV-TR criteria. Methods: AAO was defined as the first reliably diagnosed hypo/manic or depressive episode according to diagnostic criteria. We used normal distribution mixture analysis to identify subgroups of patients according to AAO. Models were chosen according to the Schwarz's Bayesian information criteria (BIC). Clinical correlates of EO were analysed using univariate tests and multivariate logistic regression models. Results: A two normal components model best fitted the observed distribution of AAO in BD1 (BIC = −1599.3), BD2 (BIC = −2158.4), and in the whole sample (BIC = −3854.9). A higher number of EO BD2 patients had a depression(hypo)mania-free interval (DMI) course, while a higher rate of (hypo)mania-depression-free interval (MDI) course was found in EO BD1. EO BD2 had also a higher rate of comorbidity with alcohol dependence compared to EO BD1. The latter finding was confirmed by multivariate logistic regression analysis. Conclusions: In conclusion, both BD1 and BD2 had bimodal AAO distributions, but EO subgroups had a diagnosticspecific clinical delineation.
Mood disorders; Diagnostic subtypes; Early onset; Retrospective study; Admixture analysis
Bipolar disorder (BD) is a heritable psychiatric illness
characterised by cyclic mood episodes of opposite
polarity alternating with intervals of well-being
. As in other psychiatric complex genetic
diseases, the relatively high clinical heterogeneity of BD
might have hindered the identification of molecular and
clinical determinants of risk as well as of predictors of
. The magnitude of clinical
heterogeneity might be reduced by studying subgroups of
BD patients sharing specific clinical characteristics such
as patterns of treatment response
(Alda et al. 2005)
(Goes et al. 2007)
, or early illness
(Jamain et al. 2014)
. Indeed, the extensive analysis
of age at onset (AAO) BD subgroups through admixture
analysis has shown clinical
(Bellivier et al. 2001; Lin et al.
2005; Manchia et al. 2008; Tozzi et al. 2011; Ortiz et al.
(Etain et al. 2006; Severino et al. 2009;
Etain et al. 2010; Belmonte et al. 2011)
specific, particularly, to early onset (EO) BD.
As the vast majority of studies investigated BD type
1 (BD1) samples, it remains to be established, however,
whether this clinical delineation of EO is present also in
BD type 2 (BD2) patients. Furthermore, the distributional
properties of AAO have never been investigated in
samples exclusively composed of BD2 patients.
The primary aim of the present study was to test whether
BD1 and BD2 differed in terms of AAO distributions.
The secondary objective was to test whether EO had
clinical characteristics specific for each diagnostic subgroup.
To this end, we (i) studied the AAO distribution of each
diagnostic subgroup with mixture modelling; (ii)
compared the AAO distributions identified in BD1 and BD2;
(iii) analyzed the pattern of associations of a set of
demographic and clinical variables with EO in each diagnostic
subgroup; and (iv) compared the clinical association
patterns between EO BD1 and EO BD2.
Patient population and assessment instruments
Our sample consisted of 515 unrelated patients with
BD. Two hundred and twenty-four were diagnosed with
BD1, while 279 had a diagnosis of BD2, and 12 had a
diagnosis BD not otherwise specified (NOS). All
subjects were of Italian ancestry. Patients were recruited
at the Anxiety and Mood Disorders Unit, University
of Turin, Italy and at the Department of Psychiatry,
University of Naples SUN, Napoli, Italy. Certified
psychiatrists with at least 4 years of postgraduate
clinical experience performed the clinical assessment of
patients. All potential interviewers met prior to study
beginning and underwent a common extensive training
prior to conducting the assessments. They were trained
in the use of a common semi-structured interview that
was used to collect (a) socio-demographic data (age,
gender, marital status, years of education, and
occupational status); (b) diagnosis (current and lifetime), which
were performed according to the Diagnostic and
Statistical Manual of Mental Disorders (DSM)-IV-Text
Revision (TR) criteria
(American Psychiatric Association
using the structured clinical interview for
DSMIV-TR Axis I disorders (SCID-I/P)
(First et al. 2002)
clinical data including AAO. In addition, a systematic
review of patients’ medical records helped clinicians to
establish AAO and corroborate data concerning
clinical characteristics of the disorder emerging from direct
interview. Age at onset was defined as the first reliably
diagnosed hypo/manic or depressive episode meeting
the diagnostic criteria. External corroboration for AAO
was obtained, whenever possible, by directly
interviewing, with patient’s consent, a first-degree family
member or other significant individuals. For the purposes of
the present study, we included only subjects for whom
it was possible to establish AAO with complete
agreement between the information provided by patients and
their relatives. Age at interview was defined as the age
at which subjects were first assessed by a clinician at
each research centre.
In the early phase of the study, inter-rater reliability
of the diagnosis of Axis I disorders with the SCID-I was
ascertained. The inter-rater reliability was found to be
good: Cohen kappa coefficient was 0.89 for the presence
of any current or lifetime Axis I disorder.
We used Gaussian distribution mixture analysis to test
whether we could identify subgroups of patients
according to the AAO. We investigated a range of number of
AAO groups (1–9). The choice of the mixture model
that best fit the distribution of AAO was made
according to the Schwarz’s Bayesian information criteria (BIC).
Specifically, the analysis performed with the “Mclust”
(Fraley and Raftery 1999; Fraley et al. 2014)
implemented in R (R Development Core Team 2008)
indicates the best model as the one with the highest
BIC among the fitted models
(Fraley and Raftery 2007)
This package estimates parameters of the model using
an expectation–maximization (EM) algorithm. Cut-off
points were derived using the Gaussian cumulative
distribution function of estimated AAO mixture function and
calculating each data point’s probability of belonging to
each class. Specifically, once the mixture model
parameters were estimated, we calculated the posterior
probability of any data point. The resulting probabilities were
then compared in order to establish which class the data
point belonged to. Gaussian mixture analysis (both
number of components and parameters estimates) was also
replicated and confirmed with the “Mixtools” R package
(Benaglia et al. 2009). We used Kolmogorov–Smirnov
(K–S) test to determine whether the Gaussian cumulative
distribution function of estimated AAO mixture
identified in BD1 patients was significantly different from the
one identified in BD2 patients. Further, we used K–S to
test for differences in the AAO Gaussian cumulative
distribution function of estimated AAO mixture between
participating research centres. We tested the
association of continuous and categorical clinical variables with
AAO subgroups using univariate analysis (t test or χ2
test as appropriate). The independent variables tested
included sex, age at interview, diagnosis, illness duration,
presence of family history of any DSM-IV-TR psychiatric
disorder, presence of family history of BD and any
DSMIV-TR mood disorders, number of manic/hypomanic,
depressive and mixed episodes, type of clinical course
cycle [i.e. (hypo)mania-depression-free interval (MDI),
depression-(hypo)mania-free interval (DMI),
irregular cycling, continuous cycling, rapid cycling], presence
of lifetime suicidal behaviour, lifetime comorbidity with
substance (other than alcohol) dependence, and lifetime
comorbidity with drug and/or alcohol dependence.
Statistical significance was set at α = 0.05. Only clinical
variables presenting a statistical significant association with
an AAO subgroup (p < 0.05) of each diagnostic sample
(BD1 and BD2) were entered into a backward stepwise
multivariate binary logistic model to account for possible
intercorrelations. All statistical analyses, except for
mixture modelling, were performed with STATA/SE 12.0.
Age at onset distribution: Gaussian mixture analysis
The BD1 sample (99 males and 125 females) had a mean
age at interview (±SD) of 47.2 years (±13.0) and a mean
AAO of 26.7 years (±9.2). A two normal components
model best fitted the observed distribution of AAO
(BIC = −1599.3) (Fig. 1). Models with three and four
components did not improve the fit (Table 1).
The EO component had a mean AAO of 22.6 years
(±4.8), while the late onset (LO) component had a mean
AAO of 35.1 years (±10.1) comprising 67% and 33% of
the population proportion, respectively. The cut-off point,
derived by the Gaussian cumulative distribution function
of the latter estimated AAO function, was at 32 years for
BD1 (EO group <32 years; LO group ≥32 years) with 169
patients in the EO group and 55 in the LO group.
The BD2 sample (114 males and 165 females) had a
mean age at interview of 50.6 years (±14.8) and a mean
AAO of 30.6 years (±12.7). The observed distribution of
AAO was also best fitted by a two normal components
model (BIC = −2158.4) (Fig. 2).
No improvement of the fit was observed with three and
four components models (Table 1). The BD2 sample had
an EO component with mean AAO of 20.9 years (±4.1)
and a LO component with mean AAO of 38.2 years
(±11.8) with population proportions of 44% and 56%,
respectively. The cut-off point, derived by the
Gaussian cumulative distribution function of this estimated
AAO function, was at 28 years (EO group <28 years; LO
group ≥28 years). The EO group comprised 142 patients,
while the LO group included 137 patients.
Kolmogorov–Smirnov test showed that the
Gaussian cumulative distribution functions of estimated AAO
mixture of BD1 and BD2 differed significantly (D = 0.18,
p < 0.00001) (Fig. 3). Conversely, there was no significant
difference between the Gaussian cumulative distributions
of estimated AAO mixture of the two participating
research centres (D = 0.05, p = 0.57). Finally, Gaussian
mixture analysis confirmed a best fitting model of two
normal components (detailed in Table 1) in the whole sample
of 515 BD patients (216 males and 299 females), which
included 12 subjects with BD NOS (Fig. 4). The cut-off
point, derived by the Gaussian cumulative distribution
function of this estimated AAO function, was at 30 years.
Clinical correlates of early onset: patterns of association in bipolar disorder type 1 and type 2 diagnostic subgroups
As shown in Table 2, a trend association (p = 0.05) was
identified for the presence of family history of any
DSMIV-TR mood disorder, with LO BD1 having a higher rate
than EO BD1. Early onset BD1 had a lower age at
interview and a longer duration of illness than LO BD1.
Early onset BD2 had a higher rate of comorbidity with
alcohol dependence, as well as a higher rate of family
history of BD in EO BD2 compared to LO BD2 (Table 2).
Further, they also had a lower age at interview and longer
illness duration than LO BD2.
We then performed a multivariate logistic
regression in BD2 confirming that family history of BD [odds
ratio (OR) = 1.97, 95% confidence interval (CI) 1.04–
3.73, p = 0.04], comorbidity with alcohol dependence
(OR = 3.2, 95% CI 1.21–8.54, p = 0.02), and illness
duration (OR = 1.03, 95% CI 1.008–1.048, p = 0.006) were
associated with EO. As age at interview was used to
calculate illness duration, and consequently significantly
correlated with it (r = −0.97), it was not included in the
logistic regression model. Multivariate analysis was not
performed in BD1 since no clinical variable was
significantly associated with AAO subgroup in univariate test.
Clinical correlates of early onset: comparison between bipolar disorder type 1 and type 2 diagnostic subgroups
The mean AAO was significantly lower in BD1 compared
to BD2 (26.7 ± 9.2 vs. 30.6 ± 12.7; t = −3.8; p < 0.0001).
The results of the univariate analysis are shown in
Table 3. Early onset BD1 patients were older than EO
BD2. Conversely, EO BD2 had a higher rate of
comorbidity with alcohol dependence. In addition, a higher
number of EO BD2 presented with a DMI course, while
a higher rate of MDI course was found in EO BD1. The
multivariate binary logistic regression confirmed the
association of comorbidity with alcohol dependence with
EO BD2 (OR = 0.4, 95% CI 0.18–0.90, p = 0.02).
Significant values are typed in italics
MDI (hypo)mania-depression-free interval, DMI depression-(hypo)mania-free interval, SD Standard Deviation, p p value
a BD1: missing data for 1 patient
b BD1: missing data for 12 patients, BD2: missing data for 4 patients
c BD1: missing data for 1 patient
d BD1: missing data for 1 patient, BD2: missing data for 4 patients
e BD1: missing data for 1 patient, BD2: missing data for 4 patients
f BD1: missing data for 1 patient, BD2: missing data for 3 patients
g BD2: missing data for 2 patients
The present study highlighted that a two normal
component model in BD1 as well as in BD2 diagnostic subgroup
best described the distribution of AAO. This finding was
not reflected, however, in similar distributional
properties of AAO, as well as in comparable pattern of
association with clinical variables between the two diagnostic
subgroups. In fact, our study found that EO BD2 patients
had a higher rate of alcohol dependence compared to
both LO BD2 and EO BD1. Further, EO BD2 patients had
more frequently a DMI type of clinical course, while the
MDI type was more frequently associated with EO BD1.
Finally, EO BD2 showed a higher familial load for BD
compared to LO BD.
The bimodal AAO distribution found in both BD
diagnostic subgroups and in the whole sample of 515
patients is consistent with some studies
(Ortiz et al. 2011;
Kennedy et al. 2005; Javaid et al. 2011)
. Conversely, the
majority of studies on mixture analysis of AAO showed
a trimodal distribution in samples comprising mainly
(Bellivier et al. 2001, 2003; Lin et al. 2006;
Severino et al. 2009; Hamshere et al. 2009; Tozzi et al.
2011; Bellivier et al. 2014; Golmard et al. 2015)
note, a recent study showed that bimodal and trimodal
distribution fit equally well the AAO of BD
(GrigoroiuSerbanescu et al. 2014). Further research is needed to
determine which distribution (bi- or tri-modal) better
describes AAO in BD and which is (or which are) the
best cut-off(s) before investigating clinical correlates and
genetic differences between subgroups based on AAO.
In fact, thresholds between subgroups found in
different studies differed [e.g. thresholds between the
intermediate and late AAO subgroups differed from 25 in one
(Tozzi et al. 2011)
to 40 years in another
et al. 2009)
] as well as percentages of patients in each
AAO subgroups [e.g. percentages of patients attributed
to the early onset subgroup varied between 21.4%
(Bellivier et al. 2001)
(Lin et al. 2006)
Discrepancies in the identified AAO distributions, cut-off points,
and proportions of patients in each AAO subgroups
may depend on diverse assessment methods, recall bias,
(Montlahuc et al. 2016)
, and differences in
characteristics of samples studied, including geographic
(Post et al. 2008; Bellivier et al. 2014)
(Bauer et al. 2015; Golmard et al. 2015)
Concerning study design, Montlahuc et al. (2016) tested whether
cross‐sectional designs (which cause right truncation),
unreliable diagnosis for individuals younger than 10 years
old (which causes left truncation), and the selection
criterion used for admixture analysis impacted the number
of identified AAO subgroups. Importantly, a combination
of left and right truncation, which is common in
previously published studies of AAO admixture analysis,
appeared to significantly influence the number of AAO
(Montlahuc et al. 2016)
Geographical location appears also to impact on AAO admixture
analysis findings. Bellivier et al. (2014) found significant
differences in the theoretical AAO functions between
USA and European BD samples, mainly led by the higher
proportion of patients in the EO subgroup and the lower
mean AAO in the USA sample. Finally, birth cohort effect
might also influence the estimation of AAO subgroups
parameters. In this regard, Golmard et al. (2015) found
that the proportion of EO cases increased substantially
among BD cases born after 1960 compared to those born
before the same year.
Several other findings deserve a comment. In our
sample, BD type 1 patients had an earlier mean AAO
(26.7 years) than BD2 patients (30.6 years), in agreement
with existing data showing that BD1 first manifest their
symptoms at an earlier age
(Merikangas et al. 2011)
keeping, admixture analyses indicated a larger
proportion of EO cases among BD1 patients (67%) compared
to BD2 patients (44%). As a consequence, the EO BD1
group had a later mean AAO (22.6 years), compared to
EO BD2 patients (20.9 years), reflecting in a higher AAO
cut-off point (32 years for BD1 and 28 years for BD2).
These distributional properties of AAO distinguishing
EO BD1 from EO BD2 resulted also in diverse patterns
of clinical correlates of EO. Indeed, EO BD2 patients
had a higher rate of alcohol dependence compared to
both LO BD2 and EO BD1. Similarly, previous
studies investigating clinical correlates of AAO subgroups
found that EO BD patients have higher rates of alcohol
(Javaid et al. 2011; Lin et al. 2006)
. However, these
studies either analysed only BD1 individuals (Lin et al.
2006) or did not specify the diagnostic stratification
(Javaid et al. 2011)
. Interestingly, a recent study from
Propper et al. (2015) did not find differences in rates of
alcohol abuse among AAO subgroups in a sample with a
BD1:BD2 ratio of 2:1.
Early onset BD2 patients had more frequently a DMI
type of clinical course, while the MDI type was more
frequently associated with EO BD1. Although not directly
comparable with our study, the findings reported by
Perlis et al. (2004) and Propper et al. (2015) in their samples
relatively balanced in terms of BD1:BD2 ratio, indicated
that very EO and EO BD patients have more frequently
onset episodes of depressive polarity compared to later
onset subgroups. In addition, our findings were
consistent with the work of
Koukopoulos et al. (2013
found that patients with a DMI illness course are more
likely to be BD2, while MDI illness course is
overrepresented in BD1 patients.
Differently from what reported in the literature, our
study failed to confirm in the BD1 subgroup the
wellestablished association of EO BD with a higher familial
load for BD and for mood disorders in general, as well as
with higher rates of suicidal behaviour
(Geoffroy et al.
. On the contrary, EO BD2 showed a higher
familial load for BD (p = 0.02) compared to LO BD. Finally,
although not statistically significant, EO BD2 showed
higher rates of family history for BD as well as for mood
disorders, compared to EO BD1. Similarly,
Baek et al.
) found higher rates of major depression, but not
of BD, in BD2 patients compared to BD1, although their
sample was not stratified according to AAO. Further,
another recent study showed that both BD1 and BD2
presented a similar familial load for mood disorders
(Dell’Osso et al. 2016)
There is compelling evidence that EO BD patients
appear also to be more frequently associated with rapid
cycling, drug abuse, higher rates of obsessive–compulsive
disorder, and possibly for psychotic features, and panic
(Geoffroy et al. 2013)
. Although our study did
not test for association most of these clinical correlates,
there were no statistically significant associations with
EO for drug dependence. Of note, most of this evidence
is derived from BD1 samples as there is a lack of data
on the analysis of AAO in BD2, while only a few
studies investigated mixed samples with both BD1 and BD2
(Perlis et al. 2004; Severino et al. 2009; Tozzi
et al. 2011; Ortiz et al. 2011; Propper et al. 2015)
Our results should be interpreted in the context of
some limitations. The retrospective assessment of AAO
might have been subject to recall bias. However, data
were gathered through direct interview of the patients as
well as with systematic review of medical charts
decreasing the probability of a systematic bias in the assessment
of AAO. An additional limitation is the lack of a
systematic approach in collecting family history data, which
might have influenced the assessment of familial load in
our sample. Moreover, external corroboration for AAO
was obtained, whenever possible, by directly interviewing
a first-degree family member or other significant
individuals. Further, our samples of BD1 and BD2 patients might
not have had an adequate statistical power to detect
association signals of small to moderate magnitude. Finally,
our study did not consider birth cohort effect in our
To our knowledge, this is the first study specifically aimed
at comparing clinical correlates of EO between BD1 and
BD2 patients’ populations using admixture analysis. Our
work found that, beside diverse distributional
properties of AAO, BD1 and BD2 EO subgroups differed also in
their clinical characteristics. Of note, our study identified
a subgroup of EO BD2 with an AAO even earlier than in
the EO BD1 subgroup, characterised by a higher genetic
load (i.e. higher familial load for BD) and at greater risk
of developing alcohol dependence. Should our findings
be replicated in other studies directly comparing AAO
BD1 subgroups with AAO BD2 subgroups, future work
will be able to focus also on the differences in the genetic
and biological makeup, possibly facilitating the search for
reliable disease biomarkers.
BD: bipolar disorder; BD1: BD type 1; BD2: BD type 2; AAO: age at onset;
EO: early onset; LO: late onset; NOS: not otherwise specified; BIC: Schwarz’s
Bayesian information criteria; DMI: depression-(hypo)mania-free interval; MDI:
(hypo)mania-depression-free interval; SCID-I/P: structured clinical interview for
DSM-IV-TR Axis I disorders; KS: Kolmogorov–Smirnov.
MM performed the statistical analysis and wrote the initial draft of the
manuscript. GM designed the study, oversaw data analysis, and critically revised
the manuscript. BC contributed to the study design, and critically revised
the manuscript. FP contributed to the study design, and critically revised the
manuscript. LS, VDA, and VS collected clinical data, and critically revised the
manuscript. MA oversaw data analysis, and critically revised the manuscript. AT
designed the study, oversaw data analysis, and performed some of the
writing. UA designed the study, oversaw data analysis, and performed some of the
writing. All authors read and approved the final manuscript.
The authors have no other relevant affiliations or financial involvement with
any organisation or entity with a financial interest in or financial conflict with
the subject matter or materials discussed in the manuscript.
After a detailed description of the study procedures, informed written consent
to participate in the study was obtained from all patients. The local Ethical
Committees from each participating centre approved the study.
This research did not receive any specific grant from funding agencies in the
public, commercial, or not-for-profit sectors.
Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
Alda M. The phenotypic spectra of bipolar disorder . Eur Neuropsychopharmacol . 2004 ; 14 ( Suppl 2 ): S94 - 9 .
Alda M , Grof P , Rouleau GA , Turecki G , Young LT . Investigating responders to lithium prophylaxis as a strategy for mapping susceptibility genes for bipolar disorder . Prog Neuropsychopharmacol Biol Psychiatry . 2005 ; 29 ( 6 ): 1038 - 45 .
American Psychiatric Association . Diagnostic and statistical manual of mental disorders (4th ed . text rev.) . 1st ed. Washington, DC: American Psychiatric Association; 2000 .
Baek JH , Park DY , Choi J , Kim JS , Choi JS , Ha K , et al. Differences between bipolar I and bipolar II disorders in clinical features, comorbidity, and family history . J Affect Disord . 2011 ; 131 ( 1-3 ): 59 - 67 .
Bauer M , Glenn T , Alda M , Andreassen OA , Angelopoulos E , Ardau R , et al. Influence of birth cohort on age of onset cluster analysis in bipolar I disorder . Eur Psychiatry . 2015 ; 30 ( 1 ): 99 - 105 .
Bellivier F , Etain B , Malafosse A , Henry C , Kahn JP , Elgrabli-Wajsbrot O , et al. Age at onset in bipolar I affective disorder in the USA and Europe . World J Biol Psychiatry . 2014 ; 15 ( 5 ): 369 - 76 .
Bellivier F , Golmard JL , Henry C , Leboyer M , Schurhoff F. Admixture analysis of age at onset in bipolar I affective disorder . Arch Gen Psychiatry . 2001 ; 58 ( 5 ): 510 - 2 .
Bellivier F , Golmard JL , Rietschel M , Schulze TG , Malafosse A , Preisig M , et al. Age at onset in bipolar I affective disorder: further evidence for three subgroups . Am J Psychiatry . 2003 ; 160 ( 5 ): 999 - 1001 .
Belmonte MP , Pirooznia M , Goes FS , Seifuddin F , Steele J , Lee PH , et al. Genome-wide association analysis of age at onset and psychotic symptoms in bipolar disorder . Am J Med Genet B Neuropsychiatr Genet . 2011 ; 156B ( 3 ): 370 - 8 .
Benaglia T , Chauveau D , Hunter DR , Young D. mixtools: an R package for analyzing finite mixture models . J Stat Softw . 2009 ; 32 ( 6 ): 1 - 29 .
Dell'Osso B , Dobrea C , Cremaschi L , Buoli M , Miller S , Ketter TA et al. Italian bipolar II vs I patients have better individual functioning, in spite of overall similar illness severity . CNS Spectr 2016 ; 1 - 8 .
Etain B , Dumaine A , Mathieu F , Chevalier F , Henry C , Kahn JP , et al. A SNAP25 promoter variant is associated with early-onset bipolar disorder and a high expression level in brain . Mol Psychiatry . 2010 ; 15 ( 7 ): 748 - 55 .
Etain B , Mathieu F , Rietschel M , Maier W , Albus M , McKeon P , et al. Genomewide scan for genes involved in bipolar affective disorder in 70 European families ascertained through a bipolar type I early-onset proband: supportive evidence for linkage at 3p14 . Mol Psychiatry . 2006 ; 11 ( 7 ): 685 - 94 .
First MB , Spitzer RL , Gibbon M , Williams JBW . Structured clinical interview for DSM-IV-TR Axis I disorders , Research Version, Patient Edition . (SCID-I/P) . New York: Biometrics Research, New York State Psychiatric Institute; 2002 .
Fraley C , Raftery A . MCLUST : software for model-based cluster analysis . J Classif . 1999 ; 16 ( 2 ): 297 - 306 .
Fraley C , Raftery A , Scrucca L . Normal mixture modeling for model-based clustering, classification, and density estimation . R 2014 ;( 4 .3).
Fraley C , Raftery AE . Model-based methods of classification: using the mclust software in chemometrics . J Stat Softw . 2007 ; 18 ( 6 ): 1 - 13 .
Geoffroy PA , Etain B , Scott J , Henry C , Jamain S , Leboyer M , et al. Reconsideration of bipolar disorder as a developmental disorder: importance of the time of onset . J Physiol Paris. 2013 ; 107 ( 4 ): 278 - 85 .
Goes FS , Zandi PP , Miao K , McMahon FJ , Steele J , Willour VL , et al. Moodincongruent psychotic features in bipolar disorder: familial aggregation and suggestive linkage to 2p11 -q14 and 13q21-33. Am J Psychiatry . 2007 ; 164 ( 2 ): 236 - 47 .
Golmard JL , Scott J , Etain B , Preisig M , Aubry JM , Henry C , et al. Using admixture analysis to examine birth-cohort effects on age at onset of bipolar disorder . Acta Psychiatr Scand . 2015 ; 133 ( 3 ): 205 - 13 .
Goodwin FK , Jamison KR . Course and outcome . In: Goodwin FK , Jamison KR , editors. Manic-depressive illness . New York: Oxford University Press; 2007 . p. 119 - 54 .
Grigoroiu-Serbanescu M , Rietschel M , Hauser J , Czerski PM , Herms S , Sun X , et al. Commingling analysis of age-of-onset in bipolar I disorder and the morbid risk for major psychoses in first degree relatives of bipolar I probands . J Affect Disord . 2014 ; 168 : 197 - 204 .
Hamshere ML , Gordon-Smith K , Forty L , Jones L , Caesar S , Fraser C , et al. Ageat-onset in bipolar-I disorder: mixture analysis of 1369 cases identifies three distinct clinical sub-groups . J Affect Disord . 2009 ; 116 ( 1-2 ): 23 - 9 .
Jamain S , Cichon S , Etain B , Muhleisen TW , Georgi A , Zidane N , et al. Common and rare variant analysis in early-onset bipolar disorder vulnerability . PLoS ONE . 2014 ; 9 ( 8 ): e104326 .
Javaid N , Kennedy JL , De Luca V. Ethnicity and age at onset in bipolar spectrum disorders . CNS Spectr . 2011 ; 16 ( 6 ): 127 - 34 .
Kennedy N , Everitt B , Boydell J , VAN Os J , Jones PB , Murray RM . Incidence and distribution of first-episode mania by age: results from a 35-year study . Psychol Med . 2005 ; 35 ( 6 ): 855 - 63 .
Koukopoulos A , Reginaldi D , Tondo L , Visioli C , Baldessarini RJ . Course sequences in bipolar disorder: depressions preceding or following manias or hypomanias . J Affect Disord . 2013 ; 151 ( 1 ): 105 - 10 .
Lin PI , McInnis MG , Potash JB , Willour V , MacKinnon DF , Depaulo JR , et al. Clinical correlates and familial aggregation of age at onset in bipolar disorder . Am J Psychiatry . 2006 ; 163 ( 2 ): 240 - 6 .
Lin PI , McInnis MG , Potash JB , Willour VL , MacKinnon DF , Miao K , et al. Assessment of the effect of age at onset on linkage to bipolar disorder: evidence on chromosomes 18p and 21q . Am J Hum Genet . 2005 ; 77 ( 4 ): 545 - 55 .
Manchia M , Lampus S , Chillotti C , Sardu C , Ardau R , Severino G , et al. Age at onset in Sardinian bipolar I patients: evidence for three subgroups . Bipolar Disord . 2008 ; 10 ( 3 ): 443 - 6 .
Merikangas KR , Jin R , He JP , Kessler RC , Lee S , Sampson NA , et al. Prevalence and correlates of bipolar spectrum disorder in the world mental health survey initiative . Arch Gen Psychiatry . 2011 ; 68 ( 3 ): 241 - 51 .
Montlahuc C , Curis E , Jonas SF , Bellivier F , Chevret S . Age-at-onset subsets of bipolar I disorders: a critical insight into admixture analyses . Int J Methods Psychiatr Res . 2016 ; 1 - 8 .
Ortiz A , Bradler K , Slaney C , Garnham J , Ruzickova M , O'Donovan C , et al. An admixture analysis of the age at index episodes in bipolar disorder . Psychiatry Res . 2011 ; 188 ( 1 ): 34 - 9 .
Perlis RH , Miyahara S , Marangell LB , Wisniewski SR , Ostacher M , DelBello MP , et al. Long-term implications of early onset in bipolar disorder: data from the first 1000 participants in the systematic treatment enhancement program for bipolar disorder (STEP-BD) . Biol Psychiatry . 2004 ; 55 ( 9 ): 875 - 81 .
Post RM , Luckenbaugh DA , Leverich GS , Altshuler LL , Frye MA , Suppes T , et al. Incidence of childhood-onset bipolar illness in the USA and Europe . Br J Psychiatry . 2008 ; 192 ( 2 ): 150 - 1 .
Propper L , Ortiz A , Slaney C , Garnham J , Ruzickova M , Calkin CV , et al. Earlyonset and very-early-onset bipolar disorder: distinct or similar clinical conditions? Bipolar Disord . 2015 ; 17 ( 8 ): 814 - 20 .
R Development Core Team. R: a language and environment for statistical computing . R 2008 .
Severino G , Manchia M , Contu P , Squassina A , Lampus S , Ardau R , et al. Association study in a Sardinian sample between bipolar disorder and the nuclear receptor REV-ERBalpha gene, a critical component of the circadian clock system . Bipolar Disord . 2009 ; 11 ( 2 ): 215 - 20 .
Tozzi F , Manchia M , Galwey NW , Severino G , Del Zompo M , Day R , et al. Admixture analysis of age at onset in bipolar disorder . Psychiatry Res . 2011 ; 185 ( 1-2 ): 27 - 32 .