.
. |
Af fective Disorders
Af fective Disorders
Research report
Temperament and character t raits predict future burden of depression
Tom Rosenström
a, d,n
, Pekka Jylhä
b,e
, C. Robert Cloninger
c
, Mirka Hintsanen
a, h
,
Marko Elovainio
a,d
, Outi Mantere
b,f ,g
, Laura Pulkki-RÃ¥back
a
, Kirsi Riihimäki
b
,
Maria Vuorilehto
b ,f
, Liisa Keltikangas-Järvinen
a
, Erkki Isometsä
b, f, g
a
IBS, Unit of Personality, Work and Health Psychology, University of Helsinki, Helsinki, Finland
b
Department of Mental Health and Substance Abuse Services, National Institute of Health and Welfare, Helsinki, Finland
c
Department of Psychiatry, Washington University School of Medicine, St. Louis, MO, USA
d
National Institute for Health and Welfare, Helsinki, Finland
e
Department of Psychiatry, Jorvi Hospital, Helsinki University Central Hospital, Espoo, Finland
f
Department of Psychiatry, Institute of Clinical Medicine, University of Helsinki, Helsinki, Finland
g
Department of Psychiatry, Helsinki University Central Hospital, Helsinki, Finland
h
Helsinki Collegium for Advanced Studies, University of Helsinki, Helsinki, Finland
article info
Article history:
Received 28 October 2013
Received in revised form
27 January 2014
Accepted 28 January 2014
Available online 11 February 2014
Keywords:
Personality
Major depressive disorder
Bipolar disorder
Mood disorders
Longitudinal data
Prevention
abstract
Background: Personality traits are associated with depressive symptoms and psychiatric disorders.
Evidence for their value in predicting accumulation of future dysphoric episodes or clinical depression
in long-term follow-up is limited, however.
Methods:Within a 15-year longitudinal study of a general-population cohort (N ¼ 751), depressive
symptoms were measured at four time points using Beck's Depression Inventory. In addition, 93 primary
care patients with DSM-IV depressive disorders and 151 with bipolar disorder, diagnosed with SCID-I/P
interviews, were followed for fi ve and 1.5 years with life-chart methodology, respectively. Generalized
linear regression models were used to predict future number of dysphoric episodes and total duration of
major depressive episodes. Baseline personality was measured by the Temperament and Character
Inventory (TCI).
Results: In the general-population sample, one s.d. lower Self-directedness predicted 7.6-fold number of
future dysphoric episodes; for comparison, one s.d. higher baseline depressive symptoms increased the
episode rate 4.5-fold. High Harm-avoidance and low Cooperativeness also implied elevated dysphoria
rates. Generally, personality traits were poor predictors of depression for specifi c time points, and in
clinical populations. Low Persistence predicted 7.5% of the variance in the future accumulated depression
in bipolar patients, however.
Limitations: Degree of recall bias in life charts, limitations of statistical power in the clinical samples, and
21– 79% sample attrition (corrective imputations were performed).
Conclusion:TCI predicts future burden of dysphoric episodes in the general population, but is a weak
predictor of depression outcome in heterogeneous clinical samples. Measures of personality appear more
useful in detecting risk for depression than in clinical prediction.
& 2014 Elsevier B.V. All rights reserved.
1. Introduction
Depression is a common disorder with a high risk of episode
recurrence over time ( Vos et al., 2012; Hardeveld et al., 2013 ).
Predicting future chronicity and recurrence of depression is clini-cally important, for targeting treatment. Preceding episodes, family
history of depression ( Hardeveld et al., 2013), and comorbidity
(Melartin et al., 20 04) predict recurrence; less obvious factors, such
as body-image dissatisfaction (Rosenström et al., 2013 ), may con-tribute. Previous studies have also found that personality traits,
such as those de fined by the Psychobiological Model of Personality
(Cloninger, 1987; Cloninger et al., 1993 ), are predictive of depressive
symptoms measured 3 months (Na i t o e t a l. , 2 0 0 0), a year (Cloninger
et al., 20 06), and 4 years ( Elovainio et al., 20 04; Farmer and Seeley,
20 09) later,suggestinga more generalbackgroundbehindaccu-mulation of depressive and dysphoric episodes. Other evidence that
personality predicts risk of depression has been obtained with
Contents lists available at ScienceDirect
journal homep age: www.elsevier.com/loc at e/jad
Journal of Affective Disorders
htt p ://dx.doi.org/10.1016/j.jad.2014.01.017
0165-0327 & 2014 Elsevier B.V. All rights reserved.
n
Correspondence to: University of Helsinki, Siltavuorenpenger 1A , P.O. Box 9,
Finland. Tel.: þ 358 9 1912 9396; fax: þ 358 9 1912 9521.
E-mail address: tom.rosenstrom@helsinki. fi (T. Rosenström).
Journal of Affective Disorders 158 (2014) 139 – 14 7
measures of coping in relation to concurrent and future depression
in community samples ( Rohde et al., 1990) and with antecedent
personality traits in never-depressed siblings of depressives com-pared to never-depressed siblings of controls (Farmer et al., 20 03 ).
Prior work with Cloninger's psychobiological model of person-ality shows that the risk of depression is associated with high
Harm Avoidance, low Self-directedness, and low Persistence
(Cloninger et al., 2012, 2010; Farmer et al., 20 03 ). Conversely,
resilience is associated with low scores in Harm Avoidance, and
high scores in Self-directedness, Cooperativeness, and Persistence
(Elye et al., 2013). A brain imaging study showed that these
personality traits can be linked with a speci fic brain circuit that
modulates mood and reward-seeking behavior (Gusnard et al.,
20 03; Cloninger et al., 2012 ). Dysfunctional attitudes that increase
the risk of depression are largely explained by low Self-directed-ness, as expected from the cognitive theory of depression, but the
other personality variables in fl uence in particular circumstances
(Luty et al., 1999; Richter and Eisemann, 20 02; Otani et al., 2013).
Dysphoric, or subclinical, symptoms are strongly associated
with functional impairment ( Karsten et al., 2010), and show no
clear empirical boundary with respect to more severe forms of
depression ( Haslam et al., 2012 ). Sample differences among gen-eral and clinical populations are likely, however. Simultaneously
studying longitudinal accumulation in clinical and general popula-tions offers the opportunity to examine which personality traits
have prognostic value under what starting points (e.g., for ran-domly chosen individual versus randomly chosen mood-disorder
patient). The potential differences among different clinical popula-tions are studied herein using two separate clinical populations;
one with bipolar disorder and another with unipolar depressive
disorder. We concentrate on the predictive value of personality
traits for future dysphoric/depressive episode accumulation rather
than on future depression at single time points. In the general
population, the outcome is rate of future dysphoric episodes; in
clinical populations, the outcome is the proportion of follow-up
with participant fulfi lling the DSM-IV criteria for a major depres-sive episode.
The aim of this study was to provide an answer to two
questions. First, are there personality traits that predispose people
to a higher or lower rate of future dysphoric episodes compared to
the base rate in the general population? Second, which personality
traits predict future burden of major depressive episode for
unipolar and bipolar mood disorder patients? These results may
have clear and immediate clinical utility, as the importance of
prevention efforts for depression has been recently emphasized
(Ghaemi et al., 2013). Personality is an attractive candidate for
detection of at-risk groups, as it is malleable, yet more stable than
the actual target of prevention— depressive episodes ( Klein et al.,
2011).
2. Methods
This study used one data set with a random sample from the
general population and two samples from clinical populations of
psychiatric patients.
2.1. Participants from Young Finns study (YFS)
YFS is an ongoing prospective study with the fi rst data collec-tion in 1980 (Raitakari et al., 20 0
. The original sample consists of
3596 healthy Finnish children and adolescents (1832 women, 176 4
men) sampled from six birth cohorts with approximately equal
frequency (born 1962, 1965, 196 8, 1971, 1974, or 1977). In order to
select a b roadly sociodemographically representative sample,
Finland was divided into fi ve areas according to locations of
university cities with a medical school (Helsinki, Kuopio, Oulu,
Tampere, and Turku). In each area, urban and rural boys and girls
were randomly selected on the basis of their unique personal
social-security number. All participants gave written informed
consent and the study was approved by the ethical committee of
the Varsinais-Suomi's hospital district's federation of municipali-ties. The sample has been followed subsequently in 8 data collec-tion waves in 1983, 1986, 1989, 1992, 1997, 20 01, 20 08, and 2012,
but only data from the four latter waves contained the required
measures of both depressive symptoms and personality. Data from
the year 1997 formed the baseline data, whereas the 20 01, 20 08,
and 2012 follow-ups were used for evaluating future dysphoria
and depressive symptoms.
Altogether 751 participants (256 men and 4 95 women) pro-vided all data needed for the intended analyses in YFS data. The
study attrition was 79% from the initial year-1980 sample, and 56%
from those with baseline data available ( n ¼ 1690). Of ten, those
who lack data in YFS have more psychopathology-related person-ality traits and depressive symptoms, and are more likely to be
young and male, compared to retained participants ( Rosenström
et al., 2012a, 2012b). Correlates of attrition are same in clinical
studies (Melartin et al., 20 0 4 ). Supplementary on-line material
presents an imputation analysis, indirectly testing the sensitivity
of the fi ndings for missing observations. For simplicity, the main
manuscript presents non-imputed estimates; both should be
provided in some form, when possible (White et al., 2011).
2.2. Participants from Vantaa Primary Care Depression Study (PC-VDS)
Baseline data collection of the PC-VDS was based on stratifi ed
sampling from two districts within the city of Vantaa, Finland,
during the year 20 02 (population 63,40 0). Primary care patients
aged 20– 69 from general practitioners' waiting rooms were
screened by using Primary Care Evaluation of Mental Disorders,
PRIME-MD ( Spitzer et al., 1994), from three health centers and two
maternity clinics. A total of 1119 participants were addressed, of
which 402 screened positive for depressive symptoms; 37 of these
refused to participate in the study and the rest gave their written
informed consent. In the second phase, a diagnosis was made by a
psychiatrist using the Structured Clinical Interview for DSM-IV
axis I disorders (SCID-I/P; First et al., 20 02). All available informa-tion from face-to-face interviews and psychiatric records was
used; if the diagnosis was uncertain, other informants were
contacted. To exclude substance-induced mood disorder, patients
who were currently abusing alcohol or other substances were
interviewed af ter 2 – 3 weeks of abstinence. The fi nal baseline
cohort consisted of 137 depressive disorder patients. Two thirds
had major depressive disorder (MDD), the rest being diagnosed
with dysthymia, subsyndromal MDD with 2– 4 symptoms (mini-mum one core symptom) and lifetime MDD, or minor depression
otherwise similar to subsyndromal MDD, but without history of
MDD. Distress or functional impairment was required. Interrater
reliability for current depressive disorder, evaluated from 20
randomly selected videotaped interviews, was perfect [ κ ¼ 1. 0
(Vuorilehto et al., 20 05, 20 09; Riihimäki et al., 2011 )].
The participants were followed again af ter 6 and af ter 18
months, and af ter 5 years from the baseline. A life chart of the
entire 5-year follow-up period was constructed for the patients by
one of the two interviewers to determine the duration of the index
episode and the timing of possible relapses and recurrences using
all available medical and psychiatric records to complement the
information. Altogether 93 participants provided the necessary
personality and depression inventories at the baseline, and the full
life-chart information. Hence, study attrition was between 32%
and 47%, depending on the unknown clinical status of refused
patients. Further details of the sample can be found from previous
T. Rosenström et al. / Journal of Af fective Disorders 158 (2014) 139 – 14 7 14 0
publications (Vuorilehto et al., 20 05, 20 09; Jylhä et al., 2011;
Riihimäki et al. , 2011).
2.3. Participants from Jorvi Bipolar Study (JoBS)
The patients for the JoBS were screened from those of the
Department of Psychiatry at the Jorvi Hospital (part of Helsinki
University Central Hospital), serving the adjacent cities of Espoo,
Kauniainen, and Kirkkonummi in Finland during the year 20 02
(population 261,10 0). All patients, excluding those with schizo-phrenia (n ¼ 1630), were screened with the Mood Disorder Ques-tionnaire (MDQ); and 546 positive screens for bipolar disorder
(BD) were found; 91 participants refused and the rest gave their
written informed consent. In the second phase, a diagnosis was
made by one of six psychiatrist using the Structured Clinical
Interview for DSM-IV axis I disorders (SCID-I/P; First et al.,
20 02). All available information from face-to-face interviews and
psychiatric records was used; if the diagnosis was uncertain, other
informants were contacted. Altogether 191 patients were assigned
a research diagnosis of DSM-IV type I or type II BD; interrater
reliability of BD and type I or II diagnoses, evaluated by 20
randomly selected interviews, was perfect ( κ ¼ 1.0). Details of
baseline methodology have been published elsewhere (Mantere
et al., 20 0 4).
The participants were followed again af ter 6 and af ter 18
months. Graphic life charts of the follow-up period were con-structed individually for each patient, as in PC-VDS. Altogether 151
patients provided the necessary personality and depression inven-tories at the baseline, and the full life-chart information. Hence,
study attrition was between 21% and 46%, depending on the
unknown diagnostic status of refused patients. Further methodo-logical details can be found from previous publications ( Jylhä et al.,
2011; Mantere et al., 20 0.
2.4. Measures
A modifi ed version of the Beck's Depression Inventory (mBDI)
was used in the general-population YFS to measure depressive
symptoms (Cronbach's α ¼ 0.91 in 1997, 0.92 in 20 01, 0.93 in 20 08,
and 0.93 in 2012). In the modifi ed version, subject rank s to what
degree (a 5-point scale from‘no’ to ‘ very much ’) he or she suffers
from the ailment presented in the second mildest symptom
description of the original Beck's Depression Inventory; such
mo dified versions of clinical scales are frequently used because
they better represent the general-population variation in the
symptoms than the original clinically oriented scales ( Rosenström
et al., 2012b ). In year 20 08, the participants also fulfilled Beck's
Depression Inventory II (BDI-II,α ¼ 0.8 for which a national
standardization has been published (Beck et al., 20 04 ); BDI and
BDI-II are highly similar measures that are strongly correlated (at
0.93) with each other (Beck et al., 1996). Using the 20 08 mBDI and
BDI-II measures, a general relationship between the mBDI and BDI-II scales was established (see beginning of the Results section).
Further psychometric analyses of relationships between mBDI and
BDI has been published elsewhere, including Item Response Theory
modeling and various attrition analyses (Rosenström et al., 2012b;
Rosenström, 2013b).
In the national standardization, BDI-II scores above 13 points
signify at least mild depression, a state referred to as dysphoric
episode herein. Via the established mBDI to BDI-II relationship, it
was possible to count the dysphoric episodes across all the three
follow-ups af ter the baseline. The total number of dysphoric
episodes within given number of assessments/follow-ups is
referred to as caseness , as in previous studies ( Jokela et al., 2011).
Caseness is, for the population-based YFS, a related measure to the
proportion of time a person suffers from a depressive episode as
measured from the life chart for the clinical data.
In the clinical data, all collected information was integrated into
a graphic of a life chart together with the patient. In addition to
symptom ratings, change points in psychopathological states were
inquired using probes related to important life-events in order to
improve accuracy of the assessment. From the life charts, propor-tions of time in the follow-up during which the participants
fulfi lled DSM-IV criteria for MDE (5 or more of the 9 symptoms;
SCID-I/P;First et al., 20 02 ) were computed ( Holma et al., 20 08;
Vuorilehto et al., 20 09). Accuracy of, or information in, life charts
must considerably exceed simple interpolation from face-to-face
follow-up assessments (see Sections 2.2 and 2.3), but cannot be
quanti fi ed further, as the patients were not under full-time
continuous surveillance. The participants also fi lled in the Beck's
Depression Inventory [BDI (Beck and Steer, 1993), 0.86 r α r 0.95].
In addition to the depression assessments, personality as
defi ned by the Psychobiological Model of Personality ( Cloninger,
1987; Cloninger et al., 1993 ) was assessed in the baseline follow-up of both the population-based YFS study and the clinical studies.
The PC-VDS and JoBS used the Revised version of Temperament
and Character Inventory (Cronbach's α ¼ 0.81 0.94), while YFS
used the Temperament and Character Inventory (internal consis-tencies below) modifi ed to correspond to the revised version with
a 5-point Likert scale (Cloninger et al., 1994). A personality trait is a
continuous measure for individual differences occurring along
certain dimension of behavior and thought. The main personality
traits that were used are briefl y described below, and more
detailed description has been published elsewhere ( Cloninger
et al., 1993).
Novelty seeking is a tendency toward excitement and activation
of behavior in response to novel stimuli, or in response to cues of
potential rewards or potential relief of punishment (40 items,
α ¼ 0.85 in YFS). Harm avoidanceis a tendency to inhibit behavior
in response to signals of aversive stimuli or frustrative non-reward
(35 items, α ¼ 0.92). Reward dependenceis a tendency to form
social attachments in response to signals of reward (especially to
signals of social approval; 24 items, α ¼ 0.80). Persistence is a
tendency to maintain or resist extinction of behavior previously
associated with intermittent rewards or relief from punishment (8
items, α ¼ 0.6 4). Self-directedness is a tendency to set and to strive
towards self-determined rather than externally infl uenced life
goals, and to attribute causes for the consequences of one's actions
to oneself rather than to other peoples or external circumstances
(4 4 items, α ¼ 0.89). Cooperativenessrefers to ability and desire to
co-operate with other people (42 items, α ¼ 0.91). Finally, Self-transcendence is a tendency to be aware of connections with what
is beyond the individual self, referring to personal qualities such as
spirituality and universal values (33 items, α ¼ 0.91).
2.5. Statistical analyses
Regression models for future accumulation of depression were
estimated, with baseline personality traits as predictors. First,
‘individual effects’ of traits and depression scores in predicting
the amount of time a person was depressed were estimated
(Model 0). Then, Model I assessed what personality traits con-tribute when adjustment is made for the baseline depressive-symptoms summary score. In YFS, we also adjusted for the
presence of a dysphoric episode as a dichotomous variable at
baseline (Model II). Finally, Model III assessed the contribution of
each variable controlling for all the other traits and/or the
depression score, that is, a full multiple regression was estimated.
Despite this conceptual division, every regression model was
adjusted for sex and age (single continuous variable in clinical
data; five cohort indicators in YFS).
T. Rosenström et al. / Journal of Af fective Disorders 158 (2014) 139 – 14 7 14 1
As the outcome variable was either a count (in YFS) or a
proportion (in PC-VDS and JoBS), generalized linear regression
models were used (Gelman and Hill, 20 07). In YFS, Quasi-Poisson
regression was applied (sensitivity analysis with Ordered Logistic
regression in on-line supplement). In the clinical data sets, two
complementary approaches were taken. Proportions are fre-quently modeled by transforming them into a continuous variable
by Logit transformation, but this does not work when 0 or
1 proportions exist in the data, as the Logit transformation for
the former is minus in fi nity and for the latter plus infi nity [the
Logit transformation is the map p-log{ p /(1-p )} from open interval
(0, 1) to the real line]. Therefore, we also applied Infl ated Beta
Regression, which can handle the extreme values as well (Ospina
and Ferrari, 2010; Stasinopoulos and Rigby, 20 07 ), thereby allow-ing the use of all eligible data. In addition to being a sensitivity
analysis, the approach with an explicit Logit transformation also
allows for presenting the coefficient of determinations (R
2
) and the
change resulting from adding an independent variable into a
regression model (ΔR
2
). R
2
value signifies the proportion of
outcome-variable variance explained by the model; herein, “ R
2
â€
always refers to the covariate-number “adjusted R
2
†(Gelman and
Hill, 20 07). Notice that change ΔR
2
for adjusted R
2
can be negative
for a bad predictor.
All analyses were performed using R-sof tware 6 4-bit version
2.15.3, and for the Infl ated Beta Regression, GAML SS R-package
(Core Team, 2012; Stasinopoulos and Rigby, 20 07). Statistical
comparisons between two linear models were based either on
the classical F-test or on the Akaike's Information Criterion [AIC
(Stasinopoulos and Rigby, 20 07)]. Continuous independent vari-ables in regression models were standardized z -scores. The statis-tical p -values in Table 1 are from two-tailed t -test of equal means.
2.6. On the interpretation of outcome variables
In the clinical samples (JoBS and PC-VDS), we were able to
explicitly compute the proportion of follow up that a participant
suffered from symptoms fulfi lling Major Depressive Episode cri-teria by using the life charts and hospital records. Hence, direct
associations between personality measured at baseline and pro-portion of time depressed could be evaluated. In the general
population (YFS), however, the participants were sampled only
in discrete time points, without knowledge of their emotional
states in between. As the temporal sampling points determined by
the study protocol can be considered unrelated to individuals'
emotional processes, the number of dysphoric states observed
during the sampling times should be monotonically related to
their general rate of occurrence. Therefore, estimated changes in
base rates due to a covariate should be reasonably comparable
across partially observed general-population trajectories and more
fully observed clinical trajectories. It should be kept in mind,
however, that quasi-Poisson models assess relative increases in
base rate of dysphoric episodes rather than absolute increases.
In addition to comparability between samples, there was
another reason for studying accumulation of discrete dysphoric
episodes instead simple sums of symptom sums over several time
points. This way one avoids confounding cases of repeatedly
elevated scores with cases of a single very high score and several
quite low scores.
3. Results
Because the distances among the levels of depressive-symptom
severity are encoded differently by mBDI and BDI-II ( Rosenström
et al., 2012b), the inclusion of a quadratic component was required
in modeling the relationship between the mBDI and the clinically
oriented BDI-II ( β
quadratic¼ 1. 4 8 , S.E. ¼ 0.07,p o 0.0 01, ΔR
2
¼ 0.085;
see Fig. 1A). Further nonlinear components were not needed in the
model (p ¼ 0.998 and ΔR
2
¼ 0.0 0 for a cubic term; other relevant
estimates were: R
2
¼ 0.695; β
linear ¼ 4.24 andβ
intercept¼ 4.02). Cor-relation between the established model estimate and measured
BDI-II was 0.83, which is close to the maximum possible [Cron-bach's (alpha) reliability of the BDI-II was 0.88]. A dysphoric
episode was defi ned by this estimated quadratic transformation
(i.e., BDI-II modeled by the mBDI) exceeding 13 points of BDI-II
score (see Measures section 2.4). Af ter the baseline-year 1997,
there were 3 non-baseline follow-ups, and hence three dysphoric
episodes were maximum number of ‘ future ’ episodes in YFS
(Fig. 1
.
As can be seen fromFig. 1A , the predictor/proxy for the BDI-II-defined mild depression had higher speci fi city (0.9 than sensi-tivity (0.65). Speci fi city implied that episodes detected by the
model almost always re flected at least mild episodes according to
BDI-II. Sensitivity implied that we missed 35% of such episodes,
suggesting that regression estimates below are underestimates
rather than overestimates. The issue of sensitivity is pertinent,
however, only so far as one prefers BDI-II over mBDI in de fining
dysphoric episodes.
Table 1 presents the basic characteristics of the studied sam-ples. Whereas the proportion of time as depressed is shown for
clinical PC-VDS and JoBS data sets, the number of dysphoric
episodes in the three non-baseline follow-ups is shown for YFS
general-population data that lacks the life chart methodology. On
average, the clinical patients had approximately 14– 17 points
higher BDI compared to the YFS participants' (BDI-II proxy).
Table 1
Basic sample characteristics and their comparison.
Variable PC-VDS YFS p-value
Median Mean s.d. Median Mean Range/s.d.
Age at the baseline 46.15 4 4.27 13.85 29 27.61 20– 35 o 0.0 01
Age at the fi nal follow up 51.15 49. 27 13.85 4 4 42.61 35– 50 –
Proportion/number of episode(s) 0.18 0.32 0.34 0 0.32 0.71 –
BDI or BDI-II proxy at baseline 17 19.54 10.51 3.79 5.63 5.30 o 0.0 01
JoBS
Age at the baseline –––37.93 38.54 11.72 0.0 01
Age at the fi nal follow up –––39.43 40.0 4 11.72 –
Proportion of derpressive episode(s) –––0.27 0.35 0.32 –
Logit of depressive-episode proportion 1.32 1. 17 1. 8 9 0.55 0.65 1.65 0.056
BDI or BDI-II proxy at baseline –––23 22.07 11.78 0.082
Note: “ PC-VDS†¼ Primary Care Vantaa Depression Study; “ YFS †¼ Young Finns (general-population) Study; “ JoBS †¼ Jorvi Bipolar Study. p -value is provided for a column-wise
t-test when such test made sense; “ s.d. †¼ standard deviation, range is given for age at YFS that had six approximately equally large cohorts.
T. Rosenström et al. / Journal of Af fective Disorders 158 (2014) 139 – 14 7 14 2
3.1. Main results for general population
Table 2 shows regression coeffi cients from quasi-Poisson mod-els predicting the number of future dysphoric episodes with the
baseline measures in general-population participants. Baseline
measures included score of depressive symptoms (mBDI), indica-tor of dysphoric episode at baseline, and personality traits; all
independent variables were standardized except dichotomous
indicator variables. For all models, low current Self-directedness
predicted the greatest increase in the rate of future dysphoric
episodes, among personality traits. Also high Harm avoidance, low
Cooperativeness, and depressive symptoms contributed strongly.
The presence of a dysphoric episode at baseline was a non-signifi cant predictor af ter adjusting for the continuous depression
Fig. 1. Outcome Variab
Research report
Temperament and character t raits predict future burden of depression
Tom Rosenström
a, d,n
, Pekka Jylhä
b,e
, C. Robert Cloninger
c
, Mirka Hintsanen
a, h
,
Marko Elovainio
a,d
, Outi Mantere
b,f ,g
, Laura Pulkki-RÃ¥back
a
, Kirsi Riihimäki
b
,
Maria Vuorilehto
b ,f
, Liisa Keltikangas-Järvinen
a
, Erkki Isometsä
b, f, g
a
IBS, Unit of Personality, Work and Health Psychology, University of Helsinki, Helsinki, Finland
b
Department of Mental Health and Substance Abuse Services, National Institute of Health and Welfare, Helsinki, Finland
c
Department of Psychiatry, Washington University School of Medicine, St. Louis, MO, USA
d
National Institute for Health and Welfare, Helsinki, Finland
e
Department of Psychiatry, Jorvi Hospital, Helsinki University Central Hospital, Espoo, Finland
f
Department of Psychiatry, Institute of Clinical Medicine, University of Helsinki, Helsinki, Finland
g
Department of Psychiatry, Helsinki University Central Hospital, Helsinki, Finland
h
Helsinki Collegium for Advanced Studies, University of Helsinki, Helsinki, Finland
article info
Article history:
Received 28 October 2013
Received in revised form
27 January 2014
Accepted 28 January 2014
Available online 11 February 2014
Keywords:
Personality
Major depressive disorder
Bipolar disorder
Mood disorders
Longitudinal data
Prevention
abstract
Background: Personality traits are associated with depressive symptoms and psychiatric disorders.
Evidence for their value in predicting accumulation of future dysphoric episodes or clinical depression
in long-term follow-up is limited, however.
Methods:Within a 15-year longitudinal study of a general-population cohort (N ¼ 751), depressive
symptoms were measured at four time points using Beck's Depression Inventory. In addition, 93 primary
care patients with DSM-IV depressive disorders and 151 with bipolar disorder, diagnosed with SCID-I/P
interviews, were followed for fi ve and 1.5 years with life-chart methodology, respectively. Generalized
linear regression models were used to predict future number of dysphoric episodes and total duration of
major depressive episodes. Baseline personality was measured by the Temperament and Character
Inventory (TCI).
Results: In the general-population sample, one s.d. lower Self-directedness predicted 7.6-fold number of
future dysphoric episodes; for comparison, one s.d. higher baseline depressive symptoms increased the
episode rate 4.5-fold. High Harm-avoidance and low Cooperativeness also implied elevated dysphoria
rates. Generally, personality traits were poor predictors of depression for specifi c time points, and in
clinical populations. Low Persistence predicted 7.5% of the variance in the future accumulated depression
in bipolar patients, however.
Limitations: Degree of recall bias in life charts, limitations of statistical power in the clinical samples, and
21– 79% sample attrition (corrective imputations were performed).
Conclusion:TCI predicts future burden of dysphoric episodes in the general population, but is a weak
predictor of depression outcome in heterogeneous clinical samples. Measures of personality appear more
useful in detecting risk for depression than in clinical prediction.
& 2014 Elsevier B.V. All rights reserved.
1. Introduction
Depression is a common disorder with a high risk of episode
recurrence over time ( Vos et al., 2012; Hardeveld et al., 2013 ).
Predicting future chronicity and recurrence of depression is clini-cally important, for targeting treatment. Preceding episodes, family
history of depression ( Hardeveld et al., 2013), and comorbidity
(Melartin et al., 20 04) predict recurrence; less obvious factors, such
as body-image dissatisfaction (Rosenström et al., 2013 ), may con-tribute. Previous studies have also found that personality traits,
such as those de fined by the Psychobiological Model of Personality
(Cloninger, 1987; Cloninger et al., 1993 ), are predictive of depressive
symptoms measured 3 months (Na i t o e t a l. , 2 0 0 0), a year (Cloninger
et al., 20 06), and 4 years ( Elovainio et al., 20 04; Farmer and Seeley,
20 09) later,suggestinga more generalbackgroundbehindaccu-mulation of depressive and dysphoric episodes. Other evidence that
personality predicts risk of depression has been obtained with
Contents lists available at ScienceDirect
journal homep age: www.elsevier.com/loc at e/jad
Journal of Affective Disorders
htt p ://dx.doi.org/10.1016/j.jad.2014.01.017
0165-0327 & 2014 Elsevier B.V. All rights reserved.
n
Correspondence to: University of Helsinki, Siltavuorenpenger 1A , P.O. Box 9,
Finland. Tel.: þ 358 9 1912 9396; fax: þ 358 9 1912 9521.
E-mail address: tom.rosenstrom@helsinki. fi (T. Rosenström).
Journal of Affective Disorders 158 (2014) 139 – 14 7
measures of coping in relation to concurrent and future depression
in community samples ( Rohde et al., 1990) and with antecedent
personality traits in never-depressed siblings of depressives com-pared to never-depressed siblings of controls (Farmer et al., 20 03 ).
Prior work with Cloninger's psychobiological model of person-ality shows that the risk of depression is associated with high
Harm Avoidance, low Self-directedness, and low Persistence
(Cloninger et al., 2012, 2010; Farmer et al., 20 03 ). Conversely,
resilience is associated with low scores in Harm Avoidance, and
high scores in Self-directedness, Cooperativeness, and Persistence
(Elye et al., 2013). A brain imaging study showed that these
personality traits can be linked with a speci fic brain circuit that
modulates mood and reward-seeking behavior (Gusnard et al.,
20 03; Cloninger et al., 2012 ). Dysfunctional attitudes that increase
the risk of depression are largely explained by low Self-directed-ness, as expected from the cognitive theory of depression, but the
other personality variables in fl uence in particular circumstances
(Luty et al., 1999; Richter and Eisemann, 20 02; Otani et al., 2013).
Dysphoric, or subclinical, symptoms are strongly associated
with functional impairment ( Karsten et al., 2010), and show no
clear empirical boundary with respect to more severe forms of
depression ( Haslam et al., 2012 ). Sample differences among gen-eral and clinical populations are likely, however. Simultaneously
studying longitudinal accumulation in clinical and general popula-tions offers the opportunity to examine which personality traits
have prognostic value under what starting points (e.g., for ran-domly chosen individual versus randomly chosen mood-disorder
patient). The potential differences among different clinical popula-tions are studied herein using two separate clinical populations;
one with bipolar disorder and another with unipolar depressive
disorder. We concentrate on the predictive value of personality
traits for future dysphoric/depressive episode accumulation rather
than on future depression at single time points. In the general
population, the outcome is rate of future dysphoric episodes; in
clinical populations, the outcome is the proportion of follow-up
with participant fulfi lling the DSM-IV criteria for a major depres-sive episode.
The aim of this study was to provide an answer to two
questions. First, are there personality traits that predispose people
to a higher or lower rate of future dysphoric episodes compared to
the base rate in the general population? Second, which personality
traits predict future burden of major depressive episode for
unipolar and bipolar mood disorder patients? These results may
have clear and immediate clinical utility, as the importance of
prevention efforts for depression has been recently emphasized
(Ghaemi et al., 2013). Personality is an attractive candidate for
detection of at-risk groups, as it is malleable, yet more stable than
the actual target of prevention— depressive episodes ( Klein et al.,
2011).
2. Methods
This study used one data set with a random sample from the
general population and two samples from clinical populations of
psychiatric patients.
2.1. Participants from Young Finns study (YFS)
YFS is an ongoing prospective study with the fi rst data collec-tion in 1980 (Raitakari et al., 20 0
. The original sample consists of
3596 healthy Finnish children and adolescents (1832 women, 176 4
men) sampled from six birth cohorts with approximately equal
frequency (born 1962, 1965, 196 8, 1971, 1974, or 1977). In order to
select a b roadly sociodemographically representative sample,
Finland was divided into fi ve areas according to locations of
university cities with a medical school (Helsinki, Kuopio, Oulu,
Tampere, and Turku). In each area, urban and rural boys and girls
were randomly selected on the basis of their unique personal
social-security number. All participants gave written informed
consent and the study was approved by the ethical committee of
the Varsinais-Suomi's hospital district's federation of municipali-ties. The sample has been followed subsequently in 8 data collec-tion waves in 1983, 1986, 1989, 1992, 1997, 20 01, 20 08, and 2012,
but only data from the four latter waves contained the required
measures of both depressive symptoms and personality. Data from
the year 1997 formed the baseline data, whereas the 20 01, 20 08,
and 2012 follow-ups were used for evaluating future dysphoria
and depressive symptoms.
Altogether 751 participants (256 men and 4 95 women) pro-vided all data needed for the intended analyses in YFS data. The
study attrition was 79% from the initial year-1980 sample, and 56%
from those with baseline data available ( n ¼ 1690). Of ten, those
who lack data in YFS have more psychopathology-related person-ality traits and depressive symptoms, and are more likely to be
young and male, compared to retained participants ( Rosenström
et al., 2012a, 2012b). Correlates of attrition are same in clinical
studies (Melartin et al., 20 0 4 ). Supplementary on-line material
presents an imputation analysis, indirectly testing the sensitivity
of the fi ndings for missing observations. For simplicity, the main
manuscript presents non-imputed estimates; both should be
provided in some form, when possible (White et al., 2011).
2.2. Participants from Vantaa Primary Care Depression Study (PC-VDS)
Baseline data collection of the PC-VDS was based on stratifi ed
sampling from two districts within the city of Vantaa, Finland,
during the year 20 02 (population 63,40 0). Primary care patients
aged 20– 69 from general practitioners' waiting rooms were
screened by using Primary Care Evaluation of Mental Disorders,
PRIME-MD ( Spitzer et al., 1994), from three health centers and two
maternity clinics. A total of 1119 participants were addressed, of
which 402 screened positive for depressive symptoms; 37 of these
refused to participate in the study and the rest gave their written
informed consent. In the second phase, a diagnosis was made by a
psychiatrist using the Structured Clinical Interview for DSM-IV
axis I disorders (SCID-I/P; First et al., 20 02). All available informa-tion from face-to-face interviews and psychiatric records was
used; if the diagnosis was uncertain, other informants were
contacted. To exclude substance-induced mood disorder, patients
who were currently abusing alcohol or other substances were
interviewed af ter 2 – 3 weeks of abstinence. The fi nal baseline
cohort consisted of 137 depressive disorder patients. Two thirds
had major depressive disorder (MDD), the rest being diagnosed
with dysthymia, subsyndromal MDD with 2– 4 symptoms (mini-mum one core symptom) and lifetime MDD, or minor depression
otherwise similar to subsyndromal MDD, but without history of
MDD. Distress or functional impairment was required. Interrater
reliability for current depressive disorder, evaluated from 20
randomly selected videotaped interviews, was perfect [ κ ¼ 1. 0
(Vuorilehto et al., 20 05, 20 09; Riihimäki et al., 2011 )].
The participants were followed again af ter 6 and af ter 18
months, and af ter 5 years from the baseline. A life chart of the
entire 5-year follow-up period was constructed for the patients by
one of the two interviewers to determine the duration of the index
episode and the timing of possible relapses and recurrences using
all available medical and psychiatric records to complement the
information. Altogether 93 participants provided the necessary
personality and depression inventories at the baseline, and the full
life-chart information. Hence, study attrition was between 32%
and 47%, depending on the unknown clinical status of refused
patients. Further details of the sample can be found from previous
T. Rosenström et al. / Journal of Af fective Disorders 158 (2014) 139 – 14 7 14 0
publications (Vuorilehto et al., 20 05, 20 09; Jylhä et al., 2011;
Riihimäki et al. , 2011).
2.3. Participants from Jorvi Bipolar Study (JoBS)
The patients for the JoBS were screened from those of the
Department of Psychiatry at the Jorvi Hospital (part of Helsinki
University Central Hospital), serving the adjacent cities of Espoo,
Kauniainen, and Kirkkonummi in Finland during the year 20 02
(population 261,10 0). All patients, excluding those with schizo-phrenia (n ¼ 1630), were screened with the Mood Disorder Ques-tionnaire (MDQ); and 546 positive screens for bipolar disorder
(BD) were found; 91 participants refused and the rest gave their
written informed consent. In the second phase, a diagnosis was
made by one of six psychiatrist using the Structured Clinical
Interview for DSM-IV axis I disorders (SCID-I/P; First et al.,
20 02). All available information from face-to-face interviews and
psychiatric records was used; if the diagnosis was uncertain, other
informants were contacted. Altogether 191 patients were assigned
a research diagnosis of DSM-IV type I or type II BD; interrater
reliability of BD and type I or II diagnoses, evaluated by 20
randomly selected interviews, was perfect ( κ ¼ 1.0). Details of
baseline methodology have been published elsewhere (Mantere
et al., 20 0 4).
The participants were followed again af ter 6 and af ter 18
months. Graphic life charts of the follow-up period were con-structed individually for each patient, as in PC-VDS. Altogether 151
patients provided the necessary personality and depression inven-tories at the baseline, and the full life-chart information. Hence,
study attrition was between 21% and 46%, depending on the
unknown diagnostic status of refused patients. Further methodo-logical details can be found from previous publications ( Jylhä et al.,
2011; Mantere et al., 20 0
.
2.4. Measures
A modifi ed version of the Beck's Depression Inventory (mBDI)
was used in the general-population YFS to measure depressive
symptoms (Cronbach's α ¼ 0.91 in 1997, 0.92 in 20 01, 0.93 in 20 08,
and 0.93 in 2012). In the modifi ed version, subject rank s to what
degree (a 5-point scale from‘no’ to ‘ very much ’) he or she suffers
from the ailment presented in the second mildest symptom
description of the original Beck's Depression Inventory; such
mo dified versions of clinical scales are frequently used because
they better represent the general-population variation in the
symptoms than the original clinically oriented scales ( Rosenström
et al., 2012b ). In year 20 08, the participants also fulfilled Beck's
Depression Inventory II (BDI-II,α ¼ 0.8
for which a national
standardization has been published (Beck et al., 20 04 ); BDI and
BDI-II are highly similar measures that are strongly correlated (at
0.93) with each other (Beck et al., 1996). Using the 20 08 mBDI and
BDI-II measures, a general relationship between the mBDI and BDI-II scales was established (see beginning of the Results section).
Further psychometric analyses of relationships between mBDI and
BDI has been published elsewhere, including Item Response Theory
modeling and various attrition analyses (Rosenström et al., 2012b;
Rosenström, 2013b).
In the national standardization, BDI-II scores above 13 points
signify at least mild depression, a state referred to as dysphoric
episode herein. Via the established mBDI to BDI-II relationship, it
was possible to count the dysphoric episodes across all the three
follow-ups af ter the baseline. The total number of dysphoric
episodes within given number of assessments/follow-ups is
referred to as caseness , as in previous studies ( Jokela et al., 2011).
Caseness is, for the population-based YFS, a related measure to the
proportion of time a person suffers from a depressive episode as
measured from the life chart for the clinical data.
In the clinical data, all collected information was integrated into
a graphic of a life chart together with the patient. In addition to
symptom ratings, change points in psychopathological states were
inquired using probes related to important life-events in order to
improve accuracy of the assessment. From the life charts, propor-tions of time in the follow-up during which the participants
fulfi lled DSM-IV criteria for MDE (5 or more of the 9 symptoms;
SCID-I/P;First et al., 20 02 ) were computed ( Holma et al., 20 08;
Vuorilehto et al., 20 09). Accuracy of, or information in, life charts
must considerably exceed simple interpolation from face-to-face
follow-up assessments (see Sections 2.2 and 2.3), but cannot be
quanti fi ed further, as the patients were not under full-time
continuous surveillance. The participants also fi lled in the Beck's
Depression Inventory [BDI (Beck and Steer, 1993), 0.86 r α r 0.95].
In addition to the depression assessments, personality as
defi ned by the Psychobiological Model of Personality ( Cloninger,
1987; Cloninger et al., 1993 ) was assessed in the baseline follow-up of both the population-based YFS study and the clinical studies.
The PC-VDS and JoBS used the Revised version of Temperament
and Character Inventory (Cronbach's α ¼ 0.81 0.94), while YFS
used the Temperament and Character Inventory (internal consis-tencies below) modifi ed to correspond to the revised version with
a 5-point Likert scale (Cloninger et al., 1994). A personality trait is a
continuous measure for individual differences occurring along
certain dimension of behavior and thought. The main personality
traits that were used are briefl y described below, and more
detailed description has been published elsewhere ( Cloninger
et al., 1993).
Novelty seeking is a tendency toward excitement and activation
of behavior in response to novel stimuli, or in response to cues of
potential rewards or potential relief of punishment (40 items,
α ¼ 0.85 in YFS). Harm avoidanceis a tendency to inhibit behavior
in response to signals of aversive stimuli or frustrative non-reward
(35 items, α ¼ 0.92). Reward dependenceis a tendency to form
social attachments in response to signals of reward (especially to
signals of social approval; 24 items, α ¼ 0.80). Persistence is a
tendency to maintain or resist extinction of behavior previously
associated with intermittent rewards or relief from punishment (8
items, α ¼ 0.6 4). Self-directedness is a tendency to set and to strive
towards self-determined rather than externally infl uenced life
goals, and to attribute causes for the consequences of one's actions
to oneself rather than to other peoples or external circumstances
(4 4 items, α ¼ 0.89). Cooperativenessrefers to ability and desire to
co-operate with other people (42 items, α ¼ 0.91). Finally, Self-transcendence is a tendency to be aware of connections with what
is beyond the individual self, referring to personal qualities such as
spirituality and universal values (33 items, α ¼ 0.91).
2.5. Statistical analyses
Regression models for future accumulation of depression were
estimated, with baseline personality traits as predictors. First,
‘individual effects’ of traits and depression scores in predicting
the amount of time a person was depressed were estimated
(Model 0). Then, Model I assessed what personality traits con-tribute when adjustment is made for the baseline depressive-symptoms summary score. In YFS, we also adjusted for the
presence of a dysphoric episode as a dichotomous variable at
baseline (Model II). Finally, Model III assessed the contribution of
each variable controlling for all the other traits and/or the
depression score, that is, a full multiple regression was estimated.
Despite this conceptual division, every regression model was
adjusted for sex and age (single continuous variable in clinical
data; five cohort indicators in YFS).
T. Rosenström et al. / Journal of Af fective Disorders 158 (2014) 139 – 14 7 14 1
As the outcome variable was either a count (in YFS) or a
proportion (in PC-VDS and JoBS), generalized linear regression
models were used (Gelman and Hill, 20 07). In YFS, Quasi-Poisson
regression was applied (sensitivity analysis with Ordered Logistic
regression in on-line supplement). In the clinical data sets, two
complementary approaches were taken. Proportions are fre-quently modeled by transforming them into a continuous variable
by Logit transformation, but this does not work when 0 or
1 proportions exist in the data, as the Logit transformation for
the former is minus in fi nity and for the latter plus infi nity [the
Logit transformation is the map p-log{ p /(1-p )} from open interval
(0, 1) to the real line]. Therefore, we also applied Infl ated Beta
Regression, which can handle the extreme values as well (Ospina
and Ferrari, 2010; Stasinopoulos and Rigby, 20 07 ), thereby allow-ing the use of all eligible data. In addition to being a sensitivity
analysis, the approach with an explicit Logit transformation also
allows for presenting the coefficient of determinations (R
2
) and the
change resulting from adding an independent variable into a
regression model (ΔR
2
). R
2
value signifies the proportion of
outcome-variable variance explained by the model; herein, “ R
2
â€
always refers to the covariate-number “adjusted R
2
†(Gelman and
Hill, 20 07). Notice that change ΔR
2
for adjusted R
2
can be negative
for a bad predictor.
All analyses were performed using R-sof tware 6 4-bit version
2.15.3, and for the Infl ated Beta Regression, GAML SS R-package
(Core Team, 2012; Stasinopoulos and Rigby, 20 07). Statistical
comparisons between two linear models were based either on
the classical F-test or on the Akaike's Information Criterion [AIC
(Stasinopoulos and Rigby, 20 07)]. Continuous independent vari-ables in regression models were standardized z -scores. The statis-tical p -values in Table 1 are from two-tailed t -test of equal means.
2.6. On the interpretation of outcome variables
In the clinical samples (JoBS and PC-VDS), we were able to
explicitly compute the proportion of follow up that a participant
suffered from symptoms fulfi lling Major Depressive Episode cri-teria by using the life charts and hospital records. Hence, direct
associations between personality measured at baseline and pro-portion of time depressed could be evaluated. In the general
population (YFS), however, the participants were sampled only
in discrete time points, without knowledge of their emotional
states in between. As the temporal sampling points determined by
the study protocol can be considered unrelated to individuals'
emotional processes, the number of dysphoric states observed
during the sampling times should be monotonically related to
their general rate of occurrence. Therefore, estimated changes in
base rates due to a covariate should be reasonably comparable
across partially observed general-population trajectories and more
fully observed clinical trajectories. It should be kept in mind,
however, that quasi-Poisson models assess relative increases in
base rate of dysphoric episodes rather than absolute increases.
In addition to comparability between samples, there was
another reason for studying accumulation of discrete dysphoric
episodes instead simple sums of symptom sums over several time
points. This way one avoids confounding cases of repeatedly
elevated scores with cases of a single very high score and several
quite low scores.
3. Results
Because the distances among the levels of depressive-symptom
severity are encoded differently by mBDI and BDI-II ( Rosenström
et al., 2012b), the inclusion of a quadratic component was required
in modeling the relationship between the mBDI and the clinically
oriented BDI-II ( β
quadratic¼ 1. 4 8 , S.E. ¼ 0.07,p o 0.0 01, ΔR
2
¼ 0.085;
see Fig. 1A). Further nonlinear components were not needed in the
model (p ¼ 0.998 and ΔR
2
¼ 0.0 0 for a cubic term; other relevant
estimates were: R
2
¼ 0.695; β
linear ¼ 4.24 andβ
intercept¼ 4.02). Cor-relation between the established model estimate and measured
BDI-II was 0.83, which is close to the maximum possible [Cron-bach's (alpha) reliability of the BDI-II was 0.88]. A dysphoric
episode was defi ned by this estimated quadratic transformation
(i.e., BDI-II modeled by the mBDI) exceeding 13 points of BDI-II
score (see Measures section 2.4). Af ter the baseline-year 1997,
there were 3 non-baseline follow-ups, and hence three dysphoric
episodes were maximum number of ‘ future ’ episodes in YFS
(Fig. 1
.
As can be seen fromFig. 1A , the predictor/proxy for the BDI-II-defined mild depression had higher speci fi city (0.9
than sensi-tivity (0.65). Speci fi city implied that episodes detected by the
model almost always re flected at least mild episodes according to
BDI-II. Sensitivity implied that we missed 35% of such episodes,
suggesting that regression estimates below are underestimates
rather than overestimates. The issue of sensitivity is pertinent,
however, only so far as one prefers BDI-II over mBDI in de fining
dysphoric episodes.
Table 1 presents the basic characteristics of the studied sam-ples. Whereas the proportion of time as depressed is shown for
clinical PC-VDS and JoBS data sets, the number of dysphoric
episodes in the three non-baseline follow-ups is shown for YFS
general-population data that lacks the life chart methodology. On
average, the clinical patients had approximately 14– 17 points
higher BDI compared to the YFS participants' (BDI-II proxy).
Table 1
Basic sample characteristics and their comparison.
Variable PC-VDS YFS p-value
Median Mean s.d. Median Mean Range/s.d.
Age at the baseline 46.15 4 4.27 13.85 29 27.61 20– 35 o 0.0 01
Age at the fi nal follow up 51.15 49. 27 13.85 4 4 42.61 35– 50 –
Proportion/number of episode(s) 0.18 0.32 0.34 0 0.32 0.71 –
BDI or BDI-II proxy at baseline 17 19.54 10.51 3.79 5.63 5.30 o 0.0 01
JoBS
Age at the baseline –––37.93 38.54 11.72 0.0 01
Age at the fi nal follow up –––39.43 40.0 4 11.72 –
Proportion of derpressive episode(s) –––0.27 0.35 0.32 –
Logit of depressive-episode proportion 1.32 1. 17 1. 8 9 0.55 0.65 1.65 0.056
BDI or BDI-II proxy at baseline –––23 22.07 11.78 0.082
Note: “ PC-VDS†¼ Primary Care Vantaa Depression Study; “ YFS †¼ Young Finns (general-population) Study; “ JoBS †¼ Jorvi Bipolar Study. p -value is provided for a column-wise
t-test when such test made sense; “ s.d. †¼ standard deviation, range is given for age at YFS that had six approximately equally large cohorts.
T. Rosenström et al. / Journal of Af fective Disorders 158 (2014) 139 – 14 7 14 2
3.1. Main results for general population
Table 2 shows regression coeffi cients from quasi-Poisson mod-els predicting the number of future dysphoric episodes with the
baseline measures in general-population participants. Baseline
measures included score of depressive symptoms (mBDI), indica-tor of dysphoric episode at baseline, and personality traits; all
independent variables were standardized except dichotomous
indicator variables. For all models, low current Self-directedness
predicted the greatest increase in the rate of future dysphoric
episodes, among personality traits. Also high Harm avoidance, low
Cooperativeness, and depressive symptoms contributed strongly.
The presence of a dysphoric episode at baseline was a non-signifi cant predictor af ter adjusting for the continuous depression
Fig. 1. Outcome Variab