Journal of Affective Disorders 265 (2020) 18–25
Contents lists available at ScienceDirect
Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad
Predictability of temperaments and negative experiences on higher-order symptom-based subtypes of depression Ji Young Choia, Min Sook Gimb, Joo Young Leec,
T
⁎
a
Department of Child Studies, Inha University, Incheon, Republic of Korea Department of Psychiatry, Sanggye Baik Hospital, Inje University c Department of Child Development and Education, Dongduk Women's University, Seoul, Republic of Korea b
A R T I C LE I N FO
A B S T R A C T
Keywords: Depression Subtype Higher-order symptoms Temperament Negative experiences Latent profile analysis
Background: The identification of subtypes of depression based on higher-order symptoms of emotional, thought, and behavioral dysfunction will broaden understanding of the heterogeneity in depression. Furthermore, exploring the ability of temperaments and negative experiences to predict each subtype is an effective way of facilitating treatment decisions. Methods: Participants were 417 patients diagnosed with depressive disorder at the psychiatry department of a major medical hospital in Seoul, Korea. A latent profile analysis was performed based on three higher-order scales of the MMPI-2-RF: Emotional/Internalizing Dysfunction, Thought Dysfunction, and Behavioral/ Externalizing Dysfunction. Four temperament dimensions were assessed by the Temperament and Character Inventory-Revised-Short, and negative experiences including recent negative life events, number of lifetime traumatic events, and severity of maltreatment, were used as covariates in a multinomial regression analysis. Results: Four classes were obtained from the latent profile analysis: a “severe mood class” (39.8%), a “moderate mood class” (37.4%), a “mild mood class” (11.3%), and a “severe mood/thought class” (11.5%). Among temperament dimensions, high harm avoidance and low persistence significantly predicted more severe mood classes. Low reward dependence, number of lifetime traumatic events, and severity of maltreatment in negative experiences were significant predictors of the severe mood/thought class. Limitations: This study could not explain the more detailed heterogeneity within depression because of overinclusiveness of the higher-order scales. Conclusions: This study identified three latent classes that differed in emotional severity and one other class with thought problems. The distinct dimensions of temperament and different types of negative experiences predicted the identified subtypes.
1. Introduction Depression is one of the most prevalent psychiatric disorders worldwide and involves very heterogeneous clinical features. In DSM-5 (APA, 2013), specifiers such as “with anxious distress,” “with mixed features,” “with melancholic features,” “with atypical features,” and “with mood-congruent/mood-incongruent psychotic features” are suggested to be included in a diagnosis. However, the diagnostic criteria are not sufficient to explain the heterogeneous clinical features of depressive disorder, and there are often overlaps among the specifiers. Thus, there have been many attempts and suggestions to differentiate the subtypes of depression by means of factor analysis, symptom severity, age at onset, family history, and biological mechanisms
(Cassano et al., 2009; Savitz and Drevets, 2009; Sharpley and Bitsika, 2014, 2013). Recently, person-centered analytical approaches have detected data-driven subtypes of depression. Latent class analysis (LCA) is one of the common statistical methods that aims to recover hidden subgroups of depressed patients with similar symptom profiles. Previous LCA studies have attempted to classify subtypes with categorical indicators of symptoms, mainly based on DSM criteria for depressive disorder or depression symptom scales (Alexandrino-Silva et al., 2013; Carragher et al., 2009; De Vos et al., 2015; van Loo et al., 2012). A number of studies have analyzed subtypes of depression such as melancholia, atypical features, and anxious depression (AlexandrinoSilva et al., 2013; Bühler et al., 2014; Rodgers et al., 2014), and some
⁎ Corresponding author at: Department of Child Development and Education, Dongduk Women's University, 60, Hwarang-ro 13-gil, Seongbuk-gu, Seoul 02748, Republic of Korea. E-mail address:
[email protected] (J.Y. Lee).
https://doi.org/10.1016/j.jad.2020.01.028 Received 18 May 2019; Received in revised form 18 November 2019; Accepted 5 January 2020 Available online 09 January 2020 0165-0327/ © 2020 Elsevier B.V. All rights reserved.
Journal of Affective Disorders 265 (2020) 18–25
J.Y. Choi, et al.
Therefore, this study aims to classify depressive patients with different patterns and severity of higher-order psychopathology of emotion, thought, and behavior, based on the EID, THD, and BXD scales of the MMPI-2-RF. This study also aims to explore the variables that predict subtypes of depression based on higher-order symptoms. Depression is a complex disease of genetic, biological, and environmental factors, and efforts to identify the impact of each factor have been ongoing for a long time (Gibb et al., 2003; Harkness et al., 2012; Kessler et al., 2001; Novick et al., 2005). Temperament is one of the genetic factors that predicts not only the severity of depression but also subtypes of depression. Harm Avoidance (HA) within the temperament dimensions based on Cloninger's model was consistently reported to be associated with depressive disorder along with the Self-directedness (SD) dimension of character (de Winter et al., 2007; Wessman et al., 2012; Zaninotto et al., 2016). Recently, temperament and character have been found to be equally heritable, and since character dimensions are consistently low in mood disorders, temperament may be more useful for subtyping depressive patients in clinical settings (Zwir et al., 2018). The relationship between Reward Dependence (RD) or Persistence (P) and depression is not consistent. RD is associated with a better response to treatment, but in some instances, depressive patients score higher in the RD subscales of intimacy and dependence (Miettunen and Raevuori, 2012). High P is associated with greater health and happiness overall, but it may lead to more negative emotions unless it is accompanied by tolerance of frustration and high self-directedness (Cloninger et al., 2012). Among environmental factors, the impact of various negative experiences on depression has been frequently explored. Childhood trauma, represented by abuse and neglect, is widely accepted as a factor that increases the risk of depression, and it is associated with more severe and more complex symptoms (Gibb et al., 2003; Kessler et al., 2001). It has been also reported that patients with depressive disorder who experience more traumatic events as well as childhood trauma manifest more frequent recurrence and are less responsive to psychotherapy (Harkness et al., 2012). Many studies have explored the effects of temperament and negative experiences as predictors of depression; few studies, however, have examined the predictive power of those variables for the data-driven subtypes of depression. To summarize, this study has two objectives: (1) to identify subtypes of depression based on higher-order symptoms; (2) to confirm that temperamental dimensions and negative experiences including recent negative life event, lifetime traumatic events, and maltreatment, can predict subtypes of depression based on higher-order symptoms.
have categorized subtypes as of a cognitive, psychosomatic, and emotional class (Carragher et al., 2009; Lee et al., 2014). Latent profile analysis (LPA) using continuous variables from symptom scales typically separates classes in relation to symptom severity (De Vos et al., 2015). In addition, studies have classified the subtypes of depression in terms of both quantitative differences in symptom severity and qualitative differences in the presence of specific symptoms such as anxiety, appetite, and insomnia (Ten Have et al., 2016; Ulbricht et al., 2015). Subtyping based on DSM criteria is limited to only describing DSMdefined features, which might not fully address the wide heterogeneity of depressive patients with thought dysfunctions such as paranoid ideation or aberrant experiences, as well as those with behavioral problems such as aggression or irritability. Therefore, it is necessary to consider higher-order symptoms of thought dysfunction and behavioral problems when classifying subtypes of depression. Depression is a typical disorder of emotional and internalizing problems, but many depressive patients have problems such as paranoid ideation, hypersensitiveness, irritability, and behavioral disinhibition. It is known that up to 20% of cases of major depressive disorder are accompanied by psychotic features (Keller et al., 2007; Ohayon and Schatzberg, 2002), and many depressive patients have negative beliefs about themselves, interpersonal sensitivities, and persecutory ideations that subsequently exacerbate depression (Moritz et al., 2017). In terms of behavioral and externalizing dimensions, up to 30% of cases of major depressive disorder (MDD) show irritability among those who do not satisfy MDD with mixed features diagnosis, which is referred to as hostile-irritable depression (Benazzi and Akiskal, 2005; Parneix et al., 2014). In other words, although there clearly exists a subgroup of depressive individuals with either thought or behavioral problems, prior studies could not empirically identify this subgroup because they relied only on categorical diagnostic criteria or depressive symptoms as indicators in their analyses. A dimensional approach to psychiatric disorders provides a more empirical and comprehensive understanding of psychopathology compared to a categorical approach. Rather than focusing on the microscopic categorical symptoms of depression, focusing on more macroscopic dimensions, such as thought dysfunction and behavioral problems, may reveal previously unidentifiable but clinically meaningful subgroups. Ulbricht et al. (2018), who systematically reviewed studies using LCA to identify subtypes of depression, suggested that incorporating depressive symptoms and function along with other dimensions would be helpful in subtyping depression. Considering that many depressive patients have accompanying psychotic features and behavioral problems, subtyping based on higher-order symptom dimensions may be effective for capturing a more realistic clinical picture of depression. The Minnesota Multiphasic Personality Inventory-2 (MMPI-2; Butcher et al., 2001) is a well-known assessment inventory that comprehensively assesses a wide range of psychopathology. The newly developed MMPI-2-Restructured Form (MMPI-2-RF) resolves the psychometric weaknesses of the MMPI-2 and improves its construct, convergent, and discriminant validity. In particular, the higher-order scales of Emotional/Internalizing Dysfunction (EID), Thought Dysfunction (THD), and Behavioral/Externalizing Dysfunction (BXD) included in the MMPI-2-RF have the advantage of summarizing three major dimensions of psychopathology (Tellegen and Ben-Porath, 2011). High EID scores reflect demoralization, low positive emotions, and negative emotional experiences. High THD scores are related to thought problems such as paranoia, hallucination, and limited reality testing. High BXD scores are related to maladaptive behavioral tendencies such as substance abuse, violence, abusive behavior, and acting out. Previous studies based on diagnostic criteria or depressive symptoms mostly ended up with different subgroups sorted by overall severity and could not effectively capture subgroups with psychotic features or behavioral problems that exist at substantial rates among depressive patients.
2. Methods 2.1. Participants and procedure Participants were 417 outpatients diagnosed with depressive disorder who visited the psychiatry department of a major medical hospital in Seoul, Korea from March 2012 to December 2016. Staff psychiatrists made diagnoses according to the DSM-IV (APA, 1994) from 2012 to 2014, and the DSM-5 (APA, 2013) from 2015 to 2016. Patients with an intellectual disability or any kind of cognitive disorder were excluded. Of the 417 patients, 320 (82.3%) were diagnosed as having major depressive disorder, 49 (11.8%) as other-specified depressive disorder, 12 (2.9%) as dysthymic disorder, and 13 (3.1%) as persistent depressive disorder. In particular, 23 (5.5%) were specified as having depressive disorder with psychotic features, while other specifiers were not diagnosed. Among 417 patients, 199 (47.7%) received more than one diagnosis, including: anxiety disorders (n = 38), trauma- and stress-related disorders (n = 33), substance use disorder (n = 33), somatic symptom related disorder (n = 22), personality disorder (n = 50), eating disorder (n = 6), and other disorders (n = 17) 19
Journal of Affective Disorders 265 (2020) 18–25
J.Y. Choi, et al.
experience, 1 = once, 2 = two or three times, 3 = more than 3 times). Negative events including severe illness or injury, financial loss, family conflict or discord, difficulties in work or school, daily mistakes, and being a victim of bullying, are listed in the scale. A higher score means more negative life events were experienced, and the range of values is from 0 to 141. Internal consistency in the present study was .92. The NLES score was used to explore environmental factors affecting the subtypes of depression.
including attention-deficit and hyperactivity disorder, obsessive-compulsive disorder, and disruptive/impulse-control/conduct disorders. The mean age of the sample was 38.56 years (SD = 16.33) with a range of 18–80 years, including 196 males (47.0%) and 221 females (53.0%). Individuals who provided informed consent to participate in the study completed the self-report measures as part of their psychological evaluation. Participants could decline to participate at any time without penalty, and there was no compensation for participation. On average, participants completed the psychological assessment within two weeks of their initial consultation. Official diagnoses were made after a followup consultation with staff psychiatrists within one week after their psychological assessment. This study was approved by the hospital's Institutional Review Board.
2.2.4. Life events checklist The Life Events Checklist is a 17-item self-report questionnaire which was originally included in the Clinician Administered PTSD Scale, a tool developed for diagnosing PTSD (Blake et al., 1995) and for screening potential traumatic events. Sixteen events that are generally associated with PTSD and one event that was not included in the 16 life events comprised items in the checklist. Each item inquires as to whether an individual has personally experienced the event, witnessed the event, heard about the event, or has become aware of the event (in addition to “do not know” and “not applicable”). A clinical psychologist who is bilingual in English and Korean translated the original checklist into Korean. Cronbach's alpha in the present study was .74. For this study, the number of experienced trauma events was calculated as lifetime traumatic events; this score was used to explore the environmental factors that affect the subtypes of depression.
2.2. Measures 2.2.1. Higher-order (H-O) scales of the minnesota multiphasic personality inventory-2 restructured Form (MMPI-2-RF), Korean version The MMPI-2-RF (Tellegen and Ben-Porath, 2011) was developed to improve the overall psychometric properties of the MMPI-2 that is widely used to assess symptoms and diagnostic possibilities in clinical populations. The MMPI-2-RF Korean version has been published with acceptable reliability and validity (Han et al., 2011). The MMPI-2-RF is composed of 338 yes/no items; the items comprise 9 validity scales, 3 higher-order (H-O) scales, 9 restructured clinical (RC) scales, 23 specific problem (SP) scales, 2 interest scales, and revised Personality Psychopathology Five (PSY-5) scales. The 3 higher-order (H-O) scales—Emotional/Internalizing Dysfunction (EID), Thought Dysfunction (THD), and Behavioral/Externalizing Dysfunction (BXD)—were used as indicators of the latent profile analysis. They were derived from exploratory factor analyses of the MMPI-2, and an equivalent three-factor structure of higher-order scales was found in Korean (Han et al., 2011) and US normative samples (Tellegen and Ben-Porath, 2011). The EID scale consists of 41 items that assess overall levels of emotional distress such as demoralization, low positive emotion, and negative emotional experiences. The THD scale contains 26 items that assess an extensive range of psychological difficulties related to thought dysfunction such as paranoia, hallucinations, or limited reality testing. The BXD scale consists of 23 items that measure overall levels of maladaptive behavioral tendencies such as acting out, conduct problems, or aggressive behavior. The validation study in Korea indicated an adequate internal consistency of .90 for the EID, .87 for the THD, and .76 for the BXD (Han et al., 2011).
2.2.5. Korean childhood trauma questionnaire The Korean version of the Childhood Trauma Questionnaire was developed by Berstein and Fink (1998) and translated by Yu et al. (2009). This questionnaire comprises five subscales: emotional abuse, physical abuse, sexual abuse, emotional neglect, and physical neglect as well as three items pertaining to the validity scale (minimization/denial scale). Each item is rated on a 5-point Likert scale, with higher scores representing a more severe degree of maltreatment. Internal consistency of the Korean version was 0.79, and Cronbach's alpha across the five subscales was 0.80 (emotional abuse), 0.82 (physical abuse), 0.79 (sexual abuse), 0.89 (emotional neglect), and 0.51 (physical neglect) (Yu et al., 2009). Cronbach's alpha values for the present study were 0.89 for physical abuse, 0.72 for physical neglect, 0.85 for emotional abuse, 0.88 for emotional neglect, and 0.80 for sexual abuse. The total score was used in the present study to indicate severity of maltreatment; this score was also used to explore environmental factors affecting the subtypes of depression. 2.2.6. Beck depression inventory Originally developed by Beck et al., (1961) to assess the degree of depression, the scale was standardized in Korean by Lee and Song (1991). The scale consists of 21 items measured on a 3-point Likert scale. Internal consistency of the Korean version was 0.78, and its test-retest reliability was 0.75 (Lee and Song, 1991). Internal consistency in the present study was 0.92. The BDI score was used as a control variable in the latent profile analysis.
2.2.2. Korean version of the temperament and character inventory-revisedshort (TCI-RS) The TCI-RS is a 140-item self-report questionnaire that assesses four dimensions of temperament and three dimensions of character, with each item rated on a 4-point Likert scale. The temperament dimensions of Novelty Seeking (NS), Harm Avoidance (HA), Reward Dependence (RD), and Persistence (P) are assumed to be independently heritable and to manifest early in development. The character dimensions, including Self-directedness (SD), Cooperativeness (CO), and Self-transcendence (ST), are assumed to be equally heritable as temperaments, but are likely to mature with age. The Korean version of the TCI-RS was standardized and validated in 2007 and demonstrated adequate validity and reliability. Cronbach's alpha values for the NS, HA, RD, P, SC, CO, and ST scales were .83, .86, .81, .82, .87, .76, and .90 respectively (Min et al., 2007). In this study, we used the NS, HA, RD, and P scales to explore heritable factors that affect the subtypes of depression.
2.3. Data analyses To investigate the predictability of higher-order symptom-based subtypes of depression from temperament dimensions and negative experiences, we first conducted a latent profile analysis with three higher-order scales (EID, THD, BXD) of the MMPI-2-RF as indicators. After identifying the optimal classes and assigning participants to their most likely classes according to posterior probabilities, we compared the demographic and clinical characteristics of those classes using chisquared tests and one-way ANOVAs. Lastly, multinomial regression analyses were executed to examine the predictability of the identified classes from the temperament dimensions and negative experiences variables. Latent profile analysis (LPA) was performed using M-plus version 7
2.2.3. Negative life events survey (NLES) The NLES is a 47-item self-report questionnaire to assess the frequency of an individual's experiences of negative life events within the past 3 months. Lee (1993) modified the Life Experiences Survey (Sarason et al., 1978) and each item is rated on a 4-point scale (0 = no 20
Journal of Affective Disorders 265 (2020) 18–25
J.Y. Choi, et al.
improvement over the four-class model (p > .05). The entropy measures were quite similar for all the models tested. We chose the fourclass model as the best-fitting class solution based on the conceptual and interpretative meaning of the classes although the three-class model would be the best-fitting model according to LRT value guidelines. See Fig. 1 for a graphical depiction of the final four-class solution. Class 1 (n = 166, 39.8%) was labeled “severe mood class” as it had relatively high severity score on EID and low severity score on THD and BXD. Class 2 (n = 156, 37.4%) was labeled “moderate mood class” due to a moderate level of EID. Class 3 (n = 47, 11.3%) was labeled “mild mood class” as it had the lowest scores in all the scales. Lastly, Class 4 (n = 48, 11.5%) was labeled “severe mood/thought class” as it had the highest EID and THD scores. Class 4 also had the highest score on the BXD scale, which suggests severe emotional problems accompanied by both thought dysfunction and behavioral problems. The mean scores of indicators for each class are presented in Table 2.
(Muthén and Muthén, 2015) based on three indicators—the EID, THD, and BXD scales of the MMPI-2-RF—and the BDI score was included as a control variable in LPA to reduce the impact of overall symptom severity on classification. Maximum likelihood estimation was used with robust standard errors (ML). Five latent class models were fitted to the data, from a one-class model to a five-class model. The optimal class solution was decided by the following fit indices: (1) lower Akaike Information Criterion (AIC) values, Bayesian Information Criterion (BIC) values, and sample-size adjusted BIC values (SSABIC), (2) a significant Lo-Mendell-Rubin Likelihood (LRT) test value and Bootstrapped Likelihood Ratio Test (BLRT) p-value in which the k class model is a better fitting model than the k-1 class model, (3) relatively higher entropy values, and (4) conceptual meaning (Nylund et al., 2007). Consistent with previous LPA research (e.g., Versella et al., 2016), model selection also was based on the size of the smallest derived class, as classes constituting less than approximately 5% of the sample may over-fit the data and, thus, be more likely to fail to replicate in independent datasets. After determining the best class solution, we first compared the demographic variables, depression severity, temperament dimensions, and negative experience variables among the latent classes by chisquare tests and one-way ANOVAs. We then tested the effects of predictors on latent class membership using multinomial regression analysis with SPSS 24.0. For this part of the analysis, the default approach for dealing with missing data is list-wise deletion in SPSS, thus reducing our sample to 342 participants. Multicollinearity among potential predictor variables was assessed using the variance inflation factor statistic (VIF) from the equivalent linear regression model, and this assessment did not indicate serious multicollinearity (VIF > 10).
3.3. Demographic and clinical characteristics of the 4 high-order symptombased latent classes Demographic characteristics, clinical features, temperament, and negative life experiences of the four classes are summarized in Table 3. The four groups significantly differed in age and gender. Patients in the severe mood class and the severe mood/thought class were significantly younger than in the other two classes, and there were more males in the severe mood/thought class. The severity of depressive mood measured by BDI was highest in the severe mood/thought class. The severe mood/ thought class showed the highest frequency of psychotic features and the highest rate of comorbidity. For co-occurring disorders, PTSD was most common in the severe mood/thought class. In terms of the temperament and negative life experiences, it was revealed that classes with more severe symptoms tended to have higher harm avoidance, lower reward dependence and persistence, and more negative life experiences including recent negative life events, lifetime traumatic events, and maltreatment.
3. Results 3.1. Descriptive statistics of study variables The mean T scores for the three indicator variables were 72.64 (SD = 14.37), 54.93 (SD = 13.72), and 50.40 (SD = 12.32) for EID, THD, and BXD, respectively. Among the three higher-order scales, only the mean EID T score was above 65T, indicating a high severity of clinical level. The mean score of BDI was 30.23 (SD = 12.22), a moderate level. The mean scores of the predictor variables were 51.20 (SD = 14.22), 66.09 (SD = 13.71), 46.58 (SD = 11.57), 38.73 (SD = 13.40), 25.87 (SD = 19.87), 1.75 (SD = 1.95), and 59.10 (SD = 15.05) for NS, HA, RD, P, negative life events, lifetime traumatic events, and maltreatment respectively.
3.4. Predictors of higher-order symptom-based subtypes of depression To evaluate whether temperament dimensions and negative life experiences could predict an individual's membership in one of the classes identified in LPA, multinomial logistic regression analyses were conducted, comparing the mild mood class with the moderate mood class, and the moderate mood class and severe mood/thought class with the severe mood class. Table 4 shows odds ratios and 95% CIs of the multinomial logistic regression. Results indicated that harm avoidance (β = -0.07, p = 0.002, OR = 0.94) in the temperament dimensions, and number of lifetime traumatic events (β = -0.65, p = 0.006, OR = 0.52) and severity of maltreatment (β = -0.07, p = 0.000, OR = 0.93) in negative life experiences, were significant in predicting the mild mood class versus the moderate mood class. In predicting the moderate mood class versus the severe mood class, harm avoidance (β = -0.07, p = 0.000, OR = 0.93) and persistence (β = 0.04, p = 0.004, OR = 1.04) were significant. Comparing the severe mood/ thought class with the severe mood class, reward dependence (β = -
3.2. Determination of the number of latent classes LPA was run with three higher-order scales of EID, THD, and BXD from the MMPI-2-RF as indicators. Fit indices of the competing latent class models are reported in Table 1. AIC, BIC, and SSABIC were markedly lower for the four-class model compared to the earlier models. Thereafter, there were only small decreases in these indices, providing weak support for the five-class solution. The LMR-LRT results further confirmed that the five-class solution did not show significant Table 1 Fit information for latent profile analysis models with a class of 2-5. Model
Log-likelihood values
AIC
BIC
SSABIC
LMRa-LRT p value
Entropy
BLRT
Smallest class proportion
2 3 4 5
−4889.891 −4805.469 −4783.154 −4768.654
9801.781 9642.937 9608.308 9589.308
9846.145 9707.467 9693.003 9694.168
9811.239 9656.694 9626.364 9611.663
259.948 (p = .0700) 163.426 (p = .0000) 43.190 (p = .0013) 28.070 (p = .4232)
0.846 0.792 0.780 0.784
<0.001 <0.0001 <0.001 <0.001
17.5% 11.8% 11.7% 5.0%
Note. AIC = Akaike information criterion; BIC = Bayesian information criterion; SSABIC = sample-size adjusted BIC; LMRa-LRT = Lo-Mendell-Rubin adjusted likelihood ratio test; BLRT = Bootstrapped likelihood ratio test. *p < .05. ⁎⁎ p < .01. ⁎⁎⁎ p < .001. 21
Journal of Affective Disorders 265 (2020) 18–25
J.Y. Choi, et al.
Fig. 1. The mean T-score of the three higher-order MMPI-2-RF scales in the 4-class model. Table 2 Mean (standard deviation) T-scores of the higher-order scales in the MMPI-2-RF for the higher-order symptoms-based 4 latent classes.
EID THD BXD
Class 1. Severe mood (n = 166)
Class 2. Moderate mood (n = 156)
Class 3. Mild mood (n = 47)
Class 4. Severe mood/thought (n = 48)
83.31 (6.31) 52.69 (8.42) 51.36 (11.19)
64.65 (6.88) 51.45 (9.33) 48.10 (10.13)
47.23 (5.58) 45.32 (6.13) 41.23 (7.28)
86.58 (6.46) 83.42 (10.65) 63.52 (15.27)
Note. MMPI-2-RF = Minnesota Multiphasic Personality Inventory-2 Restructured Form; EID = Emotional/Internalizing Dysfunction; THD = Thought Dysfunction; BXD = Behavior/Externalizing Dysfunction.
Table 3 Demographic and clinical characteristics of the higher-order symptoms-based 4 latent classes. M (SD)/n (%)
Class 1. Severe mood (n = 166)
Class 2. Moderate mood (n = 156)
Class 3. Mild mood (n = 47)
Class 4. Severe mood/thought (n = 48)
χ2/ F (two-tailed)
Age (years) Gender (women) BDI Psychotic features Comorbidity Any anxiety disorder PTSD Substance related disorder Personality disorder Novelty seeking Harm avoidance Reward dependence Persistence Negative life events Lifetime traumatic events Maltreatment
34.17 (14.79) 82 (49.4%) 35.84 (9.05) 2 (1.2%) 84 (50.6%) 15 (9.0%) 12 (7.2%) 15 (9.0%) 23 (13.9%) 52.84 (13.89) 74.45 (11.54) 39.88 (14.03) 34.57 (11.92) 27.61 (17.28) 1.48 (1.57) 60.22 (14.74)
42.96 (16.42) 92 (59.0%) 25.40 (9.30) 6 (3.8%) 62 (39.7%) 16 (10.3%) 10 (6.4%) 10 (6.4%) 18 (11.5%) 50.00 (12.55) 62.06 (8.84) 43.92 (11.29) 42.47 (11.43) 22.24 (16.81) 1.66 (1.69) 55.89 (11.73)
47.19 (16.07) 30 (63.8%) 13.79 (7.54) 6 (12.8%) 17 (36.2%) 2 (4.3%) 1 (2.1%) 2 (4.3%) 3 (6.4%) 46.04 (10.34) 53.36 (8.32) 46.58 (11.57) 47.67 (11.72) 12.95 (16.75) 1.17 (2.10) 48.47 (10.96)
31.00 (13.57) 17(35.4%) 42.65 (8.92) 9 (18.8%) 33 (68.8%) 5 (10.4%) 10 (20.8%) 7 (14.6%) 6 (12.5%) 54.34 (20.91) 72.54 (22.53) 32.54 (17.72) 32.65 (17.33) 42.90 (22.87) 3.56 (2.71) 74.43 (16.63)
p p p p p p p p p p p p p p p p
Note. BDI = Beck Depression Inventory; PTSD = posttraumatic stress disorder. a Class 1 versus class 2 was significantly different, by post-hoc test with Scheffe correction. b Class 1 versus class 3 was significantly different, by post-hoc test with Scheffe correction. c Class 1 versus class 4 was significantly different, by post-hoc test with Scheffe correction. d Class 2 versus class 3 was significantly different, by post-hoc test with Scheffe correction. e Class 2 versus class 4 was significantly different, by post-hoc test with Scheffe correction. f Class 3 versus class 4 was significantly different, by post-hoc test with Scheffe correction. 22
< < < < < < < = = < < < < < < <
.001a,b,e,f .05 .001a,b,c,d,e,f .001 .01 .01 .01 .221 .575 .01b,f .001a,b,d,e,f .001a,b,c,e,f .001a,b,e,f .001b,c,d,e,f .001b,c,d,e,f .001a,b,c,d,e,f
Journal of Affective Disorders 265 (2020) 18–25
J.Y. Choi, et al.
Table 4 Odds ratios and confidence intervals (95%) of multinomial logistic regression analyses for 4 latent classes. Variable
Mild vs. Moderate mood#
Moderate vs. Severe mood#
Severe mood/thought vs. Severe mood#
Age (years) Gender (women) Novelty seeking Harm avoidance Reward dependence Persistence Negative life events Lifetime traumatic events Maltreatment
1.01 1.42 0.96 0.94 1.03 1.04 0.97 0.52 0.93
1.02 1.09 0.99 0.93 1.02 1.04 0.98 1.10 0.98
0.99 1.68 0.98 1.00 0.94 1.02 1.02 1.44 1.04
(0.98–1.04) (0.49–4.12) (0.91–1.00) (0.90–0.98) (0.98–1.09) (1.00–1.09) (0.94–1.01) (0.33–0.83) (0.88–0.98)
⁎⁎
⁎⁎ ⁎⁎
(1.00–1.04) (0.59–2.02) (0.96–1.01) (0.91–0.96) (0.99–1.05) (1.01–1.07) (0.96–1.02) (0.91–1.33) (0.96–1.01)
⁎⁎⁎
⁎⁎
(0.96–1.03) (0.59–4.76) (0.94–1.02) (0.96–1.04) (0.90–0.98) (0.98–1.07) (1.00–1.05) (1.09–1.88) (1.01–1.07)
⁎⁎
⁎⁎ ⁎⁎
#Indicates the reference class. *p < .05. ** p < .01. *** p < .001.
reality testing and exhibit uncontrolled behavior. Supporting this assumption, significantly more patients with psychotic features belonged to this class. There were significant differences in gender and age among classes. Males and younger individuals tended to belong to the more severe classes. Although not significant in logistic regression, there were significantly more young male patients in the severe mood/thought class. Considering that subtypes may vary by gender, it would be more appropriate to analyze data in detail by the demographic characteristics of the sample (Parker et al., 2014). Depression is a complex disorder affected by both temperament and environmental risk factors, and it is difficult to quantify their relative contribution to the disease. Yet, the fact that harm avoidance consistently affected severity is in accordance with previous studies (Kampman and Poutanen, 2011; Zaninotto et al., 2016). Low persistence predicted membership in the more severe class, and this is in line with Cloninger et al. (2012), who found that high persistence reflected a disposition to be generally healthier and happier. However, high persistence may cause negative emotions in combination with high harm avoidance and low self-directedness. Further research on the role of persistence is needed. It is notable that low reward dependence significantly differentiated the severe mood/thought class from the severe mood class. Although the relationship between reward dependence and psychopathology has not yet been clearly defined, recently, high reward dependence was reported to be related to satisfactory therapeutic improvements in depression (Paavonen et al., 2018). In a meta-analysis of the relationship between axis I psychiatric disorders and temperament, low reward dependence was associated with schizophrenia (Miettunen and Raevuori, 2012), and this was related to low sociability as commonly observed in individuals with non-affective psychotic disorder. In Hori et al.’s (2017) study, the socially detached group characterized by low reward dependence showed high depression scores and high rates of antipsychotic medication. We can infer from these findings that low reward dependence may be related to certain types of thought problems because it represents low sociability and deficiency of interpersonal relations. Recent stressful events had no significant effect on the higher-order symptom-based subgroups. Lifetime traumatic events and maltreatment discriminated the mild mood and moderate mood class, but they were not effective in discriminating the moderate mood and severe mood class. Instead, lifetime traumatic events and severity of childhood maltreatment significantly differentiated the severe mood/thought class from the severe mood class. When individuals experience complex traumatic events or childhood trauma, they tend to show more complex symptoms (Choi et al., 2014; Cloitre et al., 2009). Therefore, the severe mood/thought class may be a distinct subgroup of depressive disorder with more complex symptom manifestation beyond the inherent characteristics of depression. Depressive patients who belong to this class may suffer from emotional dysregulation, interpersonal difficulties, and
0.06, p = 0.003, OR = 0.94) in the temperament dimensions, number of lifetime traumatic events (β = 0.36, p = 0.009, OR = 1.44) and severity of maltreatment (β = 0.04, p = 0.009, OR = 1.04) in negative life experiences, were proven to be significant predictors.
4. Discussion The aim of this study was to identify data-driven subtypes of depression based on broadband symptoms of mood, thought, and behavior rather than diagnostic criteria or depressive symptoms among Korean patients, and to explore predictors of the identified subtypes. The result of a latent profile analysis on the higher-order scales of the MMPI-2-RF showed four distinctive subgroups. The first class was the most predominant subgroup with severe mood problems, followed by two subgroups of moderate and mild mood problems. The last class was a subgroup with severe mood/thought problems along with a relatively high level of behavioral problems. Since this is the first empirical study to incorporate higher-order symptom indicators, it is not possible to directly compare the results with those of studies that classified melancholia and atypical subtypes, or with those that categorized cognitive, psychosomatic, and emotional subtypes. Yet, many person-centered studies which used diagnostic assessment tools usually ended up with subtypes of different levels of severity (De Vos et al., 2015; Rodgers et al., 2014), and this study also indicated three subgroups of quantitatively different symptom severity. However, we found a qualitatively different subgroup with thought and behavioral problems, and this is in line with several studies that found heterogeneous subtypes with anxiety symptoms (Ten Have et al., 2016; Ulbricht et al., 2015). In this study, three subgroups were classified by the severity of emotional problems—mild, moderate, and severe—and there was one other class containing 11.5% of patients who showed very severe thought dysfunction and moderate behavioral problems. This class tended to be accompanied by broad thought dysfunctions, and it was more likely to show disinhibited behaviors such as substance use or aggression. Previous studies were unable to capture this subgroup because they did not include appropriate indicators that could assess those characteristics. However, a substantial number of depressive patients do have a wide range of psychotic symptoms such as paranoid ideation or aberrant experience even when they are not actually diagnosed as psychotic depression. Moreover, many depressive patients manifest hostility, irritability, or impulsive behavior. Given these facts, the subtyping of depression incorporating thought and behavioral problems is very helpful in revealing disparate aspects of depression. Hori et al. (2017) classified an adaptive group (30.3%), a neurotic group similar to melancholia (46.2%), and a socially detached group (23.5%), based on temperament and character dimensions. They observed that the socially-detached group was more likely to use antipsychotic drugs, be low-functioning, and to lack reality testing. The severe mood/thought class of this study is also more likely to lack 23
Journal of Affective Disorders 265 (2020) 18–25
J.Y. Choi, et al.
Validation, Writing - review & editing.
behavioral disinhibition due to the prolonged effects of lifetime traumatic experiences and childhood maltreatment. To summarize the predictors of the subgroups, while harm avoidance was a critical temperament that discriminated the mild, moderate, and severe mood classes, low reward dependence was a significant predictor of the severe mood/thought class. Among the environmental factors, cumulative traumatic experiences including childhood maltreatment significantly predicted the severe mood/thought class. These results can provide important treatment guidelines. Helping depressive patients to understand and regulate their temperamental characteristics, and dealing with their past traumatic experiences, should be considered essential in treatment. Specifically, for those who exhibit both thought and behavioral problems, exploring and processing past traumatic events after stabilization of depressive symptoms could be an important therapeutic target. This study has a number of limitations. First, the higher-order scales of EID, THD, and BXD include very broad and heterogeneous symptoms. For example, EID assesses comprehensive and somewhat overinclusive internalizing problems such as depression, anxiety, and low positive emotions. Therefore, this study could not qualitatively classify more precise subtypes of depression. Future studies will be able to examine this heterogeneity in more detail if they use more specific measures of the emotional, thought, and behavioral dimensions. Furthermore, there is a need for replication using latent profile analysis to test the robustness of our findings. Second, this study did not include longitudinal data for recurrence of depression, duration of depressive episodes, and types of medication. Moreover, structured diagnostic assessments were not available. Third, some data were missing for 75 subjects, who were excluded from the multinomial regression analysis. This group of excluded subjects tended to be older and included more females; this could affect the generalizability of the results. In addition, this study used a sample of treatment-seeking patients; therefore, the same subtypes might not be applicable to other populations, such as community samples. Despite such limitations, this study empirically subtyped depressive patients using the macro dimensions of emotion, thought, and behavior, and it examined the predictability of both heritable and environmental factors. In conclusion, we found three distinctive classes of different levels of severity, as well as one other class with severe thought problems and moderate behavioral problems that do exist clinically but have not been identified in previous person-centered studies based on depressive symptoms. It is noteworthy that this study newly identified the higher-order symptom-based subgroups of depressive disorder and revealed the significant predictors of heritable temperaments and negative environmental experiences for those subtypes.
Declarations of Competing Interest None. Acknowledgement We thank Maumsarang Co., Ltd. for converting MMPI-2 to MMPI-2RF. References Alexandrino-Silva, C., Wang, Y.P., Carmen Viana, M., Bulhões, R.S., Martins, S.S., Andrade, L.H., 2013. Gender differences in symptomatic profiles of depression: results from the São Paulo Megacity Mental Health Survey. J. Affect. Disord. 147, 355–364. https://doi.org/10.1016/j.jad.2012.11.041. American Psychiatric Association, 1994. Diagnostic and Statistical Manual of Mental Disorders, fourth ed. American Psychiatric Association Press, Washington, DC. American Psychiatric Association, 2013. Diagnostic and Statistical Manual of Mental Disorders, fifth ed. American Psychiatric Association Press, Washington, DC. Beck, A.T., Ward, C.H., Mendelson, M., Mock, J., Erbaugh, J., 1961. An inventory for measuring depression. Arch. Gen. Psychiatry 4, 561–571. Benazzi, F., Akiskal, H., 2005. Irritable-hostile depression: further validation as a bipolar depressive mixed state. J. Affect. Disord. 84, 197–207. https://doi.org/10.1016/j.jad. 2004.07.006. Berstein, D., Fink, L., 1998. Childhood trauma questionnaire. J. Am. Acad. Child Adolesc. Psychiatry 36, 340–348. https://doi.org/10.1097/00004583-199703000-00012. Blake, D.D., Weathers, F.W., Nagy, L.M., Kaloupek, D.G., Gusman, F.D., Charney, D.S., Keane, T.M., 1995. The development of a clinician-administered PTSD scale. J. Trauma. Stress. 8, 75–90. https://doi.org/10.1007/BF02105408. Bühler, J., Seemüller, F., Läge, D., 2014. The predictive power of subgroups: an empirical approach to identify depressive symptom patterns that predict response to treatment. J. Affect. Disord. 163, 81–87. https://doi.org/10.1016/j.jad.2014.03.053. Butcher, J.N., Graham, J.R., Ben-Porath, Y.S., Tellegen, A., Dahlstrom, W.G., Kaemmer, B., 2001. MMPI-2 (Minnesota Multiphasic Personality Inventory-2): Manual for Administration, Scoring, and Interpretation, Revised Edition. University of Minnesota Press, Minneapolis, MN. Carragher, N., Adamson, G., Bunting, B., McCann, S., 2009. Subtypes of depression in a nationally representative sample. J. Affect. Disord. 113, 88–99. https://doi.org/10. 1016/j.jad.2008.05.015. Cassano, G.B., Benvenuti, A., Miniati, M., Calugi, S., Mula, M., Maggi, L., Rucci, P., Fagiolini, A., Perris, F., Frank, E., 2009. The factor structure of lifetime depressive spectrum in patients with unipolar depression. J. Affect. Disord. 115, 87–990. https://doi.org/10.1016/j.jad.2008.09.006. Choi, J.Y., Choi, Y.M., Gim, M.S., Park, J.H., Park, S.H., 2014. The effects of childhood abuse on symptom complexity in a clinical sample: mediating effects of emotion regulation difficulties. Child Abus. Negl. 38, 1313–1319. https://doi.org/10.1016/j. chiabu.2014.04.016. Cloitre, M., Stolbach, B.C., Herman, J.L., Van Der Kolk, B., Pynoos, R., Wang, J., Petkova, E., 2009. A developmental approach to complex PTSD: childhood and adult cumulative trauma as predictors of symptom complexity. J. Trauma. Stress. 22, 399–408. https://doi.org/10.1002/jts.20444. Cloninger, C.R., Zohar, A.H., Hirschmann, S., Dahan, D., 2012. The psychological costs and benefits of being highly persistent: personality profiles distinguish mood disorders from anxiety disorders. J. Affect. Disord. 136, 758–766. https://doi.org/10. 1016/j.jad.2011.09.046. De Vos, S., Wardenaar, K.J., Bos, E.H., Wit, E.C., De Jonge, P., 2015. Decomposing the heterogeneity of depression at the person-, symptom-, and time-level: latent variable models versus multimode principal component analysis. BMC Med. Res. Methodol. https://doi.org/10.1186/s12874-015-0080-4. de Winter, R.F.P., Wolterbeek, R., Spinhoven, P., Zitman, F.G., Goekoop, J.G., 2007. Character and temperament in major depressive disorder and a highly anxious-retarded subtype derived from melancholia. Compr. Psychiatry 48, 426–435. https:// doi.org/10.1016/j.comppsych.2007.04.002. Gibb, B.E., Wheeler, R., Alloy, L.B., Abramson, L.Y., 2003. Emotional, physical, and sexual maltreatment in childhood versus adolescence and personality dysfunction in young adulthood. J. Pers. Disord. 15, 505–511. https://doi.org/10.1521/pedi.15.6. 505.19194. Han, K.H., Moon, K.J., Lee, J.Y., Kim, J.H., 2011. The Korean Version of Minnesota Multiphasic Personality Inventory-2-RF Manual. Maumsarang, Seoul, Korea. Harkness, K.L., Michael Bagby, R., Kennedy, S.H., 2012. Childhood maltreatment and differential treatment response and recurrence in adult major depressive disorder. J. Consult. Clin. Psychol. 80, 342–353. https://doi.org/10.1037/a0027665. Hori, H., Teraishi, T., Nagashima, A., Koga, N., Ota, M., Hattori, K., Kim, Y., Higuchi, T., Kunugi, H., 2017. A personality-based latent class typology of outpatients with major depressive disorder: association with symptomatology, prescription pattern and social function. J. Affect. Disord. 217, 8–15. https://doi.org/10.1016/j.jad.2017.03. 053. Kampman, O., Poutanen, O., 2011. Can onset and recovery in depression be predicted by temperament? A systematic review and meta-analysis. J. Affect. Disord. 135, 20–27. https://doi.org/10.1016/j.jad.2010.12.021.
Authorship Statement All persons who meet authorship criteria are listed as authors, and all authors certify that they have participated sufficiently in the work to take public responsibility for the content, including participation in the concept, design, analysis, writing, or revision of the manuscript. Furthermore, each author certifies that this material or similar material has not been and will not submitted to or published in any other publication before its appearance in the Journal of Affective Disorders. Funding source The study was supported by INHA UNIVERSITY Research Grant. CRediT authorship contribution statement Ji Young Choi: Conceptualization, Data curation, Funding acquisition, Methodology, Writing - original draft, Writing - review & editing. Min Sook Gim: Conceptualization, Data curation, Writing - review & editing. Joo Young Lee: Conceptualization, Formal analysis, 24
Journal of Affective Disorders 265 (2020) 18–25
J.Y. Choi, et al.
Savitz, J.B., Drevets, W.C., 2009. Imaging phenotypes of major depressive disorder: genetic correlates. Neuroscience 164, 300–330. https://doi.org/10.1016/j. neuroscience.2009.03.082. Sharpley, C.F., Bitsika, V., 2013. Differences in neurobiological pathways of four “clinical content” subtypes of depression. Behav. Brain Res. 256, 368–376. https://doi.org/10. 1016/j.bbr.2013.08.030. Sharpley, C.F., Bitsika, V., 2014. Validity, reliability and prevalence of four “clinical content” subtypes of depression. Behav. Brain Res. 259, 9–15. https://doi.org/10. 1016/j.bbr.2013.10.032. Tellegen, A., Ben-Porath, Y.S., 2011. MMPI-2-RF: Minnesota Multiphasic Personality Inventory-2 Restructured Form: Technical Manual. University of Minnesota Press, Minneapolis, MN. Ten Have, M., Lamers, F., Wardenaar, K., Beekman, A., De Jonge, P., Van Dorsselaer, S., Tuithof, M., Kleinjan, M., De Graaf, R., 2016. The identification of symptom-based subtypes of depression: a nationally representative cohort study. J. Affect. Disord. 190, 395–406. https://doi.org/10.1016/j.jad.2015.10.040. Ulbricht, C.M., Chrysanthopoulou, S.A., Levin, L., Lapane, K.L., 2018. The use of latent class analysis for identifying subtypes of depression: a systematic review. Psychiatry Res. 266, 228–246. https://doi.org/10.1016/j.psychres.2018.03.003. Ulbricht, C.M., Rothschild, A.J., Lapane, K.L., 2015. The association between latent depression subtypes and remission after treatment with citalopram: a latent class analysis with distal outcome. J. Affect. Disord. 188, 270–277. https://doi.org/10.1016/j. jad.2015.08.039. van Loo, H.M., de Jonge, P., Romeijn, J.W., Kessler, R.C., Schoevers, R.A., 2012. Datadriven subtypes of major depressive disorder: a systematic review. BMC Med. https:// doi.org/10.1186/1741-7015-10-156. Versella, M.V., Piccirillo, M.L., Potter, C.M., Olino, T.M., Heimberg, R.G., 2016. Anger profiles in social anxiety disorder. J. Anxiety Disord. 37, 21–29. https://doi.org/10. 1016/j.janxdis.2015.10.008. Wessman, J., Schönauer, S., Miettunen, J., Turunen, H., Parviainen, P., Seppänen, J.K., Congdon, E., Service, S., Koiranen, M., Ekelund, J., Laitinen, J., Taanila, A., Tammelin, T., Hintsanen, M., Pulkki-Råback, L., Keltikangas-Järvinen, L., Viikari, J., Raitakari, O.T., Joukamaa, M., Järvelin, M.R., Freimer, N., Peltonen, L., Veijola, J., Mannila, H., Paunio, T., 2012. Temperament clusters in a normal population: implications for health and disease. PLoS ONE. https://doi.org/10.1371/journal.pone. 0033088. Yu, J.H., Park, J.S., Park, D.H., Ryu, S.H., Ha, J.H., 2009. Validation of the Korean Childhood Trauma Questionnaire: the practical use in counselling and therapeutic intervention. Korean J. Heal. Psychol. 14, 563–578. Zaninotto, L., Solmi, M., Toffanin, T., Veronese, N., Cloninger, C.R., Correll, C.U., 2016. A meta-analysis of temperament and character dimensions in patients with mood disorders: comparison to healthy controls and unaffected siblings. J. Affect. Disord. 194, 84–97. https://doi.org/10.1016/j.jad.2015.12.077. Zwir, I., Arnedo, J., Delval, C., Pulkki-Råback, L., Konte, B., Yang, S.S., Romero-Zaliz, R., Hintsanen, M., Cloninger, K.M., Garcia, D., Svrakic, D.M., Rozsa, S., Martinez, M., Lyytikäinen, L.-P., Giegling, I., Kähönen, M., Hernandez-Cuervo, H., Seppälä, I., Raitoharju, E., de Erausquin, G.A., Raitakari, O., Rujescu, D., Postolache, T.T., Sung, J., Keltikangas-Järvinen, L., Lehtimäki, T., Cloninger, C.R., 2018. Uncovering the complex genetics of human character. Mol. Psychiatry. https://doi.org/10.1038/ s41380-018-0263-6.
Keller, J., Schatzberg, A.F., Maj, M., 2007. Current issues in the classification of psychotic major depression. Schizophr. Bull. 33, 877–885. https://doi.org/10.1093/schbul/ sbm065. Kessler, R.C., Avenevoli, S., Ries Merikangas, K., 2001. Mood disorders in children and adolescents: an epidemiologic perspective. Biol. Psychiatry. 49, 1002–1014. https:// doi.org/10.1016/S0006-3223(01)01129-5. Lee, S.Y., Xue, Q.L., Spira, A.P., Lee, H.B., 2014. Racial and ethnic differences in depressive subtypes and access to mental health care in the United States. J. Affect. Disord. 155, 130–137. https://doi.org/10.1016/j.jad.2013.10.037. Lee, Y.H., 1993. The Relations Between Attributional Style, Life Events, Event Attribution, Hopelessness and Depression. Seoul National University, Seoul (Doctoral dissertation). Lee, Y.H, Song, J.Y., 1991. A study of the reliability and the validity of the BDI, SDS, and MMPI-D scales. Korean J. Clin. Psychol. 10, 98–112. Miettunen, J., Raevuori, A., 2012. A meta-analysis of temperament in axis I psychiatric disorders. Compr. Psychiatry 53, 152–166. https://doi.org/10.1016/j.comppsych. 2011.03.008. Min, B.B., Oh, H.S., Lee, J.Y., 2007. Temperament and Character Inventory-RevisedShort. Maumsarang, Seoul, Korea. Moritz, S., Göritz, A.S., McLean, B., Westermann, S., Brodbeck, J., 2017. Do depressive symptoms predict paranoia or vice versa? J. Behav. Ther. Exp. Psychiatry 56, 113–121. https://doi.org/10.1016/j.jbtep.2016.10.002. Muthén, L.K., Muthén, B.O., 2015. Mplus, seventh ed. Muthén & Muthén, Los Angeles. Novick, J.S., Stewart, J.W., Wisniewski, S.R., Cook, I.A., Manev, R., Nierenberg, A.A., Rosenbaum, J.F., Shores-Wilson, K., Balasubramani, G.K., Biggs, M.M., Zisook, S., Rush, A.J., 2005. Clinical and demographic features of atypical depression in outpatients with major depressive disorder: preliminary findings from STAR*D. J. Clin. Psychiatry. 66, 1002–1011. https://doi.org/10.4088/JCP.v66n0807. Nylund, K.L., Asparouhov, T., Muthén, B.O., 2007. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Struct. Equ. Model. 14, 535–569. https://doi.org/10.1080/10705510701575396. Ohayon, M.M., Schatzberg, A.F., 2002. Prevalence of depressive episodes with psychotic features in the general population. Am. J. Psychiatry. 159, 1855–1861. https://doi. org/10.1176/appi.ajp.159.11.1855. Paavonen, V., Luoto, K., Lassila, A., Leinonen, E., Kampman, O., 2018. Temperament and character profiles are associated with depression outcome in psychiatric secondary care patients with harmful drinking. Compr. Psychiatry 84, 26–31. https://doi.org/ 10.1016/j.comppsych.2018.04.001. Parker, G., Fletcher, K., Paterson, A., Anderson, J., Hong, M., 2014. Gender differences in depression severity and symptoms across depressive sub-types. J. Affect. Disord. 167, 351–357. https://doi.org/10.1016/j.jad.2014.06.018. Parneix, M., Péricaud, M., Clément, J.P., 2014. Irritability associated with major depressive episodes: its relationship with mood disorders and temperament. Turk Psikiyatr. Derg. 25, 106–113. Rodgers, S., Grosse Holtforth, M., Müller, M., Hengartner, M.P., Rössler, W., AjdacicGross, V., 2014. Symptom-based subtypes of depression and their psychosocial correlates: a person-centered approach focusing on the influence of sex. J. Affect. Disord. 156, 92–103. https://doi.org/10.1016/j.jad.2013.11.021. Sarason, I.G., Johnson, J.H., Siegel, J.M., 1978. Assessing the impact of life changes: development of the life experiences survey. J. Consult. Clin. Psychol. 46, 932–946. https://doi.org/10.1037/0022-006X.46.5.932.
25