Deconstructing depression Monden, Rei
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date:
2017
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Monden, R. (2017). Deconstructing depression: A 3D perspective. Rijksuniversiteit Groningen.
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
autumn (Painted by Miho Hashimoto) autumn (Painted by Miho Hashimoto)
Chapter 6
Simultaneous decomposition of Depression and Anxiety heterogeneity on the Person-, Symptom- and Time-level in a large cohort study:
the Netherlands Study of Depression and Anxiety (NESDA)
Rei Monden
Klaas J. Wardenaar
Alwin Stegeman
Femke Lamers
Brenda W.J.H. Penninx
Peter de Jonge
Submitted
Abstract
Background: Depression and anxiety are highly comorbid and depressive and/or anxious patients show great variation in their symptom-specific course trajectories, which hampers research into patient-specific mechanisms and treatments. Therefore, to better capture the complex phenotypical variations across depressive and/or anxious patients, the current study aimed to decompose patients’ symptom data into more homogeneous person-, symptom- and time- level components, and to evaluate the components’ external correlates.
Method: Data came from 1-month depression and/or anxiety patients (n=792), who participated in the Netherlands Study of Depression and Anxiety (NESDA). The Inventory of Depressive Symptomatology, Beck Anxiety Index and adapted Mood and Anxiety Symptoms Questionnaire were administered at baseline and at 1-year, 2-year, 4-year and 6-year follow-up. The data were analyzed with Three-mode Principal Component Analysis (3MPCA). The resulting components’ correlations with external variables (i.e. clinical, biomarkers) were calculated.
Results: The optimal 3MPCA model (explained variance=29%) had three symptom-level (‘anxious-arousal’, ‘anhedonia’, ‘mood-cognition’) and two time- level components (‘improving’, ‘persisting’). There were four person-level components, characterized by different symptom-by-time patterns and external correlations: (1) the ‘Anhedonic’ person-component was characterized by persisting ‘anhedonia’ symptoms and low extraversion (r=-0.42), the ‘Somatic’
component by persisting ‘anxious-arousal’ and increased insomnia (r=0.35), the
‘Cognitive’ component by persisting ‘mood-cognition’ symptoms and high neuroticism (r=0.70), and the ‘Recovery’ component by quick recovery of all symptoms.
Conclusion: A parsimonious set of empirically-derived components and their interactions explain a considerable part of the variation across patients in their course trajectories on various symptom-domains. Making such subtle distinctions could enable much more patient-specific etiological and clinical research.
Introduction
Depressive and anxiety disorders are highly prevalent across the world and have a large impact on both patients and society [1-3]. Although the distinction between depressive and anxiety symptoms was already introduced in the DSM-I [4] (1952) and the fact that these diagnostic categories have been widely used over the past three decades, their validity has often been questioned [5]. One reason for this is that comorbidity of depression and anxiety is the rule rather than the exception [6]. About 75% of the patients with major depression having a comorbid psychiatric disorder, mainly anxiety disorders [7], which is not surprising given the overlap between depression and anxiety disorders in their criterion symptoms (i.e. sleep disturbance, fatigue [8-10]). Moreover, treatment indications for depressive and anxiety disorders are very similar, with either prescription of antidepressants and/or cognitive behavioral therapy as a recommended first line of treatment [11, 12].
Another issue that makes many authors question the validity and usefulness of DSM-based depression and anxiety classifications is their large heterogeneity (e.g. [13, 14]). Patients with the same particular depression or anxiety diagnosis can be very different in terms of their symptom-profiles [15, 16] and the numbers of possible patterns are even higher because of the high co- occurrence of depressive and anxiety symptomatology. Moreover, large variations have been found across depression and/or anxiety patients in their age of onset [8, 17], severity levels [18], and response to treatment [19, 20]. The latter makes it hard for clinicians to find the optimal treatment for individual patients with similar diagnoses. In addition, diagnostic heterogeneity hampers research into underlying (biological) mechanisms of depression and/or anxiety because variations within diagnostic groups are likely to obscure small between-group effects of genetic variants and/or other biomarkers [21].
Because of the above described issues, diagnostic heterogeneity has been and remains to be an important topic of investigation in both depression (e.g. [22]) and anxiety (e.g. [23]) research. However, given the high level of comorbidity and strong overlap between depression and anxiety, addressing the problem of heterogeneity for each disorder group separately is suboptimal. This was already recognized in the 1990’s, leading researchers to develop cross-diagnostic dimensional models to explain interpersonal symptom variations. The ensuing research showed that both a common (General Distress), and disorder-specific symptom dimensions for depression (Anhedonic Depression) and anxiety (Anxious Arousal) can be used to describe the many possible patterns of (co)occurrence of depression and anxiety [24]. Many studies have found support for this so-called ‘tripartite’ structure of depression and anxiety [25-30] and it has
Abstract
Background: Depression and anxiety are highly comorbid and depressive and/or anxious patients show great variation in their symptom-specific course trajectories, which hampers research into patient-specific mechanisms and treatments. Therefore, to better capture the complex phenotypical variations across depressive and/or anxious patients, the current study aimed to decompose patients’ symptom data into more homogeneous person-, symptom- and time- level components, and to evaluate the components’ external correlates.
Method: Data came from 1-month depression and/or anxiety patients (n=792), who participated in the Netherlands Study of Depression and Anxiety (NESDA). The Inventory of Depressive Symptomatology, Beck Anxiety Index and adapted Mood and Anxiety Symptoms Questionnaire were administered at baseline and at 1-year, 2-year, 4-year and 6-year follow-up. The data were analyzed with Three-mode Principal Component Analysis (3MPCA). The resulting components’ correlations with external variables (i.e. clinical, biomarkers) were calculated.
Results: The optimal 3MPCA model (explained variance=29%) had three symptom-level (‘anxious-arousal’, ‘anhedonia’, ‘mood-cognition’) and two time- level components (‘improving’, ‘persisting’). There were four person-level components, characterized by different symptom-by-time patterns and external correlations: (1) the ‘Anhedonic’ person-component was characterized by persisting ‘anhedonia’ symptoms and low extraversion (r=-0.42), the ‘Somatic’
component by persisting ‘anxious-arousal’ and increased insomnia (r=0.35), the
‘Cognitive’ component by persisting ‘mood-cognition’ symptoms and high neuroticism (r=0.70), and the ‘Recovery’ component by quick recovery of all symptoms.
Conclusion: A parsimonious set of empirically-derived components and their interactions explain a considerable part of the variation across patients in their course trajectories on various symptom-domains. Making such subtle distinctions could enable much more patient-specific etiological and clinical research.
Introduction
Depressive and anxiety disorders are highly prevalent across the world and have a large impact on both patients and society [1-3]. Although the distinction between depressive and anxiety symptoms was already introduced in the DSM-I [4] (1952) and the fact that these diagnostic categories have been widely used over the past three decades, their validity has often been questioned [5]. One reason for this is that comorbidity of depression and anxiety is the rule rather than the exception [6]. About 75% of the patients with major depression having a comorbid psychiatric disorder, mainly anxiety disorders [7], which is not surprising given the overlap between depression and anxiety disorders in their criterion symptoms (i.e. sleep disturbance, fatigue [8-10]). Moreover, treatment indications for depressive and anxiety disorders are very similar, with either prescription of antidepressants and/or cognitive behavioral therapy as a recommended first line of treatment [11, 12].
Another issue that makes many authors question the validity and usefulness of DSM-based depression and anxiety classifications is their large heterogeneity (e.g. [13, 14]). Patients with the same particular depression or anxiety diagnosis can be very different in terms of their symptom-profiles [15, 16] and the numbers of possible patterns are even higher because of the high co- occurrence of depressive and anxiety symptomatology. Moreover, large variations have been found across depression and/or anxiety patients in their age of onset [8, 17], severity levels [18], and response to treatment [19, 20]. The latter makes it hard for clinicians to find the optimal treatment for individual patients with similar diagnoses. In addition, diagnostic heterogeneity hampers research into underlying (biological) mechanisms of depression and/or anxiety because variations within diagnostic groups are likely to obscure small between-group effects of genetic variants and/or other biomarkers [21].
Because of the above described issues, diagnostic heterogeneity has been and remains to be an important topic of investigation in both depression (e.g. [22]) and anxiety (e.g. [23]) research. However, given the high level of comorbidity and strong overlap between depression and anxiety, addressing the problem of heterogeneity for each disorder group separately is suboptimal. This was already recognized in the 1990’s, leading researchers to develop cross-diagnostic dimensional models to explain interpersonal symptom variations. The ensuing research showed that both a common (General Distress), and disorder-specific symptom dimensions for depression (Anhedonic Depression) and anxiety (Anxious Arousal) can be used to describe the many possible patterns of (co)occurrence of depression and anxiety [24]. Many studies have found support for this so-called ‘tripartite’ structure of depression and anxiety [25-30] and it has
6
been found useful in etiological and clinical research [31, 32], although some studies have also found different latent structures for depression and anxiety (e.g.
[33-38]). Additional work has found that a hierarchical structure underlies depressive and anxiety disorders with each group of disorders being linked to a common higher-order ‘internalizing’ vulnerability dimension and a disorder- cluster dimension (e.g. fear vs. distress disorders). In addition, each disorder shows some unique variance and features (e.g. [39-41]). Together, these findings suggest that a cross-diagnostic approach using common and specific dimensions of depression and anxiety symptomatology to capture variations in symptomatology, offer an elegant and effective approach to capture inter-personal heterogeneity.
Another line of research has been focused on explaining heterogeneity in depression and anxiety by discrete subtyping rather than dimensions. Several recent studies have incorporated both depressive and anxiety symptoms in efforts to identify discrete population subgroups to explain population heterogeneity. For instance, Ten Have et al [42] used latent class analyses (LCA) to identify distinct patient subgroups and found subgroups with similar (moderate) depressive symptom-levels that showed differences with regard to their anxiety symptoms.
In addition, Wanders et al. (2016) [43] used a hybrid discrete-dimensional subtyping approach and found that anxiety symptoms co-occurred with depressive symptoms in all subgroups. These findings offer further support for the idea that anxiety and depression should be considered together to optimally capture all relevant symptom variations across patients.
Although above described research has offered many new insights into the underlying structure of depression and anxiety heterogeneity, it does not allow for an explanation of the whole problem because it has been largely focused on describing cross-sectional person- and symptom-heterogeneity. However, several empirical studies have shown that patients can differ strongly in terms of their course-trajectories on severity scales over time (Depression: e.g. [22, 44-47]);
Anxiety: e.g. [48-50]) and that for subgroups of patients, course-trajectories can differ depending on the symptom domain (e.g. [51, 52]). To incorporate this temporal source of heterogeneity into models, a more integrative modeling approach is needed that captures how patients differ from each other with regard to their course-trajectories on different symptom-domains. However, such incorporation of all modes of a longitudinal symptom dataset (i.e. [A] symptoms assessed in [B] persons at [C] repeated time-points) into a single model that explains the heterogeneity from these different interacting sources is very complex and unfeasible with traditional latent variable models (i.e. Factor Analysis, LCA, Growth Mixture Modeling).
Fortunately, incorporating three major sources of heterogeneity in one model is possible by means of Three-mode Principal Component Analysis (3MPCA [53-59]). This technique can be used to analyze longitudinal data and simultaneously decompose symptom-level, time-level and person-level data into limited sets of more homogeneous components, while accounting for interactions between the different levels’ components. As such, 3MPCA makes it possible to capture a sizable part of the variance present in a dataset with a parsimonious, data-driven model and, as such, provides a dimensional approach to describing variations across persons in their development over time on different symptom domains. 3MPCA has previously been used to investigate depression heterogeneity and the resulting components have been shown to have clear predictive value for clinically-relevant outcomes [60, 61] and treatment-outcome [62]. However, the approach has not been applied to a larger pool of symptoms of both depression and anxiety disorders. Therefore, the current study aimed to use 3MPCA to capture the heterogeneity of depression and/or anxiety in a single model and to investigate the components’ associations with a range of external variables, such as patients’ demographics, clinical characteristics and biomedical variables. To this end, 3MPCA was applied to repeatedly assessed depression and anxiety symptom data, collected in persons with a 1-month anxiety and/or depressive disorders (N=792), who were assessed at baseline and at 1-, 2-, 4- and 6- year follow-up.
Method
Participants and sample
The data came from the Netherlands Study of Depression and Anxiety (NESDA), an ongoing longitudinal cohort study that consists of 2,981 subjects (aged 18-65 years at baseline), who were recruited from community (19%), primary care (54%) and specialized mental health care organizations (27%). The majority of participants had an anxiety and/or depressive disorder (78%), but healthy controls (22%) were also included. Exclusion criteria were: not being fluent in the Dutch language, and/or a primary clinical diagnosis of psychotic, obsessive-compulsive, bipolar or severe addiction disorder. All patients had a face-to-face session at baseline and 2-, 4- and 6-year follow-up, which included a structured psychiatric interview by a trained research assistant, filling in self- report questionnaires, taking biological measurements and a blood-draw.
The study protocol was approved by Ethical Review Boards of all participating centers and all participants gave written informed consent. The data- collection procedure has been described in detail elsewhere [63]. For the current
been found useful in etiological and clinical research [31, 32], although some studies have also found different latent structures for depression and anxiety (e.g.
[33-38]). Additional work has found that a hierarchical structure underlies depressive and anxiety disorders with each group of disorders being linked to a common higher-order ‘internalizing’ vulnerability dimension and a disorder- cluster dimension (e.g. fear vs. distress disorders). In addition, each disorder shows some unique variance and features (e.g. [39-41]). Together, these findings suggest that a cross-diagnostic approach using common and specific dimensions of depression and anxiety symptomatology to capture variations in symptomatology, offer an elegant and effective approach to capture inter-personal heterogeneity.
Another line of research has been focused on explaining heterogeneity in depression and anxiety by discrete subtyping rather than dimensions. Several recent studies have incorporated both depressive and anxiety symptoms in efforts to identify discrete population subgroups to explain population heterogeneity. For instance, Ten Have et al [42] used latent class analyses (LCA) to identify distinct patient subgroups and found subgroups with similar (moderate) depressive symptom-levels that showed differences with regard to their anxiety symptoms.
In addition, Wanders et al. (2016) [43] used a hybrid discrete-dimensional subtyping approach and found that anxiety symptoms co-occurred with depressive symptoms in all subgroups. These findings offer further support for the idea that anxiety and depression should be considered together to optimally capture all relevant symptom variations across patients.
Although above described research has offered many new insights into the underlying structure of depression and anxiety heterogeneity, it does not allow for an explanation of the whole problem because it has been largely focused on describing cross-sectional person- and symptom-heterogeneity. However, several empirical studies have shown that patients can differ strongly in terms of their course-trajectories on severity scales over time (Depression: e.g. [22, 44-47]);
Anxiety: e.g. [48-50]) and that for subgroups of patients, course-trajectories can differ depending on the symptom domain (e.g. [51, 52]). To incorporate this temporal source of heterogeneity into models, a more integrative modeling approach is needed that captures how patients differ from each other with regard to their course-trajectories on different symptom-domains. However, such incorporation of all modes of a longitudinal symptom dataset (i.e. [A] symptoms assessed in [B] persons at [C] repeated time-points) into a single model that explains the heterogeneity from these different interacting sources is very complex and unfeasible with traditional latent variable models (i.e. Factor Analysis, LCA, Growth Mixture Modeling).
Fortunately, incorporating three major sources of heterogeneity in one model is possible by means of Three-mode Principal Component Analysis (3MPCA [53-59]). This technique can be used to analyze longitudinal data and simultaneously decompose symptom-level, time-level and person-level data into limited sets of more homogeneous components, while accounting for interactions between the different levels’ components. As such, 3MPCA makes it possible to capture a sizable part of the variance present in a dataset with a parsimonious, data-driven model and, as such, provides a dimensional approach to describing variations across persons in their development over time on different symptom domains. 3MPCA has previously been used to investigate depression heterogeneity and the resulting components have been shown to have clear predictive value for clinically-relevant outcomes [60, 61] and treatment-outcome [62]. However, the approach has not been applied to a larger pool of symptoms of both depression and anxiety disorders. Therefore, the current study aimed to use 3MPCA to capture the heterogeneity of depression and/or anxiety in a single model and to investigate the components’ associations with a range of external variables, such as patients’ demographics, clinical characteristics and biomedical variables. To this end, 3MPCA was applied to repeatedly assessed depression and anxiety symptom data, collected in persons with a 1-month anxiety and/or depressive disorders (N=792), who were assessed at baseline and at 1-, 2-, 4- and 6- year follow-up.
Method
Participants and sample
The data came from the Netherlands Study of Depression and Anxiety (NESDA), an ongoing longitudinal cohort study that consists of 2,981 subjects (aged 18-65 years at baseline), who were recruited from community (19%), primary care (54%) and specialized mental health care organizations (27%). The majority of participants had an anxiety and/or depressive disorder (78%), but healthy controls (22%) were also included. Exclusion criteria were: not being fluent in the Dutch language, and/or a primary clinical diagnosis of psychotic, obsessive-compulsive, bipolar or severe addiction disorder. All patients had a face-to-face session at baseline and 2-, 4- and 6-year follow-up, which included a structured psychiatric interview by a trained research assistant, filling in self- report questionnaires, taking biological measurements and a blood-draw.
The study protocol was approved by Ethical Review Boards of all participating centers and all participants gave written informed consent. The data- collection procedure has been described in detail elsewhere [63]. For the current
6
analyses, all patients with a 1-month diagnosis of a depressive and/or anxiety disorder (N=1,392) were initially selected.
Measures
Depression and anxiety symptoms
The short Dutch 30-item adaptation of the Mood and Anxiety Scoring Questionnaire (MASQ-D30 [30, 64]), 30-item Inventory of Depressive Symptomatology self-report (IDS-SR [65]) and the 21-item Beck Anxiety Inventory (BAI [66]) were administered at baseline, 1-, 2-, 4- and 6-year follow- up. The MASQ-D30 measures three domains: ‘general distress’ (10 items), anxious arousal’ (10 items) and ‘anhedonic depression’ (10 items). The
‘anhedonic depression’ items were reverse-scored [30] (e.g. ‘I felt successful’ ‘I felt really happy’) and recoded to reflect a lack of positive affect. In the IDS-SR, appetite gain and appetite loss were combined in a compound ‘appetite change’
item and weight-loss and gain were combined in a ‘weight change’ item. The assessments at baseline, 2-, 4-, and 6-year follow-up were administered during face-to-face sessions. The 1-year follow-up questionnaires were sent to the respondents and returned by mail.
Patient characteristics
Besides sociodemographics variables (e.g. age, working status), a range of additional variables was assessed. Personality traits were measured at baseline, 2- and 4-year follow-up with the Neuroticism-extraversion-Openness Five Factor Inventory (NEO-FFI [67]). Lifestyle variables were measured at baseline, 2-, 4- and 6-year follow-up, and included smoking (Fagerstorm test for nicotine dependence [68]), a self-reported number of different used drugs, sleep behavior (Insomnia Rating Scale [69]) and physical activity level (International Physical Activity Questionnaire [70]). Psychiatric variables were measured during the interview sessions and included current and lifetime DSM-diagnoses of depressive (MDD and dysthymia) and anxiety disorders (generalized anxiety disorder, social anxiety disorder, panic disorder, agoraphobia), number of months with a disorder in the past year, psychiatric history (e.g. number of previous episodes) and medication use. These variables were all assessed with the Composite International Diagnostic Interview (CIDI [71]) version 2.1.
Biomedical measures
Body Mass Index (BMI) and blood pressure were assessed during the interview sessions. For the blood-pressure measurement, the OMRON IntelliSense Professional Digital Blood Pressure Monitor, HEM-907XL (Omron Healthcare, Inc) was used. Systolic and diastolic blood pressures were measured twice during supine rest on the right arm and were averaged over the 2 measurements.
Biomarkers Blood markers
Blood markers were determined in blood samples collected by a blood- draw prior to the interview session (between 8:00 and 9:00 AM) after an overnight fast. Venous Blood samples (50ml) were transferred to a local lab to start processing within an hour and stored at -85°C for later assaying. The following routine assays were assessed: Gamma-GT, ASAT, ALAT, glucose, cholesterol, triglyceride, HDL- and LDL-cholesterol, haemoglobin, haematocrite, erythrocytes, thyroid-stimulating hormone and free thyroxine. In addition, high- sensitivity C-Reactive Protein (CRP), interleukin-6 (IL-6), Tumor Necrosis Factor (TNF)-α, Brain Derived neurotrophic factor (BDFN) and vitamin D were assessed.
High-sensitivity CRP plasma levels were assessed in duplicate by an in- house ELISA based on purified protein and polyclonal anti-CRP antibodies (Dako, Glostrup, Denmark). The CRP assay was standardized against the CRM 470 reference agent. The lower detection limit and the sensitivity of CRP are 0.1 mg/l and 0.05mg/l, respectively. Intra- and inter-assay coefficients of variation were 5% and 10%, respectively. Plasma levels of IL-6 were assessed in duplicate using a high sensitivity enzyme-linked immunosorbent assay (PeliKine Compact™
ELISA, Sanquin, Amsterdam). The IL-6 assay was standardized against a recombinant human IL-6 standard. The lower detection limit and the sensitivity of IL-6 are 0.35 pg/ml and 0.10 pg/ml, respectively. Intra- and inter-assay coefficients of variation were 8% and 12%, respectively. TNF-α levels were assayed in duplicate using a high-sensitivity solid phase ELISA (Quantikine® HS Human TNF-α Immunoassay, R&D systems Inc, Minneapolis, U.S.). The TNF-α assay was calibrated against a highly purified Escherichia coli-expressed recombinant human TNF-α. The lower detection limit and the sensitivity of TNF- α are 0.10 pg/ml and 0.11 pg/ml, respectively. Intra- and inter-assay coefficients of variation were 10% and 15%, respectively. BDNF levels were measured by the Emax Immuno Assay system from Promega according to the manufacturer’s protocol (Madison, MI, USA). Serum samples were diluted 100 times with Greiner Bio-One high-affinity 96-well plates, and the absorbency was read in
analyses, all patients with a 1-month diagnosis of a depressive and/or anxiety disorder (N=1,392) were initially selected.
Measures
Depression and anxiety symptoms
The short Dutch 30-item adaptation of the Mood and Anxiety Scoring Questionnaire (MASQ-D30 [30, 64]), 30-item Inventory of Depressive Symptomatology self-report (IDS-SR [65]) and the 21-item Beck Anxiety Inventory (BAI [66]) were administered at baseline, 1-, 2-, 4- and 6-year follow- up. The MASQ-D30 measures three domains: ‘general distress’ (10 items), anxious arousal’ (10 items) and ‘anhedonic depression’ (10 items). The
‘anhedonic depression’ items were reverse-scored [30] (e.g. ‘I felt successful’ ‘I felt really happy’) and recoded to reflect a lack of positive affect. In the IDS-SR, appetite gain and appetite loss were combined in a compound ‘appetite change’
item and weight-loss and gain were combined in a ‘weight change’ item. The assessments at baseline, 2-, 4-, and 6-year follow-up were administered during face-to-face sessions. The 1-year follow-up questionnaires were sent to the respondents and returned by mail.
Patient characteristics
Besides sociodemographics variables (e.g. age, working status), a range of additional variables was assessed. Personality traits were measured at baseline, 2- and 4-year follow-up with the Neuroticism-extraversion-Openness Five Factor Inventory (NEO-FFI [67]). Lifestyle variables were measured at baseline, 2-, 4- and 6-year follow-up, and included smoking (Fagerstorm test for nicotine dependence [68]), a self-reported number of different used drugs, sleep behavior (Insomnia Rating Scale [69]) and physical activity level (International Physical Activity Questionnaire [70]). Psychiatric variables were measured during the interview sessions and included current and lifetime DSM-diagnoses of depressive (MDD and dysthymia) and anxiety disorders (generalized anxiety disorder, social anxiety disorder, panic disorder, agoraphobia), number of months with a disorder in the past year, psychiatric history (e.g. number of previous episodes) and medication use. These variables were all assessed with the Composite International Diagnostic Interview (CIDI [71]) version 2.1.
Biomedical measures
Body Mass Index (BMI) and blood pressure were assessed during the interview sessions. For the blood-pressure measurement, the OMRON IntelliSense Professional Digital Blood Pressure Monitor, HEM-907XL (Omron Healthcare, Inc) was used. Systolic and diastolic blood pressures were measured twice during supine rest on the right arm and were averaged over the 2 measurements.
Biomarkers Blood markers
Blood markers were determined in blood samples collected by a blood- draw prior to the interview session (between 8:00 and 9:00 AM) after an overnight fast. Venous Blood samples (50ml) were transferred to a local lab to start processing within an hour and stored at -85°C for later assaying. The following routine assays were assessed: Gamma-GT, ASAT, ALAT, glucose, cholesterol, triglyceride, HDL- and LDL-cholesterol, haemoglobin, haematocrite, erythrocytes, thyroid-stimulating hormone and free thyroxine. In addition, high- sensitivity C-Reactive Protein (CRP), interleukin-6 (IL-6), Tumor Necrosis Factor (TNF)-α, Brain Derived neurotrophic factor (BDFN) and vitamin D were assessed.
High-sensitivity CRP plasma levels were assessed in duplicate by an in- house ELISA based on purified protein and polyclonal anti-CRP antibodies (Dako, Glostrup, Denmark). The CRP assay was standardized against the CRM 470 reference agent. The lower detection limit and the sensitivity of CRP are 0.1 mg/l and 0.05mg/l, respectively. Intra- and inter-assay coefficients of variation were 5% and 10%, respectively. Plasma levels of IL-6 were assessed in duplicate using a high sensitivity enzyme-linked immunosorbent assay (PeliKine Compact™
ELISA, Sanquin, Amsterdam). The IL-6 assay was standardized against a recombinant human IL-6 standard. The lower detection limit and the sensitivity of IL-6 are 0.35 pg/ml and 0.10 pg/ml, respectively. Intra- and inter-assay coefficients of variation were 8% and 12%, respectively. TNF-α levels were assayed in duplicate using a high-sensitivity solid phase ELISA (Quantikine® HS Human TNF-α Immunoassay, R&D systems Inc, Minneapolis, U.S.). The TNF-α assay was calibrated against a highly purified Escherichia coli-expressed recombinant human TNF-α. The lower detection limit and the sensitivity of TNF- α are 0.10 pg/ml and 0.11 pg/ml, respectively. Intra- and inter-assay coefficients of variation were 10% and 15%, respectively. BDNF levels were measured by the Emax Immuno Assay system from Promega according to the manufacturer’s protocol (Madison, MI, USA). Serum samples were diluted 100 times with Greiner Bio-One high-affinity 96-well plates, and the absorbency was read in
6
duplicate using a Bio-Rad Benchmark microplate reader at 450 nm. The intra- and inter-assay coefficients of variation were 3% and 9%, respectively. Vitamin D was measured based on circulating levels of 25(OH)D, which were extracted and analyzed by XLC-MS/MSa (Spark Holland, Emmen, the Netherlands) and coupled to a Quattro Premier XE tandem mass spectrometer (Waters Corp., Milford, MA, USA). Total 25(OH)D was calculated by the sum of concentrations of 25(OH)D2 and 25(OH)D3. Intra- and inter-assay coefficients of variation for concentrations between 25 and 180 nmoll -1 were 6 and 8%, respectively.
Saliva markers
Prior to the first face-to-face assessment session, participants themselves collected saliva samples at home using Salivettes (Sarstedt AG and Co, Nürmbrecht, Germany) at 6 time points: four morning samples (at awakening [T1], at 30 [T2], 45 [T3] and 60 [T4] minutes later) and two evening samples (at 22:00 [T5] and 23:00 h [T6]). Additionally, the dexamethasone suppression test (DST, REF) was carried out by oral administration of a 0.5-mg dexamethasone pill directly after T6 and a final cortisol sampling the next morning at awakening (T7). Within 15 minutes before the sampling, participants were prohibited to eat, smoke, drink or brush their teeth. The samples were stored in refrigerators and returned by mail.
The received samples were centrifuged at 2000g for 10 minutes, aliquoted, and stored at -80°C.The saliva samples were used to assess levels of cortisol, amylase and testosterone.
Cortisol levels were measured by Competitive electrochemiluminescence immunoassay (E170, Roche, Basel, Switzerland) at a functional detection limit of 2.0 nmol/l. Intra- and inter-assay variability of coefficients were less than 10%. Several cortisol indicators were calculated based on the assessed cortisol levels. First, the cortisol awakening response (CAR) was estimated by using the T1 to T4 measurements to calculate the area under the curve with respect to the ground (AUCg), which reflects the total cortisol secretion during the first hour after awakening, and the area under the curve with respect to the increase (AUCi), which reflects the sensitivity of the system and emphasizes changes over time [72]. Second, evening cortisol was estimated by averaging the T5 and T6 values. Third, the DST cortisol-suppression effect was estimated by calculating the cortisol suppression ratio (the cortisol level at awakening (T1)/the cortisol level after the dexamethasone ingestion (T7)). To estimate testosterone levels, 75 μl of the samples collected at T1 to T4 were mixed to generate one morning sample and 150 μl of samples collected at T5 and T6 were mixed to generate one evening sample. Using the testosterone in saliva assay from Diagnostic Biochem Canada (EiAsy Testosterone Saliva, DBC: CAN-TE-300) with 100 μl of material from the morning and evening sample respectively, free
testosterone in saliva was measured in duplo. To assess amylase levels, samples were diluted 50-fold with a Hamilton Microlab 500 B/C dilutor in physiological saline solution (Versylene® Fresenius, Cat. Nr. B230551) after overnight thawing of the evening saliva samples (T5 and T6) at 4°C. Using a kinetic colorimetric assay for total amylase activity (Cat Nr. 03183742, Roche Diagnostics, Mannheim, Germany), analyses were performed on a routine clinical chemistry analyzer.
Amylase activity was measured in IU/L at 37°C.
Statistical analyses Missing data
To avoid introducing bias in a multiple imputation procedure and to retain as much of the sample as possible, participants who did not provide any item score on the MASQ-D30, IDS-SR and BAI at more than one time point were excluded from the study. This resulted in a sample of 792 subjects, which constituted 57%
of the initially selected 1,392 participants with a 1-month diagnosis of depression and/or anxiety. Of the 792 participants, 406 subjects did not have any missing score. In total, 0.8% of the data were missing and imputed 20 times by using the Amelia II package [73] in R [74]. All of the above-listed patient characteristics and questionnaire scores were used in the multiple imputation models.
Three-mode Principal Component Analysis (3MPCA)
A detailed description of the 3MPCA analytical procedure can be found elsewhere [57, 60] and is briefly summarized below in seven steps. First, to investigate if the dataset contained a non-negligible three-way interaction between the person-, symptom-, and time-levels, and thus whether 3MPCA was warranted, a three-way ANOVA was performed in each imputed dataset to evaluate the proportion of variance explained by the main effects and the interaction effects of persons, symptoms and time. The averaged results across 20 imputed datasets were inspected and the size of the ‘three-way interaction plus error term’ was evaluated.
Standard deviations across the 20 imputed datasets were calculated to provide insight into the stability of the estimations. Second, each of the imputed datasets was centered across the person-mode to assure that the 3MPCA model captured only the heterogeneity around the ‘general trend’ in the data. Also each dataset was normalized within the symptom-mode to ensure that all items were treated as equally important in the model, irrespective of their variance. Third, selection of the number of components for each mode (persons, symptoms and time points) was done based on two criteria: (i) the generalized scree test, which balances model complexity with explained variance [75, 76] and (ii) the stability [57], which was determined with a split-half procedure. Fourth, to obtain an
duplicate using a Bio-Rad Benchmark microplate reader at 450 nm. The intra- and inter-assay coefficients of variation were 3% and 9%, respectively. Vitamin D was measured based on circulating levels of 25(OH)D, which were extracted and analyzed by XLC-MS/MSa (Spark Holland, Emmen, the Netherlands) and coupled to a Quattro Premier XE tandem mass spectrometer (Waters Corp., Milford, MA, USA). Total 25(OH)D was calculated by the sum of concentrations of 25(OH)D2 and 25(OH)D3. Intra- and inter-assay coefficients of variation for concentrations between 25 and 180 nmoll -1 were 6 and 8%, respectively.
Saliva markers
Prior to the first face-to-face assessment session, participants themselves collected saliva samples at home using Salivettes (Sarstedt AG and Co, Nürmbrecht, Germany) at 6 time points: four morning samples (at awakening [T1], at 30 [T2], 45 [T3] and 60 [T4] minutes later) and two evening samples (at 22:00 [T5] and 23:00 h [T6]). Additionally, the dexamethasone suppression test (DST, REF) was carried out by oral administration of a 0.5-mg dexamethasone pill directly after T6 and a final cortisol sampling the next morning at awakening (T7). Within 15 minutes before the sampling, participants were prohibited to eat, smoke, drink or brush their teeth. The samples were stored in refrigerators and returned by mail.
The received samples were centrifuged at 2000g for 10 minutes, aliquoted, and stored at -80°C.The saliva samples were used to assess levels of cortisol, amylase and testosterone.
Cortisol levels were measured by Competitive electrochemiluminescence immunoassay (E170, Roche, Basel, Switzerland) at a functional detection limit of 2.0 nmol/l. Intra- and inter-assay variability of coefficients were less than 10%. Several cortisol indicators were calculated based on the assessed cortisol levels. First, the cortisol awakening response (CAR) was estimated by using the T1 to T4 measurements to calculate the area under the curve with respect to the ground (AUCg), which reflects the total cortisol secretion during the first hour after awakening, and the area under the curve with respect to the increase (AUCi), which reflects the sensitivity of the system and emphasizes changes over time [72]. Second, evening cortisol was estimated by averaging the T5 and T6 values. Third, the DST cortisol-suppression effect was estimated by calculating the cortisol suppression ratio (the cortisol level at awakening (T1)/the cortisol level after the dexamethasone ingestion (T7)). To estimate testosterone levels, 75 μl of the samples collected at T1 to T4 were mixed to generate one morning sample and 150 μl of samples collected at T5 and T6 were mixed to generate one evening sample. Using the testosterone in saliva assay from Diagnostic Biochem Canada (EiAsy Testosterone Saliva, DBC: CAN-TE-300) with 100 μl of material from the morning and evening sample respectively, free
testosterone in saliva was measured in duplo. To assess amylase levels, samples were diluted 50-fold with a Hamilton Microlab 500 B/C dilutor in physiological saline solution (Versylene® Fresenius, Cat. Nr. B230551) after overnight thawing of the evening saliva samples (T5 and T6) at 4°C. Using a kinetic colorimetric assay for total amylase activity (Cat Nr. 03183742, Roche Diagnostics, Mannheim, Germany), analyses were performed on a routine clinical chemistry analyzer.
Amylase activity was measured in IU/L at 37°C.
Statistical analyses Missing data
To avoid introducing bias in a multiple imputation procedure and to retain as much of the sample as possible, participants who did not provide any item score on the MASQ-D30, IDS-SR and BAI at more than one time point were excluded from the study. This resulted in a sample of 792 subjects, which constituted 57%
of the initially selected 1,392 participants with a 1-month diagnosis of depression and/or anxiety. Of the 792 participants, 406 subjects did not have any missing score. In total, 0.8% of the data were missing and imputed 20 times by using the Amelia II package [73] in R [74]. All of the above-listed patient characteristics and questionnaire scores were used in the multiple imputation models.
Three-mode Principal Component Analysis (3MPCA)
A detailed description of the 3MPCA analytical procedure can be found elsewhere [57, 60] and is briefly summarized below in seven steps. First, to investigate if the dataset contained a non-negligible three-way interaction between the person-, symptom-, and time-levels, and thus whether 3MPCA was warranted, a three-way ANOVA was performed in each imputed dataset to evaluate the proportion of variance explained by the main effects and the interaction effects of persons, symptoms and time. The averaged results across 20 imputed datasets were inspected and the size of the ‘three-way interaction plus error term’ was evaluated.
Standard deviations across the 20 imputed datasets were calculated to provide insight into the stability of the estimations. Second, each of the imputed datasets was centered across the person-mode to assure that the 3MPCA model captured only the heterogeneity around the ‘general trend’ in the data. Also each dataset was normalized within the symptom-mode to ensure that all items were treated as equally important in the model, irrespective of their variance. Third, selection of the number of components for each mode (persons, symptoms and time points) was done based on two criteria: (i) the generalized scree test, which balances model complexity with explained variance [75, 76] and (ii) the stability [57], which was determined with a split-half procedure. Fourth, to obtain an
6
interpretable 3MPCA component structure an orthogonal Joint Orthomax rotation [77] was used. Here, standard weights were used, but no weights were applied in the person-mode. Fifth, the resulting twenty 3MPCA solutions were averaged for the person-, symptom- and time-mode components, respectively. The core array, which indicates the strength of the three-way interactions, was averaged by means of a generalized Procrustes rotation [59, 78, 79]. Sixth, two types of fit percentages were calculated for the resulting 3MPCA model: the explained variance of the 3MPCA model with and without ‘general trend’ (see details in [60]). Seventh, to interpret the person-mode components, correlations (Spearman’s ρ or Pearson’s r) were calculated between the person-mode component scores and the above described patient characteristics and biomarkers.
Bonferroni correction was used to adjust calculated P-values for multiple testing.
3MPCA was conducted with Tucker3.m [57, 80]. Except for the multiple imputation procedure, all analyses were done in MATLAB [81].
Results
Descriptive statistics
The sample contained 68.2% of women and the mean age was 43.1 years (s.d.=12.3). Of the sample, 26.6% had only a 1-month depressive disorder, 37.8%
had only an anxiety disorder, 35.6% had both a depressive and anxiety disorder, and 71.3% used psychotropic medication. The mean baseline sum score on the BAI and IDS-SR were 16.6 (s.d.=9.8) and 29.0 (s.d. = 11.8), respectively, indicating moderate severity levels (BAI [82] ;IDS-SR [83]). Inspection of the mean item-scores over time revealed various trends (Figure 1), with some item- scores showing a dramatic decrease between baseline and 2-year follow-up and other items showing more persistent high or low mean scores over time, which indicated considerable heterogeneity in the dataset.
Figure 1 Mean scores on three items at each of the time points
Three-mode Principal Component Analysis Three-way ANOVA
The results of the fixed-effect three-way ANOVA (Table 1) showed the largest explained variance for the symptom-level (35.8%), followed by the ‘three-way interactions plus error’ term (24.5%). As the proportion of the explained variance by the ‘three-way interactions plus error term’, was considerable, 3MPCA was a valid approach to analyze the data. Given that the ‘time’, ‘person*time’ and
‘symptom*time’ effects showed small explained variances in Table 1, the variation explained by ‘time’ was likely to be limited compared to the other effects and interactions.
Table 1 Three-way Analysis of Variance
Effect SS std % expl. Var Std
Person 47759.68 48.01 11.40 0.0001
Symptom 149877.11 125.36 35.77 0.0002
Time 3566.27 16.93 0.85 0.0000
Person*Symptom 95657.92 43.15 22.83 0.0001
Person*Time 18326.81 35.90 4.37 0.0001
Symptom*Time 1054.92 6.39 0.25 0.0000
Person*Symptom*Time + error 102718.53 67.40 24.52 0.0002
Total 418961.24 343.15 100
interpretable 3MPCA component structure an orthogonal Joint Orthomax rotation [77] was used. Here, standard weights were used, but no weights were applied in the person-mode. Fifth, the resulting twenty 3MPCA solutions were averaged for the person-, symptom- and time-mode components, respectively. The core array, which indicates the strength of the three-way interactions, was averaged by means of a generalized Procrustes rotation [59, 78, 79]. Sixth, two types of fit percentages were calculated for the resulting 3MPCA model: the explained variance of the 3MPCA model with and without ‘general trend’ (see details in [60]). Seventh, to interpret the person-mode components, correlations (Spearman’s ρ or Pearson’s r) were calculated between the person-mode component scores and the above described patient characteristics and biomarkers.
Bonferroni correction was used to adjust calculated P-values for multiple testing.
3MPCA was conducted with Tucker3.m [57, 80]. Except for the multiple imputation procedure, all analyses were done in MATLAB [81].
Results
Descriptive statistics
The sample contained 68.2% of women and the mean age was 43.1 years (s.d.=12.3). Of the sample, 26.6% had only a 1-month depressive disorder, 37.8%
had only an anxiety disorder, 35.6% had both a depressive and anxiety disorder, and 71.3% used psychotropic medication. The mean baseline sum score on the BAI and IDS-SR were 16.6 (s.d.=9.8) and 29.0 (s.d. = 11.8), respectively, indicating moderate severity levels (BAI [82] ;IDS-SR [83]). Inspection of the mean item-scores over time revealed various trends (Figure 1), with some item- scores showing a dramatic decrease between baseline and 2-year follow-up and other items showing more persistent high or low mean scores over time, which indicated considerable heterogeneity in the dataset.
Figure 1 Mean scores on three items at each of the time points
Three-mode Principal Component Analysis Three-way ANOVA
The results of the fixed-effect three-way ANOVA (Table 1) showed the largest explained variance for the symptom-level (35.8%), followed by the ‘three-way interactions plus error’ term (24.5%). As the proportion of the explained variance by the ‘three-way interactions plus error term’, was considerable, 3MPCA was a valid approach to analyze the data. Given that the ‘time’, ‘person*time’ and
‘symptom*time’ effects showed small explained variances in Table 1, the variation explained by ‘time’ was likely to be limited compared to the other effects and interactions.
Table 1 Three-way Analysis of Variance
Effect SS std % expl. Var Std
Person 47759.68 48.01 11.40 0.0001
Symptom 149877.11 125.36 35.77 0.0002
Time 3566.27 16.93 0.85 0.0000
Person*Symptom 95657.92 43.15 22.83 0.0001
Person*Time 18326.81 35.90 4.37 0.0001
Symptom*Time 1054.92 6.39 0.25 0.0000
Person*Symptom*Time + error 102718.53 67.40 24.52 0.0002
Total 418961.24 343.15 100
6
Model complexity and fit percentages
To balance model interpretability and complexity, the generalized scree test was performed with the maximum number of components set to (5,4,3) for the person-, symptom- and time-mode, respectively. Either the (4,3,2) structure or (3,2,2) structure was suggested by the test, with fit percentage of 26.6% and 29.2%, respectively. For both structures, stability was high, as indicated by minimum congruence coefficients for all modes’ components ≥0.97. Based on the higher fit percentage and its interpretability, the (4,3,2) structure was chosen.
The estimated (4,3,2) 3MPCA model, averaged across 20 imputed datasets, explained 29.2% (s.d.=0.01) of variance when the general trend was excluded from the data and 91.8% (s.d.=0.007) if the general trend was included.
Small standard deviations indicated high consistency across the 20 imputed datasets. The large difference between the fit percentages for the model with and without the general trend indicated that most of the patients in the dataset followed the general trend to a large extent. This is not surprising given that ‘time’ effect in Table 1 was associated with relatively small explained variance, indicating little variation over time around the general trend. As a result, the variations around this trend captured by the 3MPCA model were limited in comparison.
Indeed, the (4,3,2) model showed that only a limited number of components was needed to capture this variation.
Symptom-mode components
The symptom-mode component scores are presented in Table 2. Interestingly, the symptom components consists of items from different questionnaires. The first component had high loadings on panic/anxiety-related arousal items (e.g. ‘face flushed’, ‘feeling hot’) and was therefore labeled the ‘anxious-arousal’
component. The second component had high loadings on lack of positive affective items (e.g. rescored ‘I felt like I was having a lot of fun’) and was therefore labeled the ‘anhedonia’ component. The third component had high loading symptoms that were characterized by depressive mood, thoughts and/or cognitions (e.g. ‘I blamed myself for a lot of things’ or ‘I felt inferior to others’) and was labeled the
‘mood-cognition’ component. Table 2 also illustrated that 24 out of 81 items did not load on either of the symptom components. All estimated scores had small standard deviations, indicating high consistency across the 20 imputed datasets.
Table 2 Symptom-mode components
Items Source Anxious
-arousal Anhedonia Mood-cognition
Other bodily symptoms IDS-SR 0.25 0.01 -0.05
Face flushed BAI 0.22 0.00 -0.05
Feeling hot BAI 0.21 0.04 -0.08
Hot, cold sweats BAI 0.21 0.02 -0.04
Aches and pains IDS-SR 0.20 0.10 -0.09
Had hot or cold spells MASQ-D30 0.20 0.01 -0.05
Heart pounding, racing BAI 0.20 -0.06 0.01
Numbness or tingling BAI 0.19 0.01 -0.03
Difficulty in breathing BAI 0.19 -0.03 0.01
Wobbliness in legs BAI 0.19 -0.01 0.02
Dizzy or lightheaded BAI 0.19 0.00 0.00
Had pain in my chest MASQ-D30 0.18 -0.04 0.01
Felt dizzy or light-headed MASQ-D30 0.18 -0.02 0.02
Muscles were tense or sore MASQ-D30 0.18 0.04 0.00
Shaky, unsteady BAI 0.18 -0.02 0.04
Was short of breath MASQ-D30 0.18 0.01 -0.02
Was trembling or shaking MASQ-D30 0.17 -0.02 0.04
Heart was racing or pounding MASQ-D30 0.17 -0.08 0.05
Unsteady BAI 0.17 0.01 0.02
Indigestion BAI 0.16 0.01 0.00
Had trouble swallowing MASQ-D30 0.15 -0.04 0.02
Hands trembling BAI 0.15 -0.01 0.04
Felt like I was having a lot of fun* MASQ-D30 0.01 0.28 -0.02 Felt really ‘up’ or lively* MASQ-D30 0.01 0.27 -0.01 Felt like I had a lot to look forward
to* MASQ-D30 -0.01 0.26 -0.01
Felt optimistic* MASQ-D30 -0.01 0.25 0.02
Felt really good about myself* MASQ-D30 0.00 0.25 0.04
Felt really happy* MASQ-D30 -0.01 0.24 0.03
Felt like I had a lot of energy* MASQ-D30 0.03 0.24 -0.01
Felt really talkative* MASQ-D30 -0.03 0.24 -0.04
Felt like I had accomplished a lot* MASQ-D30 -0.01 0.24 0.00
Felt successful* MASQ-D30 0.01 0.22 0.01
Interest in Sex (Please Rate
Interest, not Activity) IDS-SR 0.08 0.17 -0.05
Capacity for Pleasure or
Enjoyment (excluding sex) IDS-SR 0.03 0.15 0.08
Blamed myself for a lot of things MASQ-D30 -0.07 -0.01 0.25
Felt inferior to others MASQ-D30 -0.07 0.00 0.25
Felt worthless MASQ-D30 -0.06 0.05 0.22
Worried a lot about things MASQ-D30 -0.02 0.01 0.22
Scared BAI 0.05 -0.11 0.21
Felt hopeless MASQ-D30 -0.03 0.05 0.21
Interpersonal Sensitivity IDS-SR -0.05 0.01 0.21
Felt irritable MASQ-D30 -0.01 -0.02 0.20
Fear of worst happening BAI 0.05 -0.10 0.20
Fear of losing control BAI 0.04 -0.08 0.20
View of Myself IDS-SR -0.08 0.06 0.20
Continue to the next page
Model complexity and fit percentages
To balance model interpretability and complexity, the generalized scree test was performed with the maximum number of components set to (5,4,3) for the person-, symptom- and time-mode, respectively. Either the (4,3,2) structure or (3,2,2) structure was suggested by the test, with fit percentage of 26.6% and 29.2%, respectively. For both structures, stability was high, as indicated by minimum congruence coefficients for all modes’ components ≥0.97. Based on the higher fit percentage and its interpretability, the (4,3,2) structure was chosen.
The estimated (4,3,2) 3MPCA model, averaged across 20 imputed datasets, explained 29.2% (s.d.=0.01) of variance when the general trend was excluded from the data and 91.8% (s.d.=0.007) if the general trend was included.
Small standard deviations indicated high consistency across the 20 imputed datasets. The large difference between the fit percentages for the model with and without the general trend indicated that most of the patients in the dataset followed the general trend to a large extent. This is not surprising given that ‘time’ effect in Table 1 was associated with relatively small explained variance, indicating little variation over time around the general trend. As a result, the variations around this trend captured by the 3MPCA model were limited in comparison.
Indeed, the (4,3,2) model showed that only a limited number of components was needed to capture this variation.
Symptom-mode components
The symptom-mode component scores are presented in Table 2. Interestingly, the symptom components consists of items from different questionnaires. The first component had high loadings on panic/anxiety-related arousal items (e.g. ‘face flushed’, ‘feeling hot’) and was therefore labeled the ‘anxious-arousal’
component. The second component had high loadings on lack of positive affective items (e.g. rescored ‘I felt like I was having a lot of fun’) and was therefore labeled the ‘anhedonia’ component. The third component had high loading symptoms that were characterized by depressive mood, thoughts and/or cognitions (e.g. ‘I blamed myself for a lot of things’ or ‘I felt inferior to others’) and was labeled the
‘mood-cognition’ component. Table 2 also illustrated that 24 out of 81 items did not load on either of the symptom components. All estimated scores had small standard deviations, indicating high consistency across the 20 imputed datasets.
Table 2 Symptom-mode components
Items Source Anxious
-arousal Anhedonia Mood-cognition
Other bodily symptoms IDS-SR 0.25 0.01 -0.05
Face flushed BAI 0.22 0.00 -0.05
Feeling hot BAI 0.21 0.04 -0.08
Hot, cold sweats BAI 0.21 0.02 -0.04
Aches and pains IDS-SR 0.20 0.10 -0.09
Had hot or cold spells MASQ-D30 0.20 0.01 -0.05
Heart pounding, racing BAI 0.20 -0.06 0.01
Numbness or tingling BAI 0.19 0.01 -0.03
Difficulty in breathing BAI 0.19 -0.03 0.01
Wobbliness in legs BAI 0.19 -0.01 0.02
Dizzy or lightheaded BAI 0.19 0.00 0.00
Had pain in my chest MASQ-D30 0.18 -0.04 0.01
Felt dizzy or light-headed MASQ-D30 0.18 -0.02 0.02
Muscles were tense or sore MASQ-D30 0.18 0.04 0.00
Shaky, unsteady BAI 0.18 -0.02 0.04
Was short of breath MASQ-D30 0.18 0.01 -0.02
Was trembling or shaking MASQ-D30 0.17 -0.02 0.04
Heart was racing or pounding MASQ-D30 0.17 -0.08 0.05
Unsteady BAI 0.17 0.01 0.02
Indigestion BAI 0.16 0.01 0.00
Had trouble swallowing MASQ-D30 0.15 -0.04 0.02
Hands trembling BAI 0.15 -0.01 0.04
Felt like I was having a lot of fun* MASQ-D30 0.01 0.28 -0.02 Felt really ‘up’ or lively* MASQ-D30 0.01 0.27 -0.01 Felt like I had a lot to look forward
to* MASQ-D30 -0.01 0.26 -0.01
Felt optimistic* MASQ-D30 -0.01 0.25 0.02
Felt really good about myself* MASQ-D30 0.00 0.25 0.04
Felt really happy* MASQ-D30 -0.01 0.24 0.03
Felt like I had a lot of energy* MASQ-D30 0.03 0.24 -0.01
Felt really talkative* MASQ-D30 -0.03 0.24 -0.04
Felt like I had accomplished a lot* MASQ-D30 -0.01 0.24 0.00
Felt successful* MASQ-D30 0.01 0.22 0.01
Interest in Sex (Please Rate
Interest, not Activity) IDS-SR 0.08 0.17 -0.05
Capacity for Pleasure or
Enjoyment (excluding sex) IDS-SR 0.03 0.15 0.08
Blamed myself for a lot of things MASQ-D30 -0.07 -0.01 0.25
Felt inferior to others MASQ-D30 -0.07 0.00 0.25
Felt worthless MASQ-D30 -0.06 0.05 0.22
Worried a lot about things MASQ-D30 -0.02 0.01 0.22
Scared BAI 0.05 -0.11 0.21
Felt hopeless MASQ-D30 -0.03 0.05 0.21
Interpersonal Sensitivity IDS-SR -0.05 0.01 0.21
Felt irritable MASQ-D30 -0.01 -0.02 0.20
Fear of worst happening BAI 0.05 -0.10 0.20
Fear of losing control BAI 0.04 -0.08 0.20
View of Myself IDS-SR -0.08 0.06 0.20
Continue to the next page
