Abstract
Major depression consists of multiple phenotypic traits. Our objective was to characterize depressive phenotypes in the patient health questionnaire (PHQ)-9 using the Research Domain Criteria (RDoC) research framework. Cross-sectional data were examined from the 2013–2014 (N = 5397) and 2015–2016 (N = 5164) National Health and Nutrition Examination Survey, a large, nationally representative U.S. sample. Using both factor analysis and qualitative analysis in mapping scale items along RDoC domains, a four factor model was found to be theoretically appropriate and had an excellent model fit for the PHQ-9. The factor structure consisted of phenotypes describing Negative Valence Systems and Externalizing (anhedonia and depression), Negative Valence Systems and Internalizing (depression, guilt and self-harm), Arousal and Regulatory Systems (sleep, fatigue and appetite) and Cognitive and Sensorimotor Systems (concentration and psychomotor). High correlation between these phenotypes did indicate screening and monitoring for depression study population using a single depression score is likely useful in most circumstances. In multiple indicator multiple cause analysis, differences in the means of the phenotypic traits were found by age, race/ethnicity, sex, and number of comorbidities. Future research should explore whether phenotype expression derived from readily available self-rated depression scales can help to inform more personalized care.
Keywords: major depression, patient health questionnaire-9, research domain criteria, National Health and Nutrition Examination Survey, factor analysis, multiple indicator multiple cause modeling
1. Introduction
Major depression is a common and heterogenous condition (Brody et al., 2001; Hasler and Northoff, 2011; Leventhal et al., 2008; Pizzagalli et al., 2005). By partitioning major depression into phenotypes based on the groupings of particular key symptoms, markers representing more direct expressions of underlying genes, neurophysiology, and psychosocial characteristics might be isolated (Gottesman and Gould, 2003; Hasler and Northoff, 2011; Leventhal et al., 2008). Characterizing depressive phenotypes could also have important implications for treatment and outcomes.
In general depressed populations, the distinction between somatic symptoms (e.g. fatigue, psychomotor impairment) and cognitive/affective symptoms (e.g. depressed mood, feelings of guilt) has been the foundation for many psychometric studies (Beck et al., 1996; Lux and Kendler, 2010). The patient health questionnaire (PHQ)-9 is a commonly used depression screening tool in clinical practice (Kroenke et al., 2001; Levis et al., 2019; Mitchell et al., 2016). Previous psychometric studies within the PHQ-9 have commonly utilized either a single total score or cognitive/affective and somatic symptom domains (Dum et al., 2008; Granillo, 2012; Gunzler et al., 2015; Huang et al., 2006; Kalpakjian et al., 2009; Patel et al.; Titov et al., 2011). One can describe PHQ-9 items for anhedonia, depression, guilt and self-harm as affective symptoms and sleep, fatigue and appetite as somatic symptoms. Psychomotor and concentration are cognitive symptoms that could also be considered somatic symptoms.
As a recent example, Patel et al. (2019) performed confirmatory factor analysis (CFA) on a series of models under the assumption that the PHQ-9 consists of either a single total score or cognitive/affective and somatic symptom domains. Factor analysis is a multivariate technique for investigating whether a latent construct or constructs underlie a set of item responses and, if so, what the relationships are among the latent constructs. Exploratory factor analysis (EFA) is an unsupervised learning method used to help identify the underlying construct or constructs while CFA is used to verify the relationships between the latent constructs and a set of item responses.
Huang et al. (2006) performed EFA and found a unidimensional construct for the PHQ-9 using eigenvalue criterion. Granillo (2012) performed EFA and found affective and somatic dimensions for the PHQ-9 using model fit criterion. Gunzler et al. (2015) noted similar mixed results for EFA for the PHQ-9 depending on the evaluation criterion (eigenvalues, model fit) and used a unidimensional construct with residual correlations as a compromising solution. However, to the best of our knowledge, none of these prior psychometric studies of the PHQ-9 considered a research framework with multiple domains integrating many levels of information (e.g. genomics, circuits to behavior and self-reports) outside of the cognitive/affective and somatic symptom distinction.
Our analysis used the Research Domain Criteria (RDoC) matrix (Insel et al., 2010) as a research foundation for our study. RDoC is a research framework for approaches to investigating mental disorders such a major depression. The current version of the RDoC matrix is constructed around six domains reflecting contemporary conceptualization of major systems of emotion, cognition, motivation, and social behavior. In conjunction with the RDoC matrix we used factor analytic methods to help identify phenotypes of depression based on common variance in groupings of the items of the PHQ-9 (i.e. items for anhedonia, depressed mood, sleep, fatigue, appetite, guilt, psychomotor, concentration and self-harm).
To the best of our knowledge, no prior study has evaluated depressive phenotypes using the RDoC matrix within a PHQ-9 framework. The results of our analysis has potential to provide an analytic structure for describing phenotypic expressions within major depression in a general population being assessed with a commonly used depression screening tool. Findings may help lay the groundwork for future research investigating whether phenotypes identified among patients being screened for depression may be used to inform algorithms for clinical decision-making.
2. Methods
2.1. Design and Measures
The National Health and Nutrition Examination Survey (NHANES) is a program of the National Center for Health Statistics that began in 1960. The objective of the NHANES is to assess the health and nutritional status of individuals in the United States. The NHANES is a cross-sectional collection of surveys and other health examination data for a nationally representative sample of the resident, civilian, non-institutionalized U.S. population with approximately 5,000 individuals sampled each year (Brody et al., 2018; Control and Prevention, 2006). We used cross-sectional data from 2013–2014 (N = 5397) as a sample for performing EFA and 2015–2016 (N = 5164) as a sample for performing CFA and CFA with covariates (i.e. multiple indicator multiple cause analysis) in our study.
The inclusion criterion for our study was having a recorded PHQ-9 total score and being 18 or older. The PHQ-9 is a self-reported instrument that provides information regarding severity of depressive symptoms and treatment response. The PHQ-9 is often used to guide treatment decisions in clinical practice (Kroenke et al., 2001). On the PHQ-9, individuals specify frequency in the past 2 weeks (0 = not at all to 3 = every day) of nine symptoms, yielding a total score (range: 0–27). In particular, a PHQ-9 ≥ 10 has been previously established as a screening cutoff for depressive disorder given high sensitivity and specificity at this score (Ferrando et al., 2007; Kroenke et al., 2001). The PHQ-9 has been validated using multiple modes for administration, clinical populations, and diverse racial and ethnic groups (Pinto‐Meza et al., 2005).
Our dataset extracted from the NHANES survey included sex, age, number of comorbidities (ranging from 0 to 11) and race/ethnicity (Hispanic, Non-Hispanic White, Non-Hispanic Black and Other Race). See Table 1 for a descriptive summary of our study population.
Table 1.
Descriptive Summary of NHANES Survey Data Study Population (2013–2014 and 2015–2016)
| 2013–2014 | 2015–2016 | |
|---|---|---|
| (n=5397) | (n=5164) | |
| Anhedonia | ||
| 0 | 3967 (74 %) | 3798 (74 %) |
| 1 | 878 (16 %) | 850 (17 %) |
| 2 | 291 (5 %) | 253 (5 %) |
| 3 | 236 (4 %) | 233 (5 %) |
| Depression | ||
| 0 | 4064 (76 %) | 3906 (76 %) |
| 1 | 900 (17 %) | 865 (17 %) |
| 2 | 212 (4 %) | 207 (4 %) |
| 3 | 196 (4 %) | 156 (3 %) |
| Sleep | ||
| 0 | 3402 (63 %) | 3178 (62 %) |
| 1 | 1125 (21 %) | 1190 (23 %) |
| 2 | 349 (6 %) | 351 (7 %) |
| 3 | 496 (9 %) | 415 (8 %) |
| Fatigue | ||
| 0 | 2639 (49 %) | 2453 (48 %) |
| 1 | 1827 (34 %) | 1789 (35 %) |
| 2 | 421 (8 %) | 460 (9 %) |
| 3 | 485 (9 %) | 432 (8 %) |
| Appetite | ||
| 0 | 4020 (75 %) | 3794 (74 %) |
| 1 | 831 (15 %) | 866 (17 %) |
| 2 | 275 (5 %) | 240 (5 %) |
| 3 | 246 (5 %) | 234 (5 %) |
| Guilt | ||
| 0 | 4439 (83 %) | 4302 (84 %) |
| 1 | 616 (11 %) | 553 (11 %) |
| 2 | 154 (3 %) | 146 (3 %) |
| 3 | 163 (3 % | 133 (3 %) |
| Concentration | ||
| 0 | 4410 (82 %) | 4306 (84 %) |
| 1 | 584 (11 %) | 519 (10 %) |
| 2 | 191 (4 %) | 145 (3 %) |
| 3 | 187 (3 %) | 164 (3 %) |
| Psychomotor | ||
| 0 | 4771 (89 %) | 4598 (90 %) |
| 1 | 365 (7 %) | 347 (7 %) |
| 2 | 133 (2 %) | 98 (2 %) |
| 3 | 103 (2 %) | 91 (2 %) |
| Self-Harm | ||
| 0 | 5189 (97 %) | 4942 (96 %) |
| 1 | 121 (2 %) | 126 (2 %) |
| 2 | 31 (1 %) | 36 (1 %) |
| 3 | 31 (1 %) | 30 (1 %) |
| PHQ-9 Total Score | ||
| Mean (SD) | 3.3 (± 4.4) | 3.2 (± 4.2) |
| PHQ-9 Total Score ≥ 10 | ||
| No | 4861 (90 %) | 4719 (92 %) |
| Yes | 511 (10 %) | 415 (8 %) |
| Gender | ||
| Male | 2585 (48 %) | 2508 (49 %) |
| Female | 2787 (52 %) | 2626 (51 %) |
| Age | ||
| Mean (SD) | 47 (± 18) | 48 (± 18) |
| Race/Ethnicity | ||
| Hispanic | 2315 (43 %) | 1590 (31 %) |
| Non-Hispanic White | 1234 (23 %) | 1710 (33 %) |
| Non-Hispanic Black | 1087 (20 %) | 1096 (21 %) |
| Other Race | 736 (14 %) | 738 (14 %) |
| Number of Comorbidities | ||
| Mean (SD) | 0.77 (± 1.2) | 0.82 (± 1.2) |
Mean (standard deviation) for continuous measures and number of subjects in each category (percentage of subjects in each category) for discrete measures are reported.
We assumed that there are no time period differences in the data for the study populations from NHANES 2013–2014 and 2015–2016. This assumption appears reasonable based on the relatively similar distribution of the data across the two samples in Table 1. In Supplementary Figure 1 a plot of the empirical cumulative distribution functions of the PHQ-9 total score from 2013–2014 and 2015–2016 are visually similar.
2.2. Qualitative Analysis using the RDoC framework
The Research Domain Criteria (RDoC) framework was presented by the National Institute of Mental Health (NIMH) in a commentary published in the American Journal of Psychiatry by Insel et al. (2010). The RDoC is an alternate framework to the Diagnostic and Statistical Manual (DSM) to conceptually organize and potentially direct biological research on mental disorders (Cuthbert and Insel, 2013; Insel et al., 2010).
The RDoC framework is structured around six major domains. These domains include Negative Valence Systems, driving reactions to aversive stimuli; Positive Valence Systems, driving reactions to positive stimuli; Cognitive Systems, including various mental processes; Social Processes, responsible for interpersonal behavior and cognition; and Arousal and Regulatory Systems, involved in context-based and homeostatic regulation or neural systems (Carcone and Ruocco, 2017). Additionally, Sensorimotor Systems are responsible for the control and execution of motor behaviors, and their refinement during learning and development. There is interaction and overlap among each domain.
Research on the systems and processes included in the RDoC framework is organized around a dimensional approach incorporating and integrating the following levels or units of analysis: genes, molecules, cells, circuits, physiology, behavior and self-report. The aim of the NIMH’s RDoC initiative has been to progress further understanding of these domains such that a new diagnostic nosology can be developed (Carcone and Ruocco, 2017).
For our analysis, a panel of eight experts in psychiatry, internal medicine, nursing, biostatistics, epidemiology and clinical research convened to qualitatively describe items of the PHQ-9 within the RDoC matrix. We used a deductive coding method. A blank 9 × 6 contingency table was created with the nine PHQ-9 items to be classified into the six RDoC domains. Each of the eight experts independently mapped each of the PHQ-9 items onto the six RDoC domains described above and filled out the contingency table.
In this mapping process, given nine items and six domains, multiple items were classified by our expert consensus group in multiple domains when deemed to be theoretically appropriate. Total number of votes for each item for each domain were tallied on a summarizing contingency table. Agreement was reached on some, but not all domains and items. However, items with three or more votes within a domain were highlighted, while items with majority vote (five or more) were emphasized. Upon group discussion, it was then agreed upon that the voting had appropriately classified the items into the domains reflecting a suitable amount of expert to expert variation in agreement as well. The coded contingency table is included as Supplementary Table 1.
2.3. Statistical Analysis
We used exploratory factor analysis (EFA) first to help determine the factor structure among the nine PHQ-9 items using NHANES 2013–2014. Due to a floor effect in the items of the PHQ-9 and to account for the ordinal responses of the items included in the model (see Table 1) we used a robust weighted least square adjusted for mean and variance (WLSMV) in performing EFA and all subsequent analyses involving a measurement model. We used a geomin rotation in performing EFA, which is an oblique type of rotation (Muthén and Muthén, 2012).
In order to help determine the optimal number of factors and factor structure in the PHQ-9, we examined the eigenvalues, which represent the variance accounted for by each underlying factor, in a scree plot as well as model fit criterion. The number of eigenvalues ≥ 1 represent unique factors according to Kaiser’s rule (Costello and Osborne, 2005). A chi-square difference test between adjacent factor solutions was also performed to help determine the number of factors. Model fit for each solution was evaluated through several fit indices. Namely, we examined chi-square test statistic, comparative fit index (CFI), Tucker Lewis fit index (TLI), root mean square error of approximation (RMSEA) and standardized root mean square residual (SRMR) values (Bentler, 1990; Browne et al., 1993; Hu and Bentler, 1998; Tucker and Lewis, 1973). A nonsignificant chi-square test statistic, CFI and TLI ≥ 0.95, RMSEA ≤ 0.05 and SRMR ≤ 0.08 represent an excellent fitting model.
After determining the factor structure of the PHQ-9, we utilized confirmatory factor analysis (CFA) for evaluating reliability and internal validity using NHANES 2015–2016. The overall internal consistency of items in the factor structure was tested by using Cronbach’s alpha (Cronbach, 1951) and omega (Bentler, 1972; Bentler, 2009; Green and Yang, 2009) coefficients. Values ≥ 0.70 for both alpha and omega coefficients indicate acceptable internal consistency.
Average variance extracted (AVE) was used to measure convergent validity. Good convergent validity establishes that a factor is well explained by its observed variables. An acceptable threshold for average variance extracted is 0.50.
We expected naturally occurring correlation between multiple factors describing domains of depression. However, as a cutoff, if the correlations among any two factors was ≥ 0.85 we deemed that the factors showed poor discriminant validity (Brown, 2014; Gunzler and Morris, 2015). Discriminant validity evaluates whether domains that are not supposed to be correlated are actually uncorrelated.
We evaluated the relationship between the factors and sex (reference male), age, race/ethnicity and number of comorbidities using multiple indicator multiple cause (MIMIC) analysis. A MIMIC model is a measurement model with covariates. Thus, we used MIMIC analysis to evaluate the influence of covariates (direct effect) on the factor means of the depressive phenotypes. Race/ethnicity was coded as a series of dummy variables for Non-Hispanic White, Hispanic, Non-Hispanic Black and Other Race. We omitted the dummy variable for Non-Hispanic White in our analysis due to linear dependency, choosing it as the reference category.
We defined α = 0.05 for our level of significance in all statistical tests. All statistical tests were two-tailed. R program in the R studio environment was used for data cleaning and calculating descriptive statistics and reliability and validity assessment (Venables et al., 2002). EFA, CFA and MIMIC analysis, were conducted using Mplus Version 8.2 (Múthen and Múthen, 2019).
2.4. Procedure for describing depressive phenotypes using qualitative and quantitative analysis
The results of the qualitative analysis were to be used in conjunction with quantitative analysis for describing depressive phenotypes. That is, we gave equal weight to both qualitative criterion via the mapping of particular items to particular domains using the RDoC matrix and quantitative criterion via exploratory factor analytic methods.
In sequence, we first had our panel of experts map the PHQ-9 items on the RDoC domains. We next performed EFA.
In evaluating our EFA output, we looked for a substantive meaning in the identified factors using our qualitative results. We were willing to consider solutions beyond just two factors for the PHQ-9 (affective/cognitive and somatic) given our research framework consisted of six domains and interaction and overlap among these domains. The PHQ-9 item factor loadings can be interpreted as the shared variance between an item and the factor which the item loads on. All primary factor loadings should have a standardized estimate ≥ 0.40 (Gunzler and Morris, 2015).
After we determined the factor structure of the PHQ-9, as outlined in our statistical analysis plan we evaluated measurement properties (reliability and internal validity). We then evaluated the relationships between the depressive phenotypes and covariates of interest. In the discussion we make recommendations about depressive phenotypes in the PHQ-9 based on our analyses.
3. Results
3.1. Determining the factor structure of the PHQ-9
Overall, our qualitative and exploratory factor analysis provided evidence for a four factor solution. There was only a single eigenvalue ≥ 1 with a value of 5.62 and the scree plot visually confirmed a single factor (see Figure 2).
Figure 2.
Scree plot from the exploratory factor analysis of the nine PHQ-9 items using NHANES survey data, 2013–2014 (N = 5397).
However, model fit criterion supported a multifactor solution (see Table 2).
Table 2.
Model fit statistics and indices from exploratory factor analysis for PHQ-9 for NHANES Survey Data, 2013–2014 (N = 5397)
| Number of Factors | |||||
|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | 5 | |
| Chi-square | 642.89 (27) p < 0.001 |
185.50 (19) p < 0.001 |
83.46 (12) p < 0.001 |
8.13 (6) p =0.229 |
0.694 (1) p = 0.405 |
| Chi-square difference test* | 381.43 (8) p<0.001 |
96.55 (7) p<0.001 |
70.66 (6) p<0.001 |
7.40 (5) P=0.192 |
|
| CFI | 0.977 | 0.994 | 0.997 | 1.000 | 1.000 |
| TLI | 0.970 | 0.988 | 0.992 | 1.000 | 1.000 |
| RMSEA (90% CI) | 0.065 (0.061, 0.069) | 0.040 (0.035, 0.046) | 0.033 (0.027, 0.040) | 0.008 (0.000, 0.021) | 0.000 (0.000, 0.034) |
| SRMR | 0.053 | 0.029 | 0.021 | 0.007 | 0.002 |
Chi-square difference test for the evaluating for a difference between the current solution with k factors and the previous solution with k-1 factors.
The one factor solution had a reasonably acceptable model fit, with a slightly high RMSEA value (above the guideline threshold of 0.05). We reached criterion for acceptable fit for two factors, whereas the chi-square test statistic was still significant. We reached criterion for an excellent fit for four factors. Model fit criterion in Chi-square, RMSEA and SRMR also still improved meaningfully from the three factor solution to four factor solution. There were only trivial model fit differences between the four factor and five factor models.
The chi-square difference test for comparing the three factor to four factor solution was significant (chi-squared test statistic = 70.66 (6) p < 0.001) and four factor to five factor (chi-squared test statistic = 7.40 (5) p = 0.192) was not significant; this provided quantitative evidence for the four factor solution.
In table 3 we report the factor loadings for the one, two and four factor solutions using EFA. We note the breakdown of items with a strong loading on each of the four factors in bold (≥ 0.40) from Table 3; although we still consider loadings below this threshold for potential cross-loading.
Table 3.
One, two and four factor solutions (factor loadings, values greater than 0.40 in bold) for exploratory factor analysis using the PHQ-9 and the NHANES survey data 2013–2014 (N = 5397)
| One Factor | Two Factors | Four Factor | |||||
|---|---|---|---|---|---|---|---|
| 1 | 1 | 2 | 1 | 2 | 3 | 4 | |
| Item | Depression | Cognitive/Affective | Somatic | Negative Valence Systems and Externalizing | Negative Valence Systems and Internalizing | Arousal and Regulatory Systems | Cognitive and Sensorimotor Systems |
| Anhedonia | 0.770 | 0.395 | 0.426 | 0.835 | 0.011 | 0.021 | 0.084 |
| Depression | 0.856 | 0.764 | 0.150 | 0.295 | 0.643 | 0.033 | −0.010 |
| Sleep | 0.696 | −0.122 | 0.858 | −0.064 | −0.055 | 0.733 | 0.154 |
| Fatigue | 0.757 | 0.017 | 0.789 | 0.087 | 0.073 | 0.757 | −0.062 |
| Appetite | 0.706 | 0.187 | 0.562 | 0.072 | 0.167 | 0.425 | 0.123 |
| Guilt | 0.825 | 0.869 | 0.006 | 0.050 | 0.856 | −0.017 | 0.023 |
| Psychomotor | 0.761 | 0.401 | 0.410 | 0.091 | 0.166 | 0.075 | 0.527 |
| Concentration | 0.762 | 0.362 | 0.449 | 0.006 | 0.013 | −0.005 | 0.858 |
| Self-Harm | 0.767 | 0.873 | −0.070 | −0.107 | 0.876 | 0.014 | 0.032 |
The one factor solution was straightforward to interpret as a unidimensional depression construct. All factor loadings were high in magnitude and ranged from 0.706 to 0.856.
The two factor could be interpreted using a more classical understanding regarding depression theory and help in the description of cognitive/affective and somatic dimensions. There were several cross-loaded items (items for anhedonia, concentration and psychomotor) for the two factor solution. The cross-loading of anhedonia on a somatic dimension is not theoretically reasonable and would not be retained in any successive measurement model. The three factor model (not listed) could be similarly used to help describe affective, cognitive and somatic dimensions. However, the solution did not have any practical quantitative or interpretative advantage over the two factor model.
The RDoC matrix was useful to interpret the four factor solution, given four dimensions and expected overlap and interaction among domains. Item 2 for depression substantially loads onto the second factor. However, the factor loading of item 2 for depression on the first factor was still significant at 0.295 (even though below the threshold of 0.40). Using qualitative reasoning, there is strong evidence that this item should cross-load on the first factor.
Item 1 for anhedonia and item 2 for depression are used together as the PHQ-2 in practice as a pre-screener. One application of the PHQ-2 is that patients with an affirmative response to either question are prompted to complete the PHQ-9. Further, in applying the RDoC matrix, the contribution of these two items together in a depressive phenotype is discussed in the next section.
3.2. Describing the intermediate phenotypes of depression using the RDoC matrix
We linked the empirically derived factor structures in Table 3 to our RDoC matrix voting for evaluating interpretability of the four factor solution within the research framework. We described the four factors using the RDoC framework (see Figure 3): Negative Valence Systems and Externalizing (items 1. anhedonia and 2. depression), Negative Valence Systems and Internalizing (items 2. depression, 6. guilt, and 9. self-harm), Arousal and Regulatory Systems (items 3. sleep, 4. fatigue, and 5. appetite) and Cognitive and Sensorimotor Systems (items 7. concentration and 8. psychomotor).
Figure 3. Four factor solution for the PHQ-9 using NHANES survey data 2015–2016 (N = 5164) with standardized estimates of factor loadings and factor intercorrelations using confirmatory factor analysis.
All estimates in the figure (standardize factor loadings, factor intercorrelations) are statistically significant p < 0.001. Model fit: Chi-square 56.33 (20) p < 0.001, CFI = 0.998, TLI = 0.997, RMSEA (90% CI)=0.019 (0.013, 0.025), SRMR = 0.011.
These factor names reflect overlap and interaction amongst the RDoC matrix domains within the PHQ-9 items. For example, items 7. concentration and 8. psychomotor were classified by our expert consensus panel in both domains for Cognitive Systems and Sensorimotor Systems and we named the construct accordingly, Cognitive and Sensorimotor Systems. Our expert consensus panel classified items 1. anhedonia, 2. depression, 6. guilt and 9. self-harm in the domain for Negative Valence Systems. Thus, given that the items load onto two different factors in our model, we further used descriptors “Externalizing” and “Internalizing” to help distinguish between these two constructs. Interestingly, these two constructs were the most distinctive in the model with the lowest factor intercorrelation of any two factors (see Figure 3).
In more detail, higher endorsement on anhedonia and depression items describe negative reactions to aversive stimuli. These two items also describe symptoms relevant to the relationship of an individual to their environment (i.e. externalizing symptoms). The first phenotypic trait (Negative Valence Systems and Externalizing) was aptly named to help distinguish it from the second phenotypic trait (Negative Valence Systems and Internalizing). Externalizing symptoms are evident in interpersonal behavior; anhedonia and depression items were also mapped into the domain for Social Processes by our expert panel. Thus our first phenotypic trait describes items of the PHQ-9 that overlap among the RDoC domains of Negative Valence Systems and Social Processes.
Higher endorsement on depression, guilt and self-harm items also describe negative reactions to aversive stimuli. These three items describe symptoms relevant to the relationship of an individual to their own emotional state (i.e. internalizing symptoms). One may note that the cross-loaded item for depression contributes to both these constructs and the externalizing and internalizing of symptoms.
Sleep, fatigue and appetite items describe somatic symptoms that are inherent in arousal and regulation. Finally, concentration and psychomotor items describe mental process symptoms and motor behavior.
In Figure 3 we reported the results of the CFA for the four factor solution allowing for the cross-loading of item 2 for depression. The four factor CFA model had an acceptable to excellent fit (see Figure 3 legend). We do note that there was still a low factor loading (0.27) of the cross-loaded item for depression on the factor for Negative Valence Systems and Externalizing. However, again given our goal of identifying depressive phenotypes (which involves investigating which items influence particular phenotypes) and not scale construction, dropping the cross-loading would theoretically weaken the analyses (Asparouhov et al., 2015).
3.3. Reliability and validity of the four factor solution
We first assessed reliability and validity of the unidimensional measurement model. This model exhibited strong reliability and acceptable convergent validity (see Table 4).
Table 4.
Reliability and validity of confirmatory factor analysis results for the one and four factor models using NHANES 2015–2016 (N =5136)
| One factor | Four factors | |||||
|---|---|---|---|---|---|---|
| Depression | Negative Valence Systems and Externalizing | Negative Valence Systems and Internalizing | Arousal and Regulatory Systems | Cognitive and Sensorimotor Systems | ||
| Alpha | 0.920 | 0.811 | 0.906 | 0.786 | 0.769 | |
| Omega | 0.849 | 0.472 | 0.753 | 0.649 | 0.530 | |
| AVE | 0.577 | NA | NA | 0.560 | 0.626 | |
Alpha = Cronbach’s alpha coefficient; Omega = coefficient omega (Bentler 1972, 2009); AVE = average variance extracted. AVE is the property of items and is not available for factors with cross-loaded items (Negative Valence Systems and Externalizing and Negative Valence Systems and Internalizing).
In the four factor solution, the coefficient alpha of each factor reached the threshold for acceptable internal consistency (Table 4). Coefficient omega reached the threshold for acceptable internal consistency only for Negative Valence Systems and Internalizing (Table 4).
The factors for Arousal and Regulatory Systems and Cognitive and Sensorimotor Systems both showed acceptable convergent validity (Table 4). Factor intercorrelations higher than our threshold for poor discriminant validity (≥ 0.85) was found between Cognitive and Sensorimotor Systems and Arousal and Regulatory Systems (see Figure 3).
3.4. Evaluating for sociodemographic differences in the intermediate phenotypes of depression
The MIMIC model we used for evaluating the relationships between the depressive phenotypes and covariates of interest is displayed in Figure 4. The model fit was acceptable to excellent (see Legend for Figure 4).
Figure 4. Multiple indicator multiple cause (MIMIC) model for the relationship between depressive phenotypes and sex, race/ethnicity, age and number of comorbidities.
For visual ease pairwise factor inter-correlations between the depressive phenotypes are omitted from this figure but included in the model. The measurement part of the MIMIC model for the four depressive phenotypes was displayed in Figure 3. Model fit using NHANES survey data 2015–2016 (N = 5164): Chi-square 177.20 (50) p < 0.001, CFI = 0.994, TLI = 0.990, RMSEA (90% CI)=0.022 (0.019, 0.026), SRMR = 0.019.
The magnitude of the factor loadings and factor intercorrelations in the measurement model were relatively similar as in the CFA (see Table 5). In the structural model we regressed the four depressive phenotypes on the covariates. The means for all four depressive phenotypes significantly differed depending on the number of comorbidities, age and level of sex and race/ethnicity (Table 5).
Table 5.
Multiple indicator multiple cause analysis results using NHANES survey data, 2015–2016 (N = 5164)
| Measurement Model | ||||
|---|---|---|---|---|
| Depressive Phenotype | Item | Standardized Estimate | Standard Error | p |
| Negative Valence Systems and Externalizing | Anhedonia | 0.93 | 0.05 | <0.001 |
| Depression | 0.29 | 0.05 | <0.001 | |
| Negative Valence Systems and Internalizing | Depression | 0.66 | 0.05 | <0.001 |
| Guilt | 0.92 | 0.01 | <0.001 | |
| Self-Harm | 0.84 | 0.02 | <0.001 | |
| Arousal and Regulatory Systems | Sleep | 0.74 | 0.01 | <0.001 |
| Fatigue | 0.79 | 0.01 | <0.001 | |
| Appetite | 0.72 | 0.01 | <0.001 | |
| Cognitive and Sensorimotor Systems | Concentration | 0.81 | 0.01 | <0.001 |
| Psychomotor | 0.77 | 0.02 | <0.001 | |
| Factor Intercorrelations | ||||
| Negative Valence Systems and Externalizing | Negative Valence Systems and Internalizing | 0.66 | 0.04 | <0.001 |
| Negative Valence Systems and Externalizing | Arousal and Regulatory Systems | 0.73 | 0.04 | <0.001 |
| Negative Valence Systems and Externalizing | Cognitive and Sensorimotor Systems | 0.71 | 0.02 | <0.001 |
| Negative Valence Systems and Internalizing | Cognitive and Sensorimotor Systems | 0.76 | 0.04 | <0.001 |
| Negative Valence Systems and Internalizing | Arousal and Regulatory Systems | 0.82 | 0.02 | <0.001 |
| Arousal and Regulatory Systems | Cognitive and Sensorimotor Systems | 0.84 | 0.02 | <0.001 |
| Structural Model | ||||
| Depressive Phenotype | Covariate | Standardized Estimate | Standard Error | P |
| Negative Valence Systems and Externalizing | Number of Comorbidities | 0.26 | 0.02 | <0.001 |
| Sex | 0.15 | 0.04 | <0.001 | |
| Age | −0.11 | 0.02 | <0.001 | |
| Hispanic | 0.13 | 0.05 | 0.009 | |
| Non-Hispanic Black | 0.14 | 0.05 | 0.008 | |
| Other Race | 0.00 | 0.06 | 0.994 | |
| Negative Valence Systems and Internalizing | Number of Comorbidities | 0.24 | 0.02 | <0.001 |
| Sex | 0.09 | 0.04 | 0.021 | |
| Age | −0.17 | 0.02 | <0.001 | |
| Hispanic | 0.00 | 0.05 | 0.934 | |
| Non-Hispanic Black | −0.17 | 0.06 | 0.003 | |
| Other Race | −0.06 | 0.07 | 0.406 | |
| Arousal and Regulatory Systems | Number of Comorbidities | 0.33 | 0.02 | <0.001 |
| Sex | 0.31 | 0.03 | <0.001 | |
| Age | −0.19 | 0.02 | <0.001 | |
| Hispanic | −0.06 | 0.04 | 0.164 | |
| Non-Hispanic Black | −0.14 | 0.05 | 0.003 | |
| Other Race | −0.23 | 0.05 | <0.001 | |
| Cognitive and Sensorimotor Systems | Number of Comorbidities | 0.27 | 0.02 | <0.001 |
| Sex | 0.10 | 0.05 | 0.034 | |
| Age | −0.11 | 0.03 | <0.001 | |
| Hispanic | 0.15 | 0.06 | 0.007 | |
| Non-Hispanic Black | −0.01 | 0.06 | 0.86 | |
| Other Race | −0.13 | 0.07 | 0.087 | |
We used the WLSMV option in MPlus. We used the STDYX option to report the standardized estimates for the measurement model and for the structural model involving the continuous independent variables (number of comorbidities and age). We used the STDY option to report the standardized estimates involving the binary independent variables (sex, Hispanic, Non-Hispanic Black and Other Race).
The number of comorbidities positively influenced each of the four depressive phenotypes. Therefore, as the number of comorbidities increases the factor mean increases for each of the depressive phenotypes. Age negatively influenced each of the depressive phenotypes. Therefore, as age increases the factor mean decreases for each of the depressive phenotypes. The effect sizes for the standardized model estimates of the regression paths were moderate for number of comorbidities and small for age.
Being a female (compared to male) lead to a small increase in the factor means for Negative Valence Systems and Externalizing, Negative Valence Systems and Internalizing and Cognitive and Sensorimotor Systems. Being female (compared to male) lead to a moderate increase in the factor mean for Arousal and Sensorimotor Systems.
Being Hispanic (compared to Non-Hispanic White) lead to a small increase in the factor means for Negative Valence Systems and Externalizing and Cognitive and Sensorimotor Systems, but had no significant relationship with the other two depressive phenotypes. Being Non-Hispanic Black (compared to Non-Hispanic White) lead to a small increase in the factor mean for Negative Valence Systems and Externalizing and a small decrease in the factor means for Negative Valence Systems and Internalizing and Arousal and Sensorimotor Systems, but had no significant relationship with Cognitive and Sensorimotor Systems. Finally, being in an Other Race group (compared to Non-Hispanic White) lead to a moderate decrease in the factor mean for Arousal and Sensorimotor Systems but had no significant relationship with the other three depressive phenotypes.
4. Discussion
Major depressive disorder is an extremely heterogenous disorder with multiple phenotypic traits. The PHQ-9 is commonly implemented in clinical care as a depression screening and symptom monitoring tool. Previous psychometric studies have commonly utilized either a single summary score or two factors representing somatic and cognitive/affective domains within the PHQ-9 (Dum et al., 2008; Gunzler et al., 2015; Huang et al., 2006; Kalpakjian et al., 2009; Patel et al.; Titov et al., 2011). We used contemporary conceptualization of symptom/behavioral domains and factor analytic methods to identify four intermediate depressive phenotypes in the PHQ-9 describing Negative Valence Systems and Externalizing, Negative Valence Systems and Internalizing, Arousal and Regulatory Systems and Cognitive and Sensorimotor Systems. Our study had some important strengths in using (1) a large, nationally representative sample with recent, available PHQ-9 data for generalizability (2) the RDoC framework for multidimensional evaluation of depressive symptoms and (3) multiple causal models to rigorously evaluate psychometric properties and sociodemographic differences in the proposed depressive phenotypes.
Prior reports by Ulbricht et al. (2018), Rathbun et al. (2019) and Sun et al. (Rathbun et al., 2019; Sun et al.; Ulbricht et al., 2018) used unsupervised learning approaches to identify depression subtypes based on observed symptom patterns. These reports found distinct clusters of people varying in severity (i.e. mild, moderate and severe) and symptomology (i.e. psychomotor, catatonic, melancholic, anhedonic, physio-somatic). Thus, there has been a precedence to understand the complexities of depressive symptoms beyond affective and somatic/cognitive domains.
In particular for the PHQ-9, the cross-sectional study by Sun and colleagues (Sun et al.) evaluated a sample of 2783 women at different prenatal and postnatal periods and assessed depressive symptoms using latent class analysis. Five latent subtypes were found: “no symptoms,” “mild physio-somatic symptoms,” “severe physio-somatic symptoms and moderate anhedonia,” “moderate-to-severe symptoms,” and “severe symptoms.” Possibly indicative of physiological differences, postpartum women were more likely to belong to the severe depressive symptoms group, while pregnant women were likely to report severe physio-somatic symptoms.
Our study evaluated multidimensional aspects of depression similarly to these prior studies on subtypes, but instead assessed phenotypic traits (i.e. domains) in the PHQ-9. A distinct contrast is that our goal relied on identifying the underlying dimensions of items describing symptoms of depression (variable-centered analyses) as opposed to unsupervised clustering of people (person-centered analyses). Our approach required both an appropriate research framework and factor analytic techniques in tandem. In our analyses the RDoC matrix framework reflected contemporary knowledge about major systems of emotion, cognition, motivation, and social behavior beyond the typical affective, somatic and cognitive domain classification of the PHQ-9.
Our phenotypes overlap at least partially with depressive sub-types that have been described in the previous literature (Rathbun et al., 2019; Sun et al.; Ulbricht et al., 2018). For example, Cognitive and Sensorimotor Systems overlap with described psychomotor and catatonic groupings, Negative Valence Systems and Externalizing and Negative Valence Systems and Internalizing overlap with melancholic and anhedonic groupings and Arousal and Regulatory Systems overlap with physio-somatic groupings in this prior literature.
The four traits in our study were found to have high intercorrelations and it is likely that multiple phenotypes will present themselves within an individual screening positive for depressive symptoms. These high intercorrelations along with the good measurement properties of the unidimensional depression score indicate screening and monitoring for depression in a general study population using a single depression score is likely useful in most circumstances.
However, an implication of our study is still that use of the PHQ-9 in monitoring depressive screening should consider these four, potentially overlapping phenotypes. Individual subjects may present with some subset (i.e. one, two or three) of the four phenotypes (see Appendix) and treatment approaches can be designed to appropriately target the presence of one or more of the four identified phenotypic traits in different individuals.
We did find age, sex, race/ethnicity and number of comorbidities significantly influenced the depressive phenotypes. Namely, having more comorbidities, being younger and being female lead to higher endorsement of each of the four depressive phenotypes. Being a member of different race/ethnicity groups (Hispanic, Non-Hispanic Black, Other Race, Non-Hispanic White), however, lead to differing relationships across the depressive phenotypes. These sociodemographic differences should be studied further.
Several limitations must be considered in interpreting our study. The PHQ-9 is a screening instrument, and our study is unable to determine depression diagnoses because no clinical assessments or diagnostic interviews were performed. Respondents may over- or under-report symptom frequency on the PHQ-9 leading to over- or under-estimation of depressive symptoms in our study. Furthermore, the cross-sectional design of the NHANES survey prevents us from examining changes over time in the same participants.
RDoC is still undergoing modifications as research evolves. Other measures of depression might lead to different results in describing intermediate phenotypes. For example, the Center for Epidemiological Studies Depression Scale (CES-D)-10 (Radloff, 1977) includes positively worded items describing hope about the future and happiness. Thus, a phenotype for positive valence systems may be present in the CES-D-10 that we did not find within the PHQ-9.
Measurement (factorial) and structural invariance testing using multigroup analyses could be used for a more in depth evaluation of the factor structure across groups. The MIMIC analysis had some advantages over multigroup analyses in that it allowed us to adjust simultaneously for age, sex, race/ethnicity and number of comorbidities and to treat age and number of comorbidities more naturally as continuous measures.
Our complex model using the RDoC as guidelines included cross-loading and certain factors had high intercorrelations, thus reliability and validity assessment produced mixed results for the factor structure (with less than ideal omega coefficients for internal consistency for certain factors and poor discriminant validity between certain factors). However, our goal was to describe depressive phenotypes using both qualitative and quantitative methods in tandem and not to perform the ideal scale development. Given our goal, our model showed a good fit and had relatively good psychometric properties. Future psychometric studies of mental health measures should continue to be guided by the RDoC matrix.
5. Conclusion
Self-reported depressive symptom severity is complex and consists of multiple phenotypes. Describing these phenotypes has potential to facilitate developing personalized medicine approaches for varying patterns of depressive symptoms. For example, we may identify individuals with a high factor score (individualized score on a factor) on the Arousal and Regulatory System phenotype, but low factor scores on the other three phenotypes. We may find that depressed patients with this Arousal and Regulatory System phenotype respond differently to certain medications than depressed patients with other phenotypes or combinations of phenotypes. Future work is needed to identify case studies and general indications of the phenotypes we proposed. Reproducing these results in other general study populations and within particular clinical and sociodemographic study populations is a logical next step.
Our study suggests the usefulness of the RDoC framework in evaluating the phenotypic traits of depression. Further applications and refinement of the RDoC framework are warranted.
Supplementary Material
Figure 1.
Procedure for using qualitative and quantiative analysis for describing depressive phenotypes within the PHQ-9
Highlights.
Four depressive phenotypes found in the patient health questionnaire (PHQ)-9 in a large, nationally representative U.S sample (Negative Valence Systems and Externalizing, Negative Valence Systems and Internalizing, Arousal and Regulatory Systems and Cognitive and Sensorimotor Systems).
Research Domain Criteria (RDoC) research framework was useful in evaluating the phenotypic traits of depression.
Differences in the means of the phenotypic traits were found by age, race/ethnicity, sex and number of comorbidities.
High correlation between the four depressive phenotypes indicates screening and monitoring for depression study population using a single depression score is likely useful in most circumstances.
Describing these phenotypes has potential to provide a step towards developing personalized medicine approaches for varying patterns of depressive symptoms.
Funding Statement:
This work was supported by the National Institutes of Health/National Institute of Diabetes and Digestive and Kidney Diseases grant number R01DK112905.
Appendix
Factor scores are individualized scores on a latent variable. Factor scores can be estimated by applying Bayes’ theorem to calculate the posterior distribution of the factor scores given the observed indicators. Using this approach, factor scores are normally distributed with zero mean within a study population.
We extracted factor scores for each of the four factors in our study using the MIMIC model that adjusted for sex, race/ethnicity, age and number of comorbidities. We then categorized subjects as either less than (lower) or greater than or equal to (higher) the median factor score for each factor. In Table A.1 we observe 16 distinct patterns based on these thresholds across the four depressive phenotypic traits.
The majority of the subjects have factor scores either less than the median across all factors (pattern 1) or greater than or equal to the median across all factors (pattern 16), given the high factor intercorrelations (see Figure 2). However, a nontrivial number of subjects (787 total or 15.2% of subjects) have different patterns (patterns 2–15). For such patterns, multiple phenotypes may be more informative than a single depression score. For example, for subjects in pattern 3, clinicians may want to focus on monitoring an Arousal and Regulatory Systems phenotype (i.e. sleep, fatigue, appetite).
Table A.1.
Patterns based on the median factor score (lower than, higher than or equal to) across depressive phenotypes using the PHQ-9 and the NHANES survey data 2015–2016 (N = 5136).
| Pattern | Negative Valence Systems and Externalizing | Negative Valence Systems and Internalizing | Arousal and Regulatory Systems | Cognitive and Sensorimotor Systems | Frequency |
|---|---|---|---|---|---|
| 1 | Lower | Lower | Lower | Lower | 2222 |
| 2 | Lower | Lower | Lower | Higher | 15 |
| 3 | Lower | Lower | Higher | Lower | 67 |
| 4 | Lower | Lower | Higher | Higher | 25 |
| 5 | Lower | Higher | Lower | Lower | 16 |
| 6 | Lower | Higher | Lower | Higher | 59 |
| 7 | Lower | Higher | Higher | Lower | 113 |
| 8 | Lower | Higher | Higher | Higher | 64 |
| 9 | Higher | Lower | Lower | Lower | 78 |
| 10 | Higher | Lower | Lower | Higher | 32 |
| 11 | Higher | Lower | Higher | Lower | 49 |
| 12 | Higher | Lower | Higher | Higher | 94 |
| 13 | Higher | Higher | Lower | Lower | 21 |
| 14 | Higher | Higher | Lower | Higher | 138 |
| 15 | Higher | Higher | Higher | Lower | 16 |
| 16 | Higher | Higher | Higher | Higher | 2155 |
Footnotes
Conflicts of Interest: Author DG reports a book royalty agreement with Taylor Francis Publishing. Author MS has received grant support from Otsuka, Alkermes, Janssen, the National Institutes of Health, the Centers for Disease Control and Prevention and the International Society for Bipolar Disorders. She has served as a consultant to Bracket, Otsuka, Janssen, Alkermes, Neurocrine and Health Analytics and receives royalties from Springer Press, Johns Hopkins University Press, Oxford Press, and UpToDate. All other authors declare no conflicts of interest with the research or writing of this paper.
Contributor Information
Douglas Gunzler, Center for Health Care Research & Policy, The MetroHealth System, 2500 MetroHealth Drive, Cleveland, Ohio 44109.
Ashwini R. Sehgal, Center for Reducing Health Disparities, The MetroHealth System, Case Western Reserve University, Cleveland, OH
Kelley Kauffman, Center for Reducing Health Disparities, The MetroHealth System, Case Western Reserve University, Cleveland, Ohio.
Christine Horvat Davey, Case Western Reserve University, Cleveland, OH.
Jacqueline Dolata, Center for Reducing Health Disparities, The MetroHealth System, Case Western Reserve University, Cleveland, Ohio.
Maria Figueroa, Center for Reducing Health Disparities, The MetroHealth System, Case Western Reserve University, Cleveland, Ohio.
Anne Huml, Center for Reducing Health Disparities, The MetroHealth System, Case Western Reserve University, Cleveland, Ohio.
Julie Pencak, Center for Reducing Health Disparities, The MetroHealth System, Case Western Reserve University, Cleveland, Ohio.
Martha Sajatovic, Department of Psychiatry and of Neurology, Case Western Reserve University School of Medicine, Neurological and Behavioral Outcomes Center, University Hospitals Cleveland Medical Center, Cleveland, Ohio.
References
- Asparouhov T, Muthén B, Morin AJ, 2015. Bayesian structural equation modeling with cross-loadings and residual covariances: Comments on Stromeyer et al. Sage Publications Sage CA: Los Angeles, CA. [Google Scholar]
- Beck AT, Steer RA, Brown GK, 1996. Manual for the beck depression inventory-II. San Antonio, TX: Psychological Corporation 1, 82. [Google Scholar]
- Bentler P, 1972. A lower-bound method for the dimension-free measurement of internal consistency. Social science research 1 (4), 343–357. [Google Scholar]
- Bentler PM, 1990. Comparative fit indexes in structural models. Psychological bulletin 107 (2), 238. [DOI] [PubMed] [Google Scholar]
- Bentler PM, 2009. Alpha, dimension-free, and model-based internal consistency reliability. Psychometrika 74 (1), 137. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Brody AL, Saxena S, Mandelkern MA, Fairbanks LA, Ho ML, Baxter LR Jr, 2001. Brain metabolic changes associated with symptom factor improvement in major depressive disorder. Biological psychiatry 50 (3), 171–178. [DOI] [PubMed] [Google Scholar]
- Brody DJ, Pratt LA, Hughes JP, 2018. Prevalence of Depression Among Adults Aged 20 and Over: United States, 2013–2016. NCHS data brief(303), 1–8. [PubMed] [Google Scholar]
- Brown TA, 2014. Confirmatory factor analysis for applied research. Guilford Publications. [Google Scholar]
- Browne MW, Cudeck R, Bollen KA, Long JS, 1993. Alternative ways of assessing model fit. Sage Focus Editions 154, 136–136. [Google Scholar]
- Byrne BM, 2013. Structural equation modeling with Mplus: Basic concepts, applications, and programming. Routledge. [Google Scholar]
- Carcone D, Ruocco AC, 2017. Six Years of Research on the National Institute of Mental Health’s Research Domain Criteria (RDoC) Initiative: A Systematic Review. Frontiers in cellular neuroscience 11, 46–46. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Centers for Disease Control and Prevention, 2006. National Center for Health Statistics (NCHS). National Health and Nutrition Examination Survey Questionnaire (or Examination Protocol, or Laboratory Protocol). http://www.cdc.gov/nchs/nhanes.htm.
- Costello A, Osborne J, 2005. Best practices in exploratory factor analysis: four recommendations for getting the most from your analysis. Pract Assess Res Eval 2005; 10. URL http://pareonline.net/getvn.asp 10 (7). [Google Scholar]
- Cronbach LJ, 1951. Coefficient alpha and the internal structure of tests. Psychometrika 16 (3), 297–334. [Google Scholar]
- Cuthbert BN, Insel TR, 2013. Toward the future of psychiatric diagnosis: the seven pillars of RDoC. BMC medicine 11 (1), 126. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Dum M, Pickren J, Sobell LC, Sobell MB, 2008. Comparing the BDI-II and the PHQ-9 with outpatient substance abusers. Addictive behaviors 33 (2), 381–387. [DOI] [PubMed] [Google Scholar]
- Ferrando SJ, Samton J, Mor N, Nicora S, Findler M, Apatoff B, 2007. Patient health questionnaire-9 to screen for depression in outpatients with multiple sclerosis. International Journal of MS Care 9 (3), 99–103. [Google Scholar]
- Gottesman II, Gould TD, 2003. The endophenotype concept in psychiatry: etymology and strategic intentions. American Journal of Psychiatry 160 (4), 636–645. [DOI] [PubMed] [Google Scholar]
- Granillo MT, 2012. Structure and function of the Patient Health Questionnaire-9 among Latina and non-Latina White female college students. Society for Social Work and Research, US, pp. 80–93. [Google Scholar]
- Green SB, Yang Y, 2009. Reliability of summed item scores using structural equation modeling: An alternative to coefficient alpha. Psychometrika 74 (1), 155–167. [Google Scholar]
- Gunzler D, Perzynski A, Morris N, Bermel R, Lewis S, Miller D, 2015. Disentangling Multiple Sclerosis & Depression: An Adjusted Depression Screening Score for Patient-Centered Care. Journal of Behavioral Medicine 38 (2), 237–250. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Gunzler DD, Morris N, 2015. A tutorial on structural equation modeling for analysis of overlapping symptoms in co-occurring conditions using MPlus. Statistics in Medicine 34 (24), 3246–3280. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hallquist MN, Wiley JF, 2018. MplusAutomation: An R Package for Facilitating Large-Scale Latent Variable Analyses in M plus. Structural Equation Modeling: A Multidisciplinary Journal 25 (4), 621–638. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hasler G, Northoff G, 2011. Discovering imaging endophenotypes for major depression. Molecular psychiatry 16 (6), 604. [DOI] [PubMed] [Google Scholar]
- Hu L. t., Bentler PM, 1998. Fit indices in covariance structure modeling: Sensitivity to underparameterized model misspecification. Psychological methods 3 (4), 424. [Google Scholar]
- Huang FY, Chung H, Kroenke K, Delucchi KL, Spitzer RL, 2006. Using the Patient Health Questionnaire‐9 to measure depression among racially and ethnically diverse primary care patients. Journal of general internal medicine 21 (6), 547–552. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Insel T, Cuthbert B, Garvey M, Heinssen R, Pine DS, Quinn K, Sanislow C, Wang P, 2010. Research domain criteria (RDoC): toward a new classification framework for research on mental disorders. Am Psychiatric Assoc. [DOI] [PubMed] [Google Scholar]
- Kalpakjian CZ, Toussaint LL, Albright KJ, Bombardier CH, Krause JK, Tate DG, 2009. Patient Health Questionnaire-9 in spinal cord injury: an examination of factor structure as related to gender. The journal of spinal cord medicine 32 (2), 147–156. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kroenke K, Spitzer RL, Williams JB, 2001. The Phq‐9. Journal of general internal medicine 16 (9), 606–613. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Leventhal AM, Pettit JW, Lewinsohn PM, 2008. Characterizing major depression phenotypes by presence and type of psychomotor disturbance in adolescents and young adults. Depression and anxiety 25 (7), 575–592. [DOI] [PubMed] [Google Scholar]
- Levis B, Benedetti A, Thombs BD, 2019. Accuracy of Patient Health Questionnaire-9 (PHQ-9) for screening to detect major depression: individual participant data meta-analysis. BMJ 365, l1476. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Luk JW, King KM, McCarty CA, Stoep AV, McCauley E, 2016. Measurement Invariance Testing of a Three-Factor Model of Parental Warmth, Psychological Control, and Knowledge across European and Asian/Pacific Islander American Youth. Asian American journal of psychology 7 (2), 97–107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lux V, Kendler KS, 2010. Deconstructing major depression: a validation study of the DSM-IV symptomatic criteria. Psychological medicine 40 (10), 1679–1690. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Mitchell AJ, Yadegarfar M, Gill J, Stubbs B, 2016. Case finding and screening clinical utility of the Patient Health Questionnaire (PHQ-9 and PHQ-2) for depression in primary care: a diagnostic meta-analysis of 40 studies. BJPsych Open 2 (2), 127–138. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Muthén LK, Muthén BO, 2012. Mplus. The comprehensive modelling program for applied researchers: user’s guide 5.
- Patel JS, Oh Y, Rand KL, Wu W, Cyders MA, Kroenke K, Stewart JC, Measurement invariance of the patient health questionnaire-9 (PHQ-9) depression screener in U.S. adults across sex, race/ethnicity, and education level: NHANES 2005–2016. Depression and anxiety 0 (0). [DOI] [PMC free article] [PubMed] [Google Scholar]
- Pinto‐Meza A, Serrano‐Blanco A, Peñarrubia MT, Blanco E, Haro JM, 2005. Assessing Depression in Primary Care with the PHQ‐9: Can It Be Carried Out over the Telephone? Journal of general internal medicine 20 (8), 738–742. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Pizzagalli DA, Jahn AL, O’Shea JP, 2005. Toward an objective characterization of an anhedonic phenotype: a signal-detection approach. Biological psychiatry 57 (4), 319–327. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Putnick DL, Bornstein MH, 2016. Measurement Invariance Conventions and Reporting: The State of the Art and Future Directions for Psychological Research. Developmental review : DR 41, 71–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Radloff LS, 1977. The CES-D scale: A self-report depression scale for research in the general population. Applied psychological measurement 1 (3), 385–401. [Google Scholar]
- Rathbun A, Schuler M, Stuart AE, Shardell MD, Yau MS, Gallo JJ, Ryan AS, Hochberg MC, 2019. Depression Subtypes in Persons with or at Risk for Symptomatic Knee Osteoarthritis. Arthritis Care & Research. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sun J-W, Cao D-F, Li J-H, Zhang X, Wang Y, Bai H-Y, Lin P-Z, Zhang H-H, Cao F-L, Profiles and characteristics of clinical subtypes of perinatal depressive 2 symptoms: A latent class analysis. Journal of Advanced Nursing 0 (ja). [DOI] [PubMed] [Google Scholar]
- Timmons A, 2010. Establishing factorial invariance for multiple-group confirmatory factor analysis. KUant Guide 22. [Google Scholar]
- Titov N, Dear BF, McMillan D, Anderson T, Zou J, Sunderland M, 2011. Psychometric comparison of the PHQ-9 and BDI-II for measuring response during treatment of depression. Cognitive Behaviour Therapy 40 (2), 126–136. [DOI] [PubMed] [Google Scholar]
- Tucker LR, Lewis C, 1973. A reliability coefficient for maximum likelihood factor analysis. Psychometrika 38 (1), 1–10. [Google Scholar]
- Ulbricht CM, Dumenci L, Rothschild AJ, Lapane KL, 2018. Changes in depression subtypes among men in STAR* D: a latent transition analysis. American journal of men’s health 12 (1), 5–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Venables WN, Smith DM, Team RDC, 2002. An introduction to R. Network Theory Ltd. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.