Structure of the DSM-5 generalized anxiety disorder criteria among a large community sample of worriers

Journal of Affective Disorders 157 (2014) 18–24

Contents lists available at ScienceDirect

Journal of Affective Disorders journal homepage: www.elsevier.com/locate/jad

Research report

Structure of the DSM-5 generalized anxiety disorder criteria among a large community sample of worriers Megan J. Hobbs a,n, Tracy M. Anderson a, Tim Slade b, Gavin Andrews a a b

Clinical Research Unit for Anxiety and Depression, University of New South Wales, Australia National Drug and Alcohol Centre, University of New South Wales, Australia

art ic l e i nf o

a b s t r a c t

Article history: Received 29 October 2013 Received in revised form 25 December 2013 Accepted 27 December 2013 Available online 4 January 2014

Background: There is growing empirical and clinical consensus that many psychiatric disorders are continuous in nature. The DSM-5 however makes a categorical distinction between subthreshold and threshold cases of generalized anxiety disorder (GAD). This study tested the a priori assumption that the DSM-5 criteria identify a break in psychopathology between subthreshold and threshold cases of GAD. Methods: Respondents of the 2007 National Survey of Mental Health and Well Being who self-identified as worriers were selected for analyses (n ¼1738). The GAD criteria were assessed using the World Mental Health version of the Composite International Diagnostic Interview. Item response, latent class and factor mixture models were estimated to assess the structure of the GAD criteria. Results: The relative fit of the latent variable models suggested that a single continuous factor explains the way that worriers endorse the GAD criteria. However, the similar psychometric properties of the GAD criteria suggested that the GAD criteria impose a relatively finite threshold over this dimension of severity. Limitations: Although these structural analyses did not identify a break in psychopathology between subthreshold and threshold cases of GAD based on the way that respondents endorsed the DSM-5 criteria, it is possible that structural analyses of risk factors and other clinical correlates of GAD may identify such a break in the future. Conclusions: These data suggest that the DSM-5 GAD criteria lend themselves to making both categorical decisions about cases as well as being indices of a continuum of severity. & 2014 Elsevier B.V. All rights reserved.

Keywords: Anxiety Classification Latent Continuity Epidemiology Psychometric

1. Introduction The DSM-5 makes categorical distinctions between subthreshold and threshold cases of psychopathology. However, studies that have examined the validity of the diagnostic thresholds that serve to distinguish between subthreshold and threshold cases of some individual disorders generally support a more continuous approach to psychiatric classification. These studies have shown that the respective diagnostic thresholds do not identify distinct types of individuals with respect to risk factors or clinical correlates (Anderson et al., 2009; Angst and Merikangas, 2001; Kessler et al., 1997; Sakashita et al., 2007). These data challenge the validity of continuing to make categorical distinctions between subthreshold and threshold cases of some disorders (Regier et al., 2009).

n Correspondence to: Clinical Research Unit for Anxiety and Depression, Level 4 O’Brien Centre of St. Vincent0 s Hospital, 394-404 Victoria Street, Darlinghurst, New South Wales, Australia. Tel.: þ 61 2 8382 1408; fax: þ 61 2 8382 1401. E-mail address: [email protected] (M.J. Hobbs).

0165-0327/$ - see front matter & 2014 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jad.2013.12.041

Despite the studies that have shown that a continuous approach to classification is more appropriate than a categorical approach to classification for some of the DSM-5 disorders, few studies have investigated the categories vs. continua debate with respect to generalized anxiety disorder (GAD). One way to evaluate the validity of the categorical distinction between subthreshold and threshold cases of GAD is to assess whether the DSM-5 categorical threshold reflects a naturally occurring break in psychopathology (Kendell and Jablensky, 2003). If subthreshold and threshold cases of GAD are discrete it would be expected that these groups of individuals would be qualitatively distinct with respect to their risk factors, clinical correlates and/or associated disability. Epidemiological studies that have examined the risk factors and correlates of subthreshold and threshold cases of GAD show that these groups of individuals are not qualitatively distinct. Instead, individuals who endorse all the diagnostic criteria for GAD except, for example, that their anxiety was not excessive or that their symptoms lasted for less than 6 months are similar with respect to risk and clinical correlates to individuals who endorsed all the criteria for GAD (Angst et al., 2009; Bienvenu et al., 1998; Lee et al., 2009; Maier et al., 2000; Ruscio et al., 2005; Slade and Andrews, 2001). These

M.J. Hobbs et al. / Journal of Affective Disorders 157 (2014) 18–24

data suggest that there is a degree of continuity between subthreshold and threshold cases of GAD. Although epidemiological data suggest that subthreshold and threshold cases of GAD are continuously related, structural examinations are needed to make firm conclusions about the relative continuity of GAD. Taxometric analyses, pioneered by Meehl (1995), can be used to evaluate the structure of psychopathology without making a priori assumptions about the structure of the construct that is being examined (Haslam, 2007; Meehl, 1992, 1995; Ruscio et al., 2006). Using taxometric procedures Ruscio et al. (2001) assessed the structure of worry and found that worriers could be ordered along a severity continuum rather than being classified into distinct groups of worriers. This finding although informative of the worry process, which is one of the core clinical features of GAD, is not in and of itself indicative of the latent structure of GAD (Andrews et al., 2010). An important way of assessing whether the continuity of worry generalizes to the structure of GAD is to replicate Ruscio et al. results using other statistical methods and by examining the structure of the criteria that are used to define GAD in DSM-5. Only a limited number of studies have modelled the structure of the GAD criteria. Of the studies that have, all have used item response theory (IRT) models (Hobbs et al., 2014; Kubarych et al., 2005, 2008; Neuvo et al., 2008). IRT studies make the a priori assumption that the co-occurrence of the manifest indicators (e.g., the GAD criteria) can be explained with reference to continuous latent factor(s) (Hambleton et al., 1991). In contrast to the continuous assumptions of IRT, latent class analyses (LCA) make the a priori assumption that the underlying structure of the manifest indicators is categorical (Lazarsfeld and Henry, 1968). Factor mixture models combine the continuous assumptions of IRT and the categorical assumptions of LCA by explaining the co-occurrence of the manifest indicators using continuous latent factor(s) and explain the continuous latent variation using categorical latent variable(s) (Lubke and Muthén, 2005). Although IRT, LCA and factor mixture models each make a priori assumptions about the structure of the construct being assessed, the relative validity of these latent variable models can be tested using model fit indices. No study has used these methods to examine the structure of GAD criteria. The purpose of this study was to assess the structure of the DSM-5 GAD criteria among worriers using IRT, LCA and factor mixture models. These analyses test whether or not the criteria that are used to define threshold cases of GAD identify a break in psychopathology among worriers. Based on the available epidemiological data about the risk factors and correlates of subthreshold and threshold cases of GAD and the latent structure of worry, we hypothesized that GAD is experienced to varying degrees of severity and that the GAD criteria do not index a break in psychopathology among worriers.

2. Methods 2.1. Sample This study is based on secondary analyses of the 2007 National Survey of Metal Health and Well Being (NSMWHB), a nationally representative community survey that provides the most comprehensive assessment of the occurrence and correlates of chronic diseases and psychopathology in Australia (Slade et al., 2009). The Australian Bureau of Statistics, the federal statutory authority that conducted the survey, identified private households using a stratified, multistage area probability design. Among the 8841 fully responding households (60% response rate), one household member provided details about the household and then a household member aged between 16 and 85 years was randomly selected

19

to complete the survey interview. All respondents gave their informed consent to participate in the survey and received no compensation (Slade et al., 2009). Respondents who identified themselves as being worriers (described below) from the survey were selected for this study (n ¼1738). Selecting worriers who did (wtd%(SE) ¼41.6%(1.7)) and who did not meet threshold criteria for GAD (wtd%(SE) ¼58.4% (1.7)) meant that it was possible to test the a priori assumption that the DSM-5 criteria identify a naturally occurring break in psychopathology. The sociodemographic characteristics of the selected worriers are presented in Table 1. Worriers were mostly female and were more likely to have never been married or to be separated or divorced rather than be currently married. Worriers were more likely to be aged between 30 and 59 years than aged 60 years or more and did not differ with respect to their post high school level of education or their employment status compared to respondents who did not consider that they were worrier 2.2. Diagnostic assessment Trained lay interviewers who were employed by the Australian Bureau of Statistics interviewed all respondents using the World Mental Health version of the Composite International Diagnostic Interview (Kessler and Üstun, 2004) in a face-to-face computer assisted interview. All respondents completed the diagnostic interview in English. The WMH-CIDI uses a stem-probe interview format to reduce respondent burden and survey costs. All respondents of the 2007 NSMHWB were asked “Did you ever have a time in your life when you were a “worrier” – that is, when you worried a lot more about things than other people with the same problems as you? Did you ever have a time in your life when you were much more nervous or anxious than most other people with the same problems as you?” Respondents who endorsed at least one of these stem questions and reported that they had been anxious or worried Table 1 Sociodemographic characteristics of worriers from the 2007 National Survey of Mental Health and Well Being (n¼1738). n

wtd%(SE)

OR(95% CI)

Sex Female Male χ2 (1) ¼24.4, p o 0.01

1104 634

59.6(1.8) 40.5(1.8)

1.6(1.3–1.9) 1.0

Age 16–29 years 30–44 years 45–59 years 60þ years χ2 (1) ¼33.7, p o 0.01

349 542 455 392

20.6(1.4) 34.5(1.7) 27.4(1.8) 17.5(0.9)

1.05(0.9–1.3) 1.7(1.4–2.0) 1.4(1.1–1.7) 1.0

Education Postgraduate Bachelors Certificate or diploma No post school education χ2 (3) ¼1.8, p¼ 0.6

123 278 582 755

5.0(0.6) 14.9(1.2) 34.5(1.6) 45.6(1.8)

0.9(0.6–1.2) 1.1(0.8–1.3) 1.0(0.9–1.2) 1.0

Employment status Employed Unemployed Not in labor force χ2 (2) ¼0.64, p¼ 0.7

1087 53 598

64.2(1.6) 2.5(0.5) 33.4(1.5)

0.9(0.8–1.1) 0.9(0.5–1.6) 1.0

Marital status Never married Widowed Divorced Separated Married χ2 (4) ¼42.5, p o 0.01

592 116 272 110 648

34.5(0.6) 4.3(0.6) 11.7(1.2) 4.2(0.5) 45.3(2.0)

1.3(1.1–1.6) 1.2(0.8–1.6) 2.2(1.7–2.8) 2.4(1.6–3.5) 1.0

20

M.J. Hobbs et al. / Journal of Affective Disorders 157 (2014) 18–24

on more days than not for at least 1 month during their lives were selected for analysis. It is important to note that the DSM-IV and the DSM-5 criteria for GAD are identical and therefore the results of this study generalize to the DSM-IV GAD literature. 2.3. GAD diagnostic criteria This study examined nine of the criteria that define GAD in DSM-5 (American Psychiatric Association, 2013). These criteria are: apprehensive expectation that is excessive (“Do you think your (worry or anxiety/nervousness or anxiety/anxiety or worry)) was ever excessive or unreasonable or a lot stronger than it should have been?”), difficult to control (“Did you find it difficult to control your (worry or anxiety/nervousness or anxiety/anxiety or worry”)) and experienced on more days than not for 6 months or more (“What is the longest period of months or years in a row you ever had when you were (worried or anxious/nervous or anxious/ anxious or worried)) most days? Did you ever have a period that lasted 6 months or longer?). This apprehensive expectation is experienced with at least 3 from 6 somatic symptoms (“Did you often feel restless, keyed up, or on edge? Did you often get tired easily? Were you often more irritable than usual? Did you often have difficulty concentrating or keeping your mind on what you were doing? Did you often have tense, sore, or aching muscles? Did you often have trouble falling or staying asleep?”) 2.4. Analyses 2.4.1. Endorsement of the GAD criteria Cross tabulations were used to estimate the percentage of worriers who endorsed each of the GAD criteria. These analyses were conducted in the SUDAAN v 10.1 software package using the balanced repeated replication technique with 60 replicate weights to reflect the age and sex distribution of the Australian population and to control for the probability of being sampled for inclusion in the survey (Research Triangle Institute, 2008; Slade et al., 2009). 2.4.2. Structural analyses A series of IRT, LCA and factor mixture models were estimated using the MPlus software package v5.2 to evaluate the latent structure of the GAD criteria (Muthén and Muthén, 1998–2009). 2.4.2.1. IRT. Exploratory and confirmatory factor analyses were conducted to identify the best fitting continuous latent model of the GAD criteria. 2.4.2.2. LCA. LCA models were estimated with an increasing number of classes until there was a decrement in model fit and/or there was no longer very strong evidence, as measured by the Bayes factor (described below: Raferty (1995)), for estimating LCA models with more classes. 2.4.2.3. Factor mixture models. Two types of factor mixture models were estimated. These were: latent class factor analyses (LCFA) and factor mixture analyses (FMA) (Muthén, 2006). LCFA models explained the covariance of the GAD criteria using continuous latent factor(s) and explained the variation of the factor(s) using a categorical latent variable. The factor means varied across the classes of the categorical latent variable, the factor variances were fixed to zero and the factor loadings and symptom intercepts were invariant across classes. If the LCFA model provided the closest model fit for the GAD criteria, it would suggest that GAD is being measured in the same way across classes and that respondents0 latent severity determined their classification. Moreover, if the LCFA provided the closest fit to the structure of the GAD criteria it

would suggest that respondents classified within each class did not differ in their latent severity. The FMA modelled the latent structure of the GAD criteria in the same way as the LCFA except that the variance (and covariation) of the factor(s) were estimated and were invariant across the classes of the categorical latent variable. If the FMA provided the closest model fit for the GAD criteria, it would suggest that within each class of the categorical latent variable, respondents varied in their degree of latent severity. 2.4.3. Model estimation Exploratory analyses of the GAD criteria were estimated using an oblimin rotator that extracted 1–3 factors. Confirmatory IRT, LCA, LCFA and FMA models were all estimated using a maximum likelihood estimator with robust standard errors and final fit statistics are based on log likelihood values that were replicated across the final ten stage optimizations to avoid local maxima (Muthén and Muthén, 1998–2009). 2.4.4. Testing model fit Several statistics were used to evaluate the relative fit of the latent variable models. 2.4.4.1. Fit statistics of the exploratory IRT. Standard fit indices were used to evaluate the exploratory IRT models. These were the comparative fit statistic (CFI) (Bentler, 1990), the Tucker Lewis index (TLI) (Tucker and Lewis, 1973) and the root mean square error of approximation (RMSEA) (Steiger, 1990). As a rule of thumb CFI and TLI values Z90 reflect acceptable fit and a RMSEA r05 indicates close model fit (Browne and Cudeck, 1992). We did not use χ2 difference testing, the accuracy of which would have been compromised by our large sample. 2.4.4.2. Fit statistics of the confirmatory IRT, LCA, LCFA and FMA. The Akaike Information Criterion was used to test relative model fit where smaller AIC values reflect closer model fit (Akaike, 1974). An entropy statistic was used to test the proportion of respondents who were correctly classified in the categorical latent variable based on their posterior odds of class membership. This entropy statistic ranges from 0 to 1 where 1 indicates that 100% of the sample were correctly classified based on their posterior odds and values above 0.8 are preferred (Ramaswamy et al., 1993). We also used the Bayesian Information Criterion (BIC) (Schwartz, 1978), which balances overall model fit with the parsimony of the model and can be used to compare the relative fit of nested and nonnested models. Although smaller BIC values indicate better model fit, the magnitude of the BIC difference between two models can also be used to calculate a Bayes factor (eBICmodel i  BICmodel j ), which indexes the posterior odds of preferring the model with the smaller BIC value. As a rule of thumb, BIC differences of more than 10 provide very strong support for preferring the model with the smaller BIC value (Raferty, 1995). The MPlus software package calculates BIC values using the Schwartz method. The Schwartz criterion is equal to minus two times BIC values and because of this, the Bayes factors that are presented here are calculated using e:5 ðBICmodel i  BICmodel j Þ (Muthén and Muthén, 2008).

3. Results 3.1. Endorsement of the GAD criteria The endorsement rates of the GAD criteria are shown in Table 2. Most of the criteria were endorsed by approximately two thirds of

M.J. Hobbs et al. / Journal of Affective Disorders 157 (2014) 18–24

the sample. Difficult to control worry was the most frequently endorsed criterion whereas muscle tension was endorsed the least.

3.2. Structure of the GAD criteria among worriers The exploratory analyses of the nine GAD criteria suggested that one factor would account for the covariance of the criteria sufficiently (eigenvalues ¼3.1, 1.2, 1.0; CFI¼ 0.9, TLI ¼0.9, RMSEA ¼ 0.03). Unidimensional factor analyses confirmed the exploratory results. However, the 6 month duration criterion had a negative and low factor loading (e.g., 0.2). As a result, the duration criterion was excluded from all further analyses. A unidimensional factor model that explained the covariation of the remaining eight GAD criteria was supported by fit statistics (e.g., CFI Z0.9, TLIZ0.9 and RMSEA r0.05). The model fit indices of the confirmatory IRT, LCA, LCFA and FMA are shown in Table 3. LCA models were estimated until decrements in model fit were observed. Estimating a LCA model with three classes compared to a LCA model with two classes reduced AIC values but did not improve entropy values and the Bayes factor suggested that there was weak support for preferring a model with three classes compared to two classes. Given that two of the three statistics preferred a two class model rather than a three class model, a two class model was considered to be the best fitting LCA model. Amongst the factor mixture models, all model fit indices preferred the FMA rather than the LCFA. This suggested that the Table 2 Endorsement rates and item characteristics of the DSM-5 generalized anxiety disorder criteria among worriers from the 2007 National Survey of Mental Health and Well Being (n¼ 1738).

Excessive anxiety and worry Z6 months of anxiety and worry Difficult to control anxiety and worry Restlessness Fatigue Irritability Difficulty concentrating Muscle tension Sleep disturbance

n

wtd%(SE)

a

b

1033 1057 1387 1344 1167 1251 1211 757 1261

62.4(1.7) 60.4(1.7) 81.3(1.1) 76.7(1.2) 68.1(1.5) 72.9(1.4) 69.0(1.5) 44.0(1.7) 72.0(1.4)

0.4

 0.7

0.6 0.7 0.7 0.8 0.8 0.5 0.6

 1.8  1.2  0.8  0.9  0.8 0.4  1.1

Note: a ¼discrimination parameter; b¼ difficulty parameter.

21

worriers varied in their latent severity within each of the classes of the categorical latent variable. The best fitting IRT, LCA and FMA were then compared. There was very strong support for preferring the confirmatory IRT and the FMA over the two class LCA. A comparison between the IRT with one factor and the FMA one factor-2class model however showed that there was very strong support for preferring the confirmatory one factor IRT. Inspection of the entropy values and the number of respondents in each of the two classes also supported this conclusion. One hundred per cent of the sample was correctly classified and 100% of respondents were classified based on their posterior odds in one class of the FMA. This indicated that only one factor is needed to explain how selfidentifying worriers endorse the DSM-5 GAD criteria.

3.3. Item characteristics of the GAD criteria among worriers Within an IRT framework, binary criteria like those used to define GAD in DSM-5 are described using one person parameter and two item parameters. The person parameter (theta) describes the latent severity of the factor being measured (e.g., the covariance of the GAD criteria). The difficulty parameter (b-parameter) indexes the degree of theta that is associated with a 50% likelihood of endorsing the respective criterion and the discrimination parameter (a-parameter) describes how well the criterion distinguishes between continuous degrees of theta (Hambleton et al., 1991). The difficulty and discrimination parameters of the GAD criteria are presented in Table 2. Most of the DSM-5 GAD criteria have very similar psychometric properties. For instance, excessive anxiety and worry, fatigue, difficulty concentrating and irritability each have difficulty parameters within 0.2SD of each other. The psychometric properties of restlessness and sleep disturbance are also strikingly similar. The relationship between the item characteristics and respondents0 latent severity explains the endorsement rates of the GAD criteria (described above). The similar psychometric characteristics of excessive anxiety and worry, fatigue, difficulty concentrating, irritability, restlessness and sleep disturbance meant that these criteria were endorsed by a similar number of respondents. Given that the difficult to control criterion required the mildest latent severity to endorse and muscle tension criterion required the greatest degree of latent severity, the difficult to

Table 3 Model fit indices of the structure of the DSM-5 generalized anxiety disorder criteria among worriers from the 2007 National Survey of Mental Health and Wellbeing (n¼ 1738). BIC

0.5(BICi  BICj)

Bayes factor: strength of support for preferring modelsa

0.6 0.6

15,953.1 15,951.7

0.7

2.0

Weak support for LCA 3c vs. 2c

1.0

0.6 15,917.5

15,953.1 17.8

5.4  107

Very strong support for FMA vs. LCFA

17.8 23.9 3.8

5.4  107 2.5  1010 4.7  102

Very strong support for FMA vs. LCA Very strong support for CFA vs. LCA Very strong support for CFA vs. FMA

log(L)

k

AIC

Entropy

CFA 1f

 7892.9

16

15,817.8

15,905.2

LCA 2c 3c

 7913.1  7878.9

17 26

15,860.2 15,809.8

FMM 1f-2c LCFA 1f-2c FMA 1f-2c

 7913.1  7891.6

17 18

15,860.2 15,819.2

Best fitting CFA, LCA and FMM FMA 1f-2c vs. LCA 2c LCA 2c vs. CFA 1f FMA 1f-2c vs. CFA 1f

Note. The best fitting CFA, LCA and FMM models are shown in bold. Log(L) ¼ loglikelihood; k ¼ number of estimated parameters; AIC¼ Akaike Information Criterion; BIC¼ Bayesian Information Criterion; IRM ¼ item response models; LCA ¼latent class analyses; LCFA ¼latent class factor analyses. a

Bayes factors of 1–3¼ weak evidence; 3–20¼ positive evidence; 20–150¼ strong evidence; 4150 very strong evidence (Raferty, 1995).

22

M.J. Hobbs et al. / Journal of Affective Disorders 157 (2014) 18–24

Probability of endorsing the GAD criteria

1

0.8

0.6

0.4

0.2

0 -3

-2

-1

0

1

2

3

Latent severity (θ) Excessive

Difficult to control

Restless

Fatigue

Irritable

Difficulty concntrating

Muscle tension

Sleep disturbance

Fig. 1. Item characteristic curves of the DSM-5 generalized anxiety disorder criteria among, worriers from the 2007 National Survey of Mental Health and Well Being (n¼1738).

control criterion was the most frequently endorsed criterion whereas muscle tension was endorsed the least. The combination of these person and item parameters produces item characteristic curves for each criterion. Each item characteristic curve describes the likelihood that respondents will endorse the criterion as a function of the underlying factor and the characteristics of the criterion. The item characteristic curves of the GAD criteria are shown in Fig. 1. The close proximity of the item characteristic curves of the GAD criteria demonstrates the similar psychometric characteristics of the GAD criteria and that respondents of average latent severity have a 50% likelihood of endorsing each of the GAD criteria.

4. Discussion This is the first study to compare the relative fit of latent variable models with the purpose of examining whether the DSM-5 GAD criteria identify a break in psychopathology among worriers. We found that they do not. The DSM-5 GAD criteria index a continuum of latent severity and have similar psychometric properties among worriers. These findings have practical implications for researchers and clinicians. The first salient finding of this study was that the DSM-5 GAD criteria index a continuum of latent severity among worriers. This result supports the growing consensus that mental disorders are continuous in nature (Anderson et al., 2009; Angst and Merikangas, 2001; Kessler et al., 1997; Sakashita et al., 2007). More specific to the GAD classification, these results provide structural support for earlier epidemiological data that had suggested that subthreshold and threshold cases of GAD are continuously related (Angst et al., 2009; Bienvenu et al., 1998; Lee et al., 2009; Maier et al., 2000; Ruscio et al., 2005; Slade and Andrews, 2001). These data therefore challenge the validity of continuing to classify GAD categorically and instead support a more dimensional assessment of worriers. For instance, severity specifiers could be inserted in future GAD classifications. Given that researchers and clinicians already frequently use continuous measures to identify likely cases of GAD (e.g., the Penn State Worry Questionnaire (Molina and Borkovec, 1994) and the Generalized Anxiety Disorder-7 (Spitzer et al., 2006)), the notion

of severity specifiers for future versions of DSM may be readily embraced as both useful and structurally sound. The second important finding of this study was that the DSM-5 GAD criteria have similar psychometric properties. This means that worriers who experience similar degrees of GAD have similar likelihoods of endorsing each of the GAD criteria. The similar psychometric characteristics of the diagnostic criteria therefore impose a relatively finite threshold over the underlying dimension. The categories vs. continua debate around the classification of mental disorders has often been simplified as a debate between clinicians who find diagnostic categories useful for communicating and researchers who find the increased statistical power of continuous variables useful for delineating the epidemiology of psychopathology. Our finding that the DSM-5 GAD criteria index a continuum of psychiatric variation as well as define a relatively finite threshold over this continuum mean that the GAD criteria lend themselves to making categorical decisions about cases as well as being indices of GAD severity. There is therefore structural support for using the DSM-5 criteria to define GAD as a category or as a continuum in clinical practice and research dependent on what is more useful for the individual clinician or researcher. 4.1. Strengths and limitations This study has a number of methodological strengths. These are the first data to evaluate the validity of categorizing cases of GAD among worriers using robust latent variable methods and analyses were conducted in a large sample. However, four methodological limitations should be noted. First, these data are based on lifetime framed interview questions, which some commentators have suggested introduce age-related biases into results. The empirical support for such assertions is mixed (cf. (Andrews et al., 1999; Kessler et al., 2010)). In our recent study, we examined whether age biased respondents0 endorsement of the GAD criteria (Hobbs et al., 2014). We identified statistically significant sources of non-invariance in some criteria but these statistically significant differences had little or no effect on the prevalence of GAD. We therefore consider that the current study provides an important advance for the empirical basis of the GAD classification irrespective of its use of lifetime framed questions. Second, although we found that the GAD criteria index a finite threshold, we also found that the GAD criteria do not index a break in psychopathology among worriers. It is however possible that future structural analyses based on other risk and clinical factors could identify an empirically justified threshold for delimiting subthreshold and threshold cases of GAD (Markon, 2010). Thirdly, this study examined the structure of nine of the criteria that are used to define GAD in DSM-5. We were unable to examine how the clinical significance criterion indexes the continuum of GAD among worriers. This was because the skip-logic of the WMH-CIDI that was used in the survey meant that respondents were asked about their emotional distress and functional impairment attributable to their worry only if respondents reported that they had experienced two or more associated symptoms, essentially meeting the 3/6 associated symptom threshold for a GAD diagnosis. This violates the local independence assumption of IRT (Hambleton et al., 1991). It will be of interest for future research to examine the relationship between the clinical significance criterion and the structure of the other GAD criteria among worriers. Similarly, it will also be useful to examine the relationship between the behavioral criteria that were proposed for inclusion in DSM-5 GAD classification, which were not assessed in the survey and were not included in DSM-5 for lack of evidence about their utility to the diagnosis (Andrews and Hobbs, 2010; Andrews et al., 2010). Future structural analyses can explicitly examine this nosologically and clinically important issue.

M.J. Hobbs et al. / Journal of Affective Disorders 157 (2014) 18–24

Finally, this study selected respondents who self-identified as worriers. Worry is a core clinical feature of GAD and therefore selecting worriers facilitated an examination of how the DSM-5 GAD criteria distinguish between respondents who are the most likely in the community to meet threshold criteria for GAD. Anxiety and worry are however a part of the natural fight or flight response and accordingly there is undoubtedly a group of community dwelling adults who experience anxiety and who worry during their lives but do not consider themselves to be worriers. This means that our findings do not necessarily reflect the structural underpinnings of the full range of phenotypic variation associated with anxiety and worry. There are therefore two possible conclusions from these data. GAD criteria do not index a naturally occurring break in psychopathology or GAD criteria do index a break in psychopathology but this threshold exists between the group of people who experience low and/or fleeting periods of worry and the group of worriers selected for this study. Considered in conjunction with A. Ruscio et al. (2001) who found that worriers differ by degree rather than type, we consider that excluding individuals who experienced low and/or fleeting periods of worry is a minor limitation of this study and that there is converging evidence for concluding that GAD criteria index a latent continuum of severity.

5. Conclusions DSM-5 GAD criteria index a latent continuum of severity among worriers. Although support was found for the continuous side of the categories vs. continua debate, the GAD criteria do have similar psychometric characteristics. As a result of these similar characteristics, the DSM-5 GAD criteria impose a relatively finite threshold over the continuum of latent severity among worriers. The DSM-5 GAD criteria therefore lend themselves to making categorical decisions as well as providing indices of continuous psychiatric variation that can be used in clinical practice or research dependent on what is more useful for the individual clinician or researcher.

Role of funding source None.

Conflict of interest None.

Acknowledgments None.

References Akaike, H., 1974. A new look at the statistical model identification. IEEE Trans. Autom. Control 19, 716–723. American Psychiatric Association, 2013. Diagnostic and Statistical Manual for Mental Disorders, 5th ed. American Pychiatric Association, Washington, DC. Anderson, T.M., Slade, T., Andrews, G., Sachdev, P.S., 2009. DSM-IV major depressive episode in the elderly: the relationship between the number and the type of depressive symptoms and impairment. J. Affect. Disord. 117, 55–62. Andrews, G., Anstey, K., Brodaty, H., Issakidis, C., Luscombe, G., 1999. Recall of depressive episode 25 years previously. Psychol. Med. 29, 787–791. Andrews, G., Hobbs, M.J., 2010. The effect of the draft DSM-5 criteria for GAD on prevalence and severity. Aust. N. Z. J. Psychiatry 44, 784–790. Andrews, G., Hobbs, M.J., Borkovec, T.D., Beesdo, K., Craske, M.G., Heimberg, R.G., Rapee, R.M., Ruscio, A., Stanley, M.A., 2010. Generalized worry disorder: a review of DSM-IV Generalized Anxiety Disorder and options for DSM-V. Depress. Anxiety 27, 134–147.

23

Angst, J., Gamma, A., Baldwin, D.S., Ajdacic-Gross, V., Rössler, W., 2009. The generalized anxiety spectrum: prevalence, onset, course and outcome. Eur. Arch. Psychiatry Clin. Neurosci. 259, 37–45. Angst, J., Merikangas, K.R., 2001. Multi-dimensional criteria for the diagnosis of depression. J. Affect. Disord. 62, 7–15. Bentler, P.M., 1990. Comparative fit indexes in structural equation models. Psychol. Bull. 107, 238–246. Bienvenu, O.J., Nestadt, G., Eaton, W.W., 1998. Characterizing generalized anxiety: temporal and symptomatic thresholds. J. Nerv. Mental Disord. 186, 51–56. Browne, M.W., Cudeck, R., 1992. Alternative ways of assessing model fit. Sociol. Methods Res. 21, 230–258. Hambleton, R.K., Swaminathan, H., Rogers, H.J., 1991. Fundamentals of Item Response Theory. Sage, Newbury Park, CA Haslam, N., 2007. The latent structure of mental disorders: a taxometric update on the categorical vs. dimensional debate. Curr. Psychiatry Rev. 3, 172–177. Hobbs, M.J., Anderson, T.M., Slade, T., Andrews, G., 2014. Relationship between measurement invariance and age-related differences in the prevalence of generalized anxiety disorder. J. Affect. Disord. 152, 306–312. Kendell, R.E., Jablensky, A., 2003. Distinguishing the validity and utility of psychiatric diagnoses. Am. J. Psychiatry 160, 4–12. Kessler, R.C., Birnbaum, H., Bromet, E., Hwang, I., Sampson, N.A., Shahly, V., 2010. Age differences in major depression: results from the National Comorbidity Survey Replication. Psychol. Med. 40, 225–237. Kessler, R.C., Üstun, T.B., 2004. The world mental health (WMH) survey initiative version of the world health organization composite international diagnostic interview (CIDI). Int. J. Methods Psychiat. Res. 13, 93–121. Kessler, R.C., Zhao, S., Blazer, D.G., Swartz, M., 1997. Prevalence, correlates, and course of minor depression and major depression in the national comorbidity survey. J. Affect. Disord. 45, 19–30. Kubarych, T.S., Aggen, S.H., Hettema, J.M., Kendler, K.S., Neale, M.C., 2005. Endorsement frequencies and factor structure of DSM-III-R and DSM-IV generalized anxiety disorder symptoms in women: implications for future research, classificaton, clinical practice and comorbidity. Int. J. Methods Psychiat. Res. 14, 69–81. Kubarych, T.S., Aggen, S.H., Hettema, J.M., Kendler, K.S., Neale, M.C., 2008. Assessment of generalized anxiety disorder diagnostic criteria in the National Comorbidity Survey and Virginia Adult Twin Study of Psychiatric and Substance Use Disorders. Psychol. Assess. 20, 206–216. Lazarsfeld, P.F., Henry, N.W., 1968. Latent Structure Analyses. Plenum Press, New York, NY Lee, S., Tsang, A., Ruscio, A.M., Haro, J.M., Stein, D.J., Alonso, J., Angermeyer, M.C., Bromet, E.J., Demyttenaere, K., de Girolamo, G., de Graaf, R., Gureje, O., Iwata, N., Karam, E.G., Lepine, J.P., Levinson, D., Medina-Mora, M.E., Oakley Browne, M.A., Posada-Villa, J., Kessler, R.C., 2009. Implications of modifying the duration requirement of generalized anxiety disorder in developed and developing countries. Psychol. Med. 39, 1163–1176. Lubke, G.H., Muthén, B.O., 2005. Investigating population heterogeneity with factor mixture models. Psychol. Methods 10, 21–39. Maier, W., Gänsicke, M., Freyberger, H.J., Linz, M., Heun, R., Lecrubier, Y., 2000. Generalized anxiety disorder (ICD-10) in primary care from a cross-cultural perspective: a valid diagnostic entity? Acta Psychiatri. Scandin. 101, 29–36. Markon, K.E., 2010. How things fall apart: understanding the nature of internalizing through its relationship with impairment. J. Abnorm. Psychol. 119, 447–458. Meehl, P.E., 1992. Factors and taxa, traits and types, differences of degree and differences in kind. J. Personal. 60, 117–174. Meehl, P.E., 1995. Bootstraps taxometrics: solving the classification problem in psychopathology. Am. Psychol. 50, 266–275. Molina, S., Borkovec, T.D., 1994. The Penn State Worry Questionnaire: psychometric properties and associated characteristics. In: Davey, G.C.L., Tallis, F. (Eds.), Worrying: Perspectives on Theory, Assessment and Treatment. Wiley, New York, NY, pp. 265–283. Muthén, B.O., 2006. Should substance use disorders be considered as categorical or dimensional? Addiction 101, S6–S16. Muthén, B.O., Muthén, L., 2008. MPlus User Discussion: Confirmatory Factor Analyses, Tests of Model Fit. Muthén, L., Muthén, B.O., 1998–2009. Mplus Users0 Guide, 5th ed. Muthén & Muthén, Los Angeles, CA. Neuvo, R., Ruiz, M.A., Montorio, I., Losada, A., Márquez-González, M., 2008. A comparison of the factorial structure of DSM-IV generalized anxiety disorder between younger and older adults. J. Psychopathol. Behav. Assess. 30, 252–260. Raferty, A.E., 1995. Bayesian model selection in social research. Sociol. Methodol. 25, 111–163. Ramaswamy, V., DeSarbo, W., Reibstein, D., Robinson, W., 1993. An empirical pooling approach for estimating marketing mix elasticitys with PIMS data. Market. Sci. 12, 103–124. Regier, D.A., Narrow, W.E., Kuhl, W.E., Kupfer, D.J., 2009. The conceptual development of DSM-V. Am. J. Psychiatry 166, 645–650. Research Triangle Institute, 2008. SUDAAN v 10.1. Research Triangle Park, NC. Ruscio, A., Borkovec, T.D., Ruscio, J., 2001. A taxometric investigation of the latent structure of worry. J. Abnorm. Psychol. 110, 413–422. Ruscio, A., Lane, M., Roy-Byrne, P., Stang, P., Stein, D.J., Wittchen, H.-U., Kessler, R.C., 2005. Should excessive worry be required for a diagnosis of generalized anxiety disorder? Results from the US National Comorbidity Survey Replication. Psychol. Med. 35, 1761–1772. Ruscio, J., Haslam, N., Ruscio, A.M., 2006. Introduction to the Taxometric Method: a Practical Guide. Lawrence Erlbaum Associates, Mahwah, NJ

24

M.J. Hobbs et al. / Journal of Affective Disorders 157 (2014) 18–24

Sakashita, C., Slade, T., Andrews, G., 2007. An empirical investigation of two assumptions in the diagnosis of DSM-IV major depressive episode. Aust. N. Z. J. Psychiatry 41, 17–23. Schwartz, G., 1978. Estimating the dimension of a model. Ann. Stat. 6, 461–464. Slade, T., Andrews, G., 2001. DSM-IV and ICD-10 generalized anxiety disorder: discrepant diagnoses and associated disability. Soc. Psychiatry Psychiatr. Epidemiol. 36, 45–51. Slade, T., Johnston, A., Oakley Browne, M.A., Andrews, G., Whiteford, H., 2009. 2007 National survey of mental health and wellbeing: methods and key findings. Aust. N. Z. J. Psychiatry 43, 594–605.

Spitzer, R.L., Kroenke, K., Williams, J.B.W., Löwe, B., 2006. A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch. Intern. Med. 166, 1092–1097. Steiger, J.H., 1990. Structural model evaluation and modification: an internal estimation approach. Multivar. Behav. Res. 25, 173–180. Tucker, L.R., Lewis, C., 1973. A reliability coefficient for maximum likelihood factor analysis. Psychometrika 38, 1–10.