Journal of Affective Disorders 226 (2018) 232–238

Contents lists available at ScienceDirect

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

Research paper

A psychometric evaluation of the Posttraumatic Cognitions Inventory with Veterans seeking treatment following military trauma exposure

MARK

⁎

Minden B. Sextona,b, , Margaret T. Davisa,b,c, Diana C. Bennetta,b, David H. Morrisa,b, Sheila A.M. Rauchd,e a

VA Ann Arbor Healthcare System, Ann Arbor, MI, United States University of Michigan Medical School, Department of Psychiatry, Ann Arbor, MI, United States c Yale University School of Medicine, New Haven, CT, United States d Emory University School of Medicine, Atlanta, GA, United States e Atlanta VA Medical Center, Atlanta, GA, United States b

A R T I C L E I N F O

A B S T R A C T

Keywords: PTSD Cognitions Veterans Validity Reliability

Trauma-related beliefs have salient relationships to the development and maintenance of Posttraumatic Stress Disorder (PTSD) following stress exposure. The Posttraumatic Cognitions Inventory (PTCI) has the potential to be a standard assessment of this critical construct. However, some critical aspects of validity and reliability appear to vary by population. To date, the PTCI has not been psychometrically evaluated for use with militaryspecific traumas such as combat and military sexual trauma (MST). Based on exploratory and confirmatory analyses with 949 Veterans seeking trauma-focused treatment for military traumas, we found a four factor model (negative view of the self, negative view of the world, self-blame, and negative beliefs about coping competence) provided the best fit. In contrast, the original three factor model was not confirmed. Both models demonstrated convergent and discriminative validity. Although gender was associated with PTCI total and factor scores, differences did not persist after controlling for trauma type. MST was associated with higher PTCI scores even when controlling for gender, though the clinical magnitude of these differences is likely negligible. Internal reliability validity was demonstrated with PTCI total and subscale scores.

1. Introduction Theory and extant research have identified clinically meaningful connections between negative posttraumatic cognitions and PTSD symptoms. Cognitive models of PTSD suggest that preexisting negative beliefs about the self and others facilitate the development of PTSD by both expediting fear conditioning and maintaining the perceptions of continuing threat and inability to cope (Ehlers and Clark, 2000; Foa and Rothbaum, 1998). Additional cognitive vulnerability models postulate that the distress experienced by individuals with PTSD yields a sense of vulnerability which fosters development of negative beliefs (Shahar et al., 2013). Empirical research has supported a connection between posttraumatic cognitions and the persistence and severity of PTSD symptom. More specifically, studies find that reductions in negative beliefs are associated with decreased PTSD symptoms (Zalta et al., 2014), and that chronic PTSD symptoms are associated with worsening negative cognitions (Dekel et al., 2013). Further, Shahar et al. (2013) identified a cyclical pattern between PTSD symptoms and negative cognitions. Clinical research also suggests that the best validated ⁎

interventions for PTSD (e.g., Prolonged Exposure, Cognitive Processing Therapy) result in decreased PTSD symptoms and negative cognitions (Foa and Rauch, 2004; Iverson et al., 2015) and change in thoughts appears to drive symptom change (Kumpula et al., 2016; Schumm et al., 2015). The significance of negative trauma-specific cognitions as mediators of posttraumatic functioning seems clear. As such, it is critical that assessment instruments designed to gauge such cognitions are psychometrically robust. To this end, Foa and colleagues designed the Posttraumatic Cognitions Inventory (PTCI), a 36-item self-report questionnaire, to assess for and characterize the nature of posttraumatic cognitions (Foa et al., 1999). This widely used measure has the potential to be a standard assessment of this critical construct. Utilizing negative beliefs derived from clinical and theoretical foundations, the authors rigorously evaluated the psychometric properties of the scale with individuals without trauma histories, those with histories of trauma but not experiencing PTSD, and those with moderate or greater severity of PTSD. Their initial analyses yielded a three factor solution with 33 items comprised of three subscales: Negative Cognitions about Self

Correspondence to: Ann Arbor Veterans Healthcare Administration, Mental Health Service (116C), 2215 Fuller Rd., Ann Arbor 48105, United States. E-mail address: [email protected] (M.B. Sexton).

http://dx.doi.org/10.1016/j.jad.2017.09.048 Received 23 May 2017; Received in revised form 14 August 2017; Accepted 24 September 2017 Available online 27 September 2017 0165-0327/ Published by Elsevier B.V.

Journal of Affective Disorders 226 (2018) 232–238

M.B. Sexton et al.

women Veterans are more likely to present for treatment secondary to sexual trauma while men are more likely to seek care for combat events. To address this concern, we also investigated gender differences on PTCI total and subscale scores with and without controlling for trauma type. To further assess the psychometric performance of the PTCI with Veterans, we aimed to explore its factor structure in treatment-seeking Veterans and contrasted the performance of our empirically-derived factor model with that originally proposed by Foa et al. (1998). PTCI responses were also compared to self-reported symptom and coping scales to establish convergent validity, and internal reliability was investigated using both models. We hypothesized negative cognitions would be positively correlated with PTSD and depression symptom scales and negatively correlated with a scale of resilient coping.

(SELF), Negative Cognitions about the World (WORLD), and Self-Blame (BLAME) subscales. Three original items were included in the initial version of the scale with a recommendation for further evaluation, and not used in scoring. Their initial investigation found excellent internal consistency reliability, high convergent validity with PTSD symptom severity, and good discriminative validity to distinguish between those with and without PTSD following trauma. Since its introduction, research examining the PTCI has included participants with a wide variety of trauma histories (e.g. Daie-Gabai et al., 2011), including physical and sexual assault (Foa et al., 1999; Van Emmerik et al., 2006) and motor vehicle accidents (e.g. Beck et al., 2004), for which the interpersonal nature of the events may vary. To date, PTCI validation studies have supported three-factor structures and identified acceptable model fit for samples who have experienced assault. However, the BLAME subscale has proven less consistently associated with PTSD symptom severity following non-interpersonal traumas, suggesting that the validity of the PTCI may vary depending on trauma type (Su and Chen, 2008). In addition, results have varied widely with respect to support for using all items of the PTCI and the dependability of items to load on anticipated factors. This evidence, coupled with the frequent use of convenience populations in PTCI validation studies, may limit our conclusions regarding the psychometric constancy of the PTCI with trauma-exposed individuals. Veterans with exposure to military combat and sexual traumas experience PTSD at higher rates than non-military samples (Kang et al., 2005; Tolin and Foa, 2006), and evidence suggests that negative posttraumatic cognitions are highly prevalent among Veterans presenting for trauma-focused care (Rauch et al., 2009). While military combat traumas (MCT) often share some of the interpersonal aspects of civilian assault trauma, the PTCI has not been empirically evaluated for use in the context of combat trauma. Relatedly, though some psychometric studies are based on physical and sexual assault survivors, it is unknown whether the PTCI will retain its utility following experiences of military sexual trauma (MST). The possibility that Veterans’ experiences of sexual trauma within the context of military culture differ from those experienced by their civilian counterparts in unique ways that influence cognitions and beliefs concerning self and others warrants consideration. For example, the military prizes unit cohesiveness and bodily strength. As such, survivors of MST may be more likely to believe military peers would be unsupportive of efforts to report assaultive behaviors (Burns et al., 2014; Suris and Lind, 2008). Similarly, those who have experienced MST may be more prone to believing they are to blame for not being able to physically deter an attack. Nonetheless, researchers have been using the PTCI with combat Veterans and identified that the PTCI has clinical utility in mediating associations between PTSD and anger or pain, for example (Germain et al., 2015; Porter et al., 2013). Yet, without establishing the psychometric characteristics of the PTCI for Veterans with military-specific traumas, the generalizability and interpretability of these findings is limited. The current study aimed to address this gap in the literature by evaluating the validity and reliability of the PTCI in a sample of treatment-seeking Veterans and directly contrasted the pervasiveness of negative cognitions between those seeking care for combat and military sexual traumas while taking gender into consideration. While no currently published studies have examined gender differences in the PTCI among American military Veterans, studies of other populations suggest that significant gender differences exist, particularly for SELF. In a sample of Israeli adults with exposure to heterogeneous trauma types, for example, women reported higher scores on SELF, whereas no differences were found on the PTCI total score or the WORLD or BLAME scales (Daie-Gabai et al., 2011). Similarly, Moser et al. (2007) examined the PTCI among trauma-exposed college students and found that the associations between the WORLD and BLAME scales with PTSD symptoms were no longer significant once gender and other relevant variables were included in the model. Unfortunately, in Veteran populations, gender and trauma type are often conflated as

2. Method 2.1. Participants and procedure Participants were 949 Veterans presenting for trauma-focused treatment at a Midwestern Veterans Healthcare Administration PTSD specialty clinic between 2006 and 2013. At their initial visit, Veterans completed self-report questionnaires, (including the PTCI), as part of the intake and treatment planning process. In order to investigate the psychometric performance of the PTCI following military-specific traumas, only Veterans with complete PTCIs who endorsed primary combat or military sexual traumas were included in analyses. The VA Ann Arbor Healthcare System's Human Subject Committee approved the research protocol. 2.2. Measures 2.2.1. Posttraumatic Cognitions Inventory (PTCI) This study utilized the original 36-items of the PTCI developed by Foa et al. (1999). The initial instrument included 33 validated items and three additional items recommended for further assessment. The self-report items contain responses ranging from 1 (totally disagree) to 7 (totally agree). The original factor structure supported a three-factor model including Negative Cognitions About the Self (SELF, 21 items), Negative Cognitions About the World (WORLD, seven items), and SelfBlame (BLAME, five items). The PTCI has demonstrated adequate convergent and discriminant validity in some samples though the recommendation to retain all of the items has varied by population as previously discussed. In the present analyses, we utilized the 36-item version to explore the factor structure of the PTCI and contrasted our findings with Foa's 33-item version. 2.2.2. PTSD checklist-civilian (PCL-C) The PCL-C (Weathers et al., 1993) is 17-item self-report assessment of PTSD symptom severity based on DSM-IV-TR (APA, 2000) diagnostic criteria. Items are scored from 1 (not at all) to 5 (extremely) with participants rating how bothered they have been by the presence of the symptom during the past month with total scores ranging from 17 to 85. The PCL-C has demonstrated adequate reliability and validity (Ruggiero et al., 2003). 2.2.3. Patient health questionnaire-9 (PHQ-9) Depressive symptoms were evaluated with the PHQ-9 (Kroenke et al., 2001), a nine-item self-report screen. Items responses range from 0 (not at all) to 3 (nearly every day). Scores above nine are generally considered reflective of moderate or greater symptom severity and scores above 13 suggest a current major depressive episode. The PHQ-9 has demonstrated good reliability and validity (Beard et al., 2016; Inoue et al., 2012; Kroenke et al., 2001). 233

Journal of Affective Disorders 226 (2018) 232–238

M.B. Sexton et al.

2.2.4. Connor-Davidson Resilience Scale (CD-RISC) Resilience was assessed using the Connor-Davidson Resilience Scale (Connor and Davidson, 2003). This instrument is a 25-item, measure of stress coping ability. Scores range from 0 (not true at all) to 4 (true nearly all the time) with higher scores denoting more frequent utilization of positive coping strategies posited to buffer against exposure to traumas and other stressors. The CD-RISC has demonstrated adequate validity and internal reliability with traumatized populations (Connor and Davidson, 2003; Karaırmak, 2010).

Table 1 Demographic and service characteristics.

Age Married Caucasian Male MCT Service Era Vietnam OIF/OEF

2.3. Analytic strategy All analyses were conducted in IBM SPSS version 22.0. For the factor analyses, the original sample was divided into two datasets (S1 and S2). Cases were split such that the resulting datasets were roughly equal in size (S1 n = 480; S2 n = 469), and matched for distribution of gender (S1% male = 91.50; S2% male = 90.6) and primary trauma type (S1% combat trauma = 91.50; S2% combat trauma = 89.2). In light of the disparate findings of previous authors concerning the structure of the PTCI (as detailed above), the first analytic step was to conduct an exploratory factor analysis (EFA). More specifically, we submitted the full 36-item PTCI to EFA using sample 1. Maximum likelihood extraction and an oblique (direct oblimin) rotation was used in recognition of the fact previous research has shown the PTCI factors to be correlated. Solutions were examined using eigenvalues, fit statistics, and patterns in factor loadings (Kaiser, 1960). In the second analytic step, the selected model from step 1 (EFA following removal of non-loading items) was submitted to CFA using the second, matched sample (S2). A second CFA was also conducted in S2 using Foa et al. (1999) originally-derived three-factor structure for the PTCI (using 33 items). For both models, modification indices were examined to determine whether allowing the error terms of any items specified to load on the same factor to co-vary would substantially improve model fit (i.e. reduce model χ2 value by 10 or more). Where substantial improvement in fit was indicated and suggested paths made theoretical sense, modifications were made. Models were then re-run to confirm fit improvement (Schreiber et al., 2006). Finally, AIC values were compared to evaluate whether the EFA-based (hereafter referred to as the Veteran model) four-factor model or the original three-factor model (hereto referred to as the Foa model) provided a closer fit to the data (a difference in AIC scores of 10 or greater was interpreted as in indication of superior model fit; Hooper et al., 2008). Of note, because the models were not nested, a direct comparison of model fit (chisquare difference test) could not be performed. Individual items were used as indicators in all models. For all analyses, PTCI items were treated as continuous in keeping with prior research. Fit indices for all CFA models were evaluated based on criteria for identifying close fit (Hu and Bentler, 1999; Kline, 2005: RMSEA ≤ .06), and model fit was assessed by examining all available fit statistics for each model (Kline, 2011). CFIs and TLIs were analyzed with > .90 considered indicative of acceptable fit and ≥ .95 suggestive deemed good fit (Hu & Benter). All participants included in analyses had complete data on the PTCI. Power analysis was conducted a priori referencing criteria established by MacCallum et al. (1996) for use in structural modeling. Both samples exceeded the identified minimum sample sizes to achieve power of .80 for the degrees of freedom in each planned model. In order to assess potential mean differences due to sample or trauma characteristics, ANOVAs were used to contrast gender and military trauma type (MST vs. MCT) on mean total and factor scores for the Veteran and original Foa versions of the PTCI. Histories of dual exposure to MST and MCT were reported by 2.8% of participants Of these, 89.5% were seeking treatment associated with MST and 10.5% were seeking treatment for combat trauma. Analyses of trauma type were based on the primary trauma for which Veterans were seeking care. ANCOVAs were also conducted to evaluate the relative influence

Total sample (N = 949)

EFA (n = 480)

CFA (n = 469)

F/χ2

p

45.6 (16) 53.9% 85.7% 91% 90.3%

46.3 (16) 52.5% 86.1% 91.5% 91.5%

44.8 (16) 55.4% 85.4% 90.6% 89.2%

1.84 .78 .08 .21 1.44

.176 .378 .777 .648 .230

39.2% 39.6%

42.9% 35.8%

35.5% 41.3%

5.32 2.95

.021* .086

Notes: EFA = exploratory factor analysis group; CFA = confirmatory factor analysis group; MCT = military combat trauma is the primary identified traumatic event; OIF/ OEF = Operation Iraqi Freedom/Operation Enduring Freedom. * p < .05.

of gender and trauma type and reduce conflation of effects. Bivariate correlations were used to evaluate the associations between PTCI total and subscale scores and PTSD symptoms, depressive complaints, and resilience. Internal consistency reliability was estimated using Cronbach's alpha. 3. Results 3.1. Descriptive and service characteristics Characteristics of the total sample and the subgroups used for the factor analyses are described in Table 1. The two samples were similar on demographic and service characteristics with the exception of a significantly higher representation of Vietnam service era Veterans in the EFA group compared to the CFA subgroup. Of the 92 Veterans comprising the MST group, 50 were female. Among the 857 Veterans presenting for treatment associated with MCT exposure, 822 were male. 3.2. PTCI validation 3.2.1. Factor analyses 3.2.1.1. Exploratory factor analysis. All 36 PTCI items were first submitted to EFA in S1 (n = 480). Examination of both the scree plot and eigenvalues (referencing a cut-off of > 1 in keeping with the Kaiser-Guttman rule; Kaiser, 1960) revealed that a four-factor solution provided best fit to the data (accounting for 56.97% of variance, cumulatively). Of note, the accepted four-factor model did not result in close fit based on evaluation of chi-square alone (χ2= 1278.23, df = 492, p < .001). All items in the accepted four-factor model had salient loadings (parameter estimates > .4; Kline, 2011) on at least one factor with the exception of item 14 which loaded below .4 on Factors 1 (.37) and 4 (.37) and was removed from subsequent analyses. Item 24 cross-loaded on Factors 1 and 2 and was also omitted. See Table 2 for a full list of final factor loading values after removal of these two items. The selected four-factor Veteran model solution was similar to Foa et al. (1999) three-factor model in a number of ways (see Supplemental Table 1 for item-content mapping). First, Factors 2 (WORLD, eight items, accounting for 6.34% of variance) and 3 (BLAME; five items; accounting for 3.90% of variance) were identical to those observed in the original Foa model with one exception. That is, in the new model one of the three “experimental” items excluded from the original Foa model (item 34; “You never know when something terrible might happen”) loaded onto Factor 2. Similarly, Factor 1 (SELF; 18 items, accounting for the majority, 43.98%, of variance) largely overlapped with the Foa model Factor 1 with three notable exceptions. First, in the four-factor model another of the experimental items, consistent in content (item 32; “I will not be able to tolerate my feelings about the event, and I will fall apart”) also loaded on Factor 1. Second, as noted above, item 14 which 234

Journal of Affective Disorders 226 (2018) 232–238

M.B. Sexton et al.

factor was at most moderately correlated with other factors (highest r = .55), making it difficult to justify the items ‘belonging to’ or being merged into another factor. Finally, theoretically, we felt the content of the items, negative beliefs about coping competence, might represent a conceptually distinct and clinically meaningful construct among trauma-exposed Veterans.

Table 2 PTCI items OBLIMIN Rotated Factor Loadings (EFA analyses). Item

Item content

F1

F2

F3

F4

1 2 3 4 5 6 7 8 9 10 11 12 13

Happened because of my actions Can’t trust myself to do right I am weak Can’t control my anger Can’t deal with upset I’m always miserable Can’t trust others Need to be on guard Feel dead inside Never know who will harm you Never know what is next I am inadequate Won’t be able to control my emotions Can’t handle thinking about trauma Happened because of the sort of person I am My reactions mean I’m going crazy Never be able to feel normal emotion World is dangerous Somebody else would have stopped it Permanently changed for the worse Feel like an object not a person Someone else wouldn’t have gotten into this Can’t rely on others Feel isolated/set apart I have no future Can’t stop bad things happening to me People not what they seem My life has been destroyed Something wrong with me as a person Reactions show I am a lousy coper Something about me made (trauma) happen Not able to tolerate thoughts about it Feel like I don’t know myself anymore Never know when something will happen Can’t rely on myself Nothing good can happen anymore

− .156 .409 .427 − .027 .139 .474 .130 − .032 .729 − .077 − .104 .692 .246

.076 −.060 −.056 .116 .091 .231 .657 .740 .192 .772 .652 .033 .088

.627 .181 .201 −.045 .030 −.124 .006 .031 −.055 .068 .139 .046 .076

.049 .268 .131 .754 .684 .202 .076 .067 −.080 .041 .196 .043 .590

– .045

– .034

– .727

– .022

.470 .615

− .067 .186

.225 − .023

.248 .052

.024 .135

.675 .002

.021 .568

.035 − .036

.703 .720 .084

.105 .157 .081

−.105 .051 .653

.101 −.065 .009

.325 – .868 .467

.566 – −.026 .194

.036 – .012 −.014

−.069 – −.073 .050

.190 .679 .720

.713 .094 .039

.048 .061 .025

−.099 .024 .066

.659 .151

−.037 .021

.126 .787

.092 −.105

.560

−.076

.129

.264

14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

.617

.130

.052

.113

.092

.630

.028

.014

.695 .769

−.155 −.078

.184 .104

.120 −.025

3.2.1.2. Confirmatory factor analyses. The correlated four-factor firstorder measurement model identified through EFA was then submitted to CFA using maximum likelihood estimation in S2. This model did not initially result in close fit (χ2 = 1576.65, df = 521, p < .001; RMSEA = .06). Modification indices suggested that allowing seven pairs of error terms (all between items loading on the same factor) to correlate would substantially improve model fit. Each possible modification was examined to ensure theoretical justification existed for allowing the terms to covary. Modifications were then made one at a time and the model was re-run to ensure the predicted chi-reduction had occurred, and that in light of the changes allowing other identified error terms to co-vary would still substantially improve model fit. After each of the eight modifications were implemented, the Veteran model showed adequate fit to the data (χ2 = 1278.77, df = 513, p < .001; RMSEA = .056; RMSEA CI = .05–.06; CFI = .93; TLI = .92). Zero-order correlations among latent variables in this model ranged from .433 to .795 (See Table 3 for factor loadings). Next, the three-factor original Foa model using 33 PTCI items was submitted to CFA using S2. Table 3 PTCI Item CFA Factor Loadings. Foa Model Item 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

Notes. PTCI = Posttraumatic Cognitions Inventory; EFA = Exploratory Factor Analyses; F1 = Factor 1; F2 = Factor 2; F3 = Factor 3; F4 = Factor 4; – = omitted during the preliminary EFA due to cross-loading or failure to load ≥ .4. Bold font indicates the factor each item was retained on for the CFA.

originally loaded onto the Foa model's Factor 1 was eliminated due to cross-loading, Third, and most significantly, two items from the Foa model's Factor 1 (items 4 and 5; “I will not be able to control my anger and will do something terrible” and “I can’t deal with even the slightest upset”), and the third experimental item not included in the original model (item 14; “If I think about the event, I will not be able to handle it”) loaded onto a distinct fourth factor. Examination of the content of this fourth factor revealed a set of self-directed negative thoughts (in keeping with Foa model Factor 1) all of which focused on perceived inability to cope with adverse thoughts or emotions related to the trauma led us to label it “Negative Beliefs about Coping Competence” (COPE). Of note, Kline (2005) recommends that latent variables consist of a minimum of three indicators (although it remains accepted practice in select cases to retain two-item factors, as in the case of DSM-5 PTSD's avoidance factor; APA, 2013). We therefore considered the decision to retain the threeitem COPE factor as independent with care. Our decision to do so was based on three points. First, the factor accounted for a small, but meaningful amount of variance independently (2.8%). Second, the

F1

EFA-Derived Veteran Model F2

F3

F1

F2

.609 .576a .590a .577 .624a .759

F3

F4

.609 .583a .590a .793 .768 .750a

.786 .799

.782 .800 .792a

.793 .793 .757 .741 – .699a

.795 .762 a

.742 –

– –

–

.789

.878 –

.691a .723a

.672 .724a

.642a

.643 .679 .740a .794a

.679 .741a .793a

.736

.736

.718 – .787a .604

a

.719 .790a .606a .774a

.712 –

–

–

.776

.750 .779 .733a – .819 – .741a .773a

– .789

.747 .776 .739a –

.795 –

–

–

.794 .773 .818 .650a .749a .771a

Note. Dashes included in the table (–) indicate that item loadings were fixed to 0 where specified. a Denotes items whose covariances were permitted to vary. Rationales associated with varying indices are provided in supplemental tables.

235

Journal of Affective Disorders 226 (2018) 232–238

M.B. Sexton et al.

a notable interaction found males scored significantly higher than females subsequent to MST (F = 4.911, df = 3945, p = .03), indicating higher perceived inability to cope. Using the original scoring, no significant between-group gender differences persisted after controlling for trauma type. When contrasting total and subscale differences based on military trauma type (MST vs. MCT), Veterans seeking care associated with military sexual trauma scored higher on total scores and all subscales (Table 6). In all instances, these differences remained when controlling for gender. However, the magnitudes of these differences were small.

Table 4 PTCI total and factor score means, internal reliability analyses, and bivariate correlations with the PCL-C, PHQ-9, and CD-RISC. Mean (SD)

α

PCL-C*

PHQ-9*

CD-RISC*

4-factor model Total Factor 1 Factor 2 Factor 3 Factor 4

129.2 (41.8) 3.6 (1.4) 5.1 (1.3) 2.7 (1.5) 3.5 (1.7)

.96 .96 .91 .83 .85

.59 .59 .48 .31 .44

.65 .67 .49 .32 .47

− .65 −.69 − .43 − .31 − .54

3-factor model Total Factor 1 Factor 2 Factor 3

125.5 (40.7) 3.6 (1.4) 5.0 (1.4) 2.7 (1.5)

.96 .96 .90 .84

.59 .59 .47 .31

.65 .67 .49 .32

− − − −

.65 .69 .44 .31

3.2.3. Convergent validity In order to evaluate convergent validity, we correlated the PTCI total and factor scores with measures of PTSD and depressive symptoms and resilient coping (Table 4). Average scores on the PCL-C, PHQ-9, and CD-RISC were 61.6 (12.5), 15.0 (6.6), and 57.0 (18.5), respectively. Symptom-based scales (PCL-C and PHQ-9) exceeded clinical cutoffs suggestive of the presence of PTSD and major depression. The CD-RISC was substantially below established norms for non-clinical populations suggesting less resilient coping in our sample. As expected, higher PTCI total and subscale scores for the Veteran and Foa models were significantly and positively correlated with self-reported PTSD and depression symptoms. Similarly, significant negative associations were found between PTCI scores and resilience.

Notes. PTCI = Posttraumatic Cognitions Inventory; PCL-C = PTSD Checklist-Civilian; PHQ-9 = Patient Health Questionnaire-9; CD-RISC = Connor-Davison Resilience Scale; F1 = Factor 1; F2 = Factor 2; F3 = Factor 3; F4 = Factor 4. * Bivariate correlations were significant at p < .001 in all cases.

Identical procedures for evaluating modification indices and permitting error-covariance were utilized, and nine modifications were made. Zero-order correlations among latent variables in the final modified Foa model ranged from .433 to .727. However, neither the original (χ2 = 1694.18, df = 492, p < .001; RMSEA = .07) or modified (χ2 = 1322.9, df = 483, p < .001; RMSEA = .06; RMSEA CI = .06–.07; CFI = .92; TLI = .91) version of the original Foa model represented adequate fit to the data (See Table 3 for factor loadings). Because this Foa model was not nested within the Veteran model, chi-square difference testing could not be conducted to compare fit. Rather, model AIC values were compared with a difference of 10 or greater considered indicative of superior fit. This comparison (Veteran AIC = 1510.77; Original Foa AIC = 1544.9) confirmed that the Veteran four-factor model represented a significantly better fit to the data. See Table 4 for average total and factor scores for the four- and three-factor models.

3.2.4. Discriminative validity A cutoff score of ≥ 50 on the PCL-C was used to denote cases of PTSD in order to evaluate the discriminative ability of the PTCI total scores. As may be anticipated in a treatment-seeking sample, 81% of the Veterans reported PTSD symptom severity in the clinical range. Using the Veteran 34-item model, PTCI total score predicted the PTSD diagnostic status of 82.1% of participants correctly (β = .03, SE = .004, Wald = 70.98, p < .001, OR = 1.030 [95% CI = 1.023–1.037]). Nagelkerke's R2 suggested a medium effect for the relationship (.279). Utilizing the original 33-item model, the PTCI total score predicted the diagnostic status of 82.9% of participants correctly (β = .03, SE = .004, Wald = 71.238, p < .001, OR = 1.031 [95% CI = 1.024–1.038]; Nagelkerke R2 = .278).

3.2.2. Gender and trauma type contrasts Between-gender contrasts detected significant differences for the Veteran model on the PTCI total score and Factors 2 and 3 with women scoring significantly higher in each regard (Table 5). When using the original factor loadings, women were significantly higher on total scores and the WORLD and BLAME subscales. However, for both the Veteran and Foa models, between-group effect sizes were small and did not persist when controlling for trauma type. Only one factor demonstrated a significant interaction effect. For the Veteran COPE (factor 4),

3.3. Internal consistency reliability The internal reliability was excellent for both the Veteran and Foa model total scores (α = .96; Table 4). All factors demonstrated good or excellent internal reliability (α range = .83–.96)

Table 5 Gender differences in PTCI factors with and without controlling for trauma type. Gender

No control for trauma type

Controlled for trauma type

Males M (SD)

Females M (SD)

F

p

ƞp2

F

p

ƞp2

4-factor model Total Factor 1 Factor 2 Factor 3 Factor 4

128.3 (41.5) 3.5 (1.4) 5.1 (1.3) 2.7 (1.4) 3.5 (1.6)

138.6 (43.9) 3.8 (1.5) 5.4 (1.2) 3.2 (1.8) 3.4 (1.7)

4.75 6.74 5.01 10.52 .32

.029* .069 .025* .001** .575

.005 .003 .005 .011 < .001

.11 < .01 .12 .28 4.41

.741 .982 .727 .599 .036*

< .001 < .001 < .001 < .001 .005

3-factor model Total Factor 1 Factor 2 Factor 3

124.6 (40.3) 3.6 (1.4) 5.0 (1.4) 2.7 (.1)

134.7 (42.8) 3.9 (1.5) 5.3 (1.2) 3.2 (.2)

4.84 2.48 4.53 10.52

.028* .116 .034* .001**

.005 .003 .005 .011

< .01 .29 .03 .28

.992 .593 .868 .599

< .001 < .001 < .001 < .001

Notes: PTCI = Posttraumatic Cognitions Inventory. * p < . 05. ** p < .01.

236

Journal of Affective Disorders 226 (2018) 232–238

M.B. Sexton et al.

Table 6 Trauma type differences in PTCI factors with and without controlling for gender. Trauma Type

No control for gender

Controlled for Gender

MST M (SD)

MCT M (SD)

F

p

ƞp2

F

p

ƞp2

4-factor model Total Factor 1 Factor 2 Factor 3 Factor 4

148.7 (39.7) 4.1 (1.4) 5.7 (1.1) 3.5 (1.6) 3.9 (1.8)

127.1 (41.5) 3.5 (1.4) 5.0 (1.3) 2.7 (1.4) 3.4 (1.6)

22.63 15.70 18.41 29.45 5.73

< .001** < .001** < .001** < .001** .019*

.023 .016 .019 .030 .006

17.87 12.41 13.32 18.98 9.66

< .001** < .001** < .001** < .001** .002**

.019 .013 .014 .020 .010

3-factor model Total Factor 1 Factor 2 Factor 3

144.7 (38.7) 4.2 (1.4) 5.6 (1.1) 3.5 (1.6)

123.4 (40.4) 3.6 (1.4) 5.0 (1.4) 2.7 (1.4)

23.13 15.39 19.27 29.45

< .001** < .001** < .001** < .001**

.024 .016 .020 .030

18.28 13.15 14.69 18.98

< .001** < .001** < .001** < .001**

.019 .014 .015 .020

Notes: PTCI = Posttraumatic Cognitions Inventory; MST = military sexual trauma, MCT = military combat trauma. * p < . 05. ** p < .01.

4. Discussion

In order to further examine these key demographic factors and their influence on the PTCI, we then focused on trauma type. Across the PTCI and using both the original and new four factors, Veterans seeking treatment for PTSD secondary to MST reported more negative thoughts about the self, negative thoughts about the world, more self-blame, and more sense that they cannot cope with strong affect. When controlling for gender, all of these effects remained. Further, although some psychometric investigations have described issues with convergent validity between the BLAME scale and PTSD symptoms with mixed trauma and non-interpersonal trauma types (i.e. Beck et al., 2004; Daie-Gabai et al., 2001; Su and Chen, 2008), it appears that Veterans presenting for care following military-specific traumas (i.e., combat and sexual trauma) do appear more similar to non-Veterans endorsing sexual and physical assault events (i.e. Foa et al., 1999; Van Emmerik et al., 2006). This has important clinical ramifications for targeting salient maladaptive cognitions during trauma-focused therapy within the context of military traumas. For instance, the blame subscale in Veterans with MCT more closely approximates samples with interpersonal traumas (i.e. physical or sexual assault) than traumas with comparatively less probability for interpersonal characteristics (i.e. motor vehicle accidents). As such, emphasis on perceptions related to social networks and self-blame may be of particular importance with Veterans seeking care for militaryspecific traumas. Despite the large treatment seeking sample, several important caveats are apparent. First, these Veterans were all presenting to assessment and potential treatment in a VA PTSD clinic and, as such, may not represent the full spectrum of negative thoughts in trauma survivors or even those suffering from PTSD. Second, due to gender differences in exposure to MST and low numbers of female Veterans overall, the analyses of trauma type (MST versus MCT) and gender are somewhat conflated even with the use of statistical control procedures. Male Veterans were more likely to have MCT as their primary trauma and female Veterans were more likely to have MST as their primary trauma. In addition, while 2.8% of Veterans endorsed exposure to both MST and MCT, we did not analyze groups with dual exposure separately. Future research is also needed to ascertain whether this rate of dual exposure is consistent in other samples of trauma-seeking Veterans as well as whether the presentations or clinical care needs of Veterans with histories of MST and MCT vary based on which trauma is considered primary. Moreover, previous examinations have indicated that men and women may differ in rates and rationales for withholding disclosure of MST (Hoyt et al., 2011; Suris and Smith, 2011). Although not unique to this study, this may result in an underestimation of dual MST and MCT

The current study supports the psychometric integrity of the PTCI for a Veteran population through replication and extension of the original PTCI in a large Veteran sample. Specifically, the PTCI demonstrated convergent and discriminative validity and internal reliability using the scoring and factor validity of the BLAME and COPE subscales. As such, the current study supports the PTCI total score for use in the Veteran population. The normative means provided from this large sample represent an excellent population base for future comparison and determination of clinical significance of PTCI scores. In contrast, using Foa et al.’s (1999) initial three-factor model, we were unable to attain an adequate fit to the data with confirmatory factor analyses, even when utilizing modification indices and permitting error-covariance. However, exploratory and confirmatory factor analyses pressed on to demonstrate that slight modifications to the three factors created a stable and potentially informative fourth factor that can best be described as perceived competence to handle strong negative affect that we are labeling COPE. While only three items create this additional factor, it demonstrates internal reliability and was stable in the confirmatory analysis. In addition to providing the best fit to the data, our four-factor model similarly evidenced good-to-excellent correlations with PTSD, depression, and resilience, and was similarly robust with convergent validity, predictive potential, and internal reliability. With the central role that negative thoughts about one's ability to regulate affect can play in Veterans’ functioning, providing a subscale that uniquely taps this construct, may provide specific insights into the process of PTSD and therapeutic change in Veterans. Similarly, the original items Foa and colleagues recommended for additional evaluation appear particularly notable for Veterans seeking care for military traumas and essential to the psychometric stability of the factor structure. When examining the impact of gender and trauma type on the subscales, gender impacts the original and revised four-factor scales of negative thoughts about the world and self-blame such that women report small but significantly higher negative thoughts about the world and self-blame. However, when controlling for whether the trauma is combat or military sexual trauma, these differences disappear and only the coping factor shows that men report significantly more sense that they cannot manage intense negative affect. As with the uncontrolled gender effects, this difference is small but significant. Further research is needed to ascertain whether the effect of gender on the PTCI offers clinical value in assessment or treatment. 237

Journal of Affective Disorders 226 (2018) 232–238

M.B. Sexton et al.

303. Foa, E.B., Rauch, S.A., 2004. Cognitive changes during prolonged exposure versus prolonged exposure plus cognitive restructuring in female assault survivors with posttraumatic stress disorder. J. Consult. Clin. Psychol. 72 (5), 879. Foa, E.B., Rothbaum, B.O., 1998. Treating the Trauma of Rape: A Cognitive-behavioral Therapy for PTSD. Guilford, New York. Germain, C.L., Kangas, M., Taylor, A., Forbes, D., 2015. The role of trauma‐related cognitive processes in the relationship between combat‐PTSD symptom severity and anger expression and control. Aust. J. Psychol. Hooper, D., Coughlan, J., Mullen, M., 2008. Structural equation modelling: guidelines for determining model fit. Articles 2. Hoyt, T., Klosterman Rielage, J., Williams, L.F., 2011. Military sexual trauma in men: a review of reported rates. J. Trauma Dissociation 12 (3), 244–260. Hu, L.T., Bentler, P.M., 1999. Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct. Equ. Model.: Multidiscip. J. 6 (1), 1–55. Inoue, T., Tanaka, T., Nakagawa, S., Nakato, Y., Kameyama, R., Boku, S., Koyama, T., 2012. Utility and limitations of PHQ-9 in a clinic specializing in psychiatric care. BMC Psychiatry 12 (1), 1. Iverson, K.M., King, M.W., Cunningham, K.C., Resick, P.A., 2015. Rape survivors' traumarelated beliefs before and after Cognitive processing therapy: associations with PTSD and depression symptoms. Behav. Res. Ther. 66, 49–55. Kaiser, H.F., 1960. The application of electronic computers to factor analysis. Educ. Psychol. Meas. Kang, H., Dalager, N., Mahan, C., Ishii, E., 2005. The role of sexual assault on the risk of PTSD among Gulf War veterans. Ann. Epidemiol. 15 (3), 191–195. Karaırmak, Ö., 2010. Establishing the psychometric qualities of the Connor–Davidson Resilience Scale (CD-RISC) using exploratory and confirmatory factor analysis in a trauma survivor sample. Psychiatry Res. 179 (3), 350–356. Kline, T.J., 2005. Psychological Testing: A Practical Approach to Design and Evaluation. Sage Publications. Kline, R.B., 2011. Principles and Practice of Structural Equation Modeling, 3rd ed. Guilford Press, New York, NY. Kroenke, K., Spitzer, R.L., Williams, J.B., 2001. The PHQ-9. J. Gen. Intern. Med. 16 (9), 606–613. Kumpula, M.J., Pentel, K.Z., Foa, E.B., LeBlanc, N.J., Bui, E., McSweeney, L.B., Rauch, S.A., 2016. Temporal sequencing of change in posttraumatic cognitions and PTSD symptom reduction during prolonged exposure therapy. Behav. Ther. MacCallum, R.C., Browne, M.W., Sugawara, H.M., 1996. Power analysis and determination of sample size for covariance structure modeling. Psychol. Methods 1 (2), 130. Moser, J.S., Hajcak, G., Simons, R.F., Foa, E.B., 2007. Posttraumatic stress disorder symptoms in trauma-exposed college students: the role of trauma-related cognitions, gender, and negative affect. J. Anxiety Disord. 21 (8), 1039–1049. Porter, K.E., Pope, E.B., Mayer, R., Rauch, S.A., 2013. PTSD and pain: exploring the impact of posttraumatic cognitions in veterans seeking treatment for PTSD. Pain Med. 14 (11), 1797–1805. Rauch, S.A., Defever, E., Favorite, T., Duroe, A., Garrity, C., Martis, B., Liberzon, I., 2009. Prolonged exposure for PTSD in a Veterans Health Administration PTSD clinic. J. Trauma. Stress 22 (1), 60–64. Ruggiero, K.J., Del Ben, K., Scotti, J.R., Rabalais, A.E., 2003. Psychometric properties of the PTSD Checklist—Civilian version. J. Trauma. Stress 16 (5), 495–502. Schreiber, J.B., Nora, A., Stage, F.K., Barlow, E.A., King, J., 2006. Reporting structural equation modeling and confirmatory factor analysis results: a review. J. Educ. Res. 99 (6), 323–338. Schumm, J.A., Dickstein, B.D., Walter, K.H., Owens, G.P., Chard, K.M., 2015. Changes in posttraumatic cognitions predict changes in posttraumatic stress disorder symptoms during cognitive processing therapy. J. Consult. Clin. Psychol. 83 (6), 1161. Su, Y.J., Chen, S.H., 2008. The posttraumatic cognitions inventory-Chinese revised: validation and refinement with a traumatized college sample in Taiwan. J. Anxiety Disord. 22 (7), 1110–1119. Surís, A., Lind, L., 2008. Military sexual trauma a review of prevalence and associated health consequences in veterans. Trauma Violence Abus. 9 (4), 250–269. Suris, A., Smith, J., 2011. PTSD related to sexual trauma in the military. In: Moore, B.A., Penk, W.E. (Eds.), Treating PTSD in Military Personnel: A Clinical Handbook. Guilford Press, New York, NY. Tolin, D.F., Foa, E.B., 2006. Sex differences in trauma and posttraumatic stress disorder: a quantitative review of 25 years of research. Psychol. Bull. 132 (6), 959. Van Emmerik, A.A.P., Schoorl, M., Emmelkamp, P.M.G., Kamphuis, J.H., 2006. Psychometric evaluation of the Dutch version of the posttraumatic cognitions inventory (PTCI). Behav. Res. Ther. 44 (7), 1053–1065. Weathers, F.W., Litz, B.T., Herman, D.S., Huska, J.A., Keane, T.M., 1993. The PTSD Checklist (PCL): reliability, validity, and diagnostic utility. In: Annual Convention of the International Society for Traumatic Stress Studies. International Society for Traumatic Stress Studies, San Antonio, October. Zalta, A.K., Gillihan, S.J., Fisher, A.J., Mintz, J., McLean, C.P., Yehuda, R., Foa, E.B., 2014. Change in negative cognitions associated with PTSD predicts symptom reduction in prolonged exposure. J. Consult. Clin. Psychol. 82 (1), 171.

that may have important relationships with post-traumatic cognitions. Finally, the PTCI factors demonstrated preliminary validity with the PCL-C, which measured DSM-IV PTSD symptoms consistent with the data collection window of the study. With the modification of PTSD criteria presented in the DSM-5 (APA, 2013), additional research is warranted to confirm if our findings generalize to current guidelines. Notwithstanding these limitations, this research is the first to examine the factor structure of the PTCI with Veterans and adds to our knowledge of the psychometric performance of this measure with those seeking PTSD care subsequent to military combat and sexual traumas. Overall, several factors appeared robust for use with this population although a four-factor model offered the best fit for this study, which warrants replication and future research. The factors demonstrated good reliability and validity with measures of PTSD, depression, and resilience and provide normative data for those with military traumas. Future research is recommended to ascertain the clinical and predictive utility of the four-factor PCTI with military populations to evaluate the differential effectiveness of discrete factors for the development and course of PTSD and associated concerns associated with combat and sexual traumas. Role of funding source The research presented was supported by the Mental Health Service at VA Ann Arbor Healthcare System and the University of Michigan Department of Psychiatry. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Acknowledgements The authors would like to thank Margaret Venners, Caitlyn Authier, and Lauren McSweeney for their contributions to this research. Appendix A. Supplementary material Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jad.2017.09.048. References American Psychiatric Association, 2000. Diagnostic and Statistical Manual of Mental Disorders DSM-IV-TR, fourth edition (text revision). American Psychiatric Pub., Washington, DC. American Psychiatric Association, 2013. Diagnostic and Statistical Manual of Mental Disorders (DSM-5). American Psychiatric Pub. Beard, C., Hsu, K.J., Rifkin, L.S., Busch, A.B., Björgvinsson, T., 2016. Validation of the PHQ-9 in a psychiatric sample. J. Affect. Disord. 193, 267–273. Beck, J.G., Coffey, S.F., Palyo, S.A., Gudmundsdottir, B., Miller, L.M., Colder, C.R., 2004. Psychometric properties of the posttraumatic cognitions inventory (PTCI): a replication with motor vehicle accident survivors. Psychol. Assess. 16 (3), 289. Burns, B., Grindlay, K., Holt, K., Manski, R., Grossman, D., 2014. Military sexual trauma among US servicewomen during deployment: a qualitative study. J. Inf. 104, 2. Connor, K.M., Davidson, J.R., 2003. Development of a new resilience scale: the Connor‐Davidson resilience scale (CD‐RISC). Depress. Anxiety 18 (2), 76–82. Daie-Gabai, A., Aderka, I.M., Allon-Schindel, I., Foa, E.B., Gilboa-Schechtman, E., 2011. Posttraumatic Cognitions Inventory (PTCI): psychometric properties and gender differences in an Israeli sample. J. Anxiety Disord. 25 (2), 266–271. Dekel, S., Peleg, T., Solomon, Z., 2013. The relationship of PTSD to negative cognitions: a 17-year longitudinal study. Psychiatry 76 (3), 241–255. Ehlers, A., Clark, D.M., 2000. A cognitive model of posttraumatic stress disorder. Behav. Res. Ther. 38 (4), 319–345. Foa, E.B., Ehlers, A., Clark, D.M., Tolin, D.F., Orsillo, S.M., 1999. The posttraumatic cognitions inventory (PTCI): development and validation. Psychol. Assess. 11 (3),

238