Introduction
It is well established that people with post-traumatic stress disorder (PTSD) have negative cognitions, i.e. a negative perception of self and negative world view (e.g. Ehlers and Clark, Reference Ehlers and Clark2000; Foa and Cahill, Reference Foa, Cahill, Smelser and Bates2001; Foa and Rothbaum, Reference Foa and Rothbaum1998), which have recently been incorporated into the DSM-5 criteria for PTSD (American Psychiatric Association, 2013). Research has shown clinically meaningful connections between negative post-traumatic cognitions and PTSD symptoms (e.g. Beck et al., Reference Beck, Coffey, Palyo, Gudmundsdottir, Miller and Colder2004; Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999; Shin et al., Reference Shin, Chung, Kim, Kim and Chang2020), and demonstrated that such beliefs are predictive of PTSD severity (Brown et al., Reference Brown, Belli, Asnaani and Foa2018). According to cognitive models of PTSD, negative beliefs about the self and others can maintain perceptions of continuing threat and prevent effective coping (Ehlers and Clark, Reference Ehlers and Clark2000; Foa and Rothbaum, Reference Foa and Rothbaum1998). Furthermore, when individuals with PTSD experience distress, they may feel vulnerable, which can promote the development of negative beliefs, which may in turn maintain other PTSD symptoms (Shahar et al., Reference Shahar, Noyman, Schnidel-Allon and Gilboa-Schechtman2013).
Individuals diagnosed with serious mental illness (SMI), typically defined as any major psychiatric disorder, such as schizophrenia, schizoaffective disorder, major depression and bipolar disorder, accompanied by persistent functional impairments (Martínez-Martínez et al., Reference Martínez-Martínez, Richart-Martínez and Ramos-Pichardo2020; Substance Abuse and Mental Health Services Administration, 2017), who have experienced childhood trauma, have poorer clinical outcomes than those not exposed, including higher levels of repetitive negative thinking, dysfunctional metacognitive beliefs, and more severe symptoms (Grubaugh et al., Reference Grubaugh, Zinzow, Paul, Egede and Frueh2011; Mansueto et al., Reference Mansueto, Cavallo, Palmieri, Ruggiero, Sassaroli and Caselli2021; Struck et al., Reference Struck, Krug, Yuksel, Stein, Schmitt, Meller, Brosch, Dannlowski, Nenadić, Kircher and Brakemeier2020). Although SMI has typically been linked to schizophrenia, bipolar disorder and treatment refractory major depression (Bigdeli et al., Reference Bigdeli, Voloudakis, Barr, Gorman, Genovese, Peterson and Gill2022; Grubaugh et al., Reference Grubaugh, Brown, Wojtalik, Myers and Eack2021; Parabiaghi et al., Reference Parabiaghi, Bonetto, Ruggeri, Lasalvia and Leese2006), consensus is lacking in terms of the specific disorders included (Martínez-Martínez et al., Reference Martínez-Martínez, Richart-Martínez and Ramos-Pichardo2020). SMI categorizations in many jurisdictions in the United States include DSM diagnoses, other than the typical SMI-related conditions above, when there is accompanying persistent impairment in functioning. SMI in this study is broadly defined as a major psychiatric disorder such as schizophrenia, bipolar and major depressive disorder, with resultant serious functional impairment. Individuals with SMI are at greater risk for trauma exposure (Grubaugh et al., Reference Grubaugh, Zinzow, Paul, Egede and Frueh2011; Grubaugh et al., Reference Grubaugh, Brown, Wojtalik, Myers and Eack2021; Struck et al., Reference Struck, Krug, Yuksel, Stein, Schmitt, Meller, Brosch, Dannlowski, Nenadić, Kircher and Brakemeier2020) than people in the general population, with multiple traumatizations common (Lu et al., Reference Lu, Mueser, Rosenberg and Jankowski2008; Mueser et al., Reference Mueser, Salyers, Rosenberg, Goodman, Essock, Osher, Swartz and Butterfield2004). The high exposure to trauma in persons with SMI is associated with high rates of co-occurring PTSD, typically ranging between 25 and 48% of current PTSD (Grubaugh et al., Reference Grubaugh, Zinzow, Paul, Egede and Frueh2011), considerably higher than the estimated 3.5% current prevalence in the general population (Kessler et al., Reference Kessler, Berglund, Demler, Jin, Merikangas and Walters2005). PTSD contributes to a worse course of SMI, and more severe symptoms, including anxiety, depression and psychosis (Mueser et al., Reference Mueser, Salyers, Rosenberg, Goodman, Essock, Osher, Swartz and Butterfield2004; Seow et al., Reference Seow, Ong, Mahesh, Sagayadevan, Shafie, Chong and Subramaniam2016). PTSD can be effectively treated in people with SMI and there is evidence that changes in PTSD symptoms are mediated by changes in trauma-related cognitions in people with SMI (Mueser et al., Reference Mueser, Rosenberg, Xie, Jankowski, Bolton, Lu, Hamblen, Rosenberg, McHugo and Wolfe2008, Reference Mueser, Gottlieb, Xie, Lu, Yanos, Rosenberg, Silverstein, Duva, Minsky, Wolfe and McHugo2015), as reported in the general population (e.g. Brown et al., Reference Brown, Belli, Asnaani and Foa2018; Germain et al., Reference Germain, Kangas, Taylor and Forbes2016). Scales for measuring negative trauma-related cognitions in SMI populations are therefore important.
The Post-Traumatic Cognitions Inventory (PTCI; Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999) is a commonly used measure of trauma-related cognitions and beliefs. The PTCI contains 36 statements, with three subscales assessing negative cognitions about the self (SELF subscale; 21 items), the world (WORLD subscale; 7 items), and self-blame (BLAME subscale; 5 items). The initial three-factor model was based on the 33 items of its initial validation study (Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999). Three items were added to the original scale for exploratory purposes, and inclusion of these items in further analyses in other studies led to a four-factor solution in some of those studies (Sexton et al., Reference Sexton, Davis, Bennett, Morris and Rauch2018). The PTCI also discriminated between traumatized individuals with and without PTSD and was found to be sensitive to treatment changes (Foa and Rauch, Reference Foa and Rauch2004; Germain et al., Reference Germain, Kangas, Taylor and Forbes2016).
Since its initial development and validation, studies have examined whether the three-factor structure of the PTCI is supported across a range of trauma-exposed samples (Table 1). Supporting the original three-factor structure was a study with a Dutch sample with primarily non-sexual interpersonal violence experiences (van Emmerik et al., Reference van Emmerik, Schoorl, Emmelkamp and Kamphuis2006). However, results inconsistent with the three-factor structure were reported by Beck et al. (Reference Beck, Coffey, Palyo, Gudmundsdottir, Miller and Colder2004) in a sample of people who had experienced a serious motor vehicle accident (MVA). In the Beck et al. (Reference Beck, Coffey, Palyo, Gudmundsdottir, Miller and Colder2004) study, an adequate fit was achieved only by removing four items from the SELF factor items (2, 4, 24 and 29). While SELF and WORLD subscales had concurrent and discriminant validity, the BLAME subscale demonstrated poor convergent validity by failing to correlate with measures of PTSD, depression or quality of life, and poor discriminant validity by failing to discriminate between individuals with versus without PTSD. Similar findings were reported by Su and Chen (Reference Su and Chen2008) and Daie-Gabai et al. (Reference Daie-Gabai, Aderka, Allon-Schindel, Foa and Gilboa-Schechtman2011) in samples of participants with mixed trauma histories. One possible explanation for the discrepant findings on the factor structure of the PTCI across these studies may be differences in the traumatic events experienced by the participants. The samples of Foa et al. (Reference Foa, Ehlers, Clark, Tolin and Orsillo1999) and van Emmerik et al. (Reference van Emmerik, Schoorl, Emmelkamp and Kamphuis2006) included victims of interpersonal violence, the sample of Beck et al. (Reference Beck, Coffey, Palyo, Gudmundsdottir, Miller and Colder2004) consisted solely of survivors of MVAs, while Su and Chen (Reference Su and Chen2008) and Daie-Gabai et al. (Reference Daie-Gabai, Aderka, Allon-Schindel, Foa and Gilboa-Schechtman2011) included individuals with mixed trauma histories.
Table 1. Model fit indices of nine prior studies of PTCI (between years 1999 and 2020)
χ2, diagonally weighted least squares chi-square; d.f., degrees of freedom; CFI, comparative fit index; TLI, Tucker-Lewis index; RMSE, root mean square error of approximation; SRMR, standard root mean square residual; *p<.05,
a Foa’s three-factor model with 33 items removing items 13, 32 and 34 and five pairs of error covariances (error covariance between item 3 and 12, 4 and 5, 5 and 14, 10 and 11, 25 and 36).
b Four-factor model with 34 items removing items 14 and 24 and five pairs of error covariances (error covariance between item 3 and 12, 4 and 13, 10 and 11, 25 and 36, 32 and 33);
c NFI, normed fit index.
While most studies of the PTCI have focused on the three-factor model, a recent study among veterans found support for a four-factor model (Sexton et al., Reference Sexton, Davis, Bennett, Morris and Rauch2018). In this study, exploratory and confirmatory factor analyses of PTCI supported the presence of a stable fourth factor (labelled COPE, consisting of three items that originally loaded onto the SELF subscale) that can be described as perceived lack of competence at handling strong negative emotional states.
Although the psychometric properties of the PTCI have been examined in a range of trauma-exposed groups, they have not yet been investigated among individuals with SMI (Table 1). Prior studies suggest that clients with SMI and co-occurring PTSD score significantly higher on the PTCI (Mueser et al., Reference Mueser, Rosenberg, Xie, Jankowski, Bolton, Lu, Hamblen, Rosenberg, McHugo and Wolfe2008, Reference Mueser, Gottlieb, Xie, Lu, Yanos, Rosenberg, Silverstein, Duva, Minsky, Wolfe and McHugo2015) than sexual assault survivors (Andreu et al., Reference Andreu, Peña and de La Cruz2017), MVA victims (Beck et al., Reference Beck, Coffey, Palyo, Gudmundsdottir, Miller and Colder2004), and veterans (Sexton et al., Reference Sexton, Davis, Bennett, Morris and Rauch2018). Given that trauma is highly prevalent and disabling among people with SMI, there is a need to investigate the factor structure and construct validity of PTCI in this population. The aims of the study therefore were to explore the factor structure and psychometric properties of the PTCI among individuals with SMI. We hypothesized that the Sexton et al. (Reference Sexton, Davis, Bennett, Morris and Rauch2018) four-factor model would have the best fit, while the three-factor model would demonstrate adequate fit based on existing studies on the factor structure of the PTCI (Table 1). Prior studies have found mixed support for the three-factor model, whereas the study of Sexton et al. (Reference Sexton, Davis, Bennett, Morris and Rauch2018) in a large treatment-seeking veteran sample supported the four-factor model over the three-factor model. We further hypothesized that the PTCI subscales would show good internal consistency across three different samples of people with SMI, and that PTCI subscales would be moderately and significantly correlated with PTSD symptom severity and other psychiatric symptoms (suggesting convergent validity), but not PTSD knowledge or working alliance, constructs unrelated to post-traumatic cognitions (suggesting divergent validity).
Method
Participants
Participants for analyses came from three studies (see Table 2). The first sample was drawn from a randomized controlled trial (RCT) and conducted in New Hampshire/Vermont, USA at four community mental health centers (Study 1; n=108; Mueser et al., Reference Mueser, Rosenberg, Xie, Jankowski, Bolton, Lu, Hamblen, Rosenberg, McHugo and Wolfe2008). Participants met the following criteria: (1) >18 years old; (2) determination by New Hampshire or Vermont as having a SMI, defined as a DSM-IV Axis I disorder and persistent impairment in work, school, or ability to care for oneself; (3) diagnosis of major depression, bipolar disorder, schizoaffective disorder, or schizophrenia confirmed through Structured Clinical Interview of DSM-IV disorders (SCID; First et al., Reference First, Spitzer, Gibbon and Williams1996); and (4) current diagnosis of PTSD based on Clinician Administered PTSD Scale (CAPS; Blake et al., Reference Blake, Weathers and Nagy1990).
Table 2. Characteristics of the sample (N=432)
BDI, Beck Depression Inventory; BAI, Beck Anxiety Inventory; CAPS, Clinician-Administered PTSD Scale; PTCI, Posttraumatic Cognitions Inventory.
The second sample was drawn from participants in a RCT conducted in partial hospitalization programs and out-patient programs within a community mental health service system in New Jersey, USA (Study 2; n=199; Mueser et al., Reference Mueser, Gottlieb, Xie, Lu, Yanos, Rosenberg, Silverstein, Duva, Minsky, Wolfe and McHugo2015). Inclusion criteria were: (1) >18 years old; (2) met State of New Jersey definition of SMI; (3) diagnosis of schizophrenia, schizoaffective disorder, major depression, or bipolar disorder, based on the SCID (First et al., Reference First, Spitzer, Gibbon and Williams1996); and (4) diagnosis of severe PTSD, based on the CAPS, schizophrenia version (Gearon et al., Reference Gearon, Bellack and Tenhula2004), with a minimum total score of 65. Exclusion criteria included having a hospitalization, suicide attempt, or substance dependence within the past three months.
The third sample was drawn from a RCT for treatment of PTSD among 132 participants recruited from 12 vocational rehabilitation programs at community mental health centres in three Northeastern states in the USA (Study 3; n=125; Lu et al., Reference Lu, Yanos, Waynor, Jia, Siriram, Leong, Gill and Mueser2022). Inclusion criteria for Study 3 were: (1) >18 years old; (2) received supported employment services within the past 24 months; (3) in treatment for SMI, with diagnosis based on self-report; (4) current diagnosis of PTSD, determined by the CAPS-5 (Weathers et al., Reference Weathers, Bovin, Lee, Sloan, Schnurr, Kaloupek, Keane and Marx2018); (5) no current chart diagnosis of alcohol or drug dependence; and (6) no hospitalizations or suicide attempts in the past two months.
Across all three studies, participants had SMI diagnosis and confirmed co-occurring PTSD diagnosis using CAPS or CAPS-5. The majority of participants were female (69%) and in their late 40s (Table 2). Participants predominantly (98%) had diagnosis of schizophrenia, schizoaffective disorder, bipolar disorder, or major depressive disorder. For primary SMI diagnosis, all studies had more representation of non-psychotic (i.e. mood disorders) than psychotic disorders (16% for Study 1, 34% for Study 2, and 36% for Study 3). The racial composition differed significantly between studies, with the majority of participants in Study 1 being White, the majority being non-White in Study 2, and an approximately equal proportion of White and non-White participants in Study 3. Across all studies, 51% of participants were White and 38% were Black.
Measures
Participants in all three studies completed baseline assessments prior to randomization. Only baseline data were used for the present analyses.
Posttraumatic Cognitions Inventory (PTCI; Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999)
The PTCI was used to assess trauma-related cognitions. The PTCI is a self-report measure pertaining to common negative thoughts and beliefs about self, other people, and the world in individuals with PTSD (Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999). The PTCI consists of 36 items ranging from 1 (totally disagree) to 7 (totally agree). It has good test–retest reliability and has been shown to be particularly effective at discriminating between traumatized individuals with PTSD and those without (Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999). In detecting PTSD, sensitivity ranged from .70 to .78 and specificity ranged from .81 to .93 (Beck et al., Reference Beck, Coffey, Palyo, Gudmundsdottir, Miller and Colder2004; Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999). Both subscale scores and total scores are based on the original 33 items (Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999). In the current investigation, Cronbach’s alpha for the 36-item PTCI total score was .95, .94, .96 and .95 in Studies 1–3, and total combined sample, respectively.
PTSD Checklist for DSM-5 (PCL-5; Weathers et al., Reference Weathers, Litz, Keane, Palmieri, Marx and Schnurr2013)
The PCL-5 was used in Study 3 to assess the severity of PTSD symptoms. The PCL-5 is a 20-item self-report measure that assesses the 20 DSM-5 symptoms of PTSD on a 0–4 scale (Weathers et al., Reference Weathers, Litz, Keane, Palmieri, Marx and Schnurr2013). The PCL-5 has demonstrated strong internal consistency (α>.90; Blevins et al., Reference Blevins, Weathers, Davis, Witte and Domino2015), and convergent validity (Wortmann et al., Reference Wortmann, Jordan, Weathers, Resick, Dondanville, Hall-Clark, Foa, Young-McCaughan, Yarvis, Hembree, Mintz, Peterson and Litz2016). The internal consistency of the PCL-5 was .89 for Study 3.
Beck Depression Inventory (BDI-II; Beck et al., Reference Beck, Steer and Garbin1988; Beck, Reference Beck1996)
The BDI-II is a 21-item self-report questionnaire used in all three studies to measure depression. Average internal consistency coefficient was .86 with excellent validity (Beck et al., Reference Beck, Steer and Garbin1988). The internal consistency of the BDI-II was .93, .91 and .91 for Studies 1–3, respectively.
Beck Anxiety Inventory (BAI; Beck and Steer, Reference Beck and Steer1993)
The BAI is a 21-item scale measuring self-reported anxiety in all three studies. The BAI discriminates between anxiety and depression and has high test–retest reliability (Beck et al., Reference Beck, Steer and Garbin1988). The questionnaire describes emotional, physiological and cognitive symptoms of anxiety. The internal consistency of the BAI was .93, .94 and .91 for Studies 1–3, respectively.
Expanded version of the Brief Psychiatric Rating Scale (BPRS; Lukoff et al., Reference Lukoff, Nuechterlein and Ventura1986)
The expanded version of the BPRS was used in Studies 1 and 3 to assess psychiatric symptoms. This interview-based measure includes 24 items, rated on a 7-point Likert scale. The BPRS is a measure of the severity of psychiatric symptoms; it has excellent psychometric properties and has established factor structure in SMI population (Mueser et al., Reference Mueser, Goodman, Trumbetta, Rosenberg, Osher, Vidaver, Auciello and Foy1998).
Positive and Negative Syndrome Scale (PANSS; Kay et al., Reference Kay, Fiszbein and Opler1987)
The PANSS is a 30-item interview-based measure used in Study 2 to measure psychiatric symptoms. Each item is rated on a 7-point Likert scale (from ‘absent’ to ‘extreme’). The scale consists of three subscales: positive, negative, and general psychopathology. Internal consistency ranged from .85 to .88 (Citrome et al., Reference Citrome, Meng and Hochfeld2011; Lancon et al., Reference Lancon, Auquier, Nayt and Reine2000) and test–retest reliability ranged from .77 to .89 (Kay et al., Reference Kay, Fiszbein and Opler1987).
Structured Clinical Interview for DSM-IV Axis-II Personality Disorders (SCID II; First et al., Reference First, Spitzer, Gibbon and Williams1995)
The SCID II Borderline Personality Module was used in Studies 1 and 2 to assess interpersonal functioning with items including unstable relationships, recurrent self-injurious/suicidal behaviour, mood instability and more.
Quality of Life Interview (QOLI; Lehman et al., Reference Lehman, Kernan and Postrado1995)
The QOLI is a 74-item instrument used in Study 2 to assess quality of life on a 7-point scale. Internal consistency ranged from 0.68 to 0.85 (Lançon et al., Reference Lançon, Auquier, Toumi, Launois, Llorca, Lehman and Bebbington1999). Test–retest reliability ranged from .28 to .98, and validity is good based on confirmatory factor analyses and multivariate predictive models (Lehman, Reference Lehman1996).
PTSD Knowledge Scale (K-PTSD; Pratt et al., Reference Pratt, Rosenberg, Author, Brancato, Salyers, Jankowski and Descamps2005)
The K-PTSD was used in Studies 1 and 2 and assessed understanding of PTSD. It contains 15 multiple choice questions in relation to PTSD. Previously it has demonstrated sensitivity to changes following psychoeducational interventions about trauma for people with psychiatric disabilities (Pratt et al., Reference Pratt, Rosenberg, Author, Brancato, Salyers, Jankowski and Descamps2005). The internal consistency of the K-PTSD was .65 for Study 1 and .48 for Study 2.
Patient version of the Working Alliance Inventory (WAI; Horvath and Greenberg, Reference Horvath and Greenberg1989)
The patient version of the WAI was used in Studies 1 and 2 to rate the therapeutic alliance with case managers (i.e. not the therapist providing CBT treatment). Internal consistency estimates for the patient version ranged from .83 to .97 (Hanson et al., Reference Hanson, Curry and Bandalos2002). The measures utilized in each study are presented in the Supplementary material.
Data analysis
Confirmatory factor analyses (CFA) with maximum likelihood estimation with robust standard errors (i.e. MLR estimator) using Mplus 8.7 was conducted to identify the best fitting model. There were 15,552 data points on PTCI for the total sample (N=432), of which 306 (1.95%) were missing data. Missing data were included in the analysis by using Full Information Maximum Likelihood (FIML) in Mplus software. Missing data were handled using pairwise deletion in correlational analysis.
Model goodness of fit was evaluated using indices: model χ2 test, comparative fit index (CFI), Tucker-Lewis index (TLI), standardized root mean square residual (SRMR), and root mean square error of approximation (RMSEA). Models with reasonably good fit may have CFI and TLI values that exceed 0.90 (Kline, Reference Kline2005). Using two-index presentation strategy, a cut-off value close to .95 for CFI and TLI, a cut-off value close to .08 for SRMR, and a cut-off value close to .06 for RMSEA are needed to establish relatively good fit (Hu and Bentler, Reference Hu and Bentler1999).
Multi-group CFA was conducted to examine measurement invariance of models across the three primary diagnostic samples of SMI (schizophrenia-schizoaffective disorder, major depression, and bipolar disorder). Configural, metric and scalar invariance were used to determine the degree of invariance of PTSD between groups, as suggested by Gregorich (Reference Gregorich2006) and Meredith (Reference Meredith1993). Configural invariance determined if there was adequate fit and if the relationships between variables could be considered equivalent. Metric invariance tested if both groups contributed similar regression weights to the latent construct. Scalar invariance determined if both groups had equivalent intercept values in the latent construct on the observed variable of item scores (Gregorich, Reference Gregorich2006; Meredith, Reference Meredith1993). More stringent criteria were applied when the total N>300, including a change <.01 in CFI supplemented with a change of <.015 in RMSEA and <.03 in SRMR for metric invariance; and a change of <0.01 in CFI supplemented with a change of <.015 in RMSEA and <.01 in SRMR for scalar invariance (Chen, Reference Chen2007). Convergent validity was calculated using correlations between PTCI scales and measures of trauma symptom severity, anxiety, depression, psychiatric symptoms, and quality of life. Divergent validity was evaluated by correlations between PTCI scales, knowledge of PTSD, and working alliance.
Results
Participants in the three studies met PTSD criteria using CAPS or CAPS-5, and had moderate to severe depression on average and moderate to severe anxiety. Participants in both Studies 1 and 2 had severe PTSD on average (CAPS >65). Total PTCI scores did not differ across the three samples (see Table 2).
To examine whether the PTCI had a similar factor structure in SMI clients to the three-factor model with 33 items (excluding items 13, 32 and 34; Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999), a CFA was performed using MLR estimator (see Table 1). This initial model without error covariance did not result in good fit (χ2/d.f.=2.14, p<.001; CFI=.89, TLI=.89, RMSEA=.05, SRMR=.05). After allowing five pairs of error terms (all between items loading on the same factor; error covariance between item 3 and 12, 4 and 5, 5 and 14, 10 and 12, 25 and 36) to covary, suggested by modification indices, the model fit improved but was not ideal (χ2/d.f.=1.92, p<.001; CFI=.91; TLI=.91, RMSEA=.05, SRMR=.05, see Table 1).
Next, analysis on the four-factor model (Sexton et al., Reference Sexton, Davis, Bennett, Morris and Rauch2018) was carried out with 36 items of PTCI (excluding items 14 and 24). The initial model also did not result in a good fit (χ2/d.f.=2.05, p<.001; CFI=.90; TLI=.89, RMSEA=.05, SRMR=.05). After allowing five pairs of error terms (all between items loading on the same factor; error covariance between item 3 and 12, 5 and 14, 10 and 11, 25 and 36, 4 and 5) to covary, indicated by modification indices, the model fit improved substantially and suggested adequate fit (χ2/d.f.=1.84, p<.001; CFI=.92; TLI=.91, RMSEA=.04, SRMR=.05, see Table 1). The four-factor model performed similarly to the three-factor model, both reaching adequate fit by the commonly accepted standard of a RMSEA value less than .08, SRMR less than .08, and CFI and TLI values higher than .90 (Hu and Bentler, Reference Hu and Bentler1999; Kline, Reference Kline2005) (CFI=.92 vs .91; TLI=.91 vs .91; RMSEA=.04 vs .05; SRMR=.05 vs .05, respectively). Both established relatively good fit, using the commonly accepted standard of a cut-off value close to .95 for CFI and TLI, .06 for RMSEA, and .08 for SRMR (Hu and Bentler, Reference Hu and Bentler1999).
To examine measurement invariance of Foa’s three-factor model and Sexton’s four-factor model across the three diagnostic groups traditionally associated with SMI (schizophrenia-schizoaffective, bipolar, major depression), multi-group CFA was conducted (Table 3). We examined configural, metric and scalar invariance levels. Two out of the four fit indices for Foa’s three-factor model indicated good model fit (CFI=.86, TLI=.85, RMSEA=.06, SRMR=.068) and suggested invariance of the factor structure (configural invariance) across diagnostic groups. Sexton’s four-factor model indicated slightly better fit than Foa’s model (CFI=.87, TLI=.85, RMSEA=.000, SRMR=.066) and suggested configural invariance. Following Chen (Reference Chen2007), in the multi-group CFA, changes in CFI, RMSEA and SRMR values were used to determine between-model statistical significance; ΔCFI<−.01, ΔRMSEA<.015, and ΔSRMR<.030 for the metric invariance model or ΔSRMR<.010 for the scalar invariance model. Both Foa’s and Sexton’s models demonstrated configural, metric and scalar invariance across diagnostic groups with associated changes in CFI, RMSEA and SRMR values of less than .01 (Table 3).
Table 3. Measurement invariance across three diagnostic samples, gender, and ethnic groups for the 3-factor and 4-factor model
Diagnostic group (schizophrenia/schizoaffective n=116, bipolar n=134, and major depression n=174); gender (male n=133 vs female n=299); ethnic groups (White n=214 vs Black n=164); CFI, comparative fit index; TLI, Tucker-Lewis index; RMSEA, root mean square error of approximation; SRMR, standard root mean square residual; *p<= .05;
a Foa’s 3-factor model (33-item, removed 13, 32, 34; error covariance between item 3 and 12, 4 and 5, 5 and 14, 10 and 11, 25 and 36);
b Sexton’s 4-factor model (34-item, removed 14, 24; error covariance between item 3 and 12, 4 and 13, 10 and 11, 25 and 36, 32 and 33).
Further, multi-group CFA was conducted to examine measurement invariance of Foa’s three-factor model and Sexton’s four-factor model across gender (male vs female) and ethnic groups (White vs Black) (Table 3). Similarly, for the invariance analysis concerning gender, Foa’s three-factor model indicated acceptable model fit (CFI = .89, TLI = .89, RMSEA = .05, SRMR = .06) while Sexton’s four-factor model performed similarly (CFI = .89, TLI = .89, RMSEA = .05, SRMR = .06), indicating configural invariance regarding gender. The value differences in CFI, RMSEA, and SRMR were smaller than .01, meeting the criteria for both metric and scalar invariance for gender. For the ethnic group analysis (White vs. Black), fit indices for Foa’s three-factor model indicated acceptable model fit (CFI = .90, TLI = .89, RMSEA = .05, SRMR = .06), whereas Sexton’s four-factor model performed similarly for ethnic group invariance testing (CFI = .89, TLI = .88, RMSEA = .05, SRMR = .06), indicating configural invariance. Further, the changes in values were smaller than .01 for CFI, RMSEA, and SRMR, thus establishing metric and scalar invariance for ethnicity.
Altogether, CFA analysis and multi-group CFA indicated that four-factor models had a similar fit to the three-factor model. To evaluate convergent validity, correlations were computed between the PTCI total score and its factor scores for Foa’s three-factor model, Sexton’s four-factor model, measures of clinician-rated (CAPS, CAPS-5) and self-reported PTSD symptoms (PCL-5), depression (BDI-II), and anxiety (BAI), (Table 4). PTCI total scale and subscales for each of the three models were significantly correlated with clinician-rated PTSD symptoms, and client’s self-reported PTSD symptoms and depression severity. The PTCI subscale of SELF was more strongly correlated with BDI-II (r =.74; 95% CI: .69–.77; p < .01) than the other subscales in the models (r’s ranged from .32 to .54; 95% CI ranged from .23–.40 to .47–.61; p < .01). For both models, the correlations were of comparable magnitude. The fourth PTCI subscale of COPE reported in Sexton’s model, was significantly correlated with depression, PTSD symptoms, clinician-rated and self-reported PTSD symptoms, and anxiety (r ranged from .23 to .53). Its correlation patterns were similar to that of SELF subscale in Sexton’s model, though at a lesser magnitude. Convergent validity studies thus support both models.
Table 4. Pearson correlation between the PTCI, trauma symptom measures, and other psychiatric measures (N=432)
CAPS-5, Clinician Administered PTSD Scale for DSM-5; PTCI, Posttraumatic Cognitions Inventory; BDI-II, Beck Depression Inventory-II; BAI, Beck Anxiety Inventory; BPRS, Brief Psychiatric Rating Scale; PANSS, Positive and Negative Syndrome Scale; QOL, Quality of Life Interview; WAI, Working Alliance Inventory; SCID II, Structured Clinical Interview for DSM-IV Axis-II Personality Disorders. *p≤.05, **p≤.01, ***p≤.001;
a n=307 (Study 1+Study 2 sample only);
b n=125 (Study 3 sample only);
c n=199 (Study 2 sample only);
d n=233 (Study 1 + Study 3 samples only).
To evaluate divergent validity, we examined correlations between PTCI factor scores, working alliance (WAI) and PTSD knowledge (K-PTSD; Table 4). The PTCI total score and subscale scores in all three models were weakly or not significantly correlated with working alliance or PTSD knowledge (r = –.06 to –.17), suggesting divergent validity of both PTCI models.
The internal consistency for the overall PTCI was .95 for the combined data sets from three studies (n=432) [.95 for Study 1 (n=108); .94 for Study 2 (n=199); .95 for Study 3 (n=125)]. The internal consistencies of the PTCI subscales for the three and four factor model were also high, ranging from .77 to .94 (Table 5).
Table 5. PTCI item CFA factor loadings (N=432)
Foa’s 3-factor model was based on 33 items, excluding 13, 32 and 34 consistent with (Foa et al., Reference Foa, Ehlers, Clark, Tolin and Orsillo1999; p. 314), and five pairs of error covariances (error covariance between item 3 and 12, 4 and 5, 5 and 14, 10 and 11, 25 and 36). Sexon’s 4-factor model was based on 34 items excluding 14 and 24 consistent with Sexton et al. (2018), and five pairs of error covariances (error covariance between item 3 and 12, 4 and 13, 10 and 11, 25 and 36, 32 and 33); COPE subscale includes items 4, 5, 13.
Discussion
Our study used CFA as well as multi-group CFA to assess two different models proposed by previous literature. Sexton et al. (Reference Sexton, Davis, Bennett, Morris and Rauch2018) found the superiority of the four-factor model over Foa’s three-factor model. We did not find support for the hypotheses that Sexton’s four-factor model would be superior to Foa’s three-factor model, and instead found that the four-factor model performed similarly to the three-factor model, with both having adequate fit. Our study offers support for Sexton’s four-factor model; however, support is also found for Foa’s three-factor model. Both models have adequate fit in this sample, which contributes to their credibility for continued use, although future research is warranted.
The findings from this study strongly support the reliability and validity of the PTCI in people with SMI and co-morbid PTSD. Furthermore, we found adequate support for the three-factor structure of the PTCI initially reported by Foa et al. (Reference Foa, Ehlers, Clark, Tolin and Orsillo1999), as well as for the four-factor model of Sexton et al. (2008).
Consistent with Sexton et al. (Reference Sexton, Davis, Bennett, Morris and Rauch2018), the current study found that the four subscales in the four-factor model of PTCI had good internal consistency. Moreover, the PTCI total scale and subscales also had moderate convergent correlations with measures of PTSD symptom severity, depression and anxiety. Evidence for divergent validity of PTCI was also found with weaker correlations with measures of PTSD knowledge and working alliance (r ranged from –.07 to –.16), a construct unrelated to post-traumatic cognition.
While the three-factor solution was found to show good fit, CFA analysis and multi-group CFA across the three diagnostic groups indicated that the four-factor solution had a better fit than the three-factor solution. Both models performed similarly across gender and race (White vs Black). Sexton’s model has previously been used with veterans seeking treatment for military trauma but not among individuals with SMI until this study. Similarly, this study is the first time Foa’s model was examined among individuals with SMI and co-occurring PTSD. With regard to PTCI subscales, results are consistent with Su and Chen (Reference Su and Chen2008) in that all subscale scores had moderate to high correlations with depression as measured by the BDI. The SELF subscale was the strongest predictor of depression and a stronger predictor of PTSD severity than the other scales, in both the three- and four-factor solutions. In contrast to Su and Chen (Reference Su and Chen2008), our study found that PTSD severity (CAPS and PCL) was most weakly correlated with the PTCI BLAME subscale. These findings are consistent with Daie-Gabai et al. (Reference Daie-Gabai, Aderka, Allon-Schindel, Foa and Gilboa-Schechtman2011), who reported that the BLAME subscale had the lowest correlations with PTSD severity and depression symptoms compared with the SELF subscale, which was the strongest predictor of PTSD severity and depression, followed by the WORLD subscale.
Interestingly, the COPE subscale in Sexton’s four-factor model was moderately correlated with PTSD symptoms, psychiatric symptoms, borderline personality functioning, and quality of life. Next to SELF subscale, COPE had the second strongest correlations with depression (BDI), anxiety (BAI), and PTSD symptoms (PCL) among all the subscales (r ranged from 0.22 to 0.54). The findings suggest the importance of addressing coping mechanisms in PTSD treatment due to its strong correlations with PTSD symptoms and overall psychiatric symptoms. The COPE subscale suggests that a perceived lack of competence in coping is also a strong predictor of borderline personality functioning, quality of life, and psychiatric symptoms among individuals with SMI and co-morbid PTSD. The findings are consistent with Mueser et al. (Reference Mueser, Rosenberg, Xie, Jankowski, Bolton, Lu, Hamblen, Rosenberg, McHugo and Wolfe2008, Reference Mueser, Gottlieb, Xie, Lu, Yanos, Rosenberg, Silverstein, Duva, Minsky, Wolfe and McHugo2015)’s treatment in which cognitive restructuring was used to help people with SMI to increase their competence in coping with upsetting situations. This is consistent with PTSD treatment guidelines in which non-trauma-focused therapies (i.e. stress inoculation training, present-centered therapy, and interpersonal psychotherapy) also facilitate recovery from PTSD (Cusack et al., Reference Cusack, Jonas, Forneris, Wines, Sonis, Middleton and Gaynes2016; Watts et al., Reference Watts, Schnurr, Mayo, Young-Xu, Weeks and Friedman2013).
Several important limitations should be noted. First, our study included clients with SMI and co-occurring PTSD who sought treatment. Therefore, these participants may have more severe psychiatric symptoms and higher scores on post-traumatic cognition than those without PTSD, or not currently in treatment. In Studies 1 and 2, participants’ PTSD was assessed using DSM-IV criteria. Future research with the updated PTSD criteria in the DSM-5 (American Psychiatric Association, 2013) is needed to verify that findings are generalizable. Additionally, divergent validity was examined using correlations of PTSD knowledge and working alliance rather than measures of different symptoms or cognitions, which future studies may utilize instead.
These limitations notwithstanding, the current study is the first to examine the factor structure of the PTCI among three heterogeneous samples of people with SMI and co-occurring PTSD, and adds to the knowledge of the psychometric properties of this measure with special populations. Even though the three-factor model (SELF, WORLD, BLAME) appeared robust with SMI clients, the four-factor models (SELF, WORLD, BLAME, COPE) provided an alternative and better fit for this population. Overall, the PTCI appears to be a reliable and valid tool for assessing trauma-related beliefs in people with SMI and co-occurring PTSD.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1352465823000140
Data availability statement
At the time that the grant was awarded, the funding agency did not require grantees to share data. We do not have participant consent or IRB approval to share data publicly.
Acknowledgements
We wish to thank the following individuals for their assistance with this project: Edward Kim, Lee Hyer, Rachael Fite, Kenneth Gill, Rosemarie Rosati, Christopher Kosseff, Karen Somers, John Swanson, Avis Scott, Rena Gitlitz, John Markey, Zygmond Gray, Marilyn Green, Janice Belarmino, William Waynor, Joseph Dephillips, Kevin Harris, Carol Gao, Audra Schulman and Ke Wang.
Author contributions
Weili Lu: Conceptualization (equal), Funding acquisition (equal), Investigation (equal), Methodology (equal), Project administration (lead), Supervision (lead), Writing – original draft (equal); Kim Mueser: Conceptualization (equal), Resources (equal), Writing – review & editing (equal); Philip Yanos: Conceptualization (equal), Investigation (equal), Writing – original draft (equal), Writing – review & editing (equal); Amanda Siriram: Writing – original draft (equal), Writing – review & editing (equal); Yuane Jia: Data curation (equal), Formal analysis (equal), Software (equal); Alyssa Leong: Formal analysis (equal); Steven Silverstein: Investigation (equal), Writing – review & editing (equal); Jennifer Gottlieb: Investigation (equal); Mary Jankowski: Investigation (equal).
Financial support
This research was supported by a National Institutes of Mental Health grant (R01MH064662; R01MH064662-02) and by the NIDILRR Field Initiated Grant no. HHS-Administration for Community Living number 90IF0074.
Competing interests
All authors declare that they have no competing interests.
Ethical standards
Authors abided by the Ethical Principles of Psychologists and Code of Conduct as set out by the BABCP and BPS. Research has conformed to the Declaration of Helsinski.
Open access
Comments
No Comments have been published for this article.