Measurement Nonequivalence of the Clinician-Administered PTSD Scale by Race/Ethnicity: Implications for Quantifying Posttraumatic Stress Disorder Severity

Abstract

Research studies suggest racial/ethnic differences in posttraumatic stress disorder (PTSD) diagnosis and symptom severity. Few studies to date, however, have examined the extent to which these findings are due to differences in measurement properties of existing PTSD scales. This study examined measurement equivalence across race/ethnicity in the Clinician-Administered PTSD Scale (CAPS) by testing for differential item functioning (DIF) in the item response theory (IRT) framework. Participants were 506 trauma-exposed women (M = 39.41 years, SD = 8.94) who participated in the National Drug Abuse Treatment Clinical Trials Network Women and Trauma Study. PTSD severity score estimates were improved upon as part of IRT estimation incorporating symptom “weights” (i.e., factor loadings) and group-specific DIF. Six symptoms from the CAPS showed DIF, with the majority of differences in measurement driven by White/African American and White/Latina differences, particularly for (a) avoidance of thoughts and (b) a sense of foreshortened future. Despite both racial/ethnic minority groups being slightly (not significantly) more likely to receive a PTSD diagnosis, African Americans (p = .014; Cohen’s d =−.22) and Latinas (p < .001; d =−.73) had significantly lower PTSD severity scores than Whites as estimated under IRT with group-specific DIF. Examination of PTSD severity scores based on symptom counts revealed these differences were either dampened (White/Latina difference d =−.39) or entirely negated (White/African American difference d =−.08). The findings suggest the importance of considering differences in symptom relevance across race/ethnicity and their impact on capturing symptom severity parallel to diagnostic criteria. Implications for clinical practice are discussed.

Disparities in posttraumatic stress disorder (PTSD) diagnosis, symptom expression, and severity across racial and ethnic groups have been well documented, yet findings have also been mixed. Epidemiological studies using probability samples indicate that African Americans and Hispanics/Latinos/Latinx (from here on referred to as Latinos or Latinas) have higher rates of exposure to traumatic events, greater risk of developing PTSD, higher rates of PTSD, and greater PTSD symptom severity than Whites (Dohrenwend, Turner, Turse, Lewis-Fernandez, & Yager, 2008; Himle, Baser, Taylor, Campbell, & Jackson, 2009; Roberts, Gilman, Breslau, Breslau, & Koenen, 2011).¹ In contrast, other studies with postdisaster and clinical samples have shown similar rates of trauma exposure and PTSD diagnosis among African Americans, Latinos, and White Americans (Adams & Boscarino, 2005; Ghafoori, Barragan, Tohidian, & Palinkas, 2012). No studies to date, however, have examined whether dimensions of measurement nonequivalence in clinicians’ evaluations may contribute to these conflicting findings. This current study therefore addressed this gap in the literature by examining whether a gold-standard PTSD clinical assessment tool and the clinicians using them exhibit different interpretations of symptom presence/absence across different racial/ethnic groups.

Two theories have been proposed to account for racial/ethnic disparities in the risk for, and prevalence of, PTSD: (a) differential exposure to types of traumas that confer greater risk for PTSD (e.g., violent homicides or rapes; Alim et al., 2006) and (b) differential vulnerability to the effects of traumatic events after exposure (Perilla, Norris, & Lavizzo, 2002). For example, racial/ethnic minorities are more likely to be exposed to more frequent and severe stressors (e.g., chronic poverty, high-crime neighbor-hoods, and racism/discrimination) due to their lower socioeconomic status, which may enhance their risk for the development of PTSD after trauma exposure (Hatch & Dohrenwend, 2007). Hinton and Lewis-Fernández (2011) conducted a systematic review of studies published from 1994 to 2011 that examined trauma, PTSD, culture, ethnicity, and race and concluded that the findings are “suggestive, but not conclusive, of cultural variability in the conditional probability of PTSD” (p. 789). However, these mixed findings may also be due to measurement bias of existing PTSD assessment measures across racial and ethnic groups.

An emerging body of research has examined measurement equivalence of the PTSD construct by gender (Chung & Breslau, 2008; Contractor et al., 2013), number of traumatic events (Contractor, Caldas, Dolan, Lagdon, & Armour, 2018), and trauma type (Chung & Breslau, 2008). Measurement equivalence refers to the degree to which a measure assesses the underlying construct similarly across various groups (Putnick & Bornstein, 2016).

Fewer studies have assessed the moderating impact of race/ethnicity on PTSD diagnosis and symptoms in general or on the measurement equivalence of the Clinician-Administered PTSD Scale (CAPS; Blake et al., 1995), the gold-standard clinical interview for PTSD (Weathers, Keane, & Davidson, 2001). Furthermore, the few studies that have examined the impact of race/ethnicity on measurement equivalence have focused on self-report questionnaires (Contractor, Caldas, Dolan, Natesan, & Weiss, 2019).

In a sample of Latino and White college students, Hoyt and Yeater (2010) tested the measurement equivalence of eight theoretical models of PTSD symptom structure as assessed using the PTSD Checklist-Civilian (Weathers, Litz, Herman, Huska, & Keane, 1993) and found equivalence for only three of the eight models. While Latinos and Whites had similar factor loadings within those three models, at the item level, Whites had significantly greater mean scores on two items (B4 [upset when reminded] and C5 [feeling distant]) compared with Latinos with the same level of underlying PTSD severity, suggesting the need for consideration of cultural factors in the manifestation and assessment of those specific symptoms. In contrast, Marshall, Schell, and Miles (2009) compared Latinos, non-Hispanic Whites, and African Americans who were physically injured and seeking treatment at a trauma center. The authors found that, compared with non-Hispanic Whites, Latinos reported greater symptom severity on PTSD symptoms pertaining to hypervigilance, intrusive thoughts, flashbacks, and emotionality. The mechanisms underlying these findings remained unclear (Marshall et al., 2009). Relatedly, Contractor and colleagues (2015) utilized confirmatory factor analysis with invariance testing to examine whether race and ethnicity moderated a five-factor model of PTSD—as assessed by the self-report University of California-Los Angeles PTSD Reaction Index (UCLA PTSD-RI; Steinberg, Brymer, Decker, & Pynoos, 2004)— among children and adolescents referred for clinical treatment after trauma exposure. Results revealed there was measurement equivalence for the five-factor model and similar item-level PTSD symptom severity across racial (Caucasian vs. non-Caucasian) and ethnic groups (Latino vs. non-Latino), providing support for the use of the UCLA PTSD-RI across these subpopulations. Finally, Tiemensma and colleagues (2018) assessed the measurement invariance of a one-factor model of PTSD as assessed by the self-report Impact of Events Scale-Revised total score (IES-R; Weiss, 2004) in a sample of 545 undergraduate college students who were recently exposed to campus violence. They found equivalent performance of the IES (total and subscale scores) between Latino and non-Latino participants, indicating similar measurement of the construct of PTSD across these subgroups. These studies present a distinct constellation of findings regarding which specific PTSD symptoms, if any, exhibit measurement nonequivalence across various racial/ethnic groups.

While having important implications for clinical practice, the aforementioned studies were all based on self-report questionnaires of PTSD, which are subject to biases related to recall and differential interpretation and understanding of the questions asked for the construct being assessed (Burlew, Feaster, Brecht, & Hubbard, 2009). Self-report and clinical interview measures also have important differences in question stems, in item response scaling, in whether there are clarifying prompts that redirect patients to the target traumatic event, and in the manner in which questions are converted from frequency/severity items to symptoms and Diagnostic and Statistical Manual of Mental Disorders (DSM–5; American Psychiatric Association, 2013) diagnoses—all of which may influence participants’ responses and scores (Lunney, Schnurr, & Cook, 2014; Monson et al., 2008; Palmieri, Weathers, Difede, & King, 2007). Moreover, the combination of multiple racial groups into a non-Caucasian category (e.g., Contractor et al., 2015) and the combination of multiple racial and cultural subgroups in a non-Latino subgroup category (e.g., Tiemensma et al., 2018) potentially obscures any within-groups differences that may exist among those participants. Additionally, the focus on college and child/adolescent samples (e.g., Contractor et al., 2015; Tiemensma et al., 2018) limits the generalizability of these findings to adult psychiatric populations.

Many of the applications listed above use PTSD severity scores (i.e., sum scores) that do not account for variation in symptom weighting nor consider how different symptom weights may be needed across racial and ethnic groups. Approaches such as item response theory (IRT) allow examination of invariance at the symptom level (Embretson & Reise, 2000). This is particularly important in clinical interviews where clinicians are gathering responses from patients and making a clinical judgment about whether the frequency and intensity of patients’ symptoms meet a certain level of severity in order to establish the presence/absence of the symptom (Weathers et al., 2001). Clinician judgments are often based on prior training, experience, perceptions, and interpretation of the patient’s response, wording of questions and answers, nonverbal behavior, and/or bias (Gray-Little, 2009; Meade & Lautenschlager, 2004).

Differences or similarities in rates of PTSD diagnosis, symptom expression, and severity across racial/ethnic groups may thus be due to several factors: (a) underlying differences in trauma-related psychopathology (i.e., true differences in the expression of psychopathology linked to sociocultural variables) across racial/ethnic groups (true differences), (b) inability of the PTSD measure to assess PTSD similarly across racial/ethnic groups even if trauma-related psychopathology is similar across racial/ethnic minority groups (psychometric biases), and (c) clinician interpretation of symptoms and how they affect functionality even if trauma-related psychopathology is expressed similarly across racial/ethnic minority groups (i.e., clinician biases). Social distance theory suggests that racial/ethnic or class differences between clinicians and clients may contribute to clinician difficulties in accurately hearing and understanding clients’ symptoms despite similarities in underlying symptoms (Gray-Little, 2009). Understanding more about how race/ethnicity true differences, psychometric biases, and clinician biases influence the PTSD assessment process (at the overall PTSD diagnosis and symptom level) is critical in providing assessment and treatment that is attuned to the nuances in PTSD symptom presentation across racial/ethnic groups.

Based on these alternative explanations, we aimed to conduct a rigorous test of measurement equivalence using the CAPS across racial/ethnic groups using IRT (Embretson & Reise, 2000). The National Drug Abuse Treatment Clinical Trials Network Women and Trauma Study (CTN-0015)—a multisite clinical trial that tested the efficacy of Seeking Safety (an integrated treatment for co-occurring PTSD and substance use disorders [SUDs]; Najavits, 2002) compared with Women’s Health Education (a control condition; Miller, Pagan, & Tross, 1998)—provided a unique opportunity to examine measurement equivalence in CAPS diagnostic rates (based on DSM–IV–TR diagnostic criteria; American Psychiatric Association, 2000) and clinician-rated PTSD symptom severity scores (at the item and total symptom severity level) by race/ethnicity (Hien et al., 2009). Although there have been changes in the DSM–5 to the diagnostic criteria for PTSD, there have been no studies to the authors’ knowledge that have examined CAPS measurement equivalence with a diverse sample. Thus, the current study adds to the literature by examining this question in a racially/ethnically diverse sample of trauma-exposed women with SUDs who were assessed for PTSD by a trained clinician and provides a foundation for similar empirical studies with measures of DSM–5 PTSD.

Method

Participants

Participants in the current study were 543 women who met initial eligibility for inclusion into the randomized controlled trial (RCT) based on a brief screener using PTSD diagnostic criteria. Three hundred fifty-three of the women met inclusionary criteria for the RCT by having at least one lifetime traumatic event and a DSM–IV diagnosis of either full or subthreshold PTSD in the past 30 days. For this study, baseline data from the women in the RCT were combined with data from an additional 190 women who did not have a diagnosis of full or subthreshold PTSD (and thus did not meet inclusion criteria to enroll in the RCT) but otherwise met all other criteria (see additional inclusion criteria below). Subthreshold PTSD was defined as meeting Criterion B (reexperiencing the trauma) and either Criterion C (avoidance of trauma reminders) or D (hyperarousal) instead of both. Other inclusion criteria were (a) being a female, (b) being 18–65 years of age, (c) using alcohol or illicit substances within the past 6 months, and (d) meeting a current (within the past year) DSM–IV–TR diagnosis of substance abuse or dependence. Exclusionary criteria were presence of (a) impaired mental cognition as indicated by a Mini-Mental Status Exam (Folstein, Folstein, & McHugh, 1975) score of < 21, (b) significant risk of suicidal/homicidal behavior (current plan, plan in the past 6 months, or attempt in the past 6 months), (c) history of schizophrenia-spectrum diagnosis, or (d) active (past 2 months) psychosis. All procedures were reviewed and approved by institutional review boards associated with the lead research team and each treatment site, and all patient-participants gave written informed consent. See Hien et al. (2009) for more detailed information on the procedures and methods of the RCT.

In brief, interested individuals were assessed for eligibility during a screening assessment and then completed a baseline interview. After the baseline assessment, participants were randomized into one of two treatment groups: Seeking Safety (Najavits, 2002) or Women’s Health Education (a control condition; Miller et al., 1998). Women in both conditions completed brief, weekly assessments of PTSD symptoms, substance use, and service utilization during treatment and were reassessed using the full assessment battery 1 week, 3 months, 6 months, and 12 months posttreatment. For this study, we only utilized data from the baseline assessment. There were no significant differences in baseline demographic, psychiatric, or trauma-related symptoms between those assigned to Seeking Safety or Women’s Health Education (Hien et al., 2009).

Of the 543 overall participants, 259 were non-Hispanic Whites, 202 were non-Hispanic African Americans, 45 were Latinas, and 37 were Others (including Asian Americans, Alaska Natives/Pacific Islanders, and mixed-race participants). The Other group was excluded from analyses as the group was too heterogeneous from a measurement perspective.

Measures

Sociodemographics.

Basic demographic data, including age and self-identified race/ethnicity, were collected at the screening assessment. Marital status and education level were collected at the baseline assessment.

Life Events Checklist (LEC).

The LEC (Gray, Litz, Hsu, & Lombardo, 2004), a 17-item self-report measure, was used to assess exposure to 16 potentially traumatic events (PTE; e.g., physical assault, sexual assault, witnessing violence, life-threatening illness or injury, etc.) that might lead to PTSD and one additional item that captures any other very stressful life event or experience not assessed by the first 16 items. The LEC was administered before the CAPS at pretreatment, posttreatment, and all follow-up time points, and CAPS questions were asked in relation to up to three PTEs identified in the LEC. While there is no formal scoring criteria for the LEC (Blake et al., 1995), researchers have utilized the LEC in various ways including assessing the number of different types of PTE participants have been exposed to or the degree of exposure to a particular PTE (e.g., happened to the person, person witnessed the event, or person learned about the event). The LEC has demonstrated good convergent validity with the Traumatic Life Events Questionnaire, and temporal stability of the LEC has been established (Gray et al., 2004). In this analysis, the LEC was used to assess the total number of types of PTEs across race/ethnicity groups.

Clinician-Administered PTSD Scale for DSM–IV (CAPS-IV).

The CAPS-IV (Blake et al., 1995), a structured clinical interview, was used to measure the frequency and intensity of 17 PTSD symptoms experienced in the previous 30 days and to determine PTSD diagnosis and symptom severity. The CAPS-IV has three symptom cluster subscales: (a) reexperiencing, (b) avoidance/numbing, and (c) hyperarousal. A DSM–IV diagnosis of PTSD requires the presence of a Criterion A trauma, at least one reexperiencing symptom, three avoidance/numbing symptoms, and two hyperarousal symptoms. For this analysis, PTSD symptoms were coded in a binary fashion for presence or absence of the symptom based on a symptom frequency of at least once in the previous week and symptom intensity that was at least moderate. This rating criterion is a modification to accommodate symptom-level equivalence in endorsement to the PTSD Symptom Scale— Self-Report; Blake et al. (1995) advocated for a frequency criterion of once in the previous month for the CAPS. PTSD severity scores were estimated using IRT.

During the original RCT, independent assessors, who held at least a master’s degree with clinical experience including diagnostic skills, received a 1-day expert-led training on administering and scoring the CAPS and had weekly conference calls with the lead team to maintain competency and interrater reliability on the measure. Reliability of diagnoses were conducted by reviewing 10% of all baseline and 10% of all follow-up assessments. Kappas on diagnosis and intraclass correlations on severity ratings were computed between the independent assessors and expert raters. Raters were expected to have a .70 level of agreement. If agreement levels fell below .70, the supervisor conducted joint rated interviews with the independent assessors until .70 level was achieved in three consecutive interviews. Independent assessors were blind to the treatment condition of any participant at any time during the course of the study.

Substance use disorders.

Patients were assessed for lifetime and current (in the past month) SUDs at the screening assessment using the Composite International Diagnostic Interview for DSM–IV (Robins et al., 1988). The CIDI is a fully structured, interviewer-administered measure with demonstrated reliability and validity (Robins et al., 1988).

Data Analysis

Covariates.

Demographic covariates included age, marital status (married vs. unmarried), and educational status (completed college vs. not). These covariates were examined to see if mean differences and/or DIF across race/ethnicity were reduced after including them in the models. Focal substances (alcohol, opioids, and cocaine) were also examined as potential covariates, independent of race/ethnicity effects, on (a) latent PTSD severity and (b) each PTSD symptom conditional on latent PTSD severity (i.e., measurement bias/DIF).

Multiple-group item response theory (MG-IRT).

Mplus Version 8 (Muthén & Muthén, 1998–2017) was used to fit a series of MG-IRT models estimated under robust weighted least squares estimation (WLSMV with delta parameterization). First, an initial base MG-IRT model with varying factor loadings and thresholds across the 17 symptoms was fit, where item parameters for each symptom were constrained to equality across the four race/ethnicity groups. The factor mean was set to zero, and the variance was set to one for the group with the largest N (Whites) and freely estimated in all other groups. This model fit the data moderately well; the comparative fit index (CFI) was well above the .90 threshold for good fit (CFI = .953), while the root mean square error of approximation (RMSEA) was slightly above the .05 threshold for good fit, RMSEA = .054, 90% CI [.045, .063], suggesting a model assuming measurement equivalence across measures could be improved upon by incorporating group-specific DIF.

Next, a series of 17 models were fit where factor loadings and intercepts were allowed to vary across the four groups to test for symptom-specific DIF. Models were compared with the base model via likelihood ratio tests where −2 times the log-likelihood values from the base and the focal model was taken and evaluated against a six df χ² distribution as a test of differences in item parameters across race/ethnicity groups, analogous to an omnibus test for group differences in an analysis of variance (ANOVA) with more than two groups. These results for symptom-specific DIF are shown in Table 1, with 11 out of the 17 symptoms showing no DIF of any sort; of the six symptoms that showed DIF for either factor loadings, item intercepts, or both, group-specific item characteristic curves are shown in Figures 1–6. See also Table 2 for clinician ratings of symptoms by race/ethnicity. Note that, after inclusion of SUD, types of traumas (that showed differences across race/ethnicity), and demographic covariates for DIF tests, none of the inferences regarding DIF across race/ethnicity differed. Omnibus tests of DIF across race/ethnicity were followed by pairwise group contrasts for differences in factor loadings and item intercepts (conditional on group differences in PTSD severity factor scores) using the Mplus “model constraint” command; these tests are analogous to post hoc tests in ANOVA.

Table 1.

Omnibus Differential Item Functioning Testing

Symptom	χ2(6) – critical value = 12.59	p
Reexperiencing symptoms
Intrusive recollections	8.083	.23
Dreams	7.49	.27
Event reoccurrence	11.91	.063
Psychological cues	5.69	.45
Physiological cues	12.2	.057
Avoidance symptoms
Thought avoidance	22.597	<.001
Activity avoidance	10.47	.1
Inability to recall	15.73	.015
Diminished interest	8.55	.2
Detachment	14.53	.02
Restricted affect	10.73	.09
Foreshortened future	14.12	.02
Hyperarousal symptoms
Sleep problems	8.9	.17
Irritability	6.66	.35
Concentration problems	4.5	.6
Hypervigilance	16.04	.01
Startle	21.67	.001

Note. The items bolded are the 6 scale items that showed differential item functioning.

Figure 1.

Item characteristic curves. Predicted endorsement probabilities across race/ethnicity for thought avoidance. African Americans = Latinas. PTSD = posttraumatic stress disorder.

Figure 6.

Item characteristic curves. Predicted endorsement probabilities across race/ethnicity for exaggerated startle. Whites = African Americans. All other pairwise comparisons are significant. PTSD = posttraumatic stress disorder.

Table 2.

Symptom Endorsement Percentages and Demographics by Race/Ethnicity

Variable	Whites % (n = 259)	African Americans % (n = 202)	Latinas % (n = 45)
Intrusive recollections	55.21	47.52	51.11
Dreams	27.41	24.26	22.22
Flashbacks	16.99	13.37	6.67
Psychological cues	48.26	40.10	26.67
Physical cues	38.61	32.67	20.00
Thought avoidance	53.28	55.45	40.00
Activity avoidance	34.75	40.59	31.11
Inability to recall	36.68	26.23	31.11
Diminished interest	43.24	48.51	26.67
Detachment	59.85	63.86	31.11
Diminished interest	62.55	52.48	40.00
Foreshortened future	22.39	22.28	22.22
Sleep problems	60.23	58.91	46.67
Irritability	50.97	48.01	37.78
Concentration problems	60.23	49.50	40.00
Hypervigilance	43.62	59.41	57.78
Exaggerated startle	31.27	26.23	37.78
Age M (SD)	38.5 (9.4)	40.7 (7.9)	36.9 (9.4)
Completed college	15.38	5.63	8.33
Married	23.62	12.67	8.33
Past month alcohol disorder	36.29	49.50	57.78
Past month cocaine disorder	53.28	50.99	64.44
Past month opioid disorder	64.09	80.26	68.89

Note. M = mean; SD = standard deviation.

Results

Demographics

The mean age of the sample by race/ethnicity was as follows: non-Hispanic Whites (M = 38.51, SD = 9.40), African Americans (M = 40.73, SD = 7.96), and Latinas (M = 36.88, SD = 9.41). African American participants were significantly older than White and Latina participants (p < .05). White participants had a higher percentage of women who completed college (15.38%) and were married (23.62%) compared with African American participants (5.63% and 12.67%, respectively, ps < .05). See Table 2.

Differential Item Functioning Across Race/Ethnicity

Thought avoidance.

Figure 1 shows the item characteristic curves for thought avoidance across the three racial/ethnic groups. The item parameters did not differ between African Americans and Latinas. The factor loading for African Americans (p = .039) and for Latinas (p < .001) was significantly higher than for Whites, suggesting thought avoidance was more strongly related to PTSD severity for African American and Latina than White women.

Inability to recall.

Figure 2 shows the item characteristic curves for inability to recall across the three groups. The factor loading for Latinas (p = .002) was significantly higher than for African Americans, suggesting inability to recall was less indicative of PTSD severity for African Americans. The item intercept was higher for African Americans, suggesting that at the mean level of PTSD severity, African Americans have a significantly higher probability of clinician-rated inability to recall than Latinas (p = .04).

Figure 2.

Item characteristic curves. Predicted endorsement probabilities across race/ethnicity for inability to recall. All pairwise comparisons nonsignificant except African Americans ≠ Latinas. PTSD = posttraumatic stress disorder.

Detachment.

Figure 3 shows the item characteristic curves for detachment across the three groups. Factor loadings did not differ significantly across the three groups. However, several differences in item intercepts emerged across groups: Latinas had significantly higher probabilities of clinician-rated detachment (at the mean level of PTSD severity) than Whites (p = .030) and African Americans (p = .001). Whites had significantly higher probabilities of clinician-rated detachment (at the mean level of PTSD severity) than African Americans (p = .004).

Figure 3.

Item characteristic curves. Predicted endorsement probabilities across race/ethnicity for detachment. All pairwise comparisons significant. PTSD = posttraumatic stress disorder.

Sense of foreshortened future.

Figure 4 shows the item characteristic curves for sense of foreshortened future across the three groups. The factor loading for Whites was significantly higher than for African Americans (p = .042) and higher than for Latinas but nonsignificant (p = .067), suggesting a sense of a foreshortened future was less indicative of PTSD severity for African Americans and Latinas than Whites.

Figure 4.

Item characteristic curves. Predicted endorsement probabilities across race/ethnicity for foreshortened future. Both comparisons with Whites are significant. PTSD = posttraumatic stress disorder.

Hypervigilance.

Figure 5 shows the item characteristic curves for hypervigilance across the three groups. Factor loadings did not differ significantly across the three groups. However, item intercepts did differ such that Whites were significantly more likely to receive a clinician rating of hypervigilance than African Americans (p < .001) and Latinas (p < .001) at the mean level of PTSD severity.

Figure 5.

Item characteristic curves. Predicted endorsement probabilities across race/ethnicity for hypervigilance. Whites differ significantly from African Americans and Latinas. All other comparisons nonsignificant. PTSD = posttraumatic stress disorder.

Exaggerated startle response.

Figure 6 shows the item characteristic curves for exaggerated startle response across the three groups. The factor loading for Whites was significantly higher than for Latinas (p = .035), suggesting that exaggerated startle response was a stronger indicator of PTSD severity for Whites than for Latinas. Item intercept differences also emerged such that Latinas had significantly lower probabilities of clinician-rated exaggerated startle response (at the mean level of PTSD severity) than African Americans (p = .036).

Group Mean Differences in Trauma Exposure and PTSD Severity

There were no significant racial/ethnic differences in total number of traumatic event exposures; however, there were significant differences in type of trauma exposure. Specifically, compared with Whites and Latinas, African Americans were more likely to report exposure to physical assault, assault with a weapon, and witnessing a violent death compared with Whites. A greater proportion of Whites reported exposure to motor vehicle accidents and severe suffering than African Americans and Latinas.

Differences in PTSD severity were examined across multiple scoring approaches that varied in (a) the incorporation of different item parameters across symptoms (i.e., varying factor loadings across symptoms) and (b) whether the item parameters varied across race/ethnicity groups. The means and standard deviations for the PTSD severity scores are shown in Table 3. It is noted that the intercorrelations between the total symptom count scores, IRT measurement equivalence (IRT-ME) scores, and IRT-DIF scores are all .97 or higher. Despite whatever group differences there are across PTSD scoring approaches, there are no significant differences between groups in PTSD diagnosis rates.

Table 3.

PTSD Scores on the CAPS by Race/Ethnicity

Variable	Whites (n = 259) M (SD)	African Americans (n = 202) M (SD)	Latinas (n = 45) M (SD)	Significant post hoc at p < .05
Full PTSD diagnosis (%)	68.86	74.72	73.53	None
CAPS symptom endorsement counts	7.4556 (4.7862)	7.0941 (4.2328)	5.6889 (4.2684)	W > L
CAPS IRT score (assuming full measurement equivalence)	−0.0354 (0.8994)	−0.1518 (0.9984)	−0.6088 (1.0707)	W > L, AA > L
CAPS IRT score (with DIF incorporated for 6 symptoms)	−0.03064 (0.8999)	−0.2577 (1.0396)	−0.7725 (1.1046)	W > AA, W > L, AA > L

Note. CAPS = Clinician-Administered PTSD Scale; IRT = item response theory; DIF = differential item functioning; PTSD = posttraumatic stress disorder; W = Whites; AA = African Americans; L = Latinas.

CAPS total symptom counts.

CAPS symptom counts for Whites were significantly higher than for Latinas, b = 1.76 (.73), t = 2.41, p = .016, Cohen’s d = .38, but not African Americans, p = .39, d = .07. African Americans had higher total symptom counts than Latinas, b = 1.40 (.75), t = 1.88, p = .06, Cohen’s d = .31, but this was nonsignificant, p = .05. Nevertheless, the Cohen effect size exceeded the |.20| convention for what would be considered a small, meaningful effect.

CAPS IRT assuming measurement equivalence.

CAPS IRT-ME scores for Whites were significantly higher than for Latinas, b = .57 (.16), t = 3.68, p < .001, Cohen’s d = .58, but not African Americans, p = .19, d = .11. African Americans were higher on CAPS IRT-ME scores than Latinas, b = .46 (.16), t = 2.87, p = .004, Cohen’s d = .46.

CAPS IRT with group-specific DIF (no covariates).

CAPS IRT-DIF scores for Whites were significantly higher than for African Americans, b = .22 (.09), t = 2.45, p = .014, Cohen’s d = .22, and Latinas, b = .74 (.16), t = 4.66, p < .001, Cohen’s d = .74. African Americans were significantly higher on CAPS IRT-DIF scores than Latinas, b = .51 (.16), t = 3.17, p = .002, Cohen’s d = .51.

CAPS IRT with group-specific DIF (with covariates).

The same set of mean differences in CAPS IRT-DIF scores were examined while incorporating age, marital status, education level, trauma type, and SUD status (i.e., alcohol, opioid, cocaine, stimulants, sedatives) as covariates. Whites remained significantly higher than African Americans, b = .21 (.08), t = 2.63, p = .008, Cohen’s d = .21, with virtually no change in effect size compared with the CAPS-IRT model with no covariates, d = .21 vs. .22; this suggests that these differences are not better accounted for by demographic covariates, differences in SUD status, or trauma type. Whites also remained significantly higher than Latinas, b = .33 (.15), t = 2.29, p = .022, Cohen’s d = .33, though the effect size for White/Latino differences was substantially reduced, d = .33 vs. .74, suggesting that some of the differences are partially explained by covariates. African American/Latina differences were nonsignificant, b = .12 (.15), t = .82, p = .41, Cohen’s d = .12, in contrast to the analysis without covariates, p = .001, d = .51, suggesting that African American/Latina differences are entirely explained by covariates.

Discussion

The current study examined measurement equivalence of the CAPS-IV between White and racial/ethnic minority women. Findings revealed significant differences in item functioning across several PTSD symptoms, providing evidence that several CAPS items operate differently for African American and Latina women than White women, which, after scoring under IRT, likely account for larger differences in PTSD severity between White and African American and Latina women. While the hallmark symptoms of PTSD, such as reexperiencing symptoms, did not differ across White and racial/ethnic minority women, significant differences were seen in the avoidance/numbing and hyperarousal symptom clusters. Compared with White women, for African American and Latina women, thought avoidance was more strongly related to PTSD severity, and sense of foreshortened future was less strongly related to PTSD severity. African American women had a lower probability of receiving a clinician rating of inability to recall and detachment than Latina women, and White women also had a lower probability of clinician-rated detachment than Latina women. Compared with African American and Latina women, White women were more likely to receive a clinician rating of hypervigilance, and their exaggerated startle response was more strongly related to PTSD severity. These findings suggest potential differences in severity/threshold levels for certain clinician ratings. These specific symptom differences provide some potential guide-lines regarding differential treatment targets.

PTSD severity scores were improved upon by incorporating group-specific DIF and showed larger differences between African American/Latina women and White women than with severity scores based on symptom counts. African American and Latina women had significantly lower PTSD severity scores than White women as estimated under IRT with group-specific DIF; these same differences were either reduced, nonsignificant, or both when using symptom counts, highlighting the importance of considering differences in symptom relevance across race/ethnicity and their potential impact on capturing symptom severity. After controlling for covariates such as age, education, marital status, trauma type, and SUD in the group-specific DIF model, the PTSD severity differences between the White and African American women remained unchanged. In contrast, the differences between White and Latina women were reduced, although they remained significant, suggesting that demographic and clinical variables are partially responsible for some of the differences between White and Latina women. The differences between African American and Latina women were eliminated after controlling for covariates.

It is notable that the differences in PTSD severity scores between Whites and African Americans/Latinas emerged despite PTSD diagnostic rates that were similar across the groups. The lower severity rates among African American and Latina women compared with White women run counter to findings from epidemiological (Roberts et al., 2011) and clinical samples (Ghafoori et al., 2012) that show either higher or equal PTSD severity rates, respectively, for racial/ethnic minorities compared with Whites. However, those studies did not account for differential symptom weighting, which may have contributed to differential findings. A number of explanations may be offered for the compelling findings from the current analyses including true racial/ethnic variations in symptom presentation (type and severity), as well as possible clinician biases in the assessment and diagnostic process.

It is possible that the type and severity of PTSD symptoms themselves occur at varying frequencies and intensities in distinct racial/ethnic groups as a function of culturally relevant risk and protective factors or that their presence genuinely influences PTSD severity distinctly (Hinton & Lewis-Fernández, 2011). For example, a sense of a foreshortened future and exaggerated startle response were both more indicative of PTSD severity in White women than in African American or Latina women. Given the chronic stressors and micro- and macroaggressions faced by racial/ethnic minorities (e.g., Alim et al., 2006; Turner & Avison, 2003), individuals from these groups may be more likely to already experience these symptoms outside of the context of traumatic event exposure or PTSD. If individuals from racial/ethnic minority groups already exhibit elevated baseline levels of these symptoms, then they may not present with an exacerbation of these symptoms following traumatic event exposure, and their influence on PTSD severity would thus be minimized. This latter explanation would suggest that the way PTSD is experienced in and of itself varies across racial/ethnic groups, rather than clinician interpretation or bias per se. Conversely, researchers have argued that there may be limitations in the cultural validity of the individual PTSD symptoms themselves or in the items created to assess those symptoms, which may contribute to differences in the presence, salience, and severity of particular symptoms (Hinton & Lewis-Fernández, 2011; Rogler, 1999). For example, studies have shown cross-cultural variation in the avoidance/numbing clusters such that in non-Western cultures (e.g., Vietnam, Cambodia), there were lower endorsements of thought avoidance, which led to lower prevalence rates of PTSD within those cultures compared with Western cultures (Hinton & Lewis-Fernández, 2011). These cultures may have different ways of coping with or handling intrusive thoughts (e.g., meditation). It is interesting to note, however, that in our U.S. sample, thought avoidance was more strongly associated with PTSD for African American and Latina women compared with White women. Finally, it is possible that the CAPS scores may not capture all of the culturally relevant responses to a range of traumatic events.

Clinician biases may also partially account for some of these findings. According to social distance theory, racial/ethnic or class differences between clinicians and clients may contribute to clinician difficulties in accurately hearing and understanding clients’ symptom reports. Clinicians may also differentially weight or rate particular symptoms based on various client characteristics such as race/ethnicity, gender, age, and so forth, contributing to differences in overall severity ratings (Gray-Little, 2009). In the current study, clinicians may have inadvertently failed to recognize or accurately interpret the severity of several PTSD symptoms in African American or Latina women compared with White women. Indeed, CAPS symptom counts were significantly higher for White than for Latina women but not for African American women. There was also a trend for African American women to have greater total symptom counts than Latina women. It is possible that clinicians may be more likely to inadvertently disregard or misinterpret distress when reported by Latina women. Alternatively, the Latina women either may have had lower symptom counts secondary to culturally relevant protective factors or may have minimized particular symptoms secondary to intrapersonal concerns (e.g., stigma, fears of being perceived as weak, or cultural expectations about what is or is not appropriate to share). Racial/ethnic minority women also may be reluctant to disclose symptom severity to clinicians whom they may perceive as potentially biased, thus leading to lower symptom severity ratings (Hopkins & Shook, 2017). Some of this reluctance may also result in discrepancies between patient-reported and clinician-reported PTSD symptom severity measures, which may influence treatment outcome effect sizes (Morgan-López et al., 2020; Ruglass, Papini, Trub, & Hien, 2014). Moreover, although the differences between African American and White women remained unchanged and significant, even after controlling for various covariates (age, education, marital status, trauma type, and SUD), the fact that the difference between White and Latina women was reduced (although remained significant) highlights the importance of considering factors other than race/ethnicity that may help explain apparent group differences.

Although there have been changes in DSM–5 criteria that warrant continued research in this area, this is the first study to explore racial/ethnic differences in clinician ratings of PTSD symptom and PTSD severity using IRT. Changes made to the PTSD diagnostic criteria under the DSM–5 included (a) the separation of PTSD from other anxiety disorders; (b) the addition of a new Criterion D (negative alterations in cognitions and mood), which includes symptoms such as persistent negative emotional states and distorted cognitions regarding the cause and consequences of the traumatic event; (c) the addition of a symptom for self-destructive behaviors to the hyperarousal cluster; and (d) the removal of the requirement that the emotional response to the trauma involved fear, helplessness, or horror. However, in studies that have compared symptom endorsement and diagnosis rates, the comparison of symptoms that remain common between the DSM–IV and DSM–5 within the same patients have a concordance rate equivalent to or better than what would have been expected for test-retest reliability with either system, and the three new symptoms do not appear to contribute much in additional clinical utility above and beyond the 17 symptoms that overlap across the DSM–IV and DSM–5 (e.g., Hoge, Riviere, Wilk, Herrell, & Weathers, 2014). Thus, while the analyses presented in this article utilized the DSM–IV measure of PTSD, the specific clinical implications of this work for researchers studying current conceptualizations of PTSD under the DSM–5 are likely not compromised. Our findings represent an important foundation upon which similar empirical studies with DSM–5 PTSD criteria can build.

Limitations

Although the findings suggest some DIF in a gold-standard PTSD assessment tool by race/ethnicity, the sources of the bias remain unclear. Specifically, it remains unknown whether the observed findings occurred as a result of true racial/ethnic differences (e.g., in types of trauma exposure or cultural factors, etc.), clinician bias, psychometric bias, or some combination of both. Given the small sample size for the Latina subgroup, findings on the differences between White and Latina women should be considered exploratory. Moreover, the specific item parameters that were estimated here should not be used to score a separate racial/ethnic subsample. These findings serve as proof of concept for future research, with larger sample sizes, that may disentangle these effects through a combination of methodologies (e.g., multirater assessments and identification of potential differences in trauma type or intensity to rule out psychometric bias and use of racial/ethnically matched raters to rule out clinician bias and true racial/ethnic differences). The utilization of more fine-grained measures of racial-identity development is also recommended given the limitations of using a categorical definition of race/ethnicity. For example, depending on the level of racial-identity development of the participant, the race/ethnicity of the assessor may be more or less salient/important to that person (Helms & Carter, 1991). Examination of within-group variables such as racial/ethnic identity may also allow for a more nuanced exploration and understanding of the relationship between race/ethnicity and measurement equivalence.

Given the disproportionately high rates of PTSD in women and associated SUDs, the focus on women is a strength of this study (Olff, 2017). However, future studies are needed to determine whether similar measurement bias issues exist in diverse male samples, among those who do not identify as women, and among those without co-occurring SUDs. As with any study examining racial/ethnic differences, the need to have sample size and proper representation of a diversity of racial/ethnic minority groups is critically important. Future studies that can provide more specific comparisons by racial/ethnic minority group would be indicated. Finally, as we have discussed above, given that the results are based on the DSM–IV PTSD diagnostic structure, findings should be replicated with the CAPS based on DSM–5.

Clinical Implications

While the source of measurement invariance across racial/ethnic groups remains unclear, clinicians and clinical treatment researchers using the CAPS should implement it with the awareness that it may produce discrepant responses across racial/ethnic groups, though IRT and related approaches are designed to mitigate these very discrepancies. Clinician bias is an important clinical concern when using interviewer-rated PTSD assessments such as the CAPS. Training in understanding and reducing unconscious bias as well as increasing multicultural competence among service providers is an important remedy and has received increasing attention in the clinical field over the past decade (Burgess, Beach, & Saha, 2017; White & Stubblefield-Tave, 2017; Wilson & Tang, 2007). Increasing clinicians’ knowledge and awareness that racial/ethnic minority women may be reluctant to disclose symptom severity, and particularly so for avoidance symptoms, to those they may perceive as potentially biased is an important component as well. Clinicians are encouraged to remain mindful of these potential dynamics and to be willing to explore the interpersonal process with their clients if there are any subtle or overt indications of mistrust that may influence the assessment process (Safran, Muran, & Eubanks-Carter, 2011).

These findings also have important implications for considering treatment approaches targeting specific PTSD symptoms that vary by racial/ethnic group that can help to personalize and optimize treatment outcomes for racial/ethnic minorities. A better understanding of the mechanisms involved in the differences in racial/ethnic responses demonstrated in the current study will help improve detection of PTSD symptoms in diverse populations. These differences may also have implications for treatment interventions that target the most distressing PTSD symptoms in racial/ethnic minorities (Chapman, Delapp, & Williams, 2014; Hinton & Lewis-Fernández, 2011).

Public Significance Statement.

The current study examined measurement equivalence across race/ethnicity in the Clinician-Administered PTSD Scale by testing for differential item functioning in the item response theory framework. The findings suggest that considering differences in posttraumatic stress disorder (PTSD) symptom relevance across race/ethnicity is critical for increasing accuracy in diagnostic criteria and estimation of PTSD severity.