An Empirical Crosswalk for the PTSD Checklist: Translating DSM-IV to DSM-5 Using a Veteran Sample

Abstract

The fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) introduced numerous revisions to the fourth edition’s (DSM-IV) criteria for posttraumatic stress disorder (PTSD), posing a challenge to clinicians and researchers who wish to assess PTSD symptoms continuously over time. The aim of this study was to develop a crosswalk between the DSM-IV and DSM-5 versions of the PTSD Checklist (PCL), a widely used self-rated measure of PTSD symptom severity. Participants were 1,003 U.S. veterans (58.7% with PTSD) who completed the PCL for DSM-IV (the PCL-C) and DSM-5 (the PCL-5) during their participation in an ongoing longitudinal registry study. In a randomly selected training sample (n = 800), we used equipercentile equating with loglinear smoothing to compute a “crosswalk” between PCL-C and PCL-5 scores. We evaluated the correspondence between the crosswalk-determined predicted scores and observed PCL-5 scores in the remaining validation sample (n = 203). The results showed strong correspondence between crosswalk-predicted PCL-5 scores and observed PCL-5 scores in the validation sample, ICC= .96. Predicted PCL-5 scores performed comparably to observed PCL-5 scores when examining their agreement with PTSD diagnosis ascertained by clinical interview: predicted PCL-5, κ = 0.57; observed PCL-5, κ = 0.59. Subsample comparisons indicated that the crosswalk’s accuracy did not differ across characteristics including gender, age, racial minority status, and PTSD status. The results support the validity of this newly developed PCL-C to PCL-5 crosswalk in a veteran sample, providing a tool with which to interpret and translate scores across the two measures.

The publication of the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5;American Psychiatric Association [APA], 2013) introduced numerous revisions to the diagnostic criteria for posttraumatic stress disorder (PTSD), including the addition of new symptoms; the modification of several existing symptoms; and the introduction of four, rather than three, symptom clusters. These changes to the diagnostic criteria pose a challenge to clinicians and researchers who previously collected symptom data using measures reflecting the PTSD diagnostic criteria in the prior version of the DSM (i.e., the fourth edition, text revision; DSM-IV-TR; APA, 2000) but who wish to follow the course of PTSD symptoms over time, including after the revisions to the criteria were published. This shift may be especially challenging to longitudinal investigations of PTSD, in which continuity of symptom measurement over time is critical for many statistical analyses.

Clinicians and researchers with these continuity concerns must choose among using symptom severity measures that correspond with outdated PTSD diagnostic criteria; using measures that correspond with the updated DSM-5 PTSD diagnostic criteria; or creating idiosyncratic, unvalidated measures that simultaneously collect information about both sets of diagnostic criteria. None of these choices is ideal. Instead, researchers and clinicians would benefit from a guide that translates results of DSM-IV congruent measures to estimated results on DSM-5 congruent measures, and vice versa. Recent research has suggested that DSM-IV- congruent symptom ratings can be used to approximate a diagnosis of DSM-5 PTSD (Rosellini et al., 2015). However, there is currently no tool available to enable linking of continuous total or cluster-specific PTSD symptom severity scores derived from DSM-IV- and DSM-5-congruent measures. Therefore, the aim of the present study was to establish a translational crosswalk between symptom severity scores on the PTSD Checklist–Civilian Version for DSM-IV-TR (PCL-C) and the PCL for DSM-5 (PCL-5; Weathers, Litz, Herman, Huska, & Keane, 1993; Weathers et al., 2013), as the PCL is the most commonly used self-rated measure of PTSD symptom severity. To do so, we conducted test-equating procedures using data from both versions of the measure collected concurrently in a sample of United States military veterans.

Method

Participants

Participants were 1,003 United States Army or Marine veterans enrolled in the Veterans After-Discharge Longitudinal Registry (Project VALOR). Project VALOR is a registry of Veterans’ Affairs (VA) mental health care users with and without PTSD who were deployed in support of recent military operations in Afghanistan and Iraq. To be included in the cohort, veterans must have undergone a mental health evaluation at a VA facility. The cohort oversampled for veterans with probable PTSD according to VA medical records (i.e., at least two instances of a PTSD diagnosis by a mental health professional associated with two separate visits) at a 3:1 ratio. Female veterans were oversampled at a rate of 1:1 (female to male). A sample of 1,649 (60.8%) veterans completed the baseline assessment for Project VALOR. For the current analysis, we focused on a subsample of this group that consisted of 1,003 participants who reported experiencing a DSM-5 Criterion A traumatic event during a clinical interview and had complete data (required for the test-equating analyses) on both the PCL-C and PCL-5 during the fourth wave of study assessments (Time 4 [T4]). There were no significant differences in sex, racial minority status, or PTSD diagnostic status or symptom severity at the first wave of data collection (Time 1 [T1]) between the 1,003 participants included in this analysis and the remaining cohort members, ps = .262–.891. However, participants included in this analysis were older (M age = 38 years) compared with the remaining cohort members (M age = 36 years), t(1,647) = −3.56, p = .000, and had a higher level of educational attainment (i.e., 38% of the analytic sample had a bachelor’s degree vs. 30% of remaining cohort members), χ²(6, N = 1,642) = 15.74, p = .015.

Procedure

At T4 of Project VALOR, participants provided informed consent verbally over the telephone in accordance with the research protocol approved by the VA Boston Healthcare System institutional review boards and the Human Research Protection Office of the U.S. Army Medical Research and Material Command. Participants then completed a self-administered questionnaire (SAQ) online and, following this, completed a telephone-based diagnostic clinical interview. The SAQ consisted of a large battery of questionnaires that, in total, included over 740 questions pertaining to physical health, functional impairment, psychiatric symptoms, deployment experiences, and lifetime trauma exposure.

Measures

Demographic information.

Participant age and sex were extracted from a U.S. Department of Defense database. Race, ethnicity, and education were collected via self-report in the T4 SAQ.

PTSD symptom severity.

The PCL-C is a self-rated measure of PTSD symptom severity designed to correspond to the 17 core DSM-IV PTSD symptoms (Weathers et al., 1993). Respondents use a scale ranging from 1 (not at all) to 5 (extremely) to rate how much each symptom has bothered them in the past month. Although a military version of the PCL (the PCL-M) is available, we used the civilian version because it corresponded with the study’s clinical interview procedures, which did not restrict potential index traumatic events solely to military-related events. The PCL-C is one of the most commonly used self-rated measures of DSM-IV PTSD symptom severity, and it has demonstrated excellent psychometric properties across a range of samples and settings (for review, see Norris & Hamblen, 2004). In the current sample, internal reliability of PCL-C scores was excellent, Cronbach’s α = .96.

The PCL-5 (Weathers et al., 2013) is a self-rated measure of PTSD symptom severity designed to correspond to the 20 core DSM-5 PTSD symptoms. Respondents use a scale ranging from 0 (not at all) to 4 (extremely) to rate how much each symptom has bothered them in the past month. Like its predecessor, the PCL-5 is frequently used across a range of settings for a variety of purposes, including monitoring symptom change as well as screening for and providing a provisional diagnosis of PTSD. Data from the PCL-5 has demonstrated good test–retest reliability, r = .84, and convergent and discriminant validity (Blevins, Weathers, Davis, Witte, & Domino, 2015; Bovin et al., 2015; Keane et al., 2014; Wortmann et al., 2015). Internal reliability of PCL-5 scores was excellent in the current sample, Cronbach’s α = .96.

Major depression and PTSD diagnosis.

The PTSD and Major Depressive Episode (MDE) modules of the Structured Clinical Interview for DSM-5 (SCID-5; First, Williams, Karg, & Spitzer, 2015) were used to assess exposure to a Criterion A event and to assess current PTSD diagnostic status and presence or absence of a current MDE. Interrater agreement was evaluated for a random sample of 100 cases and was excellent for both current PTSD, κ = 0.85, and current MDE, κ = .98.

Data Analysis

To link PCL-C and PCL-5 scores, we used equipercentile equating, a test-equating procedure that is commonly used in educational measurement fields to determine comparable scores on different versions of the same exam (for a review, see Dorans, Moses, & Eigner, 2010). Equipercentile equating considers scores on two measures to be equivalent to one another if their percentile ranks in a given group are equal. This approach has a number of benefits relative to mean or linear equating methods; for example, it results in all imputed scores falling within the actual range of the scale and does not rely on the assumption of a normal distribution of test scores. Equipercentile equating methods have been used to develop crosswalks for a number of neurocognitive and psychiatric rating scales (e.g., Choi, Schalet, Cook, & Cella, 2014; Monsell et al., 2016).

Prior to performing the equating procedure, we randomly split the sample into a training sample (n = 800) and a validation sample (n = 203; a split which allows for a large sample size to be retained for the equating procedure, consistent with recommendations by Dorans et al., 2010). In the training dataset, equipercentile equating with loglinear smoothing was performed using the R package Equate (Albano, 2016). Standard errors and 95% confidence intervals of the crosswalk estimates were calculated using 10,000 bootstrapped samples. After completing the equating procedure in the training dataset, we used the resulting crosswalk to impute predicted PCL-5 scores from PCL-C scores for all participants in the validation dataset.

To evaluate the accuracy of the crosswalk in the validation sample, we examined the intraclass correlation coefficient (ICC) between predicted and observed PCL-5 scores and calculated the average difference between predicted and observed PCL-5 scores. We calculated sensitivity, specificity, efficiency (correct classification rate), quality of efficiency (i.e., Cohen’s kappa), and area under the curve (AUC) for use of crosswalk-predicted PCL-5 cut scores, using the cutoff of PCL-5 score of 33 or greater (Bovin et al., 2015) in identifying PTSD diagnosis as determined by the SCID interview. Finally, in order to evaluate whether the crosswalk demonstrated accuracy across relevant subgroups of individuals, we compared these same markers of accuracy when the sample was divided into subgroups based on education level, age, gender, racial minority status, and presence or absence of PTSD and MDE.

We used the same test-equating procedures to create crosswalks from PCL-C subscale scores to PCL-5 subscale scores, representing each of the DSM-5 PTSD symptom clusters (Cluster B = intrusion symptoms, Cluster C = avoidance symptoms, Cluster D = negative alterations in cognitions and mood, Cluster E = alterations in arousal and reactivity). These symptom clusters were approximated in the PCL-C data by summing Items 1–5 (Cluster B), Items 6 and 7 (Cluster C), Items 8–12 (Cluster D), and Items 13–17 (Cluster E). Missing data were minimal (one missing case each for variables of age, race, and education status; and three cases missing the MDE module of the SCID) and were therefore handled using pairwise deletion.

Results

The characteristics of the sample and subsample are presented in Table 1. In all, 58.7% percent of participants met criteria for current (i.e., past month) PTSD and 34.5% met criteria for current MDE. Group comparison tests revealed no significant differences among the training and validation samples on sex, race, ethnicity, education level, PCL-C or PCL-5 score, or proportion of sample with current PTSD or MDE, ps = .363–.878. The PCL-C and PCL-5 were highly correlated in both the training and validation samples, rs = .95 and .96, respectively. These correlations were well over thresholds recommended for equating procedures (i.e., .75–.86; Choi et al., 2014). A histogram of total score frequencies in the training sample is presented in Figure 1.

Table 1.

Demographic Characteristics of the Total, Test, and Validation Samples

Variable	Total Sample (n = 1,003)			Test Sample (n = 800)			Validation Sample (n = 203)
	%	M	SD	%	M	SD	%	M	SD
Sex
Female	51.1			50.8			52.7
Male	48.9			49.2			47.3
Age (years)		43.2	9.8		43.2	9.9		43.2	9.5
Racial minority status
Non-White	23.2			22.6			25.2
White	76.8			77.4			74.8
Highest education level
High school or GED	6.6			6.9			5.4
Some college	38.9			39.7			37.1
Bachelor’s degree or higher	50.8			50.1			53.7
Current PTSD	58.7			58.9			58.1
Lifetime PTSD	87.5			87.4			88.2
Current MDE	34.5			34.4			35.0
PCL-C score		49.7	17.3		50.0	17.3		50.1	17.6
PCL-5 score		36.2	20.6		36.1	20.7		36.7	20.6

Note. PCL-C = Posttraumatic Stress Disorder Checklist–Civilian Version (for DSM-IV); PCL-5 = Posttraumatic Stress Disorder Checklist for DSM-5; PTSD = posttraumatic stress disorder; MDE = major depressive episode; GED = general education development.

Figure 1.

Histograms of total Posttraumatic Stress Disorder Checklist–Civilian Version (for DSM-IV; PCL-C) and PCL for DSM-5 (PCL-5) scores in the training sample (n = 800). DSM = Diagnostic and Statistical Manual of Mental Disorders (DSM-IV = fourth edition; DSM-5 = fifth edition).

The crosswalk for converting PCL-C to PCL-5 scores based on equipercentile equating results is presented in Figure 2. The PCL-C scores were equated to lower PCL-5 scores, which is not surprising given the difference in scaling ranges between the two measures (PCL-C scores range from 17 to 85 and PCL-5 scores range from 0 to 80). For example, a score of 50 on the PCL-C was equated with a score of 36 on the PCL-5. In the validation sample, the ICC among the observed and predicted PCL-5 scores was .96. The mean difference between observed and predicted PCL-5 scores was 0.20 (SD = 6.30). Using the cutoff score of 33 or higher, the predicted PCL-5 score had similar diagnostic utility to the observed PCL-5 score in predicting PTSD diagnosis determined by clinical interview: Cohen’s κ = .55, sensitivity = .81, specificity = .74, AUC = .77, correct classification of 78% of cases for the predicted PCL-5; Cohen’s κ = .58, sensitivity = .84, specificity = .74, AUC = .79, correct classification of 80% of cases for the observed PCL-5.

Figure 2.

Crosswalk of corresponding Posttraumatic Stress Disorder Checklist–Civilian Version (for DSM-IV; PCL-C) and PCL for DSM-5 (PCL-5) total scores with 95% confidence intervals from 10,000 bootstrapped samples. DSM = Diagnostic and Statistical Manual of Mental Disorders (DSM-IV = fourth edition; DSM-5 = fifth edition).

The accuracy of the crosswalk was highly consistent across subgroups based on sex, age, racial minority status, education level, PTSD diagnostic status as determined by clinical interview, and presence or absence of current MDE (see Table 2). The ICCs between predicted and observed PCL-5 scores were very high for all subgroups, ICCs = .92–.96. There were no significant differences in the mean differences between observed and predicted PCL-5 score between any of these demographic subgroups. The kappa values between observed and predicted probable DSM-5 PTSD diagnosis were good for all subgroups examined, and the proportion of correctly classified cases did not differ significantly by subgroup.

Table 2.

Posttraumatic Stress Disorder Checklist (PCL) Crosswalk Accuracy in Clinical and Demographic Subgroups Within the Validation Sample

		Crosswalk-Predicted and Observed PCL-5 Scores	Difference Between Crosswalk-Predicted and Observed PCL-5 Scoresa		Crosswalk-Predicted and Observed Probable PTSDb
	n	ICC	M	SD	κ
Sex
Male	96	.95	−0.29	6.72	0.91
Female	107	.96	−0.08	5.84	0.85
Age (years)
< 40	116	.95	0.30	6.55	0.83
≥ 40	86	.96	−0.84	5.85	0.95
Racial minority status
Non-White	51	.96	0.75	6.23	0.88
White	151	.95	−0.49	6.27	0.88
Education level
High school or some college	93	.94	−0.65	6.71	0.90
Bachelor’s degree or higher	109	.96	0.22	5.83	0.85
PTSD diagnosis
Present	118	.92	−0.28	6.33	0.85
Absent	85	.93	−0.16	6.31	0.78
Current MDE
Present	70	.93	−0.90	6.29	0.83
Absent	130	.95	0.25	6.21	0.86

Note. n = 203. ICC = intraclass correlation coefficient; PCL-C = Posttraumatic Stress Disorder Checklist–Civilian Version (for DSM-IV); PCL-5 = Posttraumatic Stress Disorder Checklist for DSM-5; PTSD = posttraumatic stress disorder; MDE = major depressive episode.

^aIn t tests between all subgroups, ps = .190–.812.

^bProbable PTSD defined as a PCL-5 score ≥ 33.

The items comprising Clusters B and C are highly similar between the PCL-C and PCL-5, with only minor wording changes (e.g., the addition of “unwanted” to Item 1 or the addition of “strong” to Item 5 on the PCL-5). Not surprisingly then, the equipercentile-equated crosswalk for the Cluster C subscale was identical to a linear transformation of subtracting 2 points from PCL-C scores to reflect the change in scaling between the two measures. Similarly, the equipercentile-equated crosswalk for Cluster B subscale scores was nearly identical to a linear transformation involving subtracting 5 points from PCL-C scores. The ICC between equated and observed scores using these two methods was equal to .997. Additionally, the equipercentile-equated crosswalk for Cluster B did not outperform the linear transformation method in the accuracy analyses conducted in the validation sample, which suggests that the linear transformation can be used for simplicity when converting Cluster B subscale scores between the PCL-C and PCL-5. However, such a linear transformation would not be appropriate for Clusters D and E given that both clusters include new symptoms in DSM-5 relative to DSM-IV-TR. The crosswalks for Cluster D and E subscores based on equipercentile equating with loglinear presmoothing are presented in Figure 3. Predicted cluster subscores were very strongly correlated with observed cluster subscores in the validation sample for all four clusters; the ICC values between observed and predicted subscale scores were .94 for Cluster B, .88 for Cluster C, .89 for Cluster D, and .91 for Cluster E.

Figure 3.

Crosswalk of corresponding Posttraumatic Stress Disorder Checklist–Civilian Version (for DSM-IV; PCL-C) and PCL for DSM-5 (PCL-5) Clusters D and E scores with 95% confidence intervals from 10,000 bootstrapped samples. Approximated PCL-C scores for Clusters D and E were computed by summing Items 8–12 (Cluster D) and Items 13–17 (Cluster E) of the PCL-C. DSM = Diagnostic and Statistical Manual of Mental Disorders (DSM-IV = fourth edition; DSM-5 = fifth edition).

Discussion

This is the first known study that attempted to equate scores between two versions of a frequently used PTSD symptom severity measure: the DSM-IV-based PCL-C and the DSM-5-based PCL-5. The resulting crosswalk enables researchers and clinicians to interpret and translate scores across the two measures, an important consideration in longitudinal observational and clinical treatment studies that cross iterations of the DSM. A particular strength of this study was the use of both training and validation samples, which allowed us to evaluate the accuracy of the crosswalk. Supporting the validity of the crosswalk, results demonstrated a strong degree of concordance between observed and predicted PCL-5 scores (both total and cluster subscale scores) in the validation sample. Additionally, predicted PCL-5 scores performed comparably to observed PCL-5 scores when examining their agreement with PTSD diagnosis ascertained by clinical interview. Finally, the results suggest a similar degree of concordance between crosswalk–predicted and observed subscale scores and indicate that the metrics of crosswalk accuracy did not differ across subgroups.

We anticipate that the PCL crosswalk may be particularly useful for longitudinal research or for interpretation of clinical data that has been collected over a time period spanning the use of both the DSM-IV and DSM-5. It may also allow for the combining of datasets from studies using different versions of the PCL, facilitating research that requires large sample sizes, such as gene association studies. Moreover, the availability of crosswalks for computing DSM-5 symptom cluster subscale scores will allow for further study of the association between specific domains of symptoms (e.g., avoidance, arousal) and risk factors or outcomes of interest. However, it should be noted that the evolution of the diagnostic criteria from DSM-IV to DSM-5 has led to some substantive differences in how the PTSD construct is defined in each version. The strong correlation among PCL-C and PCL-5 scores (r = .95) suggests that it was statistically appropriate to use test-equating procedures to link the scales. This strong association has been demonstrated in prior studies of the PCL-5 (e.g.,Wortmann et al., 2016) and is consistent with other research suggesting a strong degree of overlap between the two DSM criteria sets (e.g., Kilpatrick et al., 2013). However, it should also be acknowledged that the resulting crosswalk cannot provide specific information about the elements of the PTSD construct that are new to DSM-5 and were not assessed in DSM-IV (i.e., distorted blame, reckless behavior), and it also does not address differences in the definition of a Criterion A traumatic event.

This study has a number of strengths for a test-equating design. We used a single-group design in which all participants completed both versions of the PCL, thus producing more reliable linking across measures. The sample was large and gender-balanced, and participants showed a wide degree of variation in PTSD symptom severity. However, the sample consisted solely of veterans serving in recent-era (post-September 11, 2001, terrorist attacks) combat operations in Afghanistan and Iraq. Although the crosswalk showed invariance to several demographic characteristics within the sample, it is not clear to what extent the results would generalize to civilian samples. We suggest caution in applying the crosswalk to these samples and encourage continued study of these results in other trauma-exposed samples. Additionally, it should be noted that the PCL-C and PCL-5 were administered in the same order for every participant, with the PCL-C administered first. Therefore, order effects may have influenced our results, and future research should examine this possibility, using a counter-balanced design.

In this study, we present a crosswalk that will allow for conversion between PCL-C and PCL-5 symptom severity scores. The results provide support for the validity of the crosswalk within a veteran sample. This tool will allow researchers and clinicians to make use of archival PCL-C data in longitudinal research, clinical settings, and beyond.