Author's Accepted Manuscript

Latent classes of PTSD Symptoms in Iraq and Afghanistan female veterans Claire Hebenstreit, Erin Madden, Shira Maguen

www.elsevier.com/locate/jad

PII: DOI: Reference:

S0165-0327(14)00255-9 http://dx.doi.org/10.1016/j.jad.2014.04.061 JAD6728

To appear in:

Journal of Affective Disorders

Received date: 21 April 2014 Accepted date: 25 April 2014 Cite this article as: Claire Hebenstreit, Erin Madden, Shira Maguen, Latent classes of PTSD Symptoms in Iraq and Afghanistan female veterans, Journal of Affective Disorders, http://dx.doi.org/10.1016/j.jad.2014.04.061 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Latent Classes of PTSD Symptoms in Iraq and Afghanistan Female Veterans

Claire Hebenstreit San Francisco VA Medical Center and University of California, San Francisco Erin Madden San Francisco VA Medical Center Shira Maguen San Francisco VA Medical Center and University of California, San Francisco

Correspondence should be addressed to Claire Hebenstreit, Ph.D., San Francisco VA Medical Center, PTSD Program (116-P), 4150 Clement St., San Francisco, CA 94121, Phone: (415) 2214810 x 2511, Fax: (415) 379-5562, Email: [email protected].

1   

Abstract Background. Recent studies have used latent class analysis (LCA) to identify subgroups of individuals who share similar patterns of PTSD symptom endorsement; however, further study is needed among female veterans, whose PTSD symptom expression may vary from that of their male counterparts. The current study examined latent PTSD symptom classes in female veterans who returned from recent military service in Iraq and Afghanistan, and explored military and demographic variables associated with distinct PTSD symptom presentations. Methods. A retrospective analysis was conducted using existing medical records from female Iraq and Afghanistan veterans who were new users of VA mental health outpatient (MHO) care, had received a PTSD diagnosis anytime during the post-deployment period, and completed the PTSD checklist within 30 days of their first MHO visit (N = 2,425). Results. The LCA results identified four latent classes of PTSD symptom profiles in the sample: High Symptom, Intermediate Symptom, Intermediate Symptom with High Emotional Numbing (EN), and Low Symptom. Race/ethnicity, age, time since last deployment, and distance from a VA facility emerged as predictors of PTSD symptom presentation. Limitations. The current study was cross-sectional and utilized administrative data. The results may not be generalizable to female veterans from other service eras. Conclusions. Longer times between end of last deployment and initiation of MHO services were associated with more symptomatic classes. Exploration of PTSD symptom presentation may enhance our understanding of the service needs of female veterans with PTSD, and suggests potential benefits to engaging veterans in MHO soon after last deployment.

2   

Keywords: Posttraumatic Stress Disorder, Women’s health services, Veterans/psychology, Female 

Latent Classes of PTSD Symptoms in Iraq and Afghanistan Female Veterans In recent years, several studies have used latent class analysis (LCA) to identify subgroups of individuals who share similar patterns of PTSD symptom endorsement. This method fits latent profile models to the data, and participants are classified as belonging to the profile with the best fit. To date, studies of PTSD symptom classes have included the application of LCA to a range of populations, including adult and adolescent community samples (Breslau et al., 2005; Ayer at al., 2011), trauma-exposed children and adolescents (Geronazzo Alman et al., 2012), Canadian treatment-seeking military veterans (Armour et al., 2014; Naifeh et al., 2010), female sexual assault survivors (Au et al., 2013), and adult survivors of childhood abuse (Steuwe et al., 2012) and natural disasters (Rosellini et al., 2014). Studies have also examined latent classes of PTSD symptoms in U.S. military veterans of multiple eras, including Vietnam veterans (Steenkamp et al., 2012; Wolf et al., 2012) and Iraq and Afghanistan veterans (Maguen et al., 2013). The results of these studies typically indicate the presence of either three (Ayer et al., 2011; Breslau et al., 2005; Steenkamp et al., 2012; Steuwe et al., 2012) or four (Geronazzo Alman et al., 2012; Au et al., 2013; Maguen et al., 2013; Rosellini et al., 2014) latent classes. Across studies, these classes are generally distinguished by symptom severity, with classes representing minimal/mild, intermediate, and severe PTSD symptoms. In addition to symptom severity, PTSD classes have also been distinguished by specific symptom elevations. For example, Geronazzo Alman et al. (2012) found that individuals in the most severe symptom class 3   

endorsed more avoidance and sleep disturbance. Several studies indicate the presence of a class of PTSD symptoms characterized by dissociation (Steenkamp et al. 2012; Steuwe et al, 2012; Wolf et al, 2012). Emotional numbing (EN) in PTSD has also been associated with latent class membership, including the finding by Breslau et al. (2005) that individuals in the pervasive disturbance symptom class reported higher levels of EN. Most recently, a study of male and female Iraq and Afghanistan veterans found that individuals in two intermediate symptom classes were distinguished by high and low EN symptom levels (Maguen et al., 2013). Although the latent structure of PTSD symptoms has been explored in multiple populations, including veterans, further study is needed among female veterans. These women represent a segment of the veteran population that is rapidly increasing: between 2000 and 2011 the number of female Veterans Health Administration (VHA) users increased from 159,000 to 337,000 (Women’s Health Evaluation Initiative, 2011), and this number is expected to double again within the next decade (National Center for Veterans Analysis and Statistics, 2011). The health needs of female veterans appear to differ from those of male veterans as well as female non-veterans, due in part to differing military experiences and stressors (Bean-Mayberry et al., 2006; Goldzweig et al., 2011). Among veterans with PTSD, women utilize services at higher rates than their male counterparts (Cohen et al., 2010; Maguen et al., 2012), and women who reported experiencing repeated exposure to violence during military service were shown to have required substantially more frequent outpatient medical appointments than singly or nontraumatized peers (Sadler et al., 2004). Female veterans’ PTSD symptom expression may also vary from that of their male counterparts. For example, King et al. (2013) found that female veterans were more likely to endorse concentration difficulties and distress related to reminders of a traumatic event, while 4   

male veterans were more likely to endorse hypervigilance, EN, and nightmares. An examination of gender differences in the factor structure of PTSD found that, while the emotional numbing model of PTSD (King et al., 1998) was statistically superior to the dysphoria model (Simms et al, 2002) in male veterans, neither model was superior in female veterans (Hall et al., 2013). Given the continued increase in the number of female veterans, it is important to examine the clinical presentation of PTSD symptoms within this population. The primary aim of this study was to examine latent PTSD symptom classes in female veterans who received mental health services. We included female veterans who returned from recent military service in Iraq and Afghanistan and had enrolled in the VA health care system following their most recent deployment. In order to better understand patterns of PTSD symptoms prior to the onset of treatment, we focused our investigation on women who had completed a PTSD symptom measure within one month of their first mental health visit. Our secondary aim was to explore demographic and military factors associated with distinct PTSD symptom presentations.

Methods Study Population We identified the study population using the VA National OEF/OIF/OND Roster, an accruing database of veterans who have returned from recent military service in Iraq and Afghanistan and have enrolled in the VA health care system. We examined administrative data from 2,425 Iraq and Afghanistan female veterans who were new users of VA health care (after their most recent deployment) from October 1, 2007 onward, were seen in mental health outpatient care, received a diagnosis of PTSD (International Classification of Diseases, Ninth Revision Clinical Modification diagnostic code 309.81) recorded at two or more clinical 5   

encounters, received the PTSD checklist (PCL) within 30 days of their first MHO visit (includes integrated primary-mental health care), and had been in the VA system for at least one year by July 31, 2013. October 1, 2007 was chosen as the start date because at that time, PCL responses began being stored in national VA databases, and we focused on those receiving the PCL within 30 days in order to capture PTSD symptom severity at initial presentation. The study was approved by the Committee on Human Research, University of California, San Francisco and the Human Research Protection Program at the San Francisco VA Medical Center. Data Source The VA OEF/OIF/OND Roster includes information on veterans' demographic characteristics as well as aspects of their military service, such as their rank and the branch of the military with which they served. The Roster was linked to two other national administrative databases: (1) the VA National Patient Care Database (NPCD) to obtain information on VA clinic visits, associated clinical diagnoses and additional race/ethnicity data, and (2) the VA Corporate Data Warehouse (CDW) to obtain PTSD checklist results. The VA National Patient Care Database includes data from outpatient and inpatient visits to any of the approximately 150 VA hospitals and over 900 VA clinics nationwide. The electronic record includes the date of the visit, a code designating the type of visit, patient race and ethnicity, and the diagnosis(es) associated with the visit classified using the (ICD-9-CM) codes. Visits to mental health outpatient services were identified using clinic stop codes (Seal et al., 2010; Cohen et al., 2010; Maguen et al., 2012). Mental health outpatient services included visits to primary care-mental health integrated care clinics. Fee basis codes designated care rendered at non-VA facilities reimbursed by VA, but did not capture all non-VA care (e.g.,

6   

private insurance). Data on the distance to and type of nearest VA medical facility based on each veteran's zip code (in Roster) were calculated by the VA Planning Systems Support Group. PTSD checklist screening results were extracted from the VA Corporate Data Warehouse (CDW), a national data repository comprising data from several Veterans Health Administration (VHA) clinical and administrative systems, including the Mental Health Assessment. Measures PTSD symptoms were assessed using the PTSD Checklist, a 17-item measure (PCL; Weathers et al., 1993; Blanchard et al., 1996). Each item was rated on a five-point Likert scale with responses ranging from not at all to extremely, and participants were asked to rate symptoms over the last month. For the purposes of this study, symptoms rated as moderately or above on the PCL were rated as present. The PCL is widely used as a screen for PTSD, has been shown to have very good internal consistency, and correlates strongly with other measures of PTSD symptoms (Weathers et al., 1993). The PCL also demonstrates high diagnostic efficiency (i.e., .90; Blanchard et al., 1996). The PCL is mainly administered in the mental health treatment setting within the VA, at the discretion of treating clinicians, and is typically used to track patient progress during the course of mental health treatment. Study Variables Dependent Variables. Four PTSD Latent Classes were identified by latent class analysis (described below) using the seventeen symptoms from the PCL: High Symptom (50% of sample), Intermediate Symptom (20.6%), Intermediate Symptom with High Emotional Numbing (20.8%), and Low Symptom (8.6%). Independent Variables. We used the OEF/OIF/OND Roster to obtain dates of birth, gender, marital status, component type (active component versus National Guard or Reserve), 7   

military rank (enlisted versus officer), service branch (Army, Air Force, Marines, Navy/Coast Guard), and whether a veteran had been deployed more than once. Data were also obtained on the distance to and type of nearest VA facility (VA medical center vs. VA community-based outpatient clinic). Race and ethnicity data comes primarily from the OEF/OIF/OND Roster and secondarily from VA inpatient and outpatient encounters data. Race/ethnicity is a composite variable with the following values: White, Black, Hispanic, Asian/Pacific Islander, or other/unknown. We merged the VA and Roster data such that Roster race/ethnicity data took precedence, unless it was missing/unknown, in which case VA race/ethnicity data was used, if it existed. We allowed the Roster race/ethnicity to take precedence because this was the source of all other demographic and military data for our study. Analysis Distinct classes of PTSD symptom profiles at initial clinical presentation were identified with latent class analysis (LCA; Hagenaars and McCutcheon, 2002). The aim of LCA is to uncover latent classes that explain the observed relationships between multivariate categorical variables (e.g., PTSD symptoms) by clustering individuals such that the within-cluster responses are statistically independent. LCA assumes that there is a single underlying latent variable that is accounting for the associations between the observed dependent variables. Each of the 17 PCL items was dichotomized by a score of moderate (3) or greater before being included as binary dependent variables in a LCA. The LCA was conducted using the robust maximum likelihood estimator. We built a series of models with two to fifteen latent classes. The following goodness-of-fit statistics were then compared across the seven models: bootstrapped likelihood ratio tests (BLRt), sample-size adjusted Bayesian Information Criterion (aBIC), and sample-size corrected Aikake Information Criterion (AICc). We focused on models 8   

with statistically significant BLRt p-values and lower aBIC values (Nylund et al., 2007; Henson, et al., 2007). Entropy, which is a measure of the quality of classification, was also compared across models. Entropy values approaching one indicate clear delineation of classes (Celeux and Soromenho, 1996) and we focused on models with entropy ≥0.80. Model selection was based on equal consideration of parsimony, practical interpretability, and goodness of fit statistics. Most-likely latent class membership was determined for each participant and used as outcome variable in multinomial logistic regression analyses to examine the associations between the latent class membership and demographic and military characteristics. The LCA was conducted with the Mplus statistical modeling software (Version 7; Muthen andMuthen, 2012). Descriptive statistics and multinomial logistic regression analyses were carried out with SAS (Version 9.3, Cary, NC).

Results The sample had a median age of 29 years (IQR: 20-66); 49.9% were White, 27.9% Black, 14.1% Hispanic, 4.2% Asian/PI, and 4.0% other/unknown race/ethnicity. Model fit statistics (aBIC) showed that the model with 9 classes was best (see appendix). However the decrease in aBIC for models with greater than 4 classes was relatively small and, upon inspection, the general patterns represented by the nine latent classes were adequately captured by the symptom profiles in the four-class solution. The four-class solution was thus selected as being most parsimonious and clinically meaningful. In Figure 1, the latent classes are presented in terms of the within-class probability of endorsing each PCL item. The first class, which was the lowest prevalent class (Low Symptom Class; 8.6% of sample), was characterized by a very low probability (< 0.3) of endorsing all of 9   

the 17 PTSD symptoms, except for Insomnia (probability=0.54) and Irritability (probability=0.37). The second (Intermediate Symptom with High EN Class; 20.8% of sample) and third (Intermediate Symptom; 20.6% of sample) classes were both characterized by symptoms of reexperiencing (particularly intrusive memories and psychological reactivity), as well as avoidance, and some symptoms of increased arousal (particularly insomnia). These two classes diverged sharply on particular symptoms of emotional numbing with the Intermediate Symptom Class having a low probability ( .9) of endorsing PTSD symptoms except for having flashbacks, amnesia and feelings of a foreshortened future. Multinomial logistic regression models were used to examine the associations between latent classes and demographic and military characteristics are shown in Table 1. All six pairwise comparisons of latent classes were captured by three models: One compared the three more symptomatic classes with the Low Symptom (reference class), a second had the Intermediate Symptom class as reference, and a third had Intermediate Symptom with high EN as reference. The strongest independent predictor of PTSD Latent Class membership was time since the end of last deployment. Longer times (≥1.2 years) were associated with increased likelihood of belonging to either the High Symptom or Intermediate Symptom with high EN classes as compared to the Low or Intermediate classes (see Table 1). 10   

Demographic factors such as age at first MHO visit and race/ethnicity and distance to the closest VA facility were also independently associated with PTSD latent class membership. Older age (30-39 vs. 18-24 years) at first MHO visit was associated with increased likelihood of belonging to the Intermediate Symptom class as compared to the Low Symptom (OR=2.08, 95% CI=[1.19, 3.60], p