Journal of Affective Disorders 176 (2015) 87–94

Contents lists available at ScienceDirect

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

Research report

PTSD symptom presentation across the deployment cycle Maria M. Steenkamp a, Alyssa M. Boasso b, William P. Nash c, Jonathan L. Larson b, Rebecca E. Lubin b, Brett T. Litz b,n a

New York University School of Medicine, United States VA Boston Healthcare System, Massachusetts Veterans Epidemiology Research and Information Center (MAVERIC), Boston University School of Medicine, United States c Boston VA Research Institute (BVARI), United States b

art ic l e i nf o

a b s t r a c t

Article history: Received 8 January 2015 Accepted 15 January 2015 Available online 29 January 2015

Background: Symptom-level variation in posttraumatic stress disorder (PTSD) has not yet been examined in the early post-deployment phase, but may be meaningful etiologically, prognostically, and clinically. Methods: Using latent class analysis (LCA), we examined PTSD symptom heterogeneity in a cohort of participants from the Marine Resiliency Study (MRS), a longitudinal study of combat Marines deployed to Iraq and Afghanistan (N¼892). Typologies of PTSD symptom presentation were examined at one month pre-deployment and again one, five, and eight months post-deployment. Results: Heterogeneity in PTSD symptom presentation was evident at each assessment point, and the degree of symptom heterogeneity (i.e., the number of classes identified) differed by time point. Symptom patterns stabilized over time from notable symptom fluctuations during the early post-deployment period to high, medium, and low symptom severity by eight months post-deployment. Hypervigilance and exaggerated startle were frequently endorsed by participants in the initial month post-deployment. Flashbacks, amnesia, and foreshortened future were infrequently endorsed. Greater combat exposure, lifespan trauma, and avoidant coping generally predicted worse outcomes. Limitations: Data were self-report and may have limited generalizability due to our lack of women and inclusion of only combat Marines. Attrition and re-ranging of data resulted in significant missing data and affected the representativeness of the sample. Conclusions: Symptom-level variability is highest in the month following deployment and then stabilizes over time. Should post-deployment assessments occur too soon, they may capture common and transient early post-deployment reactions, particularly anxious arousal. Published by Elsevier B.V.

Keywords: Afghanistan PTSD Latent class Symptoms Combat Military

1. Introduction To date, most research on posttraumatic stress disorder (PTSD) has focused on outcomes indexed as a composite PTSD symptom burden at a given cross section (e.g., a total severity score). Such emphasis on total PTSD symptom scores assumes homogenous symptom presentations across individuals and across time, and may mask important symptom-level variability. Because PTSD comprises a highly heterogeneous set of 17 symptoms (there are 20 in DSM-5), a finer-grained analysis of PTSD outcomes at the symptom level may reveal important variation in post-deployment PTSD symptom presentation. For example, some trauma survivors may be chiefly haunted by frequent nightmares, intrusive thoughts, and

n Correspondence to: VA Boston Healthcare System, 150 S. Huntington Ave, 13-B71, Jamaica Plain, MA 02130, United States. E-mail address: [email protected] (B.T. Litz).

http://dx.doi.org/10.1016/j.jad.2015.01.043 0165-0327/Published by Elsevier B.V.

avoidance symptoms, while others may have few of these symptoms but instead experience marked numbing, hypervigilance, and anger. In particular, symptom-level variation has rarely been examined in the early post-deployment phase, but may be meaningful etiologically, prognostically, and clinically. The early temporal and functional relationships between the various PTSD symptoms and symptom subclusters have not been well explained in theories of PTSD but, empirically, several studies have suggested that the presence of certain symptoms early following trauma may be particularly meaningful. For example, studies have noted the potential prominence of early hyperarousal symptoms in the development of PTSD (e.g., Bremner et al., 1996; Schell et al., 2004; Solomon et al., 2009), and in one retrospective study of veterans with PTSD, hypervigilance and exaggerated startle were the first symptoms to appear (Andrews et al., 2009). Understanding the phenomenology of early post-deployment reactions may be particularly important because some symptom endorsement is likely expectable (and potentially normative) following war. Temporary reactions and unremitting pathological reactions may

88

M.M. Steenkamp et al. / Journal of Affective Disorders 176 (2015) 87–94

mimic each other in this period, complicating prognostic prediction. Clinically, knowing which symptoms are likely to remit on their own may prevent unnecessary intervention. In this paper, we examine PTSD symptom presentation across the deployment cycle. We employed a sample of Marines in which we previously identified longitudinal trajectories of global PTSD severity (Nash et al., 2014), with the aim of deepening the granularity of analysis to symptom-level presentation. We used latent class analysis (LCA), a form of mixture modeling that identifies qualitatively distinct subgroups of individuals present in a sample (Nylund et al., 2007). Existing LCAs of military-related PTSD (Hebenstreit et al., 2014; Maguen et al., 2013; Naifeh et al., 2010; Steenkamp et al., 2012) have typically assessed PTSD years or decades after deployment, have only examined a single crosssection, and have mostly employed clinical samples. We extend this literature by examining PTSD symptom presentation in an active-duty, non-treatment-seeking cohort at four times points across the deployment cycle. We predicted that, compared to prior LCAs that examined outcomes more distally, we would find considerably greater variability because service members preparing for war and in the early stages of return from war were assumed to be in a state of flux. In line with prior findings, we expected that at each time point, the modal latent class would consist of low overall percentages of clinically significant symptoms. Because we evaluated a whole battalion of Marines (rather than samples of convenience or treatment-seekers), we predicted that a greater number of classes would be apparent than in previous studies. We also assumed that different demands associated with each time point would lead to different symptom presentations, such that the number and composition of classes would differ across time points. Finally, we examined three predictors of latent class membership – combat exposure (measured as number of prior deployments at pre-deployment), lifespan trauma, and avoidant coping – and predicted that all three would be associated with worse outcomes.

2. Method 2.1. Design and participants The data source for this study was the MRS, a longitudinal study of four battalion cohorts of active-duty male Marines deployed to Iraq and Afghanistan between 2008 and 2012 (Baker et al., 2012). Cohorts were evaluated in four separate large-scale data collections, timed to coincide with each cohort's deployment. Four assessment intervals were planned: one month prior to a 7month deployment, 1-week post-deployment, and 3- and 6months post-deployment. Overall, 2593 Marines completed the Time-0 pre-deployment assessment, 2317 (89.3%) completed the Time-1 assessment, 1901 (73.3%) completed the Time-2 assessment, and 1634 (63.0%) completed the Time-3 assessment. The MRS captured reactions to combat stress, such as PTSD, depression, and anxiety, and assessed an array of psychosocial predictors of risk and resilience, including prior lifetime trauma, avoidant coping, and combat exposure. Study procedures were approved by relevant institutional review boards. Participation at each assessment was voluntary and individual informed consent was obtained before enrollment at baseline with no senior unit leaders present. The current analyses focused exclusively on Cohort 4 (N ¼892). Cohorts 1 and 2 were not selected because the events to which PTSD scores were indexed changed from exclusively military at baseline to lifetime at post-deployment, complicating interpretability. Cohorts 3 and 4 provided interpretable over time comparisons, but Marines' training and leave schedules resulted in varied timing of assessment across cohorts. Cohort 3 and 4's modal

assessment points fluctuated by as much as 3 months, preventing cumulative assessment of these two cohorts. Cohort 4 entailed more evenly spaced modal assessment times. Cohort 4 deployed to Helmand Province in Afghanistan in late 2010 during which U.S. forces sustained their highest causality rates. To address variability around the modal postdeployment assessment times, we used the following procedures, outlined by King et al. (2006). For the three postdeployment assessments, scores on all measures were assigned to three follow-up date ranges determined by the count of days since the date of return from deployment. We aimed to minimize the dispersion of days within each date range and to maximize the number of included participants. The ranges of days that best fit the data were 20 to 40 days for T1 (M _ 30, SD _ 4), 140 to 160 days for T2 (M _ 153, SD _ 4), and 240 to 260 days for T3 (M _ 249, SD _ 5). In other words, on average, assessments occurred 1 month pre-deployment (T0; N¼859) and 1-month (T1; N¼554), 5-months (T2; N¼328), and 8-months (T3; N¼287) post-deployment. Marines who did not deploy (n¼ 4) and those who died during deployment (n¼17) were excluded from analyses. In the final sample, Marines were primarily Caucasian (83.1%). At baseline, participants had served an average of 3.10 (SD ¼ 3.15) years in the military and 51.54% had deployed at least once before. Participants' ages at baseline ranged from 18 to 43 (M¼ 23.16, SD ¼3.67); 68.1% had no more than a high school diploma, and 41.1% were married. A full table of the analysis of responders and non-responders for can be found in Table 1 of the supplemental materials, and sample information at each time point can be found in Tables 2–5. 2.2. Measures 2.2.1. Posttraumatic stress disorder The PTSD Checklist (PCL) is a psychometrically valid self-report measure (Weathers et al., 2001) of PTSD symptom severity. Symptoms endorsed as moderately (a value of 3 on a 1 to 5 scale) or above were considered to be clinically present (Weathers et al., 1993). PCL assessments were indexed to any lifetime trauma reported as currently most distressing so as not to artificially constrain the actual trauma-related burden in the sample; PTSD was not necessarily military-related. 2.2.2. Predictors of latent class membership At each time point, we assessed the influence of avoidant coping style, lifespan trauma, and war-zone exposure on latent class membership. We assessed avoidant coping strategies at each time point by aggregating subscales from the Brief COPE, a 28-item questionnaire that assesses 14 different current coping styles (Carver, 1997). To create an index of avoidant coping, we averaged scores on five 2-item subscales of the Brief COPE (Schnider et al., 2007): self-distraction, denial, behavioral disengagement, self-blame, and substance use (T0 α ¼0.81; T1 α ¼ 0.80; T2 α ¼0.81; T3 α ¼0.82). We used the Life Events Checklist (LEC) to assess lifetime exposure to 16 potentially traumatic events (Gray et al., 2004). A prior lifespan trauma composite was created by assigning a 1 to each item endorsed as happened to me or witnessed it and a 0 to all other responses, and then by summing across the 16 events. The LEC data collected at T0 included all prior lifetime events. At all post-deployment time points, the LEC data entailed only newonset events that occurred since the prior assessment. War-zone exposure was indexed as number of prior deployments at T0, and, at subsequent time points, was indexed using the Combat Experiences Scale (CES) from the Deployment Risk and Resilience Inventory (DRRI; King et al., 2006). The CES is a 16-item ‘yes/no’ scale that assesses individual or unit member exposure to

M.M. Steenkamp et al. / Journal of Affective Disorders 176 (2015) 87–94

war-zone-related stressors. The MRS employed a revised version in which response options represented frequency of exposure (0never to 4-daily or almost daily; see Vasterling et al., 2006), and items were limited to personal experiences. This measure was administered at T1 and was used to predict class membership at all post-deployment time points. A combat experiences composite was created by averaging across all items (α ¼ 0.91).

2.3. Data analysis strategy LCA is a person-centered analysis that uses binary indicators to identify patterns of responses, and assigns individuals to classes based on these patterns. All LCA models were estimated using Mplus (Version 7), and full information maximum likelihood (FIML) was employed to handle missing data. The best-fitting models were selected based on prior research, class sizes of at least 5% of the total sample, interpretability, formal fit indices, and classification quality (i.e., entropy). We interpreted fit statistics in accordance with findings from a Monte Carlo study by Nylund and colleagues (2007). At each cross-section, to substantiate the validity of class solutions, we conducted analyses of variance in SPSS to examine whether indicators of mental health such as depression, anxiety, and physical functioning predicted class membership. Next, with the most symptomatic class in each model serving as the comparison class, avoidant coping, prior lifetime trauma, and war-zone exposure (number of prior deployments at T0 and combat exposure at T1, T2, and T3) were tested as predictors of class membership and were simultaneously submitted to a multinomial logistic regression and a relative weights analysis.

3. Results Table 1 shows the model fit statistics for each cross-section. At T0, the Consistent Akaike's Information Criteria (CAIC) and the Bayesian Information Criteria (BIC) indicated a 3-class solution whereas the sample-size adjusted Bayesian Information Criteria (SSA-BIC) indicated that a 4-class solution was a comparatively better fit to the data. The CAIC and BIC tend to under-extract by one class whereas the SSA-BIC is slightly more accurate in determining the correct class solution; consequently, we selected the 4-class solution. At T1, the SSA-BIC, which is most accurate considering the sample size, indicated that the 6-class solution was a comparatively better fit to the data. At T2 and T3, both the CAIC and BIC indicated a 3-class solution, suggesting selection of the 4-class solution based on the findings of Nylund et al. (2007). However, at T3, the 4-class solution had a class size of less than 5% of the total sample; thus for T3 we selected the 3-class solution. Overall, based on comparative fit and interpretability, we selected 3 classes at T0, 6 classes at T1, 4 classes at T2, and 3 classes at T3. All class solutions had high entropy, high posterior probabilities (ranging from 0.84 to 0.99), and adequate class sizes. At each time point, classes that entailed significant percentages of clinically significant symptoms had higher percentages of clinical and subthreshold PTSD cases and greater scores on co-morbid conditions and functional impairment. Across all models, demographic characteristics differentiated between only a few class solutions (see supplemental materials Tables 2–5). Below, we describe each model in detail below, as well as the results of the multinomial logistic regression and relative weights analyses (see Table 2). LCA class solution names reflect symptom endorsement probabilities (greater than.5) across PTSD sub-clusters, as defined by the 5-factor model (Pietrzak et al., 2012), which we replicated in this sample (manuscript under review).

89

3.1. Pre-deployment At T0, the largest class (74.5%) was characterized by very low clinically significant endorsement probabilities across all PTSD symptoms (labeled Low Overall; see Fig. 1). Marines in the second largest class (12.4%; labeled Low-Moderate Overall) were also generally low overall, but higher numbers of Marines (ranging from 30% to 40%) reported experiencing clinically significant intrusions, emotional and physical reactivity, emotional avoidance, and hypervigilance. A third class (7.5%; labeled Anxious and Dysphoric Arousal) was characterized by a pattern similar to the Low-Moderate Overall class, but had elevated (above 50% of the sample) endorsement of anxious and dysphoric arousal symptoms, particularly anger and hypervigilance. The smallest class (5.6%; labeled Moderate-High Overall) had moderate to high probabilities of endorsing nearly all symptoms. Notably, across all classes, flashbacks, foreshortened future, difficulty concentrating, and exaggerated startle had considerably lower endorsements relative to other symptoms in a given cluster within a given class. The multinomial logistic regression revealed that relative to Marines in the Low Overall class, participants in the ModerateHigh Overall class endorsed more avoidant coping (b¼1.438, po 0.001) and reported more prior lifetime traumas (b ¼0.262, po 0.001). Marines in the Moderate-High Overall class also endorsed more avoidant coping (b¼ 0.735, p¼ 0.048) than those in the Low-Moderate Overall class. The results of the relative weights analysis generally parallel these findings; prior lifetime trauma and avoidant coping differentiated between the ModerateHigh Overall class and the Low Overall class. Number of prior deployments was a consistently weak predictor of class membership across both the relative weights and multinomial logistic regression analyses. Avoidant coping, prior lifetime trauma, and number of prior deployments failed to differentiate between the Moderate-High Overall and Anxious and Dysphoric Arousal class. 3.2. One month post-deployment T1 was characterized by noteworthy variability in symptom presentation, as evidenced by the 6-class solution. As with T0, the largest class (46.1%) was characterized by uniformly low endorsement probabilities across all symptoms (labeled Low Overall). We partitioned the remaining classes into ‘full symptom expression’ and ‘partial symptom expression’ classes (see Fig. 2). The two smallest classes endorsed a uniformly full expression of PTSD symptoms in that they were characterized by moderate or high endorsement probabilities across symptoms, respectively labeled Moderate Overall (7.6%) and High Overall (6.6%). Three mid-sized, similarly patterned classes were characterized by disturbances across all symptom clusters, with the exception of numbing, and were collectively termed ‘partial expression’ classes. These classes contained 21.9%, 10.4% and 7.6% of the sample, and were labeled Anxious Arousal, Re-experiencing, and High No Numbing, respectively. Consistent with the T0 model, in general across all classes, symptoms of flashbacks, foreshortened future, and amnesia had lower endorsement relative to other symptoms within a given cluster within a given class. For the multinomial logistic regression and relative weights analyses, similarly patterned classes were collapsed prior to analysis. That is, we analyzed differences between three groups: the Low Overall class, an aggregate of the partial expression classes, and an aggregate of the full expression classes. Multinomial logistic regression analyses revealed that Marines in the full expression classes engaged in more avoidant coping (b¼3.803, po 0.001), endorsed more prior lifetime traumas (b¼0.431, po 0.001), and reported fewer combat experiences (b¼ 0.705, p¼ 0.006) than participants in the Low Overall class. Compared

90

Table 1 Fit statistics for all cross-sections. Class

CAIC

Pre-deployment (N¼ 858)a 1 8083.32 2 6551.59 3 6538.93 4 6603.43 5 6693.95

SSA-BIC

BLRT

Entropy

Lowest posterior probability

Smallest class proportion

8066.32 6516.59 6485.93 6532.43 6604.95

8012.33 6405.44 6317.62 6306.96 6322.31

0.00 0.00 0.00 0.03

0.93 0.87 0.87 0.89

0.96 0.86 0.85 0.83

0.17 0.06 0.06 0.05

9381.00 7063.77 6623.38 6532.32 6503.80 6482.47 6488.69

0.00 0.00 0.00 0.00 0.00 0.00

0.93 0.90 0.88 0.89 0.89 0.89

0.95 0.92 0.92 0.91 0.84 0.84

0.28 0.15 0.06 0.06 0.07 0.04

One-month post-deployment (N¼ 554)a 1 2 3 4 5 6 7 Five-months post-deployment (N ¼ 317)a 1 5032.53 2 4117.71 3 4049.43 4 4097.80 5 4169.72

5015.53 4082.71 3996.43 4026.80 4080.72

0.91 0.89 0.90 0.92

0.97 0.93 0.90 0.91

0.30 0.10 0.08 0.07

Eight-months post-deployment (N¼ 287)a 1 3805.97 2 2960.36 3 2948.76 4 3015.41 5 3085.60

3788.97 2925.36 2895.76 2944.41 2996.60

0.95 0.11 0.93 0.93

0.98 0.91 0.93 0.88

0.28 0.91 0.04 0.04

CIAC-Consistent Akaike's Information Criteria, BIC-Bayesian Information Criterion, SSA-BIC-sample size adjusted BIC, BLRT-bootstrapped likelihood ratio test. a

Fit indices indicated by Nylund et al. (2007) to be less than 80% accurate in identifying a class solution for a given sample size were not considered and are not reported.

M.M. Steenkamp et al. / Journal of Affective Disorders 176 (2015) 87–94

BIC

M.M. Steenkamp et al. / Journal of Affective Disorders 176 (2015) 87–94

91

Table 2 Predictors by group comparisons per cross-section. W

95% CI

b

p

W

95% CI

b

p

Pre-deployment predictors by group comparison Moderate-high overall to low overall

Moderate-high overall to low-moderate overall

Avoidant coping 0.058 [0.00,0.11] 1.44 o 0.001 Prior lifetime trauma 0.059 [0.01,0.10] 0.26 o 0.001 Number of deployments 0.007 [  0.03,0.02] 0.04 0.741

0.029 0.000 0.003

One-month post-deployment predictors by group comparison Full expression to low overall Avoidant coping 0.213 [0.12,0.32] 3.80 o 0.001 Prior lifetime trauma 0.148 [0.07,0.22] 0.43 o 0.001 Combat experiences 0.055 [0.01,0.11] 0.71 0.006

Full expression to partial expression 0.128 [0.04,0.20] 1.94 o 0.001 0.058 [  0.01,0.11] 0.23 o 0.001 0.020 [  0.03,0.05] 0.28 0.173

Five-months post-deployment predictors by group comparison Moderate-high overall to low overall Avoidant coping Prior lifetime trauma Combat experiences

0.244 [0.11,0.41] 0.053 [  0.02,0.15] 0.098 [0.00,0.27]

3.36 o 0.001 0.17 0.098 0.83 0.044

Eight-months post-deployment predictors by group comparison Moderate-high overall to low overall Avoidant coping 0.154 [0.05,0.31] 2.53 o 0.001 Prior lifetime trauma 0.108 [0.01,0.28] 0.28 0.035 Combat experiences 0.144 [0.02,0.31] 1.58 0.022

[  0.01,0.12] [  0.05,0.02] [  0.03,0.04]

Moderate-high overall to arousal 0.194 [0.04,0.38] 0.006 [  0.07,0.06] 0.011 [  0.05,0.09]

0.74  0.02 0.16

0.048 0.764 0.320

anxious and dysphoric 1.87 0.07 0.25

o 0.001 0.523 0.561

W

95% CI

b

p

Moderate-high overall to anxious and dysphoric arousal 0.025 [  0.01,0.14] 0.52 0.184 0.001 [  0.03,0.04] 0.02 0.717 0.003 [  0.03,0.07]  0.07 0.664

Moderate-high overall to numb and dysphoric arousal 0.080 [  0.15,0.30] 0.96 0.095 0.018 [  0.24,0.10] 0.07 0.579 0.022 [  0.23,0.10] 0.43 0.381

Moderate-high overall to low-moderate overall 0.069 [  0.16,0.19] 1.27 0.056 0.064 [  0.13,0.17] 0.23 0.095 0.107 [  0.14,0.32] 1.13 0.096

W¼ Raw weight.

Fig. 1. Pre-deployment latent class analysis.

Fig. 2. One month post-deployment latent class analysis, full expression (left) and partial expression (right) PTSD symptom profiles.

to partial expression classes, Marines in full expression classes engaged in more avoidant coping (b¼1.940, p o0.001) and endorsed more prior lifetime traumas (b¼0.227, p o0.001).

Combat exposure was not significant. The patterns of significance for the relative weights analysis match those of the multinomial logistic regression with the exception of the marginally significant

92

M.M. Steenkamp et al. / Journal of Affective Disorders 176 (2015) 87–94

influence of the combat exposure predictor. Of note, avoidant coping accounted for a large portion of variance when the full expression classes were compared to both the Low Overall (21.3%) and the partial expression classes (12.8%). 3.3. Five months post-deployment At T2, the number of classes best fitting the data decreased to four. Again, a Low Overall class was most prevalent (57.0%). The second largest class, an Anxious and Dysphoric Arousal class (25.6%), showed elevations in arousal symptoms, in particular difficulty in sleeping and hypervigilance. A Moderate-High Overall class (9.5%) consisted of elevations on most symptoms, falling above 50% endorsement on all symptoms except amnesia and foreshortened future. A Numb and Dysphoric Arousal class (7.9%) was characterized by high endorsement probabilities on the numbing symptoms of diminished interest, detachment, and numbing, as well as the dysphoric arousal symptoms of difficulty sleeping and anger (see Fig. 3). As with the previous time points, across all classes, symptoms of flashbacks, amnesia, and foreshortened future either had very low clinically significant endorsement probabilities or had lower endorsement relative to other symptoms in a given cluster within a given class. The multinomial logistic regression revealed that participants in the Moderate-High Overall class reported more avoidant coping (b¼ 3.358, p o0.001) and reported more combat experiences (b¼ 0.834, p ¼0.044) than Marines in the Low Overall class. Marines in the Moderate-High Overall class also endorsed more avoidant coping (b¼1.866, po 0.001) compared to the Anxious and Dysphoric Arousal class. These three variables did not distinguish between the Moderate-High Overall and the Numb and Dysphoric Arousal class. The results of the relative weights analysis exactly mirror these findings. 3.4. Eight months post-deployment At T3, the largest class (64.4%) was again characterized by low endorsement probabilities across all symptoms (labeled Low Overall). The second largest class (25.0%) was characterized by moderate endorsement probabilities of dysphoric arousal and anxious arousal symptoms, and low endorsement probabilities of re-experiencing

and numbing symptoms (labeled Low-Moderate Overall). The smallest class was similarly patterned to the T2 Moderate-High Overall class, with the exception that it was characterized by slightly lower re-experiencing probabilities, higher arousal probabilities, and moderate numbing probabilities (10.6%; labeled Moderate-High Overall). Again, similar to all other cross-sections, there were low baserates of flashbacks, amnesia, and foreshortened future symptoms relative to other symptoms within the dysphoric arousal symptom cluster for a given class (see Fig. 4). For the multinomial logistic regression analysis, Marines in the Moderate-High Overall class endorsed more avoidant coping (b¼2.525, po 0.001) and reported more prior lifetime traumas (b¼0.282, p ¼0.035) and more combat experiences (b¼ 1.575, p¼ 0.022) than Marines in the Low Overall class. These differences were mirrored by the relative weights analyses; avoidant coping was the strongest predictor, accounting for 15.4% of the overall variance. None of the tested predictors differentiated between the Moderate-High Overall class and the Low-Moderate Overall class.

4. Discussion As predicted, there was heterogeneity in PTSD symptom presentation at each assessment point, and the extent of heterogeneity (i.e., the number of classes identified) differed by time point. The three classes evident at pre-deployment increased markedly to six classes at one month post-deployment, and then decreased to a four class solution at five months, and a three class solution at eight months. This suggests that a relatively wide variety of PTSD symptom presentations may be expectable in military cohorts soon after returning from war, ranging from low overall symptoms, to various partial expression classes, to high overall symptoms. Indeed, one month after return to the U.S., the percentage of Marines falling in the low symptom (‘resilience’) class had decreased starkly from 75% at pre-deployment to 46%, meaning that approximately half of Marines fell in classes endorsing at least some clinically significant PTSD symptoms. However, the progressively lower number of class solutions over the post-deployment phase implies a stabilization of this flux over time. Notably, the degree of heterogeneity in clinically significant symptoms in our sample exceeded that of prior LCAs in veterans, which examined symptom presentation more distally, typically years

Fig. 3. Five months post-deployment latent class analysis.

M.M. Steenkamp et al. / Journal of Affective Disorders 176 (2015) 87–94

93

Fig. 4. Eight months post-deployment latent class analysis.

after deployment. The three class solution evident at our final time point approximates previous LCA studies (e.g., Maguen et al., 2013; Steenkamp et al., 2012), suggesting that high symptom variability present in the first few weeks home from deployment may, with time, broadly stabilize into groups with high, medium, and low percentages of clinically significant symptoms. Our findings also illustrate that, in the immediate post-deployment flux, temporary subclinical or ‘partial expression’ presentations may mimic subclusters of more severe, unremitting profiles. For instance, the Anxious Arousal class, which was characterized by elevations in hypervigilance and exaggerated startle reactions, is a particularly noteworthy partial expression class one month post-deployment as it comprised almost one quarter of the sample (22%). Given that we also found high arousal in our ‘PTSD’ class at this time point, such prominent hyperarousal may be indicative of both PTSD and a common reaction in the initial weeks home that subsides naturally with time, underscoring the complexity of clinical prediction in this early phase; this tempers the value of using early hyperarousal symptoms in screening for later PTSD (e.g., Bliese et al., 2008). Indeed, Kimble et al. (2013) similarly found that deployment in and of itself can lead to hypervigilance, in that even veterans without PTSD can endorse noteworthy hypervigilance behaviors. What might account for returning service members only developing some clinically significant symptoms and not others? Our predictor analyses revealed that, one month post-deployment, Marines in the partial expression classes engaged in less avoidant coping and had fewer prior lifetime traumas than those in the full expression classes. Avoidant coping accounted for a large portion of variance when the full expression classes were compared to both the Low Overall (21.3%) and the partial expression classes (12.8%). This may be due either to avoidant coping being a catalyzing pre-condition for the full spectrum of PTSD symptoms to develop, or due to a greater propensity to engage in avoidant coping when dealing with a greater number of symptoms. More broadly, all three of our predictors predicted in expected directions, with prior lifetime trauma, avoidant coping, and greater combat exposure generally predicting worse outcomes. The consistently low general endorsement of flashbacks, amnesia, and foreshortened future symptoms in our sample also warrants discussion. It is unclear theoretically why these symptoms would be endorsed at lower rates than other symptoms in the syndrome. Flashbacks and amnesia can loosely be considered

dissociative symptoms, and a recent latent profile analysis (a conceptually similar, but statistically distinct, analysis to LCA) of PTSD in VA veterans proposed a dissociative subtype of PTSD (Wolf et al., 2012). However, even within the dissociative subclass identified in that study, flashbacks, amnesia, and foreshortened future remained the three lowest endorsed DSM-IV PTSD symptoms. In our sample, flashbacks were infrequently endorsed prior to deployment, but were then endorsed at a greater than 50% rate in two of the symptomatic classes at T1, particularly the high overall (‘PTSD’) class. By T2, the probability of endorsement of flashbacks among these most symptomatic Marines fell to 60% and subsequently dropped to less than 20% by T3. Future studies should examine the natural resolution of flashbacks in recently returned veterans. Amnesia and foreshortened future, in contrast, were endorsed at a greater than 50% probability of endorsement in only one class at only one time point: the high overall (‘PTSD’) class at T1 (see also Bliese et al., 2008). This suggests that, when endorsed, these two symptoms may be strong indicators of high levels of overall distress and psychopathology, and may be useful for screening purposes.

5. Limitations Limitations of the current analyses include the self-report nature of the data, which depend on participant self-awareness and insight into the potential problem being assessed, and as such may under-represent or exaggerate actual symptoms. Our sample may have limited generalizability due to our lack of women and inclusion of only combat Marines. Attrition and re-ranging of data resulted in significant missing data and affected the representativeness of the sample; Marines excluded because of attrition were older and more likely to have previously deployed, and had greater functional impairment and more prior lifetime trauma. Lastly, it is important not to reify classes: the mixture modeling approach to classification is data-driven and latent classes are merely aspects of a complex distribution. The selection of latent classes relies on indices of comparative fit, which often indicate different solutions. To minimize the impact of these limitations, we examined the interpretability of the latent classes and substantiated our class

94

M.M. Steenkamp et al. / Journal of Affective Disorders 176 (2015) 87–94

solutions by testing covariates such as depression and anxiety, which distinguished classes in anticipated directions.

6. Conclusions Despite these limitations, the current findings significantly extend previous LCAs of military-PTSD, demonstrating that symptom-level variability is highest in the month following deployment and then stabilizes over time into high, medium, and low symptom severity profiles. Clinically, our findings hold important implications for PTSD screening soon after return from deployment: should assessments occur too soon, they will capture common and transient (though still impairing) early post-deployment reactions, particularly high anxious arousal, that may remit by themselves over the following months.

Role of funding source This study was funded by VA Health Service Research and Development (SDR 09-0128), the Navy Bureau of Medicine and Surgery, and Headquarters, U.S. Marine Corps. The funding source was involved in study design and data collection and was not involved in analysis and interpretation of the data, in writing the report and in the decision to submit the article for publication.

Conflict of interest None.

Acknowledgments This study was funded by VA Health Service Research and Development (SDR 09-0128), the Navy Bureau of Medicine and Surgery, and Headquarters, U.S. Marine Corps. The authors acknowledge the Marine Resiliency Study (MRS) team who made this work possible.

Appendix A. Supporting information Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.jad.2015.01.043. References Andrews, B., Brewin, C.R., Stewart, L., Philpott, R., Hejdenberg, J., 2009. Comparison of immediate-onset and delayed-onset posttraumatic stress disorder in military veterans. J. Abnorm. Psychol. 118, 767–777. Baker, D.G., Nash, W.P., Litz, B.T., Geyer, M.A., Risbrough, V.B., Nievergelt, C.M., O’Connor, D.T., Larson, G.E., Schork, N.J., Vasterling, Hammer, P.S., WebbMurphy, J.A., MRS Team, 2012. Predictors of risk and resilience for posttraumatic stress disorder among ground combat Marines: methods of the Marine Resiliency Study. Prev. Chronic Dis. 9, E97.

Bliese, P.D., Wright, K.M., Adler, A.B., Cabrera, O., Castro, C.A., Hoge, C.W., 2008. Validating the primary care posttraumatic stress disorder screen and the posttraumatic stress disorder checklist with soldiers returning from combat. J. Consult. Clin. Psychol. 76, 272. Bremner, J.D., Southwick, S.M., Darnell, A., Charney, D.S., 1996. Chronic PTSD in Vietnam combat veterans: course of illness and substance abuse. Am. J. Psychiatry 153, 369–375. Carver, C.S., 1997. You want to measure coping but your protocol's too long: consider the Brief COPE. Int. J. Behav. Med. 4, 92–100. Gray, M.J., Litz, B.T., Hsu, J.L., Lombardo, T.W., 2004. Psychometric properties of the life events checklist. Assessment 11, 330–341. Hebenstreit, C., Madden, E., Maguen, S., 2014. Latent classes of PTSD symptoms in Iraq and Afghanistan female veterans. J. Affect. Disord. 166, 132–138. Kimble, M.O., Fleming, K., Bennion, K.A., 2013. Contributors to hypervigilance in a military and civilian sample. J. Interpers. Violence 28, 1672–1692. King, D.W., King, L.A., McArdle, J.J., Grimm, K., Jones, R.T., Ollendick, T.H., 2006. Characterizing time in longitudinal trauma research. J. Trauma. Stress 19, 205–215. King, L.A., King, D.W., Vogt, D.S., Knight, J., Samper, R.E., 2006. Deployment Risk and Resilience Inventory: a collection of measures for studying deployment-related experiences of military personnel and veterans. Mil. Psychol. 18, 89–120. Maguen, S., Madden, E., Bosch, J., Galatzer-Levy, I., Knight, S.J., Litz, B.T., Marmar, C.R., McCaslin, S.E., 2013. Killing and latent classes of PTSD symptoms in Iraq and Afghanistan veterans. J. Affect. Disord. 145, 344–348. Naifeh, J.A., Richardson, J.D., Del Ben, K.S., Elhai, J.D., 2010. Heterogeneity in the latent structure of PTSD symptoms among Canadian veterans. Psychol. Assess. 22, 666–674. Nash, W.P., Boasso, A.M., Steenkamp, M.M., Larson, J.L., Lubin, R.E., Litz, B.T., 2014. Posttraumatic stress in deployed Marines: prospective trajectories of early adaptation. J. Abnorm. Psychol. (Advance online publication). Nylund, K.L., Asparouhov, T., Muthén, B.O., 2007. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Struct. Equ. Model. 14, 535–569. Pietrzak, R.H., Tsai, J., Harpaz-Rotem, I., Whealin, J.M., Southwick, S.M., 2012. Support for a novel five-factor model of posttraumatic stress symptoms in three independent samples of Iraq/Afghanistan veterans: a confirmatory factor analytic study. J. Psychiatr. Res. 46, 317–322. Schell, T.L., Marshall, G.N., Jaycox, L.H., 2004. All symptoms are not created equal: the prominent role of hyperarousal in the natural course of posttraumatic psychological distress. J. Abnorm. Psychol. 113, 189–197. Schnider, K.R., Elhai, J.D., Gray, M.J., 2007. Coping style use predicts posttraumatic stress and complicated grief symptom severity among college students reporting a traumatic loss. J. Couns. Psychol. 54, 344–350. Solomon, Z., Horesh, D., Ein-Dor, T., 2009. The longitudinal course of posttraumatic stress disorder symptom clusters among war veterans. J. Clin. Psychiatry 70, 837–843. Steenkamp, M.M., Nickerson, A., Maguen, S., Dickstein, B.D., Nash, W.P., Litz, B.T., 2012. Latent classes of PTSD symptoms in Vietnam veterans. Behav. Modif. 36, 857–874. Vasterling, J.J., Proctor, S.P., Amoroso, P., Kane, R., Heeren, T., White, R.F., 2006. Neuropsychological outcomes of army personnel following deployment to the Iraq war. JAMA 296, 519–529. Weathers, F., Keane, T., Davidson, J., 2001. Clinician-administered PTSD scale: a review of the first ten years of research. Depress. Anxiety 13, 132–156. Weathers, F.W., Litz, B.T., Herman, D.S., Huska, J.A., Keane, T.M., 1993. The PTSD Checklist (PCL): reliability, validity, and diagnostic utility. In: Proceedings of the 9th Annual Conference of the ISTSS, San Antonio, TX. Wolf, E.J., Miller, M.W., Reardon, A.F., Ryabchenko, K.A., Castillo, D., Freund, R., 2012. A latent class analysis of dissociation and posttraumatic stress disorder: evidence for a dissociative subtype. Arch. Gen. Psychiatry 69, 698–705.