Discriminating Mild Traumatic Brain Injury and Post-Traumatic Stress Disorder Using Latent Neuroimaging and Neuropsychological Profiles in Active-Duty Military Service Members

Abstract

Objective:

Mild traumatic brain injury (mTBI) and post-traumatic stress disorder (PTSD) commonly occur among military Service Members and Veterans and have heterogenous, but also overlapping symptom presentations, which often complicate the diagnoses of underlying impairments and development of effective treatment plans. Thus, we sought to examine whether the combination of whole brain gray matter (GM) and white matter (WM) structural measures with neuropsychological performance can aid in the classification of military personnel with mTBI and PTSD.

Method:

Active-Duty U.S. Service Members (n = 156; 87.8% male) with a history of mTBI, PTSD, combined mTBI+PTSD, or orthopedic injury completed a neuropsychological battery and T1- and diffusion-weighted structural neuroimaging. Cortical, subcortical, ventricular, and WM volumes and whole brain fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AD) were calculated. Latent profile analyses were performed to determine how the GM and WM indicators, together with neuropsychological indicators, classified individuals.

Results:

For both GM and WM respectively, a four-profile model was the best fit. The GM model identified greater ventricular volumes in Service Members with cognitive symptoms, including those with a diagnosis of mTBI, either alone or with PTSD. The WM model identified reduced FA and elevated RD in those with psychological symptoms, including those with PTSD or mTBI and comorbid PTSD. However, contrary to expectation, a global neural signature unique to those with comorbid mTBI and PTSD was not identified.

Conclusion:

The findings demonstrate that neuropsychological performance alone is more robust in differentiating Active-Duty Service Members with mTBI and PTSD, whereas global neuroimaging measures do not reliably differentiate between these groups.

Introduction

Up to 20% of U.S. Service Members report traumatic brain injury (TBI) or probable TBI^1-4, with ~80% or more being mild TBI (mTBI)⁵. MTBI often results in varied symptoms including headache, decreased processing speed, memory and attention performance, and psychosocial symptoms^2,6,7. Many military personnel with mTBI will recover to normal function, but a significant proportion experience chronic cognitive complaints^2,8. Volumetric cortical and subcortical gray matter (GM) abnormalities are typically associated with moderate and severe TBI, but subtle changes are noted in the cingulate, hippocampus, amygdala, and basal ganglia after mTBI, although location and extent of volume loss are inconsistent across studies⁹. Altered GM shape characteristics have also been found in the basal ganglia and thalamus showing unique associations with attention, processing speed, memory, and subjective symptoms in Service Members with mTBI^10-12.

However, white matter (WM) microstructure is more vulnerable in mTBI as the injury causes stretching and shearing of axons, resulting in traumatic axonal injury (e.g., damage to cytoskeletal structure and axonal membranes), which disrupts cell function and impairs neurotransmission¹³. Diffusion tensor imaging (DTI) can be used to detect axonal injury by measuring alterations in global WM microstructure^14,15. Fractional anisotropy (FA), a commonly assessed scalar DTI metric, indicates the directional preference of diffusion where higher values indicate greater diffusion along the axon than across it, reflecting more intact axonal structure^14,16. Military personnel with mTBI show heterogenous and spatially distributed patterns of reduced FA in regions including the corpus callosum (CC), corona radiata (CR), superior longitudinal fasciculus (SLF), internal capsule (IC), and posterior thalamic radiation (PTR)^17-19 which are associated with declines in verbal memory and executive functioning^17,20. Other scalar diffusivity metrics can provide further WM microstructure characterization. Specifically, mean diffusivity (MD) represents the overall level of diffusion while radial diffusivity (RD) and axial diffusivity (AD) reflect diffusion perpendicular and parallel to the axon, respectively^14,16. However, the direction of change of diffusion metrics and WM regions across studies are inconsistent, potentially due to injury heterogeneity (e.g., injury mechanism, pathophysiological processes, clinical symptoms)^13,21.

Understanding the effects of mTBI in military personnel is further complicated by the presence of comorbid psychiatric conditions such as post-traumatic stress disorder (PTSD). Up to half of Veterans from Operation Enduring Freedom and Operation Iraqi Freedom (OEF/OIF) with combat-related mTBI meet criteria for PTSD^3,22, and deployment-related mTBI increases the risk of PTSD^4,22. PTSD alone contributes to cognitive compromise^2,23 and reduced health-related quality of life²⁴. MTBI co-morbid with PTSD presents challenges for diagnosis and treatment due to the high degree of symptom overlap (e.g., fatigue, mood disturbances, memory dysfunction etc.)^25,26. Moreover, the presence of mTBI and PTSD together may exacerbate psychological symptoms leading to more severe impairments in cognitive and psychological functioning^27-29. Similar brain structures are implicated in mTBI and PTSD and may contribute to these impairments. Limbic structures including the hippocampus, amygdala, cingulate, and their WM connections are vulnerable in both mTBI and PTSD as these regions are located in areas particularly susceptive to biomechanical forces and are involved in affective and cognitive processing^28,30,31. Greater GM and WM microstructure abnormalities in limbic regions are associated with more severe PTSD symptoms in comorbid mTBI/PTSD^31-33.

We previously used latent profile analysis (LPA) to characterize cognitive and psychological performance in U.S. Active-Duty Service Members with mTBI, PTSD, and mTBI with comorbid PTSD³⁴. The neuropsychological indicators identified five classes based on symptom presentation: high functioning, normal functioning, cognitive symptoms, psychological symptoms, and combined cognitive and psychological symptoms. These symptom clusters closely aligned with the diagnostic classifications, but also demonstrated heterogeneity within diagnostic groups. Critically, those with mTBI with comorbid PTSD were almost split evenly among the cognitive symptoms, psychological symptoms, combined cognitive and psychological symptoms, and normal functioning groups. Here, we extend these analyses to include GM and WM measures to determine whether the addition of these metrics can aid classification of participants. Given the heterogeneity of mTBI, PTSD, and comorbid mTBI and PTSD, focusing on the individual patient’s specific symptoms and corresponding GM/WM metrics, rather than diagnosis, may be a more useful guide for selecting and evaluating treatment strategies. We focused on whole brain, global measures of GM and WM brain structures as these measures of brain health are easier for clinicians to integrate into their assessments.

Materials and Methods

Participants

Participants (N = 156; 87.8% male) were drawn from a large sample of U.S. Active-Duty Service Members recruited from a military treatment facility, as previously described³⁴. Participants were included in the current study if they had completed a neuroimaging assessment. The study consisted of four groups including: 1) mTBI only, 2) mTBI with PTSD symptoms, 3) PTSD only, and 4) orthopedic controls (Table 1). The study protocol was approved by the local hospital IRB and the Human Research Protection Office at the U.S. Army Medical Research and Materiel Command. All participants provided informed consent prior to participation.

Table 1.

Demographic Characteristics.

	All	OI Control	PTSD	mTBI	mTBI+PTSD	Significance
N	156	60	22	30	44
Age	35.3 (8.4)	37.4 (7.3)	37.6 (6.2)	29.6 (6.5)	35.2 (10.3)	F = 7.07, p < .001 (mTBI group different)
Sex (Male/Female)	137 / 19	48 / 12	20 / 2	29 / 1	40 / 4	χ2 = 6.21, p = .10
Race
White	106 [67.9%]	38 [63.3%]	13 [59.1%]	23 [76.7%]	32 [72.7%]	χ2 = 13.70, p = .033 (Bonferroni corrected post-hocs not significant, but OI group had more Black/African-American and less Other race)
Black or African-American	29 [18.6%]	18 [30%]	3 [13.6%]	2 [6.7%]	6 [13.6%]
Other race	21 [13.5%]	4 [6.7%]	6 [27.3%]	5 [16.7%]	6 [13.6%]
Ethnicity
Hispanic or Latino	44 [28.2%]	13 [21.7%]	10 [45.5%]	9 [30%]	12 [27.3%]	χ2 = 4.57, p = .21
Not Hispanic or Latino	112 [71.8%]	47 [78.3%]	12 [54.5%]	21 [70%]	32 [72.7%]
Education
High School / GED	68 [43.6%]	15 [25.0%]	8 [36.4%]	22 [73.3%]	23 [52.3%]	χ2 = 24.23, p < .001 (mTBI group had more high school and less college and post graduate degrees)
College (Associates or Bachelor’s)	63 [40.4%]	29 [48.3%]	10 [45.5%]	8 [26.7%]	16 [36.4%]
Post Graduate Degree	25 [16%]	16 [26.7%]	4 [18.2%]	0 [0%]	5 [11.4%]
Years of Military Service	13.2 (7.6)	15.7 (6.7)	14.5 (5.9)	7.7 (5.7)	13.1 (8.9)	F = 8.81, p <.001 (mTBI group different)
Number of Deployments	2 (1)	2 (0.9)	2.1 (1.3)	1.9 (0.9)	2.1 (1.1)	F = 0.28, p = .84
PTSD Checklist Military Version	42.1 (18.5)	27.2 (12.4)	60.6 (11.7)	33.6 (7.1)	58.9 (10.2)	F = 100.09, p < .001 (All groups differed except PTSD and mTBI+PTSD)
Mechanism of Injury
Blast	-	-	-	21 [70%]	22 [50%]	χ2 = 3.75, p = .29
Vehicular	-	-	-	1 [3.3%]	6 [13.6%]
Fall	-	-	-	3 [10%]	6 [13.6%]
Other	-	-	-	5 [16.7%]	10 [22.7%]
Number of Prior TBI Reported	-	-	-	2.4 (4.2)	3.3 (4.4)	t = 0.83, p = .41
Time Since Injury (Months)	-	-	-	10.3 (5.6)	10.1 (6)	t = 0.10, p = .92
Loss of Consciousness	-	-	-	11 [36.7%]	20 [45.5%]	χ2 = 1.40, p = .50
Post-Traumatic Amnesia	-	-	-	1 [3.3%]	6 [13.6%]	χ2 = 2.21, p = .14

Note: Mean with standard deviation in parentheses. Frequency with percentage in square brackets. One OI control was missing data for deployments. Two participants (1 mTBI and 1 mTBI+PTSD) had missing data for number of prior TBI. One mTBI+PTSD had missing data for loss of consciousness.

mTBI Participants:

mTBI participants (n=74; 69 males, 5 females) were Active-Duty Service Members who took part in a large cognitive rehabilitation clinical trial study (Study of Cognitive Rehabilitation Effectiveness (SCORE!)³⁵. The diagnosis of mTBI was made using Veteran Affairs (VA)/Department of Defense Clinical Practice Guidelines⁶ following a screening interview and medical record review by experienced TBI and Concussion Center medical staff. Furthermore, each participant was required to have persistent cognitive symptoms which was defined as having a Neurobehavioral Symptom Inventory (NSI) score of 3 or higher on any of the four cognitive symptoms³⁵. Participants were included if they were 18 to 55 years old, sustained a closed head injury during deployment (OEF/OIF/Operation New Dawn [OND]) activities 3 to 24 months prior to recruitment, and were able to understand and communicate in English. Given that mTBI often coexists with PTSD, we further classified our mTBI group into mTBI with potential comorbid PTSD using the PTSD Checklist - Military Version (PCL-M) cut-off score of >44^36-38. This resulted in a mTBI-only group (n=30; 29 male) and a mTBI with PTSD group (n=44; 40 male).

PTSD Participants:

PTSD participants (n=22; 20 males, 2 females) included Active-Duty Service Members recruited through the Department of Behavioral Health Clinic. Potential participants underwent a screening interview and medical record review and were excluded if they had a previous closed head injury. PTSD control participants were required to have a deployment related Clinician-Administered PTSD Scale (CAPS; DSM-IV criteria) confirmed diagnosis of PTSD. In addition, PTSD participants were selected using similar inclusion criteria including age range, deployment history, and English communication skills.

Orthopedic Participants:

Orthopedic injury controls (n=60; 48 males, 12 females) were Active-Duty Service Members recruited through the Orthopedic Clinic. Potential participants underwent a screening interview and medical record review and were excluded if they had a previous closed head injury and/or a PTSD diagnosis according to the standardized CAPS. In addition, orthopedic injury control participants were selected using similar inclusion criteria including age range, deployment history, and English communication skills.

Participants were excluded (regardless of diagnostic group) if they had neurologic comorbidities (i.e., seizures, psychosis), history of moderate/severe TBI, spinal cord injury, were on scheduled narcotic pain medications, unable to use dominant hand, or had abnormal MRI findings. Attempts were made to include patients in each group such that the groups would be similar in age, rank, and sex distribution.

Measures

Demographic, Clinical, and Cognitive Variables.

Demographic information (e.g., age, sex, education, number of deployments, years in service) and injury history (e.g., time since injury, loss/alteration of consciousness, post-traumatic amnesia, mechanism of injury, number of prior injuries) were collected. Participants completed self-report psychological questionnaires including the Alcohol Use Disorder Identification Test (AUDIT), Neurobehavioral Symptom Inventory (NSI), PCL-M, and the Depression and Anxiety subscales from the Symptom Checklist 90 (SCL-90). Technician-administered assessments were used to assess cognitive performance: memory: Total Recall, Short Delay Free Recall, Long Delay Free Recall and Recognition Hits from the California Verbal Learning Test-II (CVLT-II); verbal fluency: Delis-Kaplan Executive Function System (D-KEFS Category and Letter Fluency subtests); attention: Paced Auditory Serial Addition Test (PASAT); executive function: time to complete Trail Making Test Part B minus Part A (TMT B-A); working memory and processing speed: Working Memory Index (Digit Span and Letter-Number Sequencing) and Processing Speed Index (Coding and Symbol Search) from the Wechsler Adult Intelligence Scale 4^th edition (WAIS-IV). Mean self-report scores or mean performance are reported in Supplemental Table 1.

Neuroimaging Acquisition.

Multimodal neuroimaging assessments were collected on a single 3T Siemens Verio Syngo scanner with 32-channel head coil. MRI and cognitive assessments were collected concurrently (same week, mode=1 day). T1-weighted and DTI sequences were analyzed in the current study.

The T1-weighted sequence was acquired with the following parameters: field of view (FOV)=256 mm, repetition time (TR)=2300 ms, echo time (TE)=2.98 ms, flip angle=9°, and slice thickness=1 mm. The DTI sequence was collected using a clinical TBI protocol with the following acquisition parameters: 64 noncollinear directions, baseline acquisition sequences at b=0 s/mm with diffusion-weighting sequences at b=1000 s/mm, 1.8 x 1.8mm in-plane resolution with slice thickness of 4mm, TR=6300 ms, TE=96 ms, FOV=203 mm. T1-weighted data were available from 156 participants and DTI data were available from 144 participants.

Volumetric Image Processing and Analysis.

The T1-weighted images were visually inspected for artifacts including motion, complete anatomy coverage, and inhomogeneities, and there were no observable lesions on the images. The raw DICOM images were processed in Freesurfer v6.0.0. (http://surfer.nmr.mgh.harvard.edu/) with the recon-all command, resulting in fully segmented and labeled images. The processed images were visually inspected for any errors (e.g., skull stripping, segmentation, labeling) and no manual editing was necessary. Volumetric data were extracted from the aseg.stats file and the variables SubCorGrayVol, CortexVol and CerebralWhiteMatterVol were used to estimate volumes of subcortical GM, cortical GM, and WM, respectively, based on the standard Freesurfer atlas ³⁹. Total ventricular volume was calculated by summing the Freesurfer variables representing the lateral, inferior lateral, third and fourth ventricles. All volumes were corrected for total intracranial volume, and thus all values inputted for analysis were a ratio.

DTI Processing and Analysis.

Enhancing NeuroImaging Genetics through Meta-Analysis (ENIGMA)-DTI protocols were implemented for processing and extraction of scalar metrics from diffusion-weighted images ⁴⁰. For each participant, diffusion images underwent standardized preprocessing steps including eddy current correction, brain extraction and masking, and tensor fitting^41,42. Data were then processed using fMRI Software Library (FSL) tract-based spatial statistics^41,42 tools though predefined ENIGMA templates and atlases were used in the registration and region of interest (ROI) mapping steps. The ENIGMA-ROI extraction protocol uses the Johns Hopkins University WM atlas⁴³ to extract ROI values as well as average scalar metrics (i.e., FA, MD, AD, and RD) across the entire WM skeleton. Only the average whole brain metrics were used in the main analyses.

Statistical Approach

LPA was used to examine the sample based on neuroimaging and neuropsychological indicators. Model 1 included the GM metrics (subcortical GM, cortical GM, ventricular volume) and 15 cognitive and psychological indicators. Ventricular volume was included in the GM model as ventricular enlargement can be an indirect measure of cerebral atrophy more generally⁴⁴. Model 2 included the WM metrics (WM volume, average FA, MD, RD, AD) and the 15 cognitive and psychological indicators. Models were fitted in MPlus Version 8.6. The final models were selected based on the model fit statistics (Akaike Information Criterion [AIC], Bayesian Information Criterion [BIC], and sample-size adjusted BIC values [SABIC]⁴⁵, Lo-Mendell-Rubin Likelihood [LMR]⁴⁶, Bootstrapped Likelihood Ratio Test [BLRT]⁴⁷, and entropy⁴⁵) and conceptual meaning (for full description of model fitting procedures, see Supplemental Material and Esopenko et al.³⁴). GM/WM and neuropsychological indicators are presented so higher values indicate higher volumes/scalar metrics and better performance, respectively.

Post-estimation provided information on the proportion of mTBI, PTSD, mTBI+PTSD and orthopedic controls for each of the identified classes. For each LPA model, one-way ANOVAs with class membership as the between-subjects variable were used to assess separation between classes. Significant ANOVAs were followed with Tukey’s HSD post-hoc tests to determine significant differences between classes (p < 0.05). Semantic labels for class profiles were assigned based on previous findings in the larger sample³⁴. Labels were determined by two neuropsychologists who were blinded to participant representation (i.e., clinical diagnosis group) and based on performance across neuropsychological indicators within each class. For example, normal cognitive function with high psychological symptoms was labelled Psychological Symptoms class.

Exploratory post-hoc analyses with GM and WM ROIs were performed to further explore the whole brain LPA results and are described in the Supplemental Material.

Results

Model 1: GM

Class Determination.

The structural GM measures and neuropsychological indicator variables were fit into latent profile models with k=2 to 5 classes. Models with 2 to 4 classes replicated the log-likelihood values and were well-identified. The 5-class model had untrustworthy standard error parameter estimates, indicating poor model identification. The 4-class model minimized the log-likelihood values, AIC, BIC, and SABIC relative to the 2- and 3-class models. The BLRT demonstrated that the 4-class model performed significantly better than a 3-class model. While the LMR for the 4-class model was not significant, the entropy values were well above the .8 threshold and were relatively stable for all k-class models up to k=4 (Table 2). One-way ANOVAs comparing class membership for the 4-class model demonstrated good separation between the classes for the neuropsychological indicators (p’s < 0.007) and for ventricular volume (p=0.04), but not for the subcortical and cortical GM indicators.

Table 2.

Model Fit Statistics for Gray Matter and White Matter Latent Profile Analyses.

Classes	Class Proportions	Loglikelihood	AIC	BIC	SABIC	BLRT	p BLRT	LMR	p LMR	Entropy
GM Model (n = 156)
2	0.56 / 0.44	−8539.163	17188.33	17356.07	17181.98	619.91	p<.001	613.52	.002	.933
3	0.33 / 0.49 / 0.19	−8435.054	17018.11	17243.80	17009.56	208.22	p<.001	206.07	.337	.933
a4	0.21 / 0.38 / 0.22 / 0.19	−8340.402	16866.80	17150.44	16856.07	189.30	p<.001	187.35	.194	.947
5	Model not well-identified
WM Model (n = 144)
2	0.54 / 0.46	−8419.603	16961.21	17142.37	16949.35	548.95	p<.001	543.74	.003	.922
3	0.51 / 0.33 / 0.15	−8316.315	16796.63	17040.16	16780.69	206.58	p<.001	204.62	.178	.933
a4	0.19 / 0.19 / 0.24 / 0.38	−8231.739	16669.48	16975.37	16649.45	169.15	p<.001	167.55	.417	.939
5	0.16 / 0.23 / 0.33 / 0.15 / 0.13	−8168.981	16585.96	16954.22	16561.85	125.51	p<.001	124.32	.348	.956
6	Model not replicated

Note: GM = Gray matter; WM = White matter; AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; SABIC = sample-size adjusted BIC; BLRT = Bootstrapped Likelihood Ratio Test; LMR = Lo-Mendell-Rubin Likelihood.

Lower values of AIC, BIC, SABIC demonstrate better model fit. p_BLRT < .05 indicates that the k-class model is a better fit to the data than the k-1 class model. p_LMR small p-values indicate that the k-class model fits better to data than the k-1 class model. Entropy value close to 1 indicates excellent classification of subjects into latent classes.

^aDenotes the selected models. Although the 5-class WM model (under WM model) is the best fit model regarding the fit indices, it does not demonstrate the appropriate separation of the classes compared to 4-class model.

Model Interpretation.

Figure 1 panel A demonstrates how the GM and neuropsychological indicators load for the 4-class model (labels adopted from Esopenko et al.³⁴). The neuropsychological indicators loaded according to the classes: 1) High Functioning, 2) Normal Functioning, 3) Cognitive Symptoms, 4) Psychological Symptoms. Military personnel demonstrating cognitive symptoms showed elevated ventricular volumes (p=.031) relative to the normal functioning group. There were no other group differences. Age and education did not differ between the classes.

Figure 1. Gray matter (GM) model.

Panel A. Standardized means plotted for the latent profile analysis with GM and neuropsychological indicators. The 4-class GM model identified high functioning, normal functioning, psychological symptoms, and cognitive deficit classes. Right inset: A magnified view of the class loadings of the gray matter indicators for the 4-class GM model. Panel B. Distribution of participants from each diagnostic classification to each class for the GM model. Left: The allocation of participants from each diagnostic class to each of the classes of the 4-class GM model. The thickness of each line between diagnostic group and class represents number of participants (thicker line = more participants). Right: The percentage of class assignments for each diagnostic classification in the 4-class GM model. GM Indicators: Subcortical GM = total subcortical gray matter volume; Cortical GM = total cortical gray matter volume; Ventricles = total ventricular volume. Psychological Indicators: AUDIT = Alcohol Use Disorder Identification Test; NSI = Neurobehavioral Symptom Inventory; PCL-M = Posttraumatic stress disorder Checklist – Military Version; Depression = Symptom Checklist 90 (SCL-90) Depression subscale; Anxiety = SCL-90 Anxiety Subscale; Cognitive Indicators: Total Recall = California Verbal Learning Test-II (CVLT-II) total recall; Short Delay = CVLT-II short delay free recall; Long Delay = CVLT-II long delay free recall; Recog Hits = CVLT-II recognition hits; Cat Fluency = Delis-Kaplan Executive Function System (D-KEFS) category fluency; Lett Fluency = D-KEFS Letter Fluency; PASAT = Paced auditory serial addition test; TMT B-A = Trail Making Test Part B minus Part A; Proc speed = Wechsler Adult Intelligence Scale-IV (WAIS-IV) – Processing Speed Index; Work Mem = WAIS-IV Working Memory Index. OI Control = orthopedic injury control; PTSD = post-traumatic stress disorder only; mTBI = mild traumatic brain injury only; mTBI+PTSD = comorbid mTBI and PTSD.

Diagnostic Classification.

Figure 1 panel B shows how participants from each diagnostic category were distributed to each class for the 4-class model, which was similar to the distribution in the neuropsychological model³⁴.

Exploratory post-hoc analyses did not significantly differentiate subcortical ROI indicators (see Supplemental Material).

Model 2: WM

Class Determination.

The structural WM measures and neuropsychological indicator variables were fit into latent profile models with k=2 to 6 classes. Models with 2 to 5 classes replicated the log-likelihood values and were well-identified; the 6-class model was not. The 5-class model minimized the log-likelihood values, AIC, BIC, and SABIC relative to the other models. The BLRT demonstrated that the 5-class model performed significantly better than a 4-class model. While the LMR for the 5-class model was not significant, the entropy values were well above the .8 threshold and were relatively stable for all k-class models up to k=5 (Table 2). One-way ANOVAs comparing class membership for the 5-class model demonstrated good separation between the classes for the neuropsychological indicators (p’s < 0.004), but not for the WM indicators (p’s > 0.089). Only the 4-class model demonstrated separation between the classes for the neuropsychological indicators (p’s < 0.015) as well as FA (p=0.024) and RD (p=0.025) and thus was selected as the best fitting model. Age and education did not differ between the classes.

Model Interpretation.

Figure 2 panel A demonstrates how the WM and neuropsychological indicators load for the 4-class model (labels adopted from Esopenko et al.³⁴). The neuropsychological indicators loaded according to the classes: 1) High Functioning, 2) Cognitive Symptoms, 3) Psychological Symptoms, 4) Combined Cognitive and Psychological Symptoms. Relative to the high functioning group, the psychological symptoms subgroup demonstrated similar WM volumes, but altered WM microstructure with reduced FA (p=0.018) and elevated RD (p=0.034). There were no other group differences.

Figure 2. White matter (WM) model.

Panel A: Standardized means plotted for the latent profile analysis with WM and neuropsychological indicators. The 4-class WM model identified high functioning, cognitive symptoms, psychological symptoms, and combined cognitive / psychological symptom classes. Right inset: A magnified view of the class loadings of the WM indicators for the 4-class WM model. Panel B: Distribution of participants from each diagnostic classification to each class for the WM model. Left: The allocation of participants from each diagnostic class to each of the classes of the 4-class WM model. The thickness of each line between diagnostic group and class represents number of participants (thicker line = more participants). Right: The percentage of class assignments for each diagnostic classification in the 4-class WM model. WM Indicators: WM Vol = Total white matter volume; FA = Fractional anisotropy; MD = Mean diffusivity; AD = Axial diffusivity; RD = Radial diffusivity. Psychological Indicators: AUDIT = Alcohol Use Disorder Identification Test; NSI = Neurobehavioral Symptom Inventory; PCL-M = Posttraumatic stress disorder Checklist – Military Version; Depression = Symptom Checklist 90 (SCL-90) Depression subscale; Anxiety = SCL-90 Anxiety Subscale; Cognitive Indicators: Total Recall = California Verbal Learning Test-II (CVLT-II) total recall; Short Delay = CVLT-II short delay free recall; Long Delay = CVLT-II long delay free recall; Recog Hits = CVLT-II recognition hits; Cat Fluency = Delis-Kaplan Executive Function System (D-KEFS) category fluency; Lett Fluency = D-KEFS Letter Fluency; PASAT = Paced auditory serial addition test; TMT B-A = Trail Making Test Part B minus Part A; Proc speed = Wechsler Adult Intelligence Scale-IV (WAIS-IV) – Processing Speed Index; Work Mem = WAIS-IV Working Memory Index. OI Control = orthopedic injury control; PTSD = post-traumatic stress disorder only; mTBI = mild traumatic brain injury only; mTBI+PTSD = comorbid mTBI and PTSD.

Diagnostic Classification.

Figure 2 panel B shows how participants from each diagnostic category were distributed to each class for the 4-class model. The distribution was similar to the GM model as well as the neuropsychological model³⁴.

Exploratory post-hoc analyses showed that the psychological symptoms and combined cognitive and psychological symptoms groups demonstrated reduced FA in the splenium of the CC, CR, and SLF compared to the cognitive symptoms and high functioning groups. Further, the psychological symptoms group had reduced FA in the IC and PTR compared to the cognitive symptoms and high functioning groups (see Supplemental Material).

Discussion

MTBI and PTSD commonly occur in OEF/OIF Veterans^3,22 and have diverse, but overlapping effects on neuropsychological outcomes^25,26 as well as on brain structure³⁰. These effects are often exacerbated in those with a history of mTBI and comorbid PTSD^27-29,31-33. We previously showed that neuropsychological performance can differentiate military personnel with mTBI and/or PTSD into distinct symptom profiles ³⁴. Here, we extended that work to examine whether the addition of whole brain GM and WM neuroimaging metrics with neuropsychological performance better cluster military personnel with mTBI and PTSD into distinct profiles.

The findings mostly reproduced the neuropsychological results reported by Esopenko et al.³⁴ in a smaller subset of the sample, demonstrating the robustness of neuropsychological measures to classify participants. However, our results indicated that the neuroimaging metrics were less able to differentiate classes, as evidenced by the stark distinction between the performance across neuropsychological indicators but less distinct differences in GM and WM metrics between the classes. Moreover, compared to the neuropsychological only model described in Esopenko et al.³⁴, the GM model omitted the combined cognitive and psychological symptoms class, whereas the WM model omitted the normal functioning class. This may be partially due to the smaller sample size resulting in 48% of Active-Duty Service Members in the combined cognitive and psychological symptoms and 36% in the normal functioning classes in the neuropsychological model not being included in the structural models due to missing neuroimaging data, likely reducing the sensitivity of the models to detect these classes. However, the neuropsychological measures were better able to differentiate classes relative to global structural variables given the limited sample size, likely due to the inclusion of multiple measures that captured subtle distinctions in cognitive and psychological function. Nonetheless, the GM model identified elevated ventricular volumes in Active-Duty Service Members with cognitive symptoms, while the WM model identified reduced FA and elevated RD in those with psychological symptoms.

Across models, the cognitive symptoms group primarily contained military personnel with mTBI, either alone or with PTSD. In the GM model, only the ventricles showed differences between the classes, with the cognitive symptoms group (~40% with mTBI; ~32% mTBI+PTSD) showing larger ventricles relative to the other classes, yet the WM pattern was less distinct for the cognitive symptoms group. This is consistent with past work indicating that ventricular size is an indicator of gross, overall brain health⁴⁴, with larger size typically associated with increased age and/or reduced health. While a longitudinal study of military Service Members with mTBI showed no differences in ventricle-to-brain ratios compared to no mTBI history⁴⁸, the results may support that ventricular expansion is predictive of cognitive complaints⁴⁹. Contrary to these results, we had expected a more definitive structural injury related to mTBI encompassing cortical and subcortical GM and WM. Exploratory post-hoc analyses did not reveal subcortical specific effects of mTBI, possibly reflecting the gross nature of GM subcortical volume measurements. The exploratory analyses also did not distinguish the region-specific FA measures in the cognitive symptoms group (see Supplemental Material). The lack of effect for WM may be due to the heterogenous effects of mTBI on these structures across individuals that becomes diluted when averaged at the group level⁹. Furthermore, we expected a worse structural injury pattern in those with mTBI and comorbid PTSD; however, the combination of diagnoses may have obscured the global neural signature. These findings underscore the difficulty of diagnosing those with both mTBI and PTSD, and further, demonstrates the heterogeneity of symptom presentation and lack of consistent neural signature.

Similar to our past work, the psychological symptoms group included mainly PTSD and mTBI with current PTSD and showed reduced FA and elevated RD; no GM effects were shown. Given past studies demonstrating reduced GM volumes in the hippocampus and amygdala in PTSD^50,51, the absence of a global GM signature may suggest a more subtle, region-specific pattern of GM abnormalities that cannot be detected using whole brain measures. As such, exploratory post-hoc analyses that examined subcortical ROIs were performed but did not show evidence of altered regional volumes in the psychological symptoms group (see Supplemental Material). The WM signature is consistent with past studies showing WM abnormalities in PTSD^52,53 and given the sensitivity of these MRI metrics may represent a better way of assessing differences between these patient groups. Similarly, the exploratory post-hoc analyses showed reduced FA in several WM regions including the splenium of the CC CR, IC, PTR, and SLF in the psychological symptoms group (see Supplemental Material). Larger samples will be needed to replicate these findings and explore WM microstructure in other ROIs. Together, these results suggest that reduced WM microstructure with evidence of psychological symptoms may be a biomarker of PTSD and/or mTBI with comorbid PTSD.

For those with mTBI and comorbid PTSD, whole brain structural neuroimaging in the current statistical models did not add any additional information beyond the information found in the neuropsychological performance when classifying patients. Thus, use of state-of-the-art neuropsychological measures are currently the most beneficial for clinicians to use to direct and track treatment strategies. Future prospective research studies that examine rehabilitation methods may want to use the latent neuropsychological features described in these manuscripts as a way of determining the efficacy of various treatment strategies as one might expect that different neuropsychological profiles warrant different or more coordinated treatment approaches. Nonetheless, the findings provide preliminary evidence supporting the potential for neuroimaging markers to distinguish mTBI and PTSD in clinical settings.

Limitations

Study limitations are discussed in detail in Esopenko et al.³⁴, but for the current work main limitations are that the overall sample size was reduced due to missing neuroimaging data, had few female Service Members, and the sample sizes were not equal across the diagnostic classifications. Moreover, mTBI with comorbid PTSD was not diagnosed prior to study enrollment. Instead, cut-off scores on the PCL-M were used to define this group. As this is a self-report measure, the outcomes on the PCL-M may not match clinician-confirmed diagnoses. However, the PCL has good sensitivity and specificity in high-risk populations and is a reasonable alternative to the gold standard diagnostic interview⁵⁴. Future studies should examine individuals diagnosed by clinicians using current diagnostic criteria. TBI assessments used standardized language and injuries were corroborated with in theatre medical records. However, assessments focused on deployment related injuries only and a validated instrument assessing lifetime TBI was not used. Further, while efforts were made to age match groups during recruitment, the mTBI group was significantly younger than the other groups (see Table 1). The effect of age was not examined in this study, but aging has been shown to significantly affect total brain, GM, and WM volumes and microstructure⁵⁵, and should be examined in future studies when sample sizes allow. Finally, whole brain measures were used, and while exploratory post-hoc analyses were performed, the study was not powered to robustly examine all ROIs that may be sensitive to the effects of mTBI and PTSD. In addition, while atrophy is a common consequence of pathology, it is possible to have dysfunctional tissue that is not necessarily atrophy. As such, future studies should assess whether the properties of specific regions provide better classification of individuals with mTBI and/or PTSD.

Conclusions

The findings demonstrate a lack of consistent global neural signature in military mTBI and PTSD which may be one mechanism underlying the heterogeneity of symptom presentation and difficulty in treating mTBI and PTSD. Clinically, these results further suggest that, unless otherwise indicated, neuropsychological measures alone may be the best tool for identifying distinct symptom profiles that clinicians can use to guide treatment planning. However, the findings shed light on the potential for neuroimaging markers to distinguish military personnel with mTBI and PTSD.