A. Research Question, Goal, or Specific Aims
Traumatic experiences cause acute posttraumatic stress symptoms (PTSS) in survivors, which can be exacerbated by comorbidly incurred mild to moderate traumatic brain injury (mTBI). Although PTSS remit in most survivors over days to weeks, these symptoms can persist over months, leading to PTSD in 8 – 44% of trauma survivors. Understanding PTSD development after acute trauma will improve early prediction and prevention of PTSD and provide insight into mechanisms that distinguish PTSD development from PTSD free recovery.
Neuroimaging studies suggest chronic PTSD patients have functional and structural changes in emotion-related brain regions. Both our studies and those by other teams suggest that early brain changes within 2 weeks after trauma may predict PTSD in subsequent months. These early changes include reduced hippocampal volumes and greater medial prefrontal cortex (mPFC) emotion appraisal activation at days after trauma. Further progressive changes over the post-trauma 3 months involve decreases in prefrontal cortical volume and mPFC emotion appraisal activation and increases in emotion responses in insular cortex. Furthermore, alterations in parietal cortical thickness and visual activation in days after mTBI caused by car accidents may contribute to deficits in emotion processing, predicting PTSS severity at 3 months later. However, these preliminary small sample studies are limited in comprehensively detecting early post-trauma brain changes associated with acute PTSS and development of PTSD, therefore existing findings are inconsistent.
We have been working on mega-analyses of brain structure in ENIGMA PTSD and TBI Working Groups and developing advanced high-resolution vertex-based cortical machine learning based analyses. Therefore, we propose a new mega-analysis of early and progressive brain structural measures after acute trauma to predict and differentiate PTSD development and recovery without PTSD. A large database of longitudinal neuroimaging data after acute trauma will be built in this project to facilitate further analyses by researchers in ENIGMA working groups.
A. Analytic Plan
1. Data collection
We will provide participating laboratories an updated data collection package to collect cortical, subcortical, and subfield data processed using FreeSurfer 6 or 7. Brain structural measures, symptoms and imaging parameters will be collected according to different post-trauma durations, e.g., Time-pre (pre-trauma), Time-1 (within 30 days after trauma), Time-2 (any time afterward), …, Time-n after trauma).
Any acute or longitudinal data collected in ENIGMA – Psychiatric Genetics Consortium (ENIGMA-PGC) PTSD Neuroimaging Working group and ENIGMA TBI Working Group will be identified and included after PI approves.
Additional data will be requested from data repositories, e.g., The Federal Interagency Traumatic Brain Injury Research (FITBIR) Informatics System (https://fitbir.nih.gov), OpenNeuro (https://openneuro.org), and NIMH Data Archive (NDA).
2. Data organization
Data will be collected at Duke in PTSD group and University of Utah (UoU) in TBI group.
ENIGMA QC protocols will be applied in participating labs. Central site will exclude the problematic measures from analyses.
We will organize and standardize data across multiple studies at the central site.
A major goal of this project is to build a large database of longitudinal neuroimaging data after acute trauma, which will facilitate future analyses. The database will be available to any researchers following the policies of PTSD and TBI working groups. We will also ask PIs for permissions to contribute data to NDA if the data has not been previously submitted to NIH depositories.
3. Data analysis
Aim 1. Identify early post-trauma alterations associated with subsequent PTSD development.
– analysis 1: a mega-analysis using a linear mixed-effects modeling approach to compare brain structural measures collected within days (less than a month) after trauma among trauma-exposed subjects who were or were not diagnosed with PTSD at month(s) after trauma. Healthy subjects without trauma exposure will be used as controls to test effects of trauma exposure. Effects of confound factors will be evaluated. Expect outcomes: This analysis of a large dataset will increase statistical powers to detect the brain measure most impacted in the early post-trauma period for prediction of PTSD development in subsequent months.
– analysis 2: using machine learning algorithms to predict PTSD diagnosis or symptom severity in subsequent months using early post-trauma brain structural measures. We will test different algorithms and approaches, e.g., SVM, random forest, multi-view learning, and our novel multi-vertex pattern and network analysis (MVPNA) to improve the accuracy of prediction and then identify structural measures and/or network connections that contribute to the prediction. Expected Outcome: The large database allows development of algorithms for effective and reliable prediction of PTSD. This work will evaluate the predictive power of early post-trauma brain structural measures for PTSD.
Aim 2. Identify different trajectories associated with PTSD development versus recovery without PTSD of brain structural measures over post-trauma periods in trauma-exposed subjects.
– analysis 1: comparison of trajectories of brain structural changes in PTSD and non-PTSD subjects after trauma exposure. The healthy subjects without trauma exposure will be used as controls. Repeated-measure ANOVA and growth curve modeling within subjects approaches and cross-subject post-trauma time course analysis will be conducted. We will also explore novel analytical approaches, e.g., machine learning multi-task survival analysis or recurrent neural network, to improve group differentiation across the post-trauma periods. Expected outcomes: The large database, including data at multiple post-trauma durations, will reveal the trajectories of brain structural changes during PTSD development. This analysis may also identify the post-trauma development of brain deficits reported in chronic PTSD patients.
analysis 2: correlation and machine learning analyses of trajectories of brain structural measures and progression of PTSD and other symptoms. Expected outcome: the analysis will identify the brain changes that are associated with PTSD symptom development.
Aim 3. Identify early post-TBI changes in brain structural measures and their progressions after acute TBI that are associated with trajectories of neuropsychological and mental health symptoms after TBI. Subjects with assessment for TBI after the index trauma will be identified.
– analysis 1: TBI subjects will be compared with non-TBI injured and with healthy subjects to identify early TBI-related brain changes. Confound factors, e.g., injury mechanism and TBI severity, will be considered.
– analysis 2: the mixed-effect and prediction analyses of early post-trauma brain structural measures proposed in above Aim 1 will be applied in subjects who were identified as sustaining or not sustaining TBI during the index trauma. The measures of neuropsychological (e.g., physical symptoms, attention, and memory deficits) and psychiatric symptoms (e.g., PTSD, depression and anxiety) at early and subsequent months after an acute head injury will be used as outcome measures in the mixed-effect analysis or machine learning-based prediction analysis of early TBI changes identified in analysis 1. Expected outcomes: this analysis will identify the early TBI-related brain structural changes associated with development of symptoms at early and prolonged post-trauma periods.
– analysis 3: the progression analyses proposed in Aim 2 will be repeated to differentiate TBI subjects from non-TBI injured or healthy subjects to examine the progression of TBI-related brain changes. The trajectories of TBI-related brain changes will also be tested with progression of symptoms during the same periods. Expected Outcomes: The proposed analyses will (1) identify early TBI-related brain structural changes that may improve the detection of TBI, particularly mild TBI; (2) explore associations between TBI effects on the brain and post-injury neurological and psychological problems, and (3) reveal the trajectories of TBI-related brain changes using unique multi-cohort data across multiple post-injury durations.
A. Analytic Personnel
We will form a team of acute trauma experts to design the variables and data collection, and then we will develop this team as an analytical/writing group for multiple publications. ENIGMA-PTSD and ENIGMA TBI Working groups are encouraged to participate in proposed analyses and initiate new analyses using the database.
The current team includes:
Xin Wang PhD, Raj Morey MD, Emily Dennis PhD, Israel Liberzon MD, Christine Larson PhD, Chiahao Shih PhD and Hong Xie PhD will contribute to design variables and interpretations and publications of results.
Raj Morey MD, Emily Dennis PhD, Israel Liberzon MD, Christine Larson PhD, and Xin Wang PhD will contribute to data collection.
Hong Xie PhD and Chiahao Shih PhD will contribute to data organization and quality control.
Chiahao Shih PhD will perform proposed statistical analyses for Aims. Emily Dennis PhD will supervise the analyses.
Erin O’Leary MS will perform machine learning analyses. Kevin Xu PhD will supervise machine learning analyses and develop new algorithms.
A. Resources Needed
Structural data will be collected including the following:
- cortical thickness of both parcellated regional mean thickness and normalized vertex thickness maps,
- cortical surface area of parcellated regions,
- volumes of parcellated cortical and subcortical regions and subfields of hippocampus, thalamus, amygdala, and brainstem.
Data of symptoms and demographics are as follows:
- age at trauma, age at initial MRI scan, gender, race, and ethnicity,
- index trauma types, post-trauma durations, additional trauma,
- PTSS and PTSD diagnosis,
- TBI symptoms, neuropsychological tests, and diagnosis,
- co-morbidity symptoms (depression, anxiety, substance/alcohol usage),
- psychotropic medications, treatment history, and
- childhood adversity history, social supports, and trauma history
Additional information on the MRI data acquisition and process:
- MRI scanner manufacture and models,
- imaging protocol, FOV, matrix, slice thickness, voxel size,
- FreeSurfer version
We have built two computing servers, one at UToledo and one at TAMU, for data analysis. They are sufficient for our novel multi-vertex pattern and network analysis (MVPNA) using machine learning approaches.