Design
This is a two-arm, randomized, double-blinded (with respect to scoring and data analysis), single-center, controlled clinical trial. All participants will provide written informed consent before participating in any study procedures. Participants will be randomly assigned to one of the two intervention conditions (ESA or EIA).
The study has three phases: pre-testing, intervention, and post-testing (see Figure 1 and Figure 2). In the pre-testing phase, we collect demographic information including psychological assessments, neuropsychological assessments, cognitive assessments, MRI data, and daily life measures (refer to Table 1 for measures list). The pre-testing is conducted at two adjacent university sites located in Dallas, TX, USA. The intervention consists of two arms (ESA and EIA) and is administered via laptop computer (assisted by an experimenter when needed during eight check-in sessions) over the course of four weeks. The experimenter checks in with each participant at the beginning and mid-point of each week during the intervention. The experimenter check-in ensures that participants are putting in sufficient time on the intervention each week and enables the participant to ask for clarifications on their performance as needed. The remote laptop delivery enables these interventions to be mobile and to occur at any location. The post-testing phase largely repeats the testing procedures carried out at our Dallas, TX study site (psychological, neuropsychological, and cognitive assessments, MRI data, and daily life measures (refer to Table 1)). When possible we employ alternate versions of tests to minimize practice effects in the post-testing session (the Trail Making Test is the only measure we use that does not enable multiple versions for test-retest purposes, though we anticipate that practice effects may occur for some measures).
Participants
We will recruit 100 participants (ages 18-55) with a history of mild-to-moderate range TBI as indexed by the Ohio State University (OSU) TBI assessment [18]. Participants will be veterans of the U.S. military. Participants will only include people who can speak and comprehend English, as not all of the standardized and experimental cognitive tests have been normed for non-English speakers. This study has been approved for research with human subjects by the Institutional Review Boards (IRB)s of the University of Texas Southwestern Medical Center at Dallas (IRB#8843) and The University of Texas at Dallas (IRB#11-43) and complies with the Declaration of Helsinki.
Inclusion Criteria: Selection criteria include: Male and female veterans, who have sustained a TBI at least 3 months previously, can comprehend simple instructions, perform the tests, and take part in the interventions. Participants will be screened and enrolled at the Center for BrainHealth® at The University of Texas at Dallas. Selection criteria also includes participants who can safely have an MRI (though this is not exclusionary), who can tolerate at least 2 hours of testing sessions at a time, and who can participate in tasks involving motor abilities such as use of at least one arm and hand. No racial/ethnic groups will be excluded, although the participants must be able to speak, read, and comprehend English sufficiently to complete testing sessions.
Exclusion Criteria: Exclusionary criteria include lack of proficiency in reading, comprehending, and speaking English, pre-existing cerebral palsy, mental retardation, autism, schizophrenia, pervasive developmental disorder, major depression, psychosis, active behavioral disorder, or uncontrolled epilepsy. Many individuals who have sustained a TBI during military service also express Post-Traumatic Stress Disorder (PTSD) symptoms. We are sensitive to these and measure their presence (using the PTSD Checklist (PCL)) [19], but endorsement of some symptom level is not exclusionary. Optimally, all participants will be capable of having an MRI, but due to the possibility of shrapnel or metal-related injuries within the sample population, this is not exclusionary to participation (without MRI). Participants cannot be actively engaged in another cognitive intervention concurrently.
Feasibility Estimate. Both performance sites: The Center for BrainHealth® and the Brain Performance Institute are centrally located in the Dallas-Fort Worth metropolitan area allowing access to a large and diverse population including numerous individuals who have sustained TBIs either in or outside of military service. Over the past six months, we have been able to recruit an average of four to five individuals per month who qualify for the study. This level of patient flow indicates that we should be able to meet our sample size within two years, meeting the goals for the project-funding period. We will make information about the study available via clinicaltrials.gov, veteran social media (Facebook, Instagram sites relevant to veteran populations), the Internet (www.project-expedition.org), and veteran events (university and college veteran center fairs/meetings).
Screening. Participants are screened for inclusion/exclusion criteria during a structured interview, which includes use of the OSU TBI assessment, the Glasgow Outcome Scale-Extended (GOS-E) [20] and a set of relevant demographic, medical, PTSD, and TBI questionnaires. Presence of PTSD symptoms is screened for in the current study, but history of PTSD symptoms is not exclusionary (as noted above). In addition to the severity level indicated by the OSU TBI assessment, we seek confirmatory evidence based on Glasgow Coma Scale (GCS) [21] documentation of prior medical evaluations when available, though lack of a documented GCS is not exclusionary. Given the limited availability of GCS scores, particularly for mild TBI participants, we ultimately rely on GOS-E functional severity and OSU TBI assessment to define TBI severity.
Primary Outcome Measures. The VMET and TBI Awareness Questionnaire are the primary outcome measures.
The VMET is an executive function measure implemented in a game-based simulation and is based on Shallice and Burgess’ Multiple Errands Test [22] that was carried out in real world environments. It requires participants to plan and execute several task goals within a set of pre-defined rules. Success depends on implementing memory, inhibitory control, and planning needs in a novel context. Greater performance on this measure for the ESA group over the EIA group would represent “near transfer” of executive function improvements in a game-based simulated context [23]. VMET data are scored by tallying the total numbers of task failures, inefficiencies, strategies, rule-breaks, and interpretation failures [12]. A task failure occurs when a subtask is not completed satisfactorily. An inefficiency occurs when a more effective strategy could have been used in order to accomplish the task. Strategies can be used that facilitate carrying out the tasks and the use of possible strategies is scored based on their effectiveness. Rule-breaks represent violation of a rules-set that are listed on the task instructions sheet. Interpretation failures are cases in which the requirements of a particular task are misunderstood.
The TBI Awareness Questionnaire [13] is a survey measure asking participants to evaluate their cognitive abilities at the current time compared to pre-injury and serves as another “far transfer” real-world evaluation of current TBI impairment. We predict the ESA will improve TBI symptom recovery compared to EIA as measured by this instrument.
Secondary Outcome Measures.
Secondary outcome measures are employed to characterize participants cognitive and emotional functioning and to address changes in cognition related to the interventions. The GOS-E is included (initial GOS-E score is used to assess TBI symptom severity, the second administration is an indication of change from initial baseline). The OSU TBI assessment is used to assess TBI history and severity. The PCL-Specific (PCL-S) is used to screen for Post-Traumatic Stress Disorder (PTSD) symptoms. We assess depression and anxiety using the Beck Depression Inventory (BDI) [24] and Beck Anxiety Inventory (BAI) [25]. The Alcohol Use Disorders Identification Test (AUDIT) [26] and the alcohol, smoking and substance involvement screening test (ASSIST) [27] are used to screen for substance use history. The Test of Premorbid Functioning (TOPF) [28] is given to assess pre-TBI intelligence. Cognitive executive function measures are gathered using the ANAM (Matching-to-sample Test, Mathematical Processing Test, Code Substitution Test, Stroop test, and Tower Puzzle) along with the Trail Making Tests (part A and B), and the Similar Situations Task.
Interventions.
In the current randomized control trial, we study the efficacy of a functional cognitive training software tool applied to individuals experiencing cognitive deficits due to chronic TBI. The program is delivered via four separate simulation modules over the course of one month. There are over one hundred different possible specific playable instances within each module enabling the intervention to remain novel for the participant throughout the month of training. All navigation occurs within game-based simulations using a mouse. All interactions are achieved by dragging, pointing, and clicking the mouse. This mouse interface is designed to be easy for participants to use and was adopted to avoid human-computer interaction challenges that some individuals may experience when using a Virtual Reality headset, or specialized gaming controller. Each simulation enables increased challenge to be delivered based on increasing the complexity of the simulations being delivered to the participant. The simulations scale in difficulty according to participant achievement. The level of challenge will be increased over the course of repeated success, while challenge level will be reduced if a participant repeatedly performs poorly on a given module over several playable instances.
Smart Phone usage is nearly ubiquitous in the United States at the time of this study. In accordance with this trend, all modules include an electronic version of a smartphone displayed in the lower left corner of each simulation. The phone serves two potential functions. First, it can help participants with information, much like a real world smart phone. Second, it can distract participants, again mimicking the capabilities of a real world smart phone. Participants are encouraged to use the smart phone for helpful purposes, but minimize its capability for distraction by selectively ignoring it when it provides no advantage to task completion.
Week one: The first week focuses on the construct prospective working memory improvement. This is implemented through the operation of gathering a set of items that meet the task goals. These are applied to the function of simulated packing for a vacation, or work-related trip. Participants navigate a small apartment (bedroom, closet, and bathroom) and are challenged to meet a set of goals for their trip by obtaining goal-relevant clothing and travel-related items from a series of closets and dressers. Items that may be selected for packing can be organized in a staging area and transferred to a set of virtual suitcases by clicking, dragging, and dropping the items.
Working memory demand increases as participants succeed on the task (through accuracy and speed). If task demands are met multiple times (based on a proficiency algorithm), difficulty scales up. This involves the addition of more items needed to successfully pack for progressively longer trips (the number of days that the trip specifies in an instructions screen) (adding to overall item load) and by greater complexity (e.g., business presentation and side trip to an exotic location requiring specific items). Refer to Figure 3.
The smart phone in Week One serves the helpful function of providing texts to remind participants of certain task requirements to focus on as they pack. The phone also distracts from successful performance by buzzing irrelevant text messages to the participant throughout the simulation.
Week Two: The second week focuses on two constructs: planning and executing a plan. This is operationalized by asking participants to generate a successful sequence and to follow that sequence. Participants are asked to accomplish the functional goal of planning a trip on an unfamiliar subway transit system and then carry out that plan by navigating on a virtual subway car. The planning phase occurs on an interactive transit map in which the participant selects a route that balances time and money (refer to Figure 4). After planning their route, participants navigate within a subway system moving to the correct platforms, boarding the correct trains, and exiting at the appropriate stops in order to carry out their plan.
Participant success, which is based on a proficiency algorithm, determines the difficulty level of subsequent trials. Week Two challenge level increases by requiring participants to accomplish a more complex plan involving multiple subway lines and with a greater number of stops in between the start and goal. Added difficulty is also implemented in the form of greater distraction during plan execution, as added noise plays on advertising screens in the virtual subway cars and on platforms.
Participants can use the smart phone to access the subway map. This device also serves as extra distraction by buzzing to indicate the arrival of task-irrelevant text messages periodically during the simulation.
Week Three. The third week is aimed at improving the construct, long-term memory. This is operationalized through learning new facts. Week Three taxes the functional goal of effective encoding of new information. Participants are asked to visit a series of virtual museums to view and learn about specified topics. They freely navigate different attractions within multiple museums. After completing their museum visit, they complete a quiz probing their memory for the specific items (Figure 5).
Difficulty is increased by requiring participants to visit multiple exhibits and multiple museums. Both of these factors increase the overall amount of information that the participants must encode and remember for subsequent memory tests.
As in the prior weeks, the in-game smart phone can be accessed. It can negatively affect performance by buzzing to indicate the arrival of irrelevant text messages to the participants.
Week Four: In the fourth week, participants are asked to take on all three scenarios in sequence, as they would in a real life travel situation (prospective working memory, planning/execution, and long-term memory). Week four serves two purposes. First, it demands greater cognitive flexibility upon participants. Second, it emulates an extended task sequence that one could encounter in daily life (pack for a trip, navigate to a location, and visit a destination). This final week may serve as an important indicator of overall success and performance in Week Four may be most predictive of gains on outcome measures.
Participants will be excluded from further participation if they fail to meet assigned time on the simulations for any given week. We will afford participants an opportunity to pick up on simulation play the following week, but if they do not meet the five hour criteria we will move to exclude them. If participants are unable to carry out the activities in any of the weeks of the intervention (or are unwilling to), we will discontinue their enrollment. We will make efforts to retain all enrolled participants by reaching out (at least two times per week during the intervention) to ensure that they understand the intervention and are able to complete it. We will also reach out via email and texting if participants are comfortable with those methods.
Neuroimaging. Each participant is transported a short distance by the experimenter to a nearby building where the MRI scanner is housed. After completing the MRI safety screening form, participants view the experimental task pre-instructions under the guidance of the experimenter. Participants are then positioned in the scanner, and then the experimenter and a technician run the experiment and scanning protocols, respectively. The protocol for this neuroimaging acquisition is derived from our previous research investigating TBI [29, 30]
Functional MRI design. The fMRI Face/Scene Selection Task [31] is performed. The experiment consists of six blocks divided into three runs. All problems follow a similar format: encoding period showing four images for 4s total, followed by a 4s delay and a 2s probe image. Two are attention blocks, where the participant is instructed attend to the faces or the scenes with the goal of judging whether the probe image matches and responds with a match/no match button press. One block is a categorization task, where the participant simply presses the right button for a scene and left button for a face.
Prior to the above experimental task, an independent localizer task is used to isolate face and scene selective areas of the brain [31]. The functional localizer includes seven 16 s blocks of grayscale images of faces and scenes.
Lastly, two runs of seven minutes (14 minutes in total) of echo-planar imaging are acquired with participants in a resting-state, for which they are instructed to lie still and remain awake in the scanner.
FMRI data acquisition. Imaging is performed on a 3T Scanner (Siemens MAGNETOM Prisma). Functional images are acquired with an echo-planar image sequence sensitive to blood oxygen level dependent (BOLD) contrast (TE 30 ms, TR 2 s, α flip angle 70°). The whole brain is scanned using an 80 × 80 matrix with approximately 37 transverse slices (4 mm thickness with a 0 mm inter-slice gap yielding a voxel size of 3 × 3 × 4 mm). Three runs of the selective attention task and one run of the localizer task are acquired. Structural images are acquired and these include a magnetization-prepared, rapid access gradient-echo image sequence (voxel size 1 × 1 × 1 mm) and a fluid-attenuated inversion recovery image sequence (voxel size 1 × 1 × 1 mm).
Diffusion tensor imaging (DTI) design—image acquisition. DTI images are obtained using a single shot, spin-echo, echo-planar imaging sequence with field of view (FOV) = 224 mm, slice thickness/gap = 2/0 mm, 75 slices, repetition time = 7,500 ms, echo time = 65.7 ms, flip angle = 90 degrees, number of excitations (NEX) = 1, posterior-to-anterior phase encoding direction, and a matrix of 112 × 112. The diffusion-sensitizing gradients are applied at a b value of 1000 s/mm2 per axis with 40 non-colinear directions and 1 b0 image. The acquisition time is 7 minutes. The voxel size is 2 × 2 × 2. Another b0 image is also acquired with opposite phase encoding direction to correct for susceptibility artifacts in the DTI images.
Assignments to counterbalanced testing-imaging components. New participants are assigned to counterbalanced testing and imaging components by fixed rotation, except when assignments are unbalanced by discontinued participants. In such cases, any underrepresented counterbalance category is filled first.
Testing and imaging. Each of the testing and neuroimaging sessions occurring at the pre-training phase and post-training phase is administered by experimenters trained and experienced in their respective procedures.
Random assignment to groups. To maintain an even balance throughout the study’s duration, participant randomization is completed in runs of 10 individuals using a computerized random number generator to establish a random order of group assignment. Enrolled participants are therefore randomly assigned to either the ESA or the EIA groups prior to their pre-testing appointment. Random group assignment is implemented via study personnel that are not involved in participant recruitment, communication, scheduling, nor the administration of pre- and post-training assessments. Group assignment is blinded to all but two of the study personnel via server-side folder-level permissions. Group assignment and participant identification numbers (i.e., randomly generated four-digit numbers that de-identify participants for confidentiality) are only present together during group assignment and within our in-house database, both of which are blinded to all other study personnel via server-side folder-level permissions.
Procedures for double blinding. Each experimenter involved in collecting the pre-testing, post-testing, and neuroimaging measures is kept blind to the treatment group of each participant. Further, the participants are not informed about the particular content of their treatment group in order to maximize the likelihood that all participants will participate as actively as possible within their particular assigned intervention. Unblinding would only be permitted if a participant’s safety was determined to be at risk.
Data and Statistical Analysis Plan
Power Analysis
We performed a sample size analysis to determine the appropriate number of subjects to enroll in order to assure that we would achieve adequate statistical power. This was conducted based on data from the Primary Outcome Measure, the VMET (task failures and strategy use) indicating that For an alpha level of .05, and an anticipated effect of 0.5 (medium), and a power of .8, we need a total of 102 (50 in each rehabilitation group) for a one-tailed (ESA > EIA) directional hypothesis. Based on these analyses we plan to enroll 100 participants. These individuals will be randomized into the two intervention arms.
Data Management
Cognitive and affective neuropsychological assessment data are scored initially by study personnel that administered the assessments. The database manager double-checks these data and enters them into an electronic in-house database. The database automatically crosschecks entered data with its expected permissible values (e.g., numerical entries must be numerical), thus limiting data entry mistakes. Paper assessments are stored in locked filing cabinets, while digitized and electronically collected data are stored on a password protected server managed by university technicians. All identifying participant information is password protected on the university server and no identifying information is stored within the in-house database. Confidential participant information is never shared with individuals or entities outside of trained study personnel. Audits may be conducted at random by the sponsoring IRBs.
Clinical trial data will also be entered into the Federal Interagency Traumatic Brain Injury Research (FITBIR; https://fitbir.nih.gov/) Informatics System in compliance with U. S. Department of Defense funding guidelines. FITBIR is a data-sharing platform designed to facilitate collaborations among the TBI research community and improve research efforts across the country. Identifying participant information is not stored within FITBIR.
Statistics for Primary Outcome Measures
Virtual Multiple Errands Test (VMET): We will evaluate intervention efficacy by comparing the performance of the ESA group to that of the EIA group on performance-based measures of executive function ability related to daily life. This first test will evaluate near-transfer of intervention skills to the VMET, a test of executive functions evaluated in a computerized format. VMET requires working memory, planning, coordinating, and prioritizing, thus is serves as a global, performance-based measure of executive function. We will evaluate change on total number of errors from the pre-intervention baseline using a 2 (group) x 2 (testing session) repeated-measures analysis of variance (ANOVA) followed by Bonferroni-correct post-hoc tests to evaluate the significance of simple mean comparisons. We predict a significant time-by-group interaction in which the ESA group will perform more accurately relative to the EIA group at the post-intervention test, but not at the pre-intervention test. In the event that this measure proves insensitive, we will compute and analyze sub-scores based on strategy errors, completion errors, and rule-violation errors applying an additional Bonferroni multiple comparison correction for the use of three sub-measures.
Secondary Outcome Measures
In addition to evaluation of near and far transfer of executive function training skills, we will evaluate a series of subcomponents that predicted to be improved by the intervention.
Performance on Neuropsychological Tests
The neuropsychological tests will be evaluated according to the procedures outlined for the use of the ANAM and the relevant scoring procedures for each additional test in the planned battery. These scores will be organized by primary domain and analyzed using 2 (group) x 2 (testing session) ANOVAs. The domains of interest for the neuropsychological battery are as follows: working memory, goal maintenance, verbal fluency and general cognition. These same groupings will be applied to the ANAM data as well with the exception of verbal fluency, as this domain is not examined in our ANAM battery.
Multiple Correspondence Analysis (MCA)
MCA is a multivariate statistical method that can analyze a large amount of data from multiple variables and reveal the pattern of relationships and the structure of the data collected from all participants. Projection plot points on each two graphs represent the observations (e.g., participants) and the variables (e.g., assessments). Projections are mapped onto principal components that account for variance in the data. Projections close together represent similarities according to those components and distance represents differences. Participants will be grouped in this analysis according to the type of head injury they sustained to determine any patterns in performance deficits (and improvements) for the measures of memory, attention, task switching, and other assessments of executive functioning. We will also include MRI-based data and demographic variables in this MCA, which will enable us to better understand the population and the degree of relatedness among brain, cognitive, life function, and demographic data elements.
Analysis of Game Performance Data Acquired During Intervention
Data from the ESA and EIA weekly modules will be analyzed by total number of task errors and by completion times for each task in order to determine the degree of individual improvement on tasks from Day 1 to Day 5 during each of the four weeks. For the long-term memory module we will also analyze total recall scores across the week. We will apply this same analysis to the overall recall scores obtained in Week 4 during the composite game-based simulation. Data will be regularly analyzed by the research team. We do not anticipate needing a data monitoring board, as internal procedures should minimize data fidelity loss. Data will be uploaded to the FITBIR website at regular annual intervals leading us to attend to the data regularly within the research team. Adverse events will be immediately reported to the IRB agencies sponsoring this trial.
Additional Neuroimaging Methods
fMRI Data Acquisition and Analysis: Functional images will be acquired on a 3T Siemens MAGETOM Prisma with an echo-planar image sequence sensitive to BOLD-contrast (TE 30ms, TR 2 s, α flip angle 70°). The volume covered the whole brain with a 64x64 matrix and 37 transverse slices (4 mm thickness with a 0 mm inter-slice gap); (voxel size 3x3x4 mm). The structural scans include a magnetization-prepared, rapid access gradient-echo image sequence with 211 sagittal slices and a fluid-attenuated inversion recovery image sequence with 179 slices.
Resting-state functional connectivity: Images are preprocessed using AFNI. Standard preprocessing methods include de-spiking, slice time correction, motion correction, spatial normalization (MNI template), temporal normalization, linear regression and band-pass filtering (0.009 < f < 0.08 Hz). In the linear regression, (1) rigid head motion profile, (2) the signals averaged over the lateral ventricles, the deep cerebral matter and the whole brain and (4) the first temporal derivatives of aforementioned parameters are regressed out. After the band-pass filtering, motion ‘scrubbing’ is performed with a frame-to-frame head movement rate of 0.5 mm and a standardized DVARSa of 1.8 to prevent potential motion artifacts. The motion 'scrubbed' images in the gray matter are then spatially smoothed at 6 mm full width at half maximum.
Seed-based Connectivity Analysis: The spatial correlation maps of fMRI images between a 10 mm diameter seed region and the rest of the brain are obtained then converted by the Fisher's Z-transform to enhance the normality. To compare groups on the correlation maps, random effects analyses are performed on group maps.
Brain Network Analysis: The brain network analysis is performed using the brain connectivity toolbox (Rubinov, 2010). Nodes are defined using 264 functional areas, and then edges are defined as the correlation coefficients of time courses between node pairs. Correlation coefficients are thresholded by strongest proportions of correlations (10 to 2%). Small-worldness, global and local efficiency, centrality and modularity are evaluated.
DTI: Image Acquisition: 1-shot, spin-echo, echo planar imaging (EPI) sequence, field of view (FOV)=224 mm, slice thickness/gap = 2/0 mm, approximately 75 slices, repetition time = 7500 milliseconds, echo time = 65.7 milliseconds, flip angle = 90°, NEX = 1, posterior-to-anterior phase encoding direction, and a matrix of 112 ×112. Voxel size is 2 ×2× 2 mm3. Images are skull-stripped, eddy current corrected, and susceptibility artifacts corrected. Fractional anisotropy (FA), mean diffusivity (MD), and DTI color maps are calculated. FA and MD maps are normalized to Montreal Neurological Institute (MNI) space using a 12 parameter linear registration and subsequent nonlinear registration.
DTI Voxel Based Analysis: Normalized FA maps for normal controls are averaged and smoothed with a 4mm full-width half maximum (FWHM) kernel to create a spatially normalized FA template. Subsequently, all FA maps (both intervention groups) are normalized and smoothed with an 8mm kernel. Voxel-wise statistics are performed on these images using FSL. A two-tailed two-sample t-test is performed between groups, restricting the analysis to WM voxels by using a normalized mask of WM derived from FSL’s MNI Avg152, T1 2x2x2. WM lesion loads (below normal FA) are calculated using t-tests between a previously collected control FA map of each individual’s FA maps. The result of each t-test is a map of voxels with below-normal FA, and these identified voxels will be summed for each participant.
The study sponsoring IRBs may audit the conduct of this study at any time, though specific scheduled audits are not planned.