The investigation of four technologies to assist in detecting mild to moderate traumatic brain injury of U.S. Military service members

1Introduction

Traumatic Brain Injury (TBI) is the result of high energy force to the face or skull occurring in a number of ways [1] and categorized according to severity from mild to moderate to severe [2]. Active duty and reserve service members are at increased risk for sustaining a mild to moderate TBI compared with their civilian peers [3] in part due to their exposure to blasts and other combat conditions during recent conflicts in Iraq and Afghanistan [4]. Mild traumatic brain injury (mTBI), more commonly known as concussion, can occur from a non-penetrating blast exposure [5]. From the year 2000 to 2006, the U.S. Army experienced a 107% increase in mTBI incidents, while the civilian sector showed a 46% decrease for this same time period [6]. TBI is therefore of considerable interest to the military services.

The symptoms of mTBI are far more subtle and difficult to diagnose than moderate to severe TBI. Rapid advances in medical technology are transforming the way military service members are being assessed and treated in our nation’s armed forces. Even so, at the time of this study there was no single standard, reliable, objective procedure for evaluating mild traumatic brain injury (mTBI) at Level I trauma centers in the United States [7–8]. Assessment and identification of mTBI is even more difficult in a military theater-of-operation, with limited medical facilities and multiple casualties [9]. However, there is an assemblage of evidence to suggest several emerging technologies may serve as valuable assessment tools to assist in screening individuals who may be experiencing a mTBI.

With blasts contributing to a high number of head injuries in theater, the need for rapid assessment is vital to identify injuries early, so appropriate treatment can commence. Following blast exposure, the initial screening measures in theater for TBI have included assessments using the Military Acute Concussion Evaluation (MACE) [10], the Glasgow Coma Scale (GCS), a neurological assessment (with eye, motor, speech, and balance components), and evaluation of pupillary response, as well as blood pressure, oxygenation, and interview/observation. If no loss of consciousness or alteration of consciousness occurs during administration of the MACE [9], the health care practitioner (typically a medic) may stop and consider other possible causes (other than concussion) for current symptoms. A Concussion Management Algorithm is used to determine the next steps in terms of the level of care needed, such as rest or evacuation to a Combat Support Hospital [9]. These clinical diagnostic tools serve as the ‘best we can do under the circumstances’, but are open to the interpretation of the individuals administering them. In addition, individuals with a mTBI often do not report their symptoms initially, but may present with symptoms days, weeks, or even months after the incident. Finally, over a third of Iraq veterans thought to have TBI related concussion syndrome had comorbidities of Post-Traumatic Stress Disorder (PTSD) or depression, and the overlapping symptoms of mTBI, PTSD, and depression further confound diagnostics [11]. As a result, a mTBI can exist undiagnosed, putting service members at risk for a second head injury and endangering the military service member and their team.

Often, those who experience a first concussion experience a second concussion. Within the sports medicine realm, this condition is referred to as second impact syndrome and although professionals debate this issue [12], some contend it may produce increased susceptibility for behavioral symptoms, greater diffuse axonal injuries (DAI) [13–18] prolonged neuropsychiatric impairment [19], and later-life cognitive impairment [19-20], including chronic traumatic encephalopathy [21]. In sports-related research, sustaining a concussion has been shown to increase the likelihood of sustaining a second concussion [22], possibly due to increased susceptibility.

There are technologies that show potential for identifying mTBI, four of which were targeted for this study. The first two are relatively well known: balance measured with a AMTI OR6-7 force plate and neurocognition and mood using the Automated Neuropsychological Assessment Metric4 (ANAM4) - configured for assessing TBI. The second two selected for study are not as well-known and have less research backing, blood flow comparisons using a Brain Acoustic Monitor (BAM), and voice using Voice Analysis software (VA).

Balance is “the process of maintaining equilibrium and center of gravity within the body’s base of support” [23]. Having a problem with balance is the second most frequently self-reported symptom in TBI patients [24]. Multiple investigators have shown postural instability in patients with TBI [25–26]. In fact, Allison has stated that nearly everyone with a TBI has some type of balance problem [27]. While the research clearly indicates balance as an issue among individuals with a TBI, it is not known how long these symptoms persist after the point of injury and which (if any) measures of balance can be used to assist with identifying whether a person has incurred a TBI. That is, most of the research on balance and TBI has focused on rehabilitation techniques to improve balance and recovery [28], rather than using balance to predict who may have a mTBI.

A constellation of long-lasting symptoms of a TBI is classified as “post-concussion syndrome” (PCS), which includes difficulties with attention and concentration, impaired memory, fatigue, headaches, anxiety, depression, irritability, or sleep disturbances [29]. Hill and colleagues [30] found that over 50% of the individuals with mTBI included in their study, reported active symptoms of sleep disturbance, irritability, memory problems, and headaches. The Automated Neuropsychological Assessment Metric (ANAM) is a computer-based tool designed “to detect speed and accuracy of attention, memory, and thinking ability” that also includes a mood scale [31]. It includes a number of assessments, which can be activated or deactivated for use with particular populations, illnesses, or injuries. The ANAM4 was specifically configured as a neurocognitive assessment focused on TBI, and it incorporates evaluations of all the symptoms of PCS. The ANAM (including ANAM4) has been administered pre and post soldiers’ deployment into harms’ way, and to assess patient function among service members with head injuries [32]. The ANAM has also been used, along with other neurocognitive assessments, to set baselines and identify potential head injuries among athletes, using selected tasks to construct the Automated Sports Medicine Battery [33–34]. Its portability and ease-of-use may contribute to its use within war zone theaters of operation [35]. While some studies have found decreased performance on neurocognitive tasks among military with a mild to moderate TBI, others have not [5]. Roebuck-Spencer and colleagues [36] showed the highest percentage of post-deployment service members with neurocognitive decline were those who reported active symptoms, as well as having sustained a TBI during deployment.

Correctly identifying a head injury early is important to initiating effective therapy. Active Signal Technologies has developed a sensor system to detect and amplify sound waves within the brain, known as the Brain Acoustic Monitor (BAM) [37–38]. Although not audible by standard measures, the arterial pulse wave produces a predictable deformation of the skull as it enters the brain and can be acoustically measured. The underlying theory for the BAM is that arterial pulse waves produce acoustically detectable, and predictable, changes that can be measured on the forehead. These sounds are altered when there is damage to the cerebrovascular system and differ from those heard at a reference point, such as the radial artery or a finger pulse [37]. In a person without a TBI, the cerebrovascular acoustic wave forms are more similar to those at the reference point than in a person experiencing a TBI [38]. Dutton and colleagues examined the effectiveness of using the BAM to identify mTBI in patients with normal and abnormal CT scans, and found the BAM sensitivity to be 100% with a specificity of 30.14% [38], while a second study by Dutton and colleagues found a sensitivity of 93%, and a specificity of 14%. Both studies examined individuals early after hospital admission, consequent to their demonstrating anatomic and functional evidence of TBI. From the first study, they concluded the BAM may serve as a useful, additional diagnostic tool, adding important information to self-reports of injury, loss of consciousness, and health care practitioner interviews [38]. From the second study, they concluded early BAM screenings may guide prehospital and emergency room diagnostic and therapeutic decision making, however they suggested further refinements are necessary to improve specificity [39].

Differences in voice and voice analysis results have been demonstrated in severe TBI in terms of irregularity, noise-to-harmonic ratio, and hypernasality [40], sometimes resulting from motor disorders [41]. Speech is currently being explored as a diagnostic tool for mTBI in athletes [42]. Past research investigating voice analysis as a diagnostic tool have included Parkinson’s disease [43], Huntington’s disease [44], and examination of various health states [45]. While these investigations have resulted in limited success, new technologies and methodologies continue to yield possibilities and the concept of wearable biomarker devices to assist with mTBI identification is appealing. Sound Health’s Bioacoustic Biology uses voice, and voice analysis, as representations of an individuals’ health and well-being, developed after examining the voices of people with similar illnesses or injuries for their similarities in sound frequencies [46]. The theory behind Sound Health’s technology is that people with similar health issues produce similar vocal biomarkers. Sound Health’s voice analysis (VA) system was used in this study.

As medical evaluation procedures and diagnostic technologies evolve, there is an opportunity to identify injuries expediently and accurately with objective and quantitative measures to avoid susceptibility to further injury and quickly implement appropriate treatment. Moreover, there is a need for rapid, evidence-based, objective diagnostics that can be employed in a combat setting to assist military health care providers in screening, identifying, and tracking the symptoms of mTBI in military personnel. At this time, none of the mentioned technologies has a sufficiently strong research backing supporting their use at, or shortly after, the point of injury in a field setting.

The purpose of this research was 1) to examine the usefulness of four technologies in differentiating individuals with and without mild or moderate, service-related TBI, and 2) if found to be useful to then build a clinical prediction rule (CPR) based on the results. The methods and technologies examined included: 1) balance using a force plate and analysis software (AMTI OR6-7-2000) developed by Advanced Mechanical Technology, Inc.), 2) neurocognitive and mood testing using the ANAM4, 3) a comparison of the acoustic wave forms from cerebrovascular blood flow compared with a reference point, using BAM developed by Active Signal Technologies, and 4) voice recordings using voice analysis software developed by the Institute for BioAcoustic Biology and Sound Health.

2Methods

2.3Technologies

2.3.1Force platform

This study examined the predictive ability of various measures of balance in the identification of mTBI. Each volunteer participated in a series of standing balance tests, i.e., measurements of postural sway first with their eyes open and then with vision occluded by wearing a mask over their eyes. The balance tests were performed on an AMTI OR6-7 force platform using integrated AMTI NetForce software. All stances were performed with the participant wearing socks. We evaluated eight separate positions with eyes open and eyes closed and compared measurements of sway displacement, velocity, and variability between groups across positions to identify a subset of variables that would appropriately discriminate between individuals with and without mTBI. The positions we evaluated can be generally categorized across three descending levels of base of support (BOS): bilateral stance, tandem stance, and single limb stance, and two levels of vision: vision and no vision. The specific balance positions selected were those used frequently in evaluations of balance. They are shown in Table 1 and were performed in the order listed. Use of the force plate yielded 418 pre/post outcome variables. These measures took approximately 45 minutes to complete.

Table 1

Balance Positions

Stance	Description	Image
Bilateral Stance	Feet side by side, apart; arms; eyes open and closed
Tandem Stance, right foot in front of left	Right foot in front of left; arms at sides; eyes open and closed
Tandem Stance, left foot in front of right	Left foot in front of right; arms at sides; eyes open and closed
Tandem Stance, dominant foot forward	Dominant foot in front on non-dominant foot, arms at sides; eyes open
Romberg Test	Bilateral stance with ankles together, arms crossed, fingers touching shoulder; eyes open and closed
Military Academy Stance Test, (MAST) right leg	One-legged stance, balancing on right leg, left leg lifted and bent in a comfortable position, arms at sides; eyes open and closed
Military Academy Stance Test, (MAST) left leg	One-legged stance, balancing on left leg, right leg lifted and bent in a comfortable position, arms at sides; eyes open and closed

2.3.2Automated Neuropsychological Assessment Metric (ANAM)

Individual cognitive assessments were conducted using the ANAM4 TBI Battery on a laptop computer. This automated battery of tests was developed specifically for use in screening for concussion/TBI [48]. The measures administered as part of the TBI Battery can be seen in Table 2.

Table 2

Automated Neuropsychological Assessment Metric components administered

Assessment	Description
Demographics	Demographic information such as age and gender
Sleepiness Scale	Measure of user’s current state of energy-fatigue level
Mood Scale	Anxiety (anxiety level), and restlessness (motor agitation)
Simple Reaction Time	Index of visuo-motor response timing
Code Substitution- Learning	Index of visual search, sustained attention, and encoding
Procedural Reaction Time	Reaction time and processing efficiency associated with simple mapping rules
Mathematical Processing	Basic computational skills, concentration, and working memory
Matching to Sample	Spatial processing and visuo-spatial working memory
Code substitution – delayed	Memory (delayed)
Simple Reaction time	Index of visuo-motor response timing

The ANAM was administered twice during the first visit to help account for learning effects and to provide a more reliable set of test scores. The first administration served as a practice session to familiarize the volunteer with the evaluations, whereas data from the second administration (given at the conclusion of the visit) were analyzed.

The ANAM Mood Scale 2 – Revised (AMS-2) is included within the ANAM software. The AMS-2 is comprised of 42 mood-related adjectives, rated on a 7-point Likert scale of mood intensity, ranging from 0 to 6. The lowest verbal anchor is 0 “not at all”, midpoint is 3 “somewhat”, with a maximum of 6 “very much”. The mood adjectives are grouped under seven mood subscales: anxiety (tension/anxiety level), depression (dysphoria), anger (negative disposition), vigor (high energy level), fatigue (low energy level), happiness (positive disposition), and restlessness (motor agitation) [49]. The full ANAM4 battery took approximately 30–45 minutes per administration and yielded 55 pre/post outcome variables.

2.3.3Brain Acoustic Monitor (BAM)

The BAM protocol used for this study closely followed the University of Maryland R. Adams Cowley Shock Trauma Center (STC) protocol using a laptop-based system. The system included a Panasonic Toughbook¹ with the conditioning electronics located on the base of the laptop, two forehead sensors secured using an elastic headband, and a finger reference sensor (Figs 1–3).

Fig. 1–3

BAM equipment (Left to Right): Panasonic Toughbook, forehead sensors with headband, and finger reference sensor.

The BAM sensor disks were applied to the front of the participant’s forehead close to the volunteer’s hairline and centered over the pupils of each eye, under an elastic cloth band. A modified pulse oximeter finger clip was applied to the middle or ring finger, which served as the reference sensor. Acoustic data from the two forehead sensors and from the reference point were then recorded for 10 seconds. This process was repeated a total of five times per visit or until an acceptable reading was obtained, which was indicated on a BAM display. An unacceptable display most often reflects incorrect placement of the sensors. The waveforms are measured in time and frequency domains. In a healthy individual the forehead signals resemble the arterial reference location, while they do not match in someone with a head injury. The BAM system has a signal processing algorithm, with allowable boundaries of divergence. The individual is considered ‘healthy’ or ‘normal’ if the readings fall within the boundaries (BAM Negative). They individual is considered suspicious for pathology if the readings fall outside the boundaries (BAM Positive). The system has a red/green display indicating the sensor output results. This assessment produces 7 outcome variables. Testing with the BAM required approximately 10 to 20 minutes [50].

2.3.4Voice Analysis (VA)

Voice data were gathered using a Sound Recorder [51], and analyzed using proprietary software developed by the Institute for BioAcoustic Biology and Sound Health [46]. A uni-directional condenser microphone (SAMSON C01U USB) was connected to a laptop computer. The laptop computer was configured with BioAcoustic Biology and Sound Health vocal analysis software and a calibrated sound card [46]. The sound recorder settings included the sound playback default device set at Sigma Tel Audio, the sound recording default device set at SAMSON C01U, the volume set at maximum, and the MIDI Playback set at Microsoft GS Wavetable SW Synth.

Three recordings were made, with each recording being 30 seconds long. The three recordings included one recording of the volunteer reading a pre-selected paragraph, one talking about personal physical health, and one talking about a mundane subject, such as what they had for breakfast or lunch.

The recorded vocal sounds were converted to numeric data using properties of frequency (hertz/cycles per second) and amplitude (decibel) by means of a Fast Fourier Transform. This translates the vocal data to numeric data creating a digital graph that represents the vocal frequencies using a range of 0–1000 cycles per second, depicting a grid of 0 to 60 on the Y axis and 0 to 1000 along the X axis. Each volunteer’s voice analysis included 24 base frequencies identified by the software as being “in stress” [52]. These 24 frequencies, along with their 24 ‘reciprocal’ values were then compared with a series of frequencies identified by the Institute for BioAcoustic Biology and Sound Health using their proprietary Army Compare database, as being associated with TBI injuries. While not true mathematical reciprocals, the ‘reciprocal’ values are mathematically related to the base frequencies via proprietary software calculations [52]. For the purpose of this initial examination, the number of each individual’s 48 values (frequencies + reciprocals) that also appeared in the Army Compare database (i.e. were thought to be associated with TBI) were used for the data analysis. Twelve outcome measures were produced. Data collection for voice analysis took approximately five to ten minutes.

3Results

Among the 98 volunteers enrolled in the study, 88 provided reasonably complete data for variables derived from force plate, ANAM, BAM, and VA data collection instruments. Of those completing the study, 35 (39.8%) had a diagnosis of mmTBI. Participants with mmTBI (30.7±7.5 years) were significantly younger (p < .001) than those with no TBI (39.1±10.2 years). Likewise, participants with mmTBI had significantly fewer (p < .001) years of military service (9.1±7.1 years) than those with no TBI (16.1±8.4 years). Gender proportions were also significantly different between groups (p = .002): in the mmTBI group only 1 (2.9%) of 35 subjects was female; in the no TBI group 15 (28.3%) of 53 participants were female. Demographic variables are summarized for each group in Table 3. Among the mmTBI subjects in our study (n = 26), the mean time since injury at time of data was 207±248 days (range 15 to 1088 days).

From a total of 492 available variables (418 force plate variables, 55 ANAM variables, 7 BAM variables, and 12 voice variables), 50 potential predictors had significantly different mean values at a relaxed alpha level (.10) comparing the mmTBI group to the no TBI group, and controlling for age, gender, and years of military service. Among those variables showing univariate groupwise discrimination were 40 force plate variables (Table 4), 8 ANAM variables, 1 BAM variable, and 1 voice variable (Table 5). Of these 50 potential predictor variables, four variables were eliminated from further consideration because of missing data due to human and/or recording failures. Another seven variables were eliminated because of between-group mean differences showing paradoxically better balance in the mmTBI group.

This left 39 variables which were entered into the multivariate stepwise logistic regression analysis based on 83 subjects (32 with mmTBI) with complete data for all 39 entered predictors. The stepwise procedure retained 3 predictors: 1) ANAM4 mean of ratings for the Vigor adjectives from the Mood Scale; 2) absolute value of average center of pressure x-axis (cm) with left foot in front of right, eyes open; 3) average displacement along y-axis (cm) on right leg, eyes open. This model correctly classified 64 (77.1%) of 83 subjects (Table 6), and was statistically significant (p < .001), with the 3 predictors explaining 37.5% of the variance in TBI status. Sensitivity (precision or True Positive Rate) is the ability to correctly identify an individual with mmTBI. Specificity (True Negative Rate) is the ability to correctly identify those who do not have a mmTBI, based on their negative scores.

The Hosmer-Lemeshow statistic established reasonable goodness of fit for the model (p = .684). Adjusted odds ratios were.44 (95% CI.28 to.67) for the mean of ratings for the Vigor adjectives in the ANAM4 Mood Scale, 2.38 (95% CI 1.16 to 4.91) for absolute value of average center of pressure x-axis (cm) with left foot in front of right eyes open, and 10.10 (95% CI 1.62 to 62.95) for average displacement along y-axis (cm) with right leg, eyes open. Areas under the ROC curves for the three retained predictors were as follows: 0.74 (95% CI 0.64 to 0.84) for the ANAM4 mean of ratings for the Vigor adjectives from the Mood Scale,.63 (95% CI 0.51 to 0.75) for absolute value of average center of pressure x-axis (cm) with left foot in front of right eyes open, and.57 (95% CI 0.44 to 0.70) for average displacement along y-axis (cm) with right leg, eyes open. Cut scores determined with the Youden Index method, defining positive test results for each of the retained predictors, were <2.42 points, >1.44 cm, and >1.00 cm, respectively.

Diagnostic accuracy profiles for each level of the 3-item CPR, using all predictors in combination, based on n = 86 with complete data for all 3 predictors, are presented in Table 6. Table 7 shows the three tests for mmTBI, including the score on each measure which denotes when it is positive for predicting ammTBI.

4Discussion

The purpose of this research was to examine the usefulness of four technologies in differentiating military service members with and without mild or moderate, service-connected TBI, and to build a clinical prediction model based on the results. Our statistical analysis of four technologies, the AMTI force plate, ANAM4, the BAM, and VA resulted in two of the four devices being helpful in predicting TBI: the AMTI Force plate for two measures and the ANAM4 mood subscale for vigor. A clinical prediction model was developed.

5Conclusions and recommendations

This study identified a regression equation predicting those who did, and did not, have a mmTBI, consisting of three measures one from a mood scale and two from use of a force platform. From this equation, a clinical prediction rule was developed to identify patients with a mild to moderate traumatic brain injury. The results indicate that the presence of at least two of the three clinical measures: the ANAM4 vigor subscale score ≤2.42, frontal plane sway ≥1.44 cm during tandem stance with eyes open and the left foot forward, and sagittal plane sway ≥1 cm while standing on the right foot with eyes open increases the likelihood of the patient having a mmTBI from 15% to 76%. The absence of all three of these predictors decreases the patient’s likelihood of having a mmTBI to approximately 2%. That is, a subject with positive results on any two or on all 3 tests would shift them from a pre-test probability of 15% to a post-test probability of 75.7% based on a PLR of 17.67. A subject with negative results on all 3 tests in the CPR would shift them from a pre-test probability of having mmTBI from 15% to a post-test probability of 2.3% based on a NLR of 0.13. These results demonstrate the increase in the ability to correctly identify those with a mmTBI and rule out those who do not have a mmTBI. Likelihood ratios of these magnitudes are characterized as sufficient to “ ...   generate large and often conclusive changes in pretest to posttest probability” for a positive test result, and “ ...   moderate shifts in pretest to posttest probability” for a negative test result [85]. That is, the model performs better at ruling in mild to moderate TBI than ruling out the condition. Figure 4 provides a quick visual reference showing the changes in pre-test probability to post-test probability, summarizing our findings.

Fig.4

Likelihood ratio nomograms^* showing post-test probabilities‡for: (A) a positive result with any 2 + tests positive on the 3-item CPR, and for (B) a negative result with 0 tests positive.^* [55, 84]. Copyright ©1975, Massachusetts Medical Society. All rights reserved. ‡Assumes 15% pre-test probability for mild to moderate TBI.

Recommendations include future studies of military service members exposed to blast, much closer to the time of initial injury. Such studies may include balance, mood scales, and the BAM, as well as other, perhaps newer technologies and tools, as they become available. The balance platform is an example an assessment that is very sensitive to balance, perhaps indicating the need for development of more sensitive measures of underlying sensorimotor and executive functioning deficits associated with TBI, such as oculomotor control [87], working memory [88], or other deficits. In addition, research suggestions include examining balance and vigor measures for their prediction consistency with a larger population, both closer to the time of injury and over the course of recovery. While voice analysis shows promise in other studies, it appears more research is required prior to examining its’ use with mild to moderate TBI.

6Limitations

This study was limited by the small number of subjects who completed the study. It is possible that a greater number of variables would have been included in the model, given a higher number of participants. Additionally, the study was limited by the inclusion of only subjects with mmTBI being seen in a tertiary care facility, and who were enrolled an average of 207 days (6.8 months) following injury. Therefore, the accuracy of this prediction rule to identify mmTBI in acutely injured individuals remains untested at this time. However, persistent deficits in balance following mTBI have been previously reported [26–8], and the results of our study, if subsequently validated, should help to inform decisions about prognosis and treatment considerations in individuals exposed to blast injury where diagnosis may have been delayed or not established due to comorbid conditions. McGinn and colleagues have proposed methodological guidelines for clinical prediction rules, suggesting that prospective validation and impact analyses should follow initial derivation studies before a clinical prediction rule is considered ready for widespread clinical use [89]. Therefore, future research with an independent sample of subjects should be conducted to ascertain the prediction rule’s validity.

Conflict of inetrest

None to report.