A Novel Tool for Evaluation of Mild Traumatic Brain Injury Patients in the Emergency Department: Does Robotic Assessment of Neuromotor Performance Following Injury Predict the Presence of Postconcussion Symptoms at Follow-up?

Abstract

Objectives

Postconcussion symptoms (PCS) are a common complication of mild traumatic brain injury (TBI). Currently, there is no validated clinically available method to reliably predict at the time of injury who will subsequently develop PCS. The purpose of this study was to determine if PCS following mild TBI can be predicted during the initial presentation to an emergency department (ED) using a novel robotic-assisted assessment of neurologic function.

Methods

All patients presenting to an urban ED with a chief complaint of head injury within the preceding 24 hours were screened for inclusion from March 2013 to April 2014. The enrollment criteria were as follows: 1) age of 18 years or greater, 2) ability and willingness to provide written informed consent, 3) blunt head trauma and clinical diagnosis of isolated mild TBI by the treating physician, and 4) blood alcohol level of <100 mg/dL. Eligible mild TBI patients were enrolled and their neuromotor function was assessed in the ED using a battery of five tests that cover a range of proprioceptive, visuomotor, visuospatial, and executive function performance metrics. At 3 weeks postinjury, participants were contacted via telephone to complete the Rivermead Post-Concussion Symptoms Questionnaire to assess the presence of significant PCS.

Results

A total of 66 mild TBI patients were enrolled in the study with 42 of them completing both the ED assessment and the follow-up; 40 patients were included in the analyses. The area under the receiver operating characteristic curve (AUC) for the entire test battery was 0.72 (95% confidence interval [CI] = 0.54 to 0.90). The AUC for tests that primarily measure visuomotor and proprioceptive performance were 0.80 (95% CI = 0.65 to 0.95) and 0.71 (95% CI = 0.53 to 0.89), respectively.

Conclusions

The robotic-assisted test battery has the ability to discriminate between subjects who developed PCS and those who did not. Additionally, poor visuomotor and proprioceptive performance were most strongly associated with subsequent PCS.

Traumatic brain injury (TBI) is responsible for nearly 30% of all injury-related deaths in the United States.¹ Affecting 1.7 million U.S. people annually, TBI is a primary cause of death and disability.² Approximately 75% of all TBIs are considered mild, defined as an acute injury to the head as a result of blunt force trauma or rapid acceleration/deceleration causing loss of consciousness (<30 minutes), posttraumatic amnesia (<24 hours), or other transient neurologic abnormalities with a Glasgow Coma Scale score of 13–15.^3,4 While the annual incidence of mild TBI is estimated to be 100 to 300 per 100,000 U.S. people, the true incidence rate is perhaps much higher because frequently mild TBI patients may decide not to seek medical care.^5,6

Serving over 1.2 million TBI patients every year,¹ the emergency department (ED) is a common access point for medical care for mild TBI patients, yet nearly 38% of isolated mild TBI patients are discharged to home from the ED without specific recommendations for follow-up.⁷ Emergency physicians and other clinicians are often challenged with the management and prognostication of mild TBI because these patients more commonly appear normal on the acute clinical examination. Nonetheless, the sequelae following injury can be profound and may severely limit functionality and participation.^7–9 Indeed, up to 80% of mild TBI patients will suffer persistent neuropsychological symptoms.¹⁰ Identification of patients at high risk for protracted postconcussion symptoms (PCS) at the time of their initial presentation to the ED is important for targeting limited clinical resources such as specialty follow-up and rehabilitation, also allowing for an enrichment strategy for improving efficiency in clinical trials.^11,12

The majority of mild TBI patients will have no acute intracranial findings on head computed tomography (HCT).¹³ Additionally, HCT poorly discriminates between those that will suffer long-term sequelae following mild TBI and those that will not.^13,14 Other neuroimaging techniques such as diffusion tensor imaging and magnetic resonance imaging are expensive and time-prohibitive and are generally unavailable in most EDs.¹⁵ Currently, clinicians have no reliable and validated mechanism for providing accurate and objective prognosis for patients with mild TBI.

The Kinesiological Instrument for Normal and Altered Reaching Movements (KINARM) End-Point system (BKIN Technologies, Kingston, Ontario, Canada) is a robotic device that is capable of objectively assessing multiple neurologic domains, including sensory, motor, and cognitive functions. It has been suggested that the device may detect subtle deficits that are not reliably observed in standard neurologic examination.¹⁶ The KINARM consists of two graspable robotic arms (see Figure 1) and a two-dimensional virtual reality display that, together, can create sophisticated visual and mechanical environments. The KINARM device has the ability to combine visual scanning, sensory stimulation, and forced feedback and has been used primarily to measure and quantify deficits in patients undergoing stroke rehabilitation.¹⁷ We refer to the capabilities provided in part by the programmable, mechanized handles as “robotic.”

Figure 1.

Simplified top view of the graspable robotic handles of the KINARM endpoint device (Image Courtesy: BKIN Technologies Inc.). KINARM = Kinesiological Instrument for Normal and Altered Reaching Movements.

The goal of this study was to determine if robotic-assisted neuromotor assessment at the time of initial ED presentation might be predictive of more severe PCS at 3-week follow-up. We hypothesized that subtle sensorimotor and cognitive impairments, detected at the time of initial ED presentation and within 24 hours of injury, could predict subsequent PCS in a mild TBI population.

Methods

Study Setting and Population

This was a prospective observational study of mild TBI patients presenting to the ED. Clinical study assistants were available 24 hours per day to perform active, realtime patient screening for study eligibility.¹⁸ After written informed consent was obtained, subjects were enrolled and tested (within 24 hours of injury) in the ED using a battery of robotic-assisted tests. Subjects were contacted via telephone 3 weeks after enrollment to complete follow-up questionnaires. The study protocol was reviewed and approved by the local institutional review board prior to start of the study.

Patients presenting to the ED within 24 hours of head injury were screened for study inclusion. The enrollment criteria were as follows: 1) age of 18 years or greater, 2) ability and willingness to provide written informed consent, 3) blunt head trauma and clinical diagnosis of isolated mild TBI by the treating physician, and 4) blood alcohol level of <100 mg/dL. The exclusion criteria were as follows: 1) presence of focal neurologic deficit on standard neurologic examination, 2) presence of comorbidities such as significant acquired or baseline visual disturbance or fractures to the upper extremities that would create logistic limitations to testing, and 3) vulnerable population (prisoners or in police custody and pregnant women). Screening and enrollment took place between March 2013 and April 2014. The last follow-up was completed in May 2014.

Robotic-assisted Testing

Enrolled patients were tested using a battery of five robotic-assisted tests (KINARM Standard Tests) during their stay in the ED. After the subject was seated appropriately in a height-adjustable chair in front of the device, the position of the chair was locked in place. Each of the following tests was then administered with verbal instructions provided at the start of each test. The total testing time was approximately 40 minutes.

Arm-matching¹⁹

In this test, the robotic handle moved one arm (passive hand) and the subject was required to mirror match the spatial position using their other arm (active hand). The device then repeated this process by randomly taking the arm through a nine-target grid, as shown in Figure 2, and iterated the entire grid six times for a total of 54 trials. The purpose of this test is to evaluate the position sense component of proprioception. An opaque barrier blocked the subject's arms and hands from his or her visual field and the device provided no visual stimulus. The subject therefore had only proprioceptive information about the arm position. Proprioceptive perception processing has been localized to the region from the anterior parietal cortex (APC) to the posterior insula and posterior parietal cortex (PPC) through the secondary somatosensory cortex (S2) and a second pathway for proprioceptive processing of action information from APC through S2 to the PPC.^20,21 Relevant parameters measured in this test included variability (consistency of matching using the active arm between trials), spatial shift (errors between active and passive arm positions), and spatial contraction/expansion (total area matched by the active arm relative to the passive arm).¹⁹ The test was first performed with the dominant hand as the passive arm (moved by the device) and then was repeated with the nondominant hand.

Figure 2.

Performance of two mild TBI patients on the arm-matching test. The robot moved the right (denoted by the green solid line) and the subject matched with the left arm (denoted by the blue solid line; this is overlaid as a dotted line to help visualize how accurate the subject was at judging the positions of his or her arm). The ellipses denote how consistent the subject was, in each trial; larger ellipses mean more variability. PCS = postconcussion symptoms; TBI = traumatic brain injury.

Visually Guided Reaching¹⁷

This test required the subject to make unassisted reaching movement to red circular targets (1-cm radius) on the screen. It evaluates complex motor and visual pathways involved in volitional hand movements,^22,23 allowing for independent differentiation of visuomotor deficits from those with proprioceptive disturbances.²⁴ Targets appeared on the screen in a fashion such that the subject started each trial by holding the virtual handle on a central target. One of eight peripheral targets arranged uniformly on the circumference of a 10-cm virtual circle was then randomly illuminated for the subject to reach to. The entire circular block, shown in Figure 3, was repeated eight times for a total of 64 trials. Movement parameters recorded in this test included reaction time (time between appearance of a target and onset of movement), postural hand speed (characterizes subject's ability to hold steady on a target), and several other attributes related to sensorimotor control.¹⁷ This test was administered twice, first with the dominant arm and second with the nondominant arm.

Figure 3.

Performance (hand paths) of two mild TBI patients on the visually-guided reaching test, where the subject reaches from the central target to one of the eight peripheral targets. PCS = postconcussion symptoms; TBI = traumatic brain injury.

Trail Making

This test included the standardized Trail Making Tests (TMT) A and B that requires subjects to connect letters and/or numbers sequentially.²⁵ It has been shown that TMT A and B are highly sensitive to brain impairments that affect visual search, executive function, and motor function.^26,27 The KINARM version of these tests allowed precise measurement of test completion times, dwell time (total time spent on targets before moving to the next), and errors.

Bimanual Object Hit²⁸

In this test, the subject was required to hit virtual circular objects (targets) falling from 10 random locations (bins) from the top of the screen, using two virtual paddles (one for each arm). The test involves participation from a larger range of brain regions, spanning across the occipital, parietal, and frontal lobes.²⁹ Subjects were instructed to hit as many objects as possible using either paddle. The speed and number of objects increased with time for a fixed testing time of just over 2 minutes and a total of 300 objects. The test characterized interlimb coordination and visuospatial attention by measuring parameters such as target hits (reported as percentage of objects missed), hand bias (quantifies relative hand use), and mean hand speed during the test.²⁸ While interlimb coordination has been traditionally localized to the supplementary motor area, recent evidence suggests a distributed brain network that depends on the complexity of the coordination task.³⁰

Bimanual Object Hit and Avoid

This test is a variant of the object-hit test where six differently shaped objects are randomly dropped from the top of the screen. It is designed to assess executive functioning, in addition to proprioceptive and sensorimotor performance. At the start of the test, subjects were presented with two shapes (targets) and instructed to hit only those two shapes and avoid the other shapes (nontargets or distractors), emphasizing rapid motor selection and attention. The functionality and parameters measured in this test were same as that of the object-hit test.

Follow-up and Outcome Assessment

The prevalence of PCS 3 weeks postinjury was assessed using the Rivermead Post-Concussion Symptoms Questionnaire (RPQ).³¹ The RPQ measures the extent of 16 PCS on a scale of 0 to 4 compared to preinjury levels (where 0 = the symptom is absent; 1 = same as preinjury; 2 = mild; 3 = moderate; and 4 = severe). In this study, clinically important PCS was considered present if three or more symptoms were rated as worse (score ≥ 2) than before sustaining a mild TBI.^32,33 Given the significant variability of PCS definition in the literature, we refer to subjects meeting this criteria as “Rivermead positive” and others as “Rivermead negative.”

Data Analysis

The robotic-assisted tests generated a total of 79 parameters for each subject (see Table 1 for details). Each test has a database of normative scores collected from healthy control subjects.³⁴ Subjects' scores were compared against the normative data corresponding to their age, sex, and handedness. A subject's performance for any given parameter is considered abnormal if the score for that parameter was outside the 5th to 95th percentiles of the normal range (see Table 2 for sample data). Each test was scored as the proportion of abnormal test parameters, so that higher proportion indicates worse performance. Receiver operating characteristic (ROC) curves were plotted for each test and the area under the ROC curve (AUC) was calculated to quantify the ability of the test to discriminate between subjects with PCS (Rivermead positive) and those without PCS (Rivermead negative). To determine if combining the results of high-performing tests enhanced the discrimination ability, a net reclassification analysis was performed. We conducted two sensitivity analyses to determine the impact of participants lost to follow-up, one assuming that all lost cases were Rivermead positive and another assuming that all lost cases were Rivermead negative. The chi-square test or Fisher's exact test was used to test for differences in proportions and independent samples t tests were used to test for differences in continuous variables, and 95% confidence intervals (CIs) were calculated.

Table 1. Overview of Parameters Measured During Robotic-assisted Testing.

Test Parameter (Unit)	Description
Arm Matching Test*
Assessment Domain: Proprioception (position sense)
Variability (m)	Mean value of the SDs of the subject's hand position in X and Y directions. The variability in the XY direction is calculated as the root mean square of the X and Y variables.
Contraction/expansion ratio	Ratio of range of movement of the arm moved by the subject compared that of the arm moved by the robot.
Shift (m)	Mean difference between the matched position of the arm moved by the subject and the position of the arm moved by the robot. Positive values denote lateral shifts and negative values denote medial shifts.
Absolute error	Mean absolute distance error across all trials.
Reaching Test*
Assessment domain: upper-limb postural control, motor response to visual stimuli, feed forward, and feedback control
Posture speed (m/s)	Median hand speed when the hand should be at rest.
Reaction time (s)	Time between when a target appears and movement onset.
Initial direction error (rad)	Angular deviation between 1) a straight line from the hand position at movement onset to the hand position after the initial phase of movement and 2) a straight line from the hand position at movement onset to the destination target.
Initial distance ratio	Ratio of 1) the distance the hand traveled during the subject's initial phase of movement to 2) the distance the hand traveled between movement onset and movement offset.
Speed maxima count	Number of maxima in hand speed between movement onset and movement offset.
Min–max speed difference (m/s)	Mean difference hand speed minima and maxima.
Movement time (s)	Total time elapsed from movement onset to movement offset.
Path length ratio	Ratio of the distance traveled by the hand between movement onset and movement offset and the straight line distance between those two hand positions.
Trail Making Test (A & B)*
Assessment domain: executive function
Total test time (s)	Total time from the start of the test to reaching the last target.
Dwell time (s)	Total time spent on dwelling/waiting on targets.
Time ratio	Ratio of time for targets 13–25 to time for targets 1–12.
Object-hit Test
Assessment domain: Motor, spatial, and temporal performance.
Total hits	Total number of object hits.
Hand bias hits	Quantifies which hand is used more often for hitting the objects (hand dominance).
Miss bias	Quantifies any bias of misses toward one side of the work space or the other.
Hand transition (m)	Shows where the subject's preference for using one hand over the other switches in the work space.
Hand selection overlap	Quantifies how effective subjects are at using both hands and how often they overlap hands.
Median error	Percentage of the way through the test when the subject made half of their errors.
Hand speed (m/s)	Mean hand speed of the subject through the entire test.
Hand speed bias	Value from −1 to 1 which describes the bias in hand speed between the hands.
Movement area (m2)	Area of space used by the subject with each hand during the test.
Movement area bias	Value from −1 to 1 which describes the bias in movement area between the hands.
Object-hit and Avoid Test
Assessment domain: attention, executive function (in addition to parameters and domains from the object-hit task).
Distractor hits	Number of distractor objects hit by the subject.

^*The test is administered twice, first with dominant arm and second with the nondominant arm. A total of 79 parameters were derived from the attributes listed in this table.

Table 2. Sample Data (Scores) of a 21-year old, Female, Right-handed, Mild TBI Patient on Trail Making and Object-hit Robotic-assisted Tests.

Test	Parameters	Score	Typical Range	Unit
Trail Making A (left hand)	Total time1	31.91*	12.7 to 30.8	s
	Time Ratio (2nd half/1st half time)1	0.92	0.54 to 1.39
	Dwell time1	11.11	4.6 to 16.9	s
	Errors	2	0.00 to 2.00	#
Trail Making B (right hand)	Total time1	57.24*	17 to 46	s
	Time Ratio (2nd half/1st half time)1	0.65	0.58 to 1.79
	Dwell time1	33.96*	7.0 to 26.9	s
	Errors	0	0.0 to 4.5	#
Total time B/Total time A Object-hit test		1.79	0.95 to 2.30
	Target hits1,2	68.7*	80.1 to 97.0	%
	Median error1	64.7*	69.2 to 87.3	%
	Miss bias	3.1	-12.5 to 9.6	cm
	Right hand speed2,3	17.71*	23 to 49	cm/s
	Movement area (right hand)2,3	929*	1058 to 2161	cm2
	Left hand Speed2,3	21.59*	22.5 to 46.9	cm/s
	Movement area (right hand)2,3	1160	973 to 2178	cm2
	Hand bias hits3	−0.117*	−0.068 to 0.177
	Hand transition3	2.5	−6.5 to 3.2	cm
	Hand selection overlap1	13.1	7.6 to 20.9	%
	Hand speed bias3	−0.099*	−0.088 to 0.120
	Movement area bias3	−0.111*	-0.089 to 0.135

Typical range: 5%–95% (scale based on

¹age,

²sex, and

³handedness.

^*Scores outside the typical range.

Study data were collected and managed using the REDCap (Research Electronic Data Capture) Web-based application.³⁵ All statistical analyses were conducted using SPSS 22.0 (IBM Corp., Armonk, NY) and R (package: pROC, Hmisc).^36–38

Results

Subject Characteristics

A total of 1,423 ED patients were screened for eligibility during the study period. The majority of the screened patients either did not meet the eligibility criteria (872 patients) or declined to participate (293 patients). Sixty-six mild TBI patients consented to participate in the study. Of these, six withdrew consent and five were unable to complete testing. There were 42 subjects who completed testing and follow-up. Of these, two were excluded after follow-up, one due to comorbidities that could impact testing and another due to machine malfunction. There were 40 subjects included in the analyses (see Figure 4 for details).

Figure 4.

Study participants: 1,423 ED patients were screened. Sixty patients enrolled and remained in the study. Of these, 40 patients had robotic testing data and outcome data.

^aOther reasons include vulnerable population, patient was unconsentable or admitted, technical problems with the device, and testing area was unavailable.

^bAdditionally, three patients (without robotic testing data) completed follow-up.

^cTwo patients were excluded, one due to missing blood alcohol level value and machine problems and one due to comorbidity (horizontal nystagmus).

There were no meaningful differences in demographic characteristics between those who completed follow-up and those who did not (Table 3). Overall, the mean (±SD) age was 37 (±17) years, 26 of 40 (65%) were male, and 20 of 40 (50%) were Caucasian. Of the 40 included subjects, 25 (63%; 95% CI = 47% to 76%) reported significant PCS (Rivermead positive) 3 weeks following mild TBI. Subjects who did not develop PCS (Rivermead negative) were more often Caucasian (73%) when compared to those who did (36%), with a difference of 33% (95% CI, 4% to 63%). Otherwise, there were no significant differences in demographic characteristics between those who were Rivermead positive and Rivermead negative (Table 4).

Table 3. Demographic Characteristics by Follow-up Completed.

	Follow-up Completed (n = 40)	Follow-up Not Completed (n = 13)	p-value
Age (y), mean (±SD)	37 (±18)	36 (±16)	0.816
White, n (%)	20 (50.0)	6 (46.2)	0.810
Non-Hispanic, n (%)	40 (100.0)	13 (100.0)	—
Male, n (%)	26 (65.0)	11 (84.6)	0.181
Past medical history, n (%)
Allergies	21 (52.5)	2 (15.4)	0.019
Hypertension	10 (25.0)	3 (23.1)	1.000
Pulmonary disorders	6 (15.0)	1 (7.7)	0.667
Gastrointestinal disorders	5 (12.5)	0 (0.0)	0.317
Learning disabilities	5 (12.5)	3 (23.1)	0.382
Neurovascular disease	4 (10.0)	0 (0.0)	0.561
Autoimmune disorders	4 (10.0)	0 (0.0)	0.561
Diabetes	1 (2.5)	0 (0.0)	1.000
Cancer	1 (2.5)	0 (0.0)	1.000
Cardiovascular disease	1 (2.5)	0 (0.0)	1.000
Liver disease	0 (0.0)	1 (7.7)	0.245
Presenting vital signs, mean (±SD)
Systolic blood pressure (mm Hg)	135 (±22)	130 (±16)	0.426
Diastolic blood pressure (mm Hg)	81 (±15)	85 (±15)	0.394
Heart rate (beats/min)	83 (±13)	83 (±19)	0.988
Respiratory rate (beats/min)	17 (±2)	16 (±2)	0.149
O2 saturation (%)	97 (±2)	97 (±2)	0.636
Proportion of abnormal KINARM scores, median (range)
Total	20% (3%–54%)	34% (3%–67%)	0.234
Matching	25% (0%–79%)	50% (0%–92%)	0.064
Reaching	22% (0%–67%)	28% (0%–50%)	0.853
Trail Making	17% (0%–92%)	42% (0%–75%)	0.216
Hit avoid	8% (0%–62%)	15% (0%–69%)	0.619
Object hit	17% (0%–67%)	8% (0%–58%)	0.678

Table 4. Demographic Characteristics by Rivermead Status.

	Rivermead Positive (n = 25)	Rivermead Negative (n = 15)	p-value
Age (y), mean (±SD)	36 (±14)	38 (±23)	0.802
White, n (%)	9 (36.0)	11 (73.3)	0.048
Non-Hispanic, n (%)	25 (100.0)	15 (100.0)	—
Male, n (%)	15 (60.0)	11 (73.3)	0.443
Past medical history, n (%)
Allergies	12 (48.0)	9 (60.0)	0.462
Learning disabilities	4 (16.0)	1 (6.7)	0.633
Hypertension	5 (20.0)	5 (33.3)	0.457
Autoimmune disorders	4 (16.0)	0 (0.0)	0.278
Pulmonary disorders	3 (12.0)	3 (20.0)	0.651
Neurovascular disease	2 (8.0)	2 (13.3)	0.622
Gastrointestinal disorders	2 (8.0)	3 (20.0)	0.345
Cancer	1 (4.0)	0 (0.0)	1.000
Diabetes	0 (0.0)	1 (6.7)	0.375
Cardiovascular disease	0 (0.0)	1 (6.7)	0.375
Liver disease	0 (0.0)	0 (0.0)	—
Presenting vital signs, mean (±SD)
Systolic blood pressure (mm Hg)	135 (±21)	136 (±23)	0.925
Diastolic blood pressure (mm Hg)	83 (±15)	76 (±15)	0.126
Heart rate (beats/min)	86 (±13)	79 (±13)	0.086
Respiratory rate(beats/min)	17 (±2)	17 (±1)	0.992
O2 saturation	98 (±2)	97 (±2)	0.553

Robotic-assisted Neuromotor Assessment

Table 5 shows the proportion of abnormal scores in each robotic-assisted test for each patient, grouped by Rivermead status. The overall AUC for the robotic-assisted test battery was 0.72 (95% CI = 0.54 to 0.90). For the five individual robotic-assisted tests, the AUC for the visually guided reaching test was 0.80 (95% CI = 0.65 to 0.95) and the AUC for the arm-matching test was 0.71 (95% CI = 0.53 to 0.89), representing the greatest ability in discriminating between subjects who were Rivermead positive and Rivermead negative (Table 6 and Figure 5). Figures 2 and 3 show matching and reaching test results respectively for two study subject exemplars. A net reclassification analysis showed that the addition of the matching test to the reaching test provided minimal improvement to the reaching test, with a net reclassification index of 0.72 (95% CI = 0.11 to 1.33).

Table 5. Proportion of Abnormal Scores in Each Robotic-assisted Test for Each Patient, Grouped by Rivermead Status.

Rivermead Status	ID	Matching	Reaching	Trail Making	Object Hit	Hit Avoid	Total
Rivermead positive	1	75%	67%	42%	33%	31%	54%
	2	50%	61%	67%	50%	23%	51%
	3	58%	67%	42%	17%	31%	47%
	4	17%	67%	58%	67%	31%	44%
	5	46%	67%	33%	33%	31%	44%
	6	46%	50%	92%	17%	0%	42%
	7	50%	39%	33%	42%	31%	41%
	8	29%	44%	83%	33%	8%	38%
	9	42%	61%	33%	8%	23%	37%
	10	79%	22%	17%	17%	15%	37%
	11	71%	11%	50%	8%	0%	33%
	12	33%	33%	50%	8%	0%	27%
	13	33%	39%	25%	8%	0%	24%
	14	13%	28%	17%	33%	23%	22%
	15	21%	6%	0%	17%	8%	11%
	16	8%	28%	0%	0%	8%	10%
	17	17%	0%	8%	17%	0%	9%
	18	4%	0%	0%	25%	23%	9%
	19	8%	0%	0%	0%	23%	6%
	20	0%	6%	0%	8%	8%	4%
	21	25%	–	0%	17%	0%	–
	22	54%	50%	—	—	—	—
	23	42%	17%	—	—	—	—
	24	25%	—	—	—	—	—
	24	—	67%	17%	50%	0%	—
Rivermead negative	26	54%	0%	75%	42%	62%	44%
	27	8%	22%	58%	50%	54%	33%
	28	33%	28%	8%	0%	8%	19%
	29	17%	6%	50%	17%	8%	18%
	30	25%	17%	17%	25%	0%	18%
	31	17%	0%	0%	33%	38%	16%
	32	4%	0%	75%	8%	8%	15%
	33	25%	17%	0%	8%	8%	14%
	34	8%	0%	8%	17%	8%	8%
	35	17%	6%	0%	0%	0%	6%
	36	0%	17%	17%	0%	0%	6%
	37	4%	0%	8%	0%	0%	3%
	38	—	61%	8%	25%	15%	—
	39	17%	61%		25%	38%	—
	40	—	17%	0%	0%	0%	—

— = patient did not complete a particular test.

Table 6. Receiver Operating Characteristic Analysis of Robotic-assisted Tests.

				Asymptotic 95% CI
Test Result Variable(s)	Area	SE	Asymptotic Sig.	Lower Bound	Upper Bound
Matching	0.710	0.092	0.049	0.529	0.892
Reaching	0.800	0.077	0.005	0.648	0.952
Trail Making	0.560	0.109	0.572	0.346	0.775
Object hit	0.602	0.109	0.340	0.388	0.816
Hit avoid	0.563	0.115	0.559	0.337	0.788
Total	0.723	0.093	0.037	0.541	0.904

Figure 5.

ROC curves for robotic-assisted tests. ROC = receiver operating characteristic.

Discussion

This study is the first of its kind to describe the deployment of a sophisticated robotic-assisted assessment tool to evaluate mild TBI patients in the ED for developing subsequent PCS.³⁹ The analysis demonstrates that poor visuomotor and proprioceptive performance on robot-assisted testing is associated with Rivermead positive (presence of PCS at 3 weeks postinjury). This is an important step as morbidity among those with PCS is high;¹⁰ therefore, identification of those at high risk for PCS is critical for guiding utilization of limited clinical resources such as specialty follow-up and neurocognitive therapy and to facilitate patient education and expectation management.⁹ Furthermore, patients at high risk for suffering PCS are the target for future acute care interventional trials and accurate identification of these patients upon entry into the medical system is fundamental to the success of these trials.

Visuomotor performance such as arm movements in the reaching test are mediated by visual pathways and proprioceptive inputs (e.g., position sense in the arm matching test) are processed by cortical pathways that are distinct from those involved in vision for action.⁴⁰ Additionally, the reaching test measures deficits in motor performance that are largely independent from those measured by the matching test.²⁴ Therefore, both visuomotor and proprioceptive deficits are perhaps independently associated with long-term sequelae following mild TBI. These results are consistent with literature showing the relationship between motor and balance deficits and mild TBI.^41,42 While the findings reported here show prognostic capabilities among a cohort of mild TBI subjects, it is also possible that the testing could refine the diagnosis of mild TBI. Subjects were included based on a clinical diagnosis for mild TBI, which is fundamentally limited, and therefore the test findings may be more reflective of improved diagnosis of mild TBI in the ED. A deeper understanding of the relationship between visuomotor and proprioceptive performance will require further investigation in regard to both diagnostics and prognostication using robotic-assisted assessment methods.

The etiology of PCS is controversial, most likely because of the significant variability in prevalence and severity of protracted symptoms in the mild TBI population. A common belief is that multiple factors in both physiological and psychological domains contribute to the development and persistence of PCS.⁴³ Application of advanced engineering solutions to the acute care environment, such as the assessment reported here, has many benefits including providing objective data and increased sensitivity to detect and identify potentially important yet previously unrecognized deficits. Most importantly, identification of those acute deficits associated with longer-term sequelae may reveal useful information about the etiology of PCS.

TBI is a heterogeneous disease process and injury-related deficits may vary greatly by mechanism of injury, specific neural structures injured, and other patient-specific factors. The use of technology such as KINARM that allows for comprehensive evaluation of multiple neurologic domains may help characterize each individual patient and allow for more detailed classification of injuries. Future studies of mild TBI patients with these techniques may permit identification of injury phenotypes, provide personalized risk stratification for specific deficits, and allow for targeted intervention potentially before an issue becomes chronically debilitating.

Limitations

While our study demonstrated the association between robotic assessment of neuromotor performance following mild TBI and the presence of PCS at follow-up, the study was intended to be exploratory in nature and there are several limitations to note. First, there was no control group for this study. However, the KINARM standard tests have normative ranges from uninjured controls allowing for comparison to expected performance. Second, patients were included based on the clinical diagnosis of mild TBI. While this may result in workup bias, this was the most representative cohort to study in this initial investigation as it represents the cohort that would be targeted for testing in clinical practice. Also, time of testing and length of ED stay were not explicitly controlled for in the study, although all subjects were enrolled and tested in the ED within 24 hours of injury.

As with other studies of mild TBI conducted in the ED, a substantial number of patients were lost to follow-up. However, our follow-up rate of 75% was higher than follow-up rates in other mild TBI studies.⁴⁴ This could be due to the fact that the study excluded subjects susceptible to drop out such as those with high blood alcohol level and focal neurologic deficits. Based on sensitivity analyses (Table 7), the results are consistent under the assumption that all patients lost to follow-up were Rivermead positive. On the other hand, if all lost to follow-up cases were assumed to be Rivermead negative, the results did not perform as well. However, the finding of Rivermead positive was common (63%), and it is very unlikely that all the lost to follow-up cases were Rivermead negative. Furthermore, we note that there were no statistical differences in the group that was lost to follow-up relative to the cohort in whom follow-up was completed (Table 3), suggesting lack of systematic bias.

Table 7. Sensitivity Analysis of Robotic-assisted Tests.

				Asymptotic 95% CI
Test Result Variable(s)	Area	SE	Asymptotic Sig.	Lower Bound	Upper Bound
Assuming that all cases lost to follow-up are Rivermead positive
Matching	0.734	0.076	0.018	0.584	0.883
Reaching	0.804	0.064	0.002	0.679	0.930
Trail Making	0.585	0.100	0.390	0.389	0.780
Object hit	0.580	0.103	0.419	0.377	0.782
Hit avoid	0.583	0.108	0.397	0.372	0.795
Total	0.753	0.080	0.010	0.596	0.909
Assuming that all cases lost to follow-up are Rivermead negative
Matching	0.537	0.087	0.673	0.367	0.707
Reaching	0.685	0.084	0.035	0.521	0.849
Trail Making	0.493	0.089	0.936	0.319	0.667
Object hit	0.588	0.085	0.315	0.421	0.755
Hit avoid	0.542	0.089	0.631	0.368	0.716
Total	0.605	0.087	0.230	0.435	0.775

Conclusion

In this study, a robotic neuromotor assessment tool was deployed in the emergency care setting for the evaluation of acutely injured mild traumatic brain injury patients. The Kinesiological Instrument for Normal and Altered Reaching Movements test battery was able to discriminate between those who did and those who did not have postconcussion symptoms 3 weeks after initial injury. Additionally, poor visuomotor and proprioceptive performance, as assessed by the reaching and matching tests, were representative of discriminatory deficits for predicting future morbidity. Further studies with similar tests may be able to identify the mild TBI cohort that is likely to benefit from novel therapies and allow for efficient triage to follow-up after initial presentation in the ED.