Abstract Text: Oculomotor testing has been a mainstay of diagnostic vestibular evaluation for decades due to its sensitivity to changes in health of the neural pathways required for normal ocular motor function that can differentiate central versus peripheral vestibular involvement. As eye-tracking technology has advanced and become more readily available, research has demonstrated the capability of oculomotor assessment to detect acute and sub-acute neurotrauma and products are now receiving FDA approval for use in concussion diagnosis. Towards this end, three commercially available eye-tracking devices (SyncThink EYE-SYNC, Oculogica EyeBOX, NeuroKinetics IPAS) were identified by the US Army Medical Research and Development Command Non-Invasive NeuroAssessment Devices (NINAD) Integrated Product Team as meeting criteria towards being operationally effective in detection of TBI in service members and thus broadening settings for TBI diagnostic or screening capability. This study was developed to perform analytical comparison of these three devices to determine viability of use with servicemembers. Participants with acute mild traumatic brain injury (mTBI) were recruited from the Womack Army Medical Center after being identified as experiencing mTBI, whereas control participants without history of mTBI were recruited at both Walter Reed National Military Medical Center and at Fort Bragg. All participants were between the ages of 18 and 45 years of age to minimize age-related effects and completed a TBI assessment protocol with all three devices, counterbalanced across participants. Respective data for each device was analyzed to determine an optimal model for classification of subjects as TBI or control, which was then used to generate Area Under the Curve (AUC) results for each device to enable comparison of device sensitivity/specificity to TBI. This analysis approach was necessary as each device utilizes proprietary tests and algorithms thus direct comparison of device results was not possible or appropriate. Overall, 63 participants identified as having mTBI were recruited as TBI subjects and 119 participants without history of TBI were recruited as control subjects. According to criteria applied internally by each device or by subject matter expert review of the collected data, data obtained from some subjects were determined to be of poor or unusable quality. Depending on device, between 6% and 43% of subjects were deemed to have insufficient quality data for analysis and were excluded from analysis for each respective devices as the presence of invalid and poor-quality data affects the respective AUCs for each device. However, some subjects with insufficient quality data for one device had to be retained in analyses due to the large number of subjects with poor data quality. The remaining TBI subjects were then matched with controls by age and gender to generate a balanced dataset for the final comparative analysis, leaving a total of 98 subjects, 49 with TBI and 49 controls. AUCs obtained for this balanced analysis ranged from 0.666 to 0.773, demonstrating varying ability to discriminate TBI subjects from controls by all three devices when data quality is sufficient. When all subjects were included in the analysis regardless of their data quality, AUCs obtained ranged from 0.583 to 0.661. An AUC of 0.850 was the target objective of the NINAD at the time of study development. Overall, these findings confirm the growing evidence that eye-tracking can be used to detect neurotrauma following injury; however, there is opportunity for capability improvement. Device-dependent variability in data quality, length of testing, and ease of use must be taken into account for NINAD objectives and DoD implementation.
