Eye Tracking Metrics Differences among Uninjured Adolescents and Those with Acute or Persistent Post-Concussion Symptoms

Abstract

SIGNIFICANCE 

Eye tracking assessments that include pupil metrics can supplement current clinical assessments of vision and autonomic dysfunction in concussed adolescents.

PURPOSE 

This study aimed to explore the utility of a 220-second eye tracking assessment in distinguishing eye position, saccadic movement, and pupillary dynamics among uninjured adolescents, those with acute post-concussion symptoms (≤28 days since concussion), or those with persistent post-concussion symptoms (>28 days since concussion).

METHODS 

Two hundred fifty-six eye tracking metrics across a prospective observational cohort of 180 uninjured adolescents recruited from a private suburban high school and 224 concussed adolescents, with acute or persistent symptoms, recruited from a tertiary care subspecialty concussion care program, 13 to 17 years old, from August 2017 to June 2021 were compared. Kruskal-Wallis tests were used, and Bonferroni corrections were applied to account for multiple comparisons and constructed receiver operating characteristic curves. Principal components analysis and regression models were applied to determine whether eye tracking metrics can augment clinical and demographic information in differentiating uninjured controls from concussed adolescents.

RESULTS 

Two metrics of eye position were worse in those with concussion than uninjured adolescents, and only one metric was significantly different between acute cases and persistent cases. Concussed adolescents had larger left and right mean, median, minimum, and maximum pupil size than uninjured controls. Concussed adolescents had greater differences in mean, median, and variance of left and right pupil size. Twelve metrics distinguished female concussed participants from uninjured; only four were associated with concussion status in males. A logistic regression model including clinical and demographics data and transformed eye tracking metrics performed better in predicting concussion status than clinical and demographics data alone.

CONCLUSIONS 

Objective eye tracking technology is capable of quickly identifying vision and pupillary disturbances after concussion, augmenting traditional clinical concussion assessments. These metrics may add to existing clinical practice for monitoring recovery in a heterogeneous adolescent concussion population.

More than 1.9 million children sustain a concussion in the United States annually, with adolescents accounting for more than 50% of these injuries.¹ Sport- and recreation-related concussion is a complex injury that affects multiple neurological domains simultaneously and can negatively impact many aspects of daily adolescent life, including school and sport.^2–4 Visual and autonomic deficits frequently occur in adolescents after concussion.^5–7 Specifically, nearly 70% of concussed adolescents evaluated in a specialty care setting presented with visual deficits, such as convergence insufficiency, accommodative disorders, and/or saccadic dysfunction.⁵ Although most children with concussion achieve symptom recovery within 4 weeks, the presence of visual deficits after injury predicts prolonged symptoms and recovery.^8,9 As such, accurate detection of visual deficits soon after a concussion may allow for earlier intervention, potentially reducing overall symptom burden or the likelihood of experiencing persistent post-concussion symptoms.

Currently, visual deficits are assessed clinically via vision-specific symptom inventories and clinical examination of saccades, smooth pursuits, accommodation, and convergence.^10–14 Although these measures have demonstrated both utility and reliability among pediatric patients with concussion, they are not used ubiquitously among athletic trainers, who are often the first health care providers to conduct concussion assessments.¹⁵ Recent advances have enabled objective eye tracking to quantify visual deficits in acutely concussed children, adolescents, and adults.^16–18 Previous reports using this eye tracking methodology have described its utility in identifying deficits in eye positioning during a smooth pursuit task, achieving 71.9% sensitivity and 84.4% specificity in identifying concussion among a cohort of pediatric patients with a moderate level of reliability among uninjured adolescent athletes.^16,19 The utility of eye tracking for characterizing saccadic movement and pupillary dynamics is of particular interest because deficits in saccadic eye movements and pupillary dynamics have been described by researchers using other methods in the pediatric and adolescent concussion populations.^10,20,21 An eye tracking assessment that includes pupillary dynamic metrics may also provide useful information for both visual and autonomic function. Measuring right and left eye position, saccadic movement, and pupillary dynamics simultaneously during an eye tracking task could more comprehensively and objectively characterize deficits that may contribute to acute and persistent symptoms.

The purpose of this study was to determine if there are differences in objective eye tracking metrics that characterize eye position, saccadic movement, and pupillary dynamics between uninjured adolescents, adolescents with acute concussion symptoms (≤28 days since injury), and adolescents with persistent concussion symptoms (>28 days since injury). We hypothesized that those with acute or persistent concussion symptoms would have disconjugacy in left and right eye position, abnormal saccadic movement, and abnormal pupillary dynamics compared with the uninjured adolescent comparison group. We also hypothesized that there would be sex-specific differences between uninjured adolescents and those with acute or persistent concussion symptoms in these objective eye tracking metrics because female adolescents demonstrate greater oculomotor and vestibular dysfunction in comparison with male adolescents after a concussion.²²

METHODS

Study Design, Setting, and Participants

Participants aged 13 to 17 years were enrolled between August 2017 and June 2021 as part of a prospective observational cohort study assessing a suite of clinical and objective measures, including objective eye tracking, approved by the Children's Hospital of Philadelphia Institutional Review Board.¹³ Participants and/or their parents/legal guardians provided written assent/informed consent. Uninjured athletes (n = 180) were recruited from a private suburban high school and completed an objective eye tracking assessment before their sports season. Concussed participants (n = 224) were recruited during a clinical care visit from the Children's Hospital of Philadelphia Minds Matter Concussion program, as well as from the high school. Thirty-two participants who enrolled in the uninjured cohort subsequently sustained a concussion and, for the purposes of this analysis, were included only in the concussed cohort. Concussion diagnosis was made by a trained sports medicine pediatrician according to the most recent Consensus Statement on Concussion in Sports.⁴ Concussed participants performed the objective eye tracking assessment during a clinical visit. If a participant with concussion had multiple eye tracking assessments across several visits, only the first assessment was used in this analysis. Participants with a concussion diagnosis were categorized into two groups for analysis: an acute group, who completed their first eye tracking assessment ≤28 days since injury, and a persistent group, who completed their first eye tracking assessment >28 days since injury. Exclusion criteria for both concussed and uninjured participants included a previous concussion within 1 month of injury or pre-injury assessment, and any ocular or neurologic condition that could affect eye tracking responses.

Of the 346 participants enrolled, valid eye tracking assessments were obtained for 178 of 180 uninjured (98.8%) and 219 of 224 (97.8%) concussed participants. Among those without a valid assessment, five had insufficient data capture during the assessment of one or both eyes. Data from two additional subjects were lost because of device error.

Instrumentation

Eye movements were recorded using EyeBOX (Oculogica, Inc, New York, NY), which uses an Eyelink 1000 (SR Research, Ottawa, Ontario, Canada), positioned at a fixed distance of 55 cm from the participant's eyes while they track during a 220-second stimulus video (Fig. 1). The stimulus video, a 220-second child's movie music video clip with a 4:3 aspect ratio, occupied approximately one-ninth of the display monitor area and moved in a smooth clockwise direction along the outer edges of the monitor at a rate of 10 seconds per edge of the monitor. The total visible span of the moving aperture was approximately 17° horizontally and 13° vertically from the middle of the screen. Eye position data from both eyes were obtained independently at 500 Hz. Because the position of the head was fixed, it is important to note that the eye position, the position of the eye in the orbit, and the gaze position, the position of the eye and head combined, are equivalent. The EyeBOX is not spatially calibrated, as the metrics of interest are derived from changes in eye position over time, allowing for independent analysis of each pupil position with respect to the moving visual stimulus. The EyeBOX pupil mode was set to center of mass. The first and last 10 seconds of the assessment were discarded to reduce potential noise from capturing the onset and completion of the eye tracking task. The eye tracking data were then automatically processed to yield 256 eye tracking metrics that quantify (a) eye position (the position of the left and right pupils independently and comparisons of their positions over time during the task), (b) saccadic movement in each eye, and (c) pupillary dynamics for each eye, both independently and together. Eye position metrics were derived from the x and y spatial coordinates of each pupil and by comparing the coordinates of one eye to the other eye at any given time point throughout the entire assessment.²³ There were a total of 183 metrics of eye position, 60 metrics of saccadic movement metrics, and 13 metrics of pupillary dynamics. The large number of eye position and saccadic movement metrics are due to the fact that these metrics are evaluated based on the side of the screen the video was on (i.e., the top, right, bottom, and left) for both the eyes individually and comparisons between the two. Although the video stimulus moves smoothly along the edges of the screen, saccadic eye movements were captured when the participant glanced between objects, such as two characters, within the video stimulus. Saccadic eye movements were automatically detected based on right and left eye position velocities.

FIGURE 1: (A) Image of eye tracker with trained research staff and mock participant completing the assessment. (B) Image of screen that participant views during the assessment with the x and y axes labeled. Some eye tracking metrics related to eye position and saccadic movement are measured in either the x and y axis directions.

Procedures

Participants completed questionnaires to capture demographic and clinical characteristics, such as age at the time of assessment, sex, race/ethnicity, and prior history of concussion. The Post-Concussion Symptom Inventory (PCSI), a self-report of symptom severity over the past 2 days (21 items on a 7-point Likert scale [0, none; 6, most severe], total score range of 0 to 126), was completed on the day of the eye tracking assessment for participants with acute or persistent concussion symptoms and within 7 days of the eye tracking assessment for uninjured participants.²⁴ Trained research staff conducted the objective eye tracking assessment in an athletic training room or sports medicine office and were not blinded to concussion status. Participants sat in a chair and placed their chin on a height-adjustable chin rest to limit movement. To confirm detection of pupils, participants were instructed to focus on the computer monitor, and research staff confirmed that the device detected both left and right pupils around all four edges of the computer screen. Participants then were instructed to focus on a 220-second video clip that traveled in a smooth clockwise rotation along the edges of the monitor while eye movements were recorded as described previously.

Statistical Analyses

Distributions of demographic and clinical characteristics for those with acute or persistent concussion symptoms and uninjured adolescents were compared using χ² statistics and Fisher exact tests for categorical variables (sex, race/ethnicity, and history of prior concussion) and analysis of covariance with Tukey honestly significant difference post hoc testing for continuous variables (age and PCSI score). All 256 eye tracking metrics were compared among uninjured adolescents and those with acute or persistent concussion symptoms, with Kruskal-Wallis tests and Dunn post hoc tests. In an exploratory analysis of differences in eye movement post-injury by sex, eye tracking metrics were compared between uninjured participants and those with acute or persistent concussion symptoms among female and male subjects separately, with Kruskal-Wallis tests and Dunn post hoc tests. Multiple comparisons were accounted for by calculating Bonferroni corrections where a P < .05/256 or .00019 was considered significant for the Kruskal-Wallis tests and P < .05/768 or .000065 was considered significant for the Dunn post hoc tests. We also calculated area under the curve (AUC) values from receiver operating characteristic (ROC) curves and Cohen d effect sizes for metrics significantly different between groups. Finally, an analytic regression model based on the discriminatory eye tracking metrics using principal components analysis was created. All analyses were conducted using R (R Foundation for Statistical Computing, Vienna, Austria).²⁵

RESULTS

Study Population

Among participants with a concussion and evaluable eye tracking assessments, 130 (59.4%) had acute concussion, and 89 (40.6%) had persistent concussion symptoms (Table 1). A total of 75.8% had a sport- or recreation-related concussion. The uninjured cohort differed from the persistent concussion symptom cohort in race and ethnicity. The uninjured cohort was also slightly younger than the persistent concussion symptom cohort. Those with acute concussion symptoms completed an eye tracking assessment within a median interquartile range of 11 (4 to 19.75) days after injury, whereas those persistent with concussion symptoms completed an eye tracking assessment within a median interquartile range of 53 (40–68) days after injury. Participants with acute and persistent concussion symptoms reported significantly greater overall symptom severity than uninjured participants. The persistent concussion symptom group had a significantly higher proportion of females (69.7%) than the acute concussion symptom group (47.7%, P = .004) and the uninjured group (53.4%, P = .02). A significantly greater proportion of the acute and persistent concussion symptom groups reported a prior history of concussion compared with the uninjured group. Demographic and clinical characteristics associated with concussion status were not subsequently included as covariates in comparisons of eye tracking metrics across groups, as these variables did not show an association with the eye tracking metrics.

TABLE 1 - Demographic and clinical characteristics of the study cohort, with continuous data presented as median (IQR)

	Acute concussion symptoms (n = 130)	Persistent concussion symptoms (n = 89)	Uninjured (n = 178)	P	Post hoc test P
Age, median (IQR) (y)	15.6 (14.5–16.9)	16.0 (14.7–17.0)	15.1 (14.3–16.3)	.01	Acute vs. persistent: .74 Acute vs. uninjured: .09 Persistent vs. uninjured: .02
Days after injury, median (IQR)	11.0 (4.0–19.75)	53.0 (40.0–68.0)
PCSI total score (IQR)*	22.0 (9.0–49.0)	28.0 (6.0–51.0)	3.0 (0.0–7.2)	<.001	Acute vs. persistent: .87 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
Sex, female, n (%)	62 (47.7)	62 (69.7)	95 (53.4)	.004	Acute vs. persistent: .002 Acute vs. uninjured: .38 Persistent vs. uninjured: .02
Race/ethnicity, n (%)				.04	Acute vs. persistent: .20
Hispanic	5 (3.8)	5 (5.6)	8 (4.5)		Acute vs. uninjured: .20
Non-Hispanic Black	9 (6.9)	11 (12.4)	16 (9.0)		Persistent vs. uninjured: .009
Other/unknown	14 (10.8)	4 (4.5)	33 (18.5)
Non-Hispanic White	102 (78.5)	69 (77.5)	121 (68.0)
Prior history of concussion,† n (%)	57 (43.8)	37 (41.6)	50 (28.1)	.008	Acute vs. persistent: .86 Acute vs. uninjured: .006 Persistent vs. uninjured: .03

Bold values are significant. *A total of 122 (93.8%) with acute concussions symptoms, 85 (95.5%) with persistent concussion symptoms, and 148 uninjured completed PCSI. †One with acute concussion symptoms and one with persistent concussion symptoms did not report prior history of concussion. IQR = interquartile range; PCSI = Post-Concussion Symptom Inventory.

Eye Tracking Metrics in Concussion

Of the 256 metrics, there were significant differences among uninjured participants and those with acute or persistent concussion symptoms after Bonferroni correction for multiple comparisons, in 13 metrics, 11 of which were related to pupillary dynamics. Table 2 includes definitions of these 13 metrics. All pupil size metrics represent the area of the pupil as calculated by the number of pixels the pupil occupies in the camera image. Participants with acute or persistent concussion symptoms had significantly larger left and right mean, median, and minimum and maximum pupil size (acute vs. uninjured, P < .001; persistent vs. uninjured, P < .001). Those with acute or persistent concussion symptoms also had significantly greater pupil asymmetry, with greater mean, median, and variance differences in left and right pupil size (mean: acute vs. uninjured, P < .001; persistent vs. uninjured, P < .001; median: acute vs. uninjured, P < .001; persistent vs. uninjured, P = .003; variance: acute vs. uninjured, P < .001; persistent vs. uninjured, P < .001). Two eye position metrics were significantly different among groups—conj.varYbot, defined as the variance between the left and right eyes in the y direction along the bottom of the screen (acute vs. uninjured, P = .003; persistent vs. uninjured, P < .001; acute vs. persistent, P = .04), and conj.varYtopbotRatio, defined as the total variance between the left and right eyes in the y axis around forced average of zero for the ratio of the segment where the visual stimulus moves across the top of the screen to the segment where the visual stimulus moves across the bottom of the screen (acute vs. uninjured, P < .001; persistent vs. uninjured, P = .001). Median and interquartile range values for each metric for each group are presented in Table 3. Violin plots of uninjured adolescents, acute cases, and persistent cases for the 13 metrics can be found in Appendix Figures A1 to A13, available at https://links.lww.com/OPX/A573. Area under the ROC curve, sensitivity, specificity, and effect sizes for the 13 metrics can be found in Appendix Tables A1 to A3, available at https://links.lww.com/OPX/A572.

TABLE 2 - Definitions of eye tracking metrics significantly different between groups

Metric	Definition
left.pupilsizemean	The average pupil size (area) of the left eye
left.pupilsizemedian	The median pupil size (area) of the left eye
right.pupilsizemean	The average pupil size (area) of the right eye
right.pupilsizemedian	The median pupil size (area) of the right eye
left.pupilsizemaxabs	The maximum pupil size (area) of the left eye
right.pupilsizemaxabs	The maximum pupil size (area) of the right eye
left.pupilsizeminabs	The minimum pupil size (area) of the left eye
right.pupilsizeminabs	The minimum pupil size (area) of the right eye
conj.pupilsizediffmean	The average difference in pupil size between the left and right eyes, a measure of pupil asymmetry
conj.pupilsizediffmedian	The median difference in pupil size between the left and right eyes, a measure of pupil asymmetry
conj.pupilsizediffvar	The variance of the differences in pupil size between the left and right eyes, a measure of pupil asymmetry
conj.varYbot	The total variance between the left and right eyes in the y direction around forced average of zero when the video moves along the bottom of the screen
conj.varYtopbotRatio	The total variance between the left and right eyes in the y axis around forced average of zero for the ratio of the segment where the visual stimulus moves across the top of the screen to the segment where the visual stimulus moves across the bottom of the screen

TABLE 3 - Eye tracking metrics (median and IQR) distinguishing uninjured adolescents from those with acute or persistent concussion symptoms

	Acute concussion symptoms (n = 130)	Persistent concussion symptoms (n = 89)	Uninjured (n = 178)	Adjusted P	Post hoc test P
left.pupilsizemean	2087.13  (1830.50–2449.87)	2247.83  (1673.73–2601.17)	1327.01  (1057.57–1660.14)	<.001	Acute vs. persistent: .46 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
left.pupilsizemedian	2079.00  (1810.00–2447.25)	2243.00  (1667.50–2609.50)	1278.00  (1052.00–1640.50)	<.001	Acute vs. persistent: >.43 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizemean	1949.90  (1619.07–2488.29)	2128.56  (1681.00–2579.42)	1264.09  (1011.58–1603.89)	<.001	Acute vs. persistent: .22 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizemedian	1930.50  (1596.75–2469.00)	2122.00  (1661.50–2608.88)	1248.00  (990.50–1587.00)	<.001	Acute vs. persistent: .20 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
left.pupilsizeminabs	1048.50  (646.25–1394.25)	1200.00  (862.00–1611.00)	762.00  (532.75–1002.00)	<.001	Acute vs. persistent: .08 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
left.pupilsizemaxabs	2865.00  (2313.25–3349.50)	2948.00  (2401.00–3555.00)	2156.00  (1749.50–2787.50)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizeminabs	1021.00  (673.00–1374.00)	1212.00  (948.00–1622.00)	715.00  (509.25–953.75)	<.001	Acute vs. persistent: .04 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizemaxabs	2652.00  (2175.75–3234.75)	2903.00  (2308.00–3539.00)	2052.00  (1631.50–2551.75)	<.001	Acute vs. persistent: .54 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.pupilsizediffmean	184.45  (113.63–269.44)	191.51  (106.27–286.01)	117.15  (64.28–196.57)	.02	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.pupilsizediffmedian	180.50  (100.00–260.00)	173.00  (95.00–270.00)	109.50  (58.00–189.25)	.03	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.pupilsizediffvar	6250.63  (3841.05–10,860.39)	6594.06  (3769.28–13,460.44)	2833.55  (1410.97–6255.91)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.varYbot	0.002  (0.001–0.005)	0.003  (0.002–0.009)	0.001  (0.001–0.003)	<.001	Acute vs. persistent: .04 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.varYtopbotRatio	1.01  (0.40–5.79)	0.93  (0.38–2.57)	2.10  (0.85–7.50)	.004	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: .001

Adjusted P values are reported by multiplying the actual P value by the number of comparisons. Bold values are significant. IQR = interquartile range.

Exploratory Analysis of Stratification by Sex

To explore changes in eye position, saccadic movement, and pupillary dynamic metric post-injury by sex, eye tracking metrics were compared between uninjured adolescents and those with acute or persistent concussion symptoms among female and male subjects separately. Demographic and clinical characteristics of male and female participants can be found in Appendix Tables A4 and A5, available at https://links.lww.com/OPX/A572. After Bonferroni correction, 12 of 13 metrics found to be significantly different between uninjured participants and those with acute or persistent concussion symptoms in the overall sample were also significantly different among female subjects according to status (uninjured, acute or persistent concussion symptoms) (Table 4). In contrast, only four eye tracking metrics were significantly associated with concussion status (Table 5).

TABLE 4 - Eye tracking metrics median and IQR values in female subjects

	Acute concussions symptoms (n = 62)	Persistent concussion symptoms (n = 62)	Uninjured (n = 95)	Adjusted P	Post hoc test P
left.pupilsizemean	2087.13  (1830.50–2449.87)	2247.83  (1673.73–2601.17)	1327.01  (1057.57–1660.14)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
left.pupilsizemedian	2079.00  (1810.00–2447.25)	2243.00  (1667.50–2609.50)	1278.00  (1052.00–1640.50)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizemean	1949.90  (1619.07–2488.29)	2128.56  (1681.00–2579.42)	1264.09  (1011.58–1603.89)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizemedian	1930.50  (1596.75–2469.00)	2122.00  (1661.50–2608.88)	1248.00  (990.50–1587.00)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
left.pupilsizeminabs	1203.50  (766.50–1443.25)	1254.50  (897.25–1639.50)	740.00  (535.00–955.50)	<.001	Acute vs. persistent: .58 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
left.pupilsizemaxabs	2929.00  (2619.25–3361.25)	2952.50  (2394.00–3523.50)	1993.00  (1652.00–2631.00)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizeminabs	1155.00  (728.25–1422.75)	1220.00  (1021.00–1676.75)	711.00  (514.00–936.50)	<.001	Acute vs. persistent: .29 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizemaxabs	2862.00  (2413.25–3341.50)	2878.50  (2293.00–3496.75)	1993.00  (1600.50–2423.50)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.pupilsizediffmean	198.90  (129.61–271.89)	195.00  (109.11–287.83)	109.54  (57.26–178.06)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.pupilsizediffmedian	189.50  (125.50–268.25)	192.00  (102.75–281.25)	98.00  (51.50–168.00)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.pupilsizediffvar	6958.52  (4566.90–11,788.72)	6953.30  (3730.72–13,544.58)	2399.70  (1011.98–5351.12)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
conj.varYbot	0.003  (0.001–0.005)	0.004  (0.002–0.009)	0.001  (0.001–0.003)	.008	Acute vs. persistent: .29 Acute vs. uninjured: .02 Persistent vs. uninjured: <.001
conj.varYtopbotRatio	0.84  (0.38–5.27)	0.96  (0.43–2.38)	2.96  (0.77–7.66)	.26

Adjusted P values are reported by multiplying the actual P value by the number of comparisons. Bold values are significant. IQR = interquartile range.

TABLE 5 - Eye tracking metrics median and IQR values in male subjects

	Acute concussions symptoms (n = 68)	Persistent concussion symptoms (n = 27)	Uninjured (n = 83)	Adjusted P	Post hoc test P
left.pupilsizemean	1882.89  (1540.90–2268.86)	2140.47  (1789.32–2869.61)	1501.37  (1194.04–1916.11)	.003	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
left.pupilsizemedian	1883.50  (1532.75–2264.25)	2137.00  (1775.00–2866.50)	1450.00  (1149.00–1883.00)	.003	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizemean	1760.48  (1442.74–2209.21)	1960.56  (1738.08–2722.14)	1393.98  (1091.12–1864.82)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
right.pupilsizemedian	1753.00  (1446.00–2145.00)	1952.00  (1715.00–2732.00)	1374.00  (1084.00–1873.50)	<.001	Acute vs. persistent: >.99 Acute vs. uninjured: <.001 Persistent vs. uninjured: <.001
left.pupilsizeminabs	988.50  (621.00–1339.75)	1000.00  (815.50–1481.50)	788.00  (545.50–1072.00)	>.99
left.pupilsizemaxabs	2599.00  (2177.75–3308.25)	2874.00  (2532.50–3775.00)	2350.00  (1844.50–2943.00)	>.99
right.pupilsizeminabs	914.50  (652.00–1303.75)	1147.00  (716.50–1455.00)	772.00  (503.50–985.50)	.13
right.pupilsizemaxabs	2456.50  (2016.00–3182.75)	2983.00  (2440.00–3624.50)	2218.00  (1698.50–2730.00)	.06
conj.pupilsizediffmean	178.12  (104.66–257.71)	152.12  (93.51–230.90)	157.70  (77.11–247.67)	>.99
conj.pupilsizediffmedian	166.50  (83.50–250.50)	130.00  (84.50–217.00)	122.00  (67.00–197.00)	>.99
conj.pupilsizediffvar	5877.04  (3199.47–10463.39)	5961.67  (3961.14–12,436.00)	3073.95  (1798.52–6950.02)	.11
conj.varYbot	0.002  (0.001–0.004)	0.003  (0.002–0.008)	0.001  (0.001–0.003)	.22
conj.varYtopbotRatio	1.15  (0.42–6.15)	0.72  (0.23–2.71)	1.93  (1.03–5.16)	>.99

Adjusted P values are reported by multiplying the actual P value by the number of comparisons. Bold values are significant. IQR = interquartile range.

Regression Model

Based on the univariate findings of the eye tracking metrics, we applied logistic regression models to determine whether eye tracking metrics can augment clinical and demographic information in the differentiation of uninjured from concussed adolescents, regardless of how many days since injury the eye tracking assessment was completed. We chose to combine our two groups of cases and conduct the regression analysis on two groups, as the univariate analyses did not show significant differences between acute and persistent cases. Any participants with missing demographics or clinical data such as PCSI total score or self-reported prior history of concussion were excluded from this analysis, resulting in 148 uninjured and 207 concussed adolescents (122 acute and 85 persistent).

To reduce redundancy and collinearity in the model because of high correlations between some of the 13 eye tracking metrics found to be significantly different between uninjured and concussed adolescents, a principal components analysis was applied. A similarity matrix calculated with Pearson r can be found in Appendix Figure A14, available at https://links.lww.com/OPX/A573. A total of four principal components described 84.5% of the variance within the 13 eye tracking metrics for this set of participants. A scree plot depicting the proportion of variance explained for each principal component can be found in Appendix Figure A15, available at https://links.lww.com/OPX/A573. A table describing the contribution of each variable to principal components 1 to 4 can be found in Appendix Table A6, available at https://links.lww.com/OPX/A572.

A null logistic regression model including only the clinical and demographics metrics of age, PCSI total score, sex, and self-reported prior history of concussion was used to predict concussion status. A second logistic regression model included the same clinical and demographics metrics and the first four principal components of the eye tracking metrics to predict concussion status. The second model that included eye tracking metrics displayed a significantly better fit based on DeLong comparison method (AUC [95% confidence interval], 0.889 [0.857 to 0.922]; sensitivity, 73.8%; specificity, 89.3%) than the null model (AUC [95% confidence interval], 0.847 [0.807 to 0.886]; sensitivity, 68.9%; specificity, 89.3%) (P < .001). Model results are reported in Appendix Tables A7 and A8, available at https://links.lww.com/OPX/A572.

DISCUSSION

Vision and autonomic dysfunctions are common sequelae of concussion in adolescents and may be a driver of both acute and persistent symptoms.^5,7,9 Objective eye tracking technology may quickly identify vision disturbances after concussion to allow for earlier recognition, referral, and treatment, potentially reducing long-term effects.^16–18 To our knowledge, this study is the first to demonstrate the utility of objective eye tracking in distinguishing uninjured adolescents from those with acute or persistent concussion symptoms based on metrics of eye position, saccadic movement, and pupillary dynamics. Prior work performed in an emergency department and a tertiary specialty care clinic used a similar experimental protocol and found different eye tracking metrics that distinguished between concussed and uninjured participants.^16,17 The findings of this study add to prior studies by including those with acute and persistent concussion symptoms. In addition, the metrics analyzed in this study expanded on those previously studied by examining not only eye position but also saccadic movement and pupillary dynamics.

Objective Measures of Vision and Autonomic Dysfunction

The increase in pupil size during the eye tracking assessment found in both those with acute and persistent concussion symptoms in comparison with uninjured adolescents may be due to an excitatory-inhibitory autonomic imbalance post-injury, resulting in excessive sympathetic tone.²⁶ These findings align with prior work that found amplification in multiple metrics quantifying monocular pupillary light reflex among adolescents injured within 28 days of injury.²¹ Of note, in our study, three pupillary dynamic metrics related to differences between left and right pupil area were larger among those with acute and persistent concussion symptoms compared with uninjured participants. Few studies have explored pupillary asymmetry post-concussion in adolescents during a dynamic eye tracking task; previous studies that have explored pupillary asymmetry post-concussion during assessments of the pupillary light reflex have not found significant asymmetry in pupillary response but were performed with the pupil in a stationary position.^27,28 In addition, the eye position metrics found to be significantly different in both those with acute and those with persistent concussion symptoms were also related to differences between the left and right eyes. These findings suggest that, in the adolescent population, the dynamic task of tracking a moving video places a workload strain on the autonomic nervous system, potentially enhancing the asymmetry to a greater degree than the static assessment of the pupillary light reflex.

Although we identified 13 metrics that were significantly different between uninjured adolescents and those with either acute or persistent concussion symptoms, after adjusting for multiple comparisons, most metrics analyzed were not significantly associated with concussion status, and almost none were significantly different between acute cases and persistent cases. This could be due to the heterogeneity of concussion, as it is likely that not all concussions had visual dysfunction after injury.⁵ It is also possible that the automatically derived metrics are not all clinically relevant in this population. Although the application of Bonferroni corrections to account for multiple comparisons is a conservative method that generally controls for false positives, there is a possibility that one or more of these metrics was falsely associated with concussion. This may be of greater relevance for the 2 metrics of eye position, as only these 2, of a total of 183 metrics of eye position, were found to be associated with concussion in comparison with 11 of a total of 13 metrics of pupillary dynamics. Of note, no saccadic movement metrics were found to distinguish between those with either acute or persistent concussion symptoms and uninjured participants. This may be because, although saccadic eye movements were quantified when the participant glanced between objects within the video stimulus, the primary eye tracking task of following the video stimulus that is moving smoothly along the edges of the screen is smooth pursuit.

Prior studies have looked at the BOX score, a binary classifier derived from a best subset regression model that combines many of the eye tracking metrics obtained from this assessment and has been used for diagnostic purposes in pediatric concussion populations and also been associated with symptom severity.^16,18,29 In this study, however, BOX score was not significantly associated with concussion status (Appendix Tables A9 to A12, available at https://links.lww.com/OPX/A572). Thus, we chose to reduce the redundancy of the eye tracking metrics found to be significantly different between groups and verify the discriminatory capability of these metrics through a principal components analysis and applied them as predictors in a logistic regression model. The model that included the transformed eye tracking metrics along with clinical and demographics information achieved a similar area underneath the ROC curve to a prior study that included concussed and uninjured participants between 4 and 21 years of age from both sexes.¹⁶ Although this model performed better than the model that included clinical and demographics data alone, the generalizability of the principal components analysis outside of this set of participants is unknown. Future work should include validating the association of these 13 metrics with concussion status in a different sample of uninjured and concussed adolescents. Future work should also explore alternative methods of distilling these 256 metrics for application to a heterogeneous concussed adolescent population and integration into existing clinical concussion protocols for monitoring recovery.

Sex-specific Differences

This study also explored the association of injury status and eye tracking metrics among female and male subjects separately. Because we found significant differences between uninjured adolescents and those with concussions among female participants for 12 of 13 metrics, but only in 4 among male participants, we suspect that the female participant data were driving the population differences between uninjured, and acute and persistent concussions in this cohort. Although the sample size, particularly of males with persistent symptoms, is fairly low, this supports recent findings that female adolescents demonstrate greater oculomotor and vestibular dysfunction in comparison with male adolescents after a concussion, and female pediatric patients take longer to recover from visual and vestibular deficits than their male counterparts.^22,30 The preliminary sex-specific findings from this study, combined with evidence that female adolescents present with greater symptom severity and longer recovery times, indicate that future work should prospectively investigate the evolution of these eye tracking metrics across recovery in both male and female adolescents with concussion to understand if acute measures can serve prognostic purposes.^30–33

Limitations

The population was limited to adolescents aged 13 to 17 and thus may not be generalizable to younger pediatric or older adult populations. In addition, adolescents with concussions were recruited from a specialized tertiary care center, and the number of days after injury the eye tracking assessment was completed was not controlled, so it is possible that the utility of metrics found to be significantly different in these populations is not applicable to individuals completing the assessment hours after injury or more than 3 months after injury. It is likely that not all concussed participants presented with visual disturbances after concussion, and by combining participants with and without visual disturbances, the ability to detect between-group differences for visual function was reduced. However, this represents a typical clinical population, and examining the utility of eye tracking in this cohort is clinically relevant. It is possible that some participants had underlying diagnosed or undiagnosed vision problems that might account for differences in objective measures of eye movements after concussion. Iris color, which may have an impact on automatic pupil detection with this eye tracker, was not collected in this study. Future work should include iris color as part of data collection to better understand any potential effect of iris color on these metrics. Only one visual stimulus, not well studied in the literature outside of the use of the EyeBOX, was used in this study. Future work should investigate other visual stimuli and paradigms to determine whether other standard assessments of oculomotor function can identify deficits in concussed children.

CONCLUSIONS

Deficits after a concussion are not specific to a single neurological domain. Eye tracking is a promising objective method that may potentially supplement current clinical assessments as a dynamic objective measure of vision and autonomic dysfunction. In this study, 13 eye tracking metrics were found to be significantly different between uninjured adolescents and those with either acute or persistent concussion symptoms, encompassing the domains of eye position, saccadic movement, and pupillary dynamics. Sex-specific differences were identified, where most eye tracking metrics were significantly different based on injury status in females, but only four in males. An analytic model that included transformed eye tracking metrics with clinical and demographics information was better able to discriminate uninjured adolescents from concussed adolescents than a model with clinical and demographics information alone. Future work should involve modifying and consolidating these metrics in a generalizable way and combining them with existing clinical measures to monitor recovery in a heterogeneous adolescent concussion population.