Longitudinal Assessment of Balance and Gait After Concussion and Return to Play in Collegiate Athletes

We recruited a total of 50 volunteers (25 concussed individuals, 25 matched healthy individuals) during the 2015 and 2016 athletic seasons from 6 sports and athletics departments of universities within the Portland, Oregon, area. Analysis of RTP dates revealed that 2 concussed participants did not RTP during the 8-week follow-up period, so we removed them from further analysis (Table 1). Athletes who had sustained concussions were referred to the study by the team athletic trainer within 24 to 48 hours postinjury. Concussed athletes were included if they were 18 years of age or older and had received a diagnosis of concussion from either their team clinician or an Oregon Health and Science University (OHSU) sports physician. Clinicians used the Sports Concussion Assessment Tool and clinical judgment to diagnose concussions in athletes. Healthy individuals serving as controls were student-athletes participating in sports in the same universities' athletics departments. They were matched for age, sex, height, mass, and, where possible, sport. Participants with a history of medical conditions that would impair cognition or mobility, including any injury or surgery within the 6 months before the study that would affect balance, were excluded. All universities gave written approval for OHSU to conduct research at their campuses. All participants provided written informed consent, and the study was approved by the OHSU Institutional Review Board.

Participants were assessed during 9 testing sessions over 8 weeks postinjury to evaluate the recovery of balance and gait. The timeline involved 2 testing sessions in the first week, followed by weekly testing for 7 weeks. Healthy participants were tested using the same time intervals for comparison.³²  Athletes completed testing sessions at their universities of enrollment. For consistency across universities, testing was conducted in a well-lit, straight hallway with a firm surface. The testing site at each university was the same for each test day.

Each testing session included an instrumented assessment of balance and gait. To complete the assessment, we positioned 3 wireless inertial sensors (Opal; APDM Inc, Portland, OR) bilaterally on the anterior and distal aspect of each shank³³  and posterior pelvis at the height of L5.³⁴  We attached each inertial sensor to the body using an elastic belt and adjusted the belt to fit snugly enough to reduce unwanted sensor movement without being uncomfortable to the participant.

Our balance-assessment procedures have been previously described.³²  Briefly, balance assessment was conducted using (1) the Balance Error Scoring System (BESS) protocol^12,35  to obtain the subjective clinical error count and (2) instrumented sway metrics. The BESS test was modified to 30 seconds of stance to obtain instrumented measures,^36,37  a requirement for collecting more reliable sway data.³⁸  We only conducted the clinical error count over the initial 20 seconds of each condition. Trials were video recorded (Bloggie Touch; Sony Corporation, Tokyo, Japan), and blinded scoring was completed by a single member of our research team (C.W.S.) who was trained in scoring the BESS. Although participants completed the entire BESS protocol, we used only the double-legged–stance, firm-surface condition for the instrumented analysis because, according to King et al,³⁷  instrumented mediolateral sway captured during this condition offers the greatest sensitivity for classifying individuals under acute concussion timelines.

We assessed gait using an instrumented 2-minute walk test under single- and dual-task conditions. Participants were instructed to walk normally back and forth down an approximately 25-m hallway. In the single-task condition, participants were instructed to walk at their normal comfortable pace. For the dual-task condition, participants were required to read words aloud from recently published sports news articles while concurrently walking at their self-selected pace. We chose this task because it is an ecological task that many people perform daily, slower gait speeds have been found when participants read text messages,³⁹  and walking affects reading speed.⁴⁰  Articles used had a Flesch reading level between 72.4 and 78.2 (ie, fairly easy to read).⁴¹  The words were printed on an A4 sheet in 12-point font and with double-spaced lines. Participants completed a timed, seated baseline reading trial and then the dual-task condition. We used an audio recorder (USB flash drive voice and audio recorder) to record the reading trials and count the number of words read. Each participant received the same articles in the same order for consistency and continued the article from where he or she finished during the previous test. The dual-task cost was calculated using the number of words read during the seated baseline and walking trials. The baseline trials (single-task gait and seated reading) allowed the calculation of DTC-G and DTC-R. The order of procedures was consistent across each testing session, and rest breaks were provided between trials.

We collected data at 128 Hz using Opal sensors and used Mobility Lab Software (version 1; APDM Inc³⁴ ) to export postural sway and gait metrics. The balance measures were the total BESS score and sway as measured by the root mean square of mediolateral sway (mm/s²) of the double-legged stance condition.³⁷  Gait measures were single- and dual-task gait speed. The dual-task costs, which were the negative changes in performance associated with adding a secondary task, were assessed for gait and reading. The Equation illustrates the formula for calculating the dual-task cost:

where DT indicates dual-task performance and ST indicates single-task performance.⁴² 

Between-groups differences for demographic variables were assessed using independent-samples t tests. Any variables that were significantly different were accounted for in further analyses.

Our approach and statistical design have been previously reported.³²  Linear mixed models were fit to each outcome measure to assess whether groups differed over the recovery period. Group was entered as a fixed factor to assess group differences. Time was included as (1) a continuous linear covariate to assess the change in each outcome measure over time and (2) a linear spline function of time, which was referred to as RTP change, to assess the change in each outcome measure after RTP. The RTP change function allowed an inflection to occur at the average RTP time. By including this inflection point, the trend before and after the average RTP day could be modeled, allowing analysis of whether a change in the trend occurred after RTP. The group × time and group × RTP change interactions were included in the model to assess whether the changes over time in each measure were different between groups. Random intercepts and random linear slopes over time were fit for each participant to account for within-subject correlations across time. The model used a full covariance matrix and Cholesky parameterization. Preliminary models included any demographic variables (eg, contact or noncontact sport type) that were significantly different between groups as covariates. If the demographic covariates had effects, they were retained in the final models. Statistical analyses were conducted using MATLAB (version R2017a Statistics and Machine Learning Toolbox; The MathWorks, Natick, MA), and the α level was set at .05.

The average RTP time was 13.7 days (range, 8–26 days) after concussion, with 65% (n = 15) returning between 9 and 15 days postinjury. Days tested and participants lost to follow-up are presented in Table 2. No group differences were found for age, height, or mass, but more people in the concussed group played contact sports (Table 1). Descriptive data for each measure are provided in Table 3.

We found no significant effects for the influence of sport type (contact versus noncontact) in any analysis model (P > .05). Fixed-effects coefficients for models assessing each outcome measure are presented in Table 4 and the Figure.

No significant effects or interactions were found for the BESS total error score.

We observed a group difference for sway, with the concussed group initially swaying more than the control group (positive effect of group, P = .004). We also noted a slight increase in sway in the control group over time (positive effect of time, P = .046). Sway decreased in the concussed group initially (negative group × time interaction, P = .02) but increased after RTP (positive group × RTP change interaction, P = .03).

Single-task gait speed was not different between groups; however, we found an increase in gait speed over time (positive effect of time, P < .001) that was more pronounced in the concussed group (positive group × time interaction, P < .001). At the RTP time point, gait speed stopped increasing and leveled out. This was present in both the control (negative effect of RTP change, P < .001) and concussed groups, with a greater change occurring in the concussed participants (negative group × RTP change interaction, P = .005).

We observed no initial group difference in dual-task gait speed but demonstrated an overall increase in gait speed (positive effect of time, P < .001), which was more prominent in the concussed group (group × time interaction, P = .042). After RTP, dual-task gait speed stopped increasing in both groups (negative effect of RTP change, P < .001; no group × RTP change interaction, P = .08).

No effects or interactions were found for DTC-G (P > .05) or DTC-R (P > .05).

This study involved a longitudinal investigation of standing balance, gait, and dual-task cost after concussion in collegiate athletes. The analysis focused on concussed athletes both before and after RTP. Our main findings were (1) We observed no differences in BESS error count between the concussed group during recovery and the control group over the 8-week period tested. (2) Between-groups differences were detected with an instrumented measure of sway, which also indicated a change in the recovery trajectory after RTP. (3) We found no between-groups differences in single-task or dual-task gait, but a pattern of increasing gait speed that tended to level off corresponding to times after RTP was present. (4) A pattern relating to a possible shift in prioritization between gait and cognition during the recovery period for the concussed athletes was evident. Albeit not statistically different, this pattern could be clinically relevant and warrants further research. Taken together, the findings indicated that a change in objective balance and gait measures may occur in relation to RTP. Although it is unclear whether these changes are good or bad, the results suggest RTP is an important clinical feature that needs to be considered when evaluating a patient's recovery.

The number of BESS errors appeared to decrease as the time since injury increased, suggesting that balance improved in both groups over time. However, analysis of this measure revealed null findings, which is perhaps unsurprising given that our results showed an average difference of only 1 to 2 errors between groups across the testing timeline. We find it notable that more persistent but subtle balance deficits in the concussed group were revealed by the instrumented sway measure, yet no similar findings (effect or interaction) were present for the BESS error count. This error count is known to have limited sensitivity^37,43  and to be influenced by practice effects,⁴⁴  which may be why our longitudinal analysis of this measure was unable to reveal any significant findings.

The concussed group exhibited more sway than the control group during quiet stance with eyes closed, a finding which is consistent with our hypothesis and previous research^12,37  indicating balance deficits in individuals after concussion. The analysis suggested that sway did not improve over the 8-week study period. The lack of improvement in the concussed group would appear to be in agreement with studies^14,15,36  showing more persistent balance deficits after concussion. The interactions suggested a difference in group trends before and after RTP, potentially indicating a regression in the recovery of balance control after RTP in the concussed group. Our findings prompted 2 clinical considerations. First, the current reliance on the BESS during clinical evaluation of concussion should be reevaluated given its insensitivity relative to objective measures of sway. Second, the potential benefit of instrumented procedures for assessing balance in these populations should be widely disseminated.

The lack of group effect for single-task gait speed indicated no differences in walking speed between the concussed and control groups during initial testing. This finding did not support our hypothesis that single-task gait speed would be slower in the concussed than the control group. Although this result contrasts with the results of several researchers^18–23  who reported slower gait speeds in concussed college-aged students and athletes, it is consistent with the reports of other investigators^{10,24,45–47}  who found no differences between concussed and control individuals. The effect of time and the group × time interaction for single-task gait speed suggested gait speed increased over time in the concussed group. Although not directly comparable with the previous findings, an increase in single-task gait speed over time has been documented for recovering concussed adults who initially walked more slowly than control adults.^17,18,21,23  The negative effect of RTP change and group × RTP change interactions suggested a change around the period of RTP, which was more prominent in the concussed group. It is not clear whether this represents a decreased rate of improvement after RTP or perhaps the initial increase in gait speed was an outcome of familiarity with the task.

We expected to see a slower dual-task gait speed in the concussed group during initial postinjury testing; however, no group effect was found. This result was in contrast to several previous studies, in which authors have shown slower dual-task gait speed in recently concussed college-aged students and athletes,^10,19–25  albeit in agreement with other studies.^23,45,46  Similar to single-task gait, dual-task gait speed increased for both groups but occurred at a greater rate in the concussed group. Recovery of dual-task gait speed is consistent with other studies.^21,23,47  Our results suggested a different trend pattern before and after RTP, with dual-task gait speed remaining relatively consistent after the RTP period. However, the lack of group × RTP change interaction suggested this pattern was not different between the control and concussed groups. Therefore, without an understanding of baseline walking speeds in these athletes, it is difficult to know whether this represents true recovery in this group, and, as with single-task gait, it is possible that initial increases in gait speed were driven by task familiarity.

Although not significant within the statistical model, an interesting pattern emerged when we qualitatively analyzed the trends of both DTC-G and DTC-R in the concussed group. A lower cost on gait was paired initially with a greater cost on reading, which shifted after RTP. After RTP, a smaller cost on reading appeared to be concurrent with a larger cost on gait. Inconsistencies in the reporting of these measures (ie, few reports of the actual dual-task cost^21,48,49 ), conflicting findings relating to cognitive performance while walking in concussed adolescent^50,51  but not adult^20,22,46,52  populations, and the lack of effects or interactions within this study for these measures make interpretation of these findings difficult. Nonetheless, given that the dynamic nature of sport participation requires continuous concurrent physical and cognitive attendance, we believe this pattern may warrant further attention and focus in future investigations.

Testing was completed across multiple universities, making our testing environment highly representative of an applied setting. We made every effort to reduce the effect of potential environmental confounders, including keeping the testing location across universities consistent (well-lit, firm-surface hallway) and constant across the timeline. However, the following limitations should be considered when interpreting the results. First, the dual-task paradigm that we used (reading while walking) was different from the tasks used in previous studies, such as question and answer, the auditory Stroop, and mental arithmetic tasks. Our protocol involved both continuous walking and continuous dual tasking, which differs from the discrete trial methods used in many early investigations.^{20,22,50–52}  Indeed, a DTC-G was found consistently across time for both control and concussed athletes when they were required to walk and read simultaneously; however, we observed no group effect. This lack of effect for DTC-G may have resulted from the variability of this measure, which ranged from −23% to −2% for the concussed group and from −17% to 1% for the control group on average across the recovery period. It is plausible that oculomotor function, vocabulary, reading skill, and engagement with the text, as well as other factors, could play a role in performing the reading task and, furthermore, in the level of variability in interference on the concurrent task. In addition, given that word calling (without comprehension) can occur at a more automatic level of information processing,⁵³  it is also possible that the reading task used could not affect cognition in the same manner as tasks used in previous studies that required a more conscious level of control.

Second, longitudinal studies can be limited by participant dropout or participants missing sessions throughout the period assessed. Linear mixed modeling, as we used in this study, can more flexibly manage datasets where the number of observations per participant differ, making it a more robust method of handling longitudinal data. Nonetheless, when interpreting our findings, readers should consider that participants were lost to follow-up.

Third, having baseline balance and gait data for the concussed group would have been highly beneficial for undertaking within-subject case analyses of preinjury versus postinjury in addition to recovery. This would also have allowed the comparison between baseline and RTP values and should be considered by future investigators in this area.

The purpose of our study was to describe the recovery of athletes after concussion while considering changes around the RTP date. Concussed athletes had greater sway but did not walk more slowly than control athletes during single-task or dual-task gait trials. Importantly, although the standard and subjective BESS error scores were not different between groups, the more sensitive inertial sensor measurement of sway appeared to improve initially and then regress without returning to normal during the 8-week study period. We cannot suggest any change occurred due to RTP, but this novel method of analysis provides incentive for a more in-depth research agenda studying the effects of RTP on concussion recovery and any effect this may have on other injuries as well. Furthermore, these results suggested that the RTP date is an important clinical decision that should be accounted for when evaluating recovery trajectories after concussion.

This study was supported by research grant R21HD080398 from the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health (Dr King). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

We thank the athletic departments of the participating universities (Portland State University, Lewis & Clark College, Pacific University, Linfield College, George Fox University, and Concordia University), the athletic trainers who contributed to recruiting participants, and the athletes for their participation.