Concussion, also referred to as mild traumatic brain injury (mTBI), is a physiological disruption in the normal functioning of the brain occurring as a result of a traumatic impact (American Congress of Rehabilitation Medicine [ACRM], 1993; see also Wood et al., 2019). Most often, concussion results from the head, face, neck, or other body part being struck or striking an object, causing rapid acceleration/deceleration of the brain (ACRM, 1993). Sports-related concussions (SRCs) are prevalent among athletes, especially those participating in contact sports (ACRM, 1993; Zuckerman et al., 2015).

SRCs can affect physical, emotional, and cognitive functioning, especially within the first few weeks following the injury (Iverson et al., 2017). Commonly reported symptoms include nausea, blurred vision, headaches, slurred speech, unsteady gait, dizziness, incoordination, and imbalance (ACRM, 1993; Catena et al., 2009; Gagnon et al., 2004). Clinical recovery from concussion refers to returning to normal activities following injury, including a return to normal neuromotor and neurocognitive functioning (Iverson et al., 2017). Although the literature previously suggested that concussion symptoms often resolve within 14 days of injury (Catena et al., 2009; Iverson et al., 2017), it is now recognized that sequelae can persist well beyond 14 days, and critically, even after a patient is no longer reporting symptoms (Catena et al., 2009; Chou et al., 2004; Parker et al., 2006). In fact, in a recent large-scale study (NCAA-DoD CARE Consortium) examining 34,709 athletes from 30 academic institutions, it was determined that a more realistic timeline for recovery is closer to 1-month post-injury (Broglio et al., 2022). Notably, the consequences of returning athletes before they have fully recovered, which would be the case if one relies solely or primarily on self-reported symptoms, are considerable and primarily involve increased risk for re-injury. Specifically, after one TBI, the risk of sustaining another injury is three times greater; after the second injury, the risk increases by a factor of eight (CDC, 1999). Even among elite athletes, the data suggest that sustaining a concussion and prematurely returning to activity increases the rate of subsequent concussions and other orthopedic injuries (e.g., Jildeh et al., 2021; Nyberg et al., 2015). Such findings highlight the importance of accurately assessing concussion and the need for a comprehensive, accurate, quantitative assessment of the sequelae of SRCs.

A recently formulated framework and Consensus Conference statement was introduced based on a meta-analysis that defined concussion subgroups and concussion-associated conditions (Lumba-Brown et al., 2020). In that analysis, one of the most prevalent concussion subtypes was labeled cognitive to reflect the most pronounced symptoms that manifest (Lumba-Brown et al., 2020). Pediatric neuropsychologists have many options (e.g., executive and/or working memory measures) to thoroughly quantify potential neurocognitive deficits. Within that same framework, Lumba-Brown and colleagues (2020) also identified two concussion subtypes that implicate gait and balance: labelled ocular-motor and vestibular. Researchers have also forwarded a rationale for identifying a gait-specific concussion subtype (see Williams et al., 2021). Moreover, in a 2017 consensus statement of the Concussion in Sport Group (CISG), gait and balance were both included as common sequelae of a concussion that merit attention in any concussion evaluation (McCrory et al., 2017). However, presently pediatric neuropsychologists have few validated instruments available to quantify gait.

This paper first provides an overview of the literature highlighting the neuromotor (specifically, gait) consequences of concussion. We also present a rationale for using quantitative assessments of neuromotor functioning that allow for both norm-referenced and self-referenced comparisons as part of a neuropsychological screen following a suspected concussion. Finally, we introduce and illustrate a method of gait analysis using accelerometers, including a presentation of test–retest reliability and recovery data.

Long-term deficits in dynamic motor function, such as postural control and balance, following a concussion are well-documented (Chou et al., 2004; Gagnon et al., 2004; Parker et al., 2006) and are thought to contribute to cases of recurrent concussions (Catena et al., 2009). For example, children aged 7–16 years old who sustained mTBI showed post-injury balance deficits for up to 12 weeks and performed significantly worse 12 weeks post-injury compared to non-injured children (medium effect size, d = 0.51) on the Bruininks-Oseretsky Test of Motor Proficiency (BOTMP), which measures static and dynamic balance without external perturbations (Gagnon et al., 2004). This study also demonstrated an apparent recovery effect in motor proficiency, as the mTBI group performed better at week 12 than they had when measured 1-week post-injury, with this change represented a medium to large effect size (d = 0.77). Furthermore, college-aged adults who sustained a concussion demonstrated decreased dynamic balance compared to controls during a divided attention walking task administered at 1-month post-injury (Parker et al., 2006). The effect was large for gait stability, especially when attention was divided from the primary motor task (i.e., walking). The motoric consequences of concussion on gait were also shown at 28 days post-concussion, as those with mTBI tended to walk slower (d = 0.18) and had less separation between their whole-body center of mass (COM) and center of pressure (COP) than controls (d = 0.28), reflecting differences in gait speed and stability between those with mTBI and controls (Parker et al., 2006). Adults with mild-to-severe TBI showed significantly slower walking speed (gait velocity), stride length, and increased mediolateral motion during walking compared to controls (Chou et al., 2004). An increased mediolateral motion was also found in a study of high school students who had sustained a concussion and were tested at five-time intervals ranging from 72 h to 2 months post-injury (Howell et al., 2014). The observed effect for increased mediolateral motion was largest immediately after the injury, but the effects persisted even after 2 months (Howell et al., 2014).

A meta-analysis (Fino et al., 2018) examining gait consequences across the lifespan (children, adolescents, and adults) found acute effects for stride width, but less consistent findings at other post-incident timeframes. In general, long-term effects were more commonly found with complex gait tasks (e.g., dual attention, obstacle). However, these findings may depend on extraneous factors such as small sample sizes that limit statistical power.

Overall, gait changes associated with TBI appear to extend well beyond the typical 2-week return-to-play window. This suggests that gait data may have a different (potentially longer) recovery trajectory, or gait variables may be more sensitive to ongoing concussion-related problems as compared to other (e.g., cognitive) domains during the latter stages of recovery (Parker et al., 2006). Moreover, the presence of gait alterations is consistent with patients’ subjective reports of “instability,” and subtle but neurologically relevant gait instabilities may not be observable with routine clinical (qualitative) gait examinations (Chou et al., 2004).

Advanced instrumentation to record gait functioning significantly improves the understanding of gait (Tao et al., 2012). Gait speed assessed using accelerometers has superior sensitivity to dysfunction relative to manually collected gait speed data, such as timing a walk over a fixed distance (Maggio et al., 2016). Moreover, accelerometers can produce reliable and objective outputs across a range of clinical populations (e.g., Byun et al., 2016; Fujiwara et al., 2020; Henriksen et al., 2004; Kluge et al., 2017; Kobsar et al., 2016; Moore et al., 2017; Werner et al., 2020).

The literature also includes several studies using accelerometers/sensors to assess various aspects of gait within the context of SRCs/mTBIs. For example, the gait features of those with mTBI were compared to matched controls using inertial sensors (accelerometers), with the mTBI group having a slower gait pace (d = 0.91) and slower turning (d = 0.82), despite the assessment occurring at an average of 1-year post-incident for the mTBI group (Martini et al., 2021). This research also found sensor-derived gait variability to relate significantly to scores on the Neurobehavioral Symptom Inventory. Parrington and colleagues (2019) used inertial sensors to show that a concussed group of collegiate athletes initially demonstrated more significant gait sway than a control group, as well as a pattern of increasing gait speed as recovery progressed that leveled off after the athlete returned to play. Finally, a 2019 meta-analysis of 22 studies with adult participants with at least one concussion found that sensor-assessed gait velocity significantly relates to neurocognitive status, with the concussed participants walking 0.12 m/sec slower than controls (Wood et al., 2019). Even after 28 days, previously concussed participants walked significantly slower than controls, with this effect declining over time (Wood et al., 2019).

Overall, the literature suggests that accelerometers can capture essential aspects of the gait cycle, and quantifying the data facilitates its relation to concussion outcomes. However, it is unknown whether such data can be collected with minimal or no equipment/instrumentation while maintaining reliability and validity with respect to SRCs, as this would make such assessments useful for broad adoption in clinical practice.

We here describe the use of noninvasive, continuous tri-axial accelerometers to visually depict gait as a waveform, referred to here as a BioKinetoGraph (BKG). Analogous to the 12-lead electrocardiogram, BKG waveforms are related to specific neuromotor gait cycle events, including cadence, foot strike, push-off, double stance time, and swing phase. Raw data are combined to generate BKG waveforms as gravitational accelerations over time. Tracings are used to obtain the range of motion, amplitudes, and timing intervals for the various components of the waveform that relate directly to the gait cycle. The adopted method for collecting and representing gait data is similar to other published studies (e.g., Godfrey et al., 2015).

The BKG and the algorithms used to extract the various gait features were initially validated using a recorded video at 240 frames per second, which is twice the sampling rate of the accelerometers. This recording was completed in a test session with four accelerometers (affixed to both ankles, one wrist, and the sacrum) that matched the recorded data and key markers identified by the algorithm (heel strike and toe-off) to what was observed in the video. The camera was an iPhone X configured to record at 240fps while attached to a rolling dolly system that followed the participant as they walked in order to keep them in the center frame. The high-speed video was used to identify the start (heel strike) and end (toe-off) features of a gait cycle and match them to the accelerometer data. Subsequent gait events were identified on the waveform within these boundary events.

Although gait data can be collected from the sensors affixed to different parts of the body, the bulk of the published research and our own work is based on the data derived from a dedicated sensor affixed with a belt at the sacrum. This sensor records inertial measurement units (IMU; a general term for accelerometers, gyroscopes, and related technology) via transmitted signals at a rate of 200 Hz to a local device, which then transmits the data to remote servers for BKG motion analysis following each trial. Raw data are saved to a persistent data store as a collection of timestamps and accelerometer readings of each axis. Raw inertial data are then processed using algorithms that detect critical markers of the gait cycle and extract the spatial, temporal, kinetic, and spectral features reported below.

The above-described BKG output can generate hundreds of variables, but only a subset of those variables will be described here. Specifically, 18 variables organized into four conceptual domains: balance (stability), stride (timing intervals), power (amplitudes), and symmetry (regularity and consistency) are discussed, in part because these variables correspond to similar variables in the gait literature that were recently validated (Lecci et al., 2023).

Balance captures stability (and was previously labelled stability) during the walk, and is defined as the range of motion (ROM) at the center of mass (sacrum) and gait cycle variability, with data coming from both the straight and turnaround portions of a walking trial. This domain includes the stability during the straight portion of the gait cycle (side balance), the stability during the turnaround (turnaround balance), the timing of side-to-side movement (sway time), the force generated with lateral movement (side power), the rhythmicity of gait patterns (gait smoothness), and the variability of time in double stance phase (support consistency). All balance values are inverted such that low scores indicate problematic functioning.

Stride captures the time it takes to complete different parts of the gait cycle and is therefore a proxy for speed. Specifically, the stride domain refers to maintaining velocity during walking and includes the average timing interval between contralateral heel strikes (stride time), the average time spent in the double stance phase when both feet are in contact with the ground (double stance), the average time spent in right and left side stances (stance phase), and the average duration of the swing phase (swing phase). (Note: Stance phase and swing phase were added to the stride domain in the current mobile assessment, after the BKG’s initial validation using the sacrum sensor.) All stride values are inverted such that low scores denote problematic functioning.

Symmetry is a measure of consistency in one's gait and refers to the uniformity and regularity of movement across the anterior–posterior and vertical dimensions. This domain includes a comparison of the forward and backward displacement from the center of mass (forward movement symmetry), a comparison of the distance of upward and downward displacement from the center of mass (vertical movement symmetry), and a comparison of the velocity of upward and downward movement from the center of mass (vertical sway symmetry). Low scores denote problematic functioning.

Power is reflected in the amplitude of one’s gait and is defined in terms of the amount and efficiency of energy generated and expended in the gait cycle. This domain includes the total net force generated at the center of mass (total power), the force generated with vertical movement (vertical power), the force produced with forward movement (forward power), the force generated by the foot strike (striking force), and force generated by the toe pushing off (pushing force). Low scores denote problematic functioning.

These four domains and their corresponding BKG variables are automatically generated by the BKG mobile assessment and produce output as T-scores (mean 50, SD = 10) based on normative data. Research has also established that the BKG can produce consistent scores over time. Specifically, a study of 60 participants aged 18 to 35 (61.7% female) with no apparent health problems established the test–retest reliability of the BKG variables based on an accelerometer attached to the sacrum. The resulting correlations between test scores taken on two separate days (with a mean retest interval of 4 days) ranged from 0.72 to 0.91, with a mean of 0.80 (Lecci et al., 2023). This suggests that BKG variables can produce consistent (reliable) output using a dedicated sacrum sensor.

This same sacrum sensor was used in a sample of 1,008 individuals (53.4% female) aged 8 to 50 years (M = 16.98, SD = 4.43) to validate the gait assessment. Specifically, 950 ostensibly healthy individuals completed the BKG as part of a standard baseline evaluation, and 58 individuals completed the BKG while undergoing post-concussion evaluations. The findings indicated that the BKG variables grouped within the above-noted conceptual domains are highly correlated with NIH 4-m gait scores (multiple R = 0.51, p < 0.001), with robust associations for power and stride (i.e., less power and lower stride scores are associated with slower walking speed) (Lecci et al., 2023). This illustrates convergent validity between the BKG data and a well-validated measure of gait speed (i.e., the time it takes to traverse 4 m is essentially a measure of speed). Although not reported in the original study, this same data set shows that the BKG variables can likewise significantly predict Balance Error Scoring System (BESS) total scores (multiple R = 0.28, p < 0.001), though in this instance the balance and symmetry variables take on a more prominent predictive role.

Most relevant to the current research, BKG data have been shown to predict concussion-related outcomes. For example, in a sample of 111 cases, BKG scores (when combined with other data from the SportGait platform) predict the remove-from-play decisions of a pediatric neurologist, achieving a classification accuracy of 91% and AUC of 1.0 when using a machine learning general linear model (Keith et al., 2019). In addition, the BKG variables predict the endorsement of CDC concussion symptoms significantly, and out-predicted by upwards of four-fold the separate and combined effects of two frequently used and well-validated measures of gait and balance (NIH 4-m gait and BESS scores) when predicting concussion symptoms (Lecci et al., 2023). Moreover, BKG variables significantly predict CDC concussion symptom endorsement over and above the NIH 4-m gait and BESS measures (Lecci et al., 2023). Thus, although gait speed and balance measures are significant predictors of concussion outcomes, sensor-based assessments of motion that quantify many gait variables in addition to speed and balance, provide a more robust assessment of gait and concussion-related sequelae, which is in keeping with the literature on mTBI (Dever et al., 2022). This initial research validated the BKG methodology and technology using a sacrum sensor.

Next, we present data on the test–retest reliability of BKG mobile (smartphone) assessment, as establishing psychometric properties in the mobile environment is critical to broadening the accessibility and use of accelerometers in quantifying gait. Moreover, test–retest reliability is crucial for interpreting data collected from repeat testing, as repeat testing and comparisons to previous data points typically occur during recovery.

A sample of 4150 individuals (35% female) ranging in age from 5 to 78 (M = 16.91, SD = 6.12) was selected from baseline testing completed between December 14, 2020 to January 9, 2022 using the SportGait Mobile App. The SportGait App includes a cognitive assessment, symptom rating, and a measure of affect, along with the BKG. For the purpose of this paper, only the BKG gait assessment data are presented. The sample was drawn from an original data set containing 4797 consecutive evaluations not previously published. (Note: Date are available from the first author.) This research was approved by the University of North Carolina Wilmington IRB; Protocol #21–0047.

Evaluations were removed if they were not baseline assessments (a total of 129 or 3%), if they were duplicates (456 or 11%), or if they failed to follow the directions (a total of 62 people or 1.5%). Some of the problems following directions included having a difference of 5 or more steps in the number of steps before and after the turnaround, or the mobile sensors on the phone detecting an improper positioning of the phone. Of the included data, extreme outliers were winsorized (i.e., capped at ± 3 SDs) for two variables, with the total number of affected data points accounting for less than 1% of the data for those variables.

Participants were instructed to download and open the SportGait App to their phones. Both iPhones (running iOS) and Androids of various generations were used. However, having different operating systems for iPhones and Androids is not inherently problematic, as the App works the same regardless of the operating system and the same algorithms are applied for all users. Although there were some cosmetic changes to the App during the data collection period, the changes primarily focus on appearance and not functionality. The oldest operating systems (6.0.1) were detected on Android devices, with a very small number dating back to 2013. The oldest Android phone was Nexus 5 which launched in 2013 with an OS version 6.0.1. The oldest operating system on an iPhone was 11.3, and the oldest iPhone was an iPhone 6 (released Sept. 2014), using iOS 12.4.1. The majority of our data come from iOS users (N = 3733, 90% of the sample), with the remaining 10% (N = 417) using Android OS. Although we do normalize the incoming data, which are subtly different from these different operating systems, past research indicates no substantive differences based on the version of the phones or operating systems (see Freund, 2021 for data illustrating comparability of different phones with respect to the cognitive task within the same App). Of course, the phone sensors can be damaged, and under those circumstances, the data can be less optimal.

The BKG gait assessment portion of the App involves three unobstructed walks of approximately 20 feet each, with participants instructed to walk at their usual pace (speed). Before beginning the BKG, participants are asked to find a location with approximately 20 continuous feet of unobstructed walking space (e.g., open room, hallway) with a firm surface (concrete, tile, hardwood floor, low carpet). Participants are also instructed to walk either barefoot or in socks. These instructions were included because previous research has shown that footwear and the walking surface can influence gait variables, at least to some degree (Lecci et al., 2023).

At the beginning of each walk, the App instructs participants to hit the start button on the screen, then hold the phone to their chest, with the phone oriented horizontally and the screen facing their body, with both hands over it. They then walk ten paces, turn around and return to the start point before hitting the stop button. After the walk, the SportGait App notifies the participant if the data were successfully collected. The same procedure was used for the second and third walks. Data extracted from the second and third walks were compared to evaluate the test–retest reliability figures, and this corresponds to the method adopted in previous research using an affixed sensor (Lecci et al., 2023). The App provides a visual illustration of the procedure, depicted in Fig. 1.

In a previous data collection, participants were asked to walk six times to identify the optimal number of walks needed to adequately sample gait biomechanics with the BKG. The first and second walks resulted in the largest differences in the obtained values, with differences between subsequent walks decreasing markedly. Thus, the BKG task instructions now require three walks, with the mean of the second and third walks serving as the primary data of interest. (Note: Future research will explore the first walk as a potential “separate” assessment involving gait under the condition of a somewhat novel task with dual attention components, due to the participant counting out the target distance.)

The rationale for focusing on two walks in such close temporal proximity is that although these represent two separate walks, variables that can impact gait (e.g., footwear, walking surface, minor idiosyncrasies in how the phone is held, the type of phone used), but are irrelevant to assessing the underlying biomechanics of gait, would be identical. Moreover, factors relevant to gait biomechanics and captured by the BKG, such as the presence of orthopedic and head injuries, would likewise be equivalent in that short time interval. Thus, an assessment of the consistency in scores across the two trials would make for an optimal evaluation of test–retest reliability for the BKG.

Before analyzing the initial (heel strike) and final (toe-off) foot–ground contact events of each gait cycle from the Mobile BKG, the signal is processed to remove extraneous data recorded before and after the gait trial, adjust for tilt and orientation of the sensor relative to gravity, and filtered to reduce noise (see Lecci et al., 2023 for a more detailed description). The raw signal data is then fourth-order-zero-phase-shift filtered (Winter, 2009), with upper and lower cutoffs of 3 Hz and 0.111 Hz, respectively. Initial and final contact events were detected algorithmically by first identifying local maxima representative of the mid-swing with toe-off and heel strike events indicated by minima occurring immediately before and after, respectively. Displacement and velocity were obtained for each axis by trapezoidal integration and subtraction of the zero-phase rolling average (Oppenheim and Schafer, 1989) equal to one gait cycle.

BKG variables from the second and third walks were correlated and are presented in Table 1. Pearson correlations for the 18 Mobile BKG variables ranged from r = 0.51 to 0.92, with an overall mean of 0.79. The four domains resulted in average correlations of 0.79 for the five power variables, 0.88 for the four stride variables, 0.76 for the six balance variables, and 0.73 for the three symmetry variables. Thus, each domain and the overall mean reflect strong test–retest reliability, and the 18 BKG variables are at least adequate when considered individually. Moreover, the obtained values are commensurate with the test–retest figures obtained in monitored, controlled environments using a static sensor attached at the sacrum (Lecci et al.,