Methodological Critique of Concussive and Non-Concussive Dual Task Walking Assessments: A Scoping Review

Objective: To understand the methodological approaches taken by various research groups and determine the kinematic variables that could consistently and reliably differentiate between concussed and non-concussed individuals. Methods: MEDLINE via PubMed, CINAHL Complete via EBSCO, EBSCOhost, SPORTDiscus, and Scopus were searched from inception until 31 December 2021, using key terms related to concussion, mild traumatic brain injury, gait, cognition and dual task. Studies that reported spatiotemporal kinematic outcomes were included. Data were extracted using a customised spreadsheet, including detailed information on participant characteristics, assessment protocols, equipment used, and outcomes. Results: Twenty-three studies involving 1030 participants met the inclusion criteria. Ten outcome measures were reported across these articles. Some metrics such as gait velocity and stride length may be promising but are limited by the status of the current research; the majority of the reported variables were not sensitive enough across technologies to consistently differentiate between concussed and non-concussed individuals. Understanding variable sensitivity was made more difficult given the absence of any reporting of reliability of the protocols and variables in the respective studies. Conclusion: Given the current status of the literature and the methodologies reviewed, there would seem little consensus on which gait parameters are best to determine return to play readiness after concussion. There is potential in this area for such technologies and protocols to be utilised as a tool for identifying and monitoring concussion; however, improving understanding of the variability and validity of technologies and protocols underpins the suggested directions of future research. Inertial measurement units appear to be the most promising technology in this aspect and should guide the focus of future research. Impact: Results of this study may have an impact on what technology is chosen and may be utilised to assist with concussion diagnosis and return to play protocols.


Introduction
Concussions are a mild traumatic brain injury (mTBI) that individuals can experience through sport that are frequently missed or underestimated resulting in individuals returning to sport earlier than they should, increasing the risk of sustaining a musculoskeletal injury [1,2] or leading to further brain damage if a second concussion is experienced in close proximity to the first event [3,4]. There is a need to have protocols that can assess the extent of the concussion experienced while also determining readiness for return to activity. Typical methods of assessing concussions are clinical assessments which consider physical and mental attributes such as balance and memory, respectively [5]. These assessments are generally tested as two separate elements, yet researchers have suggested that a dual task (DT) assessment that combines physical and mental testing provides a more accurate understanding of concussion than standalone walking and cognitive assessments [6][7][8][9][10]. Protocols incorporating a cognitive task alongside a gait assessment are becoming frequently utilised to evaluate the effects of concussion [11][12][13]. The Stroop test, which involves participants responding to an auditory or visual cue whilst undergoing locomotion, requires equipment to facilitate the test and record the accuracy and speed of responses [6,[13][14][15][16][17][18][19][20]. Other more common cognitive dual tasks include reciting the months of the year in reverse order, subtracting by sixes or sevens from a given number, or spelling common five letter words in reverse, while walking along a level walkway [7,11,[21][22][23][24][25][26][27][28][29][30][31][32]. Gait variables such as walking speed, stride length and cadence are quantified to determine any variability through the introduction of a cognitive task [11,12,23,32,33].
To measure the physical variables, 3D motion capture (3D MOCAP) and/or force plates are commonly used to differentiate between concussive diagnoses by assessing postural balance and control [6,7,9,11,[14][15][16][21][22][23][24][25][26][27][28]30,34]. Inertial measurement units [8,[31][32][33]35,36] and accelerometers [18] are other forms of technology that have been used for DT concussion gait analysis. However, whether certain technologies and/or certain variables are better suited to discriminating between concussed and non-concussed diagnoses is unknown. Of particular interest to the authors and that which provides the purpose of this scoping review was understanding the methodological approaches taken by various research groups and determining those variables that could consistently and reliably differentiate between concussed and non-concussed individuals.

Data Sources and Searches
A scoping review was conducted guided by the standards presented by the Preferred Reporting Item for Systematic Reviews and Meta-Analysis extension for Scoping Reviews (PRISMA-ScR) [37]. This review aimed to examine the methodological approaches, determine those variables that could consistently and reliably differentiate between concussed and non-concussed individuals, identify limitations of current technologies and related protocols and, finally, to outline areas of future research for concussion assessment. MED-LINE via PubMed, CINAHL Complete via EBSCO, EBSCOhost, SPORTDiscus and Scopus databases were searched for relevant articles from the inception of the databases until 31 December 2021. The search strategy included five concepts (concussion, mTBI, cognition, gait and dual task) and a combination of key words to adapt to each database. From the initial screening, 70 articles were identified, and titles and abstracts were screened to determine relevance to review. Forty-four full-text articles were examined to determine inclusion eligibility. Reference lists of included articles were searched for other potentially relevant information. A total of 23 articles were identified as being eligible for full-text review and subsequent analysis ( Figure 1).

Study Selection
Studies were included if a steady state DT walking assessment was used; the DT involved a cognitive task paired with a steady state gait task; individuals had a concussion, either through sport or other activity, or mild traumatic brain injury with healthy individuals used as control subjects; and kinematic walking measures were reported. There were no restrictions placed on the age or gender of participants. Articles were excluded if steadystate gait was not the primary dependent variable of cognitive task performance, such as reaction time, tandem gait, balance or gait termination time. Review articles and case studies were excluded. Full-text articles were retrieved and scanned when inclusion could not be determined by screening titles and abstracts. Articles that involved all healthy participants or those with a more severe brain injury were excluded from the analysis. Risk of bias was mitigated in this research, given the focus was more a technological/methodological critique rather than a review of the outcome measures and findings as such.

Data Extraction
One reviewer (C.M.) extracted the data using a customised designed standardised Excel database (version 2201, Microsoft, Redmond, WA, USA) which was validated by a secondary reviewer (J.C.). General study information (i.e., author, year), subject characteristics (i.e., sample size, age, concussion history, sport/activity), type of study (i.e., cross-sectional and prospective), methods of assessment (i.e., testing equipment, environment, protocol) and primary outcome measures (e.g., means and standard deviations of average gait velocity) were extracted. Descriptive information relating to the sport and performance level were used to categorise each of the participants. A wide array of definitions for elite, sub-elite and novice athletes exist (30). Therefore, in order to clearly differentiate between groups with concussions, skill level was grouped according to the level at which participants were competing. National or regional representatives were classified as elite athletes. Participants competing at university or collegiate (Uni/Col) were categorised as sub-elite. Recreational athletes were deemed as such (Rec) and those who did not experience a sport-related concussion were classified as NoSport. Adolescent athletes were categorised as high school (HS) athletes.

Role of the Funding Source
This scoping review was funded by Movement Solutions. The funder played no role in the design, collation, synthesis and writing of this review.

Participants
From the 1030 participants, 54% were male and 49% were female; the sex of the participants was not reported in one study [25]. There was insufficient detail provided in each of the studies to determine if there was any overlap of sample groups. The age of participants most commonly ranged between 13 and 22, often being high school and university students and athletes. One study involved adults over 64 years of age [39]. The majority of concussions experienced by participants were sport-related, with the sporting level ranging from elite athletes, intercollegiate athletes, and high school athletes to local and recreational athletes. Three studies included subjects who had sustained a concussion through activities of daily living [24,25,38]. Subjects experienced concussion through a variety of sports with the most common sports reported being football, American football and ice hockey (n = 2) [31,33].
Verbal fluency (naming as many words beginning by a given letter) -Arithmetic (counting backward by two from a given number).
Four baseline trials initially (level walking no cognitive task); then every combination of conditions repeated four times. Elements of cognitive tasks were changed across trials. ST: barefoot walking at a self-selected pace along a walkway. DT: auditory Stroop task identifying the pitch of words "high" or "low" spoken in either a high or low pitch. Eight to ten trials completed at each session. Tested at points within 72 h post-concussion, then again at 1 week, 2 weeks, 1 month and 2 months post-concussion. NC tested at similar periods.
Inertial measurement units (IMU) were utilised by five research groups [31][32][33]36,40]. IMUs were placed on the lumbosacral junction and dorsum of each foot (n = 4) and recorded data at a sampling rate of 128 Hz [31][32][33]40]. One article placed IMUs on the dorsum of each foot, forehead, lumbar spine and sternum, with the sampling rate not being specified [36]. A single research group utilised an accelerometer in combination with 3D MOCAP [18]. The accelerometer was attached at the L5 vertebrae and collected data at a sampling rate of 128 Hz.
Three articles used a microphone to record participants' responses during their respective DT [6,7,15]. A GAITRite walkway, sampling at 80 Hz, was used to collect gait data in one article [12]. One article utilised a manual stopwatch to time participants' gait [41].
Regarding cadence and double support, there was a paucity of researchers investigating the sensitivity of these measures over time, with double supporting having largely been discussed with historic concussion subjects only (n = 4) [12,33,36,39]. Two out of four articles which included double support analysis found a significant increase in double support duration for historically concussed individuals [12,36].

Reliability
None of the studies reviewed established the reliability of the specific protocols they implemented. Four of the articles reviewed referred to reliability of the equipment and protocols established in other studies (Table 4). On reviewing these studies, two research groups investigated the reliability of GAITRite walkway variables, which only one reviewed article used [12]. Montero-Odasso et al. [42] considered gait velocity, step length, stride length, step time, stride time and double support time in single and dual task walking with a cognitively impaired elderly population (average age 76.6 ± 7.3 y). The absolute consistency (coefficient of variation (CV)) ranged from 6.36-18.28% for ST and 11.02-19.27% for DT. In terms of relative consistency, intraclass correlation coefficient (ICC) ranged from 0.80-0.97 for ST and 0.93-0.97 for DT. The GAITRite walkway was also investigated by Paterson et al. [43], however, the comparison was between younger (20.08 ± 0.7 y) and older (67.93 ± 7.8 y) populations. CVs ranged from 2.33-4.08 %. In terms of relative consistency, ICCs ranged from 0.66-0.94. Step  Step velocity ICC: 0.534 Step length ICC: 0.419 Step The GAITRite walkway was also utilised in conjunction with inertial sensors to establish reliability of other technologies using continuous walking protocols. Moore et al. [44] sought to establish the reliability of a wearable accelerometer (AX3) with stroke patients. Within the variables of step velocity, step length, step time, and stance time, the absolute agreement was good (ICC: 0.744-0.797) between AX3 and GAITRite, and moderateexcellent (ICC: 0.831-0.923) between AX3 and Opal inertial sensors. Morris et al. [45] compared GAITRite with Opal inertial sensor data analysed via Mobility Lab across young adults, older adults and adults with Parkinson's disease. Gait velocity, stride length, cadence and stride time had moderate-excellent absolute agreement (ICC: 0.741-0.998); however, double support time had poor absolute agreement (ICC: 0.213-0.716).
To establish reliability of cognitive tasks while walking, Howell et al. [30] used IMUs to investigate the ST and DT gait of collegiate athletes in both contact and non-contact sports (19.2 ± 1 y) through gait speed, cadence and stride length. This research group only reported relative consistency: the ICCs ranged from 0.68-0.80 for ST and 0.73-0.85 for DT walking.

Discussion
Concussions are an increasingly common mild traumatic brain injury that are experienced in sport. To limit misdiagnosis of individuals with concussion and to assist with return to play, there is a need for assessment protocols that incorporate both cognitive and physical elements to allow for a more accurate evaluation of concussive impairment. Assessing gait whilst performing a cognitive task is one such assessment protocol and formed the focus of this review. Of particular interest were the methodological approaches taken by various research groups and determining those protocols and/or variables that could consistently differentiate between concussed and non-concussed individuals.
The participants involved across the reviewed articles were diverse in sample size (12-122), age (12-68 y), sport (football, cheerleading, horseback riding, to name a few) and competition level (recreational-elite). Sixty-one percent of the reviewed study protocols required participants to partake barefoot, which presents an interesting issue in terms of whether testing should take place with shoes or barefoot, which potentially may affect the clinical outcomes. Counting or spelling backwards seemed to be the easiest of dual tasks to implement given the ease of administration and lack of equipment required, negating the need for extensive set up time. It is suggested that these cognitive tests should be randomised to limit any learning effects.
The most widespread use of equipment involved 3D MOCAP and force plates. While the equipment may be considered to provide more precise information, the cost of the equipment and the expertise required to run, process and analyse the data is a restrictive factor for assessing concussions outside of conducting research. A significant time cost is also involved with processing the information recorded from MOCAP and force plates to generate data for analysis. Equipment that does not require as extensive proficiency or time to process and analyse collected data, such as with inertial sensor technology, may offer a more accessible tool for practitioners in diagnosing and monitoring concussion.
The most common distance that participants were assessed over with dual task gait was 8-10 m. This was largely a result of the space in which the testing was conducted and the available equipment e.g., 2-3 force plates in series and/or in ground with 3D MOCAP. The authors feel that the set-up of such equipment is a limitation, in that testing is restricted to a particular environment (i.e., sports laboratory) which may impede the initial diagnosis and subsequent monitoring of concussed individuals, thus, being detrimental for quicker return to play. More portable technologies (i.e., IMUs) may provide a more accessible and convenient tool that can be utilised within a wide range of environments. If dual task gait analysis of concussive diagnosis is to have any real-world utility, then serious consideration of other technological approaches will be needed.
None of the 3D MOCAP and force plate outcome measures reported were found to be sensitive enough to consistently determine differences between concussed and nonconcussed diagnosis during DT walking. Gait velocity, stride/step length and stride/step width were the variables that were most reported on, with significant differences being reported by 31% [11,18,22,23,38] and 25% [11,21] of the reviewed articles for gait velocity and stride/step length, respectively, but no article was found to report significant differences in stride/step width. The majority of articles found no significant differences across the gait variables of interest. Comparatively, articles that utilised IMUs to measure gait velocity and stride/step length reported significant differences in 100% [31][32][33]36,40] and 80% [32,33,36,40] of the articles, respectively. This may indicate that IMU utilisation enables increased accuracy and/or sensitivity due to a closer interaction with the gait movement patterns. It also needs to be noted that the diagnostic value of any gait analysis is enhanced when data is collected over multiple testing occasions. This historic data provides a better insight into any aberrations that may need addressing.
The emergence of inertial sensor technology [31][32][33]36,40] might provide a viable alternative to MOCAP and force plate analyses. The outcome variables reported by the articles that utilised IMUs showed promising consistency in differentiating between concussed and non-concussed diagnoses. It would be interesting to understand whether different sensor placements (e.g., in-sole sensors) offer added sensitivity and accuracy, compared to the sensor arrangements in the bulk of the studies reviewed (lumbosacral, dorsum and foot).
One of the most concerning aspects of all the articles reviewed was the absence of any reporting of the reliability of the outcome measures of interest. Understanding the "noise" or unexplained variability associated with a measure is fundamental to interpreting findings. Only one research group [12] provided evidence regarding the reliability of the GAITRite walkway in elderly and young cohorts, citing the work of Paterson et al. (2008) [43] and Montero-Odasso et al. (2009) [42]. The results were markedly different in that Paterson et al.'s [43] findings were acceptable (CV < 4.08%; ICCs 0.66-0.94), whereas the absolute consistency of Montero-Odasso et al. [42] was not (DT CVs 11.02-19.27%; ICCs 0.93-0.97). This could be attributed to the age of the participants in the latter study. Nonetheless, it needs to be noted that only Martini et al. [12] used the GAITRite walkway as a method of measuring DT gait variables and, therefore, it is problematic to make generalisations to other methodological approaches.

Conclusions
However, whether certain technologies and/or variables are better suited in discriminating between concussed and non-concussed diagnoses is unknown.
Of particular interest to the authors was understanding the methodological approaches taken by various research groups and determining those variables that could consistently and reliably differentiate between concussed and non-concussed individuals. In terms of the first foci, MOCAP and force plates were the dominant technologies used to quantify concussed and non-concussed gait. From the literature reviewed, it would seem that none of the gait parameters assessed using MOCAP and force plates used to quantify concussed and non-concussed gait impairments were consistently sensitive enough to determine significant differences between groups, particularly over various time periods/testing occasions. This may mean two things: (1) DT walking is not sufficiently sensitive enough as an assessment to determine concussive diagnosis consistently; or (2) the protocols/technologies that are being used need refining or replacing to enable better concussion detection. For example, it would be interesting to determine if longer distances/large fields of capture enabled better precision of measurement.
With regards to the consistency and reliability of data, there seems to be little attention in the research reviewed on the variability of the measures utilised to quantify gait characteristics. Fundamental to research going forwards, especially with new and innovative technology, is establishing the reliability and smallest worthwhile changes in gait parameters.
Inertial sensor technology has been used in a few studies to date with some promising results around average gait speed and stride length. However, as with the other technologies reviewed, the reliability has not been documented and there may be better placement of sensors than the lumbar and dorsum but researchers have provided a starting point for ongoing investigation. For example, it would be interesting to determine if inertial sensors that quantify the foot-ground interaction (e.g., inner sole sensors) offer any diagnostic benefits in this area.
Finally, the cost of MOCAP and force plates and the expertise required to run, process and analyse the data is a restrictive factor for assessing concussions outside of conducting research. It is believed that the advent of technological "solutions" such as inertial sensors may enable dual task testing outside of the laboratory given the portability of such devices. If the technology is found to be valid, reliable, accurate and sensitive to changes in gait characteristics, they may provide a viable assessment option that could result in higher utility of dual task walking assessments in the diagnosis of concussion. Data Availability Statement: The data in this study is openly available from the journals as listed in the reference section.