Instrumented Analysis of Spatial Temporal Gait Variability as a Marker of Falls Risk to Assist Clinical Practice: A Brief Review

Spatial temporal gait variability has developed into a measure of interest in clinical gait analysis. It is capable of providing unique insight into rhythmic stability of human gait and may be a sensitive biomarker of falls risk. Several lines of evidence support the use of spatial temporal gait variability as a clinical measures. The purpose of this review is to provide a brief, practical review, of spatial temporal gait variability. This review discusses how gait variability data is obtained, examines previous studies reporting gait variability as a marker of falls in a range of clinical populations and identifi es approaches to implement this measure into clinical practice. In summary, it is suggested that spatial temporal gait variability is a sensitive measure of gait function that can assist clinical practice and delivery of therapy services. Review Article


Introduction
Regaining independent mobility is a common rehabilitation goal for many clinical populations including stroke survivors, amputees and patients recovering from various orthopedic surgeries such as total knee arthroplasty. Sensitive measures which quantify characteristics of gait are therefore of signifi cant value for clinicians. Analysis of gait can assist clinical practice by identifying functional impairments which can be used as therapeutic targets and inform selection of appropriate rehabilitation strategies to optimize recovery of gait function. Measures of gait function can be categorized as spatial temporal (e.g. step time, step length), kinematic (the geometry of motion), kinetic (measures of force, joint moments) or energy expenditure (metabolic cost of walking).
Utilizing one, or a combination of these measures will provide valuable information on gait characteristics which is likely to assist clinical practice.
The performance of gait involves coordinated action of muscles acting at the hip, knee and ankle joints to move the centre of gravity forward in a controlled manner. It is produced by the locomotor system which integrates inputs from the primary motor cortex (M1), cerebellum and basal ganglia, is acceptable, and may in fact be benefi cial for movement in space. For example, the ability to vary gait patterns may allow for stepping over obstacles or slight changes in direction to avoid hazards [7][8][9]. However, in stable testing environments, gait patterns should be relatively consistent. It is thought that a more variable gait pattern may cause the centre of pressure to move over or beyond the base of support in a relatively uncontrolled and unstable fashion which may predispose an individual to experience a fall [10]. Falls assessment is an important focus for many health services. Intrinsic risk factors for falls include age, chronic disease, gait and balance instability, decreased vision, altered mental status and medication use [11]. Many, if not all, of these factors would likely be prevalent in many patients groups that utilize acute hospital services, rehabilitation facilities, community services and aged care facilities. Therefore, there is a strong clinical requirement for sensitive tools to assess falls risk and spatial temporal gait variability may be an appropriate biomarker.
Sensitive measures to predict falls risk or discriminate falls history have substantial value given the negative consequences and signifi cant cost on the health care system associated with falls [12,13]. Experiencing a fall has been associated with functional limitations, institutionalization, decline in independence, reduced confi dence and self-imposed restriction of activity [12,[14][15][16]. The activity limitations and participation restrictions associated with falls highlights the importance of identifying sensitive measures of gait function, such as spatial temporal gait variability, which are associated with falls history and falls risk.

Assessment and analysis of spatial temporal gait variability
Spatial temporal gait variability can be assessed using various techniques. Often considered the 'gold standard', 3D motion capture systems are able to record locomotion with high precision in purpose build laboratory settings. Motion capture systems use multiple cameras positioned to track location of refl ective markers placed on specifi c landmarks of an individual. Subsequent identifi cation of heel strike and toe off within the system software allows analysis of spatial temporal data, while kinematic and kinetic data can also be derived (analysis of gait kinetics requires additional force measurement, e.g. ground reaction force plates). While 3D motion captures systems provide a rich source of information, they are relatively expensive and often require purpose built laboratory facilities. Furthermore, 3D motion capture systems can require considerable time to both collect and process gait data. As a result, lower-cost, portable systems are frequently used. One of the most common methods currently used is computerized walkways [17]. Systems often come in a variety of lengths, are portable, and record spatial temporal parameters from embedded pressure sensors which detect foot position as the participant walks across the walkway. Alternative approaches include gyroscopes [18] or tri-axial accelerometers [19,20] worn on the participant to record spatial temporal gait parameters while they mobilize. Importantly, test-retest reliability appears similar between various assessment tools [18,21,22], providing some level of confi dence that the choice of assessment tool will not affect reliability of spatial temporal gait parameters.
Currently there does not appear to be a standardized metric used to report spatial temporal gait variability in the literature, with studies reporting a mixture of standard deviation and coeffi cient of variation as a measure of variability. As a result, caution should be applied when comparing spatial temporal variability between studies to ensure that the metrics compared (standard deviation or coeffi cient of variation) are the same.
The advantage of standard deviation is that it is not as strongly infl uenced by the mean value of the spatial temporal measure compared to the coeffi cient of variation. Coeffi cient of variation is calculated as the standard deviation divided by the mean. Therefore, a subject with larger step length but the same standard deviation would have a smaller coeffi cient of variation. However, the advantage of using coeffi cient of variation is that it is a dimensionless unit allowing comparison with other variability measures (e.g. step length variability and step time variability) and across studies. It has been suggested future studies report both standard deviation and coeffi cient of variation as measures to quantify gait variability [17]. Given relative advantages and disadvantages of these methods, this appears to be a reasonable suggestion. Furthermore, no matter which metric for variability is reported, pooling left and right step measurements to determine variability may obscure underlying asymmetries, particularly for patients groups where impairments are unilateral (e.g. stroke survivors, amputees). Reporting spatial temporal variability for each limb with respect to the relative to the underlying pathology is recommended (e.g. paretic and non-paretic limb for stroke, or amputated and non-amputated limb for amputees).
In order to reliably assess variability of spatial temporal gait parameters, a number of steps should be recorded. As a result, variability of spatial temporal parameters recorded with motion capture systems or computerized walkways involve repeated walk trials across an active data collection area. A recent review suggested a minimum of 12 steps are collected to improve reliability of data [17]. However, earlier previous studies have recommended up to 120 steps be recorded to reliably measure variability [19,23]. Increasing the number of steps recorded during data collection may be achieved by repeating individual walk trials with a participant stopping, turning 180 degrees, and again walking over or across the data collection area (see Figure 1). This approach, while practical, has technical limitations. It is generally considered that continuous walking may be more accurate, as short interrupted walks do not allow spatial temporal rhythms to stabilize. Indeed, gait variability measured from short interrupted walks has been shown to be higher than continuous walk test protocols [24] and likely refl ects stabilization of spatial temporal gait parameters at either the beginning or end of the walk trial as the participant initiates walking and decelerates at the end of the trial. It could be argued that walking during everyday activities predominantly occurs over short bursts, so testing in a manner refl ecting this provides a greater representation of everyday walking. However, if the purpose of testing is to identify abnormalities in spatial temporal rhythms, then artifi cially introducing variability may not be appropriate. Therefore, the most appropriate method may be to utilize continuous walking approaches or extending lead in and out lengths so that spatial temporal rhythms can stabilize (Figure 1).
While it is likely that a greater number of steps recorded would provide a stronger, and more reliable, estimate of spatial temporal variability, consideration should be given to the patient population to ensure that fatigue induced by the desire for high step counts does not affect the measure of gait variability [25]. For example, older adults [26], people with a high metabolic cost of walking (e.g. patients with diabetes mellitus [27]), or those with abnormal biomechanics (e.g. stroke survivors [28] or amputees [29]) are unlikely to be able to perform a high volume of walking within a short period of time and may fatigue quickly during gait analysis.
The introduction of fatigue would likely alter spatial temporal parameters and result in reduced gait speed which may result in an increased variability. Further supporting this suggestion, it is known that for an individual subject, variation in gait speed between individual walk trials is a potential confounder of gait variability fi ndings [30][31][32][33]. While previous work has investigated speed normalization of gait parameters in adults walking at a range of speeds [31,34], only one study has translated this approach to investigate the effect on variability of spatial temporal parameters [33]. This study demonstrated that variability of spatial temporal parameters was signifi cantly different compared to variability parameters normalized for gait speed, with those normalized for speed providing a more sensitive assessment of gait function. However, gait speed may differentially affect measure of spatial-temporal gait variability. In patients with Parkinson's disease it was shown that gait speed affected the measure of stride time variability, but had no effect on swing time variability [35]. Variations in speed may affect accurate interpretation of spatial temporal gait variability measures, however further work is required to thoroughly investigate the effect of gait speed on measures of spatial temporal gait variability. It may be that approaches of normalizing data for speed are particularly appropriate for populations where fatigue is likely to infl uence measures of spatial temporal variability. including older adults [36][37][38][39][40], lower-limb amputees [33,41,42], stroke [3], Parkinson's disease [2], Huntington's disease [43], patients with peripheral neuropathy [44] and mild cognitive impairment/dementia [38]. The broad nature of these pathological populations demonstrates the wide potential to utilise gait variability as a biomarker of falls risk in various clinical settings.

Evidence supporting gait variability as a marker of falls
Interestingly, the specifi c spatial temporal gait variability measures associated with falls appears quite diverse both within, and across, different patient groups (Table 1). For example, in older adults, one prospective and one retrospective study reported that stride time variability was a predictor of falls [37,38], others reported swing time and stride length variability were predictors of falls risk [36,40], while another reported step width variability was associated with falls history [39]. Similarly, retrospective studies with lower-limb amputees have reported that falls history was associated with greater amputated limb swing-time variability [42], non-amputated limb step time variability [41] and greater amputated limb step time, step length and step width variability [33]. This variance in the measures identifi ed may refl ect the strong collinearity It should also be acknowledged that analysis of spatial temporal gait variability may have wider implications for rehabilitation clinicians beyond being a biomarker for falls risk. It has previously been reported that higher levels of gait variability are associated with frailty [45] and fear of falling [46] in older adults. Recently, gait variability was found to differentiate older adults with mobility and cognitive impairment, and predict those with future cognitive decline [47]. Furthermore, increased step time variability was found to be associated with a greater burden of subclinical brain vascular abnormalities as identifi ed with magnetic resonance imaging [48]. Assessment of gait variability has the potential to be a sensitive, objective measure, of global function in rehabilitation, providing further support to suggest that it should be strongly considered in clinical practice.

Potential clinical utility
Spatial temporal gait variability is a measure which can clearly assist clinical practice. In this review it has been highlighted that spatial temporal gait variability is a sensitive measure of human motor control [1]. Previous studies have identifi ed correlations with falls history [2,3,33,38,39,[41][42][43][44] or ability to predict falls [36,37,40]. Given this signifi cant implications that experiencing a fall has on both the individual [12,[14][15][16] and also clinical services [12,13], the translational potential for spatial temporal gait variability should be investigated. It is likely that gait variability can assist clinical practice in several ways. For example, spatial temporal gait variability could be used as a tool to screen patients on admission to clinical services to help characterize falls risk and apply appropriate falls prevention strategies. In addition, screening with gait variability could be used to identify appropriate treatment targets. Abnormalities in gait would be objectively identifi ed and treatment approaches could be implemented to normalize gait function. This may assist streamlining of rehabilitation and therapy services for patients who require specifi c gait interventions. Furthermore, therapy may be applied which is capable of restoring measures of gait variability to normal levels. Previous studies have demonstrated that dual task training in older adults [49], treadmill training in Parkinson's disease [50] and pharmacological interventions in Parkinson's disease [51] were directly able to improve gait variability measures. Finally, gait variability could be used as a clinical outcome measure to demonstrate effectiveness of therapy and services. This would be achieved by quantifying changes in gait function from admission to discharge (or beyond) and may be particularly important given the increased demand for evidenced based practice and demonstrating effectiveness of have already been investigated and found to be a sensitive measure of gait function and mobility defi cits [52]. However, given the potential clinical signifi cance that gait variability has as a marker of falls, it is suggested that diffi culty in preselecting an appropriate gait measure should not preclude the analysis from being conducted.

Future directions
This review has demonstrated that spatial temporal gait variability is an important measure of gait function that should be considered for clinical use as a marker of falls risk. To continue developing and facilitating clinical utility of this tool, further research is required to identify appropriate metrics (e.g. step time variability, step length variability) and variability parameters (e.g. coeffi cient of variation, standard deviation).
Future work should also consider identifying the minimum number of steps required to obtain a reliable measure of gait variability. While it is generally acknowledged that a greater number of steps provides a stronger estimate of variability, this approach is somewhat impractical given the potential to induce fatigue in clinical populations. Finally, it is suggested that future studies should continue to investigate the predictive capacity of spatial temporal gait variability to predict falls risk with prospective studies as this is likely to have signifi cant value for clinical services. It may be that gait variability could be combined with additional outcomes measures to form an algorithm to provide a strong predicative tool to reliably inform clinicians of falls risk.

Conclusion
Spatial temporal gait variability is a sensitive measure of gait function capable of identifying falls risk or discriminating based on history of falls in a range of clinical populations. As such, gait variability has signifi cant potential to assist clinical practice as a sensitive marker of falls. Clinical services should further investigate implementing gait variability as a standard assessment to identify falls risk and potential therapeutic targets. It is suggested that this would assist delivery of therapy and identifi cation of patients requiring falls precautions.