Objective falls-risk prediction using wearable technologies amongst patients with and without neurogenic gait alterations: a narrative review of clinical feasibility

Introduction

Injuries are the fifth largest disease burden in Australia and falls alone account for 40% of injury-related admissions (1).A fall is an event in which a person inadvertently comes to rest on the ground (2). Falls are overrepresented in elderly populations and mostly occur during walking (3).A neurogenic gait alteration (NGA) is an alteration to gait due to abnormalities of the central or peripheral nervous system(s) resulting in changes to ambulation (4). The presence of an NGA doubles an elderly person’s risk of falling by altering postural stability, limb dynamics, movement speed, symmetry and smoothness (5). Multiple sclerosis, Parkinson’s disease, stroke, cerebral palsy, neurogenic claudication and peripheral neuropathy are common causes of NGA (3). Historically, tests like the Timed Up-and-Go (TUG), Berg-Balance-Scale (BBS), 10-Meter-Walk Test (10MWT) and Physiological Profile Assessment (PPA) were used for clinical assessment of falls-risk. These ‘test batteries’ require the patient to perform compound movements that assess balance and coordination during walking and completing questionnaires on existing risk-factors. However, they have high inter- and intra-observer variability and only provide an overall falls-risk without identifying which gait metrics are contributing, or where to intervene (6,7). With advances in gait-analysis technology, the gait perturbations prevalent in patients prone to falling can be assessed. Then, accurate capture of a patient’s gait metrics in a clinical setting would allow the clinician to make unbiased recommendations for intervention based on the perturbations of gait displayed

Wearable devices capture the ‘free-living’ gait of unobserved and routine walking behaviours (8). Wearable sensor technology also enables for remote monitoring of many patient metrics by feeding data from small wearable devices to a database on a phone or cloud accessible by the patient and physician (9). For gait analysis, the most common device is a single Inertial Measurement Unit (IMU) or ‘single-point’ wearable at the pelvis or sternum near the centre of mass (COM), at the ankle or feet for more precise step-recognition, or at multiple points with integrated signals (multi-point wearable sensors) (10). IMUs comprise of at least an accelerometer and gyroscope, sometimes accompanied by magnetometer. Recent IMUs employ tri-axial accelerometers which measure in 3 axes, antero-posterior (AP), medio-lateral (ML) and vertical (VT) (11).IMUs can calculate some gait metrics such as velocity, distance, step and turning occurrences and durations (12),from raw inertial measurements (and accompanying noise). To become the gold-standard clinical falls-risk assessment tool, IMUs must accurately capture and process relevant gait metrics to create falls-risk profiles from short walking bouts (WBs) to generate falls-risk profiles. These falls-risk predictions derived from IMU-derived gait metrics are consequently compared to patient-report of falls over a set retrospective period of time, acquired through a questionnaire and/or interview. This requires prior understanding of literature using quantitative gait assessment comparing fallers and non-fallers to identify which gait parameters are implicated in falls (13). Recent reviews have found that, compared to non-fallers, fallers have reduced gait velocity (GV) (13-17), cadence and step length (SL) (13,16), increased dynamic sway (13) and more variability in gait speed (GSV), step time (STV) and step length (SLV) (16). Table 1 summarises the gait variables often used to compare fallers and age- and disease-matched non-fallers using these variables and four others, step width (SW), step count (SC), static sway and turning kinematics, that have not been reviewed. For explanation of gait metrics, see Figures 1-4. We present the following article in accordance with the Narrative Review reporting checklist (available at https://dx.doi.org/10.21037/mhealth-21-7).

Figure 1 Gait bout. S₀=step that begins measurement window, turning occurs from S₈-S₁₄ (7 steps). S₁₈ and S₁₉ represent a variation in step length. SL, stride length; SW, step width; VT, vertical; ML, medio-lateral; COM, Centre of motion; AP, antero-posterior; S (n), nth step.

Figure 2 Timing of gait bout in Figure 1. S (n)=nth step, t (n)=time at nth step, t₀=time at start. t₄t₅ represents a variation in step time. ST, step time.

Figure 3 Static balance testing. Blue represents lateral movement of COM relative to the feet, orange represents anterior movement of the COM relative to the feet. ML, medio-lateral; COM, Centre of motion; AP, antero-posterior.

Figure 4 Calculations of key gait and balance variables. S’f=final step of first bout, Stn=nth step of turn, S’’i=Initial step of second bout.

Rationale

This review aims to investigate the application of wearable technology in capturing and processing relevant gait data for falls-risk analysis. In particular, this study aims to explore the presence of three barriers to the implementation of IMU devices in clinical falls-risk assessment: (I) accurately measuring gait metrics which are proven to be associated with high falls risk (mentioned above), (II) identifying differences in these metrics between groups of patients who have fallen, and those who have not, (III) accuracy of falls-prediction algorithms developed from IMU-collected gait metrics (Appendix 1).

Objectives

To perform a narrative review involving collection and critical appraisal of literature pertaining to the use of wearable devices in the prediction of falls, and identify factors limiting clinical implementation of IMU devices in falls-risk assessment.

Methods

Review protocol registered on PROSPERO (Appendix 2).

Eligibility

Included studies comprised of:

• Cohort studies or case-control studies;
• Studies with 6 or greater month follow up;
• Studies conducting gait analysis with wearable technologies measuring relevant gait metrics and/or used relevant gait metrics to assess falls-risk with analysis of diagnostic test accuracy;
• Studies written in English and available in full-text.

It should be noted that information could not be presented from papers in which methodology was not adequately discussed, in particular if they failed to provide detailed description of data capture, processing, and statistical analysis. This resulted in one paper being discounted from qualitative synthesis.

Information sources

Databases used to search for articles were Medline, PubMed, Scopus and Embase. All articles with results discussed came from a structured literature search. A similar search strategy was used across all databases for consistency, done in February 2020.

PubMed example: (((acceleromet* OR gyrometer OR magnetometer OR gait analysis) AND (sensor OR tracker OR device OR detector OR wearable) AND (gait OR posture OR ambulation)) AND (falls risk OR fall risk OR fall prediction OR falls prediction)).

Study selection

Studies were collated in EndNote X9 and duplicates removed. Studies were then screened based on inclusion/exclusion criteria by 2 independent reviewers (CB and DH) on an online reviewing software by Rayyan (Qatar Computing Research Institute). Discrepancies were resolved by cross-examination until consensus was reached.

Individual analysis of study bias

Critical appraisal of literature was conducted throughout the review process, with summaries concluding each section. For assessment of potential sources of bias, and level of evidence for each article based on criteria from the Centre for Evidence Based Medicine (CEBM) (18,19), see the available online: https://cdn.amegroups.cn/static/public/mhealth-21-7-1.xlsx.

Data extraction

The following information was collected from each article which met eligibility criteria by 2 independent reviewers (CB and DH), and documented available at available online: https://cdn.amegroups.cn/static/public/mhealth-21-7-1.xlsx:

• Number of subjects.
• Type, number, and placement of IMU device on subjects.
• Gait features that were measured by the IMU device (s) used in the study.
• Length and setting of WB used to measure gait features.
• Gait features used in the falls-risk algorithm (if any) produced by the study.
• Tool used to classify subjects into high and low falls-risk group
• Study period over which patients were assessed for falls.
• Participant demographics, including the presence of any neurological disease affecting gait (NGA).
• General data from the article such as year of publication, journal, objectives, level of evidence, risk of bias and funding sources.

Data synthesis

Following data collection, studies were sorted into 3 groups pertaining to the barrier to clinical IMU implementation that they addressed, those being device accuracy, device sensitivity, and algorithm predictive accuracy. Qualitative analysis was performed to provide an appreciation of the results of existing literature and discuss whether any barriers remain to implementation.

Results

Characteristics of included studies

Since 2002 (20), researchers have investigated the use of single-point or multi-point IMUs to measure gait metrics (Table 2). In total, 46 articles met the eligibility criteria for inclusion in this study (Figure 5). While most used single-point IMUs, they examined fewer variables, had more disagreement with gold-standard measurements, and only 5 studied NGAs. Single-point wearables are much easier to use in a clinical setting or home, but they cannot compare data from numerous sensors to reduce false-positive errors in data capture, which may be enhanced in NGAs where movement is less smooth. Inaccuracies also mean that algorithms interpreting the data must provide leeway when identifying steps to avoid false-negative errors in data processing. 17 papers compared IMUs to gold-standard gait-analysis technologies and one (21) assessed test-retest reliability to validate wearables in clinical assessment of relevant gait parameters. 25 papers investigated the ability of IMUs to detect expected gait changes between groups of fallers and age-matched non-fallers. 33 studies used IMUs from 50–100 Hz to distinguish populations of high and low falls-risk determined by fall history (22), test battery score or prospective falls. The 46 studies included in the analysis typically presented with a low-to-moderate risk of bias (available online: https://cdn.amegroups.cn/static/public/mhealth-21-7-1.xlsx)

Figure 5 PRISMA flow diagram of the article selection process showing yields at each stage of the process.

Discussion

Accuracy and reliability in measuring spatiotemporal gait metrics

Studies using multi-point sensors on the limbs and trunk verified that whether accelerometers were used alone (23,24) or with gyroscopes and magnetometers (12,25,26), they could capture gait with comparable accuracy compared to a gold-standard. GV was accurately detected by multi-point IMUs sampling at 32 Hz (24) to 500 Hz (27), SLV and STV at 130 Hz (23), dynamic balance in ML and VT directions at 130 Hz (26) and SC as low as 20 Hz (12). However, one study (27) found a significant discrepancy between IMU data at 500 Hz and force-sensitive treadmill data for cadence and SL, while other studies found IMUs 100–200 Hzto be accurate for cadence (26) and SL (20,23,24). Given otherwise similar methodology and low risk of bias, one possible reason for this discrepancy may be location of sensor placement, which was on the feet in the 500 Hz trials. This may have caused the IMU to detect noise from smaller movements of the foot during a gait cycle, compared to an IMU placed on the trunk which detects overall motion.

Studies using single-point wearables from 31.26–255 Hz (21,28) have more disagreement between gold-standard and sensor data. For example, when compared to the Centre of Pressure sway (COP) from a force plate during static posturography, IMU data (sampled at 52 Hz) captured from the COM had poor correlation (29), while data capture at 200 Hz from the sternum highly correlated (30). It is unlikely the difference sampling rates account for this discrepancy (given both are above the Nyquist sampling thresholds of around 30 Hz for detecting human motion) (31).Perhaps placing the IMU above the COM magnifies small movements in the AP and ML directions, thus being able to distinguish sway from noise. This translates into the sternum being a key location for single-point fixation wearables for gait and falls assessment. When single-point sensors were used to measure SL (8,28,32), cadence (8,28), GV (8), SC (28,33), STV (8), dynamic sway (34) and turning (35), they demonstrated good-to-excellent agreement with gold-standard methods of gait analysis. The findings of these studies thus suggest the gait metrics measured by wearables have sufficient accuracy and reliability to be used in clinical setting, but the consistency of this agreement from condition-to-condition is yet to be assessed.

Sensitivity in detecting differences between fallers and non-fallers

Regarding the ability to differentiate between groups of age- and disease-matched fallers and non-fallers, studies tended to show that wearable devices were able to detect subtle differences in relevant biomechanical gait metrics between these groups. Overall, IMUs were sensitive enough to detect differences in GV (25,36-42), SL (25,37,39,40), turning kinematics (25,41,43), GSV (39), SLV (37), STV (36,44), cadence (25,36,39-41,45) and postural sway in single (39,46,47) or multiple axes (25,38,40-42,48-50) between fallers and non-fallers. Only one retrospective study did not identify dynamic imbalance during treadmill walking as significantly different between fallers and non-fallers (51). However, it should be noted that past fallers tend to consciously widen their stance for lateral stability, and this effect is enhanced on a treadmill (52). Moreover, this study had a moderate-high risk of bias according to CEBM criteria. Interestingly, most studies assessing SC using single-point IMUs at 50–100 Hz, did not find that fallers had a significantly lower SC than non-fallers (37,39,44,49,50,53), while one did (40). This may be because waist, thigh and wrist-worn high-frequency single-point sensors underestimate SC by 2–10% (54) and SC seems to only be altered in fallers with NGAs, which only one study (50) assessed. IMU devices accurately measure gait factors related to falls-risk and appear able to differentiate between groups of fallers and non-fallers using these factors. However, for clinical implementation this sensitivity in detecting differences in gait metrics needs to be confirmed in continuous real-world data collection in home and community environments.

Classification accuracy of falls-prediction tools

Regarding the accuracy of falls-prediction algorithms developed based on biomechanical data collected by IMUs, studies were considered successful if biomechanical data was used by an algorithm to accurately sort patients into high- and low-falls-risk groups. In successful studies, these algorithms were developed post-hoc by neural network, machine learning or decision trees, trialling sets of 5–30 gait parameters to determine the most successful combination. Limiting the number of gait parameters trialled at once prevents ‘over-training’, which increases internal classification accuracy but reduces generalisability (55).Nonetheless, these post-hoc selected ‘inputs’ into the algorithms increases risk of a ‘contextualised’ model and its accuracy, reliability and validity to the generalised population of fallers is unknown without validation in external data-sets.

Additionally, there is limited evidence verifying the role of static balance in falls, several papers utilising biomechanical data algorithms to classify fallers performed static posturography. 5 papers utilised static postural stability in their algorithm yielding faller classification accuracies of 71.2–85% (25,56,57) and falls-risk classification accuracies of 90.5% (36) and 97% (58) to Tinetti and STRATIFY scores respectively. While their results seem impressive, it should be noted that multiple measures like SD and CV were used for sway, and the algorithm tested combinations of them until it yielded the highest classification accuracy. By inputting the same variable multiple times, false-positive results are more likely, and this can only be accounted for by testing the algorithm against novel data to test true classification accuracy, known as ‘cross-validation’ (24,59). The majority of these papers had a moderate risk of bias according to CEBM criteria (36,56-58).

That said, when papers did not include static posturography, classification of fallers was still accomplished. A study by Iluz et al., (38) using a 100 Hz IMU at the COM identified lower GV, and higher sway in AP and VT directions during daily-life sit-to-walk transitions in fallers compared to age-matched adults. Combining this data into a faller identification algorithm yielded a classification accuracy of 88% however, using data captured during lab-based gait yielded 71% accuracy. This may imply that algorithms should be specific to the gait-capture setting. Iluz’s paper is one of many to combine dynamic sway with metrics including GV (25,47,60), SL (25,60), turning kinematics (25,59), cadence (25,59-62), STV (60,63), SLV (60) and GSV (60) in algorithms with moderate-good classification accuracies of 68-85%, consistently out-performing existing test batteries (36,38,43,47,56,57). One study (64) used cadence, SL and GV measured by a triaxial accelerometer at 100 Hz at the COM in an algorithm which classified 6-month prospective fallers with an Odds-Ratio of 3.89 if the cut-off value was exceeded. An algorithm using SLV, cadence and turning kinematics from 102.4 Hz sensors at each shank had a predictive accuracy of 2-year prospective falls of 83.02% and another algorithm using SL, STV and turning kinematics had an accuracy of 72.97% (43).

Feasibility and limitations

No studies thus far have attempted falls-risk classification using IMU data from NGA populations. As such, limitations still exist when it comes to the assessment of falls-risk in neurological populations which make up a large portion of hospital inpatients and clinical visitors. Falls-classification tools in elderly populations used single- or multi-point IMUs with various sensor configurations, walking settings and gait-metrics. Although, IMU devices are able to assign falls risk using algorithms developed with gait metrics collected from WBs for non-neurological populations, these are yet to be extensively validated in large external populations. Current falls-prediction tools therefore require validation in external data-sets before clinical implementation Moreover current classification accuracy is limited by comparison against self-report of retrospective falls (over a set period) by patients which is not an accurate nor reliable gold-standard of objective measurement. Additionally, falls-prediction tools constructed from ‘walking bouts’ may not reflect falls-risk during ‘free-living’ gait and thus future studies employing gait metrics from continuous monitoring of walking in the community and in the home are warranted.

Despite these limitations, IMUs placed at the COM, chest or limbs, sampling at 50–250 Hz have been validated against gold-standard methods in capturing various gait metrics. They also have the potential to distinguish fallers and non-fallers based on gait deterioration patterns and use this to accurately assign falls-risk. Single-point wearables are more viable in clinical settings but have not been thoroughly tested in NGA populations nor validated for SLV, static sway or GSV. Moreover, the studies have a lot of heterogeneity in methodology, with no clear standard. Finally, our review included insufficient studies of NGA populations and more data is required to justify using single-point IMUs for gait analysis and falls prediction in NGAs, primarily those of spinal origin.

Conclusions

Currently, the literature is divided on metrics like GV and SL as predictors of falls, and advents such as turning kinetics are in the immature stage of research. In addition, the bulk of studies examine community-dwelling elderly, with few investigating NGAs and none investigating spinal NGAs. Moreover, while the use of sensor technology in falls-risk analysis has been investigated, the studies examine very few gait metrics at once, and comparing studies reveals a wide variety of sensor types in use. Future avenues of research thus include investigations of multiple gait metrics simultaneously especially in NGA and spinal NGA populations.

Acknowledgments

Special thanks to Dr Greg Smith for helping to develop the content and structure of the manuscript. Doctor Matthew Brodie for providing details of past publications and proofreading the section on wearable technology. Luke Sy for algorithm and application development for the methodology. Jessie Ding for creating the artwork used in this manuscript.

Funding: This work did not receive funding from any organisations. The work was done as part of a research project for the University of New South Wales, Department of Medicine (NSW, Australia), in collaboration with NeuroSpineClinic (Suite 7 Level 7, Prince of Wales Private Hospital, NSW, Australia) and the NSURG Research Group (Level 7, Prince of Wales Private Hospital, NSW, Australia).

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://dx.doi.org/10.21037/mhealth-21-7

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://dx.doi.org/10.21037/mhealth-21-7). The authors have no conflicts of interest to declare

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Australian Institute of Health and Welfare. Injury. Canberra, Australia: 2019.
2. World Health Organisation. Falls. World Health Organisation, Geneva, Switzerland. 2018. Available online: https://www.who.int/news-room/fact-sheets/detail/falls
3. McCrum C, Gerards MHG, Karamanidis K, et al. A systematic review of gait perturbation paradigms for improving reactive stepping responses and falls risk among healthy older adults. Eur Rev Aging Phys Act 2017;14:3. [Crossref] [PubMed]
4. Ronthal M. Neurologic gait disorders of elderly people. UpToDate. 2018. Available online: https://www.uptodate.com/contents/neurologic-gait-disorders-of-elderly-people
5. Stolze H, Klebe S, Zechlin C, et al. Falls in frequent neurological diseases--prevalence, risk factors and aetiology. J Neurol 2004;251:79-84. [Crossref] [PubMed]
6. Gor-García-Fogeda MD, Cano de la Cuerda R, Carratalá Tejada M, et al. Observational Gait Assessments in People With Neurological Disorders: A Systematic Review. Arch Phys Med Rehabil 2016;97:131-40. [Crossref] [PubMed]
7. Toro B, Nester C, Farren P. A review of observational gait assessment in clinical practice. Physiother Theory Pract 2003;19:137-49. [Crossref]
8. Brodie MA, Coppens MJ, Lord SR, et al. Wearable pendant device monitoring using new wavelet-based methods shows daily life and laboratory gaits are different. Med Biol Eng Comput 2016;54:663-74. [Crossref] [PubMed]
9. Majumder S, Mondal T, Deen MJ. Wearable Sensors for Remote Health Monitoring. Sensors (Basel) 2017;17:130. [Crossref] [PubMed]
10. Vienne A, Barrois RP, Buffat S, et al. Inertial Sensors to Assess Gait Quality in Patients with Neurological Disorders: A Systematic Review of Technical and Analytical Challenges. Front Psychol 2017;8:817. [Crossref] [PubMed]
11. Patonis P, Patias P, Tziavos IN, et al. A Fusion Method for Combining Low-Cost IMU/Magnetometer Outputs for Use in Applications on Mobile Devices. Sensors (Basel) 2018;18:2616. [Crossref] [PubMed]
12. Anwary AR, Yu H, Vassallo M. An Automatic Gait Feature Extraction Method for Identifying Gait Asymmetry Using Wearable Sensors. Sensors (Basel) 2018;18:676. [Crossref] [PubMed]
13. Creaby MW, Cole MH. Gait characteristics and falls in Parkinson's disease: A systematic review and meta-analysis. Parkinsonism Relat Disord 2018;57:1-8. [Crossref] [PubMed]
14. Fasano A, Canning CG, Hausdorff JM, et al. Falls in Parkinson's disease: A complex and evolving picture. Mov Disord 2017;32:1524-36. [Crossref] [PubMed]
15. Gillain S, Boutaayamou M, Beaudart C, et al. Assessing gait parameters with accelerometer-based methods to identify older adults at risk of falls: a systematic review. Eur Geriatr Med 2018;9:435-48. [Crossref]
16. Mortaza N, Abu Osman NA, Mehdikhani N. Are the spatio-temporal parameters of gait capable of distinguishing a faller from a non-faller elderly? Eur J Phys Rehabil Med 2014;50:677-91. [PubMed]
17. Menant JC, Schoene D, Sarofim M, et al. Single and dual task tests of gait speed are equivalent in the prediction of falls in older people: a systematic review and meta-analysis. Ageing Res Rev 2014;16:83-104. [Crossref] [PubMed]
18. Phillips B, Ball C, Sackett D, et al. Oxford Centre for Evidence-based Medicine – Levels of Evidence (March 2009). Centre for Evidence Based Medicine, Oxford, UK. 2009. Available online: https://www.cebm.net/2009/06/oxford-centre-evidence-based-medicine-levels-evidence-march-2009/
19. Centre for Evidence Based Medicine. Critical Appraisal tools. University of Oxford, Oxford, UK. 2014. Available online: https://www.cebm.net/2014/06/critical-appraisal/
20. Aminian K, Najafi B, Büla C, et al. Spatio-temporal parameters of gait measured by an ambulatory system using miniature gyroscopes. J Biomech 2002;35:689-99. [Crossref] [PubMed]
21. Saunders NW, Koutakis P, Kloos AD, et al. Reliability and validity of a wireless accelerometer for the assessment of postural sway. J Appl Biomech 2015;31:159-63. [Crossref] [PubMed]
22. Skelton D, Todd C. What are the main risk factors for falls amongst older people and what are the most effective interventions to prevent these falls? Copenhagen: World Health Organisation, Network HE;2004.
23. Rebula JR, Ojeda LV, Adamczyk PG, et al. Measurement of foot placement and its variability with inertial sensors. Gait Posture 2013;38:974-80. [Crossref] [PubMed]
24. Aich S, Pradhan PM, Park J, et al. A Validation Study of Freezing of Gait (FoG) Detection and Machine-Learning-Based FoG Prediction Using Estimated Gait Characteristics with a Wearable Accelerometer. Sensors (Basel) 2018;18:3287. [Crossref] [PubMed]
25. Qiu H, Rehman RZU, Yu X, et al. Application of Wearable Inertial Sensors and A New Test Battery for Distinguishing Retrospective Fallers from Non-fallers among Community-dwelling Older People. Sci Rep 2018;8:16349. [Crossref] [PubMed]
26. Brodie MA, Beijer TR, Canning CG, et al. Head and pelvis stride-to-stride oscillations in gait: validation and interpretation of measurements from wearable accelerometers. Physiol Meas 2015;36:857-72. [Crossref] [PubMed]
27. Donath L, Faude O, Lichtenstein E, et al. Mobile inertial sensor based gait analysis: Validity and reliability of spatiotemporal gait characteristics in healthy seniors. Gait Posture 2016;49:371-4. [Crossref] [PubMed]
28. O'Brien MK, Hidalgo-Araya MD, Mummidisetty CK, et al. Augmenting Clinical Outcome Measures of Gait and Balance with a Single Inertial Sensor in Age-Ranged Healthy Adults. Sensors (Basel) 2019;19:4537. [Crossref] [PubMed]
29. Jebelli H, Ahn CR, Stentz TL. Fall risk analysis of construction workers using inertial measurement units: Validating the usefulness of the postural stability metrics in construction. Saf Sci 2016;84:161-70. [Crossref]
30. Hsieh KL, Roach KL, Wajda DA, et al. Smartphone technology can measure postural stability and discriminate fall risk in older adults. Gait Posture 2019;67:160-5. [Crossref] [PubMed]
31. Cheng Q, Juen J, Li Y, et al. editors. GaitTrack: Health monitoring of body motion from spatio-temporal parameters of simple smart phones. Proceedings of the International Conference on Bioinformatics, Computational Biology and Biomedical Informatics, 2013.
32. Kitagawa N, Ogihara N. Estimation of foot trajectory during human walking by a wearable inertial measurement unit mounted to the foot. Gait Posture 2016;45:110-4. [Crossref] [PubMed]
33. Najafi B, Armstrong DG, Mohler J. Novel wearable technology for assessing spontaneous daily physical activity and risk of falling in older adults with diabetes. J Diabetes Sci Technol 2013;7:1147-60. [Crossref] [PubMed]
34. Brodie MA, Psarakis M, Hoang P. Gyroscopic corrections improve wearable sensor data prior to measuring dynamic sway in the gait of people with Multiple Sclerosis. Comput Methods Biomech Biomed Engin 2016;19:1339-46. [Crossref] [PubMed]
35. El-Gohary M, Pearson S, McNames J, et al. Continuous monitoring of turning in patients with movement disability. Sensors (Basel) 2013;14:356-69. [Crossref] [PubMed]
36. Gietzelt M, Feldwieser F, Gövercin M, et al. A prospective field study for sensor-based identification of fall risk in older people with dementia. Inform Health Soc Care 2014;39:249-61. [Crossref] [PubMed]
37. Del Din S, Galna B, Godfrey A, et al. Analysis of Free-Living Gait in Older Adults With and Without Parkinson's Disease and With and Without a History of Falls: Identifying Generic and Disease-Specific Characteristics. J Gerontol A Biol Sci Med Sci 2019;74:500-6. [Crossref] [PubMed]
38. Iluz T, Weiss A, Gazit E, et al. Can a Body-Fixed Sensor Reduce Heisenberg's Uncertainty When It Comes to the Evaluation of Mobility? Effects of Aging and Fall Risk on Transitions in Daily Living. J Gerontol A Biol Sci Med Sci 2016;71:1459-65. [Crossref] [PubMed]
39. van Schooten KS, Pijnappels M, Rispens SM, et al. Daily-Life Gait Quality as Predictor of Falls in Older People: A 1-Year Prospective Cohort Study. PLoS One 2016;11:e0158623 [Crossref] [PubMed]
40. van Schooten KS, Pijnappels M, Rispens SM, et al. Ambulatory fall-risk assessment: amount and quality of daily-life gait predict falls in older adults. J Gerontol A Biol Sci Med Sci 2015;70:608-15. [Crossref] [PubMed]
41. Zakaria NA, Kuwae Y, Tamura T, et al. Quantitative analysis of fall risk using TUG test. Comput Methods Biomech Biomed Engin 2015;18:426-37. [Crossref] [PubMed]
42. Brodie MA, Menz HB, Smith ST, et al. Good lateral harmonic stability combined with adequate gait speed is required for low fall risk in older people. Gerontology 2015;61:69-78. [Crossref] [PubMed]
43. Greene BR, Doheny EP, Walsh C, et al. Evaluation of falls risk in community-dwelling older adults using body-worn sensors. Gerontology 2012;58:472-80. [Crossref] [PubMed]
44. Brodie MA, Coppens MJ, Ejupi A, et al. Comparison between clinical gait and daily-life gait assessments of fall risk in older people. Geriatr Gerontol Int 2017;17:2274-82. [Crossref] [PubMed]
45. Schwesig R, Fischer D, Lauenroth A, et al. Can falls be predicted with gait analytical and posturographic measurement systems? A prospective follow-up study in a nursing home population. Clin Rehabil 2013;27:183-90. [Crossref] [PubMed]
46. Mahoney JR, Oh-Park M, Ayers E, et al. Quantitative trunk sway and prediction of incident falls in older adults. Gait Posture 2017;58:183-7. [Crossref] [PubMed]
47. Isho T, Tashiro H, Usuda S. Accelerometry-based gait characteristics evaluated using a smartphone and their association with fall risk in people with chronic stroke. J Stroke Cerebrovasc Dis 2015;24:1305-11. [Crossref] [PubMed]
48. Bischoff-Ferrari HA, Conzelmann M, Stähelin HB, et al. Is fall prevention by vitamin D mediated by a change in postural or dynamic balance? Osteoporos Int 2006;17:656-63. [Crossref] [PubMed]
49. Weiss A, Brozgol M, Dorfman M, et al. Does the evaluation of gait quality during daily life provide insight into fall risk? A novel approach using 3-day accelerometer recordings. Neurorehabil Neural Repair 2013;27:742-52. [Crossref] [PubMed]
50. Weiss A, Herman T, Giladi N, et al. Objective assessment of fall risk in Parkinson's disease using a body-fixed sensor worn for 3 days. PLoS One 2014;9:e96675 [Crossref] [PubMed]
51. Riva F, Toebes MJ, Pijnappels M, et al. Estimating fall risk with inertial sensors using gait stability measures that do not require step detection. Gait Posture 2013;38:170-4. [Crossref] [PubMed]
52. Tesio L, Rota V. The Motion of Body Center of Mass During Walking: A Review Oriented to Clinical Applications. Front Neurol 2019;10:999. [Crossref] [PubMed]
53. Brodie MA, Okubo Y, Annegarn J, et al. Disentangling the health benefits of walking from increased exposure to falls in older people using remote gait monitoring and multi-dimensional analysis. Physiol Meas 2017;38:45-62. [Crossref] [PubMed]
54. Bassett DR Jr, Toth LP, LaMunion SR, et al. Step Counting: A Review of Measurement Considerations and Health-Related Applications. Sports Med 2017;47:1303-15. [Crossref] [PubMed]
55. Pham HNA, Triantaphyllou E. The Impact of Overfitting and Overgeneralization on the Classification Accuracy in Data Mining. In: Maimon O, Rokach L. editors. Soft Computing for Knowledge Discovery and Data Mining. Boston, MA: Springer US; 2008:391-431.
56. Greene BR, McGrath D, Walsh L, et al. Quantitative falls risk estimation through multi-sensor assessment of standing balance. Physiol Meas 2012;33:2049-63. [Crossref] [PubMed]
57. Greene BR, Doheny EP, Kenny RA, et al. Classification of frailty and falls history using a combination of sensor-based mobility assessments. Physiol Meas 2014;35:2053-66. [Crossref] [PubMed]
58. Giansanti D, Macellari V, Maccioni G. New neural network classifier of fall-risk based on the Mahalanobis distance and kinematic parameters assessed by a wearable device. Physiol Meas 2008;29:N11-9.
59. Drover D, Howcroft J, Kofman J, et al. Faller Classification in Older Adults Using Wearable Sensors Based on Turn and Straight-Walking Accelerometer-Based Features. Sensors (Basel) 2017;17:1321. [Crossref] [PubMed]
60. Ihlen EAF, van Schooten KS, Bruijn SM, et al. Improved Prediction of Falls in Community-Dwelling Older Adults Through Phase-Dependent Entropy of Daily-Life Walking. Front Aging Neurosci 2018;10:44. [Crossref] [PubMed]
61. Howcroft J, Lemaire ED, Kofman J. Wearable-Sensor-Based Classification Models of Faller Status in Older Adults. PLoS One 2016;11:e0153240 [Crossref] [PubMed]
62. Rivolta MW, Aktaruzzaman M, Rizzo G, et al. Evaluation of the Tinetti score and fall risk assessment via accelerometry-based movement analysis. Artif Intell Med 2019;95:38-47. [Crossref] [PubMed]
63. Gietzelt M, Nemitz G, Wolf KH, et al. A clinical study to assess fall risk using a single waist accelerometer. Inform Health Soc Care 2009;34:181-8. [Crossref] [PubMed]
64. Mignardot JB, Deschamps T, Barrey E, et al. Gait disturbances as specific predictive markers of the first fall onset in elderly people: a two-year prospective observational study. Front Aging Neurosci 2014;6:22. [Crossref] [PubMed]