- Email: [email protected]
Accelerometry-Based Gait Characteristics Evaluated Using a Smartphone and Their Association with Fall Risk in People with Chronic Stroke
Accelerometry-Based Gait Characteristics Evaluated Using a Smartphone and Their Association with Fall Risk in People with Chronic Stroke Takuya Isho, PT, MSc,* Hideyuki Tashiro, PT, MSc,† and Shigeru Usuda, PT, PhD‡
Background: The smartphone, which contains inertial sensors, is currently available and affordable device and has the potential to provide a self-assessment tool for health management. The aims of this study were to use a smartphone to record trunk acceleration during walking and to compare accelerometry variables between poststroke subjects with and without a history of falling. Methods: This cross-sectional study was conducted in 2 day care centers for elderly adults. Twenty-four community-dwelling adults with chronic stroke (mean age, 71.6 6 9.7 years; mean time since stroke, 68.5 6 38.7 months) were enrolled. Acceleration of the trunk during walking was recorded in the anteroposterior and mediolateral directions and quantified using the autocorrelation coefficient, harmonic ratio, and interstride variability (coefficient of variation of root mean square acceleration). Fall history in the past 12 months was obtained by self-report. Results: Eleven participants (45.8%) reported at least one fall in the past 12 months and were classified as fallers. Fallers exhibited significantly higher interstride variability of mediolateral trunk acceleration than nonfallers. In the logistic regression analysis, interstride variability of mediolateral trunk acceleration was significantly associated with fall history (adjusted odds ratio, 1.462; 95% confidence interval, 1.009-2.120). The area under the receiver operating characteristic curve for interstride variability of mediolateral trunk acceleration to discriminate fallers from nonfallers was .745 (95% confidence interval, .527-.963). Conclusions: The results suggest that quantitative gait assessment using a smartphone can provide detailed and objective information about subtle changes in the gait pattern of stroke subjects at risk of falling. Key Words: Accelerometry— accidental falls—cerebrovascular disorders—gait—postural balance—walking. Ó 2015 by National Stroke Association
Introduction Falls are a major complication after stroke1 and occur more frequently in the community setting after hospital discharge than they do in the hospital setting.2 Falls can From the *Department of Rehabilitation, National Hospital Organization Takasaki General Medical Center, Takasaki, Gunma; †Department of Rehabilitation, Saitama Cooperative Hospital, Kawaguchi, Saitama; and ‡Department of Rehabilitation Sciences, Gunma University Graduate School of Health Sciences, Maebashi, Gunma, Japan. Received January 25, 2015; accepted February 5, 2015. The present address of H.T. is Department of Rehabilitation, Eniwa Hospital, Eniwa, Hokkaido, Japan. Institutions where the study was performed are the Geriatric Health Services Facility Minuma, Kawaguchi, and the Daycare Center of the Kumagaya Cooperative Hospital, Kumagaya, Saitama, Japan.
cause depression,3 fear of falling,4 and severe injuries such as hip fractures,5,6 all of which can restrict physical activities and cause further deterioration of health status.2 According to a literature review, between 23%
The authors received no grant support for this study. The authors have no conflicts of interest to declare. Address correspondence to Takuya Isho, PT, MSc, Department of Rehabilitation, National Hospital Organization Takasaki General Medical Center, 36 Takamatsu-cho, Takasaki, Gunma 370-0829, Japan. E-mail: [email protected]. 1052-3057/$ - see front matter Ó 2015 by National Stroke Association http://dx.doi.org/10.1016/j.jstrokecerebrovasdis.2015.02.004
Journal of Stroke and Cerebrovascular Diseases, Vol. 24, No. 6 (June), 2015: pp 1305-1311
1305
T. ISHO ET AL.
1306
and 73% of community-dwelling stroke survivors experienced at least 1 fall during a 3-12–month follow-up period, and walking was the most common activity that lead to falls.2 A number of previous studies have demonstrated that balance and mobility deficits measured with clinical assessment tools are important risk factors for falls.7-10 However, a few studies have reported that clinical performance measures fail to distinguish fallers from nonfallers.8,11,12 Harris et al12 showed that there was no relation between the Berg Balance Scale (BBS) score, gait speed, and fall history in the past 6 months in individuals with chronic stroke. Similarly, Belgen et al8 reported no significant difference in the Timed Up and Go (TUG) test between community-dwelling poststroke individuals with and without a history of falls. The BBS, gait speed measured over a short distance, and the TUG are valid and reliable clinical tools to assess functional ability in stroke patients.13,14 However, these tools may not be optimal to assess fall risk in these patients. The BBS predominantly tests balance in static positions and during transition movements,15 and does not contain many items that evaluate dynamic balance during walking activities. Gait speed and the TUG test are simplified quantitative mobility assessment that assess only the temporal aspect of gait, and therefore, may be insufficient to identify the gait characteristics of poststroke individuals at risk of falling. Objective assessment using technological devices plays an important role in the quantification of movement patterns. Recently, the use of inertial sensors such as accelerometers and gyroscopes for movement analysis has become popular in clinical practice because these sensors are inexpensive, small, and portable and testing is not restricted to a laboratory environment.16 Inertial sensors are contained in mobile devices such as smartphones, meaning that a smartphone may be used to assess gait. This has been attempted in a few studies.17,18 Several studies have assessed gait by measuring the acceleration of the upper body in stroke subjects. Accelerometry variables have been compared between stroke subjects and healthy controls19,20 and between stroke subjects walking with different conditions of cane use,21,22 and the association of accelerometry variables with motor and mobility functions and ambulatory independence has been evaluated in stroke subjects.19,20,23 However, no studies have yet assessed the clinical relevance of gait stability measures derived from upper body accelerometry to the risk of falling in individuals with stroke, nor have there been any studies that have used a smartphone to measure upper body acceleration during gait. The aims of the present study were to use a smartphone to record trunk acceleration during walking and to compare accelerometry variables between poststroke subjects with and without a history of falling.
Methods Participants This was a cross-sectional study conducted in 2 day care centers for elderly adults in Saitama, Japan. The Ethics Committee of the Saitama Cooperative Hospital approved all study procedures. All participants were informed about the study and gave written informed consent. Community-dwelling adults with chronic stroke receiving day care services were recruited and screened for inclusion and exclusion criteria. Inclusion criteria were (1) more than 12 months since stroke onset and (2) ability to walk 16 m independently with or without a singlepoint cane and/or an orthosis. Exclusion criteria were (1) severe cardiovascular, respiratory, musculoskeletal, or neurologic disorder other than stroke that affected gait performance; (2) unable to understand the instructions because of communication problem or moderateto-severe cognitive dysfunction (ie, 5 or more errors on the Short Portable Mental Status Questionnaire [SPMSQ]); and (3) household ambulators walked only indoors or only mobilized during rehabilitation sessions. Fall history is an independent predictor of future falls.10 Participants were divided into fallers and nonfallers according to whether or not they reported that they had experienced falling in the past 12 months. Falls were defined as ‘‘falling down to the ground or to the lower level against one’s will.’’24 Retrospective self-report of fall events in the previous 12 months has shown excellent agreement with 3-month interval recall method (% agreement, 98.8%) in Japanese community-dwelling older adults.25
Procedures Demographic and clinical characteristics were collected from medical records and interviews. All clinical assessments were carried out by a pair of trained physical therapists. The observer’s subjective assessment was judged by the consensus of both examiners. Participants performed each assessment once. Cognitive status was evaluated using the SPMSQ, which is a widely used 10-item cognitive examination26 and has shown good reliability and validity.27 A score of 5 or more errors on the SPMSQ indicates moderate-to-severe cognitive impairment; a score of 3-4 errors indicates mild cognitive impairment; and a score of 0, 1, or 2 errors indicates cognitively intact. Lower-extremity motor function was assessed according to the 6 motor stages defined by Brunnstrom,28 where lower stages indicate greater motor deficit. The test–retest reliability for lower-extremity Brunnstrom recovery stage was tested preliminarily in 17 hemispheric stroke patients with supratentorial lesions from medical record in our hospital and was weighted kappa 5 .879 (mean duration between test and retest, 2.3 6 1.4 days). Lower-extremity sensory impairment was measured with the sensory
GAIT ANALYSIS USING A SMARTPHONE
subscale of the Stroke Impairment Assessment Set (SIAS), which evaluates light touch sensation and thumb position sense on a 4-point scale.29 The total score of the sensory subscale of SIAS ranges from 0 to 6 points, with a lower score indicating a more severe sensory impairment. The inter-rater reliability of SIAS lower-extremity sensory items in patients with hemiparetic stroke was weighted kappa 5 .500 and .838 for touch and position, respectively.29 Physical performance was tested using the Mini-Balance Evaluation Systems Test (Mini-BESTest), the TUG, and the 10-m walking test (10MWT). The MiniBESTest is a clinical balance assessment tool designed to measure dynamic balance.30 It evaluates 4 different balance control systems: anticipatory postural adjustments, postural responses, sensory orientation, and balance during gait. It consists of 14 items and has a minimum score of 0 and a maximum score of 28, with higher scores indicating better balance ability. In comparison with the BBS, the Mini-BESTest contains items that evaluate functional stability during walking activities and has less ceiling effect.31 Intrarater and inter-rater reliability (intraclass correlation coefficient [ICC] 5 .97 and .96, respectively) and concurrent validity (correlation with the BBS, Spearman’s rho 5 .83) of the Mini-BESTest have been established in subjects with chronic stroke.31 The TUG is a test of functional mobility that measures the time required to rise from an armchair, walk 3 m at a comfortable and safe pace, turn around, walk back to the chair, and sit down again.32 Participants were allowed to use a single-point cane and/or an orthosis during the TUG test if necessary. For the 10MWT, participants were asked to walk a distance of 16 m at a self-selected comfortable speed with or without a single-point cane and/or an orthosis. The time taken to walk the central 10 m was measured by a digital stopwatch and used to calculate gait speed. The TUG and the 10MWT both have excellent test–retest reliability (ICC ..9) in community-dwelling individuals with stroke.33
Trunk Accelerometry Gait Analysis Trunk linear accelerations were measured along the anteroposterior (AP), mediolateral (ML), and vertical axes during the 10MWT. Acceleration was measured using the Android-based smartphone (Xperia Ray SO-03C; Sony Mobile Communications Inc., Tokyo, Japan; Android operating system 2.3; size, 111 3 53 3 9.4 mm; weight, 100 g) mounted to the torso at the level of the L3 spinous process with an elastic band. The Android operating system is commonly available on a wide variety of mobile devices. The smartphone contained a tri-axial acceleration sensor (BMA150; Bosch Sensortec GmbH, Reutlingen, Germany; range, 64 g; sensitivity, 128 LSB/g) and a recording medium. Commercially available application software was installed in the smartphone and used to record the acceleration signal measured by
1307
the embedded acceleration sensor. The sampling rate was set at SENSOR_DELAY_GAME, which is a unique mode for Android that corresponds to a sampling rate of approximately 50 Hz on the device that we used. The duration of the test set-up and measurement was 2-3 minutes.
Data Processing Acceleration time-series data in the AP and ML axes were processed in Excel 2007 (Microsoft Japan Co., Ltd., Tokyo, Japan) and R-2.15.2 (The R Foundation for Statistical Computing, Vienna, Austria) with package RSEIS, version 3.0-9. Data recorded in the vertical axis were excluded from analysis because of low reliability of some of the measures associated with these data in preliminary experiments (data not shown). The sampling rate of the acceleration data recorded in the embedded memory of the smartphone was not constant; therefore, the data were resampled at 50 Hz and filtered by a second-order Butterworth bandpass filter (.1-20 Hz) to avoid gravity component and aliasing. Initial contacts were identified by the characteristic sharp peaks in the AP acceleration signal that correspond to initial contact, as previously reported.34 Five consecutive strides (10 steps) in the central part of the walkway were used for analysis. Three outcome variables were quantified: the unbiased autocorrelation coefficient, the harmonic ratio, and interstride variability. These variables have been used in previous gait analysis studies by means of trunk accelerometry in stroke subjects and have had clinical relevance due to associations with hemiparetic severity and functional outcomes.19,20,23 The unbiased autocorrelation coefficient was calculated according to Moe-Nilssen and Helbostad35 and was used as an indicator of gait regularity. In brief, the acceleration time series was correlated with the overlapping portion of the same acceleration time series that had been phase shifted by the average stride time. Autocorrelation coefficient values range from 0 to 1, with 0 indicating no association and 1 indicating high gait regularity. The harmonic ratio was derived from the Fourier analysis of trunk acceleration, as described by Kavanagh and Menz,16 and was used as an indicator of walking smoothness, walking rhythmicity, or dynamic stability. A higher harmonic ratio indicates a smoother and more stable walking pattern. The interstride variability of trunk acceleration was quantified using the coefficient of variation of the root mean square (RMS) acceleration. RMS is a measure of the dispersion of the data relative to 0, and RMS acceleration represents the average magnitude of the acceleration.36 RMS acceleration was computed for each of the 5 strides selected for analysis, and the mean and standard deviation (SD) of these 5 strides were quantified for each participant. The coefficient of variation was then
T. ISHO ET AL.
1308
Table 1. Comparison of characteristics and clinical measures between stroke subjects with and without a history of falling
Age (y) Sex (female/male) Body mass index (kg/m2) Stroke type (hemorrhagic/ischemic) Time since stroke (mo) Hemiplegic side (left/right) Use of a single-point cane (no/yes) Use of an orthosis (no/yes) SPMSQ errors (0-10) BRS for lower extremity (III/IV/V/VI) SIAS lower extremity sensory score (0-6) Mini-BESTest score (0-28) TUG test time (s) 10MWT gait speed (m/s)
Fallers (n 5 11)
Nonfallers (n 5 13)
P value
70.5 6 12.5 7/4 26.6 6 4.3 2/9 59.7 6 39.0 6/5 2/9 7/4 2.1 6 1.2 4/1/1/5 5.3 6 0.6 15.5 6 4.1 17.9 6 5.9 .66 6 .25
72.5 6 6.9 5/8 22.5 6 3.0 4/9 75.9 6 38.3 9/4 3/10 6/7 1.5 6 1.1 6/3/1/3 4.5 6 1.7 14.5 6 3.1 21.1 6 7.9 .54 6 .18
.639 .414 .012* .649 .317 .675 1.000 .444 .266 .763 .482 .507 .271 .175
Abbreviations: BRS, Brunnstrom recovery stage; Mini-BESTest, Mini-Balance Evaluation Systems Test; SIAS, Stroke Impairment Assessment Set; SPMSQ, Short Portable Mental Status Questionnaire; TUG, Timed Up and Go; 10MWT, 10-m walking test. Values are expressed as mean 6 standard deviation or number of subjects. *Indicates statistical significance at the 5% level.
calculated for each participant as the SD divided by the mean and expressed as a percentage. Higher values of the coefficient of variation indicate greater interstride variability. The immediate test–retest reliability for accelerometry variables used in the present study was tested in preliminary experiments with 14 patients with hemiparesis after stroke and was moderate to good (ICC [1,1], .531-.900; P , .05).
Statistical Analysis Statistical analyses were performed using R-2.15.2 with package pROC, version 1.5.4. Data are expressed as mean 6 SD for continuous variables and as frequency distribution for categorical variables. The Shapiro–Wilk test was used to examine the normality of continuous variables. Demographic and clinical characteristics were compared across fallers and nonfallers using the independent t tests, Mann–Whitney U tests, and Fisher exact tests for parametric continuous, nonparametric continuous, and categorical variables, respectively. Accelerometry variables identified as significantly different between fallers and nonfallers were entered into a logistic regression analysis to examine the association of these variables with fall history. Crude and adjusted odds ratios (ORs) with 95% confidence interval (CI) were calculated. Demographic characteristics with P less than .20 in the univariate analysis were included as possible confounding variables to calculate the adjusted OR. Accelerometry variables identified as significantly associated with falling in the multiple logistic regression analysis were examined using receiver operating characteristic curve analysis to estimate the area under the curve (AUC) and its 95% CI.
The AUC is a measure of overall discriminative accuracy and ranges from .5 to 1.0, with values of .5-.7, .7-.9, and greater than .9 indicate low, moderate, and high accuracy, respectively.37 In addition, the optimal cutoff value was determined based on the best sensitivity and specificity combination. The level of statistical significance was set at P less than .05 for all statistical tests.
Results A total of 24 participants (mean age, 71.6 6 9.7 years; 12 men, 12 women) met the criteria and were enrolled in the study. Mean time since stroke was 68.5 6 38.7 months (range, 14-144 months). Sixteen participants were neighborhood ambulators, and 8 were independent community ambulators. Only 1 participant had vision impairment (hemianopia), but this participant had no history of falls in the past 12 months. Eleven participants reported falling in the past 12 months: 4 reported multiple falls and 7 reported a single fall. These 11 participants were classified as the fallers. The other 13 participants reported no falls in the past 12 months and were classified as nonfallers. Demographic and clinical characteristics and the results of clinical assessments are summarized in Table 1. With the exception of body mass index, demographic and clinical characteristics were similar in fallers and nonfallers. Clinical performance measures did not differ between fallers and nonfallers. The variables derived from trunk accelerometry are presented in Table 2. Interstride variability of ML trunk acceleration was significantly higher in fallers than in nonfallers (P 5 .024). There were no differences in other accelerometry variables between fallers and nonfallers.
GAIT ANALYSIS USING A SMARTPHONE
1309
Table 2. Comparison of accelerometry variables between stroke subjects with and without a history of falling Fallers (n 5 11) Autocorrelation coefficient AP .79 6 .11 ML .61 6 .21 Harmonic ratio AP 1.50 6 .85 ML 1.76 6 .66 Interstride variability (%) AP 5.9 6 2.9 ML 12.3 6 5.6
Nonfallers (n 5 13)
P value
.72 6 .12 .61 6 .18
.180 .947
1.20 6 .73 1.42 6 .47
.339 .160
6.3 6 2.6 7.6 6 3.8
.663 .024*
Abbreviations: AP, anteroposterior; ML, mediolateral. Values are expressed as mean 6 standard deviation. *Indicates statistical significance at the 5% level.
Univariate logistic regression analysis revealed that higher interstride variability of ML trunk acceleration was significantly associated with fall history (crude OR, 1.253; 95% CI, 1.005-1.561; P 5 .045). This association remained statistically significant after adjusting for body mass index (adjusted OR, 1.462; 95% CI, 1.009-2.120; P 5 .048). The AUC for interstride variability of ML trunk acceleration was .745 (95% CI, .527-.963; Fig 1), and the cutoff value was 11.6% with a moderate sensitivity of 72.7% (95% CI, 43.4-90.3) and specificity of 84.6% (95% CI, 57.8-95.7).
Discussion To our knowledge, this is the first study to investigate the association of trunk acceleration during gait with
Figure 1. Receiver operating characteristic curve for the interstride variability of trunk acceleration in the mediolateral direction to discriminate between fallers and nonfallers.
fall risk in individuals with chronic stroke, and also the first to use the smartphone to quantify trunk acceleration. Our results showed that the interstride variability of ML trunk acceleration was significantly different between fallers and nonfallers, whereas clinical performance measures were not. The interstride variability of ML trunk acceleration was also a significant explanatory variable for retrospective fall incidence and had moderate overall accuracy for discriminating fallers from nonfallers. These results suggest that accelerometry-based gait analysis can be used to identify poststroke individuals at risk of falling and is more sensitive than clinical performance measures for this purpose. In the logistic regression analysis, higher interstride variability of ML trunk acceleration was significantly associated with fall history. This finding supports previous reports that greater variability in spatiotemporal gait parameters was more sensitive than average or typical gait characteristics in predicting future falls in older adults38,39 and that the interstride variability of ML trunk acceleration was higher in stroke survivors who had to be supervised when walking than in stroke survivors who could walk independently.23 Interstride variability represents stride-to-stride fluctuations in walking and is influenced by cardiovascular factors and mental health in addition to physiological factors that affect gait dynamics such as neural control, muscle function, and postural control.40 Balasubramanian et al41 found that between-leg differences in step variability during hemiplegic walking, which suggest that increased interstride step variability after stroke is affected by poststroke neuromuscular impairments. Hence, greater interstride variability of ML trunk acceleration in stroke subjects at risk of falling may also relate to multiple impairments of the central nerve system including altered somatosensory inputs from hemiplegic side and hindrance of afferent feedback for motor coordination and postural control. Although interstride variability of ML trunk acceleration was higher in fallers than in nonfallers, the interstride variability of AP trunk acceleration was similar. Previous three-dimensional motion analysis and two-dimensional videographic studies demonstrated that subjects with stroke had greater amplitude and asymmetry of lateral pelvic displacement than healthy controls.42,43 Together, these results suggest that upper body dynamics during gait are affected by stroke, and the effect of stroke is greatest in the ML direction. The motion of upper body segments plays an important role in maintaining equilibrium and postural orientation during walking, and the motion of upper body segments after stroke, particularly in the frontal plane, may reflect difficulties controlling the lateral motion of the trunk segment. In receiver operating characteristic curve analysis, interstride variability of ML trunk acceleration showed moderate accuracy, sensitivity, and specificity for
T. ISHO ET AL.
1310
discriminating between fallers and nonfallers. Interstride variability of ML trunk acceleration may provide more discriminative accuracy when combined with other fall risk factors such as cognitive impairment, functional performance, vision impairment, and medication use. In spite of the significant difference in the interstride variability of ML trunk acceleration between fallers and nonfallers, there was no significant difference in the ML autocorrelation coefficient. The autocorrelation coefficient is affected by the variability of acceleration amplitude and by the temporal variability of gait cycle events, and the lack of group differences may be because of the temporal variability of gait cycle events. Similarly, there was no significant difference in the harmonic ratio of trunk acceleration between fallers and nonfallers. Iosa et al20 showed that the harmonic ratio was significantly lower in the AP direction in subjects with stroke than in young and age-matched controls and had a significant positive correlation with gait speed. The association between the harmonic ratio and gait speed may make it difficult to capture differences in the harmonic ratio between groups that have similar gait speed. There were no significant differences in clinical performance measures (Mini-BESTest score, TUG test, and gait speed) between fallers and nonfallers in the present study. These results agree with a few previous studies in which clinical performance measures failed to differentiate fallers from nonfallers,8,11,12 and suggest that these measures are insufficient to capture the subtle changes in gait patterns that are related to risk of falling. Although Tsang et al31 found that a Mini-BESTest score less than 17.5 discriminated individuals with chronic stroke who had a fall history from those who had not fallen (sensitivity, 64.0%; specificity, 64.2%; AUC, .64), participants in their study were younger (mean age, 57.1 6 11.0 years) than those in the present study. In the present study, the mean Mini-BESTest score was in the middle of the total range of possible scores (55.2 6 14.7% of the total possible score for fallers and 51.6 6 11.1% for nonfallers) and was lower than the cutoff value identified by Tsang et al.31 Future studies should include stroke subjects of varying demographic and clinical characteristics and with a wider range of MiniBESTest scores to provide further insight into the association of the Mini-BESTest with risk of falling. The findings of the present study show that quantitative gait assessment using the smartphone can provide detailed and objective information about subtle changes in the gait pattern of stroke subjects at risk of falling. This method has several advantages. First, it takes only a few minutes from set-up to completion of the measurement, and therefore, would be easy to use in clinical and community settings. Second, because smartphones are now widely available in the general population, using smartphones for gait assessment has the potential to provide a self-assessment tool that can be used for health
management. However, this would require development of the application software so that it can perform all steps of the process, from data collection to data processing, and creation of a reference database that includes normative values to allow identification of gait characteristics related to risk of falling in varied population groups. This study has several limitations. First, the fall events were determined according to retrospective self-report. Reliability of this method has not been established in the stroke population. In addition, this method may have been affected by recall bias because some participants in this study had mild cognitive impairment. Second, because of the small sample size, the results have low generalizability. Third, the study design was cross sectional, therefore, we could not evaluate the utility of accelerometry variables for predicting future falls or determine cutoff values for these variables to identify patients at risk of future falls. In conclusion, accelerometry-based gait analysis performed using a smartphone can identify differences in gait characteristics of individuals with chronic stroke with and without a history of falling. This assessment method may provide valuable complementary information to a multidisciplinary fall risk assessment in people with stroke. Acknowledgments: The authors thank the participating facilities and their rehabilitation staff for cooperation and assistance in data collection. The authors also thank the members of Professor Usuda’s laboratory for helpful discussions.
References 1. Langhorne P, Stott DJ, Robertson L, et al. Medical complications after stroke: a multicenter study. Stroke 2000;31: 1223-1229. 2. Weerdesteyn V, de Niet M, van Duijnhoven HJ, et al. Falls in individuals with stroke. J Rehabil Res Dev 2008;45: 1195-1213. 3. Forster A, Young J. Incidence and consequences of falls due to stroke: a systematic inquiry. BMJ 1995;311:83-86. 4. Watanabe Y. Fear of falling among stroke survivors after discharge from inpatient rehabilitation. Int J Rehabil Res 2005;28:149-152. 5. Ramnemark A, Nyberg L, Borssen B, et al. Fractures after stroke. Osteoporos Int 1998;8:92-95. 6. Dennis MS, Lo KM, McDowall M, et al. Fractures after stroke: frequency, types, and associations. Stroke 2002; 33:728-734. 7. Andersson AG, Kamwendo K, Seiger A, et al. How to identify potential fallers in a stroke unit: validity indexes of 4 test methods. J Rehabil Med 2006;38:186-191. 8. Belgen B, Beninato M, Sullivan PE, et al. The association of balance capacity and falls self-efficacy with history of falling in community-dwelling people with chronic stroke. Arch Phys Med Rehabil 2006;87:554-561. 9. Hyndman D, Ashburn A. People with stroke living in the community: attention deficits, balance, ADL ability and falls. Disabil Rehabil 2003;25:817-822.
GAIT ANALYSIS USING A SMARTPHONE 10. Mackintosh SF, Hill KD, Dodd KJ, et al. Balance score and a history of falls in hospital predict recurrent falls in the 6 months following stroke rehabilitation. Arch Phys Med Rehabil 2006;87:1583-1589. 11. Lamb SE, Ferrucci L, Volapto S, et al. Risk factors for falling in home-dwelling older women with stroke: the Women’s Health and Aging Study. Stroke 2003;34: 494-501. 12. Harris JE, Eng JJ, Marigold DS, et al. Relationship of balance and mobility to fall incidence in people with chronic stroke. Phys Ther 2005;85:150-158. 13. Blum L, Korner-Bitensky N. Usefulness of the Berg Balance Scale in stroke rehabilitation: a systematic review. Phys Ther 2008;88:559-566. 14. Mudge S, Stott NS. Outcome measures to assess walking ability following stroke: a systematic review of the literature. Physiotherapy 2007;93:189-200. 15. Berg K, Wood-Dauphinee S, Williams JI, et al. Measuring balance in the elderly: preliminary development of an instrument. Physiother Can 1989;41:304-311. 16. Kavanagh JJ, Menz HB. Accelerometry: a technique for quantifying movement patterns during walking. Gait Posture 2008;28:1-15. 17. Nishiguchi S, Yamada M, Nagai K, et al. Reliability and variability of gait analysis by android-based smartphone. Telemed J E Health 2012;18:292-296. 18. Yamada M, Aoyama T, Mori S, et al. Objective assessment of abnormal gait in patients with rheumatoid arthritis using a smartphone. Rheumatol Int 2012;32:3869-3874. 19. Mizuike C, Ohgi S, Morita S. Analysis of stroke patient walking dynamics using a tri-axial accelerometer. Gait Posture 2009;30:60-64. 20. Iosa M, Fusco A, Morone G, et al. Assessment of upperbody dynamics stability during walking in patients with acute stroke. J Rehabil Res Dev 2012;49:439-450. 21. Boonsinsukh R, Panichareon L, Phansuwan-Pujito P. Light touch cue through a cane improves pelvic stability during walking in stroke. Arch Phys Med Rehabil 2009; 90:919-926. 22. Boonsinsukh R, Panichareon L, Saengsirisuwan V, et al. Clinical identification for the use of light touch cues with a cane in gait rehabilitation poststroke. Top Stroke Rehabil 2011;18(Suppl 1):633-642. 23. Higashi Y, Yamakoshi K, Fujimoto T, et al. Quantitative evaluation of movement using the timed up-and-go test. IEEE Eng Med Biol Mag 2008;27:38-46. 24. Gibson MJ. Falls in later life. In: Kane RL, Evans JG, Macfadyen D, eds. Improving the health of older people: a world view. New York: Oxford Univ Press 1990:296-315. 25. Haga H, Yasumura S, Niino N, et al. An examination of two reporting methods of falls among the elderly living in the community. Nihon Koshu Eisei Zasshi 1996;43: 983-988. 26. Pfeiffer E. A short portable mental status questionnaire for the assessment of organic brain deficit in elderly patients. J Am Geriatr Soc 1975;23:433-441.
1311 27. Foreman MD. Reliability and validity of mental status questionnaires in elderly hospitalized patients. Nurs Res 1987;36:216-220. 28. Brunnstrom S. Motor testing procedures in hemiplegia: based on sequential recovery stages. Phys Ther 1966; 46:357-375. 29. Chino N, Sonoda S, Domen K, et al. Stroke impairment assessment set (SIAS). In: Chino N, Melvin JL, eds. Functional evaluation of stroke patients. Tokyo: SpringerVerlag Tokyo 1996:19-31. 30. Franchignoni F, Horak F, Godi M, et al. Using psychometric techniques to improve the Balance Evaluation Systems Test: the mini-BESTest. J Rehabil Med 2010;42: 323-331. 31. Tsang CS, Liao LR, Chung RC, et al. Psychometric properties of the Mini-Balance Evaluation Systems Test (Mini-BESTest) in community-dwelling individuals with chronic stroke. Phys Ther 2013;93:1102-1115. 32. Podsiadlo D, Richardson S. The timed ‘‘Up & Go’’: a test of basic functional mobility for frail elderly persons. J Am Geriatr Soc 1991;39:142-148. 33. Flansbjer UB, Holmback AM, Downham D, et al. Reliability of gait performance tests in men and women with hemiparesis after stoke. J Rehabil Med 2005;37: 75-82. 34. Zijlsta W, Hof AL. Assessment of spatio-temporal gait parameters from trunk accelerations during human walking. Gait Posture 2003;18:1-10. 35. Moe-Nilssen R, Helbostad JL. Estimation of gait cycle characteristics by trunk accelerometry. J Biomech 2004; 37:121-126. 36. Menz HB, Lord SR, Fitzpatrick RC. Acceleration patterns of the head and pelvis when walking on level and irregular surfaces. Gait Posture 2003;18:35-46. 37. Akobeng AK. Understanding diagnostic tests 3: receiver operating characteristic curves. Acta Paediatr 2007;96: 644-647. 38. Maki BE. Gait changes in older adults: predictors of falls or indicators of fear. J Am Geriatr Soc 1997;45:313-320. 39. Hausdorff JM, Rios D, Edelberg HK. Gait variability and fall risk in community-living older adults: a 1-year prospective study. Arch Phys Med Rehabil 2001;82: 1050-1056. 40. Hausdorff JM. Gait variability: methods, modeling and meaning. J Neuroeng Rehabil 2005;2:19. 41. Balasubramanian CK, Neptune RR, Kautz SA. Variability in spatiotemporal step characteristics and its relationship to walking performance post-stroke. Gait Posture 2009; 29:408-414. 42. Dodd KJ, Wrigley TV, Goldie PA, et al. Quantifying lateral pelvic displacement during walking. Clin Biomech (Bristol, Avon) 1998;13:371-373. 43. De Bujanda E, Nadeau S, Bourbonnais D. Pelvic and shoulder movements in the frontal plane during treadmill walking in adults with stroke. J Stroke Cerebrovasc Dis 2004;13:58-69.
Background: The smartphone, which contains inertial sensors, is currently available and affordable device and has the potential to provide a self-assessment tool for health management. The aims of this study were to use a smartphone to record trunk acceleration during walking and to compare accelerometry variables between poststroke subjects with and without a history of falling. Methods: This cross-sectional study was conducted in 2 day care centers for elderly adults. Twenty-four community-dwelling adults with chronic stroke (mean age, 71.6 6 9.7 years; mean time since stroke, 68.5 6 38.7 months) were enrolled. Acceleration of the trunk during walking was recorded in the anteroposterior and mediolateral directions and quantified using the autocorrelation coefficient, harmonic ratio, and interstride variability (coefficient of variation of root mean square acceleration). Fall history in the past 12 months was obtained by self-report. Results: Eleven participants (45.8%) reported at least one fall in the past 12 months and were classified as fallers. Fallers exhibited significantly higher interstride variability of mediolateral trunk acceleration than nonfallers. In the logistic regression analysis, interstride variability of mediolateral trunk acceleration was significantly associated with fall history (adjusted odds ratio, 1.462; 95% confidence interval, 1.009-2.120). The area under the receiver operating characteristic curve for interstride variability of mediolateral trunk acceleration to discriminate fallers from nonfallers was .745 (95% confidence interval, .527-.963). Conclusions: The results suggest that quantitative gait assessment using a smartphone can provide detailed and objective information about subtle changes in the gait pattern of stroke subjects at risk of falling. Key Words: Accelerometry— accidental falls—cerebrovascular disorders—gait—postural balance—walking. Ó 2015 by National Stroke Association
Introduction Falls are a major complication after stroke1 and occur more frequently in the community setting after hospital discharge than they do in the hospital setting.2 Falls can From the *Department of Rehabilitation, National Hospital Organization Takasaki General Medical Center, Takasaki, Gunma; †Department of Rehabilitation, Saitama Cooperative Hospital, Kawaguchi, Saitama; and ‡Department of Rehabilitation Sciences, Gunma University Graduate School of Health Sciences, Maebashi, Gunma, Japan. Received January 25, 2015; accepted February 5, 2015. The present address of H.T. is Department of Rehabilitation, Eniwa Hospital, Eniwa, Hokkaido, Japan. Institutions where the study was performed are the Geriatric Health Services Facility Minuma, Kawaguchi, and the Daycare Center of the Kumagaya Cooperative Hospital, Kumagaya, Saitama, Japan.
cause depression,3 fear of falling,4 and severe injuries such as hip fractures,5,6 all of which can restrict physical activities and cause further deterioration of health status.2 According to a literature review, between 23%
The authors received no grant support for this study. The authors have no conflicts of interest to declare. Address correspondence to Takuya Isho, PT, MSc, Department of Rehabilitation, National Hospital Organization Takasaki General Medical Center, 36 Takamatsu-cho, Takasaki, Gunma 370-0829, Japan. E-mail: [email protected]. 1052-3057/$ - see front matter Ó 2015 by National Stroke Association http://dx.doi.org/10.1016/j.jstrokecerebrovasdis.2015.02.004
Journal of Stroke and Cerebrovascular Diseases, Vol. 24, No. 6 (June), 2015: pp 1305-1311
1305
T. ISHO ET AL.
1306
and 73% of community-dwelling stroke survivors experienced at least 1 fall during a 3-12–month follow-up period, and walking was the most common activity that lead to falls.2 A number of previous studies have demonstrated that balance and mobility deficits measured with clinical assessment tools are important risk factors for falls.7-10 However, a few studies have reported that clinical performance measures fail to distinguish fallers from nonfallers.8,11,12 Harris et al12 showed that there was no relation between the Berg Balance Scale (BBS) score, gait speed, and fall history in the past 6 months in individuals with chronic stroke. Similarly, Belgen et al8 reported no significant difference in the Timed Up and Go (TUG) test between community-dwelling poststroke individuals with and without a history of falls. The BBS, gait speed measured over a short distance, and the TUG are valid and reliable clinical tools to assess functional ability in stroke patients.13,14 However, these tools may not be optimal to assess fall risk in these patients. The BBS predominantly tests balance in static positions and during transition movements,15 and does not contain many items that evaluate dynamic balance during walking activities. Gait speed and the TUG test are simplified quantitative mobility assessment that assess only the temporal aspect of gait, and therefore, may be insufficient to identify the gait characteristics of poststroke individuals at risk of falling. Objective assessment using technological devices plays an important role in the quantification of movement patterns. Recently, the use of inertial sensors such as accelerometers and gyroscopes for movement analysis has become popular in clinical practice because these sensors are inexpensive, small, and portable and testing is not restricted to a laboratory environment.16 Inertial sensors are contained in mobile devices such as smartphones, meaning that a smartphone may be used to assess gait. This has been attempted in a few studies.17,18 Several studies have assessed gait by measuring the acceleration of the upper body in stroke subjects. Accelerometry variables have been compared between stroke subjects and healthy controls19,20 and between stroke subjects walking with different conditions of cane use,21,22 and the association of accelerometry variables with motor and mobility functions and ambulatory independence has been evaluated in stroke subjects.19,20,23 However, no studies have yet assessed the clinical relevance of gait stability measures derived from upper body accelerometry to the risk of falling in individuals with stroke, nor have there been any studies that have used a smartphone to measure upper body acceleration during gait. The aims of the present study were to use a smartphone to record trunk acceleration during walking and to compare accelerometry variables between poststroke subjects with and without a history of falling.
Methods Participants This was a cross-sectional study conducted in 2 day care centers for elderly adults in Saitama, Japan. The Ethics Committee of the Saitama Cooperative Hospital approved all study procedures. All participants were informed about the study and gave written informed consent. Community-dwelling adults with chronic stroke receiving day care services were recruited and screened for inclusion and exclusion criteria. Inclusion criteria were (1) more than 12 months since stroke onset and (2) ability to walk 16 m independently with or without a singlepoint cane and/or an orthosis. Exclusion criteria were (1) severe cardiovascular, respiratory, musculoskeletal, or neurologic disorder other than stroke that affected gait performance; (2) unable to understand the instructions because of communication problem or moderateto-severe cognitive dysfunction (ie, 5 or more errors on the Short Portable Mental Status Questionnaire [SPMSQ]); and (3) household ambulators walked only indoors or only mobilized during rehabilitation sessions. Fall history is an independent predictor of future falls.10 Participants were divided into fallers and nonfallers according to whether or not they reported that they had experienced falling in the past 12 months. Falls were defined as ‘‘falling down to the ground or to the lower level against one’s will.’’24 Retrospective self-report of fall events in the previous 12 months has shown excellent agreement with 3-month interval recall method (% agreement, 98.8%) in Japanese community-dwelling older adults.25
Procedures Demographic and clinical characteristics were collected from medical records and interviews. All clinical assessments were carried out by a pair of trained physical therapists. The observer’s subjective assessment was judged by the consensus of both examiners. Participants performed each assessment once. Cognitive status was evaluated using the SPMSQ, which is a widely used 10-item cognitive examination26 and has shown good reliability and validity.27 A score of 5 or more errors on the SPMSQ indicates moderate-to-severe cognitive impairment; a score of 3-4 errors indicates mild cognitive impairment; and a score of 0, 1, or 2 errors indicates cognitively intact. Lower-extremity motor function was assessed according to the 6 motor stages defined by Brunnstrom,28 where lower stages indicate greater motor deficit. The test–retest reliability for lower-extremity Brunnstrom recovery stage was tested preliminarily in 17 hemispheric stroke patients with supratentorial lesions from medical record in our hospital and was weighted kappa 5 .879 (mean duration between test and retest, 2.3 6 1.4 days). Lower-extremity sensory impairment was measured with the sensory
GAIT ANALYSIS USING A SMARTPHONE
subscale of the Stroke Impairment Assessment Set (SIAS), which evaluates light touch sensation and thumb position sense on a 4-point scale.29 The total score of the sensory subscale of SIAS ranges from 0 to 6 points, with a lower score indicating a more severe sensory impairment. The inter-rater reliability of SIAS lower-extremity sensory items in patients with hemiparetic stroke was weighted kappa 5 .500 and .838 for touch and position, respectively.29 Physical performance was tested using the Mini-Balance Evaluation Systems Test (Mini-BESTest), the TUG, and the 10-m walking test (10MWT). The MiniBESTest is a clinical balance assessment tool designed to measure dynamic balance.30 It evaluates 4 different balance control systems: anticipatory postural adjustments, postural responses, sensory orientation, and balance during gait. It consists of 14 items and has a minimum score of 0 and a maximum score of 28, with higher scores indicating better balance ability. In comparison with the BBS, the Mini-BESTest contains items that evaluate functional stability during walking activities and has less ceiling effect.31 Intrarater and inter-rater reliability (intraclass correlation coefficient [ICC] 5 .97 and .96, respectively) and concurrent validity (correlation with the BBS, Spearman’s rho 5 .83) of the Mini-BESTest have been established in subjects with chronic stroke.31 The TUG is a test of functional mobility that measures the time required to rise from an armchair, walk 3 m at a comfortable and safe pace, turn around, walk back to the chair, and sit down again.32 Participants were allowed to use a single-point cane and/or an orthosis during the TUG test if necessary. For the 10MWT, participants were asked to walk a distance of 16 m at a self-selected comfortable speed with or without a single-point cane and/or an orthosis. The time taken to walk the central 10 m was measured by a digital stopwatch and used to calculate gait speed. The TUG and the 10MWT both have excellent test–retest reliability (ICC ..9) in community-dwelling individuals with stroke.33
Trunk Accelerometry Gait Analysis Trunk linear accelerations were measured along the anteroposterior (AP), mediolateral (ML), and vertical axes during the 10MWT. Acceleration was measured using the Android-based smartphone (Xperia Ray SO-03C; Sony Mobile Communications Inc., Tokyo, Japan; Android operating system 2.3; size, 111 3 53 3 9.4 mm; weight, 100 g) mounted to the torso at the level of the L3 spinous process with an elastic band. The Android operating system is commonly available on a wide variety of mobile devices. The smartphone contained a tri-axial acceleration sensor (BMA150; Bosch Sensortec GmbH, Reutlingen, Germany; range, 64 g; sensitivity, 128 LSB/g) and a recording medium. Commercially available application software was installed in the smartphone and used to record the acceleration signal measured by
1307
the embedded acceleration sensor. The sampling rate was set at SENSOR_DELAY_GAME, which is a unique mode for Android that corresponds to a sampling rate of approximately 50 Hz on the device that we used. The duration of the test set-up and measurement was 2-3 minutes.
Data Processing Acceleration time-series data in the AP and ML axes were processed in Excel 2007 (Microsoft Japan Co., Ltd., Tokyo, Japan) and R-2.15.2 (The R Foundation for Statistical Computing, Vienna, Austria) with package RSEIS, version 3.0-9. Data recorded in the vertical axis were excluded from analysis because of low reliability of some of the measures associated with these data in preliminary experiments (data not shown). The sampling rate of the acceleration data recorded in the embedded memory of the smartphone was not constant; therefore, the data were resampled at 50 Hz and filtered by a second-order Butterworth bandpass filter (.1-20 Hz) to avoid gravity component and aliasing. Initial contacts were identified by the characteristic sharp peaks in the AP acceleration signal that correspond to initial contact, as previously reported.34 Five consecutive strides (10 steps) in the central part of the walkway were used for analysis. Three outcome variables were quantified: the unbiased autocorrelation coefficient, the harmonic ratio, and interstride variability. These variables have been used in previous gait analysis studies by means of trunk accelerometry in stroke subjects and have had clinical relevance due to associations with hemiparetic severity and functional outcomes.19,20,23 The unbiased autocorrelation coefficient was calculated according to Moe-Nilssen and Helbostad35 and was used as an indicator of gait regularity. In brief, the acceleration time series was correlated with the overlapping portion of the same acceleration time series that had been phase shifted by the average stride time. Autocorrelation coefficient values range from 0 to 1, with 0 indicating no association and 1 indicating high gait regularity. The harmonic ratio was derived from the Fourier analysis of trunk acceleration, as described by Kavanagh and Menz,16 and was used as an indicator of walking smoothness, walking rhythmicity, or dynamic stability. A higher harmonic ratio indicates a smoother and more stable walking pattern. The interstride variability of trunk acceleration was quantified using the coefficient of variation of the root mean square (RMS) acceleration. RMS is a measure of the dispersion of the data relative to 0, and RMS acceleration represents the average magnitude of the acceleration.36 RMS acceleration was computed for each of the 5 strides selected for analysis, and the mean and standard deviation (SD) of these 5 strides were quantified for each participant. The coefficient of variation was then
T. ISHO ET AL.
1308
Table 1. Comparison of characteristics and clinical measures between stroke subjects with and without a history of falling
Age (y) Sex (female/male) Body mass index (kg/m2) Stroke type (hemorrhagic/ischemic) Time since stroke (mo) Hemiplegic side (left/right) Use of a single-point cane (no/yes) Use of an orthosis (no/yes) SPMSQ errors (0-10) BRS for lower extremity (III/IV/V/VI) SIAS lower extremity sensory score (0-6) Mini-BESTest score (0-28) TUG test time (s) 10MWT gait speed (m/s)
Fallers (n 5 11)
Nonfallers (n 5 13)
P value
70.5 6 12.5 7/4 26.6 6 4.3 2/9 59.7 6 39.0 6/5 2/9 7/4 2.1 6 1.2 4/1/1/5 5.3 6 0.6 15.5 6 4.1 17.9 6 5.9 .66 6 .25
72.5 6 6.9 5/8 22.5 6 3.0 4/9 75.9 6 38.3 9/4 3/10 6/7 1.5 6 1.1 6/3/1/3 4.5 6 1.7 14.5 6 3.1 21.1 6 7.9 .54 6 .18
.639 .414 .012* .649 .317 .675 1.000 .444 .266 .763 .482 .507 .271 .175
Abbreviations: BRS, Brunnstrom recovery stage; Mini-BESTest, Mini-Balance Evaluation Systems Test; SIAS, Stroke Impairment Assessment Set; SPMSQ, Short Portable Mental Status Questionnaire; TUG, Timed Up and Go; 10MWT, 10-m walking test. Values are expressed as mean 6 standard deviation or number of subjects. *Indicates statistical significance at the 5% level.
calculated for each participant as the SD divided by the mean and expressed as a percentage. Higher values of the coefficient of variation indicate greater interstride variability. The immediate test–retest reliability for accelerometry variables used in the present study was tested in preliminary experiments with 14 patients with hemiparesis after stroke and was moderate to good (ICC [1,1], .531-.900; P , .05).
Statistical Analysis Statistical analyses were performed using R-2.15.2 with package pROC, version 1.5.4. Data are expressed as mean 6 SD for continuous variables and as frequency distribution for categorical variables. The Shapiro–Wilk test was used to examine the normality of continuous variables. Demographic and clinical characteristics were compared across fallers and nonfallers using the independent t tests, Mann–Whitney U tests, and Fisher exact tests for parametric continuous, nonparametric continuous, and categorical variables, respectively. Accelerometry variables identified as significantly different between fallers and nonfallers were entered into a logistic regression analysis to examine the association of these variables with fall history. Crude and adjusted odds ratios (ORs) with 95% confidence interval (CI) were calculated. Demographic characteristics with P less than .20 in the univariate analysis were included as possible confounding variables to calculate the adjusted OR. Accelerometry variables identified as significantly associated with falling in the multiple logistic regression analysis were examined using receiver operating characteristic curve analysis to estimate the area under the curve (AUC) and its 95% CI.
The AUC is a measure of overall discriminative accuracy and ranges from .5 to 1.0, with values of .5-.7, .7-.9, and greater than .9 indicate low, moderate, and high accuracy, respectively.37 In addition, the optimal cutoff value was determined based on the best sensitivity and specificity combination. The level of statistical significance was set at P less than .05 for all statistical tests.
Results A total of 24 participants (mean age, 71.6 6 9.7 years; 12 men, 12 women) met the criteria and were enrolled in the study. Mean time since stroke was 68.5 6 38.7 months (range, 14-144 months). Sixteen participants were neighborhood ambulators, and 8 were independent community ambulators. Only 1 participant had vision impairment (hemianopia), but this participant had no history of falls in the past 12 months. Eleven participants reported falling in the past 12 months: 4 reported multiple falls and 7 reported a single fall. These 11 participants were classified as the fallers. The other 13 participants reported no falls in the past 12 months and were classified as nonfallers. Demographic and clinical characteristics and the results of clinical assessments are summarized in Table 1. With the exception of body mass index, demographic and clinical characteristics were similar in fallers and nonfallers. Clinical performance measures did not differ between fallers and nonfallers. The variables derived from trunk accelerometry are presented in Table 2. Interstride variability of ML trunk acceleration was significantly higher in fallers than in nonfallers (P 5 .024). There were no differences in other accelerometry variables between fallers and nonfallers.
GAIT ANALYSIS USING A SMARTPHONE
1309
Table 2. Comparison of accelerometry variables between stroke subjects with and without a history of falling Fallers (n 5 11) Autocorrelation coefficient AP .79 6 .11 ML .61 6 .21 Harmonic ratio AP 1.50 6 .85 ML 1.76 6 .66 Interstride variability (%) AP 5.9 6 2.9 ML 12.3 6 5.6
Nonfallers (n 5 13)
P value
.72 6 .12 .61 6 .18
.180 .947
1.20 6 .73 1.42 6 .47
.339 .160
6.3 6 2.6 7.6 6 3.8
.663 .024*
Abbreviations: AP, anteroposterior; ML, mediolateral. Values are expressed as mean 6 standard deviation. *Indicates statistical significance at the 5% level.
Univariate logistic regression analysis revealed that higher interstride variability of ML trunk acceleration was significantly associated with fall history (crude OR, 1.253; 95% CI, 1.005-1.561; P 5 .045). This association remained statistically significant after adjusting for body mass index (adjusted OR, 1.462; 95% CI, 1.009-2.120; P 5 .048). The AUC for interstride variability of ML trunk acceleration was .745 (95% CI, .527-.963; Fig 1), and the cutoff value was 11.6% with a moderate sensitivity of 72.7% (95% CI, 43.4-90.3) and specificity of 84.6% (95% CI, 57.8-95.7).
Discussion To our knowledge, this is the first study to investigate the association of trunk acceleration during gait with
Figure 1. Receiver operating characteristic curve for the interstride variability of trunk acceleration in the mediolateral direction to discriminate between fallers and nonfallers.
fall risk in individuals with chronic stroke, and also the first to use the smartphone to quantify trunk acceleration. Our results showed that the interstride variability of ML trunk acceleration was significantly different between fallers and nonfallers, whereas clinical performance measures were not. The interstride variability of ML trunk acceleration was also a significant explanatory variable for retrospective fall incidence and had moderate overall accuracy for discriminating fallers from nonfallers. These results suggest that accelerometry-based gait analysis can be used to identify poststroke individuals at risk of falling and is more sensitive than clinical performance measures for this purpose. In the logistic regression analysis, higher interstride variability of ML trunk acceleration was significantly associated with fall history. This finding supports previous reports that greater variability in spatiotemporal gait parameters was more sensitive than average or typical gait characteristics in predicting future falls in older adults38,39 and that the interstride variability of ML trunk acceleration was higher in stroke survivors who had to be supervised when walking than in stroke survivors who could walk independently.23 Interstride variability represents stride-to-stride fluctuations in walking and is influenced by cardiovascular factors and mental health in addition to physiological factors that affect gait dynamics such as neural control, muscle function, and postural control.40 Balasubramanian et al41 found that between-leg differences in step variability during hemiplegic walking, which suggest that increased interstride step variability after stroke is affected by poststroke neuromuscular impairments. Hence, greater interstride variability of ML trunk acceleration in stroke subjects at risk of falling may also relate to multiple impairments of the central nerve system including altered somatosensory inputs from hemiplegic side and hindrance of afferent feedback for motor coordination and postural control. Although interstride variability of ML trunk acceleration was higher in fallers than in nonfallers, the interstride variability of AP trunk acceleration was similar. Previous three-dimensional motion analysis and two-dimensional videographic studies demonstrated that subjects with stroke had greater amplitude and asymmetry of lateral pelvic displacement than healthy controls.42,43 Together, these results suggest that upper body dynamics during gait are affected by stroke, and the effect of stroke is greatest in the ML direction. The motion of upper body segments plays an important role in maintaining equilibrium and postural orientation during walking, and the motion of upper body segments after stroke, particularly in the frontal plane, may reflect difficulties controlling the lateral motion of the trunk segment. In receiver operating characteristic curve analysis, interstride variability of ML trunk acceleration showed moderate accuracy, sensitivity, and specificity for
T. ISHO ET AL.
1310
discriminating between fallers and nonfallers. Interstride variability of ML trunk acceleration may provide more discriminative accuracy when combined with other fall risk factors such as cognitive impairment, functional performance, vision impairment, and medication use. In spite of the significant difference in the interstride variability of ML trunk acceleration between fallers and nonfallers, there was no significant difference in the ML autocorrelation coefficient. The autocorrelation coefficient is affected by the variability of acceleration amplitude and by the temporal variability of gait cycle events, and the lack of group differences may be because of the temporal variability of gait cycle events. Similarly, there was no significant difference in the harmonic ratio of trunk acceleration between fallers and nonfallers. Iosa et al20 showed that the harmonic ratio was significantly lower in the AP direction in subjects with stroke than in young and age-matched controls and had a significant positive correlation with gait speed. The association between the harmonic ratio and gait speed may make it difficult to capture differences in the harmonic ratio between groups that have similar gait speed. There were no significant differences in clinical performance measures (Mini-BESTest score, TUG test, and gait speed) between fallers and nonfallers in the present study. These results agree with a few previous studies in which clinical performance measures failed to differentiate fallers from nonfallers,8,11,12 and suggest that these measures are insufficient to capture the subtle changes in gait patterns that are related to risk of falling. Although Tsang et al31 found that a Mini-BESTest score less than 17.5 discriminated individuals with chronic stroke who had a fall history from those who had not fallen (sensitivity, 64.0%; specificity, 64.2%; AUC, .64), participants in their study were younger (mean age, 57.1 6 11.0 years) than those in the present study. In the present study, the mean Mini-BESTest score was in the middle of the total range of possible scores (55.2 6 14.7% of the total possible score for fallers and 51.6 6 11.1% for nonfallers) and was lower than the cutoff value identified by Tsang et al.31 Future studies should include stroke subjects of varying demographic and clinical characteristics and with a wider range of MiniBESTest scores to provide further insight into the association of the Mini-BESTest with risk of falling. The findings of the present study show that quantitative gait assessment using the smartphone can provide detailed and objective information about subtle changes in the gait pattern of stroke subjects at risk of falling. This method has several advantages. First, it takes only a few minutes from set-up to completion of the measurement, and therefore, would be easy to use in clinical and community settings. Second, because smartphones are now widely available in the general population, using smartphones for gait assessment has the potential to provide a self-assessment tool that can be used for health
management. However, this would require development of the application software so that it can perform all steps of the process, from data collection to data processing, and creation of a reference database that includes normative values to allow identification of gait characteristics related to risk of falling in varied population groups. This study has several limitations. First, the fall events were determined according to retrospective self-report. Reliability of this method has not been established in the stroke population. In addition, this method may have been affected by recall bias because some participants in this study had mild cognitive impairment. Second, because of the small sample size, the results have low generalizability. Third, the study design was cross sectional, therefore, we could not evaluate the utility of accelerometry variables for predicting future falls or determine cutoff values for these variables to identify patients at risk of future falls. In conclusion, accelerometry-based gait analysis performed using a smartphone can identify differences in gait characteristics of individuals with chronic stroke with and without a history of falling. This assessment method may provide valuable complementary information to a multidisciplinary fall risk assessment in people with stroke. Acknowledgments: The authors thank the participating facilities and their rehabilitation staff for cooperation and assistance in data collection. The authors also thank the members of Professor Usuda’s laboratory for helpful discussions.
References 1. Langhorne P, Stott DJ, Robertson L, et al. Medical complications after stroke: a multicenter study. Stroke 2000;31: 1223-1229. 2. Weerdesteyn V, de Niet M, van Duijnhoven HJ, et al. Falls in individuals with stroke. J Rehabil Res Dev 2008;45: 1195-1213. 3. Forster A, Young J. Incidence and consequences of falls due to stroke: a systematic inquiry. BMJ 1995;311:83-86. 4. Watanabe Y. Fear of falling among stroke survivors after discharge from inpatient rehabilitation. Int J Rehabil Res 2005;28:149-152. 5. Ramnemark A, Nyberg L, Borssen B, et al. Fractures after stroke. Osteoporos Int 1998;8:92-95. 6. Dennis MS, Lo KM, McDowall M, et al. Fractures after stroke: frequency, types, and associations. Stroke 2002; 33:728-734. 7. Andersson AG, Kamwendo K, Seiger A, et al. How to identify potential fallers in a stroke unit: validity indexes of 4 test methods. J Rehabil Med 2006;38:186-191. 8. Belgen B, Beninato M, Sullivan PE, et al. The association of balance capacity and falls self-efficacy with history of falling in community-dwelling people with chronic stroke. Arch Phys Med Rehabil 2006;87:554-561. 9. Hyndman D, Ashburn A. People with stroke living in the community: attention deficits, balance, ADL ability and falls. Disabil Rehabil 2003;25:817-822.
GAIT ANALYSIS USING A SMARTPHONE 10. Mackintosh SF, Hill KD, Dodd KJ, et al. Balance score and a history of falls in hospital predict recurrent falls in the 6 months following stroke rehabilitation. Arch Phys Med Rehabil 2006;87:1583-1589. 11. Lamb SE, Ferrucci L, Volapto S, et al. Risk factors for falling in home-dwelling older women with stroke: the Women’s Health and Aging Study. Stroke 2003;34: 494-501. 12. Harris JE, Eng JJ, Marigold DS, et al. Relationship of balance and mobility to fall incidence in people with chronic stroke. Phys Ther 2005;85:150-158. 13. Blum L, Korner-Bitensky N. Usefulness of the Berg Balance Scale in stroke rehabilitation: a systematic review. Phys Ther 2008;88:559-566. 14. Mudge S, Stott NS. Outcome measures to assess walking ability following stroke: a systematic review of the literature. Physiotherapy 2007;93:189-200. 15. Berg K, Wood-Dauphinee S, Williams JI, et al. Measuring balance in the elderly: preliminary development of an instrument. Physiother Can 1989;41:304-311. 16. Kavanagh JJ, Menz HB. Accelerometry: a technique for quantifying movement patterns during walking. Gait Posture 2008;28:1-15. 17. Nishiguchi S, Yamada M, Nagai K, et al. Reliability and variability of gait analysis by android-based smartphone. Telemed J E Health 2012;18:292-296. 18. Yamada M, Aoyama T, Mori S, et al. Objective assessment of abnormal gait in patients with rheumatoid arthritis using a smartphone. Rheumatol Int 2012;32:3869-3874. 19. Mizuike C, Ohgi S, Morita S. Analysis of stroke patient walking dynamics using a tri-axial accelerometer. Gait Posture 2009;30:60-64. 20. Iosa M, Fusco A, Morone G, et al. Assessment of upperbody dynamics stability during walking in patients with acute stroke. J Rehabil Res Dev 2012;49:439-450. 21. Boonsinsukh R, Panichareon L, Phansuwan-Pujito P. Light touch cue through a cane improves pelvic stability during walking in stroke. Arch Phys Med Rehabil 2009; 90:919-926. 22. Boonsinsukh R, Panichareon L, Saengsirisuwan V, et al. Clinical identification for the use of light touch cues with a cane in gait rehabilitation poststroke. Top Stroke Rehabil 2011;18(Suppl 1):633-642. 23. Higashi Y, Yamakoshi K, Fujimoto T, et al. Quantitative evaluation of movement using the timed up-and-go test. IEEE Eng Med Biol Mag 2008;27:38-46. 24. Gibson MJ. Falls in later life. In: Kane RL, Evans JG, Macfadyen D, eds. Improving the health of older people: a world view. New York: Oxford Univ Press 1990:296-315. 25. Haga H, Yasumura S, Niino N, et al. An examination of two reporting methods of falls among the elderly living in the community. Nihon Koshu Eisei Zasshi 1996;43: 983-988. 26. Pfeiffer E. A short portable mental status questionnaire for the assessment of organic brain deficit in elderly patients. J Am Geriatr Soc 1975;23:433-441.
1311 27. Foreman MD. Reliability and validity of mental status questionnaires in elderly hospitalized patients. Nurs Res 1987;36:216-220. 28. Brunnstrom S. Motor testing procedures in hemiplegia: based on sequential recovery stages. Phys Ther 1966; 46:357-375. 29. Chino N, Sonoda S, Domen K, et al. Stroke impairment assessment set (SIAS). In: Chino N, Melvin JL, eds. Functional evaluation of stroke patients. Tokyo: SpringerVerlag Tokyo 1996:19-31. 30. Franchignoni F, Horak F, Godi M, et al. Using psychometric techniques to improve the Balance Evaluation Systems Test: the mini-BESTest. J Rehabil Med 2010;42: 323-331. 31. Tsang CS, Liao LR, Chung RC, et al. Psychometric properties of the Mini-Balance Evaluation Systems Test (Mini-BESTest) in community-dwelling individuals with chronic stroke. Phys Ther 2013;93:1102-1115. 32. Podsiadlo D, Richardson S. The timed ‘‘Up & Go’’: a test of basic functional mobility for frail elderly persons. J Am Geriatr Soc 1991;39:142-148. 33. Flansbjer UB, Holmback AM, Downham D, et al. Reliability of gait performance tests in men and women with hemiparesis after stoke. J Rehabil Med 2005;37: 75-82. 34. Zijlsta W, Hof AL. Assessment of spatio-temporal gait parameters from trunk accelerations during human walking. Gait Posture 2003;18:1-10. 35. Moe-Nilssen R, Helbostad JL. Estimation of gait cycle characteristics by trunk accelerometry. J Biomech 2004; 37:121-126. 36. Menz HB, Lord SR, Fitzpatrick RC. Acceleration patterns of the head and pelvis when walking on level and irregular surfaces. Gait Posture 2003;18:35-46. 37. Akobeng AK. Understanding diagnostic tests 3: receiver operating characteristic curves. Acta Paediatr 2007;96: 644-647. 38. Maki BE. Gait changes in older adults: predictors of falls or indicators of fear. J Am Geriatr Soc 1997;45:313-320. 39. Hausdorff JM, Rios D, Edelberg HK. Gait variability and fall risk in community-living older adults: a 1-year prospective study. Arch Phys Med Rehabil 2001;82: 1050-1056. 40. Hausdorff JM. Gait variability: methods, modeling and meaning. J Neuroeng Rehabil 2005;2:19. 41. Balasubramanian CK, Neptune RR, Kautz SA. Variability in spatiotemporal step characteristics and its relationship to walking performance post-stroke. Gait Posture 2009; 29:408-414. 42. Dodd KJ, Wrigley TV, Goldie PA, et al. Quantifying lateral pelvic displacement during walking. Clin Biomech (Bristol, Avon) 1998;13:371-373. 43. De Bujanda E, Nadeau S, Bourbonnais D. Pelvic and shoulder movements in the frontal plane during treadmill walking in adults with stroke. J Stroke Cerebrovasc Dis 2004;13:58-69.