RESNA Annual Conference - 2019

Using Kinect Camera To Quantify Gait Variables That Can Predict Falls In Older Adults With Dementia

 Sina Mehdizadeh1, Elham Dolatabadi1, Twinkle Arora1, Kimberley-Dale Ng1, Babak Taati1,2, Andrea Iaboni1,2

1Toronto Rehabilitation Institute, University Health Network, Toronto, ON, Canada; 2University of Toronto, Toronto, ON, Canada

INTRODUCTION

Dementia is a progressive neurodegenerative disorder with cognitive and motor symptoms [1]. Walking in people with dementia is of high safety concern as gait deficits are associated with increased risk of falls. People with dementia who have gait abnormalities are approximately three times more likely to fall compared to those with a normal gait [2]. Therefore, investigating gait characteristics of this population could be informative in estimating and predicting their risk of falls [3].

Previous studies have mainly quantified spatiotemporal gait variables in people with dementia [3]. However, impairment in spatiotemporal measures could reflect cautious gait arising from a fear of falling in this population and not the cause of falls. Many of these studies assess gait in a laboratory environment or at a single time point and may not reflect gait in a natural environment. Frequent, assessment of gait via ambient sensing enables longitudinal monitoring gait quality – including gait stability, symmetry, and smoothness – to determine its relationship to falls in dementia.

The aim of the present study was therefore to investigate the relationship between gait features extracted from naturalistic vision-based gait assessments to future falls in older people with dementia.

METHODS

Participants

This study took place in the Specialized Dementia Unit at Toronto Rehabilitation Institute, Toronto, Canada, a tertiary dementia care unit. Participants were 52 people with dementia admitted to this unit for treatment of behavioral and psychological symptoms of dementia. The following information and tests were gathered upon enrollment to the study: demographic data, admission neuropsychiatric inventory score (NPI), history of falls, severe impairment battery-short version (SIBS), Tinetti performance oriented mobility assessment (POMA) for balance and gait, and the Katz index of independence in activities of daily living (KATZ). Fall incidents during the inpatient admission were recorded from the medical chart.

Data recoding

A Microsoft Kinect for Windows v2 sensor (Microsoft, Redmond, WA, USA) was mounted on the ceiling of a hallway in the unit. Participants’ walking was captured automatically when they walked within the view of the sensor. The 3D joint motions during walks – as captured by the Kinect – were used for further analyses.

Data analysis

Thirty measures of gait in five categories were calculated, including spatiotemporal, variability, symmetry, stability, and trunk acceleration frequency domain measures. Separate uni- and multivariate Poisson regression analyses adjusted for age and sex and exposure variable were completed using number of falls during the admission as the dependent variable and gait and clinical variables as the predictors.

RESULTS

Results of the gait variables regression

The results of the univariate regression analyses for the gait variables revealed that step time symmetry, stability (quantified using margin of stability method), and the percent of acceleration power below 0.7 Hz (power <0.7 Hz) in mediolateral (ML) direction had the highest R2 as well as p-values <0.05. The multivariate regression analysis of the gait variables led to the step time symmetry and stability being significantly important gait variables for the final model (Table 1).

Results of the clinical variables regression

 

Table 1: Results of the multivariate Poisson regression with all important gait variables (R2=0.22). * indicates statistically significant at p<0.05.

Gait Variable
Regression Coefficient Standard Error t-Statistics p-Value
Step time symmetry* 0.07 0.03 2.22 0.02
Stability (Margin of stability)* -19.03 7.72 -2.466 0.01
Power<0.7 in ML -2.50 2.90 -0.86 0.38

Initial univariate regression analyses revealed that POMA for balance, SIBS, NPI, fall history, and number of walks had significant contributions to predict the number of falls. The multivariate analysis of the clinical data (adjusted for age and sex) showed that POMA for balance, NPI, and number of walks were statistically important variables in predicting the number of falls (Table 2). However, it was decided to include fall history in the final model as previous studies reported it is an important predictor of future falls [4].

Final model

The results of the final Poisson regression analysis (Table 3) revealed that stability (negatively), sex, NPI, and the number of walks (negatively) were statistically significant in predicting the number of falls (R2=0.65). The final model predictors were compared between males and females and the results indicated that females had significantly lower stability and higher NPI (p<0.01) but males were relatively more frequent fallers although the number of falls was the same for the two gender groups (39 falls per sex groups).

DISCUSSION

Table 2: Results of the multivariate Poisson regression with all important clinical variables (R2=0.60). * indicates statistically significant at p<0.05.

Variable
Regression Coefficient Standard Error t-Statistics p-Value
Age -0.01 0.01 -0.82 0.40
Sex 0.52 0.27 1.90 0.05
POMA for balance* -0.15 0.04 -3.76 0.0001
SIBS -0.01 0.008 -1.80 0.07
NPI* 0.01 0.008 2.20 0.02
Fall History 0.35 0.27 1.29 0.19
Number of walks* -0.02 0.006 -3.75 0.0001

In our study, we demonstrated that only gait stability (margin of stability) can successfully predict number of falls in people with dementia. To maintain walking stability, one should be able to control the body center of mass (COM) within the boundaries of the base of support (BOS). More specifically, a shorter distance between the COM and the boundaries of BOS will result in a reduced margin of stability, which implies a lower ability to deal with internal and external perturbations while walking, and consequently corresponds with an increased risk of falling.

Our results also showed that higher NPI is a predictor of number of falls in people with dementia. Higher NPI scores imply the presence of clinically significant motor agitation, impulsive or aggressive behaviors.  These types of behaviors are associated with poor judgement or risk-taking in movement and may thus be directly associated with falling.  People who are more agitated are also more likely to made sudden or unpredictable movements, which may tax their abilities to maintain stability.  Finally, agitated patients are more likely to receive psychotropic medications and at higher doses.  Future studies will examine the influence of medications on stability.

From the perspective of feasibility as a clinical tool for falls risk assessment, our study demonstrated that the a system comprising of a single Kinect camera can successfully quantify and monitor gait characteristics of older people with dementia which makes it feasible to be used in the clinical settings.

CONCLUSIONS

Our study demonstrated that a gait monitoring system comprising of a single Kinect camera can successfully quantify and monitor gait stability and fall risk of older people with dementia which makes it feasible to be used in the clinical settings.

Table 3: Results of final multivariate Poisson regression model with all important gait and clinical variables (R2=0.65). * indicates statistically significant at p<0.05.
Variable Regression Coefficient Standard Error t-Statistics p-Value
Step time symmetry 0.05 0.04 1.16 0.24
Stability (Margin of stability)* -19.25 7.18 -2.68 0.007
Age -0.02 0.01 -1.21 0.22
Sex* 0.96 0.33 2.86 0.004
POMA for balance -0.09 0.05 -1.87 0.06
NPI* 0.02 0.008 2.51 0.01
Fall History 0.47 0.28 1.66 0.09
No. of walks* -0.02 0.006 -3.28 0.001

REFERENCES

[1] Scott KR, Barrett AM. Dementia syndromes: Evaluation and treatment. Expert Rev Neurother. 2007 7: 407-22.

[2] Barba AL, Kelly Changizi B, Higgins DS, Factor SA, Molho ES. Dementia. Parkinson's Disease, 2012 338:413-33.

[3] Dolatabadi E, Van Ooteghem K, Taati B, Iaboni A. Quantitative Mobility Assessment for Fall Risk Prediction in Dementia: A Systematic Review. Dement Geriatr Cogn Disord. 2018 45:353-67.

[4] Fernando E, Fraser M, Hendriksen J, Kim CH, Muir-Hunter SW. Risk Factors Associated with Falls in Older   Adults with Dementia: A Systematic Review. Physiother Can. 2017 69:161-70.

[5] Kikkert LHJ, De Groot MH, Van Campen JP, Beijnen JH, Hortobágyi T, Vuillerme N, et al. Gait dynamics to            optimize fall risk assessment in geriatric patients admitted to an outpatient diagnostic clinic. PLoS ONE. 2017 12:1-14.