Work

Reliability and Validity of an Accelerometer-based Balance Assessment for Fall Risk Screening

Year: 2013
Type: article
Abstract: Postural sway occurs spontaneously during static stance as a result of the unstable properties of the body.In balance dysfunction, often a consequence of the normal aging process, the amplitude and frequency of this sway may exceed the limits of stability and contribute to a fall.Therefore, it is important to develop reliable and valid tools to assess postural sway and fall risk.When compared with force plate center of pressure (COP) measurements (one of the most commonly used postural sway assessment tools), accelerometers have been as good as, or better at discriminating postural sway differences between populations with and without balance deficits and between tasks requiring different levels of balance control.A wireless tri-axial accelerometer has recently come on the market that is smaller, less expensive, and more sensitive than those previously utilized in postural control studies.However, no reliability or validity studies on this device exist.Therefore, we aimed to assess the new accelerometer's 1) test-retest reliability, 2) ability to discriminate between tasks with differing sensory conditions (i.e., visual and somatosensory), 3) ability to discriminate between individuals who are fallers versus non-fallers , and 4) and its concurrent validity with other fall risk screening tools in older adult populations.iii Twenty healthy and independent living older adults (mean age 81±4 years; 8 male and 12 female) participated in the reliability and balance task discrimination study.An accelerometer was taped to the lower back of the participant with a gait belt positioned just below.Participants completed three 30-second trials of the four classic Romberg conditions (standing on a firm or compliant surface with eyes open or closed) in random order.Following a 20 minute rest, participants underwent a second block of testing for a total of 24 trials.Raw data was collected at approximately 250 Hz and low-pass filtered at 55Hz.Data were transformed to correct for pelvic tilt and adjust for low frequency.Following processing of raw data, root mean squared (RMS), with the unit gravity (g), for the anterior-posterior (A-P) and mediolateral (M-L) acceleration data was calculated.Intraclass Correlation Coefficients (ICC) were used to test for reliability, while repeated measures analysis of variance (ANOVA) models were used to test for a main effect of balance condition on the A-P and M-L acceleration RMS.To validate the accelerometer-based balance assessment (ABA) as a fall risk screening tool, we recruited 95 healthy older adults (mean age 86±6 years) residing in five different independent living facilities.Participants reported the number of falls they had in the previous six months prior to testing.Participants then completed, in random order, the Berg Balance Scale (BBS), Timed up-and-go (TUG), Activities-specific Balance Confidence (ABC) scale, and the ABA.The ABA consisted of the same four Romberg conditions used in the reliability assessment with and without a cognitive task (counting backward by 3's) for a total of eight balance conditions.The best attempt of two successful trials (the one with the lowest RMS values) for each condition was iv included in the data analysis.The fall risk discriminative power of various combinations of the TUG, BBS, ABC, and/or the ABA was evaluated using logistic regression and chisquare analyses.Spearman's rank correlation coefficients were used to investigate the relationship between RMS, BBS, and the ABC scale measures.Pearson's productmoment correlation was used to determine the relationship between RMS and TUG scores.ICCs were all good to excellent with values ranging from 0.736 to 0.972 for trialto-trial and from 0.760 to 0.954 for block-to-block.There was a significant stepwise increase in A-P and M-L acceleration RMS from conditions 1 to 4. With respect to A-P sway, the mean acceleration RMS increases from conditions 1 to 2, 2 to 3, and 3 to 4 were 0.0031±0.0008g,0.0035±0.0011g,and 0.0075±0.0022g,respectively.Similarly, the M-L mean acceleration RMS increased significantly from conditions 1 to 2, 2 to 3, and 3 to 4 (0.0025±0.0007g, 0.0068±0.0010g,and 0.0084±0.0018g,respectively).A-P and M-L acceleration RMS were statistically similar when standing on a firm surface.However, M-L acceleration RMS significantly exceeded A-P for condition 3 (F(1, 19) = 4.61, p = 0.045) and condition 4 (F(1, 19) = 6.68, p < 0.018).For the prediction of fall risk, all prediction models with the ABA, alone or in combination with the TUG, BBS, and ABC scale, had AUC's ≥ 0.80.Models without the ABA had AUC's in the range of 0.45 to 0.75.In general, most of the prediction models were better at correctly identifying non-fallers than fallers, indicated by substantially larger specificity than sensitivity values.Only for the comparison of multiple fallers to non-fallers did a model, the combination of the TUG, BBS, and ABC v scale with the ABA, exhibit both good sensitivity (85.0) and specificity (82.5).No clinical test on its own or in combination exhibited good sensitivity (all ≤ 61.1).Surprisingly, the associations between the ABA and the TUG, BBS, and ABC scale were not only insignificant but extremely weak with correlation coefficients of 0.11, -0.04, and -0.02 for the TUG, BBS, and ABC scale, respectively.As expected, the TUG, BBS, and ABC scale were all significantly correlated with each other and in the proper direction.In summary, the accelerometer exhibited good to excellent trial-to-trial and blockto-block reliability.The accelerometer was also able to discriminate between visual and supporting surface conditions and acceleration axes.Though not significantly correlated with the clinical tests, the ABA alone was better than the TUG, BBS, and/or ABC at discriminating fallers from non-fallers.vi Acknowledgments Professionally, I would to thank Dr. Devor for his guidance during my growth as a scholar and the undeniable influence he has had on my development as a teacher.The research presented here would not exist without the many contributions of a great number of people.Drs.Kloos and Kegelmeyer, thank you for helping me bridge the gap between my knowledge of exercise science and a useful and practical extension of that knowledge in my work with older adults.Panos Koutakis and Nick Hanson, thank you for helpingme turn a noisy accelerometer signal into convenient and useable information.Finally, in no particular order of contribution, the following students were integral to the success of this study as they interacted with participants, collected data, and in some cases even contributed to the study design and data interpretation: Brenna Congeni, Jessica Dicke, Jessica Fortney, Deanne Gauch, and Josh Stone.On a personal level, my family has been unbelievably supportive and patient during the last 10 consecutive years of higher education.Some colleagues have commented that it must be challenging to succeed in an academic environment with a wife and children.For me it has been just the opposite.Thank you Karyn, Kayleigh, and Brie for providing me with a much needed distraction and adding perspective.It is not in spite of you, but because of you that I have achieved this level of success.
Language: English
Open Access status: green
OpenAlex ID: W2519446483
FWCI: 0.39
Cites: 100
Cited by: 3
Related to: 10