https://orcid.org/0000-0001-8682-6054
Figures
Abstract
The activities-specific balance confidence scale (ABC) assesses balance confidence during common activities. While low balance confidence can result in activity avoidance, excess confidence can increase fall risk. People with lower limb amputations can present with inconsistent gait, adversely affecting their balance confidence. Previous research demonstrated that clinical outcomes in this population (e.g., stride parameters, fall risk) can be determined from smartphone signals collected during walk tests, but this has not been evaluated for balance confidence. Fifty-eight (58) individuals with lower limb amputation completed a six-minute walk test (6MWT) while a smartphone at the posterior pelvis was used for signal collection. Participant ABC scores were categorized as low confidence or high confidence. A random forest classified ABC groups using features from each step, calculated from smartphone signals. The random forest correctly classified the confidence level of 47 of 58 participants (accuracy 81.0%, sensitivity 63.2%, specificity 89.7%). This research demonstrated that smartphone signal data can classify people with lower limb amputations into balance confidence groups after completing a 6MWT. Integration of this model into the TOHRC Walk Test app would provide balance confidence classification, in addition to previously demonstrated clinical outcomes, after completing a single assessment and could inform individualized rehabilitation programs to improve confidence and prevent activity avoidance.
Author summary
People with lower limb amputations have a high risk of falling and typically report low balance confidence levels. Low balance confidence can prevent people from engaging in physical and social activities. Previous research demonstrated that smartphone signals collected during walking can be used with machine learning to predict clinical outcomes in these individuals (e.g., stride parameters, fall risk). In this research, we investigated if balance confidence scores, categorized into high confidence and low confidence groups, could be classified. We demonstrate that smartphone signal data can stratify people with a lower limb amputation into high and low balance confidence groups after they complete a 6-minute walk test, correctly classifying over 80% of participants. The paper also discusses the relationship between balance confidence and ability, and how identifying those with mismatched balance ability and confidence scores (i.e., high confidence, but poor balance ability) could improve outcomes.
Citation: Juneau P, Baddour N, Burger H, Lemaire ED (2024) Balance confidence classification in people with a lower limb amputation using six minute walk test smartphone sensor signals. PLOS Digit Health 3(8): e0000570. https://doi.org/10.1371/journal.pdig.0000570
Editor: Walter Karlen, University of Ulm, GERMANY
Received: November 17, 2023; Accepted: July 1, 2024; Published: August 26, 2024
Copyright: © 2024 Juneau et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The dataset in this research includes sensitive human participant data containing identifiable and personal medical information which cannot be shared publicly at this time. However, this dataset is in the process of de-identification and can be made available upon request. Requests can be sent to the University of Ottawa Research Data Management Team (rdm@uOttawa.ca).
Funding: This research was funded by Natural Sciences and Engineering Research Council of Canada (NSERC). NSERC CREATE READI: RGPIN-2019-04106, EDL, https://carleton.ca/readi/ NSERC CREATE BEST 482728-2016CREAT, NB, http://create-best.com/#focus The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
1 Introduction
Balance confidence has been defined as “an individual’s confidence in their ability to maintain their balance while performing various activities” [1]. Balance confidence can have a broad effect on quality of life; including, decreased health status, functional decline, depression and anxiety, and activity avoidance [2–5]. Clinically, the activities-specific balance confidence scale (ABC) has become a common tool to assess confidence when moving within a person’s chosen environment.
ABC is a 16-item questionnaire that evaluates a person’s confidence in completing different activities that may challenge their balance; such as, walking indoors and out-doors on different surfaces, using stairs, and reaching for items overhead under different conditions (Table 1) [6]. Participants are asked to rate their confidence performing each of these activities without becoming unsteady, on a scale of 0% (no confidence) to 100% (complete confidence). If the person normally uses a mobility aid or other support while performing these activities, they are encouraged to complete the questionnaire as if they had access to their aids while performing the activity. The ABC scale is also used to evaluate various at-risk populations and is considered a reliable balance confidence measure [7–10] for people with lower limb amputations [10]. The final score is an average across each activity out of 100%. A score of 80% or greater is considered high confidence and can indicate higher physical function and decreased fall risk [11].
People with lower limb amputations present with increased gait variability and instability that put them at elevated risk of falls during all stages of rehabilitation [12]. They are at a greater risk of falls compared to age-matched able-bodied populations [13]. These individuals report low balance confidence scores, with some research reporting average ABC scores between 62.8 and 63.8% [14,15,16]. People who had a vascular amputation reported lower average scores (50.6–54.1%) compared to those whose amputation was due to a different cause [7,14]. Asano et al. [15] demonstrated that an ABC score of <80% significantly decreased quality of life scores in people with lower limb amputations and indicated that the ability to move about one’s environment safely was moderately related to their perceived quality of life.
While high confidence can indicate higher functional ability in older adults, greater balance ability in people with lower limb amputations can increase fall risk. Moreover, high balance confidence imposes an elevated risk for this population if this confidence is mismatched with the individual’s ability. One study reported that, for people with a lower limb amputation, risk of falling was 3.7 times greater if balance ability was low, but confidence scores were high [17]. Steinberg et al. [12] concluded that excess confidence in balance and walking abilities, less caution during stair walking, and higher gait variability increased fall risk in people with lower limb amputation. Early identification of those at elevated risk of falls, particularly those with mismatched balance confidence and ability is important to ensure that preventative strategies, such as education, exercise, or monitoring, can be implemented.
In previous research, Juneau et al. [18] demonstrated that important clinical outcomes for people with lower limb amputation could be calculated using smartphone sensor signals collected during a 6-minute walk test (6MWT). Stride parameters (e.g., step time, stride time) and step-based features were used to train a random forest to classify fall risk. Participants were identified as “fall risk” if they had at least one self-reported fall in the six months prior to data collection, consistent with clinical assessments. Stride parameter calculation and fall risk classification were equivalent to results achieved when manually-labelled foot strikes were used, indicating that automated foot strikes and smartphone signals were a clinically viable approach. The 6MWT is a commonly used movement assessment that evaluates functional capacity in several populations, including people with lower limb amputations. Despite not being designed as a fall risk measure, the research in [18] illustrated that additional clinical information could be extracted from the 6MWT task.
Understanding a person’s perception of their balance abilities is an essential part of the rehabilitation process in recovery from a fall and confidence levels can affect behaviours that could lead to future falls. In this paper, we are interested in classifying an ABC score from 6MWT data. To achieve this task, we must first choose between using a classification or regression approach. The choice depends on the specific goals, categorization, or prediction. In this research, a random forest algorithm was selected. Random forests are machine learning algorithms that consist of multiple decision tree classifiers. Each decision tree in the ensemble provides an individual classification result. The final classification is the class with the most predictions across all trees. Our goal is to categorize balance confidence, thereby enhancing the information a clinician can obtain after a 6MWT. Furthermore, the relationship between the 6MWT and ABC score is likely nonlinear, which lends further support to using a classification approach.
Given the relationship between balance ability and confidence level, classification of balance confidence in people with lower limb amputations could allow for a more individualised rehabilitation approach, especially when combined with clinical measures of balance ability. In this paper, we investigated if low and high balance confidence groups could be classified using 6MWT smartphone signal data and automated foot strike detection collected from people with lower limb amputations. Classification results were analyzed alongside self-reported fall history to investigate the relationship between self-reported confidence and reported balance history.
2 Materials and methods
2.1 Recruitment and participants
Individuals with transtibial, transfemoral, and bilateral lower limb amputations were recruited from the outpatient clinic at the University Rehabilitation Institute (Ljubljana, Slovenia). All participants provided informed consent. This research was approved by the Ethic Committee of the University Rehabilitation Institute, Slovenia (# 46/2018) and re-approved for 30 additional participants (# 27/2019).
Participants provided information regarding their amputation, falls in the past six months (i.e., fall risk group), and ABC score (Table 2). An ABC score of 80% or greater was considered high confidence [11]. The inclusion criteria were: transtibial amputation or higher; ability to walk with single cane, two crutches, or without any walking aids; minimum of six months post-amputation; had a functional prosthesis; no injuries on the residual limb; was willing to participate and consented to participate. 93 people with lower limb amputations were recruited based on these criteria (Fig 1). Only participants who completed a full 6MWT trial and completed an ABC questionnaire were included in this analysis. Trials were excluded due to incomplete trial (15), fall risk status unavailable (8), ABC score unavailable (6), cell phone placed at the side of the hip instead of lower back (5), and use of a non-rolling walker (1).
2.2 Data collection
Each participant completed one 6MWT trial along a 20 m hallway with an Android smartphone placed in a belt at the lower back. Participants were video recorded during their walk test. The TOHRC Walk Test app collected smartphone accelerometer, gyroscope, and orientation data at 50 Hz [19]. Raw smartphone signal data and timestamps for each participant trial were imported into MATLAB (MathWorks) 2020b [20] for preprocessing. Since smartphone signals have a variable sampling rate, each signal was linearly interpolated to a fixed 50 Hz for a total of 18,049 data points per signal per participant over the 6MWT. Signals were then filtered with a fourth-order zero-lag Butterworth low pass filter with a 4Hz cut-off frequency [21].
2.3 Feature calculation and classification model
Automated foot strikes were used to calculate step-based features. The automated step detection model was described in [22]. Predicted foot strike labels were post-processed in MATLAB (MathWorks) 2020b to correct for model prediction errors, including extra foot strike predictions and missed steps. Typically, a single foot strike event would correspond with a peak in the anterior-posterior (AP) acceleration signal. Extra predictions were defined as instances where two or more consecutive predicted foot strikes were present on the same AP signal, usually immediately before or after the peak in the signal. The timepoints corresponding to these consecutive foot strike predictions were identified. The foot strike event corresponding to the AP acceleration peak was selected and all other predictions in this period were removed.
To identify missed steps, a method similar to that employed by Capela et al. [19] was applied. The duration between each step was recorded. Periods where the duration between two consecutive steps was at least 1.5 times greater than the previous pair of steps were identified. An adaptive locking period specific to each participant’s trial was applied and searched for potential missed steps. Within the adaptive locking period, the AP acceleration peak was identified, and a foot strike was inserted at this timestamp. Final cleaned predictions were used for feature calculations.
62 features were extracted for each step (Table 3). Then the minimum, maximum, mean, and standard deviation were calculated over all steps for a total of 248 features (62 features multiplied by 4 statistics) Table 3. These 248 features were used as input for the random forest model to classify balance confidence groups. The random forest was built and run in Weka 3.8 (Developer University of Waikato) [23]. Correlation-based feature selection (CFS) was used to reduce dimensionality. Participant level leave-one-out cross-validation was used to evaluate classification model performance.
AP = anterior-posterior; ML = medio-lateral; RMS = root-mean square; FFT = fast Fourier transform; REOH = ratio of even/odd harmonic frequencies.
3 Results
For this study, 58 participants were suitable for analysis, 36 non-fallers and 19 fallers (Table 2). Participants were majority male (84.5%), and most participants (male and female) had a transtibial amputation (87.9%). Final model performance was based on 58 individually trained models after the leave-one-out cross-validation. CFS reduced the dataset to 18 step-based features. Self-reported fall history was not used as a feature for the ABC algorithm.
The average ABC score was 62.8%, with only 19 of 58 (32.8%) participants reporting a final score of 80% or greater. Both fallers and non-fallers were distributed across both low confidence and high confidence groups (Table 4), though a greater proportion of both groups were categorized as “low confidence”.
Table 5 details the average responses of individual ABC questions. Scoring across individual ABC questions was not consistent, with some activities resulting in lower average scores than others. Questions 15 and 16 reported the lowest average scores, with the second fewest (stepping onto or off an escalator while holding onto parcels) and fewest (walking outside on icy sidewalks) responses of 100% confidence. “Standing on a chair to reach for something” had the greatest number of 0% confidence responses. The criterion that received the highest average score and the greatest number of 100% confidence responses was “reaching for a small can off a shelf at eye level”.
Table 6 displays the confusion matrix for balance confidence classification. Balance confidence classification accuracy was 81%, sensitivity 63.2%, specificity 89.7%, and precision 75.0%.
4 Discussion
This research demonstrated that balance confidence scores of people with lower limb amputations could be classified using smartphone 6MWT data and automated foot strikes when balance confidence was categorised into high and low confidence groups. Balance confidence classification was better for those in the low confidence group, which is the group that needs identification for possible interventions. Approximately two thirds (67.8%) of participants included in this analysis were low confidence, with an average score of 62.8%, which is consistent with the literature.
81% accuracy would provide correct classification for most people completing a 6MWT, thereby providing balance confidence knowledge that would not have been available without completing the questionnaire. 89.7% specificity and 75% precision showed that few people with low balance confidence were misclassified, so efforts to further investigate potential balance issues for a patient classified to "low confidence" would not be a large drain on resources due to unnecessary investigations (e.g., completing an ABC questionnaire to verify results, etc.).
Fear of falling can affect a person’s balance confidence, regardless if the fall resulted in an injury [25]. Identifying people with low balance confidence is essential, especially for those who may be restricting their social or physical activity participation due to decreased confidence. Clinicians could tailor treatment by providing strategies to improve balance and prevent future falls, but also to increase confidence and promote participation in social and physical activities. Miller et al [14] reported that limitations in activities of daily living, depression, and fear of falling were statistically related to balance confidence, highlighting the importance of addressing social and mental health factors in addition to functional ability for people with lower limb amputations.
It is important to acknowledge that the cut off value determined by Myers et al. [11] was developed for older adults. This study included data from a majority male population (84.5%) with mostly single transtibial amputations (87.9%). This reflects the distribution of males and females with lower limb amputations. However, the high percentage of male participants may limit the ability of this classification model to be applied to women with lower limb amputations, and future work should include a greater percentage of female participants. People with transtibial amputations are at a higher risk of falling during the post-operative period, while people with transfemoral and bilateral amputations are more likely to fall once they have completed their rehabilitation [26]. Including a greater number of people with transfemoral and bilateral amputations would improve the model’s generalizability and possibly allow for sub-group analysis. Additionally, this research only included people who were able to complete the full 6MWT. Excluding participants who were unable or unwilling to complete the full trial potentially excluded people with reduced mobility, less active individuals, or people with poorer balance.
For people with lower limb amputations, when high confidence is not reflected by the person’s physical ability, fall risk is greatly increased [12,17]. This is possibly because these individuals may feel confident enough to engage in more activities. For example, in [12] less caution during stair walking was identified as a risk factor for falls in people with lower limb amputation. Indeed, five people categorised in the high confidence group were self-reported fallers, one of whom reported an ABC score of 100% (Table 4). Interestingly, 23 of the low confidence participants had not fallen in the previous 6 months, indicating that there may not have been a direct relationship between previous falls and balance confidence score. The mechanisms that contribute to falls are diverse and balance confidence alone is not enough to identify individuals who are at risk of falls in this population.
Previous research [18] demonstrated that signal data collected from a smartphone during a 6MWT could be used to calculate stride parameters and provide preliminary fall risk information in people with lower limb amputations. Combining balance confidence scores with a measure of functional ability (i.e., 6MWT) might provide a clearer picture of an individual’s fall risk and would allow for clinicians to create a more individualized treatment program. It is important to note that this model does not replace the value of completing patient interviews. For example, the ABC questionnaire can identify specific activities of interest, like walking outside on icy sidewalks, that could be targeted for balance training. However, for use at the point of patient contact, these models could be integrated into the TOHRC Walk Test App for immediate reporting after completing a walk test.
5 Conclusions
This research showed that balance confidence can be classified in people with lower limb amputations using smartphone signals collected during a 6MWT. While balance ability and balance confidence may not always correspond, both affect an individual’s risk of future falls and quality of life. Identifying participants who have mismatched balance confidence and functional ability is not only essential to providing the appropriate resources and rehabilitation to prevent future falls, but also to prevent social isolation due to fear of falling. Incorporating balance confidence, in addition to stride parameter calculation and fall risk analysis, into the TOHRC Walk Test can enhance the 6MWT and provide additional information after completing a single assessment.
Acknowledgments
The authors would like to thank the Ottawa Hospital Rehabilitation Centre and the University of Ottawa for providing resources for development and training. They would also like to thank the University Rehabilitation Institute in Ljubljana, Slovenia and the volunteers who participated in this study.
References
- 1. Thompson LA, Badache M, Brusamolin JA, Savadkoohi M, Guise J, de Paiva GV, Suh P, Sanchez Guerrero P and Shetty D, "Investigating Relationships between Balance Confidence and Balance Ability in Older Adults," J Aging Res, vol. 2021, no. 3214366, November 2021.
- 2. NIA Strategic Directions, "Aging well in the 21st century: strategic directions for research on aging.," 2016. [Online]. Available: https://www.nia.nih.gov/about/aging-well-21st-century-strategic-directions-research-aging.
- 3. Schoene D, Heller C, Sieber CC, Kemmler W and Freiberger E, "A systematic review on the influence of fear of falling on quality of life in older people: is there a role for falls?," Clin Interv Aging, vol. 14, pp. 701–719, 24 April 2019.
- 4. Liu M, Hou T, Li Y, Sun X, Szanton SL, Clemson L and Davidson PM, "Fear of falling is as important as multiple previous falls in terms of limiting daily activities: a longitudinal study.," BMC Geriatrics, vol. 21, p. 350, 2021.
- 5. Howland J, Lachman ME, Peterson EW, Cote J, Kasten L and Jette A, "Covariates of fear of falling and associated activity curtailment.," The Gerontologist, vol. 38, no. 5, pp. 549–555, 1998. pmid:9803643
- 6. Powell LE and Myers AM, "The Activities-specific Balance Confidence (ABC) Scale," The Journals of Gerontology: Series A, vol. 50A, no. 1, pp. M28–M34, January 1995. pmid:7814786
- 7. Miller WC, Deathe AB and Speechley M, "Psychometric properties of the Activities-specific Balance Confidence scale among individuals with a lower-limb amputation," Archives of Physical Medicine and Rehabilitation, vol. 84, no. 5, pp. 656–661, 2003.
- 8. Botner EM, Miller WC and Eng JJ, "Measurement properties of the Activities-specific Balance Confidence Scale among individuals with stroke.," Disabil Rehabil, vol. 27, no. 4, pp. 156–163, 18 February 2005. pmid:15824045
- 9. Shah G, Oates AR, Arora T, Lanovaz JL and Musselman KE, "Measuring balance confidence after spinal cord injury: the reliability and validity of the Activities-specific Balance Confidence Scale," The Journal of Spinal Cord Medicine, vol. 40, no. 6, pp. 768–776, 2017.
- 10. Cleary KK and Skornyakov E, "Reliability and Internal Consistency of the Activities-Specific Balance Confidence Scale," Physical & Occupational Therapy In Geriatrics, vol. 32, no. 1, pp. 58–67, 2014.
- 11. Myers AM, Fletcher PC and Sherk W, "Discriminative and evaluative properties of the activities-specific balance confidence (ABC) scale," J Gerontol A Biol Sci Med Sci, vol. 53, no. 4, pp. M287–94, July 1998.
- 12. Steinberg N, Gottlieb A, Siev-Ner I and Plotnik M, "Fall incidence and associated risk factors among people with a lower limb amputation during various stages of recovery–a systematic review," Disability and Rehabilitation, vol. 41, no. 15, pp. 1778–1787, 2019.
- 13. Olenšek A, Zadravec M, Burger H and Matjačić Z, "Dynamic balancing responses in unilateral transtibial amputees following outward-directed perturbations during slow treadmill walking differ considerably for amputated and non-amputated side," J Neuroeng Rehabil, vol. 18, no. 1, p. 123, 2021.
- 14. Miller WC, Speechley M and Deather AB, "Balance Confidence Among People With Lower-Limb Amputations," Physical Therapy, vol. 82, no. 9, pp. 856–865, 2002. pmid:12201800
- 15. Asano M, Rushton P, Miller WC and Deathe BA, "Predictors of quality of life among individuals who have a lower limb amputation," Prosthetics and Orthotics International, vol. 32, no. 2, pp. 231–243, 2008. pmid:18569891
- 16. Miller WC, Deathe AB, Speechley M and Koval J, "The influence of falling, fear of falling, and balance confidence on prosthetic mobility and social activity among individuals with a lower extremity amputation," Archives of Physical Medicine and Rehabilitation, vol. 82, no. 9, pp. 1238–1244, 2001.
- 17. Wong K and Chihuri ST, "Impact of Vascular Disease, Amputation Level, and the Mismatch Between Balance Ability and Balance Confidence in a Cross-Sectional Study of the Likelihood of Falls Among People With Limb Loss: Perception Versus Reality," American Journal of Physical Medicine & Rehabilitation, vol. 98, no. 2, pp. 130–135, 2019.
- 18. Juneau P, Lemaire ED, Burger H, Bavec A and Baddour N, "Automated step detection with 6-minute walk test smartphone sensors signals for fall risk classification in lower limb amputees.," PLOS Digital Health, vol. 1, no. 8, p. e0000088, 18 August 2022. pmid:36812591
- 19. Capela NA, Lemaire ED and Baddour N, "Novel algorithm for smartphone-based 6-minute walk test application: algorithm, application development, and evaluation," J Neuroeng Rehabil., vol. 12, no. 19, February 2015.
- 20. The MathWorks Inc, "MATLAB version: 9.9.0 (R2022b)," 2020. [Online]. Available: https://www.mathworks.com.
- 21.
Robertson DGE, Caldwell GE, Hamill J, Kamen G and Whittlesey S, Research methods in biomechanics, Champaign, IL: Human kinetics, 2014.
- 22. Juneau P, Baddour N, Burger H, Bavec A and Lemaire ED, "Comparison of Decision Tree and Long Short-Term Memory Approaches for Automated Foot Strike Detection in Lower Extremity Amputee Populations.," Sensors, vol. 21, p. 6974, 2021. pmid:34770281
- 23. Frank E, Hall MA and Witten H, "The WEKA Workbench. Online Appendix for "Data Mining: Practical Machine Learning Tools and Techniques"," Morgan Kaufmann, Fourth Edition, 2016.
- 24. Sadeghi H, Allard P, Prince Fand Labelle H, "Symmetry and limb dominance in able-bodied gait: a review.," Gait & Posture, vol. 12, no. 1, pp. 34–45, 2000.
- 25. Crenna F, Rossi GB and Berardengo M, "Filtering Biomechanical Signals in Movement Analysis.," Sensors, vol. 21, no. 13, p. 4580, 2021. pmid:34283131
- 26. Finco DR (2022). Musculoskeletal and Biomechanical Adaptations Post-Amputation. The FASEB Journal, 36(S1)