In-shoe Sensor Measures of Loading Asymmetry during Gait as a Predictor of Frailty Development in Community-Dwelling Older Adults

1. Introduction

The global prevalence of frailty is over 17%, and it increases with age [1]. Frailty is a weakness syndrome associated with aging, which represents a physiological decline and is closely related to adverse events such as falls, disorders, and death [2,3]. While many frailty-specific interventions have been proposed, almost none target older adults before the onset of frailty. Interventions aimed at preventing rather than treating frailty are likely to have a greater impact on the overall health of the population [4].

One of challenges in developing primary preventions for frailty is the difficulty in identifying high-risk individuals among older adults without clear signs of poor health. Currently, technological advances are providing more accurate gait analysis methods to detect frailty risk in older people. In particular, wearable sensors are used in most studies because their advantage is the possibility of monitoring their subjects anywhere without the need for a controlled environment. Several studies [5,6,7,8] have supported the use of gait analysis systems for the evaluation of frailty risk in older people.

Ground reaction force is commonly used in gait analysis to distinguish between healthy and pathologic subjects, and plays a crucial role in identifying biomechanical factors of disease progression in older adults [9,10]. Specifically, ground force has been used as the primary metric in a biofeedback training protocol and as both an assessment tool or secondary metric in older adults [11]. Instead of expensive equipment to measure ground reaction forces, load-sensing insoles enable the collection of force-based metrics through wearable technology, thereby enhancing the accessibility of gait assessments. The new triple-sensor load measuring insoles (loadsol®, Novel Electronics, Munich, Germany) provides solutions to the difficulties of gait mechanics monitoring. They feature a wireless design using Bluetooth communication and are capable of measuring the force applied to the ground by the plantar overall region, as well as three sub-regions: forefoot, midfoot, and rearfoot. The loadsol® requires the use of an app-enabled mobile device, rendering it a portable and clinically accessible option for collecting force-based data [12,13,14,15]. It has been shown to be a valid and reliable method of collecting load kinetics in older adults [16]. Additionally, with the use of loadsol®, we previously found increased loading asymmetry on the plantar region during walking in community-living older adults with a history of falls [17]. This finding indicates the importance of assessing loading asymmetry to screen frailty risk in older adults, because frailty is closely related to fall risk. If asymmetry in loading, as measured by the loadsol®, proves to be a reliable predictor of frailty risk, this would suggest that the loadsol® is effective in detecting unnoticed problems of locomotion through its collection of load kinetics. Moreover, if a regional plantar force that contributes to loading asymmetry on net forces could be identified, it would help to establish a novel metric of gait training in a biofeedback to prevent frailty.

The primary objectives of this investigation were, therefore, (1) to test the hypothesis that increased limb loading asymmetry on net plantar force during walking predicts the development of frailty in community-dwelling older adults participating in a community health survey, and (2) to examine the relationship of loading asymmetry between net and regional plantar forces in this population.

2. Materials and Methods

2.1. Participants

We conducted a prospective cohort study through a community health survey in Minamisoma City, Fukushima Prefecture, Japan. The participants were recruited via advertisements in community newsletters. To be eligible to take part in the present study, the participants had to be 65 years or older, community dwelling, and able to be independent in activities of daily living (ADL). The exclusion criteria included meeting frailty criteria at baseline, the use of walking aids, and limitations preventing them from participating in the gait and physical function tests described below.

Ethical approval for this study was granted by the Institutional Review Board of Fukushima Medical University (reference number 2022-123). The participants were provided with an information sheet explaining the study.

2.2. Procedures

After informed consent was obtained, the subjects filled out a questionnaire and performed the gait and physical function tests at baseline. Ten months after this baseline assessment, a follow-up assessment was conducted to obtain information regarding frailty.

2.3. Assessment of Loading Asymmetry

Each participant was fit with a pair of commercial walking shoes, with the loadsol® placed within the shoes (Figure 1). The force sensors are segmented by the manufacturer into thirds based on sensor length, representing rearfoot, midfoot, and forefoot plantar regions. The force sensors were calibrated using a previously described protocol [13,14]. Participants were instructed to load the insoles with their full bodyweight in a single peg stance was then unloading the insole three times on each foot. The loading of the participant was entered in Newton (N) in the loadsol® application on an iPad mini (Apple Inc., Cupertino, CA, USA) and was recorded at 100 Hz.

Following calibration, each participant performed the 10-meter walk test (10MWT) twice. Participants were instructed to walk as quickly but as safely as possible. First, they walked along a 14-m straight walkway on a flat floor at their maximum speed. Then, they slowly turned around and walked back along the 14-m straight walkway at their maximum speed. The turning phase was used to segment the walking data.

Data were processed using loadpad analysis software (Novel Electronics). The raw plantar force data was proceeded without a Butterworth filter using a cutoff based on a previously reported analysis [14]. The calibrated plantar forces were imported as a net value and three regional values. Each participant’s steps were identified within walking using a 30 N threshold (≥ 6% of mean participant bodyweight) from net force timeseries data to avoid false positive step identification when the force sensor deflected within the shoe. Plantar forces data were extracted for the middle 10 steps within a straight walking phase and calculated to four variables of interest (contact time [CT], peak force [PF], loading rate [Lr], and impulse) of each left / right for each trial (Figure 2). CT was defined as the time in which the total force signal of the insole is equal or higher than 30 N. PF was defined as the maximum force value of each step in the period of 0 – 40 % of stance phase. Lr was defined as the slope of 20 % to 80 % into the first peak of the foce curve. Impulse was defined as the force time integral between the force curve and the time axis. CT, PF, and impulse were calculated for net, rearfoot, midfoot, and forefoot regions, and Lr was calculated for net region only. For left and right steps independently, CT, PF, and Lr were averaged over 5 steps and impulse was summed those of 5 steps. Finally, all variables were averaged over two trials, and PF, LR, and impulse were normalized to body weight.

Additionally, asymmetry parameters for each of all variables were calculated by the difference between the more loaded foot and the less loaded foot, and then divided by the sum of both feet [18]. For example, PF asymmetry was calculated using the following formula:

PF asymmetry (%) =

× 100

Mean PF on the more loaded foot + Mean PF on the less loaded foot

2.4. Evaluation of Frailty

The Kihon Checklist (KCL), which allows comprehensive assessment of frailty in daily life [19,20], was used to assess frailty at baseline and follow-up. The KCL comprises 25 items (yes/no questions) that assess important areas to frailty, including ADL (Items 1–5), physical function (Items 6–10), nutritional status (Items 16 and 17), cognitive function Items 18–20) and depressive mood (Item 21–25). The participants were classified into three groups based on the KCL. Out of a maximum of 25 points, those with scores of ≥ 8 points were defined as the frailty grading, those with 4–7 points were classified as the pre-frailty grading, and those with ≤ 3 points as the robust grading [21]. The frailty development was defined if the participant progressed from baseline at least one grading group of frailty at follow-up period. The frailty not-development was defined if the participants sustained or improved their frailty grading at follow-up period from baseline.

2.5. Assessment of Covariates

The gait speed, the Timed Up and Go (TUG) test, fall history, presence of cerebrovascular disorder, and leg pain were collected as confounding factors at baseline. The straight walking trials at their comfortable speed were timed in the middle 10 m, between marks 2 and 12 m and its data were expressed as speed (m/s) [22]. The TUG test was performed to rise from an armchair, walk 3.0 m at participant’s maximum speed to a mark, turn around, return to the chair, and sit down [23]. History of falls in the past year and presence of cerebrovascular disorder were retrospectively investigated by questionnaire. In the past year, those who reported one or more falls were defined as fallers, and those who reported no falls were defined as non-fallers. Knee and/or foot pain was assessed using a questionnaire asking whether the participants had experienced pain lasting for more than one month.

2.6. Statistical Analysis

To determine potential confounding factors, independent t-test for continuous variables and ꭓ² test for categorical variables at baseline were used to test for statistical differences between participants who developed frailty and those who did not. Analysis of covariance (ANCOVA) adjusted for potential confounding factors was performed, with the occurrence of frailty development set as the independent variable, and the net asymmetry parameters set as the dependent variables. Following convention [24], multivariate logistic regression was used to predict net asymmetry parameters associated with frailty development. Age, gender, gait speed, and KCL score were included as covariates in the model. Additionally, correlations between the predictive values of net asymmetry and rearfoot, midfoot, and forefoot asymmetry were tested by using Pearson’s correlation coefficient. A 2-sided p-value < 0.05 was considered statistically significant. Statistical analysis was performed using SPSS software (Version 29, SPSS Inc., Chicago, IL, USA).

3. Results

3.1. Subject Characteristics

Sixty-three subjects participated in the baseline of this study. None of them used walking aids. Of the participants, nine subjects did not attend the 10-month follow-up visit. Six subjects who met the frailty criteria at baseline and one subjects with missing force loading data were excluded (Figure 3). There were no significant differences regarding baseline characteristics, performance, and loading asymmetry parameters in the subjects with and without follow-up. The mean age of the remaining 47 subjects (37 women, 10 men) was 75.3 ± 5.8 years. The mean height and weight were 153 ± 7 cm and 53.2 ± 9.5 kg, respectively. Gait and balance functions were relatively high; the mean comfortable gait speed was 1.45 ± .18 m/s, and TUG time was 6.4 ± 1.1 s.

According to the KCL, the prevalence of robustness and pre-frailty at baseline were 62% (n = 29) and 38% (n = 18), respectively. At follow-up period, the 14 participants were defined as the frailty development group and the 33 participants as the frailty non-development group (Figure 4). The baseline variables of the frailty development and non-development groups are shown in Table 1. The KCL scores at baseline, that the frailty development group had significantly lower score than non-development group (P = 0.03), was used as a covariate for ANCOVA.

3.2. Loading Asymmetry and Frailty Development

The ANCOVA analyses, run with the KCL scores as baseline as a covariate, showed that two values of loading asymmetry on net plantar force were significantly increased in those who developed frailty (n = 14) compared with those who did not (n = 33), as summarized in Figure 5. Net Lr asymmetry in frailty development group were significantly higher than those in frailty non-development group (F = 4.432, p = 0.042). Also, Net impulse asymmetry in frailty development group were significantly higher than those in frailty non-development group (F = 9.647, p = 0.004).

Multivariate logistic regression analysis showed that the risk factor independently associated with frailty development included the net impulse asymmetry (OR = 1.269, 95%CI 1.016–1.585, P = 0.033) and the KCL scores (OR = 2.331, 95%CI 1.170–4.647, P = 0.016) (Table 2). The net Impulse asymmetry was associated only with PF asymmetry on midfoot (r = .323, P < 0.05). Other loading asymmetry variables on the regional plantar forces did not correlate with net impulse asymmetry (Table 3).

4. Discussion

The present study of loading asymmetry in gait has several key findings. First, among community-dwelling older adults without frailty, net impulse asymmetry in gait is a valid predictor of frailty development within ten months. Second, net impulse asymmetry in gait is associated with PF asymmetry in the midfoot region, where the medial longitudinal arch of the foot plays an important role in shock attenuation and in generating sufficient power for propulsion during gait.

The present findings are consistent with previous results obtained in slightly different settings and populations. Pinloche et al. [25], using plantar sensors for foot pressure, reported that COP velocity and projection differed between frailty and pre-frailty groups in a nursing home population. Anzai et al. [26], using smart insoles for plantar pressure, found that the classification of participants relative to their frailty state based on the KCL relied on features extracted from the plantar pressure series during walking. Additionally, with respect to the frailty condition at baseline, Fried et al. [2] reported that pre-frailty stage is identified as a high risk of progressing to frailty. The present study extends these findings in several ways. Our results indicate the feasibility of obtaining measures of loading asymmetry using a single easy-to-use instrument with a clinical walk test. Furthermore, the present study shows the relationship between loading asymmetry in gait and future development of frailty in community-dwelling older adults.

Increased net impulse asymmetry was associated with increased midfoot PF asymmetry. Considering that it is commonly accepted that the midfoot with the medial longitudinal arch is the shock attenuation and the propulsive part of the foot, it could be hypothesized that these midfoot functions have potentially declined in people with frailty risk. Age-related modification of posture with accentuation of plantar deformation makes older adults cautious, reducing symmetry during walking [27,28]. Some explanations could be given, such as tibialis posterior dysfunction [29], decreased strength, sensitivity or mobility of the foot, or alteration of the somatosensory system. The midfoot change suggests that older adults adopt a ‘pull-off’ rather than ‘push-off’ strategy to generate forward momentum during walking. This dependence on a host of musculoskeletal and biomechanical factors may help to account for the observed predictive value of loading asymmetry in gait.

The present study has several limitations. First, the relatively small sample size may have resulted in a Type II error or a failure to detect other differences between the developed and non-developed frailty groups. Other measures of loading asymmetry might have been also expected to differ significantly between the developed and non-developed frailty groups. Increasing the sample size would reduce the risk of Type II errors, and allow further detailed studies of (1) the relationship between net and regional plantar forces in relation to load asymmetry or (2) the specificity and sensitivity of various predictors of frailty risk.

In addition, we should consider a limitation related to the standards adopted in this study to identify frailty. Although the KCL is widely used in Japan and has been described as a valid frailty prediction tool in several reports [21,30], many other criteria are available to predict frailty. Ambagtsheer et al. [31] suggested that results could vary widely, especially between self-administrated methods, such as the Kihon checklist, tests administered by nurses or those by physicians. It is possible that the accuracy of the present study could have been different, either increased or decreased, if another frailty assessment tool had been used as a reference instead of the KCL.

Finally, another limitation of the present study is the non-inclusion of variables related to other diseases that include joint pain such as arthritis. Indeed, a combination of gait loading asymmetry and such diseases could strengthen accuracy scores for frailty condition prediction. Therefore, futures studies are needed to investigate asymmetry due to a history of joint pain using questionnaires and physical examinations.

5. Conclusion

Limb loading asymmetry, collected with a 3-sensor insole during a 10-m walking test, has been successfully used to identify older adults with potential frailty risk. Our results suggest that measures of limb loading asymmetry in gait could be used to help predict frailty development in older adults via a community health screening. Although further research is needed to gain a full understanding of the wide variety of factors that contribute to gait asymmetry and its relationship to frailty risk, our findings may be useful for the application of new or existing interventions using limb loading as biofeedback to prevent frailty.