Real-world data capture of daily limb loading using force-sensing insoles: feasibility and lessons learned

Abstract

Force-sensing insoles are wearable technology that offer an innovative way to measure loading outside of laboratory settings. Few studies, however, have utilized insoles to measure daily loading in real-world settings. This is an ancillary study of a randomized controlled trial examining the effect of weight loss alone, weight loss plus weighted vest, or weight loss plus resistance training on bone health in older adults. The purpose of this ancillary study was to determine the feasibility of using force-sensing insoles to collect daily limb loading metrics, including peak force, impulse, and loading rate. Forty-four participants completed a baseline visit of three, 2-minute walking trials while wearing force-sensing insoles. During month two of the intervention, 37 participants wore insoles for 4 days for 8 waking hours each day. At 6-month follow-up, participants completed three, two-minute walking trials and a satisfaction questionnaire. Criteria for success in feasibility was defined as: a) >60% recruitment rate; b) >80% adherence rate; c) >75% of usable data, and d) >75% participant satisfaction. A 77.3% recruitment rate was achieved, with 44 participants enrolled. Participants wore their insoles an average of 7.4 hours per day, and insoles recorded an average of 5.5 hours per day. Peak force, impulse, and loading rate collected at baseline and follow-up were 100% usable. During the real-world settings, 87.8% of data was deemed usable with an average of 1200 minutes/participant. Lastly, average satisfaction was 80.5%. These results suggest that force-sensing insoles appears to be feasible to capture real-world limb loading in older adults.

Introduction

Over the last two decades, technological advances have increased our ability to measure human movement in real-world environments (Adams et al. 2021; Dobkin and Martinez 2018; Wang and Adamczyk 2019). For instance, with the development of inertial measurement units, we can measure an individual’s physical activity (Bammann et al. 2021), postural control (Patel, Pavic, and Goodwin 2020), gait (Washabaugh et al. 2017), and sedentary behavior (F et al. 2019). Rather than having only a snapshot of a person’s movement at one point in time, technology can assess movement over multiple hours and days. Most sensors, however, measure kinematics such as acceleration (Chow et al. 2021; Patel et al. 2022). While understanding body position and acceleration can provide important health metrics, measuring force is also needed to understand limb and joint loading during various activities.

Force-sensing insoles provide a novel way to measure in-shoe forces outside of a laboratory setting (Cudejko, Button, and Al-Amri 2023). Unlike force plates or instrumented walkways, insoles are fitted into a person’s shoes and can be worn during everyday activities in various settings (Renner, Williams, and Queen 2019). Insoles can measure forces acting on the plantar surface of the foot during movement activities. While insoles are valid and reliable during gait and balance activities compared to force plates, previous studies largely utilized insoles in research settings (Peebles et al. 2018; Renner, Williams, and Queen 2019). Few studies have utilized the portability of insoles to measure loading in the real-world over multiple days.

The Incorporating Nutrition, Vests, Education and Strength Training (INVEST; NCT04076618) study is a randomized controlled trial that compares the effects of weight loss alone, weight loss plus weighted vest use, or weight loss plus resistance training (RT) on bone mineral density and fracture risk indicators (Miller et al. 2021). In older adults living with obesity, intentional weight loss through diet and exercise can improve overall health (Villareal, Banks, et al. 2006; Villareal et al. 2011), but weight loss is also associated with bone mineral density loss (Ensrud et al. 2003). Both RT and weighted vests aim to increase mechanical stress through musculoskeletal loading while losing weight to maintain bone health (Daly et al. 2005; Shah et al. 2011; Villareal, Fontana, et al. 2006). However, the overall amount of loading that each activity creates and thresholds needed to minimize bone mineral density loss is unclear.

The INVEST trial offers a unique opportunity to use insoles to continuously monitor lower limb loading during intentional weight loss. As an ancillary pilot study to the INVEST trial, we implemented force-sensing insoles to generate daily limb loading metrics through continuous, real-world monitoring. As such, the purpose of this study was to determine the feasibility of using insoles in older adults across the three treatment arms of the INVEST trial. We hypothesized that we would demonstrate successful feasibility, defined as: 1) at least 60% recruitment rate of participants enrolled in INVEST, 2) at least 80% adherence rate of older adults wearing insoles, 3) at least 75% usable data of real-world capture, and 4) at least 75% participation satisfaction with using the insoles.

Methods

Participants

This pilot study included 44 participants enrolled in the parent INVEST trial. Recruitment began one year following recruitment for the parent study. Full inclusion and exclusion criteria are described in Miller al. (Miller et al. 2021). Briefly, inclusion criteria for the INVEST trial were: 1) between age 65-85 years; 2) overweight or obese defined as a body mass index (BMI) of 27.0-30.0 kg/m² or 30-40kg/m², respectively, and at least one obesity-related risk factor. Participants were excluded if they: 1) weighed >450 lbs; 2) were dependent on a cane or walker; 3) Montreal Cognitive Assessment (MoCA) score <20. During the consent process, participants had the option to consent to participate in the insole sub-study. Opting out of the insole sub-study did not affect participation in the parent INVEST study.

Intervention Procedures

Full procedures of the INVEST study are described elsewhere (Miller et al. 2021). Briefly, eligible participants were randomized to one of three groups: weight loss alone (WL), WL plus weighted vest, or WL plus RT.

Weight Loss Alone:

All participants underwent a dietary WL intervention designed to elicit behavioral changes leading to decreased caloric intake sufficient to result in a 10% loss of initial body mass. The WL intervention included the use of a partial meal replacement program (OPTAVIA^® Optimal Weight 4 & 2 & 1 Plan^®) with group nutrition education (Normandin et al. 2018). Participants were guided by a study dietitian who provided dietary counseling and led behavioral counseling sessions focused on teaching changes in behavioral habits to promote WL and prevent weight regain.

Weight Loss Plus Vest (WL+Vest):

Participants completed the WL component as described above. In addition, participants received an appropriately sized weighted vest (Hyperwear, Austin, TX) that fits comfortably over or under clothing. Participants wore the vest on a daily basis, progressing to a goal of 8 hours/day (by week 2) each day during their most active part of the day. Initially, the vest was unloaded, and 2.25 oz weight blocks was then incremented according to each participant’s rate of WL based on their weekly body weight assessments. As body mass is lost, the weight of the vest was increased to replace the lost weight.

Weight Loss Plus Resistance Training (WL+RT):

Participants completed the WL component as described above. In addition, participants completed a progressive RT intervention to elicit musculoskeletal adaptations. Exercise sessions occurred on three non-consecutive days/week under the guidance of exercise specialists. The exercise prescription is based on recent guidelines for intensity, number of repetitions, exercises, sets, and days per week (Nelson et al. 2007). The training goal was to complete three sets of 10-12 repetitions for eight different exercises at 70-75% one-repetition maximum for that given exercise. The eight exercises included leg press, leg extension, leg curl, triceps pushdown, bicep curl, overhead press, chest press, and seated row.

Insole Assessments

Bilateral lower limb loading was measured using force-sensing insoles (loadsol^®, Novel, Munich, Germany) at baseline, month two of the intervention, and six-month follow-up (Figure 1). Version 1 of the loadsol with a single sensor per insole was used. Previous studies have established the validity and reliability of the loadsols compared to force plates during walking, running, and landings (Peebles et al. 2018; Renner, Williams, and Queen 2019). At baseline, participants completed three, two-minute walking trials at their usual, comfortable speed. Research staff fitted participants with their appropriate insole size and helped participants calibrate each insole. The insoles synced with a smartphone device (Samsung Galaxy S7, Suwon, South Korea) via Bluetooth to record force at 100 Hz. Insoles were placed in participants’ shoes over their current insoles. During calibration, participants entered their weight in pounds followed by a series of three cycles of unloading and loading each insole with their full body weight.

Figure 1.

Participants eligible for the insole study were randomized into one of three treatment groups. They completed three walking trials at baseline, wore the insoles for 4-days during week 8 of the intervention, and completed three walking trials at 6-month follow-up.

At month 2 of the intervention period, participants wore insoles for 4 weekdays for 8 waking hours each day. Prior to at-home wear, research staff met with each participant for insole orientation. During orientation, participants were provided a smartphone with the loadsol^® application installed, a pair of insoles, and batteries for the insoles. Participants then practiced taking batteries in and out of each insole, calibrating each insole, and using the application. Once familiarized and comfortable using the insoles, participants were instructed to wear them during their daily activities, for 8 hours per day over 4 consecutive weekdays. This included time during vest wear for those in the WL+Vest group and time during RT for those in the WL+RT group. Participants were instructed to calibrate their insoles every morning and each time they took their shoes off and on. Data were captured in 4 or more 1-2 hour increments to capture a total of 8 hours per day. They were provided an instruction booklet and asked to contact research staff should they have questions during at-home wear (see Supplementary Material). They also recorded their daily insole wear time on a log.

At 6-month follow-up, participants completed three, two-minute walking trials at their usual, comfortable speed. They also completed an eight-item questionnaire on a five-point Likert scale about their overall satisfaction wearing the insoles, ranging from “strongly agree” to “strongly disagree”. The sum of the eight questions were calculated and divided by a total score of a 40. A score of 100% represents “highly satisfied” and 0% represents “highly unsatisfied”.

Data Analysis

Successful feasibility was defined as: a) recruitment rate of >60% of participants enrolled in INVEST; b) >80% adherence rate in wearing force-sensing insoles for 4 consecutive days; c) >75% of usable data, defined as the ratio between usable minutes of data and total minutes of data, and d) >75% participant satisfaction through an eight-question satisfaction questionnaire. These criteria were determined from previous studies examining feasibility of wearable sensor use (de Bruin et al. 2008; Mc Ardle et al. 2018; Silva de Lima et al. 2017).

Insole data were extracted from smartphone devices and processed in MATLAB (Mathworks Inc, Natick, MA, USA) with custom code available online in the Supplementary Material. Following previous studies on validity and reliability of the loadsol^®, insole data were not filtered (Peebles et al. 2018; Renner, Williams, and Queen 2019). For baseline and follow-up walking trials, the first and last 5 seconds were trimmed and peak force, loading rate, and impulse was calculated for each step, then averaged across all steps in each trial and between the three trials. For at-home wear, data were divided into stepping and standing. Stepping was defined as a period of loading <2 seconds and reaching at least 50% of bodyweight during 30-70% of the loading period. Standing was defined as any period of loading >30 seconds and >25% of the person’s bodyweight. For all steps, peak force, loading rate, impulse, and step time was calculated and averaged across all steps during the four days. For all stands, peak force and impulse was calculated and averaged across the four days. Data were deemed usable if forces were >0N and provided expected mass-normalized values of peak force, loading rate, and impulse for at least one foot. Figure 2 depicts examples of data from baseline (A), real-world wear (B), calculations per step (C), and exclusion criteria during real-world wear (D).

Figure 2.

Force sensing insoles were used to collect data during a 2-minute walking trial at comfortable speed (A) and during everyday activities in the real-world environment (B). For each step, peak force, loading rate, and impulse was calculated (C). Standing was defined as loading >25% bodyweight and >30 seconds, while stepping was defined as loading <2 seconds and >50% bodyweight (D). Loading that did not follow these criteria (i.e., too low, too long) was excluded.

Results

Recruitment

A total of 163 participants were offered to participate in insole sub-study. Of these 163, 37 declined resulting in a recruitment rate of 77.3%. The top reasons for declining were disinterest (n=26), wearing corrective insoles (n=5), and foot pain or foot conditions (n=5). Of those who consented into the study, 82 participants were excluded from the parent INVEST trial. Most common reasons for exclusion were inability to tolerate meal replacement plans (n=13), MoCA<20 (n=7), and BMI>40 (n=6). A total of 44 participants completed baseline testing, 43 participants were randomized, and 37 participants completed at-home wear and follow-up testing. Between randomization and month two of the intervention, six participants dropped out of the insole sub-study due to time constraints, and one participant dropped out of the parent INVEST study. Figure 3 depicts a study flow cart, and Table 1 depicts demographic information of the 37 participants.

Figure 3.

Study flow chart for participants enrolled and included in the study.

Table 1.

Demographic information of participants who completed the insole sub-study. Continuous data are presented as mean (SD), and discrete data are presented as n (%).

Variable	Overall
Age (years)	68.1 (5.3)
Male	10 (27%)
BMI (kg/m2)	33.8 (3.3)
African American/Black	12 (32.4%)
Highest Education
High School/Equivalent	8 (21.6%)
College	18 (48.6%)
Presence of Comorbidities
Arthritis	25 (67.5%)
Cardiovascular Disease	6 (16.2%)
Diabetes	4 (18.8%)

Adherence

Participants wore the insoles between 24-34 hours over a 4-day period, with an average of 7.4 hours per day. Insoles recorded 11.9-33.7 of worn hours over a 4-day period, with an average of 5.5 hours per day. Differences between participant reported and insole captured hours are due to Bluetooth disconnections between the insoles and smartphone. The most common reason for recording <8 hours of wear time per day was rainy weather as participants were instructed to not wear the insoles in wet conditions.

Usable Data

Insole data collected at baseline and follow-up visits were 100% usable. During at-home wear, usable data ranged from 70.4% to 100% depending on the participant, with an average of 87.8%. Usable data captured ranged from 357 minutes to 2024 minutes, with an average of 1200 minutes per participant. The primary reasons for unusable data were likely calibration errors resulting in negative force values or long time periods where the force exceeded 150% body weight. The total number of steps ranged from 800 – 10,000 steps per day. The total amount of standing time ranged from 3.5 minutes to 59.6 minutes per day. No evidence of signal drift was detected as the peak force values during walking were similar throughout the collected hours each day. Peak force values during the first 50 steps averaged 1.12 N/kg while the last 50 steps averaged 1.13 N/kg. Tables 2 and 3 shows mean daily values for limb loading outcomes during walking and standing.

Table 2.

Limb loading data from walking periods through four days, normalized to bodyweight (BW). Averages (SD) are reported.

	Group 1 (n=9)	Group 2 (=17)	Group 3 (n=11)	Overall (n=37)
Number Steps per Day	6,508 (2216)	4,381 (2848)	3,549 (3241)	4,659 (2983)
Average Peak (BW/step)	1.23 (0.36)	1.30 (0.39)	1.27 (0.32)	1.27 (0.35)
Max Peak (BW/step)	1.71 (0.44)	1.81 (0.64)	1.85 (0.54)	1.80 (0.55)
Total Impulse (BW*s/hr)	267.27 (150.51)	291.96 (206.82)	171.94 (78.39)	248.89 (167.95)
Average Impulse (BW*s/step)	0.77 (0.26)	0.83 (0.21)	0.83 (0.26)	0.81 (0.23)
Average Loading Rate (BW/s/step)	6.14 (1.4)	6.57 (3.92)	6.01 (1.2)	6.29 (2.75)

Table 3.

Limb loading data from standing periods through four days, normalized to bodyweight (BW). Averages (SD) are reported.

	Group 1 (n=9)	Group 2 (=17)	Group 3 (n=11)	Overall (n=37)
Average Peak (BW/bout)	0.92 (0.40)	0.95 (0.43)	0.97 (0.30)	0.95 (0.38)
Max Peak (BW/bout)	1.02 (0.47)	1.08 (0.52)	1.05 (0.31)	1.06 (0.44)
Total Impulse (BW*s/hr)	2.43E3 (1.08E3)	3.14E3 (2.53E3)	2.86E3 (1.87E3)	2.87E3 (2.03E3)

Satisfaction

The average satisfaction was 80.5%, ranging from 57.5% to 100% (Table 4). Scores were highest for ‘the study staff answered my questions in an appropriate and timely manner’, ‘the insoles were comfortable to wear’, and ‘overall, I am satisfied with my participation in the insole study’. Scores were lowest for ‘it was easy to use the smartphone and navigate the insole application’. Participants commented that the insoles disconnecting from the smartphone was their largest frustration. Despite the intermittent disconnections, all participants were able to reconnect and continue data capture during the at-home wear periods.

Table 4.

Participants completed an 8 item satisfaction questionnaire. Each question was scored on a five-point Likert scale, with 1 indicating strongly agree and 5 indicating strongly disagree.

Question	Average (SD)
It was easy for me to set up and wear the insoles on my own.	2.35 (1.15)
The insoles were comfortable to wear.	1.45 (0.55)
Wearing the insoles interfered with my daily activities.	4.13 (0.99
It was easy for me to use the smartphone and navigate the insole application.	2.86 (1.18)
Using the insole application on the smartphone interfered with my daily activities.	3.97 (1.16)
The study staff answered my questions in an appropriate and timely manner.	1.24 (0.43)
I would participate in a future study wearing the insoles.	2.08 (1.19)
Overall I am satisfied with my participation in the insole study.	1.81 (0.84)

Discussion

The purpose of this study was to determine the feasibility of using force-sensing insoles to assess limb loading in real-world environments in older adults with obesity at risk for bone mineral density loss. We evaluated feasibility using recruitment rate, adherence to wearing the insoles, ability to capture usable data, and overall participant satisfaction. Overall, our results demonstrate that older adults: 1) will participate in a study involving at-home wear, 2) will wear the insoles for four days, 3) are able to capture usable data for analysis, and 4) report satisfaction with the insoles. Leveraging force-sensing insoles to capture real-world limb loading appears to be feasible for older adults.

To the authors’ knowledge, this is the first study to utilize force-sensing insoles outside of a laboratory setting over multiple days. While previous studies completed in-lab walking trials, our results from 4-days of wear are comparable. For example, previous studies of participants walking with the loadsol^® insoles report peak forces ranging from 1.03-1.27 bodyweights (Kong et al. 2023; Renner and Queen 2021; Renner, Williams, and Queen 2019). Our peak average of 1.27 bodyweights per step across treatment groups is in line with the previous literature. In addition, our results of average bodyweight-normalized impulse and loading rate per step are also comparable to prior studies (Kong et al. 2023; Renner and Queen 2021; Renner, Williams, and Queen 2019; Yeo, Chhabra, and Kong 2022). These results suggest that force-sensing insoles can provide accurate walking and standing measurements in real-world environments.

Force-sensing insoles offer the opportunity to measure limb loading outside of laboratory settings. During this study, the insoles measured limb loading while participants walked in their homes, the community, and while exercising, providing a snapshot of daily activity and limb loading patterns. While technology issues occurred including calibration errors, Bluetooth disconnection, and battery drainage, we were able to capture an average of 20.3 hours of data over 2.5 days of wear. Future studies may consider using force-sensing insoles to measure real-world loading. To maximize efficiency and data quality for future studies, we provide a summary of three lessons learned from the current study.

Lesson #1: Correct calibration is essential for data quality. The primary reason for unusable data was incorrect calibration. Prior to the 4-day wear, research staff met with each participant to review calibration procedures. Participants were asked to complete each step independently, with guidance from a pamphlet. Once participants successfully inserted the batteries, calibrated each insole, and started recording, they were sent home. However, even with careful instructions, ~11% of the data was calibrated incorrectly, resulting in unusable data. Given the at-home data capture methods used in this study, we were unable to identify the cause of each instance of a poor calibration. Some participants called each morning to review calibration steps, which ensured that the insoles were calibrated correctly. For future studies, we recommend staff follow-up the first morning of insole use to review calibration steps. Alternatively, creating simple data checks for participants to perform such as weight-shifting or marching in place along with education on the expected values during these activities may help participants identify inaccurate calibrations and increase the percentage of usable data.

Lesson #2: Ensure Bluetooth connection. Connection loss was another cause of missing data. Insoles disconnected from the application if participants moved too far from the smartphone. Based on participant reports, impacts to the insole battery housing (e.g., housing hits a sidewalk curb or stairs) also disrupted the connection. We recommend providing participants with a strap or clip to consistently keep the phone nearby. We also recommend teaching participants how to read the real-time force outputs from the insoles on the smartphone. If data from an insole was missing, we explained that the insole disconnected and asked participants to reconnect and recalibrate. While some participants followed these steps, others had difficulty interpreting the force outputs, resulting in missing data. One alternative solution may be to require participants to recalibrate their insoles every 2-3 hours to reduce the amount of lost data. Additionally, similar metrics can be obtained from other insole options, including the second-generation loadsol pro or insoles from Noraxon, Xsensor (Parker, Andrews, and Price 2023), Moticon (Braun et al. 2015) and Tekscan (Patrick and Donovan 2018). To the authors’ knowledge, these insoles have not been tested outside of laboratory environments to determine their feasibility in real-world settings.

Lesson #3: Consider age-related declines in manual dexterity. The insoles used for this study were powered with coin cell batteries. Due to the multi-day use, participants were asked to remove their batteries at the end of each day to preserve battery life and prevent batteries from dying during data collection. With the loss of manual dexterity with aging, older adults reported difficulties opening the battery housing, which requires fine motor skills to remove and insert each battery. During the study, Novel released a new insole, the loadsol^® pro, which uses a USB-rechargeable battery that has been validated during walking, running and landing tasks (Brindle et al. 2022; Schümperlin et al. 2023). We used the loadsol^® pro on three participants and no issues with charging the insoles were noted. Additional testing with the loadsol^® pro is needed, but for future studies we recommend using the loadsol^® pro with older adults.

Despite the challenges we encountered, we collected an abundant amount of limb loading data in real-world environments. We collected an average of 20.3 hours of data per participant over 4 days of wear. The insoles also allowed us to collect data during participants’ RT and weighted vest wear, expanding our analysis beyond traditional laboratory-based measures. With the lessons learned from this pilot study, we will be able to incorporate insoles into future clinical trials and recruit a larger sample size to compare limb loading across treatment arms. Future studies can also use insoles in real-world environments to examine movement patterns and behaviors in older adults. This study also has limitations. First, our real-world data only included a 4-day period during the second month of a 12-month intervention. Due to the pilot nature of the study, the goal was to test the insoles for a shorter period during the intervention. Future studies will aim to utilize the insoles for even longer periods of wear. Second, during real-world capture, we isolated data into walking and standing bouts and may have missed activity that did not fit this criterion. Future work should examine isolating various periods of daily activities. In conclusion, we found that levering force-sensing insoles within a clinical trial for older adults to maintain bone health was feasible.