Participants were recruited irrespective of whether they did or did not have any current musculoskeletal conditions. Similarly, we did not require participants to have any prior experience in using smartphone apps or other digital health platforms for managing musculoskeletal conditions or accessing services. This allowed us to gain a more generalized range of viewpoints from the trial. We used the following inclusion and exclusion criteria, which participants were screened against, before being onboarded into the study.

The following inclusion criteria were used:

The following exclusion criteria were used:

The research team was recruited from a wide range of community organizations and venues, including community centers, local charity groups, libraries, supported-living accommodations, and religious centers such as churches and mosques. A total of 21 local organizations were contacted with the support of the West Midlands Clinical Research Network [14]. Paper and digital format posters were distributed in these locations through relevant contacts at each organization or via an in-person visit by a member of the research team.

Participants registered their interest by providing their contact details (email or postal address) via (1) an online form, (2) email, or (3) telephone message. The participant information leaflet and consent form were then sent to those who were registered. For organizations and community sites where there were multiple participants, we coordinated via the organization’s contact a suitable time to visit the site to onboard the group of participants. Individual participants were contacted directly to arrange a suitable date and location to attend the onboarding session, which was held at the university or a local community venue.

The study received ethical approval from the University of Warwick’s Biomedical and Scientific Research Ethics Committee (ref. 147.20‐21). Participants provided written consent prior to participation in the trial. All participants who engaged with the trial received a £20 (US $27) gift voucher for their participation. Participant data were pseudonymized using participant ID numbers assigned during the consenting process. A formal data sharing agreement was put in place between the researchers’ host institution and the commercial organization that produced the app.

The key stages of the project are shown in Figure 1. The trial began with an in-person onboarding session, where the participants were screened, consented, provided with guidance on installing and navigating the app, and demonstrated the functional activities. Initial recordings of the StS and SLB tasks were taken.

After the onboarding session, participants began the 4-week home-based trial. This involved using the app at least once a week to complete the two sets of assessments. Participants were able to complete the assessments at any time convenient to them during the week. No reminders were built into the app, so it was up to the participants to remember to complete the activities each week.

After the second week, the researcher contacted the participants over the phone to check that they were able to successfully complete the first 2 weeks of assessments. Any reported issues were subsequently resolved as necessary.

At the end of the trial period, the researcher arranged a phone interview for each individual participant to provide detailed feedback on their experiences during the trial. Interviews lasted approximately 30 minutes. A semistructured approach was taken using a list of guiding questions (provided in Multimedia Appendix 1). The researcher maintained detailed notes from the interview responses. Participants also completed the SUS [15] to provide a quantitative rating on the usability of the app.

The prototype app, named Phio Measures, was developed by EQL Ltd. The app worked by recording the StS and SLB activities using the inertial measurement units built into smartphones. Thus, the raw data recorded were the 3D measures of acceleration (using the accelerometer), angular velocity (gyroscope), and magnetic field (magnetometer), sampled at 100 Hz. Inertial data and relevant metadata (eg, trial type, user ID) were automatically uploaded to a secure server. The research team accessed the data via an online dashboard.

Offline algorithms were used to identify and record the time duration to complete 5 repetitions of StS. This used a peak detection method applied to the gyroscope data to identify the repetitions of movement in this activity, which could be visually checked for accuracy. For the SLB task, the algorithm measured the magnitude of sway (using the accelerometer data in anterior-posterior and medio-lateral directions), as well as the duration of completing the task before returning the foot to the floor. Significant foot contact was identified by a clear impact spike in the sway data, which was detected using a threshold approach. In addition, participants were asked to end the recording once both feet were back in contact with the ground. In this project, we did not assess the accuracy of the algorithm but rather determined the proportion of uploaded trials that the algorithms were able to successfully analyze (ie, completed analysis without error and correctly identified the start, end, and repetitions of the movements). In the case of SLB, the algorithm typically required >11 seconds of data to calculate the measures.

The app was only available to Apple iPhone users, and this was made clear in all the recruitment literature and communications. However, 1 participant had to be excluded due to having an iPhone that was too old to install the software. After completing the consent process, participants were assisted with installing the software onto their smartphones. As this was prototype software, it involved installing an additional Apple app, TestFlight, which allowed the installation of prerelease software onto the phone. The steps for installation are shown in Figure 2. Participants were able to delete the software from their device at the end of the trial or at any time during the trial (which was subsequently noted as a wish to withdraw from the trial).

The study focused on the usability of the app and the recording processed during the assessments. A mixed methods approach was used to capture smartphone measures, usability scores, and qualitative interview data.

The primary outcome measure was the qualitative results from the feedback interviews conducted after the trial period.

The secondary outcome measures included usability, assessed using SUS (0-100) during the posttrial interview; data quality, measured by the proportion of successful algorithmic assessments from unsupervised recordings taken during the trial; and trial adherence, evaluated by the number of recordings per participant and the time intervals between sessions [15].

A fair notes approach [16] was used to record the responses from the interviews, which were facilitated by a member of the research team (CK). An inductive content analysis was used to identify the key categories and concepts that were mentioned by participants during the interviews. These were grouped and synthesized to identify the positive and negative recurring concepts along with any subcategories that emerged from the discussions. Analyses of the interview notes were undertaken independently by author MTE and subsequently cross-checked by CK, who suggested amendments to categories and content allocation.

The mean and SD of SUS scores were calculated across participants. The resulting value was translated to a percentile level compared to standardized usability scores from historic studies [17]. Descriptive statistics were used on other secondary objective measures.

A total of 25 individuals initially expressed interest in participating in the study (see Figure 4 for the CONSORT [Consolidated Standard of Reporting Trials] flow diagram for this study). Of these, 21 participants subsequently consented to take part in the study (female, n=15). One participant was unable to participate due to their Apple iPhone being too old to run the app; the remaining 3 did not respond to follow-up or complete a consent form. Finally, 17 participants completed the trial and provided valid data for analysis (female, n=12). The other 4 participants did not engage with the app, resulting in no data being uploaded (other than that collected during the initial onboarding session). One participant who completed the trial was unable to be contacted for the final interview.

A total of 192 data recordings were uploaded by the 17 participants over the duration of the trial. Of these, 142 out of 192 (74%) were valid for subsequent analysis by the algorithm. There were a lower number of valid SLB measures (81/127, 63.8%) than StS measures (58/62, 93.5%). It should be noted that there were generally double the recordings of SLB (n=127) compared to StS (n=65) due to the recording of both left and right legs. Most of the invalid uploads were due to insufficient data samples recorded to perform an analysis, likely due to the participant placing their foot on the floor (recognized by a signal spike in the sensor data) after a very short duration and stopping the recording.

Participants were requested to complete both the StS activity and SLB for both legs at least once per week over the 4-week period of the trial. Participants, on the whole, adhered to the trial requirements (Table 1), completing a mean of 3.8 (SD 1.2) sessions for each activity, with just over a 7-day gap between sessions and a total duration between the first and the last sessions of just above the expected 21 days. Using 1-sample t tests against the target values, there were no significant deviations by participants from what was instructed (P>.05). Five participants were noted to have completed more sessions than requested (5‐6 in total) by completing multiple sessions in 1 week or continuing beyond the 4-week period. In contrast, 7 participants completed fewer than the requested number of sessions (2‐3 in total).

The overall mean time to complete the 5 repetitions of StS was 19.1 (SD 2.6) seconds. The overall mean time participants maintained SLB was 13.7 (SD 0.9) seconds. It was noted that the algorithms used were unable to calculate the sway measures when the duration of SLB was <11 seconds.

The mean SUS rating across all participants was 81.2 (SD 17.5; range 45-100). This is equivalent to being rated in the 90th percentile against other studies reporting the SUS score [17].

A total of 16 telephone interviews were undertaken with participants. Five top-level categories of discussion were extracted: app, task, device, time, and personal perceptions. These categories are a mix of those that emerged directly from the subject of the questions asked, as well as those that emerged indirectly through the discussion and have been identified in the analysis. Within the top-level categories, the recorded notes and quotations were further classified into sub-concepts, as shown in Figure 5. The following section discusses the context of each category and subcategory, providing representative quotes that were identified from participant interviews.

The final theme captured any personal perceptions that completing the assessments had on participants or could have on participants, along with additional requirements identified for them to use this formally in the future.

The aim of the digital platform trialed in this study was to be able to collect quantitative, objective measures of physical function when patients complete commonly used physiotherapy performance-based tests without clinical supervision. Our focus was to investigate the feasibility and usability of such a platform when used by older adults (aged 60 y or older), the group with the highest prevalence of musculoskeletal conditions [18]. The app was evaluated in three different ways: (1) engagement, evaluating whether participants completed the full program of assessments at least once per week over the 4-week period; (2) data quality, analyzing the resulting sensor data to examine the proportion of recordings that could be assessed by the algorithm; and (3) usability, detailed feedback from participants on usability using a standard questionnaire (SUS) and a phone-based interview at the end of the trial.

Our key findings were that overall, participants adhered to using the app at the minimum scheduled frequency over the duration of the trial. The quality of the data uploaded unsupervised by the participants during home-based completions of the activities was high for the StS task (93.5%), but much lower for the SLB task (63.8%), highlighting the challenges of completing a balance task by the participants. Despite its limited functionality as a prototype, the app scored highly on the SUS scale, likely due to the simplicity of the interface. However, this effectively provided a blank canvas for participants to recognize the desired features that were missing; they subsequently highlighted specifically the need for rapid feedback on their performance/outcome following completion of the tasks and the option for more activities to be made available to be chosen either by the individual or a medical professional (eg, physiotherapist, general practitioner). Short assessments or activities are essential and allowed participants to fit in alongside their daily activities. The proposed method of holding the phone to the chest while completing activities was considered feasible, but not ideal. Suggestions were put forward to improve this approach, including the use of a smartwatch; however, such devices are less prevalent in terms of ownership compared to smartphones, risking digital exclusion.

Overall, there was a good level of engagement with the trial, with a total of 4 (19.1%) participants dropping out over the trial period. This contrasts with a meta-analysis study of mental health smartphone apps, which found an overall dropout rate of 26.2% (47.8% when accounting for publication bias) [19]. Our low dropout rate and high engagement are likely due to the short duration of the trial (4 wk). A longer-term program might result in a reduced sense of novelty and subsequently a significant dropout rate over time, without further intervention (eg, the use of incentives [20,21]). However, older adults are more likely to engage with remote digital health studies for a longer period than younger counterparts [22].

It was observed from the interview outcomes that the short, easy-to-complete activities were another driver of engagement, with participants able to fit them into their regular daily routines. While it may be considered that older adults have more leisure time than young adults to complete these activities, an Australian study found that 16% of inactive adults aged 65‐69 years reported not having enough time as the first or second highest barrier to exercise [23]. While the activities in this study are functional rather than cardiovascular-type exercises, it is important that their duration remains short. The Osteoarthritis Research Society International recommended performance-based tests for osteoarthritis [24] include StS (chair stand), along with a number of other simple activities, such as timed up and go, 6-minute walk, and stair-climb. Technically, all these activities could be recorded using standard smartphone sensors, but as reflected by our findings, each should be tested for its feasibility in unsupervised conditions to allow for unforeseen complexities in the task.

Our results further highlighted that only the StS task was practical for unsupervised monitoring using a smartphone. The low data quality of the SLB task was further reflected by the participants’ remarks, with some struggling to perform the balance task successfully. Participants also requested a variety of activities, which might involve more practical functional tasks such as a 10-m walk, timed up and go, or a stair climbing task [24].

Maintaining motivation to complete the activities is essential for longer term programs. While an increased range of exercises is one method of maintaining engagement, the other aspect highlighted by a substantial number of participants was to have feedback. The lack of feedback in the prototype was identified as a significant omission, with requests for both performance-based feedback (eg, time to complete 5 repetitions) and acknowledgments around the task (eg, task successfully completed, data uploaded). Performance-based feedback can give participants a sense of achievement, particularly if historical trends of performance are recorded [25]. However, it is noteworthy that when asked, no participants showed any desire to make these activities “competitive” (eg, using leader boards or comparing with friends).

A significant concern in collecting data for unsupervised activities using devices such as smartphones is the reduction in data quality [26,27]. The activities need to be clearly defined and simple to execute with limited opportunity for alternative interpretation. Algorithms must have a level of flexibility to ensure there is some resilience against variation in the orientation of devices and varied sensor performance across smartphone devices (usually compared using step count variance [28-30]). In this study, the data analysis algorithms were implemented offline, after the sensor was uploaded to the server (thus removing the provision of feedback that many participants desired); near real-time analysis will be required for the app to be practical in the future, and this will require suitable detection of invalid recordings to avoid false feedback, particularly for clinical assessments. Data quality remained high, however, particularly for the StS task, where 93.5% of recordings were valid and able to be analyzed. Other studies have also confirmed smartphone sensors can provide a reliable and consistent method of capturing this activity [9,31]. The SLB was more concerning, with only 63.8% of recordings being usable. This reflects the challenge of the task, highlighted by some participants in the interviews, with one participant stating it also felt unsafe. Therefore, it is recommended that the StS activity is an activity that produces robust data for smartphone assessments while also being a useful functional assessment clinically [24]. On the other hand, the SLB task in its current form used in the trial is less practical for unsupervised, digitally recorded assessments. This could potentially be resolved by gradually increasing the difficulty of such a task. For example, rather than immediately requiring participants to complete a full 30 seconds from the start, the duration could be slowly increased over time, monitoring and adjusting the difficulty as users achieve goals.

The mean SUS score provided by the participants after using the app for 4 weeks was somewhat surprisingly high. However, it has been noted that smartphone apps do typically score highly on the SUS, with a study investigating a range of apps reporting an average SUS score of 77.7 [32]. Our mean score of 81.2 is therefore in the upper range expected of apps. This is likely due to the simplicity of the interface we had in this prototype; the limited features and options implicitly made the app relatively easy to use. Adding more features (which participants strongly indicated were required) would therefore benefit from a co-design approach, such that potential end users would be involved in the design cycle to ensure usability is not lost as the complexity of the app increases.

There were some limitations emerging from this study. Unfortunately, due to information not collected on some participants and issues with the lead university not agreeing to licensing terms on the musculoskeletal questionnaire to be used, we gathered limited demographic information—particularly in terms of the overall age distribution of participants and level of function and mobility. However, in a related study in which some of the participants from this study also participated, it was reported that 80% of participants had some form of current musculoskeletal condition [26].

In terms of the app, it is likely that the prototype nature of the app meant its interface and lack of feedback had some impact on perceptions. However, this also had the benefit of being able to assess the core features required to facilitate the use of the smartphone to record these functional activities. Moreover, the process of using and holding the phone to complete the assessments without supervision was a key aspect of the study, and one that was achieved using the prototype app.

It was found that it is feasible for people to use the built-in inertial sensors in a smartphone to record physiotherapy-related functions. However, activities should be limited to simple, stable activities such as an StS task, rather than activities that lead to instability, such as SLB. It is recommended that more exercises be evaluated from a clinical, patient, and data quality perspective to establish a portfolio of activities that patients can perform independently and be automatically monitored objectively via an app. Feedback is essential and therefore requires more study into how this can be optimized to achieve long-term adherence. Overall, smartphone-based monitoring of functional activities can facilitate unsupervised, remote assessments, thus reducing the burden on physiotherapy services and increasing the ability to monitor progress objectively.