This study includes a post hoc analysis of a randomized controlled trial, titled Remote Monitoring in Specialist Health Care Study (ReMonit, ClinicalTrials.gov: NCT05031767). The primary results of the trial are reported elsewhere [25]. The ReMonit Study was a 3-armed randomized controlled trial comparing remote monitoring and patient-initiated care to usual care (prescheduled regular hospital visits) among patients with axSpA [26]. Patients were recruited from an outpatient clinic at Diakonhjemmet Hospital, Oslo, Norway, and randomized 1:1:1 ratio to receive usual care, remote monitoring, or patient-initiated care. Patients randomized to the remote monitoring and patient-initiated groups were asked to use the activity tracker for 1 year. This was a commercially available activity tracker (Garmin vívosmart 4). Physical activity data recorded by the vívosmart 4 was wirelessly transferred manually by patients each week via Garmin Software Development Kit to the MyDignio app [27].

In the ReMonit study, patients with axSpA with low disease activity (Axial Spondyloarthritis Disease Activity C-Reactive Protein Score (ASDAS-CRP) <2.1) and stable treatment with a tumor necrosis factor inhibitor (TNFi) over the past 6 months were included [26]. Since we used the Garmin Software Development Kit for the present study, patients who already used a Garmin device and the Garmin Connect app could not be included, as this interfered with synchronization of data to the MyDignio app. In addition, due to privacy regulations, direct downloading of data from the patients’ private Garmin devices was not allowed. Therefore, patients could not use their private devices in the present study.

At baseline, patients completed a digital questionnaire including age, sex, education level, and working status. BMI was calculated based on self-reported body height and weight, and information on years since diagnosis of axSpA was obtained from their medical records. The patients completed the recommended disease-specific questionnaires [28] such as Bath ankylosing spondylitis disease activity index (BASDAI) (0‐10, 10 being the worst score), Bath ankylosing spondylitis functional index (BASFI) (0‐10, 10 being the worst score), and patient global assessment (0‐10, 10 being the worst score) [29,30]. The ASDAS-CRP was calculated based on patients’ self-reported disease activity-related questions and measurement of C-reactive protein [31]. The Work Productivity and Activity Impairment item 6 (WPAI) (0‐10, 10 being the worst score) was used for measuring the impact of the disease on patients’ everyday life [32]. In addition, eHealth literacy was measured by 4 scales from the eHealth Literacy Questionnaire [33]: Using technology to process health information (Scale 1), Ability to actively engage with digital services (Scale 3), Feel safe and in control (Scale 4), and Motivated to engage with digital services (Scale 5). For measuring self-reported physical activity and exercise, we used 3 items from a population-based study measuring the frequency, duration, and intensity of exercise [34]. We then further stratified the patients into 3 groups based on whether the level of physical activity was below recommended (below 150 min of moderate-intensity exercise per week), at recommended (150 moderate or 60 vigorous minutes each week), or above recommended [35].

To assess the feasibility of using activity trackers for measuring physical activity, we categorized feasibility into 3 novel subcategories inspired by a previous feasibility study on activity trackers [18].

The patients received oral and written instructions stating that the activity tracker should be worn on the non-dominant wrist for at least 10 hours a day. It should not be worn on the outside of garments but could be used while swimming and showering. Instructions were also given regarding charging, connecting, and synchronizing the device with the app. Contact information for the study team was provided in the event of technical issues.

The Garmin vívosmart 4 activity tracker (Olathe) was integrated into the MyDignio app by the company Dignio. Physical activity data were recorded at a minute level, measuring both steps per minute and average heart rate during the concurrent minute. The Garmin vívosmart 4 uses an accelerometer for measurement of physical movement and activity and a photoplethysmography sensor for measuring heart rate [37].

In order to reduce data noise from the activity tracker, a filter was applied by Dignio, which filtered out heart rate values below 20 beats per minute. The native Garmin motivational messages and notifications were muted. Patients were instructed to weekly synchronize physical activity data manually from the device using the MyDignio app due to limited internal memory of the activity tracker. If patients failed to synchronize in time, the MyDignio app automatically sent a push notification as a reminder. If patients still did not synchronize after receiving the push notification, the study team followed up with an SMS text message reminder. The number of manual SMS text message reminders sent out was not registered. Any technical issues regarding app connectivity or the activity tracker were primarily resolved by the study team, or if necessary, by developers at Dignio.

Median values with IQR or mean values with SD were used for describing the demographical and clinical variables. Data regarding the trial, technical, and operational feasibility were described by either percentage, mean with SD, or median with IQR. Differences between patients agreeing to use a wearable compared to those declining were assessed by the Mann-Whitney U test or Student’s t-test.

To assess wear-time patterns based on the data returned from the activity tracker, we used the hourly level data, which were summed up on a weekly level. Each patient’s first week of recording was assigned as their baseline week. Unsupervised hierarchical clustering using the pheatmap package in RStudio was used for assessing different wear-time patterns based on the number of hours of physical activity data recorded [38]. The optimal numbers of predefined clusters were based on the visual inspection of the clustered heatmap and internal validation from the R-package clValid [39]. The cluster groups were named according to their observable patterns that occurred within each group. Patient characteristics in the clustering groups for wear-time patterns were compared and tested for significant differences using the Kruskal-Wallis test, and Dunn test if the initial Kruskal-Wallis test had a P-value below 0.05. The Fisher exact test was used for testing differences in the distribution of sexes between the groups. Preparation and analysis of data were conducted using RStudio and Stata (version 18.0; StataCorp LLC).

The study was conducted according to the Helsinki Declaration. All patients provided written consent to participate and were informed that they could stop using the activity tracker at any given time during the study. Data were de-identified and stored at a secure research server. The Regional Committees for Medical and Health Research Ethics South-Eastern Norway approved the study (ref: 229187).

In total, 2 patient research partners were involved in the planning of the study. They also pilot-tested the activity tracker and gave feedback on the instructions to the patients and on the study logistics. One of the patient research partners (SH) also contributed to analyzing and discussing the results and is a co-author of this manuscript.

A total of 75 (47%) out of 160 patients agreed to use the activity tracker for 1 year and were included in this study (Figure 1). Among the 85 non-participating patients, 49 (31%) declined to wear an activity tracker, 29 (18%) were already users of a Garmin activity tracker device, and 7 (4%) could not wear an activity tracker due to uniform regulations at their workplace (Figure 1). Baseline characteristics were mostly similar between patients who agreed and those who declined to use the activity tracker, with minor numerical variances in self-reported physical activity levels (Table 1). Patients who declined had significantly lower eHealth literacy scores compared with those who agreed to use the activity tracker, but the numeric differences in mean scores were small (Table 1).

Among the 75 patients who agreed to use the activity tracker, 64 (85%) had valid physical activity data and a median adherence to use the activity tracker of 66% (IQR 30‐86) (Figure 1). Over the 1-year measurement period, 8 out of 75 patients never recorded any data. In total, 3 patients returned less than 1000 minutes of data recorded (corresponding to approximately a total of 16 h of physical activity data) and were subsequently excluded from the wear-time pattern analyses during aggregation of data (Figure 1). A total of 12 million datapoints were returned, corresponding to a mean of 191,677 (SD 116,564) minutes of data per patient (Figure 1). Analyses showed that the proportion of missing data was higher for the steps (30%) compared to heart rate data (0.01%).

To remind patients to synchronize the activity tracker data, a total of 1285 (32%) out of a potential 3900 automatic push notifications were distributed, with a median of 18 (IQR 9‐25) notifications per patient. In total, 5 activity trackers required replacement during the study, of which 4 were caused by connectivity issues with the app and one due to a loss of the activity tracker (Figure 1).

Data on wear time were grouped in clusters according to the visual wear-time patterns, and 3 cluster groups were found to be the optimal number of clusters. The 3 cluster groups were categorized as “Adherent” (n=33), “Minimal use” (n=17) and “Intermittently adherent” (n=14) according to their visible wear-time patterns (Figure 2). The patients included in the Adherent group had a steady recording of physical activity data, the Minimal use group showed longer periods of no recorded data, and the Intermittently adherent group presented with a notably larger variation in the amount of recorded data (Figure 2). We found that the Adherent group had a median of 85% (IQR 72‐89) adherence to the use of the activity tracker, while the Minimal use and Intermittently adherent groups had lower adherence with 7% (IQR 4‐16) and 49% (IQR 43‐59), respectively (Table 2).

The Minimal use group had a significantly (P=.019) lower median age (35 y, IQR 32‐50) compared to the Adherent (46 y, IQR 42‐55) and Intermittently adherent groups (40 y, IQR 33‐53) (Table 2). There were nonsignificant between-group differences (P=.107) in the proportion of sex, with 23 out of 33 (70%) males in the Adherent group, 16 out of 17 (94%) in the Minimal use group, and 10 out of 14 (71%) in the Intermittently adherent group. There were no differences between the groups in disease activity, self-reported physical function, or the impact on daily activities.

This study explored the feasibility and wear-time pattern of long-term use of a commercially available activity tracker over 1 year among patients with axSpA. The results indicated a low trial feasibility and acceptable technical and operational feasibility. In total, 3 different cluster groups were identified, with the largest group demonstrating adherence to using the wearable activity tracker.

In the assessment of trial feasibility, we found a low inclusion rate for the activity tracker, with less than half of the 160 eligible patients agreeing to wear the activity tracker for 1 year. This might indicate that the willingness to use a wearable activity tracker over a long period may not be present among the majority of patients with axSpA with low disease activity. However, we believe that a higher inclusion rate could have been achieved if direct data downloads from patients’ private Garmin devices had been possible. Implementing such solutions may require specially adapted software and the use of “proxy-users” [40], potentially leading to complex data management.

We observed that a small proportion of patients either did not record any data or recorded less data than the cutoff for a valid day (>10 hours). An earlier study on patients with gout reported similar findings, showing that 33 of 44 patients had valid data, and 40% of the total data was missing [19]. Missing data, when measuring at this high-level granularity, are expected but yet remain a challenge considering the use of activity trackers in clinical settings [41]. A possible reason for the high proportion of missing steps data might be due to differences between Android and iOS phones in the interpretation of the step data, where Android phones misinterpreted inactivity as missing data instead of providing the numerical value of 0. The low number of missing heart rate data may be due to the filter that deleted observations of heart rates below 20 beats per minute, possibly resulting in a skewed representation of the number of missing heart rate data. Additionally, the discontinuation of using the activity tracker has been discussed in earlier studies, showing that perceived usefulness and inaccuracy of data could be a reason for discontinuing the use of activity trackers among patients with osteoarthritis and for healthy people [18,42]. Patients’ perception of data inaccuracy may also play a role in our study. Since we only recorded steps and heart rate as measures of physical activity, we effectively limited our recordings to step-related activities, thereby excluding popular Norwegian activities such as bicycling, cross-country skiing, and swimming.

The operational feasibility showed that a median of 18 reminders were sent out per patient, indicating a fairly low number based on the fact that reminders were sent out on a weekly level. It was also observed that the majority of tracker replacements were due to connectivity issues, which could lead to longer periods without recording physical activity.

The hierarchical clustering analysis showed that the largest cluster group was patients who showed adherence to using the activity tracker. Since this represents 33 out of a possible 160 patients, it could be argued that using activity trackers is only feasible for this specific group of patients. The between-group comparisons of patient characteristics revealed numerically small differences between the 3 wear-time cluster groups regarding age and proportion of sex. However, all 3 groups had a small sample size, which limits trust in the between-group comparisons.

Considering the potential benefits of physical activity monitoring, activity trackers may prove valuable in guiding and motivating patients to engage in higher levels of physical activity [2]. Longitudinal data on physical activity can potentially offer new insights into how axSpA impacts patients’ activity levels and can aid in developing targeted exercise and physical activity interventions. The collection of longitudinal physical activity may potentially support physiotherapists and other health care professionals in prescribing lifestyle interventions for patients with axSpA [2]. However, challenges such as low adherence and technical issues must be addressed. A potential increase in adherence might have been achieved if health care professionals had provided personalized guidance on each patient’s physical activity level and had set specific goals. These strategies warrant further research to explore their impact on enhancing adherence to using activity trackers.

The strengths of this study are the long measurement period of 1 year. This allowed for longitudinal analysis of how many patients recorded physical activity and how many discontinued their recordings. This insight further explains how activity trackers may function in clinical settings. Further strengths include the aspect of technical and operational feasibility, showing that the use of wearable activity trackers may involve an increased workload for the health care providers and require complex data management.

Limitations in the study include that we did not set cutoff values as to what constituted a valid week with physical activity measurement. This was decided because we wanted to explore the adherence to using the activity tracker. However, using >10 hours of valid day measurement has also been used by previous studies [22,40]. The hierarchical clustering analysis with the different wear-time patterns has a notable limitation, as it was conducted on a limited sample. This reduces the external validity of our study’s findings. Furthermore, the low number of patients per cluster group makes the comparison between these groups prone to biases and skewed data. Lastly, the definition of adherence used in this article may be misleading, as it uses the total amount of physical activity data that was returned. For example, technical issues related to synchronization may have led to missing physical activity data, thus not reflecting the actual use of the activity tracker.

In order to increase the feasibility of wearable activity trackers, some optimizations are needed. First, data transfer should be automated, allowing for a “passive” collection of data. In the present study, patients had to use the MyDignio app and manually transfer data each week, which might have increased the burden on patients. Second, providing additional measurements beyond steps and heart rate would provide a more comprehensive overview of physical activity levels with time spent in different intensities and perhaps would increase adherence to using the activity tracker. Lastly, using wearable activity trackers, as demonstrated in this study, results in a vast amount of data. Considering the ethical perspectives of storing massive amounts of data, a structured and well-thought-out plan for how data are to be used should be implemented before physical activity data collection is initiated.

Future research should be based upon the recently published Wearable Activity Tracker Checklist for Healthcare (WATCH), which disposes a 12-point list of aspects to consider when implementing a wearable activity tracker in health care [43]. The checklist should be used in studies, preferably conducted in a real-world clinical setting. Additionally, “bring your own device” clinical studies may also hold promise as study design by representing a more accurate resemblance of real-world settings in which some patients already are owners of activity trackers [44]. Future research should incorporate qualitative methods to explore contributing factors behind the variations in wear-time patterns and adherence to using activity trackers.

Despite the trendiness in both research on and commercial use of wearable activity trackers, long-term use of activity trackers had low to moderate feasibility over 1 year in patients with axSpA in low disease activity. The trial feasibility of using wearable activity trackers was low, while technical and operational feasibility were acceptable. Based on the wear-time patterns, we found that only 51% had consistently high activity tracker adherence. Future research should aim to ensure automated synchronization and investigate motivational factors influencing tracker adherence.