The Track-OA feasibility study [26] used a randomized, delayed-control design, whereby the randomization determined the timing of when the intervention was provided (ie, immediately vs a 1-month delay). As such, preliminary efficacy could be assessed within a conventional RCT (ie, with an intervention group and a control group) at 1 month, while all participants received the intervention after 1 month. This study design is the best suited for complex interventions with components that are likely beneficial and present a low risk to participants (eg, promoting physical activity). By ensuring that all participants receive the intervention at the end of a study, this design might promote protocol compliance.

Eligible individuals were patients who had a physician-confirmed diagnosis of knee OA, or passed 2 criteria for early OA: (1) being age 50 years or older, and (2) having experienced pain or discomfort in or around the knee during the previous year lasting 28 or more separate or consecutive days. In a community-based study by Marra et al [27], 191 of 195 (98.0%) urban-dwelling participants who met these criteria also met the American College of Rheumatology clinical criteria for knee OA [28].

We excluded individuals who (1) had a diagnosis of inflammatory arthritis, connective tissue diseases, fibromyalgia, or gout, (2) had used disease-modifying antirheumatic drugs or gout medications, (3) had knee arthroplasty, (4) were on the waitlist to receive total knee arthroplasty, (5) had acute knee injury in the past 6 months, (6) did not have an email address or daily access to a personal computer with Internet access, (7) had a body mass index of 40 kg/m² or more, (8) had received a steroid injection in the last 6 months, (9) had received hyaluronate injection in a knee in the last 6 months, (10) were using medications that impaired activity tolerance (such as β-blockers), or (11) had an inappropriate level of risk for increasing their unsupervised physical activity. Potential participants completed the Physical Activity Readiness Questionnaire (PAR-Q; 2014 version) [29]. If the PAR-Q identified a potential risk, we required physician confirmation to ensure that the person was able to be physically active without the supervision of a health care professional.

We recruited participants from 3 sources: (1) postings on Facebook, Twitter, Kijiji, Craigslist, and the Arthritis Research Canada website, (2) emails sent by the Arthritis Consumer Experts (Vancouver, BC, Canada), a nonprofit patient education organization, to their patient members, and (3) emails sent by the Vancouver Coastal Health Research Institute (Vancouver, BC, Canada) to its staff. Interested individuals were invited to contact the research coordinator, who provided details about the study, screened respondents for eligibility, and obtained their informed consent.

After completing the baseline assessment, participants were randomly assigned to the immediate group or the delayed group in a 1:1 allocation ratio. The delayed group received the same intervention as the immediate group after a 1-month wait. Random numbers were generated in variable block sizes for the random allocation.

The intervention involved participants attending a 1.5-hour session, where they received (1) a standardized group education session about physical activity, (2) a Fitbit Flex (Fitbit, Inc, San Francisco, CA, USA), and (3) individual weekly activity counselling with a PT by telephone. The education session, delivered in groups of 2 to 3 participants, addressed the benefits of physical activity, the detrimental effects of sedentary behavior, and ways to be active without aggravating OA symptoms. The counselling component followed the brief action planning approach [30], whereby the PT guided participants to identify their activity goals, develop an action plan, identify barriers and solutions, and then rate their confidence in executing the plan. The process was repeated until the confidence rating reached at least 7 out of 10, indicating that the person was confident about implementing the plan. For sedentary behavior, the PT began by asking participants to estimate their sitting time in a normal day and identify ways to break up the sitting time. They then repeated the goal setting and confidence assessment.

Participants were then provided a Fitbit to be worn at the wrist of the nondominant side to track their physical activity behavior. They were instructed to wear the fitness band 24 hours a day except during water-based activity or when charging the device. The data were wirelessly synchronized with Fitbit’s online dashboard that could be viewed only by the participants and their study PTs. During the intervention period, the PT reviewed each individual’s physical activity on the dashboard and progressively modified the activity goals during 4 weekly 20-minute telephone calls. Participants could also contact the PT via email. At the end of the intervention, they could keep the Fitbit, but no longer had access to the PT.

Guided by Bowen et al [31] and Thabane et al [32], the feasibility assessment focused on implementation, practicality, and preliminary efficacy. We measured implementation by the recruitment rate, dropout rate, adherence to the study protocol, and equipment retention. We aimed to achieve the following: (1) at least 80% of eligible individuals agreeing to participate, (2) no more than 10% dropping out, (3) at least 85% adhering to the intervention and assessment protocol, and (4) no more than 10% loss or malfunction of the 20 SenseWear accelerometer devices (BodyMedia, Pittsburgh, PA, USA) used in the study (for measuring the primary outcome; see below). We assessed practicality by self-reported adverse events and adherence to the assessment protocol. Specifically, participants were required to wear a SenseWear armband monitor for at least 20 hours during at least 4 of the 7 days of each evaluation period [33] and to complete all questionnaires within 7 days of the scheduled date. We assessed preliminary efficacy by examining outcome measures at baseline, and at the end of months 1 and 2.

The primary outcome was mean time spent in bouted MVPA per day. We defined a bout as at least 10 consecutive minutes at the level of at least 3 metabolic equivalent tasks (METs; ie, the lower bound of MVPA), with allowance for interruption of up to 2 minutes below the threshold [34]. Participants received a SenseWear Mini armband sensor by courier and wore it 24 hours a day for 7 consecutive days, with the exception of removal for water-based activities. Unlike Fitbit, which is a commercial activity tracker with important limitations in measurement accuracy [35], SenseWear is a research-based accelerometer and sensor with established measurement properties [36]. Tierney et al [37] showed that SenseWear is a valid tool for estimating energy expenditure during daily activities in people with arthritis (intraclass correlation coefficient 0.72). Additional analysis was performed with a cutoff at 4 or more METs, which reflects an activity level of brisk or faster walking (ie, purposeful activities) [38].

Secondary outcomes were the mean time spent in sedentary behavior, the Knee Injury and Osteoarthritis Outcome Score (KOOS) [39,40], and the Partners in Health Scale [41]. Compared with Fitbit, SenseWear is a superior outcome measure because of its ability to differentiate between sedentary and light activities [36]. We calculated the mean daily time spent with an energy expenditure of at least 1.5 METs, occurring in bouts of more than 20 minutes during waking hours [18,21,42,43]. The KOOS consists of 5 subscales: Pain, Symptoms, Activity of Daily Living, Sports and Recreation Function, and Knee-related Quality of Life. It was originally developed for people recovering from anterior cruciate ligament and meniscus injury and has been validated in people with OA [39,40]. The Partners in Health Scale is a 12-item measure designed to assess perceived self-management capacity via subjective knowledge of the health condition and treatment, and perceived self-management behavior (eg, adopting a healthy lifestyle) (Cronbach alpha=.82) [41]. We also tracked self-reported adverse events (falls, cardiovascular and musculoskeletal events) [44] using a monthly log completed by the participants.

With the resources available for the feasibility study, we aimed to recruit 30 participants within an 8-week period. We used descriptive statistics to summarize the feasibility variables and baseline variables of the 2 groups. We explored preliminary efficacy using intention-to-treat analysis. Q-Q plots were used to assess normality of the outcome variables. We conducted analysis of covariance to evaluate the effect of the group type (immediate vs delayed) on the outcome measures assessed at the end of 1 month (T1) and 2 months (T2), after adjusting for blocking and baseline (T0). We assessed 3 planned contrasts of changes in outcome measures over the measurement periods. The first contrast compared T0 to T1 between the 2 groups to determine whether the intervention was superior to the control. The second contrast compared T0 to T1 in the immediate group against T1 to T2 in the delayed group. The third contrast compared T0 to T1 in the immediate group against T0 to T2 in the delayed group. The last 2 models assessed whether the 1-month delay had an impact on the effect of the intervention.

We assessed the impact of missing data on the estimated effects of group assignment using imputation methods as described in van Buuren [45]. Specifically, we generated 10 imputed values using alternative random variates derived in a linear regression model, which included group, sex, baseline age, and baseline body mass index as predictors. We repeated the analyses using the 10 imputed values, and compared the conclusions and estimates against the main analysis, which assumed that data were missing at random.

The research protocol was approved by the University of British Columbia Clinical Research Ethics Board (application number: H14-02631), was registered with ClinicalTials.gov (NCT02313506), and has been published in the peer-reviewed literature [26].

Our recruitment strategy identified 46 eligible individuals; of those, 34 (74%) enrolled and none dropped out (Table 2). All but 1 participant adhered to the intervention protocol. All participants completed the assessments as per protocol at T0 and T1. Participants were required to wear a SenseWear (the primary outcome measure) for at least 4 days [46], with each day requiring less than 4 hours of off-body time to be included in the analyses. All 34 participants met these wear criteria at T0 (mean number of days worn: 5.9, SD 0.3; mean off-body time: 23.1, SD 13.6 minutes) and T1 (mean number of days worn: 5.6, SD 0.7; mean off-body time: 24.8, SD 18.0 minutes). At T2, 31 participants adhered to the wear criteria (mean number of days worn: 5.7, SD 0.6; mean off-body time: 38.4, SD 27.9 minutes). In the delayed group, 1 participant did not complete the outcome measures.

Figure 2 shows the results of outcome measures from 3 time points. Prespecified contrast analyses revealed a significant effect whereby the immediate group improved in the MVPA (≥3 METs) time at T0 to T1 compared with the delayed group (contrast coefficient –31.1, 95% CI –56.6 to –5.7; P=.02) (Table 3). We also found a significant effect in the Partners in Health Scale scores at T0 to T1 (contrast coefficient 10.9, 95% CI 2.5-19.3; P=.02). The planned contrast of the immediate group at T0 to T1 versus the delayed group at T1 to T2 showed a significant effect in sedentary time (contrast coefficient –83.6, 95% CI –154.1 to –13.1; P=.03) and the KOOS symptoms subscale (contrast coefficient 6.9, 95% CI 0.4-13.5; P=.048), favoring the delayed group at T1 to T2. We found no significant effect in any outcome measures in the contrasts comparing the immediate group at T0 to T1 with delayed group at T0 to T2.

Results from the imputation analysis (data not shown) were in line with the main missing-at-random analysis, in estimates, standard errors, and P values of group effects. This suggests that the presence of missing data did not have an important effect on the findings. No adverse event associated with the intervention was reported by participants during the study.