Participants in this secondary data analysis were enrolled in a randomized controlled trial of a 12-week physical activity intervention. Data were collected from February 2015 to July 2016. The University of California, San Diego institutional review board approved all study procedures, and all participants provided written informed consent. The trial was registered with Clinicaltrials.gov (NCT 02332876). Eligible participants were female breast cancer survivors, in the age range of 21 to 85 years, who were diagnosed less than 5 years before study enrollment, had completed chemotherapy or radiation treatment, were sedentary (defined as self-reporting less than 60 min of MVPA in 10 min bouts per week), and had access to the Internet and a Fitbit-compatible computer, tablet, or phone. Exclusion criteria included any medical condition that could make it potentially unsafe to be in an unsupervised physical activity intervention (determined by the Physical Activity Readiness Questionnaire [44]), other primary or recurrent invasive cancer within the last 10 years, and unable to commit to a 12-week intervention.

Out of 911 women who were screened for eligibility, 108 were eligible, and 97 came to the baseline visit. Most common reasons for being ineligible included being too active (n=225), unable or unwilling to attend clinic visits (n=106), breast cancer surgery more than 5 years ago (n=81), and medical exclusion (n=36). At the baseline visit, 10 women were deemed ineligible (high blood pressure, n=8; physical limitation, n=2). A total of 87 participants were randomized to the exercise arm (n=43) or the control arm (n=44). One participant from each arm was lost to follow-up, resulting in a 97.7% retention rate (exercise n=42, control n=43) [45]. The current analyses comprise data from the 42 participants who were randomized to the exercise arm and completed the study.

A detailed description of the protocol was previously published [46]. Briefly, participants were predominantly recruited via cancer registry lists. Potential participants were telephone-screened to determine eligibility. Interested and eligible women were scheduled for an in-person visit where they were given an ActiGraph GT3X+ accelerometer to wear for 7 days and bring back to the randomization visit. At the randomization visit, participants in the exercise arm were given a Fitbit One as part of the intervention. Participants were instructed to use their Fitbit to self-monitor their physical activity. As the intervention focused on Fitbit’s “Active Minutes,” which consists of MVPA and did not focus on steps, participants were encouraged to wear the Fitbit when engaging in MVPA but were not instructed to wear it for a minimum amount of time each day. Participants were informed that their Interventionist would check their Fitbit data at least once a week and that they may be contacted by the interventionist if it seemed that they were struggling or having a great week. All participants received intervention phone calls around the 2-week and 6-week time points and automatic emails every 3 days throughout the 12-week intervention, which included reminders to sync and wear their Fitbit. One week before their final study visit, participants were mailed the ActiGraph GT3X+ and asked to wear it for 7 days, concurrently with the Fitbit, and to bring the ActiGraph to the in-person visit. At the final in-person visit, participants completed a questionnaire regarding their use of the Fitbit tracker and the Fitbit app and website.

The Fitbit One, a commercially available accelerometer-based activity tracker, was used to examine patterns of physical activity throughout the 12-week intervention. Fitbit uses a proprietary algorithm to classify each minute as being in sedentary, light, moderate, or vigorous activity and provides metabolic equivalent of tasks (METs) for each minute. Data were wirelessly uploaded to the user’s fitbit.com account and then downloaded by the research team through a database called Fitabase (Small Steps Lab, San Diego, CA), which allows for collecting data at the minute level. Daily adherence to wearing the Fitbit tracker was defined as wearing the tracker for >10 hours in a day or logging at least some activity (>1 min MVPA). This definition for a valid Fitbit wear day was used because participants were not instructed to wear the Fitbit all day; rather they were instructed to use the Fitbit to track activity. Thus, wearing the tracker specifically to log MVPA was deemed to be valid wear based on these instructions. Fitbit wear time was determined by processing of minute level Fitbit data using the R function accel.weartime within the “accelerometry” package [47]. Nonwear was classified using both steps and METs. Consistent with standard protocols for ActiGraph wear time [48], greater than 90 consecutive minutes of 0 steps or METs, with 2-min tolerance (ie, for 2 min with nonzero counts during nonwear intervals) was deemed nonwear.

The ActiGraph GT3X+, a well-validated research grade accelerometer [49], provided frequency, duration, and intensity of physical activity for 7 days at baseline (prerandomization) and at week 12. Using standard guidelines, sufficient ActiGraph wear time was classified as >10 hours of wear a day for at least 5 days or >50 hours across 4 days and screened for in the ActiLife software (ActiGraph, Pensacola, FL) using guidelines outlined by Choi et al [48]. All complete and valid data were processed in ActiLife software using the low frequency extension and aggregated to 60-second epochs so that published physical activity cut points could be applied [50]. MVPA was defined as 1952 or more counts per minute (3.00-7.00 METs).

Self-report questionnaires at follow-up (12 weeks) were used to determine participants’ frequency of looking at their activity data on the Fitbit tracker itself and (in a separate question) the Fitbit app or website. These questions used an 8-point Likert scale with the following response options: more than once per day, once per day, 4-6 times per week, 2-3 times per week, once per week, 2-3 times per month, once a month or less, and never.

The distribution (mean [SD] and n [%]) of participant demographics and breast cancer characteristics were calculated at baseline for the analytic sample. ActiGraph-measured physical activity, measured at baseline and follow-up, was described using mean (SD). The agreement between ActiGraph- and Fitbit-measured MVPA was assessed by calculating the concordance correlation in days with overlapping wear.

Participants were 42 female breast cancer survivors who were predominantly diagnosed at stage 1 (62%, 26/42). About half had received chemotherapy, and about three-fourths were currently taking an aromatase inhibitor or tamoxifen. They were an average of 58 years old (SD 11.3), with the majority being non-Hispanic (81%, 34/42), white (83%, 35/42), and having a college education or greater (69% [29/42]; Table 1). The intervention group significantly increased ActiGraph-measured MVPA from baseline (93.8 min/week, SD 90.79) to 12 weeks (195.3 min/week, SD 105.9, P<.001) [45]. Minutes of ActiGraph-measured MVPA at week 12 was highly correlated with Fitbit MVPA collected on overlapping days (r=.81: ActiGraph MVPA/day mean 29.9, SD 25.90, Fitbit MVPA/day mean 25.8, SD 28.76).

Adherence to wearing the Fitbit was high, with a mean number of valid days across the 12-week intervention period of 88% (SD 14), median of 95%, and range of 31% to 100% of intervention days. Each week, participants wore the Fitbit on average 6.2 out of 7 days (88.5% per week, SD 1.8). Although adherence to wearing the Fitbit appeared to decrease in the middle of the intervention period (Figure 1), adherence did not significantly differ across the 12 weeks (P=.71).

Greater adherence to wearing the Fitbit was associated with greater increases in ActiGraph-measured MVPA (b_interaction=0.35 P<.001). Someone with the median amount of valid wear days (95%) had an expected ActiGraph-measured MVPA increase of 109.8 min/week, whereas someone with the first or third quartile amount of valid wear days (80% and 98%, respectively) had an expected ActiGraph-measured MVPA increase of 73.3 min/week and 117.2 min/week, respectively.

Across the 12 weeks, participants averaged 182.6 minutes/week (SD 143.9) of MVPA on the Fitbit. Minutes of MVPA per week significantly differed over the 12 weeks (F_11/392=1.91, P=.04; Figure 2). Weeks 3 and 9 had the highest average MVPA with 222.9 min/week (SD 173.4) and 198.4 min/week (SD 167.9), respectively. Weeks 12 and 5 had the lowest average MVPA with 159.7 min/week (SD 128.4) and 168.1 min/week (SD 123.5), respectively.

At study completion, participants answered a series of questions regarding use of the Fitbit tracker and Fitbit website or mobile app. Two participants did not answer these two items. Of the 40 participants who answered these items, 68% (27/40) reported looking at their activity data on the Fitbit app or website once a day or more; 13% (5/40) reported looking at the app or website less than once a week; 68% (27/40) of participants reported looking at the Fitbit tracker itself once a day or more, whereas 10% (4/40) reported looking at it less than once a week. Exactly half (50%, 20/40) of the participants reported looking at both the Fitbit tracker and the app or website at least once per day (Table 2).

Frequency of looking at one’s data on the Fitbit app or website, controlling for adherence to wearing the Fitbit, was not associated with change in ActiGraph MVPA (P=.36). There was a negative association between looking at one’s data on the Fitbit tracker and change in ActiGraph MVPA (b=−1.36, P=.07), controlling for adherence to wearing the Fitbit. Participants who reported looking at the tracker more frequently had smaller increases in MVPA than those who looked less often. When this analysis was carried out on the combined binary variable of looking at the Fitbit tracker and the app or website at least daily, we found no association with change in ActiGraph MVPA (P=.87).

This study is one of the first to take an in-depth look at use of a commercially available wearable activity tracker and how it relates to changes in physical activity. Using minute level data collected from the Fitbit, adherence to wearing the tracker was high and stable across the 12-week intervention period. This is generally consistent with previous research [30,33,51]; however, one recent observational study found linear decreases in Fitbit use over a year [52]. This suggests that Fitbit use long term and not within an intervention may be different than was seen in this study. One challenge of comparing Fitbit use across studies is that how adherence to wearing the tracker was calculated or defined is often not reported [30,33,51,52]. To our knowledge, this is also one of the first studies to explore the relationship between use of a commercially available activity tracker and success in a physical activity intervention where the physical activity was also measured by an ActiGraph, the gold standard measure for free-living physical activity in research. The positive association between wearing the Fitbit and increased MVPA suggests that using the real-time data to determine if someone is wearing their tracker could help to identify individuals who may need additional support, or possibly other self-monitoring methods, to support behavior change.

In our analyses, MVPA significantly varied throughout the intervention. Interestingly, the highest minutes of MVPA occurred at week 3, immediately after the intervention call, which typically occurred around the end of week 2, and at week 9, which was around when participants were contacted to confirm their final visit at 12 weeks. This highlights the importance of personal contact with participants in a physical activity intervention and is consistent with research that has found greater benefit for combining technology-based self-monitoring with counseling than using technology alone [53].

A novel aspect of wearable trackers is that they can provide objective feedback on MVPA. Previous studies with traditional pedometers could only provide feedback on steps, which captures activities of all intensities. We identified only two other published physical activity interventions in which participants set goals explicitly on Fitbit’s active minutes and examined changes in Fitbit-measured active minutes as one of the primary study outcomes [33,54]. Given the numerous benefits of MVPA [55-57], the capability of trackers to automatically collect information on MVPA may be useful in helping individuals meet physical activity guidelines [13].

While technology-based activity trackers make self-monitoring less burdensome compared with traditional tracking methods, they also do not require a person to attend to the information being collected by the monitor. Overall, self-reported viewing of activity data on the Fitbit itself, or on the Fitbit website or app was very high. Looking at activity data on the app or website was not associated with changes in ActiGraph-measured MVPA. Surprisingly, more frequent looking at data on the Fitbit tracker itself was associated with smaller changes in ActiGraph-measured MVPA. One reason for this finding may be that the Fitbit One tracker did not show minutes of MVPA; that information was only available on the app or website. These results could also indicate that checking one’s own data is not as important as being accountable to someone else for increasing physical activity. In this study, it was stressed that the Fitbit would be used so that the interventionist could see the data and provide support. It may be that being accountable was a greater motivating factor for increasing MVPA than being self-aware of one’s own activity levels. Much of the field’s understanding of the importance of self-monitoring is based on active self-monitoring, which typically require a person to think about their day and record their minutes of activity, but this record is often not easily or immediately shared. As technology-based trackers become more common place in interventions, we need to continue exploring the impact of active versus passive self-monitoring and the role of accountability on behavior change so that our understanding of the role of monitoring physical activity is consistent with new and emerging technologies.

Although this in-depth analysis of daily activity data from a commercially available activity tracker is an important addition to our understanding of how trackers are associated with behavior change, several limitations should be noted. This study comprised a small sample of a relatively homogenous group of breast cancer survivors, and results may not be generalizable. Fitbit does not share information on number of times a person checks the Fitbit app or website; therefore, checking of app or website relied on self-report, which unfortunately had little variation, with most participants reporting looking at their Fitbit or the app or website daily. Additionally, self-report questions assessing use and engagement with the tracker were only asked at the end of the intervention, limiting our ability to examine trends in engagement throughout the intervention. In addition, we used a question combining website and app use and could not examine those two modalities separately. The intervention period was relatively short—long-term use of an activity tracker and its relationship with increasing physical activity could not be assessed. Although we also used standard cut-points for determining ActiGraph-measured MVPA, future studies should consider using machine learning algorithms to classify ActiGraph-measured behaviors. Finally, the intervention included many reminders to wear the Fitbit, so adherence results may not be representative of what would happen outside of an intervention protocol.

With the continued emergence of new technologies for self-monitoring physical activity, it is important to understand how people use these new devices and how use of these devices can support behavior change. These insights may enhance our ability to effectively utilize activity trackers to promote behavior change.