Setting, eligibility, and enrollment
B-Well was conducted at Brigham and Women’s Hospital, an academic medical center in Boston, Massachusetts. The study ran from July 2014 to June 2015 and was funded to include 300 participants. All participants received the financial incentives intervention; there was no control arm. Employees were recruited via a limited number of hospital-wide emails and advertisements. Eligible individuals were older than 21 years, reported ≤30 min of exercise per week, and had access to a computer. Weekly minutes of exercise were assessed with a question asking participants if they exercised for 30 min on 0, 1, 2, 3, or ≥4 days of the week. Physicians and nurses were not eligible for participation, as they had access to other similar PA programs. Pregnant women were excluded.
Eligible employees formed three-person teams, created independently or by B-Well staff, to provide peer support. Each team met with study staff to review the study protocol, PA monitoring schedule, and financial incentive details. At this meeting, each participant was given a Fitbit Flex activity monitor (Fitbit Inc., San Francisco, California) and instructions for its use.
Trained research staff collected health data at the baseline visit. Participants provided written informed consent and completed an online questionnaire containing questions related to general health, healthcare utilization, quality of life (measured by EQ-5D), and perceptions of PA. Participants self-reported comorbidities including heart disease, hypertension, diabetes, osteoarthritis, and others. Musculoskeletal disease was assessed with a validated musculoskeletal index .
Participants also completed an assessment to determine their degree of delay discounting. Delay discounting, a measure of impulsivity, describes the extent to which individuals prefer immediate, smaller rewards to larger, long-term gains [12, 25, 26]. Based on choices between receiving a specified monetary amount now or receiving a larger amount later, participants were assigned a value on a delay discounting index, with higher values indicating greater preference for immediate rewards. We used a method developed by Kirby and Petry to estimate the index .
Intervention and financial incentives structure
Participants were asked to wear their Fitbits continuously for 26 weeks. During the first 2 weeks, participants were in the pre-intervention period and were compensated for wearing the Fitbit. They received $20 for wearing the Fitbit for at least 10 days and an additional $10 for wearing it for all 14 days. Teams in which all members met the 10 day criterion received an additional $10 per member. Results from these 2 weeks were considered baseline levels of PA.
Financial incentives, chosen to be meaningful but not coercive, were available for the subsequent 24 weeks. Participants could receive $10 per week for increasing minutes of moderate-to-vigorous PA (MVPA) by at least 10% from the preceding week or for meeting Centers for Disease Control and Prevention (CDC) PA guidelines (150 min of moderate PA per week) . Individuals meeting their MVPA goal every week for one, three, or 6 months received an additional $15, $25, or $50 respectively. If all team members successfully met their MVPA goals, each individual’s reward was doubled that week. In total, participants could receive up to $860 by meeting MVPA goals. These amounts fall within the range identified by a 2013 systematic review on financial incentives for exercise session attendance, which noted that successful programs had weekly incentives ranging from $2.79 to $46.82 .
Participants received individualized weekly emails during the intervention detailing their levels of recent MVPA, their weekly goals, and their recent rewards. While participants could view activity data (e.g. step counts) on the Fitbit website, this data did not include the minutes of MVPA that counted toward their goals. Participants could visit the B-Well study website to monitor current and past individual and team progress.
The Fitbit Flex is a commercial accelerometer that records the number of steps taken each minute. We chose the Fitbit Flex because it presents an adequate trade-off between validity and expense; it is less expensive than medical grade accelerometers and has moderate validity for measuring step count and physical activity [28,29,30]. We downloaded minute-by-minute step counts weekly and defined thresholds of 100 and 175 steps/min for moderate and vigorous PA, respectively. The threshold of 100 steps/min was selected as it corresponds with an intensity of ≥3 metabolic equivalents of task (METs) [31, 32]. As there is limited data on the association between step count and vigorous activity, we used the conservative threshold of 175 steps/min. We weighted minutes of vigorous activity as twice that of moderate activity, as CDC guidelines recommend either 150 min of moderate activity, 75 min of vigorous activity, or an equivalent combination per week . Days with fewer than 10 h of wear time were excluded from analysis. To count towards the weekly goal, activity had to be performed in bouts of at least 10 min, in accordance with the CDC PA guidelines . Within a single bout, we allowed up to two ‘grace’ minutes during which the participant’s step count could fall below the thresholds.
Our algorithm identified the start of potential bouts of moderate PA either as a minute with more than 100 steps preceded by a minute of fewer than 100 steps or as 100 steps in the first minute of the day. Bout length was defined as the number of consecutive minutes during which the step count met the 100 steps/min threshold before the two grace minutes were exceeded. Bouts of fewer than 10 min were not counted. The algorithm then conducted consecutive pair-wise comparisons of each bout of at least 10 min. When two bouts overlapped, the shorter of the two was discarded and the pair-wise comparison continued until no bouts were overlapping. The algorithm repeated with a 175 steps/min threshold to identify bouts of vigorous PA.
We assessed intervention efficacy on several domains. First, we calculated the average number of minutes spent in bouts of MVPA on a weekly basis and compared these values during the intervention period to those during the pre-intervention period. Second, we measured Fitbit adherence among enrolled participants by counting weeks in which participants wore the Fitbit for at least 10 h per day for at least 4 days of the week. We defined adherent Fitbit wearers as participants with at least 20 adherent intervention weeks (out of 24). Third, we evaluated the number and proportion of study participants who met their weekly MVPA goals or CDC PA guidelines during the intervention.
We employed generalized linear modeling to determine factors associated with the number of weeks meeting CDC PA guidelines. The multivariable analyses included demographic and clinical variables and the delay discounting index, which was divided into the highest quartile (the highest degree of discounting future benefits) vs. the lowest three quartiles combined. Based on the multivariable model results, we estimated the adjusted mean number of intervention weeks during which subjects achieved MVPA levels consistent with CDC PA guidelines. We initially built bivariate models; covariates that reached a significance level of α = 0.15 were advanced to multivariable models. We used intraclass correlation to measure within-team correlation for meeting CDC guidelines. We calculated within-team correlation for minutes of MVPA over time to examine whether within team correlation was based on selection (active participants form teams with active participants) or adaption (teams are mixed with active and inactive participants at the start and correlation increases over time). All statistical analyses were done in SAS 9.4.