Design
The study design was a pre-implementation planning phase followed by a pilot hybrid type 2 effectiveness-implementation trial (26). This approach was important, as our pre-implementation phase identified a range of implementation and effectiveness outcomes that were important to our clinic and patient stakeholders, including implementation costs and physical function. We included a convergent mixed methods (27) approach to assess implementation outcomes with patient and clinic stakeholders qualitatively and quantitatively. We used 1:1 patient-level, parallel randomization of participants to Be ACTIVE vs. enhanced usual care (instead of cluster randomization) to enhance statistical power and because contamination of the intervention was not an issue.
Population
Pragmatic eligibility criteria for participants were intentionally developed to include the bulk of the U.S. population with T2D who are presently sedentary and could safely participate. These criteria included: diagnosis of T2D, age 50–85 years, performing < 3 days/week of moderate intensity exercise for at least 20 minutes (28), and the absence of conditions that would greatly limit the safety or effectiveness of the intervention (e.g., < 6-month life expectancy, moderate-to-high risk of fall (29), severely uncontrolled hypertension/blood glucose levels, or a clinical diagnosis of dementia). Potential patients were identified from the electronic health record by the presence of a diagnosis code for type 2 diabetes mellitus, and the absence of other exclusion criteria. Research assistants contacted primary care providers to confirm eligibility, and subsequently contacted patients by mail and phone to screen for interested participants who met our sedentary behavior criterion. Participants were consented between December 2015 - January 2019. All research processes were reviewed and approved by the Colorado Multiple Institutional Review Board.
As this was a pilot trial that required objective tests of function and fitness, we recruited from clinics in close proximity to the facility where these tests could be conducted — these included two moderately large academic primary care clinics (n = 15,000 patients total; ~1500 patients with T2D). These clinics serve a racially and ethnically diverse population of patients, and had no pre-existing programs of PA counseling. Participating clinics were National Committee for Quality Assurance-certified patient-centered medical homes (30) and had a pool of existing staff members to serve as coaches who had no direct patient care responsibilities, but instead conducted outreach calls to patients with unmet clinical goals.
Intervention
Enhanced usual care
The enhanced usual care intervention consisted of a series of printed educational materials, including the current U.S. physical activity guidelines at the time of study enrollment (31), and 3 monthly mailings on diet and other diabetes self-education topics.
Development of Be ACTIVE Intervention content in a Pre-Implementation phase
The pre-implementation phase required key patient, clinician, clinic leader, and coach stakeholder input. This process was informed by our recent systematic review (7) that identified four pragmatic EBPs that significantly increased PA (32), and were also rated as highly pragmatic by the Pragmatic Explanatory Continuum Summary (PRECIS-2) (33). The recent Patient Centered Outcomes Research Center (PCORI) guidance for complex behavioral interventions highlights the need for core components of a program to outline their key functions(34, 35). The key functions of the pragmatic EBPs we identified in our review were: 1) a checklist of PA behavioral counseling techniques to simplify the counseling approach for staff; 2) training in the use of brief motivational interviewing counseling approaches (training function); 3) repeated counseling encounters with patients to set and monitor PA goals that are enjoyable (patient-centered counseling function: use of motivational interviewing and accountable tracking of patient progress towards appropriate PA goals); and 4) initial and subsequent monitoring for safety in this diseased population at-risk for injury with exercise (safe counseling function).
In a pre-implementation phase, we then engaged primary care clinicians, staff, and patients to iteratively guide the selection of implementation strategies that would deliver our key functions in a way that would meet clinic workflows and patient needs (36, 37). Initial stakeholder input identified key factors to address to make implementation strategies feasible (38) — selecting a clinic staff coach who has dedicated time to do patient outreach; provision of formal training for staff coaches in motivational interviewing for PA counseling; supporting coaches by embedding the counseling scripts in the electronic health record for ease of reference and documentation; and a resource binder for patients to track their PA. To maximize sustainability and dissemination, we also sought to identify approaches that would be sustainable and low-cost for delivery in primary care, as our review had identified this as a key gap. Clinic stakeholders identified a potential reimbursement approach of the Medicare Chronic Care Management billing code for care between clinic visits (39, 40). At the time of this intervention, this required charging a specific billing code (CPT 99490) for delivering ≥ 20 minutes per month of counseling/care management — this code was reimbursed by Medicare as 0.61 work relative value units, or $32.66/month in 2017 dollars for our region (39, 40). Of note, this billing code requires patients to have at least one other chronic disease other than diabetes, but > 95% of patients with T2D meet that criterion.
After beta-testing this Be ACTIVE delivery strategy with 5 patients with T2D, and debriefing with clinic staff, we learned we needed to share objective PA data with coaches to simplify their data review, so we added a PA tracker (FitBit©) and modeled the integration of FitBit© data with the electronic health record. Also, patients and coaches requested more specific information about motivations and methods to increase PA, so we added a “theme” to each coaching call (Fig. 1) from an evidence-based lifestyle PA program (41, 42). To pragmatically address participants’ functional limitations related to activity, we added simple multi-muscle resistance exercises recommended by the National Institute of Aging (14, 43). To capture key program costs of time, we added a field to the counseling script to capture the time spent with the patient by each coach and clinician. Ultimately, the stakeholder input in the pre-implementation phase guided the selection of several implementation strategies (38) and influenced the “form” of the Be ACTIVE intervention content and implementation strategies studied in this pilot trial to be more pragmatic, while preserving the “function” of the pragmatic EBPs that are its foundation (Fig. 1).
Outcome measures
Implementation outcomes and mixed methods analytic plan
We evaluated implementation outcomes with a convergent mixed methods approach (44) for the primary implementation outcomes of acceptability and feasibility (45) and secondary outcomes from the RE-AIM model that are related to acceptability and feasibility, including program costs. We sought to assess these outcomes among key stakeholders who completed the intervention: patients, Be ACTIVE coaches, and the clinicians who conducted the safety-focused in-person visits. In terms of qualitative data, we used semi-structured interviews to inform the acceptability of Be ACTIVE, and the areas where the program’s feasibility and acceptability were sub-optimal and could be further improved. We also qualitatively assessed key RE-AIM factors related to feasibility and acceptability: Reach (why patients chose to participate); Implementation (challenges with fidelity to program delivery and needs for adaptation related to factors such as delivery time and costs); and Maintenance (reasons why staff and patients recommend continuing/not continuing Be ACTIVE). To minimize response bias, patient interviews were conducted by a research staff member (KC) who did not deliver the intervention, and staff interviews were conducted by an independent qualitatively trained analyst (SL). Interview audio files were transcribed verbatim.
Our qualitative data analysis approach for these implementation outcomes used an iterative and team-based process guided by qualitative content analysis (46). A qualitatively trained analyst (SL) and the principal investigator (AGH) both inductively and deductively developed the codebook. Initial codes were based on themes related to Feasibility, Acceptability, interview guide domains, and the codebook was expanded based on codes emerging from the data. We used ATLAS.ti version 8 for data management. Transcripts were jointly reviewed and coded until reaching thematic saturation (i.e., no new codes identified) and strong code assignment agreement. Twenty percent of transcripts were independently read, double coded, and then merged prior to analysis. Overall, there was strong code assignment agreement among coders, and the few discrepancies in coding that emerged were resolved through discussion to consensus. Throughout the process, the analyst, principal investigator, and qualitative methodologist met regularly to check new findings, discuss emergent codes and themes, and assess the preliminary and final results.
For quantitative implementation outcome assessments, Feasibility included time spent on each counseling call, and the implementation costs per patient counseling call. We considered costs from the perspective of the clinic leader who would decide to adopt Be ACTIVE for a clinic (i.e., intra-organizational health system perspective)(47), and used a time-based activity micro-costing approach (48). Specifically, we aggregated the counseling time required across all participant coaching visits, and translated the time required to deliver a single coaching session into costs as a pro-rated portion of each counselor’s salary and benefits. Feasibility measures also included the percentage of counseling visits with fidelity to the counseling protocol based on chart review by an unblinded research assistant. Blinding to intervention group was impossible because no control participants had counseling notes — to minimize bias, fidelity was assessed by a research staff member who did not deliver the intervention. We measured the time for clinician safety visits, in order to assess whether visits exceeded their allotted time, as per the clinic leader adopter perspective. In keeping with the level of cost analysis recommended for pilot implementation studies(47), we did not estimate this clinician time in dollars, as these visits were fully reimbursed by insurance. Acceptability was assessed quantitatively as the percentage of patients who would recommend this study to a friend or family member, and the percent of staff who recommended the clinic continue to offer Be ACTIVE (i.e., Maintenance). Finally, we calculated Reach as another quantitative measure of acceptability (percent of individuals enrolled divided by the number of eligible participants).
Mixed Methods Analysis
To draw inferences from the qualitative and quantitative data on implementation outcomes, we integrated the quantitative data for feasibility and acceptability (including the RE-AIM constructs of Reach, Implementation, and Maintenance described above) with the qualitative themes, codes, and representative quotations from interviews with patients and clinic staff.
Effectiveness outcomes
All primary and secondary effectiveness outcomes were measured at baseline and immediately after the 12-week intervention. The primary outcome of objective PA was measured by the Actigraph GT3X + accelerometer (Actigraph, LLC, Pensacola, FL); analysis included ≥ 3 valid days of wear time ≥ 10 hours/day (as per standard methods); the use of linear mixed effects models allowed measurement of PA across all valid days to provide a weekly estimate of PA levels (49, 50). We considered steps/day and minutes of combined moderate-vigorous intensity exercise as the co-primary measures of PA, based on the clinical relevance of each of these PA domains to cardiovascular risk, insulin resistance/hyperglycemia, and physical function (51–54).
The secondary outcomes of physical function were selected to be clinically relevant to risk of mortality and disability/falls, and sensitive to change (23, 24, 55, 56). We included the Short Physical Performance Battery (SPPB, range of 0–12 where 12 is better function) that is sensitive to change for individuals with at least mild-to-severe baseline functional disability/frailty and the timed 400-meter walk assessment that is sensitive to change for individuals with either normal baseline physical function or mild-to-moderate impairment (23, 24, 55, 56). We also included leg extension power as a measure sensitive to change across patients with or without functional impairment (14, 57, 58). We assessed SPPB according to standard procedures, including the timed 4-meter walk, timed repetitive chair rise, and ability to stand for > 10 seconds with a tandem and semi-tandem stance (23). The standard timed walk procedure was conducted in a long corridor with a 20-meter path for patients to walk back and forth for 400 meters at maximal tolerated speed (59). The leg extension was conducted with a total of 10 trials for each leg on the Power Rig (57) and the best trial with either leg was reported. Post-intervention function assessments were not masked to intervention group for patients (who knew their intervention group assignment), or for research staff due to limited pilot funds — this lack of masking was partly mitigated by the use of objective data obtained directly from accelerometers (primary outcome), or from a stopwatch or Power Rig device (secondary outcomes).
Additional outcome measures included: glycemic control measured as Hemoglobin A1c (60), grip strength by hand-held dynamometer, cardiorespiratory fitness measured by modified Balke protocol (61) on a treadmill and MGC Diagnostics© metabolic cart (peak oxygen capacity, VO2peak) – and Anaerobic Threshold (AT) as a measure of whether the intervention affected the threshold of sustainable exercise intensity, as measured by V-slope technique during VO2peak testing (61). VO2peak tests were terminated for participant exhaustion or joint pain that limited endurance.
To better understand the potential mechanisms of Be ACTIVE, we also assessed behavioral predictors of improvements in PA and physical function outcomes. These included validated self-report measures of perceived social-environmental support for PA (62), self-efficacy for walking for incrementally longer durations without stopping (e.g., 5 minutes, 10 minutes, etc.) (63), self-efficacy to regularly make time for PA (64), and self-efficacy in persisting to be active despite chronic disease (65). We also measured depressive symptoms (Center for Epidemiologic Studies Depression scale, CESD (66)), cognitive function (behavioral dyscontrol scale (67)), and arthritis pain symptoms (68) according to the Western Ontario and McMaster Universities Arthritis Index (WOMAC), including separate subscales for pain (0–20, where higher numbers indicate greater pain), stiffness subscale (0–8), and physical function subscale (0–64).
There were no changes in the outcome assessment techniques or timing after the initiation of the research study testing procedures. In discussion with our exercise physiology study team member (JSL), after publishing the protocol on clinicaltrials.gov but before initiating testing procedures, we added leg extension power as a secondary functional outcome and eliminated grip strength as a secondary functional outcome, as leg power is typically more sensitive to change than grip strength (14, 57, 58).
Based on the effect size of prior PA counseling interventions (12), a fully-powered trial would need to complete n = 300 participants to demonstrate statistical significance for our primary outcome. Given the pilot feasibility nature of this trial, we aimed to enroll 50 participants to ensure stable effect size and an estimate of the clinical importance of observed changes in outcomes, and a relatively diverse population of participants. In lieu of interim safety analyses, we conducted annual evaluations of adverse events across the intervention/control group with a safety officer. No major adverse events occurred in either intervention group.
Parallel randomization of individuals was accomplished through a block-stratified randomization sequence to support balanced composition of age and sex. To maintain blinding to intervention allocation for all staff performing baseline assessments, the randomization key was kept in a secure drive by a staff member who did not participate in study visits, and who e-mailed the randomization code to the research staff member just prior to the randomization visit.
We used an intention to treat approach to analyze all baseline and post-intervention data from consented participants. Statistical analysis of all primary and secondary outcomes was carried out using linear mixed effects modeling to account for correlation within patients (69), allowing us to use all available data from each participant. Models for daily PA data only used data from “valid” days with ≥ 10 hours of accelerometer wear — these models contained a random intercept for subject and a random slope for visit, where possible; in the event of nonconvergence, model complexity was reduced by excluding the random slope term. Models for outcomes with only one measurement pre and post included only a random intercept for subject. Models contained fixed effect terms for visit (pre or post), treatment group (control or Be ACTIVE), and their interaction. The coefficient estimated for the interaction term represents the difference between treatment arms in mean change in outcome from pre to post, so hypothesis testing for the existence of a treatment effect is based on this coefficient. R statistical software was used to conduct all analyses (70, 71).