The CARE study used a two-arm, cluster-randomized trial to test the effectiveness of a “Healthy Lifestyles” intervention compared to a “Healthy Finances” attention control program. The study was conducted between 2015 and 2018. The study design and protocols for CARE [19] as well as a complete description of the Healthy Lifestyles intervention [20] have been described in accordance with SPIRIT and TIDieR guidelines and published elsewhere. All protocols were approved by the Institutional Review Board at the University of North Carolina at Chapel Hill and registered at www.ClinicalTrials.gov (NCT02381938).
Participants
A total of 553 child care workers from 56 child care centers located in central North Carolina participated in the study[19]. A multi-phase recruitment strategy was employed to recruit participants in four waves. Community partners helped introduce child care centers to the study by distributing announcements through existing communication channels and offering group informational sessions. The research team followed-up by phone with centers that expressed interest to review study details and confirm eligibility. In addition, the research team identified child care centers through the North Carolina Division of Child Development and Early Education’s public database of licensed providers [21] and sent announcements about the study directly (by mail and email). Announcements were followed-up with phone calls by the research team to review study details, assess interest, and confirm eligibility. Initial eligibility criteria for centers required they employ at least four staff, have been in operation for at least two years, and have no plans to close within the next 18 months. Once initial eligibility was confirmed, research team members conducted onsite center visits to recruit child care workers. To be eligible to participate, workers had to be at least 18 years of age, able to speak and read English, and either pass the Physical Activity Readiness Questionnaire (PAR-Q) screening or obtain medical permission to participate [22]. At least four workers (one administrator and three staff) had to agree to participate and sign consent for the center to be remain eligible. In addition, at least three workers (one administration and two staff) had to attend the kick-off event for the center to remain in the study and be randomized.
Randomization
Workers were randomized in clusters, based on the center where they were employed. Randomization occurred at a kick-off event, which followed baseline data collection. During these events, each center representative selected an envelope from a bowl, and the card within revealed their assignment – Healthy Lifestyles or Healthy Finances. Cards were produced based on randomization tables (generated by the study statistician) created using a block randomization approach employing a block size of two to ensure balance in the number of centers in each study arm in each wave. Results of randomization were immediately announced. Participating centers (and workers who attended the kickoff event) then adjourned to separate locations to participate in a workshop based on their assignment to Healthy Lifestyles or Health Finances.
Intervention: Healthy Lifestyles
The Healthy Lifestyles intervention has been described in detail elsewhere using TIDieR guidelines [19, 20]. Healthy Lifestyles was a six-month, multi-level, theory-guided intervention designed to increase physical activity and improve other health behaviors among child care workers. The intervention focused on increasing support for health promotion at three levels: intrapersonal (individual workers), interpersonal (interactions between co-workers), and organizational (the child care center). Strategies employed to target each level were informed by: Perceptual Control Theory (intrapersonal) [23, 24], Social Support Theory (interpersonal) [25], and Diffusion of Innovation (organizational) [26].
The intervention began with a kick-off event, which consisted of a 2-hour morning health and wellness fair integrating a variety of local community organizations and resources along with baseline assessments and a 1.5-hour afternoon educational workshop led by the study interventionist. The morning portion was attended by all participants, and randomization occurred at lunchtime. The afternoon arm-specific educational workshop introduced workers to the upcoming campaigns (Healthy Lifestyle or Healthy Finance) and the related intervention components. In the months that followed, staff participated in three successive 8-week campaigns. At the start of each campaign, centers received informational magazines to distribute to workers and materials for center displays including team activity planners.
For Healthy Lifestyles, workers were asked to set two health behavior goals – one on physical activity (e.g., getting 10,000 steps per day) and one other health behavior (i.e., increasing fruit and vegetable intake, limiting snack foods, eliminating sugar-sweetened beverages, limiting fast food, eliminating cigarette use, improving sleep, strength training or daily self-weighing). To facilitate monitoring of physical activity (i.e., steps), each worker received a pedometer. Workers were encouraged to self-monitor these behaviors every day and submit information weekly using the CARE website. Workers received tailored feedback based on this self-monitoring information and goals were adjusted accordingly (e.g., if steps goal for previous week was met, new goal would increase the steps target). Workers also received weekly email/text prompts to remind them to submit their self-monitoring information and biweekly (every other week) prompts to be active. By the end of wave 1’s first campaign, website data showed that many workers were not monitoring their behaviors. So, a center visit was added during the first week of each campaign to encourage workers to use their pedometer and log their activity, giving special attention to workers that had not been monitoring. This modification to our intervention protocol was then implemented in all 4 waves. Prize-raffles were offered as incentives for self-monitoring and goal attainment.
During each campaign, the center director received technical assistance and coaching from the study interventionist focused on critical elements of workplace health promotion. Originally, this component was designed to be delivered as a group webinar. After scheduling difficulties and lack of attendance in wave 1, delivery was modified to offer directors individual coaching calls.
Attention Control: Healthy Finances
The Healthy Finances program was designed to provide a similar level of attention as the Healthy Lifestyles arm, including three, 8-week campaigns with similar components (e.g. magazines, center displays, email/text prompts, prize raffles). The critical difference was that all messages focused on workers’ financial well-being and financial success of the center. Instead of goal-setting and self-monitoring health behaviors, workers were encouraged to take quizzes about the new financial management strategies they learned. Instead of live technical assistance and coaching, center directors were offered pre-recorded webinars on budgeting, marketing strategies for their child care program, and managing legal risk specifically designed for use by child care programs.
Outcome Measures
Outcome measures were collected at three timepoints—baseline, post-intervention, and maintenance (one-year post-intervention)—during onsite center visits conducted by trained data collectors blinded to center arm assignment. A full description of data collection protocols and measurement tools is described in detail elsewhere [19]. This paper reports on baseline to immediate post-intervention results only.
Primary Outcome. Physical activity was assessed using ActiGraph GT3X (ActiGraph, LLC, Pensacola, FL) accelerometers, which workers wore for seven consecutive days. Workers received monitors during center visits along with a postage-paid envelope for their return. Accelerometer data were downloaded using ActiLife software then processed to assess wear and non-wear time. Only participants with valid wear time (i.e., ≥7 hours of wear time on ≥4 days) were included in the primary analysis. Adult-specific cut points were then applied to compute minutes of moderate-to-vigorous physical activity (MVPA, ≥2020 cmp, primary outcome), lifestyle physical activity (≥760 cpm), and sedentary time (≤100 cpm) each day [27, 28]. Daily estimates from all valid days of wear were used to calculate average minutes per day for each level of physical activity. To account for variations in wear time, estimates were standardized to a 14-hour day. Weekday and weekend day data were also identified and used to calculate average weekday and weekend minutes per day of MVPA.
Secondary Outcomes. Workers’ health behaviors were self-reported using the Carolina Health Assessment and Research Tool (CHART) [29]. This web-based survey is divided into modules, each of which captures a specific health behavior. CHART was modified for this project to include modules on physical activity, diet, tobacco and e-cigarette use, sleep, and emotional health. CHART also included a demographics module that captured participant demographics and center characteristics. Original CHART items and all modifications drew from existing measures [30-35], as described in detail elsewhere [19]. Drawing on procedures used in source measures, CHART data were summarized to describe health behaviors, specifically times per week of muscle strengthening activities; servings per day of fruit (excluding juice), vegetables, (excluding potatoes), fruits and vegetables (excluding fruit juice and potatoes), sugar sweetened beverages, salty snacks, and fast food; eating habits score (scores range from 0 to 20, higher scores indicates healthier eating habits); current smoking status (smoker or non-smoker) and e-cigarette use (ever used or never used); hours per night of sleep and sleep quality (bad or good); and level of distress (ratings range from 0 to 10, higher scores indicate higher distress).
Biometric assessments of health and fitness indicators were taken by trained data collectors using established protocols. These measures included height, weight, and waist circumference[36]; blood pressure [37]; the six-minute walk test [38]; hand grip [39]; the 30-second chair sit and stand test [40, 41]; and the four-stage balance test [42, 43]. Height and weight measurements were used to calculate BMI. Blood pressure readings were used to calculate mean arterial pressure.
The workplace health and safety environment was assessed using a tool developed specifically for this study [19], but drawing from existing workplace environmental assessments [44-46]. Information was collected primarily through a structured interview with the director and an environmental observation conducted by data collectors. A scoring rubric was guided by a recent review existing measures of workplace environmental and policy supports for physical activity and healthy eating [47]. Data were then used to calculate scores for four domains: general infrastructure (possible range 0-27), organization policies and procedures (possible range 0-35), programs and promotions (possible range 0-65), and internal physical environment (possible range 0-27). Higher scores always indicated greater support for staff health and safety. For the latter three domains—organization policies and procedures, programs and promotions, and internal physical environment—component scores were also calculated to look at supports available for physical activity (possible range of scores being 0-7, 0-7, and 0-5, respectively) and nutrition (possible range of scores being 0-5, 0-9, and 0-11, respectively).
To document delivery and participation in the intervention, process evaluation measures (dose delivered and received) were collected throughout the study using a combination of direct observation, surveys, and field notes. In addition, a small sample of participants (n=30) from the intervention group were interviewed at the conclusion of the study to reflect on their experience.
Statistical Analysis
Descriptive statistics were used to summarize baseline demographic data for workers and centers in the intervention and control arms. Then, intent to treat (ITT) analyses were performed using all randomized participants. Missing data were addressed using maximum likelihood estimation under the assumption of missing at random. Analyses used multi-level linear mixed models (SAS PROC MIXED) for continuous outcomes and GEE-based marginal logistic regression (SAS PROC GLIMMIX) for binary outcomes to examine group differences of primary and secondary outcomes. Models included random cluster effects to account for covariance between participants within the same center as well as fixed effects for time, trial arm, time x arm interaction, and study wave (stratification variable during randomization). An unstructured working covariance was used to account for statistical covariance among repeated measurements from the same subject. The residuals of all continuous outcome variables were checked for normality. Where evidence of departure from normality was apparent the square root of the outcomes were used for analyses in order to obtain valid p-values; however, results are presented in their original scale for ease of interpretation.
To assess the impact of missing data, sensitivity analyses were conducted using multiple imputation (SAS PROC MI). Missing data were imputed 50 times for all outcome variables using variables associated with drop-out, demographic variables to be included in later regression analyses, and an indicator for child care center to account for the possibility of clustering. Analyses used linear mixed ANCOVA models to examine change in continuous outcomes and GEE-based marginal logistic regression for binary outcomes. Similar to the ITT analyses, models included random cluster effects to account for covariance between participants within the same center as well as study wave. Models were additionally adjusted for the baseline value of the given outcome variable as well as demographic characteristics identified a priori based on evidence of their predictive value for physical activity and other health behaviors (i.e., age, race, income, baseline BMI). Inferential results based on the imputed data were obtained via SAS PROC MIANALYZE. All tests were two-sided at the 0.05 level. Multiple comparisons for all secondary intervention effects were accounted for by using the false discovery rate method (i.e., the expected proportion of Type I errors among significant findings) to obtain adjusted p-values. All analyses were performed using SAS Software, version 9.4 (SAS Institute Inc., Cary, NC).