Participants and Setting. This study included 602 employees of Massachusetts General Hospital (MGH) in Boston, Massachusetts who completed a baseline visit as part of a randomized controlled trial testing a worksite dietary intervention (ChooseWell 365) between September 2016 and February 2018 (23). The study tracked participants’ workplace cafeteria purchases before, during, and after the intervention. To be included in the study, employees had to be 20–75 years old and make cafeteria purchases at least 4 times per week for at least 6 weeks during a 12-week period prior to study recruitment. Employees were ineligible if they were pregnant, wanted to gain weight, were currently participating in a weight loss study, had weight loss surgery in the past 6 months, had an eating disorder history, were employed in the cafeteria, or had plans to terminate their employment at the hospital in the next year. Employees may have worked on either weekdays or weekends, but there was no available information about which days of the week individual employees worked. The present study was a cross-sectional analysis of baseline data from the trial. Participants completed two 24-hour dietary recalls online and attended a clinical visit. All data used in this study were collected prior to randomization and any intervention procedures.
Measures.
Clinical assessment. Participants attended a clinic visit at which their height, weight, and blood pressure were measured, and blood was drawn for a fasting lipid panel, glucose, and hemoglobin A1c tests. Body mass index (BMI) was used to categorize weight status (obese = BMI ≥ 30 kg/m2, not obese = BMI < 30 kg/m2). Hypertension was defined as at least one of the following: (a) self-reported hypertension or high blood pressure diagnosis by a medical professional; (b) self-reported use of prescription antihypertensive medication; (c) study measurement of systolic blood pressure ≥ 150 mmHg; and/or (d) diastolic blood pressure ≥ 90 mmHg (24). We chose a systolic blood pressure cut-off that was higher than the guidelines (>130 mmHg) to be conservative in our definition of hypertension for participants who had not previously been diagnosed with high blood pressure.Prediabetes/diabetes was defined as at least one of the following: (a) self-reported diabetes or prediabetes diagnosis by a medical professional; (b) self-reported use of prescription medication for diabetes; and/or (c) study measurement of HbA1c ≥ 5.7 (25). Type 1 and 2 diabetes were not differentiated. Hyperlipidemia was defined as at least one of the following: (a) self-reported diagnosis of high cholesterol/hyperlipidemia; (b) self-reported use of prescription medication for high cholesterol; and/or (c) study measurement of fasting total cholesterol ≥ 220; low density lipoprotein ≥ 160; or triglycerides ≥ 180 (26).
Administrative data. Participant job type was collected from the hospital’s human resources department. Job titles were combined into four categories determined based on educational attainment needed for the type of work: (1) service workers (manual and/or unskilled laborers)/administrative assistants; (2) craft/technicians (e.g., radiology technicians); (3) management/professionals (e.g., social workers, nurses, hospital managers); and (4) MDs/PhDs (e.g., physicians, researchers).
Dietary intake. Dietary quality was assessed using Automated Self-Administered 24-hour (ASA24) dietary recall surveys. This tool was developed by the National Cancer Institute and uses multi-level probes to guide respondents through reporting their intake over the prior 24 hours (27). In most cases (93.7%), two ASA24 recalls were collected on non-consecutive days and a Healthy Eating Index (HEI) score was calculated using their average. (28). In the small number of cases where only one ASA24 was completed (6.3%), the HEI was based on that recall. The HEI score measures an individual’s compliance with dietary recommendations from the United States Department of Agriculture (USDA) Guidelines for America. The most recent HEI (HEI–2015) was used in this study. Scores range from 0—100, with higher scores signifying better compliance with dietary guidelines. Americans had an average HEI of 59 out of 100 in 2013–2014, based on data from the National Health and Nutrition Examination Study (29).
Worksite cafeteria purchases. All MGH hospital cafeterias use traffic light food labeling to provide information about the healthfulness of food and drink items (green = healthy, yellow = less healthy, red = unhealthy). The labeling algorithm was developed by hospital nutrition staff and was based on USDA dietary guidelines (30), using positive and negative nutritional criteria to rate all food/beverage items. Positive criteria included having the main ingredient be: (1) a fruit or vegetable, (2) a whole grain, and/or (3) a lean protein, plant-based meat substitute, or low-fat dairy. Negative criteria included: (1) saturated fat content of ≥ 5 grams per entrée or ≥ 2 grams per non-entrée item, condiment, or beverage, and/or (2) caloric content ≥ 500 kilocalories per entrée, ≥ 200 kilocalories per non-entrée food item, or ≥ 100 kilocalories per condiment or beverage. Items were categorized as green if they had more positive than negative criteria; those with equal positive and negative criteria, with only one negative criterion, or with no positive or negative criteria were labeled yellow; and those with multiple negative criteria and no positive criteria were labeled red. While all items available in the salad bar were rated individually, color labels were assigned to each salad purchase based on weight for study participants (green: salad < 16 ounces; yellow: salad ≥ 16 ounces) (23). Each cafeteria has permanent, highly visible signage explaining the labeling system.
Cafeteria purchases and the associated traffic light label colors were retrospectively collected from the cafeteria cash register data system for the 3 months prior to each participant’s study enrollment date. A Healthy Purchasing Score was calculated based on the weighted proportion of items purchased that were labeled red, yellow, or green to reflect overall healthfulness of a participant’s purchases over those 3 months (31). The proportion of red items was multiplied by 0, the proportion of yellow items was multiplied by 0.5, and the proportion of green items was multiplied by 1. The sum of these values was the Healthy Purchasing Score, ranging from 0 (100% red items, least healthy) to 1 (100% green items, healthiest). For example, if an employee’s 3-month baseline purchases were 20% red, 50% yellow, and 30% green-labeled items, the Healthy Purchasing Score would be: (0.2 red x 0) + (0.5 yellow x 0.5) + (0.3 green x 1) = 0.55.
International Physical Activity Questionnaire (IPAQ) Long Version. This validated and commonly used measure asks about one’s last 7 days of physical activity, split by domain in which the activity occurred: job-related, transportation, housework/house maintenance/caring for family, and recreation/sport/leisure (32). Sedentary time is assessed as time spent sitting across all domains. The present study included job-related PA (“work-related PA”), recreation/sport/leisure PA (“leisure-time PA”), sedentary time, and total PA (summed across all domains). We followed the scoring guidelines that recommend truncating time spent on each level of physical activity (walking, moderate, and vigorous) at 180 minutes per day to avoid reporting errors (33). Totals were calculated in terms of metabolic equivalents of hours per week (MET-hours), by assigning a multiplier to activity based on its intensity (walking = 3.3, moderate activity = 4, vigorous activity = 8). The IPAQ has acceptable measurement properties and high reliability (α = 0.80) (32).
Statistical Analyses. Analyses were conducted with Stata version 15.1 (Stata Corporation, College Station, TX). Medians were calculated to summarize work-related and leisure-time PA overall and by quartile. Multivariate regression models tested whether leisure-time PA, work-related PA, and sedentary time were associated with each other, and whether they each were associated with Healthy Purchasing Score and HEI, adjusting for age, sex, race, ethnicity, education, season at time of assessment, and PA (leisure-time analyses adjusted for work-related PA, work-related analyses adjusted for leisure-time PA, sedentary time analyses adjusted for total PA). Leisure-time PA, work-related PA, and sedentary time, when used as covariates, were divided into quartiles and entered into the models using the medians of each quartile to flexibly and efficiently model their non-normal distributions. Logistic regression analyses tested prevalence of obesity, prediabetes/diabetes, hypertension, and hyperlipidemia, adjusting for the same variables as above in addition to the HEI, by quartile of each type of PA. Regression-adjusted mean outcomes were calculated using Stata’s “predict” command as the mean of predicted probabilities evaluated assuming all subjects had the median value of a particular quartile while retaining their observed characteristics on other covariates. Participants with missing data were excluded from analyses that included the missing variable.