Subjects and study design
Data for this study were collected as part of the Food ‘N Mood Study, a pilot study carried out in Central Pennsylvania from September 2019 to March 2021 that aimed to examine the impact of daily FI on diet, mood and heart rate variability. Participants were recruited from selected locations serving low-income populations, such as the office for the Special Supplemental Program for Women, Infants and Children (WIC), the county assistance office, food banks, and Head-Start childcare services, in rural areas within Centre, Clearfield, Clinton, Blair, Elk, Bradford-Tioga, Huntingdon, and Snyder-Union-Mifflin counties. Healthy adults aged 20 to 50 years with a household income that fell below 185% of the Federal Poverty Line (FPL) were eligible to be enrolled. Exclusion criteria included adults who were non-English speakers, who had physical, mental or emotional disabilities, who used medications or experienced medical conditions known to affect heart rate variability, or who had disabled people as part of their household. Females who were pregnant or who had reached menopause were also excluded from the study.
An ecological momentary assessment (EMA) model on smartphones was used to collect daily FI and dietary data. Participants completed their data collection over two 3-week-long waves (from the 2nd to the 4th week of the month), covering one month in the fall season (September, October, or November) in 2019 or 2020, and another month in the winter season (February or March) in 2020 or 2021. On their devices, participants filled out a daily evening survey that asked about FI experiences in the past twenty-four hours. On Sunday, Monday, and Tuesday of the study weeks, participants were asked to report dietary intake on a food record module. The study protocol was approved by The Pennsylvania State University Institutional Review Board, University Park, Pennsylvania and the National Center for Advancing Translational Sciences (NCATS).
Assessment of daily FI
Daily FI status was measured using an adapted 6-item U.S. Adult Food Security Survey Module . Each day, participants were asked about the food situations that they encountered in the past twenty-four hours. The food situations included: due to a lack of money, the participant ‘was worried about food running out,’ ‘did not eat a balanced meal,’ ‘cut meal size or skipped a meal,’ ‘ate less,’ ‘was hungry but did not eat,’ and ‘did not eat for a whole day.’ A daily FI score was calculated as the sum of the six survey answers, yielding a total score ranging from 0 (experiencing no food insecure situations) to 6 (experiencing all food insecure situations).
Assessment of dietary intake
Food records were collected on three days (Sunday, Monday, and Tuesday) per week for three weeks in both fall and winter seasons using the provided smartphones. Built-in notifications at the end of each survey module were used to remind participants to complete their food diary. Participants were encouraged to record their food intake at the moment when they were eating or right after they ate. The application queried information on food items, food amount, food preparation, and timing and location of the food intake. Participants recorded dietary intake using free-text entries, and a trained research dietitian from the Diet Assessment Center at Penn State downloaded the dietary intake data from the research portal and re-entered into Nutrition Data System for Research software (NDSR, versions 2019 and 2020, the Nutrition Coordinating Center (NCC), University of Minnesota, Minneapolis, MN). To improve the accuracy of the dietary data, this trained research dietitian reviewed the NDSR-entered food record with the participant over the phone to confirm portion sizes and foods consumed on the day following the date of dietary data entry (i.e. a follow-up phone call was made on Monday to verify dietary intake information entered on Sunday). This interview verification process is known as ‘record-assisted recall’ and has been validated in a previous study . Dietary intake data were analyzed using NDSR. The NDSR time-related database updates analytic data while maintaining nutrient profiles true to the version used for data collection.
Diet quality was assessed using the Healthy Eating Index 2015 (HEI-2015) using the simple HEI scoring algorithm method [19,20,21,22]. The index contains 13 components, including total fruits, whole fruits, total vegetables, greens and beans, total protein foods, seafood and plant proteins, whole grains, dairy, fatty acids, refined grains, sodium, added sugars, and saturated fats. Scores are calculated proportionately according to the intakes between the minimum and maximum standards, aligning with the 2015–2020 Dietary Guidelines for Americans. The maximum possible HEI-2015 score is 100 [19,20,21].
Assessment of demographic and socioeconomic characteristics
At baseline, a background survey was administered to collect demographic, socioeconomic, and health characteristics, including age, gender, race/ethnicity, height, weight, education, employment status, marital status, household size, number of children under 18 years in the household, total annual household income, enrollment in food assistance programs, and chronic household FI in the past 12 months. Body mass index (BMI) was calculated based on self-reported weight and height by weight (kg) / height2 (m). Chronic FI status in the past 12 months was measured using the 10-item U.S. Adult Food Security Survey Module . According to the Department of Health and Human Services’ definition, poverty status was categorized as < 130% FPL or > = 130% and < 185% FPL considering the total household income and household size [24, 25].
All statistical analyses were performed in R (Version 4.0.5; R Foundation for Statistical Computing, Vienna, Austria). To summarize the characteristics of participants, means and standard deviations were reported for age, household size, the number of children under 18 in the household, BMI, HEI scores, and total energy. Proportions were reported for gender, race/ethnicity, education, employment, marital status, poverty status, and enrollment in food assistance programs. The differences in HEI scores and total energy intake were tested using the Wilcoxon Mann–Whitney nonparametric test because of their skewed distribution. Instead of analyzing FI status and dietary intake for each participant, we used person-days of information to examine the day-to-day associations between FI and dietary intake. Specifically, our study included completed person-days of information on daily FI and dietary intake three days per week (Sunday, Monday, and Tuesday) in the study months. The Generalized estimating equation (GEE) models were used to explore the associations between FI and total calorie intake, diet quality, and food groups intake in 1) fall and winter seasons, and 2) in pre-COVID-19 (September, October, and November in 2019, and February in 2020) and during-COVID-19 months (March in 2020, October and November in 2020, and February in 2021). Covariates, including gender [26, 27], race/ethnicity [28, 29], employment [30, 31], poverty status [32, 33], weekdays [34, 35], study weeks [10, 36], seasons [11, 37], and COVID-19 months [4, 38] were adjusted in the analysis, given their known associations with both FI and dietary intake. Seasons were adjusted in models examining the FI-diet associations by COVID-19 months, and COVID-19 months were adjusted in models by seasons. Total calorie intake was further adjusted in these models when estimating the associations between FI and the intake of individual food groups. The autoregressive correlation structure “AR1” was applied with GEE models accounting for within-person correlation in dietary intake . Because our study included 86.2% female participants, we also conducted sensitivity analysis to compare the FI-diet associations between all participants and female participants.