Medically Tailored Grocery Delivery for Food Pantry Clients with Diabetes.

INTRODUCTION

Food insecurity occurs when lack of money or other resources limits people’s ability to acquire adequate food for healthy living, affecting 13.5% of US households.1 Food insecurity is associated with lower intake of fruits and vegetables, and higher intake of red and processed meats and sugar sweetened beverages, contributing to an overall lower diet quality.2,3 Food insecurity and low diet quality are associated with poor management of chronic diseases such as type 2 diabetes (T2D).4–6 With scarce financial resources, people experiencing food insecurity have added challenges in following recommended T2D nutrition therapy guidelines (e.g., consuming nutritious yet high cost fruits, vegetables, legumes, and whole grains).7 Low-income households and racial and ethnic minorities are disproportionately burdened by diet-related health conditions in the US.8 Notably, among US residents, Hispanic and Pacific Islander adults experience higher rates of both food insecurity and T2D compared to non-Hispanic White adults.8–10 An evident way to address food insecurity among people experiencing high rates of T2D is to improve their access to nutritious food supportive of diabetes management.

Food Is Medicine (FIM) interventions aim to provide nutritious food to people experiencing food insecurity, and ultimately prevent or treat diet-related disease.11 Food Is Medicine intervention models may include medically tailored meals or groceries, produce prescription programs, and federal nutrition assistance programs (e.g., the Special Supplemental Program for Women, Infants, and Children) and are frequently paired with nutrition education.11 Medically tailored groceries (MTGs) include healthy fresh, frozen, or canned foods that are provided in food packages to individuals who have diet-related conditions and can prepare their own meals.11 Food banks and food pantries are often key leaders in the provision of MTGs, serving those experiencing both high rates of food insecurity and diet-related disease.12–14

Food bank-led FIM interventions that provide free diabetes-friendly groceries and/or produce have shown promise in improving participants’ dietary intake and lowering hemoglobin A1c (HbA1c).15,16 However, substantial heterogeneity across program models limits the understanding of which intervention components are most effective (e.g., type and amount of food, education components, home delivery vs on-site pick-up).15 A study of one diabetes-focused MTG intervention found that individual participants’ outcomes varied according to the extent to which they receive the intended dose of the intervention.17 The 6-month intervention offered twice-monthly MTGs from a food bank to participants experiencing food insecurity and T2D.17 Participants who picked up at least 9 of 11 food packages and attended 2 in-person nutrition education sessions showed greater reductions in HbA1c than those who did not.17 Other studies have demonstrated greater use of FIM interventions with home-delivery of meals or groceries.18,19 Home delivery addresses barriers to transportation that are associated with food insecurity and health and economic difficulties. These barriers can limit people’s ability to visit a clinic or other site for recurring food distributions and nutrition education sessions.20,21

Food Is Medicine interventions among people with T2D and food insecurity have paid limited attention to transportation barriers that may limit the dosage received by participants. The purpose of this initial investigation was to evaluate a FIM intervention intended to reduce HbA1c and improve diet quality by addressing challenges for participants to receive its full dose of food distributions and nutrition education. The intervention included home-delivered T2D-appropriate food boxes along with home-delivered and online plain language adapted diabetes self-management education and support (DSMES) materials. We hypothesized that participation in the intervention would improve two primary outcomes: glycemic control and diet quality.

METHODS

This initial investigation used a single arm pre-post design in which participants received a 3-month food delivery and DSMES intervention (Figure). The study was approved by the University of Arkansas for Medical Sciences Institutional Review Board (IRB #260304) and is registered at ClinicalTrials.gov (ID: NCT04831216). The study used a community-based participatory research approach to engage community members throughout the design of the intervention, recruitment and data collection, and interpretation and dissemination of findings.22 The study protocol includes additional details about the community-based approach, study design, intervention components, and data collection measures, briefly described herein.23

Participants were enrolled from 5 food pantries located in Benton and Washington counties in northwest Arkansas from August 2021 to February 2023. The food pantries serve a large community, including Hispanic/Latino and Marshallese Pacific Islander (i.e., Marshallese) residents that face disproportionately high rates of food insecurity and chronic diet-related disease.8–10,24,25 Food pantry clients were given a flyer describing the study and invited to participate in a free health screening collecting HbA1c, height, weight, and blood pressure. Interested clients with screening results of HbA1c ≥7.0% were invited to complete a questionnaire evaluating the other inclusion criteria, which included food insecurity (confirmed by a response of “sometimes true” or “often true” to at least 1 item from the 2-item Hunger Vital Sign screener26) and age ≥18 years. Exclusion criteria are detailed in the study protocol.23 Recruitment and consent were conducted by bilingual study personnel in English, Spanish, or Marshallese based on the participants’ preferred language. Three hundred twenty people were screened for eligibility, and 105 participants provided written informed consent and enrolled in the study. Three participants were determined to be ineligible after enrollment; 2 did not meet HbA1c (n=1) or food insecurity (n=1) criteria, and 1 was found to be medically ineligible by the study physician. One additional participant was administratively removed prior to the start of the intervention, leaving 101 participants allocated to the intervention (Supplemental Figure).

A food box was delivered each week for 12 weeks to the participant’s home address. The 12 food boxes followed the American Diabetes Association (ADA) Create Your Plate method, which emphasizes non-starchy vegetables, proteins, and grains, consistent with the meal pattern promoted in the ADA T2D 2019 nutrition therapy guidelines, which were the most recent guidelines at the time of intervention development.7 All boxes included fresh fruits and vegetables along with other ingredients needed to prepare the included recipes. The DSMES curriculum consisted of 12 modules adapted from the 8 core content areas identified within the 2017 National Standards for Diabetes Self-Management Education and Support,27 which included the American Diabetes Care and Education Specialist (ADCES) 7 Self-Care Behaviors™.28 Each module was made available in 2 formats: a 2-page plain language handout and a brief video summary in the participants’ preferred language. Each week’s food box included a handout and T2D-appropriate recipes using ingredients from the week’s food box. The handouts and videos were all posted on a publicly accessible website and were all available in English, Spanish, and Marshallese.23 Participants also received weekly text messages with links to the handouts and videos.

The Figure illustrates the data collection and study intervention timeline. For each participant, pre-intervention data collection was conducted up to 10 weeks prior to their first food box delivery. Post-intervention data collection was conducted up to 8 weeks following each participant’s 12th food box delivery. Surveys were used to collect self-reported demographic and socioeconomic information from participants, including age, household size (number of adults, number of children), sex, race/ethnicity, education, employment, marital status, annual household income, health insurance coverage, and participation in federal nutrition assistance programs. Survey questions about satisfaction and use of intervention components (e.g., food, recipes, handouts) were asked at post-intervention data collection only.

During each data collection window, HbA1c was measured by fingerstick using a Siemens DCA Vantage analyzer (Siemens Medical Solutions USA, Inc)29 in a private space at the food pantry from which the participants were recruited. Diet quality was measured by the Healthy Eating Index (HEI)-2015.30 Participants were asked to complete 3 dietary recalls via telephone (i.e., 2 weekdays, 1 weekend day), for a total of 6 dietary recalls per participant across the study. Dietary data were collected using the Nutrition Data System for Research (NDSR) software versions 2021 and 2022, to reflect the nutrient context of food available during the study period (NDSR, versions 2021 and 2022, University of Minnesota, Minneapolis, MN, 2021 and 2022). Trained interviewers used a multiple-pass approach to collect all foods and beverages consumed during the previous day.31 Healthy Eating Index scores can range from 0 to 100, with higher scores representing better alignment with the Dietary Guidelines for Americans (DGA).30 Healthy Eating Index assesses the alignment of a set of foods with the DGA by measuring 13 separate nutrient components. Separately for pre- and post- intervention data, a HEI score was calculated for each participant who completed at least 2 dietary recalls during each data collection window. Participants who completed 0 or 1 dietary recall were not included in the HEI analysis, since 2 dietary recalls at each timepoint are the minimum needed to estimate usual dietary intake.32 The National Cancer Institute’s Simple HEI scoring algorithm was used to derive and average ratios for the 13 nutrient components across each participants’ dietary recalls.30 Next, the 13 component scores were summed to create an overall HEI score for each participant for each window. Scoring of dietary data was completed in STATA 16.0 (StataCorp LLC, College Station, TX, 2019).

The intended sample size of 100 participants had 80% power to detect a small-to-medium effect size of Cohen’s d = 0.28, when we assumed a repeated-measures t-test, a Pearson correlation between repeated measures of r = 0.5, and two-sided α = 0.05. Mixed effects regression models for repeated measures were used to test for a statistically significant change between pre- and post-intervention primary outcomes (i.e., HbA1c and HEI), controlling for covariates chosen a priori based on use in prior literature: age, sex, race/ethnicity, household size, education, and employment.33,34 Analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC, 2013) with the Mixed Procedure and the restricted maximum likelihood (REML) estimation method. Statistical significance was set at two-sided α = 0.05.

Our overarching approach to missing values was to make use of all available data to minimize the potential for bias and maintain statistical power. The use of mixed effects regression models for repeated measures meant participants could be included in the analyses if they had non-missing covariates and at least 1 non-missing value for the outcome at either pre-intervention or post-intervention. To assess whether our approach to missing values impacted our results, we compared the results using all available data to those from an analysis of multiple imputed data sets.35 Multiple imputation was done using the MICE package in R (Version 4.4.1, R Core Team, Vienna, Austria, 2023) using the random forest method to create 20 complete data sets.36 The random forest method was chosen to account for potential nonlinear relationships and interactions among variables. Each complete data set was analyzed using the same mixed-effects model for repeated measures described above and results were combined by the MICE package following Rubin’s rules.35 Baseline characteristics between participants with complete data and participants with incomplete data were examined using t-tests or Chi-square tests.

All 101 participants had HbA1c at pre-intervention and 69 (68.3%) had HbA1c at post-intervention. Eighty-one (80.2%) participants had an HEI score at pre-intervention, 62 (61.4%) participants had an HEI score at post-intervention, and 56 (55.4%) had an HEI score at both pre- and post-intervention. Reasons for incomplete data included participants missing scheduled in-person survey/biometric data collection appointments or inability to reach participants for dietary recalls (e.g., disconnected phone numbers). Data for the analysis covariates had no missing values except for 1 (1%) missing value for education and 1 (1%) for employment status. After accounting for missing covariate values, data from 99 (98%) participants was included in the analysis of change in HbA1c and data from 85 (84.2%) was included in the analysis of change in HEI scores. Satisfaction and intervention use questions were summarized for the 69 participants who attended the post-intervention in-person data collection visit.

RESULTS

Participants’ mean age was 57.1 years with a mean household size of 4.4 members. Most participants were female (67.3%) and identified as Hispanic (34.7%) or Marshallese (32.7%). Other sociodemographic characteristics are presented in Table 1. Age was significantly higher among individuals with complete data compared to individuals with incomplete data for HbA1c or HEI. Education was significantly different, with more individuals with complete HEI data reporting below high school graduate compared to high school graduate and above among those with incomplete data (Supplementary Tables 1 and 2).

Ninety-eight of the 101 participants received all 12 deliveries of food, recipes, and DSMES materials. Of the participants who took the post survey (n=69) most either somewhat or strongly agreed that “This program helped me control my diabetes” (n=59, 85.5%). Sixty-five participants (94%) were satisfied or very satisfied with the food provided. Sixty-six participants (95.7%) reported using at least 1 recipe in the food box each week. Of 67 participants who reported on their video use, 26 participants (38.8%) watched all 12 videos, 23 (34.3%) participants watched between 1–11 videos, and 18 (26.9%) participants watched 0 videos.

Across all enrolled participants, mean HbA1c decreased from 9.9% (SD = 2.3) at pre-intervention to 9.1% (SD = 2.0) at post-intervention. Table 2 shows the results from the mixed effects regression analysis of change in HbA1c. Time was a significant predictor such that average HbA1c was 0.56% (units) lower at post-intervention compared to pre-intervention (P=0.01). Age was also a significant predictor suggesting that average HbA1c was expected to decrease by 0.04% (units) for each increase of 1 year in age (P=0.04). No other predictors were statistically significant. Results from the analysis of the multiple imputation data sets showed the same pattern of significant and non-significant predictors (Supplementary Table 3).

The mean HEI-2015 total score were relatively unchanged from pre-intervention (59.9; SD = 16.9) to post-intervention (59.5; SD = 13.0). Table 3 shows the results from the mixed effects regression analysis of change in HEI scores. Age was a significant predictor suggesting that average HEI scores were expected to increase by 0.39 units for each increase of 1 year in age (P=0.002). There were also significant race/ethnicity group differences with both the Hispanic (P<0.001) and Marshallese (P=0.006) groups having significantly higher HEI scores than the White group. No other predictors were statistically significant. Results from the multiple imputation analysis showed a similar pattern of significant and non-significant predictors except for education — participants with a high school education or above had significantly lower HEI scores than those with some high school and below (P=0.04) (Supplementary Table 4).

DISCUSSION

In a sample of individuals with T2D experiencing food insecurity participating in a home delivered MTG and DSMES intervention, we found a significant pre-post decrease in HbA1c and no change in HEI scores. An individual’s improvement in glycemic control of this magnitude is considered clinically significant37, reducing the risk for microvascular and macrovascular complications.38 The 0.56% (unit) improvement is similar to findings from another pre-post study including the provision of MTGs to people experiencing T2D and food insecurity, reporting a significant improvement from 8.11% to 7.96%.39 However, key differences within the present study include removal of transportation barriers by providing home delivery of food and in-language DSMES materials (e.g., Spanish, Marshallese, or English). At the same time, our study did not contain a control group, and results may be influenced by other factors (e.g., regression to the mean, medication adjustments).

The finding of significant changes in HbA1c and no change in HEI scores may be related to post-intervention dietary recalls being collected up to 8 weeks after the food deliveries were completed. By the time follow-up diet quality was assessed, participants could have experienced temporary improvements in diet quality that dissipated once food box delivery stopped. Much of the prior research of FIM interventions among individuals experiencing T2D and food insecurity have included measures of individual food groups (e.g., fruit and vegetable intake) rather than overall diet quality, and have often found improvements.15 The diet quality measure in this study (i.e., HEI-2015) is comprised of 13 separate nutrient components, some of which may be expected to improve as a result of a MTG intervention (e.g., increase in whole grains, fruits, vegetables). However, other nutrient components may decline or not be impacted, which could offset any net change. Participants reported food insecurity at the time of enrollment, and this intervention was unlikely to have an enduring effect on socioeconomic conditions that led participants to experience food insecurity, which is associated with low diet quality.2,3 Because HEI scores are related to behavior over the preceding 24h while HbA1c is related to behaviors over the preceding few months, it is possible that HbA1c changes relative to baseline may have dissipated as additional time elapsed since the food deliveries were discontinued. Behaviors beyond change in dietary intake (e.g., physical activity, blood glucose monitoring, and managing medications) were also key components of the DSMES materials in this intervention, which may have also contributed to improvements in HbA1c.23,27,28

Findings from the HEI analysis of multiple imputation data sets showed a similar pattern of significant predictors (e.g., age and race/ethnicity). However, 1 difference in the multiple imputation analysis included that the education variable became significant, whereby participants with a high school degree/GED or more had a significantly lower HEI score compared to participants with some high school or less. These findings highlight the consistency in this study of age and race/ethnicity as strong predictors of HEI scores.40,41

Hispanic and Marshallese participants showed significantly higher diet quality than White participants across pre- and post-intervention. This finding highlights the heterogeneity of dietary patterns among a sample of people actively experiencing food insecurity and T2D, who sought food at food pantries. Another study investigating diet quality across race/ethnicity in a nationally representative sample of adults also found Hispanic adults to have a higher diet quality compared to non-Hispanic White adults, in part due to differences in individual HEI components (e.g., fruits, vegetables, legumes) that are prevalent in Hispanic cultural dietary patterns.41 A prior analysis of this sample examined the variations in HEI nutrient component scores across racial/ethnic groups and identified opportunities to tailor food delivery interventions to capitalize on existing dietary strengths within groups (e.g., tailoring food to include beans for Hispanic participants and seafood for Marshallese participants).42

A success of our study is that 97% of participants received the full dose of food deliveries and DSMES, in contrast with another MTG intervention in which 18% of participants met criteria for full engagement.17 However, it is also important to recognize the missing data for primary outcomes in our study and the possibility that the intervention did not work among individuals who did not complete post-intervention data collection, which could bias the study results in a positive direction. Notably, we found age to be significantly higher among those with complete data for HbA1c and HEI. Reasons for this finding could include work and childcare commitments, although this was not captured by this study.20,21 Thus, it is important to consider that this study’s findings may be less representative of younger individuals.

It is important to emphasize that the study’s participants were recruited while visiting food pantries. Many recent FIM research studies have recruited patients while attending an appointment with a health care provider.43,44 Compared to participants enrolled from food pantries, participants enrolled during a clinic visit may have more robust access to health care, may have been experiencing acute health care challenges that led them to seek care, and may have received other treatments in addition to FIM on the day they were recruited. For these reasons, any similarities in participants’ HbA1c change from baseline to follow-up between the present study and clinic-based FIM studies may be the result of different mechanisms.

Limitations on the interpretation of findings from this initial investigation includes the lack of a control group. It is impossible to rule out the possibility that participants’ improvements in HbA1c were due to factors outside of the food delivery and DSMES materials provided by the intervention. The one group pre-post study design also limited investigation about whether the improvements in HbA1c may be attributed to the food provided versus the DSMES materials. Similarly, the present study did not provide robust data about the extent to which participants themselves ate the food delivered to their homes, used the DSMES materials provided by the intervention, or began other treatments for T2D (e.g., medications). These limitations are important considering other FIM studies in which both the control groups and intervention groups experienced improvement in HbA1c similar to the level of improvement in the present study.17,45

IMPLICATIONS FOR RESEARCH, PRACTICE, AND POLICY

Improved glycemic control among individuals experiencing food insecurity and T2D was achieved, despite no pre-post changes in diet quality. To provide additional context to diet quality results, researchers may consider evaluating changes in HEI at multiple time points (baseline, mid-intervention, follow-up) and changes over time in each of the 13 HEI nutrient components. Results from the HEI analysis of multiple imputation data sets (i.e., participants with a high school degree/GED or more had a lower HEI score compared to some high school or less) suggest that the role of education on HEI scores may need further investigation with larger studies containing similar inclusion criteria. To disentangle the impact of DSMES materials versus food delivery on HbA1c, future researchers may consider a multiple group intervention design. Further, capturing behaviors outside of dietary intake (e.g., physical activity, confidence in diabetes self-management) may help to explain other behaviors that lead to changes in HbA1c. Future FIM research may build on this study’s findings and explore the mechanisms and opportunities whereby FIM can benefit people reached outside of the health care system.

Practitioners involved in the design and implementation of FIM programs may consider strategies to improve engagement with the intervention to achieve the greatest impact on nutrition and health outcomes. The present study integrated home delivery of food and DSMES materials to address participation challenges to ensure 97% of participants received access to the intended intervention dose. Moving beyond home delivery, FIM practitioners may consider engaging their intended participants to identify and test potential solutions to improve participation in their communities (e.g., offering transportation to food pick-up sites) and identifying additional ways to engage participants in DSMES (e.g., group and/or individual education with a Registered Dietitian or Certified Diabetes Care and Education Specialist).

Lastly, practitioners may consider tailoring food and DSMES materials to build on existing dietary strengths across racially/ethnically diverse communities with varying cultural food preferences, to improve program engagement and effectiveness.