Addressing cost barriers to healthy eating with Eat Well, a prescription produce subsidy, for patients with diabetes and at risk for food insecurity: Study protocol for a type 1 hybrid effectiveness-implementation pragmatic randomized controlled trial

Abstract

Background:

Patients with diabetes at risk of food insecurity face cost barriers to healthy eating and, as a result, poor health outcomes. Population health management strategies are needed to improve food security in real-world health system settings. We seek to test the effect of a prescription produce program, ‘Eat Well’ on cardiometabolic health and healthcare utilization. We will also assess the implementation of an automated, affirmative outreach strategy.

Methods:

We will recruit approximately 2400 patients from an integrated academic health system in the southeastern United States as part of a two-arm parallel hybrid type 1 pragmatic randomized controlled trial. Patients with diabetes, at risk for food insecurity, and a recent hemoglobin A1c reading will be eligible to participate. The intervention arm receives, ‘Eat Well’, which provides a debit card with $80 (added monthly) for 12 months valid for fresh, frozen, or canned fruits and vegetables across grocery retailers. The control arm does not. Both arms receive educational resources with diabetes nutrition and self-management materials, and information on existing care management resources. Using an intent-to-treat analysis, primary outcomes include hemoglobin A1C levels and emergency department visits in the 12 months following enrollment. Reach and fidelity data will be collected to assess implementation.

Discussion:

Addressing food insecurity, particularly among those at heightened cardiometabolic risk, is critical to equitable and effective population health management. Pragmatic trials provide important insights into the effectiveness and implementation of ‘Eat Well’ and approaches like it in real-world settings.

Registration:

ClinicalTrials.gov Identifier: NCT05896644; Clinical Trial Registration Date: 2023-06-09.

1. Background

Diabetes mellitus affects over 37 million people in the United States, with type 2 diabetes (T2D) comprising 90% to 95% of all diagnosed cases [1,2]. While pharmacologic therapy, lifestyle modifications and dietary changes are highly effective approaches to diabetes management [3], approximately 20% of those with T2D live with food insecurity (FI) [4]. FI is defined by limited or uncertain access to adequate food [5,6] introducing additional barriers to accessing nutritious produce [7,8], and consequently increasing reliance on nutrient-poor and calorie-dense foods (e.g., high glycemic carbohydrates). These factors contribute to poor glycemic and metabolic control [9–12], increased healthcare utilization and expenses [13], and other deleterious outcomes, particularly among those with diabetes [14–16] and heightened cardiometabolic risk [17–19]. As a result of these challenges, there is an urgent need for interventions that not only address FI but also aim to improve overall population health outcomes and nutrition equity [20].

To address the growing need for population health management strategies to address FI, “food is medicine” initiatives [21,22] have emerged as a promising framework. These initiatives prioritize improving access to and consumption of healthy foods as a fundamental tactic for the efficient management and prevention of diabetes, and improved outcomes for other diet-sensitive chronic conditions. A notable evidence-based strategy within this domain is the “prescription produce” (RxP) model, which offers subsidized or free healthy food options to those at risk of FI and associated diet-sensitive conditions like diabetes, obesity, and hypertension through various delivery methods, including redeemable coupons at local food stores or provision of fresh produce at healthcare centers or home delivery [23–28]. Although evidence suggests RxP programs can improve food security, dietary quality, and health markers like Body Mass Index (BMI) and Hemoglobin A1c (HbA1c) [29–32], studies show mixed results and reveal gaps in understanding their long-term impact on health outcomes and healthcare utilization [29]. This variability, coupled with a lack of rigorous randomized control trials, underscores the need for more robust research to evaluate and scale RxP programs.

This pragmatic randomized control trial aims to evaluate the effectiveness of Eat Well, a RxP program targeting individuals with diabetes at risk of FI within the Duke University Health System (DUHS). Administered by Reinvestment Partners (RP), a 501c(3) non-profit based out of Durham, North Carolina, Eat Well provides monthly funds to participants on a reloadable debit card, exclusively for purchasing fresh, frozen, and canned fruits and vegetables. Building on preliminary evidence indicating dietary improvements associated with Eat Well [33,34], this trial aims to rigorously assess its effectiveness and implementation in a real-world setting, hypothesizing improvements in hemoglobin A1C levels and reductions in emergency department visits among participants. To our knowledge, this study protocol represents the largest pragmatic randomized controlled trial of an evidence-based RxP program in real-world settings to date. The outcomes are anticipated to contribute to our understanding of RxP’s effectiveness and practical application.

2. Methods

2.1. Study design and objectives

The study will be a hybrid type 1 pragmatic parallel group randomized controlled study with an intervention and control arm. We will deploy an unequal allocation strategy for randomization based on budget efficiency and recruitment considerations. Participants will be randomized to the intervention arm versus the control arm in a 2 (intervention): 1 (control) ratio. Participants will receive the benefit for up to one year following their enrollment and will be followed for the 12 months following the end of their benefit. The primary study objective is to quantify the impact of the Eat Well funds on HbA1c levels and ED visits among participants with diabetes and at risk for FI. In secondary analyses, we will assess differences in other cardiometabolic measures as well as in outpatient visits and inpatient encounters. For our primary outcomes of interest, we hypothesize that participants randomized to receive Eat Well will have a reduction in hemoglobin A1c after 12 months more than control participants not randomized to receive Eat Well. Similarly, we hypothesize that Eat Well participants will experience fewer ED visits. In addition to assessing Eat Well’s effectiveness, the hybrid type 1 approach also features the collection of exploratory data related to implementation. To do so, we will collect quantitative and qualitative data on implementation to identify strategies to enroll eligible patients using EHR-based eligibility, email recruitment, and REDCap [35] (a secure web application for building and maintaining online surveys and databases), including eConsent and randomization module.

2.2. Setting

The pragmatic trial will be conducted at DUHS. DUHS or ‘Duke Health’ consists of an integrated health system, academic medical center, and the largest primary care network in the region with family and internal medicine providers, pediatricians, and urgent care clinicians in >60 outpatient locations with almost one million patient encounters in fiscal year 2023. Of note, DUHS serves as the primary provider in Durham County, a mid-size county, that until recently has been majority-minority. The study was registered on ClinicalTrials.gov (NCT05896644) before any enrollment activities commenced. The study was reviewed and approved by the Duke Health IRB (Pro00112649).

2.3. Participants

This study aimed to enroll up to 2400 DUHS patients meeting the following criteria: aged 18 or older, active in the Duke Diabetes Registry, at least one outpatient A1c measurement in the 12 months before recruitment, and at risk for FI. The Duke Diabetes Registry is an existing database used for DUHS clinical care and population health management efforts and includes patients who do not have a current diagnosis of prediabetes and at least one of 1) DM diagnosis in their problem list; 2) DM diagnosis in at least two outpatient visits in the prior two years; 3) or has a prescription for antidiabetic medication. Participants at risk of FI must meet at least one of the following criteria: 1) responded sometimes true or often true to food insecurity screening, inability to pay for food, or hard or very hard to financial strain as part of routine screening of health-related social needs system-wide assessment and a digital community based organization referral system described elsewhere [36]); 2) has Medicaid as an insurance payer in the prior year; or 3) or lives in a census block group with an Area Deprivation index (ADI) in the 7th decile or higher. ADI was selected as an inclusion criteria based on community partner input and includes validated neighborhood-level and community measures that influence health outcomes with evidence that being at the higher end of the distribution influences health and health care outcomes [45]. Inclusion criteria are based on data from the DUHS EHR.

2.4. Power and sample size calculation

Based on the unequal allocation randomization strategy described above, we are powered to detect between-arm post-intervention versus baseline changes in HbA1c of 0.185 percentage points or larger, with 80% power and an alpha of 0.025, and assuming 20% missingness (missing completely at random) of the measurement in the post-intervention period [37]. Based on historic HbA1c measurements in the pool of candidates, we assume a standard deviation of 1.5 at both pre and post-intervention points and an expected correlation between measures of 0.7. With 80% power and an alpha of 0.025, we will be able to detect between-arm changes in participants’ ED visits (any vs none) if the odds ratio is at least 1.35 or larger.

2.5. Recruitment

We will query data from an Epic-based Duke Diabetes Registry to identify eligible patients in the Duke Diabetes Registry. Preliminary data shows that over 11,000 patients meet study eligibility criteria, live in North Carolina, and have an e-mail address. Eligible individuals will be sent an email message informing them that they are eligible to participate in the study. The recruitment messages direct patients to a REDCap electronic consent form, which can be viewed in English or Spanish on the webpage. Subsequent communication with the participants will be in the language they select. eConsent will be completed by participants via REDCap. After consenting, participants will automatically be sent a copy of their consent form by email. Participants assigned to the treatment arm will receive an email with diabetes-related educational resources and a personalized link to self-enroll to receive the Eat Well debit card. For those enrolled in the treatment arm, a follow-up email will be sent one week later as a reminder to self-enroll in Eat Well. Participants are provided with a study e-mail address to contact if they have questions or issues enrolling. Also, RP provides a support link on the self-enrollment site which requests issue details and is then immediately transferred to a participant support team to assist within 48 h. Participants assigned to the control arm will receive a message with diabetes-related educational resources. See Fig. 1 for an overview of outreach, randomization, consenting, and enrollment processes.

Fig. 1.

Study Outreach and Recruitment Process.

We will send recruitment messages to 2000 eligible subjects per week until approximately 1400 participants have consented to the treatment arm. Eligible subjects may receive up to three recruitment messages within 2 weeks from the original outreach message. In addition to this automated, mass outreach recruitment strategy, we will educate clinics, primary care providers, and others involved in the care of eligible patients to be aware of the program and be able to direct questions to the study team via a study-specific e-mail address. Waves of outreach will begin with historically marginalized communities near the DUHS medical campus and be subsequently offered to all zip codes where DUHS Diabetes Registry patients reside within the state.

2.6. Randomization

Participants will be randomized to one of two study conditions using an unequal allocation strategy to ensure sufficient enrollment into the program and based on guidance from community partners. Randomization will be stratified by the individual’s average HbA1c level in the prior year: ≥ 8% versus <8%. Block randomization will be employed with a block size of 6. The REDCap randomization module (Realtime Randomization – v1.0.1) will be employed to assign participants to a study arm following consent. See Fig. 2.

Fig. 2.

DUHS Eat Well Pragmatic Trial Randomization Overview.

2.7. Intervention

2.7.1. Eat Well RxP

Participants in the treatment arm will be offered Eat Well in partnership between RP and the DUHS. Funding for fruit and vegetable purchases comes from the North Carolina General Assembly (Grant #: 22–038-2008). Eat Well provides monthly funds to eligible DUHS patients with diabetes for the purchase of fruits and vegetables. After self-enrollment (See Recruitment Section), participants receive a debit card to use on qualified produce purchases (i.e., fresh, frozen, and canned fruits and vegetables). The debit cards are restricted to approved produce using universal product codes (UPC) and product look-up (PLU) codes. PLU and UPC are identification codes and widely used industry product classification system used for produce. Each month, $80 in Eat Well RxP funds is loaded onto the participants’ cards and can be spent across multiple shopping trips at a variety of common grocery store vendors across North Carolina (e.g., Food Lion, Walmart, and Dollar General). The RxP benefit will last up to 1 year for each participant, with possible shorter durations due to the timing of enrollment.

2.7.2. Diabetes-Related Educational Materials

Participants in both the treatment and control arm will all receive educational materials related to diabetes self-management that are used at DUHS as the standard of care. All participants receive an e-mail message with a curated selection of hyperlinks to existing approved materials including recipes, guidance on nutrition and physical activity, and other information related to diabetes management, including low blood sugar, high blood sugar, foot health, and wound care. Furthermore, the materials provide information on accessing accredited diabetes self-management and support programs, as well as the DukeWELL program, which provides support services for Duke patients between clinic visits.

2.8. Data collection

All data collection occurs via routine encounters with the health care system as part of the standard of care. The primary outcomes of interest include hemoglobin A1c, or the average blood glucose level over the past three months [38], and ED visits. Secondary outcomes and covariates include sociodemographic and health-related characteristics including age, sex, race/ethnicity, health insurance, medical history and current health status, weight, body mass index (BMI), cholesterol, and blood pressure. Data will be abstracted from the DUHS Electronic Health Record (EHR), RP administrative data, and REDCap. See Table 1 for an overview of the outcomes of interest.

Table 1.

DUHS Eat Well Pragmatic Hybrid Randomized Controlled Trial Measures.

Outcome	Description	Source	Timepoint(s)
Primary Outcomes
Hemoglobin A1c	Average blood sugar level over the past three months. Collected routinely in outpatient encounters.	EHR	18 months pre-program, during program enrollment, and up to 18 months post-program
ED visits	A count of emergency department visits.	EHR	18 months pre-program, during program enrollment, and up to 18 months post-program
Secondary Outcomes
Healthcare utilization	A count of hospitalizations and outpatient visits.	EHR	18 months pre-program, during program enrollment, and up to 18 months post-program
Cardiometabolic health	Body mass index (BMI), systolic and diastolic blood pressure (BP), cholesterol (HDL & LDL)	EHR	18 months pre-program, during program enrollment, and up to 18 months post-program
Exploratory Outcomes
	List of diabetes and hypertension medications	EHR	18 months pre-program, during program enrollment, and up to 18 months post-program
Medication adherence	Adherence to front-line medications (e. g., metformin) will be calculated using the proportion of days covered (PDC) approach	EHR-based e-prescription network data describing medication refill information.	18 months pre-program, during program enrollment, and up to 18 months post-program
Implementation Outcomes
Reach	Proportion of patients that received the Eat Well subsidy among patients randomized to be invited to enroll in the program (i.e., randomized to receive it)	RedCAP trial platform and RP administrative data	After recruitment and throughout the approximately 12 months of Eat Well program enrollment
Fidelity	Degree to which an intervention was implemented as it was prescribed in the original protocol or by the program developers [39]. Measured as initial use of the subsidy and the average benefit usage rate (e.g., the average % of the monthly benefit used by each participant)	RP administrative data	After recruitment and throughout the approximately 12 months of Eat Well program enrollment

HDL = high-density lipoprotein; LDL = low-density lipoprotein.

2.9. Data Analysis

Outcomes of interest will be analyzed in a repeated measures framework to estimate between-arm differences over time. The primary comparisons of interest will be 1) between-arm differences in the closest HbA1c measurement to the 12-month timepoint (compared to baseline) and 2) between-arm differences in total ED visits during the program (compared to the total in the 12 months pre-program). Secondary analyses will model longitudinal trends in both measures using additional time points (6 months, 18 months, 24 months post-enrollment). We will additionally explore A1c measurements in a mixed models framework allowing for irregular capture. The primary hypothesis is that there will be between-group differences in the change in HbA1c levels or differences in the change in ED visits at the 12-month timepoint (multiple primary endpoints). These estimates will be obtained from a repeated measures framework and their significance will be assessed using Wald tests with alpha = 0.025 for each outcome.

The primary analysis will follow the intention to treat principle, and all participants will be analyzed according to the group to which they were randomized. Clinical and utilization outcomes will be abstracted from the medical record based on values captured during routine clinical care encounters. Clinical outcomes may be missing due to the participant moving to a different geographical region, receiving care outside of the study health system, or because routine care is missed or postponed. We will approach the analysis of missingness depending on the drivers and scale of missingness observed in the data. We will investigate and describe missingness by arm and by other covariates, and we will assess the potential for missing completely at random (MCAR) and missing at random (MAR) mechanisms. We will use imputation or weighting methods to address missingness if appropriate.

Further, as an exploratory analysis, we will analyze the heterogeneous treatment effect of Eat Well by assessing the impact of the intervention in subgroups that are hypothesized to potentially derive greater benefit. This includes assessing the healthy equity implications of the program by examining the impact on Medicaid enrollees and minoritized populations (i.e., African American and Hispanic). This will also include assessing the impact of Eat Well on patient populations at heightened clinical risk including those with hbA1c > 8 to match the randomization schema and for patients with congestive heart failure. In addition to analyzing average and heterogeneous treatment effects from program participation, we will also conduct similar analyses to assess patterns of uptake and engagement in the program (fidelity measures of enrollment and average monthly benefit use described in Table 1) to identify opportunities to enhance equitable program design and implementation.

3. Discussion

FI imposes significant health burdens on patients, especially for those with diabetes which increases the risk of heart, vascular, and renal disease. Improving health outcomes for this population must include being responsive to FI, given that it drives poor outcomes and care utilization patterns [40–46]. RxP programs, like Eat Well, improve food and nutrition security by supporting access to healthier foods [34].

While initial findings indicate positive patient acceptance, uptake, and improved food security, there is limited knowledge regarding RxP programs, like Eat Well, that impact clinical outcomes, especially in real-world settings. Many studies of RxPs have been small, conducted in a more controlled, research context, or lacked an experimental design (i.e., observational). Further, best practices related to the implementation of RxP are not well understood despite their critical importance for informing the design and equitable scaling. Considering these gaps, our objective is to launch a large, randomized hybrid type 1 pragmatic trial of the Eat Well program to better understand its effectiveness and implementation as a population health management strategy.

Eat Well has several notable and unique features, even when compared to other RxPs. For instance, patients have full choice to purchase produce from a variety of grocery retailers without making additional trips or going out of their way, which improves ease of access and personal autonomy. This element of choice caters to individual preferences and provides agency to patients, allowing individuals and households to actively participate in food choices based on taste, cultural considerations, preparation considerations, and other factors. Further, the Eat Well program lends itself to scaling and high levels of reach as a population health strategy. Given the automated aspects of recruitment and enrollment, the infrastructure developed for this trial could be deployed efficiently without provider time or introducing cumbersome and novel practice patterns to existing clinical workflows. In addition, Eat Well is a larger monthly subsidy ($80 per month for 12 months) than is typically implemented. Given recent research that found null or mixed effects for RxPs that offered $50–$60 per month [47] or for a shorter duration (e.g., 6 months [48,49]), this trial should extend the impact of this program with increased dosage (a recognized important research frontier for RxP and ‘food is medicine’-type interventions [50]). Lastly, the impact of affirmative outreach will be a novel contribution of this research. This approach leverages existing data to determine eligibility and notify patients of their eligibility without a clinical encounter or an in-person social needs screening process. Paired with online enrollment, this process takes advantage of economies of scale and could complement more costly and labor-intensive social needs screening and response case management activities [51]. Similarly, while the Eat Well program does include extensive nutrition and self-management education, ongoing RxP studies seek to evaluate the impact of complementary education from a registered dietician or other members of the care team. For example, the LINK study, a pragmatic randomized controlled trial, is investigating the effectiveness of combined interventions including produce provision, diabetes education, and social needs resolution in improving A1c levels in food-insecure patients with type 2 diabetes, providing context for the Eat Well program’s potential impact [52]. The potential public health benefit of RxP interventions span the spectrum of prevention including managing or preventing diabetes to improving outcomes among highest risk patients when embedded within an existing care model.

A key direction of future research, as underscored in the American Heart Association’s recent Presidential Advisory [50], involves not only assessing the potential of RxP interventions as cost-effective strategies for health improvement, [50] but also evaluating their implementation. This entails enhancing enrollment and ongoing engagement in programs and optimizing program delivery to maximize their effectiveness. The advisory highlights the need for a human-centered design approach in innovation, with a focus on incorporating personalized needs and preferences into program creation [50]. It also emphasizes the importance of systematic testing to identify patients who may benefit from RxP interventions, methods to increase enrollment, and strategies to promote active engagement and retention [50]. These approaches are vital to address the limitations identified in existing research, where many interventions have not been tested in randomized trials, thus precluding definitive conclusions on efficacy or effectiveness. The descriptive findings related to implementation from this hybrid trial (e.g., data on intervention reach and fidelity) will inform efforts to design and test RxP implementation strategies. Specifically, data collected related to both uptake and ongoing fidelity to the program will yield insights into how to best optimize effectiveness and equity. To complement this trial, future research includes interviews with participants and non-participants to better design strategies to improve uptake and fidelity. For example, there may be vulnerable populations that could disproportionately benefit from Eat Well but face unique technology and redemption barriers to obtain or use the benefit (e.g., lack of internet; food environment). Future research on RxP requires a qualitative line of inquiry that leverages implementation science frameworks that incorporates a health equity lens (e.g., health equity implementation framework) and/or user centered design principles to enhance the participant experience with attention to equitable access. Future research should also use changes in utilization measures from trials like this one to estimate short- and long-term health care cost savings among higher risk patients. Emerging research on RxP cost effectiveness suggests that they may warrant additional investment from health care non-health care sector entities.

There are important limitations to this pragmatic trial that should be noted. The fact that it is a single-site trial will limit generalizability to other health systems and other parts of the country or world. Also, Area Deprivation Index is used for identifying patients at risk for FI, but this may be a flawed measure since it is a neighborhood-level construct. For example, there is evidence that ADI may overemphasize home values and not be a strong proxy for the underlying economic household conditions [53]. Additionally, beyond benefit usage, we do not have data to observe participant engagement with the online educational materials or historic grocery purchase history to assess the impact of Eat Well on these aspects of engagement. Existing research suggests the Eat Well program promotes changes to grocery purchasing patterns consistent with improvements in diet quality but impact on other diabetes self-management behaviors is understudied. Finally, as with all pragmatic trial designs, there is a risk of censoring (i.e., patients receiving care outside of DUHS) or loss of follow-up data on study participants since all clinical and utilization data is derived from routine care visits. Despite these limitations, this large pragmatic hybrid trial fills important gaps in our understanding of Eat Well, and RxP more broadly, effectiveness and implementation. The findings will add to the emerging evidence base on ‘Food is Medicine’-style interventions to ensure best practices related to intervention design and implementation are identified.

Data availability

No data is reported for the study protocol described in this article.