A Subsidized Healthy Food Prescription Program for Adults with type 2 Diabetes who are Experiencing Food Insecurity: Protocol for a Randomized Controlled Trial, Modelling and Implementation Studies

Background: The high cost of many healthy foods poses a signicant challenge to the maintenance of optimal blood glucose levels for adults with type 2 diabetes (T2DM) who are experiencing food insecurity, leading to diabetes complications and excess acute care usage and costs. Subsidized healthy food prescription programs may help to reduce food insecurity by nancially supporting patients to improve their diet quality, prevent diabetes complications and avoid acute care use. This study will use a type 2 hybrid effectiveness-implementation design to examine the reach, effectiveness, adoption, implementation and maintenance (RE-AIM) of a subsidized healthy food prescription program for adults who are experiencing food insecurity and persistent hyperglycemia. A randomized controlled trial (RCT) will investigate program effectiveness via impact on blood glucose levels (primary outcome), food insecurity, diet quality and other clinical and patient-reported outcomes. A modelling study will estimate longer-term program effectiveness in reducing diabetes-related complications, resource use and costs. An implementation study will examine all RE-AIM domains, including reasons behind program successes and failures, delity, mechanisms of impact, contextual determinants of effective implementation and sustainability. Methods: 404 adults who are experiencing food insecurity and persistent hyperglycemia, including adults who identify as Indigenous, will be randomized to a subsidized healthy food prescription intervention (n=202) or a healthy food prescription comparison group (n=202). Both groups will receive a healthy food prescription. The intervention group will additionally receive $1.50/day/household member to purchase healthy foods in supermarkets for 6 months. The implementation process will follow the Quality Implementation Framework. Outcomes will be assessed at baseline and follow-up (6 months) in the RCT and analyzed using mixed-effects linear and multinomial logistic/ordinal regression models. Longer-term outcomes will be modelled using the validated UK Prospective Diabetes Study outcomes simulation model-2. Implementation processes and outcomes will be continuously measured via quantitative and qualitative data and analyzed using descriptive statistics and theory-informed directed content analysis, respectively. Discussion: This research provide a body of with to or change over time; staff for drop-out; and commitments study partners, and participating supermarkets on to

An implementation study will support translation of ndings into practice by demonstrating reasons behind program successes and failures, mechanisms of action and determinants of effective implementation. Background Type 2 diabetes (T2DM) imposes a tremendous burden on healthcare systems worldwide, as individuals with T2DM incur twice the healthcare costs as their age-and sex-matched counterparts [1,2]. The total economic costs of diabetes were $327 billion in 2017 in the US, and $30 billion in Canada in 2019, making it among the most expensive chronic conditions in both nations [1,3,4]. The human toll on individuals and their families is also substantial in terms of reduced quality of life associated with managing the disease [5,6]. Many of these human and economic costs are avoidable, as adherence to a healthy diet within an overall diabetes management plan can yield clinically meaningful improvements in blood glucose levels, which can reduce diabetes complications over time [7][8][9][10][11][12][13][14][15]. Average blood glucose levels are most often quanti ed using the glycosylated hemoglobin level (A1C), which represents the average blood sugar level over the previous 3 months [16]. An absolute reduction of 0.5% in A1C is achievable through improving diet quality [7,8] and is considered a clinically meaningful difference [17].
Indigenous groups (constitutionally recognized as First Nations, Inuit and Métis) are a population of particular concern, given their disproportionately high rates of both T2DM and food insecurity [40][41][42]. The coexistence of food insecurity and T2DM therefore has major implications for the sustainability of healthcare systems.
Although it is well-known that food insecurity is a primary driver of acute care usage and costs, healthcare providers often lack effective strategies to address it. One approach to better address this problem is to assist patients who are experiencing food insecurity to purchase diabetes-appropriate foods through subsidized healthy food prescription programs. Preliminary evidence suggests that these programs may improve diet quality and self-reported health, while reducing food insecurity, hypertension and BMI, including within Indigenous communities [43][44][45][46][47][48][49][50][51][52][53][54][55][56][57]. Notably, one $40 fruit and vegetable subsidy delivered to 65 low-income patients with persistent hyperglycemia over 13 weeks led to a 0.71% decline in A1C [56]. Qualitative data similarly suggest patients and care providers perceive nancial, dietary and health bene ts from these programs, and support their ongoing delivery [43,50,[58][59][60]. Food prescription programs also appear to be cost-effective, with one recent modelling study indicating that a national subsidized healthy food prescription program in the US could eliminate $100.2 billion in healthcare costs if implemented over the lifetime of bene ciaries [61].
Despite some promising initial ndings, major knowledge gaps remain pertaining to the impact and optimal implementation of subsidized healthy food prescription programs. Most prior studies have been small and uncontrolled, and have examined a small number of self-reported outcomes using brief dietary [44,45,48,60,62] and/or food insecurity screeners [47,[50][51][52][53], and/or short descriptive surveys [44,58,60], rather than objective clinical outcomes. Prior programs have also subsidized the purchase of fruits and vegetables alone, without considering the relevance of entire dietary patterns to blood glucose levels and health outcomes [7,8]. Moreover, there is virtually no understanding of the effectiveness and cost-effectiveness of these programs over the longer term, nor of optimal implementation strategies [63].
We will build upon these initial ndings through three concurrent studies, including a randomized controlled trial (RCT), a modelling study and an implementation study, using a type 2 hybrid effectiveness-implementation design [64]. Collective ndings will be integrated to provide a comprehensive perspective of the reach, effectiveness, adoption, implementation and maintenance (RE-AIM) [65,66] of a subsidized healthy food prescription program among adults who are experiencing food insecurity and persistent hyperglycemia. First, the RCT will provide a basis for causal inference pertaining to program effectiveness. It will entail a broad food subsidy applicable to all major food groups, and will be powered to detect clinically meaningful changes in A1C, along with a comprehensive range of objective and self-reported health-related outcomes. A linked modelling study will provide a longer-term perspective of program effectiveness in reducing diabetes-related complications, along with healthcare use and costs. Finally, a complementary implementation study will encompass quantitative and qualitative measures of all RE-AIM domains to support translation of research ndings into practice and policy by helping to understand reasons behind program successes and failures, implementation delity, mechanisms of impact, contextual determinants of effective implementation and the likelihood of sustainability.

Overview
Ethics, privacy and con dentiality This research has been approved by the University of Calgary Conjoint Health Research Ethics Board (REB20-0543) and the University of Alberta Health Research Ethics Board Biomedical Panel (Pro00107116). Any protocol deviations will be approved in advance by the Board and updated in the clinical trials registry. All participants will provide written, informed consent prior to data collection (Additional le 1). Participant data will be anonymized and stored on a password protected University server. Only the Principal Investigators and Research Coordinators will have access to identi able participant information and the nal trial datasets.

Setting
This research will take place in Alberta, Canada. Participants will be recruited through primary care and diabetes specialty clinics located in urban and rural communities, including clinics with an explicit focus on serving people identifying as Indigenous.

Study oversight
Scienti c Steering Committee A Scienti c Steering Committee will oversee all aspects of the research, receive and review reports from the study's Advisory Boards and Subcommittees, and will have nal decision-making authority. It will be comprised of the study's ve co-Principal Investigators (DLO, ES, RFB, LL, DJTC).

Advisory Board
A multi-stakeholder Advisory Board will provide high-level oversight for the research and will advise the Scienti c Steering Committee on study conduct. Members will include policymakers, academic experts, representatives from Alberta Health Services (the provincial health authority), an Indigenous public health expert and a patient.

Indigenous Advisory Board
An Indigenous Advisory Board will ensure that research activities within Indigenous clinics proceed in a culturally sensitive, relevant, responsive, equitable and reciprocal manner that is guided by OCAP (Indigenous Ownership, Control, Access and Possession of data) principles [67] and complies with Government of Canada guidance for Indigenous Research [68,69]. The Board will include Indigenous Elders and patients, along with academic experts, policymakers, managers and front-line practitioners from the public health and healthcare sectors who are themselves Indigenous, or who work closely with Indigenous peoples.

Primary Care Clinic Subcommittee
The Primary Care Clinic (PCC) Subcommittee will include PCC managers, staff and patients. As participant recruitment and implementation of the intervention unfolds, PCC managers and staff will collect feedback from their respective clinics and will share it with the larger group as a learning tool and to inform ongoing adaptations.

Implementation Support Team
The Implementation Support Team will consist of Research Coordinators and Assistants who will execute the daily tasks required to administer, plan, support, monitor and evaluate the subsidized healthy food prescription program.

Patient and public involvement
This research has been informed by substantial prior [70][71][72][73][74] and ongoing engagement with patients experiencing nancial barriers to chronic disease care. Patients (who are not study participants) will help to pilot test infrastructural supports (e.g. healthy food prescription pamphlet, usability of the subsidy list), care pathways and implementation processes, and will be members of the Advisory Boards and PCC Subcommittee. Patients who are study participants will provide continuous program feedback via a dedicated study help-line/email, and by completing implementation delity checklists. At the conclusion of the study, participants will be invited to describe their program experiences via a post-intervention questionnaire and during in-depth, semi-structured interviews.
Lands and food hold deep cultural, symbolic and spiritual signi cance for Indigenous peoples [75]. Staff and patients from Indigenous PCCs will co-design clinical care pathways and other procedures that are contextspeci c and culturally appropriate for Indigenous patients, and that respect and promote Indigenous worldviews, particularly those surrounding food procurement and consumption. We will ensure that representatives from Indigenous clinics are involved at all stages of the research, including study design, pilot testing infrastructural supports, interpreting results and formulating conclusions, and that their agreement is obtained prior to communicating any research ndings that pertain to them. As previously described, the Indigenous Advisory Board will also oversee all aspects of the research.
Evaluation framework and theory of change RE-AIM [65,66] will provide a structured means of integrating data from the RCT, modelling and implementation studies to understand the reach, effectiveness, adoption, implementation and maintenance of the subsidized healthy food prescription program.
Our theory of change draws on Barnard et al's [76] conceptual model linking material needs insecurities with diabetes outcomes, and posits that reduced food insecurity and improved diet quality will be key mediators of improved blood glucose levels, which will help to reduce diabetes complications, and healthcare resource use and costs. Each construct will be examined to a rm or disprove the proposed pathway ( Figure 1).

Randomized controlled trial
The RCT protocol adheres to the SPIRIT and TIDieR reporting guidelines (Table 1; Additional File 2-4).

Study design and objectives
A 6 month, parallel-group RCT will examine the effectiveness of a subsidized healthy food prescription program, compared to a healthy food prescription alone, in improving the following outcomes among 404 adults who are experiencing food insecurity and persistent hyperglycemia (i.e. A1C 8-12%): 1) Primary outcome: Average blood glucose levels measured by A1C 2) Secondary outcomes: a. Blood glucose levels: Proportion of patients with elevated A1C (i.e. ≥ 9%); blood glucose measured by fructosamine b. Dietary intake: Diet quality; skin carotenoids c. Intermediate clinical outcomes: Blood lipids; blood pressure; body mass index (BMI); waist circumference; need for diabetes medication/insulin d. Patient-reported outcomes: Psychosocial well-being; self-rated health; diabetes self-e cacy; diabetes selfmanagement; diabetes distress; diabetes competing demands; perceived nancial barriers to chronic disease care; hypoglycemic episodes; household food insecurity 3) Exploratory outcomes: Subjective social status; perceived income adequacy; work productivity and activity impairment; medication and physical activity adherence Primary care clinics PCCs will be recruited, including urban, rural and Indigenous clinics. To be eligible, clinics must serve lowerincome patients, agree to allow their physicians, Registered Dietitians and/or nurses to dispense healthy food prescriptions, appoint a staff member to liaise with the Implementation Support Team, and be willing to receive training. The nal list of study sites, currently projected at 30 clinics, will be available in the clinicaltrials.gov registry.

Participants
PCCs will use information from electronic medical records to identify patients with T2DM and persistent hyperglycemia (i.e. A1C 8-12%), including patients living in rural and urban areas, and those who identify as Indigenous. Potentially eligible patients will be invited to complete a brief screening questionnaire, including the validated two-item hunger vital sign, which is 97% sensitive and 83% speci c for food insecurity [77][78][79]. Eligible patients will be adults (18-85 years) with persistent hyperglycemia (i.e. A1C 8-12%) who are experiencing food insecurity, who are not currently on intensive insulin therapy (multiple daily injections or continuous subcutaneous insulin infusion), do not reside in a facility that provides meals (e.g. shelter, long term care, prison), and can communicate in English or have someone to translatePatients will be excluded if they have an A1C < 8% or > 12% (given the recommendation for anti-hyperglycemic treatment escalation for those with A1C > 12%), have signs/symptoms of metabolic decompensation, have an eating disorder, are functionally dependent, have limited life expectancy (< 2 years), have experienced diabetic ketoacidosis or a hyperglycemic hyperosmolar emergency in the past 2 years, have a history of recurrent severe hypoglycemia or hypoglycemic unawareness, or if they have been diagnosed with dementia. Patients will also be excluded if they are pregnant or trying to conceive, breastfeeding, participating in other clinical trials, if someone in their household is currently or has previously participated, if they will no longer reside within 100km of a participating supermarket in the next 6 months, if they plan to leave Canada for more than 2 weeks in the next 6 months, or if they will not be able to complete data collection at 6 months (e.g. due to moving).
Eligible patients will be asked to provide consent to their healthcare provider to be contacted by the research team. A Research Assistant will contact patients to con rm all eligibility criteria have been met, obtain informed consent and provide instructions for collection of baseline data. Participants may elect to report baseline patientreported data immediately over the telephone, or independently via the study's online data collection platforms.
Any patients identi ed as food insecure at screening using the two-item hunger vital sign, but who do not respond a rmatively to any of the items on the full 18-item Household Food Security Survey Module during baseline data collection will be excluded. Participants will have biochemical and physical measurements performed at a community laboratory or at their PCC. Participants with an A1C outside the 8-12% range will be excluded at that point. All participants will subsequently receive a healthy food prescription pamphlet from their healthcare provider (i.e. physician, nurse, Registered Dietitian) and a brief, high-level overview of its contents using standardized teaching guidelines, either virtually or in-person during a clinic visit.

Sample size calculation
Based on local administrative data, we expect a mean baseline A1C of 9.1% (SD=1.0%) in our sample (i.e. individuals with A1C 8-12%) [80]. Assuming 5% type I error, 35% attrition [81] and potential design effects based on sampling in different clinics (25% in ation), 403 participants are required for 95% power to detect a difference in A1C of 0.5%, which is often considered a minimally important clinical difference [17].

Randomization and blinding
Following baseline data collection and delivery of the healthy food prescription pamphlet, participants will be randomized to a subsidized healthy food prescription intervention (n=202) or a healthy food prescription comparison group (n=202) with a 1:1 allocation ratio using a computer-generated, concealed, blocked randomization sequence created by an independent statistician. Blocking variables will include gender, clinic type location (urban, rural, Indigenous) and baseline A1C (8-9%, 9-12%). Allocation concealment will be ensured via secure storage of the randomization sequence separately from the participant database, which will only be accessible by the statistician. To ensure researcher blinding, randomization will be operationalized via REDCap (Research Electronic Data Capture) following completion of baseline data collection. Intervention assignment will be communicated by Research Assistants via a telephone call. Participants cannot be blinded to treatment allocation, however details of the healthy food subsidy, including its monetary value and the types of foods that are eligible, will not be divulged to participants in the comparison group. Care providers and individuals who collect biochemical and physical measurements will be blinded to group allocation.

Intervention
Development of the subsidized healthy food prescription program was informed by the social prescribing literature, EQUIP (Research to Equip Primary Healthcare for Equity) principles of equity-oriented healthcare [82], initiatives elsewhere (e.g. Wholesome Wave [83]), and stakeholder consultation. The comparison group will receive a one-time healthy food prescription pamphlet. The intervention group will receive a one-time healthy food prescription pamphlet and a 6 month subsidy of $1.50/day/household member (e.g. household of four receives $42/week) to purchase healthy foods in study-a liated supermarkets. Thus, the study is designed to test the impact of a healthy food subsidy, which is an intervention that targets economic rather than knowledge-related barriers to healthy eating. Aside from labelling the nutritional advice delivered as a "prescription" (which may have some independent impact on participants' behaviour), the healthy food prescription closely mimics current care (i.e. nutrition counselling) and is unlikely to substantially change dietary intake in the context of severe economic constraints [84][85][86][87][88][89]. The subsidy amount exceeds the bene t provided by similar US programs (e.g. [56,83]) in order to more closely bridge the gap in food spending between food secure and insecure households in Canada [90]. In their systematic review, Rao et al (2013) demonstrated that a subsidy of $1.50/day was associated with greater compliance with a Mediterranean style dietary pattern. The subsidy amount that each household receives will be determined based on household size at baseline; this amount will remain consistent throughout the intervention regardless of changes in household size. A household member is de ned as a partner or a dependent child or adult who resides at the same location at least 50% of the time. Finally, the intervention will be delivered over 6 months to allow su cient time for dietary changes to be re ected in approximately two A1C cycles [91].
The healthy food prescription pamphlet was designed by Registered Dietitians and modelled after a previous food prescription program to be a visually appealing, low literacy resource [92] (Additional File 5). The cover page contains the following pre-printed prescription "I prescribe a healthy eating pattern of minimally processed foods that have little to no added fat, sugar or salt," with space for the care provider to add their signature, date, and patient information. The inner pages outline an evidence-based healthy dietary pattern, with key messages to consume a variety of whole, minimally processed foods from all food groups with little to no added fat, sugar or salt, to spread carbohydrate foods over the day, to satisfy thirst with water, and to avoid sugary drinks, re ned grains, sweets, confectionary and desserts [7,93]. A diabetes-appropriate recipe is provided along with links to connect patients with sources of free/lower-cost food, additional recipes, nutrition information, other helpful community services and sources of emergency food assistance. Feedback from PCC staff and patients and the Advisory Boards was incorporated into the nal version of the pamphlet.
The healthy food subsidy can be used to purchase whole, minimally processed foods with little to no added fat, sugar or salt from all food groups [7,93] (Table 2). Ineligible items include beverages (except white milk), re ned grains, sweets and other foods with excess fat, sugar or salt. A pilot-tested PDF booklet was created with pictures of eligible items to assist participants to locate eligible items and Research Assistants will review the list of subsidized foods and subsidy redemption procedures with participants prior to the intervention. Participants in the intervention group can redeem their food subsidies at participating supermarkets located throughout Alberta. Subsidies will be loaded onto participants' store loyalty card weekly, and dollars-off discounts will be automatically applied on subsidy-eligible foods at the point-of-purchase. The entire food subsidy must be spent during a single shop, either in-person or using a pick-up service. Subsidies not redeemed within the week will expire. To minimize potentially stigmatizing experiences, all participants in the intervention group will receive a one-time unrestricted grocery gift card valued at $30. The grocery gift card can be used in the event that participants inadvertently exceed their subsidy amount and lack funds to cover the additional costs. Participants will also have access to a study email and telephone help-line where they can ask study-related questions. To support retention, all participants (regardless of treatment allocation) will receive $100 following completion of data collection at baseline (0 months) and again at follow-up (6 months).

Questionnaires
Electronic questionnaires will encompass sociodemographic and health-related items, dietary intake in the previous 24 hours, and a variety of patient-reported outcomes. The nal questionnaires will be reviewed by the Advisory Boards and Scienti c Steering Committee to establish face and content validity, and will be pre-tested with patients.
Sociodemographic and health-related variables will be recorded in REDCap using existing items from the Canadian Community Health Survey where available [94], including date of birth, sex at birth, gender identity, race/ethnicity, years lived in Canada, household size and composition, number of household members with T2DM, educational attainment, employment status, marital status, annual household income, main income source, access to extended health bene ts, participation in income support programs, smoking status, housing status, medication/insulin type and dose, duration of diabetes and physical activity level [95].
Patient-reported outcomes will be assessed using the following validated scales: World Health Organization wellbeing scale [96], Stanford Diabetes Self-E cacy Scale [97,98], Diabetes Self-Management Questionnaire [99,100], Problem Areas in Diabetes Scale-5 to assess diabetes distress [101], EQ-5D-5L to assess self-rated health [102], hypoglycemic episodes [30], Work Productivity and Activity Impairment Questionnaire [103], Health Canada's Household Food Security Survey Module to assess experiences of marginal, moderate and severe household food insecurity in the past 6 months [104][105][106] and medication adherence [107]. We will also assess diabetes competing demands [108] and perceived nancial barriers to chronic disease care, the latter of which has undergone testing via focus groups and cognitive interviews [109]. Subjective social status will be assessed using the MacArthur Scale of Subjective Social Status national and community ladders [110,111]. Participants will report perceived income adequacy by answering the question: "Thinking about your total monthly income, how di cult or easy is it for you to make ends meet?" [112][113][114].
Quality of dietary intake will be assessed using the online Automated Self-Administered 24-hour Dietary Recall for Canada (ASA24-Canada-2018) whereby all participants will report all foods and beverages consumed from midnight to midnight the previous day, including location of consumption and dietary supplements [115][116][117]. The ASA24 has demonstrated good correspondence with standardized interviewer administered dietary recalls and with true intakes [116,118]. Participants will receive an unannounced email/text 2-4 days later prompting them to complete a second dietary recall to provide a more precise estimate of usual intake [119]. Dietary intake data will be used to calculate a Healthy Eating Index-2015 score from 0-100 for each participant, which provides a valid assessment of overall diet quality consistent with recommendations in the healthy food prescription pamphlet [120,121].
To reduce missing data, REDCap will be con gured to require a response prior to proceeding to the next question, although 'don't know' and 'refuse to answer' will be response options. Research Assistants will also review all completed questionnaires and will telephone participants within 24 hours to request responses to any unanswered questions.

Clinical measurements
Biochemical measurements will include quanti cation of blood glucose levels via A1C (standardized to the Diabetes Complications and Control Trial [122][123][124][125]) and fructosamine, as A1C can be unreliable for some patients and fructosamine is more sensitive to acute changes [126]. Blood lipids (total-, HDL-and LDL-cholesterol, triglycerides, apolipoprotein B), serum creatinine (to calculate estimated glomerular ltration rate), albumin-tocreatinine ratio and hemoglobin concentration will also be quanti ed. Participants will provide a urine sample to detect albuminuria. All samples will be analyzed by Alberta Precision Laboratories and DynaLIFE Medical Labs.
Physical measurements will adhere to standardized measurement protocols and will be performed a minimum of two times by trained researchers/clinicians, including weight and height to calculate BMI, waist circumference, systolic and diastolic blood pressure (using oscillometric devices approved by Hypertension Canada) and heart rate. Skin carotenoids will be assessed using Pharmanex Biophotonic Scanners as biomarkers of fruit and vegetable intake [127].
Administrative health data A1C levels and information on comorbidities and diabetes complications will be obtained from Alberta Health Services' Analytics, Data Integration, Measurement and Reporting database. The secondary outcome of need for diabetes medication/insulin will be quanti ed by monitoring changes in medication/insulin use (i.e. initiation or discontinuation), type (i.e. Metformin, Sulfonylureas, Repaglinide, DPP-4 inhibitors, GLP1 receptor agonists, SGLT2-inhibitors, Acarbose, Thiazolidinediones, Statins or other lipid-lowering agents, Renin-angiotensin aldosterone antagonists and other anti-hypertensive agents) and dosage recorded in the Pharmaceutical Information Network Database [128]. We will also collect administrative data on health events and healthcare use on an ongoing basis post-intervention to support understanding of longer-term outcomes.

Implementation delity
Implementation delity will be continuously assessed using a combination of quantitative and qualitative measures as part of the implementation study (described below).

Data monitoring and adverse events
We have not established stopping guidelines or a data monitoring committee as the intervention is low-risk. In addition, physicians will be free to adjust patients' medication regimens to achieve improved glycemia, although they will be asked to initiate or uptitrate insulin therapy only when absolutely necessary. Given that serious adverse events are unlikely, adverse events will not be monitored and no provisions have been made for compensation. Interim analyses will not be conducted. At the conclusion of the study, participants will continue to be followed by their care providers.

Data analysis
Mixed-effects linear and multinomial logistic/ordinal regression will assess group differences in outcomes at follow-up, adjusted for baseline values of each outcome, blocking variables and covariates (speci c to each outcome), with subgroup analyses by gender, rural and Indigenous status, baseline A1C (8-9%, 9-12%) and insulin use. Participant-and PCC-speci c variations will be modeled using random effects. A dose-response analysis will examine whether outcomes depend on the value of food subsidies redeemed. Missing data will be handled using full information maximum likelihood under a missing at random assumption. Analyses will be intention-to-treat using all available data.
For the primary outcome of blood glucose levels, covariates will include the following: age, gender, race/ethnicity, educational attainment, years lived in Canada, number of children living in the home, marital status, change in number of household members since baseline, access to extended health bene ts, participation in other income support programs, diabetes duration, and physical activity level.

Sensitivity analyses
Sensitivity analyses will examine outcomes among patients whose diabetes medication/insulin regimen was unchanged during the 3 months prior to the study, throughout the study period, and when patients taking insulin are excluded. We will also examine the impact of excluding patients who were started on lipid-lowering or antihypertensive therapy from models assessing impact on blood lipids and blood pressure, respectively. Additional sensitivity analyses will consider the impact on ndings when food insecurity is modelled as a continuous, rather than as a categorical outcome [129], and when an indicator of energy intake misreporting (i.e. the ratio of reported energy intake to estimated energy expenditure) is included in models assessing impact on diet quality.
We will also consider the impact on ndings when models are adjusted for changes in medication/insulin type and dosage that occurred between baseline and follow-up. We propose to use a novel scoring system that attempts to match the changes made with the expected clinical impact on A1C [130]. The following changes will be assigned 1 point (expected change in A1C of ~0.5%): less than full dose of Metformin (< 2000mg/day), Sulfonylureas (Gliclazide < 60mg/day, Glyburide < 10mg/day), or Repaglinide (<5mg/day); any dose of DPP-4 inhibitor, SGLT2-inhibitor or Acarbose; initiation of basal insulin or insulin adjustment by < 20% of total daily dose.
Two points (expected change in A1C of ~1%) will be assigned for: full dose of Metformin, Sulfonylureas or Repaglinide; any dose of GLP1 receptor agonists or Thiazolidinedione; initiation of bolus insulin or insulin adjustment of > 20% of total daily dose. Points will be attributed cumulatively for all medication/insulin changes.
Addition of medication/insulin will be scored positively and reductions scored negatively to arrive at a nal cumulative medication/insulin adjustment score for each participant.
Markov chain Monte Carlo multiple imputation, inverse probability weighting and available case analysis will be used to investigate the impact of different assumptions about missing data on estimated program impacts [131][132][133]. Pattern mixture methods models [134] will be used to explore the robustness of study ndings to the assumption that data were missing not at random [135].

Study design and objectives
A mixed methods implementation study will evaluate the RE-AIM [65,66] of the subsidized healthy food prescription program, including contextual in uences, patients' experiences in the program and perceived outcomes, reasons behind program successes and failures, implementation delity, mechanisms of impact and the likelihood of sustainability ( Table 3). The Consolidated Framework for Implementation Research (CFIR) consolidates determinants of effective implementation into ve domains (intervention characteristics, inner setting, outer setting, characteristics of individuals, implementation process), and will accordingly structure our investigation of determinants of effective implementation, including barriers and facilitators, within RE-AIM's implementation domain [136]. SITUATION: In Alberta, more than 54,000 adults are experiencing food insecurity and type 2 diabetes, including 13,600 Indigenous individuals who bear a disproportionately high burden [166][167][168][169] This group of Albertans incurs $842 million/year in healthcare costs, with a small subset of 9,600 individuals with persistent hyperglycemia incurring nearly one-quarter of these costs [20,24,165,166,170]. Nevertheless, although it is well known that food insecurity is a primary driver of acute care usage and costs, primary care providers often lack effective strategies to address it.

Implementation process
The implementation process will unfold according to the four phases and action-oriented steps in the Quality Implementation Framework (QIF) [137,138].
QIF phase 1: Initial considerations regarding the host setting 1. Stakeholder buy-in: Partnership agreements will be nalized with all stakeholders.
2. Implementation Support Team: An Implementation Support Team will be formed to administer, plan, support, monitor and evaluate implementation of the intervention.
3. Training: Study personnel will be trained in principles of equity-oriented care [82,139,140] and study procedures.
4. Assess needs, t, capacity, readiness and adaptations: Implementation strategies will be tailored by clinic using a theory-informed modi ed conjoint analysis [141] in which PCC staff will complete a questionnaire to identify potential implementation barriers and facilitators within the ve domains of CFIR. Researchers will use the CFIR-ERIC (Expert Recommendations for Implementing Change compilation) matching tool to identify strategies to mitigate the barriers and leverage the facilitators identi ed by each clinic [136,[141][142][143][144][145].
5. Pre-Implementation planning and adaptations: The Implementation Support Team will develop a preimplementation plan and timeline and will execute it, including co-developing infrastructural supports, training modules, and care pathways with PCC staff and patients. The Indigenous Advisory Board will progress relationship building with Indigenous PCCs and will work with them to adapt infrastructural supports and care pathways as required.
6. Capacity building and supportive organizational climate: PCC staff will be trained in principles of equityoriented care [82,139,140] and study procedures. Training sessions and co-design processes will enhance buy-in and readiness to change. One staff designate per organization will liaise with the Implementation Support Team weekly and will serve as an organizational champion.
7. Study planning: The Implementation Support Team will develop study protocols and procure materials for all three studies. QIF phase 2: Creating a structure for implementation 1. Implementation planning and adaptations: The Implementation Support Team will use ndings from the modi ed conjoint analysis to develop a detailed implementation plan and timeline, co-customize it with PCCs, and assign speci c roles and responsibilities. Subsidy redemption procedures will be nalized.
QIF phase 3: Implementation and ongoing implementation structure 1. Program implementation and data collection: The subsidized healthy food prescription program will be implemented and data collection for the RCT and modelling study will proceed (Figure 2).
2. Technical support and communication: The Implementation Support Team will provide ongoing support to PCC staff, including via weekly meetings with staff designates. Booster training sessions will be held when new/modi ed processes are introduced and for new staff.
3. Implementation study and feedback mechanisms: The Implementation Support Team will collect data continuously for the implementation study. Ongoing monitoring and provision of feedback to PCCs will support continuous quality improvement.
QIF phase 4: Improving future applications 1. Data analysis: The research team will analyse and integrate data from all three studies.
2. Knowledge translation and learning from experience: The research team and Advisory Boards will jointly interpret and disseminate ndings. Outcomes from a deliberative dialogue [146,147] knowledge translation event will inform sustainability planning.

Data collection
Implementation processes and outcomes related to all ve RE-AIM domains, and determinants of effective implementation within CFIR domains, will be repeatedly measured via quantitative and qualitative data collected by trained Research Assistants. Participants will receive $30 for participating in interviews and $5 for completing questionnaires.
Quantitative data Quantitative data will be collected via administrative records of the numbers and characteristics of PCCs and patients that do and do not participate (reach and adoption), the number and types of healthcare providers who receive training (implementation), healthy food prescription and subsidy allocation data (implementation), audits of supermarket subsidy transaction data (implementation), and implementation delity checklists completed by patients (implementation). Quantitative data will also encompass the aforementioned questionnaire pertaining to barriers/facilitators completed as part of the modi ed conjoint analysis. A post-intervention questionnaire will be distributed to PCC staff and patients to explore all ve RE-AIM domains, including barriers/facilitators.

Qualitative data
Qualitative data pertaining to program effectiveness, implementation (including barriers and facilitators) and maintenance will be continuously collected via weekly meetings with PCC staff liaisons and through interviews with patients. The latter will entail post-intervention semi-structured interviews with 30 purposively selected patients to understand their overall experience of participating in the program, including delity of intervention delivery, barriers/facilitators, and how the program in uenced the medical management of their T2DM. An additional 30 patients will be recruited to participate in longitudinal (pre-and post-subsidy), shop-along, think aloud qualitative interviews to understand the experience of redeeming food subsidies and how they in uenced patients' food purchasing patterns. We will also record and collect qualitative observations of Indigenous Advisory Board meetings and interview Board members to understand how the program unfolded from their perspective.

Implementation delity
From the measures summarized above, objective measures of implementation delity will include subsidy allocation data and audits of supermarket subsidy transaction data. Perceived measures of implementation delity will be reported by patients via quantitative checklists and semi-structured interviews.

Data analysis
Quantitative data Quantitative ndings pertaining to all ve RE-AIM domains will be summarized using descriptive statistics and will inform areas for subsequent in-depth qualitative exploration. We will stratify our analyses by clinic type (urban, rural, Indigenous) to examine any meaningful differences between them.

Qualitative data
Qualitative data will be coded by two trained researchers using directed content analysis [148], whereby development of an initial coding scheme for each set of interviews will be informed by RE-AIM, CFIR and other frameworks as appropriate. Concurrent data collection and analysis and regular meetings between researchers will permit iterative adjustments to the interview questions and coding schemes, and continuous evaluation of the adequacy of the samples [149]. Sampling will end when new concepts are no longer being identi ed in the data.

Data integration
Quantitative and qualitative data will be integrated during the analysis stage for the purposes of expansion (e.g. qualitative data will help to elaborate and explain quantitative ndings) and convergence (e.g. to examine whether quantitative and qualitative delity ratings correspond) [150].
Modelling study

Study design and objectives
A modelling study will estimate longer-term effectiveness of the subsidized healthy food prescription program on diabetes-related complications, resource use and costs.

Data collection
The individual-level data required as inputs for the model will be captured in the baseline and follow-up assessment phase of the RCT, as previously described. Model inputs required to estimate program impact on longer-term resource use and costs will be obtained from Alberta Health Services' Analytics, Data Integration, Measurement and Reporting database, including emergency department, inpatient, specialist, general practice and urgent care costs. The costs of the intervention will include administrative costs associated with implementing the program and the costs of food subsidies at a household level. The time to administer the program will be based on the non-research hours of the study personnel and unit costs will be obtained from the Alberta Wage and Salary Survey [151].

Data analysis
The potential longer-term health and economic impacts of the subsidized healthy food prescription program will be modelled using the validated UK Prospective Diabetes Study outcomes simulation model-2 [152]. Healthrelated model outputs will include differences between the intervention and comparison groups in cardiovascular events, amputation, blindness, renal failure, diabetic foot ulcers, and mortality over 1 year, 5 year and lifetime scenarios. Model outputs related to the economic impacts of the intervention will include the incremental difference in costs and quality adjusted life years [153] between the intervention and comparison groups, the incremental cost-effectiveness ratio and the net bene t. A budget impact analysis will explore the difference in costs (e.g. prescriptions, physician visits, hospitalizations) considering the RE-AIM domains, and return on investment from the public payer perspective over 1 year, 5 year and lifetime scenarios.

Data integration and dissemination
Data from each of the three studies will be published separately, with an additional nal publication that will integrate and synthesize ndings for all RE-AIM domains to examine the following speci c outcomes: and g) PCC staff and patient-perceived quality of infrastructural supports.
Maintenance will be assessed via: a) PCC staff-reported longer-term program feasibility, acceptability, organizational t and willingness to deliver it; b) patient-reported longer-term program acceptability and willingness to participate; c) evidence of successful program integration within PCC and supermarket work ows; d) how aspects of the program are sustained or change over time; e) PCC staff and patient suggestions for program improvement; f) reasons for drop-out; and g) commitments from study partners, including Alberta Health Services, PCCs, Alberta Blue Cross and participating supermarkets to collaborate on research to assess program scale-up and longer-term sustainability.
These fully integrated data will be disseminated via technical reports, lay summaries, infographics, policy briefs, academic publications and oral/poster presentations.

Discussion
Adults who are experiencing food insecurity cannot consume the healthy foods they require to manage their diabetes if they lack su cient funds to purchase them. However, primary care providers often lack access to resources that could assist them to alleviate their patients' experiences of food insecurity. By addressing incomerelated causes of unhealthy dietary patterns and persistent hyperglycemia, subsidized healthy food prescription programs can equip clinicians with resources that assist their patients to maintain a healthier dietary pattern.
We will investigate the RE-AIM of a subsidized healthy food prescription program for adults who are experiencing food insecurity and persistent hyperglycemia. Through a RCT, modelling and implementation studies, we will generate comprehensive, in-depth and robust data pertaining to the short-and longer-term impacts of the program on glycemia, other health-related outcomes, resource use and costs, while also providing valuable implementation data to support translation of research ndings into practice and policy. Integration of ndings from these three studies acknowledges the reality that although evidence of short-term effectiveness from RCTs is valuable, such data are on their own insu cient to promote widespread and high quality implementation. In this respect, the implementation study will be critical to unpack reasons underlying the program's successes and failures, mechanisms of impact, determinants of effective implementation, and the likelihood of sustainability.
Notably, our ndings will be immediately relevant to existing programs such as Wholesome Wave, which has been providing fruit and vegetable prescriptions to US households since 2010 [83], and can inform more speci c recommendations regarding strategies to address food insecurity in patients with diabetes [10,154]. Moreover, opening up conversations around food insecurity may also provide a gateway to address other social determinants of health that constrain patients' health potential.

Potential risks and limitations
Our literature review [43, 58, 70-72, 92, 155-161] and stakeholder engagement identi ed several potential study risks. First, a single implementation model may not be effective for all PCCs, which could adversely affect implementation delity. We will therefore use a modi ed conjoint analysis to select PCC-speci c implementation strategies, co-customize and pilot test care pathways with PCCs to ensure compatibility, and monitor implementation delity. We will also provide comprehensive training with booster sessions, continuous technical support, and will meet weekly with staff designates and promptly address any concerns. Ongoing engagement, the support of operational leaders within the healthcare system and over-recruitment of PCCs will help to mitigate against and accommodate drop-out, should it occur. Primary risks at the patient level include failure to redeem subsidies or drop-out. To mitigate against these risks, participants will receive email reminders to redeem their food subsidies and will have access to a study help-line for support. Moreover, both groups will receive $100 at baseline and follow-up as compensation, which will also serve as a recruitment incentive.
Participants' anti-hyperglycemic regimen may be intensi ed by their care provider during the observation period, which can exert a signi cant impact on A1C. The tendency to intensi cation and deintensi cation should be balanced between groups owing to randomization, nevertheless, we will monitor medication/insulin type/dosage for use as adjustment variables in sensitivity analyses. In addition, need for diabetes medication/insulin has been included as a secondary outcome, whereby deintensi cation will be considered a positive, and intensi cation a negative outcome.

Conclusions
Subsidized healthy food prescription programs hold tremendous potential to improve health and reduce healthcare costs given that 20% of global morbidity and mortality is attributable to poor diet quality [162], and that food insecurity is a strong predictor of high-cost healthcare use [26,163]. In addition, the bene ts of such programs extend to all household members, including children, for whom positive health outcomes may accrue across the lifecourse. In an era of unsustainable increases in healthcare costs, the healthcare system ignores the socioeconomic needs of patients at its peril [159]. Failure to effectively address food insecurity among adults with T2DM portends unsustainable escalations in healthcare usage and costs [29,32,[36][37][38][163][164][165]. We have proposed a comprehensive investigation of the RE-AIM of a subsidized healthy food prescription program.
Ultimately, study ndings will show whether a small upstream investment in a subsidized healthy food prescription program may avert substantially higher healthcare costs to treat diabetes complications after they emerge.

Availability of data and materials
The datasets analyzed in this study will be available from the corresponding author on reasonable request.   Figure 1 Healthy food prescription program logic model