The Impact of Produce Prescriptions on Diet, Food Security and Cardiometabolic Health Outcomes: A Multi-Site Evaluation of Nine Produce Prescription Programs in the U.S.

Abstract

Background:

Produce prescriptions may improve cardiometabolic health by increasing fruit and vegetable (F&V) consumption and food insecurity yet impacts on clinical outcomes and health status have not been evaluated in large, multi-site evaluations.

Methods:

This multi-site, pre- and post-evaluation used individual-level data from 22 produce prescription locations in 12 U.S. states from 2014–2020. No programs were previously evaluated. The study included 3,881 individuals (2,064 adults aged 18+ years and 1817 children aged 2–17 years) with, or at risk for, poor cardiometabolic health recruited from clinics serving low-income neighborhoods. Programs provided financial incentives to purchase F&V at grocery stores or farmers markets (median = $63/month; duration 4 – 10 months). Surveys assessed F&V intake, food security, and self-reported health; HbA1c, BP, BMI, and BMI z-score were measured at clinics. Adjusted, multilevel mixed models accounted for clustering by program.

Results:

After a median participation of 6.0 months, F&V intake increased by 0.85 (95% CI: 0.68, 1.02) and 0.26 (95% CI: 0.06, 0.45) cups per day among adults and children, respectively. The odds of being food-insecure dropped by one-third (odds ratio [OR]: 0.63 [0.52, 0.76]) and odds of improving one level in self-reported health status increased for adults (OR: 1.62 [1.30, 2.02]) and children (OR: 2.37 [1.70, 3.31]). Among adults with HbA1c >=6.5%, HbA1c declined by −0.29 percentage points (−0.42, −0.16); among adults with hypertension, systolic and diastolic BP declined by −8.38 mmHg (−10.13, −6.62) and −4.94 mmHg (−5.96, −3.92); and among adults with overweight or obesity, BMI decreased by −0.36 kg/m² (−0.64, −0.09). Child BMI z-score did not change −0.01 (−0.06, 0.04).

Conclusions:

In this large, multi-site evaluation, produce prescriptions were associated with significant improvements in F&V intake, food security, and health status for adults and children, and clinically relevant improvements in HbA1c, BP, and BMI for adults with poor cardiometabolic health.

Introduction

One of every five deaths worldwide is attributable to suboptimal diet, more than any other risk factor including tobacco.¹ Suboptimal diet in the U.S. exerts a tremendous health burden, with more than 300,000 annual deaths from cardiovascular disease (CVD) and diabetes.^2,3 Most Americans do not meet guidelines for a healthy diet,⁴ while population disparities in diet quality and health outcomes persist or have worsened among marginalized racial, ethnic, and low-income groups.^5,6 These challenges further intersect with food insecurity, which is strongly associated with poor health outcomes^7,8 and higher healthcare utilization and costs.^9,10

Until recently, healthcare providers had few tools to address patient food and nutrition insecurity. However, rapidly evolving interest among clinicians, payers, patients, and policymakers in “food is medicine” interventions like produce prescriptions offer promising mechanisms to improve nutrition and health outcomes.¹¹ Produce prescriptions provide free or discounted produce, covered partly or fully by healthcare payers, to patients with or at risk for diet-related chronic diseases and often concomitant food insecurity.^11–13 These treatments recognize that financial incentives may be required to improve nutritional intake, especially for low-income populations. Early produce prescriptions were typically implemented at farmers’ markets with more recent programs partnering with retail grocery to increase choice, convenience, and year-round access.¹⁴

A growing body of evidence suggests that produce prescriptions increase F&V intake, reduce food insecurity, and improve quality of life.^11–13,15 Yet, existing studies are generally small pilots that evaluate a single program and could be limited by publication bias, with only studies identifying positive effects being reported. Critically, evaluations on clinical biomarkers of cardiometabolic health such as glycated hemoglobin (HbA1c), blood pressure (BP), and body mass index (BMI) for patients with diet-related illness remain limited, with varying results.^12,13,15 Thus, more robust analyses from larger samples, including pooling data across multiple programs to increase statistical power and generalizability, are critical to better understand the impact of produce prescriptions on health outcomes.

To address these gaps in knowledge and inform the design of ongoing clinical and policy efforts, we evaluated the impact of produce prescriptions programs on F&V intake, food insecurity, self-reported health status, and clinical outcomes including HbA1c, BP, and BMI in a pooled analysis of nine programs. Findings are timely and relevant for both clinical practice and policy, including produce prescription funding through the 2018 Farm Bill,¹⁶ optional coverage of produce prescriptions by Medicare Advantage plans,¹¹ and ongoing Medicaid 1115 and 1915 waivers creating produce prescription pilots in several U.S. states.^17,18

Methods

Population and Setting

Study participants included 1,817 children and 2,064 adults who participated in one of nine produce prescription programs at 22 sites across 12 U.S. states from 2014–2020. Eligibility criteria across programs included being at risk for poor cardiometabolic health for adults (e.g., overweight/obesity, diabetes, or elevated blood pressure) and having overweight / obesity for children, in addition to being food-insecure or being recruited from a health center serving a predominantly low-income neighborhood (see Table 1 for detailed descriptions by program). Participants were referred by healthcare providers (physicians, medical assistants, nurses, or other members of the care team) to receive produce prescriptions operated by Wholesome Wave,¹⁹ an organization dedicated to curbing the national burden of diet-related disease by improving affordability and access to healthy fruits and vegetables across the country. We included all produce prescription programs operated by Wholesome Wave that measured at least one biomarker in their programmatic data collection (ie, BMI / BMI z-score, HbA1c, or blood pressure). Programs were designed by Wholesome Wave in partnership with healthcare clinics to improve participant FV intake, food insecurity, and cardiometabolic health outcomes. Wholesome Wave collected data specifically to test these hypotheses but did not have the capacity to conduct formal analyses until partnering with this research team. None of the included programs have been the subject of previous research nor evaluation. This study was reviewed by the Tufts Health Sciences IRB and determined not human subjects research as the analysis used retrospective, deidentified data provided from the programs. The data that support the findings of this study are available from the corresponding author upon reasonable request.

Table 1:

Description of Participant Eligibility and Program Structure of Pooled Wholesome Wave Produce Prescription Programs

Program Name	Participants Eligibility		Incentive and Program Structure
	Population	Eligibility	US State	Healthcare Partner*	Monthly Produce Prescription Dollars	Duration (months)	Setting	Delivery Model
2016 Los Angeles Target FVRx Program	Pediatric	Children age 2–18 at risk for developing chronic disease as determined by their healthcare provider based on being overweight or obese and/or having poor nutrition or food insecurity.	CA	Eisner Health	$180 for a household of 2 $210 for a household of 3 $270 for a household of 4 $300 for a household of 5+	6	Grocery (Target) and farmers’ markets	Voucher
2017 Three-State Target FVRx Program	Pediatric	Children age 2–18 at risk for developing chronic disease as determined by their healthcare provider based on being overweight or obese and/or having poor nutrition or food insecurity.	CA, TX, FL	Eisner Health (CA), Hermann Memorial (TX), Jessie Trice Health Center (FL)	$90–120 for a household of 2 $150–210 for a household of 3 $180–270 for a household of 4 $270–300 for a household of 5+	6–9	Grocery (Target) and farmers’ markets	Voucher
2018–2019 Chobani Reward Card Program	Pediatric	Children age 2–18 with household food insecurity. Overweight or obesity was not a requirement for eligibility.	NY, ID	Chenango Memorial Health (NY), Family Health Services (ID)	$60	6–7	Grocery	Card
2014 Blue Cross Blue Shield Program	Adult	Adults with a diagnosis of diabetes in Minneapolis, MN recruited from a community health center serving a low-income neighborhood.	MN	Northpoint Health and Wellness, CentraCare Health System	$30 per household member	4	Grocery, farmers market	Voucher
2019 Humana Reloadable Gift Card Program	Adult	Eligibility criteria required food insecurity and at least one chronic disease. Participants were referred from a Federally Qualified Health Center.	FL	Community Health Centers of Pinellas	$15	8	Grocery (Walmart)	Card
2019 Ohio Department of Health Produce Voucher Program	Adult	Adults with diabetes or pre-diabetes. Although food insecurity or low-income was not a criteria, participants were recruited from health centers in low-income neighborhoods.	OH	Community Health and Wellness Partners of Logan County, Hopewell Health Centers of Athens County	$30 per household member up to four members	10	Grocery, farmers’ market	Voucher
2020 Ohio Department of Health Produce Voucher Program	Adult	Adults with diabetes or pre-diabetes. Although food insecurity or low-income was not a criteria, participants were recruited from health centers in low-income neighborhoods.	OH	Community Health and Wellness Partners of Logan County, Hopewell Health Centers of Athens County	$90	10	Grocery, farmers’ market	Voucher
2020 Hartford Healthcare Funded Program	Adult	HbA1c greater than 6.5% and receiving care at one of two Hartford Healthcare clinics. While food insecurity and income were not eligibility requirements, participants were recruited from zip codes with higher levels of poverty and many of the participants are on Medicaid.	CT	Hartford Healthcare	$60	6	Grocery	Voucher
2020 Multi-State, Weight Watchers Produce Rx Program	Adult	Adults with overweight or obesity referred by health centers in 9 states.	FL, NC, GA, MN, CO, CA, TX, NY	Broward Community Health (FL), Grady Health (GA), UNC Charlotte (NC), CEAP Health Partner (MN), Behavioral Health Services (CA), Mental Health Center of Denver (CO), Weight Watchers only (TX, NY)	$50	6–9	Grocery (Walmart) with some sites using home delivery or food panty	Card, with one site using vouchers and one site using an online app.

^*Some healthcare partners recruited participants from multiple clinics within their system.

Programs that occurred in years 2014 – 2017 had two to three in-person nutrition education and cooking demonstration events at which participants received produce prescriptions for the next two to three months. Starting in 2018, Wholesome Wave dropped the attendance of classes as a condition to receive the next round of produce prescriptions, and instead offered optional nutrition education in partnership with universities through the Expanded Food & Nutrition Education Program or through the clinic’s own nutritional programming.

Produce Prescription Intervention

After referral by a healthcare provider, participants were enrolled at the clinic or at a clinic-hosted, community event with nutrition education classes. Enrolled participants received financial incentives (paper vouchers or electronic cards) ranging from $15 - $300 / month (median = $63 / month) to purchase F&V at food retailers such as grocery stores and farmers markets (Table 1). Two of three pediatric programs and two of four adult programs scaled the produce prescription based on household size, although the dollars per household member differed by program (Table 1). Program duration ranged from 4 to 10 months. In addition to financial incentives, all pediatric and adult programs provided in-person or online nutrition education classes. Education varied from in-person, online, individual instruction and/or group lessons as well as tours of grocery stores.

Outcomes

Primary outcomes included pre- and post-changes in F&V consumption (cups/day, continuous), household food insecurity (binary), HbA1c (%) among adults with baseline values >6.5%, and systolic and diastolic BP (mmHg, continuous) among patients with baseline values ≥130mmHg (systolic) and >80 mmHg (diastolic). Secondary outcomes included changes in self-reported health status (5-point Likert scale, ordinal) body mass index (BMI) (kg/m², continuous) among adults with overweight/obesity (BMI ≥25 kg/m²) and BMI z-scores (continuous) among children with overweight/obesity (BMI for sex and age ≥ 85^th percentile). We hypothesized that among the biomarker outcomes, HbA1c and BP would be most responsive to dietary changes over a several month program period, and thus were included as primary outcomes along with F&V intake and food insecurity. F&V intake and food insecurity were included because they are the targeted pathways throughout which improvements in health outcomes might occur.

Outcomes were measured at baseline and after program participation. Consumption of F&V was assessed using semi-quantitative dietary questionnaires. One program used a 10-item F&V screener that assesses types of fruits (e.g., fruits, fruit juices) and vegetables (e.g., leafy greens, non-fried vegetables) in frequency (e.g., 3–4 times /week).²⁰ The remaining programs assessed the total daily intake of fruits and vegetables, separately, in frequency only (e.g., 3 times per day), in both frequency and portion size (e.g., 3 times per day, 0.5 cups each), or in servings per day (e.g., 3 servings per day). To harmonize data, we estimated the total intake of F&V in cups per day by combining data on frequency and portion size whenever available or assuming standard portion sizes from USDA MyPlate when portion size was not asked.²¹ Program surveys did not assess any other components of the diet.

Food security was assessed using the USDA six-item food security module²² or the validated Hunger Vital Sign.^23,24 These responses were harmonized to estimate presence or absence of household food insecurity as a binary measure.

Self-reported health status was assessed in a 5-point Likert scale (poor, fair, good, very good, excellent) using the Center for Disease Control and Prevention’s Healthy Days module²⁵ This measure has been validated in multiple studies as highly predictive of future health outcomes and total mortality.²⁶ Levels of HbA1c in blood were measured in standardized healthcare system laboratories; blood pressure, weight, and height were measured in the clinic as part of routine medical care or during a special visit to the clinic as part of program participation.

Covariates

Wholesome Wave registration surveys self-reported participants’ age (continuous), biological sex (male/female), race / ethnicity (white, non-Hispanic black, Hispanic, other), household size (number of people), and for child programs only; current household enrollment in the Supplemental Nutrition Assistance Program (SNAP) and Special Nutrition Program for Women, Infants and Children (WIC), health insurance coverage (Medicaid/CHIP, private, uninsured, other) and parent / caregiver employment (full time, part time, unemployed, other). While some programs measured additional covariates, we used all available, shared covariates measured across all programs in the analysis.

Statistical Methods

To assess changes in outcomes associated with program participation, we used hierarchical generalized linear models that account for clustering of repeated measurements (i.e., pre and post) within the same participants and clustering of participants within program locations. Both the pre- and post-measurements of the outcome were modeled as the dependent variable. We included a time variable as one of the independent variables (pre-values were time=0; post values were time=1) and therefore regression coefficients associated with time indicated outcome changes from pre- to post-program participation. For continuous outcome variables that approximated a normal distribution such as F&V intake, HbA1c, BP, BMI, BMI z-score, the link function was specified as identity; and for the binary outcome variable food insecurity the link function was specified as logit. Finally, for the ordinal outcome variable self-reported health status we used an ordered logistic model with a cumulative link function in which the odds are described as P(Y<=j|X=x)/(1-P(Y<=j|X=x) with Y being health status level, j= level 1, 2, 3, 4, 5 within the Likert scale, X is produce prescription participation, and the odds ratio is the ratio of the above odds when x=1 (after produce prescription program) over x=0 (before produce prescription program). Analyses were conducted in adult and child populations separately, while food insecurity was analyzed at the household level with only one measurement per household. Fixed effects included age, sex, race/ethnicity, and household size for outcomes assessed in adults or at the household level. For outcomes assessed in children, fixed effects additionally included SNAP enrollment, WIC enrollment, health insurance coverage, and parent / caregiver employment. For all models and outcomes, random intercepts were modeled for person and program location.

Due to high rates of missing data in some programs, we assessed patterns of missingness by comparing baseline characteristics and outcomes between participants with complete versus missing data at program end. Comparisons show small standardized mean differences of less than 0.2 for all baseline primary outcome measurements between those with complete versus missing data (Table S1). No large differences in participant characteristics, defined as a standardized mean difference greater than 0.8, between those with complete versus missing data were identified (Table S2), with nearly all comparisons reporting standardized mean differences of less than 0.5. To account for missing data, our primary analyses employ multiple imputation. Multiple imputation was conducted separately for each outcome and incorporated all available variables that might help satisfy the missing at random assumption²⁷ including shared demographic covariates; baseline values of F&V intake, health status, food insecurity, and health outcomes; and indicator variables to account for clustering by program location.²⁸ We used 30 imputations as the fraction of missingness information was less than 30 for all outcomes. To check the robustness of our multiple imputation, we also report complete-case analyses. We also conducted sensitivity analyses excluding program locations with < 50% missingness of endpoint surveys.

A final sensitivity analysis assessed program impacts on HbA1c among participants with baseline HbA1c ≥8.0%, on blood pressure among participants with baseline systolic BP ≥140 mm Hg or diastolic BP > 90 mm Hg, and on BMI among participants who were overweight or obese at baseline (BMI ≥ 30 kg/m² for adults; BMI for age and sex ≥ 95^th percentile for children). Stratified analyses were conducted to explore whether outcomes differed by (1) timing relative to the COVID-19 pandemic (whether the produce prescription program was completed prior to March 2020 or extended into the pandemic); and (2) participant characteristics including age (2–17, 18–64, 65+) , sex (as a biological variable), and race/ethnicity (as a self-reported, sociocultural construct which may reflect experiences of discrimination that worsen health), plus household size and household SNAP enrollment (characteristics that may impact a household’s food needs).

Analyses were conducted in in Stata (version SE 17.0) by one investigator (K.H.) and replicated in SAS (version 9.4) by another (M.D). Due to testing of five primary outcomes, we used a Bonferroni corrected alpha=0.01 for statistical significance for primary outcomes.

Results

Participant Characteristics

Adults had a mean (SD) age of 54.4 (14.8) years, 70.7% were female, 29.8% were non-Hispanic White, and 45.1% were Non-Hispanic Black. Children had a mean age of 9.2 (4.2) years, 51.4% were female, 75.5% were Hispanic, and 62.7% were enrolled in SNAP (Table 2). At baseline, mean (SD) F&V intake was 2.7 (1.9) cups/day among adults, and 3.4 (2.8) cups/day among children; and 56.3% of all households experienced food insecurity (Table 2).

Table 2:

Baseline Characteristics of 3,881 Participants in 9 Produce Prescription Programs across 22 US Program Locations

Characteristic*	Adult Population N=2064	Child Population N=1817
Age, years, mean (SD)	54.4 (14.8)	9.27 (4.2)
Female, N (%)	1,459 (70.7)	929 (51.4)
Race/Ethnicity, N (%)	Non-Hispanic White: 615 (29.8) Non-Hispanic Black: 929 (45.1) Hispanic: 441 (21.4) Other: 75 (3.6)	Non-Hispanic White: 159 (9.2) Non-Hispanic Black: 227 (13.1) Hispanic: 1,308 (75.5) Other: 39 (2.3)
Household Size, number of people, mean (SD)	2.7 (1.6)	4.6 (1.4)
Household SNAP Enrollment, N (%)		1,073 (62.7)
Household WIC Enrollment, N (%)		1,428 (82.6)
Insurance Status, N (%)		Medicaid, CHIP: 1,358 (79.0) Private: 67 (3.9) Uninsured: 196 (11.4) Other: 95 (5.5)
Parent / Caregiver Employment, N (%)		Full-time (40 hrs/wk): 318 (18.5) Part-time (<40 hrs/wk): 416 (24.3) Unemployed: 652 (38.0) Other: 329 (19.2)
Baseline Outcome Measurements
F&V Intake, cups per day, mean (SD)	2.7 (1.9)	3.4 (2.8)
Food Insecurity, N (%)	1,305 (65.8)	737 (49.4)
HBA1c, %, mean (SD)	8.7 (1.8)
Systolic Blood Pressure, mm Hg, mean (SD)	144.5 (13.6)
Diastolic Blood Pressure, mm Hg, mean (SD)	87.5 (7.4)
Body Mass Index, Kg/m2, mean (SD)	37.3 (8.3)
Body Mass Index z-score for sex and age, z-score, mean (SD)		1.9 (0.5)
Health Status, N (%)	Excellent: 59 (3.3) Very Good: 214 (12.0) Good: 685 (38.4) Fair: 675 (37.8) Poor: 152 (8.5)	Excellent: 117 (15.4) Very Good: 189 (24.9) Good: 325 (42.8) Fair: 99 (13.0) Poor: 30 (3.9)

Abbreviations: SNAP = Supplemental Nutrition Assistance Program, WIC = Special Supplemental Nutrition Assistance Program for Women, Infants, and Children, HbA1c = glycated hemoglobin

^*Sociodemographic variables are listed when they were measured in all adult or child programs (ie, SNAP, WIC, insurance status and employment were not assessed in all adult programs). Sample size for some sociodemographic variables do not add up to the number of participants enrolled as participants skipped survey questions in the baseline survey; sample size for participants with baseline values in outcome measures was 1290 (adults F&V intake), 1745 (children F&V intake), 1,785 (adults self-reported health), 760 (children self-reported health), 1,982 (adults food insecurity), 1491 (children food insecurity children), 741 (HbA1c), 419 (SBP), and 439 (DBP), 762, (adults BMI), 953 (children BMI z-score). No program measured all outcomes.

The median monthly produce prescription received by adults was $43 (IQR: $31 - $60) with actual, observed program participation lasting for a mean (SD) of 6.4 (1.7) months, ranging from 1.0 – 10.0 months. The average adult participant spent 73.1% of their produce prescription dollars during program enrollment (the remainder was unspent). The median monthly amount received in pediatric programs was $112 (IQR: $85 - $133) with actual, observed program participation lasting for a mean (SD) 5.4 (1.9) months, ranging 2.0 – 9.4 months. The average child participant spent 77.1% of their produce prescription dollars during program enrollment.

Outcomes in Adults

After adjusting for covariates, F&V intake increased by 0.85 cups per day (95% CI: 0.68, 1.02) (Table 3) among adults after program participation. Among all households, the odds of being food-insecure decreased by one-third after program participation (odds ratio [OR]: 0.63 [0.52, 0.76]). Assessing clinical biomarkers of cardiometabolic health, HbA1c declined by −0.29 percentage points (95% CI: −0.42, −0.16) among those with HbA1c >=6.5% and by −0.58 percentage points (−0.78, −0.38) among those with HbA1c>=8.0%. Systolic and diastolic BP declined by −8.38 mmHg (−10.13, −6.62) and −4.94 mmHg (−5.96, −3.92), respectively, among those with stage I and II hypertension at baseline; and by −11.10 mmHg (−13.84, −8.37) and −9.43 mmHg (−11.70, −7.16) among those with stage II hypertension. BMI also significantly improved with a reduction of −0.36 kg/m² (−0.64, −0.09) among adults with overweight or obesity at baseline, and −0.52 kg/m² (−0.85, −0.19) among adults with obesity. Finally, program participation was associated with an increase in self-reported health status (OR: 1.6 [1.30, 2.02]). Given health status is measured in an ordinal scale with 1 indicating “poor” health and 5 indicating “excellent” health, compared to the baseline, participants experienced a 60% increase in odds of improving one level in their health status (e.g., from poor to fair, fair to good, good to very good, or very good to excellent) at the end of the program. All adult primary outcomes were statistically significant at the 0.01 significance level after adjusting for covariates and after the Bonferroni correction.

Table 3:

Changes in F&V Intake, Food Insecurity, and Health Outcomes Associated with Produce Prescription Program Participation

Outcome	Population at Baseline	N	Estimate (95%CI)* endpoint compared to baseline	P-Value
All Adults
Fruit and Vegetable Intake, cups per day (continuous)	Adults, including parents / caregivers from pediatric programs	1,150	0.85 (0.68, 1.02) †	<0.0001
Health Status, 5-point Likert scale ranging from “poor” to “excellent” (ordinal)	Adults, including parents / caregivers from pediatric programs	1,772	1.62 (1.30, 2.02)	<0.0001
All Children
Fruit and Vegetable Intake, cups per day (continuous)	Children (all age 2–18)	1,745	0.26 (0.06, 0.45) †	0.009
Health Status, 5-point Likert scale ranging from “poor” to “excellent” (ordinal)	Children (all age 2–18)	760	2.37 (1.70, 3.31)	<0.0001
All Households
Food Insecurity, Yes or No (binary)‡	Households	3,428	0.63 (0.52, 0.76) †	0.002
Adults with Poor Cardiometabolic Health
HbA1c, % (continuous)	Adults with HbA1c ≥ 6.5%	709	−0.29 (−0.42, −0.16) †	0.008
Systolic Blood Pressure, mmHg (continuous)	Adults with systolic BP ≥ 130 mmHg	348	−8.38 (−10.13, −6.62) †	<0.0001
Diastolic Blood Pressure, mmHg (continuous)	Adults with diastolic BP ≥ 80 mmHg	368	−4.94 (−5.96, −3.92) †	<0.0001
Body Mass Index, kg/m2 (continuous)	Adults with BMI ≥ 25kg/m2	635	−0.36 (−0.64, −0.09)	0.01
Adults with Very Poor Cardiometabolic Health
HbA1c, %(continuous)	Adults with HbA1c ≥ 8.0%	418	−0.58 (−0.78, −0.38)	<0.0001
Systolic Blood Pressure, mmHg (continuous)	Adults with systolic BP ≥ 140 mmHg	186	−11.10 (−13.84, −8.37)	<0.0001
Diastolic Blood Pressure, mmHg (continuous)	Adults with diastolic BP ≥ 90 mmHg +	108	−9.43 (−11.70, −7.16)	<0.0001
Body Mass Index, kg/m2 (continuous)	Adults with BMI ≥ 30kg/m2	520	−0.52 (−0.85, −0.19)	0.002
Children with Overweight and Obesity
Body Mass Index z-score for age and sex, z-score (continuous)	Children age 2–18 with BMI percentile for age and sex ≥ 85th percentile	953	−0.01 (−0.06, 0.04)	0.71
Children with Obesity
Body Mass Index z-score for age and sex, z-score (continuous)	Children age 2–18 with BMI percentile for age and sex ≥ 90th percentile	674	−0.03 (−0.09, 0.04)	0.37

Abbreviations: HbA1c = glycated hemoglobin

^*Adult and household analyses were adjusted for age, sex, race/ethnicity, and household size. Pediatric analyses were adjusted for age, sex, race/ethnicity, household size, SNAP enrollment, WIC enrollment, insurance coverage, and parent / caregiver employment status. Estimates for outcomes labeled as continuous indicate the change in mean; estimates for food insecurity indicate odds ratio of outcome = Yes vs No (when the OR is less than 1, it indicates decreased odds of being food insecure after program participation); and estimates for health status indicate odds ratio of advancing to a higher level in the ordinal scale (when an OR is greater than 1, it means program participation was associated with an increase in the odds of a one-level increase in self-reported health status after program participation as compared to baseline, e.g., from poor to fair, fair to good, good to very good, or very good to excellent).

^†p<0.001 for the 5 co-primary outcomes, all passing the Bonferroni-corrected alpha=0.01 for statistical significance.

^‡Two programs assessed food insecurity “within the past 12 months”, meaning the endpoint surveys captured the entire program period plus several months prior to enrollment.

Hierarchal generalized linear models were used to evaluate changes in each outcome associated with program participation, with random effects for program location and for within-person repeat measures (i.e., pre and post). For continuous response variables the link function was identity; for food insecurity the link function was logit; and for health status the link function was cumulative logit. Results are from multiple imputation analyses with 30 imputations.

Outcomes in Children

After adjusting for covariates, F&V intake among children increased by 0.26 cups per day (95% CI: 0.06, 0.45) (Table 3) after program participation. Household food insecurity also improved, as described above. Self-reported health significantly improved, with a doubling of the odds of improving one level from baseline within the self-reported, 5-point Likert scale for health status after program participation (OR: 2.37 [1.70, 3.31]). No significant changes in BMI z-score were observed (−0.01 [−0.06, 0.04]) after program participation for children.

Sensitivity Analyses

Missing data varied by program and outcome, with biomarker outcomes having the most complete endpoint measurements (Table S3). Missing data occurred most frequently when participants were unable to complete final surveys. Complete-case analyses were similar to the main findings using multiple imputation (Table S4), after adjusting for covariates. After dropping program locations with over 50% missing of endpoint data (did not affect HbA1c, SBP, DBP, or BMI analyses), analyses were also similar to the main findings (Table S5). The exception was that child F&V intake no longer met our Bonferroni correction for statistical significance (p=0.04). Exploratory stratified analyses identified minimal evidence of differential effects by participant subgroups (Table S6–12). Exceptions were signals around the impact of implementation during COVID-19, with greater improvements observed in F&V intake and HbA1c after program participation within adult programs that occurred prior to COVID-19 (p for interaction all <0.05).

Discussion

In this multi-site, participant-level, pooled pre- and post- evaluation of nine produce prescription programs in 12 U.S. states, we identified statistically significant and clinically meaningful improvements in F&V intake, household food insecurity, HbA1c, BP, BMI and self-reported health status among adults; and in F&V intake, household food insecurity, and self-reported health status among children. There was no significant change in children’s BMI z-score. This investigation provides the largest evaluation and strongest evidence to-date that produce prescriptions are associated with improved nutrition, food security, and self-perceived health among adults and children as well as key health outcomes among adults with suboptimal cardiometabolic health.

Our findings build upon and extend the results of previous studies on produce prescriptions. A recent review of produce prescription studies found that 21 of 22 published studies identified improvements in F&V intake compared with baseline.¹⁵ A pooled meta-analysis of five published produce prescription studies, totaling 1,039 adults, found a 0.8 cups/day (22%) increase in F&V consumption compared with baseline.²⁹ We identified a similar 0.85 cups/day increase among adults – a meaningful improvement given that less than 10% of American adults meet national F&V recommendations to consume 5–6 cups/day,³⁰ with a mean national intake of 2.3 cups/day.³¹ The USDA estimates that an individual would need to spend $63 – 78 per month to meet the recommended daily F&V intake,³¹ and the median produce prescription value of $48 per month for adults and $112 for children across programs in this analysis would have provided financial support to reasonably increase F&V intake.

Two reviews on interventions addressing food and nutrition insecurity in healthcare settings found few studies that assessed participant food insecurity, with only five studies using validated screeners, three of which found improvements in household food insecurity.^13,32 Given the observed improvements in both diet quality and food insecurity in our study, produce prescriptions appear to advance nutrition security, defined as having consistent access, availability, and affordability of foods that promote well-being and prevent and treat disease. This concept, embraced and prioritized by USDA,³³ highlights the central importance of access to not only calories, but to healthy foods and good nutrition.

Prior interventional studies of diet patterns support causal effects of increased F&V intake on HbA1c, BP, and BMI.^34–36 The health benefits of F&V appear to be derived from a complex set of dietary fiber (prebiotics), micronutrients, and phytochemicals, as well as potential replacement of less healthful foods in the diet.³⁷ Relatively few prior produce prescription studies have evaluated HbA1c, BP, or BMI, with mixed and inconsistent results often from pilot studies with smaller samples.^13,15 A meta-analysis of prior published studies identified reductions in HbA1c (−0.8% (95%CI: −1.6, −0.1); N=5 studies, 1,064 adults) and BMI (−0.6kg/m² (95%CI: −2.8,−0.3); N=3 studies, 215 adults).²⁹ By pooling individual-level data across 22 sites in 12 states, our study provides more robust evidence that produce prescriptions could be a promising component of clinical care for food insecure and/or low-income patients with poor cardiometabolic health.

While produce prescriptions were associated with improved child F&V intake and health status, our analysis of child BMI z-score did not reveal significant change. A review of childhood obesity prevention programs found that evidence for child weight loss was strongest for combined diet-physical activity interventions delivered in schools with both home and community components.³⁸ Produce prescriptions may need to be of longer duration or combined with additional components to impact child BMI-z-score. Nonetheless, the significant increase in F&V intake, reduction in household food insecurity, and improved self-reported health among children all support potential for meaningful impact child wellbeing and longer-term health outcomes.

Among the adult programs, there is variation in produce prescription program designs within our dataset and published research,^13,15 and various programmatic factors may impact efficacy. For example, some programs increased benefits for each additional household member, and research suggests diminishing returns on F&V intake within larger households when the produce prescription value is not scaled by household size.³⁹ Some programs provided electronic cards, which may reduce stigma and increase consumer convenience.⁴⁰ The frequency, intensity, and quality of nutrition education likely also impacts efficacy, in addition to the selection of retail grocery or farmer’s markets partners (i.e., multiple store locations, year-round availability, and convenient hours will increase accessibility). Future research will need to tease out which program designs are most likely to impact health outcomes.

Implications for Clinical Care

Low F&V intake is an established risk factor for multiple poor health outcomes, including higher risk of coronary heart disease, stroke, adiposity, and cancer.^3,41 In the U.S., more than 100,000 annual deaths from cardiometabolic disease are estimated to be attributable to insufficient F&V intake.² The estimated U.S. costs of obesity related illness, including diabetes, cardiovascular diseases, and obesity-related cancers are approximately $1.7 trillion per year, equal to 9.3% of GDP.⁴² By providing a structured intervention within the healthcare system, produce prescriptions could help combat these trends, as the magnitudes of the observed improvements in F&V intake, HbA1c, BP, and BMI in our study were each clinically meaningful.

The large observed improvements in food insecurity also have clinical implications. Food insecurity is strongly associated with poor health outcomes^7,8 and higher healthcare costs.^9,10 Food-insecure individuals often employ rational coping strategies that could harm disease management, such as underusing medications⁴³ and choosing cheaper, unhealthful foods due to costs.⁴⁴ Food insecurity is also associated with worse glycemic control among individuals with diabetes⁴⁵ and Emergency Department admissions increase among low-income individuals with diabetes during the last week of the month, when food and finances are often in shorter supply.⁴⁶ With the Covid-19 pandemic, stark disparities in household food insecurity have occurred by race/ethnicity, with 7.1% of White, non-Hispanic households experiencing food insecurity compared to 21.7% percent of Black, non-Hispanic households and 17.2% of Hispanic households.⁴⁷ Our findings suggest that produce prescriptions substantially improve food security and may offset some of the adverse health inequities associated with food insecurity.

Implications for Policy

Historically, produce prescriptions have been primarily operated by community-based organizations supported by grants and donations, however, U.S. healthcare policy interest is rapidly growing, including payment for produce prescriptions within limited scopes of Medicare and Medicaid.¹² In 2020 Medicare Advantage plans began utilizing new Special Supplemental Benefits for the Chronically Ill,⁴⁸ which allow the optional coverage produce prescriptions for select members. The Agriculture Improvement Act of 2018 established the Gus Schumacher Nutrition Incentive Program (GusNIP),¹⁶ currently providing about $5 million per year in competitive grants for produce prescription implementation in healthcare settings. At the state-level, Section 1115 and 1915 demonstration waivers have allowed Medicaid programs in Massachusetts, North Carolina, Oregon, and California to pilot nutrition-focused interventions, including produce prescriptions, for high-risk patients.^17,18,49,50 Finally, the 2022 White House Conference on Hunger, Nutrition and Health highlighted produce prescriptions as a priority for policy action within the healthcare system.⁵¹

Private payers and insurance are also showing interest in produce prescriptions. Since 2019, John Hancock life insurance has offered customers premium discounts and monthly cash rebates to purchase fruits and vegetables in a national network of supermarkets.⁵² Also in 2019, Kaiser Permanente announced Food for Life, a new priority to eliminate food insecurity among its 13 million members, which includes assistance in SNAP enrollment and ongoing randomized controlled trials of produce prescriptions and medically tailored meals.⁵³ However, given the overall limited coverage of produce prescriptions nationally, this treatment is unavailable to most Americans whom might benefit. Our new findings support the need for further testing and scaling of produce prescriptions, including in clinical settings and controlled trials, across a variety of patient populations.

Strengths

To our knowledge, this is the largest evaluation of produce prescriptions and health outcomes to date, increasing statistical power to detect impacts on dietary intake, health outcomes, and other clinically relevant endpoints. We pooled patient-level data from a broad range of locations, program designs, and healthcare partners, reducing the potential for publication bias and augmenting generalizability. We evaluated both adults and children, providing a range of findings across the life course. Findings for important, self-reported outcomes were complemented by supportive results based on objectively measured clinical biomarkers. Programs documented robust participant engagement as reflected by the mean redemption rates of ~75% of total available dollars over the 4 – 10-month intervention period. Findings were robust to several sensitivity analyses.

Limitations

The primary limitation is the absence of a control group, therefore the observed improvements in health outcomes could be attributable to regression to the mean and/or to other clinical factors like medication changes. On the other hand, the observed clinically and statistically meaningful improvements in diet quality, food security, health status, and biometrics all occurring concordantly support biologic plausibility. Missing data were common in some programs, which may lead to selection bias. However, sensitivity analyses with complete case analyses and exclusion of programs with over 50% missingness supported our primary results. F&V intake was the only dietary component assessed; therefore we do not know if other dietary components improved or if detrimental substitutional effects occurred. Some programs were implemented during the COVID-19 pandemic which may have impacted their efficacy. Programs pooled together for analysis did not have identical eligibility criteria, implementation, or duration, creating heterogeneity in participant experiences across programs. Yet, all programs were designed to improve FV intake, food security, and cardiometabolic health outcomes among individuals at risk for cardiometabolic health and food insecurity; and hierarchal mixed models accounted for clustering at the program level. All programs were operated in partnership with a single nonprofit provider of produce prescriptions, and thus findings may not be generalizable to other programs.

Conclusions

In this multi-site, participant-level analysis of nine produce prescription programs across 22 sites in 12 U.S. states that enrolled low-income or food insecure patients at risk for diet-related illness, program participation was associated with improvements in F&V intake, food insecurity, and self-reported health status among both adults and children, and clinically relevant improvements in HbA1c, BP, and BMI among adults. These findings provide important new evidence from a diverse set of programs for meaningful benefits of produce prescriptions, highlighting the need for clinical, policy, and healthcare payer and provider efforts to implement larger pilots and randomized designs of produce prescriptions.

What is Known.

• Produce prescriptions may play a role in improving cardiometabolic health by increasing fruit and vegetable (F&V) consumption and food security.

• Evaluations on clinical biomarkers of cardiometabolic health such as glycated hemoglobin (HbA1c), blood pressure (BP), and body mass index (BMI) for patients with diet-related illness remain limited, with mixed results from mostly small studies.

What the Study Adds.

• This is the largest produce prescription study to date to assess health outcomes, pooling data across 22 program locations in the U.S.

• The study found that produce prescription program participation was associated with improvements in F&V intake, food insecurity, and self-reported health status among both adults and children, and clinically relevant improvements in HbA1c, BP, and BMI among adults with poor cardiometabolic health.

• These findings provide important new evidence from a diverse set of programs for meaningful benefits of produce prescriptions, highlighting the need for clinical, policy, and healthcare payer and provider efforts to implement larger pilots and randomized designs of produce prescriptions.

Non-standard Abbreviations and Acronyms

BMI

Body mass index

BMI z-score

Body mass index z-score

BP

Blood pressure

CHIP

Children’s Health Insurance Program

F&V intake

fruit and vegetable intake

HbA1c

Glycated hemoglobin

SNAP

Supplemental Nutrition Assistance Program

WIC

Special Nutrition Program for Women, Infants and Children (WIC)