Health and Economic Impacts of Implementing Produce Prescription Programs for Diabetes in the United States: A Microsimulation Study

Background Produce prescription programs, providing free or discounted produce and nutrition education to patients with diet‐related conditions within health care systems, have been shown to improve dietary quality and cardiometabolic risk factors. The potential impact of implementing produce prescription programs for patients with diabetes on long‐term health gains, costs, and cost‐effectiveness in the United States has not been established. Methods and Results We used a validated state‐transition microsimulation model (Diabetes, Obesity, Cardiovascular Disease Microsimulation model), populated with national data of eligible individuals from the National Health and Nutrition Examination Survey 2013 to 2018, further incorporating estimated intervention effects and diet‐disease effects from meta‐analyses, and policy‐ and health‐related costs from published literature. The model estimated that over a lifetime (mean=25 years), implementing produce prescriptions in 6.5 million US adults with both diabetes and food insecurity (lifetime treatment) would prevent 292 000 (95% uncertainty interval, 143 000–440 000) cardiovascular disease events, generate 260 000 (110000–411 000) quality‐adjusted life‐years, cost $44.3 billion in implementation costs, and save $39.6 billion ($20.5–58.6 billion) in health care costs and $4.8 billion ($1.84–$7.70 billion) in productivity costs. The program was highly cost effective from a health care perspective (incremental cost‐effectiveness ratio: $18 100/quality‐adjusted life‐years) and cost saving from a societal perspective (net savings: $−0.05 billion). The intervention remained cost effective at shorter time horizons of 5 and 10 years. Results were similar in population subgroups by age, race or ethnicity, education, and baseline insurance status. Conclusions Our model suggests that implementing produce prescriptions among US adults with diabetes and food insecurity would generate substantial health gains and be highly cost effective.


Systematic review and meta-analysis on effect sizes of produce prescription programs.
The intervention effects on dietary habits, BMI, and HbA1c were estimated based on a new meta-analysis of quasi-experimental or randomized controlled produce prescription interventions.We focus on programs that evaluated the impact of continued fruit and vegetable prescriptions over an extended period (more than three months) on fruit and vegetable intake, BMI, and HbA1C.A total of 20 produce prescription intervention programs were identified, including 11 studies in a recent systematic review, 6 more studies from our unpublished analysis of completed produce prescription programs, and 3 more studies identified through Pubmed searches after the screening date of the systematic review (see eTable B1 for detailed characteristics of the included studies).Among the eligible studies, most (17 of 20) enrolled adults with poor cardiometabolic health including diabetes or prediabetes, hypertension, CVD, obesity, or overweight.All programs enrolled participants who were food insecure or at high risk for food insecurity, with a weighted average of 74% being food insecure.In addition, all 10 programs that assessed the effect on HbA1c were implemented among diabetes patients, with 79% being food insecure.The average age of participants across all programs was 54 ye ars, with 69% female, 38.5% non-Hispanic Black adults, and 30.0%Hispanic adults.This population was a bit younger and had a higher percentage of females and non -Hispanic Black individuals than the national population with diabetes and food insecurity from NHANES, where the average age was 58 years, 55.5% were female, and 17% were non-Hispanic Black.However, a recent study from our team evaluating the impact of produce prescription interventions on dietary intake and cardiometabolic risk factors identified no significant evidence of differential effects by age, sex, race/ethnicity, or SNAP enrollment status.
Study-specific effect estimates were pooled using inverse-variance weighted random-effects meta-analysis.The I 2 statistic was used to assess the heterogeneity of included studies, with values <25%, 25 to 50%, and >50% corresponding to low moderate, and high degrees of heterogeneity, respectively.A high level of heterogeneity was observed for fruit and vegetable intake, and HbA1c level (I2>90%), and for BMI, a low level of heterogeneity (I2=22%).Multiple factors could influence the effect size of the produce prescription programs, including inclusion criteria, the dollar amount of food vouchers or food boxes, sample size, duration of the study period, and factors that could hardly quantify such as nutrition ed ucation provided, and convenience for accessing the fruits and vegetable purchase point.Due to the small number of studies included for each outcome (13 studies for fruits and vegetable intake, 9 studies for BMI, and 10 studies for HbA1c), and various study quality, we had little confidence to tease out the possible interactive effect of different component of produce prescription and therefore focused on evaluating the average effects across studies.
Based on the weighting factors for each study generated in the meta-analysis, we calculated the weighted mean of the dollar amounts of produce prescriptions provided and average dollar amounts redeemed per capita.

Estimation of administrative costs of national produce prescription program
We estimated the administrative costs of the national produce prescription program by reviewing the costs of the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) and the Supplemental Nutrition Assistance Program (SNAP), each of which provides nutrition assistance for eligible individuals.Both programs provide nutrition education in addition to financial benefits, with WIC required to provide specific types and numbers of nutrition education, and nutrition education being optional in SNAP with State flexibility.SNAP provides an average monthly benefit of $130 per person for 40 million participants.The administrative costs of SNAP contribute to 5-8% of total program costs according to prepandemic program cost data, including State administrative expenses, eligibility certification, nutrition education, employment and training programs, benefit and retailer redemption and monitoring, payment accuracy, EBT Systems, program evaluation and modernization, program access, and health and nutrition pilot projects.WIC provides an average monthly benefit of $61.35 per mother for 6-8 million participants each year.After excluding a minority of administrative costs related to breastfeeding promotion and education (about 1.8% of total program costs), administrative costs of WIC contribute to about 21.3% of total program costs, including participant eligibility, nutrition education, health care coordination , and referral, drug abuse education, clinic operations, food delivery and warehousing, vendor monitoring, financial management, program integrity, and systems development and operations.
A national produce prescription program may share similar administrative components with both WIC and SNAP on eligibility certification, quality control, employment and training, use of the EBT system or food delivery, benefit and retailer redemption and monitoring, and program evaluation.A national produce prescription program may have greater similarities with WIC than SNAP, as WIC includes partnerships with healthcare organizations and has higher intensity and consistency of nutritional education.Still, the administrative burden and costs could be lower in a national produce prescription program than in WIC, as the former provides only produce rather than multiple categories of foods, does not require income certification, and does not need to establish separate clinical institutions (WIC clinics).We assumed that the administrative costs of a national produce prescription program would be 15% of total program costs, about 2-3-fold higher than the administrative costs of SNAP and one-third lower than the administrative costs of WIC.In sensitivity analysis, we evaluated alternative administrative costs of 8% (upper end of SNAP costs) and 21.3% (WIC costs).We also assumed the costs to be higher in the first year of implementation due to program launching, equal to 50% of food costs (33% of total program costs).

Estimating healthcare cost savings associated with HbA1c level in the produce prescription program
HbA1c is recognized as an important predictor of healthcare costs for diabetes patients, independent of multiple diabetes comorbidities such as CVD and hypertension. 28,29Patients with good glycemic control were less likely to treat their diabetes with medication and less likely to be prescribed insulin, 51 and are associated with a reduced risk of diabetes-related microvascular (neuropathy, retinopathy, and nephropathy) and macrovascular (CHD and stroke) complications, [52][53][54] Since the cost prediction algorithm of OCD-M does not intrinsically capture the influence of HbA1c changes in the medical costs of diabetes, we further incorporated additional HbA1crelated cost savings in the model.We assumed that improved HbA1c could re duce health care costs for diabetes patients by reducing costs of diabetes treatment such as diabetes medications, physician visits, and self-testing devices, 51 and costs related to treating microvascular complications (neuropathy, retinopathy, and nephropathy), [52][53][54] and macrovascular complications (CHD and stroke).To avoid double-counting for the savings of healthcare expenditures related to CVD outcomes, which have already been captured by the fruit and vegetable and BMI pathway, we restricted our estimation to non-CVD-related healthcare costs.
To identify the best available evidence for HbA1c reduction on diabetes-related health care costs, we systematically reviewed studies assessing the impact of HbA1c reduction on diabetes-related health care costs in the US from 2000 through now.Despite diverse data sources, and the scope of healthcare costs, all studies suggested reducing HbA1c is associated with redu ced healthcare costs.
We used the evidence from the study by Maureen et al 2020, the most recent available data that provided the cost estimation related to HbA1c reduction.The study suggested that a 1 percentage point reduction in HbA1c was associated with a 13% reduction (or a $736 reduction) in diabetesrelated healthcare costs, independent of comorbidity conditions. 55We assumed that this cost reduction equally applied to both CVD and non-CVD-related diabetes-related costs.
The DOC-M predicts the marginal effect of having diabetes on health care expenditure ($3859.002per person per year), independent of age, sex, race, CVD, BMI, and blood pressure.The marginal effect of diabetes intrinsically subtracted CVD-related costs from diabetes costs.We estimate that a One percentage point reduction in HbA1c was associated with a 13%*$3859.002=$501.67 decrease in CVD-independent diabetes-related medical costs per person per year.* Baseline healthcare expenditures capture the average annual total healthcare expenditures for individuals who were aged 40 years, male, non-Hispanic Whites, BMI of 28, and had no diabetes, high blood pressure, coronary heart disease, or stroke.The marginal effects represent additional changes in health care expenditures from the baseline expenditure by the change in one unit of the predictor.For example, with all other predictors unchanged from baseline characteristics (i.e., BMI 28, male, non-Hispanic Whites, and no clinical condition), a one-unit change in age (from age 40 to age 41) would increase annual healthcare expenditures by $95 on average.Thus, the model included the healthcare costs for all conditions not directly captured in the model (e.g., asthma, joint pain, etc.) according to their average costs for a US adult of any given age, sex, race, and cardiometabolic health .* All included studies reported changes in fruit and vegetable consumption (in servings per day) pre -and post-intervention, or as a difference in differences between the control and intervention groups.Some studies reported these effect sizes for fruits an d vegetables separately, and some studies reported them in aggregate.† Values were converted to 2021 US dollars.
Table S5.Estimated age-specific etiologic effects of fruits and vegetables on cardiometabolic outcomes * Abbreviations: CHD, coronary heart disease; CI, confidence interval; RR, relative risk.* The detailed methods for reviewing and synthesizing evidence to estimate effect sizes for associations between dietary factor s and cardiometabolic endpoints have been reported. 23,24We utilized evidence from meta-analyses of prospective cohorts or randomized clinical trials evaluating direct associations of dietary factors with coronary heart disease (CHD), stroke, or type 2 diabetes, by age. 2,23,24† The available evidence suggests an effect of seafood omega -3 on fatal CHD, with less clear evidence for benefits on nonfatal CHD. 72Because the risk transitions influenced by diet in the CVD-Predict model are for the incidence of a CHD event, with subsequent transitions to death independent of dietary risk factors (see Fig 1), the current analysis will modestly overestimate the benefits of changes in seafood omega -3 consumption.Abbreviation: CVD, cardiovascular diseases; CHD, coronary heart diseases; RVSC, revascularization, including coronary artery bypass surgery and percutaneous coronary intervention.Note: Figure 1 highlights key transitions from one health state/event to another state/event using solid arrows while dotted-line arrows represent cause-or eventrelated mortality.Grey rectangles represent five different health states in which individuals can stay throughout modeled pe riods, while yellow circles show cardiovascular disease events that individuals can experience in any given year.Diabetes was defined as self -reported diabetes or one of four clinical criteria (i.e., fasting plasma glucose level ≥ 126 mg/dl, 2-hour plasma glucose level ≥ 200 mg/dl, Hemoglobin A1c level ≥ 6.5%, or use of diabetic medications) 73 .Individuals with a prior history of CVD events were defined as those who reported experiencing at least one of the four events: angina, stroke, heart attack, and coronary heart disease.For angina, we used the Rose questionnaire criteria 74 or the use of anti-angina medications.

Figure S2. Forest plot for effects of produce prescription programs on fruit and vegetable (F&V) intake.
Note: Data were pooled using random-effects meta-analysis.In most studies, combined fruit and vegetable intake was assessed, whereas effect sizes for disease outcomes were available for fruits and vegetables separately, we assumed similar average effect sizes of produce prescriptions on fruit vs. vegetable intake (0.4 servings/day each).The duration of these intervention studies ranged from 3 to 18 months.

Figure S3. Forest plot for effects of produce prescription programs on BMI.
Note: Data were pooled using random -effects meta -analysis.Pooled results are statistically significant even when most of the individual studies were insignificant, due to the improved statistical power from pooling multiple studies   The average intervention effect size and its uncertainties of the produce prescription program on fruit and vegetable intake, BMI, and HbA1c were estimated from a new meta -analysis of 20 produce prescription programs that assessed these outcomes.The 2.5 th to 97 th percentiles intervention effect sizes are equivalent to an increase of 0.083, 0.17, 0.26, 0.39, 0.53, 0.65, 0.76 servings/day for fruit and vegetables respectively, and reductions of 0.26, 0.31, 0.38, 0.44, 0.51, 0.57, and 0.64 kg/m2 for BMI, and reductions of 0.27%, 0.39%.0.49%, 0.62%, 0.75%, 0.87%, and 0.97 for HbA1c.Net changes in costs from a healthcare perspective = total intervention costs -healthcare costsavings by produce prescriptions; Net changes in costs from a societal perspective = total intervention costshealthcare cost-savings -savings from averted productivity loss.Negative values in the net costs indicate cost savings.The met monetary benefits were calculated as estimated QALYs averted* willingness to pay threshold($100,000/QALY) -net change in costs.A positive net monetary value indicates that the intervention is cost-effective.

Figure S8. Net costs by varying assumptions on administrative costs for program implementation
The weighted average monthly incentive costs for the 20 studies for estimating intervention effect sizes were $31.9/month (SE =4.4) in 2021 USD.In the base case, we estimated the administrative costs to be equal to 15% of the annual total program cost or equal to 17.5% of the incentive costs.Also, we assumed the cost is doubled in the program launching stage at the first year of intervention, equal to 30% of total program costs.We add itionally varied the administrative costs as a percentage of total program costs, for lower administrative costs at 8% of total program costs and higher at 23% of total program costs , which are also doubled in the first year of intervention.
Figure S1.Conceptual diagram of the model structure for the DOC-M.

Figure S4 .
Figure S4.Forest plot for effects of produce prescription programs on HbA1c.

Figure S6 .
Figure S6.Estimated healthcare cost savings of a national produce prescription program by population subgroups, per 100,000 population, at 5 years, 10 years, and a lifetime of interventio n

Figure S7 .
Figure S7.Net changes in costs and net monetary benefits by varying intervention effect sizes on different percentiles of its distribution.

Table S1 . Key model parameters and data sources for evaluating the health and economic impact of the produce prescription program.
ASCVD, atherosclerotic CVD; ACC/AHA, American College of Cardiology/ American Heart Association; CDC, Centers for Disease Control and Prevention; CVD, cardiovascular disease; CHD, coronary heart disease; RVSC, revascularization, including coronary artery bypass surgery (CABG) and percutaneous coronary intervention (PCI); HRQOL, health-related quality of life; MEPS, Medical Expenditure Panel Survey; SE: standard error.Where uncertainty around input parameters (e.g., cost of CABG/PCI) is not available, we assume 20% of the mean estimate as a standard error to generate parameters for probabilistic distributions.

Table S2 . Marginal effects of individual characteristics on estimated annual health care expenditures among U.S. adults aged 40-79 years, based on 73,174 individuals in MEPS 2014-16.
Abbreviations: BMI, body mass index; CI, confidence interval.MEPS, Medical Expenditure Penal Survey.

Table S3 . Percentages of missing values for individual-level characteristics in the model population.
Abbreviations: BMI, body mass index; DBP: diastolic blood pressure; HDL, high-density lipoprotein; SBP, systolic blood pressure.