Prescribed Groceries Cut HbA1c 0.40 Points in Medicaid Trial

Food as medicine has had more enthusiasm than evidence. Cities hand out produce vouchers, health plans pay for meal deliveries, and most of the proof has come from programs that compared people before and after, with no control group. A trial published in Circulation this summer is one of the few that did it properly. Researchers at Kaiser Permanente Southern California and the Food is Medicine Institute at Tufts randomly assigned 460 Medicaid insured adults with poorly controlled type 2 diabetes to usual care or to weekly home deliveries of medically tailored groceries. The deliveries lowered HbA1c more than usual care did. The gap was 0.40 percentage points.

What the trial tested

Participants were enrolled between November 2021 and July 2022 in southern California. To qualify, they needed at least two HbA1c readings of 7.5% or higher in the previous year. At baseline the average age was 59.2 years, 64.8% were women, 85.2% reported Hispanic ethnicity, 58% reported food insecurity, and the average HbA1c was 9.40%.

They were randomized one to one to one: 153 to usual care, 153 to a lower dose of groceries and 154 to a higher dose. The groceries were weekly deliveries of healthy produce scaled to household size, valued at $100 to $170 a month in the lower dose arm and $135 to $210 in the higher one. They came with matched recipes and an offer of telehealth nutrition counseling. The primary comparison pooled the two grocery arms against usual care, and the primary outcome was the change in HbA1c at six months.

The result, in absolute terms

HbA1c fell by 0.66 points from baseline in the grocery groups and by 0.25 points in the control group. The treatment difference was minus 0.40 points (95% confidence interval minus 0.73 to minus 0.08, P = 0.016). The interval matters: it runs from a barely measurable gain to a moderate one. Starting from 9.40%, the grocery groups finished at roughly 8.7% on average, which is still far above the usual target of 7%.

The secondary results are firmer. The odds of food security rose by a factor of 2.12 (95% CI 1.13 to 3.99) and the odds of nutrition security, meaning reliable access to healthy food, by a factor of 3.65 (95% CI 1.84 to 7.25). The authors report that results were consistent across prespecified subgroups, including sex, education, baseline food security and baseline HbA1c, and that adjusting for demographics, comorbidities and medication changes gave similar findings. Adherence was high: 244 intervention participants, 83.3%, said they ate most or all of the food they received.

Two null results deserve as much attention as the positive one. Hypertension and body mass index did not significantly change. And the lower and higher dose arms produced similar HbA1c reductions, so a bigger box did not buy a bigger effect. The company that delivered the program, Foodsmart, adds figures of its own in a press release, for example that the share of participants reaching an HbA1c below 7% was 8.9% in the grocery groups against 3.4% under usual care. Those numbers come from the sponsor's summary, not from the journal abstract, and they show how low the bar still was: about nine in ten people were still above 7% at six months.

Does dose matter? A second trial says yes, for food security

A separate randomized community trial in the Journal of General Internal Medicine asked how much a produce prescription needs to be worth. It randomized 242 adults to $40, $80 or $110 a month, redeemable only for fruits and vegetables at major grocery stores, for six months. The participants were mostly young and not selected for diabetes: mean age 35.2, 97.1% women, 82.6% Hispanic.

At three months the benefit redemption rate was 60.2% in the $40 group, 67.6% in the $80 group and 70.4% in the $110 group. Food insecurity was reported by 74.3%, 66.2% and 47.4%, respectively, and fruit and vegetable intake averaged 2.51, 2.66 and 2.89 cups a day. Compared with $40, the $110 dose significantly improved food insecurity (P = 0.01) and intake (P = 0.03) at three months. The two trials answer different questions. One measures blood sugar in people with diabetes, the other measures food access and diet in a general low income group, and neither shows that more money produces more HbA1c reduction.

What to make of it

The trial supports a narrow claim: for Medicaid patients with poorly controlled diabetes, a six month grocery program with recipes moves blood sugar a little and food security a lot. It does not support the stronger slogans. A 0.40 point average gain is modest, and a program that leaves the average participant near 8.7% has not solved diabetes control. Our piece on early diabetes remission shows what a far larger intervention looks like.

I read the food security result as the main finding. People who cannot reliably afford healthy food are likelier to have uncontrolled diabetes, and fixing that is worth doing even if the HbA1c effect stays small. The policy question is cost against a modest glycemic benefit, and this trial cannot answer it. The next useful studies would run longer than six months, separate the groceries from the counseling, and count medicine use, emergency visits and spending. For how to weigh a trial like this one, our five minute guide covers the checks.