The food-is-medicine movement has arrived at the moment its advocates spent a decade working toward. Section 1115 waivers now allow Medicaid dollars to reimburse produce prescriptions and medically tailored meals in more than a dozen states. Congressional staff briefings on the topic are standing-room-only. Philanthropic co-funding is at record highs. And the evidence base, honestly examined, does not support the enthusiasm.

The strongest studies in the field — the ones with a randomized comparison group and a follow-up window longer than six months — show modest, sometimes non-significant improvements in HbA1c, blood pressure, and self-reported diet quality. The effect sizes are real, but they are on the order of what a well-designed diabetes-education program produces at a fraction of the cost. The studies that show larger effects are almost uniformly single-arm pilots with high attrition, which is the study design most vulnerable to the outcome we want to see.

This is not an argument against food-as-medicine programs. It is an argument for describing what they are. They are a food-security intervention, delivered through a clinical channel because that is where the funding and the trust are. They keep people fed. They connect isolated patients to a clinician. They almost certainly improve well-being in ways the biomarkers do not capture. What they are not, on the current evidence, is a substitute for medical treatment of the conditions they are marketed against.

The distinction matters because the policy is being sold to state legislatures on a return-on-investment claim — that a dollar of produce prescription saves several dollars of downstream medical spending — that the evidence does not currently support. When the claim fails to materialize in the utilization data, the program will be defunded, and the food-security intervention that was doing real work will go with it.

The honest sale is different, and probably durable. Food-is-medicine programs are a defensible use of Medicaid dollars because they address a determinant of health that the medical system otherwise ignores, and because delivering them through the clinic reaches patients that a SNAP-only strategy does not. That case does not require the biomarker story to hold up. It is a better case anyway. It is also the one the field will need if it wants the funding still to be there in 2030.