Medicaid paid for meals as treatment. Hospitalisations fell 31% and costs nearly balanced

"Food is medicine" has been a slogan for longer than it has been a covered benefit. An evaluation published in Nature Medicine on 2 June reports on what happened when a US state actually bought the food: Massachusetts, through a Medicaid demonstration, delivered medically tailored meals to people with diet-related conditions and food insecurity, and researchers tracked what happened to their hospital use and costs [s1].

What medically tailored meals are, and what was already known

Medically tailored meals are prepared meals designed for a specific medical condition and delivered to a patient. The evidence base going into this study consisted of smaller quasi-experimental studies and short-term controlled trials showing health benefits for several conditions — but, as the paper states, medically tailored meals had not been evaluated in large-scale policy initiatives [s1]. Multiple US states are now implementing them in Medicaid, the health insurance programme for low-income individuals, without that evaluation existing [s1].

The study

The researchers examined changes in hospitalisations, emergency department visits and healthcare costs among 1,866 recipients of medically tailored meals and 1,372 comparators, from 2020 to 2023, across 11 health systems in Massachusetts [s1]. Both groups met eligibility criteria for diet-related conditions and food insecurity [s1]. Mean duration of meal receipt was 6.7 months [s1].

The analytic method was propensity overlap-weighted generalised estimating equations, comparing a six-month baseline period against the programme period [s1].

The results

Receipt of medically tailored meals was associated with 31% fewer hospitalisations (adjusted incidence rate ratio 0.69, 95% CI 0.58-0.82) and 20% fewer emergency department visits (aIRR 0.80, 95% CI 0.72-0.89) [s1].

Total healthcare costs were US$3,433 lower (95% CI -$1,215 to -$5,651) [s1]. The paper reports that these cost reductions offset 98% of the programme costs during the enrolment period [s1] — close to, but not quite, cost-neutral.

In stratified analyses, meal receipt was net cost-saving among participants with cardiovascular disease, chronic kidney disease, depression, diabetes, or high comorbidity [s1]. Findings were robust across sensitivity analyses, a negative control test, and a secondary comparison group [s1].

What the design can and cannot support

This is an observational comparison of people who received meals against people who did not, in a programme where nobody was randomised. The authors' methods are aimed squarely at that problem — overlap weighting balances the two groups on measured characteristics, and the negative control test is designed to detect residual confounding [s1]. But no statistical adjustment removes the possibility that the people enrolled in the programme differed from the comparators in ways not captured in the data.

Two specific concerns are worth naming, neither of which the study can resolve. The first is that enrolment in a meal-delivery programme requires a referral, a reachable address, and enough stability to receive deliveries for months. Those are correlates of things that also predict fewer hospital admissions. The second is the six-month baseline comparison: patients are often referred to programmes like this shortly after a period of high healthcare use, and utilisation tends to drift back toward a person's average afterwards regardless of what is done. The negative control and secondary comparison group are the authors' defences against exactly these readings [s1], and they are reasonable defences, but they are not randomisation.

What the study does establish with confidence is that a real programme, at real scale, in eleven health systems, produced these associations — which is a different and more policy-relevant claim than a trial in a single clinic.

The 98% figure

The costs-offset number is the one most likely to be repeated, and it is more specific than it sounds. It refers to healthcare cost reductions offsetting 98% of programme costs during the enrolment period [s1] — that is, during the months meals were actually being delivered, and counting the programme's own costs.

It is not a claim that the programme saved money overall, and the authors do not make one. Roughly breaking even on a benefit that also reduced hospitalisations by 31% is a strong result on its own terms; it does not need to be upgraded into a savings claim to be interesting to a state budget office.

What to watch

Whether any of the states now adding medically tailored meals to Medicaid builds a randomised or stepped-wedge evaluation into the rollout rather than evaluating after the fact. Whether the condition-specific stratified findings — cardiovascular disease, chronic kidney disease, depression, diabetes, high comorbidity [s1] — hold in other state programmes, since those are the groups that would define eligibility if the benefit were narrowed for cost reasons. And whether effects persist after meal delivery stops, which a study measuring the enrolment period cannot answer.

Sources

1. [s1] Medically tailored meals receipt and healthcare utilization and costs in Massachusetts' Medicaid demonstration. Nature Medicine, published online 2 June 2026. https://doi.org/10.1038/s41591-026-04407-5