Does food as medicine help patients with diabetes? Here’s what a new clinical trial says

Food-as-medicine programs that address food insecurity and diet-related chronic diseases have drawn widespread attention from policymakers, payers, and health care providers. But what is the evidence that such programs work as intended? Some research suggests that food as medicine can improve health and lower health care costs, but until now there has been little evidence from randomized clinical trials.

In a paper published this week in JAMA Internal Medicine, we report the results of a clinical trial of a particularly intensive food-as-medicine program. We were fortunate to partner with a large health care system whose food-as-medicine program is celebrated as a model. The approach involves the “prescription” of healthy food to patients who are food insecure and suffering from uncontrolled type 2 diabetes (an HbA1c > 8). The health care system opened brick-and-mortar clinics where patients can receive enough food for 10 meals each week for the patient and their family. While at the clinic, patients received consultations from a dietician, nurse, or community health worker. They could also participate in diabetes self-management training and cooking classes. By prescribing the intervention, the program sought to overcome stigma associated with going to a food bank and better manage patients’ diet-responsive chronic conditions. And while other programs offer healthy food for only a few weeks, this program allowed participants to stay with the program much longer, averaging about one year. The cost of the program per patient is roughly $1,900.

The trial took place from 2019 to 2022 at two such clinics, one in a small city and one in a rural area. The treatment group began the program right away, while the control group did not receive the program until six months later. The results were striking. The program was popular among the treatment group, with high engagement in food pickups, dietitian visits, and completion of education programs. There was a substantial improvement in self-reported diet in the treatment relative to the control group. Yet, this increased engagement did not translate into a significantly different HbA1c, relative to the control group, after six months.

Both groups started the study with very high HbA1c values, averaging around 10. And the treatment group did see a decline: about a 1.5-point reduction. However, the control group, which did not yet have access to the program, experienced a similar decline. When the control group began participating in the program at six months, they also had higher engagement, but their HbA1c levels had already plateaued and did not show further improvement. The program did address patients’ food insecurity and increased engagement for a socially disadvantaged group, but these changes did not improve health outcomes over the following six to 12 months as measured by laboratory tests and patient surveys. The findings were similar across a range of subgroups, including each of the two locations and for those with HbA1c above and below 9.5.

So, what can we learn from these findings?

First, the trial demonstrates the necessity of having a credible control group, especially when programs focus on patients with elevated disease markers like HbA1c. If one only looked at the treatment group before and after, one might conclude that the program reduces HbA1C. It is unclear how the control group achieved a similar reduction; we do not observe them increasing their health care utilization or using other nutrition services offered by the health care system during the period in question.

Second, we learned that in a setting where patients are followed regularly within a health care system, patients with an unusually high level of HbA1C can achieve reductions in HbA1C without an intensive food-as-medicine program.

Our study leaves a number of questions unanswered. It is possible that the program may have greater impact in other contexts. Perhaps the program would improve health outcomes among patients in a less well integrated health care system, or who have less of a connection to care. We also don’t know whether the program would have worked better without the disruption caused by the pandemic, although we also did not observe an effect on HbA1C among participants whose outcomes were measured before the pandemic began.

Most important, we can’t say whether other forms of food-as-medicine might be more effective. There are numerous credible designs of food-as-medicine programs; the two main categories are food prescription programs like the one we studied and what are known as medically tailored meals. These meals are prepared and delivered to the patient’s home. That increased convenience and time savings may improve adherence and health outcomes.

The results of this trial inform the ongoing conversation as to how to best address diet-related chronic disease. Even if the trial had a different outcome, we would be advocating for further studies and replication, and we are grateful to our partner for allowing us the opportunity to study their program. Researchers, clinicians and policymakers all share a common goal to fight food insecurity and improve population health. Randomized clinical trials are key tools for discerning what works best, for whom and why, information that we should all be hungry for.

Marcella Alsan is a professor of public policy at Harvard Kennedy School. John Cawley is a professor in the Brooks School of Public Policy, and the Department of Economics, at Cornell University. Joseph Doyle is the Erwin H. Schell professor of management and applied economics at the MIT Sloan School of Management.