MIT Sloan Health Systems Initiative

Food as Medicine: Surprising Results from a Randomized Controlled Trial

We’ve all heard that an apple a day keeps the doctor away. However, when millions of healthcare dollars and patients’ lives are at stake, it pays to have rigorous evidence informing treatment decisions. One highly regarded mid-Atlantic health system wanted evidence to support the success of their intensive food as medicine program for patients with uncontrolled type 2 diabetes. That appeared to be the case. Those who participated in the program showed improved HbA1c levels, showing a correlation. However, to prove a causal link, Prof. Joseph Doyle led a team to conduct a randomized control trial, the gold standard, to see if the resources allocated to this program had the intended benefit.

The Study

The study was a collaboration with a highly-rated health system in the mid-Atlantic with a nationally recognized food as medicine program. Two sites, one suburban and one urban, were chosen. More than 3700 type 2 diabetes patients were assessed; 500 were deemed eligible and enrolled. All those chosen to participate had an elevated HbA1c level above 8%, reported food insecurity and were affiliated with the health system.

The study population was about half female and representative of the health system’s patient population. For the clinical trial, participants were randomly assigned to the control group or the intervention group. The intervention group received enough food for 10 meals a week for every member of the household, weekly meetings with clinicians for preventive care including access to a dietitian, a nurse, and a community health worker who provided education about both their medical condition and suggestions for recipes and cooking instructions. The program was constructed such that it was seen as a prescription by the health system to help overcome stigma associated with the use of food banks.  In addition, a grocery prescription would enable participants to make their own meals, which was intended to empower the participants in ways that might affect their healthy behaviors more generally.  Ultimately, the program lowered the cost of eating healthier both in terms of spending and knowledge: participants could experiment with healthy foods such as fish and brown rice that they would find “too risky” to purchase on their own. 

After six months, those in the control group also gained access to the program.  This is a particularly intense version of a food as medicine program with an average cost of $2000/participant and an average length of time in the program of about one year.

The Results

The hypothesis was that participants in the food program would experience a greater reduction in HbA1c levels than those in the control group. HbA1c is a biomarker used to define type 2 diabetes and reflects diet and other healthy behaviors over the three months prior to the test.Intuitively, the proposed effect of the program seems to make sense. Better diet, along with education and support – just short of delivering fully prepared meals – would lead to health benefits that would be apparent in patients’ HbA1c levels. 

The results however, showed that both groups experienced a reduction is HbA1c levels by about the same amount. Furthermore, both the control and intervention groups showed no difference in related lab results such as levels of cholesterol, triglycerides, fasting glucose and blood pressure. If anything, weight increased slightly for those with access to the program. Self-reported measures such as diabetes self-efficacy, exercise, smoking and healthy attitudes were also similar. When the control group joined the treatment group in the food program, they did not experience an additional decrease in HbA1 levels, but stayed around the same. 

Participants in the food program, whether part of the intervention group or those allowed to join after six months in the control group, interacted more often with the health system, with higher engagement in the food pickup, education, and dietician visits. They also self-reported significantly improved diet than the control group.  


The main take-away from this study is the importance of randomized controlled trials, with a credible control group. An observational study – without a control group – would have concluded that the food program intervention was a success. Before the intervention, participants had high levels of HbA1c, and afterwards they posted healthier levels. These results could convince a health system to invest heavily in a food program modeled after the intervention at the expense of other possibilities. 

It isn’t clear how the control group achieved similar reductions in their HbA1c levels, but the study team can say that they don’t see engagement with healthcare services increase until they gain access to the program. What can be concluded is that patients with uncontrolled type 2 diabetes who are regularly followed in a health system can reduce their HbA1c levels without the benefit of an intensive food as medicine program.

This conclusion is similar to one from another of Doyle’s studies: Hotspotting in Camden, New Jersey. The intervention in that case was a nationally-recognized program of intensive case management to address patients’ clinical and social needs focused on those who frequented the emergency department and other high-cost care (so-called “super utilizers”). After a period of observational study, the program was considered a success and adopted by health systems around the country. Doyle and a team of MIT researchers conducted a randomized controlled trial of the program and discovered that both intervention and control groups improved. The intervention program did not have the influence that the observational study results suggested. Instead, a different type of intervention would be necessary to lower hospital utilization for the most vulnerable patients. As in the food as medicine study, a good control group was necessary to learn whether or not the program achieved its goals.

This study is a key addition to the ongoing discussion about how to best treat diet-related chronic disease. In order to make progress, we cannot assume these types of programs work, or that diabetes management lowers HbA1c by 1-2 points based on before-after comparisons.  Instead, we need a properly configured control group to uncover the causal relationship between diet interventions and disease.  At a time when there are so many questions about how to best help populations of high-need patients, a randomized controlled trial may offer definitive direction.

For more more about this project see the following articles:

  1. “The Effect of an Intensive Food-as-Medicine Program on Health and Health Care Use: A Randomized Clinical Trial,” JAMA Internal Medicine, Dec 26, 2023. Joseph Doyle, PhD; Marcella Alsan, MD, MPH, PhD; Nicholas Skelley, BA; Yutong Lu, BA; John Cawley, PhD 

  2. "Does food as medicine help patients with diabetes? Here’s what a new clinical trial says" STATNews, December 26, 2023. By Marcella Alsan, John Cawley, and Joseph Doyle

  3. Does “food as medicine” make a big dent in diabetes?, MIT News, Dec 26, 2023

  4. Webinar on Fresh Food Farmacy featuring Prof Joe Doyle

  5. A "Food Is Medicine" Approach to Disease Prevention: Limitations and Alternatives, JAMA Network, November 30, 2023. By Alyssa J. Moran and Christina Roberto  

Joseph Doyle

Joseph Doyle

Economics, Finance and Accounting

Erwin H. Schell Professor of Management

Learn More