Designing Medically Tailored Meal Interventions: The Food as Medicine for Families (FAME-F) Factorial Randomized Clinical Trial

Abstract

Background:

Medically tailored meals (MTM) have unanswered implementation questions. Providing MTM for a specific individual, rather than the entire household, could ‘underdose’ if food is shared, but household MTM is more costly. Delivery with commercial shippers rather than dedicated drivers may broaden accessibility but could worsen mental health. We sought to test these possibilities.

Methods:

2×2 factorial randomized comparative effectiveness trial (February 2024 to May 2025) of adults with control over dietary intake in New England. All participants received 10 MTM per week for 12 weeks, prepared under supervision of a registered dietitian. Participants were randomly assigned along two dimensions: dose (‘feed the family’ vs. ‘feed the individual’), and delivery (‘dedicated driver’ vs. ‘commercial shipper’). The primary ‘dose’ dimension outcome was Healthy Eating Index 2020 (HEI) diet quality score (range: 0 to 100; higher scores indicating greater adherence to dietary guidelines). The primary ‘delivery’ dimension outcome was the de Jong Gierveld loneliness scale (range: 0 to 11; greater scores indicating more loneliness). Intention-to-treat linear regression analyses assessed outcomes at 12 weeks and allowed for interactions between dimensions.

Results:

93 participants (mean [SD] age: 57.6 [12.9] years; 67% female; 48.9% non-Hispanic White) were randomized: 46 to ‘feed the family’, 47 to ‘feed the individual’, to 43 ‘dedicated driver’, and 50 to ‘commercial shipper’. For the ‘dose’ dimension, no significant differences were detected in HEI score (‘feed the individual’: 61.4; ‘feed the family’: 62.5, difference: 1.1, 95%CI: −22.6, 24.9). For the ‘delivery’ dimension, no significant differences were detected in loneliness scores (‘dedicated driver’: 7.1, ‘commercial shipper’: 6.8, difference: 0.4, 95%CI: −1.5, 2.2).

Conclusion:

Providing MTM to the entire household does not necessarily improve the focal individual’s diet quality, relative to feeding that individual alone. MTM programs could choose which delivery strategy to adopt based on considerations other than participant mental health.

Trial Registration:

NCT06160973

Food insecurity and poor nutrition are widely recognized threats to health.^1–3 For this reason, both have increasingly become targets for interventions to improve health.¹ ‘Food is medicine’ interventions are healthcare interventions that provide healthy food resources to prevent or manage specific clinical conditions.⁴ One type of food is medicine intervention is medically tailored meals (MTM), in which fully prepared meals, tailored by a registered dietitian nutritionist to an individual’s medical needs, are delivered to a person’s home.¹ With MTM, the key goal is to increase diet quality in order to improve clinical outcomes.⁵ MTM may be particularly useful in management of cardiovascular conditions, and there is now a body of evidence supporting their use in congestive heart failure, hypertension, type 2 diabetes, along with other conditions.⁶ As one example, our group has previously conducted a randomized trial demonstrating that, compared with not receiving MTM, receipt of MTM leads to large improvements in diet quality and mental health outcomes.^7,8 Despite this growing evidence however, questions about optimal MTM intervention design remain unanswered.

One unanswered question involves the ‘dose’ of the intervention—should the intervention be provided for only the index individual, or for the person’s household as well? Because families share food, providing food only for the index individual, particularly in the presence of food insecurity, could risk ‘underdosing’ the intervention—leading to less improvement in diet quality than would be expected.^1,6 On the other hand, providing food for the entire household is more expensive, and given that the meals are medically tailored for a particular individual, others in the household may not want to share them. Moreover, providing MTM to the index individual frees up household budget space that others can use, which could indirectly address the problem without needing to provide additional food.^8,9 To date, there is little evidence regarding whether a ‘feed the household’ strategy improves diet more than a ‘feed the individual’ strategy.

A second unanswered question involves how medically tailored meals are delivered. Randomized trials in senior nutrition have demonstrated the benefits daily meal delivery with a home visiting program, which combines food provision and a social visit with the meal recipient.^10–12 However, MTM organizations typically use a lighter-touch approach with weekly meal delivery and a dedicated delivery driver who hands the food to the recipient or enters the home briefly to help put the food away, but does not have a social visit. Prior qualitative research suggests that such an approach, though not as intensive as a home visiting program, can still build rapport and provide a friendly face that may be helpful for those experiencing loneliness.⁸ On the other hand, advances in logistics technology have meant that there are more options than ever to deliver foods using commercial shippers. Though this means forgoing a consistent relationship, it can expand reach into areas that are not otherwise viable for MTM organizations to serve. Whether such a strategy differs from the dedicated driver approach with regard to loneliness and other psychological outcomes is unknown.

To help answer these questions, we conducted a factorial randomized comparative effectiveness trial. In this trial, all participants received MTM, but were randomly allocated to one of two interventions along a ‘dose’ (‘feed the family’ or ‘feed the individual’) dimension, and to one of two interventions along a ‘delivery’ (‘commercial shipper’ or ‘dedicated driver’) dimension. We hypothesized that the ‘feed the family’ strategy would produce better diet outcomes than the ‘feed the individual’ strategy, and that the ‘dedicated driver’ strategy would produce better loneliness outcomes than the ‘commercial shipper’ strategy.

Methods

Study Design

This was a 2×2 full factorial randomized trial with several goals.^13,14 The overall objective was to inform the design of subsequent MTM interventions by providing information on key intervention design features. One goal was to compare the effectiveness of different ways of providing MTM with regard to diet quality and loneliness. Another goal was to pilot data collection strategies, in preparation for larger studies planned in the future. We chose a factorial design to efficiently examine several intervention design features at the same time. We did not anticipate synergistic effects between the different strategies, although our analytic approach allowed for this possibility.¹⁴

Data Sharing

Deidentified trial data will be made publicly available through the AHA’s Precision Medicine Platform.

Eligibility, Recruitment, Enrollment, and Randomization

We enrolled adults (18 years and older) without plans to move from the study area, who had control over their dietary intake, and who could complete study activities in English. Potential participants were excluded if they were unable to store provided meals, lacked a telephone, or had medical diagnoses, such as psychosis, that would prohibit participation in study activities. Participants were recruited from the referral base of Community Servings, Inc. Community Servings is a not-for-profit community-based MTM provider in the New England region of the U.S. Their service area included Massachusetts, Rhode Island, and New Hampshire. Participants provided verbal informed consent. Recruitment began in February 2024 and ended in January 2025. Data collection was completed in May 2025. The study was prospectively registered on clinicaltrials.gov as NCT06160973. The institutional review board at the University of North Carolina at Chapel Hill approved the study.

After providing informed consent, participants completed a baseline assessment and were then randomized in a 1:1:1:1 ratio to one of four interventions, described in more detail below. Randomization tables were prepared by a study member not involved with enrollment, and concealed within REDCap.¹⁵ Randomization used a permuted block algorithm with blocks of size 2, 4, and 6 in unpredictable order, and was stratified by household size (one versus more than one individual in the household).¹⁶ Study assignments were not masked after randomization.

Interventions

There were two dimensions of intervention (‘dose’ and ‘delivery’) each with two variations, which resulted in four possible interventions. In all study arms, participants received a weekly home delivery of MTM, tailored according to Community Servings standard protocol, for 12 weeks, for a total of 12 possible deliveries.

The meals provided by Community Servings provide approximately 2/3 of the total calories of a 2,000 kilocalorie per day diet. Thus, each weekly delivery provides roughly 1340 kilocalories per day in the form of scratch-made dinners (5 per week), soup and protein-based salad lunch items (5 per week), snacks, desserts, and milk. Regarding sodium, entrees do not exceed 500 mg of sodium and soups are limited to 400mg of sodium. Total saturated fat is no more than 10% of calories provided (less than 9g of saturated fat per day). In addition, the meals are medically tailored to meet a variety of dietary needs and preferences of participants. Diet offerings included 6 “core diets” with 10 “modifiers” (each core diet can have up to two modifiers). The 6 core diets were: wellness, pediatric, renal, diabetic, cardiac, or maternal health. The 10 modifiers were: mild, soft, low fiber, pescatarian, low lactose, vegetarian, no fish, no red meat, high calorie/protein, and no nuts. An example medically tailored diet would be diabetic, soft, mild (for someone with acid reflux who needs carbohydrate controlled, easy-to-chew meals).

The study arms varied in the following ways. For the ‘feed the individual’ variation of the ‘dose’ dimension, the index individual received 10 meals per week. For the ‘feed the family’ variation of the ‘dose’ dimension, the index individual and other household members each received 10 meals per week (e.g., a household with an index individual and two other members would receive 30 meals per week). Participants in the ‘feed the individual’ arm were informed the meals were for the specific enrolled individual, and the individuals in the ‘feed the family arm’ were told that additional meals were being provided for household members to consume. Because household members beyond the enrolled individual were not enrolled in the study, they were not contacted by investigators and data were not collected regarding their experiences.

For the ‘commercial shipper’ variation of the ‘delivery’ dimension, meals were delivered using a commercial logistics company. For the ‘dedicated driver’ variation, the meals were delivered by a Community Servings driver in a dedicated delivery truck. The cost for one week of MTM for one person, using either a commercial shipper or dedicated driver, was the same, $185.80. The ‘dedicated driver’ variation represented standard practice for Community Servings. This is a lighter-touch approach than home-visiting programs often used for older adults.^10–12 The dedicated driver was required to deliver meals with a face-to-face hand-off and meals could not be left unattended. The specific level of interaction with a driver varies by individual preference and was not intended to be standardized. For some participants, it may be a simple handoff. For others, a delivery driver may enter the participant’s home to assist with putting the meals away. Because we did not conceptualize any particular standard for these interactions, unlike with dedicated home visiting programs, details of the interaction were not quantified. Meals delivered via a commercial shipper were not required to meet any level of personal interaction and were permitted to be left unattended at the participant’s doorstep (with ice packs to meet food safety quality control standards).

Measures

Study participants completed a standard questionnaire at baseline (prior to randomization) and at approximately 12 weeks (the end of the intervention), containing several validated measures. To make no assumptions of literacy, a trained study team member read all items aloud to participants over the telephone.

For the ‘dose’ dimension, the primary outcome was diet quality, as a key rationale for feeding the household and not just an index individual is that, because food is often shared in households, conditions of scarcity may result in ‘underdosing’ the food provided. That is, to ‘stretch’ the food available for the household, the index individual may share their meals with others. This would result in the index individual not consuming the intended amount of food, and thus diet quality may not improve as much as hoped.

Diet quality was assessed in several ways. One way was through an approach called DietID, which uses pictorial representations of typical food consumption and has been validated against 24-hour recalls.^17,18 In this pilot study, we sought to pilot this data collection approach because it offers lower respondent burden than standard 24-hour recalls. The DietID approach yields a Healthy Eating Index 2020 (HEI) score, which indicates the extent to which food consumption adheres to the 2020 Dietary Guidelines for Americans, on a 0 to 100 scale where higher scores indicate greater adherence.^17–19 We targeted a 6-point difference between intervention arms as the minimum clinically meaningful difference. DietID also yields component scores in areas such as ‘greens and beans’ and ‘total fruits’. Because we were unsure the extent to which participants would complete DietID, we also assessed dietary patterns using the Dietary Screener Questionnaire (DSQ), which produces estimated consumption amounts for fruits, vegetables, whole grains, dairy, and other food groups and nutrients.^20–22

For the ‘delivery’ dimension, the primary outcome was the De Jong Gierveld total loneliness score (range: 0 to 11 with greater scores indicating more loneliness).^23–25 We targeted a minimum clinically important difference of 2 points. Secondary outcomes included the social and emotional loneliness components of the De Jong Gierveld scale, the EQ-5D-5L^26,27 and EconQoL^28,29 measures of health-related quality of life, the 8-item PROMIS depression scale^30,31, and food insecurity as assessed by the 18-item USDA household food security survey module, with 30-day lookback period.³²

We also assessed intervention acceptability³³, and calculated a net promotor index³⁴, which indicates the extent to which participants would recommend the intervention to others.

Statistical Analysis

We report descriptive statistics, unadjusted outcomes, and adjusted, model-based, results. As per the statistical analysis section of the study protocol, included as supplementary material, our primary approach for statistical inference was to calculate marginal estimates (also called least-squares means or predictive margins) and contrast quantities of interest.^35,36 We used an ‘ANCOVA’ (analysis of covariance) approach^37,38, rather than a ‘change score’ approach^39,40 with linear models that included terms for an individual’s ‘dose’ dimension assignment, ‘duration’ dimension assignment, their interaction, and the baseline measure of the outcome. To improve statistical precision in this small study, models also included age, sex, race and ethnicity, education, health insurance, monthly income, household size, whether a participant was receiving Supplemental Nutrition Assistance Program (SNAP) benefits, number of hospitalizations in the six months prior to enrollment, and indicators for the history of diabetes mellitus, hypertension, congestive heart failure, coronary heart disease, myocardial infarction, stroke, chronic kidney disease, and dialysis receipt.⁴¹ To account for missing data, we used multiple imputation by chained equations, with estimates pooled across 100 imputed datasets.^42–44 Estimates followed the intention-to-treat principle.

Because intervention effects may have varied by household size, we also present results stratified by whether there was one or more than one person in the household. For hypothesis testing purposes, the primary outcome for the ‘dose’ dimension was HEI score, and the primary outcome for the ‘delivery’ dimension was De Jong Gierveld total loneliness score. A p-value < 0.05 indicated statistical significance. Statistical analyses were conducted in SAS version 9.4 and R version 4.4.3.

Results

103 individuals enrolled in the study, 93 were randomized, and 74 completed follow-up interviews at 12 weeks (Figure 1, Table S1). All 93 randomized individuals were analyzed as part of the intention-to-treat analyses. Fewer participants completed the DietID assessment than the main interview—55 of 93 at baseline and 33 of 74 at follow-up (Tables S2–S3). At baseline, participant’s mean age was 57.6 (SD: 12.9) years, 66.7% were female, and 48.9% were non-Hispanic white (Table 1, Table S4). 74.2% of participants reported a history of hypertension, 66.7% reported a history of diabetes mellitus, 16.7% reported a history of congestive heart failure, and 15.6% reported a history of stroke. Participants received, on average, 90% of intended meals (mean = 10.8 of 12, SD = 2.8, median = 12) (Table S5). The mean household size in the ‘feed the family’ group was 2.33 (SD: 1.58), which translates to providing an additional 13.3 meals per week, relative to a ‘feed the individual’ strategy, on average.

Figure 1:

CONSORT Diagram

CONSORT diagram depicting flow of individuals through the study. All randomized participants were analyzed as part of the intention-to-treat analyses, using multiple imputation to account for missing data.

Table 1:

Demographics of Study Sample

	Overall	Dose Intervention		Delivery Intervention
		Feed the Individual	Feed the Family	Commercial Shipper	Dedicated Driver
	N=93	N=47	N=46	N=50	N=43
	Mean (SD) or N (%)	Mean (SD) or N (%)	Mean (SD) or N (%)	Mean (SD) or N (%)	Mean (SD) or N (%)
Age, years	57.62 (12.90)	60.25 (13.37)	54.93 (11.96)	56.45 (14.63)	58.98 (10.55)
Sex
Male	29 (31.2)	16 (34.0)	13 (28.3)	16 (32.0)	13 (30.2)
Prefer Not to Say	2 (2.2)	2 (4.3)	0 (0.0)	1 (2.0)	1 (2.3)
Female	62 (66.7)	29 (61.7)	33 (71.7)	33 (66.0)	29 (67.4)
Race and Ethnicity
Non-Hispanic White	45 (48.9)	27 (58.7)	18 (39.1)	27 (54.0)	18 (42.9)
Non-Hispanic Black	25 (27.2)	9 (19.6)	16 (34.8)	10 (20.0)	15 (35.7)
Hispanic	16 (17.4)	6 (13.0)	10 (21.7)	11 (22.0)	5 (11.9)
Non-Hispanic American Indian/Alaska Native	1 (1.1)	1 (2.2)	0 (0.0)	1 (2.0)	0 (0.0)
Non-Hispanic Asian	1 (1.1)	0 (0.0)	1 (2.2)	0 (0.0)	1 (2.4)
Non-Hispanic Native Hawaiian or Pacific Islander	1 (1.1)	1 (2.2)	0 (0.0)	1 (2.0)	0 (0.0)
Not Reported	3 (3.3)	2 (4.3)	1 (2.2)	0 (0.0)	3 (7.1)
Highest Education
Grades 1–8	1 (1.1)	0 (0.0)	1 (2.2)	0 (0.0)	1 (2.3)
Grades 9–11	10 (10.8)	7 (14.9)	3 (6.5)	6 (12.0)	4 (9.3)
Grade 12 or GED	29 (31.2)	10 (21.3)	19 (41.3)	17 (34.0)	12 (27.9)
Some College	29 (31.2)	17 (36.2)	12 (26.1)	11 (22.0)	18 (41.9)
College Degree or Greater	24 (25.8)	13 (27.7)	11 (23.9)	16 (32.0)	8 (18.6)
Health Insurance
Private Health Insurance	16 (17.2)	7 (14.9)	9 (19.6)	11 (22.0)	5 (11.6)
Medicare	19 (20.4)	10 (21.3)	9 (19.6)	8 (16.0)	11 (25.6)
Medicaid	46 (49.5)	24 (51.1)	22 (47.8)	24 (48.0)	22 (51.2)
Medicare and Medicaid	11 (11.8)	6 (12.8)	5 (10.9)	7 (14.0)	4 (9.3)
Uninsured	1 (1.1)	0 (0.0)	1 (2.2)	0 (0.0)	1 (2.3)
Monthly Household Income, $	1983.14 (1432.93)	2067.13 (1440.06)	1902.89 (1437.67)	2255.43 (1586.17)	1684.92 (1192.17)
Employment Status
Full-time Employment	8 (8.6)	4 (8.5)	4 (8.7)	4 (8.0)	4 (9.3)
Part-time Employment	8 (8.6)	3 (6.4)	5 (10.9)	6 (12.0)	2 (4.7)
Not Working due to Retirement, Disability, Caregiving, or Full-time Student	63 (67.7)	33 (70.2)	30 (65.2)	31 (62.0)	32 (74.4)
Unemployed and Looking for Work	14 (15.1)	7 (14.9)	7 (15.2)	9 (18.0)	5 (11.6)
Household Characteristics
Total Household Size	2.11 (1.36)	1.89 (1.07)	2.33 (1.58)	1.96 (1.18)	2.28 (1.53)
Number of Individuals 60 or Older	0.60 (0.65)	0.63 (0.61)	0.57 (0.69)	0.57 (0.68)	0.63 (0.62)
Number of Individuals Under 18	0.31 (0.73)	0.28 (0.67)	0.33 (0.80)	0.26 (0.71)	0.36 (0.76)
Single Person Household	37 (39.8)	19 (40.4)	18 (39.1)	22 (44.0)	15 (34.9)
Health Literacy
Low Confidence Filling out Forms	59 (63.4)	30 (63.8)	29 (63.0)	36 (72.0)	23 (53.5)
Difficulty Understanding Written Medical Information	67 (72.0)	36 (76.6)	31 (67.4)	39 (78.0)	28 (65.1)
Receive Help Reading Medical Information	55 (59.1)	30 (63.8)	25 (54.3)	32 (64.0)	23 (53.5)
History of Diabetes Mellitus	62 (66.7)	33 (70.2)	29 (63.0)	34 (68.0)	28 (65.1)
History of Hypertension	69 (74.2)	33 (70.2)	36 (78.3)	37 (74.0)	32 (74.4)
History of Congestive Heart Failure	15 (16.7)	9 (19.1)	6 (14.0)	8 (16.7)	7 (16.7)
History of Coronary Heart Disease	10 (11.5)	5 (10.9)	5 (12.2)	7 (14.9)	3 (7.5)
History of Myocardial Infarction	11 (12.0)	3 (6.5)	8 (17.4)	5 (10.0)	6 (14.3)
History of Stroke	14 (15.6)	12 (26.1)	2 (4.5)	7 (14.3)	7 (17.1)
History of Chronic Kidney Disease	26 (28.9)	15 (31.9)	11 (25.6)	16 (33.3)	10 (23.8)
Receiving Dialysis	6 (6.5)	5 (10.6)	1 (2.2)	3 (6.0)	3 (7.0)
Hospitalizations in Past 6 Months	1.86 (5.47)	1.70 (5.91)	2.02 (5.04)	2.32 (6.98)	1.33 (2.86)
Emergency Department Visits without Hospitalization in Past 6 Months	1.41 (2.26)	1.34 (2.19)	1.48 (2.35)	1.34 (2.08)	1.49 (2.47)
Received Emergency Food in Past 30 Days	42 (45.7)	23 (48.9)	19 (42.2)	21 (42.9)	21 (48.8)
Received SNAP in Past Year	65 (70.7)	35 (74.5)	30 (66.7)	35 (70.0)	30 (71.4)
Receiving SNAP at Enrollment	61 (65.6)	32 (68.1)	29 (63.0)	35 (70.0)	26 (60.5)
Received WIC in Past Year	1 (1.1)	1 (2.1)	0 (0.0)	0 (0.0)	1 (2.3)
Receiving WIC at Enrollment	1 (1.1)	1 (2.1)	0 (0.0)	0 (0.0)	1 (2.3)
Traded-off Food for Medications in Past 3 Months	39 (41.9)	20 (42.6)	19 (41.3)	21 (42.0)	18 (41.9)
Traded-off Medications for Food in Past 3 Months	31 (33.3)	13 (27.7)	18 (39.1)	14 (28.0)	17 (39.5)
Reports Transportation Barriers	38 (40.9)	19 (40.4)	19 (41.3)	19 (38.0)	19 (44.2)
Reports Housing Instability	35 (37.6)	17 (36.2)	18 (39.1)	17 (34.0)	18 (41.9)
Reports Delaying Care due to Cost in Past 3 Months	23 (25.0)	11 (23.9)	12 (26.1)	10 (20.0)	13 (31.0)
Reports Cost Related Medication Underuse in Past 3 Months	33 (35.5)	20 (42.6)	13 (28.3)	18 (36.0)	15 (34.9)
Financial Strain (How Often Do Not Have Money to Pay the Bills)
Never	14 (15.1)	8 (17.0)	6 (13.0)	7 (14.0)	7 (16.3)
Rarely	8 (8.6)	6 (12.8)	2 (4.3)	6 (12.0)	2 (4.7)
Sometimes	31 (33.3)	15 (31.9)	16 (34.8)	21 (42.0)	10 (23.3)
Often	26 (28.0)	12 (25.5)	14 (30.4)	8 (16.0)	18 (41.9)
Always	14 (15.1)	6 (12.8)	8 (17.4)	8 (16.0)	6 (14.0)

GED = General Educational Development

SNAP = Supplemental Nutrition Assistance Program

WIC = Special Supplemental Nutrition Program for Women, Infants, and Children

At baseline, 71% reported food insecurity, which decreased to 50% at follow-up and did not differ meaningfully across intervention arms (Table 2, Tables S6–S16, Figures S1–S2). In this short-term, low-risk trial, no adverse events were reported.

Table 2:

Unadjusted Baseline Outcome Values from Survey Items

	Overall	Feed the Individual	Feed the Family	Commercial Shipper	Dedicated Driver
	N=93	N=47	N=46	N=50	N=43
	Mean (SD) or N (%)	Mean (SD) or N (%)	Mean (SD) or N (%)	Mean (SD) or N (%)	Mean (SD) or N (%)
Total Fruit and Vegetable Consumption, cups/day	2.65 (0.85)	2.69 (0.93)	2.62 (0.78)	2.67 (0.77)	2.63 (0.95)
Fruit and Vegetable Consumption Not Including Fries, cups/day	2.52 (0.90)	2.56 (0.98)	2.49 (0.82)	2.54 (0.79)	2.50 (1.02)
Total Vegetable Consumption, cups/day	1.59 (0.45)	1.61 (0.51)	1.57 (0.39)	1.57 (0.39)	1.61 (0.52)
Vegetable Consumption, Not Including Fries, cups/day	1.47 (0.49)	1.50 (0.57)	1.45 (0.41)	1.46 (0.42)	1.49 (0.57)
Fruit Consumption, cups/day	1.05 (0.67)	1.09 (0.78)	1.01 (0.55)	1.12 (0.79)	0.97 (0.50)
Dairy Consumption, cups per day	1.58 (0.71)	1.65 (0.82)	1.50 (0.57)	1.57 (0.66)	1.59 (0.78)
Added Sugar Consumption, teaspoons per day	14.05 (4.88)	14.82 (5.71)	13.31 (3.87)	14.20 (5.02)	13.88 (4.79)
Added Sugar Consumption from Sugar Sweetened Beverages, teaspoons per day	5.75 (3.51)	5.98 (3.45)	5.51 (3.60)	6.02 (3.88)	5.43 (3.04)
Whole Grain Consumption, ounces per day	0.71 (0.40)	0.73 (0.43)	0.69 (0.37)	0.74 (0.44)	0.68 (0.36)
Fiber Consumption, grams per day	16.56 (4.18)	17.25 (4.66)	15.90 (3.60)	16.56 (4.11)	16.56 (4.31)
Calcium Consumption, milligrams per day	957.12 (245.12)	987.94 (290.99)	927.93 (191.43)	952.31 (219.35)	962.49 (274.18)
Total De Jong Gierveld Loneliness Score	7.15 (3.14)	6.91 (3.28)	7.40 (3.00)	7.26 (3.13)	7.02 (3.18)
Social De Jong Gierveld Loneliness Score	3.32 (1.80)	3.17 (1.77)	3.47 (1.83)	3.38 (1.90)	3.24 (1.68)
Emotional De Jong Gierveld Loneliness Score	3.84 (1.83)	3.74 (1.94)	3.93 (1.72)	3.88 (1.73)	3.79 (1.96)
Depression Score	56.12 (10.15)	54.64 (10.37)	57.62 (9.79)	56.37 (10.94)	55.82 (9.27)
EQ-5D-5L Health Related Quality of Life	0.42 (0.37)	0.42 (0.37)	0.42 (0.38)	0.43 (0.39)	0.41 (0.36)
EconQoL Score	43.07 (6.86)	43.56 (7.59)	42.55 (6.05)	44.07 (6.00)	41.91 (7.64)
Food Security Score	4.81 (3.57)	4.79 (3.96)	4.83 (3.16)	4.40 (3.59)	5.28 (3.53)
Food Insecure	66 (71.0)	32 (68.1)	34 (73.9)	33 (66.0)	33 (76.7)

For dietary outcomes, mean baseline HEI score was 63.0 (SD: 22.4), mean total fruit and vegetable consumption was 2.65 (SD: 0.85) cups per day, and mean daily whole grain consumption was 0.71 (SD: 0.40) ounces per day (Tables S17–S20). At follow-up, the adjusted mean HEI score for the ‘feed the individual’ group was 61.4, and 62.5 for the ‘feed the family’ group (difference: 1.1, 95%CI −22.6 to 24.9, p = 0.92). For the ‘commercial shipper’ group, mean HEI was 62.3 and was 61.6 for the ‘dedicated driver’ group (difference: −0.7, 95%CI −22.7 to 21.3, p = 0.95). The mean adjusted HEI was 62.3 for the ‘feed the individual + commercial shipper’ group, 60.3 for the ‘feed the individual + dedicated driver’ group, 62.3 for the ‘feed the family + commercial shipper’ group, and 62.8 for the ‘feed the family + dedicated driver’ group. There were no meaningful differences between groups for the DSQ dietary indices (Figure 2, Table 3, Tables S21–S31). There were also no meaningful differences when stratifying dietary results by household size.

Figure 2:

Adjusted Results by ‘Dose’ Group

Forest plot of adjusted dietary screener questionnaire and loneliness and other participant-reported outcomes by assignment along the ‘dose’ dimension. The figure reports the adjusted difference in group means at 12 weeks after randomization. To improve statistical precision in this small study, models were adjusted for age, sex, race and ethnicity, education, health insurance, monthly income, household size, whether a participant was receiving Supplemental Nutrition Assistance Program (SNAP) benefits, number of hospitalizations in the six months prior to enrollment, and indicators for the history of diabetes mellitus, hypertension, congestive heart failure, coronary heart disease, myocardial infarction, stroke, chronic kidney disease, and dialysis receipt.

Table 3:

Adjusted Results from Dietary Screener Questionnaire by Dose

	Feed the Individual	Feed the Family	Difference (95% CI)
Total Fruit and Vegetable Consumption, cups/day	2.76	2.90	0.14 (−0.57 to 0.85)
Fruit and Vegetable Consumption Not Including Fries, cups/day	2.58	2.77	0.19 (−0.52 to 0.89)
Total Vegetable Consumption, cups/day	1.65	1.69	0.04 (−0.28 to 0.36)
Vegetable Consumption, Not Including Fries, cups/day	1.52	1.58	0.06 (−0.28 to 0.41)
Fruit Consumption, cups/day	1.16	1.27	0.11 (−0.47 to 0.69)
Dairy Consumption, cups per day	1.62	1.74	0.12 (−0.45 to 0.69)
Added Sugar Consumption, teaspoons per day	13.64	14.85	1.20 (−2.74 to 5.15)
Added Sugar Consumption from Sugar Sweetened Beverages, teaspoons per day	6.06	6.46	0.39 (−2.78 to 3.57)
Whole Grain Consumption, ounces per day	0.95	0.94	−0.01 (−0.53 to 0.50)
Fiber Consumption, grams per day	16.75	18.03	1.28 (−1.97 to 4.54)
Calcium Consumption, milligrams per day	978.98	990.98	12.00 (−174.81 to 198.81)

^*Results represent least-squares means adjusted estimates using multiple imputation to address missing data

To improve statistical precision in this small study, models were adjusted for age, sex, race and ethnicity, education, health insurance, monthly income, household size, whether a participant was receiving Supplemental Nutrition Assistance Program (SNAP) benefits, number of hospitalizations in the six months prior to enrollment, and indicators for the history of diabetes mellitus, hypertension, congestive heart failure, coronary heart disease, myocardial infarction, stroke, chronic kidney disease, and dialysis receipt.

For loneliness and other participant-reported outcomes, baseline De Jong Gierveld total loneliness score was 7.15 (SD: 3.14). At follow-up, adjusted mean De Jong Gierveld total loneliness score was 6.77 for the ‘commercial shipper’ group and 7.12 for the ‘dedicated driver’ group (difference: 0.35, 95%CI −1.52 to 2.23, p=0.71) (Figure 3, Table 4). Follow-up adjusted mean De Jong Gierveld total loneliness score was 6.51 for the ‘feed the individual’ group and 7.37 for the ‘feed the family’ group (difference: 0.86, 95%CI −0.97 to 2.69, p=0.35) (Table S32). There were no meaningful differences for social or emotional loneliness, depressive symptoms, or health-related quality of life. There were no meaningful differences across groups defined by their factorial combination, or in results stratified by household size (Tables S33–S39).

Figure 3:

Adjusted Results by ‘Delivery’ Group

Forest plot of adjusted loneliness and other participant-reported outcomes and dietary screener questionnaire results by assignment along the ‘delivery’ dimension. The figure reports the adjusted difference in group means at 12 weeks after randomization. Total De Jong Gierveld Score is bolded as the primary outcome for this study dimension. To improve statistical precision in this small study, models were adjusted for age, sex, race and ethnicity, education, health insurance, monthly income, household size, whether a participant was receiving Supplemental Nutrition Assistance Program (SNAP) benefits, number of hospitalizations in the six months prior to enrollment, and indicators for the history of diabetes mellitus, hypertension, congestive heart failure, coronary heart disease, myocardial infarction, stroke, chronic kidney disease, and dialysis receipt.

Table 4:

Adjusted Results from Loneliness and other Participant Reported Outcomes by Delivery Method

	Commercial Shipper	Dedicated Driver	Difference (95% CI)
Total De Jong Gierveld Loneliness Score	6.77	7.12	0.35 (−1.52 to 2.23)
Social De Jong Gierveld Loneliness Score	3.14	3.04	−0.10 (−1.14 to 0.94)
Emotional De Jong Gierveld Loneliness Score	3.61	4.05	0.44 (−0.66 to 1.54)
Depression Score	56.36	55.16	−1.20 (−7.72 to 5.32)
EQ-5D-5L Health Related Quality of Life	0.45	0.43	−0.02 (−0.28 to 0.24)
EconQoL Score	43.31	40.58	−2.72 (−9.28 to 3.83)
Food Security Score	3.19	3.71	0.52 (−1.39 to 2.44)

^*Results represent least-squares means adjusted estimates using multiple imputation to address missing data

Total De Jong Gierveld Loneliness scores range from 0 to 11 with greater scores indicating greater loneliness.

Social De Jong Gierveld Loneliness scores range from 0 to 5 with greater scores indicating greater social loneliness

Emotional De Jong Gierveld Loneliness scores range from 0 to 6 with greater scores indicating greater emotional loneliness

Depression scores are from the Patient-Reported Outcomes Measurement Information System (PROMIS) Short Form v1.0 with a T-score format (population mean = 50, population SD = 10) where greater scores indicating greater depressive symptoms.

EuroQol Five Dimension - Five Level (EQ-5D-5L) health utility score ranges from 0 to 1, with 1 representing perfect health

The EconQoL score measures economic quality of life with a T-score format (population mean = 50, population SD = 10) where higher scores indicate better economic quality of life

Food security is measured using the 18-item USDA Household Food Security Survey Module with 30-day lookback period (10-items when no children are in the household). Greater scores indicate worse food security.

To improve statistical precision in this small study, models were adjusted for age, sex, race and ethnicity, education, health insurance, monthly income, household size, whether a participant was receiving Supplemental Nutrition Assistance Program (SNAP) benefits, number of hospitalizations in the six months prior to enrollment, and indicators for the history of diabetes mellitus, hypertension, congestive heart failure, coronary heart disease, myocardial infarction, stroke, chronic kidney disease, and dialysis receipt.

Overall, participants found the intervention highly acceptable and reported a high net promotor index of 57, without meaningful differences across groups (Figures S3–S8, Tables S40–S45).

Discussion

In this factorial randomized comparative effectiveness trial, we did not find meaningful differences in diet outcomes when comparing a ‘feed the individual’ and a ‘feed the family’ dosing strategy for MTM. Further, we did not find meaningful differences in loneliness and other psychological outcomes when comparing a ‘commercial shipper’ and a ‘dedicated driver’ delivery strategy. We also did not find meaningful synergy across the dimensions. Food insecurity decreased from baseline in all groups, and all groups reported that the intervention was acceptable, with a high net promotor index.

This study expands prior literature on aspects of medically tailored meal delivery. A number of studies have now shown potential benefits for MTM on important outcomes such as hemoglobin A1c control in diabetes, heart failure hospitalizations, and HIV care.^6,45–47 This has led to increasing interest in standardizing MTM interventions for implementation at scale. However, many possible variations in the design of MTM programs are plausible. This study adds evidence that can help inform design decisions. This study saw increases in the consumption of healthful food groups from baseline to follow-up, as measured by the DSQ, a well-established measure of dietary intake.²¹ This study also piloted the use of a newer dietary assessment tool, the DietID measure. Although potentially offering lower respondent burden than standard 24-hour recalls, many participants in our sample were not able to complete it owing to technology barriers. Further, the HEI scores that were measured in this study seemingly differed from those measured in a prior randomized trial we conducted in a similar setting.⁷ Whether that difference relates to differences between the instruments used or differences in the study samples is not clear. However, a recent trial from a different research group did find that HEI scores estimated using the DietID approach correlated only modestly with those estimated using 24-hour recalls in their study.⁴⁸

The findings of this study have several implications. First, the finding that a ‘feed the family’ strategy did not improve diet outcomes for an index individual relative to the ‘feed the individual’ strategy has important implications for intervention design, especially as a ‘feed the family’ strategy is more costly. This finding likely speaks to several nuances of MTM interventions within household food consumption. Though food is typically shared within households, to the extent the medically tailored diet varies from the needs and preferences of others in the household, sharing of MTM may be less desirable. Further, the design of the meal itself, as single-serving tray, may also influence sharing behavior. Moreover, by providing a substantial part of one individual’s weekly food intake, household food budget space is freed up, allowing for spillover impacts for others in the household (i.e., more money for their food) even in the absence of directly providing food for household members. It is important to remember, however, that this study specifically investigated an MTM program. Other food is medicine interventions, such as food subsidies or in-kind provision of unprepared grocery items¹, may have different dynamics as it may be easier in some cases to share those resources across a household while still according with food preferences. Also, this study did not evaluate the impact of receiving meals on other household members themselves. Our rationale for this decision was that the key justification for a ‘feed the family’ strategy, per se, is that not doing so would undercut the benefits for the index individual. However, if there is more than one individual within a household who can benefit from MTM, it would of course make sense to enroll them in the intervention individually. Thus, the results should not be interpreted as suggesting that only one individual per household should be enrolled in MTM. Rather, the implication is simply that intervention eligibility should be considered on a person-by-person basis.

A second implication of the fact that we did not find meaningful differences between a ‘commercial shipper’ and ‘dedicated driver’ strategy is that MTM providers could choose which delivery strategy to adopt based on other considerations. In dense areas, it may be more economical to employ a dedicated driver. On the other hand, for areas with lower density, or as organizations expand their service area beyond an initial hub, commercial shipping may be a feasible way to broaden reach. For larger organizations with multiple service lines, a mix of approaches may make the most sense.

The results of this study should be interpreted in light of several limitations. Missing data, particularly with regard to the dietary assessment used to calculate HEI scores, may have affected estimated outcomes. Piloting data collection instruments was a goal of the study. Learning that the dietary assessment tool used to calculate HEI scores was difficult for many participants to complete was valuable for planning subsequent studies, but does complicate interpreting results of this study. We did attempt to address the potential impact of missing data by using multiple imputation, but we nevertheless recognize missing data as an important limitation of this study. Next, this was a small, short-term study meant to inform design decisions for MTM interventions. Because meal consumption had previously been shown to be high^7,8, we did not track meal consumption in this study, to minimize respondent burden. The purpose of food is medicine interventions is to improve health, but this study was not intended to examine the efficacy of MTM, relative to usual care or another intervention, on the key health outcomes MTM are proposed for. Next, while adequate for the primary study purposes, the sample size was small when examining subgroups and interactions between study arms, and thus may have lacked power to detect small differences. However, we did not see a signal of meaningful differences based on the magnitude of the differences in outcomes, and our analytic approach allowed for the possibility of interactions. Next, although no difference in key outcomes in this study had a p-value <0.05, how to interpret p-values when multiple outcomes are of interest is an important topic.⁴⁹ ‘Splitting’ of alpha is one recommended strategy for addressing these issues, and can help control Type I error rate in cases in which one or more differences in outcomes have p-values < 0.05 (and thus might be thought to be nominally statistically significant if not considering multiple outcomes).⁴⁹ Finally, the study duration was 12 weeks, so we do not know if results would have differed over longer timeframes.

Conclusions

In this factorial randomized comparative effectiveness trial examining different versions of medically tailored meal interventions, we did not find differences in diet outcomes when comparing a ‘feed the individual’ to a ‘feed the family’ dosing strategy. We also did not find differences in loneliness or other psychological outcomes when comparing a ‘commercial shipper’ to a ‘dedicated driver’ delivery strategy. These results can help inform future medically tailored meal interventions, which can be an important part of the food is medicine approach to improving health by overcoming barriers to healthy eating.

Supplementary Material

Figures S1–S8

Tables S1–S45

What is Known?

• Medically tailored meals (MTM) can improve diet quality and health outcomes

• MTM are typically provided only to a specific individual, but many are interested in providing meals to an entire household

• MTM are typically provided by a dedicated driver, but using a commercial shipper might increase reach

What This Study Adds

• Providing MTM to the entire household did not improve diet quality, relative to providing MTM only for a single individual, so MTM organizations do not necessarily need to provide meals to the entire household

• Using a dedicated delivery driver did not improve loneliness, relative to using a commercial shipper, so MTM organizations could choose which delivery strategy to adopt based on other considerations

Non-standard Abbreviations and Acronyms

FAME-F

Food as Medicine for Families

MTM

Medically tailored meals

HEI

Healthy Eating Index 2020

EQ-5D-5L

EuroQual 5 Dimension 5 Level

PROMIS

Patient Reported Outcome Measurement Information System

ANCOVA

Analysis of covariance

DSQ

Dietary Screener Questionnaire

SNAP

Supplemental Nutrition Assistance Program