Interventions to Address Food Insecurity Among Adults in Canada and the US: A Systematic Review and Meta-analysis.

Importance: Inadequate access to food is a risk factor for poor health and the effectiveness of federal programs targeting food insecurity, such as the Supplemental Nutrition Assistance Program (SNAP), are well-documented. The associations between other types of interventions to provide adequate food access and food insecurity status, health outcomes, and health care utilization, however, are unclear. Objective: To review evidence on the association between food insecurity interventions and food insecurity status, clinically-relevant health outcomes, and health care utilization among adults, excluding SNAP. Data Sources: A systematic search for English-language literature was performed in PubMed Central and Cochrane Trials databases (inception to January 23, 2020), the Social Interventions Research and Evaluation Network database (December 10, 2019); and the gray literature using Google (February 1, 2021). Study Selection: Studies of any design that assessed the association between food insecurity interventions for adult participants and food insecurity status, health outcomes, and health care utilization were screened for inclusion. Studies of interventions that described addressing participants' food needs or reporting food insecurity as an outcome were included. Interventions were categorized as home-delivered food, food offered at a secondary site, monetary assistance in the form of subsidies or income supplements, food desert interventions, and miscellaneous. Data Extraction and Synthesis: Data extraction was performed independently by 3 reviewers. Study quality was assessed using the Cochrane Risk of Bias Tool, the ROBINS-I (Risk of Bias in Non-Randomized Studies of Interventions) tool, and a modified version of the National Institutes of Health's Quality Assessment Tool for Before-After Studies With No Control. The certainty of evidence was based on GRADE (Grading of Recommendations Assessment, Development, and Evaluation) criteria and supplemented with mechanistic and parallel evidence. For outcomes within intervention categories with at least 3 studies, random effects meta-analysis was performed. Main Outcomes and Measures: Food insecurity (measured through surveys; eg, the 2-item Hunger Vital Sign), health outcomes (eg, hemoglobin A1c), and health care utilization (eg, hospitalizations, costs).
Results: A total of 39 studies comprising 170 605 participants were included (8 randomized clinical trials and 31 observational studies). Of these, 14 studies provided high-certainty evidence of an association between offering food and reduced food insecurity (pooled random effects; adjusted odds ratio, 0.53; 95% CI, 0.33-0.67). Ten studies provided moderate-certainty evidence of an association between offering monetary assistance and reduced food insecurity (pooled random effects; adjusted odds ratio, 0.64; 95% CI, 0.49-0.84). There were fewer studies of the associations between interventions and health outcomes or health care utilization, and the evidence in these areas was of low or very low certainty that any food insecurity interventions were associated with changes in either. Conclusions and Relevance: This systematic review with meta-analysis found that providing food and monetary assistance was associated with improved food insecurity measures; however, whether it translated to better health outcomes or reduced health care utilization was unclear. Copyright 2021 Oronce CIA et al. JAMA Health Forum.

Introduction

Food insecurity, which is defined as limited or uncertain access to adequate food owing to resource limitations,[1] is an important social risk factor for poor health. Food insecurity is linked to a range of health conditions and adverse outcomes among adults, including diabetes, hypertension, worsened depressive symptoms, and greater activity limitations.[2,3] Furthermore, individuals with food insecurity are more likely to experience an emergency department visit, inpatient admission, and higher health care costs.[4] Food insecure Medicare beneficiaries incur $5527 higher health care expenditures annually compared with those who are food secure.[4] Nationally, food insecurity is estimated to contribute nearly $52 billion in excess health care spending.[5] The COVID-19 pandemic and its economic consequences have exacerbated the problem, as the prevalence of food insecurity increased from 11.1% of households in 2018 to 22.7% in April 2020.[6]

Given the scope of unmet social needs and the growth of value-based payment models that prioritize population health, there is wide interest in addressing food insecurity. For example, integrated delivery systems, hospitals, and payers have invested in programs that identify patients with food insecurity to connect them with resources.[7,8,9] Many have focused on enrolling eligible patients into the Supplemental Nutrition Assistance Program (SNAP), for which a robust evidence base demonstrates strong positive impact.[2,10,11,12] Despite the proliferation of other types of interventions addressing food insecurity, little is known about their effectiveness. To better understand the scope of interventions targeting food insecurity among adults within and outside of health care settings, as well as their association with outcomes, this systematic review with meta-analysis builds on prior work to evaluate the evidence of interventions across all settings, while excluding well-established public food assistance programs, such as SNAP.

Methods

This systematic review of randomized clinical trials and observational studies followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline. The study topic was developed in consultation with the sponsor, Humana Inc. The study was registered in PROSPERO (CRD42020172486). The University of California Los Angeles Institutional Review Board deemed this study exempt and waived informed consent because the study used only previously published research and did not meet the criteria found in the Regulations for the Protection of Human Subjects (45 CFR §46).

Data Sources and Searches

The search strategy, detailed in eAppendix 1 of the Supplement, was created by an experienced reference librarian using a broad set of terms related to food insecurity interventions. We conducted English-language searches in PubMed Central, Cochrane Trials, and Academic Search Complete, from inception to January 23, 2020; and in the Social Interventions Research and Evaluation Network (SIREN) database on December 10, 2019. The SIREN database is a University of California San Francisco project to collect, summarize, and disseminate research on interventions that address the social determinants of health.[13] For our search of SIREN, we used the social determinant “food/hunger” to filter results. We also performed a supplemental search of the gray literature using Google on February 1, 2021.

Study Selection

Three team members (C.I.O., I.M.L., and P.G.S.) independently screened the titles of retrieved citations. For titles deemed relevant by at least 1 person, abstracts were then screened in triplicate by team members. All disagreements were reconciled through group discussion. Full-text review was conducted in triplicate by 3 independent team members (C.I.O., I.M.L., and P.G.S.), with any disagreements being resolved through discussion. To be included, a study had to evaluate a food insecurity intervention and include health outcomes or food insecurity as an outcome measured at the person-level. Food insecurity was measured with self-reported surveys, most often with the US Department of Agriculture Food Security Survey Module or a short form of the module. We defined a food insecurity intervention as a program or policy that either directly addresses food needs or improves the ability to obtain food.

In consultation with the study sponsor, we did not require that a food insecurity intervention originate in a health care setting as long as it or something similar could be plausibly implemented by the health care system. We excluded studies of SNAP and similar programs, studies with fruit and vegetable intake as the sole outcome, and studies focusing on children and adolescents, as the focus of the sponsor was adults with food insecurity. Because our focus was on interventions applicable to the US, we also excluded studies conducted in low- and middle-income countries. A full description of the exclusion and inclusion criteria is available in the eAppendix 2 and eAppendix 3 of the Supplement.

Data Extraction and Quality Assessment

Data extraction was completed in triplicate (C.I.O., I.M.L., and P.G.S.). All discrepancies were resolved with full group discussion. We abstracted data on the following: study design, sample size, enrolled population, intervention, and outcomes measured. To assess risk of bias, we used the Cochrane Risk of Bias Tool,[14] ROBINS-I (Risk of Bias in Non-Randomized Studies of Interventions) tool,[15] or an adaptation of the National Institutes of Health Quality Assessment Tool for Before-After Studies With No Control.[16]

Data Synthesis and Grading

We grouped interventions into 5 broad categories: (1) food delivered to the home (eg, Meals on Wheels), (2) food offered at a secondary location (eg, a senior center), (3) monetary assistance as means-tested subsidies or similar programs (eg, tax credits or health insurance), (4) programs to address food deserts, and (5) miscellaneous interventions. We grouped outcomes into 3 domains: (1) food insecurity, (2) health outcomes (eg, blood pressure, hemoglobin A1c [HbA1c]), and (3) health care utilization. When study instruments varied, we combined food insecurity results according to the study-defined measures of food insecurity and aggregated the categories of low or very low food security. Summary statistics were extracted for each intervention and control group (or before-after intervention) for each study and outcome. To compare between group differences, results were converted to effect sizes (odds ratios [ORs] for binary outcomes and standardized mean differences [SMDs] for continuous outcomes) and plotted on a common measure. After confirming via meta-regression that 6 studies using a randomized trial design did not report systematically different effect sizes compared with the 24 observational studies, we pooled results within the categories of providing food and monetary assistance, respectively. When sufficient studies existed (≥3 studies) within each outcome and intervention category, studies were pooled using a random effects meta-analysis.[17] Outliers were assessed and removed from all pooled results.

Statistical Analysis

Tests of heterogeneity were performed using the I2 statistic.[18] Values of the I2 statistic close to 100% represent high degrees of heterogeneity. Begg rank correlation[19] and Egger regression asymmetry test[20] were used to examine publication bias. Exploratory metaregressions, which did not influence our conclusions, are described in the eAppendix 3 of the Supplement. P values were 2-tailed and statistical significance was defined as P = .05. All analyses were conducted in R, version 4.0.2 (The R Foundation for Statistical Computing).[21] We based the certainty of the evidence ratings on factors considered in the GRADE (Grading of Recommendations Assessment, Development, and Evaluation) system,[22] and supplemented these with other factors (eg, mechanistic and parallel evidence) as proposed by Howick and colleagues[23] and used by the National Academy of Medicine.[24]

Results

Description of the Evidence

We identified 4807 potentially relevant citations, with 150 undergoing full-text review and 111 of these being excluded (see eAppendix 2 of the Supplement). A total of 39 publications, including 170 605 participants, met inclusion criteria (Figure 1). Of these 39 studies, 8 were RCTs,[25,26,27,28,29,30,31,32] 1 was a quasi-experimental study,[33] 12 were before-after studies without controls,[34,35,36,37,38,39,40,41,42,43,44,45] 9 were before-after studies with controls,[46,47,48,49,50,51,52,53,54] 1 was a time-series study,[55] and 8 were cross-sectional studies.[56,57,58,59,60,61,62,63] Thirty-four studies were of interventions in the US, and 5 were of interventions in Canada.

Figure 1.

PRISMA Flow Diagram of Study Selection

a After deduplication.

b Not mutually exclusive.

Exclusion terminology: metrics refers to how food insecurity was measured; out-of-scope studies were those of hospitalized patient populations, indigenous populations in Canada, or populations that were non-Canada/US; linkage refers to screening and referral with the outcome being how many patients completed the referral process; blended SDOH were interventions addressing multiple social determinants with no separate outcome for food insecurity; healthy eating defines interventions and/or primary outcomes that promote healthy eating; SDOH denotes the Social Determinants of Health; SIREN, Social Interventions Research and Evaluation Network database; and SNAP, Supplemental Nutrition Assistance Program.

Across RCTs, most studies were judged as having a high risk of bias from outcome assessment, incomplete outcome data, and selective reporting of outcomes (eTable 1 in the Supplement). Half of the observational studies were judged to have a high risk of bias owing to missing data (eTable 2 in the Supplement). Most of the 13 before-after studies without controls had small sample sizes and high rates of loss to follow-up and thus were judged to have a high risk of bias (eTable 3 in the Supplement).

Associations With Food Insecurity

Food Delivered to Home

There were 6 studies examining the association between food provision delivered to the participant’s home and food insecurity status[25,34,47,48,49,63] (Table 1). One was a small randomized trial of medically tailored meals in which dieticians designed meals to address patients’ nutritional needs based on their medical conditions (eg, diabetes or HIV). Among those receiving medically tailored meals, 42% were food insecure vs 62% of those who were not receiving home-delivered medically tailored meals, compared with the baseline of 71% to 80%.[25] The remaining 5 interventions, which included programs, such as the Older Americans Act Nutrition Program and Meals on Wheels, delivered nonmedically tailored food to participants’ homes or places of shelter.[34,47,48,49,63] Three of these observational studies found that home delivered food interventions were associated with reduced food insecurity,[34,47,48] while 2 studies found that they were not.[49,63] Full details of these studies are described in the eAppendix 3 of the Supplement.

Table 1.

Studies of Interventions to Address Food Insecurity Among Adults

Source	Study design	Intervention	Baseline FI %	Participants, No.		Outcomes measured
				Enrolled	Follow-up	FI	Health care use	Health
Providing food delivered to home
Medically tailored meals
Berkowitz et al,[25] 2019	RCT	Home delivery of medically tailored meals	76	44	42	Yes	NR	Better dietHRQOL (various measures)HypoglycemiaHbA1cBlood pressureLipidsBMIDepressive symptoms
Berkowitz et al,[33] 2019	Retrospective matched cohort	Home delivery of medically tailored meals	NR	1020	1020	NR	Health care use and costs	NR
Gurvey et al,[46] 2013	Before-after (control group only for after)	Home delivery of medically tailored meals	NR	698	NR	NR	Health care use and costs	NR
All other kinds of foods provided
Chatterjee et al,[49] 2018	Before-after with control	Meals delivered to shelter	88	33	12	Yes	NR	Better dietObesity status (BMI)HRQOL
Frongillo et al,[47] 2010	Before-after with control	Meals delivered to home	NR	212	69	Yes	NR	Better dietWeight
Lee et al,[48] 2011	Before-after with control	Meals delivered to home	51	NR	583	Yes	NR	HRQOL
Mabli et al,[63] 2017	Cross-sectional	Meals delivered to home	23	504	NR	Yes	NR	LonelinessBetter dietDepressive symptoms
Wright et al,[34] 2015	Before-after	Meals delivered to home	40	62	51	Yes	NR	Nutritional statusQuality of life (well-being)Loneliness
Providing food at a secondary site
Medically tailored meals
Palar et al,[35] 2017	Before-after	Medically tailored meal distributed at single site	60	72	52	Yes	ED visitsHospital stays	Better dietBMIDepressive symptomsSubstance useHbA1cFasting glucoseART adherence
Medically tailored groceries (ie, “diabetic diet,” not individually prescribed)
Aiyer et al,[36] 2019	Before-after	Prescription for client-selected food at food pantry	100	242	42	Yes	Costs	Better diet
Cheyne et al,[37] 2020	Before-after	Diabetes-appropriate food, text message support, diabetes prevention program	43.6	244	192	Yes	NR	Physical activityBetter dietBMIHRQOLDepressive symptoms
Seligman et al,[38] 2015	Before-after	Medically tailored food box	83	687	NR	NR	NR	Better dietHbA1cDiabetes self-efficacyHypoglycemiaMedication nonadherence
Seligman et al,[26] 2018	RCT	Medically tailored food box	74	568	423	Yes	NR	HbA1cBetter dietDiabetes self-care proceduresDepressive symptomsHypoglycemiaMedication nonadherence
Wetherill et al,[39] 2019	Before-after	Prescription for food boxes provided at clinic	87	80	43	Yes	NR	Better dietBlood pressure
“Healthy eating” groceries (ie, more fruits and vegetables)
Feinberg et al,[45] 2018	Before-after	Food pantry at clinic	100	95	NR	NR	Costs	MortalitySerious diabetes complicationsHbA1cLipidsGlucose
Ferrer et al,[27] 2019	RCT	Food pantry at clinic, dietician-led patient education, community health worker	100	58	43	NR	NR	HbA1cBMIBetter diet
Feuerstein-Simon et al,[28] 2019	RCT	Food boxes provided at work	27	60	54	Yes	NR	Better diet
Martin et al,[29] 2013	RCT	Client-selected food at food pantry	84	241	138	Yes	NR	Better diet
All other kinds of food provided
Khan et al,[40] 2019	Before-after	Food boxes distributed at community sites	85	NR	93	Yes	NR	NR
Lee et al,[48] 2011	Before-after with control	Meals provided at congregate settings	39	NR	134	Yes	NR	HRQOL
Mabli et al,[63] 2017	Cross-sectional	Meals provided at congregate settings	16	596	NR	Yes	NR	LonelinessBetter dietDepressive symptoms
Wilkinson et al,[55] 2019	Time-series	Food boxes distributed at charitable community site	NR	50	45	NR	NR	Blood pressureHeart rateWeight
Monetary assistance
Not specifically for food (eg, Medicaid expansion, child tax credit)
Andrade et al,[50] 2019	Before-after with control	EITC Periodic Payment Pilot	34	443	278	Yes	NR	NR
Brown et al,[56] 2019	Serial cross-sectional	Canada Child Benefit	15-16a	41 455	41 455	Yes	NR	NR
Gubits et al,[30] 2018	RCT	Rental subsidies	NR	2282	1784	Yes	NR	NR
Himmelstein et al,[57] 2019	Serial cross-sectional	Insurance expansion (ACA)	17.5	16 813	7291	Yes	NR	NR
Ionescu-Ittu et al,[58] 2015	Serial cross-sectional	Canada Universal Child Care Benefit	16	29 458	11 046	Yes	NR	NR
Li et al,[59] 2016	Serial cross-sectional	Increase in Canada welfare benefits, rental subsidies	11	58 656	58 656	Yes	NR	NR
Londhe et al,[60] 2019	Serial cross-sectional	Insurance expansion (ACA)	14	361 countiesb	NR	Yes	NR	NR
McIntyre et al,[61] 2016	Serial cross-sectional	Increase in Canada senior pensions	29	8019	8019	Yes	NR	NR
Schmidt et al,[62] 2016	Serial cross-sectional	US federal safety-net, public assistance programs	33	NR	NR	Yes	NR	NR
Sonik et al,[53] 2019	Before-after with control	Supplemental Security Income	28	3933 (77 intervention, 3856 matched controls)	3933 (77 intervention, 3856 matched controls)	Yes	NR	NR
Specifically for food (vouchers for farmer’s markets/food banks)
Durward et al,[41] 2019	Before-after	Matching of SNAP benefits	69	339	138	Yes	NR	NR
Interventions targeting food deserts
Cueva et al,[42] 2018	Before-after	Mobile grocery store	57	101	92	Yes	NR	Better diet
Richardson et al,[51] 2017	Before-after with control	Full-service supermarket	33	NR	831	Yes	NR	HRQOLChronic conditions
Miscellaneous interventions
Berkowitz et al,[44] 2018	Before-after	Social needs screening with active outreach	40	141	138	Yes	NR	Better diet
Carney et al,[43] 2012	Before-after	Community garden	15.6	38	32	Yes	NR	NR
Eicher-Miller et al,[31] 2009	RCT	Nutritional education	40	236	219	Yes	NR	NR
Lohse et al,[32] 2015	Before-after (implemented as RCT but analyzed as a before-after without control)	Nutritional education	49	288	155	Yes	NR	NR
Phojanakong et al,[54] 2020	Before-after with control	Financial literacy education and support	53	372	208	Yes	NR	NR
Roncarolo et al,[52] 2016	Before-after with control	Alternative interventions, eg, community kitchens, gardens	54-90	824	450	Yes	NR	HRQOL

Reported separately for populations with and without children.

Multiple analyses with different sample population sizes.

Abbreviations: ACA, Affordable Care Act; ART, antiretroviral therapy; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); ED, emergency department; EITC, Earned Income Tax Credit; FI, food insecurity; HbA1c, hemoglobin A1c; HRQOL, health-related quality of life; NR, not reported; RCT, randomized clinical trial; SNAP, Supplemental Nutrition Assistance Program.

Food at a Secondary Site

Ten studies examined the association between food insecurity status and interventions where food was provided at a secondary location (Table 1). These were differentiated by degree of tailoring toward the patients’ comorbidities—medically tailored meals, medically tailored or appropriate food boxes, or other (no further individualization). One intervention provided medically tailored meals to individuals living with HIV or diabetes, and 4 studies evaluated medically tailored grocery interventions, such as prescriptions to a food pantry and predesigned boxes of fresh groceries and dry goods. Nearly all were tailored to populations with diabetes, prediabetes, or obesity, and were often bundled with other services to support self-management of chronic disease. Aside from 1 RCT,[26] these were before-after studies without control groups.[35,36,37,39] The other 5 studies that provided food at a secondary site were heterogeneous in type of intervention (Table 1). Two of these studies evaluated interventions that encouraged healthy food intake (ie, more fruits and vegetables),[28,29] 1 provided a supplemental food box for older adults,[40] and 2 provided meals in congregate settings for older adults.[48,63] In general, these interventions were associated with reductions in food insecurity (further details are available in the Supplement).

Thirteen of 16 interventions (81%) that provided food to participants and reported food insecurity outcomes used measures that were the same or similar. Although interventions differed across studies, we considered them all to be assessing the construct “providing regular food.” The random effects pooled estimate across these 13 interventions was an OR of 0.53 (95% CI, 0.36-0.77) in food insecurity (Figure 2), which shows an association with an absolute reduction in food insecurity of 15% given the weighted baseline average prevalence across these studies. There was significant heterogeneity (I2 statistic = 84.5%). There was disagreement among tests for possible publication bias, with Egger’s regression test being statistically significant (z  = -3.45; P = .001) but Begg’s rank correlation test being insignificant (Kendall τ, -0.359; P = .10).

Figure 2.

Association Between Food Insecurity and Intervention Type, by Studya

Odds ratios indicate the difference in the odds of being food insecure in the intervention vs the control group. The diamond marker indicates the pooled estimate and 95% CI for each type of intervention; RCT denotes randomized clinical trial.

a The study by Aiyer et al[36] was an outlier and was not included in the pooled analysis.

Monetary Assistance

Eleven studies evaluated the association between monetary assistance (eg, Medicaid Expansion, child tax credits) and outcomes (Table 1). Most of these interventions were not designed primarily as food insecurity interventions. In the only RCT, which enrolled 2282 families staying in emergency shelters, long-term rent subsidies were associated with a 10 percentage point increase in food security at 37 months.[30] The remainder were observational studies examining expansions or modifications to safety net programs, ranging from child tax credits to Medicaid eligibility. Nine interventions were associated with statistically significant decreases in food insecurity across multiple populations,[25,30,41,50,53,57,58,59,61] while 2 did not observe a difference.[56,60] Ten studies had similarly measured outcomes to permit pooling.[30,41,50,53,56,57,58,59,60,61] The random effects pooled estimate across these 10 studies was an OR of 0.64 (95% CI, 0.49-0.84) in food insecurity (Figure 2), which shows an association with an absolute reduction in food insecurity of 8% given the weighted baseline average prevalence across these studies. There was significant heterogeneity (I2 statistic = 98.8%). There was no statistical evidence of publication bias.

Food Deserts

Two studies used before-after designs to evaluate interventions to address food deserts.[42,51] One of these studies[51] was associated with a decrease of 12 percentage points compared with the control after 1 year (Figure 2).

Miscellaneous

The miscellaneous category comprised 1 RCT and 5 before-after analyses, which included interventions such as educational programs and community gardens (Table 1).[31,32,43,44,52,54] Three of the 5 before-after studies had similar outcomes that were associated with improvements in food insecurity[43,52,54] (Figure 2).

Associations With Health Outcomes

Studies that reported health outcomes were primarily food provision interventions. Fifteen reported various health outcomes: 6 reported HbA1c among patients with diabetes,[25,26,27,35,38,45] 7 reported self-reported quality of life or health status measures,[25,34,37,48,49,51,52] 6 reported body mass index (calculated as weight in kilograms divided by height in meters squared; BMI),[25,27,35,37,49,55] 5 reported depressive symptoms,[25,26,35,37,63] 3 reported blood pressure,[25,39,55] and 2 reported lipid panel results[25,45] (Table 1).

The 6 studies examining HbA1c provided food and 3 of these were RCTs. Two of these studies were of medically tailored meals,[25,35] 2 of diabetes-appropriate food boxes,[26,38] and 2 were clinic-based food pantry interventions.[27,45] While 5 studies demonstrated associations with small reductions in HbA1c, only 2 reached statistical significance, including a small RCT that provided food with dietician-led patient education and home visits from a community health worker.[27] In the random effects pooled analysis, provision of food was associated with an SMD in HbA1c of −0.13 (95% CI, −0.34 to 0.09; Figure 3). There was heterogeneity (I2 statistic = 67.8%). There was no evidence of publication bias.

Figure 3.

Association Between Food Insecurity Interventions and Health Outcomes by Study

SMD indicates standardized mean difference for each outcome. The diamond markers indicate pooled estimates and 95% CIs for each health outcome. I2 indicates the I2 statistic, and higher I2 values (closer to 100%) indicate greater study heterogeneity. BMI, denotes body mass index (calculated as weight in kilograms divided by height in meters squared); RCT, randomized clinical trial; HbA1c, hemoglobin A1c; and HRQOL, health-related quality of life.

Seven studies evaluated interventions with respect to health-related quality of life (HRQOL) or health status measures. Five studies evaluated food provision interventions,[25,34,37,48,49] 1 was a food desert intervention,[51] and 1 was a miscellaneous intervention.[52] Four of the 7 studies had similar self-reported HRQOL measures. In the random effects pooled analysis, these interventions were associated with an SMD of −0.28 (95% CI, −0.56 to 0.00) favoring improved health status (Figure 3). There was moderate heterogeneity (I2 statistic = 43.0%). There was no evidence of publication bias. One study reported its outcomes as dichotomous and could not be included in the pooled analysis, observing a significant decrease in fair or poor health from 74% to 60% after 6 months.[37]

Six observational studies evaluated the association between food provision interventions and BMI.[25,27,35,37,49,55] Four of the 6 studies had data and outcomes that were sufficiently similar to support pooling.[35,37,49,55] In the random effects pooled analysis, the provision of food was associated with an SMD in BMI of −0.05 (95% CI, −0.20 to 0.11) (Figure 3). There was no heterogeneity (I2 statistic = 0.0%). There was no evidence of publication bias. The 2 studies excluded from the pooled analysis did not observe a statistically significant association.[25,27]

Five studies reporting depressive symptoms included interventions that provided food.[25,26,35,37,63] Three of the 5 studies reported similar scales of depressive symptoms sufficient to support pooling.[35,41,63] In the random effects pooled analysis, the provision of food was associated with an SMD in depressive symptoms of −0.04 (95% CI, −0.19 to 0.11) (Figure 3). There was no heterogeneity (I2 statistic = 9.8%). There was no evidence of publication bias. The 2 studies excluded from the pooled analysis did not observe an association with depressive symptoms.[26,37]

Four studies examined blood pressure or lipid panels.[25,39,45,55] Half of these observed a small association with diastolic blood pressure and none with lipid panels.[39,55]

Associations With Health Care Utilization and Costs

Four studies, all of which directly provided food, reported health care utilization, or cost outcomes.[33,35,45,46] The heterogeneity in outcomes, such as inpatient admissions and emergency department utilization, limited us from drawing overall conclusions about their findings. Full details of these outcomes are available in eTable 4 of the Supplement.

Certainty of Evidence

The evidence that offering direct food provision is associated with reduced food insecurity was deemed to be of high certainty (Table 2). While most studies had methodologic limitations and there was heterogeneity across studies, which lowered the certainty of evidence, the overall pooled estimate strongly favored the intervention. Moreover, there is a strong mechanistic rationale that providing food to people who are worried about running out of food before they have money to buy more would alleviate their worry. We judged with moderate certainty that programs providing monetary assistance were associated with reductions in food insecurity. This was based on the inclusion of an RCT and several observational studies with a pooled estimate favoring the intervention. Furthermore, given the robust parallel evidence supporting the health benefits of SNAP and other programs, such as the Earned Income Tax Credit, there is a moderately strong rationale that providing direct monetary assistance to households would relieve food insecurity.

Table 2.

Certainty of Evidence, by Intervention Type and Outcome

Intervention or outcome	No. of studies	Study limitations	Consistency	Precision	Other factors	Overall certainty of evidence
Association of food intervention with reduced food insecurity
Food provision	4 RCTs10 Observational	Serious	Inconsistent	No serious imprecision	Very strong mechanism	High
Monetary assistance	1 RCT9 Observational	Serious	Inconsistent	No serious imprecision	Moderately strong mechanism; strong parallel evidence	Moderate
Targeting food deserts	2 Observational	Serious	Consistent	No serious imprecision	NA	Low
Association of food intervention with health outcomes and health care utilization
No change in HbA1c control	3 RCTs3 Observational	Serious	Inconsistent	Serious imprecision	NA	Low
Better HRQOL	1 RCT4 Observational	Serious	Inconsistent	Serious imprecision	NA	Low
No change in BMI	4 Observational	Serious	Consistent	No serious imprecision	NA	Low
No change in depressive symptoms	1 RCT2 Observational	Serious	Consistent	Serious imprecision	NA	Very low
Change in health care utilization	4 Observational	Serious	Inconsistent	Serious imprecision	NA	Very low

Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); HbA1c, hemoglobin A1c; and HRQOL, health-related quality of life; NA, not applicable; RCT, randomized clinical trial.

The certainty of evidence is low that interventions for food insecurity improve HbA1C, HRQOL, BMI, or depressive symptoms (Table 2). Moreover, the certainty of evidence is very low on any association of food insecurity interventions and measures of health care utilization or costs.

Discussion

Providing food or groceries, regardless of setting, has high-certainty evidence that it is associated with reductions in the prevalence of food insecurity, whereas offering income assistance or subsidies has moderate certainty evidence of an association with reduced food insecurity. We believe these associations are much more likely attributable to a causal relationship than to confounding, as these interventions focus on the 2 factors that define food insecurity: worry about inadequate food and lacking money to buy food.

There was insufficient evidence to support the association between food insecurity interventions and a variety of health outcomes, except HRQOL, although this was low certainty evidence and merits further research. Among the 15 studies that included clinically relevant health outcomes, those that evaluated associations with HbA1c had the strongest evidence (3 RCTs). Although the association was not statistically significant, the pooled analysis suggested a possible association with a modest reduction in HbA1c.

Our review complements prior work by De Marchis and colleagues,[64] which focused on interventions within health care, by including additional studies and broadening the scope of interventions outside of health care settings. We included studies examining monetary assistance and welfare expansions that may not have been initially conceived as interventions for food insecurity. Because this review included more studies than the prior review,[64] we were able to perform a meta-analysis on multiple clinical outcomes. The meta-analytic findings of this review are consistent with the narrative conclusions of the prior review. Additionally, we find that the higher risk of bias across the evidence limits the ability to draw firm conclusions regarding these interventions and any associations with health care utilization. With respect to food insecurity as an outcome, we observed that direct provision of food and monetary assistance were associated with 8% to 15% absolute reductions in food insecurity, comparable to the 11% reduction observed with SNAP in a prior evaluation.[65]

These findings have important implications as policy makers, payers, and clinicians search for opportunities to address the health-related social needs of patients and populations. First, if alleviating the anxiety and stress regarding access to food (which is intrinsic to the definition of food insecurity) is considered a beneficial health outcome, then there is enough evidence to enact policy changes now. The effectiveness of interventions to address food insecurity can encourage the adoption of the National Quality Forum’s quality measures to monitor food insecurity screening processes.[66] Likewise, payers should consider using value-based payment models to encourage data collection on food insecurity and implement interventions to address unmet needs. Expanding current programs, such as SNAP and the Commodity Supplemental Food Program for older adults, would be effective in addressing food insecurity. Additionally, the US Congress could broaden the criteria for the Medicare Special Supplemental Benefits for the Chronically Ill, expanding eligibility to those without chronic conditions if they have social needs with important health consequences, such as food insecurity.

Alternatively, if alleviating anxiety and stress concerning access to food is not considered a sufficient health outcome to support practice and policy changes, then these results along with the observational evidence that food insecurity is associated with worse health and higher health care costs, should be an urgent call to action for more evidence. In particular, rigorous evidence for interventions addressing the mechanisms that facilitate food insecurity’s health effects should be prioritized. There are many potential opportunities for further evaluation as payers seek to use traditional health care spending for social care interventions.[67] For example, recent announcements from the Centers for Medicare & Medicaid Innovation indicate that greater flexibility within the Value-Based Insurance Design model will reduce eligibility limitations in offering social services and should be a source of learning.[68] Additionally, North Carolina has implemented a Medicaid section 1115 waiver to allow the state to invest $650 million in pilot programs to address social needs and created a statewide platform to foster a high-value network of community-based organizations.[69] Such laboratories of innovation offer great promise in delivering better evidence regarding the influence of these programs on health outcomes and health care utilization.

Limitations

The main limitation of this review is the quantity and quality of the existing evidence. Most of the studies were observational and had small sample sizes, limiting the ability to draw conclusions about the effects of the interventions. Moreover, of the 39 studies included, only 15 reported health outcomes, and even fewer reported health care utilization measures. Second, the search strategy excluded interventions that identify and address multiple social determinants of health, which may include food insecurity if it was not measured as an outcome. Some interventions that address food insecurity or improve household resources to attain food may have been overlooked. Also, interventions focused on pediatric populations were excluded; therefore, pediatric-based interventions that improve household food insecurity may have been overlooked. Third, we pooled studies across both observational and RCT study designs, but only after verifying that RCTs did not systematically produce lower estimates of effect than observational studies. Furthermore, while we pooled several outcomes, these must be interpreted in the context of high heterogeneity and potential publication bias for interventions providing food. Finally, we focused on food insecurity interventions that excluded conventional assistance programs that comprise the food safety net (ie, SNAP) to focus on interventions that could be implemented by health systems or health plans.

Conclusions

In this meta-analysis, the evidence for providing food and income assistance or subsidies demonstrated a clear association with reduced food insecurity. The evidence that this translated into better health outcomes or reduced health care utilization, as suggested by observational studies, was less clear. Future efforts should prioritize evaluating food insecurity interventions on clinically important health outcomes, health care utilization, and health care spending.