Association of a Primary Care-Based Mobile Food Pantry with Child Body Mass Index: a Propensity Score Matched Cohort Study

Abstract

Objective:

To test the hypothesis that children in Food FARMacia – a six-month food insecurity intervention in May 2019 to January 2020 – would have smaller age-adjusted, sex-specific body mass index (BMIz) gains than matched counterparts.

Methods:

In this proof of concept study, we performed a difference-in-differences (DiD) analysis of a propensity-score matched cohort among pediatric primary care patients age < 6 years with household food insecurity. Children with anthropometric measures prior to and after intervention start were included. Main outcome was child BMIz from standardized clinical anthropometric measurements. We examined difference in child BMIz change between Food FARMacia participants and matched non-participants.

Results:

Among 454 children with household food insecurity, 265 were included, 44 of whom were in Food FARMacia. Mean child age was 1.48 (SD 1.46) years and most reported Hispanic/Latino ethnicity (84.5%). After propensity score matching, children in Food FARMacia had smaller increases in BMIz [unadjusted DiD −0.28 (−0.52, −0.04)] compared to non-participants in the follow-up period. After adjusting for potential confounders, findings remained statistically significant [adjusted DiD, −0.31 units (95% CI: −0.54, −0.08)].

Conclusions:

In this proof-of-concept cohort study of children in households with food insecurity, a pediatric primary care-based mobile food pantry program was associated with significant smaller increases in child BMIz over 6 months.

INTRODUCTION

Food insecurity - or lack of enough food for an active, healthy life - is a recognized upstream driver of obesity and other chronic diseases.¹ Food insecurity is common in the United States (U.S.). Households with children have higher prevalence of food insecurity at 14.5% compared to 10.5% overall.² Food insecurity together with other social determinants of health contributes to health care utilization and health outcomes.³

Innovations in electronic health record (EHR) technology allow integration of food insecurity screening and referrals into clinical care, consistent with consensus recommendations.^4–6 Enrollment in federal supplemental nutrition programs such as the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) and the Supplemental Nutrition Assistance Program (SNAP) can ameliorate but not always eliminate food insecurity because of benefits inadequecy, high food costs, and barriers to continuous enrollment.^7–11 Thus, clinical interventions that go beyond referrals to supplemental nutrition programs may be needed, especially in times of stress to economies and food systems such as during the COVID-19 pandemic. Toolkits for potential food security interventions in pediatric clinical settings exist, but evidence of intervention effectiveness to inform widespread uptake is needed.¹² Given the high and rising prevalence of obesity and household food insecurity among children, effective clinical interventions to reduce childhood obesity in households with food insecurity are needed.^13,14 Alleviating food insecurity may help reduce childhood obesity because household food insecurity has been linked to unhealthy dietary factors and childhood obesity.^15–18

We examined the effects of a 6-month pediatric primary care-based food insecurity intervention on child age-adjusted, sex-specific body mass index (BMI z score). We hypothesized that those in the Food FARMacia mobile food pantry program would have smaller increases in BMI z score compared to those not in the program. Among families participating in Food FARMacia, we also sought to secondarily explore reported changes in household food insecurity, parental obesity-related dietary factors, and parental intervention satisfaction.

METHODS

Study Design, Setting, and Participants

Applying the Obesity-Related Behavioral Intervention Trials (ORBIT) model,¹⁹ we performed a proof-of-concept study. This longitudinal cohort study used 1) EHR data to create a propensity-score matched cohort of pediatric patients and 2) survey data from parents with a child participating in the Food FARMacia at the Columbia University Irving Medical Center (Columbia) NewYork-Presbyterian (NYP) Ambulatory Care Network (ACN). NYP is an academic health care system affiliated with Columbia University Vagelos College of Physicians and Surgeons and Weill Cornell Medicine. The Columbia/NYP ACN is located in Northern Manhattan. Compared to NYC as a whole, Northern Manhattan has 2.5-fold higher proportion of Hispanic/Latino residents and lower median household income.²⁰ Food insecurity prevalence in 2018 ranged from 16.4 to 22.4% in Northern Manhattan, higher than national and NYC averages.²¹

All four pediatric primary care sites of the Columbia/NYP ACN were included. The Food FARMacia program took place at a single primary care site. Each of the four primary care sites is located in northern Manhattan urban neighborhoods with similar residential demographic characteristics. We included all four sites in the analyses because patient demographics are similar across sites and to ensure a robust sample size for the main outcome. Systematic screening for household food insecurity occurs at routine care visits using the Accountable Health Communities health-related needs screening tool that includes the 2-item Hunger Vital Signs^™, a validated screening tool for household food insecurity.^3,22 Patients age < 6 years with a routine care visit; completion of food insecurity screening questions in the pre-intervention period; and at least one anthropometric measurement in the pre-intervention (May 22, 2014 – July 16, 2019) and post-intervention (July 29, 2019 – March 20, 2020) periods were included. We excluded data from non-routine care visits. Figure 1 shows study participant flow.

Figure 1. Participant flow for Food FARMacia pilot program. Data from four sites in a multi-site ambulatory clinical practice in Washington Heights, New York City.

We used prospectively collected observational survey data among Food FARMacia participants to explore secondary outcomes. Food FARMacia participants were eligible for the prospective survey study if parents or legal caretakers were at least 18 years old and spoke English or Spanish. Study staff obtained written informed consent and administered surveys in-person or by telephone at baseline (April 29, 2019 – July 23, 2019), 2-month (July 31, 2019 – September 27, 2019), and 6-month (December 4, 2019 – January 22, 2020) follow-up. The Columbia University Institutional Review Board approved the study.

Food FARMacia Program: Mobile food pantry for families with food insecurity.

We included data from the Food FARMacia program between May 28, 2019 – January 7, 2020. Food FARMacia is a clinical-community linkage for a mobile food pantry set in primary care that was developed in response to the high prevalence of food insecurity among pediatric patients. Food FARMacia was administered at the Washington Heights Family Health Center (WHFHC), a Columbia/NYP ACN site. Choosing Healthy Active Lifestyles for Kids^™ (CHALK), a collaboration between NYP and Columbia Community Pediatrics, implemented Food FARMacia as part of routine care. West Side Campaign Against Hunger (WSCAH), a community-based organization, operated the Food FARMacia. The conceptual framework, study design, and intervention have been described in detail.²³ In brief, to be eligible for Food FARMacia, participants had 1) a positive clinical screen for food insecurity or other social need; and 2) a child receiving routine care at WHFHC under age 6 years. Continual registration took place to maximum capacity (150 individuals equating to about 50 households). Registered families attended twice-monthly food selection sessions for up to six months. Participants selected food in alignment with United States Department of Agriculture (USDA) MyPlate guidelines. Fruits and vegetables provided were predominantly fresh, local, and seasonal. Food selected provided approximately 12 meals per household member for up to 5 household members. Food FARMacia staff provided referrals and enrollment assistance in WIC and SNAP as applicable. Cooking demonstrations with nutrition education and recipe distribution occurred at food selection sessions.

Outcome Measurements

For the main outcome of child weight outcome measures, trained clinical staff used standardized practices to measure and enter data into the EHR. Weight was measured using calibrated scales and recumbent length (age < 2 years) or standing height (age 2 years or older) using an infantometer or stadiometer, respectively. We extracted longitudinal anthropometric data from the EHR. We calculated age- and sex-specific BMI z scores using World Health Organization (WHO) 2006 BMI-for-age growth charts for birth to 2 years and 2 to 5 years.²⁴ Although weight-for-length (WFLz) is traditionally used in clinical settings for children age < 2 years, extensive evidence shows that WHO BMI z score is correlated closely with WFLz, more sensitive to predicting later obesity, and a longitudinal means of tracking weight trajectories from infancy into later childhood.^25–27

For exploratory measures among Food FARMacia participants, parents responded to the USDA Six-item Short Form Food Security Survey Module with a reference period of 12 months at baseline and 2- and 6-months at respective follow-up time periods. This six-item questionnaire expands beyond the 2-item Hunger Vital Signs ^™ screening tool used in routine clinical care. The six-item questionnaire can be used to identify household food insecurity but does not ask questions about children’s food insecurity. We assigned one point for each affirmative response (possible raw score range 0–6).²⁸ Parents answered five validated questions about their vegetable, fruit, and beverage consumption from the School Physical Activity and Nutrition (SPAN) monitoring system.²⁹ We focused on parental consumption because 1) infants primarily fed with breastmilk or formula may have been the index patient and 2) parental-child dietary behaviors correlate with each other.^30,31 Questions asked about behaviors yesterday with a 4-point response scale. Parents also responded to questions about benefits enrollment and Food FARMacia program satisfaction.

Propensity Score Matching

Because of capacity limitations, not all children with household food insecurity in the study period could be registered. Registration was first-come, first-served. We used logistic regression to estimate propensity scores of program participation using pre-specified potential confounders from the EHR: child sex, race/ethnicity, age at registration, and BMI z score at registration.³² We included second-order terms for age and BMI z score at registration to reduce the likelihood that extreme values in BMI z score would influence results. Using nearest-neighbor matching, a statistical method commonly employed for propensity score matching, we matched three non-participants to each Food FARMacia participant for covariate balance (absolute standardized mean difference <0.1).³³ The matched cohort included 176 patients (44 Food FARMacia participants and 132 non-participants).

Statistical Analysis

We used descriptive statistics to examine pre-intervention BMI outcomes closest to the intervention start. For post-intervention outcomes, we used descriptive statistics to report BMI outcomes at the last eligible visit during the post-intervention period. For the main outcome of BMI z score, we used a difference-in-differences (DiD) approach among the 176 propensity-matched cohort participants. For difference-in-differences analyses, we included all BMI measurements in the post-intervention time period. We first assessed for parallel trends between Food FARMacia participants and non-participants in the pre-intervention period. To achieve this, we graphically examined data and tested for an interaction between intervention group and pre-intervention change in child BMI z score. For the main analyses, we assessed longitudinal change from pre- to post-intervention periods in individual-level BMI z score among the Food FARMacia participants compared the non-participants using multi-level, mixed effects linear regression. Regression models included random intercepts for each subject and the following terms: intervention group, time period, and the interaction between group and time period. The interaction term for intervention group × post-intervention time period indicator was used to assess the association between participation in the Food FARMacia program and the outcome. We excluded data in the first 2 months after Food FARMacia initiation to allow time for changes in food procurement and diet to affect weight outcomes. We first examined unadjusted associations of exposure and outcomes. We then adjusted for child age, sex, race/ethnicity, and baseline BMI z score.

We additionally examined child BMI and proportion of children with BMI ≥85^th percentile for age and sex (BMI p85) using mixed effects linear and linear probability models, respectively. We selected BMI p85 as a cut-off because 1) BMI p85 during infancy has higher positive predictive for later obesity than WFLz among individuals under age 2 years and 2) BMI p85 can be measured continually from birth to age 6 years.^24,25

We performed several sensitivity analyses: 1) analyses adjusted for pre-intervention linear time trends to account for the chance that we failed to detect differences in pre-intervention parallel trends;³⁴ 2) restricting the pre-intervention time period to 12 months; 3) estimating effects of early and late Food FARMacia entry; 4) as-treated effects excluding Food FARMacia registrants who were non-attendees; 5) adjusting for primary care site; and 6) restricting the comparison group to only WHFHC (location of the Food FARMacia program). We also examined alternatives to propensity score matching in sensitivity analyses: 1) unweighted multilevel regression using the entire cohort and 2) inverse-probability-of-treatment weighted analysis to estimate the average treatment effect. We estimated a linear dose-response relationship for Food FARMacia session attendance, assigning a value of 0 to 14 according to food selection attendance.

In exploratory analyses using Food FARMacia participant survey data, we examined changes in report of household food insecurity, parental vegetable consumption, and parental sugar-sweetened beverage (SSB) consumption using mixed-effects models. We compared baseline to 2-month and 6-month with a subject-level random intercept. Adjusted models additionally included parental baseline age, race/ethnicity, household size, and parental education level. In mixed effects models, we used maximum likelihood estimation for 1) all responses at each time point and 2) complete cases.

We conducted analyses using R (version 4.1.1) and SAS version 9.4 (SAS Institute Inc., Cary, North Carolina). Analyses for the main outcome of child BMI z score used 2-tailed testing with p < 0.05 denoting statistical significance. Other analyses were exploratory and not adjusted for multiple testing.

RESULTS

Among 4966 children age < 6 years with a pediatric primary care visit, 2036 completed household food insecurity screening during the pre-intervention period and 454 (22.3%) of those screened positive for household food insecurity (Figure 1).

Of those with household food insecurity, 265 met inclustion criteria and comprise the overall cohort (Supporting Information Table S1). Mean child age at baseline was 1.48 years (SD 1.46 years) and the majority of parents reported child Hispanic/Latino (84.5%) ethnicity. Overall, 64.5% of children with household food insecurity had normal weight (sex-specific BMI-for-age 5 to < 85%ile). Before propensity-score matching, the proportion of children in Food FARMacia with normal weight was higher than those not registered in Food FARMacia. After propensity-score matching, participant characteristics including proportion with normal weight were similar (Table 1; SupportingInformation Table S2). Children in the Food FARMacia program had mean 7.7 (SD 4.9) measures in the pre-intervention period and 1.9 (SD 0.9) in the post-intervention period. Post-intervention measures for Food FARMacia participants took place mean 5.1 (SD 1.1) months from intervention start. Those in the matched comparison group had 7.4 (SD 4.9) in the pre-intervention period and mean 1.6 (SD 0.9) during post-intervention. Post-intervention measures for comparison children took place mean 4.8 (SD 1.3) months after intervention start.

Table 1.

Baseline Characteristics of 176 Children in New York City in a 3:1 Retrospective Propensity-Score-Matched Sample. Data collected between May 28, 2018 and March 20, 2020.

	Food FARMaciaa (N = 44)	Matched controls (N =132)	Standardized difference before matching	Standardized difference after matching
Child Baseline Demographic Characteristic
Age, mean (SD), years	1.77 (1.55)	1.67 (1.55)	0.23	0.06
Age, n (%), years
0 to < 2	28 (64)	89 (67.4)	0.23	0.08
2 to < 6	16 (36)	43 (32.6)
Female, n (%)	22 (50)	68 (51.5)	0.05	0.03
Race/ethnicity, n (%)			0.11	0.09
Hispanic/Latino	37 (84)	114 (86.4)
Non-Hispanic/Latino	5 (11)	14 (10.6)
Declined/Missing	2 (5)	4 (3)
Child Baseline BMI Measure
BMI, mean (SD)	16.38 (1.88)	16.52 (1.66)	0.07	0.08
BMI z scoreb, mean (SD)	0.35 (0.95)	0.31 (0.97)	0.05	0.04
BMI percentilea, n (%)			0.32	0.10
<5th	1 (2)	3 (2.3)
5th to <85th	32 (73)	100 (75.8)
85th to < 97.7th	9 (20)	22 (16.7)
≥97.7th	2 (5)	7 (5.3)
Child Follow-up BMI Categories
BMI percentilea, n (%)			n/a	n/a
<5th	1 (2)	1 (0.8)
5th to <85th	28 (64)	77 (58.4)
85th to < 97.7th	9 (20)	37 (28.0)
≥97.7th	6(14)	17 (12.9)

^aFood FARMacia participants attended median 10 (range 0–14) food selection sessions.

^bWHO 2006 child growth standards (for ages 0–5 years) Note: Propensity-score matching performed using child sex, race, baseline age, baseline BMI z score, and second-order terms of baseline age and baseline BMI z score

In both the Food FARMacia group and the comparison group, BMI z score and BMI increased from pre- to post-intervention (Table 2, Figure 2). We did not find violation of the parallel trends assumption (group × pre-intervention timepoints interaction p-value = 0.28). In unadjusted and adjusted models, Food FARMacia participants had smaller increases in BMI z score (adjusted DiD −0.31; 95% CI: −0.54 to −0.08) than the comparison group from pre- to post-intervention. Findings were similar for BMI (adjusted DiD −0.68; 95% CI: −1.20 to −0.15).

Table 2.

Difference-in-Differences mixed-effects regression results for effects of the Food FARMacia on child body mass index (BMI) outcomes. Data from 176 children in New York City in a 3:1 propensity-score-matched retrospective longitudinal cohort collected between May 22, 2014 and March 20, 2020.

	Food FARMacia		Comparison group
	Pre-intervention	Post-intervention	Pre-intervention	Post-intervention	Difference-in-difference analysisa
Child Outcomeb	Mean (SD)				Unadjusted effect estimate (95% CI)c	p-value	Adjusted effect estimate (95% CI)c	p-value
BMI z-score, units	0.35 (0.95)	0.70 (0.95)	0.31 (0.97)	0.82 (1.05)	−0.28 (−0.52, −0.04)	0.02	−0.31 (−0.54, −0.08)	0.008
BMI, kg/m2	16.38 (1.88)	17.04 (1.72)	16.52 (1.66)	17.35 (1.78)	−0.64 (−1.18, −0.10)	0.02	−0.68 (−1.20, −0.15)	0.01
BMI p85d, %	25 (43.8)	34.1 (47.9)	22 (41.6)	40.9 (49.4)	−0.07 (−0.17, 0.02)	0.14	−0.09 (−0.18, 0.01)	0.07

^aMixed-effects models adjusted for child baseline age, sex, race/ethnicity, and baseline BMI z-score

^bBMI z-score, BMI, and BMI percentiles calculated using WHO 2006 child growth standards (for ages 0–5 years)

^cInteraction term of intervention status and post-intervention period indicator.

^dBMI p85 defined as BMI ≥85^th percentile

Figure 2. Pre-to-post-intervention changes in child weight outcomes by intervention group. Estimated changes within Food FARMacia and matched control groups across intervention periods for the outcomes BMIz (top left), BMI (top right), and probability of BMI ≥85th percentile (bottom left). Data from adjusted difference-in-differences multilevel models including 176 children in New York City; BMIp85 defined as BMI ≥85^th percentile.

Children in Food FARMacia had smaller increases in the proportion with BMI p85 than the comparison group (adjusted DiD −0.09; 95% CI: −0.18, 0.01) that were not statistically significant. In sensitivity analyses, results were similar to the main analyses for BMI z score, BMI, and BMI p85 (Supporting Information Tables S3 and S4).

Similar to our previously reported data on the initial Food FARMacia participant cohort,²³ mean attendance for the current study sample was 9 sessions and median attendance was 10 sessions (range 0–14 sessions; Supplemental Table S5). In the dose-response analysis, each Food FARMacia session attended was associated with a smaller increase in BMI z score (adjusted DiD −0.03; 95% −0.05 to −0.01).

Baseline demographic characteristics of children in the Food FARMacia survey study were largely similar at each time point (Supporting Information Table S6). In survey data from those who participated in the Food FARMacia (Table 3), in both unadjusted and adjusted models, the raw score for household food insecurity decreased at both 2-month (adjusted mean difference −1.05 points; 95% CI: −1.68 to −0.42) and 6-month follow-up (−0.97; 95% CI: −1.60 to −0.34). Parental report of fruit and vegetable intake increased from baseline to 2-month follow-up but not at 6-month follow-up. We also found that parental report of SSB consumption decreased from baseline to 6-month follow-up, but results were not significant at 2-months. In sensitivity analyses including only complete cases (Table 3), results were similar with the additional finding that parental SSB consumption also decreased at 2-month follow-up (−0.45 servings yesterday; 95% CI: −0.83 to −0.06). Most parents reported they were satisfied/very satisfied with food selection options and the intervention (food selection satisfaction: 94% at 2 months; 92 % at 6 months; intervention satisfaction: 95% at 2 months, 92% at 6 months).

Table 3.

Food FARMacia participants: Household Food Insecurity and Parental Dietary Factors. Survey data from 48 study participants in New York City collected between April 29, 2019 and January 20, 2020.

	Baseline (N = 48)	2-Month (N = 43)	6-Month (N = 39)	2-Month Change a	6-Month Change a
All Responses	Mean (SD)			Adjusted mean difference (95% CI) b
Household food insecurity, continuous measure	4.29 (1.76)	3.30 (2.25)	3.28 (2.06)	−1.05 (−1.62, −0.47)e	−0.99 (−1.59, −0.39)d
Parental fruit/vegetable servings yesterday, times	2.21 (1.54)	2.72 (1.35)	2.28 (1.30)	0.44 (0.06, 0.83)c	−0.05 (−0.45, 0.35)
Parental sugar-sweetened beverages yesterday, times	1.25 (1.36)	0.81 (1.24)	0.51 (0.79)	−0.37 (−0.75, 0.01)	−0.62 (−1.01, −0.23)d
	Baseline (N = 38)	2-Month (N = 38)	6-Month (N = 38)	2-Month Change a	6-Month Change a
Complete Cases	Mean (SD)			Adjusted mean difference (95% CI)b
Household food insecurity, continuous measure	4.29 (1.80)	3.24 (2.27)	3.32 (2.08)	−1.05 (−1.68, −0.42)d	−0.97 (−1.60, −0.34)d
Parental fruit/vegetable servings yesterday, times	2.29 (1.61)	2.82 (1.31)	2.26 (1.31)	0.53 (0.14, 0.92)c	−0.03 (−0.42, 0.36)
Parental sugar-sweetened beverages yesterday, times	1.13 (1.21)	0.68 (1.12)	0.50 (0.80)	−0.45 (−0.83, −0.06)c	−0.63 (−1.02, −0.25)d

^aCompared to responses in baseline survey

^bLinear mixed-effects model where reference is baseline response. Each model adjusted for baseline age, race/ethnicity, household size, and parental education level

^cp < 0.05

^dp < 0.01

^ep < 0.001

DISCUSSION

In this cohort study using propensity score methods among children younger than age 6 years with household food insecurity, we found that participation in a clinically-based mobile food pantry intervention was associated with 0.31 reduction in BMI z score change among Food FARMacia participants compared to non-participants over 6 months. According to the United States Preventive Services Task Force, a BMI z score difference of 0.20 to 0.25 is a clinically important endpoint for childhood obesity interventions in children age 6 years and older.³⁵ An abundance of evidence shows that rapid gains in weight or persistently high weight in early life are predictive of later obesity.^25,36,37 Children with obesity by age 5 years are likely to have obesity throughout childhood,³⁸ but evidence-based obesity prevention interventions are limited in early life.³⁵ Our results suggest that the Food FARMacia was associated with a clinically meaningful improvement in an objective early life risk factor for later childhood obesity.

Evidence for a causal relationship of household food insecurity with childhood obesity has been mixed, and no studies to date have examined clinically-based food security interventions on childhood obesity outcomes.¹⁸ This research is the first to demonstrate that a clinical-community partnership to deliver a mobile food pantry program in a clinical setting may help reduce childhood obesity among a population disproportionately burdened by obesity. A key feature of the Food FARMacia intervention that differentiates it from most prior food security interventions, medical nutrition therapy interventions, and current supplemental nutrition programs is the provision of a large amount of fresh, local produce and other USDA-aligned food for the entire household. Thus, the intervention addressed neighborhood food environment and cost barriers that inhibit healthy food access. Resultant changes to home food environments, parental stress, child feeding practices, parental role modeling, and improvements in child’s dietary quality may have mediated effects of the intervention.^7,15–18

Achieving child health equity requires reduction of food insecurity.³⁹ Most existing clinically-based studies of food insecurity interventions have largely focused on elderly adults, uninsured adults, or adults with specific chronic diseases such as cancer or type 2 diabetes. ^40–42 One mixed-methods evaluation among pediatric patients showed feasibility of an in-clinic pantry.⁴³ Our findings align with other research suggesting that increased access to healthy foods leads to reductions in childhood obesity risk factors.^44–46 Our study provides new information suggesting that clinical food pantry programs may be associated with improvements in objective child weight outcomes, providing support for clinically-based food security interventions as part of broader policy and community-based efforts to prevent childhood obesity.

In exploratory analyses, we found statistically significant improvements in household food insecurity among Food FARMacia participants. The clinical importance of a 1-point improvement in household food insecurity is not known. In a recent retrospective longitudinal study of adults, Berkowitz and collegues found that a 1-point improvement in a 10-item food security scale was associated with improved mental health and health utility over a one-year period.⁴⁷ Further understanding of clinically meaningful endpoints in food security measures warrants further study. Among the Food FARMacia parents, we found a small increase in parental report of fruit/vegetable consumption at 2-months and decreases in SSB beverage consumption at 6-months compared to baseline. Prior cross-sectional evidence has shown a relationship of food insecurity with diet quality and SSB beverage consumption among adults.^48,49 Household access to fresh fruit may have helped reduce cravings for sweets and therefore led to reduced SSB consumption. Another potential explanation could be that Food FARMacia participants did not need to spend as much time obtaining food from traditional food vendors because they selected free food for the whole household from the program twice monthly. Thus, Food FARMacia participants may have had less exposure to marketing for SSBs and fewer opportunities to procure them. Although we found a small increase in parental vegetable and fruit consumption at 2-month follow-up, there was no difference at 6-month follow-up. Plausible explanations for the lack of persistent increased vegetable and fruit consumption over time could be related to diminished intervention effects or sharing vegetables and fruits with others, thus reducing parental consumption.

Although we considered children under age 6 years eligible for participation, most of the patients included in this observational study were under age 2 years. The young age of study participants could be related to the more frequent routine child care visits among children age < 2 years. A recent systematic review found that food insecurity during infancy was associated with childhood obesity.¹⁸ Our findings of improvement in household food insecurity and relative reductions in BMI z score among infants and young children adds support to the need for early life interventions that target social needs such as household food insecurity in order to promote healthy child growth and reduce obesity.

Because this study was set in pediatric primary care and relatively small in sample size, we were unable to examine changes in parental BMI or perform mediation analysis to mechanisms through which the Food FARMacia may have impacted child BMI z score. For example, parental depressives symptoms have been correlated with food insecurity, child’s eating behaviors, and childhood obesity.^16,50 It is plausible that Food FARMacia-driven reductions in household food insecurity resulted in changes in parental mental health, thus leading to changes in parental-child interactions such as improved feeding patterns or changes in the home food environment. However, we cannot examine these relationships in our current proof-of-concept study. Future prospectively randomized studies to examine effects of food insecurity interventions on child weight outcomes should examine potential mechanisms of improved household food insecurity and child diet on intervention effects.

Limitations

This study was observational and intervention allocation was based on referral and voluntary participation by families. A randomized trial would be needed to provide stronger evidence of causal association between the intervention and differences in childhood obesity outcomes. We addressed potential confounding using propensity score matching and additional covariate adjustment. We also examined pre-intervention BMI z-score trends to assess the assumption in pre-intervention parallel time course in the two groups and performed additional sensitivity analyses. Residual confounding from unobserved variables such as parental motivation for dietary change or number of children in household cannot be excluded. Therefore, we cannot exclude the possibility that selection bias or regression to the mean play roles in our findings. Follow-up was limited to 6 months secondary to widespread disruptions after the start of the COVID-19 pandemic, limiting our follow-up period, sample size, and ability to draw conclusions about the intervention in current turbulent economic periods. Survey data on household food insecurity and parental dietary factors was limited to Food FARMacia participants only. Finally, the study findings may not be generalizable to less urban populations.

CONCLUSION

In this propensity-score matched cohort study among a predominantly Hispanic/Latino population of children in households experiencing food insecurity, we found that a mobile food pantry program set in pediatric primary care was associated with smaller increases in child BMI z score among participants compared to non-participants over an average follow-up period of 6 months. As systematic screening for household food insecurity becomes increasingly integrated into routine health care, effective interventions to address food insecurity may have a role in pediatric obesity prevention.