Household food insecurity and early childhood development: Systematic review and meta‐analysis

Abstract Household food insecurity (HFI) is a powerful stressor negatively associated with early childhood development (ECD). However, no comprehensive review has examined the association of HFI and ECD. Therefore, this systematic review and meta‐analysis investigated the association between HFI and ECD domains and subdomains in children under 5 years old. Peer‐reviewed and grey literature were systematically searched in electronic databases with no year or language restrictions. Studies were eligible if they assessed the association between HFI and one or more ECD domains. Data were extracted using a standard predefined protocol. Meta‐analysis was performed, and the heterogeneity across studies was explored. Nineteen studies were included in the systematic review and 14 in the meta‐analysis. Of the studies, 15 were from high income countries (HICs) and four from low–middle income countries (LMICs). For developmental risk and the cognitive/math and cognitive/school readiness and reading subdomains, the only studies available were conducted in HICs. The meta‐analysis showed that HFI was associated with developmental risk (OR 1.28; 95% CI [1.14, 1.45]), cognitive/vocabulary (OR 0.94; 95% CI [0.90, 0.98]), and cognitive/math (OR 0.84; 95% CI [0.73, 0.96]). HFI was marginally associated with cognitive/school readiness and reading (OR 0.91; 95% CI [0.82, 1.00]) and motor development (OR; 0.91, 95% CI [0.80, 1.04]). HFI was associated with poor ECD in children under 5 years old. Specifically, HFI was associated with developmental risk and poor math skills in studies conducted in HICs and with poor vocabulary skills in studies conducted in both HICs and LMICs. Prospective studies examining HFI and ECD are needed in LMICs.


adolescents.
These previous reviews were restricted to HICs, and the results focused on long-term outcomes throughout childhood and adolescence Pérez-Escamilla & Vianna, 2012;Shankar et al., 2017). These limitations called for conducting a comprehensive review to capture the entire body of evidence in HICs and low-middle income countries (LMICs) as well as to explore the association of HFI and ECD outcomes by ECD domains and subdomains, that is, cognitive (executive function, school readiness, reading, vocabulary, and math), language (communication and language comprehension and expression), motor (fine and gross motor), socio-emotional (externalizing behaviour and hyperactivity and internalizing behaviour and anxiety) and developmental risk.
Therefore, the aim of this systematic review and meta-analysis was to investigate the association between HFI and ECD domains and subdomains among children under 5 years old in HICs and LMICs.

| METHODS
The protocol for this systematic review and meta-analysis was registered on PROSPERO prior to starting the literature search (CRD42018091435).

| Eligibility criteria
This systematic review followed the guidance of the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA; Moher, Liberati, Tetzlaff, Altman, & Group, 2009). A literature search protocol was prepared on the basis of clearly defined a priori inclusion and exclusion criteria.
Studies evaluating the association between HFI and ECD in infants and toddlers under 5 years old were included. Studies were excluded if they were (a) qualitative or (b) reviews (systematic or not), letters to the editor, books, conference abstracts, study protocols and (c) studies that did not report the association between HFI and ECD. Studies were eligible if they (a) reported the association (coefficient β,and odds ratio) between ECD and HFI and (b) used an experience-based food insecurity scale to assess HFI (Pérez-Escamilla, Gubert, Rogers, & Hromi-Fiedler, 2017). Measuring ECD is complex and consensus on the use of a single tool globally has not been reached. Because tools designed to measure ECD vary widely and may cover a single or multiple ECD domains (Fernald, Prado, Kariger, & Raikes, 2017), we chose to include all studies that measured any domains or subdomains of ECD, regardless of the tool used.

Key messages
• Household food insecurity (HFI) is associated with poor early childhood development (ECD) in children under 5 years old.
• HFI is associated with developmental risk and poor math skills in high income countries (HICs) and with poor vocabulary skills in both HICs and low-middle income countries (LMICs).
• Standardized methods to assess ECD and clear definitions of ECD domains are necessary to allow global-level comparability.

Parents' Evaluation of Developmental Status
Rose-Jacobs et al. (2008) Developmental risk Children with two or more significant concerns are at developmental risk.
Parents' Evaluation of Developmental Status Drennen et al. (2019) Developmental risk Children with two or more concerns were classified as at developmental risk.

Parents' Evaluation of Developmental Status
Socio-emotional b Social-emotional skills, temperament, and personal social/adaptive skills Externalizing behaviour Hobbs & King (2018) Externalizing behaviour problems Externalizing behaviours directed toward others (e.g., aggressive behaviours, whether the child argued a lot, bullied, was disobedient, or destroyed things).
Child Behavior Checklist-(CBCL/1.5-5) Whitaker et al. (2006) Aggressive Not reported Child Behavior Checklist (CBCL/1.5-5) Nagata, Gomberg, Hagan, Heyman, & Wojcicki (2018) Oppositional defiant score Not reported Child Behavior Checklist (CBCL/1.5-5) Gill et al. (2018) High-frequency discipline Not reported "How many days in a typical week does NAME need to be disciplined for his/her behavior?" item adapted from the Home Observation Measurement of the Environment-Short Form scale. Johnson & Markowitz (2018) Conduct problems Conduct problems (e.g., how often child pushes or tantrums If two or more eligible studies used the same data, only one paper was included, which was selected either because it reported more thoroughly the data or was the most recent publication.

| Exposure
The key exposure was HFI (vs. non-HFI), assessed using experience-

| Outcome
There is no consensus on the definition of early childhood. The World Health Organization defines early childhood as ages 0-8 years (World Health Organization, 2019). The American Academy of Pediatrics divides early childhood into ages 0-1 (infants), 1-3 (toddlers), and 3-5 Studies were allocated into the predefined domains and subdomains, following a consensus process followed by two study co-

| Study selection
Articles were excluded if they did not meet the aforementioned criteria (Data S2). Two co-authors (KHDO and GMA) independently screened the titles and abstracts. Studies were included in the initial screening if they examined the association between HFI and ECD.
Next, the articles were independently assessed for eligibility using the predefined inclusion and exclusion criteria defined above. Disagreements were resolved through consensus, or when required, through consultation with a third co-author (ASM).

| Data extraction
Data were extracted independently by two co-authors (KHDO and GMA). The data extraction form included study identification (author, year of publication, and country of study, country's income level), name of the primary study and year of data collection, study design and sample size, child's age, HFI (status, tool, child's age when measured, and prevalence), ECD (tool, child's age when measured, and outcomes), effect size of the association between ECD and HFI (odds ratio or coefficient β with respective 95% confidence interval-adjusted or crude), and confounders. When necessary, authors were contacted to provide missing information or additional data to calculate the OR and CI of the association between HFI and ECD.  Hagen, Biondo, & Cummings, 2012). This quality assessment tool has been used and adapted by other systematic reviews in the maternal and child nutrition field with reliable and consistent results regarding studies quality assessment (Boccolini, Carvalho, & Oliveira, 2015; Buccini, Pérez-Escamilla, Paulino, Araujo, & Venancio, 2017).
When studies reported two or more ECD outcomes, the findings were included in the meta-analysis separately, by ECD domain and subdomain. If more than one ECD outcome per domain was presented in the same study, they were grouped according to predefined ECD domains and subdomains definitions. Domain definitions as well as how each study outcome were defined and grouped within subdomains are presented in Table 1.

| Meta-analysis
Effect measures from the meta-analysis were presented as pooled odds ratios (ORs) by ECD domain (developmental risk and motor) and, when appropriate, by subdomains (cognitive and socio-emotional). Heterogeneity was assessed by the degree of inconsistency (I 2 statistic) and its significance (p < .05), using a fixed effect model. If high heterogeneity was found (i.e., I 2 > 50%) random-effects models were used for the analysis (Deeks et al., 2008). Random-effects models take into account the variability of effects between different studies (Deeks et al., 2008). Meta-regression was considered for studies with high heterogeneity (i.e., I 2 > 50%); however, due to the relatively low number of studies, we were unable to follow this approach.
Instead, we conducted a descriptive analysis of possible sources of heterogeneity. Results were not broken down by country income due to the scarcity of studies conducted in LMICs. This decision is unlikely to have biased our results because this source of heterogeneity was considered in the interpretation of the findings, and findings when just including HICs were similar as when also including LMICs. All analyses were conducted using Stata version 14. and 14 studies were included in the meta-analysis (Figure 1). In the quality assessment, five studies were classified as weak, eight as moderate and six as strong. Figure 2 indicates that 31.6% of the studies had strong quality. The quality attributes examined that were the weakest were confounding factors followed by selection bias, study design, and withdrawals and dropout.  (Milner et al., 2018), Pakistan (Finch et al., 2018;Obradovíc et al., 2016), and Bangladesh (Saha et al., 2010). In the majority of studies, HFI and ECD were measured at the same time or in an interval up to 24 months, and six studies did not present this information (Black et al., 2012;Drennen et al., 2019;Gill et al., 2018;Greder et al., 2017;Hobbs & King, 2018;Mickens, 2019). Information regarding HFI and ECD status was usually reported by the mothers. All HFI scales probed for HFI in the context of food deprivation due to lack of money. All scales, but one (HHFS), were directly adapted from the Household Food Security Survey Module(HFSSM).

| Description of HFI and ECD outcomes
The prevalence of HFI in the US among 9-month-old infants was 12.5%, according to a nationally representative longitudinal study (Zaslow et al., 2009) and 11.7% at 48 months old (Huang et al., 2018;  Reported as ECD description adopted in this review. e Calculated by the authors as recommended by Deeks, Altman, & Bradburn (2001) and Altman (1991).
US had a mean receptive vocabulary score of 86.2 using the Peabody Picture Vocabulary Test (standard 20-160), and the reading mean score was 99.9 using Woodcock-Johnson letter-word identification (standard 100 ± 15).

| HFI and ECD associations, meta-analysis, and sources of heterogeneity
The meta-analysis was performed for the ECD domains and subdomains identified (for definitions, see Table 1 ( Figure 4). The high heterogeneity identified was possibly due to the lack of standardized (a) sample size, (b) study quality, (c) tools used to measure ECD, and (d) time lapse between assessing ECD and HFI (Table 4).
The language domain was assessed in different aspects for only two studies conducted in LMICs; thus, no meta-analysis was performed. Findings on the language domain individual studies showed that communication at 24 months was not associated with HFI (Milner et al., 2018); however, comprehensive and expressive language skills at 18 months old were significantly associated with HFI (Saha et al., 2010).

| DISCUSSION
This systematic review found that HFI is associated with poor ECD outcomes among children under 5 years old in both HICs and LMICs.
Meta-analysis showed that HFI was associated with the developmental risk and cognitive outcomes related to vocabulary and math skills.
These results are consistent with previous reviews focusing on HICs Pérez-Escamilla & Vianna, 2012;Shankar et al., 2017) and reinforce the critical role that adverse experiences, such as HFI, during early childhood have in putting at risk the future of vulnerable children.
Third, food insecurity-related psycho-emotional stress may lead to difficulties in concentrating and learning (Pérez-Escamilla & Vianna, 2012). This can also negatively affect caregivers' mental health status leading to less responsive caregiving and early childhood stimulation opportunities at home (Fuller et al., 2002;Pérez-Escamilla & Vianna, 2012). Hence, we hypothesize that HFI is associated with poverty, poor diet quality, and nutritional deficiencies as well as expose families to stress and anxiety that impairs parent-child interaction, which may put children at developmental risk.
Despite the high heterogeneity, the meta-analysis results identified HFI to be a risk factor for externalizing behavioural problems, which is consistent with previous studies (Gill et al., 2018;Hobbs & King, 2018;Johnson & Markowitz, 2018;Whitaker et al., 2006). High levels of poverty-related stressors, such as HFI Pérez-Escamilla & Vianna, 2012), could lead to behaviour problems which in turn can seriously disrupt caregiver-child interactions (McLeod & Fettes, 2007). The high heterogeneity observed among the studies examining socio-emotional development limits the strength of our conclusions in this domain.
Our meta-analysis did not find associations between HFI and motor development, which is consistent with previous studies (Hernandez & Jacknowitz, 2009;Milner et al., 2018 Only two studies examined the association of HFI with language development, thus this relationship needs to be further investigated.
In addition, further prospective studies evaluating HFI and cognitive/school readiness and reading need to be conducted to clarify this association.
Our findings should also be interpreted considering some limitations. Studies used different tools to assess HFI and ECD. This lack of uniformity, especially in the measurement of ECD and its domains, limits the interpretation of the data and must be considered when generalizing our findings. On the other hand, HFI was measured using similar experience-based HFI scales that have been shown to be reliable globally (Marques et al., 2015;Pérez-Escamilla et al., 2017). HFI experience-based scales adapted from the HFSSM have been applied F I G U R E 3 Fixed effects of metaanalysis of studies evaluating the association between household food insecurity and developmental risk (OR > 1 is considered risk factor) and motor and cognitive outcomes (OR < 1 is considered risk factor) F I G U R E 4 Random effects of meta-analysis of studies evaluating the association between household food insecurity and socio-emotional outcomes (OR > 1 is considered risk factor)   Nagata, Gomberg, Hagan, Heyman, & Wojcicki (2018) and Whitaker et al. (2006). in different socio-economic and cultural contexts and have demonstrated adequate psychometrics, reliability, and validity of the instrument properties (Marques et al., 2015;Pérez-Escamilla et al., 2017).
Our findings call for the development of standardized tools to measure ECD globally, which will allow for greater comparability between studies. Furthermore, future studies need to clearly report when and how HFI and ECD were measured. Child's age at the time of assessment can be a strong confounder when comparing findings across studies (Johnson & Markowitz, 2018); hence, it must be clearly reported.
Studies with different geographic populations and age groups were included in this review, which should be considered in the interpretation of the results. Although we considered performing a metaregression to explore the results according to the country's income, due to the very limited number of studies performed on LMICs, this was not possible. Our review and meta-analysis clearly demonstrate the need to conduct prospective studies to investigate the associations between HFI and ECD in LMICs. Despite these limitations, our meta-analysis was carefully carried out using random effects models for domains with high heterogeneity across studies (Deeks et al., 2008;Egger, Davey-Smith, & Altman , 2008). Another limitation is that the majority of studies were cross-sectional or presenting crosssectional analysis for the purposes this review, which precludes drawing causal inferences (Atkins et al., 2004). The sources of bias identified through this systematic review can help guide the design of future studies (Boutron et al., 2019).
To our knowledge, this is the first review and meta-analysis that identifies HFI as a risk factor for suboptimal ECD, which includes results from LMICs. Our findings suggest that the associations between HFI and ECD can vary across ECD domains. HFI was associated with developmental risk and poor cognitive outcomes, including vocabulary and math skills, among children under 5 years old. This finding supports the need to consider HFI as a key component of integrated ECD national strategies (Pérez-Escamilla, 2017).

ACKNOWLEDGMENT
All those who contributed to the study are already identified as authors. This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)-Brasil (Finance Code 001).

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.