Child diet and mother–child interactions mediate intervention effects on child growth and development

Abstract This study examined whether child diet and mother–child interactions mediated the effects of a responsive stimulation and nutrition intervention delivered from 2009 to 2012 to 1324 children aged 0–24 months living in rural Pakistan. Results showed that the intervention improved children's cognitive, language and motor development through child diet and mother–child interactions. Although the intervention did not improve child growth or socio‐emotional development, we observed positive indirect effects on child growth via child diet and on socio‐emotional development via both child diet and mother–child interactions. In addition, child diet emerged as a shared mechanism to improve both child growth and development, whereas mother–child interactions emerged as a distinct mechanism to improve child development. Nevertheless, our results suggest the two mechanisms were mutually reinforcing and that interventions leveraging both mechanisms are likely to be more effective at improving child outcomes than interventions leveraging only one of these mechanisms.


| INTRODUCTION
In low-and middle-income countries (LMICs), 40% of preschool-aged children are at risk of poor child development (Lu et al., 2020) and 22% are stunted (UNICEF et al., 2021). Community-based interventions integrating responsive care, stimulation, nutrition and health components delivered to parents are recognized as effective strategies to improve child growth and development in early life (Jeong et al., 2021;Prado, Larson, et al., 2019;World Health Organization, 2020). Evidence suggests that interventions providing multiple inputs are needed to improve multiple child outcomes rather than interventions providing individual inputs or the standard of care (World Health Organization, 2020). Although evidence on what interventions work to improve child growth and development is expanding, less is known about how these interventions operate, the specific aspects of nurturing care these interventions benefit, and whether these benefits translate into improved child outcomes (Obradović et al., 2016;Yousafzai et al., 2014).
Therefore, child diet and mother-child interactions remain to be unpacked as mechanisms through which multi-input interventions in LMICs improve multiple facets of child growth and development.
Child diet can affect cognitive development both directly through brain development during infancy, and indirectly by affecting child health, physical activity, and caregiver behaviour (Prado & Dewey, 2014). Although nutrition interventions can improve both linear growth and, to a lesser extent, child development (Prado, Larson, et al., 2019), it is unclear whether these impacts are achieved primarily through child diet or whether other mechanisms, such as caregiver behaviour, are also at work. Limited evidence from Bangladesh suggests that child diet mediates the effects of nutrition interventions on child growth and development (Frongillo et al., 2017;Vazir et al., 2013). Further, a recent study in Kenya showed that child diet together with parental stimulation, maternal knowledge and recall of intervention messages mediated the effects of a responsive stimulation and nutrition intervention on child cognitive and receptive language development (Luoto et al., 2021). However, to our knowledge, no studies have assessed whether responsive stimulation interventions in LMICs work to improve child growth and development exclusively through child diet.
Maternal responsive behaviours and developmentally appropriate and supportive mother-child interactions are important for child health, nutrition, and development in early life (Eshel et al., 2006). Responsive interactions involve a three-step process: observation of the child's signals, interpretation of these signals, and action/response to meet the child's signals appropriate for the developmental age of the child (Eshel et al., 2006). A caregiver may also demonstrate a range of developmentally supportive interactions, that are not necessarily responsive, but nurture development (e.g., directive interactions, language inputs). Given the complex nature of maternal responsive behaviours, culturally relevant measures with tested reliability and validity for use in LMICs are limited and no gold standard of measuring responsive behaviours exists (Bentley et al., 2011;Hentschel et al., 2021;Pérez-Escamilla & Segura-Pérez, 2020).
As the 'interpretation' step of the process is difficult to observe, most measures focus on the child's signals and maternal responses (Hentschel et al., 2021). The few studies in LMICs that have collected data on caregiver-child interactions tend to encompass both maternal responsive behaviours and supportive mother-child interactions, which are often assessed in a picture book-reading context (Aboud & Akhter, 2011;Betancourt et al., 2020;Brown et al., 2017;Knauer et al., 2020;Murray et al., 2016;Obradović et al., 2016;Scherer et al., 2019). Evidence from these studies suggests that improvements in mother-child interactions mediate the positive effects of parenting and responsive stimulation interventions on child growth and development (Aboud & Akhter, 2011;Brown et al., 2017;Eshel et al., 2006;Landry et al., 2006;Murray et al., 2016;Obradović et al., 2016). In addition, limited evidence indicates that mother-child interactions mediate the effects of nutrition interventions in early life on child growth at 4 years of age . Whether mother-child interactions mediate the effect of responsive stimulation and nutrition interventions in LMICs has not been empirically tested in children less than 2 years of age.
Understanding how multi-input interventions achieve impact on child outcomes can help increase intervention effectiveness by leveraging common and mutually reinforcing mechanisms. Therefore, in the present study we sought to deepen knowledge regarding the mechanisms through which multi-input interventions in LMICs may affect child outcomes in early life. Specifically, we used data from the Pakistan Early Child Development Scale-up (PEDS) intervention trial to explore child diet and mother-child interactions at 1 year of age as potential mediators that might explain the effects of a responsive stimulation and nutrition intervention on child growth and development at 2 years of age.

Key messages
• A responsive stimulation and nutrition intervention delivered to young children from birth to 24 months of age in rural Pakistan improved child cognitive, language, and motor development at 24 months, but not child growth or socio-emotional development.
• Child diet and mother-child interactions mediated intervention effects on child cognitive, language and motor development.
• Although we found no total intervention effects on child growth and socio-emotional development, we observed positive indirect effects on socio-emotional development through both child diet and mother-child interactions, and positive indirect effects on child growth through child diet.
• Leveraging both child diet and mother-child interactions as mechanisms may help enhance intervention effects on child outcomes. built on the standard of care by including information on the association between good nutrition and health, counselling on responsive feeding, and problem solving about feeding. Children in the EN arms also received multiple micronutrient supplementation from 6 to 24 months of age. In each intervention arm, LHWs conducted monthly home visits to deliver the intervention. The RS intervention arms also received monthly community group sessions to support the delivery of the RS package. Mother-child pairs were enroled at birth (child was <2.5 months) and received the intervention until the child reached 24 months of age. The co-primary outcomes of the trial were child development and linear growth at 24 months.
Prior work has shown that the RS and RS + EN interventions improved children's cognitive, language, and motor development at 24 months, whereas the EN intervention improved language development at 24 months and child linear growth from birth to 24 months (Yousafzai et al., 2014). The PEDS cohort was followed-up when children were 4 years old, 2 years after the intervention ended.
Using data from the PEDS trial and this follow-up, N. Brown et al. (2017) showed that improved maternal mental health and mother-child interactions mediated RS and EN effects on child growth at 4 years of age. Obradović et al. (2016) showed that the RS intervention also maintained its positive effects on child cognitive development and executive function at 4 years of age through improved parental stimulation (i.e., parental engagement in play to support children's development) and scaffolding behaviours (i.e., parental support and guidance to help a child learn new age-appropriate skills). Lastly, Jeong et al. (2019) also showed that parental stimulation mediated RS effects on cognitive and socioemotional development at 4 years of age.
We build directly onto this prior PEDS work by examining mother-child interactions as a mechanism to improve child growth and development in the first 2 years of life rather than later in life. In addition, we assessed the RS and EN interventions alone and in combination, which has not been previously done. Expanding prior work, we also examined whether the interventions worked through child diet to improve child growth and development outcomes.

| Conceptual framework
Figure 1 presents our conceptual framework, based on prior models (Brown & Pollitt, 1996;Prado et al., 2017). Positive direct intervention effects on child diet and mother-child interactions at 12 months (see #1 paths in Figure 1) were previously established . We hypothesized positive prospective associations between child diet and mother-child interactions at 12 months and child growth and development at 24 months (see #2 paths in Figure 1), a hypothesis supported by several studies in LMICs (Eshel et al., 2006;Iannotti et al., 2016;Pollitt et al., 2000;Prado, Yakes Jimenez, et al., 2019). Based on the evidence presented above, we also hypothesized positive indirect intervention effects on child growth and development at 24 months F I G U R E 1 Conceptual framework showing the Pakistan Early Child Development Scale-up Trial (PEDS) intervention effects on child growth and development through child diet and mother-child interactions. The numbers describe the direct effects comprising the indirect effects of interest we assessed: (1) direct intervention effects on mediators at 12 months of age, (2) prospective associations between the mediators at 12 months of age and child outcomes at 24 months of age through child diet and mother-child interactions. As significant indirect effects are possible in the context of overall null effects [due to competitive mediation, type 1 or type 2 error, or differential power to detect these effects (Fairchild & McDaniel, 2017;Rucker et al., 2011;Zhao et al., 2010)], we examined all co-primary outcomes regardless of whether the intervention impacted them or not.

| Participants
The analytic sample included 1350 mother-child pairs assessed at enrolment, 12 and 24 months, which represented 91% of the enrolment sample. Participants were excluded due to child death (4.4%), migration outside of the study area (4.3%), or refusal (0.8%).
Children included in the analytic sample and those excluded were generally similar in terms of enrolment characteristics (Table S1).
Therefore, the risk of bias in our sample appeared to be minimal.

| Measures
Intervention exposure was based on the intention-to-treat principle.
We included three binary variables for each intervention arm: RS, EN, RS + EN. Each intervention arm was compared with the control arm.
Child growth was assessed at enrolment, 12 and 24 months using Mother-child interactions at 12 months was assessed using the Observation of Mother-Child Interaction (OMCI) tool . The OMCI was designed to capture maternal responsive behaviours and developmentally supportive interactions, including contingent responding, emotional-affective support, support for infant foci of attention, and language inputs. It consists of 19 items: 12 items for the mother, 6 items for the child, and 1 item on mutual enjoyment. As the reciprocity of the mother-child interactions is key (Black & Aboud, 2011), we used all 19 items in the analysis. The CCDAs observed a live 5-min interaction between the mother and child using a picture book, and rated the frequency of the 19 behaviours using the following scoring criteria: 0 = never occurred, 1 = occurred infrequently (1-2 times), 2 = occurred sometimes (3-4 times), 3 = occurred frequently (5+ times). Total OMCI score (theoretical range 0-57) was calculated by summing the scores on the 19 items. Higher OMCI scores indicate higher frequency of positive interactions and behaviours. Descriptive statistics for the 19 items are shown in Table S2. Further details on the OMCI are available in Rasheed and Yousafzai (2015).
To minimize potential confounding, household wealth and food security, maternal education and mental health, and child age, sex and number of siblings (a proxy for household size) at enrolment were included as covariates. Household wealth was entered as a factor score calculated using principal components analysis based on 44 items assessing property and livestock ownership, and water and electricity access. Household food security was defined based on the Household Food Insecurity Access Scale (Coates et al., 2007).
Maternal mental health was assessed using the self-reported questionnaire (SRQ-20) (World Health Organization, 1994).
Biserial correlations for the analysed variables are shown in Table S3. We observed significant correlations between OMCI and DDS at 12 months and all outcomes at 24 months, lending support to our hypotheses of prospective relations between the mediators and the outcomes.

| Statistical analysis
We used a longitudinal cross-lagged structural equation model (Kline, 2015) to examine the total, direct and indirect effects of the PEDS intervention on child growth and development via child diet and mother-child interactions outlined in Figure 1. Assuming sequential ignorability, the indirect effects we estimate are equivalent to average causal mediation effects (Emsley et al., 2010;Imai et al., 2010;VanderWeele, 2012 comparative fit index (CFI) ≥ 0.90, root mean square error of approximation (RMSEA) ≤ 0.08, and standardized root mean squared residual (SRMR) ≤ 0.08 (Hu & Bentler, 1999).

| Ethical considerations
Ethical approval was obtained from the institutional review board of the Aga Khan University in Pakistan. The technical steering committee of The National Program for Family Planning and Primary Healthcare (also known as the Lady Health Worker program) also provided permission for the trial. Written informed consent was obtained from all participants.
3 | RESULTS  (Table 2). Nevertheless, RMSEA was still close to the prespecified cut-off for acceptable model fit.
All interventions improved cognitive, language, and motor scores at 24 months (Table 3) and DDS and OMCI at 12 months (Table S4).
None of the interventions improved socio-emotional scores or LAZ at 24 months ( Table 3). All of these results were consistent with prior findings from the trial (Yousafzai et al., 2014. Higher OMCI at 12 months predicted higher development scores in all four domains at 24 months, whereas higher DDS at 12 months predicted higher LAZ, cognitive and language scores at 24 months (Table S4).
After controlling for child diet and mother-child interactions as mediators, the RS and RS + EN interventions directly predicted lower LAZ at 24 months (Table S5). There was some indication that the RS intervention predicted lower socio-emotional scores; however, the confidence interval was just above the null. In addition, the RS and

| DISCUSSION
We showed that a responsive stimulation (RS) and enhanced nutrition (EN) intervention improved child cognitive, language and motor development among young Pakistani children through child diet and mother-child interactions. Although only the EN intervention alone Note: The null hypothesis was β = 0. Models for each child development domain were fit separately. Models controlled for the following enrolment characteristics: household wealth, household food security, maternal education, maternal mental health, child age, child sex, child length-forage z-score and number of siblings. Models accounted for clustering and missing values.
T A B L E 4 Standardized indirect intervention effects on child growth and development through child diet and mother-child interactions (as compared with the standard of care) improved children's socioemotional development, we observed positive indirect effects of all three interventions through both child diet and mother-child interactions. Similarly, none of the interventions improved child growth; however, we found positive indirect effects of all three interventions through child diet. These findings suggest that child diet mediated intervention effects on child growth, but both child diet and mother-child interactions mediated intervention effects on child development, and are in line with evidence indicating that the determinants of linear growth and development are only partially shared (Prado, Larson, et al., 2019). Overall, our findings confirmed that all paths included in our conceptual model were important, albeit not all estimates reached statistical significance.
A major contribution of the current study is showing that the RS and EN interventions appeared to have positive cross-over effects on caregiver behaviours, that is, they worked through secondary mechanisms. Specifically, the EN intervention improved child development through mother-child interactions and the RS intervention improved child growth through child diet. These findings lend support to the translation hypothesis put forth by scholars that mothers are able to translate responsiveness and supportive mother-child interactions across facets (Landry et al., 2006;Nahar et al., 2009). The RS intervention improved mother-child interactions, which may have led to better feeding techniques or more responsive feeding (Nahar et al., 2009). Similarly, the EN intervention promoted responsive feeding which may have led to the improvements in mother-child interactions we observed. However, we did not measure responsive feeding or other responsiveness aspects and were therefore unable to formally test this translation hypothesis. Moreover, the mother-child interactions tool we used only captured responsive and supportive interactions in a picture book-reading context, which is different from a feeding context that typically has clear objectives and indicators of success (e.g., child eating). Our findings should be interpreted with caution and replicated in similar samples in other LMICs. Future studies should be specifically designed to assess if, and how, caregivers apply responsive strategies across play, book-reading, feeding and other contexts, and to start building a more substantial evidence base for or against this translation hypothesis.
Another important contribution of our study is that we were able been proposed as one approach to address environmental risks (Ngure et al., 2014), no evidence to date exists on the feasibility, acceptability and effectiveness of integrating RS and baby WASH inputs. Evidence on successful integration of EN and household-level WASH inputs is emerging, but effects on child growth and development are mixed (Pickering et al., 2019;Stewart et al., 2018;Tofail et al., 2018).
Several limitations of our study should be noted. First, child diet was based on maternal report, which could be subject to social desirability bias, and was limited to a single recall over the past 24 h, which did not capture food quantity or nutrient content. Moreover, although we controlled for household food security, household wealth, and maternal education, we lacked data on other community-, household-and maternal-level factors that influence child diet, for example, food availability, seasonality, and affordability, maternal awareness of children's nutritional needs, child eating skills and acceptance. Second, the mother-child interactions tool we used does not eliminate observer bias and the observed mother-child interactions may be different from regular interactions. Importantly, the tool measured mother-child interactions only in a picture bookreading context. More work is needed to confirm its ability to measure mother-child interactions in other situations or settings. Third, omitted variables may bias our results. While we controlled for several important confounders, unmeasured and unobserved variables may confound the relations of interest, particularly those between the mediators and outcomes.
Despite these limitations, our findings have two main implications for multi-input interventions aiming to improve child growth and development in LMICs: (1) such interventions should combine RS and EN inputs, and (2) such interventions should leverage both child diet and mother-child interactions as mechanisms. First, although we observed similar indirect effects through mother-child interactions among children exposed to the RS and RS + EN interventions, we observed larger indirect effects through child diet among children , and that multi-input interventions can be more effective (World Health Organization, 2020), though we did not explicitly test for additive or synergistic effects. Second, although child diet alone mediated intervention effects on child growth, both child diet and mother-child interactions mediated intervention effects on child development. Given the growing evidence that the determinants of child growth and development are only partially shared (Prado, Larson, et al., 2019), future interventions aiming to improve both outcomes should be designed to leverage both shared (e.g., child diet) and separate (e.g., mother-child interactions) mechanisms. Currently, many multi-input interventions promote responsive stimulation and teach optimal child feeding practices. Our findings further build the case for these types of interventions by strengthening the evidence base and demonstrating that improvements in mother-child interactions and child diet translate into tangible benefits for child growth and development. The emerging evidence on cross-over effects suggests that these two mechanisms are mutually reinforcing. Thus, interventions that promote both mechanisms will likely be more effective at improving child outcomes than interventions promoting only one of the mechanisms.
Our findings build on prior PEDS work by demonstrating that child diet and mother-child interactions mediated intervention effects on child growth and development in the first 2 years of life.
Prior PEDS studies showed that maternal mental health and parental stimulation were other mechanisms through which the intervention worked Jeong et al., 2019;Obradović et al., 2016). Taking all these findings together, the PEDS interventions helped build an enabling and nurturing environment which directly benefited mothers and their children both during the intervention implementation period and after the intervention ended. These findings suggest that similar types of multi-input integrated RS and EN interventions can serve as a platform to enable, empower, and support caregivers, which in turn can improve not only child growth and development, but also child wellbeing more generally and the overall conditions in which children grow up. Future interventions in LMICs should focus on holistic approaches, which aim to improve multiple aspects of nurturing care rather than on individual child outcomes.