Maternal caregiving capabilities are associated with child linear growth in rural Zimbabwe

Abstract Between birth and 2 years, children's well‐being depends on the quality of care they receive from caregivers, primarily their mothers. We developed a quantitative survey instrument to assess seven psychosocial characteristics of women that determine their caregiving ability (‘maternal capabilities’: physical health, mental health, decision‐making autonomy, social support, mothering self‐efficacy, workload and time stress, and gender norm attitudes). We measured maternal capabilities in 4,025 mothers and growth in their 4,073 children participating in the Sanitation Hygiene Infant Nutrition Efficacy (SHINE) trial in rural Zimbabwe. We used generalized estimating equation models with exchangeable correlation structure to test the association between each maternal capability during pregnancy, and infant length‐for‐age Z (LAZ) at 18 months, accounting only for within‐cluster correlation and intervention arms in unadjusted analyses and for potential confounders in adjusted analyses to examine the association between each capability, assessed during pregnancy, with child LAZ at 18 months of age. In adjusted models, each unit increase in gender norm attitudes score (reflecting more equitable gender norm attitudes) was associated with +0.09 LAZ (95% CI: 0.02, 0.16) and a decreased odds of stunting (adjusted odds ratio [AOR]: 0.86; 95% CI: 0.74, 1.01); each unit increase in social support score was associated with +0.11 LAZ (95% CI: 0.05, 0.17, p < 0.010) and decreased odds of stunting (AOR: 0.83; 95% CI: 0.73, 0.96). Each unit increase in decision‐making autonomy was associated with a 6% reduced odds of stunting (AOR: 0.94; 95% CI: 0.89, 0.996, p = 0.04). Interventions and social programming that strengthen these maternal capabilities may improve child nutritional status.

time stress, and gender norm attitudes). We measured maternal capabilities in 4,025 mothers and growth in their 4,073 children participating in the Sanitation Hygiene Infant Nutrition Efficacy (SHINE) trial in rural Zimbabwe. We used generalized estimating equation models with exchangeable correlation structure to test the association between each maternal capability during pregnancy, and infant length-for-age Z (LAZ) at 18 months, accounting only for within-cluster correlation and intervention arms in unadjusted analyses and for potential confounders in adjusted analyses to examine the association between each capability, assessed during pregnancy, with child LAZ at 18 months of age. In adjusted models, each unit increase in gender norm attitudes score (reflecting more equitable gender norm attitudes) was associated with +0.09 LAZ (95% CI: 0.02, 0.16) and a decreased odds of stunting (adjusted odds ratio [AOR]: 0.86; 95% CI: 0.74, 1.01); each unit increase in social support score was associated with +0.11 LAZ (95% CI: 0.05, 0.17, p < 0.010) and decreased odds of stunting (AOR: 0.83; 95% CI: 0.73, 0.96). Each unit increase in decision-making autonomy was associated with a 6% reduced odds of stunting (AOR: 0.94; 95% CI: 0.89, 0.996, p = 0.04). Interventions and social programming that strengthen these maternal capabilities may improve child nutritional status.
Members of the SHINE Trial Team are listed at https://doi.org/10.1093/cid/civ844 1 | INTRODUCTION Among children under 5 years of age, 139 million (22%) are stunted, defined as having a length or height more than 2 standard deviations below the age-and sex-matched international reference population median (UNICEF, World Health Organization [WHO], & World Bank, 2017). Incident stunting is highest between conception and 2 years of age, after which there is little recovery (Shrimpton et al., 2001). Stunting is associated with reduced neurodevelopment and lower educational attainment during childhood (Dewey & Begum, 2011) and reduced economic productivity during adulthood (Hoddinott et al., 2013). Thus, stunting is an indicator of lifelong lost human capital, and interventions to prevent stunting must occur before 2 years of age.
Virtually all interventions to improve linear growth and all other nutrition and health outcomes in infants and young children (e.g. feeding, and provision of hygienic living conditions and health care) rely on implementation by a caregiver. In most contexts, this is primarily the child's mother. The UNICEF Framework on malnutrition highlighted 'nurturing care' as a critical requirement for healthy child growth and development (UNICEF, 1990).
Differences in the quality of care that children receive likely contribute to heterogeneity in nutritional status among children living in the same community with similar access to resources. Accordingly, we sought to better understand caregiving behaviour by identifying underlying psychosocial characteristics of women that determine caregiving ability. Building on the extended UNICEF model of care (Engle, Menon, & Haddad, 1999), the human capabilities theory (Nussbaum, 2001) and empirical studies, we developed a 'maternal capabilities' construct. We defined maternal capabilities as the characteristics of a mother that determine her ability to care for a young child in ways that produce positive nutrition, health and developmental outcomes. Our construct includes seven capabilities: physical and mental health, decisionmaking autonomy, social support, mothering self-efficacy, workload and time stress, and gender norm attitudes. We developed a quantitative tool to measure these maternal capabilities, primarily drawing on previously published instruments (Matare, Mbuya, Pelto, Dickin, & Stoltzfus, 2015). We defined strong maternal capabilities as good physical and mental health, high levels of social support and mothering self-efficacy, high autonomy for decision making within the household, egalitarian gender norm attitudes and low levels of perceived time stress. We hypothesized that children of mothers with strong maternal capabilities will receive better care and therefore have better nutritional status than children of mothers with weak maternal capabilities.

| METHODS
The newly developed maternal capabilities tool is available at https:// osf.io/w93hy; Table 1  and less stunting at 18 months of age.
SHINE was a cluster-randomized community-based trial conducted in two contiguous rural districts in central Zimbabwe (Chirumanzu and Shurugwi); methods have been previously described (SHINE Trial Team, 2015). Briefly, clusters were defined as the catchment area of one to four village health workers (VHWs) employed by the Ministry of Health and Child Care (MoHCC). The clusters were randomly allocated to one of four treatment groups: Standard of Care

Key messages
• The well-being of young children depends on the quality of care provided by adults-usually their mothers-for feeding, shelter, health care and nurturing stimulation.
• Maternal capabilities are skills and attributes of mothers that determine their ability to translate resources (food, health care, education and shelter) into positive nutrition, health and neurodevelopmental outcomes of their children.
• Among 4,025 rural women in Zimbabwe, holding more equitable gender norm attitudes, having greater decisionmaking autonomy and having more social support during pregnancy were each associated with better attained linear growth at 18 months of age in their children.
• Interventions that provide social support to mothers, for example, through home visits or women's groups, may improve health outcome in their children.
• Achieving global child health and nutrition goals will likely require progress toward empowerment of women within their societies and gender equity.
T A B L E 1 The maternal capabilities construct implemented in the SHINE trial: Definitions, justifications and measurement methods

Maternal capability Definition Justification Measurement
Decision-making autonomy Choice or control in decisions and access to resources that affect the mother herself or her children (Dixon-Mueller, 2013) Child nutritional status is generally associated with maternal autonomy, in particular maternal control over health care decision (Carlson, Kordas, & Murray-Kolb, 2015).
Questions were taken from the Zimbabwe Demographic and Health Survey 2015; questions ask whether a mother can make a decision on her own to purchase 7 items of varying value Gender norm attitudes Egalitarian gender norm attitudes are a belief that men and women should have equal access to resources and opportunities Mothers with more egalitarian attitudes are more likely to have an institutional delivery and a fully immunized child and to practice exclusive breastfeeding Singh, Haney, & Olorunsaiye, 2013).
Questions adapted from the Gender Norm Attitudes Scale (Nanda, 2011) Mental health Depression; loss of interest or pleasure in usual activities, reduced energy, feelings of guilt, low self-worth, disturbed sleep or appetite, poor concentration Depression reduces maternal-child interaction (Poobalan et al., 2007); child health-care seeking (Avan, Richter, Ramchandani, Norris, & Stein, 2010) has a negative impact on caregiving (WHO, 2012) and is associated with poor nutritional status (Surkan, Kennedy, Hurley, & Black, 2011).
The Edinburgh Postnatal Depression Scale was used (Cox, Holden, & Sagovsky, 1987), which has been translated and validated in Shona among Zimbabwean women (Chibanda et al., 2010) Mothering selfefficacy A woman's self-confidence in her role as a competent mother Self-efficacy is a proximal determinant of behaviour in many models (Bandura, 1977;Becker, 1974;Fishbein, 1979). There is a strong evidence that self-efficacy is an important determinant of breastfeeding (Tuthill, McGrath, Graber, Cusson, & Young, 2016).

Perceived physical health
The extent to which a woman perceives herself to be free of disease and in good health Women in low-and middle-income countries face high burdens of disease (Murray et al., 2012), which may reduce their energy levels and quality of childcare (Engle, Menon, & Haddad, 1999).
Adapted from the RAND 36-Item Health Survey (Hays, Sherbourne, & Mazel, 1993) Social support Access to relationships that provide physical, psychological and informational resources, which help the mother feel she belongs, is competent and can cope.
Questions adapted from the Interpersonal Support Evaluation List (Cohen, Mermelstein, Kamarck, & Hoberman, 1985) and the Medical Outcomes Study Social Support Survey (Sherbourne & Stewart, 1991) Time stress The extent to which a mother feels she does not have adequate time to complete all her responsibilities.
Women's work burden may negatively influence childcare because women's time in low-and middle-income countries is a zero-sum game: new activities can only be added at the expense of others (Engle, Menon, & Haddad, 1999).
Questions developed for the SHINE trial. Questions ask whether the mother feels stress, unhappiness or worry about her work load (SOC, promotion of uptake of MoHCC maternal-child health services, early breastfeeding initiation and exclusive breastfeeding to 6 months of age), WASH (all SOC interventions plus provision of a latrine, handwashing stations, soap, chlorine for point-of-use drinking water treatment and a play space to reduce infant geophagia with promotion of hygiene behaviours); IYCF (all SOC interventions plus 20 g of daily small-quantity lipid-nutrient supplement between infant ages 6 to 18 months, and promotion of optimal complementary feeding practices); or WASH+IYCF (all interventions delivered concurrently). HIV-positive women were urged to seek immediate antenatal care for prevention of mother-to-child transmission. Maternal and household characteristics were elicited using a structured questionnaire. Date of last menstrual period (LMP) was recorded. Household food security status was assessed using the Coping Strategies Index (Maxwell, Watkins, Wheeler, & Collins, 2003) and relative wealth using an asset-based index as described (Chasekwa et al., 2018 Primary results of the SHINE trial showed that the IYCF, but not the WASH, intervention improved child linear growth Prendergast et al., 2019). We also observed a higher prevalence of stunting among HIV-uninfected children born to HIV-positive mothers than among children of HIV-negative mothers (stunting prevalence was 50% compared with 32%). Accordingly, our statistical analysis plan (SAP) prespecified that we would test for effect modification between IYCF treatment and each maternal capability, and between maternal HIV infection and each maternal capability on infant LAZ at 18 months. There is also evidence that mothers care for boys differently than girls (Yount, 2004), so we prespecified testing interaction between infant sex and maternal capabilities on growth.
Several studies have reported that maternal depression during pregnancy is associated with reduced birth weight (Field et al., 2004;Nasreen, Kabir, Forsell, & Edhborg, 2010); therefore, we hypothesized that any observed associations between maternal capabilities and LAZ at 18 months could be mediated through birth weight. In summary, we prespecified our plan to test for interaction between the maternal capabilities and three factors (maternal HIV status, infant sex and the IYCF treatment arm) on infant LAZ at 18 months and to determine if any observed associations between maternal capabilities and infant LAZ at 18 months were mediated by birth weight.

| Statistical methods
All mother-infant dyads enrolled in SHINE who provided maternal capability information at baseline and child LAZ at 18 months were included in this analysis. LAZ was calculated according to WHO Child Growth Standards; stunting was defined as LAZ < −2.0. The EPDS scores were modelled as depressed (score ≥ 12 and/or suicidal ideation) or not depressed (Chibanda et al., 2010). Decision-making autonomy was calculated as the sum of five questions, each of which could be scored as 0 (no) or 1 (yes). The median (IQR) decision-making autonomy score was 5 (4, 5) out of a range of possible scores of 0-5; accordingly, we modelled this variable as a binary variable, where 1 indicated a score of 5 and 0 indicated a score < 5. For each of the other maternal capabilities, the mean of several Likert-type items was calculated to give a composite score.

| Associations between maternal capabilities and child linear growth
In unadjusted analyses, we used generalized estimating equation (

| Effect modification
We hypothesized that the association of the maternal capabilities with child LAZ may be modified by three factors: infant sex, maternal HIV status or the IYCF intervention. We used regression analysis to calculate interaction terms. Where the interaction was significant (p < 0.05), stratified analyses were conducted. For maternal capabilities that were not binary, we categorized them as high or low at the median.

| Effect mediation
We tested the association of each maternal capability with birth weight using the same GEE modelling approach as used for LAZ. If any maternal capability was significantly associated with both birth weight and LAZ, we used a mediator model to investigate whether the association between the capability and LAZ was mediated through birth weight.
All analyses were performed using STATA version 14.1 (College Station, TX: StataCorp, LP).

| Ethical considerations
The Medical Research Council of Zimbabwe and the Institutional

Review Board of the Johns Hopkins Bloomberg School of Public
Health reviewed and approved the study protocol.

| RESULTS
Among 5,280 women enrolled in SHINE, 11 were excluded and one was added owing to enrolment errors; 370 had miscarriages, had still births or died during pregnancy; 139 were lost to follow-up or exited from the trial before delivery; and 378 mothers did not provide maternal capability data at baseline (Figure 1). social support remained significantly associated with LAZ, and decision-making autonomy and social support remained significantly associated with stunting. Mothers with greater time stress during pregnancy were also more likely to have a child with lower LAZ at 18 months, although this association did not reach statistical significance (−0.04; 95% CI: −0.08, 0.004, p = 0.07). In post hoc analyses, we repeated these analyses with the outcome of LAZ and stunting at 1, 3, 6 and 12 months post-partum. Mothers who perceived themselves to be in better health compared with worse health had children with better linear growth at 1 and 12 months of age (Tables S4A and   S4B). Similar to 18-month findings, maternal decision-making autonomy, social support and time stress were associated with child linear growth at 12 months (Table S4B).
The IYCF intervention significantly modified the associations of depressive symptoms (p value of interaction term = 0.029) and gender norm attitudes (p value of interaction term < 0.001) with child LAZ at 18 months. Stratified analyses are presented in Table 2. To illustrate these interactions, we plotted the mean LAZ of children who did and did not receive the IYCF intervention stratified by those whose mothers held restrictive and egalitarian gender norm attitudes ( Figure 2A) and by those whose mothers were and were not depressed (EPDS during pregnancy; Figure 2B). Among children who did not receive the IYCF intervention, those whose mothers held restrictive compared with egalitarian gender norm attitudes and those whose mothers were depressed compared with not depressed were shorter. However, the LAZ of children who received the IYCF intervention did not differ by their mother's gender norm attitudes or depression. These graphs illustrate that the SHINE IYCF intervention ameliorated the adverse associations of child linear growth with maternal inequitable gender norm attitudes and with maternal depression. There were no other significant interaction terms between a maternal capability assessed during pregnancy and the IYCF intervention or child sex or maternal HIV status.
We found no evidence to support our hypothesis that the association between maternal capabilities assessed during pregnancy and child LAZ at 18 months is mediated through birth weight. Mothering self-efficacy was associated with birth weight (Table 3), but it was not associated with child LAZ at 18 months ( Table 2). None of the other maternal capabilities were associated with birth weight (Table 3).

| DISCUSSION
In this population of 4,025 mothers in rural Zimbabwe, those who reported greater decision-making autonomy, more egalitarian gender norm attitudes and higher levels of social support during pregnancy had children who attained better linear growth by 18 months.
Decision-making autonomy was very high among SHINE women: out of a possible score of 5, the median score during pregnancy was 5 (IQR: 4, 5). Yet even within this tight distribution, each unit increase in score was significantly associated with a 6% reduced odds of having a stunted child. Our findings on gender norms are consistent with a recent cross-sectional analysis of Demographic and Health Survey data from five east sub-Saharan African countries that reported that maternal gender norm attitudes (termed 'intrinsic agency' in that paper) were significantly associated with child height-for-age difference Z (HAZ); the effect was both direct and mediated through maternal body mass index, suggesting that holding restrictive gender norm attitudes may negatively affect self-care as well as childcare (Jones et al., 2019). Our findings on social support suggest that the positive child health effects of programmes that provide child health and F I G U R E 2 The effect modification on length-for-age Z (LAZ) outcomes in the Sanitation Hygiene Infant Nutrition Efficacy (SHINE) trial.
(A) Effect of infant and young child feeding (IYCF) intervention on LAZ stratified by gender norm attitudes (interaction term, p < 0.001). (B) Effect of IYCF intervention on LAZ stratified by depressive symptoms (interaction term, p = 0.036). Among children in the non-IYCF arms, those whose mothers are with restrictive gender norm attitudes and those whose mothers are with depression had lower LAZ than those whose mothers had stronger maternal capabilities. But among children in the IYCF arm, children of mothers with weak maternal capabilities benefited from the IYCF intervention more than those whose mothers are with strong capabilities: The IYCF intervention ameliorated the adverse association of weak maternal capabilities on linear growth Our results predict that SHINE mothers who had optimal scores for gender norm attitudes and social support (5 for both domains) during pregnancy, as compared with mothers who had the median scores Maternal depression during pregnancy has been frequently reported to predict low birth weight (Nasreen, Kabir, Forsell, & Edhborg, 2010;Rahman, Bunn, Lovel, & Creed, 2007;Wado, Afework, & Hindin, 2014). In our study, although not statistically significant, depressed mothers were 35% less likely to have a low birth weight baby (OR; 95% CI: 0.65 0.39, 1.09, p = 0.10). Our study sample had a relatively low prevalence of antenatal depression (8.8%), compared with studies that have found this association in populations with antenatal depression > 20% (Chang et al., 2014;Wado, Afework, & Hindin, 2014). We have previously demonstrated that maternal depression, gender norm attitudes and social support assessed during T A B L E 3 Association of maternal capabilities assessed during pregnancy with child length-for-age Z score and stunting at 18 months of age Unadjusted models-adjusted for cluster and intervention arms to account for within-cluster correlation and controlling for the intervention effect. b Adjusted models-adjusted for cluster, intervention arms and other baseline covariates, which were associated with the exposure and outcome at p < 0.2.
Baseline variables tested: age of mother at birth, mother's height, mother's body mass index, mother's mid-upper arm circumference, mother's level of education, receipt of water, sanitation and hygiene (WASH) intervention, receipt of IYCF intervention, child birth order, child sex, single or multiple birth, birth interval, time to water source, improved latrine, improved water source, household wealth index, presence of functioning hand washing station with water and soap, coping strategy index, gender of household head, faeces observed in yard, number of children under-fives, household size, treated water, father's level of education, mother's employment outside home, household dietary diversity and mother's HIV status at enrolment. See Table S3 for lists of covariates retained in each final adjusted model. c Interaction term between IYCF * Gender norm attitudes was significant (p < 0.001), so stratified analyses are presented showing that the association of gender norm attitude with LAZ was apparent only for children who did not receive the IYCF intervention. d Interaction term between IYCF * Depression was significant (p = 0.036), so stratified analyses are present showing that the association of depression with LAZ was apparent for children who did not received the IYCF intervention, although after adjustment, this association was not statistically significant.
pregnancy predicted several childcare behaviours (institutional delivery, breastfeeding initiation, exclusive breastfeeding, having a fully immunized child and complementary feeding) (Matare et al., 2020). In the current analysis, we demonstrate that these same capabilities in pregnancy are not associated with birth weight.
Together, these observations suggest that maternal capabilities most likely affect child growth through improved care practices rather than a biologic pathway.
We did not observe associations between linear growth and perceived health status. Notably, most mothers were in good health status (well-nourished, 80% of HIV+ women were taking antiretroviral therapy (Prendergast et al., 2019) and had very little anaemia (personal communication, JH Humphrey). It may be that some mothers perceived themselves to be in worse health than they were and that their objective health status did not constrain childcare. We also did not observe an association between linear growth and mothering self-efficacy. This may be because our tool reflected global mothering self-efficacy; it may be that self-efficacy in specific skills more directly related to child linear growth (e.g. complementary feeding) may be associated with linear growth.
In the past 5 years, four review papers have summarized the literature on maternal empowerment (defined as women's ability to make strategic life choices) and child nutritional status (Carlson, Kordas, & Murray-Kolb, 2015;Cunningham et al., 2015;Hossain, 2020;Pratley, 2016;Santoso et al., 2019): all reported that mothers' empowerment was generally associated with better child nutritional status, but with many reported associations not reaching statistical significance and many inconsistencies between studies. A major obstacle to interpreting this literature is the wide variety of definitions and tools used in assessing empowerment. In most studies 'empowerment' has included some assessment of decision-making autonomy (Aslam & Kingdon, 2012;Desai & Jain, 1994;Hossain, 2020;Shroff, Griffiths, Adair, Suchindran, & Bentley, 2009); some studies have included women's mobility (Sethuraman, Lansdown, & Sullivan, 2006), and others have relied on nonspecific indicators like education, employment and household size (Shafiq et al., 2019). In developing the set of tools used in the current study, we took a broader approach to assess a comprehensive set of skills and attributes that aimed to distinguish women who are able to take optimal advantage of available resources to care for their children from those with similar resources who are less capable. In our construct, the ability to make strategic life choices is just one determinant: a woman's physical and mental health, her self-confidence in being a mother and her role in society are also critical. Future research should test these tools in other contexts and elucidate the mechanisms linking capabilities to child health outcomes.
Our study has two important limitations. First, although the longitudinal design provides stronger evidence for evaluating the relationship between early exposures and child health outcomes than cross-sectional studies, the data are observational, subject to confounding by unmeasured factors in drawing causal inference. Our study has identified specific maternal factors that may particularly influence child health outcomes; future studies could test the impact of interventions to develop or strengthen these skill and qualities in mothers using a randomized design. Second, of the seven survey tools developed, only the tool and cut-off used for depression have been validated against psychometric measures. However, all the tools were pretested with rural Zimbabwean women to ensure face and content validity of the questions and context relevance of the response options.
In this study of 4,025 women in rural Zimbabwe, maternal decision-making autonomy, gender norm attitudes and social support during pregnancy significantly predicted attained linear growth of their children 2 years later, when the child was 18 months of age.
These findings suggest that achieving global child health goals will require improvements in the empowerment of women within their societies. Interventions to provide social support (i.e. regular and reliable interaction with caring, trusted people) may be 'lower hanging fruit', but longer term interventions are underway to address power asymmetries and achieve gender equity.

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