The prevalence of stunting in our study was “very high” as per WHO  classification and also in comparison to worldwide rates of stunting in other LMICs that range from 5 to 65 % . Recorded rates of wasting and underweight, as defined by the WHO, were categorized as “poor” and “medium prevalence,” respectively . Stunting prevalence is of particular concern since it reflects long-term structural factors of undernutrition and can serve as an indicator of a population’s well-being . The WHO  and other studies [9, 23] have found that stunting prevalence increases with age from birth until 24 to 36 months and then plateaus. In contrast, our data found that children with increasing age, above 2 years, were at highest risk of being stunted and/or underweight. The aggregated 2010 and 2014 stunting data reveals a non-linear association between stunting and child’s age (Fig. 2). This phenomenon is likely due to a variety of factors. First, Ogumaniha interventions targeted poor nutrition through promotion of exclusive breastfeeding for children 6 months and younger, training of community volunteers on the preparation of foods high in essential micronutrients, and identification of children with moderate undernutrition within communities for caloric supplementation and/or referral to a health facility for treatment if necessary. The lower odds of stunting in younger children may be evidence that they received and positively responded to these interventions, while the higher odds of stunting at later ages (36 to 59 months) may largely be due to these children having past the window of opportunity to prevent the permanent effects of undernutrition when interventions were introduced. Second, as a child is weaned from the breast, their growth and nutrition may be negatively affected by the mother’s inability to continue meeting the nutritional needs of the growing child with breast milk alone and/or by either the introduction of high-starch, low-protein foods or by a delayed introduction of supplemental foods. The median length of time for exclusive breastfeeding in our study increased between 2010 and 2014, thus potentially delaying the negative chronic effects of undernutrition as children were weaned later. This is further supported by our finding of an 80 % decreased odds of wasting in 2014 in children that exclusively breastfed for a minimum of 6 months and a five-fold increased odds of wasting in children who had delayed introduction of supplemental food.
Our finding in multivariable analysis of a protective association between receipt of vitamin A supplementation and decreased odds of stunting is consistent with similar studies in LMICs [4, 20, 23, 29–31]. These studies found that micronutrient deficiencies, including vitamin A, iron, and zinc, had a significant impact on a child’s height potential. Our survey questionnaire did not ask about zinc and iron supplementation, as they are not commonly administered in the province. In 2010, 70 % of all children reported receipt of vitamin A supplementation while only 37 % reported receiving supplementation in 2014. Although this stark difference could be attributed to missing data (40 % was missing in 2010), it is vital to increase the coverage of vitamin A supplementation or consider the implications of introducing foods fortified with vitamin A, such as sweet potatoes .
Higher proportions of children with undernutrition came from households with a higher monthly income, though no associations were detected between income and either stunting or wasting in multivariable analysis. This counterintuitive finding may result from the fact that there is likely no meaningful difference between the measured 300 versus 150 meticais per month (a difference of approximately $4 USD) in terms of purchasing power at a household level. Additionally, almost one quarter of interviewees did not know their household income, which is common among subsistence farmers and those who rely on seasonal and informal markets. In this setting, non-monetary measures of poverty, such as the multidimensional poverty index which classifies one’s poverty based on measures such as health, education, and material means may be more appropriate .
The perceived association between increased stunting and greater household dietary diversity could be explained by an inability to capture the individual child’s dietary diversity with the survey tool used. Anecdotal evidence from Zambézia Province suggests there is a hierarchical determination of diet, such that male adults within a given household may have a more diverse diet than females and children respectively, and that natural children within a household may have better diets than children taken into the home as orphans or from extended family, despite these foods being available in the household. While providing a potential measure of overall household access to resources, dietary diversity may not be a sensitive marker of under-five malnutrition, particularly in infants younger than 9 months, who may not eat all foods available in the household or who are exclusively breastfeeding.
Multivariable analysis revealed that being food insecure more than doubled the odds of being underweight in 2010 and wasted when analyzed in aggregate. This finding aligns with our expectations because wasting and underweight are traditionally used to monitor short-term deprivation of food, while stunting reveals reoccurring nutrient inadequacies over a lengthier period of time [23, 34, 35].
Hand washing practices that included the using of a cleaning agent decreased the odds of being underweight by 40 %. Although 95 % of interviewees reported washing hands the previous day, only 72 % washed their hands with soap, detergent, or ashes. The use of a cleaning agent is necessary to remove grime and bacteria from the hands, and ashes have been reported to be an effective, low-cost alternative to soap in preventing the spread of diarrheal illness . Future interventions in Zambézia Province should focus on education and economic interventions that will continue to increase the use of cleaning agents.
In our multivariable analysis, there was little evidence that the use of safe water was associated with undernutrition. Eighty percent of households used safe water and this lack of variability in the data may have made it difficult to observe a correlation between this variable and the outcome variables. Despite using safe water, roughly 30 % of children were reported as having diarrheal illness within the past month, regardless of nutritional status. We did not detect a statistically significant association between diarrheal illness and undernutrition, although recurring infectious disease and unhealthy environmental conditions have been correlated with stunting in other settings [4, 16, 23, 30, 37]. This undetectable correlation in our study could be attributed to the aforementioned lack of variability in the population. Future directions of study should include collecting bio-specimens to determine if environmental enteropathy, a subclinical condition of the gut that has been found to be an intermediary pathway connecting poor hygiene to growth deficits  plays a role in the undernutrition of this population.
Strengths of this study include a large sample size and survey design that was representative of the province. Survey interviewers were Zambézia residents that administered the surveys in one of five local languages known to the interviewee. The models in the analyses were developed a priori and tested non-linearity of covariates with restricted cubic splines. A study limitation was that the breastfeeding questions referred to the female head of household’s actions with her youngest child, which may not have been the child that was randomly selected for measurement. We assumed that practices were kept consistent among children, but there is a possibility of variance that was not accounted for in the data. Moreover, survey design was more heavily weighted towards questions that measured household level factors of undernutrition than for individual level factors. As such, it is likely we did not capture those individual level determinants that would prove important in impacting the high rates of undernutrition in this population, such as maternal undernutrition, birth weight, poor feeding practices, infectious disease specific history, and duration of disease. A final limitation of this study is the timing of the baseline and program-end surveys as they were conducted at two distinct periods of the year. Data collected in 2014 occurred at the end of the rainy, or “hunger,” season, which may have influenced the increased prevalence of stunting as well as income and food related variables, such as dietary diversity and food insecurity. However, stunting is the result of long-term food deprivation, so it is likely that the timing of surveys had a low impact on this outcome variable.