Individual and contextual predictors of children’s hemoglobin levels from Southern Brazilian municipalities in social vulnerability Preditores individuais e contextuais dos níveis de hemoglobina em crianças de municípios do Sul do Brasil com vulnerabilidade social Predictores individuales y contextuales de los niveles de hemoglobina en niños socialmente vulnerables de municipios del Sur de Brasil

Few studies have investigated the simultaneous effect of individual and contextual factors on the occurrences of anemia. This study aims to evaluate the variability of children’s hemoglobin levels from municipalities in social vulnerability and its association with factors of individual and municipal nature. This is a cross-sectional, multi-center study, with children data (12-59 months) collected from 48 municipalities of the Southern region of Brazil, that were included in the Brazil Without Poverty Plan. Individuals’ data were collected using a structured questionnaire, and secondary and ecological data of children’s municipalities were collected via national surveys and institutional websites. The outcome was defined as the hemoglobin level obtained by HemoCue. A multilevel analysis was performed using Generalized Linear Models for Location Scale and Shape using R, with a 5% significance level. A total of 1,501 children were evaluated. The mean hemoglobin level was 12.8g/dL (95%CI: 12.7-12.8), with significant variability between municipalities. Lower values of hemoglobin were observed in children who lived in municipalities with a higher urbanization rate and a lower number of Community Health Agents, in relation to the reference categories. At the individual level, lower hemoglobin values were identified for children under 24 months, not enrolled at daycares, who were beneficiaries of the conditional cash transfer program and diagnosed with underweight. The results shed light on important factors at the municipal and the individual levels that were associated to the hemoglobin levels of children living in municipalities in social vulnerability. Hemoglobin; Anemia; Child; Multilevel Analysis Correspondence D. L. F. Silva Grupo de Exposição Alimentar, Departamento de Nutrição, Universidade Federal do Paraná. Rua Poeta Francisco Ferreira Leite 40, Curitiba, PR 82900-010, Brasil. deborafrizzi@gmail.com 1 Departamento de Nutrição, Universidade Federal do Paraná, Curitiba, Brasil. 2 Departamento de Estatística, Universidade Federal do Paraná, Curitiba, Brasil. 3 Departamento de Nutrição, Universidade Federal de Ciências da Saúde de Porto Alegre, Porto Alegre, Brasil. 4 Departamento de Nutrição, Universidade Federal de Santa Catarina, Florianópolis, Brasil. 5 Coordenação Geral de Alimentação e Nutrição, Ministério da Saúde, Brasília, Brasil. 6 Departamento de Nutrição, Universidade de Brasília, Brasília, Brasil. This article is published in Open Access under the Creative Commons Attribution license, which allows use, distribution, and reproduction in any medium, without restrictions, as long as the original work is correctly cited.


Introduction
Low blood levels of hemoglobin characterize anemia, a common deficiency in children under five years of age, with biological, social, and economic implications. According to estimates from the World Health Organization, 273 million children around the world presented low levels of hemoglobin in 2011, with variable distribution among the continents; with numbers up to 12.7 million children in Europe, 17.1 million in the Americas, and 96.7 million in Southeast Asia 1 .
Despite the methodological differences observed in the studies, anemia is usually considered a moderate problem of public health in children under five years of age in Brazil 1,2 . Based on hemoglobin levels, the occurrence of anemia is frequent in this age group, and it can be a severe condition in population groups living on areas of socio-economic vulnerability 3,4,5,6,7 . Therefore, changes in life conditions and reduction of social inequalities have been prioritized in Brazil by policy makers, especially for infant's first years of life 8,9,10 .
The implementation of social programs and policies in Brazil since the 1990s has contributed to the increase of the population's income and their access to public goods and services 11,12 . However, evidence shows that the impact of these programs and policies on children's health and nutrition was not homogeneous throughout the country, even within these regions 13 . To reduce these inequalities, the Brasil Sem Miséria (Brazil Without Poverty) Plan was launched in 2011, not only to try to remove millions of Brazilians from poverty, but also to reduce the manifestations and aggravation of this condition 14 . The government designed this Plan to help the less informed, most fragile and most isolated population, which is usually the group least assisted by public policies, and that often only has access to low-quality services 15 . Among the main strategies of this Plan, there was the Brasil Carinhoso (Brazil that Cares) action, which prioritized strategies related to early childhood, involving the prevention of iron deficiency in this group 16 .
Although iron deficiency due to dietary intake is recognized as the main cause of low hemoglobin levels 1,17 , there is a growing interest in the influence of living conditions on the development of this deficiency in children 18 . Previous studies 19,20,21,22,23,24,25 have emphasized the impact of poor living conditions on the prevalence of anemia and hemoglobin levels. Nevertheless, few have investigated the simultaneous effect of individual and community factors on the occurrence of anemia 26 . Thus, this research intended to evaluate the variability of children's hemoglobin levels from the Southern municipalities included in the Brazil Without Poverty Plan and their association with factors at the municipal and individual levels.

Methods
This was a cross-sectional multi-center study on the prevalence of hypovitaminosis A and anemia in children (12-59 months) residing in municipalities from the Southern region of Brazil, comprising the states of Paraná, Santa Catarina and Rio Grande do Sul. This region has one of the highest urbanization rates, level of education, and average monthly income in Brazil 27 . It also holds the smaller coefficients of infant mortality and illiteracy 27 . Despite these good indicators, in 2014, 2.4% of the population from this region lived in poverty with a monthly income per capita less than BRL 140.00 (approximately USD 42.04) and 0.7% in extreme poverty, earning less than BRL 70.00 (approximately USD 21.02) 28 .
As for the sample size, the estimation was done considering the set of municipalities that were exclusively members of the Brazil Without Poverty Plan in 2012 that were not covered by the National Vitamin A Supplementation Program in 2014. This set was adopted because the determination of hypovitaminosis A was also a goal of the multi-center study. In each of the three Southern states, the primary sampling unit (PSU) was considered to be the municipalities. The calculation of the sample size considered a 21.5% prevalence of anemia in children less than five years old in Southern Brazil, maximum error of 5%, 95% confidence interval (95%CI), infinite population, and sampling and design effect (deff) of 1.5. Furthermore, the design effect was reduced by the adoption of at least one-third of the eligible municipalities in each PSU. Therefore, it was defined by the selection of 16 municipalities, totaling 48 municipalities in the region. As a result, the sample size was estimated at 389 children per Cad. Saúde Pública 2020; 36(12):e00166619 state, followed by the inclusion of 25% to cover possible losses, totaling 500 children, approximately 32 per municipality.
Children between 12 and 59 months old who were present in the basic health units (UBS) during the data collection days were selected to participate in the study. Exclusion criteria were children who performed blood transfusions and its derivatives; who were under immunosuppressive or corticotherapeutic therapy; who had chronic illnesses; who were HIV-infected or had any severe infectious condition; who had congenital malformation and/or who were hospitalized due to diarrhea in less than a month. Data collection was conducted by trained dieticians and took place from January to June 2015. It consisted of a face-to-face interview with the mother or legal guardian to apply a structured questionnaire in the UBS of each municipality. Then, blood samples from the children were collected and their weight and height, measured.
The questionnaire included individuals' demographic, socioeconomic, and environmental information. The blood was collected by peripheral venipuncture and a drop of the collected blood was placed on a flat surface (glass slide). Then, the tip of a microcuvette touched the sample drop to capture the necessary amount of blood, followed by the immediate reading of the hemoglobin levels in a portable hemoglobinometer (HemoCue, HemoCue Limited, Sheffield, U.K.). The test was performed in duplicate for hemoglobin values less than 11g/dL 17 , adopting the mean values. The weight and height data were collected in duplicate, following standardized procedures 29 . The indexes height-for-age (HAZ), weight-for-age (WAZ), and body mass index-for-age (BMI) were then classified according to the cut-off points of the z-score 30 . The children's hemoglobin level (g/dL) was the dependent variable in this study. The selected independent variables were defined based on a theoretical model that considered reference studies in the area. These variables were classified in two levels: individual, which covers the children's variables; and municipal, which contains variables related to the municipalities.
The children's variables with significant association in a previous bivariate analysis (p < 0.05) were included in this study: (1) children's age (< 24 months or ≥ 24 months); (2) maternal/legal guardian's race/color (white or non-white: black, brown, yellow, indigenous or other); (3) monthly household income per capita based on minimum wage (MW) in 2015 (BRL 788.00/USD 236.63) (no income or income up to ¼ of the MW, income between ¼ and ½ MW, income between ½ and 1 MW or income higher than 1 MW); (4) number of residents in the house (up to 3, 4 to 5, 6 or more); (5) mother/legal guardian's occupation (unemployed, employed with/without social security number or other (housewife, student or retired); (6) number of times the child went to the UBS in the year preceding the research (0 to 5 times, 6 to 12 times or more than 12 times); (7) enrollment at daycare center or kindergarten (yes or no); (8) Beneficiary of the Bolsa Família Program (Brazilian Income Transfer Program) (yes or no); (9) nutritional status according to the WAZ index (underweight -for both underweight and severely underweight children -, normal weight or overweight); (10) use of ferrous sulfate (never been taken, yes it was taken or yes still take it) and (11) beneficiary of the Paraná food fortification program called Leite das Crianças (Milk for Children) (yes, no or not applicable).
At the municipal level the variables were obtained or calculated using secondary and ecological data, available in national surveys and institutional websites, namely:  32 ; and (10) infant mortality rate in 2014 33 . All variables selected to compose the municipal level were categorized according to their quartiles, so that higher quartiles correspond to favorable life conditions or to better availability of public goods and services, except for the Gini coefficient and the percentage of the extremely poor population. Additionally, the infant mortality rate was not categorized by quartiles. It Cad. Saúde Pública 2020; 36(12):e00166619 was considered low when less than 5/1,000 live births, moderate when between 6-15/1,000 and high when more than 16/1,000 live births.
The characteristics of the children and the municipalities were described by absolute and relative frequencies and by measures of central tendency and dispersion. The multilevel analysis was based on the methodology of Generalized Additive Models for Location Scale and Shape (GAMLSS) 34 , considering the non-normal distribution of hemoglobin levels.
Among the available models for continuous random variables, without the inclusion of independent variables, we verified the model with the best adjustment of hemoglobin levels according to the Akaike information criterion (AIC). For this criterion, the lower the value produced by a model, the greater its ability to explain the data adequately with a smaller quantity of parameters 35 . Based on this result, the probabilistic model chosen was the Johnson's SU-distribution, defined by four parameters: mean (μ); standard deviation (σ); and two shape parameters (τ and ν) 36 . Since it is a four parameters model, Johnson's SU-distribution allows modeling data with different levels of skewness and kurtosis. In the present study, the blood hemoglobin levels presented a fat-tailed distribution (high kurtosis), which was not properly fitted by the normal or other models based on fewer parameters. Although we have selected a four parameters model, the covariates were used only to model the location parameter, μ. Some additional attempts were performed to assess the inclusion of covariates for the remaining parameters, but we did not verify any substantial improvement in the fitted model. Finally, the identity link function was applied for the location parameter, such that the covariate effects on the mean hemoglobin levels must be interpreted additively.
Afterwards, we selected the variables for the adjusted model following two steps: (1) considering the variables at the municipal level; and (2) considering the variables at the individual level, including those selected at the first step. These variables were only included in the modeling of the mean distribution, an important parameter for measuring hemoglobin level of the individuals. As for the variability between the hemoglobin levels of children from different municipalities, a random variable with normal distribution of mean zero and variance was incorporated in the analyses.
The association between the municipal variables and the hemoglobin level was initially evaluated without adjusting the effect of the remaining municipal variables and verified by the likelihood-ratio test. The variables with p < 0.20 were selected for the sequence of the analysis, with adjustment to the effect of the remaining variables at this level. After this adjustment, the variables considered to be significant were those with p < 0.05. Then, the variables at the individual level were added one by one to the model adjusted by the municipal variables. Those that produced p < 0.20 in the test of likelihoodratio were the ones which remained for the sequence of the analysis. Finally, we proceeded with a new selection in the model adjusted with all variables selected in both steps, based on the backward strategy. In this step, the variables with a smaller contribution to the adjustment, according AIC values, were removed from the adjusted model one by one. The identification of the final model occurred when the removal of any variable of the model would produce a higher AIC value. To calculate the mean hemoglobin level for each category of the variables that remained in the final model, the effect of the remaining variables was fixed to the mean.
Data were double-entered and validated on EpiData, version 3.2 (http://www.epidata.dk/) and evaluated statistically in the software R, version 3.4.2 (http://www.r-project.org) at the significance level of 5%. The package GAMLSS was applied in the choice of the probabilistic model and in the regression analysis.
The multi-center study was approved by the Ethics Committee in Research with Human Beings of the Federal University of Health Sciences of Porto Alegre (Protocol n. 722.702/2014) and all children's mothers or legal guardians signed an informed consent form.

Results
A total of 1,567 children were evaluated in the multi-center study. The dosage of hemoglobin levels was possible in 1,501 of them, which resulted in a sample loss of 4.2%. The characteristics of the children can be found in Table 1 and the characteristics of the municipalities in Table 2.   The mean hemoglobin level was 12.8g/dL (95%CI: 12.7-12.8g/dL). Figure 1 shows the variability of the hemoglobin levels among the municipalities, without and with the adjustment for the variables of the final model proposed. In both cases, a significant variability was observed in hemoglobin levels between the municipalities (p < 0.001), with a variance reduction of σ 2 m = 0.25 to σ 2 m = 0.19 in the adjusted model. Furthermore, the municipalities 3 and 14 stood out as they had the most negative effects on the average hemoglobin levels, while the municipalities 19 and 40 stood out for having the most positive effects.
As for the municipal variables included in the first model, the Municipal Human Development Index, monthly nominal income, adequate sanitation, urbanization rate, number of ACS and attendance at daycare centers remained at p < 0.20. After adjusting for other variables of this same level, only the nominal monthly income, urbanization rate, number of ACS and the attendance at daycare centers had a significant effect (p < 0.05) on the hemoglobin level, and, thus, remained in the model to adjust the variables of the individual level.
Subsequently, variables at the individual level that remained in the model were: children's age, maternal/legal guardian's race/color, monthly household income per capita, number of residents in the house, maternal/legal guardian's occupation, number of times the child went to the UBS in the year preceding the research, enrollment at daycare centers or kindergarten, beneficiary of the Brazilian Income Transfer Program, WAZ index and the use of ferrous sulfate. Finally, after adjustment of the model, the municipal factors that showed a significant effect (p < 0.05) on the mean hemoglobin level were urbanization rate and number of ACS; and the individual factors were children's age, enrollment at daycare centers, participation as a beneficiary of the Brazilian Income Transfer Program and WAZ index. Table 3 shows only the estimates of the variables that remained in the final model while their effect in the children's mean hemoglobin level can be visualized in Figure 2. Children from municipalities with higher urbanization rates had lower mean hemoglobin levels when compared to children from municipalities with lower urbanization rates (quartile 1), that is 0.18 (p = 0.016) lower than children of quartile 2, and 0.16 lower than quartiles 3 (p = 0.024) and 4 (p = 0.033). Concerning the ACS, the positive effect of this variable was significant (p = 0.003) only in the children's hemoglobin levels of quartile 2, who presented a mean level 0.22 higher than those of quartile 1. Furthermore, children older than 24 months had a mean hemoglobin level 0.37 (p < 0.001) higher than younger children. Regarding the access to social protection policies, children who were beneficiaries of the Brazilian Income Transfer Program had a mean hemoglobin level 0.23 (p < 0.001) lower than those without   the benefit. In contrast, children enrolled at daycare centers had a mean level 0.21 (p = 0.001) higher than those not enrolled in such institutions. As for nutritional status, children with normal weight or overweight had a mean hemoglobin level 0.60 (p = 0.003) and 0.74 (p = 0.001) higher than children with underweight, respectively. The other parameters of the model were estimated by σ = 1.04 and ν = -0.48, revealing slight asymmetry of the data to the left, and by τ = 1.65 3 14 1 11 41 5 28 13 4 6 25 12 20 9 30 37 17 33 21 27 47 16 39 43 18 29 8 10 22 35 36 2 7 45 44 48 24 38 32 15 34 42 46 23 26 31 40     Estimates of hemoglobin levels (g/dL) of children

Discussion
The children's mean hemoglobin level in this study (12.8g/dL; 95%CI: 12.7-12.8g/dL) was higher than the internationally adopted cut-off points for the diagnosis of anemia in children (6-60 months). It was also higher than the international estimates (11.1g/dL; 95%CI: 11.0-11.3g/dL) 1 and the national representative values (12.1g/dL ± 1.5g/dL) for children (6-59 months) 2 . In addition, this value was higher than evidenced by other studies with children from different Brazilian epidemiological scenarios, which varied between 11.1g/dL and 12.5g/dL 22,37,38 . However, these comparisons should be interpreted with caution, due to the different methodologies used by other studies, especially when measuring the hemoglobin levels.
Even though the Southern region of Brazil has one of the best indicators of social, economic and educational development in the country 27 , it is still marked by inequalities that influence the health and nutrition of the most vulnerable population. In fact, significant variability of hemoglobin levels among the municipalities was observed in this study. The combined effect of the explanatory variables contributed to the mean hemoglobin levels up to 0.55 lower and 0.48 higher than the estimated reference mean. This variability of hemoglobin levels was attributable to variables related to access to public goods and services available to the population at the municipal level, such as urbanization and the number of ACS. Likewise, variables of epidemiological importance at the individual level contributed to the variability of hemoglobin levels, such as the child's age, enrollment at daycare centers or kindergarten, being a beneficiary in the Brazilian Income Transfer Program and the nutritional status based on the WAZ index.
Residing in municipalities with a higher urbanization rate had a negative influence on children's hemoglobin levels. This could be explained by the fact that urbanized populations often depend directly on income to obtain food and meet their nutritional requirements; this population also usually makes poor food choices 20,39 . This scenario was evidenced by a study 40 that observed a direct relationship between income and the possibility of choosing food, especially among residents from the urban area. According to its results, having a higher income did not guarantee dietary quality, as a "contradictory" dietary pattern was observed. This pattern was composed of diversified and nutrientrich food, but also industrialized and ready-to-eat meals, which could compromise the fulfillment of nutritional needs, such as iron intake. In addition, an urbanized population also require greater investments in basic sanitation and health 29 . As shown in the temporal analysis 12 , the Brazilian population residing in urban areas is more exposed to factors that contribute to the intergenerational cycle of diminished opportunities, such as unemployment, social inequalities and inadequate access to basic services. These factors can expose these families to nutritional deficiencies, such as lower hemoglobin levels.
Children from municipalities with fewer ACS had lower mean hemoglobin levels when compared to children from municipalities with a higher number of these professionals, especially in the municipalities of quartile 2. The impact of the ACS on the basic health system is already wellknown 41,42,43 ; but, to our knowledge, this is the first study which shows the possible effect of these workers on the hemoglobin level of children. Therefore, it is suggested that in some municipalities the greater number of ACS may have contributed to improve the health care quality in UBS, reflecting in higher hemoglobin levels. However, it is important to consider that the maximum number of ACS in each municipality is established by a law, i.e., 1 agent per 400 inhabitants 44 .
Age was also an important factor to explain the variability in the hemoglobin levels. Children under 24 months of age had the lowest concentrations of hemoglobin as observed in previous studies conducted in other Brazilian locations 45,46 . The age group less than 24 months is considered the most biologically vulnerable to nutritional and infectious diseases. These diseases occur mainly due to the imbalance between nutrient intake to meet nutritional requirements and the little-developed functional immunity 44,47,48 . Therefore, strategies aimed at controlling and preventing nutritional diversions and infections become essential even in places where access to quality food is adequate and health services are considered efficient 18,48 .
Although more than half of the children evaluated were not enrolled at daycare centers, higher mean hemoglobin levels were observed among those that were. According to the literature, this positive effect may be associated with a significant improvement in the consumption of healthy food (rich Cad. Saúde Pública 2020; 36 (12):e00166619 in nutrients for hematopoiesis) at daycare centers 49,50 . On the other hand, a study conducted by Zuffo et al. 45 showed that these institutions have difficulty to meet the children's iron nutritional requirements. Recognizing that early childhood education is the first step towards overcoming poverty 51 , our results reinforce the importance of providing high-quality meals by educational institutions.
Unlike studies carried out in different locations 52 , this study did not verify the positive effect of the Brazilian Income Transfer Program on the children's hemoglobin levels. In fact, the establishment of a cause-and-effect relationship between the program and hemoglobin levels could not be explained by this cross-sectional study. A possible explanation for this scenario would be the timespan of participation in the Brazilian Income Transfer Program, sometimes too short to observe the effects on children's health 53 . Furthermore, the accomplishment of the conditions of the program 54 or the quality of the services provided 19,53 might not have been sufficient for the prevention and early identification of nutritional deficiencies by biochemical indicators. In addition, families that participate in this program show lower income and greater vulnerability to food deprivation 55 and to nutritional deficiencies 56 . Hence, although the studies show an improvement in the anthropometric indicators of children that are beneficiaries of the program 13,57 , their families still live in food and nutritional insecurity scenarios. This could compromise the access to quality food for preventing nutritional deficiencies 58 .
According to the WAZ index, a lower mean hemoglobin level was observed in children diagnosed as underweight. This result supports the premise that in the presence of a nutritional deficit scenario, hemoglobin levels are compromised as an adaptive response to food deprivation 59 ; however, the cross-sectional design of the study prevents any causal inference. Evidence also suggests that overweight children are subject to decreased hemoglobin levels due to their poor-quality diets 60 and the increased blood volume, iron nutritional requirements 61 , pro-inflammatory cytokines and hepcidin 62 . The association between excess body weight and low hemoglobin levels, however, was not observed in this study.
The major strength of our study refers to the use of multilevel analysis to understand the simultaneous influence of the children and municipalities variables on hemoglobin levels. However, our study has limitations that should be considered in the interpretation of our findings and in future research. First, the municipal variables selected for this study constitute a proxy for their development, income distribution, access and quality of public services, and for the possible results of the actions developed under the Brazil Without Poverty Plan. Second, some of these variables represent the situation of the municipalities prior to the implementation of the Brazil Without Poverty Plan, considering the census intervals for the collection of information about the Brazilian population, and therefore might not reflect the scenario to which the children were exposed at the time of the multicenter study. Third, the presented data may not reflect the reality in other municipalities of the same region or other Brazilian regions, even though they were part of the same public policy. Fourth, the sample was not designed to be representative of each municipality individually, but rather of the set of the 16 municipalities in each state. Fifth, we did not investigate the effect of using other supplements on hemoglobin levels in this study. Last, although this study does not allow inference of causality, the findings are a starting point for prospective research that seeks to establish clear and precise relationships between the factors conditioning different levels of determination of nutritional deficiencies in children.
Through the Brazil Without Poverty Plan, 22 million Brazilians have overcome poverty from the income perspective 14 . Considering this and the hemoglobin levels identified by this study, we suggest the possibility of a positive impact from the strategies aimed at the development of human potential in the municipalities included in this public policy, although a proper evaluation of the plan would be required to confirm this.

Conclusions
The variability of children's hemoglobin levels was explained by individual and municipal predictors. These findings reinforce the importance of access to basic public goods and services and the development of policies aimed at reducing social inequalities. The permanence of individuals factors after adjustment for the municipal variables indicates the need for improvement, monitoring and continuity of public policies for the control of nutrient deficiencies in childhood.