Agro-ecological zone and farm diversity are factors associated with haemoglobin and anaemia among rural school-aged children and adolescents in Ghana.

Understanding contextual risk factors for haemoglobin (Hb) status and anaemia of rural school-aged children (SAC) and adolescents is critical in developing appropriate interventions to prevent anaemia. We analysed secondary data from the baseline of an impact evaluation of the Ghana School Feeding Programme to determine the severity of anaemia and contextual factors associated with anaemia and Hb status among rural SAC (6-9 years; n = 323) and adolescents (10-17 years; n = 319) in Ghana. We used regression models with variable selection based on backward elimination in our analyses. The mean Hb was 113.8 ± 13.1 g/L, and the overall prevalence of anaemia was 52.3%, being 55.1% and 49.5% among SAC and adolescents, respectively. We identified child's age (β = 2.21, P < 0.001); farm diversity score (β = 0.59, P = 0.036); and agro-ecological zone (P trend <0.001) as the main predictors of Hb of SAC. Household asset index (P trend = 0.042) and agro-ecological zone (P trend <0.001) were predictors of Hb in adolescents. Agro-ecological zone and age were predictors of anaemia, but the effect of age was only significant for girls and not boys (prevalence odds ratio [POR] = 1.35, 95% CI [1.04, 1.76] vs. POR = 1.14, 95% CI [0.88, 1.46]). SAC in households with maize stock were less likely to be anaemic (POR = 0.55, 95% CI [0.32, 0.97]). Household dietary diversity score (β = 0.59, P = 0.033) was associated with Hb status for the full sample only. Anaemia is a severe public health problem among SAC and adolescents in rural Ghana irrespective of sex. Farm diversity score, availability of maize stock in the household, household asset index, and agro-ecological zone were the main predictors of Hb and anaemia among the rural SAC and adolescents.


| INTRODUCTION
Anaemia is a critical public health condition globally, particularly in low-and middle-income countries where its prevalence tends to be 3 to 4 times higher than in advanced economies (McLean, Egli, De Benoist, Wojdyla, & Cogswell, 2007;WHO, 2008). Besides pregnant women, children (both preschool and school age) are the most affected group with iron-deficiency anaemia because of the rapid growth and the general cognitive development (Kassebaum et al., 2015). Studies have shown that adolescents are also at an increased risk of developing anaemia due to increased iron demand during puberty, menstrual losses, limited dietary iron intake, and poor dietary habits (Chaparro & Lutter, 2008;WHO/FAO, 2004). Adolescent girls are particularly at high risk of developing iron deficiency and/or anaemia, especially among those who experience heavy blood losses during menstruation and corresponding decreases in ferritin levels (Black et al., 2008;Chaparro & Lutter, 2008).
It is estimated that about half of school-aged children (SAC; 6-9 years) and pregnant women and close to one third of nonpregnant women of reproductive age suffer from anaemia globally (McLean et al., 2007). In 21 countries assessed by UNICEF, more than one third of adolescent girls were anaemic (UNICEF, 2012). In Ghana, about 73.6% and 44% of rural young children (<5 years) and women are anaemic, respectively (Ghana Statistical Service/Ghana Health Service, 2015). About two thirds of Ghanaian adolescent girls and SAC are reportedly anaemic (Abizari, Azupogo, & Brouwer, 2017;UNICEF, 2012), suggesting anaemia is a severe public health problem in Ghana requiring urgent attention and context-appropriate policies.
The consequences of iron deficiency and anaemia have been well documented; these include impairment of physical and cognitive development of infants and young children (Batra & Sood, 2005;Chang, Zeng, Brouwer, Kok, & Yan, 2013;Jáuregui-Lobera, 2014), higher risk of morbidity and mortality for young children and pregnant women (Sanghvi, Ross, & Heymann, 2007;Semba & Bloem, 2008), and increased risk of low birth weight even with moderate preconception anaemia on the part of women in fertile age (Bhutta et al., 2013). The long-term effect of anaemia is reduced cognition in the early years, which is associated with lower productivity later in life (Haas & Brownlie, 2001;WHO, 2008). In fact, the negative consequences of iron deficiency or anaemia on cognitive performance may not be limited only to infants and young children but may continue into adolescence. In a randomized controlled trial with iron supplementation, Bruner, Joffe, Duggan, Casella, and Brandt (1996) showed that iron-deficient American adolescent high school girls receiving a 1,300-mg daily dose of ferrous sulphate for 8 weeks performed better on a test of verbal learning and memory than girls with a similar iron status receiving a placebo.
The primary cause of anaemia is assumed to be iron deficiency; nevertheless, this condition is seldom present in isolation (WHO, 2008). This is because anaemia is only partly caused by dietary insufficiency of nutrients and it coexists with infectious diseases such as malaria and parasitic infections, haemoglobinopathies as well as socio-economic and environmental factors such as geographical location, the source of drinking water and sewage system; these determinants are also often context specific (Darnton-Hill & Mkparu, 2015;Syed et al., 2016;Tesfaye, Yemane, Adisu, Asres, & Gedefaw, 2015;WHO, 2008).
Given the context specificity of the root causes of anaemia, understanding contextual risk factors is critical to guide the implementation of interventions to prevent or reduce anaemia. Nevertheless, there is generally a dearth of data on the major predictors of anaemia among SAC and adolescents as existing studies focus primarily on preschool children (<5 years) or women of reproductive age. Hence, this study aims to fill this critical gap by secondary analyses of data from an impact evaluation to determine the prevalence, severity, and contextual factors associated with anaemia and haemoglobin (Hb) status among rural Ghanaian SAC and adolescents and provide evidence for policy formulation and programme planning. Details of the impact evaluation study design and of the data collected have been reported elsewhere (Gelli et al., 2015).

| Study population and sample size
The study population in the impact evaluation was selected through cluster random sampling from the Greater Accra, Ashanti, Brong Ahafo, Central, Eastern, Northern, and Upper East Regions of Ghana.
In the impact evaluation, a random subsample of 717 children aged 4-17 years had measurements of Hb taken. Seventy-five (75) children not meeting the inclusion criteria (age ≥ 6 years and residing in a rural

Key messages
• Anaemia is a severe public health problem among rural SAC and adolescents in Ghana irrespective of sex.
• Farm diversity, availability of maize stock in the household, HAI, and agro-ecological zone were the main contextual factors associated with Hb and anaemia.
• A unit increase in age was associated with reduced odds of anaemia for SAC and higher odds of anaemia for, particularly adolescent girls.
• Given the severity of anaemia in other agro-ecological zones compared with the forest zone, iron and folate supplementation may be prioritized in these zones for both SAC and adolescents alongside both boys and girls. community) were excluded, which led to the final sample of 642 children aged 6-17 years for the analyses.

| Hb and anaemia
Hb was measured with HemoCue Hb 201 + using blood samples from a finger prick test by a trained phlebotomist. Anaemia was defined with the WHO criteria (WHO, 2011): Hb < 115 g/L for 6-to 11-year-old children; Hb < 120 g/L for 12-to 14-year-old children; Hb < 130 g/ L for males ≥15 years; and Hb < 120 g/L for nonpregnant females ≥15 years. We also categorized anaemia as severe (Hb < 80 g/L), moderate (Hb 80-109 g/L), and mild (Hb 110-119 g/L) using the WHO criteria (WHO, 2011).

| Helminths infestation
Urine and stool samples were collected separately from a subsample of the children (N = 311) in order to screen them for ascaris, hookworm, trichuriasis, and schistosomiasis. Analyses were carried out by microscopy and/or RDT at the General Laboratory of the Parasitology Department of NMIMR. We defined helminths infestation as the presence of any of the above parasites, and a dummy variable was created.

| Household dietary diversity
A modified household dietary diversity score (HDDS; Swindale & Bilinsky, 2006) was calculated for each household using data on consumption of 68 food items grouped into 12 food groups, over the previous 7 days. For each food group, a score of 1 was given if a household consumed at least one food item from the food group in the past week, else 0. The food group scores were summed into the modified HDDS, ranging from 0 to 12. Food groups included cereals, roots and tubers, vegetables, fruits, meat, eggs, fish and seafood, pulses and nuts, milk and milk products, oils and fats, sugar, and condiments. Furthermore, a household food variety score was constructed similar to the food variety score (Savy, Martin-Prevel, Sawadogo, Kameli, & Delpeuch, 2005), which consisted of the sum of all the unique food items consumed by a household from the 7-day recall. Similar to the HDDS, the household food variety score was a continuous score ranging 0-68. Lastly, we created a score for household animal food consumption (HAFC) consisting of a sum of all the unique animal foods consumed by the household from the 7-day food recall; the HAFC ranged 0-9. These scores did not consider minimum quantities of intakes.

| Other covariates
The household questionnaire included modules on household demographics, education, economic activities, and farm income. The data included information on household durables and agriculture assets.
We constructed household and agricultural assets indices through principal component analysis (Filmer & Pritchett, 2001) and then divided them into terciles. These indices captured household wealth in relation to the ownership of durable and agricultural assets. Furthermore, we included in our statistical models a continuous variable for the proportion of food consumed by the household from own production in the past month.
We included in the initial models the following continuous variables: the birth order of the child; the number of days the child bought food from school; the age of the household head; the household's dependency ratio; the number of months the household consumed food from own production; and parental years of schooling. Whereas Children were categorized as school age (6-9 years) and adolescents (10-17 years) based on their age (UNICEF, 2011). Lastly, school grade was categorized as lower primary and upper primary/junior high school.

| Statistical analysis
We used SAS 9.3 (SAS Institute Inc., Cary NC.) in all our statistical analyses and a two-tailed P value ≤0.05 was considered statistically significant. Population characteristics were presented as means ± SD for continuous variables and percentages for categorical variables.
We used one-way analysis of variance to assess the difference in means between groups for continuous outcome variables whereas Pearson's chi-square test was employed for categorical variables.
Based on the literature, we identified an a priori set of potential predictors of Hb and anaemia. These included child sex and age, helminths infestation, child health status (including malaria), dietary intake, household size, educational status of household head and of the mother, the occupation of the household head and of the mother, We included variables with P values ≤0.25 from bivariate analyses along with those thought to be important a priori (sex, parental education, and occupation) in backward stepwise multiple regression models using the GLMSELECT PROCEDURE (Mcgahan & Chong, 2009) and PROC LOGISTIC PROCEDURE (SAS Institute Inc, 2008) in SAS to, respectively, determine the regression coefficients (β) and the prevalence odds ratios (PORs) of the significant predictors of Hb status and anaemia in the sample. The criteria for variable selection in the GLMSELECT PROCEDURE was based on the significance level and the Predicted Residual Sum of Squares Statistic whereas the significance level and log-likelihood ratio test were used for the logistic regression. We entered interaction terms in order to explore potential non-linearities, but none of these interactions was statistically significant.
The analysis was also stratified for child's age category (SAC and adolescents). Lastly, as a robustness check, we repeated the analysis with multilevel regression with a random intercept and clustering at the agro-ecological zone using the PROC MIXED PROCEDURE in SAS (Littell, Milliken, Stroup, Wolfinger, & Schabenberger, 2006). Table 1 presents the sample characteristics. About 50.3% of the children were of school age and the majority attended lower primary school. About half were females. Children's ethnicity included Akan (38.9%), Gurma (25.9%), Mole-Dagbani (17.6%), and other (17.6%).

| Sample characteristics
About 11.5% of the children reported being sick 7 days preceding the survey, and none was found to have helminths infestation. Generally, the dependency ratio was high. Furthermore, the mean paternal and maternal years of schooling were 5.6 ± 6.5 and 3.1 ± 5.6 years, respectively. About 72% of the children belonged to male-headed households. Farming was the main occupation of both parents, but SAC children had more farmer fathers compared with adolescents. A greater proportion of SAC were in lower primary school compared with adolescents. In addition, households of SAC had more land and a higher farm diversity compared with adolescent households. Also, the proportion of food consumed from own production was lower for households of SAC, but their households consumed food from their own production for a longer duration compared with adolescent households. Lastly, more SAC than adolescents were from the Northern Savannah agro-ecological zone.
The mean Hb concentration of the children in the study was 113.8 ± 13.1 g/L and was significantly lower for SAC compared with adolescents (SAC: 111.4 ± 12.8 g/L; adolescents: 116.2 ± 12.9 g/L, P < 0.001; Table 1). The overall prevalence of anaemia was 52.3% and did not differ between SAC and adolescents (SAC: 55.1%; adolescents: 49.5%, P = 0.157). However, more SAC were moderately anaemic compared with adolescents and although about 1.2% of SAC were severely anaemic, none of the adolescents was severely anaemic (Table 1).

| Results of the bivariate analyses
We found a significant variation (P < 0.001) in the mean Hb by agroecological zone, ranging from 109.1 g/L (95% CI, [105.2, 113.0]) in the Coastal Savannah to 118.6 g/L (95% CI [117.1, 120.2]) in the forest zone (Table 2). Likewise, the mean Hb differed significantly between SAC and adolescents in each agro-ecological zone except for the Northern Savannah zone, but the prevalence of anaemia did not differ significantly between SAC and adolescents in each agroecological zone. The overall prevalence rate of anaemia varied from 36.4% in the forest zone to 63.8% in the Coastal Savannah with the odds of anaemia been significantly higher for all agro-ecological zones compared with the forest zone (Table 2).
Additionally, the results of the bivariate linear regression (Table   S1) showed that child's age, school grade, ethnicity (Gurma vs. Akan), the number of days child bought food from school, HDDS, HAFC, and agro-ecological zone were significant correlates of the children's Hb status.
Compared with the forest zone, children residing in other agroecological zones consistently had a higher odds of anaemia both at the population level and by age category (P trend < 0.001 for all models). Furthermore, HDDS (POR = 0.91, 95% CI [0.83, 1.00]) and availability of maize stock in the household (POR = 0.70, 95% CI [0.47, 1.04]) were associated with lower odds of anaemia for the full sample. Additionally, availability of maize stock in the household was significantly associated with a 45% lower odds of anaemia among SAC. Although a unit increase in HDDS was associated with a lower odds of anaemia among SAC, the effect was not statistically significant. Moreover, the middle tercile of the agriculture asset index was associated with 52% higher odds of anaemia whereas the upper tercile was associated with 26% lower odds of anaemia among SAC, but none was statistically significant. Although a unit increase in age was associ-  Paternal years of schooling 5.6 ± 6.5 5.4 ± 6.5 5.7 ± 6.6 Maternal years of schooling 3.1 ± 5.6 2.8 ± 5.4 3.5 ± 5.8 Household receipt of remittance in the past 1 year, % 29.6 26.4 32.9 Occupation of father (n = 429), % In Table 3, the proportion of variance explained (adjusted R 2 ) in the selected models for Hb was 14% for SAC and 13% for adolescents. Similarly, for anaemia (   prevalence rate (64%) than our study among 5-to 12-year-old rural SAC (Abizari et al., 2017). Although the apparent differences with our findings may be related to differences in sample size and socio- Age was not significantly associated with the Hb status of adolescents in the multiple regression; however, we found that after peaking at 11 years, the Hb of females declined ( Figure S1a), which corroborates the findings of Sabale, Kowli, and Chowdary (2013) who found that Hb decreases significantly as age increasing from 9-15 years among Mumbai school girls aged 9-19 years. The mean age of the adolescent girls (11.2 ± 1.3 years) in the present study corresponded with the mean age of the onset of menarche in Ghana (Gumanga & Kwame-Aryee, 2012;Richmond, Anthony, & Martin, 2011) and may partly explain our finding. Unfortunately, we could not further test this hypothesis, as the baseline data did not include information on age at menarche.
Furthermore, a unit increase in age was significantly associated with 24% higher odds of anaemia for adolescents in the multiple logistic regression. However, the inclusion of an interaction term for age and sex showed that the odds of anaemia for a unit increase in age was only significantly higher for adolescent girls and not boys. This is not surprising considering the growth spurt in early adolescence and the increased iron requirements for girls at menarche (WHO/ FAO, 2004). Indeed, adolescent girls are particularly at high risk of developing iron deficiency and/or anaemia, especially among those who experience heavy blood losses during menstruation and corresponding decreases in ferritin levels (Black et al., 2008;Chaparro & Lutter, 2008). The finding emphasizes the need for intervention programmes to curtail anaemia in early adolescence to compensate for the additional nutrient requirements for growth and puberty as well as the extra losses due to menstruation.
In contrast to many other studies that have reported a higher prevalence and risk of anaemia among females compared with males (Egbi et al., 2014;Salama & Labib, 2016;Tesfaye et al., 2015), though not significant, our findings pointed to a higher prevalence of anaemia among adolescent males compared with females and a similar prevalence for both school-age males and females (Figure 1). The finding is nevertheless similar to the result of Hall et al. (2009), who found a higher prevalence of anaemia for 12-to 14-year-old boys compared with girls in rural Volta region of Ghana. Although the mean Hb of adolescent males was slightly lower than that of adolescent females in the present study (boys: 115.3 g/L, girls: 117.0 g/L, P = 0.978), the WHO criteria classifies adolescent males aged ≥15 years as anaemic at a higher threshold compared with adolescent females with similar age (130 vs. 120 g/L). Thus, more adolescent males than females were classified as anaemic and may partly account for the finding.
Nonetheless, the implication of this finding is that adolescent boys should also be targeted for anaemia control programmes besides adolescent girls, the usual priority group.
In agreement with evidence that high socio-economic status is protective of anaemia (Abdel-Rasoul, El Bahnasy, El Shazly, Gabr, & Abdel-Aaty, 2015;Ngesa & Mwambi, 2014), belonging to the middle or upper terciles of the HAI was associated with a significant increase in Hb compared with those in the lower tercile of the HAI for adolescents. In addition, a unit increase in HDDS was associated with a 0.59-g/L significant increase in the Hb level of the children, but the association did not hold in the stratified analysis by age category, which could be because the subsamples were underpowered to detect any statistical certainty. Farm diversity seemed to have a weak association with Hb, but the association was clearer for SAC for whom a unit increase in farm diversity was significantly associated with a 0.59-g/L increase in Hb. We found no interaction between farm diversity and HDDS in the present study; nevertheless, studies have shown that farm diversity improves household dietary diversity (Jones et al., 2014;Sibhatu et al., 2015). The HDDS is a measure of household food security rather than dietary quality (Ruel, 2003), but it has been shown that household food security influences the quality of individual diets and micronutrient adequacy (Leung, Epel, Ritchie, Crawford, & Laraia, 2014;Tarasuk & Beaton, 1999 Several studies have reported that parental education has a protective effect against anaemia in resource-poor settings (Assefa et al., 2014;Ngesa & Mwambi, 2014;Tesfaye et al., 2015); therefore, it is surprising that this association does not hold in our study with education as either a categorical variable or a continuous variable. This may be due to the relative lack of variation in educational status in our sample, with most of the parents having completed only a few years of schooling (mothers: 3.1 ± 5.6 years, fathers: 5.6 ± 6.5 years). Likewise, we did not find any association between maternal occupation and paternal occupation with Hb status or anaemia in contrast to several other studies (Assefa et al., 2014;Mesfin et al., 2015;Tesfaye et al., 2015).
Even though helminths infestation is a key significant determinant of anaemia (Leenstra et al., 2004;Salama & Labib, 2016), remarkably, we did not find any helminths infestation among a random subsample of the children (n = 311) that was partly attributable to the mass deworming of school children, which is being implemented in Ghana by the Ghana Health Service since 2009. Ghana started its national deworming campaign for schistosomiasis and soil-transmitted helminths in that year. Indeed, schools can constitute effective platforms for reaching this population through integrated deworming, water, hygiene, and sanitation programmes (Bundy, 2011).

| Strengths and limitations of the study
The results were generally robust when modelling with backward regression, multilevel regression, and Ordinary Least Squares (OLS).
Nonetheless, some limitations inherent in the present study should be considered in the interpretation of our findings. Firstly, because these are observational, cross-sectional data, these cannot be interpreted as causal findings. We, therefore, limit our interpretation to describing associations. Even so, we thoroughly modelled several potential explanatory variables including subject and household demographics, education, livelihood, wealth, farming and production orientation, and quality of household dietary intake, as well as used different methods to test the robustness of these associations.
Furthermore, intakes of vitamin A-rich and iron-rich foods are well-recognized predictors of anaemia (Zimmermann et al., 2005), yet it was not possible to evaluate the contribution of dietary intake to the improvement of Hb in the present study as we lacked data on intakes of these foods specifically. Malaria is another factor that may contribute to anaemia (Leenstra et al., 2004;WHO, 2001), and a high prevalence of malaria among Ghanaian children has previously been reported (Abizari et al., 2017;Otupiri et al., 2012). Although we did not include malaria incidence in our statistical models, child sickness in the last 7 days preceding the survey was captured, which we presume, may have sufficiently accounted for any malaria incidence.
The selected final models accounted for smaller proportions of the variance for Hb and anaemia; nevertheless, this does not necessarily weaken the effect and the importance of the significant predictors to our outcomes. The implication of the relatively small percentage of variance explained by our models may be that, there are other unaccounted predictors contributing to the unexplained variances, which we have tried discussing. Overall, the F value (for linear regression) and log-likelihood ratio (logistic regression) for all the models were statistically significant with P values <0.001.
Our study population is rural, which limits the generalization of our findings to all SAC and adolescents in Ghana. Moreover, only   FIGURE 1 The prevalence of anaemia among the rural Ghanaian school children stratified by age category and sex. SAC: school-aged children SAC and adolescents enrolled in school were studied, but studies have shown that nonenrolled children may be more anaemic than those in school because enrolled children are in a better position to understand health and nutritional risks as this could be taught in school or sensitization programmes may be carried out in their schools. This could have underestimated the true prevalence of anaemia in our study, but considering that 36% of children in rural areas never enter school in Ghana (UNICEF-Ghana, 2015), the study population may well represent all rural children.