Age‐specific differences in the magnitude of malaria‐related anemia during low and high malaria seasons in rural Zambian children

Abstract Background Malaria causes anemia by destruction of red blood cells and inhibition of erythropoiesis. Objective We assessed whether the magnitude of the malaria‐specific effect on anemia differs by age, during low and high malaria seasons. Method In rural Zambian children participating in a pro‐vitamin A efficacy trial, we estimated differences in the prevalence of anemia (defined as hemoglobin < 110 g/L for children < 60 months. and < 115 g/L in older children) by malaria status and assessed malaria‐age interactions. Regression models (with anemia as the outcome) were used to model malaria‐age interaction in both the low and high malaria seasons, controlling for potential confounders. Results Average age was 68 months at baseline (n = 820 children). In the low malaria season, anemia prevalence was 29% in malaria‐negative children and 54% in malaria‐positive children (p < 0.001), with no malaria‐age interactions (p = 0.44). In the high malaria season, anemia prevalence was 41% in malaria‐negative children and 54% in malaria‐positive children (p < 0.001), with significant malaria‐age interactions (p = 0.02 for anemia). Age‐stratified prevalence of anemia in malaria positive versus negative children was 67.0% versus 37.1% (in children < 60 months); 57.0% versus 37.2% (in 60–69 months.); 46.8% versus 37.2% (in 70–79 months.); 37.0% versus 37.3% (in 80–89 months) and 28.0% versus 37.4% (in 90+ months). Conclusions Malarial anemia is most severe in younger children, especially when transmission is intense. Anemia control programs must prioritize this vulnerable group.

of anemia are often multifactorial and are interrelated in a complex way [9]. Historically, global anemia control efforts have predominantly focused on iron deficiency, guided by the premise that iron deficiency (ID) causes almost half of all anemia cases [9]. In recent years, attention to non-ID related anemia is growing, in part because of accruing evidence suggesting that only ∼30% of anemia is amenable to iron supplementation [10]. Of the other causes of anemia, malaria-induced anemia is of particular relevance to sub-Saharan Africa, where malaria is endemic. In this region, malarial anemia is a major cause of sickness and deaths in children [11,12]. Hence, characterization of the epidemiology of malaria-related anemia is critical in designing optimal interventions for anemia control.
Malaria-related anemia is characterized by destruction of parasitized and especially non-parasitized erythrocytes, and impaired erythrocyte production during malaria episodes [13,14]. The pathogenesis is complex, and the mechanism of malarial anemia is not fully understood. Because clinical immunity to malaria takes several years of repeated exposure to develop, malaria is typically most severe in young children, whose immune systems are immature [15][16][17][18][19]. Unfortunately, evidence is lacking on how this age-specific variation in malaria severity translates into anemia. In this study, we assessed whether the magnitude of the malaria-specific effect on anemia differs by age, during low and high malaria seasons, among children in rural Zambia, a malaria endemic region. Specific objectives include a) to compare hemoglobin values (and prevalence of anemia) in children by malaria status in both the low and high malaria seasons and b) to determine if age is an effect modifier in the association between malaria and anemia in children.

Study design and participants
This study included 4-to 8-year-old children from Mkushi District in Central Province, Zambia, an area with a high burden of malaria and anemia [20][21][22]. Data and biospecimens were collected as part of a cluster-randomized controlled trial, designed to assess the effi-  [24]. Thick and thin venous blood films were prepared using 2 and 10 μl of blood, respectively, and venous hemoglobin was assessed using a HemoCue photometer. Blood containing tubes were kept in cooler boxes containing ice packs, allowed to clot, and then transported to field laboratories for separation into serum by centrifugation. Serum was aliquoted into prelabeled cryovials, transported in liquid nitrogen, and stored at -80 • C until analyzed. All children were given an insecticide-treated bed net at baseline.

Laboratory analyses
Thick and thin malaria microscopic slides were read at the TDRC labo- All baseline assessments were repeated at the endline, 6 months later.

Definitions
Anemia was defined as hemoglobin < 110 g/L for children < 60 months and < 115 g/L for children ≥ 60 months [25]. Malaria was defined by either positive RDT, positive microscopy, or both. Malarial anemia was defined as positive test for malaria and concurrent anemia. Acute inflammation was defined as CRP > 5 mg/L, and chronic inflammation was defined as AGP > 1 g/L.

Sample size and power considerations
Because the study intervention had no impact on hemoglobin concentration, we pooled data from all children who had a hemoglobin measurement and malaria microscopy at both time points, regardless of treatment allocation (n = 820). With a sample size of 820 (274 anemic cases and 546 non-anemic cases) and 5% type-1 error rate, we had 80% power to detect a difference of 10% in anemia prevalence comparing malaria-positive children and malaria-negative children.

Statistical analyses
Indicators of nutritional status, socio-demographic status, inflammations, anemia, and malaria at baseline and endline are summarized as means or proportions (Tables 1 and 2 In the models, malaria was defined on the basis of microscopy alone. There were fewer numbers of children with malaria data at baseline (n = 801) compared to endline (820).
were constructed to visualize how the association between malaria and anemia (or hemoglobin) changed by age. Ferritin, retinol, and sTfr were log-normalized.
For all other tests, statistical significance was set at p = 0.05. All data analyses were conducted with STATA 13 software (StataCorp). In adjusted models, iron status (defined by ferritin and sTfR), vitamin A status (defined by serum retinol), and inflammation (defined by CRP and AGP) were included as covariates.

RESULTS
From 1024 children enrolled in the trial, we included a subset (n = 820) who had complete baseline and endline data for hemoglobin, ferritin, sTfR, malaria, CRP, and AGP at both baseline and endline. The children in this subsample were statistically similar to those in the overall trial with respect to age and gender. The mean (± SD) age of the study population was ∼68 months (±15) with ∼35% below the age of 5 years ( Table 1). The prevalence of reported fever, cough, and diarrhea at baseline (low malaria season) were 29%, 58%, and 6%, respectively.

TA B L E 3
Association between malaria and hemoglobin concentration (g/L) and interactive effects of age during low and high malaria season among Zambian children <0.01 -*Differences in mean hemoglobin concentration were tested using linear regression. In unadjusted models, only malaria and age were included as covariates.
In adjusted models, concentration of serum ferritin, soluble transferrin receptor, retinol, and inflammation were included as covariates, in addition to malaria and age.
The prevalence of stunting was 29%. Underweight and stunting were defined as weight-for-age Z-score less than -2 and length-for-age Zscore less than -2 standard deviation of the WHO Growth Standards respectively (WHO, 2006). Table 2 shows the distribution of anemia, malaria, and inflammation in the low and high malaria transmission seasons. The prevalence of anemia increased from 33% in the low malaria season to 40% during the high malaria season. The prevalence of malaria more than doubled from 21% in the low malaria season to 51% in the high malaria season. Prevalence of acute and chronic inflammation was 17% and 44%, respectively, in the low malaria season, increasing to 32% and 74%, respectively, in the high malaria season.
In the low malaria season, there was no statistically significant difference in the hemoglobin concentrations among children with or without malaria after controlling for age, iron status, vitamin A status, and inflammation ( Table 3, p < 0.18). In addition, no effect modification by age was observed (p-interaction = 0.45). In the high malaria season, however, both hemoglobin and anemia differed significantly by malaria status, and in addition, a significant age interaction was observed. After adjusting for covariates, mean hemoglobin was 17 g/L lower in children diagnosed with malaria compared to children without malaria (p = 0.01). As shown in Table 4, this translated into a nine-fold increase in the risk of anemia (odds ratio [OR] = 9.32, p = 0.02). Furthermore, significant age interactions were observed in the high malaria season (p = 0.03 for hemoglobin, and p = 0.07 for anemia). Figures 1 and 2 show the age-interaction plots for hemoglobin and anemia, in both the low and high malaria seasons. In the youngest children (<60 months), anemia prevalence was about 30% higher in those diagnosed with malaria ( Figure 2). The magnitude of the difference in anemia prevalence declined gradually until the differences were no longer apparent in the older children. The age distribution of anemia by malaria status is presented in Table S5 (supplemental data). Among children with malaria, anemia prevalence ranged from as high as 67% in children < 5 years to as low as 28% in children > 8 years.

DISCUSSION
This study assessed the magnitude of malarial anemia and characterized potential age-specific trends in the burden of malarial anemia among 4-to 8-year-old Zambian children, across two malaria seasons. We observed that in both high and low malaria transmission seasons, anemia prevalence was significantly higher among children with malaria. More importantly, we observed that in the high malaria season (but not in the low malaria season), the malaria-specific effect on anemia was age-specific. In the youngest age group (<60 months), we observed ∼30 percentage point difference in anemia prevalence between children with and without malaria. However, this *Differences in anemia status were tested using logistic regression. In unadjusted models, only malaria and age were included as covariates. In adjusted models, concentrations of serum ferritin, transferrin receptor, retinol, and inflammation were included as continuous covariates, in addition to malaria and age.
P (age interaction)=0.52 The observed high risk of malarial anemia among younger children could be due to a number of factors, including the degree of micronutrient deficiency, outdoor exposure, and malaria-specific immunity, which improves with age. We previously showed that both iron and vitamin A status influenced malaria and anemia risk in this population [26,27].
Because nutritional deficiencies are generally higher in younger children, it is plausible that the observed age-pattern in malarial anemia may in fact be driven by micronutrient deficiencies. However, in this paper, adjustment for both iron and vitamin A status did not change the observed age-specific patterns in malarial anemia. The more plausible explanation is that the age-related pattern in malarial anemia is driven by underlying differences in acquired clinical immunity to malaria.
In this population, although malaria prevalence increased with age (perhaps because older children engaged in more outdoor activity) [26], the parasite density was generally higher in the youngest children, particularly during the high malaria season (data not shown). This is consistent with previous studies which showed that clinical immunity to malaria develops gradually and after repeated exposures [15]. This reasoning is also supported by the fact that the age-specific patterns in malarial anemia were observed only in the high malaria season but not the low malaria season. Unfortunately, we did not assess biomarkers of malaria-specific immunity in this study. Thus, additional evidence is needed to enhance understanding of the observed patterns.
There are several strengths, but also some limitations with the current study. This study is limited by its observational nature, particularly considering the complex etiology of anemia in children. Hemoglobin levels are driven by a complex interplay of nutritional factors (particularly iron metabolism), inflammation and in the context of malaria, by factors which modulate erythrocyte destruction and synthesis [28].
Because these factors are correlated, it is challenging to elucidate cause-specific effects on anemia from an observations study. To mitigate this potential bias, we adjusted for iron status (using ferritin and sTfR) and inflammation (using CRP and AGP). Another challenge is the lack of data on other factors such as non-malarial infections, including from helminths [29] and HIV which are both prevalent in this setting [30]. It is worth mentioning that all children participating in this study were provided deworming tablets in accordance with current recommendation. Hence, it is unlikely that our estimates are significantly biased by undiagnosed helminth infections. Furthermore, we did not have data on sickle cell status, a major determinant of anemia in this population [31]. Our study is strengthened by its longitudinal nature, which allowed us to explore how changes in the malaria transmission intensity influenced age-related pattern in the malarial anemia risk. In

CONFLICT OF INTEREST
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.

ACKNOWLEDGMENTS
We would like thank the participating children, their families, and Mkushi District officials for supporting the study's implementation. We are also grateful to Dr. Mwanza and the Mkushi District Medical Office for providing bed nets and Coartem to support malaria prevention and control activities. We thank Margia Arguello, Bess Caswell, and Lauren Tanz for supporting data collection and Brian Dyer, Mitra Maithilee, and Lee Wu for their support in data management, and Dr. Douglas Norris for providing inputs to this paper.