The Proportion of Anemia Associated with Iron Deficiency in Low, Medium, and High Human Development Index Countries: A Systematic Analysis of National Surveys

Iron deficiency is commonly assumed to cause half of all cases of anemias, with hereditary blood disorders and infections such as hookworm and malaria being the other major causes. In countries ranked as low, medium, and high by the Human Development Index, we conducted a systematic review of nationally representative surveys that reported the prevalence of iron deficiency, iron deficiency anemia, and anemia among pre-school children and non-pregnant women of reproductive age. Using random effects meta-analyses techniques, data from 23 countries for pre-school children and non-pregnant women of reproductive age was pooled, and the proportion of anemia attributable to iron deficiency was estimated by region, inflammation exposure, anemia prevalence, and urban/rural setting. For pre-school children and non-pregnant women of reproductive age, the proportion of anemia associated with iron deficiency was 25.0% (95% CI: 18.0, 32.0) and 37.0% (95% CI: 28.0, 46.0), respectively. The proportion of anemia associated with iron deficiency was lower in countries where anemia prevalence was >40%, especially in rural populations (14% for pre-school children; 16% for non-pregnant women of reproductive age), and in countries with very high inflammation exposure (20% for pre-school children; 25% for non-pregnant women of reproductive age). Despite large heterogeneity, our analyses suggest that the proportion of anemia associated with iron deficiency is lower than the previously assumed 50% in countries with low, medium, or high Human Development Index ranking. Anemia-reduction strategies and programs should be based on an analysis of country-specific data, as iron deficiency may not always be the key determinant of anemia.


Introduction
Anemia is a public health problem that affects populations worldwide. To assess the global prevalence of anemia, the World Health Organization (WHO) established a global database containing population-based cross-sectional surveys and intervention studies. Using data collected from 1993 to 2005, it was estimated that about 1.6 billion people (a quarter of the world's population) suffered

Inclusion and Exclusion Criteria
We included nationally representative surveys of PSC (6-59 months of age as the primary target age range, with some flexibility to include surveys with slightly older children; 10-75 months [14]) and WRA (15-49 years) in the analyses. Studies and surveys representative only of either urban or rural populations were also included to obtain separate estimates for urban and rural settings. To be included in the analysis, each report had to present the prevalence of anemia based on low hemoglobin levels as well as IDA (defined using low serum ferritin and low hemoglobin), where ideally, ferritin was adjusted for inflammation. The reported inflammation adjustment of included surveys was made by one of the following methods: (a) the approach proposed by Thurnham [15; 4-stage adjustment model: elevated C-reactive protein (CRP) only; both CRP and alpha-1-acid glycoprotein (AGP) elevated; or AGP elevated only]; (b) presenting IDA prevalence based on serum ferritin concentrations in a subsample without inflammation; (c) increasing the serum ferritin cut-off for ID to 20 µg/L for the whole study population (one survey [16]). In case IDA was not adjusted for inflammation by the authors, but the proportion of the population with elevated CRP or AGP was reported, IDA was adjusted applying an internal correction factor (see description below), and the survey was also included in the analyses.
We excluded studies that were not population-representative but used convenience or non-random sampling to select subjects, e.g., volunteers, clinic or hospital patients, and studies restricted to ethnic minorities, immigrant populations, or non-representative subgroups. We additionally excluded studies that did not report the sampling method used, studies whose sampling methodology could not be identified after all reasonable attempts, and studies that used sampling approaches that targeted specific hemoglobin levels, serum/plasma ferritin levels, other measures of body iron, anemia, iron deficiency, or iron deficiency anemia (e.g., studies that excluded severely anemic individuals, or were restricted to anemic individuals). Further, based on the very limited number of reports available, we excluded from analysis studies conducted in countries which were ranked as having a "very high Human Development Index" [17].

Estimating ID and IDA in Presence of Inflammation
We used the clinical case definitions as put forth by the authors of the different survey reports to define anemia, ID, IDA, and inflammation (% with elevated acute phase proteins (APPs)). In order to use data from seven surveys (out of 25) that reported IDA but did not correct for inflammation, we developed an algorithm to estimate ID and IDA in the presence of inflammation in these data sets. To correct the prevalence of IDA, we assumed that the proportion of ID in subjects who presented with no inflammation was the same as that in the subjects presenting with inflammation (elevated CRP or AGP as defined by the authors) and that the subjects with inflammation had not been diagnosed with IDA (i.e., anemia and serum ferritin < 15 µg/L in women; anemia and serum ferritin < 12 µg/L in children) because of inflammation-elevated ferritin levels. Thus, the IDA prevalence reported for the whole study population only reflected the prevalence in the subsample without inflammation. Based on these assumptions, we applied the same proportion of IDA to the segment of the population with inflammation as reported for the overall population, which resulted in an increase of IDA proportional to the % of the population with elevated APP:

Grouping Countries by Inflammation Exposure
Since previous estimations indicate that the exposure to infection and chronic inflammatory conditions have an impact on the proportion of anemia associated with ID, and because the level of inflammation varies considerably between countries and population groups, we developed an inflammation-exposure index for children and women for each country using data obtained from external sources. These indices were constructed using factors known to contribute to inflammation. For children, the inflammation-exposure score was based on (a) prevalence of presumed and confirmed malaria cases [18,19]; (b) schistosomiasis prevalence [20]; and (c) an overall hygiene score based on the proportion of population using improved drinking water source and the proportion of the population using improved sanitation facilities (evenly weighted) as a proxy for the risk of enteric inflammation [21]. For women, the different factors used to estimate country-specific inflammation-exposure were (a) prevalence of presumed and confirmed malaria cases [18,19]; (b) HIV prevalence in adults [22]; (c) obesity prevalence in female adults [23]; (d) schistosomiasis prevalence [20]; and (e) an overall hygiene score based on the proportion of population using improved drinking water source and the proportion of the population using improved sanitation facilities (evenly weighted) as a proxy for the risk of enteric inflammation [21]. For 188 countries, information on the aforementioned factors was collected and combined into a single inflammation score. Using this score, countries were classified as having low, medium, high, or very high inflammation exposure (see Table S1 for a complete country list and references for the data used to calculate the index).

Data Synthesis
Data for estimating ID, IDA, and anemia prevalence were extracted from the available reports or publications into an Excel spreadsheet. We estimated the proportion of anemia associated with ID separately for PSC and WRA using the following calculation: (prevalence of IDA)/(prevalence of anemia). The country-specific proportions were then pooled using random effects meta-analysis of double arcsine transformed proportions (Freeman-Tukey transformation), which correctly estimates variances even in the presence of extreme proportions close to 0 or 1 and constrains confidence intervals to within the 0-1 range after back transformation [24,25]. Estimates were further pooled by region, inflammation exposure, setting (urban or rural), and the WHO classification of anemia burden for the country, separately for PSC and WRA. Analyses were conducted using R version 3.2.3 (The R Foundation, Vienna, Austria, 2015).

Literature Research
In the first round, about 14,000 publications were identified according to the publication title and abstract. After screening the titles, abstracts, and the identification of surveys outside the literature search, 174 publications were selected for further investigation, from which 25 surveys met the inclusion criteria. In total, 23 national and 2 urban national representative publications reporting on anemia, IDA, and elevated CRP or AGP in PSC or WRA met our inclusion criteria. Surveys from Cambodia, Cameroon, and Afghanistan included 10.0%, 5.3%, and 17.5% pregnant women, respectively, as part of the WRA sample (0.8% of total study sample). All other surveys only included non-pregnant women of reproductive age in the analyses.

Country Inflammation Exposure Categorization
Based on the inflammation exposure index for children, globally, 21 countries were categorized into the very high exposure group, 28 countries into the high group, 33 countries in the medium group, and 106 countries in the low exposure group. For women, 20 countries were classified as being in the very high inflammation exposure category, 28 countries in the high inflammation exposure category, 58 countries in the medium exposure and 82 countries in the low exposure category. All countries in the very high exposure category for PSC and WRA are located in Sub-Saharan Africa. Similarly, the high exposure category is dominated by countries from Sub-Saharan Africa with 22 countries for PSC and 24 for WRA (Table S1).

Study Characteristics
Of the 25 included studies in the meta-analysis, 23 reported on children, and 23 on women. For PSC, seven surveys were conducted in Sub-Saharan Africa and South, East, and Southeast Asia, four in Latin America and the Caribbean, three in Central Asia, and two in the Middle East. Seven studies conducted in Sub-Saharan Africa, six in South, East, and Southeast Asia, four in Latin America and the Caribbean, four in Central Asia and two in Middle East reported on WRA. The studies included in our analysis were conducted between 2003 and 2014 (Tables 1 and 2). The majority of countries included in the analysis fell into the low to medium inflammation-exposure index category, and only countries located in Sub-Saharan Africa were categorized into the very high exposure category.

Proportion of Anemia Associated with ID among PSC
The proportion of anemia associated with ID (Table 3) varied widely for PSC in the 21 countries with nationally representative data, and ranged from 0.2% for Georgia to 51% for Kenya with the pooled proportion for all countries being 24.6% (95% CI: 17.7, 32.2; Table 3, Figure S1). Regional estimates of the proportion of anemia associated with ID also varied considerably; 11.0% (95% CI: 0.2, 34.5) for Central Asia , 26 Figure S2). The proportion was slightly lower for children in rural settings than in urban settings (28.0% vs. 32.5%; Figure S3). Notably, the overall severity of anemia burden experienced by the country (estimated using the WHO classification) made little difference to the proportion of anemia associated with ID (25.7% for countries with moderate anemia burden and 21.7% for countries with severe anemia burden; Figure S4). However, after stratification for urban/rural setting, the proportion of anemia associated with ID decreased strongly with increasing anemia burden. In the rural setting, the proportion of anemia associated with ID decreased from 33.8% in countries where anemia is a moderate public health problem to 13.9% in countries where anemia is a severe public health problem. Similar results were found in the urban areas, decreasing from 39.8% to 26.0%, respectively (Figure 1). The inflammation exposure of the country did not change the proportion of anemia associated with ID ( Figure S5).

Anemia Attributable to ID
Our analyses-which focus on countries with a low, medium, and high HDI ranking-suggest that the overall proportion of anemia associated with ID is lower than commonly assumed.

Anemia Attributable to ID
Our analyses-which focus on countries with a low, medium, and high HDI ranking-suggest that the overall proportion of anemia associated with ID is lower than commonly assumed. Consequently, our analyses suggest that the extent to which iron interventions alone can decrease anemia prevalence is also lower than previously believed. In particular, the relative impact of iron interventions may be smaller in populations with both high burden of anemia and very high inflammation exposure as the proportion of anemia associated with ID, tends to be relatively low, and iron absorption is restricted in the presence of inflammation. In such populations, more context-specific combinations of programmatic approaches to reduce the anemia may be needed. Our findings suggest that countries where anemia is a severe public health problem (i.e., anemia prevalence > 40%) should include infection prevention and control as a key strategy to reduce the prevalence of anemia. Iron interventions can and should be used in these contexts, but their implementation should be based on data from population representative surveys that help identify the target groups where ID is a public health problem.
To our knowledge, this is the first multi-country study to assess the proportion of anemia associated with ID by using a direct measure of ID (i.e., low ferritin concentration). Ferritin is currently the most widely used biomarker to assess the prevalence of ID. However, it is part of the acute phase response, and serum ferritin levels are increased in the presence of inflammation. If ferritin concentrations are not corrected for inflammation, the prevalence of ID can be underestimated by more than 14% [15]. In the present study, we exclusively included surveys measuring at least one further acute phase protein (e.g., CRP or AGP) and used adjusted ferritin concentrations for acute inflammation, as recommended by WHO [50].
Other research estimating the proportion of anemia amenable to iron interventions has been based on pooled reported shifts in hemoglobin concentrations after iron supplementation [2,10,12] or fortification [9] interventions. Using data from two meta-analyses on intermittent iron supplementation trials in children [51] and women [52], the WHO calculated a hemoglobin shift of 8 mg/L and 10.2 mg/L, respectively, resulting in global estimates of 42% for PSC and 49% for non-pregnant women of anemia attributable to ID. The hemoglobin shift estimates also included studies where participants received malaria and anthelminthic treatment in addition to iron, which might have resulted in an additional shift and an overestimation of anemia amenable to iron interventions. Furthermore, the included supplementation trials predominantly targeted specific hemoglobin levels (e.g., only anemic, only iron deficient anemic, or exclusion of severely anemic children) and would thus not be representative for entire populations. Also looking at the exclusive effect of iron supplementation, Bhutta and colleagues [10] found a decrease of anemia prevalence ranging from 38% to 62% in non-malarial regions and from 6% to 32% in malarial hyperendemic areas. A recent study of children aged 12-36 months of age conducted in a malaria endemic region in Côte d'Ivoire reported that iron-fortified porridge reduced the prevalence of iron deficiency, but had no effect on anemia prevalence, as malaria anemia seemed to override anemia associated with ID [53].
Using cause-specific shifts in hemoglobin after iron-fortification interventions, Kassebaum et al. [9] estimated that, in PSC, about 60% of anemia is associated with ID globally. To calculate the anemia attributable to ID, the authors compared the hemoglobin shift after iron fortification to the total shift of hemoglobin taking into account the prevalence of the different causes of anemia in 187 countries. The calculated hemoglobin shift of 6.05 mg/L was based on the results of a meta-analysis of seven iron-fortification comparisons drawn from five studies in India, Pakistan, South Africa, Guatemala, and Chile, none of those having a very high inflammation burden. Three of the comparisons were conducted in school-aged children, two of which had markedly higher Hb shifts (14 and 17 mg/L) than other studies included in the meta-analysis. These assumptions could explain some of the differences observed between our approach and the method by Kassebaum, particularly for PSC in countries with very high exposure to inflammation. Though our analysis is not global and could not make estimations by WHO regions, some scoping comparisons between our findings and WHO's estimates can be made. Globally, the WHO [2] estimates are markedly higher than our overall estimates of 25% and 37%, and this trend is consistent with our approximations of regional estimates. For children and non-pregnant women, WHO respectively estimated the proportion of anemia attributable to ID at 32% and 41% in the Africa Region and 41% and 54% in the South-East Asia Region [2]. Our estimates for Sub-Saharan African countries were 28% and 33% for PSC and WRA, and 24% and 26% for South-East Asia (see Figures S2 and S7). As countries in these two regions have a higher disease and inflammation risk, and sizable populations with hemoglobin disorders, our findings suggest that the Hb shift methodology employed by WHO could have overestimated the proportion of anemia associated with ID in these regions.
Understanding the contribution of ID to anemia is important to national-level programmers and public health professionals. In addition to helping design anemia-reduction strategies, the proportion of anemia attributable to ID is often used by health economists when estimating disease burden using disability-adjusted life years (DALYs) or similar metrics. As data on IDA is relatively scarce, DALY calculation methods [54] have recommended estimating the prevalence of IDA as 50% of the total anemia prevalence if national-level IDA data is not available. Our analysis suggests that, if available, national-level data should be used for estimates of disease burden. If national estimates are not available, a proportion of 25% and 37% can be temporarily used for PSC and WRA in countries with a low to high HDI until a survey measuring IDA prevalence can be conducted.

Heterogeneity of Estimates of ID Associated with Anemia
Our analyses also found that the proportion of anemia associated with ID is highly heterogeneous. Regional stratification could not explain the observed heterogeneity, which is likely owed to the large variability within regions. The most distinct variability was found in the region of Central Asia, with Georgia and Tajikistan exhibiting very low, and Uzbekistan and Azerbaijan very high proportions of anemia associated with ID. Reasons for those differences are unclear, since Azerbaijan and Georgia as well as Tajikistan and Uzbekistan are neighboring countries and have similar environmental settings, similar living standards, and comparable inflammation exposures. Thus, these differences remain to be elucidated by looking more closely at dietary habits as well as the etiology of anemia in those countries. Much less variability was found in Latin America and the Caribbean countries, allowing a more solid estimate for this region, with the proportion of anemia associated with ID ranging from 17% to 34% among PSC and 53% to 63% among WRA.
Similar to this, we found a comparably low heterogeneity in anemia associated with ID among children in South, East, and Southeast Asian countries (12%-35%), which was not the case for Sub-Saharan Africa. Further subgroup analyses that were stratified by burden of anemia showed that in countries with severe burden of anemia such as Sierra Leone, Côte d'Ivoire, and Liberia, the proportion of anemia associated with ID is significantly lower, especially in the rural populations (14% for PSC; 16% for WRA). This difference for rural populations might be explained by poor sanitary conditions that are often accompanied with increased prevalence of infection.

Limitations of Analysis Methodology
Our methodology has limitations that might have impacted the precision of our estimates. First, we were not able to use a consistent approach for correcting ferritin concentrations and ID/IDA prevalence for inflammation. The most frequently used correction factor, developed by Thurnham [15], is based on a meta-analysis looking at serum ferritin during different phases of the acute phase response (incubation, early convalescence, late convalescence) using CRP and AGP. The correction factor we developed for seven surveys showed a strong correlation with Thurnham's correction factor, except when there is a high prevalence of elevated APP. In this case, our correction factor tends to be higher, resulting in a greater increase in ID and IDA prevalence. Some of the surveys included in our analysis only measured ID and IDA in subjects without inflammation. This substantially reduced the sample size of some surveys and thus might lead to biased results as their reported ID prevalence was not completely representative of the population. These different corrections may have under-or overestimated the prevalence of ID and IDA and thus the proportion of anemia associated with ID. Inflammation is a very complex process and it is perhaps simplistic to believe that one correction can fit all causes of inflammation. Currently, the Biomarkers Reflecting Inflammation and Nutritional Determinants of Anemia (BRINDA) Project is examining new approaches to adjust ferritin concentrations for inflammation. In addition to these efforts, greater research is needed to elucidate the relationship between inflammation and serum ferritin concentrations [55].
Second, our approach may have overestimated the proportion of anemia that can be resolved with iron interventions in malaria-endemic areas. To illustrate, anemia can often be multifactorial in the same person. Even in individuals where anemia is associated with ID, communicable diseases such as malarial infection can prevent anemia prevalence being decreased with iron interventions [53].
Lastly, our analyses contained sparse data from countries in the Eastern and Central Europe and the Pacific Regions, which had highest proportions of IDA according to the estimations made by Kassebaum et al [9]. As we could not identify suitable data from countries in these regions for our analysis, our overall estimate of the proportion of anemia associated with ID may be lower than global estimates.

Conclusions
Our analyses suggest that the all-encompassing estimate that 50% of anemia is attributable to ID is too high for PSC and WRA in countries with a low, medium, and high HDI ranking. Furthermore, we observed a large heterogeneity in the proportion of anemia associated with ID between countries in the same region. We therefore conclude that estimating the proportion of anemia attributable to ID using a fixed proportion (based on data from another country or hemoglobin shifts observed by supplementation studies) is inappropriate. Where possible, national surveys should measure ID and IDA prevalence so that program planners and national stakeholders can better understand the etiology of anemia in their country. Our assessment was limited to the 25 surveys that met our inclusion criteria. This relatively small number of surveys highlights the need to collect data on ID and IDA prevalence along with anemia on the national level. Ideally, this should be complemented by the assessment of other causes of anemia. Until a nationally representative survey can be conducted, public health and nutrition stakeholders can cautiously estimate that 25% and 37% of anemia is associated with ID in PSC and WRA, respectively. We argue that the current habit of assuming 50% of anemia attributable to ID should no longer be used for countries with a low to high HDI.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-6643/8/11/693/s1, Figure S1: Overall proportion of anemia associated with iron deficiency (ID) among pre-school children (PSC), Figure S2: The proportion of anemia associated with ID stratified by region among PSC, Figure S3: The proportion of anemia associated with ID stratified by rural/urban setting among PSC, Figure S4: The proportion of anemia associated with ID stratified by anemia public health significance among PSC, Figure S5: The proportion of anemia associated with ID stratified by inflammation exposure among PSC, Figure S6: Overall proportion of anemia associated with ID among women of reproductive age (WRA), Figure S7: The proportion of anemia associated with ID stratified by region among WRA, Figure S8: The proportion of anemia associated with ID stratified by rural/urban setting among WRA, Figure S9: The proportion of anemia associated with ID stratified by anemia public health significance among WRA, Figure S10: The proportion of anemia associated with ID stratified by inflammation exposure among WRA, Table S1: Inflammation exposure country categorization PSC and WRA.