点击显示 收起
1 From the Human Nutrition Laboratory, Swiss Federal Institute of Technology, Zürich (MBZ, FS-A, and RFH); the Department for Growth and Development, University Childrens Hospital, Zürich (LM); the Ministry of Health, Rabat, Morocco (NC); and the Ministry of Health, Abidjan, Côte dIvoire (PA)
2 Supported by the Thrasher Research Fund (Salt Lake City), The Nestlé Foundation (Lausanne, Switzerland), the Swiss National Fund (Bern, Switzerland), and the Swiss Federal Institute of Technology (Zürich). 3 Reprints not available. Address correspondence to MB Zimmermann, Human Nutrition Laboratory, Swiss Federal Institute of Technology Zürich, Seestrasse 72/Postfach 474, CH-8803 Rüschlikon, Switzerland. E-mail: michael.zimmermann{at}ilw.agrl.ethz.ch.
ABSTRACT
Background: Although transferrin receptor (TfR) and zinc protoporphyrin (ZnPP) are often used to define iron status in school-age children in developing countries, the diagnostic cutoffs for this age group are uncertain.
Objective: The objective was to determine the sensitivity and specificity of TfR and ZnPP in predicting iron deficiency in black and white children in Africa.
Design: Hemoglobin, C-reactive protein (CRP), serum ferritin (SF), TfR, and ZnPP were measured in children in Côte dIvoire and Morocco. We excluded children with elevated CRP and then used receiver operating characteristic (ROC) curves to evaluate TfR and ZnPP alone and in combination in screening for iron deficiency, defined as an SF concentration <15 µg/L, and iron deficiency anemia (IDA), defined as an SF concentration <15 µg/L and low hemoglobin.
Results: The sample included 2814 children aged 515 y. The sensitivity and specificity of TfR and ZnPP were limited by considerable overlap between iron-sufficient, nonanemic children and those with IDA. On the basis of ROC curves, we identified diagnostic cutoffs for TfR and ZnPP that achieved specificities and sensitivities of 6080%. Separate cutoffs for Côte dIvoire and Morocco gave the best performance; the cutoffs for both TfR and ZnPP were higher in Côte dIvoire. Moreover, a comparison of nonanemic, iron-sufficient subjects showed that Ivorian children had significantly higher TfR and ZnPP concentrations than did Moroccan children (P < 0.01).
Conclusions: New diagnostic cutoffs for TfR and ZnPP, based on ROC curve analyses, may improve the performance of these indexes in defining iron status in children. Significant ethnic differences in TfR and ZnPP suggest that separate cutoffs may be needed for black and white children.
Key Words: Iron iron deficiency hemoglobin zinc protoporphyrin transferrin receptor serum ferritin receiver operating characteristic curve school children anemia
INTRODUCTION
Iron deficiency anemia (IDA) is common among children in developing countries, where the prevalence is often 50% or more (1). In young children, iron deficiency, with or without anemia, may impair psychomotor development and cognitive function (24). Defining iron status in developing countries is a challenge. Hemoglobin is often used to screen for anemia as a proxy for iron deficiency because of its simplicity and low cost. However, hemoglobin detects only the late stages of iron deficiency and its specificity is poor, particularly in developing countries (57). Biochemical testsserum iron, total-iron-binding capacity (TIBC), transferrin saturation, serum ferritin (SF), serum transferrin receptor (TfR), and erythrocyte zinc protoporphyrin (ZnPP)are more specific than is hemoglobin (8). However, their interpretation in developing countries is limited by the confounding effects of infection, inflammation, and malnutrition. SF is the standard test for diagnosing iron deficiency. A concentration <1220 µg/L is a highly specific indicator of deficiency, and, in the presence of anemia, indicates IDA (1, 9). A major limitation of SF is its acute phase response, which can mask iron deficiency during infection (9).
TfR and ZnPP are sensitive measures of iron-deficient erythropoiesis and have been used to define iron status in children in developing countries (6, 7, 1012). TfR may have an advantage over SF because it is unaffected by the acute phase response (7, 13, 14). However, the specificity of TfR may be low because it can be increased by malaria (15), megaloblastic anemia due to vitamin deficiencies (16), and hemoglobinopathies such as sickle cell disease (17), hemoglobin H disease, and the thalassemias (18, 19). ZnPP has advantages of low cost and simplicity, but its specificity may be low as it can be increased by malaria and other infections, chronic inflammation, and hemoglobinopathies (6, 7, 2023). To improve specificity, SF is often combined with TfR, ZnPP, or both (9).
Data on the normal physiologic concentrations of TfR and ZnPP in children are scarce, and a lack of standardization limits direct comparisons obtained with different assays (16, 24). Several different diagnostic cutoffs for TfR and ZnPP to identify iron deficiency have been proposed, but they have been derived from studies in adults (21, 25). They may not apply to children because of increased erythropoiesis during growth (26, 27). Although healthy black children may have lower hemoglobin concentrations than white children (28, 29), and Allen et al (30) reported that black adults have 10% higher mean TfR concentrations than do whites, there have been no comparative studies of TfR and ZnPP in children from different ethnic backgrounds.
To help clarify these issues, we pooled and reanalyzed data from our recent surveys of iron status in school-age children from West and North Africa (7, 3137). To reduce the confounding effect of inflammation and infection, we excluded children who had elevated C-reactive protein (CRP) concentrations. Receiver operating characteristic (ROC) curves were used to describe the performance of ZnPP and TfR in screening for iron deficiency and IDA and to propose new diagnostic cutoffs for children.
SUBJECTS AND METHODS
Study sites
The studies were approved by the ethical committees at the Swiss Federal Institute of Technology Zürich or the University of Zürich Childrens Hospital and the Ministries of Health in Côte dIvoire and Morocco. Written (or oral, if illiterate) informed consent was obtained from the parents of the children, and oral consent was obtained from the children. The studies in Morocco were done in rural villages in the Rif Mountains of northern Morocco. This region is 400800 m above sea level, with a population of mixed Berber and Arab descent. The prevalence of IDA in school-age children is 1530% (3537). Because of the dry, temperate climate and clean public water supply, parasites that cause blood loss are rare, and there is no malaria. Low iron bioavailability from a legume- and cereal-based diet is the main cause of iron deficiency in this region (38). In Côte dIvoire, the studies were conducted in rural villages in the north, west, central, and southern regions of the country and in Abidjan, a large coastal city. The prevalence of IDA is 1025% in school-age children in Côte dIvoire, and the cause is multifactorial, including low dietary iron bioavailability, hemoglobinopathies, and endemic parasitoses, including hookworm, schistosomiasis, and malaria (7, 31, 34, 39).
Data collection and analysis
The screenings were done in primary schools. Children were registered, heights and weights were measured, and age and sex were recorded (7, 3137). Venous blood samples were collected in EDTA-containing tubes and transported on ice to the local laboratory. In Morocco, hemoglobin was measured with the use of an AcT8 Counter (Beckman Coulter, Krefeld, Germany) by using 3-level controls provided by the manufacturer. In Côte dIvoire, hemoglobin was measured in duplicate with the cyanomethemoglobin technique (Sigma Diagnostics kit; Sigma, St Louis) by using 3-level controls (DiaMed, Cressier sur Morat, Switzerland). Whole blood was analyzed within 12 h of blood sampling. In Morocco, anemia was defined according to the World Health Organization (WHO) criteria (1), as a hemoglobin concentration <130 g/L in boys aged 15 y, <120 g/L in children aged 12 y and in girls aged 15 y, and <115 g/L in children aged 511 y. In Côte dIvoire, because normal hemoglobin concentrations are lower in blacks (28, 29) and hemoglobin cutoffs that are 10 g/L lower in black populations may improve screening performance (1), anemia was defined as a hemoglobin concentration below the WHO cutoff values minus 10 g/L.
ZnPP was measured in red blood cells, after the cells were washed with normal saline (21), with a hematofluorometer (Aviv Biomedical, Lakewood, NJ) and 3-level control material provided by the manufacturer. Measurements were made in stored refrigerated blood within 4 d of collection and expressed as µmol/mol heme (40). The normal reference range in washed cells with this assay is 40 µmol/mol heme (21). Serum aliquots were frozen at 20°C until analyzed. Enzyme-linked immunosorbent assays were used to measure SF (41) and TfR (42). All of the Ivorian and 25% of the Moroccan samples were analyzed at the University of Kansas Medical Center in the laboratory of J Cook. The remaining samples were analyzed in our laboratory at the Swiss Federal Institute of Technology. CRP was measured by nephelometry (TURBOX; Orion Diagnostica, Espoo, Finland) with 2-level control material provided by the manufacturer; the normal reference range is <10 mg/L.
Because infection and inflammation confounds the interpretation of SF, TfR, and ZnPP, we excluded children who had signs of infection or inflammation, as defined as an elevated CRP concentration (43, 44). About 24% and 10% of children in the original surveys from Côte dIvoire and Morocco were excluded because of an elevated CRP or incomplete measurements of hemoglobin, SF, TfR, or ZnPP. Children with sickle cell disease were not included in the Ivorian sample; the Moroccan sample was not screened for hemoglobinopathies. The criteria used for definitive diagnosis of iron deficiency are as follows: decreased or absent stainable iron in a bone marrow aspirate or an increase in hemoglobin in response to iron therapy (8, 9). Because these 2 criteria were not available and the sample included only children with a normal CRP concentration, we defined iron deficiency and IDA according to WHO criteria (1), ie, iron deficiency as an SF concentration <15 µg/L and IDA as an SF concentration <15 µg/L and a hemoglobin concentration below the cutoffs from the WHO stated above. Because several investigators have argued that an SF concentration >30 µg/L should be used to define adequate iron stores in developing countries with a high prevalence of infection (7, 45), we chose an SF cutoff of >30 µg/L as being indicative of iron sufficiency.
Statistical analysis
Data processing and statistical analyses were done by using SPLUS-2000 (Insightful Corporation, Seattle) and EXCEL (Enterprise Edition; Microsoft, Seattle). Values are expressed as means ± SDs, medians (ranges), or proportions and were compared with the Mann-Whitney and chi-square tests. Logarithmic transformation was applied to SF, ZnPP, and TfR because of skewed distributions. Pearsons correlations and multiple regressions were done to examine relations between age, sex, country, weight, height, hemoglobin, SF, TfR, and ZnPP. ROC curves were used to characterize the sensitivity and specificity of ZnPP and TfR to define iron status (46). SEs of the area-under-the curve (AUC) of the ROC curves were calculated by bootstrapping (47).
RESULTS
Demographic and biochemical characteristics of the children are shown in Table 1. The sample included 2814 children; 36% were Ivorian and 46% were female. Correlations between hemoglobin and the iron-status indicators, by country, are shown in Table 2. In the total sample, the strongest associations were between ZnPP and TfR (r = 0.53, P < 0.0001), whereas associations between SF and either TfR or ZnPP were minimal. Correlations between SF and hemoglobin were also small, particularly in Côte dIvoire, reflecting the multifactorial etiology of anemia. The overlap of the distributions of ZnPP and TfR in a comparison of children with IDA with those with normal iron status, as defined as an SF concentration >30 µg/L and a normal hemoglobin concentration, is shown in Figures 1 and 2, respectively. On the log10 scale, mean (±SD) ZnPP values were 1.805 ± 0.377 and 1.555 ± 0.305 in children with IDA and normal iron status, respectively, whereas mean (±SD) TfR values were 1.053 ± 0.260 and 0.887 ± 0.149 in children with IDA and normal iron status, respectively. The considerable overlap of the distributions reduced the sensitivity and specificity of ZnPP and TfR to discriminate iron status in children.
View this table:
TABLE 1. Demographic and biochemical characteristics of the sample
View this table:
TABLE 2. Correlations (by regression) between hemoglobin, serum ferritin, transferrin receptor, and zinc protoporphyrin, by country
FIGURE 1.. Distribution of log10(zinc protoporphyrin) in 515-y-old children with iron deficiency anemia (solid heavy line; n = 447) and in iron-sufficient, nonanemic children (solid light line; n = 668) in Côte dIvoire and Morocco.
FIGURE 2.. Distribution of log10(serum transferrin receptor) in 515-y-old children with iron deficiency anemia (solid heavy line; n = 447) and in iron-sufficient, nonanemic children (solid light line; n = 668) in Côte dIvoire and Morocco.
The sensitivities and specificities of current cutoffs for ZnPP, TfR, and a combination of ZnPP and TfR in identifying iron deficiency and IDA are shown in Table 3 for both countries individually and combined. In Côte dIvoire, the current cutoffs for both TfR and ZnPP showed high sensitivity but low specificity. In contrast, in Morocco, the TfR cutoff showed low sensitivity but good specificity. A combination of TfR and ZnPP showed good specificity and sensitivity in Côte dIvoire but low sensitivity in Morocco. Proposed diagnostic cutoffs for ZnPP and TfR for identifying iron deficiency and IDA, based on the analyses of the ROC curves shown in Figures 3 and 4, are shown in Table 4. Considering the tradeoff between sensitivity and specificity, we chose cutoffs for TfR and ZnPP at a specificity of 60% to reduce the number of false-positive results while maintaining sensitivity in the range of 6080%. Discrete cutoffs for Côte dIvoire and Morocco provided the best performance. The cutoffs for both TfR and ZnPP were distinctly higher in Côte dIvoire than in Morocco. Based on the ROC curves for the entire sample, "compromise" cutoffs for ZnPP and TfR for all children from Côte dIvoire and Morocco are also shown in Table 4. The proposed new cutoffs increased the specificity of both TfR and ZnPP for defining iron deficiency and IDA in Côte dIvoire, with minimal decreases in sensitivity. Compared with the current cutoffs, at an IDA prevalence of 15%, the proposed new cutoffs in Côte dIvoire increased the positive predictive value of an elevated TfR from 23% to 28% and of an elevated ZnPP from 21% to 25%, whereas the negative predictive value of these indicators was reduced from 96% to 94% and from 95% to 92%, respectively. In contrast, in Morocco, they increased the sensitivity of TfR and ZnPP while maintaining specificity at 60%.
View this table:
TABLE 3. Sensitivities (sens) and specificities (spec) of current diagnostic cutoffs for serum transferrin receptor (TfR) and zinc protoporphyrin (ZnPP) individually and combined in school-age children, by country1
FIGURE 3.. The receiver operating characteristic (ROC) curves for transferrin receptor and zinc protoporphyrin in detecting iron deficiency in 515-y-old children from Morocco (n = 1798), Côte dIvoire (n = 1016), and the 2 countries combined (n = 2814). The mean (±SD) area under the curve (AUC) is shown for each ROC curve.
FIGURE 4.. The receiver operating characteristic (ROC) curves for transferrin receptor and zinc protoporphyrin in detecting iron deficiency anemia in 515-y-old children from Morocco (n = 1798), Côte dIvoire (n = 1016), and the 2 countries combined (n = 2814). The mean (±SD) area under the curve (AUC) is shown for each ROC curve.
View this table:
TABLE 4. Sensitivities (sens) and specificities (spec) of proposed diagnostic cutoffs for serum transferrin receptor (TfR) and zinc protoporphyrin (ZnPP) individually and combined in school-age children, by country1
The medians and 95% CIs of TfR and ZnPP in children from Morocco and Côte dIvoire who had a normal hemoglobin concentration and were iron sufficient, as defined by an SF concentration > 30 µg/L, are shown in Table 5. In these nonanemic children, multiple regression indicated that the Ivorian children had significantly higher concentrations of TfR and ZnPP than did the Moroccan children (P < 0.01); the medians were higher by 1.5 mg/L and 9 µmol/mol heme, respectively, in Côte dIvoire. There was no significant difference in TfR by age in either country, but boys in both countries had significantly higher concentrations than did girls (P < 0.01). On the log10 scale, the mean (±SD) predicted values for TfR were 0.84 ± 0.01 and 0.81 ± 0.01 for males and females, respectively, in Morocco and were 0.93 ± 0.01 and 0.90 ± 0.01 for males and females, respectively, in Côte d Ivoire. For ZnPP, there was no significant difference by age or sex in either country.
View this table:
TABLE 5. Serum transferrin receptor (TfR) and zinc protoporphyrin (ZnPP) in iron-sufficient (serum ferritin > 30 µg/L), nonanemic, school-age children (n = 2814) with normal C-reactive protein concentrations, by sex and country and by the 2 countries combined1
DISCUSSION
The main findings of the present study are that 1) the new diagnostic cutoffs for TfR and ZnPP based on ROC curve analyses may improve their performance in defining iron status in children in developing countries, 2) the sensitivity and specificity of TfR and ZnPP as indicators of iron status in Africa are limited by substantial overlap in children with IDA and iron sufficiency, and 3) there are significant ethnic differences in TfR and ZnPP concentrations in iron-sufficient, nonanemic children so that discrete cutoffs may be needed for black and white populations. The strengths of the study are as follows: 1) a large sample size in a discrete population vulnerable to iron deficiency was studied, 2) both black and white children with normal CRP values were included to reduce the confounding effect of inflammation, 3) accurate determinations were made of iron status with the use of monoclonal assays for TfR and SF and of ZnPP in washed erythrocytes by hematofluorometry, and 4) ROC curves were used to describe the diagnostic accuracy of TfR and ZnPP in discriminating iron deficiency and IDA over the complete spectrum of operating conditions.
However, the findings are also subject to several limitations. Although blood lead concentrations were not measured and may have biased the ZnPP measurements (48); 95% of the sample was from undeveloped rural areas unlikely to contain high amounts of environmental lead. We included only children with a normal CRP concentration to minimize the confounding effect of inflammation on SF, but the timing of the rise and fall of CRP during the acute phase response may differ from that of SF. It is possible that SF was elevated by the acute phase response in some children, despite a negative CRP, and the true prevalence of iron deficiency was underestimated. The use of additional acute phase proteins, such as 1-acid glycoprotein (49), could further reduce the confounding effects of inflammation, but these proteins were not measured. Although we chose an SF concentration >30 µg/L to define iron sufficiency, the use of even higher cutoffs to exclude iron deficiency would have increased specificity and may have reduced the high overlap of TfR and ZnPP in children with IDA and in iron-sufficient children. Although the combination of a low SF and a low hemoglobin concentration is definitive for IDA, the use of a low SF concentration to define iron deficiency is more problematic. SF measures the depletion of iron stores, whereas TfR and ZnPP measure iron-deficient erythropoiesis (21, 25). Although these 2 stages in the evolution of iron deficiency overlap, it is likely that some of the false-negative results (normal TfR and ZnPP but low SF) were found in children with depleted iron stores who had not yet progressed to iron-deficient erythropoiesis. Finally, because of significant interassay variability and lack of standardization (16, 48), our cutoffs are specific to the TfR and ZnPP assays used in this study and may not apply to infants or young womenthe other groups that are highly vulnerable to IDA.
The total mass of cellular TfR and, therefore, of serum TfR depends both on the number of erythroid precursors in the bone marrow and on the number of TfRs per cell, a function of the iron status of the cell (16). Therefore, normal expansion of the erythroid mass during growth (26, 27, 50), as well as diseases common in developing countries, including thalassemia, megaloblastic anemia due to folate deficiency, or hemolysis due to malaria, may increase erythropoiesis and TfR independent of iron status (51). In children, Ritchie et al (23) found that TfR was not significantly better than hemoglobin, red cell indexes, and CRP in distinguishing IDA from anemia secondary to infection. In young children, Malope et al (52) found that the combination of TfR and SF measurements was inadequate to clearly define iron status in the presence of inflammation. The diagnostic value of TfR for IDA is uncertain in children from regions where malaria is endemic (6, 15, 44, 5356). Age-related data for TfR in children are scarce (5759), and direct comparison of values obtained with different assays is difficult because of the lack of an international standard. Median (95% CIs) TfR values in 1112-y-old nonanemic, iron-sufficient Swedish boys was 7.0 (4.7, 9.2) mg/L, whereas values in men were higher5.8 (3.1, 8.5) mg/L (57). Using an immunoturbidimetric assay, Suominen et al (59) derived age-specific reference limits for children aged 04, 410, and 1016 y. There was a continuous decline in serum TfR values from birth to late adolescence; the 2.597.5% reference limits were 1.33.0 mg/L at 410 y, 1.12.7 mg/L at 1016 y, and 0.9 2.3 mg/L at >16 y. In our sample of children aged 614 y, age had no significant effect on TfR, but concentrations in boys were slightly but significantly higher than in girls, both in Côte dIvoire and in Morocco (Table 5).
In adults, ZnPP has a high sensitivity in diagnosing iron deficiency (21, 48, 6063). In infants and children, ZnPP may also be a sensitive test for detecting iron deficiency (20, 6466). Hershko et al (64) found that ZnPP correlated well with both mean cell volume and hemoglobin in children with IDA. In detecting IDA in infants, Serdar et al (65) found that erythrocyte protoporphyrin (EP) was more sensitive than was SF but it was less specific; sensitivity and specificity were 94% and 72% for EP and 85% and 94% for SF. Siegel and LaGrone (20) reported that the sensitivity of ZnPP using a cutoff of 50 µmol/mol heme was 81% in identifying young children with iron deficiency (defined as response to iron), and its positive predictive value was 72%. In children aged 15 y, Mei et al (66) reported that EP was a sensitive test for detecting iron deficiency. However, the specificity of ZnPP in identifying iron deficiency may be limited, because ZnPP can be increased by lead poisoning, anemia of chronic disease, chronic infections and inflammation, hemolytic anemias, or hemoglobinopathies (63, 6770). The effect of malaria on ZnPP in children is equivocal (6, 7, 10, 71). In African preschool children with a high prevalence of malaria, EP was more strongly correlated with hemoglobin than was either SF or TfR (6).
Direct comparisons between studies of ZnPP are difficult because of interassay variation. If ZnPP is directly measured in whole blood with a hematofluorometer, interfering substances in plasma produced by acute inflammation and hemolysis can increase ZnPP concentrations 34-fold in the absence of iron deficiency (21). These interfering substances can be removed by washing the erythrocytes, which markedly improves assay specificity (21, 72); however, this procedure is not always done because it is time-consuming (6). Incomplete oxygenation of hemoglobin may produce falsely low ZnPP values because of a shift in hemoglobin absorption (63). A reagent that converts hemoglobin to cyanomethemoglobin is available, but washing the cells to remove interfering substances oxygenates hemoglobin and eliminates the need for the reagent (24). Another problem with ZnPP is that results can be expressed as a concentration (free erythrocyte protoporphyrin, EP, or ZnPP, and these are not interchangeable) or as the ZnPP-heme molar ratio; the latter is recommended (40). Several cut-offs have been proposed for ZnPP to define iron deficiency (21, 24, 62, 68, 73, 74). Using a hematofluorometer on washed erythrocytes, Hastka et al (21) recommended a cutoff of >40 µmol/mol heme on the basis of studies in 130 healthy adults. In contrast, other authors, using unwashed erythrocytes, have proposed a cutoff of >80 µmol ZnPP/mol heme to indicate iron deficiency (24, 73). On the basis of the 97.5th percentile for ZnPP concentrations in US children, Soldin et al (74) suggested a cutoff of 55.2 µmol/mol heme in 59-y-old boys and girls and cutoffs of 61.6 and 58.0 µmol/mol heme in 1017-y-old females and males, respectively. Earlier studies of US children found no significant age or sex differences in ZnPP until about age 14 y (11, 75). Similarly, in our sample of school-age children, there were no sex or age differences in ZnPP in iron-sufficient, nonanemic subjects (Table 5).
In the United States, Allen et al (30) reported that healthy black adults had mean TfR concentrations that were 9% higher than those in whites, Asians, or Hispanics. In the present study, after adjustment for age and sex, we found significantly higher TfR concentrations in nonanemic, iron-sufficient black children in Côte dIvoire than in white children from Morocco. The median TfR was 18% higher in Ivorian children (Table 5). Similarly, median ZnPP was significantly higher (by 30%) in nonanemic, iron-sufficient Ivorian children than in Moroccan children. Diagnostic cutoffs from the ROC curves for TfR and ZnPP for detecting iron deficiency and IDA were also higher in Ivorian children (Tables 3 and 4). Although these data need to be confirmed, they suggest that physiologic concentrations of these indicators are race-specific and may be related to the well-known but unexplained difference in hemoglobin concentrations between blacks and whites (28, 29).
Although studies in adults have suggested that elevations in TfR (25) or in ZnPP (21) can be a definitive indicator of iron deficiency, in the present study there was large overlap in the distributions of these indicators in a comparison of children with IDA with those with normal iron status. This overlap may be explained by a greater variability in the erythroid mass in children than in adults (50) together with the many variables affecting children in developing countries that influence TfR and ZnPP independent of iron status, as detailed above. Because of this overlap, the sensitivity and specificity of TfR and ZnPP in identifying iron deficiency and IDA will be modest, regardless of the diagnostic cutoffs chosen.
ACKNOWLEDGMENTS
MBZ designed the study and wrote the first draft of the manuscript. LM and MBZ analyzed the data. LM, SYH, and RFH assisted in the study design and data interpretation. MBZ, FS-A, SYH, NC, and PA participated in the data collections in the field and in the revision of the manuscript.
None of the authors had any financial or personal interest in any company or organization connected in any way with the research represented in the article.
REFERENCES