Factors associated with anaemia among adolescent boys and girls 10–19 years old in Nepal

Abstract We used data from the 2016 Nepal National Micronutrient Status Survey to evaluate factors associated with anaemia (World Health Organization cut‐points using altitude‐ and smoking‐adjusted haemoglobin [Hb]) among nationally representative samples of adolescents 10–19 years. Hb, biomarkers of micronutrients, infection and inflammation were assessed from venous blood. Sociodemographic and household characteristics, dietary diversity, pica and recent morbidity were ascertained by interview. We explored bivariate relationships between candidate predictors and anaemia among boys (N = 967) and girls (N = 1,680). Candidate predictors with P < 0.05 in bivariate analyses were included in sex‐specific multivariable logistic regression models. Anaemia prevalence was 20.6% (95% confidence interval [CI] [17.1, 24.1]) among girls and 10.9% (95% CI [8.2, 13.6]) among boys. Among girls, living in the Mountain and Hill ecological zones relative to the Terai (adjusted odds ratio [AOR] 0.28, 95% CI [0.15, 0.52] and AOR 0.42, 95% CI [0.25, 0.73], respectively), ln ferritin (μg/L) (AOR 0.53, 95% CI [0.42, 0.68]) and ln retinol binding protein (RBP) (μmol/L) (AOR 0.08, 95% CI [0.04, 0.16]) were associated with reduced anaemia odds. Older age (age in years AOR 1.19, 95% CI [1.12, 1.27]) and Janajati ethnicity relative to the Muslim ethnicity (AOR 3.04, 95% CI [1.10, 8.36]) were associated with higher anaemia odds. Among boys, ln RBP [μmol/L] (AOR 0.25, 95% CI [0.10, 0.65]) and having consumed flesh foods (AOR 0.57, 95% CI [0.33, 0.99]) were associated with lower anaemia odds. Open defecation (AOR 2.36, 95% CI [1.15, 4.84]) and ln transferrin receptor [mg/L] (AOR 3.21, 95% CI [1.25, 8.23]) were associated with increased anaemia odds. Anaemia among adolescents might be addressed through effective public health policy and programs targeting micronutrient status, diet and sanitation.

. To our knowledge, only three studies, two of which were subnational, have evaluated anaemia prevalence among adolescent boys in Nepal, which reported prevalence values ranging from 24 to 52% (Baral & Onta, 2009;Chalise et al., 2018;Sinha et al., 2012).
To develop effective, evidence-based public health programming to address anaemia, it is critical to understand its context-specific determinants. Although iron deficiency is the primary cause of anaemia worldwide (Ezzati, Lopez, Rodgers, & Murray, 2004), other biological factors contribute to anaemia including other micronutrient deficiencies, infection and inflammation. Intermediate causes of anaemia, such as dietary intake, influence biological determinants, while socio-economic and other characteristics underlay many of the intermediate causes of anaemia.
Regionally, factors contributing to anaemia vary by population, highlighting the need to better understand context-specific causes.
Among women of reproductive age in rural Bangladesh, thalassemia, underweight and groundwater iron intake were associated with anaemia odds (Merrill et al., 2012). A nationally representative study of women of reproductive age in Cambodia found no factors significantly associated with anaemia in multivariable models (Wieringa et al., 2016). Little is known about context-specific determinants of anaemia among adolescents in Nepal. Existing studies in Nepal have explored bivariate relationships between anaemia and iron status, dietary indicators or sociodemographic characteristics among adolescent girls (Chalise et al., 2018;Kanodia et al., 2016;Limbu et al., 2017;Shah & Gupta, 2002;Tiwari & Seshadri, 2000); however, to our knowledge, none have examined biomarkers of infection, inflammation and micronutrient status. Additionally, only one study has conducted multivariable modelling and included adolescent boys (Chalise et al., 2018).
Nepal conducted the Nepal National Micronutrient Status Survey (NNMSS) in 2016 to inform programmatic decision-making. The NNMSS collected data on micronutrient status and many known potential causes of anaemia (Ministry of Health et al., 2018). We used nationally representative NNMSS data to identify factors associated

Key messages
• Among adolescent girls, younger age (10-14 years vs. 15-19 years), residing in the Mountain or Hill ecological zones relative to the Terai ecological zone, serum ferritin and serum RBP were associated with reduced odds of anaemia, while belonging to the Janajati ethnicity relative to the Muslim ethnicity was associated with increased odds of anaemia.
• Among adolescent boys, serum RBP and having consumed flesh, organ or blood-based foods were associated with reduced odds of anaemia, while open defecation and ln serum transferrin receptor were associated with increased odds of anaemia.
• A combination of effectively implemented strategies might reduce anaemia among adolescents 10-19 years in Nepal by addressing micronutrient status, diet and sanitation.
• Patterning of nonmodifiable factors, such as age, might explain differential success in reducing anaemia and might help inform program planning.
with anaemia among nonpregnant adolescent girls and adolescent boys 10-19 years. Census projections overestimated adolescent girls 10-19 years as a percentage of the total population. As a result, only 1,886 adolescent girls lived in the sampled households in the sampled clusters compared with the 2,160 planned for data collection. Of those, 1,850 were nonpregnant, consented to participate and were interviewed in the selected clusters (98.1%). We excluded girls with missing or invalid values for haemoglobin (Hb; n = 10), blood-based indictors (n = 8 missing; n = 29 with equivocal results on Helicobacter pylori rapid test), anthropometry (n = 117) and questionnaire-based data (n = 4). We additionally excluded n = 2 girls with biologically implausible body mass index (BMI) Z scores, per WHO guidance (de Onis et al., 2007) for a final analytic sample of 1,680 girls (89.1% of sampled girls).
Among those who consented to participate, with respect to major sociodemographic characteristics, girls who were excluded from the analytic sample were on significantly older (17.9 years vs. 13.9 years) and more likely to be married than those who were included (Table S1).
Of the 1,080 adolescent boys planned for data collection, 1,040 lived in the selected clusters (96.3%). Of those, 1,025 consented to participate and were interviewed (98.6%). We excluded boys with missing or invalid values for Hb (n = 2), blood-based indicators (n = 11), anthropometry (n = 39 missing; n = 3 with biologically implausible BMI Z scores) and questionnaire-based indicators (n = 3) for a final analytic sample of 967 boys (94.3% of sampled boys). Among those who consented to participate, boys who were excluded from the analytic sample were significantly older (17.0 years vs. 13.9 years) and more likely to be married than those who were included (Table S1).
The Nepal Health Research Council granted ethical approval for the study. Adolescents aged 10-17 years provided oral assent for interview and biological data collection, and their legal guardians or parents provided signed informed consent. Adolescents aged 18 and older provided signed informed consent.

| Data collection
The field survey team participated in an intensive 12-day training conducted by core survey team members from New ERA and CDC that included classroom instruction, demonstrations, role play and mock interviews. Enumerators had anthropometry standardization exercises on live participants, comparing measurements to experts'. Phlebotomists were given practical examinations on blood draw and field testing samples. Laboratory technicians processed the samples the phlebotomists collected to standardize technique and practice proper sample storage. Trainees who performed poorly during these practical examinations were not retained for the field work. The teams were also deployed for a 3-day pilot to test survey tools and field procedures.

| Anthropometry
Weight with one layer of light clothing was measured to the nearest 100 g using an electronic SECA digital scale. Standing height was measured without shoes to the nearest 0.1 cm using a Shorr-Board. Enumerators validated the calibration of their anthropometry equipment daily.

| Biological specimens
Following standard procedures, trained phlebotomists collected venous blood samples at the time of interview at the household to assess micronutrient, infection and inflammation status. Technicians analysed Hb (HemoCue ® Hb 301 analyser), H. pylori (QuickVue™ H. pylori Test rapid test kit) and malaria (CareStart™ malaria antigen combo rapid test kit for Plasmodium falciparum and Plasmodium vivax) in the households. Phlebotomists validated the calibration of their HemoCue daily using standard reference materials. Laboratory technicians processed blood specimens at a lab station in each cluster before transport to the National Public Health Laboratory, maintaining cold chain. Protocols on quality assurance were adhered to as outlined in the laboratory manual. Plasma and serum samples were stored in −86 C freezers until analysis.
C-reactive protein (CRP), ɑ-1-acid glycoprotein (AGP), serum ferritin, transferrin receptor (sTfR) and retinol binding protein (RBP) were measured using a sandwich enzyme-linked immunosorbent assay (Erhardt, Estes, Pfieffer, Biesalski, & Craft, 2004). For girls only, red blood cell folate was analysed using a microbiological assay (Pfeiffer et al., 2011). All laboratories conducting biological analyses were required to follow quality control procedures and participate and have acceptable performance in CDC's external quality assurance program-VITAL-EQA.

| Sociodemographic, health and other questionnaire data
In household questionnaires, the head of household or another adult respondent provided information about sociodemographic characteristics, and housing, water and sanitation characteristics and food security in enumerator-administered interviews. Household food security was ascertained using a nine-item questionnaire about access to adequate and preferred foods (Coates, Swindale, & Billinsky, 2007).
Adolescents provided information about marital status, schooling, reproductive history (girls only), consumption of foods from 10 food groups (grains, legumes, nuts, dairy, flesh foods, eggs, green leafy vegetables, vitamin A-rich fruits and vegetable, other fruits, and other vegetables) and tea, micronutrient supplement intake, pica, recent morbidity, and receipt of deworming tablets in enumeratoradministered interviews using gender-specific questionnaires.

| Biomarkers of nutrition status
To correct for inflammation's influence on biomarkers of iron status, we regression-adjusted ferritin and sTfR to a pooled country reference using CRP and AGP (ferritin) or AGP only (sTfR; Namaste, Aaron, Varadhan, Peerson, & Suchdev, 2017). Iron deficiency by ferritin was defined as adjusted ferritin <15.0 μg/L (WHO a , 2017). We defined iron-deficiency anaemia as adjusted Hb

| Infection and inflammation
CRP and AGP were included as continuous variables. We included malaria, H. pylori, and fever, diarrhoea and cough during the 2 weeks preceding the survey as binary variables (yes/no).

| Dietary intake
We defined pica as any consumption of clay, earth, termite mounds, ice, uncooked rice or starch during the 7 days before the survey. Food and Agriculture Organization (FAO)'s Minimum Dietary Diversity for Women of Reproductive Age (MDD-W) indicator was classified as intake from five or more of 10 food groups the day preceding the survey (FAO and FHI 360, 2016). We included consumption of flesh, organ or blood-based foods, legumes, green leafy vegetables, vitamin A-rich fruits and vegetables and tea the day preceding the survey as binary variables (yes/no).

| Reproductive and other health variables
We included intake of any micronutrient supplements (multivitamin, vitamin A, iron, folic acid and/or zinc) the week preceding the survey, intake of iron-folic acid (IFA) tablets during the 6 months preceding the survey and receipt of deworming tablets during the 6 months preceding the survey as binary variables (yes/no). For girls, lactation status and giving birth during the 5 years preceding the survey were included as binary variables (yes/no).

| Sociodemographic variables
Age in years was included as a continuous variable. Ethnicity was classified according to the Government of Nepal Central Bureau of Statistics: Brahmin/Chettri, Dalit, Janajati, other Terai ethnicities (including Terai/Madhesi but not including Terai or Madhesi Brahmin or Chettri), Newar and Muslim (Government of Nepal Central Bureau of Statistics, 2014). We categorized marital status as married or cohabitating as married vs. other. Schooling was categorized as never vs. ever attended school. We defined household location according to Nepal administrative classifications for rurality (rural vs. urban) and ecological zone (Mountain, Hill or Terai [Plains]). Using principal components analysis of housing characteristics and assets, we created a household wealth score and divided it into tertiles. We calculated a categorical indicator of household food insecurity (access), in accordance with the Household Food Insecurity Access Scale Indicator Guide, which classifies households into four levels of food insecurity: food secure, mildly food insecure, moderately food insecure and severely food insecure (Coates et al., 2007). We classified households as increasingly food insecure when they responded affirmatively to more severe conditions (e.g. no food to eat of any kind in the household because of lack of resources to get food) or experience of those conditions more frequently. Unimproved water source was defined as any source other than piped water, tube well borehole, protected well or spring, stone tap, rainwater or bottled water (WHO and UNICEF, 2017). We included unimproved water source, open defecation and earth floor as binary variables (yes/no) and household food security as a multilevel categorical variable.

| Statistical methods
We log transformed nonnormally distributed continuous variables.
We assessed bivariate relationships between candidate predictors and anaemia status by sex using Rao-Scott chi square tests for categorical variables and linear contrast tests for continuous variables. Variables with multiple categories (e.g. household food security) were tested as a group. All candidate predictors of anaemia with P < 0.05 in bivariate analyses were included in the sex-specific multivariable logistic regression models. To identity collinearity, we used eigenvalues <0.01 and conditionality index >30.
All analyses accounted for complex sampling design and were analysed using SAS v.9.4 (SAS Institute Inc., Cary, North Carolina). We set statistical significance at two-sided P < 0.05.
Adolescents have increased nutritional demands during periods of rapid growth (Peduzzi, Concato, Kemper, Holford, & Feinstein, 1996) and, in resource constrained contexts, may have low access to nutrient-rich diets (WHO, 2018). Girls are at additional risk for iron deficiency and anaemia due to the onset of menarche (WHO, 2018).
Among adolescent girls, ferritin was inversely associated with anaemia odds. The burden of iron deficiency in adolescent girls was lower in our analyses compared with existing data. Thirty-three percent of girls with anaemia had iron deficiency defined by ferritin in our analyses compared with 52% in the single study among adolescent girls aged 11-19 years in Dharan, Nepal, reporting both iron (ferritin) and anaemia status (Limbu et al., 2017 (Dary & Hurrell, 2006); it is also critical to consider the mix of public health interventions to best reach the population. Programs aimed at promoting nutrient-rich diets and enhanced bioavailability of micronutrients through food preparation and processing could help improve iron and vitamin A status (WHO a , 2017). Tools exist to model intervention scenarios with different interventions and potential impact (Brown, Engle-Stone, Kagin, Rettig, & Vosti, 2015), which may be useful to countries as they review policies and their mix and reach of programs.
Age was associated with anaemia among girls but not boys. Older age in girls was associated with higher anaemia odds in the multivariable model. Findings concerning age and anaemia in Nepal have been equivocal. Kanodia et al. reported higher anaemia prevalence among girls in early adolescence (10-13 years) relative to girls in middle (14--15 years) and late (16-19 years) adolescence in Dharan, Nepal (Kanodia et al., 2016), while three other studies in Dharan, urban Kathmandu and Morang District reported no differences by age (Limbu et al., 2017;Shah & Gupta, 2002;Tiwari & Seshadri, 2000), though these four studies only assessed bivariate relationships between age and anaemia. The single study with multivariable analyses reported 75% higher odds of anaemia (95% CI [1.44, 2.13]) j During the 6 months preceding the survey. k Biomarker was regression-adjusted to a pooled country reference to adjust for inflammation, using CRP and AGP (ferritin) or AGP only (sTfR; Namaste et al., 2017). l Iron deficiency defined as inflammation-adjusted serum ferritin <15.0 μg/L (WHO a , 2017). m Iron deficiency by sTfR defined as inflammation-adjusted serum sTfR >8.3 μg/L (WHO a , 2017). n We defined vitamin A deficiency as RBP <0.64 μmol/L. To find the RBP cut-point equivalent of retinol <0.70 μmol/L (WHO, 1996) among adolescents, we regressed RBP on retinol in an NNMSS subsample of 100 women 15-49 years for whom serum retinol was assessed using HPLC from the same blood draw as RBP. o Folate cutoff based on the risk of megaloblastic anaemia defined as RBC folate <305.0 nmol/L (Institute of Medicine 1998). p Minimum dietary diversity defined as intake from ≥5 of the 10 main food groups (grains, legumes, nuts, dairy, flesh foods, eggs, green leafy vegetables, vitamin A-rich fruits and vegetables, other fruits and other vegetables) the day preceding the survey based on FAO recommendations for minimum dietary diversity for women of reproductive age (MDD-W) (FAO and FHI 360, 2016). q Reported micronutrient supplement intake includes multivitamin, vitamin A, iron tablets or syrup, folic acid and/or zinc tablets consumed the week preceding the survey. among adolescent boys and girls 15-19 years relative to adolescents 10-14 years in nationally representative samples (Chalise et al., 2018). Lower odds of anaemia among younger girls in our study might be explained by shorter reproductive history (WHO a , 2017); however, no data were available on onset of menses. Even though age is nonmodifiable, information about anaemia patterning might be used to target anaemia prevention and control initiatives to older girls in Nepal.
Anaemia was patterned by ecological zone among girls. Girls residing in the Mountain or Hill ecological zones had lower odds of anaemia relative to those residing in the Terai ecological zone.
Similarly, Chalise et al. reported 80% higher odds of anaemia among adolescent boys and girls 10-19 years residing in the Terai zone relative to the Mountain zone in a nationally representative sample (Chalise et al., 2018). Geographic patterning of congenital blood disorders might explain regional differences in anaemia; however, residing in the Mountain or Hill ecological zones was associated with reduced odds of anaemia among nonpregnant women 15-49 years in Nepal after adjustment for blood disorders, suggesting that regional differences in anaemia could be due to other factors (Ford, Paudyal, Pokharel, et al., 2018). Chronic exposure to arsenic via contaminated groundwater might explain regional differences in anaemia. Studies in the Terai have documented groundwater arsenic levels exceeding the upper limit for drinking water per the WHO guidelines (>10 μg/L; WHO b , 2017; Pokhrel, Bhandari, & Viraraghavan, 2009). Arsenic exposure can lead to anaemia through increased erythrocytes hemolysis (Mahmud, Foller, & Lang, 2008) and reduced heme metabolism (Hernandez-Zavala et al., 1999). Future research might explore arsenic exposure and anaemia.
Among girls, the Janajati ethnicity had higher odds of anaemia relative to the Muslim ethnicity. Two studies, one nationally representative and one from Kathmandu, examined anaemia by ethnicity and reported no significant differences (Chalise et al., 2018;Tiwari & Seshadri, 2000); however, classifications of ethnicity varied across studies and are thus not directly comparable to our findings. Ethnic differences might be explained in part by congenital blood disorders, dietary practices or other practices not captured by this survey.
Among boys, anaemia status varied by household sanitation characteristics. In bivariate analyses, a higher percentage of boys with anaemia resided in a house without a toilet facility and in homes with a dirt floor relative to boys without anaemia. Open defecation was associated with more than double the odds of anaemia in the multivariable model; however, dirt floor was no longer significant after adjusting for other variables. We had similar findings among girls for open defecation in the bivariate analyses; however, this indicator was not significant in the multivariable model. Lack of toilet facility and dirt floors can expose household members to faecal matter, worms, protozoa and other parasites (WHO a , 2017), leading to infection. Similarly, a nationally representative study of adolescent boys and girls aged 10-19 years in Nepal found higher odds of anaemia among those who reported walking barefoot relative to those who wore shoes (Chalise et al., 2018). Although the NNMSS collected data on recent morbidity and measured biomarkers of inflammation, we did not have biological data on soil transmitted helminth infection among adolescents. Some evidence suggests that community-level sanitation variables could play a role in health (Headey, Hoddinott, & Park, 2017). One study in Ecuador reported that community-level sanitation coverage was a stronger predictor of child stunting than the household sanitation status (Fuller,  Estimates are unadjusted and adjusted odds ratios and 95% confidence intervals from bivariate and multivariable logistic regression, respectively. All analyses account for weighting and complex sampling design. Anaemia defined as altitude-and smoking-adjusted Hb <11.5 g/dL for girls 10-11 years and altitude-and smoking-adjusted Hb <12.0 g/dL for girls 12-19 years (WHO a , 2017 Biomarker was regression-adjusted to a pooled country reference to adjust for inflammation, using CRP and AGP (Namaste et al., 2017).
k Biomarker was regression-adjusted to a pooled country reference to adjust for inflammation, using CRP and AGP (ferritin) or AGP only (sTfR; Namaste et al., 2017). l Iron deficiency by ferritin defined as inflammation-adjusted serum ferritin <15.0 μg/L (WHO a , 2017). m Iron deficiency by sTfR defined as inflammation-adjusted serum sTfR>8.3 μg/L (WHO a , 2017). n We defined vitamin A deficiency as RBP <0.64 μmol/L. To find the RBP cut-point equivalent of retinol <0.70 μmol/L (WHO, 1996) among adolescents, we regressed RBP on retinol in an NNMSS subsample of 100 women 15-49 years for whom serum retinol was assessed using HPLC from the same blood draw as RBP.
Eisenberg, 2016). To our knowledge, no studies have explored the role of community-level sanitation on anaemia specifically. Future research might explore the role of community-level sanitation on anaemia in Nepal.

| Strengths and limitations
To our knowledge, this analysis is the first to examine a wide range of known potential causes of anaemia in a nationally representative sample of adolescent boys and girls in Nepal. We used data on multiple potential biological causes of anaemia, not often included from largescale, population-based surveys, such as multiple biomarkers of micronutrient status and inflammation. Due to the cross-sectional study design, we were unable to establish causality between candidate predictors and anaemia status; however, our study contributes to the limited evidence base in this understudied population group.
Among adolescents, the NNMSS did not collect data on some micronutrients for which deficiency could lead to anaemia, such B 12 , or on other potentially important biomarkers (WHO a , 2017). RBP is not the WHO-recommended indicator to assess vitamin A status (WHO, 1996). Dietary recall questions were limited in scope, and we used a dietary diversity tool created for women of reproductive age.
There is no internationally accepted tool to measure adolescent diets.
Although interview could have introduced recall and social desirability bias for reported dietary and micronutrient intake and household food insecurity, it was not likely to be differential by anaemia status because adolescents completed the survey questionnaire prior to having Hb assessed. Sample sizes were calculated to estimate nationallevel prevalence of anaemia and iron deficiency; thus, we may have been underpowered to detect risk factors with small effect sizes in multivariable models. Finally, adolescents excluded from the analyses were on average older than those who were included, potentially reducing generalizability of the findings to older adolescents.

| CONCLUSION
More than 1 in 10 boys and 1 in 5 girls had anaemia. Our findings suggest that strategies to improve iron status, vitamin A status, diet and sanitation could potentially reduce anaemia among adolescents in Nepal. Nonmodifiable factors such as age, ethnicity and ecological zone could explain differential success in reducing anaemia and might help provide context to program monitoring and evaluation and inform program strategies.

ACKNOWLEDGMENTS
We thank all who contributed to the successful completion of the NNMSS including the aetiology of anaemia component and who provided feedback on the aetiology of anaemia analyses.

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest. o Minimum dietary diversity defined as intake from ≥5 of the 10 main food groups (grains, legumes, nuts, dairy, flesh foods, eggs, green leafy vegetables, vitamin A-rich fruits and vegetables, other fruits, other vegetables) the day preceding the survey based on FAO recommendations for minimum dietary diversity for women of reproductive age (MDD-W) (FAO and FHI 360, 2016). p Reported micronutrient supplement intake includes multivitamin, vitamin A, iron tablets or syrup, folic acid and/or zinc tablets consumed the week preceding the survey.

T A B L E 4
Bivariate and multivariable logistic regression predicting anaemia among adolescent boys 10-19 years, Nepal National Micronutrient Status Survey, Nepal, 2016 (n = 967) a Estimates are unadjusted and adjusted odds ratios and 95% confidence intervals from bivariate and multivariable logistic regression, respectively. All analyses account for weighting and complex sampling design. We defined anaemia as altitude-and smoking-adjusted Hb <11.5 g/dL for boys 10-11 years, Hb <12.0 g/dL for boys 12-14 years and Hb <13.0 for boys 15-19 years (WHO a , 2017). Candidate predictors were those where P < 0.05 in bivariate analyses. b Biomarker was regression-adjusted to a pooled country reference to adjust for inflammation, using CRP and AGP (Namaste et al., 2017).