Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Is iron deficiency a major cause of the high prevalence of anemia in non-pregnant Cambodian women of… Karakochuk, Crystal D. 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2017_february_karakochuk_crystal.pdf [ 2.4MB ]
Metadata
JSON: 24-1.0340471.json
JSON-LD: 24-1.0340471-ld.json
RDF/XML (Pretty): 24-1.0340471-rdf.xml
RDF/JSON: 24-1.0340471-rdf.json
Turtle: 24-1.0340471-turtle.txt
N-Triples: 24-1.0340471-rdf-ntriples.txt
Original Record: 24-1.0340471-source.json
Full Text
24-1.0340471-fulltext.txt
Citation
24-1.0340471.ris

Full Text

IS IRON DEFICIENCY A MAJOR CAUSE OF THE HIGH PREVALENCE OF ANEMIA IN NON-PREGNANT CAMBODIAN WOMEN OF REPRODUCTIVE AGE? EVIDENCE FROM A CROSS-SECTIONAL SURVEY AND A RANDOMIZED CONTROLLED TRIAL by  Crystal D Karakochuk  B.Sc. Dietetics, The University of British Columbia, 2001 M.Sc. Nutritional Sciences, The University of Toronto, 2010  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Human Nutrition)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  December 2016 © Crystal D Karakochuk, 2016   ii Abstract  Despite a high prevalence of anemia among non-pregnant Cambodian women, recent reports suggest iron deficiency prevalence is low. If true, iron supplementation will not reduce anemia. In Phase I, we investigated factors associated with hemoglobin, ferritin, and soluble transferrin receptor (sTfR) concentrations in 450 women 18-45 years in Prey Veng province. Overall, 54% had a genetic hemoglobin disorder, 29.5% had anemia (hemoglobin <120 g/L), and 2% and 19% had iron deficiency based on ferritin (<15 µg/L) and sTfR (>8.3 mg/L), respectively. The hemoglobin E homozygous genotype was associated with 50% (95% CI: 14%, 96%) and 51% (95% CI: 37%, 66%) higher mean ferritin and sTfR concentrations as compared to normal hemoglobin structure. These findings challenged the diagnostic accuracy of ferritin and sTfR to estimate iron deficiency. In Phase II, we measured the effect of oral iron (Fe) with or without multiple micronutrients (MMN) on hemoglobin concentration as a direct way to determine the extent to which iron deficiency (or other micronutrient deficiencies) was a cause of anemia. A total of 809 non-pregnant women 18-45 years with hemoglobin ≤117 g/L (HemoCue®) were recruited from Kampong Chhnang province to a 2x2 factorial double-blind randomized trial. Women received 12 weeks of daily Fe (60 mg), MMN (14 other micronutrients), Fe+MMN, or placebo capsules. Baseline anemia prevalence was 58% (Sysmex analyzer). Mean (95% CI) hemoglobin at 12 weeks did not differ in the Fe and Fe+MMN groups (121 [120, 122] vs. 123 [122, 124] g/L); both were higher than MMN and placebo (both 116 [115, 117] g/L, P<0.05). Mean (95% CI) increase in hemoglobin was 5.6 (3.8, 7.4) g/L (P<0.001) among women who received Fe (n=383) and 1.1 (-0.7, 2.9) g/L (P=0.24) among women who received MMN (n=382), with no interaction between interventions (P=0.61). At 12 weeks, 19% and 30% of women had a hemoglobin response ≥10 g/L in Fe and Fe+MMN groups, compared to 8% and   iii 5% in MMN and placebo, respectively. Daily iron supplementation for 12 weeks increased hemoglobin in non-pregnant Cambodian women; however, MMN did not confer additional benefit. Only ~25% of our predominantly anemic study population was responsive to iron.      iv Preface This research was the result of an international collaboration between the University of British Columbia, the Cambodia Ministry of Health, the United Nations Children’s Fund (UNICEF), and Helen Keller International. Several additional partners assisted with laboratory analyses: VitMin Laboratory in Willstaett, Germany, The University of Otago in Dunedin, New Zealand, ViennaLab Diagnostics in Vienna, Austria, the Molecular Genetics Laboratory at BC Children’s Hospital in Vancouver, Canada, Agat Laboratories in Burnaby, Canada, and the National Institute of Public Health Laboratory in Phnom Penh, Cambodia.   Research was undertaken in two different provinces in Cambodia (Prey Veng and Kampong Chhnang). Ethics approval for the research was obtained from the University of British Columbia Clinical Research Ethics Board in Vancouver, Canada (H12-00451 and H15-00933) and the National Ethics Committee for Health Research in Phnom Penh, Cambodia (010-NECHR and 110-NECHR). The trial was registered at clinicaltrials.gov (NCT-02481375).   As result of my PhD work, I have published five manuscripts and have one additional manuscript under review:  A version of Chapter 2 has been published. Karakochuk CD, Whitfield KC, Barr SI, Lamers Y, Devlin AM, Vercauteren SM, Kroeun H, Talukder A, McLean J, Green TJ. Genetic hemoglobin disorders rather than iron deficiency are a major predictor of hemoglobin concentration in women of reproductive age in rural Prey Veng, Cambodia. Journal of Nutrition 2015; 145(1): 134-42. CDK and TJG designed the research; CDK drafted the research protocol and TJG   v contributed to the review and editing of the protocol to the final stage; CDK conducted the research and managed the data; TJG provided oversight and input into all aspects of the study; CDK and TJG conducted the statistical analysis of the data; CDK drafted the research manuscript; all authors contributed to the data interpretation and the review and editing of the manuscript to its final stage; CDK and TJG had primary responsibility for the final content; and all authors read and approved the final manuscript.   A version of Chapter 3 has been published. Karakochuk CD, Whitfield KC, Rappaport AI, Barr SI, Vercauteren SM, McLean J, Prak S, Hou K, Talukder A, Devenish R, Green TJ. The homozygous hemoglobin EE genotype and chronic inflammation are associated with high serum ferritin and soluble transferrin receptor concentrations among women in Cambodia. Journal of Nutrition 2015; 145(12): 2765-73. CDK and TJG designed the research; CDK drafted the research protocol and TJG contributed to the revision of the protocol to the final version; CDK conducted the research and managed the data; KCW and KH supervised data collection in the field; AIR assisted with laboratory analyses; RD conducted the capillary hemoglobin electrophoresis; SMV contributed to the interpretive diagnosis of the hemoglobin disorders; TJG provided oversight and input into all aspects of the study; CDK conducted the statistical analysis of the data and drafted the manuscript; all authors contributed to the review and editing of the manuscript to the final version; CDK and TJG had primary responsibility for the final content; and all authors read and approved the final version of the manuscript.  A version of Chapter 4 has been submitted for publication. Karakochuk CD, Barker MK, Whitfield KC, Barr SI, Vercauteren SM, Devlin AM, Hutcheon JA, Houghton LA, Prak S, Hou K, Chai TL,   vi Stormer A, Ly S, Devenish R, Oberkanins C, Puringher H, Harding KB, De-Regil LM, Kraemer K, Green TJ. The effect of oral iron with or without multiple micronutrients on hemoglobin concentration and hemoglobin response among non-pregnant Cambodian women of reproductive age: A 2x2 factorial double-blind randomized supplementation trial. CDK drafted the protocol and SIB, AMD, SMV, KBH, LMD-R, KK, and TJG contributed to revisions; CDK oversaw implementation and data management; KH, SP, and TLC provided operational support; MKB and TLC conducted DNA extractions and hepcidin assays; LAH oversaw micronutrient analyses; RD conducted hemoglobin electrophoresis; CO and HP conducted -globin genetic analyses; CDK conducted statistical analyses and drafted the manuscript; all authors contributed to data interpretation and manuscript revision; CDK and TJG had primary responsibility for content; and all authors read and approved the final manuscript.  A version of Appendix A has been published. Karakochuk CD, Janmohamed A, Whitfield KC, Barr SI, Vercauteren SM, Kroeun H, Talukder A, McLean J, Green TJ. Evaluation of two methods to measure hemoglobin concentration among women with genetic hemoglobin disorders in Cambodia: a method-comparison study. Clinica Chimica Acta 2015; 441: 148-55. CDK and TJG designed the research; CDK drafted the research protocol and TJG and SMV contributed to the revision of the protocol to its final stage; KCW assisted with data collection and supervision in the field; CDK managed the data and conducted the statistical analysis; TJG provided oversight and input into all aspects of the study; CDK drafted the research manuscript; all authors contributed to the data interpretation and the revision of the manuscript to its final stage;   vii CDK and TJG had primary responsibility for the final content; and all authors read and approved the final manuscript.  A version of Appendix B has been published. Karakochuk CD, Whitfield KC, Rappaport AI, Barr SI, Vercauteren SM, McLean J, Hou K, Talukder A, Houghton LA, Bailey KB, Boy E, Green TJ. Comparison of four immunoassays to measure serum ferritin concentrations and iron deficiency prevalence among non-pregnant Cambodian women and Congolese children. Clinical Chemistry and Laboratory Medicine 2017; 55(1): 65-72. CDK and TJG designed the research; CDK drafted the protocol and TJG revised the protocol to its final stage; LAH and KBB assisted with laboratory analyses in New Zealand; CDK conducted the statistical analysis; TJG provided oversight and input into all aspects of the study; CDK drafted the research manuscript; all authors contributed to the data interpretation and the revision of the manuscript to its final stage; CDK and TJG had primary responsibility for the final content; and all authors read and approved the final manuscript.  A version of Appendix C has been published. Karakochuk CD, Murphy HM, Whitfield KC, Barr SI, Vercauteren SM, Talukder A, Porter K, Kroeun H, Eath M, McLean J, Green TJ. Elevated levels of iron in groundwater in Prey Veng province in Cambodia: A possible factor contributing to high iron stores in women. Journal of Water and Health 2015; 13(2): 575-86. CDK and HMM designed the research and drafted the protocol; CDK and ME collected the data and water samples; Agat Laboratories (Canada) analyzed the samples; all authors contributed to data interpretation and manuscript revision; CDK had primary responsibility for content; and all authors read and approved the final manuscript.   viii Table of Contents Abstract .......................................................................................................................................... ii Preface ........................................................................................................................................... iv Table of Contents ....................................................................................................................... viii List of Tables .............................................................................................................................. xiii List of Figures ...............................................................................................................................xv List of Abbreviations ................................................................................................................. xvi Acknowledgements .................................................................................................................. xviii Chapter 1: Introduction, Literature Review, and Objectives and Hypotheses .......................1 1.1 Overview of the Dissertation ............................................................................................. 1 1.2 Introduction ........................................................................................................................ 3 1.3 Literature Review............................................................................................................... 5 1.3.1 Cambodia .................................................................................................................... 5 1.3.1 Hemoglobin................................................................................................................. 7 1.3.1.1 Structure and Function ......................................................................................... 7 1.3.1.2 Measurement ........................................................................................................ 8 1.3.1.3 Low Hemoglobin (Anemia) ............................................................................... 10 1.3.2 Prevalence of Anemia among Women ..................................................................... 11 1.3.3 Consequences of Anemia for Women ...................................................................... 12 1.3.4 Potential Causes of Anemia in Cambodian Women ................................................. 12 1.3.4.1 Iron Deficiency .................................................................................................. 12 1.3.4.2 Other Micronutrient Deficiencies ...................................................................... 15 1.3.4.3 Infection and Inflammation................................................................................ 16   ix 1.3.4.4 Menstruation and Blood Loss ............................................................................ 17 1.3.4.5 Genetic Hemoglobin Disorders.......................................................................... 17 1.3.5 Measurement of Iron Biomarkers to Estimate the Prevalence of Iron Deficiency ... 19 1.3.5.1 Ferritin................................................................................................................ 19 1.3.5.1.1 Structure and Function ................................................................................ 19 1.3.5.1.2 Measurement ............................................................................................... 20 1.3.5.1.3 Factors Associated with Ferritin Concentration ......................................... 22 1.3.5.2 Soluble Transferrin Receptor ............................................................................. 24 1.3.5.2.1 Structure and Function ................................................................................ 24 1.3.5.2.2 Measurement ............................................................................................... 25 1.3.5.2.3 Factors Associated with Soluble Transferrin Receptor Concentration ....... 25 1.3.5.3 Other Biomarkers of Iron Status ........................................................................ 26 1.3.5.4 Measurement of Hemoglobin Response to Iron Therapy .................................. 28 1.3.6 Policies and Guidelines for Anemia Reduction ........................................................ 29 1.4 Rationale and Significance .............................................................................................. 30 1.5 Research Objectives and/or Hypotheses .......................................................................... 31 Chapter 2: Factors Associated with Hemoglobin Concentration in Women of Reproductive Age in Prey Veng ..........................................................................................................................34 2.1 Summary .......................................................................................................................... 34 2.2 Introduction ...................................................................................................................... 35 2.3 Methods............................................................................................................................ 36 2.3.1 Study Design and Participants .................................................................................. 36 2.3.2 Recruitment and Eligibility ....................................................................................... 37   x 2.3.3 Data Collection and Anthropometric Measurements ................................................ 38 2.3.4 Blood Collection, Processing, and Assessment ........................................................ 38 2.3.5 Statistical Analysis .................................................................................................... 40 2.4 Results .............................................................................................................................. 42 2.4.1 Participant Characteristics ........................................................................................ 42 2.4.2 Prevalence of Genetic Hemoglobin Disorders .......................................................... 43 2.4.3 Factors Associated with Hemoglobin Concentration ............................................... 46 2.4.4 Prevalence Estimates of Anemia and Iron Deficiency ............................................. 50 2.5 Conclusions and Discussion ............................................................................................ 54 2.6 Next Steps ........................................................................................................................ 59 Chapter 3: Factors Associated with Ferritin and Soluble Transferrin Receptor Concentrations in Women of Reproductive Age in Prey Veng ...............................................61 3.1 Summary .......................................................................................................................... 61 3.2 Introduction ...................................................................................................................... 62 3.3 Methods............................................................................................................................ 64 3.3.1 Study Design and Participants .................................................................................. 64 3.3.2 Blood Collection, Processing, and Analysis ............................................................. 64 3.3.3 Data Preparation and Statistical Analysis ................................................................. 64 3.4 Results .............................................................................................................................. 67 3.4.1 Participant Characteristics ........................................................................................ 67 3.4.2 Factors Associated with Ferritin and Soluble Transferrin Receptor Concentrations 67 3.4.2.1 Ferritin Concentration ........................................................................................ 70 3.4.2.2 Soluble Transferrin Receptor Concentrations .................................................... 71   xi 3.4.3 Prevalence Estimates of Anemia and Iron Deficiency ............................................. 71 3.4.4 Adjustment of Ferritin Concentration for Inflammation .......................................... 74 3.5 Conclusions and Discussion ............................................................................................ 77 3.6 Next Steps ........................................................................................................................ 82 Chapter 4: The Effect of 12 weeks of Daily Oral Iron with or without Multiple Micronutrients on Hemoglobin Concentration and Hemoglobin Response: A 2x2 Factorial Double-Blind Randomized Controlled Supplementation Trial in Kampong Chhnang ........83 4.1 Summary .......................................................................................................................... 83 4.2 Introduction ...................................................................................................................... 84 4.3 Methods............................................................................................................................ 86 4.3.1 Study Design and Participants .................................................................................. 86 4.3.2 Randomization and Masking .................................................................................... 87 4.3.3 Procedures ................................................................................................................. 88 4.3.4 Statistical Analyses ................................................................................................... 91 4.4 Results .............................................................................................................................. 93 4.4.1 Baseline Characteristics ............................................................................................ 93 4.4.2 Adherence and Adverse Side Effects ...................................................................... 103 4.4.3 Mean Hemoglobin Concentration at 12 Weeks ...................................................... 103 4.4.4 At the Margins 2x2 Factorial Analysis at 12 Weeks .............................................. 105 4.4.5 Ferritin, STfR, Hepcidin, RBP, Vitamin B12 and Folate at 12 Weeks .................... 109 4.4.6 Proportions of Women as Hemoglobin Responders (≥10 g/L) at 12 Weeks .......... 109 4.5 Conclusions and Discussion .......................................................................................... 113 Chapter 5: Conclusions, Discussion, and Future Research Directions .................................121   xii 5.1 Discussion of Key Findings ........................................................................................... 121 5.1.1 High Serum Ferritin Concentrations in Women ..................................................... 121 5.1.2 High Soluble Transferrin Receptor Concentrations in Women .............................. 125 5.1.3 Discrepancy in Iron Biomarkers used to Estimate Iron Deficiency Prevalence ..... 127 5.1.4 Hemoglobin Response to Iron Therapy .................................................................. 129 5.1.5 Multiple Micronutrients .......................................................................................... 133 5.1.6 Is Iron Deficiency a Major Cause of Anemia in Cambodian Women? .................. 136 5.2 Strengths and Limitations .............................................................................................. 137 5.3 Significance and Contribution of the Research ............................................................. 139 5.4 Future Research Directions ............................................................................................ 140 Bibliography ...............................................................................................................................144 Appendices ..................................................................................................................................163 Appendix A: Evaluation of Two Methods to Measure Hemoglobin Concentration .............. 163 Appendix B: Comparison of Immunoassays to Measure Serum Ferritin Concentration ....... 187 Appendix C: Iron in Groundwater in Prey Veng .................................................................... 210    xiii List of Tables Table 2-1 Participant characteristics of 450 Cambodian women 18-45 years in Prey Veng ....... 42 Table 2-2 Frequency of hemoglobin genotypes detected in Cambodian women in Prey Veng ... 44 Table 2-3 Prevalence of anemia, iron deficiency anemia, iron deficiency, inflammation, and hemoglobin genotypes in Cambodian women in Prey Veng by pregnancy status ....................... 46 Table 2-4 Factors associated with hemoglobin concentration in Cambodian women using a multivariable linear regression model ........................................................................................... 48 Table 2-5 Anemia prevalence and hematological and other indicators in the seven most commonly detected hemoglobin genotypes in non-pregnant Cambodian women in Prey Veng . 51 Table 2-6 Prevalence of iron deficiency with and without anemia among non-pregnant Cambodian women with and without genetic hemoglobin disorders ........................................... 53 Table 3-1 Geometric mean ratios for ferritin and soluble transferrin receptor concentrations among Cambodian women in Prey Veng ..................................................................................... 68 Table 3-2 Iron deficiency prevalence among non-pregnant women with and without genetic hemoglobin disorders, and with and without chronic inflammation ............................................ 73 Table 3-3 Geometric mean ratios for ferritin across the stages of inflammation among Cambodian women in Prey Veng ................................................................................................. 74 Table 3-4 Comparison of the correction factors for ferritin from the Thurnham et al. meta-analysis and study-generated correction factors for Cambodian women by inflammation stage. 76 Table 4-1 Capsule formulations .................................................................................................... 88 Table 4-2 Schedule of study visits ................................................................................................ 89 Table 4-3 Baseline characteristics of enrolled Cambodian women 18-45 years in Kampong Chhnang by supplement group ..................................................................................................... 95   xiv Table 4-4 Baseline concentrations of nutrition, inflammation and hematological indicators and prevalence rates of iron deficiency, inflammation and anemia in enrolled Cambodian women in Kampong Chhnang by supplement group ..................................................................................... 99 Table 4-5 Baseline characteristics in enrolled Cambodian women by anemia status ................ 102 Table 4-6 Mean hemoglobin, ferritin, soluble transferrin receptor, hepcidin, and micronutrient concentrations at 12 weeks by supplement group in enrolled Cambodian women in Kampong Chhnang ...................................................................................................................................... 104 Table 4-7 At the margins factorial analysis: adjusted mean increase in hemoglobin concentration at 12 weeks in enrolled Cambodian women in Kampong Chhnang ........................................... 106 Table 4-8 Adjusted mean increase in hemoglobin concentration at 12 weeks for each treatment group compared to placebo in enrolled Cambodian women in Kampong Chhnang .................. 108 Table 4-9 Proportion of enrolled Cambodian women in Kampong Chhnang defined as hemoglobin responders (increase ≥10 g/L at 12 weeks) by supplement group .......................... 112   xv List of Figures Figure 1-1 Provinces of Cambodia ................................................................................................. 7 Figure 1-2 Hemoglobin structure .................................................................................................... 8 Figure 4-1 Flow diagram of trial enrolment ................................................................................. 94 Figure 4-2 Concordance of hemoglobin concentration between HemoCue® and Sysmex methods at baseline...................................................................................................................................... 97     xvi List of Abbreviations AGP  -1 Acid Glycoprotein AI  Adequate Intake ANOVA Analysis of Variance B  Beta Coefficients (Unstandardized) β  Beta Coefficients (Standardized) BMI  Body Mass Index CI  Confidence Interval CMIME Cambodian Ministry of Industry, Mines, and Energy CRP  C-Reactive Protein CS  Constant Spring (Hemoglobin Variant)  CV  Coefficients of Variation DRI  Dietary Reference Intakes EAR  Estimated Average Requirement ELISA  Enzyme-Linked Immunosorbent Assay Fe  Iron (Study Treatment Arm) Fe+MMN Iron and Multiple Micronutrients (Study Treatment Arm) IS  International Standard LSD  Least Significant Differences MCH  Mean Corpuscular Hemoglobin MCV  Mean Corpuscular Volume MMN  Multiple Micronutrients (Study Treatment Arm) PCR  Polymerase Chain Reaction   xvii PLP  Plasma Pyridoxal-5’-Phosphate RBP  Retinol Binding Protein RDA  Recommended Dietary Allowance RDL  Reporting Detection Limit RDW  Red Cell Distribution Width RR  Relative Risk SD  Standard Deviation SEM  Standard Error of the Mean STfR  Soluble Transferrin Receptor UL  Tolerable Upper Intake Level UNICEF United Nations Children’s Fund WHO   World Health Organization    xviii Acknowledgements I sincerely thank my supervisor and mentor, Dr. Timothy J Green. Tim, you have provided me with endless opportunities throughout the past three years, for which I am forever grateful. You are brilliant (and humble) and have taught me so much. Your entourage of past, present, and future students clearly elucidates to your generosity and likeability as a supervisor. Your brilliant research ideas have been the impetus for several of my publications. I have thoroughly enjoyed working with you and learning from you. You are responsible for much of the success (and liver damage) I have encountered throughout my PhD. Despite our distance across time zones and continents, I know we will keep working together for years to come and I am so excited about that!  I am incredibly grateful to my (sometimes official, sometimes unofficial) co-supervisor and long-time mentor, Dr. Susan I Barr. Susan, words cannot express my gratitude for all that you have done for me. I have been in awe of your wisdom, patience, and thoughtfulness, since I was an undergraduate student in your Lifespan course (15 years ago!). Since then, you have supported me with numerous employment references, graduate school applications, graduate student award applications, presentations, abstracts, manuscripts, magazine articles, … and the list still continues to grow. The skills that I have acquired in critical thinking, academic writing, and professionalism, have been transferred from you. I am still so grateful for any opportunity to work with and learn from you. I cannot thank you enough for always making time for me.   I feel so fortunate to have had the support and incredible technical expertise in hematology from Dr. Suzanne M. Vercauteren, my supervisory committee member. Suzanne, thank you for your   xix insightful ideas and contributions to my research protocols and manuscripts. I always look forward to our meetings together, it was simply a joy to work with and learn from you. I am so excited about the possibilities of future research and projects together.        Thank you to all of my academic mentors. Dr. Stanley Zlotkin, I have been inspired by you and your work since I was an undergraduate student in Dietetics (circa 1999). I am so very grateful that you accepted me as a Master’s student and took me under your wing. You have supported me throughout my Master’s degree and beyond – I would not be where I am today without your encouragement and generosity. Dr. Angela Devlin, I am grateful for your collegiality and generosity over the past few years. You’ve shared your time, lab space, advice, and I am so excited to work with you in the future. Dr. Judy McLean, thank you for your endless support and for being such an inspirational leader in the field of international nutrition. Dr. Jennifer Hutcheon, your friendship and your brilliance in statistics helped me throughout my Masters degree (huddled in your West End apartment crunching sample size calculations in Stata) and my PhD degree (… more Stata!). I look forward to ‘nerding out’ with you for years to come. Lastly, Dr. Klaus Kraemer, thank you for hearing out my elevator pitch in Addis Ababa, believing in my research ideas, and inviting me to work on such incredible projects with you over the past few years. I look forward to more collaborations in the future!  Thank you to my colleagues at Helen Keller International: Hou Krouen, Ngik Rem, Many Eath, Sokhoing Ly, Ame Stormer, Keith Porter, and Zaman Talukder. It was such a pleasure to work with you. Robyn Devenish, Christian Oberkanins, Rika Aleliunas, Karimah Naguib, Yvonne   xx Lamers, Lisa Houghton, and Karl Bailey, thank you for your technical advice, collaboration, and assistance with our laboratory analyses.  My fellow lab mates and grad students, your friendship has made this journey an incredible one and I am grateful for each of you: Abeer Aljaadi, Allison Daniel, Amynah Janmohamed, Aviva Rappaport, Becca Mercer, Jen Foley, Kaela Barker, Kristina Michaux, Kyly Whitfield, Philip Chebaya, Sarah Harvey, Theresa Schroder, Tze Lin Chai, Vashti Verbowski, and Zach Daly. Last, but certainly not least, I thank my friends and family who have been there to support me throughout this journey. Mom, Dad, Shell, Kyla, and Miko, I love you all.  Thank you to the organizations and funding agencies for their generous funding throughout my PhD. Thank you for believing in me and believing in my research. I received a Vanier Scholarship and a Michael Smith Foreign Study Supplement from the Canadian Institutes of Health Research, a Doctoral Research Award from the International Development Research Centre, an Ursula Knight Abbott Travel Scholarship, a BC Children’s Hospital Research Institute Travel Grant, and a Career Enhancement Award from the Canadian Child Health Clinician Scientist Training Program. Research funding was provided by the International Development Research Centre and the Department of Foreign Affairs, Trade and Development of Canada, the Micronutrient Initiative, Sight and Life Foundation, and the Canadian Institutes of Health Research.    1 Chapter 1: Introduction, Literature Review, and Objectives and Hypotheses  1.1 Overview of the Dissertation In Chapter 1, I review information pertaining to whether iron deficiency is the primary determinant of anemia in non-pregnant Cambodian women. Following a short description of Cambodia, an overview of hemoglobin and its structure and function are provided, and the common methods used to measure hemoglobin concentration are discussed. This is followed by a description of factors that influence hemoglobin concentration and anemia prevalence, the cut-offs for anemia applied to different populations, life stages, and settings, and the potential causes of anemia in Cambodian women. I then review the common biomarkers used to estimate iron deficiency (i.e., ferritin and soluble transferrin receptor concentration) and other methods used to estimate iron deficiency prevalence. Finally, I examine the current global and national policies for anemia reduction among women of reproductive age.  In Chapters 2 and 3, I describe a cross-sectional survey using baseline data from 450 women of reproductive age in Prey Veng recruited for inclusion in a randomized controlled trial designed to evaluate an improved model of homestead food production. My aim was to assess the prevalence of genetic hemoglobin disorders and determine the factors associated with each of hemoglobin, ferritin and soluble transferrin receptor concentrations. The findings from this survey raised new research questions about the diagnostic accuracy of the iron biomarkers to estimate iron deficiency and the potential contribution of iron in groundwater in Cambodia to the high levels of ferritin concentrations observed among women in Prey Veng.     2 In the next stage of my PhD, I redirected the focus of my research to develop and plan an iron supplementation trial that addressed the outstanding research questions from the cross-sectional survey in Prey Veng: ‘Is iron deficiency a major cause of the high prevalence of anemia among non-pregnant Cambodian women of reproductive age?’ I planned and initiated this trial in a new province, Kampong Chhnang. In Chapter 4, I describe a 2x2 factorial double-blind randomized supplementation trial, which included 809 non-pregnant women of reproductive age. Women were recruited and randomized to one of four interventions (iron with or without 14 other micronutrients, or a placebo) for 12 weeks.   In Chapter 5, I review the key findings of my research conducted in Prey Veng and Kampong Chhnang provinces and elaborate on the comparability of my results to the published literature. I discuss the overall strengths and limitations of the research design in each study. This is followed by a summary of the significance and overall contribution of this research to the current published literature. I conclude with potential future research directions.  In addition, I initiated, implemented, analyzed, and published two method-comparison studies that investigated the methods I had employed in my research to measure hemoglobin and ferritin concentrations. I also conducted an exploratory study to assess groundwater iron concentrations in water samples from 22 households of the same women that participated in the cross-sectional study in Prey Veng. I describe this work in detail in Appendices A to C.    3 1.2 Introduction Anemia is a worldwide public health problem among women of reproductive age (1). Defined as a hemoglobin concentration <120 g/L in women of reproductive age (2), anemia can increase the risk of adverse pregnancy outcomes (3–5), and impair work capacity and productivity of women (6). In Cambodia, approximately 45% of women of reproductive age are anemic (7). Globally, iron deficiency is thought to be the most common cause of anemia (1,8). Iron deficiency anemia can result from inadequate iron intake, impaired absorption, or loss of iron from the body (9). Iron deficiency is often assumed to be the major cause of anemia in Cambodia because the traditional diet consists mainly of rice and is low in iron-rich animal food sources (10). However, there are causes of anemia other than iron deficiency, such as poor nutrition status (e.g., micronutrient deficiencies other than iron) (11–13), inflammation and disease (14–17), blood loss due to menstruation and child birth, as well as genetic hemoglobin disorders (e.g., thalassemia) (18,19).   Since 2009, the World Health Organization (WHO) has recommended intermittent iron and folic acid supplementation (60 mg elemental iron weekly) for menstruating women in areas of anemia prevalence ≥20% (20). In 2011, the Cambodian Ministry of Health recommended weekly iron and folic acid supplementation for all women of reproductive age, a policy consistent with WHO guidelines. However, if iron deficiency is not a major cause of anemia, then at best supplementation is a waste of valuable resources and at worst could cause harm in certain individuals. Women most at risk for harm are those with -thalassemia major or hemoglobin E heterozygous disorders, both of which are found among the Cambodian population, with approximately 0.5% and 33% prevalence, respectively (18,21,22).    4 The gold standard measure of iron deficiency is the assessment of iron stores on a bone marrow aspirate stained with Prussian Blue (2,23); however, this is invasive and painful, and therefore rarely used for the diagnosis of iron deficiency. Ferritin or soluble transferrin receptor (sTfR) concentrations are more commonly used biomarkers reflecting the depletion of body iron stores and tissue iron deficiency, respectively (2). A ferritin concentration <15 μg/L is indicative of depleted iron stores in non-pregnant women of reproductive age, whereas a sTfR concentration >8.3 mg/L is indicative of tissue iron deficiency. However, these biomarkers are influenced by many factors, including inflammation and genetic hemoglobin disorders, both of which are common in Cambodia (24,25). This raises questions about the diagnostic accuracy of these biomarkers to estimate iron deficiency in this population. Further, some organizations, such as the WHO, have pushed for supplementation with other micronutrients in this population, despite little evidence of their efficacy. In 2015 (after this research study was initiated), the Cambodia Ministry of Health placed the weekly iron and folic acid supplementation program on hold because of the evidence showing that ferritin levels were high and thus that iron deficiency was not prevalent. However, because of questions raised about the diagnostic accuracy of ferritin in this population, there remains an urgent need to clarify whether iron or other micronutrient deficiencies are a major cause of anemia in Cambodia. Hence, a more direct way to determine if the anemia that is prevalent in Cambodia occurs as a result of iron deficiency (versus other causes) is to conduct a supplementation trial to measure hemoglobin responsiveness to iron therapy.    5 1.3 Literature Review 1.3.1 Cambodia Cambodia is a country of approximately 13 million people located in Southeast Asia, and is bordered by Vietnam, Laos People’s Democratic Republic, and Thailand (7). Cambodia has had a tumultuous history of political turmoil and unrest. In 1953, after decades of colonization by France, Cambodia became an independent nation. However, the country was overtaken by the extremist regime of the Khmer Rouge from 1975 to 1979, which led to a horrific period of violence and a genocide of over two million Cambodians (26). In 1993, Cambodia was proclaimed a constitutional monarchy, which has resulted in significant progress in economic and political stability (27). Despite this, Cambodia continues to be listed as one of the least developed countries in the world based on the four categories of economy and growth, health, education, and environment (7).  Approximately 80% of Cambodians live in rural farmlands and the remainder live in urban settings (7). Agriculture (predominantly rice production) and small-scale subsistence farming are the major livelihoods in the rural farmlands, and garment factories and tourism are the major sectors of employment in the urban settings (7). The traditional diet in Cambodia consists mainly of white polished rice and is low in iron-rich animal food sources (10); vegetables, fruit and fish are consumed in small portions and are often added to soups and other cooked dishes. The Cambodia Ministry of Health has reported that the diet is primarily plant-based, limited in variety, and is low in energy, fat, and bioavailable sources of micronutrients, particularly iron and vitamin A (28). According to the 2015 Global Hunger Index (based on the four indicators of undernourishment, child stunting, child wasting and child mortality), Cambodia scored 22.6 on   6 the 100-point scale (where zero is the best score [no hunger], and 100 is the worst score [extremely alarming]), indicating a ‘serious’ hunger problem (29). A study by MacDonald et al. in 2014 reported a high prevalence of food insecurity among households (n=900) in Prey Veng province (30): approximately 33%, 37%, and 12% of households were determined to be mildly, moderately, and severely food insecure, respectively, according to the Household Food Insecurity and Access Scale (31).   Cambodia is geographically divided into 24 provinces, including the capital city of Phnom Penh (Figure 1-1). The work described in this dissertation was conducted in two provinces: Prey Veng, a farming province in southeastern Cambodia, and Kampong Chhnang, an industrial province located in central Cambodia.       7  Figure 1-1 Provinces of Cambodia (source: Cambodia demographic and health survey 2014. Phnom Penh, Cambodia, and Rockville, Maryland, USA; 2015)  1.3.1 Hemoglobin  1.3.1.1 Structure and Function Hemoglobin has a quaternary protein structure that contains four globin protein subunits and four iron-containing heme molecules (9). Normal adult hemoglobin (hemoglobin A) is composed of two α-globin proteins, two β-globin proteins, and four heme molecules (18) (Figure 1-2). The α-globin and β-globin proteins are encoded by 4 α-genes and 2 β-genes, respectively (18). Deletions and/or point mutations in these α- and β-genes can result in decreased production of hemoglobin or variant forms of hemoglobin, that are collectively referred to as genetic hemoglobin disorders (18); these are described in more detail in section 1.3.4.5.    8  Figure 1-2 Hemoglobin structure (figure reproduced with permission of themedicalbiochemistrypage, LLC)    The primary function of hemoglobin is to transport oxygen throughout the body (19); however, it has other physiological roles at the cellular level (e.g., erythrocyte metabolism and carbon dioxide transport) (18,32).  1.3.1.2 Measurement  In laboratory settings, hemoglobin can be measured in a sample of blood using an automated hematology analyzer, which uses spectrophotometry to quantify hemoglobin concentrations (33). This method is considered the gold standard in clinical settings and has minimal error due to the automated process, calibration, and quality control checks (34,35). However, these analyzers are expensive, and require trained technicians for regular maintenance and electricity for operation. In the field setting, particularly in large surveys and research studies where blood requires refrigerated transport over long distances, this machine is usually not feasible for use. Portable   9 hemoglobinometers, such as the HemoCue device, have become increasingly popular in the past decade in field settings and large surveys, as they are easy to use, inexpensive, portable, battery-operated, and provide an immediate digital hemoglobin measurement (2). This device is often used in nation-wide surveys to determine anemia prevalence in populations (36). The accuracy and precision of HemoCue to measure hemoglobin in clinical laboratory settings has been confirmed against hematology analyzers (37–40). However, in field settings, the HemoCue device has shown bias and higher variability of hemoglobin measures compared to hematology analyzers (41–44).   A method-comparison of two methods to measure hemoglobin concentration among women with genetic hemoglobin disorders in Cambodia was undertaken to investigate these methodological issues and is reported in full in Appendix A: Evaluation of Two Methods to Measure Hemoglobin Concentration. Two methods were evaluated among individuals with and without hemoglobin disorders using a hemoglobinometer (HemoCue) and an automated hematology analyzer (Sysmex XT-1800i). Bias and concordance between the methods were determined among Cambodian women (18-45 years) in Prey Veng. In summary, bias and concordance appeared similar between methods among women with no hemoglobin disorders (n=195, bias=2.5, c=0.68), women with hemoglobin E variants (n=133, bias=2.5, c=0.78), and women with other hemoglobin variants (n=92, bias=2.7, c=0.73). It was concluded that bias and concordance were similar across groups, suggesting the two methods of hemoglobin measurement were comparable for use in women with and without genetic hemoglobin disorders. The overall bias was 2.6 g/L, resulting in a difference in anemia prevalence of 11.5% (41.0% using HemoCue and 29.5%   10 using Sysmex, P<0.001). Based on the WHO classifications for anemia severity, this changes the prevalence of anemia from a severe to a moderate public health problem in Cambodia. This is important as policies and programs could potentially be impacted by this shift and scarce resources (e.g., funding or staffing) may not be appropriately dispersed to the highest need.  1.3.1.3 Low Hemoglobin (Anemia) Anemia is defined by a hemoglobin concentration below a defined threshold established for a specific age, sex, or life stage. The cut-off values for each age, sex, or life stage were determined by convention as the 5th percentile of hemoglobin concentration based on the normal distribution of a healthy, iron-replete population (45). For the population we studied (non-pregnant women of reproductive age) a hemoglobin concentration <120 g/L is indicative of anemia (2). Four studies provided the reference data used to estimate this cut-off, which were based on hemoglobin concentrations of predominantly European women in the late 1960’s (46). This cut-off was later compared to hemoglobin data from the Second National Health and Nutrition Examination Survey conducted in the United States and was reported to be similar (46).  Several other factors can influence hemoglobin concentration and must be considered when measuring and interpreting values among individuals or when determining population-level hemoglobin cut-offs for anemia. These factors include age, sex, pregnancy, altitude, cigarette smoking, and African ethnicity (47). Individuals in our studies were similar in age and sex, did not live at high altitudes (<1,000 m), did not smoke, and were not of African ethnicity. As such, altitude, cigarette smoking, and ethnicity were not measured or included in the assessment of hemoglobin concentrations in our studies. The hemoglobin concentration of pregnant women,   11 who were included in the Prey Veng study only, was assessed using a hemoglobin cut-off of 110 g/L (rather than 120 g/L for non-pregnant women). Hemoglobin cut-offs are lower for pregnant women because blood volume is increased during the second and third trimesters of pregnancy leading to hemodilution and thus decreased hemoglobin concentration (9,19).   1.3.2 Prevalence of Anemia among Women  In 2011, an estimated 29% (95% CI: 24%, 35%) of non-pregnant women of reproductive age (15-49 years) worldwide were anemic, which translates to approximately 500 million anemic women (48). In Cambodia, the prevalence of anemia among women is even higher. The 2014 Demographic and Health Survey indicated that 45% of women of reproductive age were anemic (hemoglobin <120 g/L for non-pregnant women and <110 g/L for pregnant women) (7). This nationally representative survey was conducted among 11,286 women aged 15-49 years (7). Based on the WHO classification of anemia severity, anemia is a severe public health problem in Cambodian women (2). The prevalence of anemia among women of reproductive age was higher in Prey Veng (47%, n=749) and Kampong Chhnang (53%, n=418), as compared to the national prevalence (45%) (7). Over the past 14 years, the prevalence of anemia has not been reduced despite numerous policies and guidelines focused on anemia reduction, the majority of which include iron supplementation and/or fortification interventions (7). The reasons for this are largely unknown; however, there are many causes of anemia and it is possible that inadequate iron intake is not the primary cause of anemia in Cambodia.    12 1.3.3 Consequences of Anemia for Women  The consequences of anemia are potentially serious for women. Anemic women are at increased risk of adverse pregnancy outcomes, such as intrauterine growth restriction and low birth weight infants (3–5). In a 2016 meta-analysis that investigated the risk of adverse pregnancy outcomes among anemic women in low- and middle-income countries, Rahman et al. showed a significantly higher risk of low birth weight (Relative risk [RR]: 1.31 [95% CI: 1.13, 1.51], n=17 studies) and preterm birth (RR: 1.63 [95% CI: 1.33, 2.01], n=13 studies) among anemic pregnant women compared to non-anemic pregnant women (49). Further, anemia can impair work capacity and productivity of women, due to decreased oxygen delivery to working muscle (6). Decreased work capacity and productivity reduce household income and may impact food security and nutrition at the household level. On a larger scale, decreased work capacity and productivity as a result of anemia can lead to reduced gross national product, a measure of a country’s economic performance (50).   1.3.4 Potential Causes of Anemia in Cambodian Women  The causes of anemia are multifactorial and can include iron or other micronutrient deficiencies (11–13), infection and disease (e.g., hookworm) (14–17), excessive blood loss due to menstruation or blood loss during child birth, as well as genetic hemoglobin disorders (e.g., thalassemia) (18,19).   1.3.4.1 Iron Deficiency Iron is an essential mineral in the human body and is required for oxidative energy metabolism, red blood cell production, and oxygen transport, as well as other important functions (9). Iron   13 deficiency is characterized by the reduction or depletion of iron stores (51). The level of storage iron is most commonly estimated by measuring ferritin concentration in plasma or serum (2,52). In the early stages of iron deficiency, iron stores are depleted (low ferritin) but the supply of iron to the red blood cells is still adequate (normal hemoglobin). Once the iron supply to red blood cells is compromised, the iron deficiency progresses to a stage of iron-deficient erythropoiesis (51). At this stage, iron stores are depleted (low ferritin) and tissue iron deficiency occurs (elevated sTfR). This can progress to the third and final stage of iron deficiency anemia, where iron stores are depleted (low ferritin), the supply to the tissues and red blood cells is compromised (elevated sTfR), and red blood cell production decreases, resulting in anemia (low hemoglobin) (9). At this stage, decreases in hematocrit also may occur (9). Lastly, when the anemia is caused by chronic disease or inflammation (16,51), iron is sequestered in the macrophage (due to the binding of ferroportin) and the iron supply to the red blood cells is compromised (elevated sTfR) but ferritin concentrations may be normal to high as a result of the inflammation (51). This is also known as functional iron deficiency.  Iron deficiency is often assumed to be a major cause of anemia (1,8). The authors of the 2011 WHO Global Prevalence of Anemia Report suggest that approximately 49% (95% CI: 43%, 53%) of anemia in non-pregnant women of reproductive age (15-49 years) is responsive to iron supplementation (53). This estimate is similar to that for women of reproductive age in Southeast Asia: 45% (95% CI: 35%, 53%) (53). The global estimates were based on hemoglobin concentrations from approximately 257 surveys, mostly nationally representative, conducted between 1990-2012 (53). In an earlier meta-analysis, Fernandez-Gaxiola et al. reported that a hemoglobin increase of 8.6 (95% CI: 3.9, 13.4) g/L was observed in a subgroup analysis of only   14 non-pregnant menstruating anemic females (n=352 from two trials) receiving intermittent iron supplements (120 mg elemental iron once weekly) for four to five months (48). Using statistical modeling, the authors of the WHO Global Report applied this hemoglobin increase of 8.6 g/L to the estimated mean hemoglobin concentrations and reported that approximately ~50% of the anemia resolved among women (53). However, the 95% CI around the 8.6 g/L estimate was wide and the analysis included only n=352 adolescents and young women 12-19 years of age from two small trials conducted in Kenya (54) and Bangladesh (55). As only two trials contributed to this subgroup analysis the results should be interpreted with caution. Further, we recognize that these estimates are only based on hemoglobin concentrations and statistical modeling techniques, and do not include important information on inflammation, infection, disease, or genetic hemoglobin disorders. More rigorous research is needed to determine the proportions of women that are truly responsive to iron in Southeast Asia.  Iron deficiency anemia is caused by inadequate dietary iron intake, impaired absorption of iron, or increased loss of iron from the body (9). Iron deficiency impairs hemoglobin function because it is an integral component of the hemoglobin structure (18,19). Impaired absorption and increased loss of iron can also result from infection and disease (9,16,56), such as dengue fever, malaria, hookworm, and parasites, which are prevalent in some areas of Cambodia (56–59). Iron deficiency is often assumed to be a major cause of anemia in Cambodia due to iron-poor diets that consist mainly of rice and are low in iron-rich animal food sources (10). However, naturally-existing iron is also found in groundwater (60). Researchers in other parts of Asia (e.g., Bangladesh) have observed that groundwater iron is positively associated with iron stores in women (61). Elevated levels of iron in groundwater in Cambodia have been previously reported   15 (62); however, it is not known if the groundwater in Prey Veng is high in iron or if it is acidic (which could increase the bioavailability of the iron), and thus it is not known if the groundwater could possibly be contributing to dietary iron intakes in individuals consuming groundwater. There is a need for a comprehensive assessment of iron status in Cambodia, including biochemical and dietary assessments of individuals, as well as environmental assessments of the iron content (and bioavailability) in groundwater.  1.3.4.2 Other Micronutrient Deficiencies  Deficiencies of other micronutrients such as vitamin B12, folate, vitamin A, and riboflavin can also cause anemia (12,63–67). Vitamin B12 and folate metabolism are linked and therefore deficiencies of either of these nutrients can reduce the number of red blood cells produced, leading to macrocytic anemia (characterized by larger than normal red blood cells) (64). Vitamin A supplementation has been shown to improve hematological indicators and enhance the efficacy of iron supplementation (11). Retinol binding protein (RBP, an indicator of vitamin A status) is also influenced by infection and/or inflammation resulting in depressed serum RBP concentrations (68), thus confounding the association between anemia and vitamin A status. A study in Thailand found that suboptimal vitamin A status was a major negative predictor of hemoglobin concentration in school-aged children (69). The authors speculated that the mechanisms of interaction between vitamin A status and iron may be the result of a decrease in the mobilization of iron from stores into the circulation, causing anemia (69). Riboflavin status has also been shown to be associated with anemia (12,66). In a randomized controlled trial in China, anemic pregnant women (n=366) who received iron and folic acid plus riboflavin (60 mg iron, 400 μg folic acid, and 1 mg riboflavin) showed a significantly reduced prevalence of   16 anemia after two months of supplementation, as compared to anemic pregnant women who only received iron and folic acid (12). In addition, vitamin C (70,71) and vitamin B6 (72,73) have been implicated in the development of anemia but this association has not yet been examined in Cambodia, as vitamin C and B6 are rarely measured in nutrition surveys. Organizations, such as the WHO, have pushed for supplementation with other micronutrients, despite the fact that efficacy has not yet been proven. Depending on the context, diets could be low in one or more micronutrients that could contribute to anemia. However, the efficacy of multiple micronutrient supplementation in reducing anemia has not been fully assessed in most settings, including Cambodia, where micronutrient intakes are thought to be low. As such, Cambodia is an appropriate setting in which to assess the efficacy of iron supplementation, with or without multiple micronutrients, to increase hemoglobin concentrations in non-pregnant women.  1.3.4.3 Infection and Inflammation Hookworm, parasites, and malaria are prevalent in some areas of Cambodia (56–59) and can contribute to both iron deficiency and anemia (16,17). Infectious pathogens, metabolic stress, and tissue damage activate the inflammatory response (52). Cytokines are released, stimulating the production of hepcidin, which functions as the main regulator of iron metabolism (19). Hepcidin binds to and degrades ferroportin, a transport protein on the wall of the macrophage, sequestering iron in the macrophage and making it unavailable for erythropoiesis (16,74). This is thought to be a protective mechanism to prevent pathogenic organisms from using iron in circulation (16,17), and results in increased ferritin concentrations (increased storage iron) (75). The released cytokines increase inflammatory biomarkers such as C-reactive protein (CRP) and   17 -1 acid glycoprotein (AGP), two biomarkers that are commonly measured to estimate levels of inflammation (2,76).   1.3.4.4 Menstruation and Blood Loss Among women of reproductive age, menstruation and pregnancy increase daily iron requirements (77–79), which if not met, can contribute to anemia. A study by Harvey et al. among 90 healthy premenopausal British women (18-45 years) found that menstrual iron loss (measured using the ‘gold standard’ alkaline haematin method) was negatively correlated with serum ferritin concentrations, indicating that a high menstrual blood loss was associated with decreased iron stores (P<0.001) (79). Based on their findings, the authors estimated that a 1 mg/day increase in menstrual iron loss (represented as iron loss averaged over a 28-day cycle) was associated with a 6.9 μg/L decrease in serum ferritin concentration, and advocated that identifying and targeting women with high menstrual losses for iron supplementation would be an effective strategy to prevent iron deficiency (79). However, to date, feasible methods have not yet been established to do this; as such, we did not measure menstrual losses among women in our studies.   1.3.4.5 Genetic Hemoglobin Disorders Genetic hemoglobin disorders are autosomal recessive disorders that are categorized into two main groups: hemoglobin variants and thalassemias. Hemoglobin variants result from a nucleotide substitution in one of the globin chains of hemoglobin (e.g., Hemoglobin E) (18). Thalassemias result from a nucleotide deletion in either the - or -globin chains of hemoglobin (i.e., - or -thalassemia) (18). Collectively, hemoglobin disorders can result in a decreased or   18 defective hemoglobin production, leading to an increased risk of anemia and other serious health problems (18,19). Hemoglobin variants and thalassemia can be detected or quantified in blood samples using several methods such as capillary hemoglobin electrophoresis, isoelectric focusing, or high performance liquid chromatography, however, DNA analysis is sometimes required for definitive diagnosis (80).  Genetic hemoglobin disorders can result in diverse clinical outcomes depending on the severity of the disorder. Individuals who inherit only one affected allele are termed heterozygous (also referred to carriers or traits), whereas individuals who inherit two affected alleles are termed homozygous, which usually results in a more severe phenotype (e.g., more severe anemia) (18). These disorders may be asymptomatic, or they can range from to mild-to-severe anemia to death (i.e., Hydrops Fetalis) (18). Certain hemoglobin disorders (e.g., Hemoglobin EE/-thalassemia) are associated with an increased risk of iron overload and oxidative stress (21). Individuals with -thalassemia major require life-long blood transfusions to prevent further morbidity and/or mortality (21). In Cambodia, genetic hemoglobin disorders are common and affect over 50% of the population, most prevalent are the hemoglobin E variant and -thalassemia (24,25,81). Although the prevalence of genetic hemoglobin disorders among individuals in Cambodia has been previously reported, the extent to which each of the common hemoglobin disorders influence hemoglobin, ferritin, and sTfR concentrations among Cambodian women of reproductive age has yet to be investigated. There is a need to investigate the impact of these genetic disorders on prevalence estimates of anemia and iron deficiency.     19 1.3.5 Measurement of Iron Biomarkers to Estimate the Prevalence of Iron Deficiency The gold standard measure of iron deficiency is the assessment of iron stores on a bone marrow aspirate stained with Prussian Blue (2,23); however, this is invasive and painful, and therefore rarely used for the diagnosis of iron deficiency. Instead, biomarkers such as ferritin and sTfR are commonly often used as indicators of iron deficiency (2,52). Other less commonly measured biomarkers of iron status include reticulocyte count and hepcidin concentration. To improve accuracy of diagnosis the WHO recommends the measurement of at least two biomarkers of iron deficiency (2).   1.3.5.1 Ferritin 1.3.5.1.1 Structure and Function The liver is the major storage site of iron and it holds approximately 25% of total body iron stores. Other storage sites include the spleen, bone marrow, and muscle tissue, and in smaller quantities, iron is also found in serum and plasma. Ferritin is a 24-subunit protein that functions to store approximately two-thirds of total body iron (82). In healthy individuals, concentrations of ferritin found in serum or plasma are proportional to the total amount of storage iron in the body and can indicate a deficient, excess or normal iron status (9). A concentration <15 μg/L in non-pregnant women of reproductive age is indicative of iron deficiency (9,67,83). However others have suggested that increasing the cut-off from 15 to 30 μg/L can increase the specificity of ferritin as an indicator of response to iron supplementation (84,85); hence, there is some ambiguity over which cut-off should be used. Serum ferritin concentrations are considered to be the most sensitive and specific tests used for the determination of iron deficiency, in absence of inflammation (51).    20 1.3.5.1.2 Measurement  Historically, immunoradiometric assays (using labeled antibodies) and radioimmunoassays (using labeled ferritin) were the primary methods of ferritin measurement (2). Over the last few decades, automated immunoassay analyzers have been developed, eliminating the need for immunoradiometric methods. In 2004, Erhardt et al. developed a sandwich enzyme-linked immunosorbent assay (s-ELISA), which concurrently measures ferritin, sTfR, CRP, AGP, and RBP concentrations (86). This low-cost method has shown low intra- and inter-assay variability and high sensitivity for ferritin (86); as such, the method has become increasingly popular worldwide. The quantification of ferritin concentration in these immunoassays is based on the detection of specific antibodies (87). However, many challenges in the traceability of this method have been identified, as laboratories differ in terms of which ferritin isoforms are measured (e.g., isoforms found in the liver are different from those in the spleen), the antibodies selected for ferritin detection, and the reference ranges established and utilized (88,89). In 1985, the WHO established the 1st international standard (IS) for ferritin (liver, 80/602) as a reference for methods to be calibrated against in attempt to improve global traceability of methods (90). Since then, the 2nd IS was released in 1993 (spleen, 80/578) (91) and more recently the 3rd IS was released in 1997 (recombinant, 94/572) (92). An evaluation of the 3rd IS by 18 laboratories in nine countries showed adequate stability in accelerated degradation studies and acceptable traceability to the 2nd IS (93). However, calibration to the 3rd IS is not globally mandated or monitored and reference ranges continue to differ across laboratories. Many laboratories are still tracing ferritin methods to the 1st and 2nd IS, despite the fact that these materials ceased production in the mid-1990s and have been since superseded.    21 This lack of standardization of ferritin assays could have implications for the accurate measurement and comparability of ferritin concentrations using different methods, and of greater concern, on iron deficiency prevalence rates in at-risk populations. Due to these reasons, a method-comparison of four methods to measure ferritin concentration was conducted, and is reported in full in Appendix B: Comparison of Immunoassays to Measure Serum Ferritin Concentration. Our aim was to measure serum ferritin concentrations and compare iron deficiency prevalence estimates using four different immunoassays with varying quality controls and traceability. These included three automated immunoassay analyzers (Abbott AxSYM™, Siemen ADVIA Centaur® XP, and Roche Elecsys® 2010) and Erhardt’s s-ELISA in two groups of individuals: Cambodian women of reproductive age and Congolese children. In summary, serum ferritin concentrations were measured using four immunoassays: the s-ELISA and the AxSYM™ analyzer were compared among 420 non-pregnant Cambodian women; the Centaur® XP analyzer, s-ELISA, and AxSYM™ analyzer were compared among a subset of 100 Cambodian women; and the s-ELISA and the Elecsys® 2010 analyzer were compared among 226 Congolese children aged 6-59 months. Median ferritin concentrations (adjusted for inflammation) ranged between 48-91 μg/L among Cambodian women and between 54-55 μg/L among Congolese children. Iron deficiency prevalence ranged from 2-10% among Cambodian women and 5-7% among Congolese children. Bias between methods varied widely (-9 to 45 μg/L) among women, and was 43 μg/L among children (among whom only two methods were compared). Bias was lower, but not eliminated, when ferritin values outside of the s-ELISA measurement range (>250 μg/L) were excluded. It was concluded that the observed differences in ferritin concentrations likely reflect different ferritin isoforms, antibodies, and calibrators used across assays and by different laboratories. However, despite differences in ferritin   22 concentrations, iron deficiency prevalence was relatively similar and low across all methods. It is important to note that ferritin concentrations were relatively high among Cambodian women. The lack of differences in iron deficiency prevalence, despite the differences in ferritin concentrations, may be attributed to this. If ferritin concentrations were lower (closer to the 15 μg/L cut-off for women), we suspect it may have resulted in larger differences in iron deficiency prevalence rates across groups.   1.3.5.1.3 Factors Associated with Ferritin Concentration Apart from the methodological differences in ferritin concentration across select methods, ferritin concentration is also influenced by other factors. Ferritin concentration can vary by age, sex, and life stage (67). Ferritin concentrations tend to be high in infants at birth and continue to rise in the first two to four months of age, after which they tend to drop until approximately one year, after which they tend to increase until adulthood (9,94). As compared to females, males tend to have higher ferritin concentrations during adolescence and adulthood (67). Ferritin concentration tends to decrease during pregnancy as a result of an increased red blood cell production and blood volume expansion (hemodilution), which is greatest in the second and third trimesters of pregnancy (2). Further, iron requirements are increased during pregnancy, which if not met, may lead to decreased iron stores (78,95). Lastly, iron stores are also utilized during pregnancy and fetal growth, which contributes to a reduction in serum ferritin concentration.   Several genetic disorders can result in abnormally low or high ferritin concentrations. For example, iron refractory iron deficiency anemia is an inherited autosomal recessive disorder where a mutation in the TMPRSS6 gene inhibits the signaling pathway that activates hepcidin (a   23 peptide produced in the liver that acts as the main regulator of iron metabolism), resulting in abnormally high concentrations of hepcidin and low-to-normal concentrations of ferritin (96). This genetic mutation is thought to be rare, but has not been comprehensively investigated in low-resource countries (where iron deficiency anemia is most common) because the diagnosis requires complicated analytical procedures (51). Conversely, abnormally high concentrations of serum ferritin (>1000 μg/L) occur in hereditary hemochromatosis, an inherited autosomal recessive disorder where a mutation in the HFE gene can predispose individuals to increased iron absorption and iron overload (97). Individuals with this condition often require regular blood letting (phlebotomy) to reduce body iron levels in order to prevent organ damage (97). Prevalence of this inherited disorder is thought to be more common among individuals with northern European ancestry (~1-2% prevalence) (97,98); of note, it is not suspected to be common among individuals of Southeast Asian descent (99,100).  Lastly, ferritin is an acute-phase protein, and is elevated in the presence of some diseases (e.g., liver disease) (2,101), as well as inflammation or infection (e.g., malaria) (67,76,102). The WHO has proposed multiple ways to handle the confounding factor of inflammation on ferritin concentrations, depending on other available data on inflammation prevalence in the population. In the case where inflammation prevalence is unknown, but suspected to be very high, the cut-off for ferritin to diagnose iron deficiency can be raised (e.g., in women of reproductive age, from 15 to 30 μg/L) in order to account for the inflammation. In the case where inflammation biomarkers were measured in individuals, ferritin values from those individuals with inflammation can be selectively excluded from iron deficiency prevalence calculations. Alternatively, the inflammation biomarkers can be used to estimate correction factors used to adjust ferritin   24 concentrations for levels of inflammation (76,103). CRP and AGP are often measured and commonly used for this purpose. Acute and chronic inflammation are defined as CRP > 5mg/L and AGP >1 g/L, respectively (104). Thurnham et al. have developed correction factors based on the stages of inflammation (incubation [elevated CRP], early [elevated CRP and AGP] and late [elevated AGP] convalescence) using data from 32 studies of apparently healthy individuals in the absence of overt disease in over ten countries (76). These established correction factors (0.77, 0.53, and 0.75 for inflammation in the incubation, early, and late convalescence stages, respectively) are applied to ferritin concentrations to account for the presence of inflammation (76). These correction factors have been published based on apparently healthy individuals in the absence of overt disease; however, the accuracy of these correction factors has not yet been studied among populations with a high prevalence of genetic hemoglobin disorders, some of which have shown to increase ferritin and sTfR concentrations (e.g., hemoglobin E homozygous) (22,25).  1.3.5.2 Soluble Transferrin Receptor 1.3.5.2.1 Structure and Function STfR is a transmembrane glycoprotein that functions to transport circulating iron to red blood cells (101). STfR concentration (measured in serum or plasma) is a measure of tissue iron deficiency and reflects the demand for iron or increased erythropoietic activity (105). STfR is sensitive to the rate of erythropoiesis due to any cause, which limits its ability to specifically indicate iron deficient erythropoiesis (105). As a result, sTfR concentration is only a reliable measure of iron status when iron stores are depleted and there are no other causes of increased   25 erythropoiesis (105). A concentration >8.3 mg/L in non-pregnant women of reproductive age is indicative of tissue iron deficiency (2).   1.3.5.2.2 Measurement  STfR concentration is measured using immunological methods (e.g., s-ELISA), immunonephelometry, and immunoturbidimetry, and in the past decade, is more commonly measured using automated immunoassay analyzers (e.g., Siemen ADVIA Centaur® XP analyzer) (106). As previously mentioned in Section 1.3.5.1.2, Erhardt et al. developed an s-ELISA, which concurrently measures ferritin, CRP, AGP, sTfR, and RBP concentrations (86). Erhardt et al. observed low intra-assay variability for five replicates analyzed in the same sTfR assay (5.6%) and low inter-assay variability across eight days (7.5%) (86). Further, the authors found no bias (mean difference in sTfR concentrations between methods) when the results were compared between the s-ELISA and the Ramco sTfR assay (86). Pfeiffer et al. conducted a method-comparison study that measured the performance of a Roche automated immunoassay and two manual ELISA kits and observed strong positive correlations among all methods (r >0.8); however, sTfR concentrations were approximately 30% lower using the Roche automated immunoassay as compared to the manual ELISA kits (107). The authors concluded that the lack of agreement between the methods was likely due to a lack of standardization across assays (similar to the case of ferritin assays) (107).   1.3.5.2.3 Factors Associated with Soluble Transferrin Receptor Concentration STfR concentration can vary by age, sex, and life stage (9). STfR concentration is thought to be only minimally influenced by pregnancy status (108), although studies have shown conflicting   26 results: one study observed no difference between pregnant and non-pregnant women in the third trimester (109), while another study reported sTfR concentrations were significantly higher among healthy, non-anemic pregnant women as compared to non-pregnant women (110). However, higher sTfR concentrations could be due to increased erythropoiesis, which can occur during pregnancy (111) as a result of normal blood volume expansion (78), and/or as a result of increased renal oxygen consumption due to an increased glomerular filtration rate (112). Male and female adults who smoke ~10 or more cigarettes per day have been shown to have significantly lower sTfR concentrations, as compared to non-smokers (113).  Current evidence indicates that sTfR is not an acute-phase protein; thus, it is not strongly influenced by the presence of inflammation (2,108). Potential causes of increased sTfR concentrations include malarial infection, glucose-6-phosphate dehydrogenase deficiency (an inherited blood enzyme deficiency), sickle cell anemia (another type of genetic hemoglobin disorder, more commonly found in Africa), folate or vitamin B12 deficiency, hemolytic anemia, or other hemolytic conditions (85,101,114). STfR concentrations are decreased in individuals with bone marrow hypoplasia (which can occur as a result of chemotherapy for cancer treatment) (105), and in individuals with aplastic anemia and/or chronic renal failure (9).   1.3.5.3 Other Biomarkers of Iron Status Alternative biomarkers of iron status (e.g., reticulocyte count and hepcidin concentration) can also be used to estimate iron deficiency or to measure the responsiveness to iron therapy. Reticulocytes are immature red blood cells and the reticulocyte count in the peripheral blood is an indication of the ability of the bone marrow to produce red blood cells (115). An increase in   27 reticulocyte count is seen as an early response to iron therapy in individuals with iron deficiency anemia (116). Usually, the reticulocyte count will increase within three to four days and will rise rapidly until approximately seven days after the start of iron therapy (23,115). Therefore, between four to seven days is the opportune time to conduct a reticulocyte count and determine the response to iron therapy. Reticulocyte count can easily be performed given the automated methods (e.g., hematology analyzers) that include the multiple reticulocyte indices and therefore has strong potential as a biomarker to indicate iron therapy response (116).   Hepcidin is a peptide synthesized in the liver (117) that regulates iron metabolism through its participation in three mechanisms: dietary iron absorption in the gut, iron recycling, and the macrophage release of storage iron (74,118). Expression of hepcidin can be induced and inhibited by factors such as iron stores, changes in the rate of erythropoiesis, hypoxia, oxidative stress, and inflammation (119–121). When hepcidin expression is low, its inhibitory effects diminish, iron absorption is increased, and more iron is made available in the body from the diet (via gut absorption) and from the storage pool (121). In some severe forms of thalassemia (e.g., -thalassemia major), hepcidin expression can be abnormally low despite high iron stores, thus contributing to iron overload (122,123). Therefore, it may not be the most sensitive indicator to predict the responsiveness to iron therapy, but it does show promise (124).  Other biomarkers can provide additional information about an individual’s iron status, including transferrin saturation, total iron binding capacity, serum iron, and zinc protoporphyrin. Transferrin is an iron transport protein in serum or plasma that delivers iron to the cells (9). Transferrin concentrations increase in the case of iron deficiency and decrease with protein   28 deficiency (2). Total iron binding capacity is a measure of the binding capacity of iron in the blood. Transferrin saturation (reflected as a percentage and calculated as serum or plasma iron divided by total iron binding capacity) is a proxy measure of iron status; however, because it includes the measure of serum or plasma iron it has the disadvantages of being influenced by inflammation, infection, and diurnal variation (2). Zinc protoporphyrin is reflective of the adequacy of the iron supply to red blood cells in bone marrow (9). When iron stores are depleted, zinc protoporphyrin concentrations increase in the blood and reflect a decrease in circulating iron in bone marrow (9). Zinc protoporphyrin has been identified as a cost-effective screening test to determine iron deficient erythropoiesis and it has the advantage to ferritin that it is not influenced by inflammation. However, these iron biomarkers are rarely measured in low-income settings, such as Cambodia, due to a lack of resources in most local laboratories.  1.3.5.4 Measurement of Hemoglobin Response to Iron Therapy  Measuring hemoglobin concentration before and after iron supplementation to determine the response to iron therapy is a direct and practical method (2,23,125,126) to confirm if iron deficiency anemia exists in a population. Some have suggested an increase in hemoglobin of 10 g/L after two to three months of therapy (83,127) would indicate iron deficiency anemia; however,  individuals with mild anemia may require a much smaller increase in hemoglobin to resolve iron deficiency anemia (84). Response to iron therapy will vary among populations based on several factors: the prevalence of iron deficiency, the prevalence of other micronutrient deficiencies potentially contributing to anemia, and compliance to the iron therapy (23,126). There is a need to rigorously measure and quantify the hemoglobin response to iron supplementation and to determine the true proportion of anemic women that are responsive to   29 iron supplementation in Southeast Asia, as a direct method to estimate the prevalence of iron deficiency anemia in this population in light of the poor diagnostic accuracy of the common iron biomarkers.  1.3.6 Policies and Guidelines for Anemia Reduction  Since 2009, the WHO has recommended intermittent iron and folic acid supplementation (60 mg elemental iron weekly) for menstruating women in areas of anemia prevalence ≥20% (20). In 2016, complementary guidelines were released recommending daily iron and folic acid supplementation (30-60 mg elemental iron daily) for three consecutive months of the year among menstruating women and adolescents in areas of anemia prevalence ≥40% (128). These recommendations are based on the assumption that approximately 50% of anemia is due to iron deficiency (128). In 2011, the Cambodia Ministry of Health adopted the 2009 global WHO guidelines into the National Policy and Guidelines for Micronutrient Supplementation to Prevent and Control Deficiencies in Cambodia (28), and recommended weekly oral iron and folic acid supplementation (60 mg elemental iron and 2.8 mg folic acid) for all non-pregnant women of reproductive age until they become pregnant. In addition, they also recommended daily oral iron and folic acid supplementation (60 mg elemental iron and 400 μg folic acid) for any woman diagnosed as anemic (for 90 days), for pregnant women (for 90 days), and for postpartum women (for 42 days). However, the 2014 Demographic and Health Survey reported that the national prevalence of anemia and iron deficiency (based on ferritin) were 45% and 3% among Cambodian women of reproductive age, respectively (7). Given these recent findings of the surprisingly low prevalence of iron deficiency among women, the Cambodia Ministry of Health has put the weekly iron and folic acid supplementation program on hold because of the lack of   30 evidence of iron deficiency among non-pregnant women and are awaiting more evidence before developing new policy recommendations. There is an urgent need to clarify whether iron or other micronutrient deficiencies are a major cause of anemia in Cambodian women.   Cambodia’s National Nutrition Policy also outlines several other strategies to prevent and control iron deficiency and anemia among women of reproductive age, such as increasing dietary diversity to include more iron-rich sources of food and decreasing the consumption of iron absorption inhibitors (e.g., tea) (7). Recently, the Ministry of Planning and Ministry of Health in Cambodia declared the mandatory fortification of all fish and soy sauce with iron, with the aim to reduce the population-level prevalence of iron deficiency and anemia (129). Lastly, there are multiple other interventions that are promoted in Cambodia by other organizations, including the Lucky Iron Fish®, which is an iron ingot (made from old car parts) that is placed into the household’s cooking pot and slowly releases iron with the ultimate goal to increase the iron content of the cooked food (130). However, these iron supplementation and fortification initiatives will only be effective in anemia reduction if the anemia is caused by iron deficiency.   1.4 Rationale and Significance   In this research, we will determine the prevalence of genetic hemoglobin disorders among women of reproductive age in Cambodia, and determine the impact of the common genetic hemoglobin disorders on each of hemoglobin, ferritin, and sTfR concentrations. We will then comprehensively determine the factors associated with each of hemoglobin, ferritin, and sTfR concentrations among women. Further, we will measure the extent to which iron supplementation with or without multiple micronutrients can increase hemoglobin   31 concentrations, and thus, the extent to which it can reduce anemia among non-pregnant Cambodian women of reproductive age. The rigorous study design and the comprehensive assessment of factors related to hemoglobin concentration and iron status will allow us to determine if iron deficiency is indeed a major cause of anemia in this population. This research will provide the evidence to help inform appropriate strategies to prevent and treat anemia, particularly to reduce the risk of anemia and the burden of disease among women in Cambodia. Given the high prevalence of anemia, genetic hemoglobin disorders, and chronic inflammation throughout Southeast Asia, this research has potential to influence policy and programming in multiple countries in the region that face similar challenges in the diagnosis, treatment and prevention of anemia.   1.5 Research Objectives and/or Hypotheses i. Relationships between genetic hemoglobin disorders, hemoglobin concentration, and biomarkers of iron deficiency. Although the prevalence of genetic hemoglobin disorders among individuals in Cambodia has been previously reported, the extent to which each of the common hemoglobin disorders influence hemoglobin concentrations among Cambodian women of reproductive age has yet to be investigated. Our aims were to genotype women for genetic hemoglobin disorders, to investigate the factors associated with hemoglobin concentration, and to investigate the relationships between genetic hemoglobin disorders and biomarkers of iron deficiency (serum ferritin and sTfR) among women of reproductive age (18-45 years) in Prey Veng. These objectives will be addressed in Chapters 2 and 3. Hypotheses were not formulated for these objectives, as   32 our aims were to simply examine and report descriptive data from the Prey Veng survey, rather than to test one or more specific theories.   ii. Inflammation correction factors for ferritin concentrations. The published meta-analysis by Thurnham et al. (76) reported inflammation correction factors for ferritin based only on data from apparently healthy women in the absence of overt disease. We questioned if these correction factors would be appropriate for use in the current Cambodian population, which has a high prevalence of genetic hemoglobin disorders and inflammation (where as a result, ferritin concentrations may be elevated). As such, our next aim was to compare study-generated correction factors for ferritin with those from the published meta-analysis (76). This objective will be addressed in Chapter 3. Hypothesis: Study-generated inflammation correction factors for ferritin will be significantly different than the correction factors from a published meta-analysis.   iii. The effect of daily oral iron with or without multiple micronutrients on hemoglobin concentration. Lastly, we queried whether daily oral iron with or without other micronutrients would increase hemoglobin concentrations in non-pregnant Cambodian women screened as having anemia. Our final aim was to determine the effect of 12 weeks of daily oral iron with or without multiple micronutrients, as compared to a placebo group, on hemoglobin concentration and hemoglobin response among non-pregnant women of reproductive age (18-45 years) screened as having anemia in Kampong Chhnang. This objective will be addressed in Chapter 4. Hypotheses: The iron-alone supplemented group will have a significantly higher mean hemoglobin concentration than   33 the placebo group after 12 weeks. The multiple micronutrients with iron group will have a significantly higher mean hemoglobin concentration than the iron-alone supplemented group, indicating that the addition of other micronutrients conferred a benefit in reducing anemia.   34 Chapter 2: Factors Associated with Hemoglobin Concentration in Women of Reproductive Age in Prey Veng  Acknowledgement: A version of this chapter has been published. Karakochuk CD, Whitfield KC, Barr SI, Lamers Y, Devlin AM, Vercauteren SM, Kroeun H, Talukder A, McLean J, Green TJ. Genetic hemoglobin disorders rather than iron deficiency are a major predictor of hemoglobin concentration in women of reproductive age in rural Prey Veng, Cambodia. Journal of Nutrition 2015; 145(1): 134-42.  2.1 Summary Anemia is common in Cambodian women. Potential causes include micronutrient deficiencies, genetic hemoglobin disorders, inflammation and disease. The aim was to investigate factors associated with hemoglobin concentration in rural Cambodian women 18-45 years and to investigate the relationships between hemoglobin disorders and other iron biomarkers. Blood samples were obtained from 450 women. A complete blood count was conducted and serum and plasma were analyzed for ferritin, sTfR, folate, vitamin B12, RBP, CRP, and AGP. Hemoglobin electrophoresis and multiplex polymerase chain reaction were used to determine the prevalence and type of genetic hemoglobin disorders. Overall, 54% of women had a genetic hemoglobin disorder, which included 25 different genotypes (most commonly: hemoglobin E variants and 3.7-thalassemia). Of the 420 non-pregnant women, 30% had anemia (hemoglobin <120 g/L), 2% had depleted iron stores (ferritin <15 µg/L), 19% had tissue iron deficiency (sTfR >8.3 mg/L), <3% had folate deficiency (<6.8 nmol/L) and 1% had vitamin B12 deficiency (<150 ρmol/L).   35 Prevalence of iron deficiency anemia was 14.2% and 1.5% in those with and without hemoglobin disorders, respectively. There was no biochemical evidence of vitamin A deficiency (RBP <0.7 µmol/L). Acute and chronic inflammation were prevalent among 8.3% (CRP >5 mg/L) and 25.5% (AGP >1 g/L) of non-pregnant women, respectively. Using an adjusted linear regression model, the strongest predictors of hemoglobin concentration were hemoglobin E homozygous disorder and pregnancy status. Other significant predictors were two other heterozygous traits (hemoglobin E and Constant Spring [CS]), parity, RBP, log ferritin and vitamin B12. Multiple biomarkers for anemia and iron deficiency were significantly influenced by the presence of hemoglobin disorders, hence, reducing their diagnostic sensitivity. Further investigation of the unexpectedly low prevalence of iron deficiency anemia in Cambodian women is warranted.  2.2 Introduction The accurate diagnosis of iron deficiency anemia in the developing world is challenging. Diagnostic criteria for women of reproductive age include hemoglobin concentration <120 g/L in conjunction with serum ferritin <15 µg/L or sTfR >8.3 mg/L (2,9). However, serum ferritin becomes elevated in the presence of inflammation limiting its diagnostic sensitivity (2,102). The WHO recommends to correct ferritin values for inflammation using inflammation biomarkers (commonly CRP and AGP) (2,76). Alternatively, sTfR, an indicator of tissue iron deficiency, is less influenced by inflammation (108) and has been suggested as a more sensitive indicator of iron deficiency in populations with high prevalence of inflammation and/or disease (131).   However, not all anemia is caused by iron deficiency and when iron deficiency is not a major cause, iron interventions such as fortification and/or supplementation are not effective to reduce   36 or prevent anemia. In Cambodia, genetic hemoglobin disorders affect over 50% of the total population, which can result in a decreased or defective hemoglobin production, leading to an increased risk of anemia and other serious health problems (18,19). Both serum ferritin and sTfR concentrations have been reported to be elevated in individuals with genetic hemoglobin disorders (25,69,132,133). This limits the diagnostic sensitivity of serum ferritin and sTfR to determine iron deficiency in individuals with hemoglobin disorders. However, the impact that genetic hemoglobin disorders have on anemia in rural Cambodian women is not known. Further, it is not known if other micronutrient deficiencies (e.g., vitamin B12 or folate) are associated with anemia in Cambodian women of reproductive age. This has important implications for the design and implementation of effective nutrition interventions in this population.  The aims were to comprehensively genotype women for genetic hemoglobin disorders, to investigate the factors associated with hemoglobin concentration, and to investigate the relationships between genetic hemoglobin disorders and the common biomarkers of iron deficiency (serum ferritin and sTfR) in women of reproductive age in Cambodia.  2.3 Methods 2.3.1 Study Design and Participants This cross-sectional study was conducted in women 18-45 years in rural Prey Veng province. Socio-demographic and health data, anthropometric measurements and a venous blood sample were collected from women in July 2012. Ethical approval for the study was granted by the Clinical Research Ethics Board at the University of British Columbia (Canada) and the National   37 Ethics Committee for Health Research (Cambodia). Written informed consent was obtained from all women before enrollment in the study.   2.3.2 Recruitment and Eligibility This cross-sectional analysis used baseline data that were obtained from 450 women in four districts (Mesang, Kamchay Mear, Svay Anthor and Bar Phnom) of Prey Veng province for inclusion in a larger randomized controlled trial. The data were collected prior to the implementation of a trial designed to evaluate an improved model of homestead food production and aquaculture. Women were recruited using a two-stage stratified cluster design. In consultation with provincial government officials and the Ministry of Planning, a list of all villages in the province was obtained. Villages (n=6) were systematically excluded if they were participating in other development programs and thus receiving products or interventions that could have biased outcomes in the larger trial. In the first stage, 90 villages were sequentially selected from a randomly ordered list of all eligible villages in the four districts. Within each selected village, households were sequentially selected from a randomly ordered list of all eligible households and visited to determine if they met the eligibility criteria. Selection from the list continued until ten eligible households were recruited. To be eligible for inclusion the woman had to be between 18-45 years, have at least one child <5 years, and live in farming households with some access to land for agriculture or aquaculture activities. Lastly, 450 of the 900 households were randomly selected, from which women provided blood samples.     38 2.3.3 Data Collection and Anthropometric Measurements Data were collected in the local language by trained Cambodian research staff using interviewer-administered questionnaires that included the women’s age, education level, occupation, household size, parity and pregnancy status. Health status indicators included the use of micronutrient or other supplements, medications, and disease-state. All data were obtained by self-report from women at the household. Weight and height were measured for each woman at the household by trained research staff using standardized techniques (134) and calibrated equipment. Duplicate measurements were taken and the average value of the two measurements was used. Body mass index (BMI) was calculated based on the weight and height measurements: weight (kg) divided by height (m2).    2.3.4 Blood Collection, Processing, and Assessment A 3-hour fasting venous blood sample was collected in the morning by trained phlebotomists at health centers in Prey Veng province. Blood was collected in two evacuated 3.5 mL tubes (Becton Dickinson, Franklin Lakes, NJ, USA), one of which contained an anticoagulant. Samples were placed on ice and transported daily to the laboratory in Phnom Penh for processing. After processing, serum samples were aliquoted into multiple vials and stored at -70 oC until shipment to the appropriate laboratories for analysis. A complete blood count was performed using an automated hematology analyzer (Sysmex XT-1800i, Sysmex Corp, Kobe, Japan) at the laboratory in Cambodia. Serum was analyzed for ferritin, sTfR, RBP, AGP and CRP using an s-ELISA (86) at the Erhardt laboratory in Willstaett, Germany. Serum folate was analyzed using a 96-well plate microbiological assay (135) and serum B12 was analyzed using an auto-analyzer with appropriate controls (Abbott AxSym, Chicago, Illinois). Plasma pyridoxal-5’-  39 phosphate (PLP) (as an indicator of vitamin B6 deficiency) was measured in a random subset of n=99 women using high-performance liquid chromatography according to the method by Ubbink et al. (136) with modifications, at the University of British Columbia.   Genetic hemoglobin disorders were identified using methods of hemoglobin electrophoresis and polymerase chain reaction (PCR) (80). Capillary hemoglobin electrophoresis was conducted using a Sebia MINICAP analyzer (Sebia, Lisses, France) by a trained external consultant in Cambodia. This automated technique quantifies the different types of hemoglobin in blood for interpretive diagnosis. It can detect normal hemoglobin (hemoglobin A, A2, and F), hemoglobin variants (hemoglobin E, H and CS) and -thalassemia. The Sebia analyzer has an advantage over other methodology as it can distinguish between hemoglobin A2 and hemoglobin E, both of which are common in Cambodia (137). Genomic DNA was extracted from buffy coat using the QiaAmp Blood DNA kit (Qiagen Ltd., Hilden, Germany) and a multiplex PCR assay (138) was used to detect heterozygosity and homozygosity of the seven most common -globin gene deletions in cases of -thalassemia (SEA, 20.5, MED, FIL, THAI, 3.7, and 4.2-thalassemia) at the Molecular Genetics Laboratory at BC Children’s Hospital in Canada.    Anemia was defined as hemoglobin <120 g/L for non-pregnant women and <110 g/L for pregnant women (2). Microcytic anemia was defined as hemoglobin <120 g/L and mean corpuscular volume (MCV) <80 fL. Depleted iron stores was defined as ferritin <15 µg/L and tissue iron deficiency was defined as sTfR >8.3 mg/L. Iron deficiency anemia was defined as hemoglobin <120 g/L and either ferritin <15 µg/L or sTfR >8.3 mg/L. Acute and chronic inflammation were defined as CRP >5mg/L and AGP >1 g/L, respectively. The thresholds for   40 defining deficiencies of other micronutrients were as follows: serum vitamin B12 <150 ρmol/L, serum folate <6.8 nmol/L, vitamin A as indicated by serum RBP <0.7 µmol/L and vitamin B6 as indicated by plasma PLP <20 nmol/L.  2.3.5 Statistical Analysis The sample size for this cross-sectional survey was estimated using multiple regression models and in consultation with a biostatistician.  Using a two-sided P-value of 0.05, 80% power, eight independent variables, and an expected multiple regression coefficient of R2=0.2, the ideal sample size for this cross-sectional study was 367 women. Since blood samples and ethical consent were obtained from 450 women as the baseline survey of the larger trial, data from all 450 women were included in the statistical analysis.  Ferritin and RBP values were corrected for inflammation using AGP and CRP biomarkers in accordance to published methods by Thurnham et al. (68,76).  Based on subclinical inflammation stages of incubation (elevated CRP), early convalescence (elevated CRP and AGP), and late convalescence (elevated AGP) among women, correction factors were applied to ferritin concentrations (0.77, 0.53, 0.75) and RBP concentrations (1.13, 1.24, 1.11), respectively (68,76). Ferritin, sTfR, folate, CRP and AGP were not normally distributed in the population based on Shapiro-Wilks tests of normality (P<0.05) and were therefore natural log transformed before statistical analyses. The K-nearest neighbors imputation method was used (139) to generate data for women with missing vitamin B12 values (n=24) due to incomplete laboratory analyses. This method input the median values of the ten most similar samples in the data set to the samples with missing values. The dataset used for this imputation included seven variables   41 (among all 450 women); hence six variables were used to impute the B12 values: four continuous variables (adjusted ferritin, sTfR, RBP, and hemoglobin concentration) and two binary categorical variables (pregnancy status and whether or not a woman had a genetic hemoglobin disorder). No other values were missing from the dataset.  A predictive model of linear regression analyses was used to measure the association between hemoglobin concentration (continuous variable, g/L) and multiple independent variables. We included variables in the model that either had a bivariate linear relationship (r >0.2) or that were commonly known to be associated with anemia (i.e., inflammation markers AGP and CRP) even if there was a weak bivariate relationship detected in our data. Log ferritin, RBP, log sTfR, vitamin B12, log folate, log AGP, log CRP, age, parity and BMI were analyzed as continuous variables in the regression model. The five most common genetic hemoglobin disorders detected in the women were included in the model as nominal variables (coded as 0,1 according to whether they had or did not have the genetic hemoglobin disorder). Unstandardized beta estimates (B) with 95% CI and standardized beta estimates () were used to describe the magnitude of associations. Standardized beta estimates result from separate analyses on standardized independent variables so that each variable has a variance of 1. T-tests, analysis of variance (ANOVA), and chi-square tests were used to assess baseline characteristics and conduct pairwise comparisons between groups using least significant differences (LSD) to adjust for multiple comparisons where required.  Two-sided P-values less than 0.05 indicated statistical significance. IBM SPSS software v.22 (Armonk, NY, USA) was used to conduct statistical analyses.    42 2.4 Results 2.4.1 Participant Characteristics A total of 420 non-pregnant and 30 pregnant women 18-45 years were recruited (Table 2-1).   Table 2-1 Participant characteristics of 450 Cambodian women 18-45 years in Prey Veng Values are n (%) or mean  SD. AGP, -1 acid glycoprotein; BMI, body mass index; CRP, C-reactive protein, Hb, hemoglobin; RBP, retinol binding protein; sTfR, soluble transferrin receptor. 1 Values were corrected for inflammation using Thurnham et al. correction factors (68,76).  Non-pregnant Pregnant Total, n (%) 420 (93.3) 30 (6.7) Age, years, mean  SD 29.6  6.5 28.5  4.6 Micronutrient Deficiencies, n (%)         Vitamin B12, <150 ρmol/L 4 (1.0) 3 (10.0)       RBP (vitamin A)1, <0.7 µmol/L 0 (0) 0 (0)       Folate, <6.8 nmol/L 11 (2.6) 0 (0)       Ferritin1, <15 µg/L 9 (2.1) 4 (13.3)       STfR, >8.3 mg/L 79 (18.8) 5 (16.7) Inflammation, n (%)         Acute, CRP >5 mg/L 35 (8.3) 5 (16.7)       Chronic, AGP >1 g/L 107 (25.5) 2 (6.7) Parity (number of children ever born), n (%)         1 child ever born  132 (31.4) 18 (60.0)       2-3 children ever born   211 (50.2) 10 (33.3)       4 children ever born  77 (18.3) 2 (6.7) Body Mass Index (BMI), n (%)         Underweight: BMI <18.5 kg/m2 62 (14.8) n/a       Normal: BMI = 18.5-24.9 kg/m2 318 (75.7) n/a       Overweight: BMI = 25-29.9 kg/m2 33 (7.9) n/a       Obese: BMI 30 kg/m2 7 (1.7) n/a Anemia, n (%)         Hb <120 g/L (non-pregnant) and <110 g/L (pregnant) 124 (29.5) 13 (43.3)   43 Mean age, parity and the prevalence of micronutrient deficiencies, inflammation and anemia were reported for all women. BMI was reported for only non-pregnant women. Vitamin B6 deficiency (plasma PLP <20 nmol/L) was detected in two women in a random subset of n=99 women (~2% deficiency). Of the non-pregnant women with anemia, 68.5% had mild anemia (hemoglobin 110-120 g/L), 31.5% had moderate anemia (80-110 g/L), no women had severe anemia (<80 g/L), and 62.9% had microcytic anemia (hemoglobin <120 g/L and MCV <80 fL). Of the pregnant women with anemia, 76.9% had mild anemia (hemoglobin 100-110 g/L), 23.1% had moderate anemia (70-100 g/L), no women had severe anemia (<70 g/L) and 61.5% had microcytic anemia (hemoglobin <110 g/L and MCV <80 fL).  2.4.2 Prevalence of Genetic Hemoglobin Disorders Using capillary hemoglobin electrophoresis and multiplex PCR, we determined the frequency of all hemoglobin types detected in the 450 Cambodian women in the study (Table 2-2). Overall, 54% of women in the study had a genetic hemoglobin disorder. Twenty-six different genotypes were identified including normal hemoglobin, single or co-inherited deletions and/or point mutations. The most common abnormal genotypes included hemoglobin E trait, affecting ~15% of women, and 3.7-thalassemia trait, affecting ~12% of women.       44 Table 2-2 Frequency of hemoglobin genotypes detected in Cambodian women in Prey Veng  n % Normal hemoglobin  207 46.0 Abnormal hemoglobin 243  54.0   Hb E trait 67 14.9   3.7-thalassemia trait 52 11.6   Hb E trait and 3.7-thalassemia trait 35 7.8   Hb E with raised Hb F 22 4.9   Hb CS trait 15 3.3   3.7-thalassemia homozygous 8 1.8   Hb F raised in isolation 6 1.3   +-thalassemia trait2 4 0.9   Hb E homozygous  4 0.9   Hb E trait and 3.7-thalassemia homozygous 3 0.7   Hb E trait and Hb CS trait 3 0.7   +-thalassemia trait2 with raised Hb F 3 0.7   Hb E trait with raised Hb F 3 0.7   SEA-thalassemia trait 3 0.7   Hb E trait and SEA-thalassemia trait 2 0.4   Hb E trait and 4.2-thalassemia trait 2 0.4   Hb CS trait and 3.7-thalassemia trait 2 0.4   Hb E homozygous and 3.7-thalassemia trait 2 0.4   Hb E trait and SEA-thalassemia trait and 3.7-thalassemia trait  1 0.2   Hb E trait and Hb CS trait and 3.7-thalassemia trait 1 0.2   +-thalassemia trait2 and Hb CS trait 1 0.2   Hb Bart and 3.7-thalassemia trait 1 0.2   Hb E homozygous and 3.7-thalassemia homozygous 1 0.2   Hb E homozygous and 4.2-thalassemia trait 1 0.2   Hb E homozygous and SEA-thalassemia trait with raised Hb F 1 0.2 Total n=450. Values are n and %. CS, constant spring; Hb, hemoglobin; SEA, Southeast Asia.    45 The prevalence of anemia, iron deficiency anemia, iron deficiency, inflammation, and the most common hemoglobin genotypes are presented for non-pregnant and pregnant women in Table 2-3. Overall, we found contradictory estimates of iron deficiency prevalence among non-pregnant women using ferritin (2%, n=9) and sTfR concentration (19%, n=79). Women were categorized into groups representing the six most common hemoglobin genotypes, which accounted for 90% (n=407) of all women in the study. The remaining 10% (n=43) of women had rare abnormal hemoglobin genotypes with a sample size of seven or less for each type; therefore, we could not combine these into groups. As such, we did not have adequate statistical power to analyze these rare genotypes independently. ‘Hemoglobin E homozygous’ in Table 2-2 (n=4 women) refers only to women with independent hemoglobin E homozygous disorders. ‘Hemoglobin E homozygous group’ in Table 2-3, Table 2-4, and Table 2-5 (n=31 women) refers to a group of women with any form of hemoglobin E homozygous disorders, including forms co-inherited with other traits.      46 Table 2-3 Prevalence of anemia, iron deficiency anemia, iron deficiency, inflammation, and hemoglobin genotypes in Cambodian women in Prey Veng by pregnancy status Total n=450. Values are n (%). AGP, -1 acid glycoprotein; CRP, C-reactive protein; CS, constant spring; Hb, hemoglobin; sTfR, soluble transferrin receptor. Chi-square or fisher’s exact tests (when expected cell frequencies in the 2x2 table were five or less) were used to compare proportions between non-pregnant and pregnant women. 1 Hemoglobin <110 g/L for pregnant women. 2 Values were corrected for inflammation using Thurnham et al. correction factors (76). 3 Women were categorized into groups representing the six most common hemoglobin genotypes, which accounted for 90% (n=407) of all women in the study. The last category included the remaining 10% (n=43) of women with rare hemoglobin genotypes.   2.4.3 Factors Associated with Hemoglobin Concentration Using a multivariate linear regression model, we determined the predictors of hemoglobin concentration in the 450 Cambodian women (Table 2-4). Variables were added to the model in a  Non-pregnant Pregnant P Total, n (%) 420 (93.3) 30 (6.7) <0.001 Anemia        Hb <120 (non-pregnant) & <110 g/L (pregnant) 124 (29.5) 13 (43.3) 0.1 Iron Deficiency Anemia        Hb <120 g/L & ferritin <15 µg/L 4 (1.0) 1 (3.3)1 0.5     Hb <120 g/L & sTfR >8.3 mg/L 34 (8.1) 3 (10.0)1 0.003 Iron Deficiency        Ferritin2 (depleted iron stores), <15 µg/L 9 (2.1) 4 (13.3) 0.008     sTfR (tissue iron deficiency), >8.3 mg/L 79 (18.8) 5 (16.7) 0.8 Inflammation biomarkers        CRP >5 mg/L 35 (8.3) 5 (16.7) 0.1     AGP >1 g/L 107 (25.5) 2 (6.7) 0.02 Hb genotype3        Normal Hb 195 (46.4) 12 (40.0) 0.5     Hb E trait 56 (13.3) 11 (36.7) 0.001     Hb E homozygous 31 (7.4) 0 (0) 0.3     3.7-thalassemia trait 50 (11.9) 2 (6.7) 0.5     Hb E trait & 3.7-thalassemia trait 35 (8.3) 0 (0) 0.4     Hb CS trait 15 (3.6) 0 (0) 0.6     Any other abnormal Hb genotype 38 (9.0) 5 (16.7) 0.2   47 forward stepwise conditional approach. Current pregnancy status and parity were self-reported by women. Values for ferritin and RBP in the model were corrected for inflammation using Thurnham et al. correction factors (68,76). This was because log AGP and log CRP were not significant in the model when including the unadjusted values for RBP and ferritin. The correlation matrix and variance inflation factors showed no signs of multicollinearity between variables included in the model.   The largest change in R2 occurred with the addition of the hemoglobin E homozygous group variable (13.4%) and the second largest change with pregnancy status (9.7%). Hence, hemoglobin E homozygous disorder and being pregnant were the strongest negative predictors in the model. The hemoglobin E homozygous group (n=31) included women with any hemoglobin E homozygous form, including other co-inherited disorders (3.7-thalassemia trait, SEA-thalassemia trait, etc.). PLP (vitamin B6) was not included in the model as we only collected data from a sub-sample of 99 women and deficiency prevalence was low (2%). Furthermore, a bivariate correlation showed no association between hemoglobin concentration and PLP (Pearson’s r=0.02).   48 Table 2-4 Factors associated with hemoglobin concentration in Cambodian women using a multivariable linear regression model  Unstandardized coefficients  Standardized coefficients  Significance   B (95% CI) SE  P       (Constant) 108.09 (101.45, 114.73) 3.38 - <0.0001 Genetic Hb Disorders           Hb E homozygous group1       Hb E trait       Hb CS trait  -18.24 -4.33 -5.57 (-21.74, -14.73) (-7.06, -1.59) (-10.45, -0.70) 1.78 1.39 2.48 -0.41 -0.12 -0.09 <0.0001 0.002 0.025 Current pregnancy status2 -11.99 (-15.60, -8.39) 1.84 -0.27 <0.0001 RBP3, µmol/L 3.09 (1.77, 4.40) 0.67 0.19 <0.0001 Log ferritin3, µg/L 2.19 (0.86, 3.51) 0.67 0.13 0.001 Vitamin B12, ρmol/L 0.01 (0.00, 0.01) 0.003 0.09 0.026 Parity2, n -0.64 (-1.25, -0.03) 0.31 -0.08 0.039 Total n=450. Values are coefficients (95% CI), standard error (SE), and P-values for significance. Total n=450 women. R2=0.32. Adjusted R2=0.31. Hb, hemoglobin; CS, constant spring; RBP, retinol binding protein. Variables were added to the model in a forward stepwise conditional approach. Variables that were not significant in the model (P<0.05) and therefore not included in the model were: age, BMI, log sTfR, log folate, log CRP, log AGP, and two other genetic hemoglobin disorders: 3.7-thalassemia trait and co-inherited hemoglobin E trait and 3.7-thalassemia trait.  1 The largest change in R2 occurred with the addition of hemoglobin E homozygous group (13.4% change) and pregnancy status (9.7%) variables.  2 Current pregnancy status and parity were self-reported by women.  3 Values were corrected for inflammation using Thurnham et al. correction factors (68,76). This was because log AGP and log CRP were not significant in the model when including the unadjusted values for RBP and ferritin.   49 Other variables negatively associated with hemoglobin concentration in the model included parity (number of children previously born) and two other genetic hemoglobin disorders: hemoglobin E trait and hemoglobin CS trait. Other variables positively associated with hemoglobin concentration in the model included: log ferritin (adjusted for inflammation), RBP, vitamin B12. However, these other variables had less magnitude as compared to hemoglobin E homozygous disorder and pregnancy status variables.   Overall, adjusted R2=0.31 in the model, indicating that ~31% of the variance in the outcomes could be explained by the model. A bivariate correlation matrix confirmed that no two variables in the model were related beyond an acceptable level. In addition, variance inflation factors were calculated and indicated no multicollinearity between variables included in the model. This suggests that there were no interactions among the independent variables, which might compromise the fit of the model. Variables that were not significant in the model and therefore not included in the model were: age, BMI, log sTfR, log folate, log CRP, log AGP, and two other genetic hemoglobin disorders: 3.7-thalassemia trait and co-inherited hemoglobin E trait and 3.7-thalassemia trait.   Ferritin was adjusted for levels of subclinical inflammation using AGP and CRP biomarkers before inclusion in the model. This was because log CRP and log AGP were not significant in the model when including the unadjusted log ferritin variables. Replacing unadjusted ferritin in the model with adjusted ferritin did not change the fit of the model (no change in R2).    50 2.4.4 Prevalence Estimates of Anemia and Iron Deficiency Anemia prevalence and hematological indicators were investigated in the seven most commonly detected hemoglobin types in the non-pregnant women in the study (Table 2-5). Table 2-5 does not include data for rare hemoglobin variants affecting less than seven women in each category; accordingly, data are presented only for the 389 women with normal hemoglobin, hemoglobin E trait, 3.7-thalassemia trait, co-inherited hemoglobin E trait and 3.7-thalassemia trait, hemoglobin E homozygous, hemoglobin CS trait, and lastly, 3.7-thalassemia homozygous.   51 Table 2-5 Anemia prevalence and hematological and other indicators in the seven most commonly detected hemoglobin genotypes in non-pregnant Cambodian women in Prey Veng  Normal Hb Hb E trait 3.7-thalassemia  trait Hb E trait & 3.7-thalassemia trait  Hb E    homozygous group1 Hb CS    trait 3.7-thalassemia homozygous Total, n (%)   195 (46.4) 56 (13.3) 50 (11.9) 35 (8.3) 31 (7.4) 15 (3.6)  7 (1.7) Anemia, n (%)   22 (11.3) 21 (37.5) 15 (30.0)  11 (31.4)  30 (96.8)  3 (20.0) 4 (57.1) Hb, g/L 130  8.9  123  8.6* 124  9.3* 126  8.9* 109  7.3* 121  7.0* 110  13.4* MCV, fL 87.3  4.1 75.6  4.1* 80.9  5.3* 79.2  3.3* 59.2  3.8* 80.1  5.3* 69.3  4.0* MCH, ρg 28.3  1.5    24.5  1.4* 25.5  1.8* 25.5  1.2* 19.9  1.4* 25.3  1.9* 22.1  1.1* RDW, % 13.3  1.1  14.6  1.2* 13.8  1.1* 14.3  1.0* 17.6  2.0* 13.7  1.1* 15.4  1.3* RBC, x10-9/L 4.6  0.3  5.0  0.4 * 4.9  0.4* 4.9  0.4* 5.4  0.4* 4.8  0.3  5.0  0.7*  Ferritin2, µg/L 95.8  56.2 85.1  41.7 79.1  44.0 92.7  56.3 129  90.6* 90.5  37.0 101  47.9 STfR, mg/L 6.4  1.9 7.0  2.3* 6.8  2.0 6.2  1.5 9.5  3.5* 9.5  3.1* 6.5  1.5 RBP2, µmol/L 2.54  0.72 2.57  0.57 2.49  0.74 2.57  0.82 2.56  0.67 2.56  0.69 2.47  0.55 AGP, g/L 0.88  0.27 0.88  0.31 0.91  0.28 0.83  0.20 0.75  0.18 0.90  0.28 0.79  0.24 CRP, mg/L 2.13  4.00 1.62  3.12 1.92  4.26 2.07  3.05 1.32  1.43 1.45  1.61 2.52  5.81 Total n=389. Values are n (%) or mean  SD. AGP, -1 acid glycoprotein; CRP, C-reactive protein; CS, constant spring; Hb, hemoglobin; MCV, mean corpuscular volume; MCH, mean corpuscular hemoglobin; RBP, retinol binding protein; RDW, red cell distribution width; sTfR, soluble transferrin receptor. Data are not shown for rare hemoglobin disorders affecting less than seven women in each category. *Pairwise comparisons show a significant difference (P<0.05, least significant difference adjusted for multiple comparisons) between women with normal hemoglobin and women with a hemoglobin disorder. 1 This group includes all homozygous forms of hemoglobin E including those with other co-inherited traits.  2 Values were corrected for inflammation using Thurnham et al. correction factors (68,76).   52 In pairwise comparisons, statistically significant differences (P<0.05, LSD adjusted for multiple comparisons) were detected between hematological variables in women with normal hemoglobin compared to women with the common genetic hemoglobin disorders. Most remarkable was the significant difference in mean hemoglobin concentration in women in the hemoglobin E homozygous group (mean hemoglobin  SD: 109  7.3 g/L) compared to women with normal hemoglobin (mean hemoglobin  SD: 130  8.9 g/L). Anemia was also most prevalent among the hemoglobin E homozygous group (including those with other co-inherited compound and heterozygous traits, such as hemoglobin CS trait, SEA-thalassemia trait, etc.): 30 of 31 women (97%) had anemia.   MCV and mean corpuscular hemoglobin (MCH) concentrations were significantly lower among women with any of the six most common genetic hemoglobin disorders compared to women with normal hemoglobin. Low MCV and MCH concentrations are defined as microcytic and hypochromic anemia, respectively, and can be a consequence of iron deficiency anemia, genetic hemoglobin disorders, chronic disease, or other factors (12). Red cell distribution width (RDW) was elevated in all six of the most common genetic hemoglobin disorders.   Ferritin concentrations were significantly higher in women in the hemoglobin E homozygous disorders group compared to women with normal hemoglobin. No other genetic hemoglobin disorders statistically increased ferritin concentration. STfR concentrations were significantly elevated in three of the six genetic hemoglobin disorders (hemoglobin E trait, hemoglobin E homozygous group and hemoglobin CS trait). There were no significant differences in mean RBP, AGP and CRP concentrations across different hemoglobin disorders.     53 The prevalence of iron deficiency with and without anemia among women with or without genetic hemoglobin disorders was also compared (Table 2-6). Overall, those women with any form of genetic hemoglobin disorder had an iron deficiency prevalence of ~27% compared to ~11% among women with no genetic hemoglobin disorder. Those women with any form of genetic hemoglobin disorder had an anemia prevalence of ~45% compared to ~11% among women with no genetic hemoglobin disorder. Lastly, those women with any form of genetic hemoglobin disorder had an iron deficiency anemia prevalence of ~14% compared to <2% among women with no genetic hemoglobin disorder. However, when only using ferritin cut-offs for iron deficiency anemia, the prevalence among women with no genetic hemoglobin disorder was only ~1%.  Table 2-6 Prevalence of iron deficiency with and without anemia among non-pregnant Cambodian women with and without genetic hemoglobin disorders  No genetic Hb disorder Any genetic Hb disorder n (%) 195 (46.4) 225 (53.6) Iron deficiency (ferritin <15 µg/L or sTfR >8.3 mg/L)  22 (11.3) 60 (26.7)      Ferritin, <15 µg/L 5 (2.6) 4 (1.8)      STfR, >8.3 mg/L 20 (10.3) 59 (26.2) Anemia        Hb <120 g/L 22 (11.3) 102 (45.3) Iron deficiency anemia (Hb <120 g/L and ferritin <15 µg/L or sTfR > 8.3 mg/L) 3 (1.5) 32 (14.2)      Ferritin <15 µg/L and Hb <120 g/L 2 (1.0) 2 (<1.0)      STfR >8.3 mg/L and Hb <120 g/L 2 (1.0) 32 (14.2) Total n=420. Values are n (%). Hb, hemoglobin; sTfR, soluble transferrin receptor.    54 We also compared the proportion of women with low ferritin concentrations in non-pregnant women without genetic hemoglobin disorders between women with and without chronic inflammation, but did not detect a difference between groups. However, there was a significant difference using sTfR as a biomarker. A significantly higher proportion of women with chronic inflammation (17.5%) had an elevated sTfR concentration (>8.3 mg/L) compared to women without chronic inflammation (7.2%) (P=0.03). Although it is currently thought that sTfR is not influenced by inflammation (2,108), our data suggest otherwise.  2.5 Conclusions and Discussion The current study highlighted the complexity and diverse heterogeneity of hemoglobin disorders in Cambodia. Normal hemoglobin consists of - and -globin chains, encoded by four -genes and two -genes (18). The complexity of these disorders is a result of the many possible combinations of gene variations or deletions that can occur in single, compound and/or co-inherited forms in these six genes. The frequencies of genotypes detected in the women in our study are similar to other reports in Cambodian children (24,25,81). Of all the genetic hemoglobin disorders, hemoglobin E homozygous was the strongest (negative) predictor of hemoglobin concentration in our regression model.   George et al. also found that hemoglobin E homozygous was the most significant predictor of hemoglobin concentration among 2,329 children in Cambodia (25). In contrast to our study, George et al. also found that log sTfR, age, and AGP were significant (negative) predictors of hemoglobin (25). However, the George et al. study was conducted among children 6-59 months in a difference province in Cambodia, as compared to women 18-45 years from Prey Veng in our   55 study. Hence, the differences observed in outcomes may be a result of the different population groups or geographical locations in each of the two studies. The 3.7-thalassemia trait and co-inherited hemoglobin E trait and 3.7-thalassemia trait were not significant predictors and therefore not included in the model, despite our finding that women with these hemoglobin disorders had significantly lower hemoglobin concentrations compared to women with normal hemoglobin (Table 2-3). We speculate this could be due to the small sample size of some of the rare disorders. For similar reasons, we only included the five most common hemoglobin disorders in the regression model. However, there are rare hemoglobin disorders that likely would be strong predictors of hemoglobin. For example, -thalassemia major is known to result in severe anemia (18), however the frequency of this genotype in Cambodia is very low. No women with this genotype were detected in our study, therefore it was not included in our model. However, it still remains important in consideration of its hematological consequences and risk of anemia.  It is not surprising that ferritin, RBP, and vitamin B12 were all positively associated with hemoglobin concentration. This is because of potential role that these biomarkers and micronutrients play in the etiology of anemia (11,13). We suspect these variables may have been stronger predictors in the model if indeed there were higher rates of deficiencies in this population. It is also not surprising that pregnancy and parity were negative predictors of hemoglobin in the model. Blood volume is increased substantially during pregnancy and results in a diluted hemoglobin concentration in women (9,19). Furthermore, in the second and third trimesters of pregnancy, daily iron requirements increase (77,78), which in turn can contribute to the risk of anemia. The demands of multiple, consecutive pregnancies can negatively influence   56 maternal nutrition status (namely anemia), especially in closely spaced pregnancies where there is limited time for recovery and iron stores repletion between births (140).   The most unexpected finding in this study among rural Cambodian women was the extremely low prevalence of iron deficiency anemia (~1% based on ferritin). Comparatively, ferritin concentrations in the Cambodian women in our study were more than double than those seen in a representative sample of Canadian women of similar age (ferritin ~41 µg/L) (141) who presumably are consuming diets with higher iron content. We expected that ferritin concentrations would be low in Cambodian women. This is because of iron-poor diets that consist mainly of rice (low in iron content) and are low in iron-rich animal food sources. Based on self-reported data obtained from the women via questionnaires at the time of blood collection, women were not taking iron or micronutrient supplements, using iron cooking pots, or consuming iron-fortified fish or soy sauce. Twenty-four hour dietary recalls were collected and suggest that women did not have high intakes of dietary iron (142). We speculate that naturally-existing iron in ground water could be contributing to dietary intakes among women in Cambodia, as observed in other parts of Asia (60,61).   The prevalence of other micronutrient deficiencies we assessed (folate, vitamins A, B12 and B6) ranged from very low to non-existent. Unfortunately, we were not equipped to measure any biomarkers of riboflavin status in the women, which has been shown to be associated with anemia (12,66). A recent study measured erythrocyte glutathione reductase activity (a functional indicator of riboflavin status) among women in Cambodia and reported an activity coefficient ≥1.3 among ~92% (n=143/156) of women 20-45 years, suggesting a very high prevalence of   57 suboptimal riboflavin status (143). Although there is some discrepancy in the literature regarding which cut-offs should be used for this functional test (144), we acknowledge the potential problem of suboptimal riboflavin status in Cambodia and conclude that more research is needed to substantiate these findings.   Investigation of the biomarkers showed, most notably, that hemoglobin was significantly lower across all of the common genetic hemoglobin disorders. Ferritin and sTfR concentrations were elevated in individuals with certain genetic hemoglobin disorders. In neighbouring Thailand, Thurlow et al. (69) also demonstrated that ferritin concentrations were significantly elevated in children (n=567) with hemoglobin E genotypes. In our study, ferritin was significantly elevated among the hemoglobin E homozygous group (compared to women with normal hemoglobin) (P<0.05). Similar to our findings, George et al. (25) found that sTfR concentrations were significantly elevated in Cambodian children (n=2,329) with hemoglobin E homozygous, hemoglobin E trait, -thalassemia trait, and the co-inherited form of the latter two. Evidence of elevated sTfR in pregnant Thai women (n=113) with -thalassemia and hemoglobin E trait has also been demonstrated (133). Elevated sTfR in individuals with hemoglobin disorders is thought to be a consequence of increased erythropoiesis stimulated by ineffective erythropoiesis (145). Furthermore, we found conflicting evidence of iron deficiency with elevated ferritin (high iron stores) and elevated sTfR concentrations (tissue iron deficiency) in women. We conclude that serum ferritin and sTfR are poor diagnostic indicators of iron deficiency in populations with hemoglobin disorders and inflammation. It highlights the importance of finding other methods to increase the validity of these biomarkers (potential use of correction factors), or alternatively   58 exploring other biomarkers to assess iron deficiency in populations with high prevalence of hemoglobin disorders and inflammation (potentially reticulocyte count).  The main limitations of the study were typical of cross-sectional surveys in which causation cannot be inferred and findings cannot be extrapolated beyond the geographical area and population group included in the model. We did not measure disease and infection, which could be contributing to anemia. Hookworm and malaria exist in Cambodia and are known to increase inflammation and contribute to anemia (16). However, despite the prevalence of chronic inflammation among women in our study (~25%), AGP (a biomarker of chronic inflammation) was not significant as a predictor of hemoglobin and therefore not included in our regression model.   We conclude that genetic hemoglobin disorders, rather than iron deficiency, are a major predictor of hemoglobin concentration in women 18-45 years in Prey Veng. Genetic hemoglobin disorders were prevalent in over half of the women in our study and have serious potential consequences of morbidity. More work is necessary to collect nation-wide data on the frequencies of hemoglobin disorders and to develop cheaper and simpler methods of screening and management. Future research is also warranted to investigate the causes of high serum ferritin concentrations and the potential risk of iron overload, particularly in women with hemoglobin E homozygous disorders. If iron deficiency is not a problem, then interventions such as iron fortification and supplementation should be reassessed and carefully monitored.    59 2.6 Next Steps The findings from this survey showed that serum ferritin concentrations were surprisingly high among women. We expected that ferritin concentrations would be low in Cambodian women. This is because diets consist mainly of rice and were thought to be low in iron content. Due to these issues, it was queried whether other environmental sources could be influencing iron status in these women. Iron in groundwater had recently been shown to contribute to iron status in women in Bangladesh (61). As such, we queried whether there was naturally existing iron in groundwater in Cambodia that could be contributing to iron stores in these women. An exploratory study was undertaken to determine the levels of iron in groundwater from 22 households in Prey Veng. This study is detailed in full in Appendix C: Iron in Groundwater in Prey Veng: A Possible Factor Contributing to High Iron Stores in Women. In sum, groundwater ranged from 134 to 5,200 µg/L (mean: 1,422 µg/L). Based on a daily consumption of ~3 L of groundwater, this equates to ~0.4-15.6 mg iron per day (mean: 4.3 mg per day), which may be contributing to iron stores in women in Prey Veng. We did not measure bioavailability of the iron from groundwater, however we note that the groundwater was also acidic, which increases the solubility and bioavailability of the iron. Further details of this exploratory study are outlined in Appendix C: Iron in Groundwater in Prey Veng; however, we conclude that iron in groundwater may be one of several potential factors contributing to the high ferritin concentrations observed in women in Prey Veng.  In this survey in Prey Veng, we observed that in women with some genetic hemoglobin disorders (such as hemoglobin E homozygous), ferritin and sTfR concentrations were significantly higher as compared to women with no hemoglobin disorder. My next steps were to investigate this   60 observation further using a linear regression model to determine the magnitude of effect of each of the six most common hemoglobin disorders on both ferritin and sTfR concentrations. This work is outlined in full in the next chapter.     61 Chapter 3: Factors Associated with Ferritin and Soluble Transferrin Receptor Concentrations in Women of Reproductive Age in Prey Veng  Acknowledgement: A version of this chapter has been published. Karakochuk CD, Whitfield KC, Rappaport AI, Barr SI, Vercauteren SM, McLean J, Prak S, Hou K, Talukder A, Devenish R, Green TJ. The homozygous hemoglobin EE genotype and chronic inflammation are associated with high serum ferritin and soluble transferrin receptor concentrations among women in Cambodia. Journal of Nutrition 2015; 145(12): 2765-73  3.1 Summary Ferritin and sTfR concentrations are commonly used to assess iron deficiency; however, they are influenced by multiple factors. We assessed associations of numerous variables with each of serum ferritin and sTfR concentration among Cambodian women and compared iron deficiency prevalence using study-generated correction factors for ferritin with those from a published meta-analysis. Venous blood from 450 women 18-45 years was assessed for hemoglobin, ferritin, sTfR, retinol binding protein, folate, vitamin B12, CRP, AGP and genetic hemoglobin disorders. Linear regression was used to calculate geometric mean ratios (95% CI) for ferritin and sTfR concentrations. The hemoglobin E homozygous genotype was associated with 50% (95% CI: 14%, 96%) and 51% (95% CI: 37%, 66%) higher geometric mean ferritin and sTfR concentrations as compared to normal hemoglobin; and a 1-g/L increase in AGP was associated with 99% (95% CI: 50%, 162%) and 48% (95% CI: 33%, 64%) higher concentrations in the same parameters, respectively. The prevalence of iron deficiency among non-pregnant women   62 (n=420) was 2.1% (n=9) using ferritin <15 µg/L and 18.8% (n=79) using sTfR >8.3 mg/L [reported in Table 2-3]. Iron deficiency prevalence using sTfR was higher among women with the hemoglobin E homozygous genotype (n=17, 55%) compared to normal hemoglobin (n=20, 10%), and among women with normal hemoglobin and chronic inflammation (n=10, 18%) compared to women without (n=10, 7%) (P<0.05). No differences in iron deficiency prevalence were found using ferritin among hemoglobin E homozygous and normal hemoglobin genotypes (P=1.0) or by chronic inflammation status (P=0.32). There were no differences in mean ferritin concentrations among all 450 women comparing study-generated correction factors with those from the meta-analysis (P=0.87). Compared to sTfR, ferritin concentrations appear to more accurately reflect true iron deficiency in this population. The correction factors for ferritin from a published meta-analysis were appropriate for use in this population with a high prevalence of hemoglobin disorders and inflammation.  3.2 Introduction Iron deficiency is a serious public health problem that affects over two billion people worldwide (8). There is a tacit assumption that the majority of anemia in Southeast Asia is due to iron deficiency; however, recent evidence has shown surprisingly low rates of iron deficiency among women of reproductive age in Cambodia (22) and in neighboring Vietnam (146). When iron deficiency is not the cause of anemia, iron supplementation and fortification will neither treat nor reduce the burden of anemia. At best, these interventions are a waste of valuable resources, and at worst, they may cause harm.    63 The gold standard diagnostic test for iron deficiency is the assessment of iron stores on a bone marrow aspirate (2,23); however, this is rarely used. Instead, ferritin and sTfR concentrations are used as biomarkers of iron deficiency (2,52). However, these biomarkers can be influenced by multiple factors. Ferritin concentration can drop in pregnancy as a result of hemodilution (2). Conversely, ferritin can be elevated in the presence of some diseases (2,101), as well as inflammation or infection (67,76,102). To account for the elevation of ferritin in the presence of inflammation, Thurnham et al. (76) have developed correction factors based on the stages of inflammation (incubation, early and late convalescence) using data from 32 studies of apparently healthy individuals in the absence of overt disease in over 10 countries. However, the accuracy of these correction factors has not yet been studied among populations with a high prevalence of genetic hemoglobin disorders, some of which (e.g., hemoglobin E homozygous) have shown to increase ferritin and sTfR concentrations (22,25). Further, sTfR concentration may be elevated as a result of increased erythropoiesis, which has been observed in individuals with thalassemia (133,145), malaria (114,147) and megaloblastic anemia as a result of folate deficiency (101,108). Therefore, the presence of these other factors can reduce the diagnostic accuracy of iron deficiency, which has implications that can influence the design and implementation of nutrition programming.   The aims were to assess associations of numerous variables with each of serum ferritin (µg/L) and sTfR concentration (mg/L) among Cambodian women and to compare study-generated inflammation correction factors for ferritin with those from a published meta-analysis (76).    64 3.3 Methods 3.3.1 Study Design and Participants The 450 women included in this study were the same women described in Chapter 2. In sum, women were between 18-45 years, had at least one child <5 years of age, and lived in households with some access to land for agriculture or aquaculture activities.   3.3.2 Blood Collection, Processing, and Analysis Details of the blood collection, processing, and analysis were previously described in Chapter 2. In sum, serum was analyzed for ferritin, sTfR, RBP, AGP and CRP using an s-ELISA (86). Quality control checks were completed for each of the micronutrient and inflammation markers analyses and were within acceptable ranges and limits. The inter-assay coefficients of variation (CV) for ferritin, sTfR, CRP and AGP were 3.3, 2.0, 4.5, and 5.5%, respectively, and measured values were certified against the Centre for Disease Control (Atlanta, USA) quality control samples and BioRad (Hercules, USA) Liquichek™ immunoassay controls. All reported values were the mean of an independent double measurement.  3.3.3 Data Preparation and Statistical Analysis The K-nearest neighbors imputation method was used (139) to generate data for women with missing vitamin B12 values (n=24) due to incomplete assays. This method input the median values of the ten most similar samples in the data set to the observations with missing values. The dataset used for this imputation included seven variables (among all 450 women); hence six variables were used to impute the B12 values: four continuous variables (adjusted ferritin, sTfR, RBP, and hemoglobin concentration) and two binary categorical variables (pregnancy status and   65 whether or not a woman had a genetic hemoglobin disorder). No other values were missing from the dataset.  Multivariable linear regression was used to assess associations of a number of independent variables with each of serum ferritin (µg/L) and sTfR concentration (mg/L) as continuous outcome variables. The distributions of ferritin and sTfR concentration were right-skewed; therefore, both outcome variables were natural log-transformed before inclusion in the regression models. The interpretation of a model with a log-transformed outcome variable is similar to that of a model with a normally distributed outcome variable, where a 1-unit change in the independent variable results in a change in the outcome variable (β coefficient), holding all other variables constant. However, in a model with a log-transformed outcome variable, the β coefficients are interpreted in terms of the proportional change in the outcome variable. To calculate the proportional change, we exponentiated (expβ1) the unstandardized β coefficients to estimate the expected proportional change in the geometric mean of the outcome variable (148). The proportional change in the geometric mean of the independent variable in the model is then divided by the geometric mean of the reference to provide a geometric mean ratio (similar to a ratio of medians) (148). The geometric mean ratios (95% CI) are reported. A ratio >1 indicates a positive association between the explanatory variable and outcome variable, and a ratio <1 indicates a negative association (149). Ratios are interpreted somewhat similar to risk ratios. For example, a geometric mean ratio of 1.1 would indicate a positive association between the outcome and explanatory variable, and a 1-unit increase in the explanatory variable is associated with a 10% higher geometric mean value of the outcome variable. Given that the interpretation of log-transformed regression output is not always straightforward, we calculated geometric   66 mean ratios for the standardized β coefficients, which present the results so that each independent variable has a variance of 1.  We included explanatory variables in the models that were commonly known to be associated with iron status even if there was a weak bivariate relationship detected in our data. Age, vitamin B12, ferritin, sTfR, RBP, folate, CRP and AGP were analyzed as continuous variables in the regression models. RBP and ferritin concentrations (for the sTfR model) were included in the models as unadjusted variables (not corrected for inflammation) because CRP and AGP inflammation biomarkers were also included in the models. Due to a heavily right-skewed distribution, parity was included as a categorical variable, coded as 0, 1 or 2 depending on if the woman had previously given birth to 1, 2-3, or ≥4 children, respectively. An eligibility criterion for the study was that a woman had at least one child; therefore, all women in the study were either primiparous or multiparous. Pregnancy status was self-reported by women and included as a categorical variable, coded as 0 or 1 according to if they were non-pregnant or currently pregnant at the time of the survey, respectively. The common hemoglobin genotypes were included in the models as non-ordered categorical variables, coded as 0 or 1 according to absence or presence of each hemoglobin genotype, respectively. The normal hemoglobin genotype was coded as the reference group; therefore, women in each of the abnormal hemoglobin genotype groups are compared against the normal hemoglobin genotype. As our analysis was limited to data (e.g., independent variables) that were collected ‘a priori’ for a different trial, we acknowledge there are some variables that are known to be associated with serum ferritin and sTfR that were not examined in this analysis (e.g., menstrual blood loss).    67 To determine whether the use of the correction factors for ferritin proposed by Thurnham et al. (76) to adjust for inflammation were appropriate for use in this population, we used linear regression to calculate the geometric mean ratios (95%) for ferritin concentration among women in our study for each stage of inflammation (incubation, early and late convalescence), relative to no inflammation. We then calculated the corresponding correction factors (equal to the value of 1 divided by the geometric mean ratio). We compared these ‘study-generated’ correction factors to the Thurnham et al. proposed correction factors for all women, as well as women sub-grouped according to normal or abnormal hemoglobin genotype.  T-tests were used to compare means and chi-square or fisher’s exact tests (when expected cell frequencies in the 2x2 table were five or less) were used to compare proportions between groups. Two-sided P values less than 0.05 indicated statistical significance. Stata software version SE/13.1 for Mac (Stata Corp, College Station, Texas) was used to conduct statistical analyses.   3.4 Results 3.4.1 Participant Characteristics A total of 420 non-pregnant women and 30 pregnant women (18-45 years) were included in this cross-sectional analysis. Baseline data were reported in Table 2-1.  3.4.2 Factors Associated with Ferritin and Soluble Transferrin Receptor Concentrations Using linear regression, we determined the factors associated with serum ferritin and sTfR (Table 3-1) as independent outcome variables among 407 Cambodian women (not including the n=43 women with rare hemoglobin genotypes).   68 Table 3-1 Geometric mean ratios for ferritin and soluble transferrin receptor concentrations among Cambodian women in Prey Veng  Ferritin concentration  STfR concentration   Unstandardized Standardized1 Unstandardized Standardized1   Ratio (95% CI) P Ratio Ratio  (95% CI) P Ratio (Constant)  17.30 (9.99, 30.0) <0.001 N/A 3.97 (3.21, 4.91) <0.001 N/A Age, years 1.01 (1.00, 1.03) 0.04 1.13 1.00 (0.99, 1.01) 0.83 1.01 Parity (reference=1 child)             2-3 children 1.04 (0.88, 1.22) 0.62 1.03 1.03 (0.96, 1.09) 0.44 1.04     ≥4 children 1.00 (0.78, 1.28) 0.99 1.00 1.03 (0.94, 1.13) 0.52 1.04 Pregnancy status (reference=no)             Currently pregnant (yes) 0.64 (0.47, 0.85) 0.002 0.85 1.11 (0.99, 1.24) 0.08 1.09 Micronutrient biomarkers             Vitamin B12, ρmol/L 1.00 (1.00, 1.00) 0.12 1.08 1.00 (1.00, 1.00) 0.40 0.96     Folate, nmol/L 1.01 (1.01, 1.02) <0.001 1.21 1.00 (1.00, 1.00) 0.39 0.96     RBP (unadjusted), µmol/L 1.27 (1.15, 1.40) <0.001 1.25 1.06 (1.02, 1.11) 0.002 1.15     Ferritin (unadjusted), µg/L N/A N/A N/A N/A 1.00 (1.00, 1.00) 0.17 0.94     STfR, mg/L  0.95 (0.92, 0.98) 0.001 0.84 N/A N/A N/A N/A Inflammation biomarkers              CRP, mg/L 1.01 (0.99, 1.03) 0.42 1.04 1.00 (0.99, 1.01) 0.77 0.99     AGP, g/L 1.99 (1.50, 2.62) <0.001 1.31 1.48 (1.33, 1.64) <0.001 1.45 Hb genotype (reference=normal)             Hb E trait 0.97 (0.81, 1.16) 0.72 0.98 1.06 (0.99, 1.14) 0.11 1.08     Hb E homozygous 1.50 (1.14, 1.96) 0.003 1.17 1.51 (1.37, 1.66) <0.001 1.46     3.7-thalassemia trait 0.84 (0.68, 1.02) 0.08 0.92 1.00 (0.89, 1.09) 0.93 1.00     Hb E trait & 3.7-thalassemia trait 0.97 (0.74, 1.27) 0.83 0.99 0.99 (0.89, 1.09) 0.89 0.99     Hb CS trait 1.26 (0.89, 1.79) 0.19 1.06 1.44 (1.26, 1.64) <0.001 1.27   69 Total n=407 women; not including n=43 women with rare genotypes with sample size of ≤7 in each group. AGP, -1 acid glycoprotein; CRP, C-reactive protein; Hb, hemoglobin; N/A, not applicable; RBP, retinol binding protein; sTfR, soluble transferrin receptor. Multivariable linear regression was used to calculate the reported unstandardized β coefficients, which were exponentiated (expβ), resulting in the geometric mean ratio (95% CI).  1 Geometric mean ratios for the standardized β coefficients, which present the results on an equal scale, where each independent variable has a variance of 1.   70 3.4.2.1 Ferritin Concentration  Folate, RBP and AGP concentrations, age, and having the hemoglobin E homozygous genotype were significant positive predictors of ferritin concentration (µg/L). A 1-unit increase in folate, RBP and AGP concentrations was associated with 1% (95% CI: 1%, 2%), 27% (95% CI: 15%, 40%) and 99% (95% CI: 50%, 162%) higher geometric mean ferritin concentrations, respectively. A 1-year increase in age was associated with a 1% (95% CI: 0%, 3%) higher mean ferritin concentration. Lastly, the hemoglobin E homozygous genotype was associated with a 50% (95% CI: 14%, 96%) higher ferritin concentration, as compared to normal hemoglobin. Conversely, currently being pregnant and sTfR concentration were significant negative predictors of ferritin concentration. Being pregnant was associated with a 36% (95% CI: 15%, 53%) lower geometric mean ferritin concentration. A 1-mg/L increase in sTfR concentration was associated with a 5% (95% CI: 2%, 8%) lower mean ferritin concentration. The adjusted model predicted ~18% of the variance in serum ferritin concentration.   However, it is important to interpret these proportionate changes in the geometric mean with consideration of the units presented for each independent variable. For example, a 1 g/L increase in AGP represents a concentration of ~4 SD units higher. Only one woman in our study fell into this category and had an AGP concentration that was 1 g/L above the mean AGP concentration of 0.87 g/L (the mean of non-pregnant women). Therefore, we have also presented the geometric mean ratios using the standardized β coefficients, which present the results on an equal scale, where each independent variable has a variance of 1. For example, a 1 SD increase in AGP concentration was associated with a 20% (95% CI: 12%, 20%) higher mean ferritin concentration.   71 3.4.2.2 Soluble Transferrin Receptor Concentrations RBP and AGP concentrations, and having the hemoglobin E homozygous or CS trait were significant positive predictors of sTfR concentration (mg/L). A 1-unit increase of RBP and AGP concentrations was associated with 6% (95% CI: 2%, 11%) and 48% (95% CI: 33%, 64%) higher geometric mean sTfR concentrations, respectively. Alternatively, a 1 SD increase in AGP concentration was associated with an 11% (95% CI: 8%, 15%) higher mean sTfR concentration. Having the hemoglobin E homozygous or the CS trait was associated with 51% (95% CI: 37%, 66%) and 44% (95% CI: 26%, 64%) higher mean sTfR concentrations, as compared to normal hemoglobin, respectively. The adjusted model predicted ~28% of the variance in sTfR concentration.  3.4.3 Prevalence Estimates of Anemia and Iron Deficiency We compared iron deficiency prevalence among non-pregnant women with and without normal hemoglobin genotypes, and with and without chronic inflammation (Table 3-2). Differences in iron deficiency prevalence were observed among women using sTfR. Specifically, prevalence was higher among those with abnormal (26.2%, n=59) versus normal hemoglobin (10.3%, n=20), and among those women with normal hemoglobin and chronic inflammation (17.5%, n=10) versus without chronic inflammation (7.3%, n=10) (P<0.05). No differences in iron deficiency prevalence based on ferritin were detected. Among the 420 non-pregnant women, iron deficiency prevalence did not differ among women with normal hemoglobin (2.6%, n=5) as compared to the hemoglobin E homozygous genotype (0%, n=0) (P=1.0); or among those 195 women with normal hemoglobin with chronic inflammation (0%, n=0) and those without chronic inflammation (3.6%, n=5) (P=0.32). Notwithstanding this, women in our study had surprisingly   72 high mean ferritin concentrations, even after adjustment for levels of subclinical inflammation using methods proposed by Thurnham et al. (76). In fact, adjustment for inflammation did not change the prevalence of iron deficiency in any of the groups. The discrepancy in iron deficiency prevalence using ferritin and using sTfR was most prominent among women with the hemoglobin E homozygous genotype. This was not surprising, because as reported previously (22), women with the hemoglobin E homozygous genotype had significantly higher mean ferritin and sTfR concentrations as compared to women with normal hemoglobin (ferritin: 129  90.6 vs. 95.8  56.2 µg/L [P<0.05], and sTfR: 9.5  3.5 vs. 6.4  1.9 mg/L [P<0.05], respectively).   73 Table 3-2 Iron deficiency prevalence among non-pregnant women with and without genetic hemoglobin disorders, and with and without chronic inflammation Total n=420. Values are n (%). Hb, hemoglobin; sTfR, soluble transferrin receptor. Chi-square or fisher’s exact tests were used to determine if the proportion of iron deficiency among groups were significantly different (each compared to the reference group of no hemoglobin disorders or no inflammation). Values that do not share a common superscript letter in the column are significantly different from the reference group (P<0.05). 1 Values were corrected for inflammation using Thurnham et al. correction factors (76). 2 Including all types of abnormal hemoglobin genotypes (e.g., homozygous and/or heterozygous E, CS, -thalassemia). 3 Defined as AGP <1 g/L. Iron deficiency prevalence  Unadjusted ferritin  (<15 µg/L) Adjusted ferritin1 (<15 µg/L) STfR  (>8.3 mg/L) Among 420 non-pregnant women:          Women with normal Hb (n=195) 5 (2.6)a 5 (2.6)a 20 (10.3)a      Women with any abnormal Hb genotype2 (n=225) 3 (1.3)a 4 (1.8)a 59 (26.2)b      Women with the Hb E homozygous genotype (n=31) 0 (0)a 0 (0)a 17 (54.8)b Among 195 non-pregnant women with normal Hb:        Women with no chronic inflammation3 (n=138) 5 (3.6)a 5 (3.6)a 10 (7.3)a      Women with chronic inflammation (n=57) 0 (0)a 0 (0)a 10 (17.5)b   74 3.4.4 Adjustment of Ferritin Concentration for Inflammation Geometric mean ratios (95% CI) for ferritin were determined for the three stages of inflammation, relative to those with no inflammation (Table 3-3).   Table 3-3 Geometric mean ratios for ferritin across the stages of inflammation among Cambodian women in Prey Veng Total n=450. Inflammation stages defined as: incubation (CRP >5 mg/L); early convalescence (CRP >5 mg/L and AGP >1 g/L); and late convalescence (AGP >1 g/L). AGP, α-1 acid glycoprotein; CRP, C-reactive protein; Hb, hemoglobin; SE, standard error. Linear regression was used to calculate the geometric mean ratio (95% CI). Ratios compare women in each inflammation stage to women with no inflammation (reference group).  We then compared the study-generated correction factors to the Thurnham et al. (76) correction factors for all women, as well as women sub-grouped according to normal or abnormal hemoglobin genotype (Table 3-4). The study-generated correction factors for all 450 women were essentially identical to the Thurnham et al. (76) correction factors for incubation and late convalescence, and tended to be higher in early convalescence, but the Thurnham et al. factors still fell within the 95% CI of our study-generated correction factors. Among all 450 women,  n  Ratio (95% CI) SE P For all women (n=450)            Incubation 18 1.36 (0.98, 1.88) 0.23 0.07      Early convalescence 22 1.43 (1.06, 1.92) 0.22 0.02      Late convalescence 87 1.35 (1.15, 1.59) 0.11 <0.001 For women with the normal Hb (n=207)            Incubation 8 1.50 (0.91, 2.49) 0.38 0.11      Early convalescence 13 1.44 (0.97, 2.15) 0.29 0.07      Late convalescence 45 1.38 (1.09, 1.75) 0.17 <0.001 For women with any abnormal Hb genotype (n=243)            Incubation 10 1.25 (0.81, 1.93) 0.28 0.32      Early convalescence 9 1.40 (0.88, 2.21) 0.33 0.15      Late convalescence 42 1.32 (1.04, 1.66) 0.15 0.02   75 adjusted ferritin concentrations were 93.1 ± 56.8 µg/L and 93.8 ± 56.8 µg/L (P=0.87), using study-generated and Thurnham et al. correction factors, respectively. In the subset of non-pregnant women (n=420), adjusted means were 95.7 ± 57.0 µg/L and 95.0 ± 57.0 µg/L (P=0.86), respectively. In the subset of pregnant women (n=30), adjusted ferritin was 66.2 ± 46.9 µg/L and 66.9 ± 47.8 µg/L (P=0.95), respectively. As such, it was not surprising that the prevalence of iron deficiency based on ferritin concentration <15 µg/L did not change when comparing both sets of correction factors (n=9, 2.1% among non-pregnant women, and n=4, 13.3% among pregnant women).   76 Table 3-4 Comparison of the correction factors for ferritin from the Thurnham et al. meta-analysis and study-generated correction factors for Cambodian women by inflammation stage  Incubation Early  convalescence Late  convalescence From the Thurnham et al. meta-analysis (76) 0.77 0.53 0.75 Using study-generated correction factors:       All women (n=450) 0.74 (0.53, 1.02) 0.70 (0.52, 0.94) 0.74 (0.63, 0.86)    Women with the normal Hb (n=207) 0.67 (0.40, 1.10) 0.69 (0.46, 1.03) 0.72 (0.57, 0.92)    Women with any abnormal Hb genotype (n=243) 0.80 (0.52, 1.23) 0.71 (0.45, 1.14) 0.76 (0.60, 0.96) Total n=450. Inflammation stages defined as: incubation (CRP >5 mg/L); early convalescence (CRP >5 mg/L and AGP >1 g/L); and late convalescence (AGP >1 g/L). AGP, α-1 acid glycoprotein; CRP, C-reactive protein; Hb, hemoglobin. Correction factors were calculated as the value of 1 divided by the geometric mean ratio (presented in Table 3-3).   77 3.5 Conclusions and Discussion The most interesting finding we observed was among women with the hemoglobin E homozygous genotype (7%, n=31), which is a disorder caused by a mutation in the -globin gene from both parents, resulting in a more pronounced phenotype (i.e., lower hemoglobin concentration) than the heterozygous form (132,150). Having the hemoglobin E homozygous genotype was associated with 50% higher mean ferritin and sTfR concentrations, as compared to those with normal hemoglobin. These findings could have serious diagnostic implications for individuals with the hemoglobin E homozygous genotype, as elevations in serum ferritin and sTfR that do not result from changes in iron status could underestimate or overestimate the prevalence of iron deficiency, respectively. In our study, iron deficiency prevalence using sTfR was higher among non-pregnant women with the hemoglobin E homozygous genotype as compared to normal hemoglobin, and among non-pregnant women with normal hemoglobin and chronic inflammation as compared to without chronic inflammation. No differences in iron deficiency prevalence were found among hemoglobin genotype and inflammation groups using ferritin. However, women in our study had surprisingly high mean ferritin concentrations (~95 µg/L). If ferritin concentrations had been lower (closer to the cut-off for iron deficiency diagnosis) these factors could have had more substantial implications on prevalence estimates. Further research is required to substantiate this hypothesis. The hemoglobin E heterozygous trait, which is more common in Cambodia (~33% prevalence) but typically results in a less severe phenotype (18,22), was not significantly associated with ferritin concentration, as compared to normal hemoglobin (ratio of 0.97 [95% CI: 0.81, 1.16], P=0.72). The association between the hemoglobin E heterozygous trait and sTfR concentration was also non-significant (ratio of 1.06 [95% CI: 0.99, 1.14], P=0.11).   78 The question that remains is whether or not these ferritin and sTfR concentrations are actually reflecting iron status, or are elevated for other reasons. It seems improbable that a single disorder (e.g., the hemoglobin E homozygous genotype) could be simultaneously associated with both a lower risk of iron deficiency (higher ferritin) and an increased risk of iron deficiency (higher sTfR). In individuals with the hemoglobin E homozygous genotype, elevated serum ferritin or sTfR may not be a result of changes in iron status. Whether or not ferritin and/or sTfR concentrations in our study are reflective of iron deficiency cannot be ascertained from the data, as bone marrow biopsies were not conducted. However, based on the known risk of iron overload among those with the hemoglobin E homozygous genotype (18), it seems more likely that sTfR concentrations do not reflect true iron deficiency among Cambodian women with the hemoglobin E homozygous genotype and chronic inflammation. Ferritin appears to be a better indicator of iron deficiency in this population.  Current pregnancy was a significant negative predictor of ferritin and was associated with a 36% (15, 53) lower ferritin concentration. Pregnancy-related hemodilution can result in low concentrations of ferritin and other nutrients (2,78). Further, daily iron requirements are also increased during pregnancy, which if not met, may lead to decreased iron stores (78,95). For these reasons, WHO has advised that ferritin cut-off value should be used with caution in pregnant women (2). In our study, pregnancy status was trending towards a positive association with sTfR concentration (ratio of 1.11 [95% CI: 0.99, 1.24]). Compared to ferritin concentration, sTfR concentration is thought to be less influenced by pregnancy status (108), although studies have shown conflicting results: one study observed no difference between pregnant and non-pregnant women in the third trimester (109) while another study reported sTfR concentrations   79 were significantly higher among healthy, non-anemic pregnant women as compared to non-pregnant women (110). However, higher sTfR concentrations could be due to increased erythropoiesis, which can occur during pregnancy (111) as a result of hemodilution (78) and/or an increase of renal oxygen consumption due to an increased glomerular filtration rate (112).  AGP was a significant positive predictor of both ferritin and sTfR concentrations. Infectious pathogens, metabolic stress and tissue damage activate the inflammatory response (52). Cytokines are released, stimulating the production of hepcidin, which functions as a regulator of iron metabolism (19). Hepcidin binds to and degrades ferroportin, a transport protein on the wall of the macrophage, sequestering iron in the macrophage and making it unavailable for erythropoiesis (16,74). This results in ‘functional iron deficiency’ and is thought to be a protective mechanism to prevent pathogenic organisms from using iron in circulation (16,17), leading to increased storage iron such as ferritin (75). The released cytokines increase acute-phase proteins (e.g., AGP and CRP), the biomarkers used to estimate the response of ferritin in the different phases of the inflammatory response (2,76). Conversely, sTfR is not an acute-phase protein and is thought to be less influenced by inflammation (2,108). Yet in our study, a significantly higher proportion of non-pregnant women with chronic inflammation (18%) had an elevated sTfR concentration (>8.3 mg/L) than did women without chronic inflammation (7%, P=0.003). Based on our available data, it is difficult to ascertain the potential cause of elevated sTfR. Potential causes include malaria or other hemolytic conditions, which are known to be associated with increased sTfR concentrations (85,114).     80 In Cambodia, anemia prevalence among women of reproductive age is 45%, according to the recent 2014 Demographic and Health Survey (7). Global WHO guidelines recommend weekly iron and folic acid supplementation (60 mg iron and 2800 μg folic acid) for all menstruating women among populations where the prevalence of anemia is >20% (20). The 2011 Cambodian National Nutrition Policy (28) currently supports this recommendation and at the time this research was initiated, the Cambodia Ministry of Health was piloting the weekly iron and folic acid supplementation program in some districts in addition to multiple other interventions to reduce iron deficiency (e.g., the national food fortification initiative for iron-fortified fish and soy sauces) (28). However, if iron deficiency is not a major cause of anemia, these interventions could be a waste of resources. Worse, they could be harmful, in particular for those individuals with certain genetic hemoglobin disorders. There is a high risk of iron overload in individuals with hemoglobin E homozygous/-thalassemia major (18,19,132) and a moderate risk in individuals with a hemoglobin E heterozygous trait (21), both of which are found in the Cambodian population, with ~0.5% and ~33% prevalence, respectively (22). Regardless of hemoglobin genotype, excess iron can cause oxidative stress and cell damage and is associated with other diseases (95,151). This risk of iron overload has yet to be studied among apparently iron-sufficient women with genetic hemoglobin disorders in Cambodia, where multiple iron interventions are being implemented on a national scale.   The correction factors for ferritin from Thurnham et al. (76) fell within the 95% CI of our study-generated correction factors from Cambodia and we did not observe any differences in adjusted ferritin concentrations after correcting for inflammation using either method. This is an interesting finding as Thurnham’s correction factors were generated based on studies of   81 apparently healthy individuals in the absence of overt disease. In individuals with inflammation caused by infection (e.g., malaria or helminth), serum ferritin and sTfR concentrations may respond differently as compared with those individuals with inflammation in absence of overt disease (152). Unfortunately, we did not directly measure the level of infection among individuals in our study, but we note that malaria prevalence is reported as very low in Prey Veng (153).   The main limitation of this study was inherent in the cross-sectional design, in which causation cannot be inferred and findings cannot be extrapolated beyond the model. Women were recruited by convenience sampling; however, we do not feel this restricts the interpretation of our data as the recruitment criteria represent the majority of Cambodian women of reproductive age that live in rural areas. We did not have sufficient data on dietary intakes to include estimates of dietary iron intake. Dietary iron intake and/or bioavailability of dietary iron sources are likely associated with iron stores. Menstrual blood loss is also a strong determinant of iron stores in women of reproductive age (79) that was not captured in our analysis.   We conclude that the hemoglobin E homozygous genotype and chronic inflammation were associated with high ferritin and sTfR concentrations among women of reproductive age in Prey Veng. Compared to sTfR, ferritin concentrations appear to more accurately reflect true iron deficiency in this population. The accuracy of sTfR as a biomarker of iron deficiency among individuals with the hemoglobin E homozygous genotype and chronic inflammation is in question.    82 3.6 Next Steps The findings from this analysis demonstrated that certain hemoglobin disorders and chronic inflammation are associated with increased ferritin and sTfR concentrations. These findings could have serious diagnostic implications, as elevations in serum ferritin and sTfR that do not result from changes in iron status could underestimate or overestimate the prevalence of iron deficiency, respectively. The reliance on iron biomarkers to estimate iron deficiency anemia may not be accurate in Cambodia as the biomarkers are confounded by multiple factors, such as inflammation, infection, and genetic hemoglobin disorders. As such, the only way to assess the true prevalence of iron deficiency anemia is to measure the hemoglobin response to iron supplementation. Our next step was to design and conduct an iron supplementation trial that would address this research question. This work is outlined in full in the next chapter.   83 Chapter 4: The Effect of 12 weeks of Daily Oral Iron with or without Multiple Micronutrients on Hemoglobin Concentration and Hemoglobin Response: A 2x2 Factorial Double-Blind Randomized Controlled Supplementation Trial in Kampong Chhnang  Acknowledgement: A version of this chapter has been submitted for publication. Karakochuk CD, Barker MK, Whitfield KC, Barr SI, Vercauteren SM, Devlin AM, Hutcheon JA, Houghton LA, Prak S, Hou K, Chai TL, Stormer A, Ly S, Devenish R, Oberkanins C, Puringher H, Harding KB, De-Regil LM, Kraemer K, Green TJ.  4.1 Summary Despite a high prevalence of anemia among non-pregnant Cambodian women, recent reports suggest iron deficiency prevalence is low. If true, iron supplementation will not be an effective anemia reduction strategy. We measured the effect of daily oral iron (Fe) with or without multiple micronutrients (MMN) on hemoglobin concentration in Cambodian women. In this 2x2 factorial double-blind randomized trial, non-pregnant women 18-45 years with hemoglobin ≤117 g/L (based on the HemoCue® using capillary blood) were recruited from 26 villages in Kampong Chhnang province and randomized to receive 12 weeks of Fe (60 mg), MMN (14 other micronutrients), Fe+MMN, or placebo capsules. Generalized linear models were used to assess the primary outcome: mean hemoglobin across groups at 12 weeks (intention-to-treat analyses). In July 2015, 809 women were recruited and 760 completed the trial. Baseline anemia prevalence was 58% (based on the Sysmex analyzer using venous blood). Mean (95% CI)   84 hemoglobin at 12 weeks did not differ in the Fe and Fe+MMN groups (121 [120, 122] vs. 123 [122, 124] g/L); both were higher than MMN and placebo (both 116 [115, 117] g/L, P<0.05). Mean (95% CI) increase in hemoglobin was 5.6 (3.8, 7.4) g/L (P<0.001) among women who received Fe (n=383) and 1.1 (-0.7, 2.9) g/L (P=0.24) among women who received MMN (n=382), with no interaction between interventions (P=0.61). At 12 weeks, 19% and 30% of women had a hemoglobin response ≥10 g/L in Fe and Fe+MMN groups, compared to 8% and 5% in MMN and placebo, respectively. Daily iron supplementation for 12 weeks increased hemoglobin concentrations in non-pregnant Cambodian women; however, MMN did not confer additional significant benefit. Of the women who received Fe (with or without MMN), overall only ~25% (n=95/383) responded to iron. In the subset of women determined as anemic at baseline based on hemoglobin measurement by the Sysmex analyzer, ~37% (n=85/232) responded to iron. As we recruited women with hemoglobin ≤117 g/L (based on the HemoCue®), we predict that even fewer women would respond to iron in the wider population.   4.2 Introduction In 2011, an estimated 500 million non-pregnant women globally were anemic (53). It is often assumed that 50% of anemia is due to iron deficiency in low-income countries, which has been the impetus for global WHO recommendations for iron supplementation among menstruating women and adolescents (20,128). Despite this, recent surveys have shown a surprisingly low prevalence of iron deficiency (≤8% based on ferritin <12-15 μg/L) among women of reproductive age in Cambodia (22,154), Vietnam (146), Bangladesh (155), Nepal (156), the Democratic Republic of the Congo (157), and Sierra Leone (158).   85 In Cambodia, the 2014 Demographic and Health Survey reported that the national prevalence of anemia and iron deficiency (based on ferritin) were 45% and 3% among women of reproductive age, respectively (7), and genetic hemoglobin disorders (e.g., -thalassemia or hemoglobin E disorders) (22,25) were found in 60% of women (7). These inherited conditions can result in reduced or abnormal globin chain synthesis in the hemoglobin molecule, resulting in mild to severe anemia regardless of iron stores (18). Further, it has been shown that some of these hemoglobin disorders are associated with increased ferritin (21,22) and sTfR (22,133,145) concentrations. This is concerning because ferritin and sTfR concentrations that are elevated as a result of a hemoglobin disorder, rather than as a result of iron status, will underestimate and overestimate iron deficiency prevalence, respectively. This raises questions about the diagnostic accuracy of ferritin and sTfR biomarkers in this population.   If iron deficiency is not a major cause of anemia, then at best, untargeted iron supplementation is a waste of resources, and at worst, it could cause harm, especially in individuals with certain hemoglobin disorders who are already at risk of iron overload and oxidative stress (21). In 2011, the Cambodia Ministry of Health adopted the recommendation for weekly iron and folic acid supplementation for all women of reproductive age, consistent with the global WHO guidelines (20). However, given the recent findings on the lack of iron deficiency, the Ministry of Health put the weekly iron and folic acid supplementation program on hold in 2015 and is awaiting additional evidence to inform policy. Further, the efficacy of supplementation of other micronutrients (e.g., vitamin B12, folate, or zinc) to reduce anemia among non-pregnant women of reproductive age is also uncertain. As such, the aim was to conduct a randomized controlled   86 supplementation trial to determine whether iron with or without other micronutrients increases hemoglobin concentrations in non-pregnant Cambodian women screened as having anemia.  4.3 Methods 4.3.1 Study Design and Participants This was a 2x2 factorial double-blind placebo-controlled randomized trial of oral iron supplementation with or without multiple micronutrients. The study took place in two districts (Kampong Tralach and Sameakki Mean Chey) in the rural province of Kampong Chhnang at sea level in central Cambodia. To note, this trial was conducted in a different province than the survey reported in Chapters 2 and 3. Kampong Chhnang was selected for the supplementation trial as it had one of the highest prevalence rates of anemia in the country (7) and our study aimed to recruit only anemic women. Further, we were interested to investigate the factors associated with anemia among women in a new geographical area of Cambodia.  Ethics approval was obtained from the University of British Columbia Clinical Research Ethics Board in Canada (H15-00933) and the National Ethics Committee for Health Research in Cambodia (110-NECHR). This trial was registered at clinicaltrials.gov (NCT-02481375).  The planned study population included anemic non-pregnant women of reproductive age in Kampong Chhnang. To be eligible, women had to meet the inclusion criteria and provide written informed consent before enrolment. Inclusion criteria included healthy, non-pregnant women 18-45 years with a hemoglobin ≤117 g/L based on a finger prick capillary blood sample using a HemoCue® Hb 301 (HemoCue AB, Sweden). We used a cut-off of 117 g/L (rather than the WHO recommended 120 g/L cut-off for anemia (2)) to enable recruitment of more women   87 capable of a hemoglobin response ≥10 g/L. We excluded women who were taking medications or food supplements.   4.3.2 Randomization and Masking The provincial health authority provided a list of all villages in Kampong Tralach and Sameakki Mean Chey districts and 26 villages were selected using a computer-generated list of random numbers. Women were recruited via a convenience sampling method. Eligible women were enrolled and randomized stratified by anemia severity (mild [hemoglobin 110-120 g/L], moderate [hemoglobin 80-110 g/L], and severe [hemoglobin <80 g/L]) and by village, to one of four treatment groups in equal allocation on a weekly-rolling basis using a computer-generated randomization list prepared by the study coordinator that was concealed until time of randomization. Women were assigned to receive one of 12 weeks of daily oral iron ([Fe], 60 mg elemental iron as ferrous sulphate), 14 other micronutrients not including Fe (MMN), Fe and the 14 other micronutrients (Fe+MMN), or placebo (maltodextrin) capsules. The policy in Cambodia is to treat anemia (hemoglobin <120 g/L) with 60 mg elemental iron twice daily for 12 weeks (28). We chose 60 mg as the daily dose in our study, as a higher quantity may have elicited unnecessary adverse side effects and a lower quantity may have left the question of iron’s efficacy unresolved. The MMN formulation was based on the UNICEF/WHO/UNU formulation for pregnant and lactating women (UNIMMAP) (159) with an increased iron content (from 30 to 60 mg) for comparability to the Fe group. The capsule formulations are outlined in Table 4-1.       88 Table 4-1 Capsule formulations Micronutrients Fe+MMN MMN Fe Placebo Vitamin A, μg RE 800 800 - - Vitamin D, IU 200 200 - - Vitamin E, mg 10 10 - - Vitamin C, mg 70 70 - - Vitamin B1, mg 1.4 1.4 - - Vitamin B2, mg 1.4 1.4 - - Niacin, mg 18 18 - - Vitamin B6, mg 1.9 1.9 - - Vitamin B12, μg 2.6 2.6 - - Folic acid, μg 400 400 - - Iron (elemental), mg 60 - 60 - Zinc, mg 15 15 - - Copper, mg 2 2 - - Selenium, μg 65 65 - - Iodine, μg 150 150 - - Fe, iron; Fe+MMN, iron with multiple micronutrients; MMN, multiple micronutrients.  The capsules were manufactured in January 2015 by DSM Nutritional Products Ltd. (Isando, South Africa) as gel capsules identical in size and colour, and differing only by a code on the capsule package. The Project Coordinator from DSM Nutritional Products Ltd. was responsible for blinding and did not share the code until statistical analysis of the primary outcome was completed. Investigators, research staff, and participants were all blinded to group assignment.  4.3.3 Procedures Screening and recruitment commenced on July 5, 2015 and enrolment continued for five weeks, with completion of the trial on November 5, 2015. One study coordinator, two managers, and four research officers oversaw the study implementation, data collection, monitoring, and counseling. Community sensitization began with consultations with health center staff, then village chiefs and village health volunteers. Women were recruited via a convenience sampling   89 method and those 18-45 years were invited for hemoglobin screening to determine eligibility for the study. Enrolled women attended a total of six study visits: at screening, baseline, 1, 4, 8, and 12 week time-points (Table 4-2).   Table 4-2 Schedule of study visits   Study visits and assessment time points Visits (V1-5) V1 V2 V3 V4 V5 V6 Time per session, hours 0.5 0.25 0.25 0.25  0.25 0.5  Study day 0 7 14 35 63 91 Eligibility assessment X      Deworming treatment X      Randomization  X     Questionnaire  X    X Adverse event reporting  X X X X X Blood collection   X    X Capsule dispense  X  X X  Capsule count   X X X X V, visit.  At screening, women were provided with one deworming tablet (Mebendazole, 500 mg) for the preventive treatment of helminth infection. At baseline, research staff administered questionnaires at each woman’s household to collect socio-demographic and health data (including information on other factors that are associated with hemoglobin and hematological indicators, such as reported history of malaria, enteropathies, diarrhea, or other medical conditions, and history of medication use). Research staff and village health volunteers conducted regular monitoring throughout the duration of the study. Adverse events (i.e., a new illness, worsening of a concomitant illness, a side effect of the intervention, or a combination of two or more of these factors) were monitored and recorded at each visit.    90 Phlebotomists collected a morning fasting venous blood sample at baseline, 1, and 12 weeks in the village at a central location. Village chiefs and village health volunteers assisted with mobilization of women during screening and for study visits. Blood was collected in a 6 mL trace element free tube, a 6 mL evacuated tube containing EDTA, and a 2 mL tube containing EDTA (Becton Dickinson, Franklin Lakes, NJ, USA). Blood was placed in racks in a covered icebox and transported within 2-4 hours of collection to the National Institute of Public Health Laboratory for processing. Research officers provided women with one bottle (30 capsules) after baseline blood collection, and at 4 and 8 weeks, and conducted capsule counts at 4, 8, and 12 weeks to monitor adherence. Adherence was based on the average of the three capsule counts and women were defined as adherent if they consumed ≥80% of the capsules.  A complete blood count was performed using an automated hematology analyzer (Sysmex XN-1000, Sysmex Corp, Kobe, Japan) to measure hemoglobin (g/L), MCV (fL), MCH (ρg), and RDW (% of RBC). Plasma, buffy coat and serum were stored in 2 mL vials at -80 oC until shipment on dry ice to Canada for analysis or shipment to other laboratories. Serum was assessed for the following indicators using an s-ELISA (86): ferritin (µg/L), sTfR (mg/L), AGP (g/L),  CRP (mg/L), and RBP (µmol/L). Serum vitamin B12 (ρmol/L) was measured in a randomly selected subset of women (n=399) using an Elecsys® 2010 immunoassay (Roche Diagnostics, Risch-Rotkreuz, Switzerland). Serum folate (nmol/L) was measured in a randomly selected subset of women (n=400) using a microbiological assay according to the methods of O’Broin and Kelleher (135) and Molloy and Scott (160) in a 96-well plate (Costar 3596, Corning Inc., Corning, NY, USA) using chloramphenicol-resistant Lactobacillus rhamnosus (ATCC 7469). Serum hepcidin (nmol/L) was measured using a Hepcidin-25 Bioactive immunoassay kit (DRG   91 International Inc., Springfield Township, NJ, USA) (161). DNA was extracted from buffy coat using a QiaAmp Blood DNA kit (Qiagen Ltd., Hilden, Germany) and assessed for 21 -globin gene deletions and point mutations using the -globin StripAssay® kit (ViennaLab Diagnostics, Vienna, Austria). Hemoglobin electrophoresis for detection of hemoglobin variants (E, CS, H, Bart, or F) was conducted using a Sebia MINICAP analyzer (Sebia, Lisses, France) by a trained external consultant in Cambodia. Ferritin and RBP values were adjusted for inflammation using Thurnham et al. (68,76) correction factors.  4.3.4 Statistical Analyses Sample size was estimated for the primary outcome (hemoglobin, g/L) using G*Power Statistical Program v.3.1.9.2 for Mac (G*Power, Dusseldorf, Germany). A total of 628 women (n=157 per group) was calculated, assuming 90% power to detect a mean difference in hemoglobin of 3 g/L, considering a SD for hemoglobin of 10 g/L and a two-tailed significance level of 0.05. The 3 g/L has previously been reported as a clinically important difference in hemoglobin (83). To account for participant drop-out or loss to follow-up, we rounded up to n=200 per group (total n=800).  The primary outcome was mean hemoglobin (95% CI) at 12 weeks using intention-to-treat analyses with adjustments for baseline hemoglobin and village clusters. A generalized linear mixed-effects model was used to predict the marginal means (95% CI) of hemoglobin (g/L) across groups at 12 weeks (intention-to-treat) with adjustments for baseline hemoglobin (fixed-effects) and village clusters (random-effects). Bonferroni-adjusted pairwise comparisons were   92 made across groups (P<0.05 indicated significant differences). Stata SE v.13.1 (Stata Corp., Texas, USA) was used for statistical analyses.  Secondary outcomes were:  i) At the margins factorial analysis: hemoglobin increase at 12 weeks among all women who received Fe (with or without MMN) compared with those women who did not receive Fe; and all women who received MMN (with or without Fe) compared with those women who did not receive MMN. We tested for an interaction between the two interventions, as the marginal analysis is only appropriate when there is no significant interaction detected (162); ii) Adjusted mean increase in hemoglobin concentration at 12 weeks among women in each treatment group (Fe, MMN, and Fe+MMN) compared to placebo. Post-estimation linear comparisons of the beta coefficients from the adjusted model were used to determine if there was a significant additive effect of MMN to Fe;  iii) Mean serum ferritin, sTfR, hepcidin, RBP, vitamin B12, and folate concentrations among women at 12 weeks; and  iv) The proportions of women defined as ‘hemoglobin responders’ (hemoglobin increase ≥10 g/L) (127) at 12 weeks across the four treatment groups. Analyses were conducted for all women, and among women subgrouped by baseline anemia status, by iron status (ferritin), by adherence, and by the presence or absence of a genetic hemoglobin disorder.     93 4.4 Results 4.4.1 Baseline Characteristics Between July 5th and August 2nd, 2015, 2,846 adult women were screened in 26 villages, of whom 1,889 women did not meet eligibility criteria, 52 declined to participate, 4 became pregnant, 1 became seriously ill, and 91 were lost to follow-up after eligibility screening. The remaining 809 women were randomized and 760 (94%) completed the 12 week trial (Figure 4-1). Primary outcome data were available for 759 women (data for n=1 missing due to a clotted sample). Overall, women had a mean ± SD age of 30 ± 8 years, 67% were married, 56% completed primary school, and 38% had 1 or 2 children (Table 4-3).      94  Figure 4-1 Flow diagram of trial enrolment. Fe, iron; MMN, multiple micronutrients.           95 Table 4-3 Baseline characteristics of enrolled Cambodian women 18-45 years in Kampong Chhnang by supplement group Total n=809. Fe, iron; Hb, hemoglobin; IFA, iron and folic acid; MMN, multiple micronutrients.  2 IFA consumption for any given duration. Of those women who reported consuming IFA during their last pregnancy, 56% (n=257/461) reported consuming ≥90 tablets as per the current recommendation in Cambodia.  Fe MMN Fe+MMN Placebo Total enrolled, n (%) 201 (25) 202 (25) 206 (26) 200 (25) Age, y, mean ± SD  31 ± 8 30 ± 8 30 ± 8 30 ± 8 Household size, mean ± SD 4.8 ± 1.7 4.8 ± 1.8 4.7 ± 1.7 4.7 ± 1.7 Marital status, n (%)         Single 50 (25) 52 (26) 45 (22) 52 (26)     Married 133 (66) 139 (69) 142 (69) 130 (65)     Widowed 11 (6) 4 (2) 8 (4) 6 (3)     Separated/divorced 7 (4) 7 (4) 11 (5) 12 (6) Completed education, n (%)         None 25 (12) 24 (12) 25 (12) 30 (15)     Primary (grades 1-5) 105 (52) 115 (57) 119 (58) 111 (56)     Lower (grades 6-9) 66 (33) 54 (27) 51 (25) 47 (24)     Upper (grades 10-12) 5 (3) 9 (5) 11 (5) 11 (6)     Higher education/university  0 (0) 0 (0) 0 (0) 1 (<1) Main water source, n (%)         Hand pump ground well 33 (16) 35 (17) 31 (15) 36 (18)     Ring well 88 (44) 86 (43) 78 (38) 84 (42)     Pond, river or rainwater 55 (27) 49 (24) 61 (30) 46 (23)     Other 25 (12) 32 (16) 36 (18) 34 (17) Lactating, n (%) 28/191 (15) 29/191 (15) 33/192 (17) 36/186 (19) Taking birth control, n (%) 57/201 (28) 70/202 (35) 67/206 (33) 68/200 (34) Parity (children born), n (%)         0  60 (30) 57 (28) 56 (27) 62 (31)     1-2 70 (35) 78 (39) 83 (40) 75 (38)     3-4 52 (26) 51 (25) 50 (24) 43 (22)     ≥5 19 (10) 16 (8) 17 (8) 20 (10) Women with parity ≥1, who reported to receive, n (%)         IFA during last pregnancy2 108/141 (77) 118/145 (81) 128/150 (85) 107/138 (78)     Deworming during last pregnancy 75/141 (53) 81/145 (56) 98/150 (65) 78/138 (57)     Vitamin A postpartum 21/141 (15) 39/145 (27) 47/150 (31) 30/138 (22)   96 Of those women who had previously given birth, 80% reported taking iron and folic acid supplements during their last pregnancy (for any quantity and/or duration), and of those women, 56% (n=257/461) reported consuming ≥90 tablets as per the current recommendation in Cambodia. At baseline, a total of 17% women were lactating and 32% were taking oral contraceptives (birth control).  Nutrition, inflammation and hematological indicators and prevalence rates of iron deficiency, inflammation and anemia at baseline are presented in Table 4-4. Among all women screened in Kampong Chhnang (n=2,856), anemia prevalence was ~39% (n=1,112/2,856) based on the WHO cut-off for non-pregnant women (hemoglobin <120 g/L) using the HemoCue®. Despite enrolling only women with hemoglobin ≤117 g/L (based on a capillary finger prick blood sample using the HemoCue®), overall baseline anemia prevalence as assessed based on fasting venous blood samples using an automated Sysmex hematology analyzer revealed that only 58% (n=468/808) of women were anemic (<120 g/L). The concordance plot of hemoglobin concentration based on HemoCue® and Sysmex methods at baseline (Figure 4-2) shows large variation in hemoglobin between the two methods. However, we note that this figure includes only women identified as anemic using the HemoCue®, rather than all women initially screened, as such, the figure should be interpreted with this in mind. The differences observed could simply be due to regression to the mean, depending on the magnitude of intra-individual variation among women.    97  Figure 4-2 Concordance of hemoglobin concentration between HemoCue® and Sysmex methods at baseline  Among women with anemia, 52% (n=244/468) had mild anemia (hemoglobin 110-120 g/L), 45% (n=210/468) had moderate anemia (hemoglobin 80-110 g/L), and 3% (n=14/468) had severe anemia (hemoglobin <80 g/L). Of all anemic women, 77% (n=362/468) had microcytic anemia (hemoglobin <120 g/L and MCV <80 fL), 22% (n=104/468) had normocytic anemia (hemoglobin <120 g/L and MCV 80-98 fL), and the remaining <1% (n=2/468) had macrocytic anemia (hemoglobin <120 g/L and MCV >98 fL).   Overall, 74% of women (n=596/808) had some form of a genetic hemoglobin disorder (either a hemoglobin variant or -thalassemia). A total of 57% (n=457/809) of women had an abnormal hemoglobin variant (E, CS, H, Bart, or F). Hemoglobin E was most common variant (52%, n=420/809), of which 76% (n=319/420) of women were heterozygous and 24% (n=101/420)   98 were homozygous. A total of 42% (n=341/804) of women had one or more of 21 different -gene deletions or point mutations, the majority of which (62%, n=211/341) had 3.7-thalassemia.   99 Table 4-4 Baseline concentrations of nutrition, inflammation and hematological indicators and prevalence rates of iron deficiency, inflammation and anemia in enrolled Cambodian women in Kampong Chhnang by supplement group  Fe MMN Fe+MMN Placebo Total enrolled, n (%) 201 (25) 202 (25) 206 (26) 200 (25) Hematological indicators       Hemoglobin, g/L 116 ± 14 115 ± 12 116 ± 15 117 ± 13       Anemia prevalence, Hb <120 g/L 122/201 (61) 121/201 (60) 123/206 (60) 102/200 (51)   Hepcidin, nmol/L, mean ± SD 7.3 ± 6.8 7.7 ± 6.7 7.4 ± 7.3 7.0 ± 6.3       median (IQR)  5.6 (1.9, 11.0) 6.4 (2.3, 10.9) 5.3 (1.1, 10.7) 5.5 (1.7, 10.8)   Hematocrit, % 36.0 ± 3.7 35.9 ± 3.3 36.3 ± 4.0 36.4 ± 3.3   MCV, fL 75.8 ± 10.4 76.1 ± 9.5 76.4 ± 10.1 77.6 ± 9.3   MCH, ρg 24.4 ± 3.9 24.4 ± 3.5 24.4 ± 3.7 25.0 ± 3.5   RDW, % 15.0 ± 2.5 14.9 ± 2.4 14.9 ± 2.7 14.8 ± 2.9 Genetic hemoglobin disorders       Any (either Hb variant or -thalassemia) 159/201 (79) 148/200 (74) 148/206 (72) 140/200 (70)   Hb variant (E, CS, H, Bart, or F) 127/201 (63) 118/202 (58) 115/206 (56) 111/200 (56)   -thalassemia mutation  91/199 (46) 84/200 (42) 83/205 (41) 83/200 (42) Nutrition indicators        Storage iron, ferritin1, g/L, mean ± SD  54.0 ± 46.1 55.4 ± 45.4 55.1 ± 48.8 49.7 ± 43.7        median (IQR)  37.9 (17.9, 80.0) 41.8 (17.8, 83.4) 39.2 (16.9, 84.9) 37.5 (15.4, 65.7)        ID prevalence, ferritin1 <15 g/L  40/200 (20) 43/202 (21)  47/206 (23) 47/200 (24)        IDA, ferritin1 <15 g/L and Hb <120 g/L 35/200 (18) 38/202 (19) 41/206 (20) 39/200 (20)   Tissue iron, sTfR, mg/L, mean ± SD  7.9 ± 5.8  7.4 ± 5.2 8.1 ± 5.5 7.7 ± 5.5        median (IQR)  6.0 (4.9, 8.3) 5.8 (4.8, 7.7) 6.2 (5.0, 8.9) 5.9 (4.7, 7.7)        ID prevalence, sTfR >8.3 mg/L  50/200 (25) 44/202 (22)  60/206 (29)  44/200 (22)        IDA, sTfR >8.3 mg/L and Hb <120 g/L 46/200 (23) 39/201 (19) 52/206 (25) 38/200 (19)   RBP (vitamin A)1, mol/L 1.63 ± 0.50 1.70 ± 0.54 1.68 ± 0.51 1.69 ± 0.54   100  Fe MMN Fe+MMN Placebo        Vit A deficiency, RBP1 <0.7 mol/L 1/200 (<1) 0/202 (0) 0/206 (0) 0/200 (0)   Vitamin B122, ρmol/L 542 ± 250 606 ± 260 557 ± 278 532 ± 227        B12 deficiency2, B12 <150 ρmol/L 0/99 (0) 1/100 (1) 2/100 (2) 1/100 (1)   Folate, nmol/L3 13.9 ± 6.9 13.4 ± 6.4 13.2 ± 6.6 14.6 ± 7.5        Folate deficiency3, Folate <6.8 nmol/L 11/99 (11) 12/100 (12) 17/100 (17) 10/101 (10) Inflammation markers       CRP, mg/L 0.35 (0.18, 0.88) 0.40 (0.22, 0.92) 0.43 (0.21, 1.01) 0.43 (0.23, 1.05)        Acute inflammation, CRP >5 mg/L 9/200 (5) 10/202 (5) 10/206 (5) 5/200 (3)   AGP, mg/L 0.55 (0.45, 0.72) 0.55 (0.44, 0.70) 0.60 (0.48, 0.75) 0.56 (0.47, 0.69)        Chronic inflammation, AGP >1 g/L 15/200 (8) 19/202 (9) 18/206 (9) 14/200 (7) Total n=809. Values are mean ± SD or median (IQR) or n (%). AGP, -1 acid glycoprotein; CRP, C-reactive protein; Hb, hemoglobin; ID, iron deficiency; IDA, iron deficiency anemia; MCH, mean corpuscular hemoglobin; MCV, mean corpuscular volume; RDW, red cell distribution width; RBP, retinol binding protein; sTfR, soluble transferrin receptor. 1 Values were corrected for inflammation using Thurnham et al. correction factors (68,76). 2 Vitamin B12 analyses were conducted on a randomly selected subset of n=399 women. 3 Folate analyses were conducted on a randomly selected subset of n=400 women.    101 Nutrition, inflammation and hematological indicators were also compared by baseline anemia status (Table 4-5). Prevalence rates of iron deficiency, chronic inflammation, and any genetic hemoglobin disorder were significantly higher in anemic as compared to non-anemic women (P≤0.001). Compared to non-anemic women, anemic women had higher rates of iron deficiency (based on ferritin and sTfR, both P<0.001) and folate deficiency (P=0.05). Hepcidin concentrations were significantly lower in anemic women as compared to non-anemic women (P<0.001). Vitamin B12 and A deficiencies were ≤1% overall, and differences were not observed between anemic and non-anemic women (P=0.54 and P=0.39, respectively). Anemic women had a higher prevalence of chronic inflammation (AGP >1 g/L) as compared to non-anemic women (P<0.001).        102 Table 4-5 Baseline characteristics in enrolled Cambodian women by anemia status Total n=808. Values are mean ± SD or median (IQR) or n (%). AGP, -1 acid glycoprotein; CRP, C-reactive protein; Hb, hemoglobin; ID, iron deficiency; IDA, iron deficiency anemia; MCH, mean corpuscular hemoglobin; MCV, mean corpuscular volume; RDW, red cell distribution width; sTfR, soluble transferrin receptor.  1 Values were corrected for inflammation using Thurnham et al. correction factors (68,76). 2 Vitamin B12 analyses were conducted on a randomly selected subset of n=399 women. 3 Folate analyses were conducted on a randomly selected subset of n=400 women.   Non-anemic (Hb ≥120 g/L) Anemic (Hb <120 g/L) P Total, n (%)  340 (42) 468 (58) -  Hematological indicators      Hepcidin, nmol/L 7.0 (3.8, 12.2) 4.3 (0.6, 9.8) <0.001   Hematocrit, % 39.1 ± 1.9 34.0 ± 2.9 <0.001   MCV, fL 81.5 ± 7.3 72.8 ± 9.9 <0.001   MCH, ρg 26.6 ± 2.6 23.0 ± 3.6 <0.001   RDW, % 13.5 ± 1.3 15.9 ± 2.9 <0.001 Genetic hemoglobin disorders       Any (either Hb variant or -thalassemia) 227/340 (67) 369/468 (79) <0.001    Hb variant (E, CS, H, Bart, or F) 167/340 (49) 304/468 (65) <0.001    -thalassemia mutation 135/340 (40) 206/464 (44) 0.18 Nutrition indicators        Storage iron, ferritin1, g/L  50.9 (30.9, 87.4) 29.9 (11.7, 67.9) <0.001       ID prevalence, ferritin1 <15 g/L  24/339 (7) 153/468 (33) <0.001       IDA, ferritin1 <15 g/L and Hb <120 g/L 0/339 (0) 153/468 (33) <0.001    Tissue iron, sTfR, mg/L  5.1 (4.3, 6.3) 6.9 (5.4, 11.2) <0.001       ID prevalence, sTfR >8.3 mg/L  22/339 (7) 175/468 (37) <0.001       IDA, sTfR >8.3 mg/L and Hb <120 g/L 0/339 (0) 175/468 (37) <0.001    RBP1, mol/L 1.8 ± 0.6 1.6 ± 0.5 <0.001       Vitamin A deficiency1, <0.7 mol/L 0/340 (0) 1/468 (<1) 0.39    Vitamin B122, ρmol/L 572 ± 256 551 ± 253 0.47        Vitamin B122 deficiency, <150 ρmol/L 1/160 (<1) 3/240 (1) 0.54    Folate3, nmol/L 14.9 ± 7.0 13.0 ± 6.6 0.006        Folate deficiency3, <6.8 nmol/L 14/160 (9) 36/240 (15) 0.05 Inflammation markers      CRP, mg/L 0.43 (0.21, 0.98) 0.40 (0.19, 0.91) 0.39       Acute, CRP >5 mg/L 9/339 (3) 24/468 (5) 0.08   AGP, mg/L 0.55 (0.44, 0.68) 0.57 (0.46, 0.74) 0.04       Chronic, AGP >1 g/L 15/339 (4) 50/468 (11) 0.001   103 4.4.2 Adherence and Adverse Side Effects In the Fe, MMN, Fe+MMN, and placebo groups, 86%, 89%, 78%, and 88% of women reported consumption of ≥80% of the capsules, and at 4 weeks, 49%, 53%, 69%, and 44% of women reported adverse side effects, respectively. The proportion of women who were ≥80% adherent differed by treatment group (chi-square, P=0.014), and similarly, the proportion of women who reported adverse side effects at 4 weeks differed by treatment group (chi-square, P<0.001). Adherence appeared lowest and reported side effects highest in the Fe+MMN group. Over the period of the trial, symptoms appeared to subside with less overall women reporting adverse side effects at 8 weeks (24%, n=184/777) than at 4 weeks (54%, n=423/789). The most common reported adverse effects were nausea (38%), stomach cramping (30%), and headache (25%). Other less common effects were tiredness, fever, cough, and difficulty breathing.   4.4.3 Mean Hemoglobin Concentration at 12 Weeks Adjusted mean hemoglobin concentration at 12 weeks did not significantly differ between Fe and Fe+MMN groups; both were higher than MMN and placebo groups (Table 4-6). A generalized linear mixed-effects model was used to predict marginal means (95% CI) for each group with adjustments for baseline values and village clusters. Bonferroni-adjusted pairwise comparisons were made across groups. Data are presented for all women, for anemic women at baseline, and for adherent women (women who consumed ≥80% capsules).     104 Table 4-6 Mean hemoglobin, ferritin, soluble transferrin receptor, hepcidin, and micronutrient concentrations at 12 weeks by supplement group in enrolled Cambodian women in Kampong Chhnang  n Fe  MMN Fe+MMN Placebo Primary outcome: Hb, g/L         Baseline Hb, mean ± SD 808 116 ± 14 115 ± 12 116 ± 15 117 ± 13    12 wk unadjusted Hb, mean ± SD 759 121 ± 12 117 ± 13 123 ± 12 116 ± 13    12 wk adjusted Hb 759 121 (120, 122)a 116 (115, 117)b 123 (122, 124)a 116 (115, 117)b    12 wk adjusted Hb for only anemic   women at baseline 443 121 (120, 122)a 114 (113, 115)b 123 (122, 124)a 113 (112, 114)b    12 wk adjusted Hb for only women ≥80% adherent1 645 121 (120, 123)a 116 (115, 117)b 123 (122, 124)a 116 (115, 117)b Secondary outcomes:         12 wk adjusted ferritin2, g/L 756 89.1 (83.5, 94.7)a 54.5 (49.0, 60.0)b 86.7 (80.9, 92.6)a 51.0 (45.7, 56.3)b    12 wk adjusted sTfR, mg/L 756 6.4 (5.5, 7.2)a 7.3 (6.6, 8.0)b 6.6 (5.9, 7.4)a 7.3 (6.5, 8.2)b    12 wk adjusted hepcidin, nmol/L 755 32.2 (30.9, 33.5)a 18.8 (17.6, 20.1)b 32.2 (30.9, 33.6)a 15.2 (14.1, 16.4)b    12 wk adjusted RBP2, mol/L 756 1.75 (1.70, 1.80)a 1.86 (1.81, 1.92)a 1.83 (1.79, 1.88)a 1.67 (1.62, 1.72)b    12 wk adjusted B12, ρmol/L 399 597 (554, 639)a 610 (573, 647)b 583 (538, 628)a,b 539 (498, 580)a    12 wk adjusted folate, nmol/L 400 12.6 (11.5, 13.6)a 46.4 (45.4, 47.4)b 36.4 (35.3, 37.4)c 14.8 (13.8, 15.9)a Total n=808. Marginal means (95% CI) unless otherwise indicated. Fe, iron; Hb, hemoglobin; MMN, multiple micronutrients; RBP, retinol binding protein; sTfR, soluble transferrin receptor; wk, weeks. A generalized linear mixed-effects model was used to predict marginal means (95% CI) for each group with adjustments for baseline values and village clusters. Bonferroni-adjusted pairwise comparisons were made across groups. Values that do not share a common superscript letter in the row are statistically different (P<0.05).  1 Includes women who consumed ≥80% of capsules based on three capsule counts at 4, 8, and 12 weeks.            2 Values were corrected for inflammation using Thurnham et al. correction factors (68,76).   105 4.4.4 At the Margins 2x2 Factorial Analysis at 12 Weeks At the margins analysis showed the adjusted mean (95% CI) increase in hemoglobin at 12 weeks was significantly higher among all women who received Fe (with or without MMN, n=383) as compared to women who did not receive Fe, but not significantly higher among women who received MMN (with or without Fe, n=382) as compared to women who did not receive MMN (Table 4-7). The interaction between the two interventions was found to be non-significant among all women (coefficient [95% CI] = 0.7 [-1.9, 3.2], P=0.61) and among only anemic women (-0.6 [-4.5, 3.3], P=0.75).    106 Table 4-7 At the margins factorial analysis: adjusted mean increase in hemoglobin concentration at 12 weeks in enrolled Cambodian women in Kampong Chhnang   Received Fe  (with or without MMN)  Received MMN (with or without Fe)        Yes No  Yes  No All women, total n=759            n  383 376  382 377      Unadjusted Hb at 12 wk, g/L, mean ± SD  122 ± 12 116 ± 13  120 ± 13 119 ± 12      Adjusted mean (95% CI) increase in Hb  5.6 (3.8, 7.4)  -  1.1 (-0.7, 2.9) -      P value  <0.001 -  0.24 - Anemic women only, total n=443            n  232 211  232 211      Unadjusted Hb at 12 wk, g/L, mean ± SD  118 ± 12 109 ± 12  114 ± 12 113 ± 13      Adjusted mean (95% CI) increase in Hb  9.0 (6.2, 11.8) -  2.0 (-0.8, 4.8) -      P value  <0.001 -  0.17 - Total n=759. Fe, iron; Hb, hemoglobin; MMN, multiple micronutrients; wk, weeks. A generalized linear mixed-effects model was used to determine mean (95% CI) increase in hemoglobin with adjustments for baseline hemoglobin and village clusters. The interaction between the two interventions was found to be non-significant among all women (coefficient [95% CI] = 0.7 [-1.9, 3.2], P=0.61) and among only anemic women (-0.6 [-4.5, 3.3], P=0.75).    107 We also determined the mean (95% CI) increase in hemoglobin at 12 weeks comparing each of the Fe, MMN, and Fe+MMN groups to the placebo group as the reference (Table 4-8). Differences in the mean increase in hemoglobin at 12 weeks were statistically significant in the Fe and Fe+MMN groups (5.6 [3.8, 7.4] g/L, P<0.001, and 7.3 [5.5, 9.2] g/L, P<0.001, respectively), but was not significant in the MMN group (1.1 [-0.7, 2.9] g/L, P=0.24). Post-estimation linear comparisons of the beta coefficients from the adjusted model showed that the addition of MMN to Fe did not significantly increase hemoglobin at 12 weeks among all women (n=759, P=0.07) or among only anemic women (n=443, P=0.39); hence, there was no significant additive effect of MMN.   108 Table 4-8 Adjusted mean increase in hemoglobin concentration at 12 weeks for each treatment group compared to placebo in enrolled Cambodian women in Kampong Chhnang  Fe   MMN  Fe+MMN All women, n=759      Adjusted mean (95% CI) increase in Hb2, g/L  5.6 (3.8, 7.4)  1.1 (-0.7, 2.9)  7.3 (5.5, 9.2)       SE 0.93  0.93  0.92      P value <0.001  0.24  <0.001 Anemic women only, n=443     Adjusted mean (95% CI) increase in Hb2, g/L   9.0 (6.2, 11.8)   2.0 (-0.8, 4.8)  10.3 (7.5, 13.1)      SE  1.44  1.44    1.43     P value  <0.001   0.17   <0.001 Total n=759. Fe, iron; Hb, hemoglobin; MMN, multiple micronutrients. A generalized linear mixed-effects model was used to determine mean (95% CI) increase in hemoglobin concentration for each treatment group with adjustments for baseline hemoglobin and village clusters. Post-estimation linear comparisons of the beta coefficients showed that the addition of MMN to Fe did not significantly increase hemoglobin at 12 weeks among all women (n=759, P=0.07) or among only anemic women (n=443, P=0.39); hence, there was no significant additive effect of MMN.   109 4.4.5 Ferritin, STfR, Hepcidin, RBP, Vitamin B12 and Folate at 12 Weeks Table 4-6 also includes secondary outcomes: ferritin, sTfR, hepcidin, RBP, vitamin B12, and folate concentrations at 12 weeks by group. Adjusted mean ferritin and hepcidin concentrations at 12 weeks did not significantly differ between Fe and Fe+MMN groups; both were higher than MMN and placebo groups. Similarly, adjusted mean sTfR concentration at 12 weeks did not significantly differ between Fe and Fe+MMN groups; both were lower than MMN and placebo groups. We did not investigate the risk of iron overload but note at baseline, overall 5% (n=41/809) of women had ferritin concentrations >150 g/L, of which 60% were in the Fe and Fe+MMN groups, and at 12 weeks, 8% (n=64/767) had ferritin concentrations >150 g/L, of which 75% were in the Fe or Fe+MMN groups.   Adjusted mean RBP concentrations at 12 weeks did not significantly differ between Fe, MMN, or Fe+MMN groups; all were higher than the placebo group. Mean vitamin B12 concentrations at 12 weeks did not significantly differ between MMN and Fe+MMN groups; only MMN was higher than the placebo group. Mean folate concentrations at 12 weeks were significantly higher in the MMN than the Fe+MMN group; both were higher than the Fe and placebo groups.   4.4.6 Proportions of Women as Hemoglobin Responders (≥10 g/L) at 12 Weeks We compared the proportions of women defined as hemoglobin responders (hemoglobin increase of ≥10 g/L at 12 weeks) across groups for all women, and among women subgrouped by baseline anemia status, by iron status (ferritin), by adherence, and by the presence or absence of a genetic hemoglobin disorder (Table 4-9). Overall, the proportions of hemoglobin responders were higher in the Fe and Fe+MMN groups (19% and 30%) as compared to MMN and placebo   110 (8% and 5%), respectively. Among anemic women (n=443), the proportions of hemoglobin responders were substantially higher in the Fe and Fe+MMN groups (31% and 42%) as compared to MMN and placebo (12% and 6%), respectively.   A similar trend was observed among women with depleted iron stores (ferritin <15 g/L) at baseline (n=171). We note that only 21% (n=171/808) of women had depleted iron stores at baseline (ferritin <15 g/L). Of these women, only 64% (n=25/39) and 70% (n=32/46) had a hemoglobin response ≥10 g/L at 12 weeks in the Fe and Fe+MMN groups, respectively. Further, we found that iron-depleted hemoglobin non-responders had significantly higher mean baseline concentrations of hepcidin as compared to iron-depleted hemoglobin responders in the Fe (n=39, 1.75 vs. 0.54 nmol/L, P=0.001) and Fe+MMN (n=46, 1.15 vs. 0.54 nmol/L, P=0.04, Wilcoxon rank-sum test) groups, respectively. There were no significant differences between hemoglobin responders and non-responders in baseline AGP (P=0.27 and P=0.66 for the Fe and Fe+MMN groups, respectively) or CRP concentrations (P=0.12 and P=0.5 for the same two groups, respectively). In addition, the iron-depleted hemoglobin non-responders had significantly lower baseline sTfR (P=0.05) and higher hemoglobin (P<0.001) and ferritin (P=0.009) concentrations in the Fe group, and significantly higher baseline hemoglobin (P=0.003) and ferritin (P=0.003) concentrations in the Fe+MMN group, as compared to iron-depleted hemoglobin responders.  We also compared proportions of women defined as hemoglobin responders (≥10 g/L at 12 weeks) by the presence or absence of a genetic hemoglobin disorder. Overall, 74% (n=596/808) of women had one or more genetic hemoglobin disorders of any kind. The proportion of hemoglobin responders was substantially lower among women with a genetic   111 hemoglobin disorder in the Fe and Fe+MMN groups, as compared to women with no hemoglobin disorder (Fe: 15% vs. 38%, Fe+MMN: 23% vs. 48%, respectively); there were no significant differences in the proportions of hemoglobin responders and non-responders observed in the MMN and placebo groups (MMN: 9% vs. 8%, placebo: 5% vs. 5%, respectively).  We defined a hemoglobin response a priori to the initiation of the trial as an increase of hemoglobin ≥10 g/L at 12 weeks. However, of interest, we calculated the proportions of hemoglobin responders using two less conservative definitions: i) an increase of hemoglobin ≥5 g/L at 12 weeks, and ii) an increase of hemoglobin ≥10 g/L or resolved anemia (hemoglobin >120 g/L) at 12 weeks. Among all women (n=759), the proportions of women with a hemoglobin increase ≥5 g/L at 12 weeks in the Fe, MMN, Fe+MMN, and placebo groups were 42%, 30%, 50%, and 19%, and the proportions of women with a hemoglobin increase ≥10 g/L or with resolved anemia (hemoglobin >120 g/L) at 12 weeks were 58%, 42%, 65%, and 41%, respectively. Among only anemic women at baseline (n=443), the proportions of women with a hemoglobin increase ≥5 g/L at 12 weeks in the Fe, MMN, Fe+MMN, and placebo groups were 58%, 41%, 64%, and 25% and the proportions of women with a hemoglobin increase ≥10 g/L or with resolved anemia (hemoglobin >120 g/L) at 12 weeks were 44%, 22%, 50%, and 13%, respectively.    112 Table 4-9 Proportion of enrolled Cambodian women in Kampong Chhnang defined as hemoglobin responders (increase ≥10 g/L at 12 weeks) by supplement group  n Fe MMN Fe+MMN Placebo All women         Hb increase ≥10 g/L  759 37/191 (19) 16/190 (8) 58/192 (30) 9/186 (5) Subgroup of women by anemia status         No anemia at baseline (Hb ≥120 g/L) 316 1/76 (1) 2/75 (3) 9/75 (12) 3/90 (3)    Anemia at baseline (Hb <120 g/L) 443 36/115 (31) 14/115 (12) 49/117 (42) 6/96 (6)    Moderate or severe anemia at baseline (Hb <110 g/L) 214 27/56 (48) 4/49 (8) 32/59 (54) 4/50 (8) Subgroup of women by storage iron status         Iron-replete stores at baseline (ferritin ≥15 g/L) 587 12/151 (8) 12/150 (8) 26/146 (18) 7/140 (5)    Depleted iron stores at baseline (ferritin <15 g/L) 171 25/39 (64) 4/40 (10) 32/46 (70) 2/46 (4) Subgroup of women by adherence         Only women ≥80% adherent1  645 33/163 (20) 14/169 (8) 46/150 (31) 8/163 (5)    Only women <80% adherent 114 4/28 (14) 2/21 (10) 12/42 (29) 1/23 (4) Subgroup of women by the presence or absence of a genetic hemoglobin disorder          No genetic hemoglobin disorder  212 15/40 (38) 4/51 (8) 27/56 (48) 3/57 (5)    Genetic hemoglobin disorder (Hb variant or -thalassemia)2 596 22/151 (15) 12/138 (9) 31/136 (23) 6/130 (5) Total n=759. Values are n (%). Fe, iron; Hb, hemoglobin; MMN, multiple micronutrients.  1 Includes women who consumed ≥80% of capsules based on three capsule counts at 4, 8, and 12 weeks.    2 Includes women with one or more genetic hemoglobin disorders of any kind (women with a Hb variant [E, CS, H, Bart, or F] or one or more of 21 different -globin gene deletions and point mutations [-thalassemia]).      Among all women (n=759), the proportions of women with a hemoglobin response were higher in the Fe and Fe+MMN groups as compared to placebo (both P<0.001); no differences were observed in the proportions of women with a hemoglobin response in the MMN and placebo groups (P=0.16). Statistical comparisons were not made across subgroups due to small sample sizes.  113 4.5 Conclusions and Discussion In this 2x2 factorial double-blind randomized trial, we observed significantly higher hemoglobin concentrations among non-pregnant Cambodian women of reproductive age after 12 weeks of daily oral Fe and Fe+MMN, as compared to MMN and placebo groups. Mean hemoglobin at 12 weeks did not differ between Fe and Fe+MMN groups, and post-estimation linear comparisons showed that the addition of MMN to Fe did not confer a significant increase in hemoglobin. Further, marginal analysis showed that Fe (with or without MMN) significantly increased hemoglobin at 12 weeks, but MMN (with or without Fe) did not.    Despite screening for anemic women using the HemoCue® and an inclusion criterion of hemoglobin ≤117 g/L, the automated hematology analyzer (considered a more accurate and precise method of hemoglobin measurement) later showed that only 58% of women were anemic (hemoglobin <120 g/L) at baseline. This discrepancy may, in part, be due to regression to the mean, and could also be a result of differences in hydration status and/or the biological differences between capillary and venous blood. Non-fasting HemoCue® screening was conducted at any time of the day and fasting venous collection was conducted in the morning, and measurements were taken one week apart. Regardless, we were adequately powered to allow stratification of results by anemia status.  Among anemic women, 31% and 42% had a hemoglobin response ≥10 g/L at 12 weeks in the Fe and Fe+MMN groups, respectively. This suggests that approximately one third of the anemia burden in our study population was due, at least partially, to iron deficiency. The remaining two thirds is likely attributable to the high prevalence of genetic hemoglobin disorders (57% had a   114 hemoglobin variant and 42% had -thalassemia), and, to a small extent in this study population, to inflammation and/or infection. Also of note, 21% of women had depleted iron stores at baseline (ferritin <15 g/L). Of these women, only 64% and 70% had a hemoglobin response ≥10 g/L at 12 weeks in the Fe and Fe+MMN groups, respectively. It is surprising that one third of women with depleted iron stores at baseline were non-responsive, especially considering that the supplement contained a substantial amount of iron (60 mg). Adherence is not likely a factor contributing to these observed differences, as no distinct trends were observed in the proportions of hemoglobin responders across groups when women were divided into those who were adherent or non-adherent. Further, adherence rates were similar among these women who were receiving Fe (with or without MMN) in the responsive and non-responsive groups: 77% and 75% of women consumed ≥80% capsules in these groups, respectively. We queried whether these iron-depleted non-responders had high levels of serum hepcidin concentrations, thus potentially inhibiting absorption of the oral iron supplements. Interestingly, we found that their mean baseline concentrations of hepcidin were significantly higher than those of iron-depleted hemoglobin responders, in both the Fe and Fe+MMN groups. We suspected that higher baseline hepcidin concentrations observed in the non-responders might be attributed to higher levels of inflammation in the non-responder group. This is because inflammation stimulates the production of hepcidin, which binds to and degrades ferroportin, a transport protein on the wall of the macrophage, sequestering iron in the macrophage and making it unavailable for erythropoiesis (16,74). This leads to increased ferritin concentrations (storage iron) (75) and inflammatory biomarkers such as CRP and AGP. However, baseline AGP and CRP concentrations did not differ between hemoglobin responders and non-responders in the Fe and Fe+MMN groups. This suggests the differences observed in baseline hepcidin concentrations in   115 the responders and non-responders may not be due to inflammation, at least not on inflammation defined by elevated AGP and CRP concentrations. In addition, the iron-depleted hemoglobin non-responders had significantly lower baseline sTfR and higher hemoglobin and ferritin concentrations in the Fe group, and significantly higher baseline hemoglobin and ferritin concentrations in the Fe+MMN group, as compared to iron-depleted hemoglobin responders. We conclude that the observed differences in baseline hepcidin concentrations in the hemoglobin responders and non-responders did not appear to be associated with inflammation levels (at least, as defined by AGP and CRP biomarkers), and did not appear to be associated with the prevalence of genetic hemoglobin disorders.   It is not surprising that among the subgroup of women with any genetic hemoglobin disorder (Hb variant or -thalassemia) who received Fe (with or without MMN), we observed lower proportions of women with a hemoglobin response (≥10 g/L at 12 weeks) as compared to women without a hemoglobin disorder (Fe: 15% vs. 38%, Fe+MMN: 23% vs. 48%, respectively). Of note, there were no significant differences in the proportions of hemoglobin responders and non-responders observed in the MMN and placebo groups (MMN: 9% vs. 8%, placebo: 5% vs. 5%, respectively). Among women with a hemoglobin disorder who received Fe (with or without MMN), it is likely that the anemia is, at least in part, due to the genetic disorder itself. But the fact that we saw differences in hemoglobin response among hemoglobin responders and non-responders in the Fe and Fe+MMN groups but not in the MMN and placebo groups (who did not receive Fe), suggests that some of the anemia was also likely due to iron deficiency. It is likely that some women have both a genetic hemoglobin disorder and iron deficiency, as only some   116 genetic hemoglobin disorders (e.g., -thalassemia major) cause elevated ferritin concentrations and iron overload.  For simplicity, we subgrouped women together if they had any type of genetic hemoglobin disorder. We acknowledge that some genetic hemoglobin disorders have a more severe phenotype than others; as such, combining women with any disorder into one group limited our ability to detect the independent effect of each genotype on hemoglobin response. Given the diverse heterogeneity of genotypes among our n=809 women, it is difficult to statistically analyze these groups independently given both the co-inheritance and rarity of multiple genotypes, thus limiting the statistical power for independent analyses. In our next steps (outlined in section 5.4), we will investigate this further using receiver operating characteristic (ROC) analyses to determine the specificity and sensitivity of numerous hematological indicators to predict hemoglobin response among women with and without hemoglobin disorders.   Overall, only ~25% (n=95/383) of all women who received Fe (with or without MMN) responded to iron. In the subgroup of women determined as anemic at baseline based on hemoglobin measurement by the Sysmex analyzer, ~37% (n=85/232) of anemic women who received Fe (with or without MMN) responded to iron. As we recruited women with hemoglobin ≤117 g/L (based on the HemoCue®), we predict that even fewer women would respond to iron in the wider population, based on our definition of a hemoglobin response (≥10 g/L) and assuming that the prevalence of anemia among the women we screened was similar to that of the country as a whole. The outcomes in our study were observed using a daily 60 mg iron dose, coupled with deworming treatment at baseline, regular follow-up, and a comprehensive   117 community-level education strategy. Hence, benefits of a weekly iron and folic acid supplementation program (weekly 60 mg iron dose), considering previous reports of poor access and adherence to these programs in Cambodia (163), would likely result in a lower effect size of anemia reduction than observed in our controlled study.    We acknowledge that the response to oral iron would vary among populations based on the prevalence of iron deficiency, inflammation, and other micronutrient deficiencies potentially contributing to anemia, as well as the effectiveness of communication strategies, and access and adherence to supplements (126). In our study, adherence was high and access was not an issue. Baseline deficiencies of vitamin A (based on RBP concentrations) and B12 were both <1% among women and biochemical evidence of inflammation was relatively low among women in Kampong Chhnang (5% of women had acute inflammation and 8% had chronic inflammation) as compared to other findings among Cambodian women of reproductive age (e.g., 10% of women had acute inflammation and 36% had chronic inflammation in the Wieringa et al. national nutrition survey among n=470 women (164)).   Overall, baseline iron deficiency prevalence was higher in Kampong Chhnang, as compared to previous reports in Prey Veng (22) and a nationally representative survey (154); however, our population was recruited based on hemoglobin ≤117 g/L using HemoCue®. As such, we expected lower ferritin concentrations in the study population. Overall, inflammation-adjusted mean ± SD ferritin concentrations were substantially higher among unscreened women in Prey Veng (95 ± 57 g/L, n=420) (22) than women screened with hemoglobin concentrations ≤117   118 g/L in Kampong Chhnang (54 ± 46 g/L, n=808). Serum ferritin was analyzed in the same lab using the same s-ELISA method (86) (VitMin Lab, Germany) in both surveys.   Overall, baseline serum vitamin B12 concentrations in a randomly selected subset of n=399 women were surprisingly high (mean ~560 ρmol/L). Based on the cut-off of <150 ρmol/L (165), B12 deficiency was <1%. These results were surprising as we thought the traditional Cambodian diet was low in animal-source foods. Unfortunately we did not collect dietary intake data that would be required to investigate this further.   We did observe some biochemical evidence of folate deficiency (12.5%, n=50/400 based on serum folate <6.8 nmol/L); however, we note that there was no evidence of megaloblastic anemia (anemia caused by folate or B12 deficiency, characterized by a low hemoglobin and a high MCV concentration (9)). Overall, less than 1% (n=2/468) of anemic women had a high MCV (>98 fL). Wieringa et al. also recently reported evidence of folate deficiency (17.8%, based on serum folate <10 nmol/L) among Cambodian mothers of reproductive age (15-49 years) in a nationally representative survey (154). Wieringa et al. used a higher cut-off value to indicate serum folate deficiency than us (10 vs. 6.8 nmol/L, respectively), which may have contributed to the higher prevalence of deficiency observed in the Wieringa et al. study (154). These results are surprising, as our survey in Prey Veng showed <1% deficiency prevalence among non-pregnant women based on serum folate concentrations <6.8 nmol/L. This evidence of folate deficiency is interesting and requires further follow up, as it is important for women of reproductive age (who may become pregnant) to have optimal folate status, as folate is required for neural tube defect prevention in infants during the first trimester of pregnancy (166). The   119 collection and analysis of dietary intake data would also be interesting to confirm if the traditional Cambodian diet was low in folate, given the findings on the biochemical evidence of folate deficiency among women.  Strengths of this research are the rigorous factorial design and comprehensive assessment of numerous factors related to anemia (nutritional, hematological [including hemoglobin genotype] and inflammation-related) among 809 predominantly anemic non-pregnant women. The communication strategies and frequent follow-up by our research staff and the Ministry of Health resulted in high retention and adherence rates. Limitations are that we did not measure infection (e.g., helminth) or menstrual blood loss, both of which can be associated with anemia and iron stores. We recruited women via a convenience sampling method from only one province in Cambodia, which limits the generalizability of our findings. Lastly, we used oral supplements, which may have underestimated the magnitude of the hemoglobin response among women in our study. Bregman et al. (124) have shown that hemoglobin response can vary by route of iron administration and that a lack of hemoglobin response to oral therapy does not always rule out iron deficiency anemia. Notwithstanding this, the aim of our study was to measure hemoglobin response to oral iron capsules, as this route of administration is the standard of practice.   We conclude that based on our estimates that suggest <10% of non-pregnant women in the general population would likely benefit from daily iron supplementation, the justification of the weekly iron and folic acid supplementation program in Cambodia is in question. More work is needed to investigate the potential risk of untargeted blanket iron supplementation among   120 predominantly anemic yet iron-replete populations with a high prevalence of co-existing inflammation, infection and/or genetic hemoglobin disorders.   121 Chapter 5: Conclusions, Discussion, and Future Research Directions In this section I review the key findings of my research conducted in Prey Veng and Kampong Chhnang provinces and elaborate on the comparability of my results to the published literature. I discuss the overall strengths and limitations of the research design in each study. This is followed by a summary of the significance and overall contribution of this research to the current published literature. I conclude with potential future research directions.  5.1 Discussion of Key Findings 5.1.1 High Serum Ferritin Concentrations in Women  The most surprising finding of the Prey Veng survey conducted in 2014 was the extremely low prevalence of iron deficiency among non-pregnant women (2%, n=9/420) based on ferritin concentration (<15 µg/L). Median (IQR) inflammation-adjusted ferritin concentration in these women was 91 (59, 143) µg/L. Similar findings of high ferritin concentrations and a low prevalence of iron deficiency (based on inflammation-adjusted ferritin) were later confirmed in 2015 in a nationally representative survey in Cambodia (8.1% prevalence, median [IQR] ferritin concentration: 64 [35, 101] µg/L, n=2,112 non-pregnant women 15-39 years) (154). The low prevalence of iron deficiency was also confirmed in the 2014 Demographic and Health Survey that was released in late 2015 (2.6% prevalence, n=485 mothers 15-49 years with at least one child) (7). Ferritin concentrations among women enrolled in our trial in Kampong Chhnang were lower (n=808, median [IQR]: 39 [17, 79] µg/L) as compared to the Prey Veng survey (22) and the nationally representative survey (154); however, women in Kampong Chhnang were screened and recruited based on hemoglobin ≤117 g/L using HemoCue®. As such, we expected lower mean ferritin concentrations among these women in Kampong Chhnang.   122 Based on self-reported data from survey questionnaires, the women in Prey Veng were not taking iron supplements, using iron cooking pots, or consuming iron-fortified fish or soy sauce. Data from 24-hr dietary recalls suggest that women in Prey Veng did not have high intakes of dietary iron (estimated iron intakes of ~10-11 mg/day), however, the author suspected under-reporting of overall dietary intake among women, which could mean that the actual dietary iron intakes of women were higher than those reported (142). The estimated average requirement cut-point method and probability approach were applied to assess the prevalence of inadequate iron intake, which showed that the prevalence of inadequate iron intake among these women was low (6-8%) based on the Institute of Medicine distribution percentiles (18% bioavailability of iron). However, the prevalence of inadequate iron intakes increased to 43-50% when calculated based upon 10% bioavailability of the dietary iron (142). The author of this dietary intake study noted that the lack of Cambodian food composition data and the lack of evidence on the bioavailability of iron in the Cambodian diet makes the interpretation of these dietary intake results challenging (142). In conclusion, despite the limitations of this data, we believe that the women in Prey Veng did not have high intakes of dietary iron.  Our findings in Prey Veng showed that inflammation-adjusted mean ferritin concentrations among women with the homozygous hemoglobin E disorder were significantly higher as compared to women with no genetic hemoglobin disorder (129 vs. 96 µg/L, respectively, P<0.05). Further, the homozygous hemoglobin E disorder was associated with a 50% (95% CI: 14%, 96%) higher mean ferritin concentration, as compared to women with a normal hemoglobin structure. This, in part, could explain at least some of the reason for high ferritin concentrations among the ~7.4% (n=31) of women who had the homozygous hemoglobin E   123 disorder. We also showed that chronic inflammation was associated with increased ferritin concentrations. A 1 SD increase in AGP concentration was associated with a 20% (95% CI: 12%, 20%) higher mean ferritin concentration. However, the increase in ferritin as a result of inflammation was accounted for with the use of correction factors (using AGP and CRP biomarkers). In Prey Veng we showed that the study-generated correction factors for all 450 women were similar to the Thurnham et al. (76) correction factors for incubation and late convalescence, and tended to be higher only in early convalescence, but the Thurnham et al. factors still fell within the 95% CI of our study-generated correction factors. We did not observe any differences in adjusted ferritin concentrations after correcting for inflammation using both methods. This is an interesting finding as Thurnham et al. correction factors were generated based on studies of apparently healthy individuals in the absence of overt disease. Thus, we conclude that inflammation was not likely a factor contributing to the high ferritin concentrations in women because we used appropriate correction factors to adjust ferritin and account for the inflammation.  Our exploratory research showed that iron in groundwater in Prey Veng is a possible factor contributing to the higher than expected ferritin concentrations observed among women. Based on a daily consumption of 3 L groundwater, this would equate to a dietary intake of ~4.3 mg iron, ranging from 0.4 to 15.6 mg. Depending on bioavailability, these quantities of iron could potentially increase iron intake over time.   We queried whether the analytical methods used to measure serum ferritin concentrations were accurate. In both the survey in Prey Veng and the trial in Kampong Chhnang, serum ferritin was   124 analyzed by the same s-ELISA at the Erhardt laboratory in Germany (86). Our ferritin method comparison study showed that despite substantial differences in ferritin concentrations among Cambodian women between the s-ELISA and AxSYMTM automated analyzer, iron deficiency prevalence was relatively similar and low (2.1% vs. 6.7%, respectively) across both methods. Based on this data, we infer that the analytical methods may have contributed to the observed high ferritin concentrations but had little impact on our prevalence estimates of iron deficiency.  Although there are many factors that could be contributing to high ferritin concentrations in women in Prey Veng, the exact cause is unknown, especially among women with no genetic hemoglobin disorder, inflammation, or known disease. The iron in groundwater (and perhaps soil) may be a contributing factor depending on the bioavailability of the iron in groundwater (which has not yet been studied in Cambodia). Measurement of the bioavailability of iron in groundwater is complicated, as both diet composition and an individual’s iron status can influence bioavailability, and it is currently thought that an individual’s iron status is the key factor that determines bioavailability (167).   Of interest, recent surveys in the past year around the globe have shown a surprisingly low prevalence of iron deficiency (≤8% based on ferritin <12-15 μg/L) among women of reproductive age in Vietnam (146), Bangladesh (155), Nepal (156), the Democratic Republic of the Congo (157), and Sierra Leone (158). This is important and requires further follow up, as if iron deficiency is not a major cause of anemia, then national policies and programs for anemia reduction may need to be re-evaluated, especially among populations with a high prevalence of genetic hemoglobin disorders and inflammation.    125 5.1.2 High Soluble Transferrin Receptor Concentrations in Women Among non-pregnant women of reproductive age in Prey Veng, the median (IQR) sTfR concentration was 6.4 (5.4, 7.6) mg/L. The prevalence of iron deficiency based on sTfR (>8.3 mg/L) was 18.8% (n=79/420). Wieringa et al. reported lower sTfR concentrations than us, in a nationally representative survey of reproductive-aged women in Cambodia (9.3% prevalence, median [IQR] sTfR concentration: 5.5 [4.7, 6.7] µg/L, n=2,112 non-pregnant women 15-39 years) (154). Conversely, the 2014 Demographic and Health Survey showed a substantially higher prevalence of iron deficiency based on sTfR (33.9%, n=485 mothers 15-49 years with at least one child) (7). Median (IQR) concentrations of sTfR were not reported in the latter survey. STfR concentrations among women enrolled in our trial in Kampong Chhnang were 5.6 (4.8, 8.2) µg/L and the prevalence of iron deficiency among women based on sTfR was 24.5%, however, of note, these women were screened and recruited based on hemoglobin ≤117 g/L using HemoCue®, hence, comparisons between these populations should be made with caution. This is, however, surprising as the median sTfR concentration of women in our trial (with hemoglobin ≤117 g/L using HemoCue®) was similar to the value observed in the Wieringa et al. nationally representative survey (154).  We showed that sTfR concentrations among women with both the heterozygous and homozygous hemoglobin E disorder and the heterozygous CS trait in Prey Veng were significantly higher as compared to women with no genetic hemoglobin disorder (P<0.05). This, in part, could explain some of the reason for high ferritin concentrations among the 23% (n=102) of women who had one of these disorders. We also showed that chronic inflammation was associated with increased sTfR concentrations. A 1 SD increase in AGP concentration was   126 associated with an 11% (95% CI: 8%, 15%) higher mean sTfR concentration. This is contrary to most published literature which reports that sTfR is largely unaffected by levels of inflammation (2). However, in individuals with inflammation specifically caused by infection (e.g., malaria, helminth, or hemolytic conditions), sTfR concentrations can be elevated (152). Unfortunately, we did not directly measure the level of infection among individuals in our study, but we note that malaria prevalence is reported as very low in Prey Veng (153).   In summary, iron deficiency based on sTfR ranged from 9.3% to 33.9% in Cambodian women of reproductive age in our survey in Prey Veng, the national survey by Wieringa et al. (154), and the most recent national Demographic and Health Survey. Analytical methods for measuring sTfR concentrations were conducted at the same laboratory in Germany for all three aforementioned surveys, as such, it is unlikely that the observed differences are a consequence of differences in sTfR analytical methods. Interestingly, in the national survey by Wieringa et al. (154), it was reported that the prevalence of iron deficiency based on sTfR ranged from 5.2% to 18.0% based on different geographical regions in Cambodia. The lowest prevalence rate (5.2%) was observed in the Southwest region, whereas the highest rate (18.0%) was observed in the North (154). The reason for the variation in sTfR concentrations across geographical regions is largely unknown but we speculate it could be a consequence of many factors that would affect iron metabolism (e.g., access to and availability of iron-rich foods or groundwater, access to medical services [including iron and folic acid supplements], or the prevalence of disease and/or inflammation).    127 5.1.3 Discrepancy in Iron Biomarkers used to Estimate Iron Deficiency Prevalence  As discussed in the previous sections, ferritin concentrations among women in Prey Veng suggested a very low rate of iron deficiency, whereas sTfR concentrations suggested higher rates of iron deficiency. Iron deficiency among non-pregnant women in Prey Veng was 2.1% based on ferritin and 18.8% based on sTfR. This perplexing discrepancy in prevalence estimates was also observed in the 2014 Demographic and Health Survey (7) (2.6% based on ferritin and 33.9% based on sTfR), yet, only slight differences were observed in the Cambodian national survey (164) (8.1% based on ferritin and 9.3% based on sTfR) and at baseline in the trial in Kampong Chhnang (21.9% based on ferritin and 24.5% based on sTfR). Of interest, a similar discrepancy in iron biomarkers was also observed in women of reproductive age 15-49 years (n=300) in the Democratic Republic of the Congo: iron deficiency prevalence was 5.3% (95% CI: 3.1%, 8.5%) based on ferritin and 20.7% (95% CI: 16.2%, 25.7%) based on sTfR, using the same cut-offs (157).   Typically in iron deficiency anemia, there is a sequential progression from depletion of iron stores (low ferritin), to a compromised supply of iron to the tissues and red blood cells (elevated sTfR), to a reduction in red blood cell production, resulting in anemia (low hemoglobin) (9). Based on this model, one would expect the prevalence of iron deficiency based on ferritin to exceed the prevalence of iron deficiency based on sTfR, which in turn would exceed the prevalence of iron deficiency anemia. Yet, in several of the aforementioned surveys (including our Prey Veng and Kampong Chhnang surveys, the Wieringa et al. survey, and the 2014 Demographic and Health Survey) the prevalence of iron deficiency based on sTfR exceeded the prevalence of iron deficiency based on ferritin. However, another important consideration to note   128 is that when the anemia is caused by chronic disease or inflammation (16,51), iron is sequestered in the macrophage (due to the binding and internalization of ferroportin) and the iron supply to the red blood cells is compromised (elevated sTfR), but ferritin concentrations may be normal to high as a result of inflammation (51). This is also known as functional iron deficiency. We suspected that this might be a factor contributing to the discrepancy observed between these two iron biomarkers. One would expect functional iron deficiency to be more apparent in a population with a high prevalence of chronic disease and/or inflammation. In Prey Veng, the prevalence of chronic inflammation among non-pregnant women was 25.5% (n=107/420). However in Kampong Chhnang, the prevalence of chronic inflammation was much lower (8.2%, n=66/809) as compared to Prey Veng. Despite the differences in the prevalence of chronic inflammation in these two populations, for those women who had biochemical evidence of acute and/or chronic inflammation, ferritin values were adjusted accordingly using correction factors, so the occurrence of functional iron deficiency wouldn’t appear to be a logical explanation for the discrepancy among women in either province.   The question that remains is whether or not these elevated ferritin and sTfR concentrations reflect true iron status. In individuals with the hemoglobin E homozygous genotype, elevated serum ferritin or sTfR may not be a result of changes in iron status. Whether or not ferritin and/or sTfR concentrations in our study are reflective of iron deficiency cannot be ascertained from the data, as bone marrow biopsies were not conducted. However, based on the known risk of iron overload among those with the hemoglobin E homozygous genotype (18), it seems more likely that elevated sTfR concentrations do not reflect true iron deficiency among Cambodian   129 women with the hemoglobin E homozygous genotype and chronic inflammation. More work is warranted to investigate other reasons for this occurrence.   5.1.4 Hemoglobin Response to Iron Therapy As we have observed in the previous section, the reliance on iron biomarkers to estimate iron deficiency anemia may not be accurate in Cambodia as a large discrepancy between prevalence rates based on ferritin and sTfR has been observed in some studies, and further, these iron biomarkers may be confounded by other factors, such as inflammation, infection, and genetic hemoglobin disorders. As such, we measured the hemoglobin response to iron, as the only true way to assess the prevalence of iron deficiency anemia. In our trial in Kampong Chhnang, 31% and 42% of anemic women had a hemoglobin response ≥10 g/L at 12 weeks in the Fe and Fe+MMN groups, respectively. This suggests that approximately one third of the anemia burden in our study population was due, at least partially, to iron deficiency. The remaining two thirds is likely attributable to the high prevalence of genetic hemoglobin disorders (58% had a hemoglobin variant and 42% had -thalassemia), and, to a small extent in this study population, to inflammation and/or infection. In fact, among the subgroup of women with genetic hemoglobin disorders (Hb variants or -thalassemia) who received Fe (with or without MMN), we observed lower proportions of women with a hemoglobin response (≥10 g/L at 12 weeks) as compared to women without a hemoglobin disorder (Fe: 15% vs. 38%, Fe+MMN: 23% vs. 48%, respectively). This also suggests that genetic hemoglobin disorders are a plausible cause of anemia among some women and explain, at least in part, some of the observed non-response. Based on our estimates that suggest less than 10% of non-pregnant women in the general   130 population would likely benefit from daily iron supplementation, the justification of the weekly iron and folic acid supplementation program in Cambodia is in question.   We do not doubt that weekly iron and folic acid supplementation programs would be efficacious in iron-deficient populations with adequately resourced programs and communication strategies; yet, we emphasize that the impact of these blanket supplementation programs likely varies by population and context. The response to oral iron would vary among populations based on the prevalence of iron deficiency, inflammation, and other micronutrient deficiencies potentially contributing to anemia, as well as the effectiveness of communication strategies, and access and adherence to supplements (126).   A recent systematic review by Low et al. concluded that there is evidence from ten trials (including 3,273 non-pregnant menstruating women [12-50 years]) that suggest daily oral iron supplementation reduces the prevalence of anemia (RR: 0.39 [95% CI: 0.25, 0.60]) (168). These ten trials included all women of menstruating age (irrespective of anemia or iron status) and at varying doses and durations of iron therapy (between 4-12 weeks). The evidence in this systematic review was used to inform the recent 2016 WHO guidelines recommending daily iron and folic acid supplementation (including 30-60 mg elemental iron) among menstruating women and adolescents girls for three consecutive months of each year in areas of anemia prevalence ≥40% (128). The evidence in the review fails to address potential risks of iron supplementation (e.g., oxidative stress, lipid peroxidation, iron overload, or all-cause mortality) (168), as no studies included in the review measured risk-related outcomes other than the adverse gastrointestinal side effects of iron supplementation (e.g., constipation or abdominal pain) and   131 the studies were not of sufficient duration to investigate all-cause mortality. It would certainly be justified to include potential risk-related outcomes in these meta-analyses to better inform policy. We did not investigate the risk of iron overload in the trial in Kampong Chhnang but we report that 8% (n=64/767) of women had ferritin concentrations >150 g/L at 12 weeks, of which 75% were in the Fe or Fe+MMN groups.   In Vietnam, Pasricha et al. conducted an iron supplementation trial among rural non-pregnant Vietnamese women of reproductive age (n=221) to measure the change in hemoglobin concentration after 12 weeks of weekly oral iron and folic acid supplementation (60 mg ferrous sulphate and 400 µg folic acid) and deworming treatment (84). Only 37% of women (n=81/221) were determined to be anemic (hemoglobin <120 g/L) at baseline, of which 67% (n=50/75 of those anemic women for whom hemoglobin, ferritin, and sTfR data were available) had a hemoglobin response after 12 weeks of treatment. Pasricha et al. defined women as hemoglobin responders if they had a hemoglobin response ≥10 g/L or recovered from anemia at 12 weeks. Pasricha et al. concluded that although 67% (n=50/75) of anemic women were hemoglobin responders, only 33% (n=26/75) of these women were iron deficient at baseline (assessed by ferritin <15 µg/L) (84). In our study, we observed a lower proportion of anemic women with a hemoglobin response (31% [n=36/115] among women receiving Fe, and 42% [n=49/117] among women receiving Fe+MMN), despite that the iron content in the supplements in the Pasricha et al. study provided women with only 60 mg of elemental iron weekly, whereas our study provided 60 mg daily. However, Pasricha et al.’s definition of a hemoglobin responder included women with a hemoglobin increase ≥10 g/L or women who recovered from anemia (those women who had a hemoglobin concentration ≥120 g/L after the 12 week intervention), whereas   132 our definition (decided a priori to the initiation of the trial) was more conservative and included only women with an overall hemoglobin response ≥10 g/L at 12 weeks. This likely contributed to the differences observed in the proportions of women determined as hemoglobin responders in each study. Of interest, we calculated the proportions of hemoglobin responders using the same definition as Pasricha et al. (84): among only anemic women at baseline (n=443) in the Fe, MMN, Fe+MMN, and placebo groups, the proportions of women with a hemoglobin increase ≥10 g/L or with resolved anemia (hemoglobin >120 g/L) at 12 weeks were 44%, 22%, 50%, and 13%. Even with the less conservative definition of a hemoglobin response, our observed proportions of hemoglobin response were lower in both the Fe and Fe+MMN groups (44% and 50%, respectively), as compared to the proportion observed in the Pasricha et al. study (67%) (84). The Pasricha et al. study was conducted in Vietnam, a country bordering Cambodia, with comparable prevalence rates of inflammation and genetic hemoglobin disorders among women. The major limitation of this study is that it did not include a control group (placebo), which would have provided stronger evidence that the outcomes were directly caused by iron and folic acid supplementation and not just temporal effects related to other factors such as the deworming treatment. Furthermore, we cannot conclude if the hemoglobin improvement was a result of the iron (alone), the folic acid (alone), or the combination of the two. Lastly, there are other micronutrients associated with anemia that could have been included (e.g., riboflavin, vitamin B6, or vitamin B12) (10,39) for a more comprehensive investigation of the potential causes of anemia in this population. We conclude that hemoglobin response to iron therapy would likely vary by population, more specifically, depending on the proportion of the population that has iron deficiency or anemia, the method of supplement delivery (weekly vs. daily), quantity of iron   133 in the supplement (30 vs. 60 mg), adherence to the supplements, and according to the chosen definition of a hemoglobin response.   5.1.5 Multiple Micronutrients In the trial in Kampong Chhnang, we observed significantly higher hemoglobin concentrations among non-pregnant Cambodian women of reproductive age after 12 weeks of daily oral Fe and Fe+MMN, as compared to MMN and placebo groups. Mean hemoglobin at 12 weeks did not differ between Fe and Fe+MMN groups, and post-estimation linear comparisons showed that the addition of MMN to Fe did not confer a significant increase in hemoglobin. Further, marginal analysis showed that Fe (with or without MMN) significantly increased hemoglobin at 12 weeks, but MMN (with or without Fe) did not. Baseline deficiencies of vitamin A (RBP <0.7 µmol/L) and B12 (<150 ρmol/L) among women in Kampong Chhnang were both <1%. We did observe some biochemical evidence of folate deficiency (12.5%, n=50/400 based on serum folate <6.8 nmol/L); however, we note that there was no evidence of megaloblastic anemia (< 1% of anemic women had a MCV >98 fL). Given that there was no biochemical evidence of vitamins A or B12 deficiency and only some evidence of folate deficiency (in the absence of megaloblastic anemia), it is not surprising that the addition of MMN did not confer a significant benefit to Fe alone. In the Prey Veng survey, non-pregnant women similarly had very low prevalence rates of vitamin A (0%, n=420, serum RBP <0.7 µmol/L), B12 (1%, n=420, serum B12 <150 ρmol/L), folate (2.6%, n=420, serum folate <6.8 nmol/L), and B6 deficiencies (2%, based on plasma pyridoxal -5’-phosphate <20 nmol/L in a randomly selected subset of n=99).    134 A recent survey by Wieringa et al. confirmed our findings of a very low prevalence of vitamin A and B12 deficiencies and a moderately high prevalence of folate deficiency among women of reproductive age (164). This nationally representative survey was in follow up to the 2014 Demographic and Health survey, where one in every six households was later visited for collection of a blood, urine, and stool sample from women and one child under six years of age. Wieringa et al. found that among mothers of reproductive age (15-49 years) deficiencies of vitamin A (RBP <0.7 µmol/L) and B12 (<150 ρmol/L) among women in Kampong Chhnang were both <4% (164). Yet, Wieringa et al. also reported that these same women had a high prevalence of folate (17.8%, based on serum folate <10 nmol/L) and zinc deficiency (26.3%, based on serum zinc <7.65 µmol/L). Wieringa et al. used a higher cut-off value to indicate serum folate deficiency than us (10 vs. 6.8 nmol/L, respectively), which may have contributed to the higher prevalence of deficiency observed in the Wieringa et al. study (154). We interpret the Wieringa et al. findings of zinc deficiency with caution, however, as serum or plasma zinc concentrations can be influenced by recent dietary intake of zinc, the timing of the last meal, the timing of the blood collection, and can also be decreased in the presence of inflammation (169). The authors of the study did not mention any of these considerations in their assessment and interpretation of serum zinc concentrations. Yet, levels of inflammation among women were relatively high (9.7% had acute inflammation based on CRP >5 mg/L, and 35.8% had chronic inflammation based on AGP >1 mg/L), which may have resulted in an overestimation of zinc deficiency prevalence. We did not measure zinc concentrations in our trial in Kampong Chhnang, but we conclude that the evidence of deficiency reported in the Wieringa et al. survey requires follow-up.     135 We recognize that although we did not see a benefit of MMN on hemoglobin in our trial, it is important for women to have optimal micronutrient status during their reproductive years, as there are other potential benefits of multiple micronutrients that our study did not assess (e.g., pregnancy outcomes, in the case that the women later become pregnant). In a 2015 Cochrane review, Haider et al. investigated the effect of multiple micronutrients (including iron) and iron and folic acid supplementation in reducing the risk of maternal anemia in the third trimester of pregnancy (170). This review included 17 trials (n=137,791 pregnant women) and showed no significant difference in the risk reduction of maternal anemia in the third trimester among women receiving either multiple micronutrients (including iron) or iron and folic acid supplements relative to placebo (RR: 0.98 [95% CI: 0.86, 1.11]) (170). There are, however, potential benefits of multiple micronutrient supplementation during pregnancy on other health and nutrition related outcomes beyond anemia. Although no effect was shown for maternal anemia, this review showed that multiple micronutrient (including iron and folic acid) supplementation resulted in a decrease in the number of infants born with low birthweight (RR: 0.88 [95% CI: 0.85, 0.90]), small for gestational age (RR: 0.91 [95% CI: 0.84, 0.97]), and stillbirth (RR: 0.92 [95% CI: 0.86, 0.99]), as compared to control (iron with or without folic acid). Haider et al. concluded that there is a strong basis to replace iron and folic acid with multiple micronutrients in pregnant women in developing countries where micronutrient deficiencies are common in women of reproductive age. We emphasize that the impact of multiple micronutrients on hemoglobin concentrations (and potentially other outcomes) is likely modified by the baseline prevalence of deficiencies in the population.     136 In our studied population of non-pregnant women, vitamin A and B12 deficiencies were very low. We did observe some evidence of folate deficiency; however, <1% of anemic women had megaloblastic anemia (anemia caused by folate or B12 deficiency). The low prevalence of vitamin A and B12 deficiencies and/or the lack of evidence of megaloblastic anemia may have been possible reasons that we did not see a significant increase in hemoglobin with MMN. However, we conclude that although we did not see a benefit of MMN on hemoglobin in our trial, it is important for women to have optimal micronutrient status during their reproductive years, especially in the case that the women later become pregnant.  5.1.6 Is Iron Deficiency a Major Cause of Anemia in Cambodian Women? The overall goal of our research was to determine if iron deficiency was a major cause of anemia in non-pregnant Cambodian women. In the trial in Kampong Chhnang, 31% and 42% of anemic women had a hemoglobin response ≥10 g/L at 12 weeks in the Fe and Fe+MMN groups, respectively. This suggests that approximately one third of the anemia burden in our study population was due, at least partially, to iron deficiency. The remaining two thirds is likely attributable to the high prevalence of genetic hemoglobin disorders, and, to a small extent in this study population, to inflammation and/or infection. We conclude that roughly 30-40% of the anemia burden in non-pregnant Cambodian women of reproductive age is likely due to iron deficiency. Our estimate is somewhat lower than what was proposed by authors of the 2011 WHO Global Prevalence of Anemia Report, who suggested that 45% (95% CI: 35%, 53%) of anemia is caused by iron deficiency in non-pregnant women of reproductive age in Southeast Asia.    137 5.2 Strengths and Limitations Strengths of this research are the comprehensive genetic and biochemical assessment of factors related to hemoglobin and iron status in women of reproductive age in two provinces of Cambodia. Our DNA analysis for genetic hemoglobin disorders was comprehensive in both the survey in Prey Veng and the trial in Kampong Chhnang. We assessed for all potential hemoglobin variants using capillary hemoglobin gel electrophoresis and we genotyped women to detect heterozygosity and homozygosity for the most common -globin gene deletions and point mutations indicating -thalassemia. In Kampong Chhnang, we used a rigorous 2x2 factorial design, which allowed us to concurrently measure the effect of iron, with or without MMN, on hemoglobin concentration with more power and a smaller sample size. The communication strategies and frequent follow-up by our research staff and the Ministry of Health resulted in high retention (94% at 12 weeks) and adherence rates (78-89% at 12 weeks) among women in the trial.   We acknowledge some limitations of our research. In the survey in Prey Veng, women were recruited for inclusion in a larger randomized controlled trial and eligibility criteria included woman 18-45 years with at least one child <5 years who lived in farming households with some access to land for agriculture or aquaculture activities. Although we suspect these criteria would only marginally limit eligibility (most women in Cambodia have at least one child by 18 years, and most households have some access to land), we note that this may have reduced the generalizability of our results to women in the wider province of Prey Veng. In the trial in Kampong Chhnang, we recruited women via a convenience sampling method from only one province in Cambodia, which also limits the generalizability of our findings.    138 Dietary intake data were collected among women in Prey Veng and analyzed by another graduate student (142), however, we did not measure dietary intake among women in the trial in Kampong Chhnang. It would have been interesting to investigate dietary intakes of women to explore the sources of vitamin B12 in the diet given that serum B12 concentrations were so high. The traditional Cambodian diet was thought to be low in B12 due to low intakes of animal source foods. Similarly, the collection and analysis of dietary intake data would also be interesting to confirm if the traditional Cambodian diet was low in folate, given the findings on the biochemical evidence of folate deficiency among women. However, we acknowledge the challenges and limitations in the assessment of dietary intake in Cambodia, given the lack of food composition tables and lack of information about bioavailability of iron and other micronutrients.   We did not measure infection (e.g., helminth) in either study, which can be associated with anemia and iron stores. Baseline inflammation was surprisingly low in women enrolled in the trial in Kampong Chhnang; as such, we query whether the deworming tablet that was provided one week before baseline blood collection could have influenced the concentrations of inflammation biomarkers (AGP and CRP). In retrospect, it would have been ideal to collect the baseline blood sample before administering the deworming tablet, but we suspected that the deworming tablet would have a negligible effect on hemoglobin concentration (our primary outcome). A study by Soukhathammavong et al. reported that the efficacy of one dose after 3 weeks of treatment appeared to be low (only ~18-36% reduction in helminth infection) in school-aged children in neighboring Lao PDR (171). We recognize that the efficacy of deworming   139 treatment is also likely to depend on the type of parasitic infections present, and unfortunately, we did not collect information in our study about the presence or type of parasitic infection.  5.3 Significance and Contribution of the Research The findings of this research will contribute to the growing body of evidence in Cambodia on the factors associated with hemoglobin and the causes of anemia that is needed to inform policy and programs regarding iron supplementation and fortification among non-pregnant women of reproductive age. In 2015, the Cambodia Ministry of Health put the weekly iron and folic supplementation program on hold and is awaiting more evidence to inform policy. This research has provided three overall key findings in this regard. First, we highlighted the complexity and diverse heterogeneity of hemoglobin disorders in Cambodia, and showed that iron deficiency prevalence among women was low (based on ferritin concentrations). Second, we demonstrated that some hemoglobin disorders were associated with increased concentrations of ferritin and sTfR, thus reducing their diagnostic accuracy to estimate iron deficiency prevalence. Last, our trial in Kampong Chhnang showed that overall only ~25% of our study population responded to 12 weeks of daily iron supplementation. Of further interest, a lower proportion of hemoglobin response was observed among women with genetic hemoglobin disorders as compared to women with no hemoglobin disorders. Based on our results that only approximately one third of anemic women responded to iron supplementation, we estimate that less than 10% of women in the wider population would benefit at most, based on the definition of a hemoglobin response (≥10 g/L at 12 weeks) and assuming that the prevalence of anemia among the women we screened was similar to that of the country as a whole.    140 Overall, this research provides critical information to guide policy and programs on anemia reduction among women of reproductive age in Cambodia. Given limited resources, the Ministry of Health needs to prioritize the most effective nutrition interventions for women and children. We query whether the weekly iron and folic acid supplementation program is an effective approach for the reduction of the population-level anemia among non-pregnant women in Cambodia, and possibly across Southeast Asia or in other countries where anemia is largely due to other causes. However, we acknowledge that folate supplementation is still important and warranted among non-pregnant women of reproductive age, as there is convincing evidence that folic acid supplementation during pregnancy reduces the risk of a neural tube defect-affected pregnancy (166).   5.4 Future Research Directions More work is needed to investigate the potential risk of untargeted blanket iron supplementation among predominantly anemic yet iron-replete populations with a high prevalence of co-existing inflammation, infection and/or genetic hemoglobin disorders. The rationale for blanket supplementation programs should be ascertained with data on both the associated risks and benefits of the intervention. The assessment of risk is one area requires further investigation in this population. One potential study would be to quantify the degree of lipid peroxidation or to measure other products of oxidative stress after iron supplementation in iron-replete women that have hemoglobin disorders and/or inflammation or infection.  Ideally, we would use targeted approaches for iron supplementation in individuals rather than blanket approaches. However, there are many challenges to targeted approaches given the need   141 to first assess the iron status of the individual. The assessment of ferritin concentration requires the collection of venous blood (for serum or plasma), centrifugation, and processing with more complex analytical methods than are usually found in the field in low-resource countries. The personnel and consumables required for the collection, and the potential transport of blood for centrifugation and analysis, require substantial resources. Until a simple and accurate point-of-care device to measure ferritin is developed, the challenges will continue. In addition to ferritin, hepcidin has shown promise as a biomarker to predict iron responsiveness and could potentially guide iron supplementation for those who would benefit most (172). Unfortunately, the same challenges for collection and assessment exist for hepcidin as for ferritin: it requires serum or plasma and has complex analytical methods (e.g., an immunosorbent assay). One of our next steps will be to investigate which of the hematological indicators (hemoglobin, ferritin, sTfR, or hepcidin) has the strongest diagnostic ability (sensitivity and specificity) to predict responsiveness to iron supplementation after 12 weeks using receiver operating characteristic (ROC) analyses. The area under the curve of the ROC (AUCROC) values and 95% CI will be compared among hemoglobin responders and non-responders. Collaborative efforts among engineers, chemists, and laboratory technicians would help drive this work forward with hope for improved technology and point-of-care devices for individual iron assessment in the future.  We observed significant and large differences in anemia prevalence when measuring hemoglobin the HemoCue® and automated hematology analyzer. In Prey Veng, the overall bias in hemoglobin measurement between the two methods was 2.6 g/L, resulting in a difference in anemia prevalence of 11.5% (41.0% using the HemoCue [capillary blood] and 29.5% using the Sysmex analyzer [venous blood], P<0.001). In Kampong Chhnang, despite screening for anemic   142 women using the HemoCue® (capillary blood) and an inclusion criterion of hemoglobin ≤117 g/L, the Sysmex analyzer (venous blood; and considered a more accurate and precise method of hemoglobin measurement) later showed that only 58% of women were anemic (hemoglobin <120 g/L) at baseline. This discrepancy may, in part, be due to regression to the mean, and could also be a result of differences in hydration status and/or the biological differences between capillary and venous blood. These results highlight the importance of standardized procedures and methods for blood collection for the assessment of hemoglobin concentration. More research is warranted to investigate the other potential factors influencing hemoglobin concentration (e.g., hydration status or humidity) and the accuracy of the HemoCue® to estimate hemoglobin concentrations.    One challenging issue in interpreting our findings on hemoglobin response is our assumption that the arbitrary global cut-off for anemia is appropriate for Cambodian women. The cut-off for non-pregnant women of reproductive age was determined by convention as the 5th percentile of hemoglobin concentration based on the normal distribution of a healthy, iron-replete population (45). Four studies provided the reference data used to estimate this cut-off, which were based on hemoglobin concentrations of predominantly European women in the late 1960’s (46). This brings into question whether the cut-off is appropriate in Southeast Asia, where a high prevalence of genetic hemoglobin disorders and inflammation exist among individuals. The WHO has recently initiated an international consultation on revisiting the hemoglobin cut-offs, which we feel is highly warranted for populations living in regions such as Southeast Asia. In lieu of using reference data sets to estimate the cut-offs for anemia, the appropriate hemoglobin   143 cut-offs for non-pregnant women in Cambodia could be assessed with implementation of larger cohort trials that evaluate outcomes such as work capacity or productivity among women.     144 Bibliography 1.  Kassebaum NJ, Jasrasaria R, Naghavi M, Wulf SK, Johns N, Lozano R, Regan M, Weatherall D, Chou DP, et al. A systematic analysis of global anemia burden from 1990 to 2010. Blood. 2014;123:615–24.  2.  World Health Organization. Assessing the iron status of populations. 2nd ed. Geneva: World Health Organization; 2007.  3.  Allen LH. Anemia and iron deficiency: effects on pregnancy outcome. Am J Clin Nutr. 2000;71:1280–4.  4.  Xiong X, Buekens P, Alexander S, Demianczuk N, Wollast E. Anemia during pregnancy and birth outcome: a meta-analysis. Am J Perinatol. 2000;17:137–46.  5.  Stoltzfus RJ, Mullany L, Black RE. Iron deficiency anemia. In: Ezzati M, Lopez AD, Rodgers A, Murray CJL, editors. Comparative quantification of health risks: global and regional burden of disease attributable to selected major risk factors. Geneva: World Health Organization; 2004. p. 163–209.  6.  Haas JD, Brownlie TI. Iron deficiency and reduced work capacity: a critical review of the research to determine a causal relationship. J Nutr. 2001;131:676–90.  7.  Cambodia demographic and health survey 2014. Phnom Penh, Cambodia, and Rockville, Maryland, USA; 2015.  8.  de Benoist B, McLean E, Egli I, Cogswell M. Worldwide prevalence of anaemia 1993-2005: WHO global database on anaemia. Geneva: World Health Organization; 2008.  9.  Gibson RS. Principles of nutritional assessment. 2nd ed. New York: Oxford University Press; 2005.  10.  Cambodia Council for Agriculture and Rural Development (CARD). Strategic framework   145 for food security and nutrition in Cambodia 2008-2012. Phnom Penh, Cambodia; 2012.  11.  Fishman SM, Christian P, West KP. The role of vitamins in the prevention and control of anaemia. Public Health Nutr. 2000;3:125–50.  12.  Ma AG, Schouten EG, Zhang FZ, Kok FJ, Yang F, Jiang DC, Sun YY, Han XX. Retinol and riboflavin supplementation decreases the prevalence of anemia in Chinese pregnant women taking iron and folic acid supplements. J Nutr. 2008;138:1946–50.  13.  Zimmermann MB, Hurrell RF. Nutritional iron deficiency. Lancet. 2007;370:511–20.  14.  World Health Organization. The prevalence of anaemia in women: a tabulation of available information. 2nd ed. Geneva: World Health Organization; 1992.  15.  Tomkins A. Assessing micronutrient status in the presence of inflammation. J Nutr. 2003;133:1649–55.  16.  Weiss G, Goodnough LT. Anemia of chronic disease. N Engl J Med. 2005;352:1011–23.  17.  Gangat N, Wolanskyj AP. Anemia of chronic disease. Semin Hematol. 2013;50:232–8.  18.  Bain BJ. Haemoglobinopathy diagnosis. 2nd ed. Oxford: Blackwell Publishing Ltd; 2006.  19.  Greer JP, Foerster J, Rodgers GM, Paraskevas F. Wintrobe’s clinical hematology. 12th ed. Baltimore: Lippincott Williams & Wilkins; 2009.  20.  World Health Organization. Guideline: intermittent iron and folic acid supplementation in menstruating women. Geneva: World Health Organization; 2011.  21.  Zimmermann MB, Fucharoen S, Winichagoon P, Sirankapracha P, Zeder C, Gowachirapant S, Judprasong K, Tanno T, Miller JL, Hurrell RF. Iron metabolism in heterozygotes for hemoglobin E (HbE), alpha-thalassemia 1, or beta-thalassemia and in compound heterozygotes for HbE/beta-thalassemia. Am J Clin Nutr. 2008;88:1026–31.  22.  Karakochuk CD, Whitfield KC, Barr SI, Lamers Y, Devlin AM, Vercauteren SM, Kroeun   146 H, Talukder A, McLean J, Green TJ. Genetic hemoglobin disorders rather than iron deficiency are a major predictor of hemoglobin concentration in women of reproductive age in rural Prey Veng, Cambodia. J Nutr. 2015;145:134–42.  23.  Pasricha S, Flecknoe-Brown SC, Allen KJ, Gibson PR, McMahon LP, Olynyk JK, Roger SD, Savoia HF, Tampi R, et al. Diagnosis and management of iron deficiency anaemia: a clinical update. Med J Aust. 2010;193:525–32.  24.  Carnley BP, Prior JF, Gilbert A, Lim E, Devenish R, Sing H, Sarin E, Guhadasan R, Sullivan SG, et al. The prevalence and molecular basis of hemoglobinopathies in Cambodia. Hemoglobin. 2006;30:463–70.  25.  George J, Yiannakis M, Main B, Devenish R, Anderson C, An US, Williams SM, Gibson RS. Genetic hemoglobin disorders, infection, and deficiencies of iron and vitamin A determine anemia in young Cambodian children. J Nutr. 2012;142:781–7.  26.  Nike H, Quigley J, Robinson K. Genocide in Cambodia. Philadelphia: University of Pennsylvania Press; 2000.  27.  Chandler D. A history of Cambodia. 4th ed. Boulder, Colorado: Westview Press; 2008.  28.  Cambodia Ministry of Health. National policy and guidelines for micronutrient supplementation to prevent and control deficiencies in Cambodia. Phnom Penh; 2011.  29.  Von Grebmer K, Bernstein J, de Waal A, Prasai N, Yin S, Yohannes Y. 2015 Global hunger index [Internet]. Washington DC (USA); 2015. Available from: http://ebrary.ifpri.org/cdm/ref/collection/p15738coll2/id/129681 (Retrieved December 15, 2016). 30.  McDonald C, McLean J, Kroeun H, Talukder A, Lynd L, Green T. Household food insecurity and dietary diversity as correlates of maternal and child undernutrition in rural   147 Cambodia. Eur J Clin Nutr. Nature Publishing Group; 2015;69:242–6.  31.  Swindale A, Bilinsky P. Development of a universally applicable household food insecurity measurement tool: process, current status, and outstanding issues. J Nutr. 2006;136:1449S–1452S.  32.  Giardina B, Messana I, Scatena R, Castagnola M. The multiple functions of hemoglobin. Crit Rev Biochem Mol Biol. 1995;30:165–96.  33.  Fujimoto K. Principles of measurement in hematology analyzers manufactured by Sysmex Corporation. Sysmex J Int. 1999;9:31–44.  34.  Gulati GL, Hyun BH. Quality control in hematology. Clin Lab Med. 1986;6:675–88.  35.  Mohandas N, Kim Y, Tycko D, Orlik J, Wyatt J, Groner W. Accurate and independent measurement of volume and hemoglobin concentration of individual red cells by laser light scattering. Blood. 1986;68:506–13.  36.  ICF International/Demographic and Health Surveys Calverton. Biomarker field manual: demographic and health surveys methodology. Washington, DC; 2012.  37.  Von Schenck H, Falkensson M, Lundberg B. Evaluation of “HemoCue,” a new device for determining hemoglobin. Clin Chem. 1986;32:526–9.  38.  Bridges N, Parvin R, van Assendelft O. Evaluation of a new system for hemoglobin measurement. Am Clin Prod Rev. 1987;6:22–5.  39.  Neville RG. Evaluation of portable haemoglobinometer in general practice. Br Med J (Clin Res Ed). 1987;294:1263–5.  40.  Cohen AR, Seidl-Friedman J. HemoCue system for hemoglobin measurement. Evaluation in anemic and nonanemic children. Am J Clin Pathol. 1988;90:302–5.  41.  Hudson-Thomas M, Bingham KC, Simmons WK. An evaluation of the HemoCue for   148 measuring haemoglobin in field studies in Jamaica. Bull World Health Organ. 1994;72:423–6.  42.  Morris SS, Ruel MT, Cohen RJ, Dewey KG, de la Brière B, Hassan MN. Precision, accuracy, and reliability of hemoglobin assessment with use of capillary blood. Am J Clin Nutr. 1999;69:1243–8.  43.  Sari M, De Pee S, Martini E, Herman S, Sugiatmi, Bloem MW, Yip R. Estimating the prevalence of anaemia: a comparison of three methods. Bull World Health Organ. 2001;79:506–11.  44.  Neufeld L, García-Guerra A, Sánchez-Francia D, Newton-Sánchez O, Ramírez-Villalobos MD, Rivera-Dommarco J. Hemoglobin measured by Hemocue and a reference method in venous and capillary blood: a validation study. Salud Publica Mex. 2002;44:219–27.  45.  WHO nutritional anaemias: report of WHO scientific group. Technical report series No. 405. Geneva: World Health Organization; 1968.  46.  World Health Organization. Haemoglobin concentrations for the diagnosis of anaemia and assessment of severity. Geneva: World Health Organization; 2011.  47.  Sullivan KM, Mei Z, Grummer-Strawn L, Parvanta I. Haemoglobin adjustments to define anaemia. Trop Med Int Heal. 2008;13:1267–71.  48.  Stevens GA, Finucane MM, De-Regil LM, Paciorek CJ, Flaxman SR, Branca F, Peña-Rosas JP, Bhutta Z, Ezzati M. Global, regional, and national trends in haemoglobin concentration and prevalence of total and severe anaemia in children and pregnant and non-pregnant women for 1995-2011: a systematic analysis of population-representative data. Lancet Glob Heal. 2013;1:16–25.  49.  Rahman M, Abe S, Rahman S, Kanda M, Narita S, Bilano V, Ota E, Gilmour S, Shibuya   149 K. Maternal anemia and risk of adverse birth and health outcomes in low- and middle-income countries: systematic review and meta-analysis. Am J Clin Nutr. 2016;103:495–504.  50.  Stein AJ, Qaim M. The human and economic cost of hidden hunger. Food Nutr Bull. 2007;28:125–34.  51.  Camaschella C. Iron-deficiency anemia. N Engl J Med. 2015;372:1832–43.  52.  Northrop-Clewes CA, Thurnham DI. Biomarkers for the differentiation of anemia and their clinical usefulness. J Blood Med. 2013;4:11–22.  53.  World Health Organization. The global prevalence of anaemia in 2011. Geneva: World Health Organization; 2015.  54.  Leenstra T, Kariuki SK, Kurtis JD, Oloo AJ, Kager PA, ter Kuile FO. The effect of weekly iron and vitamin A supplementation on hemoglobin levels and iron status in adolescent schoolgirls in western Kenya. Eur J Clin Nutr. 2009;63:173–82.  55.  Ahmed F, Rahman Khan M, Jackson AA. Concomitant supplemental vitamin A enhances the response to weekly supplemental iron and folic acid in anemic teenagers in urban Bangladesh. Am J Clin Nutr. 2001;74:108–15.  56.  Crompton DWT, Nesheim MC. Nutritional impact of intestinal helminthiasis during the human life cycle. Annu Rev Nutr. 2002;22:35–59.  57.  Kasper MR, Blair PJ, Touch S, Sokhal B, Yasuda CY, Williams M, Richards AL, Burgess TH, Wierzba TF, Putnam SD. Infectious etiologies of acute febrile illness among patients seeking health care in south-central Cambodia. Am J Trop Med Hyg. 2012;86:246–53.  58.  Vlieghe ER, Phe T, De Smet B, Veng HC, Kham C, Lim K, Koole O, Lynen L, Peetermans WE, Jacobs JA. Bloodstream infection among adults in Phnom Penh,   150 Cambodia: key pathogens and resistance patterns. PLoS One. 2013;8:e59775.  59.  Muth S, Sayasone S, Odermatt-Biays S, Phompida S, Duong S, Odermatt P. Schistosoma mekongi in Cambodia and Lao People’s Democratic Republic. Adv Parasitol. 2010;72:179–203.  60.  Merrill RD, Labrique AB, Shamim AA, Schulze K, Christian P, Merrill RK, West KP. Elevated and variable groundwater iron in rural northwestern Bangladesh. J Water Health. 2010;8:818–25.  61.  Merrill RD, Shamim AA, Ali H, Jahan N, Labrique AB, Schulze K, Christian P, West KP. Iron status of women is associated with the iron concentrations of potable groundwater in rural Bangladesh. J Nutr. 2011;141:944–9.  62.  Feldman PR, Rosenboom J-W, Saray M, Navuth P, Samnang C, Iddings S. Assessment of the chemical quality of drinking water in Cambodia. J Water Health. 2007;5:101–16.  63.  Kraemer K, Zimmerman M, editors. Nutritional anemia. Basel: Sight and Life Press; 2007.  64.  de Benoist B. Conclusions of a WHO technical consultation on folate and vitamin B12 deficiencies. Food Nutr Bull. 2008;29:S238-44.  65.  Suharno D, West CE, Muhilal, Karyadi D, Hautvast JG. Supplementation with vitamin A and iron for nutritional anaemia in pregnant women in West Java, Indonesia. Lancet. 1993;342:1325–8.  66.  Powers HJ, Hill MH, Mushtaq S, Dainty JR, Majsak-Newman G, Williams EA. Correcting a marginal riboflavin deficiency improves hematologic status in young women in the United Kingdom (RIBOFEM). Am J Clin Nutr. 2011;93:1274–84.  67.  World Health Organization. Serum ferritin concentrations for the assessment of iron status   151 and iron deficiency in populations. Geneva: World Health Organization; 2011.  68.  Thurnham D, McCabe G, Northrop-Clewes C, Nestel P. Effects of subclinical infection on plasma retinol concentrations and assessment of prevalence of vitamin A deficiency: meta-analysis. Lancet. 2003;362:2052–8.  69.  Thurlow RA, Winichagoon P, Green T, Wasantwisut E, Pongcharoen T, Bailey KB, Gibson RS. Only a small proportion of anemia in northeast Thai schoolchildren is associated with iron deficiency. Am J Clin Nutr. 2005;82:380–7.  70.  Ajayi OA, Nnaji UR. Effect of ascorbic acid supplementation on haematological response and ascorbic acid status of young female adults. Ann Nutr Metab. 1990;34:32–6.  71.  Mao X, Yao G. Effect of vitamin C supplementations on iron deficiency anemia in Chinese children. Biomed Env Sci. 1992;5:125–9.  72.  Reinken L, Kurz R. Activity studies of an iron-vitamin B6 preparation for euteral treatment of iron deficiency anemia. Int J Vitam Nutr Res. 1975;45:411–8.  73.  Hisano M, Suzuki R, Sago H, Murashima A, Yamaguchi K. Vitamin B6 deficiency and anemia in pregnancy. Eur J Clin Nutr. 2010;64:221–3.  74.  Nemeth E, Tuttle MS, Powelson J, Vaughn MB, Donovan A, Ward DM, Ganz T, Kaplan J. Hepcidin regulates cellular iron efflux by binding to ferroportin and inducing its internalization. Science (80- ). 2004;306:2090–3.  75.  Baumann H, Gauldie J. The acute phase response. Immunol Today. 1994;15:74–80.  76.  Thurnham DI, McCabe LD, Haldar S, Wieringa FT, Northrop-Clewes CA, McCabe GP. Adjusting plasma ferritin concentrations to remove the effects of subclinical inflammation in the assessment of iron deficiency: a meta-analysis. Am J Clin Nutr. 2010;92:546–55.  77.  Hallberg L, Asp NG. Iron requirement, iron balance and iron deficiency in menstruating   152 and pregnant women. In: Hallberg L, App NG, editors. Iron nutrition in health and disease. London: John Libbey and Co. Ltd.; 1996. p. 165–82.  78.  Bothwell TH. Iron requirements in pregnancy and strategies to meet them. Am J Clin Nutr. 2000;72:257–64.  79.  Harvey LJ, Armah CN, Dainty JR, Foxall RJ, Lewis DJ, Langford NJ, Fairweather-Tait SJ. Impact of menstrual blood loss and diet on iron deficiency among women in the UK. Br J Nutr. 2005;94:557–64.  80.  Hartwell SK, Srisawang B, Kongtawelert P, Christian GD, Grudpan K. Review on screening and analysis techniques for hemoglobin variants and thalassemia. Talanta. 2005;65:1149–61.  81.  Jack SJ, Ou K, Chea M, Chhin L, Devenish R, Dunbar M, Eang C, Hou K, Ly S, et al. Effect of micronutrient sprinkles on reducing anemia: a cluster-randomized effectiveness trial. Arch Pediatr Adolesc Med. 2012;166:842–50.  82.  Wang W, Knovich MA, Coffman LG, Torti FM, Torti S V. Serum ferritin: past, present and future. Biochim Biophys Acta. 2010;1800:760–9.  83.  World Health Organization. Iron deficiency anaemia: assessment, prevention and control. Geneva: World Health Organization; 2001.  84.  Pasricha S, Casey GJ, Phuc TQ, Mihrshahi S, MacGregor L, Montresor A, Tien N, Biggs B. Baseline iron indices as predictors of hemoglobin improvement in anemic Vietnamese women receiving weekly iron-folic acid supplementation and deworming. Am J Trop Med Hyg. 2009;81:1114–9.  85.  Mast AE, Blinder MA, Gronowski AM, Chumley C, Scott MG. Clinical utility of the soluble transferrin receptor and comparison with serum ferritin in several populations.   153 Clin Chem. 1998;44:45–51.  86.  Erhardt JG, Estes JE, Pfeiffer CM, Biesalski HK, Craft NE. Combined measurement of ferritin, soluble transferrin receptor, retinol binding protein, and C-reactive protein by an inexpensive, sensitive, and simple sandwich enzyme-linked immunosorbent assay technique. J Nutr. 2004;134:3127–32.  87.  Cox KL, Devanarayan V, Kriauciunas A, Manetta J, Montrose C, Sittampalam S. Immunoassay methods. In: Sittampalam GS, Coussens NP, Nelson H, Arkin M, Auld D, Austin C, Bejcek B, Glicksman M, Inglese J, et al., editors. Assay guidance manual. Betheseda (MD): Eli Lilly & Company and the National Center for Advancing Translational Sciences; 2004.  88.  Ferraro S, Mozzi R, Panteghini M. Revaluating serum ferritin as a marker of body iron stores in the traceability era. Clin Chem Lab Med. 2012;50:1911–6.  89.  Garcia-Casal MN, Peña-Rosas JP, Pasricha S-R. Rethinking ferritin cutoffs for iron deficiency and overload. Lancet Haematol. 2014;1:e92–4.  90.  Proposed international standard of human ferritin for the serum ferritin assay. International Committee for Standardization in Haematology (Expert Panel on Iron). Br J Haematol. 1985;61:61–3.  91.  WHO expert committee on biological standardization. World Health Organ Tech Rep Ser. 1994;840:1–218.  92.  World Health Organization. WHO international standard (ferritin, human recombinant; code 94/572): instructions for use. Potters Bar, Hertfordshire; 2008.  93.  Thorpe SJ, Walker D, Arosio P, Heath A, Cook JD, Worwood M. International collaborative study to evaluate a recombinant L ferritin preparation as an international   154 standard. Clin Chem. 1997;43:1582–7.  94.  Domellof M, Dewey K, Lonnerdal B, Cohen R, Hernell O. The diagnostic criteria for iron deficiency in infants should be reevaluated. J Nutr. 2002;3680–6.  95.  Beard J. Effectiveness and strategies of iron supplementation during pregnancy. Am J Clin Nutr. 2000;71:1288S–1294S.  96.  Finberg KE, Heeney MM, Campagna DR, Aydinok Y, Pearson H a, Hartman KR, Mayo MM, Samuel SM, Strouse JJ, et al. Mutations in TMPRSS6 cause iron-refractory iron deficiency anemia (IRIDA). Nat Genet. 2008;40:569–71.  97.  Mohamed M, Phillips J. Hereditary haemochromatosis. Br Med J. 2016;353:i3128.  98.  Worwood M. Inherited iron loading: genetic testing in diagnosis and management. Blood Rev. 2005;19:69–88.  99.  Rochette J, Pointon JJ, Fisher CA, Perera G, Arambepola M, Arichchi DSK, Silva S De, Vandwalle JL, Monti JP, et al. Multicentric origin of hemochromatosis gene (HFE) mutations. Am J Hum Genet. 1999;64:1056–62.  100.  Cullen LM, Gao X, Easteal S, Jazwinska EC. The hemochromatosis 845 GA and 187 CG mutations: prevalence in non-caucasian populations. Am J Hum Genet. 1998;62:1403–7.  101.  Zimmermann MB. Methods to assess iron and iodine status. Br J Nutr. 2008;99 Suppl 3:S2–9.  102.  Thurnham DI, Mccabe GP. Influence of infection and inflammation on biomarkers of nutritional status with an emphasis on vitamin A and iron. World Health Organization report: priorities in the assessment of vitamin A and iron status in populations. Geneva: World Health Organization; 2012. p. 63–80.  103.  Thurnham DI, Northrop-clewes CA, Knowles J. The use of adjustment factors to address   155 the impact of inflammation on vitamin A and iron status in humans. J Nutr. 2015;145:1137–43S.  104.  Calder PC, Ahluwalia N, Albers R, Bosco N, Bourdet-Sicard R, Haller D, Holgate ST, Jönsson LS, Latulippe ME, et al. A consideration of biomarkers to be used for evaluation of inflammation in human nutritional studies. Br J Nutr. 2013;109 Suppl:S1-34.  105.  Cook J, Dassenko S, Skikne B. Serum transferrin receptor as an index of iron absorption. Br J Haematol. 1990;75:603–9.  106.  Pfeiffer CM, Cook JD, Mei Z, Cogswell ME, Looker AC, Lacher DA. Evaluation of an automated soluble transferrin receptor assay on the Roche Hitachi analyzer and its comparison to two ELISA assays. Clin Chim Acta. 2007;382:112–6.  107.  Pfeiffer C, Cook J, Mei Z, Cogswell M, Looker A, Lacher D. Evaluation of an automated soluble transferrin receptor (sTfR) assay on the Roche Hitachi analyzer and its comparison to two ELISA assays. Clin Chim Acta. 2007;382:112–6.  108.  Beguin Y. Soluble transferrin receptor for the evaluation of erythropoiesis and iron status. Clin Chim Acta. 2003;329:9–22.  109.  Carriaga M, Skikne B, Finley B, Cutler B, Cook J. Serum transferrin receptor for the detection of iron deficiency in pregnancy. Am J Clin Nutr. 1991;54:1077–81.  110.  Choi JW, Im MW, Pai SH. Serum transferrin receptor concentrations during normal pregnancy. Clin Chem. 2000;46:725–7.  111.  Akesson A, Bjellerup P, Berglund M, Bremme K, Vahter M. Serum transferrin receptor: a specific marker of iron deficiency in pregnancy. Am J Clin Nutr. 1998;68:1241–6.  112.  Davison J, Hytten F. Glomerular filtration during and after pregnancy. J Obstet Gynaecol Br Commonw. 1974;81:588–95.    156 113.  Raya G, Henny J, Steinmetz J, Herbeth B, Siest G. Soluble transferrin receptor (sTfR): biological variations and reference limits. Clin Chem Lab Med. 2001;39:1162–8.  114.  Menendez C, Quinto LL, Kahigwa E, Alvarez L, Fernandez R, Gimenez N, Schellenberg D, Aponte JJ, Tanner M, Alonso PL. Effect of malaria on soluble transferrin receptor levels in Tanzanian infants. Am J Trop Med Hyg. 2001;65:138–42.  115.  Beck N. Diagnostic hematology. 6th ed. London: Springer; 2009.  116.  Brugnara C. Reticulocyte cellular indices: a new approach in the diagnosis of anemias and monitoring of erythropoietic function. Crit Rev Clin Lab Sci. 2000;37:93–130.  117.  Park CH, Valore E V, Waring AJ, Ganz T. Hepcidin, a urinary antimicrobial peptide synthesized in the liver. J Biol Chem. 2001;276:7806–10.  118.  Anderson GJ, Frazer DM, McLaren GD. Iron absorption and metabolism. Curr Opin Gastroenterol. 2009;25:129–35.  119.  Franchini M, Montagnana M, Lippi G. Hepcidin and iron metabolism: from laboratory to clinical implications. Clin Chim acta. 2010;411:1565–9.  120.  Nemeth E, Valore E V, Territo M, Schiller G, Lichtenstein A, Ganz T. Hepcidin, a putative mediator of anemia of inflammation, is a type II acute-phase protein. Blood. 2003;101:2461–3.  121.  Nicolas G, Chauvet C, Viatte L, Danan JL, Bigard X, Devaux I, Beaumont C, Kahn A, Vaulont S. The gene encoding the iron regulatory peptide hepcidin is regulated by anemia, hypoxia, and inflammation. J Clin Invest. 2002;110:1037–44.  122.  Kattamis A, Papassotiriou I, Palaiologou D, Apostolakou F, Galani A, Ladis V, Sakellaropoulos N, Papanikolaou G. The effects of erythropoetic activity and iron burden on hepcidin expression in patients with thalassemia major. Haematologica. 2006;91:809–  157 12.  123.  Origa R, Galanello R, Ganz T, Giagu N, Maccioni L, Faa G, Nemeth E. Liver iron concentrations and urinary hepcidin in β-thalassemia. Haematologica. 2007;92:583–8.  124.  Bregman D, Morris D, Koch T, He A, Goodnough L. Hepcidin levels predict nonresponsiveness to oral iron therapy in patients with iron deficiency anemia. Am J Hematol. 2013;88:97–101.  125.  Beutler E, Hoffbrand AV, Cook JD. Iron deficiency and overload. Hematol Am Soc Hematol Educ Progr. 2003;2003:40–61.  126.  Freire W. Hemoglobin as a predictor of response to iron therapy and its use in screening and prevalence estimates. Am J Clin Nutr. 1989;50:1442–9.  127.  Garby L, Irnell L, Werner I. Iron deficiency in women of fertile age in a Swedish community. III. Estimation of prevalence based on response to iron supplementation. Acta Med Scand. 1969;185:113–7.  128.  World Health Organizaiton. Guideline: daily iron supplementation in adult women and adolescent girls. Geneva; World Health Organization; 2016.  129.  Laillou A, Pfanner S, Chan T, Chea C, Mam B, Sambath P, Vonthanak S, Wieringa F. Beyond effectiveness—the adversities of implementing a fortification program. A case study on the quality of iron fortification of fish and soy sauce in Cambodia. Nutrients. 2016;8:nu8020094.  130.  Charles C V, Dewey CE, Daniell WE, Summerlee AJS. Iron-deficiency anaemia in rural Cambodia: community trial of a novel iron supplementation technique. Eur J Public Health. 2011;21:43–8.  131.  Infusino I, Braga F, Dolci A, Panteghini M. Soluble transferrin receptor (sTfR) and   158 sTfR/log ferritin index for the diagnosis of iron-deficiency anemia. A meta-analysis. Am J Clin Pathol. 2012;138:642–9.  132.  Fucharoen S, Winichagoon P. Haemoglobinopathies in Southeast Asia. Indian J Med Res. 2011;134:498–506.  133.  Uaprasert N, Rojnuckarin P, Bhokaisawan N, Settapiboon R, Wacharaprechanont T, Amornsiriwat S, Sutcharitchan P. Elevated serum transferrin receptor levels in common types of thalassemia heterozygotes in Southeast Asia: A correlation with genotypes and red cell indices. Clin Chim Acta. 2009;403:110–3.  134.  Cogill B. Anthropometric Indicators Measurement Guide. Washington, DC; 2003.  135.  O’Broin S, Kelleher B. Microbiological assay on microtitre plates of folate in serum and red cells. J Clin Pathol. 1992;45:344–7.  136.  Ubbink JB, Serfontein WJ, De Villiers LS. Stability of pyridoxal-5-phosphate semicarbazone: applications in plasma vitamin B6 analysis and population surveys of vitamin B6 nutritional status. J Chromatogr B Biomed Sci Appl. 1985;342:277–84.  137.  Giordano PC. Carrier diagnostics and prevention of hemoglobinopathies using capillary electrophoresis. Evry: Sebia; 2007.  138.  Tan A, Quah T, Low P, SS C. A rapid and reliable 7-deletion multiplex polymerase chain reaction assay for alpha -thalassemia. Blood. 2001;98:250–1.  139.  Chen J, Shao J. Nearest neighbor imputation for survey data. J Off Stat. 2000;16:113–31.  140.  King JC. The risk of maternal nutritional depletion and poor outcomes increases in early or closely spaced pregnancies. J Nutr. 2003;133:1732S–1736S.  141.  Cooper M, Greene-Finestone L, Lowell H, Levesque J, Robinson S. Iron sufficiency of Canadians. Health Rep. 2012;23:41–8.    159 142.  Verbowski VC. The effect of plant-based homestead food production with and without small-scale aquaculture on dietary intakes of women farmers and their children in Prey Veng, Cambodia [Internet]. cIRcle: University of British Columbia’s Digital Repository: Electronic Theses and Dissertations (ETDs); 2015. Available from: https://open.library.ubc.ca/cIRcle/collections/ubctheses/24/items/1.0166792 (Retrieved December 15, 2016). 143.  Whitfield KC, Karakochuk CD, Liu Y, McCann A, Talukder A, Kroeun H, Ward M, McNulty H, Lynd LD, et al. Poor thiamin and riboflavin status of women of childbearing age in rural and urban Cambodia. J Nutr. 2015;145:628–33.  144.  Hoey L, Strain J, McNulty H. Studies of biomarker responses to intervention with riboflavin: a systematic review. Am J Clin Nutr. 2009;89:1960S–80S.  145.  Rees DC, Williams TN, Maitland K, Clegg JB, Weatherall DJ. Alpha thalassaemia is associated with increased soluble transferrin receptor levels. Br J Haematol. 1998;103:365–9.  146.  Nguyen P, Nguyen H, Pham H, Truong T, Nguyen S, Martorell R, Ramakrishnan U. Multicausal etiology of anemia among women of reproductive age in Vietnam. Eur J Clin Nutr. 2015;69:107–13.  147.  Verhoef H, West CE, Kraaijenhagen R, Nzyuko SM, King R, Mbandi MM, Van Laatum S, Hogervorst R, Schep C, Kok FJ. Malarial anemia leads to adequately increased erythropoiesis in asymptomatic Kenyan children. Blood. 2002;100:3489–94.  148.  Pardoe I. Regression model building I. Applied regression modeling. 2nd ed. Hoboken, New Jersey: Wiley & Sons, Inc; 2012. p. 151–9.  149.  Kirkwood B, Sterne J. Essential Medical Statistics. 2nd ed. Oxford: Blackwell Publishing   160 Ltd; 2003.  150.  Fucharoen S, Sanchaisuriya K, Fucharoen G, Panyasai S, Devenish R, Luy L. Interaction of hemoglobin E and several forms of a-thalassemia in Cambodian families. Haematologica. 2003;88:1092–8.  151.  Gutteridge J. Iron and free radicals. In: Hallberg L, Asp N, editors. Iron nutrition in health and disease. London: John Libbey and Co. Ltd.; 1996. p. 239–46.  152.  Thurnham DI. Inflammation and biomarkers of nutrition. Sight and Life Magazine. Sight and Life Press; 2015;29(1):51–9.  153.  World malaria report 2014. Geneva: World Health Organization; 2014.  154.  Wieringa F, Sophonneary P, Whitney S, Mao B, Berger J, Conkle J, Dijkhuizen M, Laillou A. Low prevalence of iron and vitamin A deficiency among Cambodian women of reproductive age. Nutrients. 2016;8:197.  155.  Merrill RD, Shamim AA, Ali H, Labrique AB, Schulze K, Christian P, West KP. High prevalence of anemia with lack of iron deficiency among women in rural Bangladesh: a role for thalassemia and iron in groundwater. Asia Pac J Clin Nutr. 2012;21:416–24.  156.  Chandyo RK, Henjum S, Ulak M, Thorne- Lyman AL, Ulvik RJ, Shrestha PS, Locks L, Fawzi W, Strand TA. The prevalence of anemia and iron deficiency is more common in breastfed infants than their mothers in Bhaktapur, Nepal. Eur J Clin Nutr. 2016;70:456–62.  157.  Harvey-Leeson S, Karakochuk CD, Hawes M, Tugirimana PL, Bahizire E, Akilimali PZ, Michaux KD, Lynd LD, Whitfield KC, et al. Anemia and micronutrient status of women of childbearing age and children 6–59 months in the Democratic Republic of the Congo. Nutrients. 2016;8:98.    161 158.  Wirth JP, Rohner F, Woodruff BA, Chiwile F, Yankson H, Koroma AS, Russel F, Sesay F, Dominguez E, et al. Anemia, micronutrient deficiencies, and malaria in children and women in Sierra Leone prior to the Ebola outbreak - findings of a cross-sectional study. PLoS One. 2016;11:e0155031.  159.  UNICEF, WHO, UNU. Composition of a multi-micronutrient supplement to be used in pilot programmes among pregnant women in developing countries. New York: UNICEF; 1999.  160.  Molloy AM, Scott JM. Microbiological assay for serum, plasma, and red cell folate using cryopreserved, microtiter plate method. Methods Enzymol. 1997;281:43–53.  161.  Ganz T, Olbina G, Girelli D, Nemeth E, Westerman M. Immunoassay for human serum hepcidin. Blood. 2008;112:4292–7.  162.  McAlister FA, Straus SE, Sackett DL, Altman DG. Analysis and reporting of factorial trials: A systematic review. JAMA. 2003;289:2545–53.  163.  Lacerte P, Pradipasen M, Temcharoen P, Imamee N, Vorapongsathorn T. Determinants of adherence to iron/folate supplementation during pregnancy in two provinces in Cambodia. Asia Pacific J Public Heal. 2011;23:315–23.  164.  Wieringa F, Dahl M, Chamnan C, Poirot E, Kuong K, Sophonneary P, Sinuon M, Greuffeille V, Hong R, et al. The high prevalence of anemia in Cambodian children and women cannot be satisfactorily explained by nutritional deficiencies or hemoglobin disorders. Nutrients. 2016;8:nu8060348.  165.  Selhub J, Jacques PF, Dallal G, Choumenkovitch S, Rogers G. The use of blood concentrations of vitamins and their respective functional indicators to define folate and vitamin B12 status. Food Nutr Bull. 2008;29:S67-73.    162 166.  Shane B. Folate and vitamin B12 metabolism: overview and interaction with riboflavin, vitamin B6, and polymorphisms. Food Nutr Bull. 2008;29:S5–16.  167.  Hurrell RF. Bioavailability of iron. Eur J Clin Nutr. 1997;51:S4–8.  168.  Low MSY, Speedy J, Styles CE, De-Regil LM, Pasricha SR. Daily iron supplementation for improving anaemia, iron status and health in menstruating women (Review). Cochrane Database Syst Rev. 2016;CD009747.  169.  King JC, Brown KH, Gibson RS, Krebs NF, Lowe NM, Siekmann JH, Raiten DJ. Biomarkers of nutrition for development (BOND): zinc review. J Nutr. 2016;146:858S–885S.  170.  Haider BA, Bhutta ZA. Multiple-micronutrient supplementation for women during pregnancy. Cochrane database Syst Rev. 2015;11:CD004905.  171.  Soukhathammavong P, Sayasone S, Phongluxa K, Xayaseng V, Utzinger J, Vounatsou P, Hatz C, Akkhavong K, Keiser J, Odermatt P. Low efficacy of single-dose albendazole and mebendazole against hookworm and effect on concomitant helminth infection in Lao PDR. PLoS Negl Trop Dis. 2012;6:e1417.  172.  Pasricha SR, Atkinson SH, Armitage AE, Khandwala S, Veenemans J, Cox SE, Eddowes LA, Hayes T, Doherty CP, et al. Expression of the iron hormone hepcidin distinguishes different types of anemia in African children. Sci Transl Med. 2014;6:1–7.    163 Appendices  Appendix A: Evaluation of Two Methods to Measure Hemoglobin Concentration   Acknowledgement: A version of this appendix has been published. Karakochuk CD, Janmohamed A, Whitfield KC, Barr SI, Vercauteren SM, Kroeun H, Talukder A, McLean J, Green TJ. Evaluation of two methods to measure hemoglobin concentration among women with genetic hemoglobin disorders in Cambodia: a method-comparison study. Clinica Chimica Acta 2015; 441: 148-55.  Summary Background: Genetic hemoglobin E variants are common in Cambodia and result in an altered and unstable hemoglobin molecule. We evaluated two methods to measure hemoglobin concentration among individuals with and without hemoglobin variants using a hemoglobinometer (HemoCue) and a hematology analyzer (Sysmex XT-1800i). Methods: We determined the bias and concordance between the methods among 420 Cambodian women 18-45 years. Results: Bias and concordance appeared similar between methods among women with no hemoglobin disorders (n=195, bias=2.5, c=0.68), women with hemoglobin E variants (n=133, bias=2.5, c=0.78), and women with other hemoglobin variants (n=92, bias=2.7, c=0.73). The overall bias was 2.6 g/L, resulting in a difference in anemia prevalence of 11.5% (41.0% using HemoCue and 29.5% using Sysmex, P<0.001). Based on visual interpretation of the   164  concordance plots, the HemoCue device appears to underestimate hemoglobin concentrations at lower hemoglobin concentrations and to overestimate hemoglobin concentrations at higher hemoglobin concentrations (in comparison to the Sysmex analyzer). Conclusions: Bias and concordance were similar across groups, suggesting that the two methods of hemoglobin measurement are similar among women with and without hemoglobin disorders. We caution field staff, researchers and policy makers in the interpretation of data and the impact that bias between methods can have on anemia prevalence rates.  Introduction Anemia is a serious public health problem affecting over 1.6 billion people worldwide, which is almost one quarter of the world’s population (1). Anemia is defined as a low hemoglobin concentration and hemoglobin cut off levels vary among populations based on multiple factors such as age, sex and altitude level (2). Hemoglobin concentration is one of the most commonly measured indicators of health and nutrition. Anemia prevalence rates among populations can have strong policy and programming implications for the treatment, prevention and management of anemia, as the condition has serious health consequences for women (3–5) and children (6,7). In laboratory settings, hemoglobin can be measured in a sample of blood using an automated hematology analyzer, which uses spectrophotometry to quantify hemoglobin concentrations (8). This method is considered the gold standard and has minimal error due to the automation of laboratory processes, calibration and quality control checks (9,10). However, these analyzers are expensive and require trained technicians. In the field setting, particularly in large surveys and   165 research studies where blood requires refrigerated transport over long distances, this method is usually not feasible.   Portable hemoglobinometers, such as the HemoCue device, have become increasingly popular in the past decade in field settings and large surveys, as they are easy to use, inexpensive, portable, and provide an immediate digital hemoglobin measurement (2). This method is often used in nation-wide Demographic and Health Surveys to estimate the anemia prevalence of large populations (11).   Multiple studies have confirmed the accuracy and precision of HemoCue compared to hematology analyzers to measure hemoglobin in laboratory settings (12–15). However, in field settings, the HemoCue device has shown bias and higher variability of hemoglobin measures compared to hematology analyzers (16–21). This higher variability in the field setting could be a result of inadequately trained or supervised field staff, heat or humidity in the natural environment, or simply less precise and accurate methods than one would find in the clinical laboratory setting. Other researchers have detected poor agreement and correlation between HemoCue devices and hematology analyzers, notably among pregnant women in Sudan (22) and among pregnant women in Tibet living at high altitudes (23). In these studies, researchers suggested that the HemoCue method is not an acceptable method to use among the populations studied. Issues relating to measurement error have been reported in the majority of studies; therefore, recommendations have been published to standardize blood collection and measurement practices (24,25). There has been some investigation of the accuracy of   166 HemoCue to measure hemoglobin in some clinical disease states, e.g., gastrointestinal bleeding (26) and as a screening tool for blood donations (27,28), however to our knowledge there has been no exploration of whether HemoCue is as accurate as the gold standard hematology analyzer for measuring hemoglobin in individuals with genetic hemoglobin disorders.  In Cambodia, genetic hemoglobin disorders affect ~50% of the population, the most prevalent of which are hemoglobin E variants and α-thalassemia (29–31). Hemoglobin E variants, in particular, result in an altered structure of the β-globin chain of hemoglobin resulting in an unstable hemoglobin molecule (32,33). These hemoglobin E disorders are autosomal recessive, so can be inherited in either the heterozygous (also known as a ‘trait’) or homozygous form (which have more serious consequences) (34). We speculated that if there were differences in measurement of hemoglobin among individuals with hemoglobin variants, it would be most apparent in the hemoglobin E variants, given their high molecular instability. The aim of this method-comparison study was to determine the bias, precision and concordance of two commonly used methods to measure hemoglobin concentration and to explore if there were any differences among women with no genetic hemoglobin disorders, women with hemoglobin E variants, and women with other hemoglobin variants.  Methods Study Design and Participants We evaluated two methods of measuring hemoglobin concentration: a portable hemoglobinometer (HemoCue Hb 201+ Hemocue AB, Angelholm, Sweden) and an automated hematology analyzer (Sysmex XT-1800i, Sysmex Corporation, Kobe, Japan). The study used   167 data that were collected in July 2012 from 420 non-pregnant women in Prey Veng province in Cambodia as part of the baseline survey for a larger trial (not yet published). Women recruited were 18-45 years and had at least one child <5 years of age. The objective of the larger trial was to evaluate an improved model of homestead food production and aquaculture in rural Cambodia. Ethical approval for the study was granted by the Clinical Research Ethics Board at the University of British Columbia (Canada) and the National Ethics Committee for Health Research (Cambodia).   Blood Collection and Analyses A 3-hour fasting capillary blood sample was taken from each woman at her home and processed immediately using the HemoCue device. Standard HemoCue 201+ microcuvettes were used to collect 2-3 drops of blood and were immediately inserted into the device for analysis. These specially designed microcuvettes act as a blood collection vessel and also contain sodium deoxycholate, which disintegrates the erythrocyte membrane and releases hemoglobin. Sodium nitrate then converts the hemoglobin iron from ferrous to ferric state to form methemoglobin, which then combines with azide to form azidemethemoglobin. This compound is then measured by a spectrophotometer. Phlebotomists were trained on procedures as per guidelines in the HemoCue 201+ operating manual (35) and standardized procedures (25).   The following morning, a 3-hour fasting venous blood sample was collected from the same women at health centers in Prey Veng by trained phlebotomists from the Cambodian National Institute of Public Health Laboratory. Venous blood was collected in an evacuated 3.5 mL tube (Becton Dickinson) containing an anticoagulant (EDTA), placed on ice and transported daily to   168 the laboratory in Phnom Penh for analysis. Venous blood was analyzed using a Sysmex hematology analyzer (Sysmex XT-1800i). A complete blood count was performed to determine hemoglobin concentrations (36). This system uses sodium laurel sulphate to convert hemoglobin to a colored compound that is measured by an automated spectrophotometer (8).   Quality control tests using both the HemoCue device and Sysmex analyzer showed that both complied with minimum standards with the use of quality control solutions. Tests on the HemoCue device were conducted using HemoTrol (Level II) quality control solution (Eurotrol BV, Ede, The Netherlands) at three different times and all control values were within acceptable levels (±6 g/L as defined by HemoTrol). Quality control tests on the Sysmex analyzer were conducted by technicians at the laboratory in Cambodia using three different levels of Sysmex e-Check control solution (Sysmex Corporation, Kobe, Japan) and showed that all values were within acceptable limits (range of CV=0.4-0.7%).  Genetic hemoglobin disorders were identified using hemoglobin electrophoresis and polymerase chain reaction (37). A detailed methodology and the frequencies of identified genotypes in the 420 women have been published elsewhere (31). Women were categorized into three groups based on the presence of hemoglobin disorders and type of hemoglobin disorders present: no hemoglobin disorders (n=195), hemoglobin E variants (including heterozygosity or homozygosity with or without any other co-inherited variant, n=133), and other hemoglobin disorders (e.g., hemoglobin Constant Spring (CS), α- and β-thalassemia, n=92). Altman & Bland have suggested that a minimum sample size of n=50 is required for calculation of bias and precision in method comparison studies (38). Therefore, we did not further segregate the   169 hemoglobin E variants into categories of heterozygosity and homozygosity due to the rarity of hemoglobin E homozygotes and their corresponding small sample size (n=31). Data Analyses and Statistical Methods Hemoglobin concentration (g/L) is presented as the mean  SD. The prevalence of anemia in each group of women was determined using the hemoglobin cut-off for non-pregnant women of reproductive age (hemoglobin <120 g/L) (2) and presented as the total number of women with anemia and the proportion of women with anemia among the total study population (%).   Bland & Altman have suggested the determination of bias (agreement) and precision (limits of agreement) and the plots of these values as the most appropriate analysis for a method-comparison study of two clinical measurements (39). Bias was defined as the difference in means between the two measures of hemoglobin concentration (g/L) and was reported as the mean  standard error of the mean (SEM). Precision was defined as the limits of agreement, or more commonly termed the 95% CI of the bias, and was reported as 1.96 SD. Precision plots are interpreted visually to compare discrepancies between methods (bias) and the width of the limits of agreement (precision and clinical significance), and any trends as the mean increases (consistency of variability).   Lin has suggested the concordance correlation coefficient (c) as the most appropriate method to determine the reproducibility between two measured values, as it measures the departure of the measured values from a 45-degree line (40). Reproducibility is also an important component of method-comparison studies, as if one or both methods do not provide repeatable results, then agreement is not useful to report alone (41). Concordance was calculated and presented as the   170 coefficient factor (c). In addition, concordance was plotted for three groups: women with no genetic hemoglobin disorders, women with hemoglobin E variants, and women with other hemoglobin variants (not including hemoglobin E variants). Pearson’s correlation coefficient (r) was not determined as it is potentially misleading to use Pearson’s coefficient in this comparison of two clinical measurements as it only measures the strength of the association and fails to detect agreement between values (departure from the 45 degree line) (39,40).   Accuracy in this study refers to the comparability of hemoglobin measurements using the HemoCue device and using a hematology analyzer, which is considered to be the gold standard method in clinical settings. If the bias is small and the limits of agreement are narrow (considering clinical significance) then it is suggested that the two methods are equivalent (39–41). T-tests, analysis of variance, and chi-square tests were used to conduct pairwise comparisons and measure statistically significant differences between groups. LSD was used to adjust for multiple comparisons when required. Two-sided P-values less than 0.05 indicated statistical significance. Stata software version SE/13.1 for Mac (Stata Corp, College Station, Texas) was used to conduct statistical analyses.  Results Data were available for 420 non-pregnant women 18-45 years from rural Prey Veng province in Cambodia. Using the HemoCue device, 172 women (41.0%) were determined to have anemia based on the cut-off for non-pregnant women of reproductive age (hemoglobin <120 g/L) as per WHO guidelines. Of those 172 women, 82 (47.7%) had mild anemia (110-120 g/L), 90 (52.3%) had moderate anemia (80-110 g/L), and no women had severe anemia (<80 g/L). Using the   171 Sysmex analyzer, 124 women (29.5%) were determined to have anemia. Of those 124 women, 85 (68.5%) had mild anemia, 39 (31.5%) had moderate anemia, and no women had severe anemia. We detected a significant difference in anemia prevalence among women overall (n=420) using HemoCue and Sysmex methods (41.0% vs. 29.5%, respectively, P<0.001).   Table 1 presents the hemoglobin concentration, anemia prevalence, bias, precision, and Pearson’s and concordance coefficients among the three groups of women using the HemoCue and Sysmex methods. Bias, precision and concordance appeared similar among all groups. Bland & Altman’s bias and precision plots (Figure 1) and Lin’s concordance plots (Figure 2) are presented for A) women with no genetic hemoglobin disorders, B) women with hemoglobin E variants, and C) women with other hemoglobin variants. In all three groups, the reduced major axis (the fitted line) on the concordance plot is not aligned to the line of perfect concordance, but rather tilted slightly clockwise. The HemoCue method appears to underestimate hemoglobin concentrations in capillary blood as compared to Sysmex (venous blood) at lower hemoglobin concentrations, and to overestimate hemoglobin concentrations in capillary blood as compared to Sysmex (venous blood) at higher hemoglobin concentrations.    172 Table 1 Hemoglobin concentration, anemia prevalence, bias, precision and concordance coefficients among Cambodian women 18-45 years using two different methods of measurement (HemoCue, hemoglobinometer and Sysmex XT-1800i, hematology analyzer) Population group n Hemoglobin concentration  Anemia prevalence1  Bias  Precision  (95% CI) Concordance   n  mean g/L  SD  n (%) mean g/L  SEM  1.96SD c   HemoCue Sysmex HemoCue Sysmex             All women 420  121.9  13.3 124.5  10.9 172 (41.0)* 124 (29.5)* 2.6  0.39 -13.07, 18.18 0.77**          A: Women with no genetic Hb disorders  195  127.0  11.9 129.5  8.9 46 (23.6)* 22 (11.3)* 2.5  0.59 -13.53, 18.50 0.68**          B: Women with Hb E variants2   C: Women with other Hb variants3 133    92  117.4  12.9   117.8  12.7 119.8  10.7   120.7  10.3 74 (55.6)   52 (56.5)* 69 (51.9)   33 (35.9)* 2.5  0.66   2.7  0.84 -12.54, 17.45   -12.97, 18.70 0.78**   0.73** 1 Based on non-pregnant women of reproductive age (hemoglobin <120 g/L). Hb, hemoglobin. 2 Includes women with hemoglobin E variants including heterozygosity or homozygosity with or without any other co-inherited variant. 3 Includes women with hemoglobin CS variants, α- and β-thalassemias, and other rare hemoglobin disorders (not including hemoglobin E). * Pearson’s chi square test showed a statistically significant difference in anemia prevalence between the HemoCue and Sysmex methods (P<0.001). ** Statistically significant (P<0.05).    173  Figure 1 Bias and precision plots among women with A) no genetic hemoglobin disorder, B) hemoglobin E variants, and C) other hemoglobin variants (not including hemoglobin E).    Figure 2 Concordance plots among women with A) no genetic hemoglobin disorder, B) hemoglobin E variants, and C) other hemoglobin variants (not including hemoglobin E).   174 Figure 3 displays the box plots for hemoglobin concentration using both the HemoCue and Sysmex methods for all women (n=420). The dashed horizontal line indicates the hemoglobin cut-off level for anemia classification for non-pregnant women of reproductive age (hemoglobin <120 g/L) (2). The box plot for HemoCue hemoglobin measurements has a wider spread (larger interquartile range and wider whiskers) compared to the box plot for Sysmex hemoglobin measurements.    Figure 3 Box plots for hemoglobin measurement using Hemocue and Sysmex methods for all women (n=420). The red dashed horizontal line indicates the hemoglobin cut-off level for anemia classification for non-pregnant women (<120 g/L).  Based on the cut-off value for anemia in non-pregnant women of reproductive age (hemoglobin <120 g/L), an additional n=48 women were considered anemic using the HemoCue device, compared to the Sysmex analyzer. In this case, an overall bias of 2.6 g/L (difference in hemoglobin means between methods) was detected (n=420), which resulted in a significant  175 difference in anemia prevalence of 11.5% between methods (41.0% using HemoCue and 29.5% using Sysmex, P<0.001).  Discussion and Conclusions We hypothesized that the measurement of hemoglobin may be compromised using HemoCue devices in women with hemoglobin variants due to the altered structure and chemically unstable hemoglobin molecule. However, hemoglobin measurements of women with no genetic hemoglobin disorders, women with hemoglobin E variants, and women with other hemoglobin variants (not including hemoglobin E) showed similar bias and agreement, suggesting that the two methods of hemoglobin measurement are similar among these three groups. Despite this, we cannot conclude that either or both methods were accurate, as hemoglobin variants may be measured comparably yet with limited accuracy in both methods, which cannot be determined from this study.   Given the rarity of hemoglobin E homozygotes, we only had a small sample size of women (n=31) in our study. Since this was a secondary analysis, we were not able to collect more samples to obtain larger groups of women with these rare hemoglobin disorders and therefore we could not statistically analyze these groups independently. In hemoglobin E heterozygotes, the variant hemoglobin usually comprises ≤30% of the total hemoglobin, compared to >40% in hemoglobin E homozygotes [37,47]. However, in the women with hemoglobin E homozygosity in this study (n=31) the variant hemoglobin comprised ~85-95% of total hemoglobin (31). Combining these groups to achieve more statistical power may have compromised our ability to  176 detect differences in agreement between methods in women with more ‘severe’ forms, notably for women with hemoglobin E homozygosity.  We explored the hemoglobin variants among two major groups of women: hemoglobin E variants and other hemoglobin variants. We analyzed hemoglobin E variants as an independent group, as they are the most common hemoglobin variant in Cambodia and therefore we had a large enough sample size (n=133). We speculated that if there were indeed differences in hemoglobin measurement among individuals with hemoglobin variants, it would be most apparent in individuals with hemoglobin E variants, given the high molecular instability of hemoglobin E molecules. The second group in our study contained women with other hemoglobin variants (n=92), including numerous different hemoglobin variants: hemoglobin CS, α- and β-thalassemias, and other rare hemoglobin disorders. Due to small sample size of these other hemoglobin variants, we could not categorize them independently. Analyzing these other hemoglobin variants as a group may have compromised our ability to detect differences overall, as different variants may result in differing agreement between methods (due to varying stability and physical form). For example, hemoglobin CS is also common in Cambodia but less prevalent as compared to hemoglobin E (29), and evidence suggests that hemoglobin CS variants have increased membrane rigidity and stability in red blood cell membranes (42). In our study, we detected a small number of women (n=15) with hemoglobin CS variants. We did not categorize women with hemoglobin CS variants as an independent group due to the small sample size. Furthermore, in other regions of the world, other hemoglobin variants exist. For example, hemoglobin S variants are prevalent in Africa and in other parts of the world (34). However,  177 hemoglobin S variants are not common in Cambodia (no women in our study had an hemoglobin S variant), therefore this variant could not be included in our method-comparison study.  Overall, we detected poor agreement between HemoCue and Sysmex methods to measure hemoglobin concentration in Cambodian women (bias >2 g/L in all three groups). The poor agreement between methods is likely attributed to several factors. Firstly, the two methods are measuring blood taken from the same woman but from different access points: capillary blood from the finger and venous blood from the arm. The lower hemoglobin concentration in capillary blood could be due to biological factors. Kupke et al. demonstrated that total protein concentration in capillary serum is approximately 3.3% lower than venous serum in fasting healthy adults (43). The small size of capillary blood vessels and the lower red cell volume in capillary blood may have contributed to the lower hemoglobin concentrations detected by the HemoCue device (capillary blood) in our study. However, other studies have shown conflicting results indicating identical (18) or higher (19,23) hemoglobin concentrations in capillary blood compared to venous blood. Therefore there are inconsistencies in the body of published evidence on this issue.   Secondly, poor agreement may be due to procedure or measurement errors that are common in point-of-care testing devices. Bias in our study could have been attributed to the fact that the HemoCue method itself is more susceptible to measurement error.  For example, it has been speculated that ‘milking’ the finger to stimulate blood flow after the finger prick can result in more plasma being pushed onto the microcuvette, causing a dilution effect on hemoglobin concentration (17). However, it is important to note that we took extensive efforts to adequately  178 train and test our field staff on proper HemoCue testing procedures using standardized techniques (25) and quality controls (9,35). Therefore, we believe this highlights the concern of using this method even among highly qualified and trained field staff.   Thirdly, the large variation in mean hemoglobin concentration in women in our study could have contributed to the poor agreement observed. Hemoglobin concentration has a natural biological variation in the blood, which is also common among multiple other biochemical indicators [52]. Morris et al. also showed that capillary blood from the left and right hand of the same individual varied greatly in regards to hemoglobin concentration (within subject CV=6.3%, c=0.69) (17). The wide limits of agreement suggested that blood samples from each finger could differ in hemoglobin concentration by as much as 20 g/L (17). Other factors have also been shown to influence variability in hemoglobin concentrations when measured using the HemoCue device such as the time of blood collection (44), body position during blood collection (45), and air humidity in tropical regions (46).  In our study, we observed larger CV (%) among mean hemoglobin concentrations in each group of women using the HemoCue device (capillary blood) as compared to the Sysmex method (venous blood). The higher variation was also visually apparent in the box plots. This is consistent with findings of other studies comparing variability between these two methods (17,18). This variation could be a cause of the aforementioned biological variation (as it uses capillary blood), or could also be due to measurement error inherent with the use of the field-based HemoCue device. In order to explore this further, we examined the precision of the HemoCue device in a smaller study using a sample of Cambodian adults (n=5) testing capillary  179 blood two times a day over three days by the same phlebotomist.  Using the HemoCue device, individuals showed minimal within-subject variation both within the same day (mean CV=2.6%, max CV=6.2%) and across three days (mean CV=3.1%, max CV=5.3%). We also compared the precision of three HemoCue devices using the same sample of adults (n=5) and also found minimal within-subject variation across devices (mean CV=3.1%, max CV=3.7%). This suggested that imprecision based on HemoCue measurement error and instrument error was low. However, this study was conducted in our Phnom Penh office while the larger study was conducted in rural field conditions, which may result in some bias in outcomes.   One of the most interesting findings is shown in the concordance plots. The HemoCue method appears to underestimate hemoglobin concentrations in capillary blood as compared to Sysmex (venous blood) at lower hemoglobin concentrations, and to overestimate hemoglobin concentrations in capillary blood as compared to Sysmex (venous blood) at higher hemoglobin concentrations. Therefore, hemoglobin measured using HemoCue (capillary blood) could result in underestimations of hemoglobin concentration, resulting in false positives in the diagnosis of anemia in individual women.  Despite the likelihood of measurement error, the difference between hemoglobin concentrations in capillary and venous blood, and the high variability among hemoglobin measurements, the bias between methods has important implications. Hemoglobin concentration is used as proxy indicator to determine the prevalence of anemia among populations. Without cautious interpretation and careful consideration of these factors, the prevalence of anemia can be largely under or overestimated.   180 Comparing the hemoglobin concentrations between methods, we observed a change in the WHO anemia severity classification from a moderate to a severe public health problem (2). This is because hemoglobin measurements are normally distributed and the mean hemoglobin among women is relatively close to the cut-off point for diagnosis of anemia. Therefore, a small shift of the mean hemoglobin (bias) results in a substantial shift in the anemia prevalence rate. Furthermore, of the women with anemia, a larger proportion of women were determined moderately anemic using HemoCue (52%, n=90/172) compared to the Sysmex method (32%, n=39/124), suggesting that both the overall prevalence and the severity of anemia classification are influenced by the methods. This is important when considering population data from a public health perspective. Policies and programs could potentially be impacted by this bias and scarce resources (e.g., funding and staffing) may not be appropriately dispersed to the highest prioritized needs.  A strength of this study is that it is the first to our knowledge to assess agreement between methods to measure hemoglobin concentration among individuals with genetic hemoglobin variants. The limitations are that we did not measure venous blood samples in the HemoCue device, which would have provided some insight to the level of measurement error potentially occurring in HemoCue blood collection procedures. However, this was not the aim of our study. In field practice, venous blood is used for hematology analyzer analysis and capillary blood is used for HemoCue analysis. Therefore, we aimed to compare these two methods using standard blood collection procedures. We also did not determine differences in hemoglobin measurements between these methods in men or in individuals with other hemoglobin variants that exist outside of Cambodia (e.g., hemoglobin S).   181 We conclude that hemoglobin measurements of women with no genetic hemoglobin disorders, women with hemoglobin E variants, and women with other hemoglobin variants (not including hemoglobin E) showed similar bias and agreement, suggesting that the HemoCue and Sysmex measurements of hemoglobin are similar among these three groups. The HemoCue is a practical and low-resource device to measure hemoglobin in field settings. It remains a commonly used method to determine hemoglobin concentration in national surveys and large studies that cannot afford the required time, money and/or technical expertise to assess hemoglobin concentration using other laboratory based methods. We emphasize the importance of following standardized protocols, providing adequate training of staff, and ensuring appropriate calibration and quality control of equipment to minimize measurement error when using the HemoCue device (25). We highlight our current findings with the intention to caution field staff, researchers and policy makers in the interpretation of data using different methods to measure hemoglobin concentration and the impact that a slight difference in means between methods can have on the prevalence rates of anemia. In our case, the bias detected between methods showed a significant difference in anemia prevalence (41.0% vs. 29.5%, P<0.001).  References 1.  McLean E, Cogswell M, Egli I, Wojdyla D, de Benoist B. Worldwide prevalence of anaemia, WHO Vitamin and Mineral Nutrition Information System, 1993-2005. Public Health Nutr. 2009;12(4):444–54.  2.  World Health Organization. Assessing the iron status of populations. 2nd ed. Geneva: World Health Organization; 2007.  3.  Allen LH. Anemia and iron deficiency: effects on pregnancy outcome. Am J Clin Nutr.  182 2000;71:1280–4.  4.  Xiong X, Buekens P, Alexander S, Demianczuk N, Wollast E. Anemia during pregnancy and birth outcome: a meta-analysis. Am J Perinatol. 2000;17(3):137–46.  5.  Stoltzfus RJ, Mullany L, Black RE. Iron deficiency anemia. In: Ezzati M, Lopez AD, Rodgers A, Murray CJL, editors. Comparative quantification of health risks: global and regional burden of disease attributable to selected major risk factors. Geneva: World Health Organization; 2004. p. 163–209.  6.  Hurtado EK, Claussen AH, Scott KG. Early childhood anemia and mild or moderate mental retardation. Am J Clin Nutr. 1999;69:115–9.  7.  Black RE, Victora CG, Walker SP, Bhutta Z a, Christian P, de Onis M, et al. Maternal and child undernutrition and overweight in low-income and middle-income countries. Lancet. 2013 Aug 3;382(9890):427–51.  8.  Fujimoto K. Principles of measurement in hematology analyzers manufactured by Sysmex Corporation. Sysmex J Int. 1999;9(1):31–44.  9.  Gulati GL, Hyun BH. Quality control in hematology. Clin Lab Med. 1986 Dec;6(4):675–88.  10.  Mohandas N, Kim Y, Tycko D, Orlik J, Wyatt J, Groner W. Accurate and independent measurement of volume and hemoglobin concentration of individual red cells by laser light scattering. Blood. 1986;68(2):506–13.  11.  ICF International/Demographic and Health Surveys Calverton. Biomarker field manual: demographic and health surveys methodology. Washington, DC; 2012.  12.  Von Schenck H, Falkensson M, Lundberg B. Evaluation of “HemoCue,” a new device for determining hemoglobin. Clin Chem. 1986;32(3):526–9.   183 13.  Bridges N, Parvin R, van Assendelft O. Evaluation of a new system for hemoglobin measurement. Am Clin Prod Rev. 1987;6:22–5.  14.  Neville RG. Evaluation of portable haemoglobinometer in general practice. Br Med J (Clin Res Ed). 1987;294(6582):1263–5.  15.  Cohen AR, Seidl-Friedman J. HemoCue system for hemoglobin measurement. Evaluation in anemic and nonanemic children. Am J Clin Pathol. 1988;90(3):302–5.  16.  Hudson-Thomas M, Bingham KC, Simmons WK. An evaluation of the HemoCue for measuring haemoglobin in field studies in Jamaica. Bull World Health Organ. 1994;72(3):423–6.  17.  Morris SS, Ruel MT, Cohen RJ, Dewey KG, de la Brière B, Hassan MN. Precision, accuracy, and reliability of hemoglobin assessment with use of capillary blood. Am J Clin Nutr. 1999;69(6):1243–8.  18.  Sari M, De Pee S, Martini E, Herman S, Sugiatmi, Bloem MW, et al. Estimating the prevalence of anaemia: a comparison of three methods. Bull World Health Organ. 2001;79(6):506–11.  19.  Neufeld L, García-Guerra A, Sánchez-Francia D, Newton-Sánchez O, Ramírez-Villalobos MD, Rivera-Dommarco J. Hemoglobin measured by Hemocue and a reference method in venous and capillary blood: a validation study. Salud Publica Mex. 2002;44(3):219–27.  20.  Bhaskaram P, Balakrishna N, Radhakrishna K V, Krishnaswamy K. Validation of hemoglobin estimation using Hemocue. Indian J Pediatr. 2003;70(1):25–8.  21.  Nkrumah B, Nguah SB, Sarpong N, Dekker D, Idriss A, May J, et al. Hemoglobin estimation by the HemoCue® portable hemoglobin photometer in a resource poor setting. BMC Clin Pathol. 2011;11(1):5.   184 22.  Adam I, Ahmed S, Mahmoud MH, Yassin MI. Comparison of HemoCue® hemoglobin-meter and automated hematology analyzer in measurement of hemoglobin levels in pregnant women at Khartoum hospital, Sudan. Diagn Pathol. BioMed Central Ltd; 2012;7(1):30.  23.  Zhou X, Yan H, Xing Y, Dang S, Zhuoma B, Wang D. Evaluation of a portable hemoglobin photometer in pregnant women in a high altitude area: a pilot study. BMC Public Health. 2009;9:228.  24.  Nestel P. Adjusting hemoglobin values in program surveys. Washington, DC: ILSI Human Nutrition Institute; 2002.  25.  Burger S, Pierre-Louis J. A procedure to estimate the accuracy and reliability of HemoCueTM measurements of survey workers. Washington, DC; 2003.  26.  Van De Louw A, Lasserre N, Drouhin F, Thierry S, Lecuyer L, Caen D, et al. Reliability of HemoCue in patients with gastrointestinal bleeding. Intensive Care Med. 2007;33(2):355–8.  27.  Rudolf-Oliveira RCM, Gonçalves KT, Martignago ML, Mengatto V, Gaspar PC, Ferreira JDS, et al. Comparison between two portable hemoglobinometers and a reference method to verify the reliability of screening in blood donors. Transfus Apher Sci. Elsevier Ltd; 2013;49(3):578–82.  28.  Rippmann CE, Nett PC, Popovic D, Seifert B, Pasch T, Spahn DR. Hemocue, an acurate bedside method of hemoglobin measurement? J Clin Monit. 1997;13:373–7.  29.  Carnley BP, Prior JF, Gilbert A, Lim E, Devenish R, Sing H, et al. The prevalence and molecular basis of hemoglobinopathies in Cambodia. Hemoglobin. 2006;30(4):463–70.  30.  George J, Yiannakis M, Main B, Devenish R, Anderson C, An US, et al. Genetic  185 hemoglobin disorders, infection, and deficiencies of iron and vitamin A determine anemia in young Cambodian children. J Nutr. 2012;142:781–7.  31.  Karakochuk CD, Whitfield KC, Barr SI, Lamers Y, Devlin AM, Vercauteren SM, et al. Genetic hemoglobin disorders rather than iron deficiency are a major predictor of hemoglobin concentration in women of reproductive age in rural Prey Veng, Cambodia. J Nutr. 2015;145(1):134–42.  32.  Frischer H, Bowman J. Hemoglobin E, an oxidatively unstable mutation. J Lab Clin Med. 1975;85(4):531–9.  33.  Yuthavong Y, Ruenwongsa P, Benyajati C, Suttimool W. Studies on the structural stability of haemoglobin E. J Med Assoc Thai. 1975;58(7):351–6.  34.  Bain BJ. Haemoglobinopathy diagnosis. 2nd ed. Oxford: Blackwell Publishing Ltd; 2006.  35.  Hemocue. HemoCue Hb 201+ Operating Manual. Angelholm, Sweden; 1995.  36.  Sysmex. Sysmex XE-2100 Operator’s Manual. Kobe, Japan: Sysmex Corporation; 2007.  37.  Hartwell SK, Srisawang B, Kongtawelert P, Christian GD, Grudpan K. Review on screening and analysis techniques for hemoglobin variants and thalassemia. Talanta. 2005;65:1149–61.  38.  Altman DG, Bland JM. Measurement in medicine: the analysis of method comparison studies. J R Stat Soc. 1983;32(3):307–17.  39.  Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;8:307–10.  40.  Lin LI. A concordance correlation coefficient to evaluate reproducibility. Biometrics. 1989;45(1):255–68.  41.  Hanneman SK. Design, analysis and interpretation of method-comparison studies. AACN  186 Adv Crit Care. 2008;19(2):223–34.  42.  Fucharoen S, Winichagoon P. Haemoglobinopathies in Southeast Asia. Indian J Med Res. 2011;134:498–506.  43.  Kupke IR, Kather B, Zeugner S. On the composition of capillary and venous blood serum. Clin Chim Acta. 1981;112(2):177–85.  44.  Boulton FE, Nightingale MJ, Reynolds W. Improved strategy for screening prospective blood donors for anaemia. Transfus Med. 1994;4(3):221–5.  45.  Gore CJ, Scroop GC, Marker JD, Catcheside PG. Plasma volume, osmolarity, total protein and electrolytes during treadmill running and cycle ergometer exercise. Eur J Appl Physiol Occup Physiol. 1992;65(4):302–10.  46.  Nguyen H. High humidity affects Hemocue cuvette function and Hemocue haemologin estimation in tropical Australia. J Pediatr Child Heal. 2002;38:427–8.    187 Appendix B: Comparison of Immunoassays to Measure Serum Ferritin Concentration  Acknowledgement: A version of this appendix has been published. Karakochuk CD, Whitfield KC, Rappaport AI, Barr SI, Vercauteren SM, McLean J, Hou K, Talukder A, Houghton LA, Bailey KB, Boy E, Green TJ. Comparison of four immunoassays to measure serum ferritin concentrations and iron deficiency prevalence among non-pregnant Cambodian women and Congolese children. Clinical Chemistry and Laboratory Medicine 2017; 55(1): 65-72.  Summary Background: Global standardization of ferritin assays is lacking, which could have direct implications on the accurate measurement and comparability of ferritin concentration and iron deficiency prevalence rates in at-risk populations. Methods: We measured serum ferritin concentrations using four immunoassays: the s-ELISA and the AxSYM™ analyzer were compared among 420 non-pregnant Cambodian women; the Centaur® XP analyzer, s-ELISA, and AxSYM™ analyzer were compared among a subset of 100 Cambodian women; and the s-ELISA and the Elecsys® 2010 analyzer were compared among 226 Congolese children 6-59 months. Results: Median ferritin concentrations (adjusted for inflammation) ranged between 48-91 μg/L among Cambodian women and between 54-55 μg/L among Congolese children. Iron deficiency prevalence ranged from 2-10% among Cambodian women and 5-7% among Congolese children. Bias between methods varied widely (-9 to 45 μg/L) among women, and was 43 μg/L among children. Bias was lower when ferritin values outside of the s-ELISA measurement range (>250 μg/L) were excluded. Conclusions: The observed differences in  188 ferritin concentrations likely reflect different ferritin isoforms, antibodies, and calibrators used across assays and by different laboratories. However, despite differences in ferritin concentrations, iron deficiency prevalence was relatively similar and low across all methods.   Introduction Iron deficiency is thought to affect over 2 billion people worldwide (1,2) and can have severe consequences for women during pregnancy (3,4) and for children’s early brain development and growth (5,6). The gold standard test for iron deficiency is the assessment of iron stores through a bone marrow aspirate (7,8); however, this is an invasive and painful test that is rarely used in practice to diagnose iron deficiency. Serum or plasma ferritin concentration is a more commonly measured biomarker reflecting the depletion of iron stores in the body (7). Historically, immunoradiometric assays (using labeled antibodies) and radioimmunoassays (using labeled ferritin) were the primary method of ferritin measurement (7). Over the last few decades, automated immunoassay analyzers have been developed (e.g., Abbott AxSYM™), eliminating the need for immunoradiometric methods. In 2004, Erhardt et al. (9) developed a s-ELISA, which concurrently measures ferritin, CRP, AGP, sTfR, and RBP concentrations in a small sample of plasma or serum. This low-cost method has shown low intra- and inter-assay variability and high sensitivity (9); as such, the method has become increasingly popular worldwide.  The quantification of ferritin concentration in these immunoassays is based on the detection of specific antibody binding (10). However, many challenges in the traceability of this method have been identified, as laboratories differ in terms of which ferritin isoforms are measured (e.g.,  189 isoforms found in the liver are different from those in the spleen), the antibodies selected for ferritin detection, and the reference ranges established and utilized (11,12). In 1985, the WHO established the 1st international standard (IS) for ferritin (liver, 80/602) as a reference for methods to be calibrated against in attempt to improve global traceability of methods (13). Since then, the 2nd IS was released in 1993 (spleen, 80/578) (14) and more recently the 3rd IS in 1997 (recombinant, 94/572) (15). An evaluation of the 3rd IS by 18 laboratories in nine countries showed adequate stability in accelerated degradation studies and acceptable traceability to the 2nd IS (16). However, calibration to the 3rd IS is not globally mandated or monitored and reference ranges continue to differ across laboratories. Many laboratories are still tracing ferritin methods to the 1st and 2nd IS, despite the fact that these materials ceased production in the mid-1990s and have been since superseded.  This lack of standardization of ferritin assays could have direct implications on the accurate measurement and comparability of ferritin concentrations using different methods, and of greater concern, on iron deficiency prevalence rates in at-risk populations. The aim of this study was to measure serum ferritin concentrations and compare iron deficiency prevalence estimates using four different immunoassays with varying quality controls and traceability. These include three automated immunoassay analyzers (Abbott AxSYM™, Siemen ADVIA Centaur® XP, and Roche Elecsys® 2010) and Erhardt et al.’s s-ELISA in two groups of individuals: Cambodian women of reproductive age and Congolese children.     190 Materials and Methods  Study Design and Participants This method-comparison study included data collected from baseline surveys in two trials: n=420 non-pregnant women 18-45 years from Prey Veng province who were selected for inclusion in a separate trial evaluating the outcomes of an improved homestead food production and aquaculture project (17); and n=226 children 6-59 months representatively sampled from South Kivu and Bas Congo provinces, Democratic Republic of the Congo, as part of a larger micronutrient survey (18). Work was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). Ethics approval was obtained from the University of British Columbia (Canada), the Université de Kinshasa (Democratic Republic of the Congo), the Université Catholique de Bukavu (Democratic Republic of the Congo), and the National Ethics Committee for Health Research (Cambodia). All participants (or caregivers on behalf of the children) provided written informed consent to participate in the studies.  Blood Collection In Cambodia, a 3-hour fasting venous blood sample was collected from women at health centers in Prey Veng in June 2012 by trained phlebotomists from the National Institute of Public Health Laboratory. Blood was collected in a trace element free 3.5 mL tube (Becton Dickinson), allowed to clot, placed indirectly on ice, and transported daily to the laboratory in Phnom Penh. In the Democratic Republic of the Congo, an overnight fasting venous blood sample was collected from children from health centers in South Kivu and Bas Congo provinces between June and October 2014 by trained phlebotomists. Blood was collected in a trace element free 8  191 mL tube (Becton Dickinson), allowed to clot, placed indirectly on ice, and transported to the primary health center for processing. Samples from both Cambodia and Democratic Republic of the Congo were centrifuged within 2-4 hours from the time of collection. Serum was separated and aliquoted into cryovials for storage at -80 °C until shipment to the University of British Columbia, where it was again stored at -80 °C until secondary shipment to respective laboratories for further analysis. During transport, samples were shipped on dry ice and thawed only at the time of sample analysis. Analyses for the s-ELISA were conducted within two months of sample collection for the Cambodia study and within four months for the Democratic Republic of the Congo study. Analyses for the Abbott AxSYM™ were conducted within six months of sample collection for the Cambodia study, within 11 months for the Siemen ADVIA Centaur® XP for the Cambodia study, and within six months for the Roche Elecsys® 2010 for the Democratic Republic of the Congo study.  Laboratory Analyses Serum ferritin concentration was analyzed using a total of four different methods in four different laboratories. A summary of the methods and the respective laboratories where assays were conducted is provided in Table 1.  192 Table 1 Summary of the methods and the respective laboratories where assays were conducted1   s-ELISA Abbott AxSYM™ ADVIA Centaur® XP Roche Elecsys® 2010 Overview of method Sandwich linked immunosorbent assay  Automated immunoassay analyzer Automated immunoassay analyzer Automated immunoassay analyzer Company/establishment Erhardt et al. Willstaett, Germany Abbott Labs,  North Chicago, USA Siemens AG, Munich, Germany Roche Diagnostics,  Basel, Switzerland Principle of method Uses a 96-well plate and ferritin antibody detection (Code A0133, Dako) to quantify ferritin concentrations (9) Uses a combination of three analytical techniques: microparticle enzyme assay, fluorescence polarization immunoassay, and ion-capture immunoassay for antibody detection (19) Uses a two-site sandwich immunoassay using direct chemiluminometric technology with two anti-ferritin antibodies (20) Uses an electrochemiluminescence detection cell and a kit containing ferritin specific reagents (21) Sample required ~30 μL of serum or plasma ~100 μL of serum or plasma ~25 μL of serum or plasma ~125 μL of serum or plasma Measurement range 0-250µg/L 0-2000 µg/L 0.5-1650 μg/L 0.5-2000 µg/L Traceability to WHO IS Not traced2 Indirectly to the WHO  1st IS (08/602)3 Directly to the WHO  2nd IS (80/578) Indirectly to the WHO  1st IS (80/602)4 Quality controls used CDC In-house Controls5  & BioRad Liquichek™ Immunology Control Level 3 (Ref # 593) Abbott AxSYM™  Ferritin Controls  Levels 1, 2 and 3 (Ref # 7K45-10) BioRad Lypochek™ Immunoassay Plus Controls Levels 2 and 3 (Ref # 362, 363) Roche Elecsys® PreciControl Varia Levels 1 and 2 (Cat # 05618860190) Laboratory where ferritin analyses were conducted Human Nutrition Lab, Willstaett, Germany Innis Lab, Child and Family Research Institute, Vancouver, Canada Lifelabs Inc., Burnaby, Canada Houghton Lab, University of Otago, Dunedin, New Zealand 1 CDC, Centers for Disease Control and Prevention (Atlanta, USA); IS, international standard; WHO, World Health Organization.  2 The lab uses serum-based CDC pools, which are assigned concentrations using the Roche Elecsys® assay using the E170 analyser.  3 The AxSYM™ ferritin standard calibrators are matched to internal standards, which were originally traceable to the WHO 1st IS (80/602).  4 The Elecsys® ferritin assay (#04491785) has been standardized against the previous one (#11820982), which was standardized against the Enzymun-Test for ferritin. This in turn has been standardized against the WHO 1st IS (80/602).  5 The laboratory uses two levels of controls: CDC controls are used for low serum ferritin values (up to ~80 µg/L) and BioRad controls are used for high serum ferritin values (>250 µg/L).   193 Data Analyses Serum ferritin concentration (μg/L) is presented as median (IQR). The prevalence of iron deficiency in each group was determined using the cut-offs for non-pregnant women of reproductive age (ferritin <15 μg/L) and children (ferritin <12 μg/L) (7). We present results for unadjusted serum ferritin and inflammation-adjusted serum ferritin (adjusted for levels of inflammation based on CRP and AGP biomarkers using correction factors suggested by Thurnham et al. (22). CRP and AGP concentrations were measured using Erhardt et al.’s s-ELISA in both the Cambodia and Congolese samples.  We calculated Bland & Altman’s bias (agreement) and limits of agreement (23), and plotted the clinical measurements of serum ferritin concentrations. Bias was defined as the difference in means between the two measures of ferritin concentration (μg/L) and was reported as the mean  SD. The limits of agreement (95% CI of the bias) was reported as  1.96 SD.   Calculation of the limits of agreement is based on the assumption that the differences between the methods are normally distributed (24). This assumption was tested and confirmed before proceeding with the statistical test. Limit of agreement plots provide a visual comparison of discrepancies between two methods (bias), the width of the limits of agreement, and also show potential trends in the data (consistency of the variability). If the bias was small and the limits of agreement were narrow (considering clinical significance) between any two methods, those two methods were then interpreted as equivalent (23,25). These methods are in concordance with the 2013 Clinical and Laboratory Standards Institute measurement procedure guidelines for method comparison studies (26), with the exception that not all values were measured in duplicate. Only  194 for the s-ELISA were values measured in duplicate and the mean of two independent measurements reported.  We used Lin’s (27) concordance correlation coefficient (c) to measure the reproducibility between two measured values (the departure of the measured values from a 45-degree line). Pearson’s correlation coefficient (r) was also reported for interest of comparison; however, we acknowledge that it only measures the strength of the association and fails to detect the agreement between values (23). We used all available data for the primary analyses (bias and concordance) and also conducted sub-group analyses using only those measurements that fell within the measurement range of the defined laboratory/method. As such, in the sub-group analyses we excluded all ferritin measurements that exceeded 250 μg/L in the s-ELISA method (Cambodian: n=13; Congolese n=17). Stata version SE/13.1 for Mac (Stata Corp, College Station, Texas) was used to conduct statistical analyses.  Results Intra-assay Variation among Methods The intra-assay CV for ferritin was 3.3% using Erhardt et al.’s s-ELISA (US Centers for Disease Control and Prevention [Atlanta, USA] QC controls and BioRad [Hercules, USA] Liquichek™ immunoassay controls), and was 2.6% using the Elecsys® 2010 method (Roche Elecsys® PreciControl Varia controls). The intra-assay CV data for the Abbott AxSYM™ and ADVIA Centaur® XP methods were not available.    195 Serum Ferritin Concentrations and Iron Deficiency Prevalence across Methods Table 2 presents the median (IQR) and min/max serum ferritin concentrations (adjusted for inflammation using correction factors proposed by Thurnham et al. (22)) and the prevalence of iron deficiency among non-pregnant Cambodian women 18-45 years and Congolese children 6-59 months using the different methods. We observed differences in ferritin concentrations across methods. However, iron deficiency prevalence rates were relatively similar and surprisingly low: ranging between 2-7% across two methods among all 420 Cambodian women, between 5-10% across three methods among a subset of n=100 Cambodian women, and between 5-7% across two methods among all 226 Congolese children.   196 Table 2 Serum ferritin concentration and the prevalence of iron deficiency among non-pregnant Cambodian women 18-45 years and Congolese children 6-59 months using different methods of measurement  Participants & Methods Unadjusted ferritin  concentration (μg/L) Adjusted ferritin concentration (μg/L)1 Iron deficiency prevalence1,2  Median (IQR)3 Median (IQR)3 n (%)     Cambodian women (n=420)    s-ELISA 90.8 (59.0, 142.6) 84.2 (52.6, 126.3) 9 (2.1) AxSYM™ 52.0 (53.8, 73.9) 47.6 (31.3, 65.9) 28 (6.7)     Subset of Cambodian women (n=100)    s-ELISA 92.1 (57.9, 147.8) 91.0 (53.8, 125.6) 5 (5.0) AxSYM™ 53.8 (33.8, 77.2) 51.4 (31.0, 66.9) 10 (10.0) Centaur® XP 57.4 (37.7, 91.2) 51.5 (34.8, 85.8) 6 (6.0)     Congolese children (n=226)    s-ELISA 68.4 (34.2, 142.2) 55.4 (27.8, 96.1) 16 (7.1) Elecsys® 2010 70.1 (34.9, 153.3) 53.8 (27.0, 98.1) 12 (5.3)     1 Ferritin concentrations adjusted for inflammation using correction factors proposed by Thurnham et al.  2 Based on ferritin <15 μg/L for non-pregnant women or <12 μg/L for children. 3 IQR= Interquartile range (25th percentile, 75th percentile).  197 Bias and Concordance between Methods Table 3 presents the bias (difference in means), limits of agreement, and the Pearson’s and concordance correlation coefficients (95% CI) among the groups. We confirm the differences in methods were normally distributed among all groups, which is an assumption required to be met for the calculation of the limits of agreement (24). Method comparison analyses were conducted on unadjusted ferritin concentrations. Bias between methods was large and varied from ~35-45 μg/L for all method comparisons, with the exception of the AxSYM™ and Centaur® XP comparison among the subset of n=100 Cambodian women. In this latter comparison, bias was only ~9 μg/L and the concordance coefficient (95% CI) was very high, indicating near perfect agreement (0.96 [0.94, 0.97]) between the AxSYM™ and Centaur® XP methods.    198 Table 3 Bias, limits of agreement, and correlation coefficients among non-pregnant Cambodian women 18-45 years and Congolese children 6-59 months using different methods of measurement1 Participants Bias  (difference in means) Limits of agreement  Pearson’s coefficient Concordance coefficient (95% CI)  mean μg/L  SD  1.96 SD r c  ( 1.96 SD)      Cambodian women       s-ELISA vs. AxSYM™ (n=420) 45.4  28.6 -10.7, 101.6 0.94 0.62 (0.58, 0.66) s-ELISA vs. AxSYM™ (n=407)2  44.0  27.1 -9.1, 97.2 0.93 0.53 (0.49, 0.57) Subset of Cambodian women      s-ELISA vs. AxSYM™ (n=100) 43.2  27.5 -10.7, 97.1 0.94 0.69 (0.62, 0.75) s-ELISA vs. Centaur® XP (n=100) 34.6  25.6 -15.5, 84.8 0.93 0.78 (0.72, 0.84) AxSYM™ vs. Centaur® XP (n=100) -8.6  12.8 -33.7, 16.6 0.98 0.96 (0.94, 0.97)      Congolese children       s-ELISA vs. Elecsys® 2010 (n=226) -42.6  194.8 -424.4, 339.2 0.76 0.45 (0.40, 0.50) s-ELISA vs. Elecsys® 2010 (n=209)3 -11.5  29.8 -69.8, 46.9 0.96 0.91 (0.89, 0.93)      1 Bias and concordance were assessed using unadjusted ferritin concentrations. 2 Excluding n=13 ferritin values exceeding 250 μg/L in the s-ELISA method (outside of the defined measurement range). 3 Excluding n=17 ferritin values exceeding 250 μg/L in the s-ELISA method (outside of the defined measurement range).  199 Visual interpretation of the limits of agreement provided additional observations. Figure 1 (A-E) shows Bland limits of agreement plots for the five comparisons. Among all 420 Cambodian women, high bias was observed among the s-ELISA and the AxSYM™ method (Figure 1A), as the reduced major axis was not aligned to the line of perfect concordance. Similarly, this was also shown among the subset of n=100 Cambodian women between the s-ELISA and the AxSYM™ method (Figure 1B) and s-ELISA and the Centaur® XP method (Figure 1C). However, in the same subset of n=100 Cambodian women, the AxSYM™ and Centaur® XP showed very high agreement (Figure 1D). Among the 420 Cambodian women, n=13 measurements exceeded 250 μg/L in the s-ELISA (the upper limit of the method’s measurement range) and n=2 measurements exceeded 250 μg/L in the AxSYM™ method. When the 13 highest measurements (as measured using the s-ELISA method) were excluded, the bias among the remaining n=407 decreased slightly from 45.4 to 44.0 μg/L and concordance (95% CI) decreased from 0.62 (0.58, 0.66) to 0.53 (0.49, 0.57).   Among the 226 Congolese children, the s-ELISA method and the Elecsys® 2010 method showed poor concordance (Table 3) and poor agreement (Figure 1E). An overall bias of ~43 μg/L was detected, and concordance was poor, further indicating that the two methods did not produce similar results. Of note, n=17 measurements exceeded 250 μg/L in the s-ELISA and n=29 measurements exceeded 250 μg/L in the Elecsys® 2010 method. When the 17 highest measurements (as measured using the s-ELISA method) were excluded, the agreement and concordance improved. Among the n=209 children remaining in the dataset, the overall bias decreased from -42.6 to -11.5 μg/L and the concordance coefficient (95% CI) increased from 0.45 (0.40, 0.50) to 0.91 (0.89, 0.93). Concordance plots are presented in Figure 2.  200   Figure 1 Limits of agreement plots for A) n=420 Cambodian women using the s-ELISA and AxSYM™; B) a random subset of n=100 Cambodian women using the s-ELISA and AxSYM™; C) a random subset of n=100 Cambodian women using the s-ELISA and Centaur® XP; D) a random subset of n=100 Cambodian women using the AxSYM™ and Centaur® XP; and E) n=226 Congolese children using the s-ELISA and Elecsys® 2010.    201    Figure 2 Concordance plots. (A) n=420 Cambodian women using the s-ELISA and AxSYM™; (B) a random subset of n=100 Cambodian women using the s-ELISA and AxSYM™; (C) a random subset of n=100 Cambodian women using the s-ELISA and Centaur® XP; (D) a random subset of n=100 Cambodian women using the AxSYM™ and Centaur® XP; and (E) n=226 Congolese children using the s-ELISA and Elecsys® 2010.     202 Discussion and Conclusions The observed differences in ferritin concentration in our study are likely a reflection of the different ferritin isoforms, antibodies, and calibrators used in the immunoassays and across the different laboratories. However, despite these differences, no major differences in iron deficiency prevalence were observed across methods in our study, in which the samples of women and children had unexpectedly low iron deficiency prevalence. Further research is warranted to ascertain if our findings are replicable in populations with higher prevalence rates of iron deficiency.  There were n=6 measurements of ferritin concentration that exceeded 750 μg/L using the Elecsys® 2010 method among the Congolese children. No measurements exceeded 750 μg/L using the s-ELISA method either among the Congolese children or among the Cambodian women. We note that we did not pre-define nor exclude extreme outliers from our original dataset; however, we query whether or not these high measurements were perhaps extreme outliers. It seems unlikely that children in this age group would have such a high ferritin concentration, even if they had a condition such as hemochromatosis or thalassemia. Regardless, the exclusion of the highest 17 ferritin concentrations from the dataset (which also included the exclusion of those six high values >750 μg/L as measured by the Elecsys® 2010 method) did not substantially change the iron deficiency prevalence across groups. Overall, ferritin concentrations were relatively high among Cambodian women and Congolese children. The lack of differences in iron deficiency prevalence, despite the differences in ferritin concentrations, may be attributed to this. If ferritin concentrations were lower (closer to the 15 μg/L and 12 μg/L  203 cut-offs for women and children, respectively (28)), we suspect it may have resulted in more substantial differences in iron deficiency prevalence rates across groups.   Traceability varied among methods with none of our analyses directly traceable to the WHO 3rd IS for ferritin. The Abbott AxSYM™ and Roche Elecsys® 2010 analyzer were indirectly traceable to the WHO 1st IS, the Siemen ADVIA Centaur® XP was directly traceable to the WHO 2nd IS, and Erhardt et al.’s s-ELISA was not traceable to any of the WHO IS. This issue poses challenges to the interpretation and comparability of results between assays and likely contributes to the differences observed among methods.   The s-ELISA has a unique advantage over the other automated analyzers, as ferritin, CRP, AGP, sTfR and RBP concentrations can concurrently be measured in one small sample of serum or plasma. This is useful as ferritin is increased and RBP is decreased in the presence of inflammation (22). As such, the WHO advises the correction of ferritin and RBP concentrations for levels of inflammation using CRP and/or AGP biomarkers (7,28). Hence, this method allows for comprehensive assessment and interpretation of population-level vitamin A and iron status. Further, this method requires only a small amount (30 μL) of serum or plasma. This allows the convenient and economical option to collect capillary rather than venous blood, which is useful for blood collection in remote locations and in populations where venous blood collection is difficult (e.g., young children). These considerations have contributed to the increasing popularity of this method in low-resource settings and in large study populations at risk for iron deficiency.    204 The quality controls for the s-ELISA are selected with the objective to accurately diagnose iron deficiency with higher sensitivity at lower ferritin concentrations (personal communication, Juergen Erhardt). Thus, caution is warranted in the interpretation of ferritin measurements exceeding ~250 μg/L using this method. In fact, the lab recommends exclusion of ferritin values above ~250 μg/L due to the decreased accuracy of measurement at these higher values (personal communication, Juergen Erhardt). We concur that analytical accuracy is most important at the lower range in consideration of accurate iron deficiency diagnosis, however, it is also important at the higher range to measure and monitor conditions of iron overload in individuals with genetic hemoglobin disorders, or other conditions associated with high ferritin concentrations (e.g. inflammation, infection, and/or hemochromatosis). For example, the co-inherited β-thalassemia/hemoglobin E homozygous genotype is prevalent in areas of Southeast Asia and individuals with this genotype have severe anemia and often require regular blood transfusions, resulting in iron overload (29). More research is needed to evaluate the accuracy of ferritin measurement at higher ranges using these methods (as a biomarker of iron overload).  We acknowledge some limitations. First, none of the four methods in our study were directly traceable to the WHO 3rd IS. Method comparison studies ideally assay a set of specimens against a candidate reference material. As such, we cannot determine which of the four methods in our study was most accurate. We note that the aim of our study was to simply compare median ferritin concentrations and the prevalence of iron deficiency among the four immunoassays in populations thought to be at risk. Second, not all four of the methods were used to measure ferritin concentrations among all groups. This would have provided a more thorough comparison. However, these automated assays are expensive and we did not have the funds to  205 conduct analyses on all available samples. We recognize that reference ranges for different laboratories may be established based on the types of samples typically received and analyzed. For example, the laboratory in Germany (s-ELISA) receives the majority of samples from children and women living in developing countries under very different geographical and socioeconomic circumstances than individuals whose samples are analyzed by the laboratory in Vancouver, Canada, which receives the majority of samples from the Canadian population. Lastly, we used Thurnham et al. (22) correction factors to adjust for levels of inflammation in our populations studied. These correction factors were demonstrated to be accurate for use in the same Cambodian population of n=420 women in this study (which also used the s-ELISA to measure the inflammation biomarkers used for adjustments) (30), however, the robustness of these correction factors for use in other ferritin assays has not yet been established.   We conclude that the accurate diagnosis and treatment of iron deficiency is important, especially for pregnant women and young children (3–5). However, overestimations of iron deficiency prevalence can result in unnecessary iron supplementation, which also can be harmful to some individuals (31). Therefore, more work on the global standardization of ferritin assays and reference materials is warranted. It is well known that traceability is an essential component of laboratory medicine and for methods to be comparable they must be evaluated against an established reference material (32). Mandating the use of the WHO 3rd IS for calibration and the use of standard controls may help to reduce the variability in assay results across laboratories and methods. The matrix of the calibrators can influence the results; hence, it may be justified to establish global controls for each matrix given the proposed type of sample for analysis. The WHO, in collaboration with the Centre for Disease Control, has recently commissioned two  206 systematic reviews to investigate the diagnostic accuracy of ferritin as an indicator of not just deficiency but also of iron overload, and to compare laboratory assays for the determination of ferritin concentration (12). We anticipate that the findings from the WHO work will play a critical role in the harmonization of ferritin assays to accurately assess iron deficiency and iron overload, in order to guide effective nutrition and public health policy globally.  References 1.  de Benoist B, McLean E, Egli I, Cogswell M. Worldwide prevalence of anaemia 1993-2005: WHO global database on anaemia. Geneva: World Health Organization; 2008.  2.  Kassebaum NJ, Jasrasaria R, Naghavi M, Wulf SK, Johns N, Lozano R, et al. A systematic analysis of global anemia burden from 1990 to 2010. Blood. 2014;123(5):615–24.  3.  Friedman A, Chen Z, Ford P, Johnson C, Lopez A, Shander A, et al. Iron deficiency anemia in women across the life span. J Women’s Health. 2012;21(12):1282–9.  4.  Allen LH. Anemia and iron deficiency: effects on pregnancy outcome. Am J Clin Nutr. 2000;71:1280–4.  5.  Lozoff B. Iron deficiency and child development. Food Nutr Bull. 2007;28(Suppl 4):S560–71.  6.  Georgieff MK. Long-term brain and behavioral consequences of early iron deficiency. Nutr Rev. 2011;69(Suppl 1):43–8.  7.  World Health Organization. Assessing the iron status of populations. 2nd ed. Geneva: World Health Organization; 2007.  8.  Pasricha S, Flecknoe-Brown SC, Allen KJ, Gibson PR, McMahon LP, Olynyk JK, et al.  207 Diagnosis and management of iron deficiency anaemia: a clinical update. Med J Aust. 2010;193(9):525–32.  9.  Erhardt JG, Estes JE, Pfeiffer CM, Biesalski HK, Craft NE. Combined measurement of ferritin, soluble transferrin receptor, retinol binding protein, and C-reactive protein by an inexpensive, sensitive, and simple sandwich enzyme-linked immunosorbent assay technique. J Nutr. 2004;134:3127–32.  10.  Cox KL, Devanarayan V, Kriauciunas A, Manetta J, Montrose C, Sittampalam S. Immunoassay methods. In: Sittampalam GS, Coussens NP, Nelson H, Arkin M, Auld D, Austin C, et al., editors. Assay guidance manual. Betheseda (MD): Eli Lilly & Company and the National Center for Advancing Translational Sciences; 2004.  11.  Ferraro S, Mozzi R, Panteghini M. Revaluating serum ferritin as a marker of body iron stores in the traceability era. Clin Chem Lab Med. 2012;50(11):1911–6.  12.  Garcia-Casal MN, Peña-Rosas JP, Pasricha S-R. Rethinking ferritin cutoffs for iron deficiency and overload. Lancet Haematol. 2014;1(3):e92–4.  13.  Proposed international standard of human ferritin for the serum ferritin assay. International Committee for Standardization in Haematology (Expert Panel on Iron). Br J Haematol. 1985 Sep;61(1):61–3.  14.  WHO expert committee on biological standardization. World Health Organ Tech Rep Ser. 1994 Jan;840:1–218.  15.  World Health Organization. WHO international standard (ferritin, human recombinant; code 94/572): instructions for use. Potters Bar, Hertfordshire; 2008.  16.  Thorpe SJ, Walker D, Arosio P, Heath A, Cook JD, Worwood M. International collaborative study to evaluate a recombinant L ferritin preparation as an international  208 standard. Clin Chem. 1997;43(9):1582–7.  17.  Karakochuk CD, Whitfield KC, Barr SI, Lamers Y, Devlin AM, Vercauteren SM, et al. Genetic hemoglobin disorders rather than iron deficiency are a major predictor of hemoglobin concentration in women of reproductive age in rural Prey Veng, Cambodia. J Nutr. 2015;145(1):134–42.  18.  Harvey-Leeson S, Karakochuk C, Hawes M, Tugirimana P, Bahizire E, Akilimali P, et al. Anemia and micronutrient status of women of childbearing age and children 6–59 months in the Democratic Republic of the Congo. Nutrients. 2016;8(2):98.  19.  Smith J, Osikowicz C, Tayi R, Walker D, Martin R, Vaught J, et al. Abbott AxSYM random and continuous access immunoassay system for improved workflow in the clinical laboratory. Clin Chem. 1993;39(10):2063–9.  20.  Siemens ADVIA Centaur XP immunoassay systems package insert (version 10629858). Muenchen, Germany; 2011.  21.  Roche Elecsys 2010 immunoassay analyzer. [Internet]. Basel, Switzerland; Available from: https://usdiagnostics.roche.com/products/04491785160/PARAM368/overlay.html 22.  Thurnham DI, McCabe LD, Haldar S, Wieringa FT, Northrop-Clewes CA, McCabe GP. Adjusting plasma ferritin concentrations to remove the effects of subclinical inflammation in the assessment of iron deficiency: a meta-analysis. Am J Clin Nutr. 2010;92:546–55.  23.  Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;8:307–10.  24.  Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res. 1999;8(2):135–60.  25.  Hanneman SK. Design, analysis and interpretation of method-comparison studies. AACN  209 Adv Crit Care. 2008;19(2):223–34.  26.  Clinical and Laboratory Standards Institute (CLSI). Measurement Procedure Comparison and Bias Estimation Using Patient Samples; Approved Guideline (EP09-A3). CLSI. 2013.  27.  Lin LI. A concordance correlation coefficient to evaluate reproducibility. Biometrics. 1989;45(1):255–68.  28.  World Health Organization. Serum ferritin concentrations for the assessment of iron status and iron deficiency in populations. Geneva: World Health Organization; 2011.  29.  Zimmermann MB, Fucharoen S, Winichagoon P, Sirankapracha P, Zeder C, Gowachirapant S, et al. Iron metabolism in heterozygotes for hemoglobin E (HbE), alpha-thalassemia 1, or beta-thalassemia and in compound heterozygotes for HbE/beta-thalassemia. Am J Clin Nutr. 2008;88:1026–31.  30.  Karakochuk CD, Whitfield KC, Rappaport AI, Barr SI, Vercauteren SM, Mclean J, et al. The homozygous hemoglobin EE genotype and chronic inflammation are associated with high serum ferritin and soluble transferrin receptor concentrations among women in rural Cambodia. J Nutr. 2015;145:2765–73.  31.  Gutteridge J. Iron and free radicals. In: Hallberg L, Asp N, editors. Iron nutrition in health and disease. London: John Libbey and Co. Ltd.; 1996. p. 239–46.  32.  Vesper HW, Thienpont LM. Traceability in laboratory medicine. Clin Chem. 2009;55(6):1067–75.    210 Appendix C: Iron in Groundwater in Prey Veng  Acknowledgement: A version of this appendix has been published. Karakochuk CD, Murphy HM, Whitfield KC, Barr SI, Vercauteren SM, Talukder A, Porter K, Kroeun H, Eath M, McLean J, Green TJ. Elevated levels of iron in groundwater in Prey Veng province in Cambodia: A possible factor contributing to high iron stores in women. Journal of Water and Health 2015; 13(2): 575-86.  Summary Iron is a natural element found in food, water and soil and is essential for human health. The aim was to determine the levels of iron and 25 other metals and trace elements in groundwater from 22 households in Prey Veng, Cambodia. Water analyses were conducted using inductively coupled plasma mass spectrometry and optical emission spectrometry. Compared to the WHO Guidelines for Drinking Water Quality, aluminum, iron and manganese exceeded maximum levels (in 5%, 73% and 41% of samples, respectively). Compared to the Cambodian Drinking Water Quality Standards, iron and manganese exceeded maximum levels (in 59% and 36% of samples, respectively). We found no evidence of arsenic contamination. Guidelines for iron were established primarily for aesthetic reasons (i.e., taste), whereas other metals and elements have adverse effects associated with toxicity. Iron in groundwater ranged from 134 to 5200 μg/L (mean ~1422 μg/L). Based on a daily consumption of 3 L groundwater, this equates to ~0.4 to 15.6 mg iron (mean ~4.3 mg/day), which may be contributing to high iron stores and the low prevalence of iron deficiency anemia in Prey Veng women. Elevated levels of manganese in groundwater are a concern and warrant further investigation.  211 Introduction Iron is an abundant element and is found naturally occurring in soil, water and food. It is an essential mineral in the human body and is required for oxidative energy metabolism, red blood cell production and oxygen transport, as well as other important functions (1). The majority of iron that is absorbed and utilized in the body is obtained from dietary food sources. However, in many parts of the developing world, diets are low in iron and cannot provide adequate amounts of iron to meet daily requirements (2).  In Cambodia, the diet is thought to be iron-poor and low in iron-rich animal food sources (3). Impaired absorption and loss of iron can also result from infection and disease (1,4) such as dengue, malaria, hookworm and parasites, which are prevalent in some areas of rural Cambodia. Therefore, it is often assumed that iron deficiency is prevalent among women in Cambodia. However, in a recent study in Prey Veng province (5) we detected very low prevalence of iron deficiency in women (n=420). Iron stores, based on serum ferritin concentrations, were unexpectedly high (mean unadjusted ferritin ~103 μg/L). Serum ferritin is an acute phase protein, which means that in the presence of inflammation, levels are elevated. Therefore, it is recommended that serum ferritin concentrations be adjusted for levels of inflammation using inflammation biomarkers (6). However, even after adjustment for inflammation, serum ferritin concentrations remained high in the women (mean inflammation-adjusted ferritin ~93 μg/L). The prevalence of iron deficiency anemia among women was very low (~1%) based on common biomarkers (hemoglobin <120 g/L and ferritin <15 μg/L, for non-pregnant women of reproductive age). Based on self-reported data, the women in the study were not taking iron supplements, using iron cooking pots, nor consuming iron fortified food products. With the  212 published evidence of elevated iron levels in groundwater in Cambodia (7), it is possible that groundwater could be a contributing factor to the high iron stores in the women in Prey Veng.    In addition to iron, elevated levels of arsenic and manganese have also previously been detected in groundwater in some areas in Cambodia (7–10). The Southeastern provinces of Kandal, Prey Veng, and Kampong Cham are the most severely affected areas for arsenic contamination (levels in groundwater exceeding 50 μg/L), with the worst affected areas along the Mekong, Bassac and Tonle Sap riverbanks (11,12). This evidence of metal and trace element contamination in groundwater is a public health concern. Inorganic arsenic, the form most often found in drinking water, is classified as a human carcinogen (13). Chronic exposure to inorganic arsenic through contaminated water or food has been associated with an increased risk of morbidity and mortality (14). Elevated levels of manganese (400 μg/L) in drinking water have been associated with adverse neurological effects in children (15,16). Elevated levels of iron in drinking water (300 μg/L), however, have not previously been associated with significant health risks. The maximum levels of iron in drinking water are set by the WHO because of undesirable aesthetic properties that are associated with high iron content in water (i.e., poor taste, colour and smell) (17).  One successful approach to reduce the iron content and to improve microbial water quality at the household level has been the implementation of point-of-use water filtering systems. In Cambodia, the BioSand filter is a commonly used household filter. It is a slow sand filter that was originally designed to reduce microbial contamination in drinking water. The BioSand filter, when installed correctly, can achieve reductions of up to ~99% removal of both bacteria and iron  213 (18). However, the availability and regular use of filtering systems in Prey Veng province is not well known.  The primary aim of this study was to determine the level of iron in groundwater in a subset of households including women participating in the larger study in Prey Veng (5).  We hypothesized that groundwater could be a contributing factor to the high iron stores of these women. The secondary aims were to determine the levels of 25 other metals and trace elements in groundwater and compare those levels to recommended minimum and maximum levels based on both chemical and nutritional quality, and to investigate the availability of and compliance with BioSand filtering systems in households.  Methods Water samples were collected from groundwater wells in households participating in a larger trial in Prey Veng province (5). Ethical approval for the study was granted by the Clinical Research Ethics Board at the University of British Columbia (Canada) and the National Ethics Committee for Health Research (Cambodia).   Water samples were collected from 22 households across ten villages in four districts (Ba Phnum, Kamchay Mear, Me Sang, Svay Antor) of rural Prey Veng province in May 2014 at the end of the dry season. The ten villages were randomly selected from a list of all villages in the four districts in Prey Veng province. From each of the ten villages, between two to three water samples were collected from randomly selected households within the catchment area of the village. If a household did not have a groundwater well, or if no family members were home at  214 the time of visit, the next nearest household was visited. A handheld global position satellite device (Garmin eTrex 10) was used to record coordinates of each groundwater well in order to confirm the geographical location of each well. Water consumption and treatment practices of the household were recorded at the time of the water sample collection, including: availability of a water filtering system, compliance and regular use of the filtering system if it was available, and descriptions of the aesthetic properties of the water (i.e., taste, colour, and smell).    A total of 22 water samples were taken directly from the well at each household. Before collection of the water samples, the wells were pumped for 90 seconds to initiate continuous water flow and avoid contamination of samples due to stagnant water in the pipe. An estimated quantity of 125 mL water was collected in an acid-washed polyethylene plastic container provided by the laboratory conducting the analyses (Agat Laboratories, Burnaby, Canada). In addition, filtered water samples were taken from a BioSand filter at three random households that reported use of a filter. The water samples from these filters were collected immediately and directly from the spout of the BioSand filter. For these households, untreated water samples were also collected directly from the well.   Temperature and pH were recorded for all 25 water samples. Temperature of the water sample was recorded to the nearest 1 o C using a standard glass thermometer (Thermo Scientific ERTCO precision model). The pH of the water sample was recorded to the nearest 0.1 value using pH indicator test strips (Merck). The pH strip was immersed in the pumped water for 15 seconds, removed and left to stand for 30 seconds. It was then compared to the pH colour strip  215 by two different field researchers (for confirmatory purposes) to estimate the pH level reading in the water sample.   After temperature and pH were recorded, concentrated nitric acid (2 mL) was immediately added to preserve the concentrations of metals (19). Water samples were placed in an icebox and kept at 4 o C during transportation. They were then stored in a 4 o C refrigerator at the Helen Keller International office in Phnom Penh. Within seven days of collection, the samples were shipped to Canada and analyzed for metal and trace element content by Agat Laboratories (Burnaby, Canada). Inductively coupled plasma mass spectrometry and optical emission spectrometry were used to determine metal and trace element content of the water as per standardized methods (20). A total of 26 metals and trace elements were analyzed: aluminum, antimony, arsenic, barium, beryllium, boron, cadmium, calcium, chromium, cobalt, copper, iron, lead, lithium, magnesium, manganese, molybdenum, nickel, selenium, silver, sodium, thallium, titanium, uranium, vanadium and zinc.   Detected levels of the 26 metals and trace elements were compared to the Cambodian Ministry of Industry, Mines, and Energy (CMIME) and WHO water quality guidelines to evaluate the chemical quality of the groundwater. The maximum standard levels are set based on expected risk of adverse health outcomes or undesirable aesthetic properties (i.e., taste, colour, and smell) associated with high levels in drinking water.   Of the total 26 metals and trace elements analyzed, 13 also have dietary reference intakes (DRIs) established for women 19-30 years of age including at least one of the following: adequate  216 intakes (AI), estimated average requirements (EAR), recommended dietary allowances (RDA) or tolerable upper intake levels (UL). The DRIs are nutrient reference values developed by the Institute of Medicine of the National Academies (21). They are established based on scientific evidence and provide reference values for minimum requirements and maximum limits for safety, and are specified on the basis of age, gender and life stage. The EAR is the average daily nutrient intake level estimated to meet the requirements of ~50% of a healthy population. The RDA is an average daily dietary intake level that is sufficient to meet the nutrient requirements of ~97.5% or more of a healthy population and is calculated using the EAR value. The AI is an estimated value that suggests a recommended intake when there is insufficient scientific evidence to establish an EAR. The UL is the highest level of daily nutrient intake that is likely to pose no risk of adverse health effects to ~99% in the general population (21).   For the 13 metals or trace elements with an established EAR, AI, RDA or UL, estimated levels of daily intakes (based on consumption of ~3 L groundwater daily) were compared to the available DRIs. We speculate that the average Cambodian woman would consume ~3 L of groundwater daily. Unpublished data on the dietary intakes of women in Prey Veng (using 24 hour dietary recall methods and survey questionnaires) suggest that the mean water intake of women in Prey Veng is ~7.9 cups per day (~2 L water) solely as drinking water (personal communication, Vashti Verbowski, The University of British Columbia). However, this does not include water from tea, soups or other cooking sources using groundwater. In Cambodia, rice is a staple food and is consumed on a daily basis in most households. It requires water for cooking and rice is usually prepared with one part rice to four parts water. Most of the water will be evaporated during the cooking process, however the iron in the water does not freely evaporate  217 and remains in the rice. Therefore, the groundwater used to cook rice is also included in our estimation of daily water consumption of women. Therefore, we speculate the average Cambodian woman would consume ~3 L groundwater daily, but recognize that this value is an estimation and would vary among individuals. Our results should be interpreted with these considerations in mind.  IBM SPSS software v.22 (Armonk, NY, USA) was used to conduct statistical analyses. Median values, mean values and standard deviations were calculated and used to describe the data.  Results A total of 25 water samples were collected from 22 households in ten villages in Prey Veng province. Data on the consumption, water treatment and filtering practices were obtained from women caregivers at the household level (self-report) at the time of water sample collection (Table 1). Data on whether or not a household had a filtering system was missing for four households (18%) and data on whether or not a household reported red sediment in the water after one day was missing for six households (27%).      218 Table 1 Water consumption and treatment characteristics of 22 households in ten villages in Prey Veng1,2 1 Values are n (%).  2 Reported by women caregivers in households. 3 Data were missing for four households (filtering system available) and six households (red sediment in water).  The groundwater samples (unfiltered) were analyzed for a total of 26 metals and trace elements. Table 2 presents the units of measure, reporting detection limit (RDL), minimum value observed, maximum value observed, mean value and standard deviation, median value of each metal and trace element, and comparisons to the CMIME and WHO water quality guidelines. Quality assurance tests confirmed that control samples were within acceptable limits for all metals and trace elements analyzed. The values of the three filtered groundwater samples were not included in this summary table.  n % Water treatment for drinking source in household (n=22)         No treatment (drinks directly from well) 17 77%       Boils well water  2 9%       Filters well water 3 14% Household has a filtering system available (n=18)3          No  12 67%       Yes  6 33% Household reports red sediment in water after one day (n=16)3         No 9 56%       Yes 7 44%  219 Table 2 Summary of metals and trace elements analyzed for chemical quality in 22 samples of groundwater from household wells in Prey Veng province and comparisons to CMIME and WHO water quality guidelines1,2 Metals and trace elements  Unit of measure RDL Min. value Max. value Mean  SD  Median value Global WHO guideline 2011 No. of samples exceeding WHO guideline Cambodia CMIME guideline 2004 No. of samples exceeding CMIME guideline Aluminum μg/L 5 5 122 25.9  35.0 5 100 1 - - Antimony μg/L 0.5 0.5 0.5 0.5  0.0 0.5 20 0 - - Arsenic μg/L 0.1 0.1 7.7 2.1  2.2 1.1 10 0 50 0 Barium μg/L 0.5 24.9 349.0 111.0  82.8 85.6 700 0 1000 0 Beryllium μg/L 0.05 0.05 0.58 0.13  0.16 0.05 - - - - Boron μg/L 5 5 18 10.2  4.9 8 500 0 - - Cadmium μg/L 0.01 0.01 0.09 0.03  0.02 0.015 3 0 10 0 Calcium mg/L 0.05 1.2 50.5 12.0  11.9 9.4 - - 200 0 Chromium μg/L 0.5 0.5 1.5 0.6  0.2  0.5 50 0 50 0 Cobalt μg/L 0.05 0.05 7.38 2.22  2.30 1.16 - - - - Copper μg/L 0.5 0.5 4.4 0.85  0.92 0.5 2000 0 1500 0 Iron μg/L 10 134 5200 1422  1296 1150 300 16 1000 13 Lead μg/L 0.05 0.05 6.06 0.51  1.29 0.13 10 0 10 0 Lithium μg/L 0.5 8.1 26.4 17.2  4.0 16.6 - - - - Magnesium mg/L 0.05 0.6 17.8 6.1  4.8 5.5 - - 150 0 Manganese μg/L 1 47 827 393.9  236.1 352.5 400 9 500 8 Molybdenum μg/L 0.1 0.1 0.6 0.2  0.1 0.1 70 0 - - Nickel μg/L 0.5 0.1 9.5 2.5  3.0 0.8 20 0 - - Selenium μg/L 0.5 0.5 0.5 0.5  0.0 0.5 10 0 10 0 Silver μg/L 0.02 0.02 0.02 0.02  0.0 0.02 - - - - Sodium mg/L 0.1 2.46 67.0 28.4  4.5 23.8 200 0 - - Thallium μg/L 0.02 0.02 0.26 0.04  0.06 0.02 - - - - Titanium μg/L 1 2 7 3.6  1.5 3 - - - - Uranium μg/L 0.01 0.01 0.28 0.05  0.06 0.015 - - - - Vanadium μg/L 1 1 2 1.1  0.2 1 - - - -  220 Metals and trace elements  Unit of measure RDL Min. value Max. value Mean  SD  Median value Global WHO guideline 2011 No. of samples exceeding WHO guideline Cambodia CMIME guideline 2004 No. of samples exceeding CMIME guideline Zinc μg/L 5 5 26 10.7  6.7 8 4000 0 1500 0 Hardness ugCaCO3/L 0.1 5.8 166.0 55.4  46.9 47.1 - - 300 0 Temperature degrees oC - 30 34 31.4  1.0 31 - - - - pH - - 4 7 5.85  1.0 6.0 - - - - 1 o C, Celsius; CMIME, Cambodia Ministry of Industry, Mines and Energy; RDL, reporting detection limit; SD, standard deviation; WHO, World Health Organization. 2 Values of the three filtered water samples (filtered by BioSand filters) were not included in this summary table as the process of BioSand filtering removes the metals and trace elements.    221 When compared to the WHO guidelines for drinking water quality, a total of three metals and trace elements exceeded the recommended maximum levels in one or more samples of unfiltered well water: aluminum (1 of 22 samples, or 5%), iron (16 of 22 samples, or 73%), and manganese (9 of 22 samples, or 41%). When compared to the Cambodia drinking water quality standards, a total of two metals and trace elements had eight or more samples that exceeded the recommended maximum levels: iron (13 of 22 samples, or 59%) and manganese (8 of 22 samples, or 36%).   Of the total 26 metals and trace elements analyzed, 13 are recognized as essential for human health and have DRIs established for women 19-30 years including at least one of the following: AI, EAR, DRA or UL. For these 13 metals and trace elements, estimations of daily intake (based on consumption of 3 L groundwater daily) and comparisons to the DRIs are presented in Table 3. Of the 13 metals and trace elements, 10 have recommended intakes (either an RDA or an AI) for women 19-30 years: calcium, chromium, copper, iron, magnesium, manganese, molybdenum, selenium, sodium and zinc. None of the groundwater samples had levels that met the recommended intakes (RDA or AI) for these elements, although an intake of 3 L would provide about a quarter of RDA for iron and two-thirds of the AI for manganese.  For other minerals, 3 L of groundwater would provide <8% of the intake recommendations. Of the 13 elements, all but chromium have a UL established for women 19-30 years (21). On average, an intake of 3 L would provide ~10% of the UL for iron and manganese, however much less (<2%) for the remaining metals and trace elements. None of the individual groundwater samples came close to levels that exceeded the UL for these 12 elements.   222 Table 3 Estimated daily intakes of 13 metals and trace elements (based consumption of 3 L groundwater) and comparisons  to established dietary reference intakes (DRIs)1,2 Metals and trace elements Unit of measure Estimated daily intake  (3 L water)3 Daily Dietary Reference Intakes (DRIs)  EAR  RDA AI UL Boron μg/day 30.6 - - - 20000 Calcium mg/day 36 800 1000 - 2500 Chromium μg/day 1.8 - - 25 ND Copper μg/day 2.55 700 900 - 10000 Iron μg/day 4266 8100 18000 - 45000 Magnesium mg/day 18.3 225 310 - 3504 Manganese μg/day 1181.7 - - 1800 11000 Molybdenum μg/day 0.6 34 45 - 2000 Nickel μg/day 7.5 - - - 1000 Selenium μg/day 1.5 45 55 - 400 Sodium mg/day 85.2 - - 1500 2300 Vanadium μg/day 3.3 - - - 1800 Zinc μg/day 32.1 6800 8000 - 40000 1 AI, adequate intakes; DRI, dietary reference intakes; EAR, estimated average requirement; L, liter; ND, not determined; RDA,  recommended dietary allowance; UL, tolerable upper intake level. 2 DRI values based on females 19-30 years are reported for comparisons. 3 Estimated daily intakes of minerals and trace elements were calculated using mean values and based on consumption of  3 L groundwater daily (drinking water, tea, soups, or other cooking sources using groundwater, namely rice). 4 The UL for magnesium applies only to magnesium consumed as a supplement or fortificant.  223 In a subset of three households, water samples were taken directly from the groundwater well and also from the BioSand filter (after filtering). Levels of iron (Table 4) were compared before and after filtering with the BioSand filter. Water filtered with the BioSand filter had significantly less iron content: 98-99% of iron was removed through the filtering process.   Table 4 Iron levels in groundwater before and after household BioSand filtering in three households1 Household Iron level from groundwater well (μg/L) Iron level after BioSand filtering (μg/L) Change in iron level 1  1150 24 98% decrease 2  1020 10 99% decrease 3  5200 10 >99% decrease 1 Iron levels from groundwater wells all exceed the recommended maximum levels, however after BioSand filtering all iron levels are decreased to below the recommended maximum levels.   Discussion and Conclusions We detected elevated levels of iron in groundwater from wells in rural Prey Veng province in Cambodia. The groundwater iron levels ranged from 134 to 5200 μg/L across households, with a mean value of 1422 μg/L and a median value of 1150 μg/L. Iron was elevated in 73% and 59% of samples based on recommended maximum levels of WHO and CMIME water quality guidelines, respectively. These maximum levels are recommended based on undesirable aesthetic properties that are associated with high iron content in water (i.e., poor taste, colour and smell). However, iron is an essential mineral in the body and a daily intake of 18 mg/day (RDA) is recommended for women 19-30 years (21) as iron is required for required for oxidative energy metabolism, red blood cell production and oxygen transport, as well as other important functions (1,22). Therefore, it was the aim of our study to determine levels of iron in groundwater and  224 assess if the water source could be contributing to the high iron stores observed in women in Prey Veng. We speculated that the average Cambodian woman would consume ~3 L of groundwater daily: including drinking water, tea, soup, or other cooking sources using groundwater. This would equate to ~0.4 mg to 15.6 mg of iron consumed daily, with a mean value of 4.3 mg iron and a median value of 3.5 mg iron.  This is a large contribution of iron to daily dietary intakes (~24% of the RDA or ~53% of the EAR for women 19-30 years), assuming that the iron in the groundwater is bioavailable.   Bioavailability of iron is suspected to be high given the form of iron in groundwater. The groundwater aquifer environment is reducing, which means the majority of iron is in the ferrous form (Fe2+) which is more dissolvable and bioavailable than the ferric form (Fe3+) (23,24). In addition, a low pH was detected in our water samples, indicating an environment. An acidic environment increases iron bioavailability and absorption in the gastrointestinal tract (25). It is likely that some of the iron is oxidized to Fe3+ during the pumping of the well water, and during storage if the water is not consumed immediately. This theory can be substantiated by the fact that some women would see red precipitate (Fe3+) in their water storage containers after a day of storage. Although Fe3+ isn’t as bioavailable as Fe2+ (23), once consumed, Fe3+ can be reduced to Fe2+ in the body with the help of stomach acid, the contents of the intestine and the cell membrane enzyme ferric-reductase (26).  Worwood et al. (27) demonstrated that iron bioavailability was high in naturally occurring mineral water (~300,000 μg iron/L) in England. Absorption studies in adults (n=13) using water labelled with ferrous sulphate indicated absorption rates of ~23% among individuals who  225 consumed 10 mL of the mineral water on an empty stomach. Furthermore, Worwood et al. demonstrated higher absorption rates (~40%) among individuals with low iron stores (ferritin <10 μg/L) compared to absorption rates of ~10% among individuals with high iron stores (ferritin >200 μg/L). As expected, the rates of iron absorption increase when iron body stores are deficient. Although the iron content of the mineral water in the England study (~300,000 μg/L) was substantially higher than the groundwater in our study (mean ~1422 μg/L), the groundwater in our study was very acidic, which increases the solubility and bioavailability of the iron. We suspect the iron in our study would be comparatively bioavailable.  Researchers in other parts of Asia have observed that increased groundwater iron is positively and significantly associated with increased iron stores in women. In rural Bangladesh, the median daily iron intake from groundwater was estimated at ~41 mg/day (24). Merrill et al. (28) then found that increased groundwater iron content was significantly correlated with plasma ferritin (r=0.36) and total body iron (r=0.35) in women in Bangladesh. Although median levels of iron in groundwater were much lower in our study (~3-4 mg/day) as compared to the Bangladesh study (~41 mg/day), the iron could still be accumulating and contributing to iron stores in women in our study, but at a slower rate than observed in the Bangladesh study. We did not attempt to correlate groundwater iron content with iron status of individual women in our study as the number of unfiltered water samples collected was too small (n=22) to detect statistically significant associations.  As noted previously, the recommended maximum levels of iron in drinking water are based on aesthetic concerns (17). Iron content in groundwater has not previously caused concern in terms  226 of potential health risks. However, there is a potential risk of iron overload in populations with genetic hemoglobin disorders (29). In Cambodia, these disorders are common and affect ~50% of the population (30). Some forms of genetic hemoglobin disorders can result in accumulated iron stores in the body (high serum ferritin concentrations) (5,31,32). Some forms of severe thalassemias can also result in decreased erythropoiesis which further enhances iron absorption in the gastrointestinal tract and thereby increases iron stores even more (33). These severely affected individuals often require life-long blood transfusions and other serious medical treatment (29). However, many of these individuals go undiagnosed and/or untreated in rural Cambodia. In these affected individuals, large quantities of iron from untreated and unfiltered groundwater sources may potentially increase the risk of iron overload and associated adverse effects.   Manganese levels in groundwater: The other metal that was elevated in the groundwater samples and is of potential concern was manganese. These findings are consistent with the findings of other researchers in Cambodia (7). Elevated manganese in drinking water is a concern as it has been associated with neurological impairment in children (15,16,34). A recent systematic review and meta-analysis (including 17 studies on manganese toxicity) published evidence that elevated manganese (detected in hair) was associated with decreased IQ levels in children 6-13 years. The authors also reported that manganese exposure was associated with attention deficit disorder with hyperactivity in children (35).    227 However, manganese, similar to iron, is also an essential nutrient in the human body. A daily intake of 1.8 mg/day (AI) is recommended for women 19-30 years (21) as manganese is required for optimal bone health, energy and protein metabolism, and regulation of cell metabolism. Although none of the groundwater samples had levels of manganese that met the AI, some samples were relatively close suggesting that groundwater provides a substantial contribution to daily recommended intakes of manganese. Of particular interest, none of the samples were close to reaching the UL, not even the samples that exceeded the CMIME and WHO water quality standards.   Bouchard et al. (16) speculated that manganese from groundwater was significantly more bioavailable than manganese from food sources based on correlations between manganese in hair samples and manganese in water. However, some researchers have highlighted several limitations of the Bouchard study (36) indicating that these findings would ideally be substantiated with studies that quantify the bioavailability of manganese from both water and foods sources. Hence, more research is warranted in this area before we can draw conclusions that excess manganese presents a greater risk of toxicity when consumed from water rather than food sources. Further research is also warranted to investigate levels of manganese in groundwater in other geographical regions and the consequences of elevated manganese levels in drinking water in populations of differing age and geographical location.   Other metals and trace element levels in groundwater: Aluminum was elevated (100 μg/L) in one sample. However, the majority of samples (18 of the 22) were well below 50 μg/L, suggesting that aluminum in groundwater is not a major concern.  228 Some researchers have speculated that elevated aluminum levels in drinking water are a risk factor for the development of Alzheimer’s Disease. However, there is insufficient evidence to prove an association. Therefore, it has been concluded that elevated aluminum in drinking water does not currently pose any significant health risk to humans (37).    None of the water samples contained arsenic levels that exceeded WHO recommended maximum limits (>10 μg/L) despite previous research indicating arsenic in groundwater as a major public health problem in Cambodia (7,38). Sthiannopkao et al. (9) reported arsenic contamination in well water in Prey Veng and neighbouring Kandal province, where 15 of 28 samples (54%) exceeded the WHO maximum limits (>10 μg/L). It has been speculated that although high iron and manganese concentrations are often found in areas of high arsenic concentration (further to the reducing environment of aquifers), statistical correlations between arsenic, iron and manganese are of only moderate strength (7). We did not detect high arsenic concentrations in our groundwater samples, despite the aforementioned studies in Cambodia that did. We conclude that the relationship between iron, arsenic and other chemical parameters is complicated. There are many factors which contribute to the determination of arsenic in groundwater samples that we neither investigated nor discussed in this manuscript. Further to arsenic, none of the other remaining 22 metals and trace elements tested in our samples were elevated beyond the maximum limits according to either WHO or CMIME guidelines: antimony, barium, beryllium, boron, cadmium, chromium, cobalt, copper, lead, lithium, magnesium, molybdenum, nickel, selenium, silver, sodium, thallium, titanium, uranium, vanadium, and zinc.    229 Other than manganese and iron, none of the other 11 metals and trace elements that have established DRIs provided a substantial contribution to recommended dietary intakes or indicated a risk of excess intake in the groundwater samples.  For simpler interpretation (Table 3) the levels of the 13 metals and trace elements were compared to DRIs for women 19-30 years. However, women in our study were between the ages of 18-45 years. Therefore, it is important to note that for some metals and trace elements, the DRIs are slightly higher or lower for women outside of the 19-30 year age category (21). In brief, women 14-18 years have a slightly higher EAR for calcium (1100 mg/day), magnesium (300 mg/day) and zinc (7.3 mg/day), a slightly lower EAR for copper (685 ug/day), iron (7.9 mg/day) and molybdenum (34 ug/day), and a slightly lower AI for chromium (24 ug/day) and manganese (1.6 mg/day). Women 31-50 years have a slightly higher EAR for magnesium (265 mg/day).  BioSand Filters: As expected, the BioSand filter was successful in removing iron from the water (98-99% iron removal) in the three households in our study. Murphy et al. (39) previously reported ~99% iron removal from Cambodian groundwater using BioSand filters. Other studies have found that these filters are effective at reducing levels of bacteria (18,40). However, in one evaluation study, it was reported that the BioSand filters do not consistently remove bacteria to levels considered acceptable by the WHO under field conditions (39). Although 33% of the households reported having a water filtering system available, only 14% of households (n=3) reported using any type of filter before consuming water from groundwater wells. This rate of compliance (42%) is lower  230 than observed in another cross sectional study of 336 households across five provinces in Cambodia evaluating compliance and use of BioSand filters (~88% compliance) (40). The low compliance rate of BioSand filters in our study highlights a potential problem and warrants further investigation.  Strengths of this study include that it is the first study to our knowledge to evaluate both the chemical and nutritional quality of groundwater in Cambodia. Limitations of this study include that it was conducted in small sample of 22 households and findings cannot be extrapolated to other geographical regions. We did not measure the oxidation-reduction potential in the ground water samples collected, which could have provided more information about the form and bioavailability of the iron in the samples. We did not investigate microbiological contamination of the water which is another important factor related to water quality. Microbiological contamination of drinking water can cause an increased risk of diarrhea or other co-morbidities, and thereby increase the risk of mortality in infants and children (41). Furthermore, the aim of our study was to investigate the levels of iron in groundwater. We did not attempt to detect an association between groundwater iron content in the household and biochemical iron stores (serum ferritin concentration) in women in the study. Therefore, we can only speculate that higher levels of iron in groundwater could contribute to iron stores in women, based on similar research conducted in Bangladesh (28).   We conclude that groundwater from wells in our study in the province of Prey Veng, Cambodia contained elevated levels of iron and manganese, but not arsenic. Iron in the well water from some households was considerably elevated (up to 5.2 mg/L). It is theorized that iron found in  231 the households’ drinking water could be contributing up to 15.6 mg iron/day to some women’s diets. Consequently, it is likely that the iron was contributing to the high serum ferritin concentrations (and therefore a lack of iron deficiency anemia) observed in a recent study (5) including women from the same households where groundwater samples were collected.   A daily intake of 3 L groundwater would provide about a quarter of the RDA for iron (and about a half of the EAR for iron) and two-thirds of the AI for manganese.  Other than manganese and iron, none of the other 11 metals and trace elements that have established DRIs provided a substantial contribution to recommended dietary intakes or indicated a risk of excess intake in the groundwater samples.   Further research is needed to characterize the groundwater quality in other geographical regions of Cambodia, particularly the occurrence of iron and manganese. Studies linking the presence of elevated iron in groundwater with elevated iron stores in women are needed to substantiate the potential of groundwater as a contribution to dietary iron intakes. Investigation of the potential risk of iron overload from groundwater in those individuals with severe forms of genetic hemoglobin disorders cannot be ignored and warrants further research.   References 1.  Gibson RS. Principles of nutritional assessment. 2nd ed. New York: Oxford University Press; 2005.  2.  de Benoist B, McLean E, Egli I, Cogswell M. Worldwide prevalence of anaemia 1993-2005: WHO global database on anaemia. Geneva: World Health Organization; 2008.   232 3.  Cambodia Council for Agriculture and Rural Development (CARD). Strategic framework for food security and nutrition in Cambodia 2008-2012. Phnom Penh, Cambodia; 2012.  4.  Weiss G, Goodnough LT. Anemia of chronic disease. N Engl J Med. 2005;352:1011–23.  5.  Karakochuk CD, Whitfield KC, Barr SI, Lamers Y, Devlin AM, Vercauteren SM, et al. Genetic hemoglobin disorders rather than iron deficiency are a major predictor of hemoglobin concentration in women of reproductive age in rural Prey Veng, Cambodia. J Nutr. 2015;145(1):134–42.  6.  Thurnham DI, McCabe LD, Haldar S, Wieringa FT, Northrop-Clewes CA, McCabe GP. Adjusting plasma ferritin concentrations to remove the effects of subclinical inflammation in the assessment of iron deficiency: a meta-analysis. Am J Clin Nutr. 2010;92:546–55.  7.  Feldman PR, Rosenboom J-W, Saray M, Navuth P, Samnang C, Iddings S. Assessment of the chemical quality of drinking water in Cambodia. J Water Health. 2007;5:101–16.  8.  Buschmann J, Berg M, Stengel C, Sampson ML. Arsenic and manganese contamination of drinking water resources in Cambodia: Coincidence of risk areas with low relief topography. Environ Sci Technol. 2007;41(7):2146–52.  9.  Sthiannopkao S, Kim KW, Sotham S, Choup S. Arsenic and manganese in tube well waters of Prey Veng and Kandal Provinces, Cambodia. Appl Geochemistry. 2008;23(5):1086–93.  10.  Luu T, Sthiannopkao S, Kyoung-Woong K. Arsenic and other trace elements contamination in groundwater and a risk assessment study for the residents in the Kandal province of Cambodia. Environ Int. Elsevier Ltd; 2009;35(3):455–60.  11.  A study of options for safe water access in arsenic affected communities in Cambodia. Phnom Penh; 2012.   233 12.  Berg M, Stengel C, Trang PTK, Hung Viet P, Sampson ML, Leng M, et al. Magnitude of arsenic pollution in the Mekong and Red River Deltas - Cambodia and Vietnam. Sci Total Environ. 2007;372(2-3):413–25.  13.  Toxicological profile for aresenic. Atlanta, GA: US Public Health Service, Department of Health and Human Services; 2007.  14.  Berg M, Tran HC, Nguyen TC, Pham H V, Schertenleib R, Giger W. Arsenic contamination of groundwater and drinking water in Vietnam: a human health threat. Environ Sci Technol. 2001;35(13):2621–6.  15.  Wasserman G, Liu X, Parvez F, Ahsan H, Levy D, Factor-Litvak P, et al. Water manganese exposure and children’s intellectual function in Araihazar, Bangladesh. Environ Health Perspect. 2006;114(1):124–9.  16.  Bouchard MF, Sauve S, Barbeau B, Legrand M, Brodeur M-E, Bouffard T, et al. Intellectual impairment in school-age children exposed to manganese from drinking water. Environ Health Perspect. 2011;119(1):138–43.  17.  Guidelines for drinking water quality. 4th ed. Geneva: World Health Organization; 2011.  18.  Stauber CE, Printy ER, McCarty FA, Liang KR, Sobsey MD. Cluster randomized controlled trial of the plastic BioSand water filter in Cambodia. Environ Sci Technol. 2012;46(2):722–8.  19.  BC environmental laboratory manual. Surrey, BC: Environmental laboratory technical advisory committee, water and air monitoring and reporting section; 2009.  20.  Determination of trace elements in waters and wastes by inductively coupled plasma - mass spectrometry (method 200.8). Washington, DC: US Environmental Protection Agency; 1994.   234 21.  Otten JJ, Hellwig JP, Meyers LD, editors. Dietary reference intakes: the essential guide to nutrient requirements. Washington, DC: The National Academies Press; 2006.  22.  Ganz T. Molecular control of iron transport. J Am Soc Nephrol. 2007;18(2):394–400.  23.  Hurrell RF. Bioavailability of iron. Eur J Clin Nutr. 1997;51(1):S4–8.  24.  Merrill RD, Labrique AB, Shamim AA, Schulze K, Christian P, Merrill RK, et al. Elevated and variable groundwater iron in rural northwestern Bangladesh. J Water Health. 2010;8(4):818–25.  25.  Hallberg L. Bioavailability of dietary iron in man. Annu Rev Nutr. 1981;1:123–47.  26.  Aster J. The hematopoietic and lymphoid system. In: Kumar V, Abbas A, Fausto N, RN M, editors. Robbins basic pathology 8th ed. Philadelphia: Sauders Elsevier; 2007.  27.  Worwood M, Evans WD, Villis RJ, Burnett AK. Iron absorption from a natural mineral water (Spatone Iron-Plus). Clin Lab Haematol. 1996;18(1):23–7.  28.  Merrill RD, Shamim AA, Ali H, Jahan N, Labrique AB, Schulze K, et al. Iron status of women is associated with the iron concentrations of potable groundwater in rural Bangladesh. J Nutr. 2011;141:944–9.  29.  Bain BJ. Haemoglobinopathy diagnosis. 2nd ed. Oxford: Blackwell Publishing Ltd; 2006.  30.  Carnley BP, Prior JF, Gilbert A, Lim E, Devenish R, Sing H, et al. The prevalence and molecular basis of hemoglobinopathies in Cambodia. Hemoglobin. 2006;30(4):463–70.  31.  Thurlow RA, Winichagoon P, Green T, Wasantwisut E, Pongcharoen T, Bailey KB, et al. Only a small proportion of anemia in northeast Thai schoolchildren is associated with iron deficiency. Am J Clin Nutr. 2005;82:380–7.  32.  George J, Yiannakis M, Main B, Devenish R, Anderson C, An US, et al. Genetic hemoglobin disorders, infection, and deficiencies of iron and vitamin A determine anemia  235 in young Cambodian children. J Nutr. 2012;142:781–7.  33.  Rees DC, Williams TN, Maitland K, Clegg JB, Weatherall DJ. Alpha thalassaemia is associated with increased soluble transferrin receptor levels. Br J Haematol. 1998;103(2):365–9.  34.  Ljung K, Vahter M. Time to re-evaluate the guideline value for manganese in drinking water? Environ Health Perspect. 2007;115(11):1533–8.  35.  Rodriguez-Barranco M, Lacasana M, Aguilar-Garduno C, Alguacil J, Gil F, Gonzalez-Alzaga B, et al. Association of arsenic, cadmium and manganese exposure with neurodevelopment and behavioural disorders in children: a systematic review and meta-analysis. Sci Total Environ. Elsevier B.V.; 2013;454-455:562–77.  36.  Chen H, Copes R. Manganese in drinking water and intellectual impairment in school-age children. Environ Health Perspect. 2011;119(6):240–1.  37.  Aluminum. International programme on chemical safety. Geneva: World Health Organization; 1997.  38.  Gault AG, Rowland HAL, Charnock JM, Wogelius RA, Gomez-Morilla I, Vong S, et al. Arsenic in hair and nails of individuals exposed to arsenic-rich groundwaters in Kandal province, Cambodia. Sci Total Environ. 2008;393(1):168–76.  39.  Murphy H, McBean E, Farahbakhsh K. A critical evaluation of two point-of-use water treatment technologies: can they provide water that meets WHO drinking water guidelines? J Water Health. 2010;8(4):611–30.  40.  Sobsey MD, Stauber CE, Casanova LM, Brown JM, Elliott MA. Point of use household drinking water filtration: a practical, effective solution for providing sustained access to safe drinking water in the developing world. Environ Sci Technol. 2008;42(12):4261–7.   236 41.  Lanata CF, Fischer-Walker CL, Olascoaga AC, Torres CX, Aryee MJ, Black RE. Global causes of diarrheal disease mortality in children <5 years of age: a systematic review. PLoS One. 2013;8(9):e72788.   

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0340471/manifest

Comment

Related Items