A response to a Change Our Mind Contest entry on iron fortification programs in India

By GiveWell @ 2023-10-25T21:34 (+91)

This is Andrew Martin, a Senior Research Associate at GiveWell, writing from GiveWell's EA Forum account.

We at GiveWell would like to thank Akash Kulgod for his deep engagement with our analysis of iron fortification programs in India in his Change Our Mind contest entry. We appreciate his work, which led us to investigate some questions about iron fortification programs in India that we had not previously considered.

Below, we discuss Akash's arguments, the additional research we've conducted in response to the issues he raised, and our current views.

Summary

Background

GiveWell has made three grants to Fortify Health for its work on iron fortification of wheat flour in India: $0.3 million in 2018, $1 million in 2019, and $8.2 million in 2021.[1] We may also consider making additional grants to iron fortification programs in India and other countries going forward.

In our intervention report, we discuss the evidence of effectiveness for iron fortification programs. One of the primary benefits of iron fortification in our cost-effectiveness model is a reduction in morbidity due to anemia, a condition in which hemoglobin concentrations in blood are lower than normal, leading to fatigue, dizziness, weakness, and other symptoms.[2] Our intervention report also discusses the evidence for short-term cognitive benefits of iron fortification among children and adults with anemia.[3]

In his post, Akash raises a set of criticisms of GiveWell's model for iron fortification in India:

We decided to conduct additional research and write a public response based on the importance of the issues Akash raised. The first three issues could have an impact on our cost-effectiveness estimates, and the final issue could also be relevant for GiveWell's future grantmaking decisions.

Akash's post references the 2019 version of GiveWell's cost-effectiveness analysis of Fortify Health—we also published a more recent cost-effectiveness analysis for our 2021 grant to Fortify Health. For this post, we are also making a new CEA version public here. Our preliminary discussion of the updates we've made between the 2021 CEA and the current version is here.

Anemia prevalence and disease burden in India

Background

The case for GiveWell's grants to Fortify Health relies on estimates of the prevalence and disease burden of anemia in India. Our cost-effectiveness model uses estimates of anemia prevalence and Years Lived with Disability (YLDs),[4] a measure of disease burden, from the Institute for Health Metrics and Evaluation (IMHE)'s Global Burden of Disease (GBD) project.[5] GBD estimates the prevalence of anemia at 44% for 0 to 19 year-olds and at 42% across all ages in India as of 2019.[6]

The World Health Organization (WHO) sets commonly used standards (including by GBD) for diagnosing anemia, based on concentrations of hemoglobin in blood samples.[7] As Akash notes in his post, WHO hemoglobin cutoffs for defining anemia have been largely unchanged since 1968, and were based on a set of studies among primarily white populations in North America and Europe (though cutoffs distinguishing mild, moderate, and severe anemia were updated more recently, and the 1968 cutoffs were later validated in surveys of multi-ethnic populations in the United States).[8]

Our understanding is that WHO's original process for defining the anemia cutoff was to identify the fifth percentile of hemoglobin concentrations in otherwise healthy individuals in the available studies.[9] In his post, Akash points to Sachdev et al. 2021, which uses data from India's 2019 Comprehensive National Nutrition Survey (CNNS) to propose new hemoglobin concentration cutoffs for 0-19 year-olds in India using a similar methodology to the original WHO process. Sachdev et al. 2021 excludes CNNS data from individuals with some identified health issues and calculates the fifth percentile of hemoglobin concentrations among the remaining sample.[10]

Sachdev et al. 2021's process leads to lower hemoglobin concentration cutoffs for anemia in 0-19 year-olds in India than the WHO standards. Sachdev et al. conclude that if their cutoffs were used, estimates of anemia prevalence among the full CNNS sample of 0 to 19 year-olds in India would drop by 19.2 percentage points.[11] If we applied this finding to the IHME estimate for anemia prevalence among 0 to 19 year-olds in India, it would drop from 44% to 25%.[12] Akash also includes some rough estimates of expected declines in anemia prevalence among adult populations in his post, extrapolating from the findings of Sachdev et al. 2021.[13]

This is important because a large drop in anemia disease burden estimates in India would lead to substantial reductions in our cost-effectiveness estimates for iron fortification programs. We estimate three main benefits of iron fortification programs in our CEA: anemia morbidity averted (46% of total benefits), cognitive benefits for children (5%), and cognitive benefits for adults (49%).[14] Our estimates of anemia morbidity averted are sensitive to baseline anemia YLDs—if baseline YLDs are cut in half (as Akash estimates as the impact of applying the Sachdev et al. 2021 hemoglobin concentration cutoffs),[15] our estimate of the benefits of iron fortification for averting anemia morbidity would be cut in half as well.[16] Updating hemoglobin cutoffs for anemia could also affect our estimates of cognitive benefits, since we only apply cognitive benefits to the subset of the population that is anemic at baseline.[17]

Our research process and updated views

To investigate the concerns raised by Akash about anemia prevalence in India, we:

A summary of our updated views:

Should hemoglobin concentration thresholds for defining anemia be lowered for India?

The anonymous iron and anemia expert we spoke to expressed skepticism about lowering hemoglobin concentration thresholds for defining anemia based on the findings of Sachdev et al. 2021:

How much would lowering hemoglobin concentration thresholds affect cost-effectiveness?

Akash estimates the impact of applying the Sachdev et al. 2021 hemoglobin concentration cutoffs would be to cut baseline anemia YLDs in India in half, which appears to be roughly proportional to the decline in anemia prevalence with the Sachdev et al. cutoffs.[25] Cutting anemia YLDs in half leads to a 54% reduction in our cost-effectiveness estimate.[26]

We are uncertain about the assumption that reductions in anemia YLDs would be roughly proportional to reductions in anemia prevalence because Sachdev et al. 2021 does not propose new thresholds for defining moderate and severe anemia. In this spreadsheet, we compare the prevalence of mild, moderate, and severe anemia in India with the number of YLDs attributed to mild, moderate, and severe anemia, according to GBD data.[27] Across all ages, mild anemia accounts for nearly 50% of anemia cases, but only accounts for 5% of YLDs attributable to anemia. This difference is due to the disability weight for mild anemia being very low (0.004).[28] Moderate anemia has a disability weight (0.052) that is 13 times higher than the weight for mild anemia, and severe anemia's disability weight (0.149) is ~37 times higher.[29]

WHO's hemoglobin levels to diagnose anemia at sea level (grams per deciliter)[30]

Population Non-anemia Mild anemia Moderate anemia Severe anemia
Children 6-59 months of age 11.0 or higher 10.0 to 10.9 7.0 to 9.9 Lower than 7.0
Children 5-11 years of age 11.5 or higher 11.0 to 11.4 8.0 to 10.9 Lower than 8.0
Children 12-14 years of age 12.0 or higher 11.0 to 11.9 8.0 to 10.9 Lower than 8.0
Non-pregnant women (15 years of age and above) 12.0 or higher 11.0 to 11.9 8.0 to 10.9 Lower than 8.0
Pregnant women 11.0 or higher 10.0 to 10.9 7.0 to 9.9 Lower than 7.0
Men (15 years of age and above) 13.0 or higher 11.0 to 12.9 8.0 to 10.9 Lower than 8.0

We have not been able to create our own estimate of changes in anemia YLDs under Sachdev et al.'s proposal because Sachdev et al. 2021 does not include updated thresholds for defining moderate and severe anemia. If thresholds for moderate and severe anemia remained relatively unchanged, the impact of adopting Sachdev et al. 2021's hemoglobin concentration thresholds for defining anemia might be a large reduction in anemia prevalence with a much smaller reduction in anemia YLDs, since the majority of cases re-defined as non-anemic would be mild cases. We expect that this would have a fairly small impact on our cost-effectiveness estimate, since mild anemia only accounts for 5% of anemia YLDs. It is possible, however, that implementing a proposal like Sachdev et al.'s would also involve re-defining cutoffs for moderate and severe anemia. We would be open to investigating this issue further if we encounter proposals for updating hemoglobin concentration thresholds for moderate and severe anemia.

Impact on cognitive benefits

Updating anemia prevalence estimates could also have an impact on our estimates of the cognitive benefits of iron fortification programs since, based on our evidence review, we only apply those benefits to individuals with anemia at baseline.[31] We include reductions in cognitive benefits in our estimate of the impact of Akash's criticisms on our cost-effectiveness model.[32] We are uncertain, however, whether this adjustment would be appropriate even if we were to adopt Akash's recommendation, since it would mean that we'd be using different definitions of anemia for populations reached by iron fortification programs today and individuals who participated in the research we rely on for estimating cognitive impacts.[33]

Implications for our cost-effectiveness model

We think the arguments laid out by the anonymous iron and anemia expert we consulted with seem reasonable, but we remain open to changing our views based on new evidence. We also would plan to revisit this issue if WHO updates its guidelines—we expect that this would be a strong signal that expert views on this issue have shifted.

The proportion of anemia cases in India caused by iron deficiency

Background

Iron deficiency is considered to be the most common cause of anemia globally.[34] Other causes of anemia include other micronutrient deficiencies, diseases including malaria and parasitic infections, inflammation, and genetic disorders.[35] Underlying causes of anemia are an important issue for our cost-effectiveness analysis, since we expect that iron fortification programs would only be able to potentially address anemia caused by iron deficiency. (Fortify Health's fortification programs also include vitamin B12, which we expect also has some impact on anemia, though much smaller than the impact of iron.)[36]

In his post, Akash points to research that suggests that areas with high levels of infectious disease may have a lower than average proportion of anemia caused by inadequate iron intake (and a higher proportion caused by inflammation as a response to infections).[37] Akash also notes that the CNNS data (discussed in the section above) has also been analyzed to estimate the proportion of 0 to 19 year-olds in India with iron deficiency—Kulkarni et al. 2021 estimates the prevalence of iron deficiency in India at ~32% for 1 to 4 year-olds, ~30% for adolescent girls, with lower rates (11%-15%) for adolescent boys and 5 to 9 year-olds.[38]

Akash doesn't suggest a specific adjustment to GiveWell's CEA for iron fortification programs in India, but notes that he believes that GBD may be overestimating the contribution of iron deficiency to anemia.[39]

Our research process

Our current views

We do not currently think that we should apply a downward adjustment to our CEA to account for the proportion of anemia cases caused by iron deficiency being lower than commonly estimated in India, for the following reasons:

Remaining uncertainties

Additional potential negative impacts of iron fortification

Akash notes that there may be additional potential negative impacts of iron fortification beyond those that we've already identified, namely an increased risk of type 2 diabetes and non-alcoholic fatty liver disease.

Below are brief summaries of the two papers Akash linked on these topics:

Based on our review, we think that the current evidence for iron fortification causing increased risk of type 2 diabetes or non-alcoholic fatty liver disease appears fairly weak. The main reason is that these non-experimental studies aren't able to establish that high serum ferritin concentrations cause increased type 2 diabetes risk or liver fat accumulation—it may also be the case that the causation is reversed, or that a third factor influences both serum ferritin concentrations and disease risk. The anonymous iron expert we spoke with noted that he believes that the most likely explanation for the association between high serum ferritin and metabolic disease or liver disease is "reverse causality."[51]

We agree with Akash that it's worth taking risks of iron fortification seriously, and we may continue investigating potential negative impacts of the intervention as we consider additional grants going forward. We are also open to revising our views on the impact of iron fortification on type 2 diabetes and non-alcoholic fatty liver disease based on new evidence.

Ethical considerations and reputational risk

In his post, Akash raises ethical concerns and notes potential reputational risk involved in funding iron fortification programs in India. Quoting from his post:

We agree that these are important issues, and we thank Akash for raising them. Below are our current responses to these concerns:

Continued research on external validity

We expect to continue to research external validity issues when considering future grants to iron fortification programs in India—some potential areas for research are listed below:

Notes

  1. We have also granted funding to Evidence Action for iron and folic acid supplementation programs in India: a $0.3 million grant in 2018, a second $3.4 million grant in 2018, and a $0.8 million grant for an impact evaluation in 2019. ↩︎

    • The cost-effectiveness model linked from our intervention report is here.
    • "Anaemia is a condition in which the number of red blood cells or the haemoglobin concentration within them is lower than normal. Haemoglobin is needed to carry oxygen and if you have too few or abnormal red blood cells, or not enough haemoglobin, there will be a decreased capacity of the blood to carry oxygen to the body’s tissues. This results in symptoms such as fatigue, weakness, dizziness and shortness of breath, among others." WHO, "Anaemia"
    ↩︎

  3. See the "Effect on cognitive outcomes" section of our intervention report. ↩︎

  4. "YLD is an abbreviation for years lived with disability, which can also be described as years lived in less than ideal health. This includes conditions such as influenza, which may last for only a few days, or epilepsy, which can last a lifetime. It is measured by taking the prevalence of the condition multiplied by the disability weight for that condition. Disability weights reflect the severity of different conditions and are developed through surveys of the general public." IHME, "About GBD" ↩︎

  5. See the most recent version of our CEA here (not previously published). As of June 2023, we have not yet updated our intervention report for iron fortification to account for our CEA updates. The published version of our CEA that Akash referenced in his post is here. ↩︎

  6. For comparison, global anemia prevalence rates are 28% for 0 to 19 year-olds and 23% for all ages. IHME, GBD Results tool, Rate of anemia from all causes, 2019 ↩︎

    • For example, for children 6-59 months of age, a hemoglobin concentration of 110 grams per liter or higher is defined as non-anemic, 100-109 g/l is defined as mild anemia, 70-99 g/l is defined as moderate anemia, and lower than 70 g/l is defined as severe anemia. WHO, Haemoglobin concentrations for the diagnosis of anemia and assessment of severity, 2011, p. 3, Table 1.
    • "Anaemia is defined by decreased blood concentration of haemoglobin. We estimated unique, continuous distributions of elevation-adjusted haemoglobin concentrations (g/L), anaemia prevalence, and years lived with disability (YLDs) by severity and 37 underlying causes of anaemia annually from 1990 to 2021 for 204 countries and territories, 21 GBD regions, male and female sexes, and 25 age groups (0–6 days, 7–27 days, 1–5 months, 6–11 months, 12–23 months, 2–4 years, 5–94 years in five-year age bins, and ≥95 years). Anaemia severity levels (mild, moderate, and severe) were defined using specific haemoglobin concentration thresholds that vary by age, sex, and pregnancy status (table)...Published WHO thresholds were used for males and females aged 6 months and older; thresholds for those younger than 6 months were imputed as described in the appendix (p 6)." GBD 2021 Anemia Collaborators 2023, p. 715.
    ↩︎
    • "The anaemia cut-offs presented in Table 1 were published in 1968 by a WHO study group on nutritional anaemias, while the cut-offs defining mild, moderate and severe anaemia were first presented in the 1989 guide Preventing and controlling anaemia through primary health care and then modified for pregnant women, nonpregnant women, and children less than five years of age in The management of nutrition in major emergencies. The overall anaemia cut-offs have been unchanged since 1968, with the exception that the original age group of children 5-14 years of age was split, and a cut-off of 5 g/l lower was applied to children 5-11 years of age to reflect findings among non-iron deficient children in the USA. Although these cut-offs were first published in the late 1960s, they have been included in numerous subsequent WHO publications and were additionally validated by findings among participants in the Second National Health and Nutrition Examination Survey (NHANES II) who were unlikely to have iron deficiency based on a number of additional biochemical tests." WHO, Haemoglobin concentrations for the diagnosis of anemia and assessment of severity, 2011, p. 3.
    • "Appropriate guidelines for measurement of haemoglobin and definition of anaemia are crucial for both clinical and public health medicine, but require consideration of the range of complexities across different populations. Haemoglobin thresholds to define anaemia were first proposed by WHO in 1959. Current thresholds recommended by WHO for men, women, young children, and pregnant women (table) were first proposed in 1968 after technical meetings of a group comprising clinical and public health experts working with data from five studies of predominantly white populations in Europe and North America (appendix). Data from other countries, races, and ages (ie, infants, young children, adolescents, and elderly people) were not available to the panel." Pasricha et al. 2018, p. e60.
    • "The current WHO cutoff levels were derived from mainly White adults but were validated in a multiethnic sample from a single country (US)." Addo et al. 2021, p. 8.
    • Addo et al. 2021 cites Looker et al. 1997 for the claim that the WHO cutoffs were validated against a multiethnic sample in the United States. We have not reviewed the original study.
    ↩︎

  9. "More than 95% of normal individuals are believed to show haemoglobin levels higher than the values given, which are appropriate for all geographic areas; however, the values must be modified for persons who reside at higher altitudes." WHO, Nutritional Anemias, 1968, pp. 9-10. ↩︎

  10. "For this population-based study, we constructed age-specific and sex-specific haemoglobin percentiles from values reported for a defined healthy population in the CNNS, which used rigorous quality control measures during sample collection and in the laboratory analyses. To obtain a healthy population, we excluded participants with iron, folate, vitamin B12, and retinol deficiencies; inflammation; variant haemoglobins (haemoglobin A2 and haemoglobin S); and history of smoking. We considered age-specific and sex-specific 5th percentiles of haemoglobin derived for this healthy population as the study cutoff to define anaemia. We compared these with existing WHO cutoffs to assess significant differences between them at each year of age and sex for quantifying the prevalence of anaemia in the entire CNNS sample." Sachdev et al. 2021, p. e822. ↩︎

  11. "Between Feb 24, 2016, and Oct 26, 2018, the CNNS survey collected blood samples from 49 486 individuals. 41 210 participants had a haemoglobin value, 8087 of whom were included in our study and comprised the primary analytical sample. Compared with existing WHO cutoffs, the study cutoffs for haemoglobin were lower at all ages, usually by 1–2 g/dL, but more so in children of both sexes aged 1–2 years and in girls aged 10 years or older. Aanemia prevalence with the study cutoffs was 19·2 percentage points lower than with WHO cutoffs in the entire CNNS sample with valid haemoglobin values across all ages and sexes (10·8% with study cutoffs vs 30·0% with WHO cutoffs)." Sachdev et al. 2021, p. e822. ↩︎

  12. 44% - 19.2% = ~25%. See here for GBD anemia prevalence estimates for India. ↩︎

  13. See this section of Akash's post ↩︎

  14. See the most recent version of our CEA here (not previously published). The published version Akash referenced in his post is here. See this section of the CEA for the proportional breakdown of the three main benefits. ↩︎

  15. "The YLDs from anemia is halved (conservative simplification)." From this section of Akash's post. ↩︎

  16. See "Anemia morbidity averted" section here. If "Adjusted annual anemia YLDs per person, all ages" is reduced by 50%, "Annual anemia YLDs per person (all ages) averted by iron fortification" is reduced by 50% as well. For details on our views on the evidence for iron fortification, see our intervention report. ↩︎

  17. See the "Cognitive benefits in children" and "Cognitive benefits in adults" sections of our CEA here. ↩︎

  18. "It is uncertain why Indians should have lower Hb thresholds than essentially any other population: inherited red cell disorders (globin gene mutations for example) are common in India, but need to be diagnosed for clinical reasons. The authors argue that the reason relates to having a muscle-thin but high adipose body composition; this hypothesized physiologic variance should currently be considered a cause of anemia until it has been confirmed the effect on Hb is not pathological." Report for GiveWell by an anonymous iron and anemia expert, p. 5. ↩︎

  19. "Hemoglobin concentrations are extremely sensitive to inflammation, and if a person becomes anemic, it can take time to recover from the anemia. Individuals were excluded if they reported a chronic illness or a known acute infection (/fever) were excluded; however, it is still highly likely that some of the individuals included still had had an illness or infection which, even if it had resolved, still caused reduced hemoglobin and hence anemia, which may not have recovered by the time the blood testing had occurred. Indeed, the authors mention this in the Discussion: 'Fourth, even though exposure to infections in the preceding 2 weeks was an exclusion criteria, lingering post-infectious altered erythropoiesis could not be ruled out.' This is a critical point in a setting where there is a high prevalence of infections. Malaria was not assessed in this study." Report for GiveWell from an anonymous iron and anemia expert, p. 5. ↩︎

  20. "The survey was done in a way to be representative of the Indian population; the Indian middle class is not the middle class of the West; some definitions (eg Pew) are that this is a group earning $10-20/day, i.e. having risen out of poverty… the numbers are not well defined. The population living a lifestyle akin to the West is much smaller. The majority of participants in this cross-sectional survey designed to assess diseases of poverty are not well off." Report for GiveWell from an anonymous iron expert, p. 4. ↩︎

  21. "It is crucial to identify healthy populations for reference studies since the goal is to identify the 5th centile – so if the population below the 5th centile is not all ‘healthy’ then these thresholds will be reduced." Report for GiveWell from an anonymous iron expert, p. 5. ↩︎

  22. Addo et al. 2021:

    "To define the healthy population, persons with iron deficiency (ferritin <12 ng/mL for children or <15 ng/mL for women), vitamin A deficiency (retinol-binding protein or retinol <20.1 μg/dL), inflammation (C-reactive protein >0.5 mg/dL or α-1-acid glycoprotein >1 g/L), or known malaria were excluded. Survey-specific, pooled Hb fifth percentile cutoffs were estimated. Among individuals with Hb and sTfR data, Hb-for-sTfR curve analysis was conducted to identify Hb inflection points that reflect tissue iron deficiency and increased erythropoiesis induced by anemia." p. 1.

    "In this cross-sectional study, data were collected and evaluated from 30 household, population-based nutrition surveys of preschool children aged 6 to 59 months and nonpregnant women aged 15 to 49 years during 2005 to 2016 across 25 countries. Data analysis was performed from March 2020 to April 2021." p. 1.

    "To define the healthy population, persons with iron deficiency (ferritin <12 ng/mL for children or <15 ng/mL for women), vitamin A deficiency (retinol-binding protein or retinol <20.1 μg/dL), inflammation (C-reactive protein >0.5 mg/dL or α-1-acid glycoprotein >1 g/L), or known malaria were excluded. Survey-specific, pooled Hb fifth percentile cutoffs were estimated. Among individuals with Hb and sTfR data, Hb-for-sTfR curve analysis was conducted to identify Hb inflection points that reflect tissue iron deficiency and increased erythropoiesis induced by anemia." p. 1. ↩︎

    • Addo et al. 2021, Figure 1, p. 7.
    • "There was low intersurvey variance when analyzing individual-level Hb data of 39 325 apparently healthy individuals, but high interstudy heterogeneity from meta-analysis highlighting the limitation of meta-analyses to directly address this study objective." Addo et al. 2021, p. 9.
    ↩︎

  24. "The Addo analysis (Addo JAMA Network Open 2021) cited in the Sachdev paper demonstrates a wide heterogeneity in Hb thresholds in different countries where the authors used a simple post-hoc approach to exclude biochemical inflammation and iron deficiency; our analyses of this dataset suggests that inflammation is not being well excluded as the cutoff is negatively correlated with the prevalence of inflammation in the population...This highlights the need to ensure populations are ‘healthy’, which cannot be done without some clinical screening of populations." Report for GiveWell by an anonymous iron and anemia expert, p. 5. ↩︎

    • "The YLDs from anemia is halved (conservative simplification)." From this section of Akash's post.
    • Sachdev et al. report a decline of greater than 50% in anemia prevalence: "Anemia prevalence with the study cutoffs was 19·2 percentage points lower than with WHO cutoffs in the entire CNNS sample with valid haemoglobin values across all ages and sexes (10·8% with study cutoffs vs 30·0% with WHO cutoffs)." Sachdev et al. 2021, p. e822.
    ↩︎
    • See the "Summary tab" of our CEA for the percent change in overall cost effectiveness from this change. .
    • See the highlighted row of the "Anemia morbidity averted" section in the "CEA [AK Anemia prevalence and YLDs]" tab of our CEA.
    ↩︎

  27. For WHO's hemoglobin concentration cutoffs defining mild, moderate, and severe anemia, see WHO, Haemoglobin concentrations for the diagnosis of anemia and assessment of severity, 2011, p. 3, Table 1. ↩︎

  28. The disability weight for mild anemia is 0.004. IHME, Global Burden of Disease Study 2019 (GBD 2019) Disability Weights ↩︎

  29. IHME, Global Burden of Disease Study 2019 (GBD 2019) Disability Weights ↩︎

  30. Adapted from WHO, Haemoglobin concentrations for the diagnosis of anemia and assessment of severity, 2011, p. 3, Table 1. ↩︎

  31. See the "Cognitive benefits in children" and "Cognitive benefits in adults" sections of the "Iron fortification CEA [GW]" tab of our CEA. ↩︎

  32. See the highlighted rows of the "CEA [AK Anemia prevalence and YLDs]" tab of our CEA. ↩︎

  33. See our write-up on cognitive impacts of iron in children here and our write-up on adults here. ↩︎

    • "Anaemia may be caused by several factors: nutrient deficiencies through inadequate diets or inadequate absorption of nutrients, infections (e.g. malaria, parasitic infections, tuberculosis, HIV), inflammation, chronic diseases, gynaecological and obstetric conditions, and inherited red blood cell disorders. The most common nutritional cause of anaemia is iron deficiency, although deficiencies in folate, vitamins B12 and A are also important causes." WHO, "Anaemia"
    • IHME estimates that greater than 50% of anemia is caused by dietary iron deficiency globally. IHME, GBD 2019, "Anemia—Level 1 impairment," Figure 1.
    ↩︎

  35. "Anaemia may be caused by several factors: nutrient deficiencies through inadequate diets or inadequate absorption of nutrients, infections (e.g. malaria, parasitic infections, tuberculosis, HIV), inflammation, chronic diseases, gynaecological and obstetric conditions, and inherited red blood cell disorders. The most common nutritional cause of anaemia is iron deficiency, although deficiencies in folate, vitamins B12 and A are also important causes." WHO, "Anaemia" ↩︎

  36. We include a rough adjustment for the impact of vitamin B12 in the Fortify Health CEA here. ↩︎

  37. See this section of Akash's post. ↩︎

  38. "Results: ID prevalence was higher in 1- to 4-y-old children (31.9%; 95% CI: 31.0%, 32.8%) and adolescent girls (30.4%; 95% CI: 29.3%, 31.5%) but lower in adolescent boys and 5- to 9-y-old children (11%–15%)." Kulkarni et al. 2021, p. 2422. ↩︎

  39. "I'm unsure how to assess the impact of this section on GiveWell's cost-effectiveness model as I wasn't able to pin down the estimate for iron-deficiency's contribution to anemia in the model. Since the BRINDA method and other work were published from 2018 onwards, and the GBD's data sources are much older, about a decade prior on average, it seems likely that GBD overestimates ID's contribution to anemia. Like before, the CNNS data is restricted to age groups b/w 0-19 yrs, but intuitively should extend to adult age groups as well." From this section of Akash's post. ↩︎

  40. We believe that Kulkarni et al. 2021 is the paper Akash is quoting from here: "This is the first study from India providing estimates of ID prevalence in a representative sample of children and adolescents at the national and state levels using multiple inflammation-adjusted ID indicators. In preschool children and adolescent girls, ID [iron deficiency] based on SF [serum ferritin] adjusted for inflammation by the modified BRINDA method was a public health problem of 'moderate' proportions (∼30%–32%), whereas in 5- to 9-y-old children (15%) and adolescent boys (11%) it was a public health problem categorized as 'mild.'" ↩︎

  41. "The proportion of anemia associated with iron deficiency was lower in countries where anemia prevalence was >40%, especially in rural populations (14% for pre-school children; 16% for non-pregnant women of reproductive age), and in countries with very high inflammation exposure (20% for pre-school children; 25% for non-pregnant women of reproductive age). Despite large heterogeneity, our analyses suggest that the proportion of anemia associated with iron deficiency is lower than the previously assumed 50% in countries with low, medium, or high Human Development Index ranking." Petry et al. 2016, p. 692. ↩︎

  42. "These indices were constructed using factors known to contribute to inflammation. For children, the inflammation-exposure score was based on (a) prevalence of presumed and confirmed malaria cases; (b) schistosomiasis prevalence; and (c) an overall hygiene score based on the proportion of population using improved drinking water source and the proportion of the population using improved sanitation facilities (evenly weighted) as a proxy for the risk of enteric inflammation. For women, the different factors used to estimate country-specific inflammation-exposure were (a) prevalence of presumed and confirmed malaria cases;(b) HIV prevalence in adults; (c) obesity prevalence in female adults; (d) schistosomiasis prevalence; and (e) an overall hygiene score based on the proportion of population using improved drinking water source and the proportion of the population using improved sanitation facilities (evenly weighted) as a proxy for the risk of enteric inflammation." Petry et al. 2016 p. 693. ↩︎

  43. See the Inflammation-exposure index ratings by country in Petry et al. 2016, Supplementary Materials, "Inflammation exposure country categorization PSC and WRA": ↩︎

  44. See here. ↩︎

  45. "Oral iron supplementation is very effective in a trial context in improving anemia (eg twice weekly iron reduced anemia 16.8% vs 35.3%, RR = 0.47) in Indian Adolescent girls; this is not a result inconsistent with appropriate thresholds (as anemia would not have responded if the girls weren’t anemic). Likewise, adolescents with anemia given iron-supplement-containing bars responded dramatically to the iron [anemia prevalence at 90 d was lower for intervention (29.2%) than for control participants (98.6%) (OR: 0.007; 95% CI: 0.001, 0.04).] (Mehta AJCN 2015). These changes would indicate anemia in India is indeed responsive to therapy. " Written response to GiveWell from an anonymous iron expert (unpublished). ↩︎

    • Mehta et al. 2017, p. 746.
    • Additional methodological details: "Design: The Let’s be Well Red study was a 90-d, pair-matched, cluster-randomized controlled trial. A total of 361 nonpregnant women (age 18–35 y) were recruited from 10 sites within Mumbai and Navi Mumbai, India. All participants received anemia education and a complete blood count (CBC). Random assignment of anemic participants to intervention and control arms occurred within 5 matched site-pairs. Intervention participants received 1 iron-supplement bar (containing 14 mg Fe)/d for 90 d, whereas control subjects received nothing. CBC tests were given at days 15, 45, and 90. Primary outcomes were 90-d changes from baseline in hemoglobin concentrations and hematocrit percentages. Linear mixed models and generalized estimating equations were used to model continuous and binary outcomes, respectively." Mehta et al. 2017, p. 746.
    ↩︎
    • "In a community-based cluster randomized controlled trial, we randomly assigned clusters of anemic women and adolescent girls to either “directly observed home-based daily iron therapy” (DOHBIT; n = 524 in 16 villages) or unsupervised self-treatment at home (n = 535 in 16 villages) for a period of 90 days. Those in the DOHBIT group, when compared with those in the unsupervised self-treatment group, had significantly lower relative risk (RR) of anemia (16.8% vs 35.3%, RR = 0.47 [95% confidence interval (CI) = 0.33-0.65]; P < .0001), higher hemoglobin (Hb) rise of ≥2 g/dL (70.2% vs 42.2%, RR = 1.56 [95% CI = 1.31-1.87]; P <.0001), and nonsignificant trend for lower side effects (3.5% vs 6.7%, RR = 0.49 [95% CI = 0.22-1.08; P < .08) on intention-to-treat analyses. On linear mixed model analysis, the subjects in the intervention group demonstrated higher mean Hb levels (13.01 vs 12.32 g/dL; P < .0001) and higher adherence to iron therapy (93% vs 60%; P < .0001). DOHBIT is effective in lowering the prevalence of anemia in rural women and adolescent girls." Bharti et al. 2015, p. 1333.
    • "All anemic rural women (young unmarried, pregnant, lactating, reproductive age group, as well as menopausal women) and adolescent girls aged 13 years and older were eligible for trial." Bharti et al. 2015, p. 1334.
    ↩︎

  48. Nicholas Kassebaum, Adjunct Associate Professor, Global Health, Institute for Health Metrics and Evaluation, University of Washington, conversation with GiveWell, November 15, 2022. (unpublished) ↩︎

  49. "A total of 12 case–control and cohort studies were analyzed. Of the 12 studies, 11 described the correlation between serum ferritin levels and type 2 diabetes. The median and high serum ferritin concentrations were significantly associated with the risks of type 2 diabetes (odds ratio [OR] 1.20, 95% confidence interval [CI] 1.08–1.33 and OR 1.43, 95% CI 1.29–1.59, respectively). However, the low concentration was not correlated with the risk of type 2 diabetes (OR 0.99, 95% CI 0.89–1.11). No significant association was observed between serum soluble transferrin receptor and type 2 diabetes, whereas the soluble transferrin receptor-to-ferritin ratio was significantly inversely related to the risk of type 2 diabetes in the median and high ratio subgroups (OR 0.71, 95% CI 0.51, 0.99 and OR 0.65, 95% CI 0.45–0.95)." Liu et al. 2020, p. 946. ↩︎

  50. "An overview of the study human cohorts and omics analyses pipeline can be found in Figure S1. Serum ferritin was measured in three cohorts: (a) a discovery cohort of subjects with obesity (n = 49); (b) a validation cohort of subjects with obesity from Italy and Spain (n = 628); and (c) an independent cohort of subjects with and without obesity from Spain (n = 130)...In both discovery and replication cohorts, serum ferritin increased with the severity of liver fat accumulation (Fig. 1a, b)." Mayneris-Perxachs et al. 2021, pp. 2-3. ↩︎

  51. "In my view, this is not a relevant argument. It is very well known that elevated ferritins alone do not portend iron loading and that liver disease is the main determinant. This paper simply documents a well-recognized association between metabolic disease, liver disease, and elevated ferritin. In other words, this is likely reverse causality (high ferritin is being caused by, and is not the cause of, metabolic disease). Ferritin is stored in the liver, and metabolic conditions (especially if they drive fatty liver disease) cause ferritin to leak from hepatocytes." Anonymous iron expert, report for GiveWell, p. 6. ↩︎

  52. "Fortify Health pays for and installs the equipment needed to fortify flour and pays for premix (which contains the iron compound that is used as a fortificant) so that its partner mills can fortify flour at no additional cost. It partners with privately-owned mills that produce flour that is sold at market prices to consumers." GiveWell, "Fortify Health — General Support (2019)" ↩︎

    • "Begin partnering with mills producing for schools in Maharashtra and explore partnerships with mills producing for the public distribution system (PDS) over 5 years. Fortify Health plans to provide premix and equipment to millers serving atta in the Amravati division of Maharashtra. Fortify Health has told us that having 5 years of funding is necessary for cultivating partnerships with the government, so we have recommended committing 5 years of funding, rather than 3 years, for this government partnerships work. We have not vetted this claim from Fortify Health, and it’s possible we should instead provide funding over 3 years to start and provide the additional 2 years of funding once Fortify Health meets milestones for the number of partnerships with mills producing for schools. The cost of this component is $1.9 million over 5 years." GiveWell, "Fortify Health – Support for Expansion (December 2021)"
    • See here for more information on India's Public Distribution System.
    ↩︎

  54. See our discussion of potential negative impacts of iron fortification programs here. ↩︎

  55. See our conversation notes here. ↩︎

  56. "Perspectives on fortification in India. The general public in India mostly does not hold strong views on food fortification, but it is somewhat contentious in some circles. Groups who are critical of food fortification in India tend to argue that it is driven by corporate profit-seeking, rather than genuine concern for public health." GiveWell's non-verbatim summary of a conversation with Advait Deshpande, May 19, 2023. ↩︎

  57. "The trajectory of government-implemented fortification in India. Fortification started in 1953 but gained momentum in 2016 when the Food Safety and Standards Authority of India (FSSAI) published standards for fortified food. In 2021, Prime Minister Modi announced that by 2024 all rice provided through social safety net programs will be fortified." GiveWell's non-verbatim summary of a conversation with Advait Deshpande, May 19, 2023. ↩︎

  58. "NGOs that receive funding from sources outside of India play a vital role in supporting fortification initiatives. As long as these NGOs adhere to the appropriate rules and regulations in India, they can operate effectively and contribute significantly to the fortification efforts, irrespective of the origin of their funding." GiveWell's non-verbatim summary of a conversation with Advait Deshpande, May 19, 2023. ↩︎

  59. Nikita Patel, Chief Executive Officer, Fortify Health, email to GiveWell, September 18, 2023 (unpublished) ↩︎

  60. One example of a study including some discussion of tea as an iron absorption inhibitor is Thankachan et al 2008. ↩︎

  61. One example of a study including some discussion of ascorbic acid as an iron absorption enhancer is Thankachan et al 2008. ↩︎

  62. See descriptions of trial participants in the "Field et al. 2021 meta-analysis" tab of our CEA. ↩︎

Ulrik Horn @ 2023-10-26T04:15 (+11)

I just want to say that I am really happy you looked into the points raised by Akash. I am not an expert in health or nutrition but did along with people close to me look into the extension of "western" datasets into non-western populations and found that for maternal health, there are somewhat large issues around gestational periods. In particular, we found that there is emerging evidence that the gestational period of persons of either African or South Asian decent might be about 1 week shorter than that of persons of European descent, on which WHOs recommendations were based. This could have large impacts as one might be too late in administering treatments to induce labor for patients of African or South Asian descent. In general, I got a feeling that the use of western datasets in recommendations for non-western populations is quite widespread and we are only starting to realize the issues this introduces in healthcare. Another data point was how blood oxygen saturation measured by those fingertip devices might have significant bias towards people with higher levels of melatonin in their skin. I think it is especially important to be aware of such issues as EA is majority white and a majority of our interventions are in non-white populations and am really happy you are also aware of reputational risks posed by using data with bias.

Akash Kulgod @ 2023-11-02T10:33 (+3)

Pretty cool to see this! I'll try and provide a response with some substantive thoughts in a couple weeks, though I'm unlikely to have capacity to do the deep dive that the complexity of the topic requires. Really hoping the issue of 'external validity ' especially w.r.t to Global South populations gets more discourse bandwidth. Big plus one to Ulrik's comment, this seems to extend beyond iron and anemia.

If anyone is interested in further research in anemia and adolescent cognitive development in India specifically, please do reach out. My father Shashikant Kulgod, a gastroenterologist and founder of the non-profit Rajlakshmi Children Foundation, and I are interested in working with a growing dataset we have of public school students from multiple districts in North Karnataka. We have lacked the expertise/capacity to do much with the data already collected (pre/post COVID) but likely have the ability to collect data from thousands of students (health and cognitive data points). My email is akashkulgod@berkeley.edu