Serum or plasma ferritin concentration as an index of iron deficiency and overload

Maria Nieves Garcia-Casal; Sant-Rayn Pasricha; Ricardo X Martinez; Lucero Lopez-Perez; Juan Pablo Peña-Rosas

doi:10.1002/14651858.CD011817.pub2

. 2021 May 24;2021(5):CD011817. doi: 10.1002/14651858.CD011817.pub2

Serum or plasma ferritin concentration as an index of iron deficiency and overload

Maria Nieves Garcia-Casal ¹, Sant-Rayn Pasricha ², Ricardo X Martinez ³, Lucero Lopez-Perez ⁴, Juan Pablo Peña-Rosas ^5,^✉

Editor: Cochrane Tobacco Addiction Group

PMCID: PMC8142307 PMID: 34028001

Abstract

Background

Reference standard indices of iron deficiency and iron overload are generally invasive, expensive, and can be unpleasant or occasionally risky. Ferritin is an iron storage protein and its concentration in the plasma or serum reflects iron stores; low ferritin indicates iron deficiency, while elevated ferritin reflects risk of iron overload. However, ferritin is also an acute‐phase protein and its levels are elevated in inflammation and infection. The use of ferritin as a diagnostic test of iron deficiency and overload is a common clinical practice.

Objectives

To determine the diagnostic accuracy of ferritin concentrations (serum or plasma) for detecting iron deficiency and risk of iron overload in primary and secondary iron‐loading syndromes.

Search methods

We searched the following databases (10 June 2020): DARE (Cochrane Library) Issue 2 of 4 2015, HTA (Cochrane Library) Issue 4 of 4 2016, CENTRAL (Cochrane Library) Issue 6 of 12 2020, MEDLINE (OVID) 1946 to 9 June 2020, Embase (OVID) 1947 to week 23 2020, CINAHL (Ebsco) 1982 to June 2020, Web of Science (ISI) SCI, SSCI, CPCI‐exp & CPCI‐SSH to June 2020, POPLINE 16/8/18, Open Grey (10/6/20), TRoPHI (10/6/20), Bibliomap (10/6/20), IBECS (10/6/20), SCIELO (10/6/20), Global Index Medicus (10/6/20) AIM, IMSEAR, WPRIM, IMEMR, LILACS (10/6/20), PAHO (10/6/20), WHOLIS 10/6/20, IndMED (16/8/18) and Native Health Research Database (10/6/20). We also searched two trials registers and contacted relevant organisations for unpublished studies.

Selection criteria

We included all study designs seeking to evaluate serum or plasma ferritin concentrations measured by any current or previously available quantitative assay as an index of iron status in individuals of any age, sex, clinical and physiological status from any country.

Data collection and analysis

We followed standard Cochrane methods. We designed the data extraction form to record results for ferritin concentration as the index test, and bone marrow iron content for iron deficiency and liver iron content for iron overload as the reference standards. Two other authors further extracted and validated the number of true positive, true negative, false positive, false negative cases, and extracted or derived the sensitivity, specificity, positive and negative predictive values for each threshold presented for iron deficiency and iron overload in included studies.

We assessed risk of bias and applicability using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS)‐2 tool. We used GRADE assessment to enable the quality of evidence and hence strength of evidence for our conclusions.

Main results

Our search was conducted initially in 2014 and updated in 2017, 2018 and 2020 (10 June). We identified 21,217 records and screened 14,244 records after duplicates were removed. We assessed 316 records in full text. We excluded 190 studies (193 records) with reasons and included 108 studies (111 records) in the qualitative and quantitative analysis. There were 11 studies (12 records) that we screened from the last search update and appeared eligible for a future analysis. We decided to enter these as awaiting classification.

We stratified the analysis first by participant clinical status: apparently healthy and non‐healthy populations. We then stratified by age and pregnancy status as: infants and children, adolescents, pregnant women, and adults.

Iron deficiency

We included 72 studies (75 records) involving 6059 participants.

Apparently healthy populations

Five studies screened for iron deficiency in people without apparent illness. In the general adult population, three studies reported sensitivities of 63% to 100% at the optimum cutoff for ferritin, with corresponding specificities of 92% to 98%, but the ferritin cutoffs varied between studies. One study in healthy children reported a sensitivity of 74% and a specificity of 77%. One study in pregnant women reported a sensitivity of 88% and a specificity of 100%. Overall confidence in these estimates was very low because of potential bias, indirectness, and sparse and heterogenous evidence. No studies screened for iron overload in apparently healthy people.

People presenting for medical care

There were 63 studies among adults presenting for medical care (5042 participants). For a sample of 1000 subjects with a 35% prevalence of iron deficiency (of the included studies in this category) and supposing a 85% specificity, there would be 315 iron‐deficient subjects correctly classified as having iron deficiency and 35 iron‐deficient subjects incorrectly classified as not having iron deficiency, leading to a 90% sensitivity. Thresholds proposed by the authors of the included studies ranged between 12 to 200 µg/L. The estimated diagnostic odds ratio was 50.

Among non‐healthy adults using a fixed threshold of 30 μg/L (nine studies, 512 participants, low‐certainty evidence), the pooled estimate for sensitivity was 79% with a 95% confidence interval of (58%, 91%) and specificity of 98%, with a 95% confidence interval of (91%, 100%). The estimated diagnostic odds ratio was 140, a relatively highly informative test.

Iron overload

We included 36 studies (36 records) involving 1927 participants. All studies concerned non‐healthy populations. There were no studies targeting either infants, children, or pregnant women.

Among all populations (one threshold for males and females; 36 studies, 1927 participants, very low‐certainty evidence): for a sample of 1000 subjects with a 42% prevalence of iron overload (of the included studies in this category) and supposing a 65% specificity, there would be 332 iron‐overloaded subjects correctly classified as having iron overload and 85 iron‐overloaded subjects incorrectly classified as not having iron overload, leading to a 80% sensitivity. The estimated diagnostic odds ratio was 8.

Authors' conclusions

At a threshold of 30 micrograms/L, there is low‐certainty evidence that blood ferritin concentration is reasonably sensitive and a very specific test for iron deficiency in people presenting for medical care. There is very low certainty that high concentrations of ferritin provide a sensitive test for iron overload in people where this condition is suspected. There is insufficient evidence to know whether ferritin concentration performs similarly when screening asymptomatic people for iron deficiency or overload.

Plain language summary

How accurate are tests to measure the level of ferritin (a protein that stores iron) in the blood at diagnosing iron deficiency and overload?

Key messages

‐ Tests that measure the level of ferritin in the blood may be reasonably accurate for diagnosing iron deficiency (low iron levels) in people:

‐ seeking medical care; and

‐ whose doctors suspect iron deficiency.

‐ The accuracy of ferritin blood tests for diagnosing iron overload (high iron levels) is unclear, due to a lack of robust evidence.

‐ To strengthen the evidence, we need future studies to:

‐ investigate a wider range of populations; and

‐ identify the levels of ferritin in the blood that are the best indicators of iron deficiency and overload.

Why is it important to diagnose iron deficiency and overload?

Iron is a mineral found in every cell of the body. It comes from iron‐rich foods like red meat, beans and fortified cereal (to which iron has been added artificially), among others, or from supplements (iron tablets, micronutrient powders, drops or syrups). The body needs iron to make:

‐ red blood cells; and

‐ haemoglobin, a protein in blood that carries oxygen from the lungs to the rest of the body.

An iron test can show if someone has too little iron (iron deficiency) or too much iron (iron overload). It is important to test iron levels because:

‐ iron deficiency can cause anaemia (low levels of red blood cells or haemoglobin), tiredness and weakness. It can be a sign that someone has a serious health problem, such as internal bleeding; while

‐ iron overload can damage the liver, heart and other organs permanently.

What tests can be used to diagnose iron deficiency and overload?

There are several different tests available to check the level of iron in the body. The most accurate tests involve using a needle to collect a small sample of:

‐ bone marrow fluid (to diagnose iron deficiency); or

‐ tissue from the liver (to diagnose iron overload).

However, these tests are expensive and can be risky for people in poor health.

A simpler test involves measuring the level of ferritin (a protein that stores iron) in the blood, to estimate the amount of iron in the body.

What did we want to find out?

We wanted to find out if ferritin blood tests accurately diagnose iron deficiency and iron overload.

What did we do?

We searched for studies that compared ferritin blood tests against:

‐ tests of iron levels in the bone marrow, to diagnose iron deficiency; and

‐ tests of iron levels in the liver, to diagnose iron overload.

We compared and summarised the results of the studies and rated our confidence in the evidence, based on factors like study methods and sizes.

What did we find?

We included 72 studies of 6059 people investigating the ability of ferritin blood tests to diagnose:

‐ iron deficiency in people who sought medical care and whose doctor suspected iron deficiency (70 studies, 5709 people);

‐ iron deficiency in people without any sign of disease (five studies, 350 people); and

‐ iron overload suspected by a doctor (36 studies, 1927 people).

Evidence suggests that ferritin blood tests may be reasonably accurate for diagnosing iron deficiency in people seeking medical care. For example, in studies where people with fewer than 30 micrograms of ferritin in one litre of blood were diagnosed with iron deficiency, ferritin blood tests correctly identified:

‐ iron deficiency in four out of five people who did have iron deficiency; and

‐ no iron deficiency in 19 out of 20 people who had normal levels of iron.

The evidence was not robust enough to determine if ferritin blood tests accurately diagnose:

‐ iron deficiency in people without any sign of disease; or

‐ iron overload suspected by a doctor.

What are the limitations of the evidence?

The studies were:

‐ small;

‐ conducted in ways that may have introduced errors into their results; and

‐ focused on specific populations (like children, young people and pregnant women).

For these reasons, we have limited confidence in the evidence.

How up‐to‐date is this evidence?

The evidence is up‐to‐date to June 2020.

Summary of findings

Background

In 2020, anaemia is estimated to affect 29.6% of non‐pregnant women, 36.5% of pregnant women, and 60.2% of children 6‐59 months of age worldwide, equivalent to 269 million children with anaemia (WHO 2021). Iron deficiency is one of the major causes of anaemia (Chaparro 2019) and can also exist in the absence of anaemia (WHO 2011a). Specific assessment of iron status is therefore critical to defining the burden of iron deficiency in a population, and identifying the proportion of anaemia that is likely to be responsive to interventions that aim to improve iron status (Garcia‐Casal 2019). Conversely, iron overload due to genetic aberrations in genes regulating iron absorption can result in serious clinical consequences including cardiomyopathy, endocrinopathies and hepatic cirrhosis and cancer (Brissot 2017). Clinical detection of iron overload is thus of crucial importance, whilst ensuring public health iron intervention programmes are not contributing to the burden of iron loading is an emerging concern. Serum and plasma ferritin is a commonly utilised assay for evaluation of iron stores in individuals across the life course. Ferritin is an iron storage protein and measurement in the plasma or serum reflects iron stores in healthy individuals; low ferritin indicates iron deficiency while elevated ferritin reflects iron overload. However, ferritin is also an acute‐phase protein and its levels are elevated in inflammation and infection. Ferritin is a continuous index and, as such, thresholds below which to define iron deficiency and above which to define iron overload are required.

Reference standard measures of iron deficiency (i.e. bone marrow aspirate, which requires intramedullary sampling of the hematopoietic cells) and iron overload (liver biopsy) are generally invasive, and are inappropriate for widespread community screening in epidemiologic studies and clinical use in unselected patients (Pasricha 2010). Thus, if measurement of serum ferritin in the peripheral blood can indicate iron deficiency or iron overload, it can save the individual considerable cost and distress. At the population level, having a biomarker that can be used as an indicator of iron deficiency or iron overload can be useful for understanding the prevalence, magnitude and distribution of the problem. Nonetheless, there is considerable uncertainty and variability concerning the optimal thresholds of ferritin to use, thus making it of importance to determine relevant thresholds for clinical practice and epidemiological assessments.

A previous systematic review included 55 studies that compared laboratory tests with histologic examination of the bone marrow and concluded that serum ferritin was a useful test for the diagnosis of iron‐deficiency anaemia with an area under the receiver operating characteristic curve (AUC^ROC), a plot of true positives versus false positives, of 0.95 (Guyatt 1992), that is, high accuracy because the agreement with the reference standard is almost perfect. The authors found that a ferritin < 15 ng/mL had a positive likelihood ratio for iron deficiency of 51.9. However, this review is over 20 years old and newer data are likely to be available; furthermore, the authors did not provide specific analysis of the utility of ferritin by itself in different populations or contexts. The authors included only anaemic adults (i.e. excluded studies in non‐anaemic individuals and children). The authors also did not assess the diagnostic value of a particular threshold for ferritin, or include in their analysis evaluation of the currently used thresholds (i.e. lower than 12 µg/L or lower than 15 µg/L). Methodologies for measurement of ferritin have evolved over the past two decades. The above‐mentioned review did not address the diagnostic value of ferritin in other important conditions, namely iron deficiency in the absence of anaemia, and iron overload, thus thresholds to identify these conditions have not been evaluated. Furthermore, it did not report ferritin thresholds for use in clinically relevant subgroups such as in children and in women during pregnancy.

Development of modern guidelines and recommendations for use in public health policy, clinical laboratories and, hence, routine practice requires a complete, up‐to‐date appraisal of the literature, including recent studies. It is important to ensure that evidence covers the full range of potential conditions. It is thus essential to now undertake a new systematic review to identify the diagnostic test accuracy of serum and/or plasma ferritin as a diagnostic indicator in both iron deficiency and iron overload.

Target condition being diagnosed

Iron is an essential element with important physiologic functions in processes including oxygen transport (as a key component of haemoglobin), mitochondrial function, enzymatic activity (especially cytochromes), and muscle metabolism, particularly myoglobin (Ordway 2004; Netz 2013).

Iron deficiency

Iron deficiency exists when body iron stores are inadequate to meet the needs for metabolism. Progressive iron deficiency can result in iron‐deficient erythropoiesis and eventually, iron deficiency anaemia (BloodSafe eLearning Australia 2011). However, even in the absence of anaemia, iron deficiency appears to be associated with clinical impairments including fatigue (Verdon 2003), impaired physical performance (Pasricha 2014), decreased work productivity (Li 1994), and sub optimal brain development (Lozoff 2007).

Iron deficiency involves inadequate iron intake, excess iron (i.e. blood) loss, and increased iron utilisation, and may occur due to physiologic, environmental, pathologic, drug‐related, genetic or iron‐restricted erythropoietic causes (Camaschella 2015). Inadequate iron intake may result from a diet with a low iron content, and/or which contains iron in a biologically inaccessible form or a diet containing high concentrations of iron absorption inhibitors. Iron may also fail to be absorbed in individuals with intestinal disorders such as celiac disease, and perhaps Helicobacter pylori infection (Papagiannakis 2013). Inflammation can also impair iron absorption, which may mediate iron deficiency in athletes (Peeling 2008). The most common causes of blood loss is menstruation, which is the chief reason iron deficiency is more common in females; in low‐income settings other important causes include chronic blood loss from hookworm and schistosomiasis. In high‐income settings, blood losses from blood donation (Spencer 2013), and bleeding from intestinal lesions (Bull‐Henry 2013) must be considered. Iron requirements are increased during rapid growth (especially in infants and preschool children) and during adolescence when growth accelerates again and coincides with the onset of menarche in females (Pasricha 2010; Pasricha 2013), while iron requirements during pregnancy are increased due to iron needs for maternal and foetal erythropoiesis and foetal growth (Pasricha 2010; Pasricha 2013).

When considering these factors, it is unsurprising that iron deficiency and iron deficiency anaemia are most common in preschool children and women of reproductive age, and that overall, iron deficiency is most common in low‐income settings where dietary iron content and availability is low, and where parasitic infections are highly prevalent (Pasricha 2013). As such, it is estimated that approximately about a third of the world’s population is anaemic, with iron deficiency the leading cause, and that anaemia accounts for almost 9% of the world's years lived with a disability burden (Kassebaum 2014). Since 2000, the global prevalence of anaemia in children under five has slowly decreased over the years, from 48.0% (95% UI 45.1%, 51.0%) to 39.8% (95% UI 36%, 43.8%), and from 2010, it has been stagnant. The global prevalence of anaemia in women of reproductive age has been stagnant, while the prevalence of anaemia in pregnant women has decreased slightly (WHO 2021).

Iron overload

Because elemental iron is toxic to the body (due to its propensity to initiate redox reactions and generate free radicals, causing tissue damage), it must be chaperoned and stored in the body by binding proteins (i.e. transferrin, ferritin). There is no physiologic mechanism to excrete iron, hence homeostatic regulation of iron stores is entirely mediated by changes in iron absorption (chiefly via modulation of the hepatic hormone, hepcidin) (Ganz 2013). Iron overload results from excess iron absorption caused by generally autosomal recessive genetic conditions (hereditary haemochromatosis, caused chiefly by mutations in the High FE2+ (HFE) gene, but also less commonly by mutations in genes that encode haemojuvelin (HJV), transferrin receptor 2 (TFR2) and ferroportin (SLC40A1)), conditions associated with ineffective erythropoiesis (for example, thalassaemia intermedia and haemoglobin E‐beta thalassaemia), and iron accumulation from repeated red cell transfusions, usually to treat inherited (e.g. thalassaemia and other congenital conditions) or acquired (e.g. aplastic anaemia, myelodysplasia and other causes of bone marrow failure) anaemia (Brissot 2009; Piga 2009). We can distinguish two types of iron overload syndromes: inherited conditions ‐ primary iron overload, and acquired conditions ‐ and secondary iron overload.

Over time, iron overload results in excess iron accumulation in organs, especially the liver (resulting in cirrhosis, liver failure and hepatocellular carcinoma), endocrine organs (causing pituitary and gonadal failure), pancreas (causing diabetes), skin (causing pigmentation), and heart (resulting in cardiomyopathy, heart failure and arrhythmia) (Allen 2008). Untreated, patients with severe iron overload succumb to cardiac or hepatic complications. Thus, early diagnosis and noninvasive tests for monitoring of treatment, are essential to optimal management of iron loading (Bacon 2011).

The global burden of iron overload is uncertain and varies by population (ethnicity, age and sex) screened, as well as the methodology used to make the estimate (screening with ferritin, transferrin saturation or genetic testing). The prevalence of hereditary haemochromatosis in Caucasian populations has been estimated to be perhaps 3.5 to 4.5 per 1000 population, greater in males (Baer 1995; Phatak 1998). Worldwide, approximately 21,000 children are born with haemoglobin E‐beta thalassaemia (about half of whom are transfusion dependent) and approximately 23,000 are born with thalassaemia major annually; a further 14,000 are born with Haemoglobin H disease (HbH), a form of alpha thalassaemia. Thus, genetic conditions associated with risk of iron overload are prevalent conditions worldwide (Weatherall 2011).

Index test(s)

Ferritin can be measured in the serum or plasma, as well as in the erythrocyte. Ferritin is an iron storage protein, chiefly found in the liver, and comprises two isoforms of poli peptidic chains (L‐light, and H‐heavy). Assays for ferritin generally do not distinguish between these forms. Synthesis of ferritin is regulated in part by iron‐responsive proteins, which bind to iron responsive elements in the ferritin messenger ribonucleic acid (mRNA). Increased iron levels suppress binding of iron binding proteins to the iron responsive element, facilitating translation and thus increasing the quantity of ferritin synthesised; conversely, iron deficiency suppresses ferritin synthesis. However, ferritin is also an acute‐phase protein and is increased in acute and chronic inflammation due to infection, inflammatory illnesses and malignancy. As such, coexistent iron deficiency and inflammation can result in relative increases in ferritin when compared to similar body iron stores without inflammation. Elevated ferritin can reflect either iron overload or inflammation (or coexistence of both). Marked elevations in ferritin may also reflect hepatic damage with release of intracellular ferritin. This circulating form does not carry iron (apoferritin) and is predominantly glycosylated, extending its circulating half life compared with non‐glycosylated forms (Ferraro 2012). The small fraction of ferritin in the serum contributes little to overall iron storage, but is in equilibrium with the body’s depot iron and hence acts as an indicator for the level of the iron stores.

Measurement of ferritin can be performed in commercial laboratories using a range of automated methods including immuno‐turbidimetry and latex agglutination. Assays require separation of serum or plasma from the cellular component of whole blood. The numerous platforms and the widespread availability of ferritin assays, raise the potential for discrepancies between results in different laboratories. This problem is addressed through calibration of instruments using standards provided by the manufacturer which have themselves been calibrated against international reference standards; furthermore, accredited laboratories are generally required to participate in external quality assurance activities (which ensure that results for samples fall within a similar range to those reported by other laboratories), and also undergo regular inspections. A recent systematic review concluded that the laboratory methods most used to determine ferritin concentrations have comparable accuracy and performance (Garcia‐Casal 2018).

The World Health Organization (WHO) currently defines iron deficiency in adults as ferritin lower than 15 μg/L, and in children under five years of age when ferritin is lower than 12 µg/L, unless inflammation is coexistent, in which case iron deficiency may exist when ferritin is lower than 30 μg/L (WHO 2011b). Iron overload is considered to exist when ferritin exceeds 200 μg/L in males and 150 μg/L in females (WHO 2001). These thresholds were not derived from up‐to‐date systematic reviews and meta‐analyses, and various authorities and individual laboratories recommend different thresholds for defining pathology without clearly explaining their rationale; for example, the Royal College of Pathologists of Australasia has recommended ferritin thresholds below 30 μg/L to identify iron deficiency and a ferritin level exceeding 200 μg/L and 300 μg/L in females and males, respectively, as indicating iron overload (RCPA 2010).

Clinical pathway

Evaluation of iron status may be performed clinically, for individual patients, or across a population. Measurement of iron status in individuals is important to correctly define iron status and provide appropriate treatment for iron deficiency, to prompt further testing and management if iron overload is suspected, and to monitor interventions for both iron deficiency and iron overload. Measurement of iron status in populations is important to determine the prevalence and distribution of iron deficiency and overload, and thus to decide appropriate interventions, and to monitor and evaluate the impact and safety of implemented public health programmes.

A clinical pathway for investigation of iron deficiency is proposed in Figure 1.

The clinical pathway for iron overload with ferritin is depicted in Figure 2. In both cases, false negative cases may see patients undiagnosed, resulting in an ongoing undetected risk of disease, whilst false positive results may result in unnecessary, expensive further testing and treatment.

In a population, measurement of iron indices may be undertaken together with measurement of haemoglobin concentrations. Often, field testing of haemoglobin is measured in isolation from red cell indices using instruments for point‐of‐care testing with a small sample of capillary blood. In population‐based testing, strategies for timely transport and separation of samples must be considered, as point‐of‐care tests for ferritin are not presently appropriate for these uses.

Ferritin determinations are widely used for monitoring renal anaemia when iron utilisation and distribution disorders are present during therapy with erythropoietin.

Prior tests

Individuals may undergo measurement of ferritin with no prior tests, with concurrent testing of other iron and haematologic parameters (i.e. haemoglobin, red cell indices), or following measurement of haematologic indices that have prompted further investigation into the iron status of the individual.

In a population, ferritin testing to ascertain the prevalence of (and risk factors for) iron deficiency is usually performed together with measurement of haemoglobin to assess anaemia prevalence. Where possible, concomitant measurement of an inflammatory marker (for example, C‐reactive protein) is recommended (WHO 2001). Additional iron indices such as soluble transferrin receptor may also be performed (WHO/CDC 2007).

Role of index test(s)

Ferritin measurements are presently undertaken as the key test to define iron status in an individual or to establish the prevalence of iron deficiency (or overload) in a population, as serum or plasma ferritin concentrations are thought to reflect body iron stores (WHO 2011b). It is, therefore, important to determine if serum or plasma ferritin concentrations accurately reflect iron statuses of relevance to population health (deficiency and overload) and to identify the thresholds that define iron deficiency and overload in clinical care.

Alternative test(s)

Other indices of iron status can be measured in addition to ferritin for iron status, although they evaluate different components of iron metabolism (Pasricha 2010). Ideally, the final diagnosis of iron status needs to consider the whole clinical picture incorporating all laboratory results, rather than relying on single test findings (Camaschella 2015).

Iron indices

Serum iron: reduced in iron deficiency and inflammation
Transferrin: increased in iron deficiency, reduced in inflammation
Transferrin saturation: increased in iron overload, reduced in iron deficiency and inflammation. In cases where elevated ferritin is due to inflammation or liver disease (for example, hepatitis or fatty liver disease), transferrin saturation will not be raised
Total iron binding capacity: reduced in iron overload, increased in iron deficiency and normal to increased in inflammation
Soluble transferrin receptor: increased in iron deficiency (reflecting tissue iron), increased in raised erythropoiesis (i.e. physiologic states such as pregnancy, childhood; disease states including thalassaemia and haemolysis)

Haematologic indices

Haemoglobin – falls in intermediate to late stages of iron deficiency, but may only cross the threshold for anaemia in late stages of iron deficiency. Numerous alternative causes for anaemia
Red cell indices (particularly: mean cell volume – reduced in iron deficiency, mean cell haemoglobin – reduced in iron deficiency, red cell distribution width – increased in iron deficiency)
Reticulocyte haemoglobin – reduced in iron deficiency

Other indices

Zinc protoporphyrin – increased in iron deficiency, haemoglobinopathy and anaemia of inflammation
Hepcidin: reduced in iron deficiency and erythropoiesis, increased in iron repletion and inflammation, pathologically reduced (or in the normal range despite iron overload) in hereditary haemochromatosis

Inflammatory indices

Because ferritin (as well as other iron indices, especially serum iron) is influenced by inflammation, concurrent measurement of an inflammatory marker (for example, C‐reactive protein, alpha‐1 glycoprotein) may be necessary to identify inflammation and enable interpretation of ferritin in clinical settings where individuals may have concurrent inflammatory conditions, and in population studies in low‐income settings where infection burden is high.

Liver function tests

Ferritin concentrations are elevated in both acute and chronic liver disease due to hepatic cell necrosis (the liver is the primary storage organ for ferritin). Hepatocellular damage can be reflected by measurement of alanine transaminase and aspartate transaminase levels.

Rationale

Current WHO threshold values for ferritin concentrations to indicate depleted iron stores are provided for individuals younger than five years of age and five years of age or older, for males and females, and for individuals under five years of age with concurrent infection. These are based on consultations in 1987 and 2004, neither of which included a systematic review and meta‐analysis. A 1987 consultation by the International Nutritional Anaemia Consultative Group concluded that, at all ages, a serum ferritin value below 10‐12 μg/L was indicative of depletion of iron stores (WHO 1989). The thresholds for adults were derived largely from the clinical literature, specifically from studies examining the highest ferritin concentration among patients with microcytic iron deficiency anaemia who also either show a therapeutic response to iron or who have no stainable iron in the bone marrow (WHO 2011b). These thresholds were therefore adopted by WHO and revised in 1993.

In 2004, WHO and the United States Centers for Disease Control and Prevention (CDC) held a technical consultation on the assessment of iron status at the population level, and selected five out of 16 indicators: haemoglobin, zinc protoporphyrin, mean cell volume, transferrin receptor, serum ferritin along with C‐reactive protein (due to effects on ferritin concentration of inflammation) (WHO/CDC 2007). In preparation for that consultation, a non‐systematic review sought to identify the most efficient indicators to evaluate the impact of interventions to control iron deficiency and detect a true change in iron status of a population using the fewest and simplest tests. Serum ferritin and haemoglobin were determined to be the most efficient indicators of population response to iron interventions (Mei 2005). This review included data from nine randomised controlled trials in seven countries to evaluate the performance of haemoglobin, ferritin, transferrin receptor, zinc protoporphyrin, mean cell volume, transferrin saturation and total body‐iron stores in measuring a change due to an iron intervention. They expressed the change in each biomarker as response to the iron intervention in standard deviation units for the intervention group compared with the control group, concluding that haemoglobin and ferritin showed the biggest changes.

However, WHO policy now requires more rigorous processes to ensure that recommendations are informed by the best available evidence, including formal evaluation of its quality. Reviews of evidence must include the most recent available literature.

Following the EURopean micronutrient RECommendations Aligned Network of Excellence methods for assessment of micronutrient status (Hooper 2009), the Grading of Recommendations Assessment, Development and Evaluation (GRADE) methodology (GRADE Working Group 2004; Guyatt 2008), and the previous WHO/CDC meeting (WHO/CDC 2007), the group concluded that “There is an urgent need for better information on the iron status of populations to enable the right interventions to be chosen for combating both iron deficiency and anaemia, and then, once programmes are in place, to have the right indicators to monitor their impact”.

This review therefore seeks to evaluate the accuracy of ferritin as an index of iron deficiency (where iron deficiency is defined by a reference standard method), and to assess the quantity and quality of evidence available to support the selection of thresholds of ferritin that define iron deficiency in outpatients, otherwise well individuals without inflammation, as well as individuals with (or at risk of) inflammation, We are also seeking to identify thresholds for iron overload in individuals with hereditary haemochromatosis or non‐transfusion dependent thalassaemia (i.e. mediated by excess iron absorption), and those with transfusion‐induced iron overload.

Objectives

To determine the diagnostic accuracy of ferritin concentrations (serum or plasma) for detecting iron deficiency;
To determine the diagnostic accuracy of ferritin concentrations (serum or plasma) for detecting iron overload in primary and secondary iron‐loading syndromes.

Secondary objectives

To determine the diagnostic accuracy of ferritin to determine iron deficiency at current WHO threshold values (WHO 2011b): < 12 µg/L; < 15 µg/L and < 30 µg/L; as well as other commonly utilised thresholds;
To assess the diagnostic accuracy of ferritin to determine iron overload at putative threshold values such as 150 µg/L in females and 200 µg/L in males (Bacon 2011; WHO 2001).

Methods

Criteria for considering studies for this review

Types of studies

We included all study designs seeking to evaluate ferritin as an index of iron status where it had been compared against an eligible reference standard (Bossuyt 2008). We expected most studies to be cross‐sectional (where recruited individuals with and without the target condition are included in proportion to their prevalence in the overall sample), with essentially concurrent measurement of the index tests (ferritin).

Ferritin could have been assessed alone or together with (and/or in comparison with) other tests. Studies must have measured the index (ferritin) and reference test simultaneously, or at the very least before any treatment (for iron deficiency, defined as oral iron supplementation or other iron treatment e.g. intravenous iron) was initiated, to ensure the tests being compared reflected the same status. We did, however, include studies where this was not explicitly stated.

We included both prospective and retrospective studies in the analysis. We assessed the effect of these studies as outlined in Sensitivity analyses.

We also included case studies that measured simultaneously ferritin and the reference standard and summarised the distribution of ferritin concentrations at different stages of iron deficiency and iron overload.

We included studies of participants where recruitment was based in the laboratory (i.e. a consecutive series of bone marrow examinations were reviewed, with concurrent or retrospective measurement of serum ferritin) as well as studies where recruitment occurred in the clinic or in the population. We anticipated clinical studies were likely to be prospective and laboratory‐based studies retrospective; in this case, we evaluated the effects of clinic and laboratory‐based studies using sensitivity analysis.

Iron deficiency

We included studies that measured the diagnostic accuracy of ferritin as an index of iron deficiency defined by the reference standard of bone marrow iron stores, as measured by assessing marrow macrophage iron content on the aspirate via an iron stain.

Iron overload

We included studies that measured the diagnostic accuracy of ferritin against a reference standard of hepatic iron content, where the hepatic tissue was obtained via liver biopsy, and the iron content assessed using either:

liver tissue iron measurements (generally via measurement of non‐heme iron content), or
liver iron overload (defined by Perl’s (Prussian blue) stain) using an accepted scoring system.

Participants

Iron deficiency

We included participants of any sex, age (i.e. adults, children and infants), pregnancy status, hospitalisation status, living in any country. We stratified the populations according to apparently healthy and non‐healthy status. We aimed to assess the differences between infants and children, adolescents, pregnant women and adults. This latter category included all nonpregnant adults, since studies on pregnant women were analysed separately.

Iron overload

We included patients at risk of iron overload from primary (non‐transfusion‐induced or IV iron‐induced) causes, including participants of any sex, age, pregnancy status, hospitalisation status, living in any country.

Specific exclusion criteria

Renal failure causes very specific changes to iron homeostasis and interpretation of ferritin levels in this context is a specialised skill, with a different set of accepted conventions. We thus considered studies exclusively recruiting participants with end‐stage renal failure to be outside the scope of this review. We therefore excluded participants with end‐stage renal failure (as defined by study authors, or with a glomerular filtration rate reported among the study population < 60 mL/min), or in patients receiving haemodialysis.

We also excluded studies exclusively recruiting patients who had undergone chronic red cell transfusions. Such patients are likely to have undergone specific, complex investigations and management (including iron chelation). Thus, clinicians caring for these patients were likely to already have a high clinical suspicion of iron overload, and would not rely on a ferritin alone to diagnose liver iron loading.

Index tests

Serum or plasma ferritin concentration measured by any current or previously available quantitative assay i.e. enzyme‐linked immunosorbent assay, radioimmunoassay, chemiluminescence immunoassay or nephelometry.

Target conditions

Iron deficiency (as diagnosed by bone marrow iron stores);
Iron overload (as diagnosed by liver biopsy and measured by tissue iron quantification or histology).

Reference standards

Iron deficiency

The definitive diagnosis of iron deficiency is by demonstration of absent iron stores on a Perl's (Prussian blue) stain of the bone marrow. Bone marrow aspirates are invasive but generally safe procedures (Melempati 2009). Aspirated bone marrow particles are spread on a slide and stained. Siderotic material (haemosiderin) is a water‐insoluble complex of ferric iron, lipid, protein and carbohydrate found chiefly in macrophages. The ferric ions in haemosiderin react with potassium ferrocyanide to form a blue compound, ferriferrocyanide (the Prussian blue, or Perl's reaction). In a normal bone marrow, iron can be detected. However, in iron deficiency, staining is markedly reduced or absent. The absence of stainable iron on a bone marrow aspirate that contains particles is diagnostic of iron deficiency. Bone marrow examination for iron deficiency is not a quantitative test although reporting systems often utilised categories to indicate absent, marginal, normal and excess iron (Beutler 1958; Riley 2009). Table 5 and Table 6 present two examples of grading systems for bone marrow iron.

1. Bone marrow iron content (Perl's Prussian blue stain method).

Grade	Features of Prussian blue stain
0	No stainable iron
+	Small intracellular iron stores using oil objective
++	Small, sparse intracellular iron particles at low power
+++	Numerous small intracellular iron particles
++++	Larger particles with a tendency to aggregate into clumps
++++++	Dense, large clumps
++++++	Very large clumps and extracellular iron

Grade	Interpretation
0	No iron observed
Trace	Rare intracellular iron granules
+	Iron granules seen, but fewer than normal
++	Normal iron content (10% of fields)
+++	Considerable amounts of iron in every field
++++	Considerable degree of iron excess

Grade	Ease of observation and magnification required
0	Granules absent or barely discernible at x 400
1	Granules barely discernible at x 250 and easily confirmed at x 250
2	Discrete granules resolved at x 100
3	Discrete granules resolved at x 25
4	Masses visible at x 10, or naked eye

Interpretation	Serum ferritin (µg/L)
	Less than five years of age		Five years of age or older
	Male	Female	Male	Female
Depleted iron stores	< 12	< 12	< 15	< 15
Depleted iron stores in the presence of infection	< 30	< 30	‐	‐
Severe risk of iron overload (adults)	‐	‐	> 200	> 150

Patients/population:		Healthy persons at risk of iron deficiency.
Settings		Community, outpatient, primary health care.
Prior testing		No prior test needed with concurrent testing of other iron and haematologic parameters or following measurement of haematologic indices.
Index test		Serum or plasma ferritin concentration measured by any available quantitative assay: i.e. enzyme‐linked immunosorbent assay, radioimmunoassay, etc.
Reference standard		Aspirated bone marrow examination to identify the iron stores presence or absence on a Perl's (Prussian blue) stain.
Importance		Avoid the challenges of an invasive, and occasionally risky bone marrow procedure, as well as elevated costs, and distress.
Studies		5 prospective/ambispective cohorts.
Test Accuracy	# of studies (# of participants)	Mean DOR = 1.6				Certainty of Evidence (CoE): Risk of bias: Crucial limitations¹ Indirectness: Serious² Inconsistency: Serious³ Imprecision: Serious⁴ Publication bias/other considerations: Undetected
		Effect per 1000 participants for prevalence of iron deficiency 41%**
		Outcome	Specificity = 60%	Specificity = 75%	Specificity = 90%
Sensitivity	5 (143)	TPs	203	111	30	⊕⊝⊝⊝ Very Low^1,2,3
		FNs	206	298	378
Specificity	5 (207)	FPs	237	148	59
		TNs	355	444	532

Patients/population		Non‐healthy patients at risk of iron deficiency.
Settings		Community, outpatient, primary health care.
Prior testing		No prior test needed with concurrent testing of other iron and haematologic parameters or following measurement of haematologic indices.
Index test		Serum or plasma ferritin concentration measured by any available quantitative assay: i.e. enzyme‐linked immunosorbent assay, radioimmunoassay, etc.
Reference standard		Aspirated bone marrow examination to identify the iron stores presence or absence on a Perl's (Prussian blue) stain.
Importance		Avoid the challenges of an invasive, and occasionally risky bone marrow procedure, as well as elevated costs, and distress.
Studies		66 prospective/ambispective cohorts, 3 retrospective cohorts, and one case‐control.
Test Accuracy	# of studies (# of participants)	Mean DOR = 49.2				Certainty of Evidence (CoE): Risk of bias: Crucial limitations¹ Indirectness: No Inconsistency: Serious² Imprecision: No Publication bias/other considerations: Undetected
		Effect per 1000 participants for prevalence of iron deficiency 36%**
		Outcome	Specificity = 75%	Specificity = 85%	Specificity = 95%
Sensitivity	70 (2051)	TPs	349	338	300	⊕⊕⊝⊝ Low^1,2
		FNs	10	21	59
Specificity	70 (3658)	FPs	256	160	64
		TNs	384	481	577

Patients/population:		Non‐healthy *adults at risk of iron deficiency.
Settings		Community, outpatient, primary health care.
Prior testing		No prior test needed with concurrent testing of other iron and haematologic parameters or following measurement of haematologic indices.
Index test		Serum or plasma ferritin concentration measured by any available quantitative assay: i.e. enzyme‐linked immunosorbent assay, radioimmunoassay.
Reference standard		Aspirated bone marrow examination to identify the iron stores presence or absence on a Perl's (Prussian blue) stain.
Importance		Avoid the challenges of an invasive, and occasionally risky bone marrow procedure, as well as elevated costs, and distress.
Studies		59 prospective/ambispective cohorts + 3 retrospective cohorts + 1 case‐control.
Test Accuracy	# of studies (# of participants)	Mean DOR = 49.8				Certainty of Evidence (CoE): Risk of bias: Crucial limitations¹ Indirectness: No Inconsistency: Serious² Imprecision: No Publication bias/other considerations: Undetected
		Effect per 1000 participants for prevalence or iron deficiency of 35%**
		Outcome	Specificity = 75%	Specificity = 85%	Specificity = 95%
Sensitivity	63 (1765)	TPs	331	315	250	⊕⊕⊝⊝ Low^1,2
		FNs	19	35	100
Specificity	63 (3277)	FPs	162	97	32
		TNs	487	552	617

Patients/population		Non‐healthy (adults, pregnant women) at risk of iron deficiency.
Settings		Community, outpatient, primary health care.
Prior testing		No prior test needed with concurrent testing of other iron and haematologic parameters or following measurement of haematologic indices.
Index test		Serum or plasma ferritin concentration measured by any available quantitative assay: i.e. enzyme‐linked immunosorbent assay, radioimmunoassay, etc.
Reference standard		Aspirated bone marrow examination to identify the iron stores presence or absence on a Perl's (Prussian blue) stain.
Importance		Avoid the challenges of an invasive, and occasionally risky bone marrow procedure, as well as elevated costs, and distress.
Studies		3 prospective/ambispective cohorts.
Test Accuracy	# of studies (# of participants)	Mean DOR = 57.9				Certainty of Evidence (CoE) Risk of bias: Crucial limitations¹ Indirectness: Serious² Inconsistency: Serious³ Imprecision: Serious⁴ Publication bias/other considerations: Undetected
		Effect per 1000 participants for prevalence of iron deficiency 53%**
		Outcome	Specificity = 80%	Specificity = 85%	Specificity = 90%
Sensitivity	3 (79)	TPs	503	464	352	⊕⊝⊝⊝ Very Low^1,2,3
		FNs	27	66	178
Specificity	3 (69)	FPs	94	71	47
		TNs	376	400	423

Patients/population		Non‐healthy (infants and children) less than 5 years old with infection/inflammation at risk of iron deficiency.
Settings		Community, outpatient, and primary health care.
Prior testing		No prior test needed with concurrent testing of other iron and haematologic parameters or following measurement of haematologic indices.
Index test		Serum or plasma ferritin concentration measured by any available quantitative assay: i.e. enzyme‐linked immunosorbent assay, radioimmunoassay, etc.
Reference standard		Aspirated bone marrow examination to identify the iron stores presence or absence on a Perl's (Prussian blue) stain.
Importance		Avoid the challenges of an invasive, and occasionally risky bone marrow procedure, as well as elevated costs, and distress.
Studies		4 prospective/ambispective cohorts.
Test Accuracy	# of studies (# of participants)	Mean DOR = 7.3				Certainty of Evidence (CoE) Risk of bias: Crucial limitations¹ Indirectness: Serious² Inconsistency: Serious³ Imprecision: Serious⁴ Publication bias/other considerations: Undetected
		Effect per 1000 participants for prevalence of iron deficiency 40%**
		Outcome	***** Spec = 60%**	***** Spec = 70%**	***** Spec = 80%**
Sensitivity	4 (207)	TPs	355	300	193	⊕⊝⊝⊝ Very Low^1,2,3,4
		FNs	45	100	207
Specificity	4 (312)	FPs	240	180	120
		TNs	360	420	480

Patients/population	*Adults with and without inflammation at risk of iron deficiency.
Settings	Community, outpatient, hospital.
Prior testing	No prior test needed with concurrent testing of other iron and haematologic parameters or following measurement of haematologic indices.
Index test	Serum or plasma ferritin concentration measured by any available quantitative assay: i.e. enzyme‐linked immunosorbent assay, radioimmunoassay, etc.
Reference standard	Aspirated bone marrow examination to identify the iron stores presence or absence on a Perl's (Prussian blue) stain.
Importance	Avoid the challenges of an invasive, and occasionally risky bone marrow procedure, as well as elevated costs, and distress.
Studies	32 studies (2538 participants) with inflammation and 5 studies (269 participants) without inflammation.
Between‐Study level Covariate: inflammation vs non‐inflammation association
HSROC Model and assumptions over parameters	Mean DOR inflammation	Mean DOR no inflammation	Mean threshold parameter inflammation	Mean threshold parameter no inflammation	P value for difference in the fit of the model		Certainty of Evidence (CoE) Risk of bias: Crucial limitations¹ Indirectness: No Inconsistency: Serious² Imprecision: No Publication bias: Undetected
Different accuracy, shape and threshold parameters	0	12.42	‐0.72	‐0.07			⊕⊕⊝⊝ Low ^1,2
Same shape	0	17.33	‐0.60	‐0.99	0.17 ( > 0.05 )	(removal of covariate for different shape)
Same shape and accuracy	0	36.16	‐0.59	‐1.07	0.4 ( > 0.05)	(removal of covariate for different shape and accuracy)

Disease group	N of studies	Estimated mean DOR	95% LCI	95% UCI
Alcoholism/liver cirrhosis/renal failure	7	78.38	13.91	441.69
Blood disorders (SS and/or SC or leukaemia)	9	36.71	8.35	161.33
Diverse chronic or acute diseases	19	44.48	15.00	131.90
Infectious diseases	2	9.91	1.462	67.21
Rheumatoid arthritis	11	59.18	21.02	166.67
All five disease groups of non‐healthy *adults	48	52.30	30.36	90.1

Analysis	N (studies)	Sensitivity and 95% confidence Interval (%)	Specificity and 95% confidence Interval (%)
Non‐healthy *adults	9	78.7 (57.6, 90.9)	98.0 (90.6, 99.6)
Non‐healthy *adults excluding low quality	8	81.7 (65.0, 91.5)	97.5 (91.6, 99.3)
Non‐healthy *adults excluding retrospectives	8	78.24 (57.9, 90.4)	96.92 (90.3, 99.1)

Studies	Study characteristics: Target population / Data Specifications / Serum Ferritin used threshold (SF_th)	Bone marrow equal to 0 group (BM = 0) (or as defined by author as iron‐depleted or diseased)			Bone marrow 1 or more group (BM = 1+) (or as defined by author as iron‐repleted or non‐diseased)			BM = 0 vs BM = 1+ comparison	Serum ferritin vs BM association
Studies		TP+FN=n₁*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	FP+TN=n₂*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	Mean difference (μg/L) in serum ferritin concentration: iron‐repleted group mean minus the iron‐depleted group mean	Pearson's correlation (r), Regression coefficient R², P‐value
Jonker 2014	Healthy, both sexes, 6 to 66 months. Sens, spec, D+, and D‐ given for the authors proposed threshold. Sens and spec for other two alternative thresholds was given. SF_th=18.	32+11=43	19	interquartile range: 12 to 36; median: 19	10+34=44				r = 0.39 (Kendall rank r)
Hallberg 1993	Healthy, 38‐year‐old women, and randomly selected from the population in Goteborg. TP, FN, FP, and TN given for the authors proposed threshold. Table of sens and spec for other 10 alternative thresholds was presented. SF_th=16.	52+17=69	12.5 (11.2)	geometric mean: 9.4	2+103=105	54.1 (40.5)	geometric mean: 44.1	41.6
Milman 1983	Healthy, both sexes, aged from 20 to 90 years, Danish medical students randomly selected. PPV, NPV, D+, and D‐ for the authors proposed threshold (among more than one threshold mentioned). SF_th=20.	10+6=16	15	4 to 30	2+35=37	65		50	significantly correlated (P < 0.001)
Puolakka 1980	Healthy, females, mean age of 26 years old, and in the second semester of pregnancy. Tables with D+, D‐, TP, FN, FP, and TN values for the authors proposed threshold. SF_th=70.	7+1=8		geometric mean: 111	0+8=8		geometric mean: 115		significantly correlated (P < 0.05)
Sorbie 1975	Healthy, both sexes, 17 to 30 years old, Canadian medical students selected randomly. TP, FN, FP, and TN extracted from figure, and tables given for the authors proposed unique threshold (among more than one threshold mentioned). SF_th=40.	7+0=7	18	7 to 38	1+12=13	92	59 to 170	74	significantly correlated (P < 0.05)
*Iron‐depleted sample size (n₁) could be different to the sum of TP + FN because the studies generally reported the numbers separately (some reasons were exclusion of values, subgroups etc.). Similarly, for iron‐repleted sample size (n₂) and the corresponding sum FP + TN.

Studies	Study characteristics: Target population / Data Specifications / Serum Ferritin used threshold (SF_th)	Bone marrow equal to 0 group (BM = 0) (or as defined by author as iron‐depleted or diseased)			Bone marrow 1 or more group (BM = 1+) (or as defined by author as iron‐repleted or non‐diseased)			BM = 0 vs BM = 1+ comparison	Serum ferritin vs BM association
Studies		TP+FN=n₁*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	FP+TN=n₂*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	Mean difference (μg/L) in serum ferritin concentration: iron‐repleted group mean minus the iron‐depleted group mean	Pearson's correlation (r), Regression coefficient R², P‐value
Aguilar 2012	Non‐healthy with Infectious diseases (malaria, tuberculosis, and other), and anaemia, both sexes, infants 6 to 59 months, attending the MDH emergency department in a rural hospital. Sens, spec, TP, FN, FP, and TN given for WHO threshold value. Sens and spec presented for two alternative thresholds. SF_th=30.	21+117=138			0+35=35
Jonker 2013	Non‐healthy with high incidence of infectious diseases (malaria, tuberculosis, and other), and severely anaemic, both sexes, children from 6 to 60 months. Specificity, sensitivity, D+, and D‐ for WHO threshold. SF_th=30.	18+24=42			4+191=195
Meira 2005	Non‐healthy with HIV and blood disorders, both sexes, children from 6 to 59 months. Patient's data available to compute TP, FN, FP, and TN for WHO threshold. SF_th=30.	1+2=3	32.8 (15.2)	17 to 47; median: 34.5	0+3=3	117.7 (89.1)	51 to 219; median: (82.6)	84.9
Phiri 2009	Non‐healthy and severely anaemic children in high infection pressure area (malaria endemic), both sexes, children with mean age of 20 months. Specificity, sensitivity, TP, FN, FP, and TN given for WHO threshold value, and specificity, sensitivity, D+, D‐, and accuracy for authors proposed threshold. SF_th=273.	18+6=24			19+60=79
*Iron‐depleted sample size (n₁) could be different to the sum of TP + FN because the studies generally reported the numbers separately (some reasons were exclusion of values, subgroups etc.). Similarly, for iron‐repleted sample size (n₂) and the corresponding sum FP + TN.

Studies	Study characteristics: Target population / Data Specifications / Serum Ferritin used threshold (SF_th)	Bone marrow equal to 0 group: B = 0 (or as defined by author as iron‐depleted or diseased)			Bone marrow 1 or more group: BM = 1+ (or as defined by author as iron‐repleted or non‐diseased)			BM = 0 vs BM = 1+ comparison	Serum ferritin vs BM association
Studies		TP+FN=n₁*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	FP+TN=n₂*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	Mean difference (μg/L) in serum ferritin concentration: iron‐repleted group mean minus the iron‐depleted group mean	Pearson's correlation (r), Regression coefficient R², P‐value
Oluboyede 1980	Non‐healthy, pregnant females, blood disorders: haemoglobin SS or SC disease. TP, FN, FP, and TN extracted from figure for the authors proposed threshold (among more than one threshold mentioned). SF_th=100.	7+2=9	44.8 (2.7)		5+8=13	129.7		84.9	significantly correlated (P < 0.001)
Puolakka 1980	Non‐healthy, pregnant females, before iron treatment in the second semester of pregnancy. Tables with D+, D‐, TP, FN, FP, and TN values for the authors proposed threshold (among more than one threshold mentioned). SF_th=35.	18+9=27		geometric mean: 52	0+6=6		geometric mean: 157		significantly correlated (P < 0.05)
Van den Broek 1998	Non‐healthy, pregnant females, almost half with HIV. Table with Sens, Spec, Acc, PPV, NPV, and LR for two thresholds: WHO and the authors proposed corresponding thresholds. SF_th=30.	39+4=43			7+43=50				significantly correlated (P < 0.005)
*Iron‐depleted sample size (n₁) could be different to the sum of TP + FN because the studies generally reported the numbers separately (some reasons were exclusion of values, subgroups etc.). Similarly, for iron‐repleted sample size (n₂) and the corresponding sum FP + TN.

Patients/population		Non‐healthy patients at risk of iron overload.
Settings		Community, outpatient, primary health care.
Prior testing		No prior test needed with concurrent testing of other iron and haematologic parameters or following measurement of haematologic indices.
Index test		Serum or plasma ferritin concentration measured by any available quantitative assay: i.e. enzyme‐linked immunosorbent assay, radioimmunoassay.
Reference standard		Liver biopsy test measured either using atomic absorption spectrophotometry, calorimetry, similar, or by using Perl's (Prussian blue) staining.
Importance		Avoid the challenges of an invasive, and occasionally risky biopsy procedure, as well as elevated costs, and distress.
Studies		27 prospective/ambispective cohorts + 2 retrospective cohorts + 7 case‐controls^1.
Test Accuracy	# of studies (# of participants)	Mean DOR = 8.1				Certainty of Evidence (CoE): Risk of bias: Very Serious² Indirectness: Serious³ Inconsistency: Serious⁴ Imprecision: No Publication bias/other considerations: Undetected
		Effect per 1000 participants for prevalence of iron overload 42%**
		Outcome	Specificity = 65%	Specificity = 75%	Specificity= 85%
Sensitivity	36 (803)	TPs	332	304	258	⊕⊝⊝⊝ Very Low^1,2,3,4
		FNs	85	113	159
Specificity	36 (1124)	FPs	204	146	87
		TNs	379	437	496

Analysis	N (studies)	Estimate of the mean DOR
Non‐healthy	70	49.16
Non‐healthy excluding low quality	61	47.56
Non‐healthy excluding retrospectives	63	46.92
Non‐healthy *adults	63	49.80
Non‐healthy *adults excluding low quality	56	50.70
Non‐healthy *adults excluding retrospectives	56	51.28

Domain	Yes	No	Unclear
*Subjects selection*
1. Consecutive or random sample enrolled?	If the study specifically states that consecutive patients or a random sample was selected.	If the study clearly states that the selection of patients was not consecutive or random, or if this can be easily inferred from the design.	If not reported or cannot be determined.
2. Was a case‐control design avoided?	If the study is cross‐sectional, prospective or retrospective using consecutive patients or a random selection.	If the study specifically states that it used a case‐control design or this can be easily inferred from the description.	If not reported or cannot be determined.
3. Did the study avoid inappropriate exclusions?	If there are no subjects inappropriately excluded	If there is inappropriate exclusion (e.g. suspicious ferritin or bone marrow results)	If not reported or cannot be determined.
Applicability: Is there concern that the included patients do not match the review question?
*Index test*
1. Were the index tests results interpreted without knowledge of the results of the reference standard?	If the study states that the ferritin concentration was performed independently from the assessment of bone marrow aspirates and liver biopsies OR the study specifically states that the interpretation of the index test was blinded to the reference test.	If the study clearly states that interpretation of the index test was not blinded to the reference test.	If not reported or cannot be determined.
2. If a threshold was used, was it prespecified?	If the thresholds were prespecified.	If the thresholds were not prespecified.	If not reported or cannot be determined.
Applicability: Is there concern that the index test, its conduct, or interpretation differ from the review question?
*Reference standard*
1. Is the reference standard likely to correctly classify the target condition?	If bone marrow aspirates were used to classify iron depletion and repletion AND liver biopsies were used to classify iron overload.	If other markers of iron status were used OR bone marrow aspirates and liver biopsies were not used.	If not reported or cannot be determined.
2. Were the reference standard results interpreted without knowledge of the index test?	If the study clearly states that the index test was not used as part of the reference standard interpretation AND the assessment of the reference criteria was blinded to the index test result.	If the study clearly states that the index test was used as part of the reference criteria OR the assessment of the reference criteria was not blinded to the index test result.	If not reported or cannot be determined.
Applicability: Is there concern that the target condition as defined by the reference standard does not match the review question?
*Flow and timing*
1. Was there an appropriate time interval between the index test and reference standard?	If the index test for each subject was obtained from the same sampling moment as the reference test.	If the index test for each subject was performed from different sampling moments than the reference test. (subclassify if the time interval between samplings varies widely).	If not reported or cannot be determined.
2. Did all subjects receive a reference standard?	If all subjects (or a random subset) who received the index test were tested for the reference standard.	If not all subjects received a reference standard OR a nonrandom subset of subjects were evaluated by the reference test.	If not reported or cannot be determined.
3. Did all subjects receive the same reference standard?	If bone marrow aspirates or liver biopsies were performed in all patients.	If other tests were performed OR bone marrow aspirates or liver biopsies were not performed in all patients.	It is not clear if bone marrow aspirates or liver biopsies were performed in all patients.
4. Were all subjects included in the analysis?	If all the subjects enrolled were included in the analysis.	If not all the subjects enrolled were included in the analysis.	It is not clear if all subjects enrolled were included in the analysis.

Studies	Study characteristics: Target population / Data Specifications / Serum Ferritin used threshold (SF_th)	Bone marrow equal to 0 group: BM = 0 (or as defined by author as iron‐depleted or diseased)			Bone marrow 1 or more group: BM = 1+ (or as defined by author as iron‐repleted or non‐diseased)			BM = 0 vs BM = 1+ comparison	Serum ferritin vs BM association
Studies		TP+FN=n₁*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	FP+TN=n₂*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	Mean difference (μg/L) in serum ferritin concentration: iron‐repleted group mean minus the iron‐depleted group mean	Pearson's correlation (r), Regression coefficient R², P‐value
Barron 2001	Non‐healthy, both sexes, 50 years or older, with blood disorders. D+, and D‐, and tables given for the authors proposed threshold. Sens and spec given for two alternative thresholds. SF_th=100.	29+6=35		2 to 544; median: 29	0+0=0
Rao 1984	Non‐healthy, both sexes, mixed age with 56.8 years as mean age, with blood disorders. D+, D‐, TP, TN values for the authors unique proposed threshold. SF_th=30.	5+12=17	112	15 to 1056	1+42=43	326	43 to 1821	214
Ruivard 2000	Non‐healthy (mixed with healthy), both sexes, 60 years as mean age, with diverse chronic mainly blood disorders. Table with Sens, Spec, Acc, PPV, NPV, and LR for the authors proposed threshold (among more than one threshold mentioned). SF_th=60.	16+5=21	85 (165)		1+32=33	778 (2012)		693
Witte 1986	Non‐healthy, both sexes, 60 years as mean age, with blood disorders. Figure, D+, D‐ data for the authors unique proposed threshold. SF_th=12.	11+1=12		12 to 126	0+85=85		8 to 205		r = 0.7752
Brown 1988	Non‐healthy, both sexes, mixed age with diverse chronic mainly blood disorders. Figure for the authors unique proposed threshold. SF_th=15.	1+7=8		12 to 100	0+51=51		20 to 1500
Forman 1980	Non‐healthy, both sexes, mixed age with blood disorders. Figure, D+, D‐ data for the authors unique proposed threshold. SF_th=15.	4+7=11		4 to 500	4+38=42		11 to 850		"good correlation"
Solomon 1981	Non‐healthy, both sexes, 50 years and older with blood disorders. Patient's data available to compute TP, FN, FP, and TN at the authors unique proposed threshold. SF_th=40.	0+3=3	278 (212)	136 to 522; median: 175	0+9=9	427 (418)	45 to 1170; median: 238	149
Mast 2002	Non‐healthy, both sexes, over 20 years old with blood disorders. Sens, Spec, PPV, NPV for the authors unique proposed threshold (among more than one threshold mentioned). SF_th=50.	12+16=28			3+47=50
Coenen 1991	Non‐healthy, both sexes, mixed age with blood disorders. Figure, D+, D‐ data for the authors proposed threshold (among more than one threshold mentioned). SF_th=70.	14+12=26			5+42=47
Adults: this group was composed of adults excluding those studies focused exclusively on pregnant women. *Iron‐depleted sample size (n₁) could be different to the sum of TP + FN because the studies generally reported the numbers separately (some reasons were exclusion of values, subgroups etc.). Similarly, for iron‐repleted sample size (n₂) and the corresponding sum FP + TN.

Studies	Study characteristics: Target population / Data Specifications / Serum Ferritin used threshold (SF_th)	Bone marrow 0 group: BM = 0 (or as defined by author as iron‐depleted or diseased)			Bone marrow 1 or more group: BM = 1+ (or as defined by author as iron‐repleted or non‐diseased)			BM = 0 vs BM = 1+ comparison	Serum ferritin vs BM association
Studies		TP+FN=n₁*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	FP+TN=n₂*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	Mean difference (μg/L) in serum ferritin concentration: iron‐repleted group mean minus the iron‐depleted group mean	Pearson's correlation (r), Regression coefficient R², P‐value
Kim 2000	Non‐healthy, both sexes, mixed age (mean 50.1 years old) with rheumatoid arthritis. Patient's data available to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=12.	4+1=5	17.7 (19.5)		1+12=13	72.2 (66.3)		54.5	correlated (P < 0.118)
Baumann Kurer 1995	Non‐healthy, both sexes, mixed age (mean 57.5 years old) with rheumatoid arthritis. Sens, spec, D+, and D‐ for the authors unique proposed threshold. SF_th=30.	12+2=14		interquartile range: 4 to 28; median: 12.5	3+28=31		interquartile range: 46.5 to 165.5; median:99		R² = 0.721 (P < 0.0001)
Shroff 1991	Non‐healthy, both sexes, 20 to 49 years old with rheumatoid arthritis. Sens, spec, D+, and D‐ for the authors unique proposed threshold. SF_th=32.	15+2=17	23.9 (11.5)		2+11=13	69.9 (24.7)		46	r = 0.802 (P < 0.01)
Mulherin 1996	Non‐healthy, both sexes, 50 years and older with rheumatoid arthritis. Sens, spec, D+, and D‐ for the authors proposed threshold (among more than one threshold mentioned). SF_th=40.	20+2=22	24	4 to 101	10+6=16	181	12 to 451	157	significantly correlated (P < 0.0001)
Vreugdenhil 1990	Non‐healthy, both sexes, 50 years and older (mean 63.4 years old) with rheumatoid arthritis. Sens, spec, PPV, D+, and D‐ for the authors proposed threshold (among more than one threshold mentioned). SF_th=50.	20+4=24		10 to 36; median: 17	3+17=20		10 to 246; median: 75.6
Hansen 1983	Non‐healthy, both sexes, 50 years and older with rheumatoid arthritis. Table, figure, sens, spec, and D+, D‐ at the authors unique proposed threshold. SF_th=60.	12+2=14	29.9 (24.1)	8 to 98; median: 22	2+14=16	189.7 (124.7)	38 to 410; median 125	159.8	r = 0.8358
Nielsen 1990	Non‐healthy, both sexes, 50 years and older with rheumatoid arthritis. Figure, D+, D‐ for the authors unique proposed threshold. SF_th=60.	19+5=24	51.6 (57.7)		1+10=11	257.4 (228.1)		205.8
Suominen 2000	Non‐healthy, both sexes, mixed age (26 to 79 years old) with rheumatoid arthritis. Table, figure, and D+, D‐ for the authors unique proposed threshold. SF_th=60.	11+1=12	24.5 (16.7)	10 to 70	8+9=17	120.1 (123.4)	20 to 200	95.6
Porter 1994	Non‐healthy, both sexes, 50 years and older (mean age 58 years old) with rheumatoid arthritis. Sens, spec, D+, D+ for the authors proposed threshold Table of sens, and spec given for other 6 alternative thresholds. SF_th=75.	45+4=49	29 (33)		7+45=52	95 (37)		66
Ravindran 2008	Non‐healthy, both sexes, mixed age (mean age 56 years old) with rheumatoid arthritis. Sens, spec, D+, D+, LR for the authors proposed threshold (among more than one threshold mentioned). SF_th=82.	17+1=18	57.9 (20.1)		0+32=32	201.5 (104.6)		143.6	iron‐depleted r = ‐0.09 (P = 0.57) iron‐repleted r = 0.79 (P < 0.0001)
Smith 1977	Non‐healthy, both sexes, mixed age (mean age 55.3 years old) with rheumatoid arthritis. Figure, and D+, D‐, for the authors unique proposed threshold. SF_th=100.	21+0=21	43.6		0+14=14	333		289.4	"poor correlation"
adults: this group was composed of adults excluding those studies focused exclusively on pregnant women. *Iron‐depleted sample size (n₁) could be different to the sum of TP + FN because the studies generally reported the numbers separately (some reasons were exclusion of values, subgroups etc.). Similarly, for iron‐repleted sample size (n₂) and the corresponding sum FP + TN.

Studies	Study characteristics: target pospulation / data pecifications / serum ferritin used threshold (SF_th)	Bone marrow equal to 0 group: BM = 0 (or as defined by author as iron‐depleted or diseased)			Bone marrow 1 or more group: BM = 1+ (or as defined by author as iron‐repleted or non‐diseased)			BM = 0 vs BM = 1+ comparison	Serum ferritin vs BM association
Studies		TP+FN=n₁*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	FP+TN=n₂*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	Mean difference (μg/L) in serum ferritin concentration: iron‐repleted group mean minus the iron‐depleted group mean	Pearson's correlation (r), Regression coefficient R², P‐value
Balaban 1993	Non‐healthy, males, 50 and older or mixed with several diseases, mainly malignancies (carcinoma, lymphoma, myeloma), and with chronic infections around 10% of the total. Only a few of the total were anaemic too. Sens, Spec, PPV, NPV, Figure, and D+, D‐ for the authors unique proposed threshold. SF_th=70.	17+11=28	63.7	22 to 258	9+83=92	212	31 to 2000	148.3	r = 0.58 (P < 0.001)
Intragumtornchai 1998	Non‐healthy, both sexes, mixed age (> 15 years) with alcoholism or liver cirrhosis or renal failure. Figure, and D+, D‐ for the authors unique proposed threshold. SF_th=50.	15+13=28	169.6 (324.4)		1+41=42	682.1 (508.6)		512.5	significantly correlated (P < 0.00007)
Isa 1988	Non‐healthy (mixed with healthy), both sexes, mixed age (30 to 69 years old) with alcoholism, liver cirrhosis or renal failure. Figure, and D+, D‐ for the authors unique proposed threshold. SF_th=50.	0+8=8	367.5		0+34=34	469		101.5
Krause 1980	Non‐healthy, both sexes, mixed age with diverse chronic diseases: iron deficiency, liver/renal disease, malignancies, and chronic inflammation. Figure, and D+, D‐ for the authors unique proposed threshold. SF_th=20.	87+17=104	17.7 (32.9)	2 to 165; median: 7	0+89=89	90.0 (140.6)	50 to 450; median:85	72.4
Kalantar‐Zadeh 1995	Non‐healthy, both sexes, mixed age (44 to 84 years) with alcoholism or liver cirrhosis or renal failure. Sens, Spec, figure, and D+, D‐ for the authors unique proposed threshold. SF_th=200.	5+5=10	83.0 (9)		2+13=15	795.0		712	iron‐repleted significantly correlated (spearman rank)
Milman 1983	Non‐healthy (mixed with healthy), both sexes, mixed age with alcoholism, liver cirrhosis or renal failure. PPV, NPV, D+, and D‐ for the authors proposed threshold (among more than one threshold mentioned). SF_th=60.	14+1=15	23.7		1+34=35	158.2		134.5	significantly correlated (P < 0.001)
Nelson 1978	Non‐healthy, males, mixed age with chronic inflammatory conditions including alcohol‐related conditions, liver disease, and malignancies. Figure, and D+, D‐ for the authors unique proposed threshold. SF_th=30.	11+1~11**	37.2 (85.5)	4 to 280; median: 9	1+60=61	341.7 (324.2)	20 to 1500; median:260	304.5	r = 0.75 (P < 0.00005)
Adults: this group was composed of adults excluding those studies focused exclusively on pregnant women. *Iron‐depleted sample size (n₁) could be different to the sum of TP + FN because the studies generally reported the numbers separately (some reasons were exclusion of values, subgroups etc.). Similarly, for iron‐repleted sample size (n₂) and the corresponding sum FP + TN.

Disease/Condition	Mean serum ferritin for iron‐deficient individuals (μg/L)	Mean serum ferritin for non‐iron‐deficient individuals (μg/L)	Means difference (μg/L)	Studies sample size (with available mean serum ferritin values for iron deficiency and non‐iron‐deficient groups)	Sample size (n) for iron‐deficient individuals	Sample size (n) for non‐iron‐deficiency individuals
Non‐healthy *adults with blood disorders	158.3	510.3	352	3	41	85
Non‐healthy *adults with rheumatoid arthritis iron deficiency	33.6	168.9	135.3	9	182	184
Non‐healthy *adults with Infectious diseases	17.7	91.9	74.2	1	39	16
Non‐healthy *adults with alcoholism or liver cirrhosis or renal failure or malignancies	108.9	392.6	283.7	7	205	368
Non‐healthy *adults (older adults population mean age > 62 years old) with diverse chronic diseases, mixed or unknown	38.9	412.8	373.9	5	135	166
Non‐healthy *adults (50 years or older or mean age > 50 years, but not older persons) with diverse chronic diseases, mixed or unknown	53.5	313.1	259.5	6	227	210
Non‐healthy *adults with diverse chronic diseases, or mixed/unknown	150.9	633.3	482.4	7	208	528
All non‐healthy adults	66.871	317.195	250.3	37	1037	1557
All healthy adults	15.167	70.367	55.2	3	46	70
*Adults: this group was composed of adults excluding those studies focused exclusively on pregnant women.

Studies	Study characteristics: target population / data specifications / serum ferritin used threshold (SF_th) / Iron‐overload threshold mg/g dry liver weight (dLw) or Perl's staining (Ps) grade: 0, 1, etc.	Non‐iron‐overload group: Non‐iron‐overloaded (liver iron concentration equal or lower than author's threshold, or Perl's staining author's threshold)			Iron‐overload group: IO (liver iron concentration higher than author's threshold, or Perl's staining author's threshold)			Non‐IO vs IO comparison	Serum ferritin vs reference standard association
Studies		FP+TN=n₁*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	TP+FN=n₂*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	Mean difference (μg/L) in serum ferritin concentration: iron‐overload group mean minus the non‐iron‐overload group mean	Pearson's correlation (r), Regression coefficient R², P‐value
**Halliday 1977	Non‐healthy patients within families with hereditary haemochromatosis (mean age: 41.1 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors proposed threshold among more than 1 threshold. SF_th=500. dLw=1.8.	2+3=5	347 (317)	40 to 709; median:180	4+8=12	597.1 (551.2)	56 to 1460; median:312	250.1	r = 0.63
Holmström 2002	Non‐healthy patients with hereditary haemochromatosis (mean age: 47.7 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=500. Ps=Grade 1 or more.	0+1=1	364		3+3=6	411.2 (202.6)	205 to 671; median:393.5	47.2
**Jensen 1994	Non‐healthy patients with hereditary haemochromatosis (26 to 52 years old). Patient's data extracted from figure to compute TP, FN, FP, and TN for the authors proposed threshold Three alternative thresholds were available too. SF_th=500. dLw=3.2.	2+1=3	1241.6 (1279.1)	310 to 2700; median:715	11+4=15	2604.7 (2241.7)	226 to 6560; median:1430	1363.1	r = 0.64
**Kaltwasser 1990	Non‐healthy patients with hereditary haemochromatosis (aged between 22 and 70 with mean 39 years old). Patient's data extracted from table and figure to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=500. dLw=1.8.	0+2=2	37	median:37	6+2=8	1480 (1643)	57 to 5133; median:934.5	1443	r = 0.85
Lawrence 1996	Non‐healthy patients with hereditary haemochromatosis within an army medical centre (aged between 26 and 70 with mean 55 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=500. dLw=1.8.	0+0=0			9+1=10	1542.6 (1019.2)	807 to 3635; median:1047.5		r = 0.08
Leggett 1990	Non‐healthy patients with hereditary haemochromatosis selected from one banking and one insurance company (large corporations) (aged between 21 and 45 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=300. dLw=2.	0+1=1	215		9+1=10	778.2 (411.06)	237 to 1550; median:690	563.2	r = 0.357
**Lim 2004	Non‐healthy patients with hereditary haemochromatosis (aged between 27 and 75 with mean 51 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors proposed threshold (among more than one threshold mentioned). SF_th=500. dLw=1.8	1+2=3	357 (257.0)	123 to 632; median:316	10+5=15	1121.2 (1347.7)	209 to 5730; median:8066	764.2	"poor correlation" r = 0.286
**Lombard 1989	Non‐healthy patients with hereditary haemochromatosis (aged between 27 and 73 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=500. Ps=Grade 2 or more.	0+9=9	54.4 (58.4)	9 to 200; median:36	6+5=11	1080.6 (1008.6)	70 to 3250; median:560	1026.2	r = 0.744
Macfarlane 1995	Non‐healthy patients with idiopathic haemochromatosis (mean age 50 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=500. dLw=1.8	0+0=0			7+1=8	1263.9 (895.3)	330 to 3069; median:987.5		"poor correlation" r = 0.560
Olynyk 1999	Non‐healthy patients with hereditary haemochromatosis (aged between 35 and 74 with mean 55.8 years old). Patient's data extracted from figure to compute TP, FN, FP, and TN for the authors proposed threshold Three alternative thresholds were available too. SF_th=150. dLw=3.2	0+0=0			7+0=7	914.1 (678.7)	187 to 2290; median:731		r = 0.944
Ortega 2005	Non‐healthy patients with haemochromatosis (mean age 50.48 years old). Patient's data extracted from figure to compute TP, FN, FP, and TN for the authors proposed threshold. Three alternative thresholds were available too. SF_th=500. dLw=2.	21+1=22	1131.3 (588.4)	500 to 2600; median:910	23+1=24	1414.6 (723.5)	500 to 3000; median:1050	283.3	r = 0.34
Phatak 1998	Non‐healthy patients with haemochromatosis (aged between 23 and 75). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold (among more than one threshold mentioned). SF_th=200. dLw=1.8.	0+0=0			25+0=25	810.9 (542.4)	213 to 2160; median:583		r = 0.353
Pietrangelo 1999	Non‐healthy patients, and their relatives with haemochromatosis (aged between 20 and 85 years). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors proposed threshold (among more than one threshold mentioned). SF_th=500. dLw=1.8.	0+0=0			13+0=13	2671.2 (1913.5)	650 to 5846; median:2165		r = 0.490
Rowe 1977	Mixed individuals (relatives of a dead serious case of IO) and their relatives with haemochromatosis (aged over 14 years old). Patient's data extracted from table, and figure to compute TP, FN, FP, and TN for the authors proposed threshold (among more than one threshold mentioned). SF_th=200. Ps=Grade 1 or more	0+10=10	21.14 (13.5)	8.1 to 45; median:16.1	1+18=19	72.8 (43.4)	22 to 202; median:57.6	51.66	r = 0.38
Schöniger‐Hekele 2002	Non‐healthy patients with hereditary haemochromatosis (mean age 48.3 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors proposed threshold (among more than one threshold mentioned). SF_th=200. Ps=Grade 1 or more.	0+3=3	40.3 (29.5)	10 to 69; median:42	8+1=9	738 (616.8)	110 to 1970; median:573	697.7	r = 0.701
Sham 1997	Non‐healthy patients with hereditary haemochromatosis (between 28 and 80 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors proposed threshold among more than 1 threshold. SF_th=300. dLw=1.8	3+0=3	1122.3 (419.3)	640 to 1400; median:1327	31+2=33	802.9 (477.1)	248 to 2246; median:634	‐319.4	r = 0.46
Smith 1997	Non‐healthy patients with hereditary haemochromatosis (mean age 46 with range between 27 and 57 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=500. dLw=1.8.	1+3=4	550 (679.7)	41 to 1539; median:310	5+0=5	1187.4 (985.5)	638 to 2933; median:706	637.4	r = 0.025
Summers 1990	Non‐healthy patients with hereditary haemochromatosis (mean age 34 with range between 12 and 78 years old). Patient's data extracted from table, and figure to compute TP, FN, FP, and TN for the authors proposed threshold (among more than one threshold mentioned). SF_th=300. dLw=1.8.	0+3=3	173		3+0=3	653.2 (721.4)	60 to 1856; median:319	480.2	r = 0.471
Thomsen 1992	Non‐healthy blood donors with haemochromatosis (aged between 20 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=500. dLw=1.8.	1+4=5	402.6 (88.3)	300 to 513; median:382	7+4=11	617.6 (226.4)	316 to 1031; median:599	215	r = 0.700 (P < 0.01)
Wands 1976	Non‐healthy patients with hereditary haemochromatosis (mean age 34.2 with range between 19 and 67 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=200. Ps=Grade 2 or more.	1+4=5	87.6 (69.7)	23 to 202; median:55.5	0+13=13	70.9 (29.8)	28 to 130; median:75.5	‐16.7	r = 0
Non‐iron‐overload sample size (n₁) could be different to the sum of FP + TN because the studies generally reported the numbers separately (some reasons were exclusion of values, subgroups etc.). Similarly, for iron‐overload sample size (n₂) and the sum TP + FN. *Studies that had two proposed thresholds: one for both sexes, and another one for each sex (males, females). Here we present the data as one threshold for both sexes.

Studies	Study characteristics: target population / data specifications / serum ferritin used threshold (SF_th) / Iron‐overload threshold mg/g dry liver weight (dLw) or Perl's staining (Ps) grade: 0, 1, etc.	Non‐iron‐overload group: Non‐IO (liver iron concentration equal or lower than author's threshold, or Perl's staining author's threshold)			Iron‐overload group: IO (liver iron concentration higher than author's threshold, or Perl's staining author's threshold)			Non‐IO vs IO comparison	Serum ferritin vs reference standard association
Studies		FP+TN=n₁*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	TP+FN=n₂*	Mean (Standard deviation) serum ferritin (μg/L)	Range value for serum ferritin; other measures (μg/L)	Mean difference (μg/L) in serum ferritin concentration: iron‐overload group mean minus the non‐iron‐overload group mean and/or OR**	Pearson's correlation (r), Regression coefficient R², P‐value
Chapman 1982	Non‐healthy patients alcoholics with hepatitis and/or cirrhosis (mean age 44.7 years old). Patient's data extracted from figure to compute TP, FN, FP, and TN for the authors proposed threshold. Three alternative thresholds were available too. SF_th=500. dLw=1.8.	13+28=41	509.5 (387.5)	50 to 1800; median:387	9+4=13	785.4 (391.6)	50 to 1800; median:700	275.9	r = 0.28
Guyader 2007	Non‐healthy patients with chronic hepatitis C (mean age 41.6 years old). D+, D‐, TP, TN for the authors unique proposed threshold. SF_th=300 males and 250 females. Ps=Grade 1 or more.	85+378=463	130		73+50=123	370	median:406	240	r = 0.461 (P < 0.0001)
Pascoe 1999	Non‐healthy patients with alcoholic liver disease (aged between 32 and 66). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors proposed threshold (among more than one threshold mentioned). SF_th=200. dLw=1.8.	9+12=21	264.9 (308.5)	24 to 1090; median:83	16+0=16	838.8 (720.6)	230 to 2980; median:570	573.91	r = 0.53
Sebastiani 2006	Non‐healthy patients with chronic hepatitis C (mean age 47.8). Sens, Spec, PPV, NPV, D+, D‐, and Table for the authors unique proposed threshold. SF_th=300 males and 200 females. Ps=Grade 2 or more.	45+171=216	124.7 (122.8)		17+9=26	346.4 (284.6)		221.7 and **OR = 13.2 (P < 0.00001)
Sebastinani 2012	Non‐healthy patients with hepatitis C, alcoholic liver disease (mean age 43.8). Sens, Spec, PPV, NPV, D+, D‐, and Table for the authors unique proposed threshold. SF_th=300 males, and 200 females. Ps=Grade1 or more.	19+114=133	135.7 (121.2)		22+50=72	280.3 (200.2)		144.6 and **OR = 4.3 (P < 0.008)
Szurowska 2010	Non‐healthy patients with liver cirrhosis (mean age 64 years with range between 22 and 81 years old). Patient's data extracted from figure to compute TP, FN, FP and TN for the authors proposed threshold. Three alternative thresholds were available. SF_th=400. Ps=Grade 1 or more.	1+14=15			10+8=18
Thorburn 2002	Non‐healthy patients with viral hepatitis infections (mean age 35 years old). Patient's data extracted from table to compute TP, FN, FP, and TN for the authors unique proposed threshold. SF_th=300. dLw=1.8.	1+2=3	153 (159.7)	6 to 323; median:130	1+1=2	334.5 (222.7)	177 to 492; median:334.5	181.5	r = 0.297
Non‐iron‐overload sample size (n₁) could be different to the sum of FP + TN because the studies generally reported the numbers separately (some reasons were exclusion of values, subgroups etc.). Similarly, for iron‐overload sample size (n₂) and the sum TP + FN. *Odds Ratio (OR)

State or disease/condition	Mean serum ferritin for non‐iron‐overload individuals (μg/L)	Mean serum ferritin for iron‐overload individuals (μg/L)	Means difference (μg/L)	Studies sample size (with available mean serum ferritin values for iron‐overload and non‐iron‐overload groups)	Sample size (n) for non‐iron‐overload individuals	Sample size (n) for iron‐overload individuals
Non‐healthy with haemochromatosis	409.6	908.7	499.1	15	78	192
Non‐healthy with chronic hepatitis, alcoholic cirrhosis, alcoholic liver disease, and hepatic fibrosis.	219.6	492.6	272.9	6	629	501
Non‐healthy mixed/unknown: hereditary haemochromatosis, alcoholic disease, chronic hepatitis, nonalcoholic fatty liver disease	1244.3	1671.3	427	4	11	24
All non‐healthy	497.6	930.8	433.3	25	718	717

PERMALINK

Serum or plasma ferritin concentration as an index of iron deficiency and overload

Maria Nieves Garcia-Casal

Sant-Rayn Pasricha

Ricardo X Martinez

Lucero Lopez-Perez

Juan Pablo Peña-Rosas

Abstract

Background

Objectives

Search methods

Selection criteria

Data collection and analysis

Main results

Authors' conclusions

Plain language summary

Summary of findings

Background

Target condition being diagnosed

Iron deficiency

Iron overload

Index test(s)

Clinical pathway

1.

2.

Prior tests

Role of index test(s)

Alternative test(s)

Iron indices

Haematologic indices

Other indices

Inflammatory indices

Liver function tests

Rationale

Objectives

Secondary objectives

Methods

Criteria for considering studies for this review

Types of studies

Iron deficiency

Iron overload

Participants

Iron deficiency

Iron overload

Specific exclusion criteria

Index tests

Target conditions

Reference standards

Iron deficiency

1. Bone marrow iron content (Perl's Prussian blue stain method).

2. Bone marrow iron content (other classification).

Iron overload

Liver biopsy

3. Interpretation of iron content of liver biopsies: histological grade of iron storage.

Search methods for identification of studies

Electronic searches

International databases

Regional databases

Searching other resources

Data collection and analysis

Selection of studies

Data extraction and management

Assessment of methodological quality

Statistical analysis and data synthesis

Definitions of key criteria

Descriptive plots

Handling of multiple thresholds

4. Relative extent of iron stores on the basis of serum ferritin concentration.

Investigations of heterogeneity

Sensitivity analyses

Assessment of reporting bias

Results

Results of the search

3.

Iron deficiency

Participants, and target populations

Covariates distribution in iron deficiency

Types of studies, and settings

Reference standards

Risk of iron overload

Studies	Study characteristics: target population	Non‐iron‐overload group: Non‐IO (liver iron concentration <= 3.2 µmol/100 mg dry liver weight or as defined by author)			Iron‐overload group: IO (liver iron concentration > 3.2 µmol/100 mg dry liver weight or as defined by author)			Non‐IO vs IO comparison	Serum ferritin vs reference standard association
*Cecchin 1990	Patients receiving haemodialysis and intravenous iron infusions, and patients with hereditary haemochromatosis (mean age 48 years old).	137.4 (95.82)	40 to 411; median: 131.5	18	1015.1 (509)	315 to 2120; median: 985	30	877.7	r = 0.9473
Hellerbrand 2003	Patients with hereditary haemochromatosis, or cirrhosis and/or pathogenesis of hepatocellular carcinoma (mean age: 61.2 years old).								r = 0.35
Hofer 2004	Patients with chronic hepatitis C (mean age: 40.8 years old).								r = 0.335
Karam 2008	38 sickle‐cell disease patients, and one with thalassaemia, all with chronic transfusion, 6 to 23 years old with mean of 12.7 years old.	940 (358)	515 to 1270; median: 987.5	4	3021.5 (1482)	975 to 6076; median: 2930	35	2081.5	r = 0.46
Kowdley 2012	Patients with nonalcoholic fatty liver disease with mean age of 40 years old.							***OR =2.2 95% CI = (1.78 to2.76)
**Kreeftenberg 2000	Patients at risk of iron overload with clinical suspicion and risk of haemochromatosis.								r = 0.759
Smith 2014	Sickle‐cell patients who underwent chronic transfusion.	292 (150.6)	64 to 480; median: 315	7	3542.9 (2573.30)	430 to 13576; median: 3150	252	3250.9	r = 0.74
Iron‐overload threshold = liver iron concentration > 3.6 µmol/100 mg dry liver weight. Iron‐overload threshold = liver iron concentration > 7 µmol/100 mg dry liver weight. **Odds Ratio (OR).

Test	No. of studies	No. of participants
1 Iron deficiency apparently healthy (All)	5	350
2 Iron deficiency apparently healthy (*adults)	3	247
3 Iron deficiency non‐healthy (All)	70	5709
4 Iron deficiency non‐healthy (infants and children)	4	519
5 Iron deficiency non‐healthy (*adults)	63	5042
6 Iron deficiency non‐healthy (adults, pregnant women)	3	148
7 Iron overload (All)	36	1927
8 Iron overload (All: threshold per gender: males/females)	9	1292

*Study characteristics*
Patient Sampling	The study was undertaken as part of a case‐control study on the aetiology and risk factors of anaemia in children less than 5 years of age. Consecutive children aged 1 to 59 months, attending the Manhiça District Hospital (MDH) emergency department between October 2008 to August 2010 with anaemia (haemoglobin lower than 110 g/L), and with no history of blood transfusion in the preceding 4 weeks were recruited (n = 443). A subsample of 173 had their parents consent, and completed their corresponding bone marrow and serum ferritin measurements.
Patient characteristics and setting	The study was carried out at the Centro de Investigacao em Saude de Manhica (CISM) in Manhica District, southern Mozambique, an area where malaria transmission is of moderate intensity most of the year with some seasonality. More than 95% of the malaria infections are due to Plasmodium falciparum. Adjacent to the CISM is the Manhica District Hospital (MDH), a 110‐bed health facility. The main causes of hospital attendance and admission among children in the area are pneumonia, malaria, anaemia, malnutrition and HIV‐related diseases. HIV prevalence in pregnant women was 29% in 2010. Patients who entered the study were anaemic infants of both sexes, and aged between 6 to 59 months with several infectious diseases (malaria, tuberculosis, HIV, and other), attending the MDH emergency department in a rural region, southern Mozambique.
Index tests	Ferritin was measured in an enzyme immunoassay ADVIA Centaur analyser (Siemens Healthcare, Barcelona, Spain). Serum ferritin concentration lower than 30 µg/L was classified as iron deficiency.
Target condition and reference standard(s)	Bone marrow iron content was semi‐quantitatively estimated classifying the amount of blue stained haemosiderin Perls in bone marrow fragments (aggregates of bone marrow cells) according to 4 categories: 0 (absent), 1 (diminished), 2 (normal) and 3 (abundant). The categories 0 and 1 were considered indicative of iron store deficiency.
Flow and timing	A bone marrow aspiration was performed after complete clinical examination was performed and the information was entered onto standardised questionnaires together with demographic data. Venous blood was collected by venipuncture for malaria parasitaemia examination, bacterial culture, full blood count and biochemical and molecular determinations. However, the flow and timing of these three types of tests was not described.
Comparative
Notes	Ferritin and other biomarkers of iron status were compared to bone marrow findings in 180 anaemic children attending a rural hospital in southern Mozambique. Eighty per cent (144/180) of the children had iron deficiency by bone marrow examination, 88% (155/176) had an inflammatory process, 66% (119/180) had moderate anaemia, 25% (45/180) severe anaemia, and 9% (16/180) very severe anaemia. The findings of this study showed that the majority (80%) of the anaemic children were iron deficient by direct assessment of iron stores, and that sTfR and TfR‐F index adjusted by CRP are the most sensitive markers with specificities of at least 50% to identify ID in this study population. However, even with these markers, 17% and 25% of children, respectively, will not be diagnosed with ID and, therefore, adequately treated.
*Methodological quality*
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient Selection
Was a consecutive or random sample of patients enrolled?	Unclear
Was a case‐control design avoided?	Yes
Did the study avoid inappropriate exclusions?	No
Could the selection of patients have introduced bias?		Unclear risk
Are there concerns that the included patients and setting do not match the review question?			Low concern
DOMAIN 2: Index Test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	Yes
If a threshold was used, was it pre‐specified?	Yes
Could the conduct or interpretation of the index test have introduced bias?		Low risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			Low concern
DOMAIN 3: Reference Standard
Is the reference standards likely to correctly classify the target condition?	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	Unclear
Could the reference standard, its conduct, or its interpretation have introduced bias?		Low risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			Low concern
DOMAIN 4: Flow and Timing
Was there an appropriate interval between index test and reference standard?	Yes
Did all patients receive the same reference standard?	Yes
Were all patients included in the analysis?	Yes
Could the patient flow have introduced bias?		Low risk

*Study characteristics*
Patient Sampling	248 adequate bone marrow specimens sequentially collected over a 18‐month period from hospital patients with pathologies representative of common pathologies in a general hospital in Canada
Patient characteristics and setting	The specimens were from hospital patients of both sexes, mixed ages, and with clinical diagnosis ranging from thrombocytopenia, liver disease, hypochromic anaemia, renal disease, myeloma, macrocytic anaemia, collagen disease, pyrexia of unknown origin, myeloproliferative disorder, acute leukaemia and lymphoproliferative disorder. The specimens came from St. Joseph's General Hospital in Hamilton, Ontario, Canada.
Index tests	Serum ferritin concentrations were measured by radioimmunoassay using ferritin labelled with iodine‐125 and rabbit anti‐ferritin antibody. Goat anti‐rabbit y‐globulin antibody and polyethylene glycol were used as separating agents. The working range of the method used was up to 500 µg ferritin per litre and required a sample of 75 µL of serum or plasma for the assay at two dilutions. Serum ferritin concentration lower than 12 µg/L was classified as iron deficiency.
Target condition and reference standard(s)	Air dried films of bone marrow aspirates were fixed in methanol for 10 to 20 minutes, then stained by May‐Grimwald‐Giemsa's stain for morphologic study and by the Prussian blue reaction for evaluation of iron stores. Iron stores were graded as absent, present, or increased by one observer who was not aware of the serum ferritin value. Iron was considered to be absent when no iron could be demonstrated in any of the bone marrow fragments and increased when every fragment contained iron deposits covering more than 50% of the area of the fragment. Grading of the iron content between these two extremes was not attempted. The absence of iron in BM was classified as iron deficiency.
Flow and timing	Haematological and biochemical analyses (such as serum ferritin), bone marrow aspiration measurements, and reviews of medical records were made. However, the flow and time of these three types of tests was not described. However, it was noted that blood samples were collected at the time of the bone marrow aspiration.
Comparative
Notes	Serum ferritin prospectively correlated with bone marrow iron stores in a mixed hospital population. 248 adequate bone marrow specimens from patients with pathologies representative of the clinical and haematologic problems likely to be encountered in a general hospital, were sent to St. Joseph's Hospital's haematology laboratory for evaluation over 18 months. Of the 248 patients, 69 were found to have no iron in the bone marrow, 116 had normal iron stores and 63 had increased iron stores. Of the 69 patients with no iron in the bone marrow, 20 had a serum ferritin value of more than 12 µg/L. Of these 20 patients, 2 were receiving iron therapy and 2 had received a blood transfusion, 6 had alcoholic liver disease, 2 had chronic active hepatitis, 5 had renal failure due to chronic glomerulonephritis, 2 had active Hodgkin's disease and 1 had multiple myeloma. All the patients with normal or increased iron stores had a serum ferritin value of more than 12 µg/L. The data reported in this study confirm the usefulness of taking measurements. Therefore, a low serum ferritin value probably indicates iron depletion, while an elevated value does not exclude that possibility.
*Methodological quality*
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient Selection
Was a consecutive or random sample of patients enrolled?	Yes
Was a case‐control design avoided?	Yes
Did the study avoid inappropriate exclusions?	Yes
Could the selection of patients have introduced bias?		Low risk
Are there concerns that the included patients and setting do not match the review question?			Low concern
DOMAIN 2: Index Test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	Unclear
If a threshold was used, was it pre‐specified?	Yes
Could the conduct or interpretation of the index test have introduced bias?		Low risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			Low concern
DOMAIN 3: Reference Standard
Is the reference standards likely to correctly classify the target condition?	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	Yes
Could the reference standard, its conduct, or its interpretation have introduced bias?		Low risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			Low concern
DOMAIN 4: Flow and Timing
Was there an appropriate interval between index test and reference standard?	Yes
Did all patients receive the same reference standard?	Yes
Were all patients included in the analysis?	Yes
Could the patient flow have introduced bias?		Low risk

*Study characteristics*
Patient Sampling	120 patients with iron‐deficiency anaemia (n =40), anaemia of chronic disease (n = 40) and anaemic rheumatoid arthritis patients (n = 40) were selected for study.
Patient characteristics and setting	Iron‐deficiency anaemia (n = 40) was diagnosed when the ferritin level was < 12 µg/L. The patients with chronic disease anaemia (n = 40) and rheumatoid arthritis had at least a 6‐month history of chronic illness with elevated erythrocyte sedimentation rates to confirm inflammation. The 40 patients with rheumatoid arthritis were selected as a model group of patients with an inflammatory condition. The study was conducted at the Department of Haematology, Southern General Hospital, Glasgow, Scotland. As a source of bone marrow, aspirates were collected from 18 patients with rheumatoid arthritis undergoing total hip replacement at the Southern General Hospital and Victoria Infirmary, Glasgow, Scotland.
Index tests	Serum ferritin levels were performed using the Access® Immunoassay System analyser (Beckman Coulter Inc, Chaska, USA).
Target condition and reference standard(s)	Bone marrow aspirations were assessed as the gold standard in 20 of the patients with chronic disease anaemia. Seven patients with chronic disease anaemia had no stainable marrow iron [mean sTfR 7.43 mg/L (3.26–12.23), mean ferritin 112 µg/L (26–261)]. Of the seven patients with iron‐deficiency anaemia, six of these had elevated sTfR concentrations (> 3.3 mg/L), and one had a sTfR concentration in the upper normal range (sTfR higher than 3.26 mg/L). Thirteen patients with anaemia of chronic disease had marrow iron present [mean sTfR 3.41 mg/L (3.26–12.23), mean ferritin 209 µg/L (50–1500)]. The marrow smears were examined for the presence of iron using Perl’s Prussian blue stain. Iron stores were graded as 0 to 6+, grades of 1+ to 3+ being considered normal. The normal control range (n = 40) was generated using a group of 10 healthy volunteers and 30 non‐anaemic patients.
Flow and timing	Over a period of 5 months, patients had haematological measurements, including Hb, serum ferritin and sTfR along with other diagnostic methods including bone marrow aspirates. An elevated erythrocyte sedimentation rate of at least 6 month's duration was used to confirm inflammation.
Comparative	The aim of the study was to evaluate the usefulness of the IDeA sTfR immunoenzymometric assay (Orion Diagnostica, Espoo, Finland) in the differential diagnosis between iron‐deficiency anaemia and anaemia of chronic disease. The sensitivity, specificity and efficiency of the parameters at detecting iron deficiency when compared with bone marrow iron stains in 20 patients with anaemia of chronic disease and at detecting iron deficiency when compared with bone marrow iron stains in 18 patients with rheumatoid arthritis.
Notes	The study had ethical approval and all data collected followed the standard practice for confidentiality.
*Methodological quality*
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient Selection
Was a consecutive or random sample of patients enrolled?	Unclear
Was a case‐control design avoided?	Yes
Did the study avoid inappropriate exclusions?	No
Could the selection of patients have introduced bias?		Unclear risk
Are there concerns that the included patients and setting do not match the review question?			Low concern
DOMAIN 2: Index Test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	Unclear
If a threshold was used, was it pre‐specified?	Yes
Could the conduct or interpretation of the index test have introduced bias?		Low risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			Unclear
DOMAIN 3: Reference Standard
Is the reference standards likely to correctly classify the target condition?	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	Unclear
Could the reference standard, its conduct, or its interpretation have introduced bias?		Unclear risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			Low concern
DOMAIN 4: Flow and Timing
Was there an appropriate interval between index test and reference standard?	Yes
Did all patients receive the same reference standard?	Unclear
Were all patients included in the analysis?	No
Could the patient flow have introduced bias?		Unclear risk

*Study characteristics*
Patient Sampling	120 anaemic (Hb < 140 g/L) male patients with diagnosis of neoplasm inflammation, infection, or hepatic disease in Dallas, Texas, USA. No more details about the sampling procedure were described.
Patient characteristics and setting	Hospitalised male patients aged 50 years or older with several diseases, mainly malignancies (carcinoma, lymphoma, myeloma), and chronic infections (fewer than 10%)
Index tests	Serum ferritin (SERFER) was determined by radioimmunoassay (RAMCO, Houston, TX). Red blood cell ferritin (RBCFER) was determined by centrifuging blood samples to obtain the packed red cell fraction and diluting to the original volume with 0.9% saline. Red cell suspensions were then prepared by isotonic/percoll gradient density separation and ferritin determined by the same radioimmunoassay. A serum ferritin concentration lower than 70 µg/L was classified as iron deficiency.
Target condition and reference standard(s)	Bone marrow examinations were obtained by iliac crest trephine aspiration and biopsy and evaluated by acid ferrocyanide (Prussian Blue) staining of aspirate smears. Zero iron was defined as no identified iron staining after examination under oil immersion (X 100). Trace staining was defined as iron stain seen only after examination under oil immersion (X 100). One to 4+ staining was based on the observed intensity of iron stain after examination under low power lens (X 40), with 1+ being less intense and 4+ being the most intense. A marrow was considered to have decreased iron stores (iron deficiency) if the averaged score was zero or trace.
Flow and timing	Haematological and biochemical analyses (such as serum ferritin), bone marrow aspiration measurements, and reviews of medical records reviewing were made. However, the flow and timing of these three types of tests were not described.
Comparative
Notes	120 anaemic (Hb < 140 g/L) male patients, mainly with a diagnosis of neoplasm inflammation, infection, or hepatic disease. The correlation between SERFER and bone marrow iron staining was r = 0.58 (P < 0.001). This correlation was better than that between RBCFER and bone marrow iron content (r = 0.36; P < 0.001). The potential to predict an iron deficient state (positive predictive value) with this combination (0.82) was better than the positive predictive value of having either a low SERFER or RBCFER. The overall accuracy of knowing both the SERFER and RBCFER (0.94) was better than the accuracy of either test alone.
*Methodological quality*
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient Selection
Was a consecutive or random sample of patients enrolled?	No
Was a case‐control design avoided?	No
Did the study avoid inappropriate exclusions?	Unclear
Could the selection of patients have introduced bias?		High risk
Are there concerns that the included patients and setting do not match the review question?			Low concern
DOMAIN 2: Index Test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	Unclear
If a threshold was used, was it pre‐specified?	No
Could the conduct or interpretation of the index test have introduced bias?		Unclear risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			Low concern
DOMAIN 3: Reference Standard
Is the reference standards likely to correctly classify the target condition?	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	Unclear
Could the reference standard, its conduct, or its interpretation have introduced bias?		Low risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			Low concern
DOMAIN 4: Flow and Timing
Was there an appropriate interval between index test and reference standard?	Unclear
Did all patients receive the same reference standard?	Yes
Were all patients included in the analysis?	Yes
Could the patient flow have introduced bias?		Low risk

*Study characteristics*
Patient Sampling	Of 108 consecutive bone marrow specimens that had a report of absent bone marrow hemosiderin and that were re‐reviewed, 19 (18%) did not allow an accurate assessment of iron stores and hence were deemed inadequate. An additional 15 specimens (14%) revealed decreased or increased, but not absent, stainable iron and were also excluded from further analysis. Therefore, only 74 specimens (69%) were thought to be accurately reported. Of these, only 37 had adequate laboratory and historical information. Two of them were excluded and two groups of 17 and 18, respectively, were made. No further details about the sampling procedure were described.
Patient characteristics and setting	35 patients with blood disorders underwent bone marrow aspirations and blood tests at the Mayo Clinic, Rochester, Minnesota, United States. They were classified in two groups: group A (clinically not consistent with iron deficiency) and group B (clinically consistent with iron deficiency). Group A consisted of 17 patients (median age: 65 years; 12 of them female) and group B of 18 patients (median age: 60 years; 8 of them female). Group A had lymphomas, leukaemia, myelomas, myeloproliferative disorders, and inflammatory disorders, etc. and group B presented mainly with bleeding.
Index tests	Serum ferritin concentrations (method not specified). Normal ranges: ferritin 20–300 μg/L. Serum ferritin concentration lower than 100 µg/L was classified as iron deficiency.
Target condition and reference standard(s)	Bone marrow hemosiderin was measured on the amount of stainable iron in the bone marrow. Iron stains (Prussian blue) were performed on one slide. For each batch of iron studies examined, a quality control of the reagents was performed on a smear previously read as showing increased iron stores. The slides were fixed with methyl alcohol for 1 min, decanted, and allowed to air‐dry. Afterward, they were placed into a Coplin jar containing one part (2%) hydrochloric acid to two parts potassium ferrocyanide, followed by incubation in an incubator at 37°C for 20 mins. After being washed with distilled water, the slides were placed into a Coplin jar and counterstained with Nuclear Fast Red for 15 mins. The slide was then washed with distilled water, fan‐dried, and examined by a pathologist. All the test slides in the current study were re‐reviewed by one of the authors. Patients in whom the absence of bone marrow hemosiderin was confirmed (iron deficient) were classified into two groups on the basis of an extensive historical and laboratory review.
Flow and timing	It was unclear. A review of the pathologic reports revealed 19 inadequate specimens and 15 with decreased, but not absent, iron stores. Thus, only 74 of the 108 cases had been accurately reported. In 35 of these cases, adequate clinical and laboratory data were available and allowed further analysis of two groups, divided according to the participants' clinical history. Haematological and biochemical analyses (such as serum ferritin), bone marrow aspiration measurements, and reviews of medical records were made. However, the flow and timing of these three types of tests were not described.
Comparative
Notes	We suspected the inaccuracy of bone marrow hemosiderin staining in the assessment for IDA in an unselected group of patients with suspected haematologic disease. In addition to potential inaccuracies attributable to methodology, we discovered that visual inspection for bone marrow hemosiderin by experienced haematopathologists was not always reproducible. In the current study, only 69% of the reports were believed to be accurate on re‐review. An interobserver difference of 31% is unacceptably high for a supposed 'gold standard', and further consequences in patient convenience and cost may not be trivial. Furthermore, subclassification of the study population according to serum ferritin concentration revealed seven patients (20%) with values above 100 μg/L, which confirmed the clinical evidence that IDA was doubtful. At the same time, the finding that 85% of the patients with low serum ferritin were believed clinically to have IDA suggests the potential superiority of measuring serum ferritin over assessing bone marrow hemosiderin in detecting IDA in certain circumstances.
*Methodological quality*
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient Selection
Was a consecutive or random sample of patients enrolled?	No
Was a case‐control design avoided?	No
Did the study avoid inappropriate exclusions?	Unclear
Could the selection of patients have introduced bias?		High risk
Are there concerns that the included patients and setting do not match the review question?			High
DOMAIN 2: Index Test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	Unclear
If a threshold was used, was it pre‐specified?	Yes
Could the conduct or interpretation of the index test have introduced bias?		Low risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			High
DOMAIN 3: Reference Standard
Is the reference standards likely to correctly classify the target condition?	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	No
Could the reference standard, its conduct, or its interpretation have introduced bias?		High risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			High
DOMAIN 4: Flow and Timing
Was there an appropriate interval between index test and reference standard?	No
Did all patients receive the same reference standard?	Yes
Were all patients included in the analysis?	No
Could the patient flow have introduced bias?		High risk

*Study characteristics*
Patient Sampling	Prospective observational single‐centre study
Patient characteristics and setting	54 patients were selected from those admitted to a tertiary nephrology centre. Inclusion criteria were age over 18 years, chronic kidney disease of stages 3–5 with non‐dialysis, and anaemia (Hb under 110 g/L 2 times, 2 weeks apart). Patients with anaemia of a specific cause (e.g. vitamin B₁₂ or acid folic deficiency, haemolytic anaemias), acute inflammatory conditions, recent bleeding, neoplasm, active liver disease (transaminases 2 times normal), severe secondary hyperparathyroidism (PTH higher than 800 pg/mL), or previous iron and epoetin therapy were excluded. Participants had a median age of 64.5 (range from 30 to 90) years, 56% males, 19% diabetes mellitus (but none with diabetic glomerular nephropathy as cause of chronic kidney disease), 50% stage 5, 20% stage 4, and 30% stage 3 chronic kidney disease with moderate anaemia (mean 99 (95% CI 94–102) g/L).
Index tests	Serum ferritin and serum transferrin were measured by immuno turbidimetric methods (Good Biotech – Taiwan; Giesse Diagnostics – Italy) on an autoanalyzer Olympus® AU400.
Target condition and reference standard(s)	Bone marrow was collected by aspiration from the iliac crest. The smears were prepared from bone marrow aspirate stained with Perls’ Prussian blue. The occurrence of siderotic granules in macrophages was graded on a scale from 0 to 6. The percentage of sideroblasts, e.g. erythroblasts with green blue particles, was also calculated. The pattern of bone marrow iron distribution was classified as normal (macrophage iron 2–4, sideroblasts in a normal percentage), iron deficiency (macrophage iron 0 or 1, sideroblasts absent or present in a very low percentage), and anaemia of chronic conditions (macrophage iron ≥ 5, sideroblasts absent or present in a very low percentage). According to bone marrow iron distribution of the 54 non‐dialysis chronic kidney disease patients, with investigations of epoetin and iron‐naive, only 7 had normal iron distribution, 26 had iron deficiency and 21 had anaemia of chronic disorders.
Flow and timing	Bone marrow was first used to classify patients, then measurements were made of hepcidin and ferroportin expression, and peripheral parameters of iron status and inflammation. The flow and timing were not described in detail.
Comparative	Using bone marrow iron distribution, the patients were classified as having normal iron distribution, iron deficiency or anaemia of chronic disorders. Comparisons were made between iron‐deficiency anaemia and anaemia of chronic disorders groups.
Notes	The local Ethical Committee (of the ‘Dr Carol Davila’ Teaching Hospital of Nephrology, Bucharest, Romania) approved the study protocol. ANEMIAWORKING GROUP ‐ Romania supported the costs of some lab reagents and publishing.
*Methodological quality*
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient Selection
Was a consecutive or random sample of patients enrolled?	Yes
Was a case‐control design avoided?	Yes
Did the study avoid inappropriate exclusions?	Unclear
Could the selection of patients have introduced bias?		Unclear risk
Are there concerns that the included patients and setting do not match the review question?			Unclear
DOMAIN 2: Index Test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	No
If a threshold was used, was it pre‐specified?	No
Could the conduct or interpretation of the index test have introduced bias?		High risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			Unclear
DOMAIN 3: Reference Standard
Is the reference standards likely to correctly classify the target condition?	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	Yes
Could the reference standard, its conduct, or its interpretation have introduced bias?		Low risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			Low concern
DOMAIN 4: Flow and Timing
Was there an appropriate interval between index test and reference standard?	No
Did all patients receive the same reference standard?	Yes
Were all patients included in the analysis?	Yes
Could the patient flow have introduced bias?		High risk

*Study characteristics*
Patient Sampling	Between December 1992 and August 1993, blood tests, bone marrow smears and trephine biopsies (posterior iliac crest) were prospectively collected from 46 anaemic (haemoglobin concentrations < 120 g/L for women and < 140 g/L for men) patients with chronic inflammatory rheumatic diseases in Zürich University Hospital. A subsample of 45 patients had available clinical history and serum ferritin measurements.
Patient characteristics and setting	45 anaemic patients (37 women, 8 men) with mean age of 57.5 years old (aged between 26 and 87 years old). 28 of them had rheumatoid arthritis according to the American Rheumatism Association criteria, nine had connective tissue disease (systemic lupus erythematosus, Sjogren syndrome, mixed connective tissue disease, polymyalgia rheumatica), seven had seronegative spondathropathies (psoriasis, Crohn’s disease and Reiter syndrome) and one woman was suffering from gout arthritis. All patients took NSAIDs and most of them were also on a second‐line drug (gold, methotrexate, sulphasalazine. d‐Penicillamin and chloroquine) with or without prednisone. No patient had received either iron supplementation or blood transfusions in the preceding 6 weeks. The study included patients from the Department of Rheumatology of the University Hospital of Zurich, Switzerland.
Index tests	Serum ferritin was measured by a radioimmunoassay (Bio‐Rad, Glattbrugg, Switzerland). Serum ferritin concentration lower than 30 µg/L was classified as iron deficiency.
Target condition and reference standard(s)	A May Gruewald Giemsa stain was carried out on four smears, an iron stain (Prussian blue) was used in addition to a nonspecific esterase (a‐naphthyl butyrate) reaction. The trephine biopsies were embedded in methacrylate, cut at 1 pm thickness and stained with May Griinwald Giemsa and ‘kemecht rot nach Romeis’. The smears as well as the biopsies were examined by three experienced observers and reviewed blindly by two of them. The iron stores in the smears were classified semi‐quantitatively from 0 to 4+. The iron grading of the biopsies was adapted from the smears. Zero and 0.5 (trace) iron was classified as iron deficiency and 1 to 4 as replete iron stores.
Flow and timing	Blood tests, bone marrow smears and trephine biopsies (posterior iliac crest) were prospectively collected from 45 patients. However, the flow and timing of these three types of tests were not described.
Comparative
Notes	45 anaemic patients (37 women, 8 men, mean age 57.5 years (range 26‐87)) with chronic inflammatory rheumatic diseases from the Department of Rheumatology of the University Hospital of Zurich, Switzerland. The combination of serum ferritin and CRP (as well as ESR) in its predictive capacity for bone marrow iron stores was examined. The relationship between iron in bone marrow and serum ferritin and CRP, transferrin, transferrin saturation, soluble transferrin receptor, erythrocyte porphyrins and percentage of hypochromic/microcytic erythrocytes was investigated. Stainable bone marrow iron was taken as standard to separate iron‐deficient from iron‐replete patients. 14 patients (31%) were lacking bone marrow iron. Regression analysis showed a good correlation between ferritin and bone marrow iron (adjusted R2 = 0.721, P < 0.0001). The combination of ferritin and CRP (ESR) did not improve the predictive power for bone marrow iron (adjusted R2 = 0.715) in this cohort of patients with low systemic inflammatory activity. With respect to the bone marrow iron content, the best predictive cut‐off value of ferritin was 30 µg/L (86% sensitivity, 90% specificity).
*Methodological quality*
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient Selection
Was a consecutive or random sample of patients enrolled?	No
Was a case‐control design avoided?	Yes
Did the study avoid inappropriate exclusions?	No
Could the selection of patients have introduced bias?		High risk
Are there concerns that the included patients and setting do not match the review question?			Low concern
DOMAIN 2: Index Test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	Unclear
If a threshold was used, was it pre‐specified?	No
Could the conduct or interpretation of the index test have introduced bias?		Unclear risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			Low concern
DOMAIN 3: Reference Standard
Is the reference standards likely to correctly classify the target condition?	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	Unclear
Could the reference standard, its conduct, or its interpretation have introduced bias?		Unclear risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			Low concern
DOMAIN 4: Flow and Timing
Was there an appropriate interval between index test and reference standard?	Yes
Did all patients receive the same reference standard?	Yes
Were all patients included in the analysis?	No
Could the patient flow have introduced bias?		Unclear risk

*Study characteristics*
Patient Sampling	1000 consecutive patients were referred to Tygerberg Hospital for bone marrow examination as part of the diagnostic evaluation of their disease between July 1978 and January 1981. The initial clinical diagnosis was obtained either from the hospital records or from the Haematology Clinic records at Tygerberg Hospital. A subsample of 400 patients had available bone marrow and serum ferritin data. No more details about the sampling procedure were given.
Patient characteristics and setting	1000 patients of mixed age with anaemic conditions, malignant lesions, infective conditions, or general systemic disorders, (285 white males, 227 white females, 260 non‐white males and 228 non‐white females), referred to the Tygerberg Hospital, in South Africa, underwent bone marrow examination and diagnostic tests. A subsample of 400 of them had available bone marrow and serum ferritin data.
Index tests	Serum ferritin evaluated with a Gamma Dab ferritin radioimmunoassay kit (Travenol Laboratories). The upper cut‐off point of the test is 500 µg/L. Serum ferritin concentration lower than 99 µg/L was classified as iron deficiency.
Target condition and reference standard(s)	Smears of the bone marrow aspirates were evaluated with a Prussian blue reaction, according to the method of Dacie. The bone marrow smears had values ranging from 0 to 6. The corresponding findings in the bone marrow were carried out on an IBM 370/158 computer. Participants graded with absent or traces of their corresponding bone marrow smears staining were classified as iron deficient (the specific grades were not specified).
Flow and timing	1000 consecutive patients were referred to Tygerberg Hospital for bone marrow examination as part of the diagnostic evaluation of their disease between July 1978 and January 1981. Haematological and biochemical analyses (such as serum ferritin), bone marrow aspiration measurements, and reviews of medical records were made. However, the flow and timing of these three types of tests were not described.
Comparative
Notes	Case series of bone marrow examinations in South African hospitals. When the ability of serum ferritin to predict bone marrow smears values was tested statistically, it was found that a good estimate could often be made, especially in iron deficiency. It was demonstrated that with the use of readily available clinical information and SF values, it is frequently unnecessary to perform bone marrow aspiration for the assessment of BMS only. The prediction of absence of BMS can be made with an extreme degree of statistical assurance when the serum ferritin level is below 99 µg/L.
*Methodological quality*
Item	Authors' judgement	Risk of bias	Applicability concerns
DOMAIN 1: Patient Selection
Was a consecutive or random sample of patients enrolled?	Unclear
Was a case‐control design avoided?	Unclear
Did the study avoid inappropriate exclusions?	No
Could the selection of patients have introduced bias?		High risk
Are there concerns that the included patients and setting do not match the review question?			High
DOMAIN 2: Index Test (All tests)
Were the index test results interpreted without knowledge of the results of the reference standard?	No
If a threshold was used, was it pre‐specified?	Yes
Could the conduct or interpretation of the index test have introduced bias?		High risk
Are there concerns that the index test, its conduct, or interpretation differ from the review question?			High
DOMAIN 3: Reference Standard
Is the reference standards likely to correctly classify the target condition?	Yes
Were the reference standard results interpreted without knowledge of the results of the index tests?	Unclear
Could the reference standard, its conduct, or its interpretation have introduced bias?		Low risk
Are there concerns that the target condition as defined by the reference standard does not match the question?			Unclear
DOMAIN 4: Flow and Timing
Was there an appropriate interval between index test and reference standard?	Unclear
Did all patients receive the same reference standard?	Yes
Were all patients included in the analysis?	Yes
Could the patient flow have introduced bias?		Unclear risk