Anemia is the result of insufficient production of healthy red blood cells in the body, which lowers the oxygen-carrying capacity of blood and has a significant impact on morbidity and mortality, especially for groups at high risk, including young children and pregnant persons (1). Anemia is a symptom of an underlying condition, not a diagnosis, but is an essential starting point for a more in-depth health evaluation (2, 3). For this reason, being able to accurately identify anemia is crucial for anyone treating patients. Despite its well-known health impacts, the same individuals are not necessarily identified as having anemia using various existing clinical practice guidelines.

Hemoglobin and hematocrit are the two most common biomarkers used in the clinical setting to identify anemia, and the two modalities are different in important ways. Hemoglobin is a molecule inside red blood cells; the laboratory test for hemoglobin is a direct measure of the number of hemoglobin molecules present in the blood. Hematocrit is the proportion of red blood cells to whole blood and can be determined by two primary methods: (a) a complete blood count in which an autoanalyzer determines the size and number of red blood cells, and (b) centrifugation in a tube where the spun red blood cells are measured in relation to the other components of whole blood (1, 4). Hematocrit can also be calculated by multiplying hemoglobin results by three or multiplying hemoglobin by the mean corpuscular hemoglobin concentration (4–6).

Clinical guidelines in the United States such as UpToDate (7) and institutional guidelines such as the Centers for Disease Control and Prevention’s (CDC) 1998 recommendations (8) consider both hemoglobin and hematocrit as acceptable measures of anemia, but this recommendation is sometimes questioned, and authors have noted pros and cons of each. Ali-Baya et al. and Keen both note how external factors can influence hematocrit measurements, while Graitcer et al. and Nguyen et al. describe the benefit of direct measurement and better correlation to anemia diagnosis with the use of hemoglobin (9–12). Brunken et al. suggest hematocrit as the preferred measure due to its cost and simplicity (13). The American College of Obstetricians and Gynecologists suggests using hematocrit as the basis to identify anemia in pregnancy (14) while the American Gastroenterological Association (15) and the American Academy of Pediatrics (16) define anemia based on hemoglobin concentration.

Although hemoglobin and hematocrit both have support in the literature and clinical guidelines as the biomarker of choice for identifying anemia, there is evidence that the two measures are not concordant. This has important implications for the identification and treatment of patients with anemia in the United States and globally. In 1981, Graitcer et al. (10) concluded that hemoglobin and hematocrit are not comparable in their ability to detect anemia in a study among US children aged 12–23 months with 20%–50% of children with anemia being missed when using hematocrit testing alone and between 25%–57% being missed when using hemoglobin alone. More recent studies from the United States (17) and Tanzania (5) are consistent with this finding.

Based on reviews and studies using hemoglobin and ferritin measurements, young children and pregnant persons are at highest risk for poor outcomes secondary to anemia, such as impacts to behavior and neural plasticity, preterm birth, preeclampsia, low birth weight, and neonatal and perinatal death (18–21). The accurate identification and treatment of individuals with anemia is important considering the health sequelae if it goes untreated. The aim of this study was to assess the degree of concordance, sensitivity, specificity, and percent agreement between hemoglobin and hematocrit for identifying anemia in children aged 1–<5 years and pregnant persons in the United States.

From the 10 NHANES cycles, 7,052 children and 1,437 pregnant persons met the inclusion criteria of having both hemoglobin and hematocrit test results. The mean age for children was 2.4 years, and 48% were female (Table 1). Mean age of pregnant persons was 27.9 years. Data on month of pregnancy were available for 1,119 (77.9%) individuals: 200 (17.4%) persons in their first trimester, 479 (41.8%) in their second trimester, and 440 (38.3%) in their third trimester (Table 1).

In both the children and pregnant populations, more individuals were identified as having anemia using hematocrit compared to hemoglobin (Tables 2 and 3). Among children aged 1–<2 years, anemia prevalence by hemoglobin was 6.3% (95% CI: 4.8–7.8) and 8.6% (95% CI: 7.0–10.2) by hematocrit. Among children aged 2–<5 years, anemia prevalence by hemoglobin was 3.0% (95% CI: 2.4–3.5) and 4.6% (95% CI: 3.8–5.4) by hematocrit (Table 2). Among pregnant persons, anemia prevalence by hemoglobin was 7.7% (95% CI: 5.9–9.5) and 12.4% (95% CI: 10.1–14.6) by hematocrit. Prevalence was highest among persons in their third trimester by both hemoglobin and hematocrit, 12.1% (95% CI: 7.5–16.7) and 15.2% (95% CI: 10.1–20.4), respectively (Table 3).

The overall kappa for children aged 1–<5 years was 0.68 (95% CI: 0.64–0.72), and the results by age, sex, race/ethnicity, and low-income threshold fell within the moderate concordance range, according to McHugh’s (2012) interpretation (0.60–0.79) (Table 2). Females (0.64; 95% CI: 0.58 – 0.71), children 2-<5 (0.64; 95% CI: 0.59 – 0.70), Mexican American children (0.64; 95% CI: 0.55–0.73), and non-Hispanic white children (0.64; 95% CI: 0.54–0.73) had the lowest kappa scores, while the highest kappa was seen in children aged 1–<2 years (0.75; 95% CI: 0.69–0.81). In children, overall sensitivity of hematocrit to identify anemia using hemoglobin as the standard was 81.5%, and specificity was 97.4% (Table 2). By child characteristics, sensitivity of hematocrit ranged from 77.6% and 77.5% among female children and children aged 2–<5 years, respectively, to 88.9% among 1–<2-year-old children; specificity was high by all characteristics (>97%). Among young children identified as having anemia on at least one test, 237 of 510 (46.5%) individuals were identified as having anemia based on only one test, 62 by hemoglobin only, and 175 by hematocrit only (Table 4).

The overall kappa for pregnant persons was 0.73 (95% CI: 0.68–0.79) (Table 3). The lowest kappa was among pregnant persons in their second trimester (0.53; 95% CI: 0.36–0.70; weak concordance), and all others, including the highest kappa among pregnant Mexican American persons (0.78; 95% CI: 0.67–0.88), showed moderate concordance. In pregnant persons, overall sensitivity of hematocrit to identify anemia based on hemoglobin as the standard was 93.6%, and specificity was 94.4%. By trimester, sensitivity ranged from 82.4% among persons in their second trimester to 100% among those in their first trimester (Table 3). Among pregnant persons identified as having anemia on at least one test, 82 of 213 (38.5%) individuals were identified by only one test, 9 by hemoglobin only and 73 by hematocrit only (Table 5).

While hemoglobin is a direct measure of the iron-containing protein and the oxygen carrying capacity in red blood cells, both hemoglobin and hematocrit are regularly used interchangeably to screen for anemia (3, 7). The two biomarkers measure different hematologic processes, and the literature documents differences in who is identified as having anemia based on the results of one biomarker or the other (5, 10, 17). This study aimed to examine the concordance, sensitivity, specificity, and percent agreement using thresholds from the 1998 CDC guidelines between hemoglobin and hematocrit for identifying anemia among children aged 1–<5 years and pregnant persons, two populations at high risk for anemia. We found a greater number of individuals identified as having anemia based on hematocrit for both populations, and the kappa results showed mostly moderate levels of agreement between hemoglobin and hematocrit in identifying anemia based on McHugh’s (26) interpretation. Using hemoglobin as the standard, overall sensitivity and specificity of hematocrit was 81.5% and 97.4% among children, respectively, and 93.6% and 94.4% among pregnant persons, respectively. The percent disagreement by both tests (46.5% in young children, 38.5% in pregnant persons) among those with at least one positive test suggests the biomarkers are identifying a substantial number of different people as having anemia and suggests they are not interchangeable.

Physiologic changes in pregnancy could potentially explain the weak level of concordance seen during the second trimester of pregnancy. Plasma volume expands by about 50% during pregnancy while red blood cell mass only expands by 25% (7). This leads to physiologic anemia by dilution and a lowered hematocrit, occurring most notably when the blood volume is undergoing the biggest transformation during the second trimester. CDC anemia cutoff values for the second trimester are lower to reflect this change. More generally, components used to measure hematocrit - red blood cell count (RBC), mean corpuscular hemoglobin concentration (MCHC), and mean corpuscular volume (MCV) (28) could also be affected by physiologic factors such as sex and age, as well as other factors like micronutrient deficiencies notably RBC folate (28) and vitamin B12. Evidence shows that MCV increases over the lifespan with the most rapid increase in the first 25 years of life (29). These physiologic changes could also affect the level of agreement between the two biomarkers. Further, hemoglobin and hematocrit were derived from the same sample of blood and measured by the same hematology analyzer. It is unknown whether the discordant findings would vary if other analyzers were used and more evidence from different analyzers could inform discussions around the use of hemoglobin and hematocrit. We are not well positioned to highlight which biomarker is the better measure of the two.

A limited number of studies have looked at the concordance between hemoglobin and hematocrit. Results from Quinto et al. (5) showed differences in who is identified as having anemia based on using hemoglobin vs. hematocrit among children aged 2–18 months in two locations in Tanzania. In Ifakara, 74% and 10% of children were identified with mild and moderate anemia, respectively, as measured by hemoglobin, and 42% and 3% with mild and moderate anemia, respectively, using hematocrit. Similar results were seen in children from Manhiça, Mozambique (5).

Sharma et al.(17) examined percent agreement among persons in their first trimester with at least one positive test for anemia using electronic health record data. They found 115 more women were identified as having anemia using the hematocrit marker compared to hemoglobin, in line with the findings from this study. Their results showed 45.2% percent agreement, lower than the results of this study (61.5% pregnant persons overall, 53.5% children aged 1–<5 years). We were not able to compare the percent agreement for the first trimester in this study due to the small sample size (n=11). According to McHugh “many texts recommend 80% as a minimum percent agreement” for healthcare settings (26). Neither study met that threshold.

The percent agreement between hemoglobin and hematocrit indicates that the two biomarkers are identifying different people as anemic. This is problematic because many clinical guidelines (including the guideline from the CDC) suggest that they are interchangeable (3, 7). The data presented here show that may not always be true. Understanding discordance, in addition to concordance, is also clinically relevant as it will be important to understand who might be missed when considering follow-up diagnostics and treatment options. In this study, hematocrit identified more people as having anemia than did hemoglobin. It will be helpful to understand how and if those identified as having anemia via hematocrit only or hemoglobin only differ in meaningful ways or in their health outcomes. This is relevant because consistently using both biomarkers to capture all possible individuals with anemia could have important resource implications. Further, if hemoglobin and hematocrit are not interchangeable, it could influence the way clinicians make diagnostic testing and treatment decisions. If only one biomarker is used, there is risk of excess or insufficient follow-up and treatment.

In addition to the clinical implications of missing individuals with anemia, there are wider policy and surveillance implications if hemoglobin and hematocrit are not concordant. For example, the US Department of Agriculture Food and Nutrition Services Supplemental Nutrition Program for Women, Infants, and Children (WIC) includes anemia among the medical conditions that could be used to meet the eligibility requirements for the program (30). When programs use anemia screening as an eligibility criterion, it may be possible that applicants are denied services depending on how anemia screening is conducted. When anemia data are used for surveillance, such as with electronic health records, anemia prevalence may be different and result in different healthcare policy decisions by those who use the data. Other additional criteria that are relevant to the considerations of selecting biomarkers include feasibility, acceptability, resources, cost effectiveness, and health equity.

There are a few limitations of this study. Trimester data were not available for all years examined. The sample sizes for pregnant persons were small, limiting the ability to disaggregate by trimester, race/ethnicity, and income. There are no standard cutoff points for hematocrit severity, so it was not possible to investigate differences by anemia severity. Data for pregnant adolescents aged 15–19 years and persons aged 45–49 years were not available in public domain in cycles 2007–2020. However, it is unlikely conclusions would be impacted by this potential selection bias given that the same method (hematology auto-analyzers) was used for all participants over the cycles. We were unable to adjust hemoglobin values for elevation of participants residing at or above 500m (24).

Strengths of this study include that NHANES used venous blood and laboratory methods did not change during the 1999–2020 cycles making the assessment of hemoglobin and hematocrit consistent throughout the years of this study. In addition, this study used nationally representative data making the findings generalizable to the US population.

All groups had moderate levels of agreement in their hemoglobin and hematocrit kappa results (McHugh’s interpretation), except for pregnant persons in their second trimester (kappa = 0.53) where agreement was weak. Among those who were identified as having anemia on at least one test, about 50% of children and 40% of pregnant persons were identified by only one of the two biomarkers, showing substantial discordance. Because hemoglobin and hematocrit are often used interchangeably in the clinical setting, individuals with anemia may be missed and not receive treatment, and therefore be at higher risk of adverse pregnancy, birth, and developmental outcomes.