Participants were women aged 15–40 y who were not pregnant or lactating and participated in a population-based biomarker survey in Chittoor, Andhra Pradesh, in southern India. The study design has been previously described (NCT04048330) ( 32, 33). Briefly, a total of 4124 households (n=3124 rural , n=1000 urban) were selected for the biomarker survey, which was designed to be population representative for women of reproductive age in this setting. Women were eligible for inclusion in this analysis if they had paired venous samples analyzed by HemoCue (HemoCue 301) and AHA (Coulter Counter HMX). Programmatically, Hb analysis via HemoCue was implemented after the start of data collection; as a result, Hb data for paired venous samples were available for 896 of the total 980 participants in the biomarker survey.

The study protocol for the biomarker survey was approved by the Institutional Review Board at Cornell University, and the Institutional Ethics Committees at Arogyavaram Medical Centre and St. John’s Research Institute (32, 33). The protocol was reviewed in accordance with CDC human research protection procedures and determined to be a nonresearch, routine surveillance activity. A nondisclosure agreement for personally identifiable information and data sharing agreement for de-identified data were established. This study received clearance from the Indian Council of Medical Research Health Ministry Screening Committee. Written informed consent (≥18 y) or assent (15 to <18 y) was obtained from all participants before the start of data collection. Women who had severe anemia (Hb <8.0 g/dL) were referred to a local clinic for follow-up per standard of care (33).

All data were collected at Arogyavaram Medical Centre by trained nurse enumerators via interviewer-administered questionnaires on electronic tablets (34). Data collection procedures (32, 33) included sociodemographic, anthropometric (e.g., weight, height; midupper arm, hip, and waist circumferences), dietary (e.g., 24-hour recall), health (e.g., signs, symptoms), and reproductive history data; and biological specimens (i.e., blood, saliva, urine). BIA (BC-148 MA; Tanita Corporation) was conducted among adults (≥18 y).

Venous blood samples were collected in 3 vacutainers (i.e., red-top, purple-top dipotassium EDTA (K2EDTA), and blue-top metal-free K2EDTA) by a trained phlebotomist (MPR) using standardized protocols (33). Blood samples (nonfasted) were collected from the antecubital vein from participants in the supine position. Venous whole blood samples collected via K2EDTA vacutainers were analyzed for Hb in real time, using both a HemoCue 301 (HemoCue) and an AHA (Coulter Counter HMX) as reference. HbA1c was assessed via nephelometry (Agappe Diagnostics). Plasma, serum, and RBCs were processed and stored at ≤ −80°C until batch analysis after the end of data collection. Blood samples were stored in a portable freezer unit (set to 4–6°C) immediately after collection until they were processed (≤4 h). After being allowed to reach room temperature, the K2EDTA vacutainers were remixed by inversion (10 times) and then aliquots were taken for analysis via both the AHA (700–800 μL) and the HemoCue device (1 drop, ~30 μL). A single drop of blood was taken from the K2EDTA vacutainer with a pipette and then placed on a glass slide. The microcuvettes were filled (10 μL) from the glass slide in 1 continuous process, and excess blood was wiped off. Microcuvettes were analyzed within 40 s of preparation.

Serum ferritin (SF) concentrations were measured by electrochemiluminescence (E411, Roche Diagnostics). Serum soluble transferin receptor (sTfR), C-reactive protein (CRP), and α−1 acid glycoprotein (AGP) concentrations were analyzed via the Roche COBAS Integra 400 plus analyzer (Roche Diagnostics). Red blood cell (RBC) folate and serum folate concentrations were measured using the WHO-recommended microbiologic assay (Bangalore, India). Serum total vitamin B-12 concentrations were assessed via chemiluminescence (E411, Roche Diagnostics). Plasma methylmalonic acid (MMA) and homocysteine (Hcy) were assessed by GC-MS (5975, Agilent Technologies).

One HemoCue device (i.e., HemoCue 301) was used for all sample analyses. The same HemoCue and AHA instruments were used for the duration of the study, and standardized protocols recommended by the instrument manufacturer were followed for instrument calibration and performance maintenance. Quality controls were performed each month for the HemoCue (Eurotrol Hb 301 controls Low/normal/high) and AHA. Protocols for repeating controls and adjusting as needed were in place if any controls registered outside of the acceptable range. No control results were outside of the acceptable range for the HemoCue or AHA during the study.

Hb concentrations were adjusted for self-reported smoking status, as recommended by the WHO (14). Continuous biomarker variables were ln-transformed before analyses. Biomarker results outside the assay limits of detection (LOD) were set to half the LOD (if below the LOD) or 2 times the LOD (if above the LOD). SF concentrations were adjusted for inflammation, using Biomarkers Reflecting Inflammation and Nutritional Determinants of Anemia (BRINDA) methods (53, 54). Geometric means (GMs) and 95% CIs were calculated to facilitate statistical inference.

Diagnostic accuracy parameters, including sensitivity, specificity, NPV, PPV, and accuracy, were calculated to evaluate the performance of HemoCue compared to the AHA reference. HemoCue diagnostic accuracy was examined using receiver operating characteristic curves.

Bland–Altman methods (i.e., mean difference, SD of differences (SDd), limits of agreement (mean difference ± 2 SDd)) (55) were used to evaluate the agreement between Hb concentrations measured via HemoCue and the AHA reference. Schuirmann’s two 1-sided tests method was used to evaluate equivalency between Hb concentrations assessed via HemoCue and AHA. A ± 7% threshold (16, 56) was used to examine differences in paired Hb results and differences in the overall mean, evaluated by HemoCue and AHA. Additional threshold differences of 0.5 g/dL and 1.0 g/dL were also considered.

Sensitivity analyses were performed by restricting the sample to HemoCue analyses conducted by the main laboratory technician. Analyses were also conducted within subsets of participants including stratification by rural residence, self-reported smoking status, Hb ≥8.0 g/dL (via AHA), and age (≥18 y, <18 y). Statistical analyses were conducted using SAS version 9.4 (SAS Institute, Inc.). Analyses were reproduced by Cornell Results Reproduction (R²) at the Cornell Institute for Social and Economic Research.

Table 4 presents Hb screening results analyzed via HemoCue and AHA methods, and Figure 2 presents the distributions of the Hb concentrations by method. The same laboratory technician (MPR) conducted most of the analyses via HemoCue (91.1%; data not shown). Hb concentrations evaluated by HemoCue were higher compared to AHA (n = 896; HemoCue: GM: 12.1 g/dL; 95% CI: 12.0, 12.2 g/dL compared with AHA: GM: 11.9 g/dL; 95% CI: 11.7, 12.0 g/dL; two 1-sided tests: not equivalent) (Table 4), with an overall mean difference (HemoCue – AHA) of 0.21 g/dL (Figure 2, Table 4). The mean difference in paired Hb results ranged from −5.6 to +4.1 g/dL. The limits of agreement were −1.8 to 2.3 g/dL, and the magnitude of the differences appeared to increase at higher Hb concentrations (Table 4, Figure 3). Although 65.8% of paired samples were within the recommended ±7% threshold, 27.7% of results differed by a magnitude of ≥1.0 g/dL and 56.5% differed by ≥0.5 g/dL (Table 4). Findings from sensitivity analyses were similar to those of the overall analyses (Supplemental Tables 1–3).

Table 4 also presents anemia screening results analyzed via HemoCue and AHA methods. The prevalence of anemia (Hb <12.0 g/dL) as evaluated by HemoCue was significantly lower than the AHA reference (36.3% compared with 41.6%; P <0.001). The prevalence of severe anemia (Hb <8.0 g/dL) as evaluated by HemoCue was not significantly different from the AHA reference (2.3% compared with 2.9%; P = 0.13) (Table 4).

Table 5 presents diagnostic accuracy parameters for anemia screening for HemoCue compared to the AHA reference. The HemoCue had 74.8% sensitivity, 91.2% specificity, 85.8% PPV, 83.5% NPV, and 84.4% accuracy for anemia screening, compared to the AHA reference (Supplemental Figures 1–3). When assessed via receiver operating characteristic curves (sensitivity plotted against 1 − specificity), the AUC was 0.893 for anemia and 0.997 for severe anemia (Supplemental Figure 4).

In this study, Hb concentrations assessed via HemoCue were significantly higher than those via the AHA reference. Although the overall mean difference was small (i.e., 0.2 g/dL), it ranged from −5.6 to +4.1 g/dL and this difference was ≥1.0 g/dL in 27.7% of paired samples. Findings are consistent with previous studies comparing screening methods of paired venous samples that reported higher Hb concentrations via HemoCue 301 than via the AHA reference (25–31), including studies in India in blood donors in Uttarakhand (n = 115) (27), women not accepted for blood donations in Mumbai (n = 147, 21–54 y) (25), and adults in a community-based study in Puducherry and Kolkata (n = 680, 18–60 y) ( 28); as well as in routine hematology analyses of whole blood samples in South Africa (n = 60) (26), and studies in children in Laos (n = 129, 15–32 mo) ( 29), pregnant women (n = 499, 18–45 y) ( 30) and children in the Gambia (n = 371, 6–24 mo) ( 30), and a refugee population in the United States (n = 299, 10 mo–60 y) ( 31). These studies reported higher Hb concentrations when evaluated by HemoCue 301 than by the AHA reference, although some did not report significance testing (25–28, 30) or magnitude of differences (57).

In the current study, the estimated prevalence of anemia via HemoCue 301 was significantly lower than that via the AHA reference (i.e., 36.3% compared with 41.6%). On an individual level, women with anemia may be incorrectly classified as not anemic (false negative) by portable hemoglobinometer and this has implications for follow-up. From a public health perspective, the prevalence of anemia may be underestimated at the population level. The estimated prevalence of anemia in this population is consistent with previous population-based studies in WRA in India when assessed by CMH or AHA (28–58.7%) (58, 59), but lower than recent state-level NFHS data for nonpregnant WRA (15–49 y; HemoCue 201+; NFHS-5 2019–2020: 59%; NFHS-4 2015–2016: 60.2%) (13). The lower estimated anemia prevalence via HemoCue observed here is consistent with previous studies evaluating venous blood via HemoCue 301 in comparison to the AHA reference, including whole blood samples submitted for routine laboratory analyses in South Africa (Hb <12.0 g/dL; HemoCue compared with AHA: 44% compared with 49%) (26), children in Laos (Hb <11.0g/dL; 36.4% compared with 65.9%) (29), pregnant women (Hb <11.0 g/dL; 37.1% compared with 54.5%) and children (Hb <11.0 g/dL; 63.9% compared with 84.4%) from the Gambia (30), and a refugee population in the United States (8.7% compared with 12.5%) (31); although significance testing was not reported in some studies (26, 30, 31).

Studies to date evaluating paired venous blood samples using the HemoCue 301 have consistently reported higher Hb concentrations and lower estimated anemia prevalence via HemoCue than via the AHA reference (25–31). In contrast, findings from studies evaluating paired venous samples on other HemoCue models (e.g., HemoCue B-Hb, HemoCue 201+, HemoCue 801, HB Donor Checker, or unspecified HemoCue model) have been heterogeneous, including higher (31, 60–68), lower (27, 68, 69), or not significantly different (60, 67, 68) Hb concentrations via HemoCue compared to the AHA and lower (31, 61), similar (60, 69), or higher estimated prevalence of anemia via HemoCue compared to the AHA, although statistical significance was not reported in some studies (27, 31, 60, 61, 63, 64, 69). A laboratory study of paired venous samples analyzed on the HemoCue 201+ compared with the HemoCue 301 found higher (2.6%) Hb concentrations on the HemoCue 301 than on the HemoCue 201+ (70). Methodological differences in Hb assessment between HemoCue models, such as measuring Hb absorbance in whole blood as opposed to lysed RBCs—which relies on cuvettes that are more susceptible to temperature and humidity (70, 71)—constrain the comparability of findings across studies using different HemoCue models (71).

In the current study, HemoCue sensitivity was 74.8% and specificity was 91.2%. To date, 13 studies (25, 26, 28, 31, 60, 64, 68, 69, 72–76) have evaluated diagnostic accuracy parameters of anemia screening methods, of which 8 used a HemoCue 301 (25, 26, 28, 31, 64, 72, 75, 76) and 4 (25, 26, 28, 31) analyzed paired venous blood samples using the HemoCue 301 compared to the AHA reference, with which our results are directly comparable. In studies in populations in India, South Africa, and the United States, the reported sensitivity (85.3%–99%) (25, 26, 28) and specificity (97.6%–100%) (26, 28, 31) of the HemoCue 301 were higher than observed here; in contrast, reported sensitivity (54.7%) (31), and specificity (52%) (25) were lower in studies in a refugee population in the United States (31) and in women not accepted for blood donations in India (25).

This analysis of paired venous samples from a population-based biomarker survey in WRA is among the largest of its kind, in a population at high risk of anemia. To our knowledge, this is among the first studies to date to evaluate diagnostic accuracy parameters of the HemoCue 301 by nutritional and metabolic characteristics of importance for WRA. This study leverages a well-characterized population with gold-standard biomarker assessment and comprehensive assessment of nutritional and metabolic characteristics. The use of paired venous samples, quality control measures, and analyses by the same laboratory instruments (e.g., the same HemoCue 301 device) and technician minimized variability between samples.

This study has several limitations. The cross-sectional study design does not enable evaluation of effects of nutritional or metabolic characteristics on anemia assessment. The WHO defines WRA as women aged 15–49 y; the current study focuses on a subset of this age range: women aged 15–40 y who were not currently pregnant or lactating. This in an important study limitation that may constrain comparability with other studies with heterogeneous definitions of WRA. The low prevalence of severe anemia (2.9%) limited the evaluation of HemoCue performance for severe anemia. The observed differences in variability between HemoCue and AHA at lower and higher Hb concentrations could not be fully investigated, owing to the small number of participants at the lower end of the Hb distributions, and this is an important study limitation. Analysis of both capillary and paired venous samples, inclusion of RBC morphology (microcytosis, macrocytosis), additional replicates of sample analyses prospectively by different methods, and evaluation in populations with wider variation in Hb concentrations would improve generalizability.

In summary, in this analysis of paired venous samples from a population-based biomarker survey in WRA in southern India, the estimated anemia prevalence via portable hemoglobinometer (HemoCue 301) was lower than via the AHA reference and may underestimate the prevalence of anemia at the population level. HemoCue sensitivity for anemia was higher in WRA with ID and lower in WRA with metabolic risk factors. The substantial burden of anemia and ID in WRA in this population and the variation in anemia assessment by screening method suggest an opportunity for prospective evaluation of screening methods and interventions for prevention of anemia.