Introduction

According to WHO guidelines, paediatric blood transfusion should be performed on patients with haemoglobin less than 4.0 g/dl with clinical signs of severity [1]. However, few facilities in sub-Saharan countries have access to reliable haemoglobin tests with a turnaround time between 1 and 8 hours [2]. As a result, around 40–65% of requested transfusions are based on clinical judgment, such as severe pallor [2]. This clinical judgment is often inaccurate in identifying severe anaemia (haemoglobin less than 5.0 g/dl), leading to unnecessary transfusions [3, 4]. This does not only expose children to transfusion-related risks but may also contribute to blood scarcity for patients who are most in need [4, 5]. Therefore, the ability to provide rapid diagnosis of severe anaemia is critical for life-saving interventions and the prioritisation of blood units for those in most need. A study conducted in Kenya, Tanzania, and Uganda revealed that a delay in blood transfusion for severely anaemic children within the first 8 hours resulted in a 52% mortality rate, with 90% of these deaths occurring within the initial 2.5 hours [6]. This outcome was associated with insufficient blood units, emphasising the importance of managing and prioritising the already scarce blood resources to prevent avoidable deaths [6]. Conversely, in a Kenyan hospital, children receiving transfusion on the same day, compared to the following day, were associated with a 42% reduction in mortality [7]. Therefore, accurate measurement of haemoglobin concentrations is important in the screening and management of anaemia in individuals, but also in the assessment of population iron status in surveys to evaluate the efficacy of intervention studies or public health interventions. Automated haematology analysers are generally considered the reference standard. These analysers are expensive and require specific ambient temperatures, cold storage for reagents, and a stable source of electricity, which poses challenges in field settings [8]. To address these challenges, various battery-operated point-of-care photometers have been introduced. They rely on either pooled or single capillary blood samples, making them convenient for use in field settings and are widely used for country-wide anaemia prevalence studies [9].

One such photometer is the HemoCue-301 (HemoCue, Ängelholm, Sweden), which was specifically developed to withstand challenging environments and does not utilise any hygroscopic ingredients [10]. The HemoCue-301 utilises a double-wavelength method, measuring the absorbance of whole blood at 506nm the Hb/HbO2 isosbestic point (i.e., the wavelength where Hb and HbO2 have the same absorbance or intensity), and at a wavelength of 880nm to compensate for turbidity. However, this technique is sensitive to oxygen, resulting in a 1.3% increase in measured haemoglobin concentration per minute the sample is left in open air [11]. Therefore, readings must be completed within 20–30 seconds to ensure accuracy [10].

Despite their benefits, variability in haemoglobin measurements between these point-of-care devices and automated haematology analysers has been reported [2]. A systematic review of 8 studies revealed higher haemoglobin concentration measurements of capillary blood using HemoCue-301 compared to measurement of venous blood in automated haematology analysers [11], with differences ranging from 0.5 to 6.1 g/L [11]. However, most of these studies determined the difference in haemoglobin measurement between the techniques using Bland-Altman analysis [12–20]. Unlike Bland-Altman analysis, Passing-Bablok regression analysis is a non-parametric method that does not assume constant measurement errors between methods and is robust against outliers [21].

We aimed to evaluate the accuracy of the HemoCue-301 in determining haemoglobin concentration among anaemic children aged 6–12 months in a field setting in rural Africa. As a reference, we used the Sysmex XN-1500 (Sysmex Corporation, Kobe, Japan), an automated haematology analyser that utilises a cyanide-free sodium lauryl sulphate reagent to lyse red blood cells, which, once released, forms a stable-coloured complex that is then measured photometrically [22].

Materials and methods

Results

A total of 227 participants had their haemoglobin assessed using both the HemoCue-301 and Sysmex XN-Series 1500 and were included in the analysis (S1 File). The participants had a mean age of 7.5 months (SD: 1.6 months); 52% were female. Of 227 participants, 44% and 33% were mildly and moderately anaemic respectively, whereas, 22% were classified as nomal according to the Sysmex haematology analyser. One participant was classified as severely anaemic (haemoglobin <7.0 g/dl) according to the Sysmex haematology analyser but not the HemoCue-301.

The median haemoglobin concentrations for HemoCue-301 and Sysmex XN-Series were 10.2 g/dL (IQR: 9.8, 10.7 g/dL) and 9.8 g/dL (IQR:9.2, 10.4 g/dL), respectively. At haemoglobin concentration of 6.5 g/dL, Hemocue measurements showed an absolute bias of 1.72 g/dL (95% CI: 1.37 to 2.10 g/dL) and a proportional bias of 26.5% (95% CI: 21.1% to 32.3%). For haemoglobin concentration of 8.0 g/dL, the absolute bias was 1.12 g/dL (95% CI: 0.91 to 1.37 g/dL) and the proportional bias was 14.0% (95% CI: 11.4% to 17.1%), while haemoglobin concentration of 10.0 g/dL showed an absolute bias of 0.32 g/dL (95% CI: 0.26 to 0.42 g/dL) and a proportional bias of 3.2% (95% CI: 2.6% to 4.2%). No bias was found at haemoglobin concentration of 11.0 g/dL (absolute bias: -0.08 g/dL; 95% CI: -0.18 to 0.07; proportional bias: -7.3%; 95% CI: -6.5 to 0.6) (Table 1, Fig 1). A modest positive correlation coefficient of 0.31, as indicated by Kendall’s tau, indicates limited agreement between the measurement methods.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Comparison of haemoglobin concentrations measured by HemoCue 301 vs Sysmex XN-1500.

HemoCue data were truncated at haemoglobin concentration <11.0g/dL (see text). Panel A:The black line represents the Passing-Bablok equation: ; the red dashed line represents the identity line; the grey shade area indicates the 95% confidence band. Panel B: The black line represents the Passing-Bablok equation: ; the red line signifies the zero bias, the blue dashed line represent ±7% acceptable difference and the grey shade indicates the 95% confidence interval. Panel C: The black line represents the Passing-Bablok equation: where the Intercept represents the constant bias at zero Sysmex (reference method), Slope reflects the proportional bias, and Xc denotes the specific Sysmex value for assessing percentage bias, the numerator combines the constant and proportional biases and divide it by Xc to normalizes the bias; The red dashed line signifies the zero bias, the blue dashed line represent ±7% acceptable difference, and the grey shade indicates the 95% confidence interval.

https://doi.org/10.1371/journal.pone.0310577.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Presents the absolute and proportional bias with their corresponding 95% confidence intervals, for haemoglobin concentration measurements obtained by HemoCue-301.

The absolute bias is expressed in grams per decilitre (g/dL), and the proportional bias is presented as a percentage (%).

https://doi.org/10.1371/journal.pone.0310577.t001

Discussion

We demonstrated that haemoglobin measurements between HemoCue-301 and the Sysmex XN-1500 vary across different haemoglobin concentrations and the magnitude of bias is inversely proportional to haemoglobin concentration. There was strong evidence of a absolute bias in haemoglobin concentrations above the acceptable difference of ±7% set by the College of American Pathologists and Westgard Clinical Laboratory Improvement Amendments in moderate and severly anaemic participants [26].

The diagnostic accuracy of HemoCue-301 compared to automated hematology analyzers has been examined in several studies. While some studies have reported acceptable differences within ±7%, others have shown higher difference [11, 28]. We found that bias between methods was inversely associated with hemoglobin concentration. Similarly, a study conducted in Laotian children reported a higher bias in children with lower haemoglobin concentrations [29]. In blood donors, no bias was observed at a haemoglobin concentration of 12.0 g/dl when compared to the WHO reference standard haemoglobin-cyanide technique [17]. A systematic review of 8 studies among adults indicated higher HemoCue-301 haemoglobin concentration measurements compared to automated analysers. This study concluded that the difference fell within the acceptable limit of ±7% set by the College of American Pathologists and Westgard Clinical Laboratory Improvement Amendments [11]. However, it is important to acknowledge that differences in haemoglobin measurements vary between methods at different concentrations, as demonstrated by our study, which was not accounted for in this study.The HemoCue-301 operates on spectrophotometric principles, where the absorbance of light passing through a blood sample is directly proportional to the concentration of haemoglobin at the Hb/HbO2 isosbestic point. Due to the stringent adherence to the haemoglobin measurement protocol within our clinical trial, the observed higher haemoglobin concentrations, particularly in anaemic children, are likely attributable to device bias arising from increased absorbance at lower haemoglobin concentrations rather than preanalytical errors. In contrast, a study among women in Peru has indicated lower haemoglobin concentrations measured by HemoCue-301 compared to automated analysers, with 50% of these women classified as anaemic according to HemoCue-301 [30]. However, in our analysis, only 7.4% of participants were misclassified as anaemic by the HemoCue-301, while they were classified as normal (haemoglobin concentrations greater than 11 g/dL) according to the Sysmex XN-1500.

The WHO is currently involved in ongoing research to comprehensively understand preanalytical and analytical factors influencing haemoglobin measurement using these point-of-care devices. This research aims to inform the development of new guidelines in this area [31] and our study will contribute to the existing evidence that can be utilised in these guidelines’ development. This study is part of an ongoing clinical trial, thus ensuring stringent monitoring and adherence to trial protocols throughout the blood sampling process and haemoglobin measurements. This study has several limitations as it was a secondary data analysis using a convenience sampling method. Consequently, the findings may not be generalisable to rural Gambian children aged between 6 and 12 months. Additionally, capillary and venous blood sampling for haemoglobin measurement were not performed on the same day. The retrospective nature of our study limited our ability to explore the potential impact of different blood sample types (venous vs capillary) on haemoglobin concentration measurements. Furthermore, our study design did not allow for a direct comparison between point-of-care tests and automated haematology analysers regarding anaemia prevalence. This is because only children identified as anaemic by HemoCue-301 underwent haemoglobin measurements using the automated analyser. Another limitation is the comparison of different blood types using samples collected on different days.

Conclusion

Our study indicated that bias due to HemoCue-301 depends on the actual Hb concentration, and seems unacceptably high at Hb values below 8.0 g/dL. While this method may offer convenience in field surveys for estimating anaemia prevalence, caution is required when using it for a screen-and-treat approach.

Supporting information

Acknowledgments

We thank the IDeA3 study team at MRCG@LSHTM for their dedicated support, the MRCG Keneba laboratory, the data management team and the MRCG Clinical Trial Unit. We also thank the participants and communities of Kiang East and Jarra West.