Agreement between Noninvasive Hemoglobin and Laboratory Hemoglobin Measurements in Neonates: A Systematic Review and Meta-Analysis

Hemoglobin (Hb) is essential for the sustenance of life by carrying oxygen for the tissue. Anemia and polycythemia are common morbidities in sick neonates, and Hb testing is a point-of-care investigation, frequently ordered by physicians. Hb assessment is helpful in the differential diagnosis of many neonatal diseases [1] and is a prognostic indicator of preterm survival [2]. Packed red blood cell (PRBC) transfusion in sick neonates is common and is based on Hb level [3]. Older preterm neonates require frequent laboratory testing for timely diagnosing the anemia of prematurity and judicious transfusion management [4, 5].

Invasive venous or arterial sampling for Hb estimation using an automated laboratory analyzer is the gold standard. Preterm neonates have a lesser pool of blood volume, and multiple phlebotomies may cause iatrogenic anemia [6]. Frequent venipuncture is a risk for infection by iatrogenic skin breakdown; again phlebotomy for laboratory testing is a painful event and increases parental anxiety.

Noninvasive spectroscopic Hb measuring devices (noninvasive Hb [NHb]) are a bedside tool to estimate Hb instantaneously with a sensor attached to one of the limbs. Recently, the clinical utility of NHb devices among adult and pediatric populations has been summarized by a systematic review and meta-analysis, excluding neonates and infants [7]. Evaluation of the diagnostic accuracy of NHb devices for neonatal Hb measurement is the primary objective of this systematic review and meta-analysis.

The protocol of this systematic review was developed following the “Preferred Reporting Items for Systematic Reviews and Meta-Analyses” (PRISMA) guidelines [8] and was registered with the International Prospective Register of Systematic Reviews – PROSPERO (Registration no CRD42022306608). The PRISMA 2020 statement checklist [9] is given as online supplementary material S1 (for all online suppl. material, see www.karger.com/doi/10.1159/000526100). The PRISMA flowchart is given in Figure 1.

Databases such as Medline, PubMed Central, PubMed, Web of Science, Google Scholar, and Scopus were searched for full-text articles in the English language. The use of NHb measurement among neonates is a recent phenomenon; therefore, literature published from January 2000 to January 2022 was searched. The search was made using a combination of MeSH (Medical Subject Headings) terms as well as keywords for identifying appropriate publications. Keywords characterizing “participants,” “measurements,” and “methods” were used in permutation and combination to get the relevant citations. For “participants,” terms like newborn, neonate, and Infant were used. For “measurements,” terms such as total Hb, total hemoglobin, hemoglobin, and Hb were used. For “methods,” terms such as non-invasive, spectroscopy, laboratory methods, validity, and reliability were used. Standard Boolean operators were employed while searching in the database. Full-text publications of conference proceedings and bibliographies of review articles were searched again in the above databases. The search strategy for this systematic review is given in online supplementary material S2.

Full-text articles reporting simultaneous Hb measurement by laboratory and noninvasive methods; and assessing accuracy by correlation coefficients, or mean difference (Bland-Altman difference plots) [10] were included in the review. The studies where the described time difference between NHb and laboratory Hb (LHb) was beyond 24 h were excluded for analysis. Case series, case reporting, and review literature about neonatal Hb measurements were excluded from the review.

The titles and abstracts were reviewed independently by two authors using the predefined selection criteria, and unique records were prepared. The full-text review was done for each unique record for the data extraction. The description of the study population, i.e., gestational age, birth weight, postnatal day of Hb estimation, clinical condition of neonate, number of neonates, number of paired observations, blood sampling technique, NHb device, total Hb (mean ± SD) using both the methods NHb and LHb, device failure rate, correlation coefficient, and bias (precision), was extracted by two investigators and was matched for finalization.

The “Quality Assessment of Diagnostic Accuracy Studies - QUADAS-2” [11] was used for bias assessment of included studies. The quality domains such as participant selection, index and standard tests, and 'flow and timing' were examined for each study independently by two authors. The consensus was reached after discussions to resolve the disagreements.

This study estimated two pooled statistical parameters in the meta-analysis. The strength of the relation between NHb and LHb was reported as pooled correlation coefficient with a confidence limit using MetaXL (v.5.0) software. It is a free “add-in” Microsoft Excel program available at https://epigear.com website, used for implementing the inverse variance heterogeneity model during metanalysis [12, 13]. The degree of heterogeneity was assessed by Cochran’s Q and I² statistics. Pooled bias (Bland-Altman difference plots) between invasive and noninvasive methods was estimated using a method suggested by Peyton and Chong [14]. Subgroup data analysis was done separately for sick and hemodynamically stable neonates. Considering the number of included studies as ten with substantial heterogeneity, tests for funnel plot asymmetry (publication bias) were not assessed [15].

This study comprises ten prospective comparative studies from different parts of the world, i.e., four studies from India [16‒19], two studies from Germany [20, 21], one study each from Republic of Korea [22], UK [23], USA [24], and Turkey [25]. Masimo radical-pulse co-oximetry (MRPCO) was used in all enrolled studies except two studies [20, 23] where Mediscan-2000 was used. The NHb level was compared with the LHb level in 1,477 paired observations from 1,047 neonates. Study sample characteristics, i.e., gestational age, birth weight, the timing of testing, clinical conditions of neonates, types of sample for LHb testing, the time gap between NHb and LHb, and device failure rate in enrolled studies, are described in Table 1.

Seven studies included both preterm and term neonates, and three studies [22‒24] enrolled neonates with average gestational age below 32 weeks including extreme low birth weight neonates (birth weight <1,000 g). Hb testing was conducted within the first postnatal week in three studies [16, 17, 24], the rest of seven studies were conducted beyond the first postnatal week of life, whereas three studies [20, 21, 23] included mostly preterm neonates beyond 2 weeks of age. The clinical condition of neonates during Hb testing was informed in all studies except Jung et al. [22]. Stable neonates, i.e., without any circulatory insufficiency, were enrolled in seven studies [16, 17, 19‒21, 23, 25], and sick neonates were enrolled for Hb estimation in rest three studies [17, 18, 24]. Bhat et al. [17] informed separately about sick and stable neonates. The validity of NHb was represented in together for both sick and stable neonates in Lingaldinna et al. [18]. Sick neonates were described as neonates requiring ventilator support or need of circulatory support.

Eight out of 10 studies had taken a venous sample for LHb estimation except Nicholas et al. [24] (either arterial or capillary sample) and Rabe et al. [23] (both capillary and venous blood). LHb was measured in an automated analyzer in all studies except the arterial blood gas analyzer in Nicholas et al. [24]. The maximum time gap between NHb and LHb testing in the included studies was 14 h [20].

The mean (SD) of Hb level measured by NHb and LHb methods was mentioned in five studies and the mean (range) in one study. The average NHb level was higher than the average LHb in three studies [18, 22, 24] and lower in three studies [16, 19, 25].

The majority of studies are of good quality, with a low risk of bias in patient selection, index testing, reference testing, 'flow and timing', and applicability concerns as presented in Table 2. The majority of studies have restricted only to stable neonates, considering the high rate of device failure in circulatory insufficiency neonates. A high risk of bias in patient selection has been found in a few studies. Similarly, LHb testing using nonvenous blood sampling in some studies may pose a concern in generalizability. Most of the studies have taken maximum care for simultaneous testing. A time gap of about 4 h between two tests has been noted in Rabe et al. [20] and 11–14 h in another study by Rabe et al. [23].

The strength of the relationship between NHb and LHb was informed as a correlation coefficient in seven studies, and the intra-class correlation coefficient (ICC) was reported by Lingaldinna et al. [18]. There was moderate agreement between NHb and LHb in sick neonates in Lingaldinna et al. (ICC = 0.56), and NHb was moderately correlated with LHb in Nicholas et al. (r = 0.67). The estimated pooled correlation coefficient between NHb-LHb was r = 0.94 (95% CI: 0.83–0.98) with significant heterogeneity (I² = 99%) (Fig. 2a). The subgroup analysis for stable neonates yielded a higher pooled correlation coefficient between NHb and LHb, i.e., 0.95 (95% CI: 0.88–0.98) (Fig. 2b).

Agreement statistics between NHb and LHb by Bland-Altman analysis were reported as bias (precision) in nine studies. In five studies, NHb underestimates LHb, and NHb overestimates LHb in three studies. In Rabe et al. [23], the bias was “nil,” with the narrowest limit of agreement (LOA) among all studies. Overall pooled bias (precision) between NHb and LHb measurements was −0.013 (1.4) gm/dL.

Five studies informed about bias between NHb and LHb in exclusively stable healthy neonates. Bhat et al. informed bias (precision) between NHb and LHb, separately for stable and sick neonates. Rabe et al. estimated bias (precision) between NHb and LHb for venous and capillary samples separately in stable neonates. The pooled bias (precision) for stable neonates was −0.26 (0.91) gm/dL. Three studies were not included in the stable and sick subgroup analysis because of limited information about neonatal conditions in Jung et al. and Lingaldinna et al. Bias (SD) value was not mentioned in the text by Rabe et al. (2005) [20]. The pooled bias (precision) in sick neonates was calculated from 207 NHb-LHb paired observations from 101 babies in two studies, i.e., Nicholas et al. and Bhat et al. For the comparison, the pooled bias (precision) in subgroups (all vs. stable vs. sick) neonates is given in Table 3.

In this review, five studies [Lingaldinna et al., Jung et al., Jamal et al., Vora et al., and Lingaldinna et al.] informed about the accuracy of the NHb device for anemic neonates. Two studies informed the bias (precision) between NHb and LHb in anemic neonates, i.e., Vora H et al., −0.2 (1.1) gm/dL in 5 neonates, and Lingaldinna et al., 2.69 (1.87) gm/dL in 72 neonates. Jung et al. and Jamal et al. had informed the correlation coefficient between NHb and LHb as r = 0.513 in 73 neonates and r = 0.907 in three neonates, respectively, in anemic neonates, and ICC was 0.284 between NHb and LHb in anemic neonates in Lingaldinna et al. Similarly, two studies informed the correlation coefficient between NHb and LHb in a total of 27 neonates for a higher range of Hb >18 gm/dL, i.e., Jamal et al. (r = 0.077, 17 neonates) and Jung et al. (r = 0.294, 10 neonates).

In this systematic review, the diagnostic accuracy of spectroscopic NHb devices was evaluated by comparing them with gold standard LHb in 10 single-center studies. Overall NHb device is very strongly correlated with LHb, and the average difference (pooled bias) is very minimal but the LOA is wider in Bland-Altman’s analysis. The NHb is more useful in stable neonates over sick neonates with a higher correlation coefficient and lesser average difference (bias) with better precision between NHb and LHb.

This systematic review is the first of its kind to look at the diagnostic accuracy of NHb devices in neonatal populations including sick and stable neonates. Low risk of bias in four key domains with certain exceptions has been observed in all the enrolled studies. However, a severe degree of heterogeneity in the included studies, consequently wider edges of the diamond, has been noted during the pooled estimation. The degree of heterogeneity may be explained by the presence of various confounders in the included studies such as cardio-respiratory status, Hb level, and NHb measuring devices. Further, the inclusion of small number of studies in this meta-analysis may have biased the estimated I² value [26].

Considering the higher accuracy of NHb estimation over clinical assessment, NHb can be useful for ordering timely LHb estimation [27]. The overall strong correlation between NHb and LHb measurements supports the usefulness of NHb in serial monitoring of Hb in neonates.

The enrolled studies used either correlation coefficient or bias (precision) in the Bland-Altman plot or both for the validity assessment of NHb devices. The correlation coefficient informs the strength of the linear relationship between two continuous measurements without providing their agreement. The Bland-Altman analysis describes the distribution of differences between two measurements by LOA and the average difference between them as bias [28]. The information obtained in Bland-Altman difference plots as “bias” could reasonably help the clinician with the likely prediction of LHb from the NHb reading. The pooled bias of NHb in this meta-analysis is comparable to the accuracy of NHb monitoring (by Masimo devices) in adult and pediatric populations [29‒34], in whom NHb is useful to restrict unnecessary transfusion.

The clinical interest in NHb devices in the field of neonatology lies in the PRBC transfusion decision in sick neonates and hemodynamically stable older preterm babies. Hb level could be estimated from various anatomical sites of the vascular system; i.e., the noninvasive device measures the Hb from the capillaries distributed over the skin, whereas LHb detects venous or arterial Hb [35]. Changes in microcirculation influence the presence of Hb in the capillaries [36]. Hence, the available NHb devices need good circulatory status of the newborn in order to obtain adequate signals for Hb measurements. The PRBC transfusion decision in neonates with cardio-respiratory support demands a sensitive and aggressive approach [3]. Simultaneously, transfusion-related complication compels physicians to consider lower Hb thresholds for transfusion decisions [37]. The lower precision (SD = 1.66 gm/dL and wider LOA) of NHb devices in sick neonates may lead to unjustified clinical transfusion decisions resulting in either unnecessary transfusion or delay in transfusion.

The accuracy of the NHb device was tested in older ex-preterm neonates (stable preterm neonates beyond 2 weeks of postnatal days) in two studies of this metanalysis [21, 23]. In Rabe et al. [23], the better precision (SD, 0.6 gm/dL) of the NHb device (Mediscan-2000) looks promising, but the lower precision (SD, 1.58 gm/dL) in Wittenmeier et al. [21] measured by MRPCO measurement has raised clinician’s concern. Considering less chance of device failure, lower bias/precision observed in stable neonates (over sick neonates) and the older preterm neonates are the best beneficiary of NHb by eliminating recurrent phlebotomy; further multicentric studies should focus on these populations with either of these devices.

There are limited data to summarize the accuracy of NHb in the higher (e.g., polycythemia) or lower Hb (e.g., anemia) range. For summary statistics, at least two studies with an adequate sample size are mandatory for the external validity of the study result [38]. Hence, further studies with more comparison samples are needed for the accuracy of NHb devices in the extreme Hb range, i.e., very low Hb levels or polycythemia, in neonates.

The review has a few limitations; first, only two studies were available with Mediscan-2000, and the rest eight studies used MRPCO, limiting the estimation of device-specific accuracy. Second, the use of a weight-appropriate sensor (as per manufacturer instruction, i.e., R1 25L for children <3 kg; R1 20L for neonates ≥3 kg) of MRPCO was restricted to only two studies [21, 25]. The NHb device Mediscan-2000 has not been explored in sick neonates and extreme ranges of Hb levels [20, 23]. Third, the inclusion of multiple readings from each enrolled patient might have introduced the dependency on data. Lastly, the estimates provided for sick neonates are based on limited data. The device’s accuracy is validated in a limited number of extreme preterm neonates. They have a lesser pool of blood volume, a higher likelihood of circulatory insufficiency, and required multiple testing; validating NHb devices in these subpopulations is an unmet need.

The currently available NHb devices measure the Hb level reasonably accurate in neonates. NHb measurements should be interpreted with caution in sick neonates and anemic conditions. It may be used as a screening tool at point-of-care or serial monitoring for timely requesting the LHb investigation. The clinical decision for transfusion practice should not be guided by NHb alone and need laboratory confirmation. Multicentric studies involving both sick and older stable preterm neonates are warranted. Further innovation in NHb devices is needed for better accuracy in critically sick neonates.

An ethics statement is not applicable because this study is based on published literature only.

The authors have no conflicts of interest to declare.

The authors received no funding for this study.

Santosh Kumar Panda: concept, design, data acquisition, interpretation, first draft, and review; Alpana Mishra: concept, design, data acquisition, interpretation, and review; Pratap Kumar Jena: concept, design, data acquisition, data analysis, data interpretation, and review.

The data used in this study are available from the published literature. All the extracted data used in this study for analysis are given in the result section itself.