Journal of Clinical Neonatology

: 2022  |  Volume : 11  |  Issue : 1  |  Page : 7--12

Accuracy of neonatal venous blood glucose measurements using blood gas analyzer compared with central laboratory chemistry analyzer

Waricha Janjindamai, Nichanan Tiwawatpakorn, Anucha Thatrimontrichai, Supaporn Dissaneevate, Gunlawadee Maneenil, Manapat Phatigomet 
 Department of Pediatrics, Faculty of Medicine, Prince of Songkla University, Hat Yai District, Songkhla, Thailand

Correspondence Address:
Waricha Janjindamai
Department of Pediatrics, Faculty of Medicine, Prince of Songkla University, 15, Kanjanavanit Raod, Tambon Korhong, Hat Yai District, Songkhla 90110


Background: Hypoglycemia is a serious problem in infants at risk and creates long-term consequences. Therefore, rapid and accurate measurement of blood glucose is of clinical importance. Objectives: The objective of this study was to evaluate the agreement of venous blood glucose measurements from hypoglycemic high-risk neonates, using blood gas analyzer (BGA), compared to central laboratory chemistry analyzer (CL). Methods: A prospective study of all high-risk neonates for hypoglycemia and neonatal intensive care units (NICUs) was enrolled. Point-of-care glucose was performed, and if <40 mg/dL, venous blood would be collected for CL and ABG. For analysis of the agreement of CL and BGA, Bland–Altman (BA) analysis, with multiple observations per individual, including limits of agreement (limits of agreement [LOA] ±1.96 standard deviation [SD]), was used. Results: One hundred and forty-five paired glucose values were analyzed. There were strong correlations between CL and BGA in all glucose measurements and hypoglycemic range (r = 0.81, P < 0.001, and 0.73, P < 0.01, respectively). A weak correlation was demonstrated in hyperglycemic ranges (r = 0.35, P = 0.15). For BA analysis of all glucose measurements and hypoglycemic ranges, LOA (±1.96 SD) of CL and BGA were −9.5 (±46.5) mg/dL and −11.1 (±10.9) mg/dL, respectively. The hyperglycemic range illustrated higher LOA, with LOA (±1.96 SD) of −32.9 (±124) mg/dL. Conclusions: In NICU settings where normoglycemic and hypoglycemic ranges are targeted, venous BGA glucose could be used as a reliable test instead of CL. No similar correlation was found in the hyperglycemic range.

How to cite this article:
Janjindamai W, Tiwawatpakorn N, Thatrimontrichai A, Dissaneevate S, Maneenil G, Phatigomet M. Accuracy of neonatal venous blood glucose measurements using blood gas analyzer compared with central laboratory chemistry analyzer.J Clin Neonatol 2022;11:7-12

How to cite this URL:
Janjindamai W, Tiwawatpakorn N, Thatrimontrichai A, Dissaneevate S, Maneenil G, Phatigomet M. Accuracy of neonatal venous blood glucose measurements using blood gas analyzer compared with central laboratory chemistry analyzer. J Clin Neonatol [serial online] 2022 [cited 2022 Jan 16 ];11:7-12
Available from:

Full Text


Glucose is a major substrate for energy metabolism, as the brain is an obligatory consumer of glucose. The long-term impact of symptomatic hypoglycemia in newborns can cause neurologic problems.[1],[2] Furthermore, prolonged or recurrent periods of hypoglycemia can lead to long-term neurocognitive impairment.[3] During the newborn period, the frequency of hypoglycemia is increased, but hyperglycemia is common, especially in critically ill newborns. Hyperglycemia develops as a result of the inability of the newborn to decrease endogenous glucose production during parenteral glucose infusion, or the inability to increase peripheral glucose use. As of date, a cutoff plasma value of 40 and over 150 mg/dL, in neonatal hypoglycemia and hyperglycemia patients, respectively, regardless of gestational age, has been developed by the American Academy of Pediatrics.[4],[5] Moreover, the Canadian Pediatric Society guidelines have accepted the definition of hypoglycemia as a blood glucose level of <47 mg/dL.[6] However, the clinical manifestations of neonatal hypo-hyperglycemia at this range are non-specific and are similar to many other disorders in neonates. Hence, the accurate measurement of blood glucose in newborn infants is of clinical importance. Due to the limitations with unreliable rapid point-of-care (POC) glucose,[7],[8],[9],[10] and a falsely long lag time from blood sampling to obtaining results from central laboratory chemistry analyzers (CLs),[4] along with world expert guidelines on pediatric parenteral nutrition recommendations, the glucose blood gas analyzer (BGA) is the best method for combining quick and accurate results for glucose measurements.[11],[12],[13],[14],[15] However, only a few studies concerning correlation and agreement of BGA and CL in high-risk neonates have been demonstrated. These also have some limitations, such as a small sample size and a limited number of study populations in hypoglycemia ranges.[14],[15] In addition, one study used arterial blood sampling, which is not routine practice,[15] while the other study had compared BGA and CL by using capillary samples.[14] A variation between capillary and venous plasma glucose could vary within a 10% range, with higher in capillary than venous plasma glucose readings.[16] There is still a lack of venous blood sampling and systematic reviews in assessing the agreement of glucose measurements by BGA and standard CL neonates; in addition, most of the studies have been performed in adults.[11],[12] Hence, the primary objective of this study was to evaluate the agreement between BGA and CL of venous blood glucose measurements in neonates at risk for hypoglycemia.


A prospective study was conducted in a single, 15-bed, neonatal intensive care units (NICUs), and nursery unit of a university-affiliated teaching hospital in the south of Thailand, from 1 January 2020 and 30 September 2020. This study followed STARD guidelines: at the time of the study, it was our policy to send a sample for CL under such circumstances. All high-risk neonates for hypoglycemia (such as small or large for gestational age newborns, those born to mothers who have diabetes, late or preterm newborns, and critically ill infants),[6] who obtained POC glucose at the NICU, at the nursery unit and all sick NICU neonates with indwelling umbilical venous catheters (UVCs) or central venous catheters (CVCs), were enrolled (to avoid intent to venipuncture for only taking blood sampling for this study). Neonates without intravenous access, the collection time between methods, being more than 5 min, nonphysiologic serum from vascular line contamination (such as a sodium level <120 or >160 mEq/L and a potassium level <3 or >8 mEq/L), insufficient blood sample for BGA, and the parents' decision not to participate were excluded. This study was approved by the Ethics Committee Board of the Faculty of Medicine, Prince of Songkla University (REC 62-375-1-1).

After enrollment, POC glucose (ACCU-CHEK® Performa: Roche Diagnostics, Mannheim, Germany) would be performed in all infants meeting the inclusion criteria. If the whole-blood glucose levels from POC glucose were <40 mg/dL, venous blood from the UVC or a peripheral line (0.7 mL) would be collected simultaneously in sodium fluoride (NaF) (0.5 mL) and heparinized tubes, by one nurse. The NaF microtube would be sent immediately to the CL to determine plasma glucose, by the enzymatic method with hexokinase (Roche/Hitachi Cobas c 701/702), and the heparinized tube (whole blood) would be used for BGA, which is accomplished by spectrophotometry, using the hexokinase method (ABL800 FLEX BGAs: Radiometer Medical ApS, BrØnshØj, Denmark). All measurements were achieved in accordance with the requirements of the manufacturer. In infants who received intravenous fluids in the NICU, only venous blood from UVC or CVC would be obtained in both tubes, by a double syringe technique. This was conducted at the same time as routine laboratory testing, per TPN guidelines.

Statistical analysis

The sample size calculation depended on mean bias and the limits of agreement (LOA) (±1.96 standard deviation [SD]). When comparing blood glucose levels between BGA and CL from a previous study, this was 0.1 ± 10.8 mg/dL.[14] Allowing for a 20% SD error, the final sample size was 136 infants. Epicalc package (R package, Songkhla, Thailand, 2016), the R program version 3.2.2 (R Foundation for Statistical Computing, Vienna, Austria, 2016) was used to develop a database of categorical, continuous variables and to undertake analysis. The data along with clinical parameters were expressed as mean ± SD, or median and interquartile range (IQR). Paired t-test or Wilcoxon rank-sum test was used to compare parameters and characteristics between BGA and CL. The Pearson correlation coefficient (r) between the different methods was determined by linear regression. For analysis of the agreement of CL and BGA, Bland–Altman (BA) analysis, including mean bias and LOA (±1.96 SD), was used and did not exclude outliers.[17] P < 0.05 was accessed as indicating statistical significance.


During the study period, 158 eligible pair glucose values were obtained. Thirteen pairs were excluded, due to collection times between methods being more than 5 min (5 pairs), insufficient volume for BGA specimens (3 pairs), serum sodium levels of more than 160 mEq/L (2 pairs), serum potassium <2 mEq/L (2 pairs), and the other one was serum potassium of more than 8 mEq/L. Finally, 145 pair glucose values were included in this study.

Baseline characteristics of the study population

Sixty percent (87/145) were male infants, and sixty-two infants (62/145, 42.8%) were preterm infants (<37 weeks' gestation). The mean ± SD birth weight was 2390 ± 935.7 g. Twenty-five infants (25/145, 17.2%) were small for gestational age, and 12 infants (8.3%) were large for gestational age. The paired glucose values were mostly collected within 1 h after birth. Thirty-one infants (31/145, 21.4%) were infants of maternal diabetic mellitus. The mean ± SD of hematocrit (Hct) was 46% ±7.3%. Fifty infants (34.5%) were hypoglycemic infants (CL plasma glucose <47 mg/dL), while 18 infants (12.4%) were hyperglycemia (CL plasma glucose >150 mg/dL). The median (IQR) CL and BGA plasma glucose levels of all infants were 79 (32, 109) mg/dL and 77 (42, 111) mg/dL, respectively, which were statistically significant in difference (P < 0.01). For the hypoglycemic range (n = 50, 34.5%), the mean ± SD of glucose levels from CL and BGA was 24.4 ± 11.4 (range, 2–44) mg/dL and 35.5 ± 15.9 (range, 1–75) mg/dL, respectively. These were also statistically significant in difference (P < 0.01). The median (IQR) of CL and BGA in the hyperglycemic range (n = 18, 12.4%) was 173.5 (158, 215) (range, 152–431) mg/dL and 183 (161.2, 289.8) (range, 113–534) mg/dL, respectively, which were not statistically significant in difference between methods (P = 0.41) [Table 1].{Table 1}

When comparing between CL and BGA, there was a strong correlation between all glucose measurements and hypoglycemic ranges by CL and BGA (r = 0.81, P < 0.01, and 0.73, P < 0.01, respectively). However, a weak correlation was demonstrated in the hyperglycemic range (r = 0.35, P = 0.15). For all glucose measurements, the BA analysis, the mean bias (LOA) of CL and BGA was −9.5 (±46.5) mg/dL: 96.6% of these values were within the 95% LOA [Figure 1]. For hypoglycemic range, the mean bias (LOA) was −11.1 (±10.9) mg/dL: 90% were within the 95% LOA [Figure 2]. The hyperglycemic range illustrated higher LOA than the hypoglycemic range, with a mean bias (LOA) being −32.9 (±124) mg/dL, and with 88.9% of all values being within the 95% LOA [Figure 3] and [Table 1].{Figure 1}{Figure 2}{Figure 3}

Receiver operating characteristic analysis for neonatal hypoglycemia and hyperglycemia

Receiver operating characteristic (ROC) analysis revealed a high value of area under the ROC curve (AUC) of BGA glucose at the critical low (hypoglycemia range), normal and high (hyperglycemia range) cutoff glucose levels compared with CL (AUC of 0.88, 0.88, and 0.94, respectively), and had a strong effect of positive likelihood ratio (37.05) in the detection of neonatal hypoglycemia [Figure 4]. Furthermore, ROC analysis revealed the highest accuracy in the hypoglycemia curve (cutoff of <47 mg/dL) for both methods, with BGA demonstrating an excellent specificity (98%) at this threshold (AUC = 0.88). Concordance within the hyperglycemia (CL plasma glucose level >150 mg/dL), ROC curve from the BGA glucose performed a desirable confirmation test of AUC 0.94 (specificity 98%), along with a positive likelihood ratio of 56.44. Moreover, at the normal glucose level (CL plasma glucose level between 47 and <150 mg/dL), BGA glucose also had high accuracy, with a specificity and AUC of 81% and 0.88, respectively [Figure 4] and [Table 2].{Figure 4}{Table 2}


Appropriate hypoglycemic detection is extremely important for infants at risk. The critical glucose concentration at which neuroglycopenia develops has been suggested to be <47 mg/dL, because of altered somatosensory-evoked potentials.[18] This present study has demonstrated that in NICU settings where normoglycemic and hypoglycemic ranges are targeted, using venous BGA glucose could be used as a reliable test instead of CL. BGA glucose is an easily performed bedside test for all busy NICUs, which uses small blood volumes, and promptly detects both low and normal glucose values. This can provide timely and desirable management for the prevention of complications in neonatal hypoglycemia. However, previous studies in the accuracy of BGA glucose measurements in newborn infants still have some limitations, such as small sample sizes, a limited number of hypoglycemia ranges, and the use of arterial blood sampling, which is not routine practice.[15] This study demonstrated that for hypoglycemic and normoglycemic ranges of glucose values, measured by venous BGA glucose compared to CL, there was a strong correlation between BGA and CL (r = 0.81, P < 0.001). Although this concurred with the study by Newman et al.,[14] the correlation between BGA and CL in hyperglycemic ranges had a weak correlation (r = 0.35, P = 0.15). The small number of the population in the hyperglycemic range (18/145, 12.4%), due to low prevalence of hyperglycemia in newborns, contributed to this weak correlation and the lack of statistical significance at this range. This was also in concordance with a higher mean bias (LOA) in the hyperglycemic range −32.9 (±124 mg/dL) compared to the hypoglycemic range −9.5 (±46.5 mg/dL) [Figure 4]. For those suspected of hyperglycemia (plasma sugar >150 mg/dL) from the POC glucose screening test, BGA could demonstrate a high accuracy of hyperglycemic diagnosis (AUC = 0. 94). However, due to weak correlation and higher mean bias interpretation of high plasma levels with venous BGA in newborn infants, this should be considered with caution.

BGA and CL are both calibrated against the same hexokinase glucose-6-phosphate dehydrogenase reference method. However, the BGA measures glucose molality and, on average, gives results slightly higher than that of the plasma concentration.[18] In this study, the BGA gave results that were overestimated to CL in all ranges of glucose levels and hypoglycemic and hyperglycemic ranges of glucose, with a mean bias of −9.5, −11.1, and −32.9 mg/dL in the BA plot, respectively. A significant contribution between these two particular methods may vary due to the fact of the time delay in the processing of the CL method. In addition, the sodium fluoride (NaF) sample tube of CL does not completely prevent red cell glycolysis, which reduces glucose levels in delayed samples sent to a distant CL:[19] glucose levels can fall as much as 5–6 mg/dL during this period.[20] Moreover, this study demonstrated higher mean bias between methods, as compared with Peet et al.[15] (5.4 mg/dL) and Newman et al. (1.8 mg/dL), respectively.[14] However, in this study, the mean bias (−9.5 mg/dL) was within 96.6% of paired values: meeting the LOA. The differences could be from the number of the study population, and differences in BGA manufactures. This study could be used to design a BGA method for use as a reliable test for infants whose POC glucose screening tests are suspected of having hypoglycemia. The results demonstrated a high specificity and ROC of the test, with a strong effect of positive likelihood ratio, especially for hypoglycemic detection (37.5): AUC 0.88.

This study has several strengths. First, the data were collected prospectively from venous blood sampling, for which the paired samples were collected at the same time. The glucose values were obtained in infants from the NICU, with high risk of abnormal glucose values.[21] Second, the sample size of hypoglycemic infants (50/145, 34.5%) was higher than in previous neonatal studies.[15] Third, this study excluded contaminated samples, such as nonphysiologic serum from vascular line contamination (sodium level is <120 mEq/L and potassium levels of <3 or >8 mEq/L). In addition, all values of paired glucose measurements in this study could be applied to general practice in NICU settings. However, this study also had some limitations, such as the sample size was too small in infants within hyperglycemia ranges (18/145, 12.4%). Hence, a larger sample size would be required to improve the power. The other limitation was the potential for discrepant results between the BGA and CL plasma venous samples, due to glucose consumption during sample transport to the CL creating overestimation of BGA to CL. This study aimed to reduce this potential error by using a time frame cutoff of 1 h.


In general practice, within NICU settings where normoglycemic and hypoglycemic ranges are being targeted, there is a statistical agreement between venous blood glucose measured using CL and BGA. Venous BGA glucose could be used as a reliable test instead of CL, which provides rapid results, and utilizes easy to use equipment, which is available in the NICU, with good correlation. In addition, these results enhance rapid decision management, so as it is made in a timely manner. No similar correlation was found in the hyperglycemic range; therefore, BGA should be interpreted with caution.

Financial support and sponsorship

This study was financially supported by the Faculty of Medicine, Prince of Songkla University, Hat Yai, Songkhla, Thailand.

Conflicts of interest

There are no conflicts of interest.


1Devaskar SU, Garg M. Disorders of carbohydrate metabolism in the neonate. In: Martin RJ, Fanaroff AA, Walsh MC, editors. Fanaroff and Martin's Neonatal-Perinatal Medicine Diseases of the Fetus and Infant. 10th ed. Philadelphia: Saunders; 2015. p. 1435-60.
2Werny D, Taplin C, Bennett JT, Pihoker C. Disorders of carbohydrate metabolism in the neonate. In: Gleason CA, Juul SE, editors. Avery's Diseases of the Newborn. 10th ed. Philadelphia: Elsevier; 2018. p. 1403-16.
3Lucas A, Morley R, Cole TJ. Adverse neurodevelopmental outcome of moderate neonatal hypoglycaemia. BMJ 1988;297:1304-8.
4Committee on Fetus and Newborn, Adamkin DH. Postnatal glucose homeostasis in late-preterm and term infants. Pediatrics 2011;127:575-9.
5Sunehag AL, Haymond MW. Glucose extremes in newborn infants. Clin Perinatol 2002;29:245-60.
6Narvey MR, Marks SD. The screening and management of newborns at risk for low blood glucose. Paediatr Child Health 2019;24:536-54.
7Beardsall K. Measurement of glucose levels in the newborn. Early Hum Dev 2010;86:263-7.
8Ho HT, Yeung WK, Young BW. Evaluation of “point of care” devices in the measurement of low blood glucose in neonatal practice. Arch Dis Child Fetal Neonatal Ed 2004;89:F356-9.
9Wei H, Lan F, He Q, Li H, Zhang F, Qin X, et al. A comparison study between point-of-care testing systems and central laboratory for determining blood glucose in venous blood. J Clin Lab Anal 2017;31:e22051.
10Woo HC, Tolosa L, El-Metwally D, Viscardi RM. Glucose monitoring in neonates: Need for accurate and non-invasive methods. Arch Dis Child Fetal Neonatal Ed 2014;99:F153-7.
11Inoue S, Egi M, Kotani J, Morita K. Accuracy of blood-glucose measurements using glucose meters and arterial blood gas analyzers in critically ill adult patients: Systematic review. Crit Care 2013;17:R48.
12Liang Y, Wanderer J, Nichols JH, Klonoff D, Rice MJ. Blood gas analyzer accuracy of glucose measurements. Mayo Clin Proc 2017;92:1030-41.
13Mesotten D, Joosten K, van Kempen A, Verbruggen S, ESPGHAN/ESPEN/ESPR/CSPEN Working Group on Pediatric Parenteral Nutrition. ESPGHAN/ESPEN/ESPR/CSPEN guidelines on pediatric parenteral nutrition: Carbohydrates. Clin Nutr 2018;37:2337-43.
14Newman JD, Pecache NS, Barfield CP, Balazs ND. Point-of-care testing of blood glucose in the neonatal unit using the AVL Omni 9 analyser. Ann Clin Biochem 2002;39:509-12.
15Peet AC, Kennedy DM, Hocking MD, Ewer AK. Near-patient testing of blood glucose using the Bayer Rapidlab 860 analyser in a regional neonatal unit. Ann Clin Biochem 2002;39:502-8.
16Kuwa K, Nakayama T, Hoshino T, Tominaga M. Relationships of glucose concentrations in capillary whole blood, venous whole blood and venous plasma. Clin Chim Acta 2001;307:187-92.
17Bland JM, Altman DG. Agreement between methods of measurement with multiple observations per individual. J Biopharm Stat 2007;17:571-82.
18ETTNO Investigators. The 'effects of transfusion thresholds on neurocognitive outcome of extremely low birth-weight infants (ETTNO)' study: Background, aims, and study protocol. Neonatology 2012;101:301-5.
19Chan AY, Swaminathan R, Cockram CS. Effectiveness of sodium fluoride as a preservative of glucose in blood. Clin Chem 1989;35:315-7.
20Gupta S, Gupta AK, Verma M, Singh K, Kaur A, Chopra B, et al. A study to compare the plasma glucose levels obtained in sodium fluoride and citrate buffer tubes at a tertiary care hospital in Punjab. Int J Appl Basic Med Res 2016;6:50-3.
21Balion C, Grey V, Ismaila A, Blatz S, Seidlitz W. Screening for hypoglycemia at the bedside in the neonatal Intensive Care Unit (NICU) with the Abbott PCx glucose meter. BMC Pediatr 2006;6:28.