Abstract
Background:
Hypoglycemia in infants is common, is difficult to recognize, and may lead to permanent neurologic impairment. Low glucose concentrations and high hematocrits in newborns pose significant analytic challenges for whole blood glucose meters.
Objective/Methods:
Three Bayer glucose monitoring systems were evaluated using 211 blood samples from 162 neonates (age range 5 hours to 29 days, median age 3 days). Hematocrit and whole blood glucose were determined in heparinized whole blood, and plasma glucose was determined using the Roche Cobas® 6000. Accuracy was evaluated against plasma concentrations using ISO 15197:2013 and CLSI POCT 12-A3 criteria.
Results:
Glucose imprecision on the Cobas system was 1.8-2.6% (CV) from 26-610 mg/dL. Imprecision across all meter systems was 2.8% (CV) at 130 mg/dL. Glucose concentrations, hematocrit, and total bilirubin ranged from 20-150 mg/dL, 18 -75%, and 0.5-19.6 mg/dL, respectively. Linear regression analysis of whole blood versus plasma for the 3 combined systems yielded an average slope of 1.06 and correlation coefficient greater than 0.980. Bias between the Contour and Cobas was not significantly correlated with hematocrit. Greater than 99% of meter results were within 15 mg/dL and 20% of plasma results at glucose concentrations ≤ 75 and > 75 mg/dL, respectively. Of meter results, 97% were within 12.5 mg/dL of plasma results at concentrations ≤ 100 mg/dL, while 96% of meter results were within 12.5% of plasma at concentrations > 100 mg/dL.
Conclusions:
The Bayer CONTOUR Blood Glucose Monitoring Systems exceed ISO 15197:2013 and CLSI criteria in neonatal blood samples.
Whole blood glucose meters are utilized in a variety of settings such as homes, physician offices, and hospital settings. While not recommended for the diagnosis of diabetes, such meters are sanctioned for monitoring response to therapy. In addition, whole blood glucose monitors are increasingly employed in so-called “tight glycemic control” protocols which prescribe the maintenance of circulating glucose concentrations within a narrow window (eg, 80-100 mg/dL) in nondiabetic, critically ill patients.1-3 The clinical utility of these protocols remains highly controversial. Such new applications of glucose meters have led to increased regulatory scrutiny and lead to new FDA guidance that proposes more rigorous accuracy and precision requirements for these in vitro diagnostic devices. 4
Glucose monitoring in newborns poses other significant challenges to whole blood glucose monitoring systems. Neonatal blood samples typically contain lower glucose concentrations and have higher hematocrits (55-65%) than blood from adult males (40-50%) or females (35-45%). Imprecision of glucose meters has historically been greatest at glucose concentrations common in newborns (<70 mg/dL). High hematocrits necessitate a robust correction for sample water content. Despite these challenges, accurate and timely recognition of severe neonatal hypoglycemia (<40 mg/dL) is essential to prevent long-term neurocognitive deficits.5,6
The Bayer Contour system utilizes FAD-linked glucose dehydrogenase and amperometry to determine a plasma equivalent glucose concentration using whole blood. We challenged the accuracy, imprecision, and hematocrit correction of 3 strip/meter combinations, the Contour NEXT, NEXT EZ, and PLUS, in a population of neonates at an academic tertiary care pediatric medical center. Performance was judged against current standards set by International Organization for Standardization (ISO) and the Clinical and Laboratory Standards Institute (CLSI).7,8
Methods
Patients and Samples
The objective of this study is to define the analytic accuracy of whole blood glucose meters. Per protocol approved by the Washington University Human Research Protection Office, no patient identifiers or indications for phlebotomy were retained with the laboratory data. 211 heparinized capillary blood specimens from 162 infants up to 30 days of age were obtained over a 10 week period as part of routine care in the Barnes-Jewish Hospital nurseries (75%) or St. Louis Children’s Hospital (25%). No patient supplied more than 2 samples for the study. An aliquot of less than 50 µL was removed for whole blood glucose analysis and hematocrit determination. Centrifugation of the remaining specimen commenced within 10 minutes of aliquot removal. The resulting plasma was used for glucose analysis in singlicate. When determined as part of clinical care, total plasma bilirubin concentration was also captured and recorded. A complete patient demographic summary is presented in Table 1.
Patient Specimen Characteristics.
Analytical
All procedures were carried out by certified medical technologists. Hematocrit was determined using a Statspin microhematocrit rotor (Iris Sample Processing, Westwood, MA). Plasma total bilirubin was determined using a modified diazo method on the Cobas 6000 (Roche Diagnostics, Indianapolis, IN). Plasma glucose was determined using a coupled hexokinase procedure on a Cobas 6000 Chemistry System. Performance of the laboratory glucose method (Cobas) was monitored daily at 2 concentrations (85 and 280 mg/dL) and weekly with 6-point control sera spanning a concentration range from 26 to 610 mg/dL. Whole blood glucose was determined randomly from a pool of 30 meters, 10 Contour NEXT, 10 Contour NEXT EZ, and 10 Contour PLUS. The chemistry and quantitation algorithms of each meter type are identical. The meter types differ only in size, appearance, screen size, and data management capability. The order of meter application was rotated every 10 specimens and 3 strip lots were assigned for each specimen such that 1 lot was used for half of the specimens and the remaining 2 lots were applied to 25% each. Integrity and imprecision of each meter type used during the study was monitored weekly with a control solution containing 130 mg/dL glucose. Laboratory temperature ranged from 25-27ºC and relative humidity varied between 17-35% throughout the study.
Data Analysis
Descriptive statistics, linear least squares regression, and ANOVA analyses were generated using Excel 2010 (Microsoft, Redmond, WA).
Results
Imprecision of Glucose Measurement
Data summarizing the imprecision of glucose measurement in plasma and whole blood are contained in Table 2. Pooled estimates of meter imprecision were derived from weekly liquid QC analyses on each of the 10 NEXT, NEXT EZ, or PLUS meter types (100 replicates each). Meter imprecision ranged from 2.3-3.2% (CV) at 130 mg/dL. A pooled estimate of plasma glucose imprecision was derived from 48 replicates of a 6-level control performed on 2 instruments over the course of the study. Imprecision ranged from 1.8-2.6% (CV) across a concentration range of 26-610 mg/dL.
Imprecision of Glucose Measurement.
Sample Characteristics
A slight majority of specimens (53%) were derived from male neonates. The average glucose concentration in this population was 70 mg/dL (range 23-150 mg/dL). A significant majority of specimens were from infants less than 1 week of age including 25 from patients less than 24 hours of age. Specimens were obtained from healthy infants in the well-baby nursery (27%), mild to moderately ill infants in the special care nursery (48%), as well as critically ill infants in the intensive care units of St. Louis Children’s Hospital (16%). Samples contained a broad range of hematocrit (18-75%) and total bilirubin concentrations (0.5-19.6 mg/dL).
Accuracy of Glucose Measurement
Glucose concentrations determined by all 3 meter systems were highly correlated with plasma measurements (r = .983). A Scatter plot of these data is shown in Figure 1. Linear regression of glucose concentrations across the 3 meter platforms versus plasma glucose is given by the following equation: contour glucose (mg/dL) = 1.06 × plasma glucose (mg/dL) – 1.1. Independent regression analyses for each of the 3 meter systems yielded virtually identical results. Bias between the plasma and meter results across all 3 meter platforms is shown in Figure 2 and averaged 3.2 mg/dL (range −9.0 to 18.4). In percentage terms, average bias between meter and plasma results was 4.5% (range −15% to 24%).
Regression analysis of contour whole blood glucose with plasma concentrations. Whole blood glucose concentrations from the Contour Next (○), XT/NEXT EZ (Δ), and PLUS (□) are displayed against corresponding plasma concentrations. Dotted lines indicate ISO 15197:2013 accuracy criteria. Regression equations in mg/dL units: (1) [NEXT glucose] = 1.06 [plasma glucose] – 0.9, r = .983, Syx = 3.8; (2) [XT/NEXT EZ glucose] = 1.06 [plasma glucose] – 1.6, r = .983, Syx = 3.8; (3) [PLUS glucose] = 1.06 [plasma glucose] – 0.7, r = .973, Syx = 3.8.
Bias of contour meter results versus plasma glucose. Absolute (left) and percentage (right) bias of Contour NEXT (○), XT/NEXT EZ (Δ), and PLUS (□) glucose concentrations versus plasma glucose concentration are displayed. Dotted lines indicate acceptance criteria specified by ISO 15197:2013.
Bias between meter and plasma glucose was independent of meter type by 1-way ANOVA (P = .08), but there were differences noted in the degree of meter bias in different patient populations. Average concentration bias between meter and plasma glucose was 2.0 mg/dL (range −4.9 to 11.9), 4.1 mg/dL (range −2.0 to 15.7), and 4.3 mg/dL (range −5.0 to 18.4) in samples from the well-baby nursery (n = 174), special care nursery (n = 303), and the neonatal intensive care unit (n = 102), respectively. The residual glucose bias in these patient groups was statistically significant as determined by single factor ANOVA (P = 10−9). Mean meter versus plasma bias in the well-baby nursery was statistically lower than that observed in both the special care nursery (P = 10−9) and intensive care units (P = 10−4) by 2-tailed t test.
Accuracy of meter results was also assessed against both ISO 15197 and CLSI POCT 12-A3 criteria. Combined results from all 3 meters are presented in Table 3. When assessed against ISO specifications, greater than 99% of meter results were within 15.0 mg/dL and 15% at glucose concentrations < 100 or ≥ 100 mg/dL, respectively. When assessed against 2 sets of current CLSI specifications, greater than 95% of meter results were within 12.5 mg/dL and 12.5% at concentrations below/above 100 mg/dL, respectively, while at glucose concentrations below/above 75 mg/dL, greater than 99% meter results were within 15 mg/dL and 20%, respectively. Contour results meet or exceed requirements for all CLSI and ISO standards.
Accuracy of Meter Glucose Concentrations.
Influence of Bilirubin and Hematocrit on Meter Glucose Measurement
The most significant challenge to accurate whole blood glucose measurement in neonates is the high variation of sample water content dictated by hematocrit. In this study, the Contour meters were challenged by sample hematocrit ranging from 18-75%. The bias between each of the 3 meters as a function of sample hematocrit is shown in Figure 3 and yielded correlation coefficients of 0.0626 (P = .36), 0.0156 (P = .82), and 0.0132 (P = .85) for the NEXT, NEXT EZ, and PLUS systems, respectively. The performance of the 3 Contour meter systems was independent of sample hematocrit.
Effect of hematocrit on bias between contour and plasma glucose concentration. Absolute (left) and percentage (right) bias of Contour NEXT (○), XT/NEXT EZ (Δ), and PLUS (□) glucose concentrations versus plasma glucose concentration as a function of sample hematocrit are displayed.
Bilirubin is likewise an important variable in the neonate population. In the first week after birth, bilirubin concentrations may reach 15-20 mg/dL necessitating dietary changes, phototherapy, or exchange transfusion in extreme circumstances. Hyperbilirubinemia typically resolves by 1 month of age. There was a slight but statistically significant correlation of meter-plasma glucose bias as a function of total plasma bilirubin (r = –.283, P < .001), but the slope of this relationship (–.0009) has little clinical significance. An 18.0 mg/dL increase in bilirubin concentration, for example, decreases the average difference between meter and plasma glucose by 5.0 mg/dL. This change in glucose concentration represents only a small fraction of the observed 30 mg/dL range of residual bias observed across all patient populations and meter types.
Detection of Hypoglycemia
While there is far from consensus on the subject, a common threshold for treatment in neonatal hypoglycemia is 40 mg/dL. The capacity of the 3 Bayer glucose monitoring systems to accurately detect clinically significant hypoglycemia was examined. There were 24 plasma glucose values less than 40 mg/dL in the study. Using a meter cutoff of 40 mg/dL, meters detected all of these specimens (100% sensitivity). Three specimens among the 609 with plasma glucose ≥ 40 mg/dL were inaccurately classified as hypoglycemic at the same cutoff value yielding specificity of 99.5%. In this data set, the range of residual bias between glucose concentrations 30 and 50 mg/dL varied from −5.0 to +5.0 mg/dL. By comparison, CLSI and ISO criteria would allow variation of ±12.5 and 15.0 mg/dL, respectively, in this concentration range.
Discussion
Controversy continues to surround the requirements for monitoring glucose using near patient whole blood instrumentation. More stringent performance criteria proposed by the FDA are focused on performance characteristics in hospital applications such as tight glycemic control regimens. In this study, each meter system exceeded existing CLSI and ISO accuracy criteria but fell short of the newly proposed FDA criteria that require 99% of meter results agree within 10% of plasma values ≥ 70 mg/dL or within 7.0 mg/dL at concentrations < 70 mg/dL. Less scrutiny has been placed on performance characteristics of glucose meters near the hypoglycemic threshold (eg, 40 mg/dL) that might prompt clinical intervention in neonates. Thorough and accurate performance data at such key clinical decision points are crucial to efficiently screen and capture all infants for which intervention to raise circulating glucose is indicated. At present, any suspicious screening results should still be confirmed with definitive plasma measurements.
Confounders of accurate whole blood glucose measurement in neonates are many. Those already alluded to include hematocrit and bilirubin but numerous hemodynamic, endocrine, and therapeutic covariates may impact electrochemical glucose measurements differently than the optical techniques employed in plasma glucose determination. The present study did detect significantly greater meter bias in babies undergoing treatment in the special care nursery or intensive care unit than babies with uncomplicated hospital courses. Despite the bias in acutely ill children, the Contour systems were very accurate (100% sensitivity, 99.5% specificity) at a concentration of 40 mg/dL. A practical clinical cutoff, however, must account for positive bias (5 mg/dL) near this critical concentration that might cause hypoglycemia to go undetected and untreated. Using a cutoff concentration of 45 mg/dL, the Contour blood glucose monitoring systems in this study reliably identified neonatal hypoglycemia. However, given the modest number of hypoglycemic infants in this study, further validation of this screening cutoff is warranted.
Footnotes
Abbreviations
CLSI, Clinical and Laboratory Standards Institute; FAD, flavin adenine dinucleotide; FDA, US Food and Drug Administration; ISO, International Organization for Standardization.
Declaration of Conflicting Interests
The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: DAS, SP, MP, and JF are employees of Bayer Healthcare, Diabetes Care.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by Bayer Healthcare, Diabetes Care.