Evaluation of the impact of hematocrit and other interference on the accuracy of hospital-based glucose meters

DIABETES TECHNOLOGY & THERAPEUTICS
Volume 10, Number 2, 2008
Mary Ann Liebert, Inc.
DOI: 10.1089/dia.2007.0257
Evaluation of the Impact of Hematocrit and Other Interference on the Accuracy of Hospital-Based BRAD S. KARON, M.D., Ph.D.,1 LAURIE GRIESMANN, A.S.,1 RENEE SCOTT, B.S.,1 SANDRA C. BRYANT, M.S.,2 JEFFREY A. DUBOIS, Ph.D.,3 TERRY L. SHIREY, Ph.D.,3 STEVEN PRESTI, B.S.,3 and PAULA J. SANTRACH, M.D.1 ABSTRACT
Background: Most glucose meter comparisons to date have focused on performance specifi-
cations likely to impact subcutaneous dosing of insulin. We evaluated four hospital-based glu-cose meter technologies for accuracy, precision, and analytical interferences likely to be en-countered in critically ill patients, with the goal of identifying and discriminating glucose meterperformance specifications likely to impact intensive intravenous insulin dosing.
Methods: Precision, both within-run and day-to-day, was evaluated on all four glucose me-
ters. Accuracy (bias) of the meters and analytical interference were evaluated by comparing re-sults obtained on whole blood specimens to plasma samples obtained from these whole bloodspecimens run on a hexokinase reference method.
Results: Precision was acceptable and differed little between meters. There were significant
differences in the degree to which the meters correlated with the reference hexokinase method.
Ascorbic acid showed significant interference with three of the four meters. Hematocrit also af-fected the correlation between whole blood and plasma hexokinase glucose on three of the fourglucose meters tested, with the magnitude of this interference also varying by glucose metertechnology.
Conclusions: Correlation to plasma hexokinase values and hematocrit interference are the
main variables that differentiate glucose meters. Meters that correlate with plasma glucose mea-sured by a reference method over a wide range of glucose concentrations and minimize the ef-fects of hematocrit will allow better glycemic control for critically ill patients.
INTRODUCTION
travenous insulin; however, target glucose con-centrations are narrower for this patient popu- SEVERAL RECENT STUDIES have suggested that lation than for patients with diabetes using
tight glucose control (maintenance of blood handheld meters to dose subcutaneous insulin.
glucose between 80 and 110 mg/dL), accom- In addition, patients in the intensive care unit plished via intensive intravenous insulin ther- (ICU) are on multiple medications, and often apy, decreases mortality in critically ill pa- have abnormal hematocrit and/or oxygen ten- tients.1,2 Use of handheld glucose meters allows sion, all of which may affect the performance for rapid treatment decisions for patients on in- 1Department of Laboratory Medicine and Pathology and 2Division of Biostatistics, Mayo Clinic College of Medi- cine, Rochester Minnesota; and 3Nova Biomedical, Waltham, Massachusetts.
KARON ET AL.
Oxygen tension and pH may affect a limited correlation to definitive methods that use mass number of glucose meter technologies.3–5 Var- spectrometry.11 Four glucose meter technolo- ious medications used in the critical care set- gies were chosen as representing the major ting and patient hematocrit have been found to hospital-based technologies currently avail- affect the performance of almost all glucose me- able: Accu-Chek® Inform® (Roche Diagnos- ter technologies available.6–9 Multiple studies tics), which uses a glucose dehydrogenase- have found that glucose meters demonstrate a based amperometric strip; Precision PCx® positive bias at low hematocrit and a negative (Abbott Diabetes, Alameda, CA), which uses bias at high hematocrit, regardless of the me- glucose dehydrogenase amperometric detec- ter technology used.7–9 One new glucose meter tion; SureStepFlexx® (LifeScan, Milpitas, CA), technology with hematocrit measurement and which uses a photometric glucose oxidase de- correction was recently introduced to address tection system; and StatStrip (Nova Biomed- ical, Waltham, MA), which uses a modified glu- Besides analytical interference, the other ma- cose oxidase-based amperometric test system jor concern in monitoring tight glycemic control in the ICU is the accuracy of glucose measure-ment when tighter ranges of glucose control are desired. Since hexokinase glucose methods Within-run precision. Venous heparinized have been found to be suitable for use as refer- whole blood was drawn 12–24 h in advance of ence methods for glucose determination,11 mul- performing the study. Aerated blood was di- tiple studies have examined the correlation be- vided into three 2-mL aliquots, which received tween glucose meter whole blood and plasma different volumes of a concentrated glucose so- hexokinase glucose. The degree to which glu- lution such that the aliquots had 20–60, cose meters correlate with plasma hexokinase 200–300, and 450–550 mg/dL glucose. Each ali- measurement of glucose varies tremendously quot was then tested 20 times on each meter.
between glucose meter technologies,12 and cor-relation with laboratory hexokinase measure- Day-to-day precision. Two levels of control ment in the hypoglycemic and hyperglycemic material manufactured by each glucose meter ranges is poor with most meters currently avail- vendor were tested in duplicate, three times able.13 Thus there is still significant concern per day for 5 days (total of 30 readings for each about the use of glucose meters for manage- control) on each meter. The controls used in- ment of tight glycemic control in the ICU.14 cluded: Nova Biomedical StatStrip glucose The aim of the current study was to compare controls (lot 0413911081, range 48–78 mg/dL; four hospital-based glucose meter technologies lot 0414611083, range 247–317 mg/dL), J & J for accuracy compared to a reference plasma LifeScan SureStepFlexx controls (lot 6C1F64, hexokinase method, determine drug interfer- range 35–59 mg/dL; lot 6C4F67, range 270– ences, and measure the effect of hematocrit on 404 mg/dL), Abbott PCx controls (lot 17692, the correlation between glucose meter and hex- range 34–64 mg/dL; lot 17697, range 224–374 okinase glucose over a wide range of glucose mg/dL); and Roche Accu-Chek controls (lot 60250, range 46–76 mg/dL; lot 60251, range287–389 mg/dL).
RESEARCH DESIGN AND METHODS
Patient specimens for method correlations rinized (23.5 units/mL) arterial whole blood The reference assay was plasma glucose us- specimens obtained for blood gas analyses.
ing the hexokinase method on the Roche Inte- These specimens were obtained within 90 min gra 400 Analyzer (Roche Diagnostics, Indi- of collection in the ICU. Hematocrit (%) values anapolis, IN). This was chosen as the reference for these specimens were calculated from he- moglobin values obtained from the ABL 725 been found to be suitable for use as reference blood gas analyzer (Radiometer, Westlake, methods for glucose determination with close OH). One hundred thirty-three whole blood COMPARISON OF HOSPITAL GLUCOSE METERS
specimens (unspiked) were analyzed by the For the studies using variable hematocrit lev- four handheld glucose analyzers (assembly- els, 30 mL of fresh whole blood from a single line set up) and immediately (within 5 min) donor was allowed to sit at room temperature spun down in order to obtain the plasma sam- for 12–24 h before division into three aliquots ples for analysis on the Roche Integra 400 An- of 5 mL per specimen. The three 5-mL aliquots alyzer. Specimens from 52 additional patients were each brought to a different concentration were spiked with various volumes of glucose of glucose using the concentrated glucose so- concentrate (20,000 mg/dL) and analyzed sim- lution. Each of these three 5-mL aliquots were ilarly to the unspiked specimens in an effort to further divided into five aliquots of 1 mL. Cen- extend the glucose range for method compari- trifugation, using a Fisher Scientific (Pitts- son purposes. The study design was approved burgh, PA) Mini-Centrifuge, and plasma ad- by the Mayo Clinic Institutional Review Board.
justments (taking some plasma from one tubeand putting it into another) resulted in five ali-quots with different hematocrit (%) levels for Interference and hematocrit studies from donor each concentration of glucose. All specimens were rocked for at least 10 min and then rapidly For the interference studies, freshly drawn, analyzed, within 10 min, on each of the glucose heparinized venous blood drawn from healthy meters (assembly-line fashion) in replicates of donors was allowed to sit at room temperature six for each glucose meter and strip device.
for 12–24 h before concentrated solutions of Hematocrit (%) values were obtained for each glucose and/or interfering substances were of the prepared specimens in this study using added. The concentrated solutions of glucose and the interfering substances were gravimet- trifuge (Separation Technology, Altamonte rically prepared. These concentrates were pre- Springs, FL). Specimens were centrifuged in or- pared as follows: 20,000 mg/dL glucose in wa- der to remove a plasma sample for duplicate analysis on the Roche Integra 400 Analyzer.
1,000 mg/dL ascorbic acid in water, 10,000mg/dL D(ϩ)-maltose monohydrate in water, 2,000 mmol/L lactate in water, 3,000 mmol/Lbeta-hydroxybutyrate in water, 12,000 IU/dL For interference experiments, results are ex- heparin in water, and 100 mg/dL epinephrine pressed as mean change from baseline glucose in water. Immediately prior to each interfer- (meter glucose with interfering substance – me- ter glucose at baseline) in mg/dL for experi- spiked with glucose concentrate bringing them ments where glucose concentration was ad- into predetermined ranges. This was followed justed to less than 100 mg/dL. Mean change by division of each of these aliquots into three from baseline glucose in percent [(meter glu- volumes, two of which were then spiked with cose with interfering substance – meter glucose the interfering material. Concentrations of each at baseline)/meter glucose at baseline ϫ 100] interferant tested were chosen to reflect (at was used when glucose was adjusted to Ͼ100 maximum concentration tested) five to 10 times mg/dL (n ϭ 6 all experiments). A clinically sig- the therapeutic drug level, similar to what has nificant interference effect was defined as any been described previously.6 Lactate and beta- concentration of interferant that changed the hydroxybutyrate were tested at concentrations mean baseline (no interfering substance added) that would reflect extreme acidosis in critically glucose value by more than 10 mg/dL (glucose ill patients. All samples were rocked for 10–20 Ͻ100 mg/dL) or 10% (glucose Ͼ100 mg/dL).
min. Glucose concentration was then analyzed in each specimen with six strips from each of manipulated, the mean glucose difference (me- the four measuring technologies. After the test- ter glucose – reference glucose, for glucose ing on the strips was completed, the specimens Ͻ100 mg/dL) or glucose percent difference were immediately centrifuged and sent for du- [(meter glucose – reference glucose)/reference plicate analysis on the Roche Integra 400 Ana- glucose ϫ 100), for glucose Ͼ100 mg/dL] was calculated for each meter technology. Statisti- KARON ET AL.
TABLE 1. WITHIN-RUN AND DAY-TO-DAY PRECISION FOR GLUCOSE METERS, IN PERCENT CV n ϭ number of replicates tested.
cal significance of the effect of hematocrit was glucose value was 168 mg/dL for the entire assessed using the two-sided unpaired t test, sample set (n ϭ 185), and the range of glucose comparing mean glucose difference or mean values covered was 39–574 mg/dL. Linear re- glucose percent difference between the lowest gression analysis demonstrated a slope of 0.90 and an intercept of Ͻ10 mg/dL glucose for theStatStrip and the Accu-Chek meters, while thePCx and SureStepFlexx meters had lower slopes and higher intercepts (Table 2). Medianbias from the reference method (meter value minus reference value) was of lower absolutemagnitude for the StatStrip and SureStepFlexx Within-run and day-to-day precision assessed meters than for either the Accu-Chek or PCx at multiple glucose levels as described above re- meters (Table 2). There were also significantly sulted in coefficient of variation (CV) values of less than 5% for all meters tested with the ex- method on the StatStrip (170 of 185) compared ception of day-to-day precision at low glucose to the SureStepFlexx (134 of 185), Accu-Chek on the PCx meter, which was 5.1% (Table 1).
(127 of 185), or PCx (79 of 185) methods.
Exclusion of the 52 samples spiked with ex- ogenous glucose resulted in slightly higher Correlation between glucose meter results slopes and lower intercepts for all four me- and the plasma hexokinase reference method ters (Table 2). Exclusion of the spiked sam- was performed by analyzing 133 fresh lithium ples also resulted in median bias that was heparin arterial blood specimens and an addi- of similar absolute magnitude for the Accu- tional 52 lithium heparin arterial blood speci- Chek, PCx, and SureStepFlexx. Median bias mens that were spiked with exogenous glucose on the StatStrip was smaller than median bias for a total of 185 specimens. Mean reference on the three other meter technologies whether TABLE 2. CORRELATION DATA FOR GLUCOSE METERS VERSUS REFERENCE PLASMA HEXOKINASE METHOD (N ϭ 185 OR 133) The n ϭ 185 data set includes 133 unaltered clinical specimens and 52 specimens spiked with exogenous glucose; then n ϭ 133 data set includes only unaltered clinical specimens.
COMPARISON OF HOSPITAL GLUCOSE METERS
or not the 52 spiked samples were excluded and highest hematocrit tested. Changes in mean glucose percent difference between low-est and highest hematocrit tested were also sta- Effect of hematocrit on glucose meter accuracy tistically significant (P Ͻ 0.001) for the Accu-Chek, PCx, and SureStepFlexx technologies at higher glucose levels and marginally signifi- ally adjusting the hematocrit of donor sodium cant (P ϭ 0.0203) at a glucose concentration of heparin blood at glucose concentrations ad- 483 mg/dL for the StatStrip (Fig. 1).
justed to 54, 247, and 483 mg/dL. At low glu-cose (54 mg/dL), mean glucose difference Effect of hematocrit on percentage bias in changed by more than 10 mg/dL between the lowest and highest hematocrit values tested onthe PCx and SureStepFlexx meters. At higher To further investigate the effect of hematocrit (247 and 483 mg/dL) glucose concentrations, on glucose meter accuracy, glucose meter per- the Accu-Chek, PCx, and SureStepFlexx meters cent bias versus hematocrit was plotted for the demonstrated greater than 10% change in the 133 patient specimen correlation data set de- mean glucose percentage difference between scribed previously (Fig. 2). There is a clear trend the lowest and highest hematocrit values (Fig.
for negative bias associated with increasing 1). At low glucose changes in mean glucose dif- hematocrit for the PCx and SureStepFlexx me- ference were statistically significant for the PCx ters (Fig. 2). Regression analysis on this data set and SureStepFlexx (P Ͻ 0.001) between lowest resulted in slopes and intercepts (percent bias erence (%)
erence (mg/dL)
Mean glucose diff
Mean glucose diff
Hematocrit (%)
Hematocrit (%)
(a) Mean glucose difference (meter glucose Ϫ
erence (%)
reference glucose) and (b) and (c) mean glucose percent
difference [(meter glucose – reference glucose)/reference
glucose ϫ 100) as a function of hematocrit at glucose con-
centrations of (a) 54 mg/dL, (b) 247 mg/dL, and (c) 483
mg/dL. Each point represents the mean Ϯ standard de- viation of the mean glucose difference or mean glucose Mean glucose diff
Hematocrit (%)
KARON ET AL.
Hematocrit (%)
Hematocrit (%)
FIG. 2a, c.
vs. hematocrit) that were significantly different dium heparin blood with glucose concentra- from zero (P Ͻ 0.0001) for the PCx and tions that had been adjusted to 44, 145, 244, and SureStepFlexx meters. For the StatStrip and 341 mg/dL, respectively. No concentration of Accu-Chek meters, the slope of percent bias acetaminophen changed the mean baseline (no versus hematocrit was not significantly differ- acetaminophen added) glucose level by more ent from zero (P Ͼ 0.05). The calculated slopes, than 10 mg/dL (for experiments performed at intercepts, and correlation coefficient (r2) val- 44 mg/dL glucose) or 10% (for experiments ues for the regression of percent bias versus performed at 145, 244, and 341 mg/dL, re- hematocrit are shown in Figure 2. Inclusion of spectively) on any of the meters. Thus acet- the 52 specimens spiked with exogenous glu- aminophen did not produce a clinically signif- cose did not change the slopes or intercepts for icant interference on any of the four meter percent bias versus hematocrit but significantly decreased the correlation coefficient (r2) for Lactate (final sample concentrations of 0, 10, and 20 mmol/L) was added to donor sodiumheparin blood with glucose concentration ad- Effect of interfering substances on glucose meter justed to 29, 143, 255, and 357 mg/dL, respec- tively. Similar to acetaminophen, lactate (0–20 Acetaminophen (final sample concentrations mmol/L) did not change mean baseline glu- of 0, 5, and 10 mg/dL) was added to donor so- cose by more than 10 mg/dL (at 29 mg/dL glu- COMPARISON OF HOSPITAL GLUCOSE METERS
Hematocrit (%)
Hematocrit (%)
Glucose percent bias [(meter glucose minus reference glucose)/reference glucose ϫ 100] for the four strip methods are plotted against sample hematocrit values for the 133 patient sample set (no sample manipulation):
(a) StatStrip, (b) Accu-Chek, (c) PCx, and (d) SureStep. Slope and intercept of the best-fit line for percent bias versus
hematocrit and correlation coefficient are also shown.
cose) or 10% (at higher glucose) on any of the trations, ascorbic acid produced a clinically sig- four meter technologies tested. Beta-hydroxy- nificant (Ͼ10%) interference on the Accu-Chek butyrate (final sample concentrations 0, 7.5, and PCx glucose meters (Fig. 3). None of the and 30 mmol/L) also did not significantly af- meter technologies was significantly affected fect any of the four glucose meter technologies when added to donor blood adjusted to 32, 135, Maltose (final sample concentrations of 0, 100, and 200 mg/dL) was added to donor so- Ascorbic acid (final sample concentrations of dium heparin blood with glucose concentra- 0, 5, and 10 mg/dL) was added to donor so- tions adjusted to 40, 109, 208, and 300 mg/dL, dium heparin blood with glucose concentration respectively. Maltose produced a clinically sig- adjusted to 70, 141, 237, and 352 mg/dL, re- nificant interference only on the Accu-Chek spectively. At low glucose (70 mg/dL), ascor- bic acid produced a clinically significant (Ͼ10 threefold, twofold, 1.5-fold, and 1.4-fold in- mg/dL) interference with the Accu-Chek, PCx, creases in mean baseline (no maltose added) and SureStepFlexx glucose meters (Fig. 3). At glucose at 40, 109, 208, and 300 mg/dL glucose higher (141 and 237 mg/dL) glucose concen- KARON ET AL.
e in baseline glucose (%) -10
e in baseline glucose (mg/dL)
Ascorbic acid (mg/dL)
Ascorbic acid (mg/dL)
e in baseline glucose (%) -10
Ascorbic acid (mg/dL)
(a) Change from baseline glucose (meter glucose in presence of ascorbic acid – meter glucose at baseline)
versus concentration of ascorbic acid using a specimen adjusted to 70 mg/dL. (b) and (c) Change in baseline glucose
expressed in percent change [(meter glucose with ascorbic acid minus meter glucose at baseline)/meter glucose at
baseline ϫ 100] using specimens adjusted to (b) 141 mg/dL and (c) 237 mg/dL glucose. Each point represents the
mean of six measurements (strips). The effect of ascorbic acid on the reference method is also shown.
While epinephrine levels up to 1 ␮g/dL had ters correlated with a plasma hexokinase refer- little effect on any of the four glucose meters, ence method differed between meters, as has the reference hexokinase procedure was af- been observed previously.12 The StatStrip and fected significantly (Ͼ10 mg/dL at 32 mg/dL Accu-Chek meter technologies demonstrated glucose and Ͼ10% at the 105 mg/dL glucose the closest correlation with hexokinase plasma level) by epinephrine. Heparin at 20 units/mL glucose based on assessment of the slope and had Ͻ10% effect on any of the technologies intercept calculated by linear regression, while tested at a glucose level of 101 mg/dL.
the StatStrip and SureStepFlexx meters dem-onstrated the lowest absolute median bias(Table 2). Since the StatStrip, Accu-Chek, and DISCUSSION
SureStepFlexx use different measurementtechnologies, it appears that calibration of the Precision of glucose meters was acceptable strips by the individual manufacturers, rather and differed little between meter technologies than measurement technology, impacts the de- (Table 1). The extent to which the glucose me- gree to which whole blood measurement cor- COMPARISON OF HOSPITAL GLUCOSE METERS
relates with laboratory hexokinase methods.
on the Accu-Chek within 10% of the reference Our findings are consistent with one previous hexokinase method, consistent with the results study of the PCx device,15 which found that the obtained from capillary blood samples in pa- slope and intercept of whole blood capillary tients undergoing intravenous insulin therapy.
versus venous plasma (hexokinase) glucose The improved performance of the StatStrip, as were 0.85 and 12 mg/dL, nearly identical to the results obtained in our study using the non- plasma hexokinase assay, should result in fewer insulin dosing errors for patients on both Correlation with the reference method was subcutaneous and intravenous insulin.
adversely impacted by inclusion of samples Hematocrit effect on glucose meter accuracy spiked with exogenous glucose for all meter (correlation with hexokinase plasma values) technologies (Table 2). This may be due to the was examined in two different experiments.
wider range of glucose concentrations covered Using sodium heparin blood pools that were by these experiments, but analytical artifacts or manipulated to obtain hematocrit values be- interferences created by the spiking procedure tween 25% and Ͼ60%, and glucose concentra- cannot be excluded. For this reason laboratory- tions between 54 and 483 mg/dL, it is clear that based evaluation of meter devices should spec- the four meter technologies have differing sen- ify whether samples have been manipulated or sitivity to hematocrit (Fig. 1). The hematocrit spiked, and data summaries for both spiked effect can also be observed in the experiment and nonspiked data sets may be useful. Al- performed with fresh arterial whole blood though laboratory-based experiments are pow- specimens, where a significant trend between erful because of the range of glucose, hemat- hematocrit and percent bias was observed for ocrit, and interfering substance concentrations the PCx and SureStepFlexx meters (Fig. 2).
that can be included, both laboratory and clin- Since hematocrit can vary widely in critically ical evaluation of devices is necessary to get a ill patients, glucose meter technologies that are complete picture of device performance.
insensitive to the effects of hematocrit should The clinical significance of differences be- also improve accuracy and decrease insulin tween whole blood and laboratory plasma glu- cose can be demonstrated by the number of Finally, the effects of various medications samples within 10% or 15% of the reference commonly used in the critical care arena were method. Using Monte Carlo simulation Boyd tested for analytical interference on all the glu- and Bruns16 previously demonstrated that at cose methods, similar to experiments that have 10% total error, 16–45% of sliding-scale insulin been published previously.6 Lactate and beta- doses would be in error, though small dosing hydroxybutyrate had no impact on glucose me- errors would predominate. Larger dosing er- ter performance, similar to previous studies rors were common when total error exceeded that showed no effect of sample pH on most 10–15%.16 Significantly more samples on the glucose meters.5 Acetaminophen, at levels up StatStrip (170 of 185) fell within 10% of the ref- to five to 10 times the therapeutic level, also did not significantly impact the glucose meters.
(127 of 185), PCx (79 of 185), or SureSteppFlexx This differs from one previous report on acet- (134 of 185). In addition, significantly fewer val- aminophen effects,6 though that study used ues on the StatStrip differed by more than 15% higher concentrations of acetaminophen and from the reference method (two of 185) com- was performed on a previous generation of glu- pared to the Accu-Chek (26 of 185), PCx (58 of 185), or SureSteppFlexx (11 of 185) meters.
Ascorbic acid has been reported to interfere One recent study of the Accu-Chek meter in with all glucose meter technologies that have patients on intravenous insulin found that 74% been tested.6 We found that ascorbic acid in- of capillary whole blood samples fell within terfered with each of the glucose meters tested 10% of a reference plasma hexokinase assay, re- with the exception of the StatStrip (Fig. 3). Mal- sulting in common small dosing errors.17 Our tose interference has been reported with glu- study demonstrated 127 of 185 (69%) samples cose dehydrogenase technologies18 and was KARON ET AL.
found to be significant for the Accu-Chek me- 7. Chance J, Li D, Jones K, Dyer K, Nichols J: Technical evaluation of five glucose meters with data manage- In conclusion, we evaluated glucose meter ment capabilities. Am J Clin Pathol 1999;111:547–556.
8. Louie R, Tang Z, Sutton D, Lee J, Kost G: Point-of- correlation with a reference hexokinase method care glucose testing. Arch Pathol Lab Med 2000;124: and analytical interferences likely to be ob- served in critical care patients on four currently 9. Tang Z, Lee J, Louie R, Kost G: Effects of different available hospital glucose meter technologies.
hematocrit levels on glucose measurements with Correlation of whole blood glucose to plasma handheld meters for point-of-care testing. ArchPathol Lab Med 2000;124:1135–1140.
hexokinase reference methods continues to 10. Rao L, Jakubiak F, Sidwell J, Winkelman J, Snyder M: vary between glucose meter manufacturers.
Accuracy evaluation of a new glucometer with auto- Hematocrit had a significant impact on the cor- mated hematocrit measurement and correction. Clin relation between whole blood and plasma glu- cose on most of the meters. The StatStrip glu- 11. Pelletier O, Arratoon C: Precision of glucose mea- surements in control sera by isotope dilution/massspectrometry: proposed definitive method compared with reference method. Clin Chem 1987;33:1397–1402.
range of glucose concentrations and was least 12. Chen E, Nichols J, Duh S, Hortin G: Performance eval- significantly impacted by sample hematocrit uation of blood glucose monitoring devices. Diabetes and other interfering substances. This should allow for better management of critically ill pa- 13. Khan A, Vasquez Y, Gray J, Wians F, Kroll M: The variability of results between point-of-care testing tients on tight glycemic control protocols.
glucose meters and the central laboratory analyzer.
Arch Pathol Lab Med 2006;130:1527–1532.
14. Dungan K, Chapman J, Braithwaite SS, Buse J: Glu- ACKNOWLEDGMENTS
cose measurement: confounding issues in setting tar-gets for inpatient management. Diabetes Care 2007; viding StatStrip, PCx, and SureSteppFlexx glu- 15. Miendje Deyi VY, Philippe M, Alexandre KC, De Nayer P, Hermans MP: Performance evaluation of thePrecision PCx point-of-care blood glucose analyzerusing discriminant ratio methodology. Clin Chem REFERENCES
16. Boyd JC, Bruns DE: Quality specifications for glucose 1. Van der Berghe G, Wouters P, Weekers F, Verwaest meters: assessment by simulation modeling of errors C, Bruyninckx F, Schietz M, Vlasselaers D, Ferdi- in insulin dose. Clin Chem 2001;47:209–214.
nande P, Lauwers P, Bouillon R: Intensive insulin 17. Karon BS, Gandhi GY, Nuttall GA, Bryant SC, Schaff therapy in critically ill patients. N Engl J Med HV, McMahon MM, Santrach PJ: Accuracy of Roche Accu-Chek Inform whole blood capillary, arterial, 2. Lewis K, Kane-Gill S, Bobek M, Dasta J: Intensive in- and venous glucose values in patients receiving in- sulin therapy in critically ill patients. Ann Pharma- tensive intravenous insulin therapy after cardiac sur- gery. Am J Clin Pathol 2007;127:919–926.
3. Tang Z, Louie R, Payes M, Chang K, Kost G: Oxygen 18. FDA Alert: FDA reminds healthcare professionals effects on glucose measurements with a reference an- about falsely elevated glucose levels; http://ww.fda.
alyzer and three handheld meters. Diabetes Technol gov/cdrh/ovid/news/glucosefalse.html. Updated No- 4. Tang Z, Louie R, Lee J, Lee D, Miller E, Kost G: Oxy- gen effects on glucose meter measurements with glucose dehydrogenase- and oxidase-based test strips for point-of-care testing. Crit Care Med 2001;29:1062–1070.
5. Tang Z, Du X, Louie R, Kost G: Effects of pH on glu- Department of Laboratory Medicine cose measurements with handheld glucose meters and a portable glucose analyzer for point-of-care test- ing. Arch Pathol Lab Med 2000;124:577–582.
6. Tang Z, Du X, Louie R, Kost G: Effects of drugs on glucose measurements with handheld glucose metersand a portable glucose analyzer. Am J Clin Pathol2000;113:75–86.

Source: http://www.novabiomedical.org.cn/wp-content/uploads/reports/184.pdf