Technical Evaluation of Five Glucose Meters With Data Management Capabilities

Transcription

1 Clinical Chemistry / LABORATORY EVALUATION OF GLUCOSE METERS Technical Evaluation of Five Glucose Meters With Data Management Capabilities Jeffrey]. Chance, PhD, Dai J. Li,MD, PhD, Kerrie A. Jones, MS, Karen L. Dyer, MT(ASCP), and James H. Nichols, PhD Key Words: Glucose; Glucose meter; Biosensor; Point-of-care; Diabetes Abstract Technical performance and data management features are prominent criteria in the selection of an appropriate meter for a point-of-care glucose testing program. We evaluated the technical performance of 5 currently available glucose meters with data management capabilities. The performance of all 5 meters was technically equivalent. Linear regression slopes vs the reference method are in the range of 0.94 to 1.02 and indicate correlation more to plasma values than to whole blood values. The percentage of glucose meter results within ± 15% of the laboratory value was at least 90%; however, the percentage within ± 10% was 75% to 7% for most meters. Within-day and between-day precision ranged from 2% to 5% coefficient of variation. Evaluation of linearity with glucose-spiked venous specimens demonstrated that the linearity of each meter agreed with the manufacturer's stated range in most cases. Meter glucose values tended to bias negative as the hematocrit increased, an effect that may be more pronounced at higher glucose concentrations. volume effects were noted between 5pL and 40 pl. The results suggest that all meters tested will likely satisfy technical performance criteria in a hospital setting and that selection of a system for point-of-care glucose testing will be influenced by the institution's data management requirements. Near-patient glucose testing by nurses and other medical personnel is an important component in managing the care of the diabetic patient. By providing faster turnaround times, the clinician can respond rapidly to changes in the patient's glucose levels, resulting in optimum insulin regimens and shorter, less expensive hospitalizations.1 Owing to its perceived success, point-of-care glucose testing has been cited as a model for bringing more tests to the bedside2 and has been the subject of several review articles.27 Small portable meters are now used widely for glucose testing and in 1996 had an estimated worldwide market of $425 million. The selection of an appropriate meter for routine use in the hospital setting must at a minimum involve an evaluation of the technical performance of the meter. In addition to the clinical requirement for accurate test results, the meter also should have features that facilitate the manipulation and communication of results and assist meeting regulatory and institutional requirements for quality control and documentation. In response, commercial vendors have developed glucose monitoring systems with a wide variety of data management capabilities. In this article, the Advantage (Boehringer-Mannheim, Indianapolis, IN), HemoCue Glucose (HemoCue, Mission Viejo, CA), One Touch II Hospital (Lifescan, Milpitas, CA), Precision G (Medisense, Bedford, MA), and SureStep Pro (Lifescan) whole blood glucose monitoring systems were evaluated for technical performance and data management capabilities. Correlation with laboratory methods, linearity, effect of hematocrit, effect of volume, and precision were studied. Materials and Methods All meters, test strips, and quality control materials were supplied to the hospital by the manufacturers for this evaluation. One meter from each manufacturer was evaluated. HemoCue supplied Boehringer Mannheim Glucose BG controls for use American Society of Clinical Pathologists Downloaded from Am J Clin Pathol 1999;111:

2 Chance et al / LABORATORY EVALUATION OF GLUCOSE METERS with its meter because HemoCue does not manufacture specific controls. Test strips were the latest strip version being marketed at the time of manuscript submission: the original cuvettes for the HemoCue and test strips for the One Touch II and SureStep; the Advantage H strip with Food and Drug Administration-approved code key modification as of December 1997 ("optimized" Advantage H); and the Precision G2b test strip. Reagents were stored as recommended. Arterial specimens were drawn in lithium heparin syringes and analyzed within 30 minutes of phlebotomy to minimize loss of blood gases and effects of glycolysis. Venous specimens were collected in lithium heparin tubes (Vacutainer, Becton Dickinson, Rutherford, NJ). Meter accuracy was determined by comparison with laboratory reference methods. In the initial study, arterial samples (n = 102) were analyzed on the va Stat Profile 5 (glucose oxidase, va, Waltham, MA) simultaneously with all 5 meters. The plasma was then immediately separated and analyzed within 30 minutes on the Boehringer-Mannheim Hitachi 747 (hexokinase, Boehringer-Mannheim). A subsequent study was conducted to evaluate the performance of new test strips being introduced for 2 of the meters. Arterial patient samples (n = 120) were analyzed on the Chiron 65 (glucose oxidase, Chiron, Emeryville, CA) simultaneously with the Advantage meter (optimized Advantage H strip) and the Precision G meter (new Precision G2b strip). The plasma was then separated and analyzed as described in the preceding text. The va Stat Profile 5 and the Chiron 65 are blood gas-electrolyte analyzers that accept a whole blood specimen but produce results comparable to plasma concentrations, as the sample is not hemolyzed during analysis. The data reported in this article for the Precision G meter reflect a 70-data point subset from the Precision G2b study (n = 120), corresponding to readings in which analysis occurred within 20 minutes of phlebotomy, to avoid interference from "microdot" formation. Meters were evaluated for intra-assay (within-day) and interassay (between-day) precision. The low (approximately 2. mmol/l [50 mg/dl]), medium (approximately mmol/l [ mg/dl]), and high (approximately mmol/l [ mg/dl]) control solutions for each meter were run 20 times in succession on the respective meter to establish intra-assay precision using controls. medium level control was available for the One Touch Hospital II, however. Intra-assay precision using actual patient specimens was then evaluated using a low glucose and a high glucose arterial specimen, each run 10 times on all 5 meters. Here, only 10 sequential readings were performed on each meter to provide a fair estimate of the precision, free from undue interference from declining glucose concentration due to time-dependent glycolysis. Interassay precision was determined from 20 or more analyses of each 4 Am J Clin Pathol 1999:111: Downloaded 5from control level during a minimum of 10 days (each control level run twice per day). Linearity was evaluated by using multiple venous spiked specimens. A venous sample was collected and held for 4 hours to allow the glucose concentration to approach zero. An aliquot was spiked to a glucose level of 3.9 mmol/l (700 mg/dl) and then diluted with unspiked blood. All solutions were analyzed in duplicate within 30 minutes of dilution. Aliquots were centrifuged and the plasma analyzed on the Hitachi 747 to confirm the clinical laboratory concentration. Hematocrit effects were determined by using venous specimens. After being held for 4 hours, 1 aliquot of a venous specimen was spiked to a glucose level of 2. mmol/l (50 mg/dl), and a second aliquot was spiked to a level of.7 mmol/l (300 mg/dl). Each aliquot was then centrifuged, and the RBCs, intermediate leukocyte layer, and a small amount of plasma were separated from the remaining plasma in each tube. The RBC mixture was mixed to produce a solution at approximately 0% (0.0) hematocrit, which was then combined with the plasma to prepare samples at 70%, 56.25%, 42.5%, 2.75%, and 15% hematocrit (0.70, 0.56, 0.42, 0.29, 0.15, respectively). These solutions were analyzed in duplicate on the meters and also with both laboratory reference methods. For the Precision G and Advantage meters, hematocrit effects also were evaluated by determining the hematocrit of the patient correlation samples. Hematocrit was verified by spun microhematocrit. Volume effects were evaluated by testing 5 to 40 ul of a patient sample on the meters with the exception of the HemoCue, which accepts only a fixed 5-LtL volume. In all cases, patient samples, linearity and hematocrit solutions, and controls were mixed thoroughly before analysis. All studies, including all glucose meter readings, were performed by 3 trained operators from our institution. Data management capabilities were obtained from the manufacturer's literature. Results Data for the technical performance of the meters are listed in ITable II. Intra-assay (within-day) precision was less than 5% (coefficient of variation [CV]) in most cases for control solutions. There was no consistent trend for increasing or decreasing imprecision as a function of glucose concentration. The One Touch Hospital II exhibited the best precision with CV percentages in the range of 1.1 to 2.5. The intra-assay precision using patient (whole blood) samples was less than or equal to a CV of 5% in all cases and was not substantially different from that using the control solutions. The interassay (between-day) precision evaluated during a minimum period of 10 days was also less than 5% (CV) in most cases.

3 Clinical Chemistry / ORIGINAL ARTICLE Table II Technical Performance* Parameter Precision (coefficient of variation, %) Intra-assay control Intra-assay patient Interassay control Correlation va/chiron Hitachi 747 Correlation coefficient va/chiron Hitachi 747 Sy/x (mmol/l) va/chiron Hitachi 747 Percentage of laboratory agreement Percentage within ±15% va/chiron Hitachi 747 Percentage within ±10% va/chiron Hitachi 747 Linearity (mmol/l) Manufacturer range Advantage HemoCue One Touch I I SureStep Pro Precision G WB WB x x X x X x X x x x WB = whole blood control from HemoCue. * For names and locations of manufacturers, see the introductory section. To convert mmol/l to mg/dl, divide by The accuracy of each meter was evaluated by comparison of the meter results with laboratory reference methods. The raw data and the least squares linear regression line for the va Stat Profile 5 or Chiron 65 correlation (dashed line) and the Hitachi 747 correlation (solid line) are plotted in Figure II. In general, all 5 meters correlated extremely well with both reference methods for the glucose concentration range of the patient samples. The regression lines are only very slightly offset from the theoretical 1:1 line, and the cluster of the data is generally tight. The linear regression equations show good correlation with the reference methods for all 5 meters, with little bias. The correlation coefficients (r) are in the range of to 0.993, and the standard errors of the estimate (Sy/x) are in the range of to 0.26 mmol/l (7-15 mg/dl). The percentage of glucose meter results that were within ±15% and ±10% of the reference method (percentage of laboratory agreement) also are given in Table 1. Linearity of each meter was evaluated by using spiked glucose specimens in a wider (mainly higher) concentration range than that seen in the patient samples IFigure 21. We used the polynomial regression method to evaluate linearity.9 The Advantage meter was found to be linear to 33.3 mmol/l (600 mg/dl), the SureStep to 27. mmol/l (500 mg/dl), and the HemoCue to 22.2 mmol/l (400 mg/dl), using the Hitachi 747 as the reference method. These ranges are in agreement with those specified by the manufacturer (Table 1). The One Touch II and the Precision G became nonlinear at high American Society of Clinical Pathologists Downloaded from glucose concentrations (lower than the manufacturer-stated range). The One Touch II was linear up to 27. mmol/l (500 mg/dl), and the Precision G was linear up to 22.2 mmol/l (400 mg/dl). However, the result for the Precision G meter may be an unavoidable artifact of the protocol the specimen was held for 4 hours to allow the glucose concentration to approach zero, whereas according to the manufacturer, the test strip should be used only with fresh specimens. With the 0 mmol/l solution, some of the meters reported a result (approximately 0 for HemoCue, One Touch II, and SureStep), while others indicated "<20" (mg/dl, Precision G) or displayed an error message (Advantage). The effect of hematocrit on the glucose meter reading at low and high glucose concentrations is shown in IFigure 31. The salient feature at both glucose levels is a decline in the reported glucose concentration at an increasing hematocrit for all of the meters and laboratory methods. Part of this feature is undoubtedly an artifact of the method used in preparing the hematocrit solutions (see the "Methods" section). Reestablishment of osmotic equilibrium after mixing glucose-spiked plasma with RBCs at a lower osmotic potential may dilute the plasma compartment. In addition, the increasing proportion of RBCs and leukocytes added to the plasma to prepare the higher hematocrit solutions will increase the extent of glycolysis in these solutions before analysis.10 This cell-dependent glycolysis would be expected to have a relatively more pronounced effect at Am J Clin Pathol 1999; 111:547-55B 549

4 C h a n c e e t al / LABORATORY EVALUATION OF GLUCOSE METERS Laboratory Reference Glucose (mmol/l) Laboratory Reference Glucose (mmol/l) 4 12 Laboratory Reference Glucose (mmol/l) Laboratory Reference Glucose (mmol/l) IFigure I I Patient correlation data and least squares linear regression lines for evaluation meters. Correlation against the Boehringer-Mannheim Hitachi 747 (open squares) yielded the following regression parameters: A, Advantage = 0.9x , r = 0.95, S y x = 0.694, n = 120; B, HemoCue = 0.96x , r = 0.9, Sy.x = 0.526, n = 102; C, One Touch II = 0.9x , r = 0.991, S y x = 0.477, n = 102; D, SureStep Pro = 0.94x ; r= 0.993, S y x = 0.391, n = 102; E, Precision G = 0.96x , r= 0.974, S = 0.26, n = 70. Correlation against the ' yx n ' va Stat Profile 5 (closed circles, n = 102) yielded the following regression parameters: B, HemoCue = x , r= 0.91, S y x = 0.676; C, One Touch II = 1.02x-0.32, r = Laboratory Reference Glucose (mmol/l) 0.990, S y x = 0.506; D, SureStep Pro = 0.9x , r = 0.992, S = Correlation against the Chiron 65 (closed circles, n = 120) yielded the following regression parameters: A, Advantage = x , r= 0.96, S = 0.677; E, Precision G = 0.94x , r = 0.974, S y x = 0.. For names and locations of manufacturers, see the introductory section. 550 Am J Clin Pathol 1999; 111: Downloaded from

5 Clinical Chemistry / ORIGINAL ARTICLE lower glucose concentrations. Figure 3 shows that the relative decline in glucose concentration with increasing hematocrit (eg, for the Hitachi 747 method) is in fact more pronounced in the lower vs the higher glucose range. An examination of the absolute difference between the meters and the laboratory methods, rather than individual changes, is therefore necessary. In the low glucose range (Figure 3A), there is some variation in the performance of the 3 laboratory methods, and the hematocrit plots for the 5 meters are not significantly different from the laboratory method plots. In the high glucose range (Figure 3B), however, all 3 laboratory methods have similar hematocrit plots, while the meters (except for the HemoCue) do not provide a reading at high hematocrit (SureStep Pro) or systematically underestimate the glucose value as the hematocrit increases above 50% (0.50) (Advantage, One Touch, and Precision G). In the second study (evaluation of new strips for Advantage and Precision G meters), the hematocrit of each patient specimen was determined, and these data are plotted in IFigure 41 as a function of the percentage of difference between the glucose meter reading and the result obtained on the Hitachi 747. While there is still a trend for both meters to underestimate the glucose value as the hematocrit increases (trend greater for the Advantage meter), the trend is slight, and the bulk of the data are within ±15% of the Hitachi 747 for the range of hematocrit observed in the patient samples (20%-60% [ ] hematocrit). The evaluation of hematocrit effect for the Precision G is probably more relevant than that shown in Figure 3, since here the specimens were analyzed within 20 minutes of phlebotomy, according to the manufacturer's instructions. 40 Hitachi 747 Glucose (mmol/l) IFigure 21 Verification of high end linearity according to manufacturer's stated linear range, using the Hitachi 747 as the laboratory reference (x). Regression statistics were based on duplicates at each concentration level and represent the optimal least squares fit. HemoCue = 0.935x 0.07, r = 1.000, Sy.x = 0.174; One Touch II = 0.51 x , r= 0.99, S yx = 0.29; SureStep Pro = 0.954x , r = 0.999, S yx = 0.531; Advantage = 0.905x , r = 0.99; Sy x = 0.635; Precision G = 0.20x , r = 0.994, S = For names and locations of manufacturers, see the introductory section. Glucose (mmol/l) e D s ^ ^ 3-2 o o * 0 "^s ls? Advantage HemoCue SureStep Pro One Touch Precision G --- Chiron va Hitachi ^ \ ^ \ ^-^, I C3 o Advantage HemoCue o SureStep Pro - One Touch -- Precision G Chiron va Hitachi 0 Hematocrit (%) Hematocrit (%) IFigure 31 Effect of hematocrit on glucose result at low (A) and high (B) glucose concentration, using spiked solutions. For names and locations of manufacturers, see the introductory section. Ameri can Society of Clinical Pathologists Downloaded from Am J Clin Pathol 1999;111:

6 C h a n c e e t al / LABORATORY EVALUATION OF GLUCOSE METERS I B 45 o 30 I I b '^^jf^s^tl l oo ««o 0. - _ * 0 * P02 (mm Hg) Hematocrit (%) Figure 41 Effect of hematocrit and po2 on glucose result for the Advantage (open circles, solid line; n = 120) and Precision G (closed circles, dashed line; n = 70, all specimens < 20 minutes) using the actual patient correlation specimens. For names and locations of manufacturers, see the introductory section. The partial pressure of oxygen (P0 2 ) of each specimen was also determined in the second study, and the results are shown in Figure 4 (lower plot). As with the hematocrit plot, there is a slight trend for the meter result to bias negative as the P0 2 increases (trend greater with the Precision G meter). However, the data are still within ±15% of the Hitachi 747. For the amount of sample applied, no effect related to the volume of blood applied to the strip (5u.L^40 LiL) was noted for any of the 5 meters (data not shown). A list of some of the more important data management features for the meters is given in ITable 21. Most commercial glucose meter systems for use in the hospital setting are now available with an extensive menu of hardware, software, and connectivity options. The 5 meter systems we tested all had the more important features and were similar except in terms of memory capacities, the details of data downloading, and ability to customize data analysis. Some of the data management deficiencies with the One Touch II Hospital system are being addressed by the recent introduction of the One Touch II DataDock system. When the vendor software cannot directly generate customized reports, the data can be exported and then imported into third-party software that will customize the presentation of data. Of particular interest to our institution was whether the system had the ability to transfer data automatically from the nursing unit to our Performance Improvement Office. While all 5 meters have a method for uploading data to a nearby PC or laptop in the nursing unit, as well as software to receive the data, at the time of the present study, the vendor software was only able to transfer the data to a centrally located PC (eg, Performance Improvement Office or laboratory information 552 Am J Clin Pathol 1999;111: Downloaded from systems) via ethernet or modem in 2 of the meters (Advantage, Precision G). Our present point-of-care glucose testing program requires an employee to visit (with a laptop computer) more than 70 nursing, intensive care, and outpatient sites and collect the data. The ethernet/modem capability of the Advantage and Precision G meter systems was, therefore, influential in selecting these 2 meters for further evaluation in our institution. Discussion Methods for point-of-care glucose determination have evolved substantially since first introduced approximately 20 years ago. Small portable meters providing quantitative measurement of glucose are now in widespread use and represent one of the unique success stories in the application of biosensor technology. Use of these devices in the hospital setting mandates consideration of the technical performance and data management capabilities of the meter. After meeting minimum requirements for technical performance (ie, accuracy of results), selection of an appropriate meter will be influenced by the extent to which data management and related features facilitate implementation of regulatory and institutional requirements for quality control, operator proficiency, and online availability of results. Evaluation of the technical performance of glucose meters often is based on accuracy goals defined by the American Diabetes Association (ADA). In 197, the ADA developed a consensus statement that recommended that glucose concentrations determined with portable meters

7 Clinical Chemistry / ORIGINAL ARTICLE Table 21 Data Management Features* Memory Quality control, patient tests Overwrite w h e n full Operator identification limits Test results Alert if patient panic value Action/comment codes Store w i t h patient identification Store w i t h operator identification Store with date/time Multiple patient tests w i t h single operator identification Data downloading RS-232 to PC Modem/telephone Direct laboratory information systems interface Lot information Strip lot no. Strip expiration date Strip expiration lockout Control lot no. Control expiration date Control expiration lockout Quality control statistics Mean/SD/coefficient of variation/ no. of measurements Percentage of in-range data Range customizable Levey-Jennings plot Quality control lot trend plots Other Patient and operator reports Customized reports Time quality control lockout Bar-code ability Advantage HemoCue One Touch I I SureStep Pro Precision G 1,000 total 4, ,300 total limits 900 total w i t h warning 1,999 2,500 with warning limits 200, 550 with warning limits, Rals-G Infrared to PC Under development Under development From PC (cuvette) /// /// /// /// /// Both Both Both Both * For names and locations of manufacturers, see the introductory section. should fall within ±15% of the laboratory values for meters available at that time and that the goal of all future meters should be to reduce this variability to within ±10% at glucose concentrations between 1.67 and 22.2 mmol/l ( mg/dl) 100% of the time.11 The 1994 and 1996 ADA consensus statements further refined this recommendation to within ±5% of laboratory values for future glucose meters.12'13 To our knowledge, these guidelines have never been met in any independent evaluation of glucose meter performance. Bain et al14 observed glucose meter readings exceeding ±15% in 35.4%, 27.0%, and 43.4% of samples with 3 different whole-blood glucose meters. In a more recent evaluation of 4 portable glucose meters, Nichols et al15 observed meter results exceeding ±15% of the laboratory reference method in the range of 3.7% to 41.7%. ADA recommendations also were not met in other recent evaluations of 1 glucose meter and 6 glucose meters.17 In a College of American Pathologists Q-Probes study involving 242 institutions and 6,653 paired bedside glucose monitor and clinical laboratory results, 26% of the bedside glucose Downloaded from monitor results were in excess of ±15% of the corresponding laboratory results, and 44% exceeded ±10%.1 The authors of that study also concluded that the testing accuracy remained unchanged between 1991 and The present study shows that while the 5 meters we evaluated also do not meet these stringent guidelines all of the time, some improvement in performance may be indicated. The percentage of meter results exceeding ±15% of the corresponding laboratory method values were in the range of 2.9% to 14.7%, an improvement since a previous study in 1995 by Nichols et al.15 However, the percentage exceeding ±10% was increased at 6.9% to 36.3%. If the National Committee for Clinical Laboratory Standards (NCCLS) guidelines are invoked,20 9.6% to 100% of the meter results are within the recommended ±20% of the laboratory value for glucose concentrations higher than 5.5 mmol/l (100 mg/dl). The percentage deviation of meter results from a reference plasma method has been shown to increase for values less than 4 mmol/l (72 mg/dl).21 While the range of glucose concentrations in the patient specimens Am J Clin Pathol 1999; 111:

8 et al / LABORATORY EVALUATION OF GLUCOSE METERS studied did not allow for a rigorous examination of meter performance in the hypoglycemic range, greater deviation is nevertheless seen as the glucose concentration declines. Using the Hitachi 747 as the reference method, 11 of the 496 total meter readings in the present study differed by more than 20%. Nine of these occurred when the Hitachi 747 glucose determination was less than 2. mmol/l (50 mg/dl), 1 occurred between 2. and 5.5 mmol/l ( mg/dl), and 1 occurred at a concentration higher than 5.5 mmol/l (100 mg/dl). Guidelines that specify an absolute rather than percentage difference target (eg, the NCCLS recommendation that meter results less than 5.5 mmol/l [100 mg/dl] be within 0. mmol/l [15 mg/dl] of the laboratory method) may be more appropriate in the hypoglycemic range. Any statement about improvement in meter performance in this article also is qualified, since the present study involved only 3 trained operators who evaluated the meters simultaneously under optimal conditions. Use of the meters under normal conditions, as in the Q-Probes study, likely introduces greater variability in test results. This increased variability is due primarily to the use of the meters by multiple operators, with concomitant variability in user technique. Operator errors such as failure to perform quality control, inaction on out-of-range quality control results, and use of mismatched or expired control solutions and/or test strip lots also will contribute to increased test result variability. However, feedback received from split-sample testing has been shown to have a definite effect in maintaining accuracy in blood glucose monitoring.22 However, the exact accuracy recommendations for clinical management have not been defined rigorously.11 It is recognized that the relative accuracy required in point-of-care glucose testing is not as strict as in other clinical tests, a factor that contributed to the waived status assigned to glucose meters under the Clinical Laboratories Improvement Amendments of While the percentage of deviation of results from laboratory values has been suggested as a more clinically useful measurement of accuracy,24 linear regression parameters, precision data, and linearity and hematocrit studies are also relevant contributors to overall meter performance. Our present evaluation indicates that all 5 meters correlate well with plasma-based laboratory reference methods and have within-run and between-run precision in the range of 2% to 5% (CV). The manufacturer-indicated linear range was verified for 3 of the meters. Two of the meters became nonlinear at glucose concentrations lower than the upper end specified by the manufacturer. However, in both cases, the deviation did not exceed the NCCLS-recommended 20% of the reference method value. Therefore, the ranges specified by each manufacturer are still valid as working ranges for all 5 Downloaded from Am J Clin Pathol 1999; 111: meters. Hematocrit biases are small or negligible for normal patient hematocrit ranges (25%^15% [ ]) but may cause unacceptable underestimation of glucose concentration at hematocrits exceeding 50% (0.50). Although volume effects between 5 LIL and 40 LIL were negligible for all meters, care should be taken in applying a sufficient amount of blood to the test strip, since it has been shown that very small amounts (eg, 1 LIL) can lead to falsely decreased glucose results in meters that allow a result to be displayed (ie, do not produce an ERROR message).25 When comparing the performance of the first-generation and newer strip advances for the Advantage and Precision G meters, the new Advantage H and Precision G2b strips were noticeably improved in precision, regression data, and percentage of deviation from the laboratory reference methods. Consistent with the manufacturer's recommendation, there was an improved quality of the data when the Precision G2b test strips were used with fresh specimens (ie, tested within 20 minutes of phlebotomy). While most meters are indicated by the manufacturer to be calibrated to whole blood, the linear regression data obtained in the present study indicate that the meters actually correlate more closely to plasma than to whole blood. Correlation to plasma glucose concentration is, in our opinion, a more relevant goal, since the rationale of point-of-care testing is to provide a near-patient substitute for the laboratory method, which is invariably based on testing of serum or plasma. While the reason for the traditional use of wholeblood calibration by glucose meter manufacturers is unclear, some manufacturers now are investigating conversion to plasma correlation.26'27 We encourage this conversion, to provide a uniform standard that avoids the confusion of test results and allows for direct comparison of meter performance between meters and against recommended performance criteria such as NCCLS accuracy goals. Data management features are likely to be a determining factor in the selection of a glucose meter system for point-of-care testing in a hospital or physician office laboratory setting. In addition to the fundamental requirement for data storage and manipulation, sufficient documentation is necessary to provide an audit trail that links a competent operator and a properly performing instrument and reagents to a patient result.2 Manual recording of data in log books has a number of disadvantages, including errors in entry and transcription, falsification of data, and difficulty detecting trends and integrating large volumes of data.12 Glucose meter systems with data management capabilities can potentially avoid many of the problems associated with the traditional use of log books. A summary of data management features for the 5 meters tested in the present study is given in Table 2. This information was obtained from the manufacturer and represents

9 Clinical Chemistry / ORIGINAL ARTICLE the most recently available information at the time of publication. Features such as operator and patient identification, entry and storage of expiration dates, notification or lockout with expired lot numbers, and quality control statistics are all important in ensuring accuracy of results and monitoring trends in meter performance. The requirements for memory storage, data downloading, and interface options depend on the size of the institution. By facilitating automated data capture, there is the potential for connection to the laboratory and hospital information systems with online availability of test results. User statistics and exception reports that document the frequency of outliers, flagged results, and corrective actions will contribute to an overall quality assurance program by improving data review and analysis. In summary, we observed that current meters provide results comparable to laboratory values for the majority of patient samples. The technical performance of all 5 meters was deemed acceptable for our institution, based on withinrun and between-run precision of less than 5% and the conformance of 99% to 100% of the results to NCCLS recommendations above and below the 5.5 mmol/l (100 mg/dl) level. However, central laboratory testing is recommended for specimens with extremes of hematocrit and/or glucose measurements. The results of evaluations such as the present study demonstrate that Food and Drug Administration approval and manufacturer data are indicative only of what an instrument is capable of achieving under optimum operating conditions. Each institution should set minimum standards for meter performance and independently verify that the meter is capable of achieving those standards within the institution. A comprehensive quality assurance program including proficiency testing is essential to ensure that patient results are not compromised by factors other than the technical performance of the meter. Selection of appropriate data management capabilities also will depend on the needs and practices of each institution and should be made in consultation with the institution's information systems staff and end users (nursing staff). Manufacturers that offer a wide selection of features, as well as flexibility in the presentation of results and interfacing of computer systems, will be more competitive in this expanding market. From the Department of Pathology, Johns Hopkins Medical Institutions, Baltimore, Maryland. The authors have accepted reagents, controls, and an equivalent financial grant from each glucometer manufacturer; Boehringer-Mannheim, HemoCue, LifeScan, and Medisense, to support the labor and supplies involved in conducting this study. Address reprint requests to Dr Nichols: Department of Pathology, Johns Hopkins Hospital, 600 N Wolfe St, Meyer B-125, Baltimore, MD Downloaded from References 1. American Diabetes Association. Bedside blood glucose monitoring in hospitals. Diabetes Care. 1995;1(suppl 1): Jones BA. Testing at the patient's bedside. Clin Lab Med. 1994;14: Nichols JH. Cost analysis of point-of-care laboratory testing. Adv Pathol. 1996;9: Nichols JH. Management of near-patient glucose testing. AACC Endocrinology and Metabolism In-Service Training and Continuing Education. 1994;12: Kost GJ. Guidelines for point-of-care testing: improving patient outcomes. Am J Clin Pathol. 1995;104(suppl 1):S111-S Kost GJ, Hague C. The current and future status of critical care testing and patient monitoring. Am J Clin Pathol. 1995;104(suppl 1):S2-S Seamonds B. Medical, economic, and regulatory factors affecting point-of-care testing: a report of the Conference on Factors Affecting Point-of-Care Testing, Philadelphia, PA, 6-7 May Clin Chim Acta. 1996;249: Rouhi AM. Biosensors send mixed signals. Chem EngNews. 1997;75: Kroll MH, Emancipator K. A theoretical evaluation of linearity. Clin Chem. 1993;39: Threatte GA, Henry JB. Carbohydrates. In: Henry JB, ed. Clinical Diagnosis and Management by Laboratory Methods. 19th ed. Philadelphia, PA: Saunders; 1996: American Diabetes Association. Consensus statement on selfmonitoring of blood glucose. Diabetes Care. 197;10: American Diabetes Association. Self-monitoring of blood glucose. Diabetes Care. 1994;17: American Diabetes Association. Self-monitoring of blood glucose. Diabetes Care. 1996;19(suppl):S62-S Bain O, DeVille PJ, McPherson RA. Performance characteristics of three whole-blood glucose monitors. Lab Med. 1991;22: Nichols JH, Howard C, Loman K, et al. Laboratory and bedside evaluation of portable glucose meters. Am ] Clin Pathol. 1995;103: Stenger P, Allen ME, Lisius L. Accuracy of blood glucose meters in pregnant subjects with diabetes. Diabetes Care. 1996;19: Trajanoski Z, Brunner GA, Gfrerer RJ, et al. Accuracy of home blood glucose meters during hypoglycemia. Diabetes Care. 1996;19: Jones BA, Howanitz PJ. Bedside glucose monitoring quality control practices: a College of American Pathologists QProbes study of program quality control documentation, program characteristics, and accuracy performance in 544 institutions. Arch Pathol Lab Med. 1996;120: Howanitz PJ, Jones BA. Bedside glucose monitoring: comparison of performance as studied by the College of American Pathologists Q-Probes program. Arch Pathol Lab Med. 1996;120: Barr JT, Betschart J, Bracey A, et al. Ancillary (bedside) Blood Glucose Testing in Acute and Chronic Care Facilities. National Committee for Clinical Laboratory Standards; 1994;14:1-14. NCCLS document C30-A. 21. Zenobi PD, Keller A, Jaeggi-Groisman SE, et al. Accuracy of devices for self-monitoring of blood glucose including hypoglycemic blood glucose levels. Diabetes Care. 1995;1:57-5. Am J Clin Pathol 1999; 111:

10 Chance et al / LABORATORY EVALUATION OF GLUCOSE METERS 22. Jones HE, Cleave B, Zinman B, et al. Efficacy of feedback from quarterly laboratory comparison in maintaining quality of a hospital capillary blood glucose monitoring program. Diabetes Care. 1996;19: Medicare, Medicaid and CLIA programs: regulations implementing the Clinical Laboratories Improvement Amendments of 19 (CLIA), 57 Federal Register (1992). 24- Clarke WL, Cox D, Gonder-Frederick LA, et al. Evaluating clinical accuracy of systems for self-monitoring of blood glucose. Diabetes Care. 197;10: Velazquez FR, Bartholomew D. Effect of small sample volume on five glucose monitoring systems. Diabetes Care. 1996;19: Lane A, Geadah D, Laferriere M, et al. Performance of the new plasma-compatible Advantage blood glucose test strips. Clin Biochem. 1997;30:465^. 27. Kempe KC, Czeschin LI, Yates KH, et al. A hospital system glucose meter that produces plasma-equivalent values from capillary, venous, and arterial blood. Clin Chem. 1997;43: Kiechle FL. At the bedside: POCT data management. Adv Lab. 1997;6: Am J Clin Pathol 1999; 111: Downloaded from