Validity and Reliability of a Glucometer Against Industry Reference Standards

Abstract

As an appealing alternative to reference glucose analyzers, portable glucometers are recommended for self-monitoring at home, in the field, and in research settings.

Objective:

The purpose was to characterize the accuracy and precision, and bias of glucometers in biomedical research.

Research Design and Methods:

Fifteen young (20-36 years; mean = 24.5), moderately to highly active men (n = 10) and women (n = 5), defined by exercising 2 to 3 times a week for the past 6 months, were given an oral glucose tolerance test (OGTT) after an overnight fast. Participants ingested 50, 75, or 150 grams of glucose over a 5-minute period. The glucometer was compared to a reference instrument.

Results:

The glucometer had 39% of values within 15% of measurements made using the reference instrument ranging from 45.05 to 169.37 mg/dl. There was both a proportional (−0.45 to −0.39) and small fixed (5.06 and 0.90 mg/dl) bias.

Conclusions:

Results of the present study suggest that the glucometer provided poor validity and reliability results compared to the results provided by the reference laboratory analyzer. The portable glucometers should be used for patient management, but not for diagnosis, treatment, or research purposes.

Portable glucose analyzers or glucometers are recommended by the American Diabetes Association (ADA) for self-monitoring at home, in the field, or in clinical settings. ¹ The Food and Drug Administration requires that these devices meet or exceed the accuracy requirements set forth by the International Organization for Standardization (ISO), which requires glucometers to produce measurements within 20% of the reference value at glucose concentrations above 75.68 mg/dl.^1-3 However, the ADA recommends that these devices should produce measurements within 5% of reference values.^1,3 If the measurement error of these glucometers is less than 5%, they are an appealing alternative to reference glucose analyzers, particularly for small research programs. While there have been dramatic improvements in performance regarding sample blood volume and analysis time, only modest improvements in accuracy have been reported since 1989.^4,5 More recently, studies have compared the performance of these glucometers to the ISO standard or assessed the clinical significance of the measured differences.^2,3,5,6 Few studies have precisely quantified the accuracy, reliability, and bias of these glucometers if these instruments are to be used in the research laboratory. Therefore, to address these gaps in the literature, the purpose of this study was to more precisely quantify the validity, reliability, and bias of a glucometer for biomedical research.

Methods

Fifteen young (20-36 years; mean = 24.5 years), moderately to highly active men (n = 10) and women (n = 5) participated in the study. Highly active individuals were defined as those who participated in physical activity at least 3 times a week for the past 6 months and were moderately active if they participated 2 to 3 times a week for the past 6 months. ⁷ No participants were previously diagnosed with diabetes, and all had a fasting glucose level less than 86.5 (±12) mg/dL (range, 76.4-99.9 mg/dL), as determined by the reference instrument. All subjects read and signed an informed consent. The Institutional Review Boards at Wheaton College and Northern Illinois University approved all procedures, in compliance with the World Medical Association’s Declaration of Helsinki.

Instruments

As a reference instrument to determine the validity of the Bayer Contour glucometer we used the Yellow Springs Instruments (YSI) 2300 Stat Plus Glucose and Lactate analyzer (Yellow Springs, OH). This reference laboratory analyzer uses an electrochemical probe to measure glucose concentration in whole blood and uses a membrane-bound enzyme electrochemical methodology, where D-glucose is oxidized in the presence of glucose oxidase. A new membrane was used for each data collection session. The analyzer was initially calibrated using a 180 mg/dL and 900 mg/dL solution. In addition, an automated quality control was performed in triplicate every 45 minutes using a 180 mg/dL solution. Blood samples were collected from a finger stick into 2 or 3 heparinized capillary tubes. Blood was then mixed in a micro centrifuge tube. Two 25 µL samples were sequentially aspirated and measured by the analyzer for duplicate samples with the YSI. One probe on the YSI was set to analyze whole blood glucose. Therefore, 1 value was reported for each sample, and we took the average of the duplicate samples. All values reported are plasma glucose being corrected by multiplying the whole blood glucose value by 1.11 as recommended by the International Federation of Clinical Chemistry. ⁸

The glucometer uses an electrochemical glucose oxidase biosensor to measure glucose concentration in a 0.6 µl sample of whole blood. A quality control solution (106.31-147.75 mg/dl) was used to ensure proper operation of the meter as recommended by the manufacturer. Samples for the first 10 participants were taken directly from the finger as instructed in the device manual. Samples for the last 5 participants were not taken directly from the finger, but from the mixed blood sample used by the YSI 2300 reference instrument.

Oral Glucose Tolerance Test

Participants performed an oral glucose tolerance test (OGTT) after an overnight fast. Participants ingested 50, 75, or 150 grams of glucose over a 5-minute period (eg, Sun-Dex, Fisher Healthcare, Houston, TX, or glucose tablets, Walgreens, USA). Blood samples were obtained before ingestion (0 minutes) and 10, 20, 30, 60, and 90 minutes postprandial. For the first 10 participants all samples were taken specifically as follows: (1) handheld directly from finger, (2) capillary tubes for YSI, and (3) handheld directly from finger. For the last 5 participants, all blood was collected into capillary tubes. Two sequential samples using the glucometer were taken from the blood mixed in the micro centrifuge prior to measurement by the reference instrument. The duplicate measurements from each instrument were averaged and these mean values compared to determine the accuracy of the glucometer.

Data Analysis

Precision was determined from at least 3 serial measurements from each instrument using calibration solutions and expressed as coefficient of variation (CV). Three methods were used to assess validity. First, we determined the clinical accuracy using the ISO 15197:2013 performance criteria by calculating the percentage of individual glucometer readings falling within ±15% mg/dL of the corresponding mean reference values above 75 mg/dL and ±15 points for values below 75 mg/dL. ⁹ Second, a Bland–Altman plot was constructed to afford a more direct comparison of the data to that of previous studies, since this approach is typically used to assess agreement between methods. However, because fixed and proportional biases cannot be determined independently from these plots, ordinary least products (OLP) regression analysis was employed. Validity was determined from the correlation coefficient in combination with the presence and degree of bias. The degree of fixed bias was determined by the 95% confidence intervals of the y-intercept. If the confidence interval for the intercept includes a value of 0, then there is no fixed bias. Proportional bias was determined from the 95% confidence interval for the slope. If the confidence interval for the slope includes the value of 1.0, then there is no proportional bias. OLP regression provides different slopes and y-intercepts than ordinary least squares (OLS) regression because error is assumed in both glucometer and reference analyzers. Reliability was determined using OLP regression to quantify the relationship between sequential measurements for both instruments.

Results

Instrument Precision

The CV for the glucometer was 6.6% at a glucose concentration of 110 mg/dL and 5.3% at a glucose concentration of 445 mg/dL. The CV for the reference instrument was 1.2% at 180 mg/dL.

Validity

As measured by our reference instrument, blood glucose concentrations during the OGTT ranged from 50 to 188 mg/dL. Of the 166 total measurements made using the glucometer, 136 (82%) met the ISO standard. When glucometer and reference instrument blood samples were taken directly from the finger the mean difference between instruments was 8.427 mg/dL (Figure 1), thus indicating the glucometer consistently provided values greater than our reference instrument. According to the OLP analysis, the fixed bias is −7.148 mg/dL (Figure 2), an underestimation of plasma glucose. However, the confidence intervals overlap with zero. Perhaps more important, there is a significant (+13%) proportional bias as indicated by the slope of the regression line being greater than 1.0.

Figure 1.

Mean differences for glucose measurements between the glucometer and reference instrument.

Figure 2.

Actual fixed bias in glucose measurements between glucometer and reference analyzer instruments.

Reliability

Serial measurements by our reference instrument were strongly related as depicted in Figure 3. There was no proportional bias as the confidence interval for the slope of the relation included the line of identity. However, there was a small fixed bias, 5.123 and 0.905 mg/dl. The glucometer showed no fixed bias, but a small proportional bias (Figure 3, bottom pane).

Figure 3.

Proportional and fixed bias of the glucose measurements.

Discussion

The purpose of this study was to quantify the validity, reliability, and bias of the glucometer to the benchmark glucose analyzer for biomedical research. Our study found that the portable glucometers had poor validity and reliability compared to the benchmark laboratory analyzer. Furthermore, glucose readings from the glucometer showed fixed and proportional bias.

It is important to note that we chose to use OLP regression to characterize the relation between the Bayer Contour glucometer and our reference analyzer. Most studies have employed a combination of Bland–Altman plots, t tests, or least squares regression to determine the degree of agreement between various glucometers and a corresponding reference analyzer. The mean difference between analyzers is determined by the interaction of any fixed and proportional bias. Therefore, the mean difference between methods, as determined through Bland–Altman plots or t tests, does not solely reflect the accuracy or fixed bias of the device, but in some cases, the presence of a proportional bias or loss of linearity. As reflected in the slope of the linear relation, the use of least squares regression to characterize the level of proportional bias is skewed because all error is assigned to the dependent variable, in this case the glucometer. The use of least products regression to compare methods avoids both of these issues, allowing independent and more accurate determination of any fixed or proportional bias.^10-12

Of the glucometer readings, 82% fell within the ISO criteria for clinical accuracy. While we compared 166 separate glucometer readings to the mean reference values ranging from 50 to 190 mg/dL, it must be noted that we did not strictly follow the ISO protocol. The Bland–Altman plots showed an 8.427 mg/dL mean difference within subjects, which is lower than the bias reported in plasma-calibrated glucometers. ¹³ This difference between analyzers is determined by the interaction of any fixed and proportional bias that may exist. As revealed by the OLP analysis, the fixed bias is −7.148 mg/dL. In addition there was approximately a +13% proportional bias. While the fixed bias was not different zero (±95% CI includes zero), there was a small proportional bias. However, this bias may be as small as 5%, having little impact on the accuracy of this instrument. This is reflected by the fact that 81.2% of the readings less than 75.0 mg/dL met the ISO criteria, compared to 82% across the whole range of glucose values. The total bias, or mean difference between the glucometer and reference instrument, is equal to approximately 7% of the average glucose value. This is greater than the 5% recommended by the ADA.^1,3 Only 31% of the measurements by our glucometer met this standard. These data indicate that even though these glucometers may be suitable for clinical monitoring and insulin management, the measurement error is greater than acceptable for use in biomedical research.

The reference instrument contained a small fixed bias, amounting to approximately 3.6% of the average glucose value, while the glucometer displayed a fixed bias almost twice that of the reference instrument. Though it must be noted, the bias was not statistically significant, as the confidence intervals overlapped zero. We believed the larger bias may be due to the use of separate blood drops used for the duplicate glucometer samples. Therefore, for the last 5 study participants, we used the same blood sample as the reference analyzer; however, this did not reduce the fixed bias, indicating that the fixed bias was not due to sampling technique. While there was no significant fixed bias, and only a small proportional bias, the confidence intervals were 5-fold greater for the glucometer compared to the reference instrument. So while little or no bias exists in the glucometer, there was large trial-to-trial variability in readings, perhaps too large to use in biomedical research laboratory.

Over the past few decades, there have been improvements in glucometer accuracy. ⁴ According to the results of this study, however, using portable glucometers in a lab-based study yielded lower reliability when compared to the reference analyzer. A review of portable glucometers claimed reliable glucose reading values within 10% of reference values were only made in one-half of the analyzers tested. ⁵ As such, our study is the latest to confirm that many glucometers do not meet either the ISO or ADA standard.

Possible errors for portable glucometers are a result of many factors, such as: improper calibration, failure to clean, not an adequate amount of blood on the strip, improper storage of strips (such as temperature, humidity, and moisture), and sometimes-patient medications. ⁶ Difference in the testing strips used for glucometers or proper calibration may diminish the analytical quality. Studies have revealed the variability between the test strips alone for glucometer systems is poor.^14,15

To minimize any sources of error, the following study was conducted in a controlled laboratory environment where the temperature and humidity were controlled for both the data collection and storage of strips. All testing was done on healthy active participants to mitigate the chance of drugs interfering with the results. All equipment was properly calibrated prior to testing as recommended by the manufacturer. The reference instrument only performed 1 quality control solution (180 mg/dL) in triplicate every 45 minutes instead of 2 solutions, which could be a limitation. It should be noted that the use of a single glucose concentration (180 mg/dL in triplicate every 45 minutes) for quality control of the reference instrument may not have allowed us to verify that it was operating properly at lower glucose concentrations. As one can see in Figure 1, however, the difference between the glucometer and reference instrument was consistent across the entire range of glucose values.

The strips used for the portable glucometer were newly ordered and within the expiration date provided by the manufacturer. However, whether the strips in the order were all from the same lot number could not be verified. All methods performed consisted of duplicate samples. However, most users of self-monitoring blood glucose monitors (SMBG) do not take duplicate measures for an accurate reading.

SMBG has surged in popularity among patients with diabetes to help them achieve their goals and is recommend by the ADA for all diabetes patients for management of the disease. The monitoring of glycemic status is meant to access the patient’s physiological status and nutritional therapy. ¹ The accuracy of SMBG is user dependent and also important for health care providers to monitor the users’ technique. ¹ Researchers have claimed that confidence and accuracy of the results is dependent on how the information obtained is used. ⁶ The data are consistent with those of previous studies that have assessed the clinical reliability of these glucometers.^2,3,5,6 However, our data continue to show significant bias in accuracy (Figure 2, 13% proportional bias) and a relatively large imprecision (Figure 2, 5-fold larger confidence intervals). Taken together, our data suggest that glucometers may be suitable for clinical monitoring and insulin management but are not currently acceptable for use as laboratory instruments. Due to the lack of reliability and validity compared to reference lab analyzers, the results of this study indicate SMBG may only be used for patient maintenance, but this should be done with great caution.

To maximize the accuracy of these devices patients must use SMBG in accordance with manufacturer recommendation. Harrison and colleagues showed that under highly controlled conditions and supplies approximately 99% of all readings of the Bayer Contour were within 15% of the reference instrument. ¹⁶ The portable glucometer was not highly correlated to the reference analyzer, as the criteria from the ADA were not met in glycemic and hyperglycemic ranges. The SMBG or portable glucometer can still be employed for patient management, but laboratory and clinical measures should be used for patient diagnosis and treatment purposes, as well as for research given the significant measurement errors.