Clinical Accuracy of a Glucose Oxidase–Based Blood Glucose Test-Strip Across Extremes of Oxygen Partial Pressure

Abstract

Background:

Glucose oxidase (GOx)-based blood glucose monitors (BGMs) are influenced by the partial pressure of oxygen (Po₂) within the applied sample. Limited in-clinic data exists regarding the quantitative effect of Po₂ in unmanipulated capillary fingertip blood samples across physiologically representative glucose and Po₂ ranges.

Method:

Clinical accuracy data were collected as part of a BGM manufacturer’s ongoing post-market surveillance program for a commercially available GOx-based BGM test-strip. The data set comprised 29 901 paired BGM-comparator readings and corresponding Po₂ values from 5 428 blood samples from a panel of 975 subjects.

Results:

A linear regression-calculated bias range of 5.22% (+0.72% [low Po₂: 45 mm Hg] to −4.5% [high Po₂: 105 mm Hg]); biases calculated as absolute at <100 mg/dL glucose was found. Below the nominal Po₂ of 75 mm Hg, a linear regression bias of +3.14% was calculated at low Po₂, while negligible impact on bias (regression slope: +0.002%) was observed at higher than nominal levels (>75 mm Hg). When evaluating BGM performance under corner conditions of low (<70 mg/dL) and high (>180 mg/dL) glucose, combined with low and high Po₂, linear regression biases ranged from +1.52% to −5.32% within this small group of subjects and with no readings recorded at <70 mg/dL glucose at low and high Po₂.

Conclusions:

Data from this large-scale clinical study, performed on unmanipulated fingertip capillary bloods from a diverse diabetes population, indicate Po₂ sensitivity of the BGM to be markedly lower than published studies, which are mainly laboratory-based, requiring artificial manipulation of oxygen levels in aliquots of venous blood.

Introduction

The oxygen sensitivity of electrochemical test-strips for blood glucose monitors (BGMs), incorporating glucose oxidase (GOx) enzyme, has been widely reported.^{1 -4} Such systems generally employ a mediator that serves as an electron acceptor for the GOx-catalyzed oxidation of glucose,^5,6 with subsequent reoxidation of the mediator at a suitably poised electrode, generating a glucose proportional signal. In physiological systems, dioxygen naturally present within the surrounding medium may also serve as an electron acceptor, with concomitant generation of hydrogen peroxide, which should remain inert at the applied BGM electrode potential. Artificial mediators are preferred to oxygen in BGMs, and indeed in certain continuous glucose monitors (CGMs), because they facilitate signal generation at lower electrode operating potentials, thereby reducing non-glucose-specific signal contributions from electroactive interference within the sample. ⁷ However, an often-quoted drawback of such systems is that oxygen present within the sample may compete with the artificial mediator for electrons within the GOx complex, thereby influencing the mediator-generated glucose-specific signal.^{1 -4},^8,9

The International Organization for Standardization (ISO) standard ISO 15197:2013 mandates inclusion of labeling within a device’s instructions for use should an interfering substance be found to influence accuracy. ¹⁰ Consequently several studies have sought to quantify the extent of GOx-based BGM oxygen interference. Bench studies, as opposed to clinical studies, have received particular attention because they allow manipulation of partial pressure of oxygen (Po₂) levels through incubating samples with added air, thereby facilitating systematic assessments of Po₂ interference across defined Po₂ ranges.^{11 -14} These studies rely on the collection of venous samples, which have relatively low Po₂ (30-40 mm Hg), ¹⁵ requiring careful manipulation to increase Po₂ levels to those consistent with capillary blood (40-100 mm Hg).^16,17 While laboratory studies may provide a higher degree of control of those variables not under assessment, a more clinically appropriate indication of behavior would be obtained by directly measuring the intended sample matrix, as indicated by a device’s instructions for use (capillary blood for many BGMs), thereby lowering the risk of introducing experimental artifacts arising from artificial sample manipulation.

Despite this need, few studies are reported in which the Po₂ sensitivity of GOx-based systems, performed on devices tested with unmanipulated blood samples, has been measured.^18,19 This is understandable, given that large studies are required to separate naturally occurring variation from the effects of Po₂ within wider patient populations. To evaluate this effect, reported here is an extensive clinical data set permitting the large-scale evaluation and impact of naturally occurring Po₂ levels within fingertip capillary blood on the performance of a commercially available GOx-based BGM.

Materials and Methods

Test-Strip

OneTouch Select Plus test-strips are manufactured by LifeScan Inc. and have a capillary blood indication. A reagent chemistry comprising GOx enzyme and ferricyanide mediator is formulated within a polymeric complex and deposited onto a substrate pre-patterned with a five-electrode carbon design. The unique design of the test-strip evaluates the sample during measurement and dynamically adapts algorithm parameters to deliver a hematocrit (Hct)-independent response across the 30%-55% v/v Hct and 20-600 mg/dL blood glucose range.

Clinical Accuracy Measurements

Data were gathered as part of LifeScan’s ongoing clinical post-market surveillance (PMS) activities reported elsewhere. ²⁰ Manufactured batches of test-strips were randomly and routinely sampled and sent to 1 of 3 UK clinics, with BGM readings obtained in singlicate with a target of 100 individual subjects with diabetes. Lancing and filling of test-strips with capillary blood were performed by clinical staff with all aspects of testing performed as per instructions for use. Capillary blood samples were collected in separate capillary tubes for singlicate determination of Po₂ (Siemens Rapidlab 348 blood gas analyzer; tubes containing observable air bubbles discarded) and Hct (Hettich Hematocrit 210 capillary centrifuge). Capillary blood plasma fractions were obtained concurrently from the same lancing site, and glucose levels were determined in duplicate on 2 laboratory comparator devices (YSI STAT PLUS; Yellow Springs Instrument Co Inc, Yellow Springs, OH, USA). All measurement instruments were subject to daily QC checks.

Data Analysis

The ISO 15197:2013 definition of clinical accuracy was applied, where at least 95% of all values must fall within either ±15 mg/dL of the average measured comparator value at glucose concentrations <100 mg/dL or ±15% at glucose concentrations ≥100 mg/dL. Bias was calculated as absolute (Bias(Abs) for glucose <100 mg/dL) and relative (Bias(%) for glucose ≥100 mg/dL):

$$\mathrm{Bias}{(\mathrm{Abs})}_{(\mathrm{glucose}<100\mathrm{mg}/\mathrm{dL})}=\mathrm{BGM}i-\mathrm{Comparator}i$$

$$\mathrm{Bias}{\left(\%\right)}_{(\mathrm{glucose}\ge 100\mathrm{mg}/\mathrm{dL})}=\frac{\mathrm{BGM}i-\mathrm{Comparator}i}{\mathrm{Comparator}i}\times 100\%$$

Results

Po₂ Distributions

A total of 29 901 paired BGM-comparator readings were recorded, corresponding to 5 428 individual blood Po₂ measurements from 975 individual subjects, because multiple tests were performed on a given subject’s blood at each clinic visit. Based on meeting study inclusion criteria, no paired data values were excluded. A mean Po₂ of 75.96 mm Hg (SD: ±12.05; range: 34.7-147.4 mm Hg) was recorded. Based on an assumption of normal distribution, skewness was calculated as +1.14 indicating a moderate distribution asymmetry, weighted toward higher Po₂ (>105 mm Hg), while the kurtosis of 5.31 indicated a greater proportion of values within the distribution tails than a true normal distribution. These distribution parameters agree with the manufacturer’s full clinical PMS Po₂ data set (n = 28 533). ²¹ The only other comparable published study found was Freckmann et al ²² in which the capillary fingertip blood Po₂ levels of mainly diabetes subjects (n = 110) were assessed, yielding a mean Po₂ of 71.1 mm Hg (SD: ±6.9; range 49-86 mm Hg). No relationship between Po₂ levels and Hct was seen: Po₂ (mm Hg) = −0.01882 × hct + 10.90).

Clinical Accuracy by Dissolved Oxygen Range

ISO 15197:2013 defined the clinical accuracy of the 29 901 paired values as shown in Table 1. Accuracy was evaluated across the full Po₂ range and across the low (<45 mm Hg), mid (45-105 mm Hg), and high (>105 mm Hg) Po₂ ranges, at low (<100 mg/dL) and high (≥100 mg/dL) glucose. Acceptable clinical accuracy was achieved in each category (noting that no datapoints were recorded under conditions of low Po₂ and low glucose, indicating the infrequency of this clinical condition). The data indicate that overall, extremes of Po₂ do not have a significant detrimental effect on the BGM’s system accuracy.

Table 1.

Test-Strip Clinical Accuracy Across Different Dissolved Oxygen Ranges, Analyzed According to the ISO 15197:2013 Definition of Clinical Accuracy.

Dissolved oxygen range (mm Hg)	Proportion of subjects (%)	Clinical accuracy (%)	Clinical accuracy (<100 mg/dL glucose) (%)	Clinical accuracy(≥100mg/dL glucose)
All (34.7-147.4)	100	97.7	96.8	97.8
Low (<45)	0.15	96.9	none	96.9
Mid (45-105)	99.1	97.7	96.8	97.8
High (>105)	0.75	98.4	100	98.2

Surveillance error grids (SEGs) of the data set, split by Po₂ levels, are shown in Figure 1a-c. At low Po₂ (<45 mm Hg), all values (98 data-pairs) were within the SEG risk category ‘None’; in the mid Po₂ range (45-105 mm Hg), 98.3% (28 827 data-pairs) were within the risk category “None,” 1.7% (495 data-pairs) were within the risk category “Slight, Lower,” and 6 data-pairs were within the risk category ‘Slight, Higher’; at high Po₂ (>105 mm Hg), 97.7% of values (417 data-pairs) were within the risk category “None” and 2.3% of values (10-data-pairs) were within the risk category “Slight, Lower.” No data-pairs were recorded within any of the higher SEG risk categories.

Figure 1.

Surveillance error grid plots of blood glucose monitor data shown by Po₂ range: (a) low Po₂ (<45 mm Hg), (b) mid Po₂ (45-105 mm Hg), and (c) high Po₂ (>105 mm Hg). Abbreviation: Po₂, partial pressure of oxygen.

Relationship Between BGM Bias and Po₂

The overall linear regression equation governing the relationship between BGM glucose bias (n=29 901) and corresponding Po₂ was

$$\mathrm{Bias}\left(\%,\mathrm{Abs}\right)=\left(-0.0{87\mathrm{Po}}_2\right)+4.641.$$

(1)

As expected, a relationship between BGM bias and Po₂ was evident, quantitatively described as an incremental 10 mm Hg increase in Po₂ corresponding to a 0.87 (%, Abs) decrease in BGM bias. Corresponding linear fit biases were calculated at different Po₂ thresholds, yielding biases of −1.89 (%, Abs) at the nominal Po₂ of 75 mm Hg and biases of +0.72 and −4.50 (%, Abs) at Po₂ extremes of 45 and 105 mm Hg, respectively.

An additional analysis was undertaken to evaluate the oxygen effect on the BGM bias at Po₂ levels above and below the nominal Po₂ of 75 mm Hg. The following linear regression equations were recorded:

$$<75\mathrm{mm}{\mathrm{Hg}\mathrm{Po}}_2:\mathrm{Bias}\left(\%/\mathrm{Abs}\right)=\left(-0.{186\mathrm{Po}}_2\right)+11.520$$

(2)

$$>75\mathrm{mm}{\mathrm{Hg}\mathrm{Po}}_2:\mathrm{Bias}\left(\%,\mathrm{Abs}\right)=\left(+0.00{2\mathrm{Po}}_2\right)-3.033$$

(3)

The BGM exhibits a greater sensitivity to Po₂ levels below the nominal Po₂, with a 2.14-fold increase in slope compared with the full data set (−0.186 vs 0.087). In context, ISO 15197:2013 describes clinical accuracy as BGMs in which >95% of results have a bias less than ±15 mg/dL/±15% to comparator, for which equation (2) allows the linear fit bias of the BGM at low Po₂ (45 mm Hg) to be calculated as +3.14 (%, Abs). The behavior of the BGM at greater than the nominal Po₂ was unexpected, based on the theory relating to oxygen effect on GOx-based oxygen-sensitive BGMs; the slope of +0.002 is negligible (a 10 mm Hg change in Po₂ altering the bias by an average 0.01 [%, Abs]), demonstrating the BGM to be essentially insensitive to Po₂ above the nominal 75-mm Hg population based on the full glucose range encountered.

Relationship Between Glucose Concentration and Po₂

The relationship between BGM bias and Po₂ levels was also evaluated by glucose level: low (<70 mg/dL), mid (70-180 mg/dL), and high (>180 mg/dL)—these glucose levels being selected as representing hypoglycemia, euglycemia, and hyperglycemia, respectively. ²³ The resultant linear regression relationships within each glucose range and at low, nominal, and high Po₂ are shown in Table 2 and Figure 2. While ISO 15197:2013 defines the clinical accuracy of a system as the proportion of individual bias values within ±15 mg/dL or ±15% of reference, the standard also defines the acceptable limit for the effect of an individual interference. In this case, the mean bias of the test system should not exceed ±10 mg/dL or ±10% of reference (at glucose <100 mg/dL and ≥100 mg/dL, respectively). These values are represented as dotted lines in Figure 2.

Table 2.

Linear Regression Relationships for Low, Mid, and High Glucose (<70, 70-180, >180 mg/dL, Respectively) and Corresponding Bias at Low, Nominal, and High Po₂ (45, 75, and 105 mm Hg Respectively).

Glucose (mg/dL)	Slope (%, abs/mm Hg)	c (%, Abs)	Bias (abs) at 45 mm Hg	Bias (%, abs) at 75 mm Hg	Bias (%) at 105 mm Hg
<70	−0.176	12.480	+4.57 (0.00%)	−0.71 (1.45%)	−5.99 (0.00%)
>= 180	−0.066	3.083	+0.10 (0.05%)	−1.90 (40.42%)	−3.89 (0.60%)
70-180	−0.114	6.643	+1.52 (0.28%)	−1.90 (56.37%)	−5.32 (0.83%)

Numbers in parentheses record the percentage of values within a given glucose + Po₂ category.

Figure 2.

Relationship between bias and Po₂ across the low (<70 mg/dL), mid (70-180 mg/dL), and high (>180 mg/dL) glucose ranges, based on linear regression relationships recorded in Table 2. The percentage of paired-data values in each category is recorded within the plot. No samples having low glucose in combination with low (45 mm Hg) and high (105 mm Hg) Po₂ were recorded. Negative biases were seen at high Po₂ for both mid and high glucose, accounting for 0.60% and 0.83% of all values recorded. Dashed lines represent ±10 mg/dL or % bias from comparator. Abbreviation: Po₂, partial pressure of oxygen.

At low glucose, no values were recorded at either low or high Po₂; low glucose in the nominal Po₂ range (1.45% of total BGM readings) yielded a bias of −0.71 mg/dL. The mid glucose range at low Po₂ (0.05% of readings) yielded a bias of +0.10 (%, Abs), compared with −3.89 (%, Abs) at high Po₂. At high glucose, the effect of Po₂ on bias at low and nominal Po₂ was minimal (+1.52%, −1.90%, respectively), while the high glucose/Po₂ combination (0.83% of readings) yielded a bias of −5.32%. For context, the ISO 15197:2013 standard describes a noninterfering effect as a less than ±10 mg/dL or ±10% (<100 mg/dL and ≥100 mg/dL glucose, respectively) difference between a test and nominal condition.

Discussion

This large-scale clinical study, using unmanipulated fingertip capillary blood samples, indicates the Po₂ sensitivity of the BGM system to be markedly lower than published studies. While the postulated oxygen interfering mechanism of electrochemical GOx-BGMs has been described earlier, BGMs are calibrated to nominal Po₂, which is most representative of the wider end-user population. Therefore, blood samples having a higher than nominal Po₂ will tend to underestimate glucose levels and vice versa. Based on the mechanistic interpretation of Po₂ interference, several researchers have sought to quantify this oxygen effect, with investigations having mostly been laboratory rather than clinic-based, as they allow collection of venous blood samples of sufficient volume to allow (relatively) controlled manipulation of Po₂ levels. However, careful control of the bench-manipulation of blood samples, and maintaining sample parameters within acceptable limits, is essential for gathering BGM data representative of the performance obtained with unmanipulated blood samples, if relevant conclusions are to be drawn.^24,25

Baumstark et al ¹¹ have evaluated the Po₂ effect on five anonymized GOx-based BGMs. Aliquots of venous blood were collected (n = 26 subjects) and Po₂ levels immediately determined. For low Po₂, only bloods below a value of 45 mm Hg were evaluated. To raise Po₂ levels to designated nominal and high Po₂ targets (70 and ≥150 mm Hg, respectively), air was introduced prior to incubation. Once Po₂ levels were within target, glucose measurements were performed on each BGM, and mean biases of low and high Po₂ to nominal were calculated. Mean biases at <45 mm Hg varied from +6.1 to +22.6% (min. +3.0%, max. +37.9% by system); mean biases at ≥150 mm Hg varied from −7.9 to −14.9% (min. −5.1%, max. −28.9%), yielding maximum/minimum bias ranges of 37.5% and 14.0%. In a subsequent study in the same laboratory, Schmid et al ¹² reported similar behaviors for five GOx systems (mean biases at <45 mm Hg: +11.8-+44.5%, min. +5.5%, max. +71.5%; mean biases at ≥150 mm Hg: −15.8 to −21.2%, min. −5.1%, max. −28.9%; maximum/minimum bias ranges: 65.7%/26.4%). Similar oxygen interfering observations have been reported in other earlier studies, where again venous blood samples were manipulated to alter Po₂ levels.^13,14 (Note: The launch of the OneTouch Select Plus BGM system postdated both of these studies.)

Baumstark et al, ¹⁸ recognizing the limitations of bench-based Po₂ manipulation studies, performed capillary blood measurements on two anonymized GOx-based BGMs in which no manipulation of sample Po₂ levels had been undertaken. Linear regression equation–calculated relative biases varied from +11.7% to −2.6% and +5.4% to −4.3% by BGM across the Po₂ range, yielding maximum and minimum bias ranges of 14.3% and 9.7%, generally much narrower compared with most bench studies, although sample Po₂ ranges were narrower (mean: 67.1 mm Hg; range: 52-85 mm Hg). The authors concluded that large-scale studies were required to verify Po₂ influences.

The greater bias gradient seen at lower Po₂ in this study were in accordance with other studies,^11,12 although the linear regression–calculated bias of +3.1 (%, Abs) at 45 mm Hg was lower than previously reported. However, a different behavior was seen at Po₂ levels >75 mm Hg, where overall the Select Plus system exhibited minimal oxygen sensitivity. This reduced oxygen insensitivity at higher Po₂ in the OneTouch Select Plus GOx-ferricyanide BGM is in accordance with Chun et al ¹⁹ who, studying native Po₂ levels in subjects receiving elevated oxygen therapy, observed a similar level of oxygen insensitivity in arterial blood samples, albeit at Po₂ levels >150 mm Hg. Similarly, the clinical arm of Tang’s study, ¹⁴ performed on non-oxygen manipulated venous blood samples, incubated in collection vessels for up to 40 minutes, also suggests possible reduced oxygen sensitivity as Po₂ levels naturally increased in the incubated samples. However, these observations challenge the bench studies^11,12 where biases of up to −21.2% were recorded. The study by Baumstark et al, ¹⁸ based on unmanipulated capillary samples, was of limited size with few native Po₂ recorded >75 mm Hg, preventing definitive conclusions from being drawn regarding system behavior at elevated Po₂, a limitation that the authors recognized.

The design of a GOx-based BGM will, in part, dictate the device’s sensitivity to Po₂. For example, the physicochemical nature of a reagent formulation will influence the diffusional behavior of each reactant material. This may explain why a more modest overall bias is seen with the studied BGM versus published data. However, this explanation seems less likely with respect to the observed oxygen insensitivity of the Select Plus BGM at higher Po₂ levels.

An explanation for this behavior may lie in the fact that different methodologies were employed in the two study approaches. The venous blood studies were laboratory-based, requiring venipuncture, collection of aliquots of venous blood into heparinized tubes, and subsequent artificial manipulation of the mid and higher Po₂ samples. The clinical study reported here was based on unmanipulated capillary fingertip blood samples (as per indicated use) and collection by lancet, as per instructions for use. The significant size of the collected data set reported here (n = 29 901 samples) allowed the systematic bias due to Po₂ to be observed against the other drivers of variability (eg, from medications and endogenous interference), thus overcoming the risk incurred by smaller studies of bias being driven by an uncontrolled source of variation.

A further consideration is that, in the native environment, tissues are protected from hypoxic and hyperoxic effects via the sigmoidal relationship between hemoglobin-bound oxygen and free oxygen in the blood. This equilibrium relationship exists for the subject group studied here, where steady state conditions would prevail, but would not be the case in the laboratory studies in which Po₂ levels were manipulated. As such, it may be argued that the data presented here are more representative of actual clinical practice.

The essential insensitivity of the studied BGM to oxygen at above nominal levels across the wider glucose range was unexpected and does not agree with theory. The cause of this behavior is not yet understood but may be assumed representative in that it is supported by ~30 000 datapoints. Oxygen is sparingly soluble in aqueous media; even at the highest recorded Po₂ level (147.4 mm Hg), calculations show that there is very considerable excess of mediator in comparison with oxygen molecules, as per BGM design. As such, features designed for the limitation of glucose flux to the reagent chemistry are not required, unlike the case for glucose sensing systems, including CGMs, that similarly rely on oxygen as the mediating species. ²⁶ In dried form, the enzyme/mediator chemistry is held within a polymer complex, which partially solubilizes during blood wetting and ensuing measurement sequence. A mechanistic explanation based on oxygen limitation from the bulk medium due to the dissolution of this complex would appear to be an unlikely cause for the behavior seen. Indeed, the general linearity of the bias/Po₂ relationship would suggest nonlimitation of the oxygen substrate, with little evidence of deviation from linearity as the upper extreme of Po₂ is reached.

The magnitude of this study addresses certain limitations evident in previously reported investigations, allowing representative oxygen relationships to be observed across both full and defined glucose/Po₂ ranges, as well as against other potentially interfering factors including Hct and exogenous/endogenous substances. However, while BGM samples were obtained as indicated in the instructions for use (unmanipulated fingertip capillary blood), the greater sample volume required for subsequent Po₂ measurement required capillary blood to be collected into a capillary tube for direct introduction to the dissolved gas analyzer, raising the possibility that such sample handling may increase sample Po₂ levels. Po₂ distributions, evaluated against an assumption of normality, did exhibit skewness (+1.14), indicating a weighting toward higher Po₂. As no exclusions were made for subjects on oxygen therapy, this latter factor may also be a contributor to the positive skewness seen.

Conclusions

This data set, based on the manufacturer’s ongoing PMS program, allowed a population-level comparison of the clinical accuracy and Po₂ levels of a panel of unmanipulated capillary fingertip blood samples from diabetes subjects. The size of the data set (n = 29 901) allows the systematic bias engendered by variable Po₂ to be observed against the wider range of variables that may routinely be found in clinic, thus allowing a more representative assessment of the effect of Po₂ on performance of a commercially available GOx-based BGM. Furthermore, the size of the study reasonably includes samples from subjects with comorbidities and thus is representative of the wider BGM user population. While the BGM showed a systematic bias of ~+3% at low Po₂ levels (<45 mm Hg), this was modest in comparison with reports on the performance of other GOx-based BGMs. The low Po₂ sensitivity of the BGM reported here challenges current perceptions regarding the oxygen sensitivity of GOx-based BGMs that have hitherto been mainly based on laboratory studies designed to mimic the behavior of capillary bloods at extremes of Po₂ which, through venous blood sample manipulation, potentially markedly overstate GOx-based BGM Po₂ sensitivity.