Materials and Methods

Horses—Samples of jugular venous blood were obtained from 3 groups of horses that were chosen to allow thorough documentation of the performance of 2 ctCO₂ analyzers. The first group consisted of 6 healthy untrained adult mares (median age, 8 years [range, 4 to 27 years]) from the Purdue University Department of Veterinary Clinical Sciences teaching herd. The horses were housed in a paddock and maintained on pasture with free access to water. Supplemental feed concentrate was fed as needed to any horse that was not maintaining its body weight through feeding on pasture alone. All horses were judged to be healthy on the basis of physical examination findings and had had no clinical signs of disease within 3 months prior to the study.

The second group consisted of 6 Standardbreds (4 fillies and 2 geldings; age, 4 to 6 years) obtained from local sources. The horses were trained 5 times/wk for 6 weeks on a high-speed treadmill and were fed a typical training diet with a dietary cation-anion difference approximately equivalent to 200 mEq/kg of dry matter.

The third group consisted of 135 trained Standardbreds from the Indiana Downs racetrack in Shelbyville, Ind, as part of the routine prerace testing to detect the administration of alkalinizing agents. Blood samples were obtained by personnel of the Indiana Horse Racing Commission within 2 hours before the start of the race. The study protocol was reviewed and approved by the Purdue University Animal Care and Use Committee.

Sample collection and storage—In the first group of healthy untrained horses, 4 of the 6 horses were given 200 g of NaHCO₃^a in 3 L of deionized water via nasogastric tube and jugular venous blood samples were obtained approximately 1 hour later to obtain plasma samples with a ctCO₂ near the action threshold (37.0 mmol/L). For sample collection from all 6 horses, the venipuncture site over the left jugular vein of each was swabbed with 70% alcohol and venous blood was collected with a 0.9 × 25.4-mm needle^b and tube holder into eight 4-mL partially evacuated plastic tubes containing lyophilized lithium heparin.^c Blood was collected into each tube until it reached the designated line on the tube because failure to completely fill the tubes would have promoted escape of CO₂ from serum into the partial vacuum above the sample, thereby resulting in measured tCO₂ values that underestimate the true value by as much as 2 mmol/L.^18,19 Care was taken to remove the last tube from the needle before removing the needle from the jugular vein to prevent aspiration of atmospheric air into the tube. Each tube was gently inverted 8 times immediately after collection to ensure appropriate mixing of blood and anticoagulant in accordance with the manufacturer's recommendations. Filled tubes were stored vertically during transport (10 minutes) to the laboratory in an insulated container that had an interior temperature of approximately 4°C.

In the group of 6 trained Standardbreds, NaHCO₃ was administered by nasal intubation at doses of 0, 0.5, and 1.0 g/kg in 3 L of distilled water in a randomized crossover design 4 hours before the horses underwent an SRP on a treadmill. Jugular venous blood samples were collected into a 12-mL polypropylene syringe through a 1.65 × 75-mm IV catheter immediately before treatment and at specific points up to 4 hours after the start of the SRP. A 1.27 × 25.4-mm needle was connected to the syringe, and the needle was inserted into 4-mL partially evacuated plastic tubes containing lyophilized lithium heparin.^c The syringe contents were then allowed to flow into the blood collection tube, and the tube was inverted 8 times.

Jugular venous blood samples were collected from the 135 trained Standardbreds before racing with the same method as described for the 6 healthy untrained horses, except that blood was collected into 3-mL partially evacuated plastic tubes containing lyophilized lithium heparin.^c Filled tubes were stored vertically during storage and transportation (2 hours) to the laboratory in an insulated container that had an interior temperature of approximately 4°C.

At the laboratory, all blood collection tubes were centrifuged for 10 minutes at 1,040 × g in a fixed-angle centrifuge as recommended by the manufacturer. Centrifugation was completed within 30 minutes after blood collection for the 6 healthy adult horses and 6 trained Standardbreds undergoing an SRP. Centrifugation was completed within 8 hours after blood collection for the 135 trained Standardbreds from the racetrack, and all samples were analyzed within 12 hours after collection.

Plasma ctCO₂ analysis—Two analyzers were used in the study: commonly used analyzer A^d and less expensive analyzer B.^e Both analyzers measure ctCO₂ through ion-selective electrode (potentiometric) methodology. The analyzers acidify a known volume of plasma or serum; acidification converts all free and protein-bound CO₂ forms to CO₂. The liberated CO₂ then diffuses through a gas permeable membrane into an internal filling solution and combines with water to form H₂CO₃, which dissociates into H⁺ and HCO₃⁻. The increase in H⁺ activity in the internal filling solution is then detected by a pH electrode²⁰; as such, the measurement of ctCO₂ is indirect in a chemical sense.²¹ The ctCO₂ (in mmol/L) is calculated from the rate of change in the mV output from the pH electrode by use of proprietary algorithms based on the assumption that the pH decrease after acidification is proportional to the number of millimoles of tCO₂ in the acidified sample volume. It is likely that this algorithm assigns a standard plasma protein concentration of 70 g/L to the sample volume.^21,22

Analyzer A was maintained and calibrated in accordance with the manufacturer's recommendations by use of the supplied low (0 mmol/L) and high (30 mmol/L) ctCO₂ calibrators.^d The unit is reported by the manufacturer to measure plasma or serum ctCO₂ from 5 to 40 mmol/L with a within-day CV < 3.0% and a total CV < 4.5%. Once the manufacturer's recommended calibration requirements were met, the stopper was removed from the tube and 0.5 mL of plasma was gently aspirated with a plastic transfer pipette and transferred to the sample cup on the tCO₂ analyzer. The sample cup had a height (20 mm)-to-diameter (7 mm) ratio of 2.86 with a reported measured free water loss of 2.6%/h to 2.9%/h at room temperature.^23,24 This was equivalent to a decrease of < 0.2% in free water during the 4-minute measurement period used in this study. Plasma ctCO₂ was measured in quadruplicate or triplicate at approximately 30-second intervals by automatic aspiration of 50 μL from the sample cup. Total CO₂ measurements in plasma were completed within 4 minutes after initially removing the cap from the tube, and the mean ctCO₂ value of replicates was calculated. The decrease in plasma ctCO₂ due to CO₂ loss during the 4-minute measurement period was estimated to be < 1.0% (14.4 %/h²⁵). The net effect of free water loss and CO₂ loss during the 4-minute measurement period was therefore estimated to produce a decrease in ctCO₂ of < 0.9%, equivalent to < 0.33 mmol/L at an action threshold of 37.0 mmol/L.

Analyzer B was maintained and calibrated in accordance with the manufacturer's recommendations by use of the supplied ctCO₂ calibrators.^e The unit is reported by the manufacturer to measure ctCO₂ from 5 to 50 mmol/L, with a within-day CV of < 3.0% and a day-today CV of < 5.5% and recommended fixed performance limits of mean ± 3.0 mmol/L. A 2-point calibration of the sensor was performed every 2 hours. Once the manufacturer's calibration requirements were met, the stopper was carefully removed from the tube and 220 μL of plasma was directly aspirated from the tube by the tCO₂ analyzer in stat mode at approximately 1-minute intervals (analytic time, 52 seconds), starting immediately after transfer of 0.5 mL of plasma to the sample cup on analyzer A. The cap was replaced on the tube between successive aspirations of each plasma sample by analyzer B. Direct aspiration of the plasma sample from the tube in stat mode was used because ctCO₂ decreases when equine plasma is sequentially aspirated from the blood collection tube into the sample cup.¹⁹ Total CO₂ concentration was measured in quadruplicate or triplicate by the tCO₂ analyzer within 4 minutes after initially removing the stopper from the tube, and the mean ctCO₂ value of replicates was calculated.

Instrument validation—A 4-month familiarization period was used for both tCO₂ analyzers. Method comparison was conducted in accordance with guidelines^26–29 established by the CLSI and involved characterization of 3 factors related to instrument validation: linearity over a specified ctCO₂ range of interest, total analytic error, and method comparison and bias estimation. Of these 3 factors, linearity up to a ctCO₂ of 40 mmol/L and estimation of bias through the use of Bland-Altman plots were considered the primary methods for instrument validation, whereas total analytic error was determined to facilitate comparisons of repeatability and reproducibility with other instrumentation.

Four aqueous tCO₂ standards (A^f: 27, 30, 33, 36, and 39 mmol/L; N^g: 10, 20, 30, and 40 mmol/L; P^h: 5, 10, 20, 25, 40, and 50 mmol/L; and Vⁱ: 5, 10, 20, 30, and 40 mmol/L) were used to evaluate linearity once daily for 4 days. Each of the 4 aqueous linearity standards was from the same batch number. The 4 linearity standards were used because they have been used, or potentially could be used, to demonstrate linearity for analyzer A. The mean of quadruplicate ctCO₂ measurements for each aqueous tCO₂ standard (dependent variable) was plotted against the nominal value (independent variable), and linear regression was performed.

Three aqueous tCO₂ standards (A^f: 27, 30, 33, 36, and 39 mmol/L; N^g: 10, 20, 30, and 40 mmol/L; and V^h: 5, 10, 20, 30, and 40 mmol/L) were used to evaluate linearity once daily for over a period of approximately 3 months. Each of the 3 aqueous linearity standards was from the same batch number. The mean of triplicate ctCO₂ measurements for each aqueous standard was determined. Linear regression equation estimates for the coefficient and intercept values for 3 aqueous linearity calibrators for the 2 analyzers were calculated, provided that the R² value for the linear regression equation exceeded 0.990.

The presence of curvilinearity was explored by fitting quadratic and cubic equations to the data. As indicated in CLSI guidelines,²⁷ curvilinearity was deemed to exist when the coefficient for the quadratic or cubic term was not equal to 0. Acceptable linearity standards for both tCO₂ analyzers were identified on the basis of the presence of linearity to 40 mmol/L, the range of R² values, and a low mean CV value for the ctCO₂ value nearest an action threshold of 37.0 mmol/L.

Imprecision, which is a measurement of the random error that occurs in a method, reflects the closeness of analytic results obtained from a series of replicate measurements. Imprecision is commonly measured in repeatable and reproducible conditions. Repeatable conditions exist when a sample is analyzed multiple times on the same day with the same instrument in the same laboratory. In contrast, reproducible conditions exist when a sample is analyzed multiple times on different days or with different instruments or different laboratories.

Total analytic error reflects the sum of random error (imprecision) and bias (systematic error, derived from constant error and proportional error). No clear consensus exists as to best method to calculate total analytic error for instrument validation.^28–31 For the study, total analytic error was estimated by calculating the S_T in accordance with CLSI recommendations for 1 run/d through the use of estimates for S_r, the SD of the daily means (B) that provides an estimate of reproducibility, and the number of replicates per run (n), whereby S_T = (B² + [{n − 1}/n] × S_r²)^½.²⁸

A short-term repeatability study was performed over 4 consecutive days to provide an estimate of the random error, summarized as the S_r. Repeatability within an 8-hour period on each of the 4 days was determined by analyzing plasma samples 16 times from a horse fed a typical diet (regular ctCO₂) and by analyzing plasma samples 16 times from a horse 1 hour after administration of NaHCO₃ by intragastric intubation (high ctCO₂ = regulatory threshold). The S_r estimates included the effect that variations in the filling of partially evacuated tubes had on measured ctCO₂¹⁹ and therefore address the variability in racing jurisdictions that require collection of multiple blood samples for testing. Repeatability within each 8-hour period was also determined by analyzing an aqueous trueness control standard (A; ctCO₂ = 36.0 mmol/L) 20 times. Mean values for SD and CV for the 4 consecutive days were calculated.

A longer-term reproducibility assessment was conducted by analyzing an aqueous trueness control standard (A; ctCO₂ = 36.0 mmol/L) once daily, analyzing 3 stabilized human serum standards for analyzer A (nominal ctCO₂, 12, 21, and 31 mmol/L^d) twice daily, and analyzing 2 bovine albumin standards for analyzer B (nominal ctCO₂, 13 and 23 mmol/L^e) twice daily for an approximately 3-month period.

The methodology used for instrument validation was in accordance with CLSI recommendations.²⁶ The test method (analyzer B) was compared with the reference method (analyzer A) by analyzing in triplicate jugular venous plasma samples obtained from 6 horses with a large range of ctCO₂ values. A scatterplot of the ctCO₂ values was created for the test method (y-axis) and reference method (x-axis) from the mean values determined with triplicate measurements by both tCO₂ analyzers at different time points of plasma samples from the 6 trained Standardbreds that orally received NaHCO₃ (0.5 or 1.0 g/kg in 3 L of deionized water) or a placebo (3 L of deionized water) and underwent an SRP. Three approaches to method comparison through the use of this data set were used: calculation of the Pearson product moment correlation coefficient from ordinary least squares regression,^j weighted Deming linear regression analysis,^k and difference (Bland-Altman) plots.^k The difference plots were adjusted for multiple observations per individual as described elsewhere.^32,33

Ordinary linear regression was performed with all ctCO₂ values from both tCO₂ analyzers for each horse, and the mean and 95% CI for the intercept (constant bias) and slope (proportional bias) values for each of the 6 horses were determined; point estimates of bias from ordinary linear regression are reliable (but not the 95% CI for the estimate) when the correlation coefficient is > 0.975,³⁴ and the point estimates of bias and 95% CIs are considered reliable when the correlation coefficient is > 0.990.³⁵

Weighted Deming linear regression analysis was performed for each horse because the SD for results of both tCO₂ analyzers was proportional to the ctCO₂. Mean estimates and 95% CIs for constant bias (intercept) and proportional bias (slope) for the 6 horses were calculated; however, the 95% CI for constant and proportional bias determined by Deming regression is unreliable when the SDs of the methods' results vary as a function of analyte concentration,³⁴ as occurred in the study reported here.

A difference plot (Bland-Altman plot) was created for the difference between the test method and comparison method for all data from the 6 horses, with the difference (expressed as a percentage of the mean resulting from both methods) on the y-axis and the mean of both methods on the x-axis.^32,33,36 The horizontal line of the difference plot identified the least squares mean difference percentage bias between the 2 analyzers and was calculated as described³³ to adjust for the multiple observations per horse. The percentage bias value was regarded as providing the best estimate of bias between the 2 analyzers because the SD was proportional to the mean value.^32,33,37

Linear regression equation estimates for the coefficient and intercept values for the aqueous linearity calibrators for the 2 analyzers were used to correct the measured ctCO₂ values, provided that the R² value for the linear regression equation exceeded 0.990.^j This approach has been commonly used in Australia and Hong Kong in an attempt to provide analytic consistency (precision and accuracy) when analyzer A is used to measure ctCO₂ in equine plasma.^6,8,11

Statistical analysis—Data are reported as mean ± SD; values of P < 0.05 were considered significant for all analyses. Normality of data distribution was assessed by visual inspection of a histogram and normal probability plots, performance of the Shapiro-Wilk test, and determination of the degree of kurtosis and skewness.^l Multivariable polynomial regression analysis^k was performed to determine the linear, quadratic, and cubic associations between ctCO₂ values and the nominal value for aqueous standards. A linear relationship was deemed to exist when nonsignificant coefficient values were obtained for the quadratic and cubic terms.²⁷ Generalized linear mixed model ANOVA^m was used to determine whether the order of analysis (3 to 4 replicates) for ctCO₂ determination was significant, with data reported as least squares mean and SE.

Method comparison was performed through the use of difference plots, with bias as the mean percentage difference between results for the 2 analyzers and the 95% limits of agreement as the interval defined from observed bias ± 1.96 × SD. The SD was estimated through generalized linear mixed model analysis as described³³ because of within-individual correlation between successive measurements.

Linear regression analysis^k was performed to determine the association of the difference between the measured value of ctCO₂ for the test method and reference method with the plasma protein concentration as determined by refractometry.ⁿ The difference in the measured ctCO₂ values was expressed as a percentage of the mean value for the ctCO₂ measured by the 2 analyzers. Two data sets were used for linear regression analysis: prerace plasma samples from the 135 Standardbreds and 103 plasma samples from the 6 trained Standardbreds undergoing the SRP. The purpose of these analyses was to determine whether the difference in the ctCO₂ measured by the 2 analyzers was associated with the plasma protein concentration^38,39; demonstration of an association or lack of an association would facilitate identification of the preferred method to correct the measured ctCO₂ on analyzer B to a traceable standard.

Results

Analyzer A—A linear relationship was evident for results obtained from analysis of all 4 aqueous tCO₂ standards (A, N, P, and V) when tested with analyzer A during 4 consecutive days with ctCO₂ (Table 1). A linear relationship was identified between the measured ctCO₂ and all 3 aqueous tCO₂ standards (A, N, and V) when tested for approximately 3 months (R² > 0.990).

The SD for values from most replicate analyses increased proportionally to the mean values, resulting in a relatively constant CV (Tables 2 and 3). Measured values for the CV for fresh equine plasma were within the manufacturer's claims for human plasma and were < 2.0% for midrange and high values for ctCO₂ and the 4 aqueous standards when tested over 4 consecutive days. The CVs were > 2.0% for some values of ctCO₂ < 20 mmol/L in 2 aqueous standards (N and V). Values for SD and CV were numerically greater when measured over approximately 3 months rather than over 4 consecutive days.

When results were used from 818 ctCO₂ measurements for plasma samples collected over approximately 3 months from 8 horses, no significant (P = 0.77) effect of measurement order was identified of order; the first, second, and third ctCO₂ measurements were 31.45, 31.45, and 31.06 mmol/L (least squares mean SE, 0.97).

Analyzer B—A linear relationship was evident for results obtained from analysis of the 3 aqueous nominal tCO₂ standards (A, P, and V) with analyzer B when tested for 4 consecutive days (Table 1); however, analysis of 1 aqueous standard (N) resulted in marked curvilinearity, particularly when the ctCO₂ exceeded 30 mmol/L. A linear relationship existed between the measured ctCO₂ and 2 nominal aqueous tCO₂ standards (A and V) when tested over 3 months (R² > 0.990). Curvilinearity was evident in results pertaining to the remaining aqueous tCO₂ standard (N).

The SD for results of the replicate analyses increased proportionally to the mean values, and the CV was relatively constant (Tables 2 and 3). Measured CV values for fresh equine plasma were within the manufacturer's claims and were < 1.5% with midrange and high values for ctCO₂ and < 2.0% for all aqueous standards when tested over 4 consecutive days. The exception was that the CV value was higher for the 40.0 mmol/L aqueous standard N (5.0% within-day CV; 5.8% between-days CV). As expected, values for SD and CV were numerically greater when measured over approximately 3 months rather than over 4 consecutive days.

Evaluation of total analytic error through the use of an aqueous standard of 36.0 mmol/L^f measured in 20 replicates for 4 consecutive days yielded an S_T of 0.58 mmol/L (CV, 1.6%; S_r, 0.62 mmol/L; and SD of the daily means, 0.23 mmol/L). Evaluation of total analytic error with a human serum standard^a (nominal ctCO₂ of 31.0 mmol/L) measured twice a day for 28 days over approximately 3 months yielded a mean ctCO₂ of 30.5 mmol/L and S_T of 1.43 mmol/L (CV, 4.7%; S, 1.24 mmol/L; and SD of the daily means, 1.13 mmol/L).

Evaluation of total analytic error by use of an aqueous standard of 36.0 mmol/L measured in 20 replicates for 4 consecutive days yielded an S_T of 0.49 mmol/L (CV 1.4%; S_r, 0.49 mmol/L; and SD of the daily means, 0.24 mmol/L). Evaluation of total analytic error with a bovine albumin standard (nominal ctCO₂ of 23.0 mmol/L) measured twice a day for 28 days over approximately 3 months yielded a mean ctCO₂ of 23.5 mmol/L and S_T of 0.61 mmol/L (CV, 2.6%; S_r, 0.54 mmol/L; and SD of the daily means, 0.47 mmol/L).

When results were used from 847 ctCO₂ measurements for plasma samples collected over approximately 3 months from 6 trained Standardbreds undergoing an SRP, no significant (P = 0.90) effect of measurement order on ctCO₂ values was identified. The first, second, and third ctCO₂ measurements were 32.38, 32.26, and 32.10 mmol/L (least squares mean SE, 0.95).

Analyzer comparison for uncorrected plasma ctCO₂ values—Analytic values were compared for the 103 plasma samples from 6 trained Standardbreds, with a large range in ctCO₂ from 9.9 to 44.8 mmol/L. Ordinary linear regression of values for each of the 6 horses (R² = 0.90 to 0.99) provided mean estimates for the intercept (2.43 mmol/L; 95% CI, −0.83 to 5.68 mmol/L) and slope (0.96; 95% CI, 0.84 to 1.08) that were not significantly different from the line of identity (intercept of 0 and slope of 1).

Weighted linear Deming regression analysis of values for each of the 6 horses provided mean estimates for constant bias (intercept, 1.8 mmol/L; 95% CI, −0.8 to 4.3 mmol/L) and proportional bias (slope, 0.98; 95% CI, 0.88 to 1.08) that were not significantly different from the line of identity (Figure 1).

Scatterplot (A) and difference (Bland-Altman) plot (B) for the comparison of uncorrected jugular venous ctCO₂ as measured by 2 tCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) in 103 plasma samples from 6 trained Standardbreds undergoing an SRP. Plasma ctCO₂ as measured with analyzer A (standard method) ranged from 9.9 to 44.8 mmol/L. A—The dotted line is the line of identity, and the solid line is the mean Deming weighted linear regression line for the 6 horses (y = 0.98x + 1.8), which was not significantly (ie, P ≥ 0.05) different from the line of identity. B—Differences are expressed as a percentage of the mean value, against the mean value for measured ctCO₂. The horizontal solid line identifies the mean percentage bias between the 2 analyzers (2.7%), and the horizontal dashed lines reflect the 95% limits of agreement (mean percentage bias ± 1.96 × SD = −9.4% to 14.8%), which is equivalent to the range of differences that contains 95% of future measurements.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Scatterplot (A) and difference (Bland-Altman) plot (B) for the comparison of uncorrected jugular venous ctCO₂ as measured by 2 tCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) in 103 plasma samples from 6 trained Standardbreds undergoing an SRP. Plasma ctCO₂ as measured with analyzer A (standard method) ranged from 9.9 to 44.8 mmol/L. A—The dotted line is the line of identity, and the solid line is the mean Deming weighted linear regression line for the 6 horses (y = 0.98x + 1.8), which was not significantly (ie, P ≥ 0.05) different from the line of identity. B—Differences are expressed as a percentage of the mean value, against the mean value for measured ctCO₂. The horizontal solid line identifies the mean percentage bias between the 2 analyzers (2.7%), and the horizontal dashed lines reflect the 95% limits of agreement (mean percentage bias ± 1.96 × SD = −9.4% to 14.8%), which is equivalent to the range of differences that contains 95% of future measurements.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

Scatterplot (A) and difference (Bland-Altman) plot (B) for the comparison of uncorrected jugular venous ctCO₂ as measured by 2 tCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) in 103 plasma samples from 6 trained Standardbreds undergoing an SRP. Plasma ctCO₂ as measured with analyzer A (standard method) ranged from 9.9 to 44.8 mmol/L. A—The dotted line is the line of identity, and the solid line is the mean Deming weighted linear regression line for the 6 horses (y = 0.98x + 1.8), which was not significantly (ie, P ≥ 0.05) different from the line of identity. B—Differences are expressed as a percentage of the mean value, against the mean value for measured ctCO₂. The horizontal solid line identifies the mean percentage bias between the 2 analyzers (2.7%), and the horizontal dashed lines reflect the 95% limits of agreement (mean percentage bias ± 1.96 × SD = −9.4% to 14.8%), which is equivalent to the range of differences that contains 95% of future measurements.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

A difference plot indicated a bias of 2.7% between the 2 analyzers and an estimated SD of 3.1%. The limits of agreement were consequently −9.4% to 14.8%, with 1.9% (2/103) of the measured values falling outside the limits of agreement. Because the rules of racing in most jurisdictions require expression of ctCO₂ to 1 decimal place, the difference plot estimate for bias indicated that a ctCO₂ value of 38.0 mmol/L as measured with analyzer B (test method) in equine plasma with a mean protein concentration of 63 g/L was equivalent to a value of 37.0 mmol/L as measured with analyzer A (reference method).

Analyzer comparison for effect of plasma protein concentration—Analytic values for prerace plasma samples from 135 trained Standardbreds were compared between analyzers, revealing a mean ± SD ctCO₂ of 33.7 ± 1.9 mmol/L, mean difference in ctCO₂ between analyzers B and A of −0.2 ± 1.3 mmol/L, and mean plasma protein concentration of 63 ± 4 g/L. The results of linear regression analysis for the difference in ctCO₂ versus sample plasma protein concentration (R² = 0.05, Figure 2) indicated that plasma protein concentration had a significant impact on the difference in measured ctCO₂, as evidenced by estimated values for the intercept of 3.9 mmol/L (95% CI, 0.7 to 7.2 mmol/L; SE, 1.64 mmol/L; P = 0.018) and slope of −0.065 (95% CI, −0.117 to −0.014; SE, 0.026; P = 0.013) of the regression equation.

Scatterplot of the difference in the prerace plasma ctCO₂ values as measured with 2 ctCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) for plasma samples from 135 trained Standardbreds, plotted against the plasma protein concentration determined by refractometry. The thick solid line is the linear regression line (y = 3.92 − 0.065x). The dashed lines indicate the 95% CI for the linear regression line, and the outer solid lines are the 95% CI for prediction.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Scatterplot of the difference in the prerace plasma ctCO₂ values as measured with 2 ctCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) for plasma samples from 135 trained Standardbreds, plotted against the plasma protein concentration determined by refractometry. The thick solid line is the linear regression line (y = 3.92 − 0.065x). The dashed lines indicate the 95% CI for the linear regression line, and the outer solid lines are the 95% CI for prediction.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

Scatterplot of the difference in the prerace plasma ctCO₂ values as measured with 2 ctCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) for plasma samples from 135 trained Standardbreds, plotted against the plasma protein concentration determined by refractometry. The thick solid line is the linear regression line (y = 3.92 − 0.065x). The dashed lines indicate the 95% CI for the linear regression line, and the outer solid lines are the 95% CI for prediction.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

Analytic values for the 103 plasma samples from 6 trained Standardbreds undergoing the SRP were compared between analyzers, revealing a mean ctCO₂ of 32.6 ± 8.5 mmol/L, mean difference in ctCO₂ between analyzers B and A of 0.8 ± 2.0 mmol/L, and mean plasma protein concentration of 67 ± 7 g/L. The results of linear regression analysis for the difference in ctCO₂ versus sample plasma protein concentration (R² = 0.22) indicated that plasma protein concentration had a significant impact on the difference in measured ctCO₂, as evidenced by estimated values for the intercept of 5.2 mmol/L (95% CI, 1.4 to 9.0 mmol/L; SE, 1.9 mmol/L; P = 0.008) and slope of −0.060 (95% CI, −0.114 to −0.005; SE, 0.27; P = 0.032). These results were similar in magnitude and direction to those obtained for plasma samples from the 135 trained Standardbreds.

Analyzer comparison for aqueous standard ctCO₂ values—The difference plot for the comparison between analyzers of 148 aqueous standard A ctCO₂ values (nominally, 27, 30, 33, 36, and 39 mmol/L) measured over an approximately 3-month period revealed a large bias of 13.4% (95% CI, 12.6% to 14.2%; SE, 0.40%), which was equivalent to 5.0 mmol/L at a ctCO₂ of 37 mmol/L. The limits of agreement were 3.8% to 22.9%, and 5.4% (8/148) of the measured values fell outside the limits of agreement. Mean standard A–corrected values for the concentration of an aqueous trueness control standard (A; ctCO₂, 36.0 mmol/L) as measured over a 3-month period were 35.6 ± 0.9 mmol/L (CV, 2.6%; n = 17) for analyzer A and 35.7 ± 0.7 mmol/L (CV, 1.9%; 17) for analyzer B.

The difference plot for the comparison between analyzers of 148 aqueous standard V ctCO₂ values (nominally, 5, 10, 20, 30, and 40 mmol/L) measured over an approximately 3-month period yielded a large bias of 12.9% (95% CI, 11.9% to 13.9%; SE, 0.50%). The limits of agreement were 2.2% to 23.6%, and 5.0% (6/119) of the measured values fell outside the limits of agreement. Mean standard V–corrected values for the concentration of an aqueous trueness control standard (A; ctCO₂, 36.0 mmol/L) as measured over a 3-month period were 36.1 ± 1.1 mmol/L (CV, 3.0%; n = 22) for analyzer A and 36.8 ± 1.0 mmol/L (CV, 2.7%; 22) for analyzer B.

Analyzer comparison for corrected plasma ctCO₂ values—The linear regression equations obtained through analysis of aqueous linearity standards A and V were used to correct measured plasma ctCO₂ values to provide a corrected ctCO₂ value for both analyzers.

When standard A–corrected ctCO₂ values were used for 91 plasma samples from 6 horses measured over an approximately 3-month period, weighted linear Deming regression analysis of results for each of the 6 horses provided mean estimates for constant bias (intercept, −2.1 mmol/L; 95% CI, −4.7 to 0.5 mmol/L) and proportional bias (slope, 0.98%; 95% CI, 0.92% to 1.03%) that indicated equivalency over the measurement range (Figure 3). A difference plot indicated a large bias of −11.9% between the 2 analyzers. Limits of agreement were −35.4% to 11.5%, with 0% (0/91) of the measured values falling outside the limits of agreement.

Scatterplot (A) and difference (Bland-Altman) plot (B) for the comparison of aqueous linearity standard–corrected jugular venous ctCO₂ as measured by 2 ctCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) in 91 plasma samples obtained over 3 months from 6 trained Standardbreds undergoing an SRP. A—The dotted line is the line of identity, and the solid line is the Deming weighted linear regression line (y = 0.98x − 2.14). B–The horizontal solid line identifies the mean percentage bias between the 2 analyzers (–11.9%), and the horizontal dashed lines reflect the 95% limits of agreement (mean percentage bias ± 1.96 × SD = −35.4% to 11.5%), which is equivalent to the range of differences that contains 95% of future measurements.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Scatterplot (A) and difference (Bland-Altman) plot (B) for the comparison of aqueous linearity standard–corrected jugular venous ctCO₂ as measured by 2 ctCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) in 91 plasma samples obtained over 3 months from 6 trained Standardbreds undergoing an SRP. A—The dotted line is the line of identity, and the solid line is the Deming weighted linear regression line (y = 0.98x − 2.14). B–The horizontal solid line identifies the mean percentage bias between the 2 analyzers (–11.9%), and the horizontal dashed lines reflect the 95% limits of agreement (mean percentage bias ± 1.96 × SD = −35.4% to 11.5%), which is equivalent to the range of differences that contains 95% of future measurements.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

Scatterplot (A) and difference (Bland-Altman) plot (B) for the comparison of aqueous linearity standard–corrected jugular venous ctCO₂ as measured by 2 ctCO₂ analyzers (commonly used analyzer A and less expensive analyzer B) in 91 plasma samples obtained over 3 months from 6 trained Standardbreds undergoing an SRP. A—The dotted line is the line of identity, and the solid line is the Deming weighted linear regression line (y = 0.98x − 2.14). B–The horizontal solid line identifies the mean percentage bias between the 2 analyzers (–11.9%), and the horizontal dashed lines reflect the 95% limits of agreement (mean percentage bias ± 1.96 × SD = −35.4% to 11.5%), which is equivalent to the range of differences that contains 95% of future measurements.

Citation: American Journal of Veterinary Research 74, 8; 10.2460/ajvr.74.8.1091

• Download Figure
• Download figure as PowerPoint slide

When standard V–corrected ctCO₂ values were used for 103 plasma samples from the 6 horses measured during an approximately 3-month period, weighted linear Deming regression analysis for each of the 6 horses provided mean estimates for constant bias (intercept, 0.5 mmol/L; 95% CI, −0.7 to 1.6 mmol/L) and proportional bias (slope, 0.92%; 95% CI, 0.88% to 0.96%) that indicated the presence of bias. A difference plot indicated a large bias of −7.9% between the 2 analyzers. The limits of agreement were −22.0% to 6.2%, with 0% (0/103) of the measured values falling outside the limits of agreement.

Discussion

The major finding of the study reported here was that analyzer B provided adequate precision and linearity (when specific aqueous standards were used) and may be a suitable replacement for the more expensive, commonly used analyzer A for measurement of ctCO₂ in equine plasma. Other major findings were that values for ctCO₂ produced by analyzer B were higher than those produced by analyzer A and were affected by changes in the plasma protein concentration. We also found that A and V aqueous solutions were satisfactory for demonstrating linearity for analyzer B over the concentration range of interest; however, because these aqueous solutions do not contain protein, they should not be used to provide corrected ctCO₂ values when analyzer B is used because they do not have a plasma protein concentration that approximates the reference range for horses in training and racing.

The total analytic error for measurement of 1 aqueous standard (A, 36 mmol/L) was determined, and the linearity of analyzer results for 4 aqueous standards was examined. The ctCO₂ values obtained with analyzer A are routinely adjusted in Australia⁸ and were adjusted in at least 1 study⁴⁰ in the United States by applying a linear regression equation developed by use of results from analysis of aqueous linearity standards.

The use of aqueous linearity standards as calibrators is not consistent with longstanding recommendations that only protein-containing control solutions should be used to verify the potentiometric measurement of pH and electrolytes in plasma samples, whereas protein-free quality-control samples can be used to verify linearity.⁴¹ Linearity was evaluated in the study reported here through the use of 4 aqueous solutions because appropriate traceable plasma or serum-based linearity solutions were not available, although 1 linearity standard (A) was reported to match plasma viscosity.^8,9,12 Another standard^o has been used to confirm linearity and calculate corrected ctCO₂ values for equine plasma measured with analyzer A,¹¹ but this linearity standard is no longer available. As discussed previously, linearity should ideally be evaluated with a matrix similar to the samples being analyzed (ie, plasma or serum).^27,41,42 Correction of plasma ctCO₂ values by use of the results of linear regression analysis of aqueous standards A and V in this study did not appear to be helpful in improving precision and overcorrected plasma ctCO₂ values obtained with analyzer B (Figure 3). These findings, coupled with our finding that values for N and P aqueous ctCO₂ solutions were not linear when > 30 mmol/L and analyzer B is used strongly, suggest analytic differences between analyzers A and B in measuring ctCO₂. Such differences were particularly evident in the study reported here when aqueous solutions were measured. We believe the major analytic difference is related to the degree of sample dilution during analysis.

The difference in the ctCO₂ values obtained with analyzers B and A increased as the plasma protein concentration decreased. The estimated intercept values for the linear regression equation relating the difference in prerace ctCO₂ values with plasma protein concentration was 3.9 mmol/L for 135 trained Standardbreds. The intercept estimate (equivalent to measurement of ctCO₂ in an aqueous solution) should be compared with our measured mean difference value of 3.9 mmol/L for ctCO₂ measured by the 2 analyzers with the use of aqueous standard A at a concentration of 33 mmol/L, and measured mean difference values of 3.1 and 4.1 mmol/L for ctCO₂ measured by the 2 analyzers with the use of aqueous standard V at concentrations of 30 and 40 mmol/L, respectively (Table 2). Taken together, these results indicated that the value for ctCO₂ obtained with analyzer B was affected by changes in the plasma protein concentration (and therefore the volume of water in plasma) to a much greater degree than occurs with analyzer A. The most likely explanation for this finding is that analyzer B minimally dilutes the sample before measurement, whereas analyzer A extensively dilutes the sample.

Analyzer B is reported by the manufacturer to directly deliver the plasma sample to the flow cell, where it is minimally diluted with an acid, providing a sample to acid ratio of 3:1, which is equivalent to 1:1.33 dilution.^p A ctCO₂ value obtained with analyzer B is primarily dependent on the total number of millimoles of bicarbonate in the aspirated sample volume, which is dependent on the bicarbonate concentration in the free water component of plasma. The plasma protein concentration, in turn, affects the amount of free water in plasma. A decrease in the protein concentration in plasma or serum will therefore increase the value for ctCO₂ obtained with analyzer B. This is because even though the bicarbonate concentration in plasma water may be minimally changed, the volume of plasma water aspirated in the sample volume relative to protein is increased and therefore the number of millimoles of bicarbonate that is aspirated in the sample volume is increased.

Analyzer A is reported to mix 50 μL of sample with 1.0 mL of diluent, whereby the sample is delivered to the flow cell and further diluted with 1.0 mL of sulfuric acid (0.16M).¹² The extensive 1:41 dilution of the plasma sample that occurs in this analyzer before the sample is acidified means that changes in plasma protein concentration exert a relatively small effect on the ctCO₂ values it yields. For example, if one assumes that analyzers A and B produce identical values for a ctCO₂ of 30 mmol/L when the protein concentration is 70 g/L, then dilution of the sample during measurement would produce a plasma protein concentration of 52.5 g/L for analyzer B (70/1.333) and 1.7 g/L for analyzer A (70/41). Consequently, the volume of plasma water in a standardized plasma sample after dilution is 947.5 mL/L of plasma (analyzer B) and 998.3 mL/L of plasma (analyzer A). Standard multiplication factors of 1.055 (1/0.9475; analyzer B) and 1.002 (1/0.9983; analyzer A) would therefore need to be applied to relate the measured ctCO₂ value back to the fixed aspirated volume of the original sample with an assumed standard plasma protein concentration of 70 g/L.²¹ As a consequence, analysis of an aqueous solution lacking protein will result in measured values 5.5% higher for analyzer B and 0.2% higher for analyzer A, whereas analysis of a plasma sample with a protein concentration of 70 g/L would be expected to produce equivalent results.

The large difference in sample dilution before analysis therefore provides an explanation as to why ctCO₂ values are similar when analyzer A is used to measure aqueous or protein-containing solutions and why values are higher when analyzer B is used to measure aqueous solutions. However, the calculated difference for analyzer B is approximately half that obtained when aqueous solutions are analyzed (Table 3). The remaining difference is likely to be due to a combination of 2 things. The first is a true analyzer difference of 2.5% (calculated from the estimated value of 2.7% in this study for mean plasma protein concentration of 67 g/L and subtracting 0.079% in measured ctCO₂ for each 1 g/L decrease in plasma protein concentration lower than 70 g/L), and the second is a standard adjustment to the measured ctCO₂ by analyzer B of +1.5% because of the binding of bicarbonate to plasma protein.⁴³ Taken together, analyzer B yields an increase in the measured value for ctCO₂ of 9.8% (1.055 × 1.025 × 1.015) when used to analyze aqueous solutions; this estimate approximates that observed in the present study.

A decrease in the prerace plasma protein concentration of 1 g/L in plasma samples from 135 trained Standardbreds was associated with an analyzer B value for ctCO₂ of 0.065 mmol/L, which was higher than that obtained with analyzer A. Interestingly, the observed increase of 0.065 mmol/g of plasma protein was similar to that observed (approx 0.06 mmol/g of plasma protein) when ctCO₂ was measured in human plasma by means of photometric titrimetry, which is a coupled enzymatic method that involves the use of phosphoenol-pyruvate carboxylase, or continuous flow analysis.⁴⁴ All 3 of these analytic methods test a fixed volume of undiluted plasma and would therefore be influenced by changes in the sample plasma protein concentration in a manner similar to the situation with analyzer B.

The dependence of ctCO₂ on plasma protein concentration when measured with analyzer B means that the plasma protein concentration should be measured whenever the ctCO₂ obtained with analyzer B exceeds the action threshold. Our results suggested an additional allowance of 0.065 mmol/g of plasma protein for ctCO₂ values obtained with analyzer B should be provided for every g/L decrease in plasma protein concentration from the assumed standard plasma specimen of 70 g/L, which approximates a mass concentration of plasma water of 0.93 kg/L of plasma.^21,22 A resulting equation is as follows: calculated ctCO₂ (mmol/L) = (analyzer B ctCO₂ [mmol/L]) − 0.065 × (70 − plasma protein concentration [g/L]). Application of this equation to the plasma samples from the 1 trained Standardbred that had a prerace ctCO₂ > 37.0 mmol/L as indicated by analyzer B had a clear impact on the categorization of this test result in that consideration of the plasma protein concentration (60.5 g/L) for this horse led to a decrease in the calculated ctCO₂ by 0.6 mmol/L, from 37.1 to 36.5 mmol/L. Analyzer A yielded a prerace ctCO₂ of 35.8 mmol/L for the same horse.

The manufacturer of analyzer B recommends that only analyzer-specific reagents, calibrators, cleaning agents, and control solutions be used because these are the only materials that have been fully characterized and tested. Although not evaluated in our study, aqueous linearity standards of relatively low cost are available from the manufacturer of analyzer B at nominal ctCO₂ values of 8, 15, 30, and 40 mmol/L. Equivolume addition of the 15 and 30 mmol/L standards and the 30 and 40 mmol/L standards could be used to provide additional nominal standards of 22.5 and 35 mmol/L, respectively. On the basis of our study results and our understanding of the determinants of ctCO₂ in plasma and serum, it would appear prudent to use aqueous tCO₂ standards to verify linearity and report the uncorrected value for ctCO₂ when measuring plasma or serum ctCO₂ with analyzer B, provided that the analyzer meets the manufacturer's requirements for performance and linearity to 40 mmol/L is demonstrated. Development of accurate plasma or serum-based ctCO₂ standards would appear to be a priority as well as complete characterization of the effect that changes in plasma protein concentration have on the measured ctCO₂ values when analyzer A or B is used.

The analyzer A used for this study was old and consequently represented a challenge to maintain in calibration. Although our values for S_r of 0.85 mmol/L when measuring an aqueous standard of 36.0 mmol/L were numerically higher than previous estimates of 0.60 mmol/L¹² and 0.28 mmol/L,¹⁴ the other 2 estimates appeared to involve standard A–corrected ctCO₂ values, whereas our values were uncorrected. Moreover, our measured values for CV were < 2.0% for fresh equine plasma with midrange and high ctCO₂ values and for the 4 aqueous standards when tested over 4 consecutive days. These values remained within the manufacturer's claims for human plasma of a within-day CV of < 3.0% and total CV of < 4.5%.

An ongoing challenge with automated methods for plasma or serum ctCO₂ measurement is that some of the dissolved CO₂ will be lost from the sample into the atmosphere because anaerobic conditions cannot be maintained during and after placement of the sample on the analyzer.⁴⁵ The loss of dissolved CO₂ from the sample cup decreases the measured value for ctCO₂, whereas the loss of free water because of evaporation from the sample cup increases the measured value for ctCO₂. We minimized the loss of CO₂ from the sample cup by keeping the interval from opening of the partially evacuated tube to the final measurement of tCO₂ to < 5 minutes and did not identify a significant decrease in ctCO₂ between sequential measurements that occurred at approximately 30-second intervals for analyzer A and at approximately 1-minute intervals for analyzer B.

The lack of a significant decrease was different from the findings of another study¹⁹ in which we aspirated plasma at approximately 1-minute intervals into a sample cup for measurement on analyzer B. It would appear that excessive aspiration can facilitate CO₂ loss from the plasma sample and should therefore be avoided whenever possible. At room temperature (approx 20°C), the predicted decrease in ctCO₂ for plasma stored in the 0.5-mL sample cups used for analyzer A has been estimated to be 14.4%/h,²⁵ which is equivalent to a decrease of 0.077 mmol/L for each minute of storage in the sample cup on analyzer A when the mean ctCO₂ is 32 mmol/L. We identified a numeric decrease in least squares mean ctCO₂ from 31.45 mmol/L to 31.06 mmol/L between the first and third analyses on analyzer A, which typically required 4 minutes. Although not significant, the magnitude of the decrease in ctCO₂ (0.39 mmol/L) was similar to that predicted for plasma placed in 0.5-mL sample cups for 4 minutes (0.31 mmol/L). This observation emphasizes that the interval between placement of plasma in 0.5-mL sample cups and analysis should be kept as short as possible if accurate values for ctCO₂ are to be obtained.

The present study demonstrated that analyzer B provided adequate precision and linearity and is therefore a suitable replacement for analyzer A for measurement of ctCO₂ in equine plasma. An uncorrected ctCO₂ value of 38.0 mmol/L obtained with analyzer B was equivalent to an uncorrected value of 37.0 mmol/L obtained with analyzer A in a group of Standardbreds with a mean plasma protein concentration of 63 g/L; 37.0 mmol/L is the action threshold for ctCO₂ used in many racing jurisdictions. Aqueous standards A and V appear suitable for use with analyzer B to evaluate linearity; however, our results did not support the use of protein-free aqueous standards for correction of plasma ctCO₂ values obtained with analyzer B. Plasma protein concentration should be measured whenever the plasma ctCO₂ value obtained with analyzer B exceeds the action threshold, given that lower-than-usual plasma protein concentrations will result in a higher measured value for ctCO₂.