204.full.pdf

ABSTRACT Five methods of measuring glucose in

the horse were compared with plasma glucose reference values generated with an Autoanalyzer3 (Seal Analytical Ltd., Hampshire, UK). Plasma glucose was measured with a One Touch Ultra (LifeScan, Milpitas, CA) hand-held glucose meter and a YSI 2300 Stat Plus Glucose and Lactate Analyzer (YSI; YSI Inc., Yellow Springs, OH). Glucose also was measured in whole blood using One Touch Ultra and YSI. Finally, interstitial glucose measurements were obtained with a SEVEN continuous glucose monitor-ing device (Dexcom, San Diego, CA). Glucose measurements were obtained on 2 consecutive days in 6 American Quarter Horses maintained on native prairie hay and water. Each morning, 2 baseline glucose measurements were obtained 30 min apart. After a meal of sweet feed, glucose measurements were obtained every 30 min for 4 h. Patterns of postprandial glucose increases and subsequent decreases were similar across all methods but were displaced vertically from one another. Of the 5 methods, the YSI method with plasma appeared to have the best reproducibility of Autoana-

lyzer3 glucose values based on a random coefficients model with intercept not statistically different from zero (P = 0.08) and slope not statistically different from one (P = 0.08) and by having the highest Lins concordance coefficient (r = 0.77). The other methods had biased random coefficients models with inter-cepts not equal to zero (P > 0.05) and slopes not equal to one (P > 0.05) and Lins concordance coefficient values that ranged from 0.39 to 0.64. Technical diffi-culties with the SEVEN device limited its utility as a reliable method for evaluating glucose in the equine.

Key words: continuous glucose monitor , equine , glucometer , glucose

INTRODUCTION Several disease conditions in the

horse have been linked to insulin resistance: equine metabolic syndrome (Johnson, 2002), laminitis (Coffman and Colles, 1983; Treiber et al., 2006), pituitary adenoma (Garcia and Beech, 1986), hyperlipidemia (Jeffcott and Field, 1985), and osteochondrosis dis-secans (Ralston, 1996). Other condi-tions, such as polysaccharide storage myopathy, render the horse unable to store glycogen normally. To manage these conditions, afflicted horses often are fed low-starch or low-glycemic-in-dex diets (Johnson, 2002; Kronfeld et

al., 2005). Low-starch feeds are being marketed as an aid in management of disease conditions, but they are also being marketed to the general public as safer alternatives to traditional equine rations.

To develop and assess low-starch diets, there is ongoing research that requires the measurement of equine blood glucose concentrations over a period of time. Using laboratory equipment to measure glucose can be cumbersome and labor intensive. Sev-eral point-of-care glucose monitoring devices are convenient, relatively inex-pensive, and readily available for use in humans, but most have not been validated for use in the equine. One of the newer technologies available to diabetic humans is the continuous glucose monitoring (CGM) device (Wentholt et al., 2007). Typically, CGM devices obtain glucose read-ings through a sensor probe inserted subcutaneously. Sensors detect an electrical current generated by a reac-tion between interstitial glucose and glucose oxidase. Sensors transmit a signal to a receiver that records data at predetermined intervals. Aside from its near-continuous collection of glucose data, the CGM dramatically reduces the number of whole blood samples required for effective glucose monitoring, and it collects continu-ous glucose data for up to 7 d. This

The Professional Animal Scientist 27 ( 2011 ):204214

A comparison of methodologies for measuring glucose concentrations in the horse1 T. L. Slough,*2 C. D. Gunkel,* L. W. Murray, and J. S. Drouillard* *Department of Animal Sciences and Industry, and Department of Statistics, Kansas State University, Manhattan 66506

2011 American Registry of Professional Animal Scientists

1 This is contribution No. 11-088-J from the Kansas Agricultural Experiment Station, Manhattan. 2 Corresponding author: [email protected]

Methods of measuring glucose in horses 205

study was designed to assess whether glucose readings obtained from clini-cally normal horses through the use of either a CGM device or a hand-held glucose monitor (both designed for use in human patients) accu-rately reproduce those obtained with laboratory analyses using a reference technique.

MATERIALS AND METHODSAll animal procedures were ap-

proved by the Kansas State Universi-ty Institutional Animal Care and Use Committee. Six American Quarter Horses were used for this study, including 3 mares and 3 geldings. All horses were 4 yr of age and weighed 488 20 kg. Horses were maintained on an ad libitum diet of native prairie hay and water. None of the horses had a history of metabolic disorders and all were considered to be clini-cally normal.

Preliminary Experiments

Preliminary experiments were con-ducted to identify potential problems with the use of a CGM in horses. In the initial experiment, sensors for the SEVEN CGM device (Dexcom, San Diego, CA) were applied in the adipose tissue overlying the latis-simus dorsi muscle of the horse, just posterior to the scapula. The receiver, which the manufacturer recommends remain within 1.8 m of the sensor, was fastened to the halter of each horse. During this study, the SEVEN system had a large number of errors. Many of the receivers required reini-tialization, some failed and had to be replaced, and numerous readings were missed. Customer service representa-tives from Dexcom indicated that SEVEN system receivers do not func-tion optimally in close proximity to metal, such as belt buckles on human patients. Thus, the metal buckles on the halters may have interfered with receiver function. In addition, when the sensor was located in this posi-tion, many of the interstitial probes (consisting of a 330-m filament) were bent to almost 90 upon removal.

Perhaps rapid contraction of the large muscle mass underlying the sensor in this location resulted in physical bending of the probes, thus rendering them nonfunctional.

Due to the large number of miss-ing data points, a follow-up experi-ment was conducted to determine optimum placement of the sensor and receiver. Sensors were placed adjacent to the poll of the horse, posterior to the scapula, on the crest of the neck, posterior to the olecranon, and adjacent to the tail head. Receivers were then attached to braided mane and tail hair, as well as to a harness device (without any metal buckles) around the girth of the horse. Wied-meyer et al. (2003) reported a Min-iMed CGM worked well when the sensor was placed over the masseter muscle and the receiver was attached to a halter on the head of the horse, but the SEVEN system produced the fewest errors when the sensor was placed near the base of the tail of the horse and the receiver was attached to braided tail hair (Figure 1). Thus this positioning of the sensor was used in subsequent experiments.

Current Experiment

On the day before the experiment, each horse was equipped with an indwelling 14-gauge 13-cm jugular catheter and a SEVEN CGM device. The SEVEN sensor and receiver were initialized, installed, and calibrated following the manufacturers instruc-tions. Hair was closely shaved from a patch of skin overlying the adipose tissue adjacent to the base of the tail. The shaved area was cleaned with alcohol, and after insertion, the sensor was firmly attached to the skin of the horse with an adhesive-backed pad provided by the manufacturer, with additional cyanoacrylate adhesive applied to the pad to secure it to the skin. The sensor measured glucose concentration in interstitial tissue and transmitted a signal to a wire-less receiver every 5 min. The receiver was attached to braided tail hair. The SEVEN receiver was calibrated every 12 h (0700 and 1900 h, preprandially)

with the One Touch Ultra (OTU; LifeScan, Milpitas, CA) hand-held glucose meter. Glucose measurements were obtained with the hand-held glu-cose monitor following manufacturer directions. Briefly, approximately 10 mL of blood was drawn through the jugular catheter into a syringe, one drop (approximately 10 L) was extracted and applied to a test strip, and a glucose reading was obtained. The glucose reading obtained with the OTU was then manually entered into the SEVEN receiver to complete the calibration.

On the morning of each experiment, 2 baseline glucose measurements were obtained 30 min apart. After acquisi-tion of the second baseline sample, horses were fed 1.125 kg of commer-cial sweet feed concentrate (Sweet Stuff 12, Cargill Animal Nutrition, Minneapolis, MN) and allowed to eat uninterrupted for 30 min. These horses were unaccustomed to grain meals, so a modest quantity of sweet

Figure 1. The SEVEN sensor (A; Dexcom, San Diego, CA) was inserted into the adipose tissue adjacent to the tail head and the receiver (B) was attached to braided tail hair. This maintained the sensor and receiver in close proximity to each other and avoided possible interference from metal buckles on the halters of the horses.

Slough et al.206

feed was offered to elicit a postpran-dial glucose response without causing digestive upset in the horses. After feeding, glucose measurements were obtained at 30-min intervals for 4 h. At each sampling, the most recent glucose reading of the CGM (which would have occurred within 5 min preceding sampling) was recorded and 10 mL of whole blood was obtained via the indwelling jugular catheter and placed into heparinized vacuum tubes. A single drop (approximately 10 L) of the venous whole blood was applied to a test strip, which was inserted into the OTU test port, as recommended by the manufacturer, to obtain a glucose reading.

Whole blood from each sample also was subjected to glucose analysis by a YSI 2300 Stat Plus Glucose and Lactate Analyzer (YSI; YSI Inc., Yellow Springs, OH). All whole blood glucose measurements were obtained within 30 min of sampling and the YSI produced an intraassay CV of 0.12%. Whole blood was centrifuged at 876 g for 10 min at approxi-mately 13C and plasma was removed. A drop of plasma (10 L) was applied to a test strip and analyzed using the OTU to measure plasma glucose concentration. Remaining plasma was frozen at 20C until analysis by an Autoanalyzer3 (AA3; Seal Analytical Ltd., Hampshire, UK) according to procedures described by Fingerhut et al. (1965). An intraassay CV of 0.35% was generated with the AA3, and when the YSI was used to measure plasma glucose, an intraassay CV of 0.03% was recorded. The entire sampling process was repeated on the following day, yielding a total of 120 discrete blood samples.

Analysis of Plasma Constituents

Each time blood was collected, glucose was measured using 6 meth-ods for each horse. These included 1) interstitial glucose concentration as measured by the SEVEN system, 2) whole blood glucose concentration as measured by the OTU, 3) plasma glucose concentration as measured by

the OTU, 4) whole blood glucose con-centration as measured by the YSI, 5) plasma glucose concentration as measured by the YSI, and 6) plasma glucose concentration as measured by the AA3, the reference method. All glucose measurements were obtained according to manufacturer directions for each instrument used. For the YSI and AA3, standards were run with each set of blood samples to ensure quality control. Quality control was also monitored in the OTU hand-held glucose meter by testing the control solution provided by the manufacturer each morning and each time a new lot of test strips was used.

Statistical Analysis

Perfect reproducibility between 2 methods occurs if the measured glu-cose is identical for the methods, that is, when the measurements are plotted in a bivariate graph, the points fall exactly on the 45 line through the origin (Lin, 1989). Statistical repro-ducibility occurs when the plotted line is not statistically different from the 45 line through the origin.

Three statistical techniques were used to assess whether glucose mea-surements obtained by methods 1 through 5 described in the previous section statistically reproduce those obtained by the AA3, our reference technique. These 3 techniques were used because they each assess repro-ducibility slightly differently. All cal-culations were performed using SAS version 9.2.1 software (SAS Institute Inc., 2007).

First, a random coefficients re-gression model was fitted (Littell et al., 1996) with AA3 values as the response (Y) and one of the 5 other methods as predictor (X) in a simple linear regression. In random coef-ficients regression models, an overall fixed regression model (i.e., the aver-age relationship over all horses on all days) is fitted, which (unlike ordinary regression models) takes into account the inherent random variability due to different horses and different days (i.e., the horseday combination). For each model, the following parameters

were estimated: overall (fixed) inter-cept and slope; variance components for random intercept, random slope, and residual; and covariance compo-nent between random intercept and random slope. In addition, an au-toregressive parameter was fitted to account for serial correlation within each horseday combination. Models were fitted using the SAS MIXED procedure with REML estimation. Normality of residuals was checked using the SAS UNIVARIATE proce-dure.

To assess reproducibility, t-tests were performed to check that the intercept was not statistically differ-ent from zero and that the slope was not statistically different from one for each model. The t-test results, with 95% CI on the regression parameters, are reported. For succinctness, in this paper we will refer to a model with both intercept not different from zero and slope not different from one as an unbiased model.

For an overall comparison of model fit among the 5 predictor methods, Akaikes information criterion (AIC) was used. Akaikes information crite-rion is used to compare models with complex variancecovariance struc-tures (Littell et al., 1996; Burnham and Anderson, 1998). For all methods having an unbiased regression model, the predictor method with the model having the smallest AIC value is considered the best for modeling AA3 values. Any predictors with AIC val-ues within 2 of the minimum AIC are considered equally good at modeling AA3 values, and AIC differences of 7 or more indicate that the predictors are different in ability to model AA3, with the higher AIC being less desir-able (Burnham and Anderson, 1998).

The second statistical technique was described by Bland and Altman (1986). BlandAltman calculations reported here are the mean difference between measurements of 2 methods, the SD of differences, and the number and percent of observed differences outside the 95% prediction interval based on the t-distribution. Assum-ing that differences are approximately normally distributed, then only 5% of

Methods of measuring glucose in horses 207

values should be outside the predic-tion interval. If data are not ap-proximately normal, BlandAltman results should be treated with cau-tion. Therefore, normality of differ-ences was evaluated using the SAS UNIVARIATE procedure. The mean difference reported is the AA3 glucose mean minus the mean glucose of a predictor method, so positive mean differences indicate that a predictor method has lesser glucose measure-ments, on average, than the AA3, whereas negative mean differences indicate the opposite. For these data, BlandAltman calculations were per-formed on both original observations and on residuals plus estimated means from the SAS MIXED procedure to account for variability due to horseday combinations. Results between the 2 were not different because the horseday variance component was typically very small relative to resid-ual variability. Therefore, BlandAlt-man results based on the original data are reported. The authors note that, as Bland and Altman (1986) point out, when possible this statistical analysis should be used in conjunc-tion with knowledge of the clinical (as opposed to statistical) significance of mean differences.

The third technique was Lins concordance coefficient (Lin, 1989),

which is a modification of the Pearson correlation coefficient that not only evaluates the strength of the linear as-sociation between 2 variables (X and Y) like Pearsons, but also evaluates the degree of departure from the 45 line through the origin. Lins concor-dance coefficient is reported with a 95% CI for each predictor. If there is reproducibility between 2 methods, the 95% CI should contain one. As with the BlandAltman analysis, con-cordance coefficient calculations were performed on both the raw data and the residuals plus estimated means to account for the horseday variability. Unlike BlandAltman results, there were some differences between concor-dance coefficients based on raw data or based on residuals, so concordance coefficients based on residuals are reported here.

The SEVEN system (X) also was assessed for its reproducibility of values obtained by OTU used on both plasma (Y1) and whole blood (Y2), because a hand-held glucose meter is generally used in calibrating CGM devices. The comparison was done using all 3 of the previously described statistical analyses.

There were some missing measure-ments in the current study, but these had no effect on validity of the sta-tistical analyses other than the issue

of smaller sample size with several predictor methods, as discussed in the results. Data points were excluded only if there was a mechanical or technical issue with the measurement device.

RESULTS AND DISCUSSIONTechnical Issues with Devices

One of the issues of most concern was the number of missing observa-tions, especially with the SEVEN system (Table 1). Although the greatest number of missing observa-tions (33) occurred in whole blood tested with the YSI, these errors were easily identified and are simple to remedy. In those samples, air bubbles in the sample tubes caused the YSI to generate obviously unrealistic glucose values. Had the presence of the air bubbles been noticed before separat-ing the plasma, the bubbles would have been removed and these missing data points would have been avoided.

It is important to note that there were 12 missing data points observed with the SEVEN system at the time of blood sampling, but because the CGM records glucose measurements every 5 min, there were far more missing data points that occurred throughout the period. It just so

Table 1. A summary of glucose measurements collected1,2,3

Item

AA3 OTU YSI SEVEN

Plasma Blood Plasma Blood Plasma Interstitial fluid

Range (mM) Low 3.820 1.389 4.000 1.470 2.630 2.167 High 8.780 9.111 12.889 6.870 9.215 11.111Mean (mM) 5.824 5.803 8.018 4.147 5.920 4.887Missing data points4 0 0 0 33 0 12Percent of values within 20% of AA3 N/A 70.8 21.7 21.7 84.2 37.51Methods of measuring glucose included measurement of plasma glucose with an Autoanalyzer3 (AA3; Seal Analytical Ltd., Hampshire, UK), glucose in whole blood and plasma with a One Touch Ultra (OTU; LifeScan, Milpitas, CA) hand-held glucometer, glucose in whole blood and plasma with a YSI 2300 Stat Plus Glucose and Lactate Analyzer (YSI; YSI Inc., Yellow Springs, OH), and interstitial glucose with a Dexcom SEVEN continuous glucose monitor (SEVEN; Dexcom, San Diego, CA).2For each method, the following values are given: maximum and minimum glucose value, mean glucose concentration, number of missing data points, and the percent of glucose values within 20% of the values generated by the AA3.3Note that this is a summary of raw data and ignores the autocorrelation among values for a given horse over time.4There were 120 total possible observations with each method tested.

Slough et al.208

happened that only 12 of those oc-curred at the time of blood sampling. Although changing the position of the SEVEN system sensor resulted in a considerable improvement com-pared with the 54 missing data points recorded during the preliminary experiment, missing observations with the SEVEN system are not easily remedied and appear to be due to mechanical errors within the system itself when used in the horse.

Another technical difficulty arose during later experiments when ambient temperatures and humidity increased. As the days became more hot and humid, both the OTU and SEVEN collected condensation inter-nal to the device. The receiver screen on the SEVEN and the output screen on the OTU both became difficult to read, and in many cases the devices did not display any output. These problems, however, were short-lived and most devices returned to normal function when weather conditions im-proved. However, 2 SEVEN receivers never regained normal function and were replaced.

Reproducibility Results

A graph of raw data (Figure 2) re-vealed that, as expected, each method of glucose measurement detected a postprandial increase, followed by a decrease, in glucose concentra-tions. Overall, glucose concentrations recorded using the different methods followed the same trajectory, which was fairly consistent, but they were displaced vertically from one another, with some methods reading high and others reading low relative to the AA3 reference method. Further informa-tion on the raw data is presented in scatter plots (Figures 3 and 4), which illustrate the relationship between measurements obtained by the various methodologies.

Descriptive statistics based on raw data are summarized in Table 1 for all methods. Note that these statistics do not account for horseday variability, autocorrelation over time, or correla-tions between method readings made on individual samples. Therefore,

means and ranges should be treated with caution in nonstatistical compar-isons of methods. However, regarding reproducibility of AA3 readings, the methods of the YSI with plasma and the OTU with whole blood had, re-spectively, 84.2 and 70.8% of samples with glucose readings within 20% of the AA3 readings, whereas fewer than 50% of the glucose measurements obtained by the OTU with plasma, the YSI with whole blood, and the SEVEN system were within 20% of the numerical glucose values obtained by the AA3. In 2003 the International Organization for Standardization established criteria to evaluate the accuracy of glucose monitors: If the glucose concentration is

Methods of measuring glucose in horses 209

system with respect to Lins concor-dance coefficient, but only slightly better for the random coefficients models. Note that it makes sense only to compare AIC among the unbiased random coefficients models to pick a best model. Because there is only one unbiased random coefficients model, AIC values are not compared here.

Because the SEVEN system was calibrated with the OTU, random coefficients regression models, BlandAltman calculations, and Lins con-

cordance coefficient were also used to evaluate whether the SEVEN system was able to reproduce the OTU using whole blood or plasma. Neither of the random coefficients models were close to having an unbiased relationship with (and hence reproducibility of) OTU values. BlandAltman calcula-tions indicated that the difference in means was closer to zero and the prediction interval was narrower when whole blood was used rather than plasma, but both methods generated

the same number of observations outside the prediction interval (

Slough et al.210

It is important to note that the reference values obtained by the AA3 in this experiment were obtained

with plasma samples; however, this is common practice in studies designed to evaluate the accuracy of hand-held

glucometers in measuring glucose in whole blood (Hussain and Sharief, 2000; Kozar et al., 2008; Kong et al., 2010). One might expect glu-cose concentrations to be higher in plasma than in whole blood, simply because of the dilution effect of the red blood cells in whole blood, as well as the fact that blood cells metabo-lize glucose at a rate of 5 to 10% per hour (Stockham, 1995). Researchers have confirmed that plasma glucose values often are larger than values obtained from whole blood (Hussain and Sharief, 2000), as was the case in this experiment, but some attribute these higher values to a hematocrit-dependent error where glucose readings tend to be overestimated in samples with low hematocrit (Louie et al., 2000). Others have cited a differ-ence in glucose distribution in various species, with humans having 50% of glucose distributed within erythro-cytes and 50% within plasma (Cold-man and Good, 1967), whereas dogs have a glucose distribution of 12.5% in erythrocytes and 87.5% in plasma (Coldman and Good, 1967).

Hollis et al. (2008) observed more accurate readings when their Accu-Chek hand-held glucometer was used with equine plasma than when it was used with whole blood. However, one must use caution in comparing results obtained by differing brands of glucometers. Forbes and Brayton (2008) reported higher glucose values obtained by an OneTouch Ultra2 glucometer compared with the Accu-Chek when both models were used in mice. These contrasting findings might be explained by differences in the indicators used by the 2 systems. The OTU uses a carbon electrode indicator and the Accu-Chek Ad-vantage uses a palladium electrode. The OTU also uses a glucose oxidase reaction (Hones et al., 2008), whereas the Accu-Chek Advantage does not (Russell et al., 2007). As a result, the OTU is more sensitive to differ-ences in oxygen tension in samples. Both the Accu-Chek and the OTU glucometers use hexacyanoferrate III/hexocyanoferrate II in the media-tor reaction. This system is prone to

Figure 3. Scatter plots illustrating the agreement in raw glucose concentrations obtained between measurement methods. For each, the vertical axis (Y) represents glucose measurements obtained in plasma using the Autoanalyzer3 (AA3; Seal Analytical Ltd., Hampshire, UK). The horizontal axis represents corresponding glucose measurements obtained in plasma with the One Touch Ultra (OTU; LifeScan, Milpitas, CA; a), whole blood with the OTU (b), plasma with the YSI 2300 State Plus Glucose and Lactate Analyzer (YSI; YSI Inc., Yellow Springs, OH; c), whole blood with the YSI (d), or interstitial fluid with the Dexcom SEVEN continuous glucose monitor (SEVEN; Dexcom, San Diego, CA; e).

Methods of measuring glucose in horses 211

interferences by uric acid, bilirubin, acetaminophen, dopamine, and ascor-

bic acid (Hones et al., 2008). In the normal, healthy, nonfasting, relatively

sedentary horse, one would not expect to see large quantities of these inter-fering substances (Newton and Davis, 1922; Stillions et al., 1971; Snow and Frigg, 1989; Engelking, 1993).

The major advantage of the OTU is its convenience. Although it may be a better predictor of reference values when plasma is used, the data presented here do not unequivocally indicate that using plasma with the OTU would increase accuracy of readings relative to those of the AA3. One caveat to using plasma with the OTU is that the process of separating plasma from whole blood reduces the convenience associated with the hand-held glucose meter and it introduces a potential source of error.

Hand-held glucose meters intended for use in humans are designed for samples containing 30 to 55% hema-tocrit (LifeScan informational package insert, 2009). Under normal circum-

Figure 4. Scatter plots illustrating the agreement in raw glucose concentrations obtained between measurement methods. For each, the vertical axis (Y) represents glucose measurements obtained in interstitial fluid using the Dexcom SEVEN continuous glucose monitor (SEVEN; Dexcom, San Diego, CA). The horizontal axis represents corresponding glucose measurements obtained in whole blood with the One Touch Ultra (OTU; LifeScan, Milpitas, CA; a) and plasma with the OTU (b).

Table 2. Estimated regression coefficients comparing equine glucose concentrations obtained by various methodologies to glucose values obtained in plasma by the AA31,2

Item

AA33

OTU3

Blood Plasma

OTU4 YSI4 SEVEN4 SEVEN4

Blood Plasma Blood Plasma Interstitial fluid Interstitial fluid

Intercept 3.71 2.83 4.35 1.41 4.60 3.97 5.70 SE 0.45 0.43 0.65 0.72 0.38 0.45 0.49 P-value (H0: 0 = 0)

Slough et al.212

stances, the relaxed equine will have a similar hematocrit range. Hackett and McCue (2010) reported that a veteri-nary glucose meter tested in horses was accurate when median packed cell volume was 38%, but when packed

cell volume was 66%, glucose val-ues were only about 52% of values generated by their reference method. However, Hollis et al. (2008) reported that packed cell volume had no effect on the accuracy of the Accu-Chek

hand-held glucose meter in measur-ing glucose in equine whole blood. As Hollis et al. (2008) and Russell et al. (2007) pointed out, equine whole blood may impair the ability of test strips to effectively filter erythrocytes

Table 3. BlandAltman comparisons of equine glucose concentrations obtained by various methodologies to values obtained by the AA31,2

Item

AA33

OTU3

Blood Plasma

OTU4 YSI4 SEVEN4 SEVEN4

Blood Plasma Blood Plasma Interstitial fluid Interstitial fluid

Observations (n) 120 120 87 120 108 108 108diffmean 0.0206 2.1937 1.6681 0.0962 0.9570 0.9339 3.2515SD 1.18 1.31 1.19 0.73 1.60 1.63 1.64No. outside diffmean 2 SD

5 4 4 4 2 4 4

% outside diffmean 2 SD

4.2 3.3 4.5 3.3 1.9 3.7 3.7

1Methods of measuring glucose included measurement of plasma glucose with an Autoanalyzer3 (AA3; Seal Analytical Ltd., Hampshire, UK), glucose in whole blood and plasma with a One Touch Ultra (OTU; LifeScan, Milpitas, CA) hand-held glucometer, glucose in whole blood and plasma with a YSI 2300 Stat Plus Glucose and Lactate Analyzer (YSI; YSI Inc., Yellow Springs, OH), and interstitial glucose with a Dexcom SEVEN continuous glucose monitor (SEVEN; Dexcom, San Diego, CA).2BlandAltman calculations performed on mean of Y minus mean of X (diffmean) with SD of differences.3Y variable.4X variable.

Table 4. Lins concordance coefficient comparisons of equine glucose concentrations obtained by various methodologies to values obtained by the AA31,2

Item

AA33

OTU3

Blood Plasma

OTU4 YSI4 SEVEN4 SEVEN4

Blood Plasma Blood Plasma Interstitial fluid Interstitial fluid

Observations (n) 120 120 87 120 108 108 108Lower CL 0.46 0.55 0.22 0.69 0.23 0.32 0.45Concordance coefficient 0.58 0.64 0.41 0.77 0.39 0.47 0.59Upper CL 0.68 0.72 0.57 0.84 0.52 0.60 0.701Methods of measuring glucose included measurement of plasma glucose with an Autoanalyzer3 (AA3; Seal Analytical Ltd., Hampshire, UK), glucose in whole blood and plasma with a One Touch Ultra (OTU; LifeScan, Milpitas, CA) hand-held glucometer, glucose in whole blood and plasma with a YSI 2300 Stat Plus Glucose and Lactate Analyzer (YSI; YSI Inc., Yellow Springs, OH), and interstitial glucose with a Dexcom SEVEN continuous glucose monitor (SEVEN; Dexcom, San Diego, CA).2Lins concordance coefficient calculations performed on mean + MIXED residuals accounting for horseday variability, which was usually not significant. Upper and lower confidence limits (CL) are shown.3Y variable.4X variable.

Methods of measuring glucose in horses 213

because of the tendency of rouleaux formation in equine red blood cells, or other factors such as microclot forma-tion, hemolysis, viscosity, or inflam-matory markers.

There is considerable variability in the literature regarding how accu-rately glucose values obtained from interstitial fluid reflect blood glucose values. Whereas some have reported a 5-min delay between blood and interstitial glucose concentrations (Re-brin et al., 1999; Kulcu et al., 2003), others have reported that interstitial glucose concentrations are approxi-mately 60% lower in human adipose tissue compared with arterialized plasma concentrations (Regittnig et al., 2003). Several published reports, however, indicate that interstitial glucose in adipose tissue tends to be 85 to 100% of arterialized plasma concentrations (Lonnroth et al., 1987; Jansson et al., 1988; Bolinder et al., 1989; Moberg et al., 1997; Rosdahl et al., 1998; Niklasson et al., 2000). Others point out that a hypoglycemic environment may develop in inter-stitial tissue preceding a decline in blood glucose values (Pickup et al., 1999). Some speculate that differences between interstitial glucose readings and blood glucose measurement may be due to factors other than sensor sensitivity, such as insulin sensitivity of the subject and vascularization of the sensor site (Kulcu et al., 2003).

The potential to use a CGM to measure glucose concentration every 5 min is extremely appealing. Horses in the current experiment exhibited only mild and transient signs of discomfort during sensor placement, and symp-toms of irritation around the insertion site were minimal. Although there were horses that rubbed their sen-sors on the fence, those occurrences were relatively rare. If the CGM obtained accurate measurements and performed dependably, it would be immensely useful as a research or clinical tool. Although conclusions drawn in this experiment regarding the ability of the SEVEN system to estimate AA3 values might vary from one statistical analysis to another, the reproducibility of measurements ob-

tained was inferior to that of the YSI when plasma was used. Others have reported accurate readings obtained with CGM devices in horses, although previous researchers validated CGM readings in only 36 samples and used a different CGM device (Wiedmeyer et al., 2003).

The ability to gather near-continu-ous glucose data in the equine could be invaluable to researchers, but the technical problems and frequent system failures experienced in this research were an impediment to usage of the CGM in horses. Sensors are advertised to last >5 d in humans, and although some sensors functioned for longer than this in individual horses, many of them ceased normal function by d 3. This required the insertion and initialization of new sen-sors. The devices are expensive, and each initialization takes several hours. Consequently, some of the logistical issues in implementing these devices may limit their usefulness to equine researchers or clinicians.

Because of the need to obtain imme-diate glucose readings for calibration, readings obtained by the OTU using whole blood were used for calibra-tion of the SEVEN system. Using a more accurate method of measuring glucose concentrations for the cali-bration of the SEVEN system may further decrease the bias of the read-ings obtained by the CGM relative to AA3 readings, as might further study into the effect of timing of calibration (Buckingham et al., 2006). Calibrat-ing the SEVEN system with plasma glucose measured by the OTU would be considerably less convenient than using whole blood, and there is not overwhelming evidence, based on this experiment, that there would be any advantage in doing so.

IMPLICATIONSIn this work, none of the methods

tested met the 2003 criteria estab-lished by the International Organiza-tion for Standardization (ISO 15197). Consequently, if absolute glucose values or statistically significant data are required, as in clinical emergency

cases, a standard laboratory refer-ence technique will provide the most accurate glucose measurements. However, all methods tested detected a postprandial increase and then decrease in glucose concentrations. If the objective is simply to determine trends or make comparisons, then the hand-held glucometer or CGM may be sufficient. If the SEVEN system is to be used, technical issues will have to be addressed.

ACKNOWLEDGMENTSThe authors would like to thank

the staff at the K-State Beef Cattle Research Center and the K-State Horse Unit (Kansas State University, Manhattan) for their assistance with horses and facilities used in the proj-ect. As well, students enrolled in ASI 661 Special Problems at Kansas State University were tremendously helpful in the collection of glucose samples.

LITERATURE CITEDBland, J., and D. Altman. 1986. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1:307.

Bolinder, J., E. Hagstrom, U. Ungerstedt, and P. Arner. 1989. Microdialysis of subcu-taneous adipose tissue in vivo for continuous glucose monitoring in man. Scand. J. Clin. Lab. Invest. 49:465.

Buckingham, B. A., C. Kollman, R. W. Beck, A. Kalajian, R. Fiallo-Scharer, M. J. Tansey, L. A. Fox, D. M. Wilson, S. A. Weinzimer, K. J. Ruedy, W. V. Tamborlane, and D. S. Grp. 2006. Evaluation of factors affecting CGMS calibration. Diabetes Technol. Ther. 8:318.

Burnham, K. P., and D. R. Anderson. 1998. Model Selection and Inference. Springer-Verlag, New York, NY.

Coffman, J. R., and C. M. Colles. 1983. Insulin tolerance in laminitic ponies. Can. J. Comp. Med. 47:347.

Coldman, M. F., and W. Good. 1967. The distribution of sodium, potassium and glucose in the blood of some mammals. Comp. Bio-chem. Physiol. 21:201.

Engelking, L. R. 1993. Equine fasting hyper-bilirubinemia. Adv. Vet. Sci. Comp. Med. 37:115.

Fingerhut, B., R. Ferzola, W. H. Marsh, and A. B. Miller. 1965. Automated methods for blood glucose and urea. Clin. Chem. 11:803.

Slough et al.214

Forbes, N. A., and C. F. Brayton. 2008. Glu-cometer validity in mouse research. J. Am. Assoc. Lab. Anim. Sci. 47:167.

Garcia, M. C., and J. Beech. 1986. Equine in-travenous glucose-tolerance testGlucose and insulin responses of healthy horses fed grain or hay and of horses with pituitary-adenoma. Am. J. Vet. Res. 47:570.

Grosenbaugh, D. A., J. E. Gadawski, and W. W. Muir. 1998. Evaluation of a portable clini-cal analyzer in a veterinary hospital setting. J. Am. Vet. Med. Assoc. 213:691.

Gunkel, C. D., J. S. Drouillard, and T. L. Slough. 2009. The use of a handheld glucom-eter for measuring glucose concentrations from whole blood collected from the horse. J. Anim. Sci. 87(E-Suppl. 2):772.

Hackett, E. S., and P. M. McCue. 2010. Evaluation of a veterinary glucometer for use in horses. J. Vet. Intern. Med. 24:617.

Hollis, A. R., B. L. Dallap Schaer, R. C. Boston, and P. A. Wilkins. 2008. Compari-son of the Accu-Chek Aviva point-of-care glucometer with blood gas and laboratory methods of analysis of glucose measurement in equine emergency patients. J. Vet. Intern. Med. 22:1189.

Hones, J., P. Muller, and N. Surridge. 2008. The technology behind glucose meters: Test strips. Diabetes Technol. Ther. 10:S10.

Hussain, K., and N. Sharief. 2000. The inac-curacy of venous and capillary blood glucose measurement using reagent strips in the newborn period and the effect of haematocrit. Early Hum. Dev. 57:111.

ISO (International Organization for Stan-dardization). 2003. In vitro diagnostic test systemsRequirements for blood-glucose monitoring systems for self-testing in managing diabetes mellitus. ISO 15197. http://www.techstreet.com/standards/ISO/15197_2003?product_id=1095349.

Jansson, P. A., J. Fowelin, U. Smith, and P. Lonnroth. 1988. Characterization by microdialysis of intracellular glucose level in subcutaneous tissue in humans. Am. J. Physiol. 255:E218.

Jeffcott, L. B., and J. R. Field. 1985. Cur-rent concepts of hyperlipemia in horses and ponies. Vet. Rec. 116:461.

Johnson, P. J. 2002. The equine metabolic syndrome peripheral Cushings syndrome. Vet. Clin. North Am. Equine Pract. 18:271.

Kong, G. W. S., W. H. Tam, M. H. M. Chan, W. Y. So, C. W. K. Lam, I. P. C. Yiu, K. M. Loo, and C. Y. Ki. 2010. Comparison in the performance of glucose meters in blood glu-cose monitoring during pregnancy. Gynecol. Obstet. Invest. 69:264.

Kozar, J., A. M. Simundic, N. Nikolac, M. Zirovic, and E. Topic. 2008. Accu Chek Compact Plus blood glucometer evaluation. Biochem. Med. 18:361.

Kronfeld, D. S., K. H. Trieber, T. M. Hess, and R. C. Boston. 2005. Insulin resistance in the horse: Definition, detection, and dietetics. J. Anim. Sci. 88:E22.

Kulcu, E., J. A. Tamada, G. Reach, R. O. Potts, and M. J. Lesho. 2003. Physiological differences between interstitial glucose and blood glucose measured in human subjects. Diabetes Care 26:2405.

Lin, L. I.-K. 1989. A concordance coeffi-cient to evaluate reproducibility. Biometrics 45:255.

Littell, R., G. Milliken, W. Stroup, and R. Wolfinger. 1996. SAS for Mixed Models. SAS Inst. Inc., Cary, NC.

Lonnroth, P., P. A. Jansson, and U. Smith. 1987. A microdialysis method allowing char-acterization of intercellular water space in humans. Am. J. Physiol. 253:E228.

Louie, R. F., Z. P. Tang, D. V. Sutton, J. H. Lee, and G. J. Kost. 2000. Point-of-care glu-cose testingEffects of critical care variables, influence of reference instruments, and a modular glucose meter design. Arch. Pathol. Lab. Med. 124:257.

Moberg, E., E. Hagstrom-Toft, P. Arner, and J. Bolinder. 1997. Protracted glucose fall in subcutaneous adipose tissue and skeletal muscle compared with blood during insulin-induced hypoglycaemia. Diabetologia 40:1320.

Newton, E. B., and A. R. Davis. 1922. Combined uric acid in human, horse, sheep, pig, dog, and chicken blood. J. Biol. Chem. 54:603.

Niklasson, M., A. Holmang, M. Sjostrand, L. Strindberg, and P. Lonnroth. 2000. Muscle glucose uptake is effectively activated by ischemia in type 2 diabetic subjects. Diabe-tes 49:1178.

Pickup, J., L. McCartney, O. Rolinski, and D. Birch. 1999. In vivo glucose sensing for diabetes management: Progress towards non-invasive monitoring. BMJ 319:1289.

Ralston, S. L. 1996. Hyperglycemia/hyper-insulinemia after feeding a meal of grain to young horses with osteochondritis dissecans (OCD) lesions. Pferdeheilkunde 12:320.

Rebrin, K., G. M. Steil, W. P. van Antwerp, and J. J. Mastrototaro. 1999. Subcutaneous glucose predicts plasma glucose independent of insulin: Implications for continuous moni-toring. Am. J. Physiol. 277:E561.

Regittnig, W., M. Ellmerer, G. Fauler, G. Sendlhofer, Z. Trajanoski, H. J. Leis, L. Schaupp, P. Wach, and T. R. Pieber. 2003. Assessment of transcapillary glucose ex-change in human skeletal muscle and adipose tissue. Am. J. Physiol. Endocrinol. Metab. 285:E241.

Rosdahl, H., K. Hamrin, U. Ungerstedt, and J. Henriksson. 1998. Metabolite levels in human skeletal muscle and adipose tissue studied with microdialysis at low perfusion flow. Am. J. Physiol. 274:E936.

Russell, C., J. E. Palmer, R. C. Boston, and P. A. Wilkins. 2007. Agreement between point-of-care glucometry, blood gas and laboratory-based measurement of glucose in an equine neonatal intensive care unit. J. Vet. Emerg. Crit. Care 17:236.

SAS Institute Inc. 2007. SAS OnlineDoc 9.1.2. SAS Inst. Inc., Cary, NC. http:// support.sas.com/onlinedoc/912/ docMainpage.jsp.

Snow, D. H., and M. Frigg. 1989. Oral-admin-istration of different formulations of ascorbic-acid to the horse. J. Equine Vet. Sci. 9:30.

Stillions, M. C., S. M. Teeter, and W. E. Nelson. 1971. Ascorbic acid requirement of mature horses. J. Anim. Sci. 32:249.

Stockham, S. L. 1995. Interpretation of equine serum biochemical profile results. Vet. Clin. North Am. Equine Pract. 11:391.

Treiber, K. H., D. S. Kronfeld, and R. J. Geor. 2006. Insulin resistance in equids: Pos-sible role in laminitis. J. Nutr. 136:2094S.

Wentholt, I. M., J. B. Hoekstra, and J. H. Devries. 2007. Continuous glucose monitors: The long-awaited watch dogs? Diabetes Technol. Ther. 9:399.

Wiedmeyer, C. E., P. J. Johnson, L. A. Cohn, and R. L. Meadows. 2003. Evaluation of a continuous glucose monitoring system for use in dogs, cats, and horses. J. Am. Vet. Med. Assoc. 223:987.