Point-of-Care Blood Gases, Electrolytes, Chemistries, Hemoglobin, and Hematocrit Measurement in Venous Samples from Pet Rabbits
Point-of-care testing is an attractive option in rabbit medicine, because it permits rapid analysis of a panel of electrolytes, chemistries, blood gases, hemoglobin, and hematocrit, requiring only 65 μL of blood. The purpose of this study was to evaluate the performance of a portable clinical analyzer for measurement of pH, partial pressure of CO2, Na, chloride, potassium, blood urea nitrogen, glucose, hematocrit, and hemoglobin in healthy and diseased rabbits. Blood samples obtained from 30 pet rabbits were analyzed immediately after collection by the portable clinical analyzer (PCA) and immediately thereafter (time <20 sec) by a reference analyzer. Bland-Altman plots and Passing-Bablok regression analysis were used to compare the results. Limits of agreement were wide for all the variables studied, with the exception of pH. Most variables presented significant proportional and/or constant bias. The current study provides sufficient evidence that the PCA presents reliability for pH, although its low agreement with a reference analyzer for the other variables does not support their interchangeability. Limits of agreement provided for each variable allow researchers to evaluate if the PCA is reliable enough for their scope. To the authors’ knowledge, the present is the first report evaluating a PCA in the rabbit.
Introduction
Rabbits are among the most common pets in the households of Europe and the US, representing a significant part of most small animal practices’ patients in Great Britain.1–4 Nevertheless, reference blood gas analysis values and their variations during the course of diseases have never been properly reported in pet rabbits. The absence of previous studies focusing on blood gas in rabbits are possibly due to the cost related with an in-house reference analyzer (RA) and the difficulty in obtaining adequate volume samples in rabbits for shipping them to a clinical laboratory.
A point-of-care portable clinical analyzer (PCA) could overwhelm those limitations. PCAs have become fairly commonplace in emergency and critical care settings in both human and veterinary hospitals.5 The PCA decreases sample processing time, which possibly decreases the effect of delay in estimation of blood gases.6, 8 Although minimal training is mandatory to adhere to standard operating procedures, previous studies have found that the operator independence of PCAs renders them easy to use even by personnel untrained in laboratory techniques.5,9
Recently, a cartridge that combines analysis of blood gas (pH, partial pressure of CO2 [PCO2], bicarbonate total CO2, base excess in the extracellular fluid, anion gap, and hematocrit [Hct], hemoglobin [Hb], selected electrolytes (Na, chloride [Cl], potassium [K]), and chemistries (blood urea nitrogen [BUN], glucose) measurement has been produced.10 The volume of blood required for the analysis was 65 µL, a volume that can be safely harvested from almost any rabbit. Indeed, such a cartridge seems a promising diagnostic tool to assist rabbits’ clinicians in managing emergencies and in monitoring response to critical care and fluid therapy.11,12
One of the major concerns regarding the use of PCAs is the validity of measurements by point-of-care devices compared with primary, bench-top instruments used in the clinical laboratory.9,13–16 An adequate agreement between instruments ensures comparability of test results.15 For that purpose, method comparison studies testing analytical performances of PCAs have been performed in several species, including dogs, cats, horses, sheep, cattle, mice, northern elephant seals (Mirounga angustirostris), viscachas (Lagostomus maximus), white-tailed deer (Odocoileus virginianus), and chickens.5,17–25 Method comparison studies compare type and magnitude of systematic errors (i.e., constant and proportional bias) between an instrument and a “true value”.26 Unfortunately, in veterinary clinical pathology, certified species-specific reference material or reference methods are seldom available, and routinely applied methods are frequently used as the method to which a new method is compared.26 To the authors’ knowledge, no data evaluating the performance of a PCA in rabbits are currently available.
The purpose of the current study was to compare the performance of a PCA, using a specific cartridge, to a reference bench-top analyzer for blood gas analysis and selected electrolyte, chemistry, Hct, and Hb measurement in venous samples collected from pet rabbits. The specific hypothesis was that the PCA and the RA, as in previous comparative researches, provided interchangeable results.17,20,25
Materials and Methods
Study Design and Animals
An observational prospective cross-sectional study was planned. The primary outcome was the agreement between the PCA and the RA. Pet rabbits (n = 30) undergoing blood sampling in the Clinic for Exotic Animals, Veterinary Specialty Center for unrelated diagnostic reasons were planned to be included in the current study. To cover a wide range of analyte concentrations, both healthy and diseased rabbits were included. The healthy rabbits presented to the clinic either for the summer pension service or were rabbits undergoing elective surgery (castration/spaying), and underwent blood analysis for a routine preliminary assessment of their clinical status. If those rabbits did not have any abnormalities on physical examination and other clinical pathology results, they were considered healthy (n = 13). All the other rabbits included in the current study in which blood collection was preliminary to either semielective or urgent surgeries or was necessary to confirm the diagnosis of a pathologic status were considered diseased. Diseased rabbits (n = 17) underwent blood sampling for several primary purposes, including hematology, and/or biochemistry, and/or serologic testing.
The predefined exclusion criterion was that the rabbit did not need blood collection. All the rabbits presented to the clinic in the study period were included in the analysis until the predefined sample size was achieved.
The study was performed in compliance with the directive 2010/63/EU of the European parliament and of the European council. The owners provided written informed consent to the inclusion of samples of their animals in the study.
Instrumentation
A PCAa with disposable cartridgesb and a RAc were used side by side. For quality control of the output results, PCA cartridges were stored in a refrigerator (3–8°C), and cartridges were only used prior to the expiration date provided by the manufacturer. The temperature of each cartridge was allowed to equilibrate with the ambient temperature of the room (20°C) before the cartridge pouch was opened for 5 min, although a study proved evidence that results of analyses were not substantially affected by use of cold, rather than warm, test cartridges.17 The PCA aqueous control solutionsd were used to assess one cartridge out of each cartridge batch as recommended by the manufacturer. The ampule containing the control solution was equilibrated for 4 hr at room temperature. Plain syringes were used to transfer the aqueous control from the ampule to the cartridge as recommendede. Calibration verification of the PCA was not performed because it was explicitly not recommended by the manufacturer.
The RA was maintained according to manufacturer’s recommendations. The analyzer used an internal two-point calibration to measure pH, PCO2, Na, K, Cl, BUN, glucose, and Hct sensor slopes and to verify sensor performancef. The two-point calibration was automatically performed at 2 hr intervals. Hb was manually calibrated with two external calibrators once daily. The RA was in Mode A, so all analytes were checked with a one-point calibration every each sample analysis. The one-point calibration was performed by exposing the electrodes to a known standard and comparing the new value to the value obtained during the two-point calibration. All samples were handled by a licensed veterinarian trained to use the instruments (N.D.).
Analytes Measured
Analytic variables measured from both the PCA and RA during analysis included pH, PCO2, Na, K, Cl, BUN, glucose, and Hct. Hb was measured by the RA and calculated by the PCA.
The PCA used direct (undiluted) electrochemical methods. pH and PCO2 were measured by direct potentiometry, and Na, K, and Cl were measured by ion-selective electrode potentiometry. In the calculation of results for pH, PCO2, Na, K, and Cl, concentration was related to potential through the Nernst equation. Glucose was measured amperometrically. Oxidation of glucose, catalyzed by the enzyme glucose oxidase, produced hydrogen peroxide (H2O2). The liberated H2O2 was oxidized at the electrode to produce a current, which was proportional to the sample glucose concentration. Urea was hydrolyzed to ammonium ions in a reaction catalyzed by the enzyme urease. The ammonium ions were measured potentiometrically by an ion-selective electrode. In the calculation of results for urea, concentration was related to potential through the Nernst equation. Hct was determined conductometrically. The measured conductivity, after correction for electrolyte concentration, was inversely related to the Hct. The PCA provided a calculated Hb result that assumed a normal mean corpuscular Hb concentration in humans using the following equation: Hb (g/dL) = Hct (% packed cell volume) × 0.34.
The RA used in this study used direct (undiluted) electrochemical methods. pH was measured using a hydrogen ion-selective glass membrane. PCO2 was measured with a modified pH sensor composed of a gas-permeable membrane mounted on a combined measuring/reference electrode. Na, K, and Cl were measured by ion-selective electrode potentiometry. Glucose was measured based on the level of H2O2 produced during the enzymatic reaction among glucose and O2 in presence of the glucose oxidase enzyme. The H2O2 was oxidized by means of a constant potential of 0.7 Volts. The current produced was proportional to the sample glucose concentration. Urea was hydrolyzed to ammonium ions in a reaction catalyzed by the enzyme urease. The ammonium ions were measured potentiometrically by an ion-selective electrode. In the calculation of results for urea, concentration was related to potential through the Nernst equation. The Hct was determined by measuring electrical resistance of the blood sample and corrected for the concentration of Na. Hb was measured by combining a conductivity measurement and a photometric measurement.
Blood Sample Type and Handling
Two experienced operators performed venipuncture and analysis of samples. The rabbits were restrained in lateral recumbency, and venipuncture site over the left saphenous vein was cleaned with a 70% alcohol swab. Venipuncture and the subsequent analyses were performed by the same operator to minimize procedure time due to sample handling. Meanwhile, the second operator prevented bleeding and hematoma formation by applying gentle digital pressure on the venipuncture site.
Blood samples were collected with a 23-gauge needle and 1 mL plastic syringes containing 25 IU of dry balanced heparing and were immediately analyzed with the PCA. The whole blood sample was introduced into the cartridge via the heparinized syringe, and a portion of the same blood sample was analyzed with the conventional RA immediately thereafter. To ensure only a minimal delay in blood analysis, the analyzers were located immediately adjacent to each other and immediately adjacent to the table where blood collection was performed.
Sample collection and handling were performed according to the manufacturers’ guidelines to avoid preanalytical errors. The most important of those was the prevention of contamination of the syringe sample with air and the immediate and proper mixing of blood syringes before introducing each sample. Appropriate measures were taken to avoid touching the contact pads (which could have interfered with data transmission) or exerting pressure over the center of the cartridge (which could have caused premature release of the calibrating solution).
Sample Size Estimation
No previous reports of similar comparisons in rabbits are present in literature. Therefore, a pilot analysis of data was planned after the first 10 samples. The purpose of the analysis was to calculate the sample size necessary to obtain correlation coefficients presenting α error <0.01 and β error <0.01.27 The variables that were included in this pilot analysis were pH, PCO2, bicarbonate, and BUN. The lowest coefficient of correlation was 0.947 (r range, 0.947–0.983) for the PCO2. Thus, the minimum sample size was calculated to be 11 individuals. Sampling was planned on 30 individuals to include both healthy and diseased animals.
Statistical Analysis
Summary statistics were compiled for rabbits’ body temperature and measured variables. Data were analyzed for nonnormality by means of the D’Agostino-Pearson test. In the case of system failures of either single or multiple values, results of the remaining variables of those cartridges were included in the statistical analysis. Method comparison was performed with Bland-Altman bias plots and Passing-Bablok regression analysis. The Bland-Altman analysis estimates how much two methods differ in the quantitative measurement and therefore aids in making the decision if one method can be substituted for another. Bland-Altman plots were constructed for each variable by plotting the difference between the measurements on the vertical axis against the respective mean of those measurements on the horizontal axis. The limits of agreement were determined by ±1.96 standard deviations (SDs) centered on the mean difference. In addition, the association between the difference and the analyte concentration was examined by standard regression analysis of the difference between the two methods on their average. Briefly, a change in bias related to analyte concentration (proportional bias) was shown by the significant slope of the regression line. If bias increases as analyte concentration increased, the analysis described above would give limits of agreement that included most differences, but they would be wider apart than necessary for small values and rather narrower than they should be for large values.28,29 Under those circumstances, logarithmic (log) transformation of both measurements before analysis enabled the standard approach to be used. The limits of agreement derived from log transformed data can be back transformed to give limits for the ratio of the actual measurements.29 The log transformation is the only transformation giving back-transformed differences which are easy to interpret.28
Passing-Bablok regression, a nonparametric model, allowed measurement error (imprecision) in both methods, did not require the measurement error to be normally distributed, and was insensitive to outliers. First, linearity of data were examined by visual inspection of the scatter plot, and the coefficient of correlation r was calculated as a prerequisite of the Passing-Bablok regression.30,31 Secondly, the cusum linearity test was performed.30 The null hypothesis tested by the cusum test was a random distribution of residuals around the fitted regression line.32 Lastly, the regression equation was calculated. Constant bias was considered present if the 95% confidence interval (CI) for the y-intercept did not include the value 0. Similarly, proportional bias was considered present if the 95% CI for the slope did not include the value 1.30
Data were analyzed using commercial softwareh. Values were reported as mean ± SD (range), unless otherwise specified. Two tailed P < .05 were considered significant.
Results
Mean body temperature of the rabbits was 38.8 ± 0.7°C (37.1–40.2°C). Rabbits ranged from 1 to 10.5 yr of age (median age, 3 yr). There were 15 males and 15 females. Tentative diagnoses for diseased rabbits included dental disease (n = 4), encephalitozoonosis (n = 2), gastrointestinal stasis of unknown cause (n = 2), uterine tumor (n = 1), coccidiosis (n = 1), corneal perforation (n = 1), epiphora secondary to dental disease (n = 1), grade 3 pododermatitis (n = 1), intestinal obstruction (n = 1), liver lobe torsion (n = 1), open pneumothorax (n = 1), and syphilis (n = 1).
All samples were analyzed in both analyzers within 20 sec after blood collection. Analysis of pH, PCO2, Hb, Na, K, BUN, glucose, and Hct provided results in all 30 cartridges (100%), whereas system failures were noted in 3 cartridges (10%) for the Cl and consequently for the calculated anion gap. The system failures were observed at Cl values (as determined by the RA) of 108.5, 103.5, and 102 mmol/L. The ranges over which the variables were evaluated via the RA and PCA were presented in Tables 1 and 2, respectively.
*Median and 95% CI for the median were reported due to nonnormality of this variable.
BUN, blood urea nitrogen; CI, confidence interval; Hb, hemoglobin; Hct, hematocrit; ISE, ion selective electrodes; K, potassium; PCA, portable clinical analyzer; PCO2, partial pressure of CO2; RA, reference analyzer; SD, standard deviation.
*Median and 95% CI for the median were reported due to nonnormality of this variable.
CI, confidence interval; ISE, ion selective electrodes; K, potassium; SD, standard deviation.
The difference between PCA and RA measurements of each variable were displayed graphically as Bland-Altman agreement plots in Figure 1. Mean difference ± SD (95% CI) between RA and PCA were for pH, 0.007 ± 0.023 (−0.001–0.016); PCO2, −8.87 ± 2.42 (−9.78 to −7.97); Hb, 0.06 ± 0.92 (−0.28–0.4); Na, 0.19 ± 2.85 (−0.87–1.26); K, 0.015 ± 0.55 (−0.19–0.22); Cl, 1.17 ± 1.85 (0.44–1.91); BUN, 0.43 ± 3.34 (−0.81–1.68); glucose, 13.46 ± 13.13 (8.56–18.37); and Hct, 1.8 ± 2.59 (0.83–2.76). Limits of agreement for each variable were reported in Table 3. In the plots of PCO2, Cl-, glucose, and Hct, the line of equality was not included in the 95% CI of mean of differences (Figure 1), indicating a constant difference.
Citation: Journal of the American Animal Hospital Association 50, 5; 10.5326/JAAHA-MS-6114
*Variables presenting differences proportional to the mean underwent logarithmic transformation. Antilogs of the limits of agreement calculated on the log-transformed data are presented as a percentage.29
All P values for r were <.01.
CI, confidence interval; RSD, residual standard deviation.
The slope of the regression fit of the difference versus average was significant in the Bland-Altman plots of pH (−3.16 ± 0.61; P < .001), PCO2 (−2.21 ± 0.4; P < .001), Hb (−1.5 ± 0.17; P < .001), K (−0.95 ± 0.19; P < .001), Cl (−1.04 ± 0.33; P = .005), glucose (2.09 ± 0.42; P < .001), and Hct (−1.51 ± 0.21; P < .001). Thus, a proportional bias was present between the PCA and the RA for those values.
Results of the Passing-Bablok regression were listed in Table 3. That analysis demonstrated an at-least-proportional difference (slope ≠ 1) between pH, PCO2, Hb, Cl-, glucose, and Hct (Figure 2). An at-least-constant difference (y-intercept ≠ 0) was present between pH, Hb, K, Cl, BUN, glucose, and Hct. The cusum test for linearity confirmed that all compared variables fitted a linear model.
![FIGURE 2. Passing-Bablok regression for two selected analytes. The regression line is indicated by the solid line, with the CIs marked as dashed lines. The identity line (x = y) is indicated as the dotted line. A: K; presence of proportional bias. Notice the two values (rabbit 16 [PCA, 7.9 mmol/L; RA, 5.77 mmol/L], rabbit 22 [PCA, 7.1 mmol/L; RA, 5.42 mmol/L]) that exceeded the upper limit of the normal range with the PCA, but were inside the reference range according to the RA. B: Glucose, presence of constant and proportional bias.](/view/journals/aaha/50/5/305fig2.png)
![FIGURE 2. Passing-Bablok regression for two selected analytes. The regression line is indicated by the solid line, with the CIs marked as dashed lines. The identity line (x = y) is indicated as the dotted line. A: K; presence of proportional bias. Notice the two values (rabbit 16 [PCA, 7.9 mmol/L; RA, 5.77 mmol/L], rabbit 22 [PCA, 7.1 mmol/L; RA, 5.42 mmol/L]) that exceeded the upper limit of the normal range with the PCA, but were inside the reference range according to the RA. B: Glucose, presence of constant and proportional bias.](/view/journals/aaha/50/5/full-305fig2.png)
![FIGURE 2. Passing-Bablok regression for two selected analytes. The regression line is indicated by the solid line, with the CIs marked as dashed lines. The identity line (x = y) is indicated as the dotted line. A: K; presence of proportional bias. Notice the two values (rabbit 16 [PCA, 7.9 mmol/L; RA, 5.77 mmol/L], rabbit 22 [PCA, 7.1 mmol/L; RA, 5.42 mmol/L]) that exceeded the upper limit of the normal range with the PCA, but were inside the reference range according to the RA. B: Glucose, presence of constant and proportional bias.](/view/journals/aaha/50/5/inline-305fig2.png)
Citation: Journal of the American Animal Hospital Association 50, 5; 10.5326/JAAHA-MS-6114
Discussion
In the current study, agreement of blood gas analysis, selected electrolyte values, selected chemistries values, Hct, and Hb between the PCA and the RA was poor, the only exception made by pH. The inadequate agreement found in the current study it is not totally unexpected. In the veterinary literature, a poor agreement between PCAs and RAs for certain assays has already been described.16,21 Although the need to evaluate point-of-care units in exotic species has already been pointed out, to the best of the authors’ knowledge the present is the first report employing a point-of-care analyzer in the rabbit.11 In addition, no previous reports of blood gas analysis results in pet rabbits were found in the contemporary literature.
When comparing two methods the choice of acceptability limits becomes critical. As Jay (2011) suggests, “Limits of acceptability are contentious and there are no hard and fast rules - at least ones that you can get two clinical chemists to agree on.”33 Thus, acceptability of results is subjective and depends on the analytical performance needed by the operator. Nevertheless, agreement is an objective parameter that can be easily measured.28 Limits of agreement presented for each variable in this study permit clinicians and researchers to evaluate if the PCA is reliable enough for their scopes.
In the current study, the pH presented an acceptable agreement between instruments. Although pH measured by the PCA presented proportional and systematic differences with results obtained by the RA, such analytical fluctuations lead to clinical insignificant differences being the limits of agreement between −0.05 and 0.07. The interpretation of the clinical status of the rabbit would be adequate, even at extreme pH values (i.e., rabbit 16 [RA, 7.16; PCA, 7.11] and rabbit 8 [RA, 7.59; PCA, 7.65]).
The PCO2 analysis presented proportional bias, systematic bias, and wide limits of agreement (from −13.63 to −4.12 mm Hg). Although reference ranges for venous PCO2 in pet rabbits are lacking, the bias influenced clinical interpretation of PCO2 both at high and low pressures. Rabbits presenting PCO2 values determined by the RA between 43.1 and 34.8 mm Hg (i.e., values in the normal range for unanesthetized dogs and cats) yielded results >45 mm Hg by the PCA, which is considered outside the normal ranges for venous sample of dogs and cats.34 By contrast, most RA values were <30 mm Hg, indicating decreased PCO2 in dogs and cats resulted in PCA values inside the normal range for dogs and cats.34
Although in rabbits both hyponatremia and hypernatremia have been experimentally induced and corrected, Na disorders are rarely observed in pet rabbits’ clinical practice.35,36 In the current study, the agreement was measured on values of Na included in the normal range reported for the rabbit (i.e., 131–155 mmol/L).37 Although neither constant nor proportional biases were found, the agreement between instruments was poor (−5.40–5.79 mmol/L). That result was not unexpected considering that in venous blood samples from cattle the limits of agreement for Na of the PCA compared with a RA was found to be around −8–1 mmol/L.20 Nevertheless, the poor agreement would have never changed clinical decision basing on the current reference range for Na in rabbits.
K concentration in rabbits is evaluated for several purposes, among others it is suggested to relate to kidney disease.38 Nevertheless, true hyperkalemia seems to be a rare condition in pet rabbits.38 One case report described a pet rabbit presenting with generalized weakness, tetraparesis, ventroflexion of the neck; and decreased postural reactions, proprioception, and spinal reflexes in the four limb, concurrent with hyperkalemia.39 In that case, correction of K imbalance led to resolution of the symptoms. Rabbits in which acute hyperkalemia was experimentally induced developed weakness, increase in respiration, tremors, and occasionally urination. Untreated animals developed severe convulsions and ventricular fibrillation.40 Therefore, K disorders in rabbits, as in other animals, need a prompt diagnosis and adequate treatment.
All results obtained by the RA were inside the upper reference range limit for healthy rabbits (6.9 mmol/L).37 In contrast, the PCA gave two samples (i.e., rabbit 16 [PCA, 7.9 mmol/L; RA, 5.77 mmol/L] and rabbit 22 [PCA, 7.1 mmol/L; RA, 5.42 mmol/L]), exceeding the upper limit of the reference range. Therefore, the use of the PCA would have overdiagnosed hyperkalemia. Hypokalemia was not diagnosed in any rabbit of the current study, with the exception of rabbit 13 in which K was below the lower reference range limit (3.6 mmol/L) by both analyzers.37 According to the Bland-Altman procedure, the difference between the PCA and the RA for K concentrations was proportional (i.e., varied with the magnitude of measurement). According to the Passing-Bablok analysis, the K assay did not show a proportional difference. That analytical difference may be secondary to the presence of two extreme values for which the Passing-Bablok nonparametric procedure is not affected.
The Cl was slightly overestimated by the PCA (mean difference, 1.17; 95% CI, 0.44–1.91). The overestimation led to potential clinical misinterpretation of Cl values at the upper limit of the normal range. Therefore, rabbit 23 (RA, 111 mmol/L; PCA, 113 mmol/L) and rabbit 7 (RA, 111.6 mmol/L; PCA, 114 mmol/L) exceeded the reference range (92–112 mmol/L) with the PCA, but were inside the range according to the RA.37
Strong evidence of the clinical value of BUN measurement during either kidney or liver disorders in pet rabbits is still lacking. Nevertheless, BUN is often measured in clinical practice as an aid in the evaluation of renal and hepatic function.38 In the current study, no proportional or systematic bias were found with the Bland-Altman plot of the BUN, even if the limits of agreement were wide (−6.13–6.99 mg/dL). Most BUN values obtained in the current study were inside the normal range for rabbits (i.e., between 13–29 mg/dL).37 The two instruments agreed in detecting two rabbits outside the upper reference range limits, although an important difference was present in one of those (PCA, 34 mg/dL; RA, 41 mg/dL). The PCA found five rabbits to present BUN values outside the lower reference range limits. Of those five rabbits, only two rabbits were below the BUN limits according the results of the RA.
Similarly, glucose was mainly underestimated by the PCA. In some cases, reliance on results of the PCA would have resulted in alteration of clinical decisions. There is some recent evidence that extremely high glucose values (i.e., around 390 mg/dL) in pet rabbits presenting gut stasis could be indicative of intestinal obstruction, necessitating surgical intervention.41 In such cases, glucose underestimation can influence decision-making and patient outcome. Again, hypoglycemia could be overdiagnosed, leading to erroneous clinical suspicion and stressful, time-consuming tests (e.g., pancreas ultrasound for suspected insulin-secreting tumors or pancreatitis).41 Nevertheless, in the current study, minimum glucose value was of 117 mg/dL, so response of PCA to low glucose values in blood samples can only be inferred. It should be noticed that standards in human medicine recommend that the accuracy of a PCA be within 15% of the reference value.42
Evaluation of hematologic parameter (Hct and Hb) in pet rabbits provides useful clinical information in rabbits, as in other species.43 In the current study, only one rabbit presented with anemia (i.e., Hb concentration <10 g/dL) according to the RA, whereas three rabbits would be diagnosed to have anemia according to PCA results.37 Similarly, the Hct was mainly underestimated by the PCA, provoking overdiagnosis of low Hct (i.e., Hct < 33%).37 That said, low Hct would be diagnosed in eight rabbits via PCA results, but only one rabbit was confirmed to have a Hct <33% via the RA.
In the current study, rabbits suffering several disease conditions were sampled. Some previous studies comparing the performances of a PCA with a RA focused on either healthy or presumed healthy rabbits.20,25 Other studies were centered on exercising individuals or on animals in undetermined health conditions.19,44,45 Ideally, the specimens to be tested should cover a wide range of analyte concentrations to make valid use of linear regression analysis from which the best estimate of constant and proportional errors could be obtained.46 Therefore, inclusion of diseased animals permitted a wider range of the variables to be evaluated.
The statistical analysis in research papers reporting comparisons of methods of measuring the same quantity is often inappropriate, with the correlation coefficient r often used in studies of method comparison.47 The deficiencies of r in method comparison have been thoroughly described in medical literature.47,48 Although correlation coefficient is proved to not respond satisfactorily to the agreement question, it is still used in instrument comparison.29,49,50 In the current study, the authors calculated the Bland-Altman limits of agreement using the mean of the measurements by the two methods as the best estimate.29 A regression analysis was also used to interpret the data because it was widely known and its interpretation was often more accessible. In previous comparative studies between point-of-care analyzers and bench-top analyzers, deming regression was used to assess the degree of agreement.20,25,51 Whereas the ordinary linear regression method assumes that only the y measurements were associated with random measurement errors, the Deming regression takes measurement errors for both methods into accounts.52 The nonparametric procedure of Passing-Bablok was selected over the Deming regression because shows reliable results in all of most situations.31 Because of the robustness of that procedure, the problem of either including or excluding extreme data points (outliers) did not arise.31 It should be considered that outliers in comparison method studies were not necessarily gross measurement errors. Instead, they could have been caused by different properties of the methods with respect to either specificity or susceptibility to interferences. Therefore, they should not be removed from the calculation without experimental reason.31
Conclusion
The current study provides sufficient evidence that the PCA is acceptable for clinical pH measurement in venous blood samples of pet rabbits. Nevertheless, considering the bias between the PCA and RA for the other variables, instrument-specific reference intervals are needed and a single analyzer should be used if the intent is to evaluate serial results in a patient over time. Furthermore, researchers aware of the limits of agreement for each variable of the cartridge can decide if it is reliable enough for their scopes.
Bland-Altman agreement plots for venous blood samples collected from pet rabbits and analyzed with a portable clinical analyzer (PCA) and a reference analyzer (RA). The middle solid horizontal line represents the mean difference between the pairs of measurements, essentially representing the mean “measurement error.” The upper and lower horizontal dashed lines represent the 95% limits of agreement (i.e., the range of measurement errors that would occur on 95% of occasions). The 95% confidence intervals (CIs) of mean of differences are depicted as dashed and dotted lines. If the 95% CIs of mean of differences do not include the 0 value, there is a significant systematic difference. Regression line with its 95% CI is depicted to assist the detection of a proportional difference.
Passing-Bablok regression for two selected analytes. The regression line is indicated by the solid line, with the CIs marked as dashed lines. The identity line (x = y) is indicated as the dotted line. A: K; presence of proportional bias. Notice the two values (rabbit 16 [PCA, 7.9 mmol/L; RA, 5.77 mmol/L], rabbit 22 [PCA, 7.1 mmol/L; RA, 5.42 mmol/L]) that exceeded the upper limit of the normal range with the PCA, but were inside the reference range according to the RA. B: Glucose, presence of constant and proportional bias.
Contributor Notes
