Comparison of biochemical and hematologic values obtained via jugular venipuncture and peripheral intravenous catheters in dogs

Abstract Background Sampling from a peripheral intravenous catheter (PIVC) might be a more efficient and less traumatic collection of blood for serum biochemistry (SB) or CBC than direct venipuncture (DV). Agreement between results of samples obtained by these methods has not been evaluated in dogs. Objectives The primary objectives were to determine whether sampling from PIVC could be used in place of DV for dogs. We hypothesized DV and PIVC samples would have clinically equivalent SB and CBC results. Animals Sixty‐one client‐owned dogs were included in each study arm. Methods This was a partially randomized method‐comparison study. Paired DV and PIVC samples obtained within 1 to 2 minutes after, or approximately 24 hours after, placement of a PIVC in a cephalic vein were evaluated for agreement and bias using percentage difference plots (with a priori application of consensus total allowable error), Bland‐Altman analysis, Passing‐Bablok regression analysis, Wilcoxon signed rank test, and McNemar's test. Results There was statistically and clinically acceptable agreement and no bias between sampling methods for the majority of results. Analytes with the most frequent disagreement were aspartate aminotransferase, total bilirubin, potassium, bicarbonate, and leukocyte differential counts, as well as red blood cell count, hemoglobin, hematocrit, and packed cell volume in the hospitalized PIVC sampling group. Few observed differences would change clinical decision making. Conclusions and Clinical Importance PIVC sampling can provide generally acceptable SB and CBC results for most dogs, but clinicians should be aware of a few values for which disparate results might occasionally be obtained.


| INTRODUCTION
For many dogs presented for medical evaluation, it could be necessary to collect a blood sample to perform serum biochemistry (SB) and CBC. It is possible to obtain blood samples directly from a peripheral intravenous catheter (PIVC), either freshly placed (fPIVC) or after a period of hospitalization (hPIVC) 1,2 but whether these samples provide reliable SB and CBC results is unclear.
In humans, collection of blood samples from PIVC rather than direct venipuncture (DV) is recommended for patients who are pediatric, have difficult vascular access, have coagulopathies, or need repeat testing. 3,4 Using PIVC sampling limits patient discomfort, decreases needle injuries, and decreases ecchymoses compared to DV. 5,6 Human studies evaluating SB and CBC found no clinically important or statistically significant differences between collection methods for most or all analytes, 4,[7][8][9][10][11][12] though some studies found clinically important or statistically significant differences for bicarbonate (HCO 3 ), 8,13 potassium (K), 10,12,13 or glucose (GLU). 6,13 Clinical confidence in PIVC blood sampling for SB and CBC might reduce the need for DV in dogs. DV commonly requires physical restraint of the dog and utilizes 2 or more people, whereas collecting blood from a PIVC may require minimal restraint and a single individual. In dogs with cardiovascular compromise, respiratory distress, coagulopathies, small size, or fractious behavior, DV could be difficult, time-consuming, or dangerous for the dog and staff. Obtaining blood from PIVC rather than DV could help improve staff efficiency as well as the dog's safety and comfort. 5 The primary objectives of this study were to determine whether sampling from fPIVC or hPIVC approximately 1 day after placement can reliably be used for SB and CBC testing. The hypothesis was that SB and CBC results from fPIVC and hPIVC would be clinically equivalent to those from DV.

| Study design and groups
This was a 2-arm prospective clinical study involving client-owned dogs. The first arm was randomized and involved comparison of SB and CBC results from samples collected by DV and from fPIVC. The

| Procedures
Direct venipuncture, PIVC placement, and PIVC blood collection was performed by one author (A. L. Guarino) or a licensed veterinary technician, and all procedures were witnessed by one of the authors. For both arms, DV and PIVC blood collection for paired samples were performed contemporaneously. In arm 1, the order of DV and fPIVC placement/ sampling was randomized using a random number generator. Order of sampling was not randomized in arm 2 to prevent unnecessary DV for dogs in which the hPIVC did not provide an adequate blood sample.
For DV, a 6-mL syringe with a 20-gauge needle was used to collect 4-6 mL of blood from a jugular vein with a "clean stick," defined as no redirections of the needle once inserted into the vein and blood observed in the needle hub. Redirections through the skin were allowed. After successful DV, 1.3 mL of blood were expelled into an uncapped micro ethylenediaminetetraacetic acid tube (Sarstedt Inc, Newton, North Carolina) 14  For fPIVC samples, the area over a cephalic vein was clipped and sterilely prepared and a 20-gauge, 1.25 00 over-the-needle catheter (Terumo Medical Corporation, Somerset, New Jersey) was placed and secured with tape. Before flushing with saline, a 6-mL syringe was attached to the catheter hub and 4 to 6 mL of blood were withdrawn and placed into tubes as previously described. If blood flow was slow, 2 syringes were used to collect 3 mL at a time instead. To assist blood collection, the vein proximal to the fPIVC was occluded with pressure from a finger.
For hPIVC, patency was assessed with a saline flush 24 ± 4 hours from time of placement. Blood collection from hPIVC was similar to a previously described study technique. 15 All IV fluids and medications administered via the hPIVC were discontinued for 5 minutes before sampling. The t-set clamp was adjusted to be as close as possible to the t-set hub. After a 1 mL waste sample of blood was removed and discarded, blood for SB and CBC was collected as described above for fPIVC except that a 20-gauge needle was attached to the syringe(s) and inserted into the injection port. Waste sample volume was determined by using saline to measure dead space from a 20-gauge, 1.25 00 catheter and attached t-set hub. This volume (0.23 mL) was multiplied by 300% and rounded up to obtain a waste volume of 1 mL. 15

| Laboratory analysis
Samples were submitted to the in-hospital clinical pathology laboratory within 30 minutes after collection. Trained laboratory staff processed samples by routine methods for SB and CBC analysis. Samples from each pair were processed and analyzed together. Blood cell counts were assessed using an Advia 2120 Hematology Analyzer (Siemens Healthcare Diagnostics, Tarrytown, New York) and SB panels were performed using an AU480 biochemistry analyzer (AU480) (Beckman Coulter, Inc, Tokyo, Japan). Values below the linearity of the AU480 were excluded. Values above the linearity of the AU480 were reported after dilution. One author (S. S. K. Beatty, a clinical pathologist), masked to sample source, performed a manual blood film review of all hematology slides to confirm the leukocyte differential counts (DIFF) and describe cell morphology.
Quantitative values, ranging from 0 to 6, for lipemia, icterus, and hemolysis (LIH) were provided as part of the AU480 output. These were converted to binary values of 0 (AU480 output of 0) or 1 (AU480 output of 1-6) to indicate absence or presence, respectively, for statistical analysis. If quantitative values were not available, the lab technician's gross assessment (absence or presence of LIH) was used. Missing data points were excluded.  In those cases, the mid-point of the estimated range was reported. If there was clumping noted with a normal or increased automated PLT, the automated PLT was still reported as a minimum value for PLT.
On blood smear review, platelets were characterized as "normal/ adequate," "decreased," or "increased," as well as "clumped" or "not clumped." For statistical analysis, red blood cell morphology changes, leukocyte toxicity, and large or clumped platelets were converted to binary values (0 for absence or 1 for presence of the morphologic change). To be considered "present," the abnormality had to have been noted at least once per every other 100Â microscopic field.
Acanthocytes, echinocytes, and keratocytes were also reported on a conventional 1+ to 4+ scale for statistical analysis.
Quality control assessment of analyzer variability and ongoing performance were assessed as described in Data S1, Supporting Information.

| Statistical analysis
Descriptive statistics were reported for each analyte, demographic data, and number of hours between hPIVC placement and sampling. Percentage difference plots, inspired by Bland-Altman plots, were used to assess for significant differences between sampling methods.
The difference between sample pair values divided by the mean of the values was calculated for each sample pair. Differences between sample pair values were classified as significant if greater than previously published cTEa values applied as the bounds of acceptable agreement. 16,17 Bland-Altman analysis was also used to assess bias for measured analytes. For analytes with normally distributed differences, bias was estimated from the mean of the differences using a paired samples t test. For analytes with non-normally distributed differences, bias was estimated from the median of the differences using a Wilcoxon signed rank sum test. A P-value ≤.05 was considered significant. Additionally, constant and proportional bias were estimated using Passing-Bablok regression analysis. Constant bias was considered statistically significant if the 95% confidence intervals for the intercept did not include 0. Proportional bias was considered statistically significant if the 95% confidence intervals for the slope did not include 1. Statistically significant biases were reviewed subjectively to determine clinical significance.
F I G U R E 1 Selected percentage plots, inspired from Bland-Altman plots, of serum biochemistry data with a priori bounds of acceptable agreement based on cTEa listed in Table 1 applied. 16 The zero line indicates 0% difference between the paired samples.  Forty-six dogs were clinically ill, 7 had nonurgent conditions necessitating elective procedures (such as surgical correction of an angular limb deformity), and 8 were healthy and undergoing routine procedures such as ovariohysterectomy.

| Biochemistry
The percentage of sample pairs within cTEa bounds for each analyte is reported in Table 1  Descriptive statistics for biochemistry data and linearity of the AU480 are reported in Table S4. Normality assessments for biochemistry data are reported in Table S5.
Bias between sampling methods is reported in Table 2. There was statistically significant bias for several analytes, but only the bias for glucose was potentially clinically important.

| Hematology
The percentage of sample pairs within cTEa bounds for each analyte is reported in Table 3. There was at least 1 disagreeing sample pair for almost all CBC parameters, but only the DIFF had less than 95%  Indicates that bias (median of the differences for non-normally distributed differences) determined by Bland-Altman analysis was statistically significant (P ≤ .05) by a Wilcoxon signed rank sum test. Bias is reported with 95% confidence intervals from the median. Significant values are bolded.
c Indicates that bias determined by Passing-Bablok regression was considered statistically significant because the 95% confidence intervals did not include 0 (for constant bias) or 1 (for proportional bias). Significant values are bolded. samples within cTEa (Table S6). Subjectively, very few differences would have affected clinical decision-making. Descriptive statistics for CBC data are reported in Table S7. Normality assessments for CBC data are reported in Table S8.
Bias between sampling methods is reported in Table 4. There was statistically significant bias for several analytes, none of which were considered clinically important.
A summary of blood cell morphologic characterization is reported in Table 5. There was a significant difference (P = .05) in the presence of acanthocytes between DV and fPIVC samples (more frequent acanthocytes in the fPIVC samples) when the data was reported on a 0 to 4+ scale, but this statistical significance did not persist when the data was converted to binary terms (presence or absence) (P = .06).
There were no other statistically significant differences in red blood cell morphology. One sample pair had "normal" platelets in the DV sample and "increased" platelets in the fPIVC sample.

| Biochemistry
The percentage of sample pairs within cTEA bounds for each analyte is reported in Table 1. Analytes with at least 1 disagreeing sample pair included ALP, AST, TBIL, CA, CREA, BUN, GLU, MG, K, and HCO 3 (Table S3). Each disagreeing sample pair was subjectively assessed individually to determine the effect on clinical decision-making. There were no decision-altering pairs for ALP, CA, CREA, BUN, MG, or K. Descriptive statistics for biochemistry data are reported in Table S4.
Normality assessments for biochemistry data are reported in Table S5.
Bias between sampling methods is reported in Table 2. There was statistically significant bias for several analytes, but only bias for glucose was considered potentially clinically important.

| Hematology
The percentage of sample pairs within cTEA bounds for each analyte is reported in Table 3. There was at least 1 disagreeing sample pair for almost all CBC parameters (Table S6). The DIFF, RBC, HB, HCT, PCV, and PLT had less than 95% samples within cTEa. Subjectively, few differences would have affected clinical decision-making. Descriptive statistics for CBC data are reported in Table S7. Normality assessments for CBC data are reported in Table S8.
Bias between sampling methods is reported in Table 4. There was statistically, and possibly clinically, significant proportional bias (Passing-Bablok regression) for BAND ( Figure 2B). The remainder of the statistically significant biases were not deemed to be clinically important. There were no other statistically significant differences in red blood cell, leukocyte, or platelet morphology ( Many disagreements between sample pairs were likely related to increments of measurement, particularly at the lower ends of measurement scales. For example, the SB analyte with the greatest proportion of sample pair differences outside cTEa was TBIL, including 18% and 28% for fPIVC ( Figure 1C) and hPIVC groups, respectively.
Because of the small magnitude of TBIL measurements in all of these sample pairs relative to the measured increments, differences are magnified when calculating error. A difference between TBIL measurement of 0.1 or 0.2 mg/dL is unlikely to change clinical decisionmaking; however, the error for this sample pair is well outside cTEa of 25%. 16  There was statistically and potentially clinically significant bias for paired GLU measurements (Figures 2A and 3A) 6 In that study, fluids, including dextrose-containing solutions, were only discontinued for 1 minute before sample collection. Another study of adult humans found GLU was 1.7 mg/dL higher on average in DV samples than PIVC samples 2 minutes after a saline bolus through the PIVC. 13 The authors deemed glucose measurements between sampling methods were not equivalent.
However, the reported laboratory error of 2.4 mg/dL was greater than the mean difference, making this data difficult to interpret. 13 There was also statistically and possibly clinically significant proportional bias for BAND from hPIVC samples, with more BAND identified in DV than PIVC samples ( Figure 2B). This bias was not found for fPIVC samples. The hPIVC BAND bias might be secondary to differences in the ability of these cells to pass through hPIVC because of kinks, fibrin deposition, or both. Differences in marginalization of BAND between jugular and cephalic veins are less likely because bias was not present for fPIVC, and differences in BAND concentrations between jugular and cephalic venous samples was not identified in a previous study of dogs. 21 There were no significant differences in presence of lipemia or hemolysis between sampling methods. A limitation is that the AU480 is not validated to quantify the degree of lipemia, hemolysis, or icterus in canine samples. It is possible that a relationship between the degree of interfering substances, especially HB, could exist between sampling methods. However, this study was not designed to detect this difference. Another limitation was the inability to statistically analyze the effect of hemolysis on abnormal biochemistry values. This was because of the lack of validated quantitative LIH data and the low number of disagreeing pairs, which might have resulted in type II error.
Interestingly, the only significant difference in RBC morphology between DV and PIVC sampling methods was for acanthocytes from fPIVC samples. This statistically significant difference did not persist for hPIVC. It is possible that the fPIVC acanthocyte difference represents type I error, or the lack of acanthocyte difference for hPIVC represents type II error. While acanthocytes could form secondary to fragmentation injury as RBCs travel through the fPIVC, there were no other indications of increased shear forces.
It is difficult to determine an appropriate number of samples to adequately power a study utilizing Bland-Altman analysis, but a higher number of paired samples provides more confidence for conclusions about whether there is clinically acceptable agreement. 18 To maintain adequate power (when α = .05 and β = 80%) with 1 disagreeing sample pair, the recommended sample size is n ≥ 59. 18 For the analytes with 2 to 3 pair sample disagreements in this study, a larger sample size might more cleanly determine whether we should conclude that the sampling methods are clinically equivalent or not.