Data of 362 canine complete blood cell counts (CBCs) including 14 follow-up samples was analyzed retrospectively. Follow-up samples are defined as blood samples, which have been taken from the same diseased dog but at different dates. The blood samples have been collected between July 2006 and April 2007 in the department of veterinary clinical sciences, Giessen, Germany and between June 2009 and February 2010 in the department of Global Non-Clinical Safety, Merck Serono, Darmstadt, Germany. Overall, 169/362 samples were obtained from diseased dogs and 193/362 samples were taken from healthy dogs respectively.
Analysis of r-PLTs and PLT parameters
Measurement was performed with ethylenediamine tetra-acetic acid (EDTA) anticoagulated blood samples using the XT-2000iV hematology analyzer and the veterinary software version 00–08 and 00–10. The retrospective analysis and gating was performed using the veterinary software version 00–11. To detect the number of r-PLTs in the blood samples, we created one gate to quantify the whole platelet population (Fig. 1 and 2, PLT-O total) and another gate located inside the first gate (Figs. 1 and 2, red gate) containing the r-PLTs. A modified version of the “Gelain-gate” was created to optimize the separation between mature and r-PLTs as well as to exclude erythrocytes and reticulocytes. This modified gate (“Oellers-gate”, Fig. 1c) was compared with the gates published by Pankraz et al. and Gelain et al. (“Pankraz-gate” Fig. 1b and Fig. 2b and e, “Gelain-gate” Fig. 1a and Fig. 2a and d). The risk of small erythrocytes and/or reticulocytes causing clinically important interferences and thus a false r-PLT count was given when gate 4 (Fig. 1) contained more than 5 % of the sum of dots measured in gates 1 and 4. The suspicion of potential interferences of small erythrocytes/reticulocytes with the r-PLT count was confirmed by visual control of the matching scattergram.
Retrospective analysis of each CBC included the following variables: platelet count by optical (PLT-O) and impedance methods (PLT-I), MPV, platelet distribution width (PDW), platelet large cell ratio (P-LCR), and plateletcrit (PCT).
The analyzer-specific canine reference intervals established previously for the XT-2000iV [19] were used as cut-off values to define anemia or regeneration of erythrocytes (i.e., anemia was diagnosed when the RBC count was < 5.1 × 1012/L and a reticulocyte count > 150.1 × 109/L was considered regenerative).
Reference population
The CBCs obtained from healthy dogs (n = 193, one sample per dog) included mainly samples taken from healthy Beagle dogs (153/193) obtained from one breeder (Marshall BioResources, USA). The Beagle dogs were kept under defined housing conditions: For 20 h a day, they were housed in groups of three to five dogs. Each dog spent the remaining four hours separated from the others in a standard kennel of 6.54 m2 and natural light. The Beagle dogs were fed once a day with 300 g of a commercially available pelleted dry food (Provimi Kliba AG, Kaiseraugust, Switzerland). Water was offered ad libitum. All Beagle dogs were dewormed and vaccinated against canine distemper, leptospirosis, canine parvovirus, parainfluenza, Bordetella, canine adenovirus 2, and rabies.
The other 40/193 healthy dogs kept as pet dogs were of various breeds including mixed-breed dogs (n = 17), Border Collies (n = 6), Great Danes (n = 3), West Highland White Terrier (n = 2), and one Beagle dog, Berger Blanc Suisse, Cocker Spaniel, Dalmatian, German Shepherd, Golden Retriever, Groenendael, Jack Russel Terrier, Labrador Retriever, Old German Shepherd Dog, Rough Collie, and Small Munsterlander each.
Statistical analysis
Data was analyzed using Microsoft Excel®, Reference Value Adviser© version 2.1 for Microsoft Excel®, Analyse-it® version 2.04 for Microsoft Excel®, and GraphPad Prism® version 6. Data distribution was assessed visually with a histogram.
Spearman’s rank correlation, Passing-Bablok-Analysis, and Bland-Altman-Analysis were performed to determine correlation and bias between r-PLTs measurements obtained with the various gates. A Kruskal-Wallis-Test was used to assess the impact of the gating method on r-PLT results, whereby the P values were adapted for multiple comparisons. Level of significance was set at P < 0.05.
Intra-assay repeatability was assessed by 25 repeated measurements of one blood sample. For platelet indices, the XT-2000iV did not report results in 18/25 measurements so that calculation of intra-assay CV was based on 7 results.
Reference intervals for r-PLTs and platelet indices were generated for all healthy dogs independent of the breed (n = 193), and also for the subgroups “Beagles” (n = 153) and “non-Beagles” (n = 40). As recommended by the American Society of Veterinary Clinical Pathology (ASVCP) [11], the statistical method used for calculation of the respective reference intervals was chosen based on the number of dogs included in the respective reference population and data distribution. The whole group of healthy dogs (n = 193) and the group of Beagle dogs (n = 153) included ≥ 120 individuals, so that the nonparametric method was chosen to calculate reference intervals [11]. The group of pet dogs contained 40 (r-PLT, PLT-O, PLT-I) and 38 (MPV, PDW, P-LCR, PCT) dogs, respectively. When the number of samples was 40 ≤ × ≤ 120, the robust method was chosen for symmetrical (Gaussian) data distribution [11]. In case of non-Gaussian distribution, a Box-Cox transformation was performed prior to calculation of reference intervals. Depending on data distribution after Box-Cox transformation, a the parametric method was chosen for data with Gaussian distribution, while a nonparametric method was used in case of non-Gaussian distribution. As the XT-2000iV did not report MPV, PDW, P-LCR and PCT for 3/193 healthy dogs, data of merely 38/40 samples was available in the group of “non-Beagles” for these variables. As recommended previously for reference populations of < 40 individuals [11], the parametric method and robust method were used in case of Gaussian and non-Gaussian distribution, respectively. For all reference intervals, the 90 % confidence interval (CI) of the upper and lower ends of the reference limits was calculated using bootstrap methods.